Gene Knockdown by EpCAM Aptamer–siRNA Chimeras Suppresses Epithelial Breast Cancers and Their Tumor-Initiating Cells

RNA interference (RNAi) offers the opportunity to treat disease by knocking down disease-causing genes (1). Recent early-phase clinical trials have shown vigorous (75%–95%), sustained (lasting up to several months) and safe knockdown of a handful of gene targets in the liver using lipid nanoparticle-encapsulated or GalNAc-conjugated siRNAs (2–5). The liver, the body's major filtering organ, traps particles and, hence, is relatively easy to transfect. The major obstacle to harnessing RNAi for treating most diseases, however, has yet to be solved, namely efficient delivery of small RNAs and gene knockdown in cells beyond the liver. In particular, the delivery roadblock is a major obstacle to harnessing RNAi to treat cancer (6).

Triple-negative breast cancer (TNBC), a heterogeneous group of poorly differentiated cancers defined by the lack of estrogen, progesterone, and Her2 receptor expression, has the worst prognosis of any breast cancer subtype (7–9). Most TNBCs have epithelial properties and are classified as basal-like, although a sizable minority are mesenchymal. TNBC afflicts younger women and is the subtype associated with BRCA1 genetic mutations. No targeted therapy is available. Although most TNBC patients respond to chemotherapy, within 3 years about a third develop metastases and eventually die. Thus, new approaches are needed.

Here we develop a flexible, targeted platform for gene knockdown and treatment of basal-like TNBCs that might also be suitable for therapy against most of the common (epithelial) cancers. We deliver siRNAs into epithelial cancer cells by linking them to an RNA aptamer that binds to EpCAM, the first described tumor antigen, a cell surface receptor overexpressed on epithelial cancers, including basal-like TNBCs. Aptamer-linked siRNAs, known as aptamer–siRNA chimeras (AsiC), have been used in small animal models to treat prostate cancer and prevent HIV infection (10–18). We chose EpCAM for targeting basal-like TNBC because EpCAM is highly expressed on all epithelial cancers. A high affinity EpCAM aptamer was previously identified (19). EpCAM also marks tumor-initiating cells (TIC, also known as cancer stem cells; refs. 20–27). Although the cancer stem cell hypothesis may not apply to all tumors, most solid cancers are heterogeneous and the less differentiated TIC subpopulation is probably responsible for initiating tumors, resistant to conventional cytotoxic drugs and responsible for recurrence and metastases. Devising therapies to eliminate TICs is an important unmet goal of cancer research (28).

In normal epithelia, EpCAM is only weakly expressed on basolateral gap junctions, where it may not be accessible to drugs (29). In epithelial cancers, it is not only more abundant (by several orders of magnitude), but is also distributed along the cell membrane. Ligation of EpCAM promotes adhesion and enhances cell proliferation and invasivity. Proteolytic cleavage of EpCAM releases an intracellular fragment that increases stem cell factor transcription (30, 31). EpCAM's oncogenic properties may make it difficult for tumor cells to develop resistance by downmodulating EpCAM. In one study, about 2/3 of TNBCs, presumably the basal-like subtype, stained strongly for EpCAM (25). The number of EpCAM⁺ circulating cells is linked to poor prognosis in breast cancer (32–36). An EpCAM antibody has been evaluated clinically for epithelial cancers, but had limited effectiveness on its own (37–39). EpCAM expression identifies circulating tumor cells in an FDA-approved method for monitoring metastatic breast, colon, and prostate cancer treatment (32–36). Moreover, about 97% of human breast cancers and virtually 100% of other common epithelial cancers, including lung, colon, pancreas, and prostate, stain brightly for EpCAM (23), suggesting that the platform developed here could be adapted for RNAi-based therapy of common solid tumors.

Here we show that epithelial breast cancer cell lines uniformly stain brightly for EpCAM, while immortalized normal breast epithelial cells, fibroblasts, and mesenchymal tumor cell lines do not. EpCAM-AsiCs cause targeted gene knockdown in luminal and basal-A TNBC cancer cell lines (which resemble basal-like TNBC primary tumors), and human breast cancer tissues in vitro, but not in normal epithelial cells, basal-B TNBC cell lines (which resemble mesenchymal TNBC primary tumors) or normal human breast tissues. Knockdown is proportional to EpCAM expression. Moreover, EpCAM-AsiC–mediated knockdown of PLK1, a gene required for mitosis, suppresses in vitro TIC functional assays (colony and mammosphere formation) of epithelial breast cancer lines. Ex vivo treatment specifically abrogates tumor initiation. Subcutaneously injected PLK1 EpCAM-AsiCs are taken up specifically by EpCAM⁺ xenografts of poor prognosis basal-A and Her2 breast cancers and cause rapid tumor regression.

MDA-MB-468 (MB468) cells transduced with a Firefly luciferase reporter (MB468-luc) were kindly provided by Andrew Kung (Columbia University, New York NY). BPE and BPLER cells, provided by Tan Ince (University of Miami, Miami, FL), were maintained in supplemented WIT-T medium (Stemgent). Other human cell lines were obtained from ATCC and grown in MEM (MCF7, BT474), McCoy's 5A (SKBR3), RPMI1640 (HCC1806, HCC1143, HCC1937, HCC1954, HCC1187, MB468, T47D), or DMEM (MB231, BT549, MB436) media supplemented with 10% FBS (Gemini Bioproducts), 100 U/mL penicillin G and 100 μg/mL streptomycin sulfate, 6 mmol/L HEPES, 1.6 mmol/L l-glutamine, 50 μmol/L β-mercaptoethanol (Sigma-Aldrich). MCF10CA1a cells, provided by Karmanos Cancer Institute (Detroit, MI), were grown in supplemented DMEM. MB231 cells stably expressing Firefly luciferase and mCherry (MB231-luc-mCherry) were selected after infection with pLV-Fluc-mCherry-Puro lentivirus (provided by Andrew Kung) using puromycin. The authors have not authenticated the cell lines used. Cell lines were obtained in 2012 and were expanded briefly before aliquoting and freezing. Cells used for experiments were thawed from low passage aliquots. All cell lines were verified to be free of mycoplasma.

The long strand of the AsiC synthesized with 2′-fluoropyrimidines (TriLink Biotechnologies) was annealed to the short antisense strand (Integrated DNA Technologies) using a 2-fold molar excess of the short strand. The long strand was heated to 95°C for 10 minutes before adding the short strand that was annealed at 65°C for 7 minutes. The mixture was allowed to cool at room temperature for 20 minutes. The annealed AsiC duplexes were purified further using Illustra MicroSpin G-25 columns (GE Healthcare Life Sciences). siRNAs, conjugated or not with cholesterol on the 3′ end as described (15), were from IDT. RNA sequences are provided in Supplementary Table S1. For some experiments the 3′ end of the antisense strand was conjugated to Cy3, Alexa Fluor 647 or 750. The stability of RNAs incubated in human or mouse serum was assessed with aliquots removed during a 36-hour incubation at 37°C in 50% serum. Samples were analyzed by densitometry after polyacrylamide gel electrophoresis and ethidium bromide staining.

Cells were incubated with 1 μmol/L Cy3-labeled EpCAM aptamer in microscopy chamber wells for indicated times and then counter stained with CellMask Deep Red Plasma Membrane Stain (Life Technologies) and imaged live. Images were acquired with a spinning disk confocal head (Yokogawa) coupled to a fully motorized epifluorescence microscope (Axio Observer) equipped with a 63× lens (Plan Apochromat, 1.4 NA, Carl Zeiss). Three 50 mW solid-state lasers (491, 561, and 660 nm; Cobalt Laser) were used as light sources. The imaging system operates under control of SlideBook 5.0 and an EM-CCD camera (Quant-EM, Hamamatsu) was used to acquire images.

For in vitro gene silencing experiments, cells were used immediately after plating at 10,000 cells per well in 96-well plates. For transfections, cells were incubated with lipoplexed siRNAs or AsiCs without lipid at 100 nmol/L or 4 μmol/L concentrations, respectively, unless otherwise indicated, in WIT-T or OptiMEM medium, respectively. For lipid transfections, cells were transfected with Dharmafect I according to the manufacturer's protocol. Gene knockdown was assessed by measuring protein levels using flow cytometry 72 to 96 hours after treatment and by measuring mRNA by qRT-PCR 24 to 48 hours after treatment. Cell viability was assessed by CellTiter-Glo (Promega) or by Trypan Blue staining 24 to 48 hours after treatment.

MB468 cells (10⁶ cells/well of a 6-well plate) were treated with 4 μmol/L PLK1 AsiCs or unconjugated PLK1 siRNA for 48 hours. The cytoplasmic lysates of AsiC/siRNA–treated cells were incubated with protein G Dynabeads (Life Technologies) that were coated with 2 μg anti-pan-Ago antibody (clone 2A8, Millipore) or mouse IgG overnight at 4°C in the presence of RNase inhibitors (Life Technologies) and protease inhibitors (Roche). The beads were washed 5 times with NT2 buffer (50 mmol/L Tris pH 7, 150 mmol/L NaCl, 0.05% NP-40) and eluted by incubation for 30 minutes at 55°C with 200 μL SDS-TE buffer (5 mmol/L Tris pH 7.5, 0.5 mmol/L EDTA, 0.5% SDS). RNA was extracted from the unbound supernatants and bead-bound protein–RNA complexes eluted from the beads using TRIzol LS (Life Technologies). TaqMan small RNA assay kits that included predesigned primers (Life Technologies) were used to quantify PLK1 siRNA and miR-16 by qRT-PCR according to the manufacturer's manual. The amount of PLK1 siRNA in the supernatant and immunoprecipitates was normalized to miR-16.

Total RNA was extracted with TRIzol (Invitrogen) and cDNA prepared from 1,000 ng total RNA using the Thermoscript RT Kit (Invitrogen) as per the manufacturer's directions. qPCR of cDNAs was performed using SYBR Green Master Mix (Applied Biosystems) and a Bio-Rad C1000 Thermal Cycler (Bio-Rad). Relative C_t values were normalized to GAPDH. Primers are provided in Supplementary Table S1.

Cells were stained in PBS containing 0.5% fetal calf serum, 1 mmol/L EDTA, and 25 mmol/L HEPES with the following antibodies: EpCAM (Clone EBA-1 from BD Biosciences or clone 9C4 from BioLegend), AKT1 (clone 55/PKBa/Akt from BD Biosciences). Flow cytometry analysis was performed on a FACS-Canto II (BD Biosciences) using FlowJo (Treestar Inc.) software.

Fresh breast cancer and normal breast tissue biopsies, obtained from the UMASS Tissue Bank, were cut into 3 × 3 × 3 mm cubes and placed in 96 well plates containing 100 μL RPMI. Samples were incubated for 24 hours with 4 μmol/L Alexa647-labeled siRNA, 100 nmol/L Alexa647-labeled, chol-siRNA, or 4 μmol/L Cy3-labeled EpCAM-AsiC, all targeting eGFP, and photographed. To make single-cell suspensions, triplicate samples were pooled, sonicated using a gentleMACS dissociator (Miltenyi, spleen program for 30 minutes at 37°C), incubated at 37°C for 30 minutes with shaking in 10 mL RPMI containing 1 mg/mL collagenase II (Sigma-Aldrich), and then sonicated again. The resulting cell suspension was passed through a 70 μm cell strainer (BD Falcon), washed with 30 mL RPMI, and stained for flow cytometry.

For colony forming assay, 1,000 cells were treated for 6 hours with medium, 4 μmol/L AsiC or 100 nmol/L paclitaxel (Sigma) in round-bottom 96-well plates and transferred to 10 cm plates in serum-containing medium. Medium was replaced every 3 days. After 8 to 14 days, cells were fixed in methanol (−20°C) and stained with crystal violet. For sphere formation assay, cells (1,000/mL), treated as above for 6 hours in round-bottom 96-well plates, were cultured in suspension in serum-free DMEM/F12 1:1 (Invitrogen), supplemented with EGF (20 ng/mL, BD Biosciences), B27 (1:50, Invitrogen), 0.4% bovine serum albumin (Sigma), and 4 μg/mL insulin (Sigma). Spheres were counted after 1 to 2 weeks.

All animal experiments were performed with the approval of the Harvard Medical School and Boston Children's Hospital Animal Care and Use Committees. Eight-week-old female Nu/J mice (Stock # 002019, Jackson Laboratories) were used for all experiments.

To assess tumor initiation, mice were injected subcutaneously with MB468-luc (5 × 10⁶ viable cells) that had been pretreated for 24 hours with medium or 4 μmol/L EpCAM-AsiCs targeting eGFP or PLK1. Cells were trypsinized with TrypLE Express (Invitrogen), resuspended in WIT media, and injected subcutaneously in the flank. After intraperitoneal injection of 150 mg/kg d-luciferin (Caliper Life Sciences), luminescent images of the whole body were taken every 5 days for 15 days using an IVIS Spectrum using Living Image software (Caliper Life Sciences). The mice were sacrificed on day 15.

To assess AsiC uptake, MB468-luc (5 × 10⁶) and MB231-luc-mCherry (5 × 10⁵) cells trypsinized with TrypLE Express (Invitrogen), were resuspended in 1:1 WIT-Matrigel and injected subcutaneously in opposing flanks. Five days later, when tumors were easily palpated, mice were injected subcutaneously in the neck with Alexa750-EpCAM-AsiC-GFP (0.5 mg/kg in PBS). Animals were maintained on an alfalfa-free diet (Research Diets, Inc) to reduce autofluorescence and imaged using the IVIS Spectrum 15 minutes after the injection and twice thereafter at 2-day intervals. Animals were sacrificed 4 days after treatment and tumors were excised and imaged.

To assess tumor inhibition, MB468-luc (5 × 10⁶) and MB231-luc-mCherry (5 × 10⁵) cells trypsinized with Tryple Express (Invitrogen), resuspended in 1:1 WIT-Matrigel, were injected subcutaneously in opposite flanks. Five days later, when tumors were palpable, mice bearing tumors of comparable size were randomized into groups and treated subcutaneously every 3 days in the scruff of the neck with PBS containing nothing or 5 mg/kg EpCAM-AsiCs directed against eGFP or PLK1, EpCAM aptamer, or PLK1 siRNA. Mice were sacrificed on day 13. A similar protocol was followed in mice injected with 4 × 10⁴ MCF10-CA1a-Luc cells, except that treatment was initiated 1 day after tumors were implanted, the injections were performed in the flank near the tumor, but not intratumorally, and the mice were sacrificed on day 15.

Student t tests were used to analyze the significance between the treated samples and the controls where the test type was set to two-tailed distribution and two-sample equal variance. To assess innate immune stimulation, one-way ANOVA with Bonferroni multiple comparison test was performed using GraphPad Prism 4 software (GraphPad Software). P < 0.05 was considered statistically significant. Correlations were analyzed by Pearson coefficient.

Mice were injected sc with eGFP EpCAM-AsiCs (5 mg/kg) or intraperitoneally with Poly(I:C; 5 or 50 mg/kg). Serum samples, collected at baseline and 6 and 16 hours after treatment, were stored at −80°C before measuring IFNβ, IL6, and IP10 using the ProcartaPlex Multiplex Immunoassay (Affymetrix/eBioscience). Spleens, harvested at sacrifice 16 hours after treatment, were stored in RNAlater (Qiagen) before extracting RNA by adding TRIzol (Invitrogen) to single-cell suspensions generated using the gentleMACS Dissociator (MACS Miltenyi Biotec). cDNA was synthesized using Superscript III and random hexamers (Invitrogen) and PCR was performed using SsoFast EvaGreen Supermix and a Bio-Rad CFX96 Real-Time PCR System (Bio-Rad Laboratories) using the following primers:

• Gapdh forward: 5′- TTCACCACCATGGAGAAGGC-3′,

• Gapdh reverse: 5′- GGCATGGACTGTGGTCATGA-3′,

• ifnb forward: 5′-CTGGAGCAGCTGAATGGAAAG-3′,

• ifnb reverse: 5′- CTTGAAGTCCGCCCTGTAGGT-3′,

• il-6 forward: 5′-TGCCTTCATTTATCCCTTGAA-3′,

• il-6 reverse: 5′-TTACTACATTCAGCCAAAAAGCAC-3′,

• ip-10 forward: 5′-GCTGCCGTCATTTTCTGC-3′,

• ip-10 reverse: 5′-TCTCACTGGCCCGTCATC-3′,

• oas-1 forward: 5′-GGAGGTTGCAGTGCCAACGAAG-3′,

• oas-1 reverse: 5′-TGGAAGGGAGGCAGGGCATAAC-3′,

• stat1 forward: 5′-TTTGCCCAGACTCGAGCTCCTG-3′,

• stat1 reverse: 5′-GGGTGCAGGTTCGGGATTCAAC-3′.

We first examined EpCAM expression in breast cancer cell lines. On the basis of gene expression data in the Cancer Cell Line Encyclopedia (40), EpCAM mRNA is highly expressed in basal-A TNBC and luminal breast cancer cell lines, but poorly in basal-B (mesenchymal) TNBCs (Fig. 1A; in all figure labels in this article, basal-A TNBC cell lines are labeled in red, basal-B TNBC in green, and luminal cell lines in blue.) Surface EpCAM staining, assessed by flow cytometry, was 2 to 3 logs brighter in all luminal and basal-like cell lines tested, than in normal epithelia immortalized with hTERT (BPE; ref. 41), fibroblasts, or mesenchymal TNBCs (Table 1). Thus, EpCAM is highly expressed in epithelial breast cancer cell lines compared with normal cells or mesenchymal tumors.

A 19 nucleotide (nt) aptamer that binds to human EpCAM with 12 nmol/L affinity (19) was identified by SELEX (42, 43). A handful of EpCAM-AsiCs that linked either the sense or antisense strand of the siRNA to the 3′-end of the aptamer by several linkers were designed and synthesized with 2′-fluoropyrimidine substitutions and 3′-dTdT overhangs to enhance in vivo stability, avoid off-target knockdown of partially complementary genes, and limit innate immune receptor stimulation. To test RNA delivery, gene knockdown, and antitumor effects, siRNAs were incorporated to knock down a marker gene (eGFP), a ubiquitous, endogenous nonessential gene (AKT1), and PLK1, a kinase required for mitosis, whose knockdown is lethal to dividing cells (Supplementary Table S1). The AsiC that performed best in dose response studies of gene knockdown joined the 19 nt EpCAM aptamer to the sense (inactive) strand of the siRNA via a U-U-U linker (Fig. 1B). The EpCAM-AsiC was produced by annealing the chemically synthesized approximately 42–44 nt long strand (19 nt aptamer + linker + 20–22 nt siRNA sense strand) to a 20–22 nt antisense siRNA strand. The 2′-fluoropyrimidine modification led to RNase resistance and stability in 50% human serum (t_1/2 >> 36 hours, Supplementary Fig. S1). EpCAM-AsiCs did not trigger innate immunity when injected sc into tumor-bearing mice (Supplementary Fig. S2).

To verify selective uptake by EpCAM⁺ tumor cells, we first used confocal fluorescence microscopy to compare internalization of the EpCAM aptamer, fluorescently labeled at the 5′-end with Cy3, in EpCAM⁺ MDA-MB-468 TNBC cells and BPE, EpCAM^dim immortalized breast epithelial cells (Fig. 1C). Because AsiCs contain only one aptamer, they do not crosslink the receptor they recognize. As a consequence, cellular internalization is slow as it likely occurs via receptor recycling, rather than the more rapid process of activation-induced endocytosis. Only MDA-MB-468 cells took up the aptamer. Uptake was clearly detected at 22 hours, but increased greatly after 43 hours. To test whether EpCAM-AsiCs are specifically taken up by EpCAM bright cell lines, the 3′ end of the antisense strand of the AsiC was fluorescently labeled. EpCAM⁺ BPLER, a basal-A TNBC cell line transformed from BPE by transfection with human TERT, SV40 early region and H-RAS^V12, took up Alexa-647 EpCAM-AsiCs when analyzed after a 24-hour incubation, but BPE cells did not (Fig. 1D). Previous studies have shown that AsiCs are processed within cells by Dicer to release the siRNA from the aptamer (10, 12, 15). To verify that the released siRNA was taken up by the RNA-induced silencing complex (RISC), we used qRT-PCR to amplify PLK1 siRNA immunoprecipitated with Ago when MDA-MB-468 cells were incubated with PLK1 EpCAM-AsiCs (Supplementary Fig. S3). No PLK1 siRNA bound to Ago when the same cells were incubated with PLK1 siRNAs.

Next to assess whether gene knockdown was specific to EpCAM⁺ tumors, we compared eGFP knockdown by eGFP EpCAM-AsiCs and lipid transfection of eGFP siRNAs in BPE and BPLER cell lines, which stably express eGFP (Fig. 1D). Although transfection of eGFP siRNAs knocked down gene expression equivalently in BPE and BPLER, incubation with EpCAM-AsiCs in the absence of any transfection lipid selectively knocked down expression only in BPLER. AsiC knockdown was uniform and comparable to that achieved with lipid transfection. Next, we compared the knockdown of AKT1 by AKT1 AsiCs and transfected AKT1 siRNAs in 6 breast cancer cell lines and normal human fibroblasts (Fig. 1E). AKT1 was selectively knocked down by EpCAM-AsiCs targeting AKT1 only in EpCAM^bright luminal and basal-A TNBCs, but not in mesenchymal basal-B TNBCs or fibroblasts. As expected, AsiCs targeting eGFP had no effect on AKT1 levels and transfection of AKT1 siRNAs comparably knocked down expression in all the cell lines studied. Moreover, EpCAM-AsiC knockdown of AKT1 correlated with EpCAM expression (Fig. 1F). Similar results were obtained when AKT1 protein was analyzed by flow cytometry in stained transfected cells (Fig. 1G and H). Thus, in vitro knockdown by EpCAM-AsiCs is effective and specific for EpCAM^bright tumor cells.

To explore whether EpCAM-AsiCs could be used as antitumor agents in breast cancer, we examined by CellTiterGlo assay the effect of AsiCs directed against PLK1, a kinase required for mitosis, on survival of 10 breast cancer cell lines that included 5 basal-A TNBCs, 2 luminal cell lines, and 3 basal-B TNBCs. EpCAM-AsiCs targeting PLK1, but not control AsiCs directed against eGFP, decreased cell proliferation in the basal-A and luminal cell lines, but did not inhibit basal-B cells (Fig. 2A). Lipid transfection of PLK1 siRNAs suppressed the growth of all the cell lines. The antiproliferative effect strongly correlated with EpCAM expression (Fig. 2B). To determine whether ligation of the EpCAM aptamer contributed to the antiproliferative effect of the EpCAM-AsiC, we compared survival of cells treated with PLK1 EpCAM-AsiCs with cells treated with the aptamer on its own (Fig. 2C). The aptamer by itself did not reproducibly affect survival of any breast cancer cell lines, possibly because as a monomeric agent it does not cross-link the EpCAM receptor to alter EpCAM signaling. Thus, the PLK1 EpCAM-AsiC asserts its specific antitumor effect on EpCAM⁺ breast cancer cells by gene knockdown.

To determine whether EpCAM-AsiCs specifically target EpCAM⁺ cells when mixed with EpCAM^dim nontransformed epithelial cells, we incubated cocultures of GFP⁻ TNBC cells and GFP⁺ BPE cells with PLK1 EpCAM-AsiCs or medium and used GFP fluorescence to measure their relative survival by flow cytometry 3 days later (Fig. 2D and E). EpCAM-AsiCs targeting PLK1 greatly reduced the proportion of surviving EpCAM⁺ basal-A tumor cells, but had no effect on survival of an EpCAM⁻ basal-B cell line. Thus, PLK1 EpCAM-AsiCs are selectively cytotoxic for EpCAM⁺ tumor cells when mixed with normal cells.

We next examined whether EpCAM-AsiCs concentrate in human breast tumors relative to normal breast samples within intact tissues. Paired normal tissue and breast tumor biopsies from 3 breast cancer patients (2 ER⁺PR⁺HER2⁻ luminal breast cancers, 1 TNBC) were cut into approximately 3 mm sided cubes and placed in 96-well plates. The tumor sample cells were all EpCAM^bright and the normal tissue cells were EpCAM^dim (Fig. 3A). Fluorescently labeled Alexa647-siRNAs (not expected to be taken up by either normal tissue or tumor), Alexa647-cholesterol–conjugated siRNAs (chol-siRNAs, expected to be taken up by both), or Cy3-EpCAM-AsiCs were added to the culture medium and the tissues were incubated for 24 hours before harvest. The Cy3 signal of the AsiC, visible to the naked eye, concentrated only in the tumor specimens and was not detected in normal tissue (Fig. 3B). To quantify RNA uptake, flow cytometry analysis was performed on washed single-cell suspensions of the tissue specimens [representative tumor-normal tissue pairs (Fig. 3C), mean ± SEM of triplicate biopsies from 3 EpCAM^bright-paired breast tumor–normal tissue samples (Fig. 3D)]. The EpCAM-AsiC was significantly taken up by the tumors, but not normal tissue, while neither took up the unconjugated siRNA and both weakly took up the chol-siRNA. Thus, within intact tissue, EpCAM-AsiCs are selectively delivered to EpCAM^bright tumors relative to normal tissue.

EpCAM was chosen for targeting in part because EpCAM marks TICs and metastasis-initiating cells (20, 22, 26, 27, 31). To investigate whether EpCAM-AsiCs inhibit TICs, we compared colony and mammosphere formation (TIC functional in vitro assays) after mock treatment, treatment with paclitaxel or with EpCAM-AsiCs against eGFP or PLK1. PLK1 EpCAM-AsiCs more strongly inhibited colony and mammosphere formation of EpCAM⁺ basal-A TNBCs and luminal cell lines than paclitaxel, but were inactive against EpCAM⁻ basal-B TNBCs (Fig. 4A–C). TIC inhibition was specific, since eGFP AsiCs had no effect. To evaluate the effect of EpCAM-AsiCs on tumor initiation, EpCAM⁺ MB468 cells stably expressing luciferase were treated overnight with medium or PLK1 or eGFP EpCAM-AsiCs and equal numbers of viable cells were then implanted subcutaneously in nude mice (Fig. 4D). PLK1 EpCAM-AsiCs completely blocked tumor formation assessed by in vivo luminescence (Fig. 4E and F). Thus, PLK1 EpCAM-AsiCs inhibit in vitro TIC assays and tumor initiation selectively for EpCAM⁺ breast cancers.

To be clinically useful, EpCAM-AsiCs need to be taken up by disseminated tumor cells. Intravenously injected AsiCs do not accumulate efficiently within subcutaneous tumors implanted in the flanks of nude mice, probably because their size (∼25 kDa) is below the threshold for kidney filtration and they are rapidly excreted. Linkage to polyethylene glycol greatly enhanced the circulating half-life, tumor accumulation, and antitumor therapeutic effect of PSMA-AsiCs in a mouse xenograft model of prostate cancer (11). However, to see if this modification could be bypassed, we examined by live animal epifluorescence imaging whether sc injection of Alexa750-labeled eGFP EpCAM-AsiCs in the scruff of the neck of 7 mice led to accumulation in distant EpCAM⁺ MB468 and EpCAM⁻ MB231 TNBCs implanted subcutaneously in each flank (Fig. 5). Within a day of injection, EpCAM-AsiCs concentrated in the EpCAM⁺ tumor and persisted there for at least 4 days. The EpCAM-AsiCs were detected around the injection site on day 2, but were only found within the EpCAM⁺ tumor on day 4.

Because subcutaneously injected EpCAM-AsiCs concentrated in distant EpCAM⁺ tumors, we next looked at whether subcutaneous injection of PLK1 EpCAM-AsiCs could selectively inhibit the growth of an EpCAM⁺ TNBC xenografted tumor. EpCAM⁺ MB468-luc cells were implanted in Matrigel in one flank of a nude mouse and EpCAM⁻ MB231-luc-mCherry cells were implanted on the opposite flank. Once the luciferase signal of both tumors was clearly detected above background, groups of 5 to 6 mice were mock treated or injected subcutaneously with 5 mg/kg of EpCAM-AsiCs targeting PLK1 or eGFP every 3 days for 2 weeks (we chose biweekly dosing because gene knockdown in siRNA-transfected rapidly dividing cancer cells usually only persists for about 5 days). Tumor growth was followed by luminescence. All the EpCAM⁺ tumors rapidly completely regressed only in mice that received the PLK1-targeting AsiCs (Fig. 6). The EpCAM⁺ tumors in mice treated with eGFP-targeting AsiCs and all the EpCAM⁻ tumors continued to grow. This experiment was repeated with similar results after injection of PLK1 AsiCs into mice bearing Her2⁺ MCF10A-CA1a (Supplementary Fig. S4). Thus, subcutaneously injected PLK1 EpCAM-AsiCs show specific antitumor activity against EpCAM⁺ human xenografts.

Here we show that EpCAM-AsiCs can be used to knock down genes selectively in epithelial breast cancer cells and their stem cells, sparing normal epithelial cells and stroma, to cause tumor regression and suppress tumor initiation. In one very aggressive TNBC xenograft model, the EpCAM-AsiCs caused complete tumor regression after only 3 injections. However, we did not keep the mice alive off therapy to see if the tumors recurred after treatment was stopped. This is a flexible platform for targeted therapy, potentially for all the common epithelial cancers, which uniformly express high levels of EpCAM. Although we used EpCAM-AsiCs targeting PLK1, in principle, the siRNA could be varied to knock down any tumor dependency gene that would be customized to the tumor subtype or the molecular characteristics of an individual patient's tumor. AsiC cocktails targeting more than one gene would be ideal for cancer therapeutics to lessen the chances of developing drug resistance. Targeted cancer therapy so far has relied on using tumor-specific antibodies or small-molecule inhibitors to oncogenic kinases. Although the AsiC platform is not new, using EpCAM as an AsiC ligand and developing RNAi therapy to target cancer stem cells is novel. No one before has shown that an unconjugated AsiC can have potent antitumor effects or that AsiCs could be administered subcutaneously (moreover, preliminary studies of sc administered CD4-AsiCs in humanized mice (L.A. Wheeler and J. Lieberman) showed strong knockdown in CD4 cells in the spleen and distant lymph nodes, suggesting that AsiCs targeting receptors on cells located elsewhere in the body could also be administered subcutaneously). There is currently no targeted therapy for TNBC or for TICs. Targeted delivery has the advantage of reduced dosing and reduced toxicity to bystander cells. Developing targeted therapy for TNBC and developing ways of eliminating T-ICs are important unmet goals of cancer research.

Multiple groups have already used AsiCs to demonstrate impressive therapeutic effects (so far only in mice) to knock down gene expression in challenging types of cells, including cancer cells and lymphocytes. AsiCs are cleaved within cells by Dicer to liberate the siRNA from the aptamer (10, 12, 15). The U-U-U linker used here may be a particularly good Dicer substrate. AsiCs are an attractive method for gene knockdown outside the liver. It is a flexible platform that by modifying the aptamer can be used to target any cell type and by changing the siRNA can be used to knock down any gene or combinations of genes. Over a thousand aptamers to human proteins have already been identified (44). Methods to select aptamers have become streamlined in recent years by combining deep sequencing and bioinformatics to more rapidly identify the features of active sequences (45). AsiCs are ideal for personalized therapy that can be altered to suit the molecular characteristics of an individual tumor or tumor subtype. The AsiC does not appear to activate the receptor, presumably because it does not crosslink it (15). AsiCs are a single chemical entity that is stable in the blood, simple to manufacture and not likely to be toxic. We did not see any evidence of toxicity or weight loss in treated mice, but we have not done formal testing. AsiCs, unlike liposomes and nanoparticles, do not get trapped in the liver and other filtering organs and should be able to readily penetrate tissues as well as small-molecule drugs (46, 47). EpCAM-AsiCs did not stimulate an innate immune response when injected in vivo and tested in the most sensitive immune cells (splenocytes) and at the time of a peak immune response with sensitive qRT-PCR assays. RNAs on their own do not induce antibodies, although the absence of AsiC antibodies needs to be formally tested. An aptamer drug, pegatanib (Macugen) is approved for macular degeneration and at least 8 aptamer drugs are in clinical trials, so aptamer RNAs on their own are well tolerated (48, 49).

The small size of the EpCAM aptamer used here is ideal for an AsiC drug, as RNAs <60 nt can be efficiently synthesized. Despite their promise, there is still considerable room to improve AsiCs, to optimize circulating t_1/2, cellular uptake, and endosomal release and to reduce the needed dose (although it is currently acceptable: approximately 1–5 mg/kg; refs. 11, 13, 15, 18). Recent studies have shown that the efficiency of GalNac-conjugated siRNAs in nonhuman primates can be improved by as much as 50-fold by optimizing the chemical modifications of the active strand to enhance stability and activity within the RNA-induced silencing complex (M. Manoharan, personal communication). Experiments to optimize the EpCAM-AsiCs are planned.

One potential source of toxicity of EpCAM-AsiCs could be targeting of epithelial tissue stem cells, especially in the gut. Because the EpCAM aptamer does not target mouse EpCAM, mouse experiments cannot adequately assess this possible toxicity. We will therefore need to use human tissues (or possibly primates) to look for EpCAM-AsiC toxicity to human tissue stem cells. An alternate drug development strategy would be to select for an EpCAM aptamer that cross-reacts with mouse and human EpCAM.

In addition to their potential therapeutic use, EpCAM-AsiCs could also be a powerful in vivo research tool for identifying the dependency genes of tumors and TICs to define novel drug targets. In principle, aptamer chimeras could be designed to deliver not only siRNAs but also miRNA mimics or antagomirs, antisense oligonucleotides that function by other mechanisms besides RNAi, or even longer mRNAs or noncoding RNAs (50, 51). They could also be designed to incorporate more than one aptamer, multiple siRNAs, or even toxins or small-molecule anticancer drugs.

J. Lieberman has ownership interest in Alnylam stock and is a member of the Alnylam Scientific Advisory Board. No potential conflicts of interest were disclosed by the other authors.

Conception and design: L.A. Wheeler, J. Lieberman

Development of methodology: P. Hamar, M.T.N. Le, L.A. Wheeler

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A. Gilboa-Geffen, P. Hamar, M.T.N.Le, L.A. Wheeler, A. Wittrup

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A. Gilboa-Geffen, P. Hamar, M.T.N. Le, F. Petrocca, A. Wittrup, J. Lieberman

Writing, review, and/or revision of the manuscript: P. Hamar, F. Petrocca, J. Lieberman

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): P. Hamar, M.T.N. Le

Study supervision: L.A. Wheeler, J. Lieberman

Other (provided methodology for flow cytometry analysis of tissue samples and performed immunohistochemistry): R. Trifonova

This work was supported by a Department of Defense Breast Cancer Research Program (DOD BCRP) grant, (W81 XWH-09-1-0058; to J. Lieberman), Fulbright Visiting Scholar grant (1213204; to P. Hamar), Jane Coffin Childs Fund fellowship (to M. Le), Cancer Research Institute, Adelstein Fellowship, Harvard Medical School MD/PhD Program and Point Foundation fellowships to (L.A. Wheeler), Harvard CFAR scholar award (to R. Trifonova), DOD BCRP fellowship (to F. Petrocca), and Swedish Research Council fellowship (to A. Wittrup).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.