NOD.CB17–Prkdc^scid/J mice were purchased from The Jackson Laboratory (Bar Harbor, ME and Sacramento, CA, USA). Animals were housed in micro-isolator cages with autoclaved bedding, and fed autoclaved food and water. All experiments were approved by Stanford's Institutional Animal Care and Use Committee and in accordance with all Administrative Panel on Laboratory Animal Care (APLAC) regulations at Stanford University and were in compliance with the National Institutes of Health Guide for Care and Use of Animals. A total of 20 NOD/SCID (non-obese diabetic–severe combined immunodeficiency) female mice had 8 × 10⁶ MDA-MB-231 breast cancer cells in 100 μl phosphate-buffered saline (PBS; pH 7.4) plus 100 μl of matrigel (BD Biosciences, San Jose, CA, USA) injected into their left second mammary fat pads. Control mice were injected with PBS into the same location. At 55 days after injection, when the average tumour volume was 1.5 cm³, mice were euthanised for isolation of CTCs. To assess the tumour formation capacity of CTCs, mammary fat pads of 10 NOD–SCID mice were injected with a total of 8 × 10⁶ MDA-MB-231 cells or CTCs in a total volume of 200 μl, consisting of 100 μl PBS and 100 μl matrigel. Tumour size was measured over a period of 3–6 weeks.

After the mice were euthanised, their organs and tumours were fixed in formalin, routinely processed, embedded in paraffin, and 5-μm sections were placed on glass slides for haematoxylin & eosin staining and immunohistochemistry (IHC). Microwave heat-induced epitope retrieval in citrate buffer was used for IHC as previously described (Higgins and Warnke, 1999). Tumour and lung sections were stained for the intrinsic marker GLUT1 (ab15309, Abcam, Cambridge, MA, USA), ATF3 (sc-188, Santa Cruz Biotechnology, Santa Cruz, CA, USA), ATF4 (ab31390, Abcam), and HIF1α (ab51608, Abcam). At 4 h before being killed, mice were injected through the intraperitoneal route with the hypoxia marker pimonidazole, and tumour and lung sections were stained using Hypoxyprobe-1 (NPI, Belmont, MA, USA).

As soon as the mice were euthanised, blood was drawn by means of a cardiac puncture, recently shown to be the method of choice in obtaining CTCs from mouse blood (Eliane et al, 2008). Mouse blood (200–900 μl per mouse) was pooled and collected into a 2-ml EDTA tube (BD, Franklin Lakes, NJ, USA). Blood was diluted to 6 ml with PBS and labelled with magnetic beads functionalised with epithelial cell adhesion molecule antibody (Dynabeads, Invitrogen, Carlsbad, CA, USA). Magnetically labelled blood samples were transferred to a six-well plate and brought up to a final volume of 10 ml using PBS. The magnetically labelled samples were then processed by the newly developed MagSweeper that isolates CTCs by multiple rounds of magnetic capture, wash, and release as previously described (Talasaz et al, 2009). Briefly, magnetic capturing rods covered by plastic sheaths were robotically swept through wells containing magnetically labelled mouse blood. Circulating tumour cells captured on the covered magnetic rods were transferred and washed in a well containing PBS, and then released into another well containing PBS by removal of the magnetic rod from its sheath. After a second cycle of capture, wash, and release, individual CTCs were microscopically visualised and transferred to a new six-well plate containing Dulbecco's modified Eagle's medium high glucose (Invitrogen, Carlsbad, CA, USA) supplemented with 10% (vol/vol) foetal calf serum, penicillin (100 U ml⁻¹), and streptomycin (100 μg ml⁻¹). The purified CTCs in the six-well plates were cultured in a 5% CO₂ incubator at 37°C.

Hypoxia was generated using the anaerobic jar HP0011A (Oxoid, Cambridge, UK and distributed by Remel, Lenexa, KS, USA) and oxygen levels were monitored using the Analox Mini O₂ DII monitor (Amron International, Vista, CA, USA). Cells were maintained in Dulbecco's modified Eagle's medium high glucose (Invitrogen) supplemented with 10% (vol/vol) foetal calf serum, penicillin (100 U ml⁻¹), and streptomycin (100 μg ml⁻¹). The passage number of CTCs was <7. For hypoxia experiments, 3 × 10⁵ of MDA-MB-231 cells or CTCs were seeded in 5-cm plates and incubated with 70–80% confluency in the humidified jar. A gas mixture of 5% CO₂/95%N₂ was purged into the jar for 20 min until the O₂ monitor indicated <0.1% O₂, after which additional gas was purged into the jar for an extra 20 min. When experiments were terminated after 24 h, the O₂ monitor indicated oxygen levels of 0.2%.

A total of 400 cells were seeded in 5-cm Petri dishes and incubated for 10 days in either normoxia or hypoxia (n=5 for each condition). After 10 days, the colonies were rinsed with PBS and stained with Giemsa stain-modified solution (Sigma-Aldrich, St Louis, MO, USA) for 20 min. Stained colonies were rinsed with water, and visually visible colonies, which consisted of approximately 30 cells, were counted.

The ATF3, ATF4 and GLUT1 antibodies used for western analysis were the same as those used for immunostaining. The HIF1α antibody used for immunoblot analysis was from BD transduction laboratories (610958, BD Biosciences). CAIX (ab15086) and BNIP3 (ab10433) were from Abcam. The apelin antibody (sc-33804) was from Santa Cruz Biotechnology. Cell extracts were generated in a cold room (4°C). The cell lysis buffer used for preparing total cell extracts was a urea-denaturing buffer (6.7 M urea, 10 mM Tris–HCl (pH 6.8), 5 mM dithiothreitol, 1% sodium dodecyl sulphate, and 10% glycerol) supplemented with Complete mini-protease inhibitor cocktail tablets (Roche Diagnostics, Indianapolis, IN, USA). Cultured cells were washed rapidly once with ice-cold PBS. A volume of 100–200 μl of urea lysis buffer was added directly to the plates of cells, scraped into an Eppendorf tube, and sonicated on ice by using a Vibra Cell ultrasonic processor (Sonics&Materials, Newton, CT, USA). A detergent-compatible bicinchoninic acid protein assay kit (Thermo Scientific, Rockford, IL, USA) was used to estimate the protein concentration of extracts according to the manufacturer's protocol. Total cell extracts (30–60 μg per lane) were subjected to reducing sodium dodecyl sulphate–polyacrylamide gel electrophoresis by first mixing equal volumes of protein extract and Laemmli sample buffer containing dithiothreitol and β-mercaptoethanol and running them on a 12% readymade gel system (Bio-Rad, Hercules, CA, USA). The resolved proteins were then electroblotted (semidry) onto immune blot polyvinylidene fluoride membrane (Bio-Rad). Bound antibodies were detected using the chemiluminescent substrate ECL+Plus (Amersham Biosciences, Buckinghamshire, UK).

Tumour hypoxia and anoxia have been shown to select for an aggressive subtype of tumour cells. Such cells would metastasise to distant organs through the blood stream. We hypothesised that tumours containing regions of hypoxia and anoxia would give rise to CTCs that, on the basis of their selection by hypoxia, should be more aggressive than non-CTCs, and would show an altered response to hypoxia. To examine this, we first generated a tumour model containing profound regions of hypoxia and anoxia. We used the human breast cancer cell line MDA-MB-231 to make orthotopic xenografts and grew them to an average tumour volume of 1.5 cm³. Staining for pimonidazole (Figure 1A) and the intrinsic hypoxia marker GLUT1 (Figure 1B; Airley et al, 2003) confirmed that all tumours contained hypoxic regions, particularly around necrotic areas. Although GLUT1 expression and pimonidazole uptake were more pronounced in tumour cells within perinecrotic regions, the distribution of HIF1α, ATF3, and ATF4 expression rarely coincided with the pimonidazole and GLUT1 staining pattern. (Figure 1Ci).

As GLUT1 is an HIF1α target gene, one would expect a similar pattern of expression. As tumour cells in perinecrotic regions are more likely to experience chronic hypoxia, we wanted to determine whether MDA-MB-231 cells subjected to chronic hypoxia in vitro would also demonstrate disparate induction of HIF1α and GLUT1. Whereas HIF1α and GLUT1 were both induced in acute hypoxia (48 h), chronic hypoxia (7 days) induced only GLUT1 but not HIF1α (Figure 1Cii).

Our xenograft model, based on injection of large numbers of tumour cells (8 × 10⁶) and an extended growth period (55 days), created profound regions of hypoxia and anoxia in the primary tumour. Lung metastases developed in all mice. Hypoxia and anoxia factors expressed in primary tumours were also expressed in pulmonary metastases (Figure 2).

To determine whether CTCs have an altered hypoxia response, they had to be isolated and grown in culture. We demonstrated that our MDA-MB-231 xenograft model shed CTCs into mouse circulation using a new technology that enabled viable human CTC capture from mouse blood (Talasaz et al, 2009). Magnetic beads attached to the epithelial cell adhesion molecule antigen on CTCs appeared as dark clusters of beads on the cell membrane surrounding translucent cells. In contrast, there were no dark cell/magnetic bead clusters visible in the blood of control mice. Instead, only small dark circles representing individual magnetic beads were observed in control mouse blood (Figure 3Ai). Captured CTCs were viable and could be subsequently grown in culture (Figure 3Aii).

It is known that anoxia exerts a selection pressure for aggressive cells with diminished apoptotic potential. Therefore, anoxia-selected cells in vivo could enter the blood stream as CTCs and demonstrate a distinct expression and/or induction of anoxia factors. To examine this, we compared anoxia factor expression in CTCs with that of parental MDA-MB-231 cells.

The CTCs demonstrated greater expression and/or induction of the anoxia-induced factors ATF3 (Figure 3B), ATF4 (Figure 3C and D), and the ATF4 target gene, ASNS (Figure 3D), than parental MDA-MB-231 cells. Western blot analysis demonstrated that ATF3 protein was expressed constitutively higher in CTCs compared with MDA-MB-231 cells, and hypoxia did not dramatically induce further ATF3 expression as confirmed by densitometry measurements. In contrast, MDA-MB-231 cells induced ATF3 2.4-fold in hypoxia (Figure 3B). Western blot analysis and densitometry measurements demonstrated that the ATF4 protein was not induced in hypoxic MDA-MB 231 cells but was induced 2.4-fold in hypoxic CTCs (Figure 3C). Western blot analysis showed that the ATF4 target gene ASNS was also induced to a greater extent in hypoxic CTCs than in hypoxic MDA-MB-231 cells (Figure 3 D). At a higher cell passage number (passage >7), CTCs demonstrated some reversion back to the parental phenotype with respect to ATF3 expression and increased induction in hypoxia (data not shown). In summary, CTCs demonstrated an anoxia phenotype by constitutively expressing and/or inducing anoxia factors at higher levels compared with parental MDA-MB 231 cells under normoxia and/or hypoxia.

In addition to anoxia factor expression, tumour aggression has been related to enhanced expression of hypoxia-induced factors. Therefore, we questioned whether the hypoxic response of CTCs is altered. To examine this, we compared the expression of hypoxia factors in CTCs and parental MDA-MB-231 cells. Although MDA-MB-231 cells and CTCs did not differ in their response to hypoxia with respect to HIF1α induction, CTCs demonstrated a distinct expression of proteins regulated by HIF1α under hypoxic conditions. Apelin was induced at greater levels in CTCs than in parental MDA-MB-231 cells (Figure 3E). In hypoxia, CTCs did induce GLUT1 (Figure 3E), CAIX and BNIP3 proteins (Figure 3F) but the inductions were lower than those in parental MDA-MB-231 cells. In summary, CTCs demonstrated an altered response to hypoxia, showing both pronounced (apelin) and reduced induction (GLUT1, CAIX, and BNIP3) of specific hypoxia/HIF1-regulated genes.

The anoxia factors expressed in CTCs are known to be associated with cell proliferation, aggression, and survival. Therefore, we questioned whether CTCs possess an enhanced colony formation capacity in hypoxia. To test this, we performed colony formation assays. A greater number of colonies was observed with CTCs than with parental MDA-MB-231 cells, when grown under conditions of chronic hypoxia for 10 days (Figure 4A). In normoxia, both CTCs and MDA-MB-231 cells resulted in a similar number of colonies after 10 days. Results were similar when different numbers of cells and/or different Petri dish sizes and/or different incubation periods were used for the colony formation assays. Therefore, CTCs had a greater colony formation capacity in chronic hypoxia compared with MDA-MB 231 cells.

The greater clonogenicity under chronic hypoxic conditions suggested that CTCs were more aggressive than their parental MDA-MB-231 cells. To examine tumourigenicity and metastatic potential of CTCs in vivo, we generated 10 orthotopic xenografts from cultured CTCs and compared tumour growth with that of 10 parental MDA-MB-231 xenografts. Within 3 weeks, two mice carrying CTC tumour xenografts died. Of the remaining 8 mice that had CTC xenografts and 10 mice that had MDA-MB-231 xenografts, five were euthanised from each tumour group at week 3. Tumour xenografts in the remaining mice (five with MDA-MB-231 xenografts and three with CTC xenografts) continued to grow and were measured from week 3 to week 6, after which the mice were euthanised. As indicated in Figure 4Bi and ii, xenografts generated from CTCs grew larger and heavier than xenografts from MDA-MB-231 cells. Pulmonary micrometastases became evident at week 3 only in mice with xenografts grown from CTCs. By week 6, mice with parental cell xenografts developed pulmonary metastases. In summary, the more aggressive phenotype of CTCs as observed in vitro by the expression of anoxia factors and greater colony formation in chronic hypoxia was also reflected in vivo by the development of more rapidly growing and metastasising tumours.