Noninvasive Molecular Imaging of Hypoxia in Human Xenografts: Comparing Hypoxia-Induced Gene Expression with Endogenous and Exogenous Hypoxia Markers

Although the existence of tumor hypoxia has been suggested by Thomlinson and Gray (1) more than 50 years ago, its true importance in cancer management has only become known within the last 15 years. This is in part due to recent advances in methods to measure hypoxia, improved understanding of the molecular events associated with tumor hypoxia, and clinical findings correlating tumor hypoxia and treatment outcome (2–9). Importantly, numerous clinical studies, using physical probes to measure PO₂ levels, has shown that existence of tumor hypoxia is negatively correlated with local control and survival (2, 3, 10, 11). Because hypoxic cells are 3-fold more radioresistant than aerated cells, such clinical results from radiotherapy trials are not surprising. However, Hockel and colleagues (10) reported similar findings for surgical patients who did not receive radiation treatment; the underlying reasons are now beginning to be understood.

It is now known that tumor hypoxia is associated with more aggressive tumor phenotypes (12–14),which may lead to poorer prognosis for patients with hypoxic tumors whether they are treated with surgery or radiation (5, 15). Not only is the hypoxic environment selective for genetically unstable tumor cells, hypoxia is also associated with tumors that are more likely to metastasize (3, 5, 16–18). Such tumor phenotypes are due in part to the hypoxia-associated molecular pathways, which regulate the expression of many genes, including those involved in tumor progression and malignant phenotypes. At the molecular level, hypoxia-inducible factor 1 (HIF-1) has been identified as the “switch” (19, 20). It is up-regulated in response to oxygen deprivation, binds to the hypoxia responsive element (HRE), and controls the transcriptional activity of numerous downstream genes that are important to the development and growth of solid tumors (21).

Its importance in cancer management has motivated much research in measuring, characterizing, and understanding tumor hypoxia. Direct tumor PO₂ measurement with physical probes has been used in clinical studies and is considered a reference standard. However, it is invasive and labor intensive. Another method is immunohistochemical (IHC) analysis of tumor biopsies based on hypoxia-associated endogenous proteins such as HIF-1α (22) and CA9 (23, 24), or exogenous markers such as pimonidazole and EF5 that are preferentially trapped in hypoxic cells (25, 26). The IHC approach is also invasive but can yield tumor hypoxia information with excellent spatial resolution for direct comparison with histology. A third approach involves nuclear imaging techniques such as single-photon emission computed tomography or positron emission tomography (PET), using hypoxia radiotracers, such as ¹⁸F-FMISO, ¹⁸F-EF5, ⁶⁴Cu-ATSM, and ¹²⁴I-IAZGP (9, 27–30).

Reporter gene imaging is emerging as a valuable tool for studying tumor biology and deciphering genetic and biological processes in unperturbed microenvironment within experimental animals (31–36). In our previous efforts, applying this approach to the study of tumor hypoxia, Wen and colleagues (29) transduced Dunning rat prostate R3327-AT tumor cells with the vector 8HRE:tkeGFP, such that the fusion gene of HSV1-tk and enhanced green fluorescent protein (eGFP) was under the regulation of a hypoxia-inducible promoter containing eight tandem repeats of HRE (29). Similarly, Serganova and colleagues (37) used the same technique in C6 rat glioma cells. In hypoxic tumors, the reporter gene was expressed and the gene product TK phosphorylated the marker substrate ¹²⁴I-FIAU to generate microPET images. However, neither the transduced R3327-AT nor the C6 cells expressed eGFP at sufficient levels for unequivocal visualization in tumor sections, and hence, no such data were presented in reference 29 and 37. This was a major deficiency with these models because in vivo eGFP data, with its superb spatial resolution in optical imaging, is a valuable tool for studying microscopic tumor hypoxia distribution, as will be shown in this paper. Furthermore, both previous studies involved rodent tumor cells.

In this study, we developed a xenograft model using human tumor cells, to ensure the closest correspondence between our preclinical and clinical investigations of hypoxia imaging in colorectal tumor. Incorporating a 9HRE element provided by Dr. Dennis Deen (University of California, San Francisco, CA), we designed and constructed a new and improved 9HRE:tkeGFP vector, and transfected it into human colorectal adenocarcinoma HT29 cells. We then conducted a rigorous process of selection and enrichment, by repeated fluorescence-activated live cell sorting (FACS), to generate clones with the high level of hypoxia-induced TKeGFP expression. In in vitro studies, we verified the hypoxia-induced expression of the TKeGFP fusion protein by fluorescence microscopy and by quantifying TK-mediated trapping of ¹⁴C-FIAU. In in vivo studies, PET and autoradiography were used to map the distributions of tumor hypoxia, based on hypoxia-induced TK expression and TK-mediated trapping of ¹²⁴I-FIAU. For direct comparison, similar distributions were obtained with the hypoxic cell radiotracer ¹⁸F-FMISO in the same animals. In addition, fluorescence microscopy was used to visualize the eGFP distribution in tumor sections, in comparison of those of CA9, EF5, and pimonidazole.

Our results show that with this human xenograft reporter system, we can image hypoxia-induced molecular events and map their distribution in vivo with noninvasive PET imaging. This is based on hypoxia-induced increase in HIF-1α level, the downstream transcriptional activation of the TKeGFP fusion gene, and the TK-mediated phosphorylation and trapping of the marker substrate ¹²⁴I-FIAU in vivo. Thus, our model system provides a link between the molecular events of tumor hypoxia and noninvasive hypoxia imaging. It also allows us to examine the intratumoral distribution of various endogenous and exogenous hypoxia markers using dual-label phosphor plate imaging, IHC, and fluorescence microscopy. This human xenograft reporter model provides a valuable tool to study tumor hypoxia and in validating existing and future exogenous markers for tumor hypoxia.

Cell culture. Human colorectal carcinoma HT29 cells, obtained from American Type Culture Collection, were routinely cultured in Mc Coy's 5A medium, 10% fetal bovine serum, 2 mmol/L l-glutamine, 1.5 g/L sodium bicarbonate, 0.1 mmol/L nonessential amino acids, 1.0 mmol/L sodium pyruvate, and antibiotics.

Generation of hypoxia-inducible dual reporter expression vectors. The hypoxia-inducible, dual reporter expression plasmid p9HRE-TKeGFP contains the Herpes Simplex Virus type 1-thymidine kinase (HSV1-tk) and eGFP fusion gene under the regulation of a hypoxia-inducible promoter, and a constitutively expressed neomycin-resistance (neo^r) gene (Fig. 1A). The hypoxia-inducible promotor consists of an artificial nine tandem repeats of the HRE (9HRE) from the enhancer region of the human erythropoietin EPO gene, and was linked to the SV40 minimal promoter (SV40 min). The plasmid, pDsRed2-1 containing the constitutively expressed neo^r gene was purchased from BD Biosciences-Clontech. The tkeGFP cDNA fragment was excised from the retroviral vector dxHRE-tke/GFP-cmv Red2XPRT (37), placed under the control of the artificial 9HRE linked to the SV40 min, and cloned into pDsRed2-1, replacing the DsRed2-1 gene. The resultant vector, designated as p9HRE-TKeGFP, was used to stably transfect human tumor HT29 cells. The DNA fragment of 9HRE is a generous gift from Dr. Dennis F. Deen (Brain Tumor Research Center of the Department of Neurological Surgery, University of California, San Francisco, CA; refs. 38, 39). Its DNA sequence is GCTGCAGGAATTCGATGCACGCGTCCGGGTAGCTGGCGTACGTGCTGCAGCCGGGTAGCTGGCGTACGTGCTGCAGCCGGGTAGCTGGCGTACGTGCTGCAGCTCGAGACTTGACGCGTCCGGGTAGCTGGCGTACGTGCTGCAGCCGGGTAGCTGGCGTACGTGCTGCAGCCGGGTAGCTGGCGTACGTGCTGCAGCTCGAGACTTGACGCGTCCGGGTAGCTGGCGTACGTGCTGCAGCCGGGTAGCTGGCGTACGTGCTGCAGCCGGGTAGCTGGCGTACGTGCTGCAG, with the italics marking the HRE tandem repeat.

Generation of stable cell lines containing the hypoxia-inducible dual reporter gene: transfection, clonal isolation, and enrichment. The HT29 cells were stably transfected with the dual reporter gene construct p9HRE-TKeGFP using calcium phosphate precipitation method (40–42). Colonies were selected and further enriched by repeated FACS gated for hypoxia-induced eGFP expression (Moflo cell sorter; Dako). In brief, the transfected cells were cultured in 400 μg/mL G418 for 3 to 4 wk. The drug-resistant cells were pooled and subjected to three cycles of FACS selection as follows. The cells were exposed to hypoxic condition (0.1% O₂) in an In vivo 2-400 hypoxic workstation for 24 h (Biotrace, Inc.), and then sorted by FACS based on the hypoxia-induced eGFP expression. The cells selected from the third sorting experiment were plated on 96-well plates (3 cells/mL; 200 μL per well). Individual clones (∼60) were isolated, expanded, and transferred to mini vials and stored at −80°C.

Subsequently, we evaluated a number of the individual clones to identify the ones that have the maximum hypoxia inducibility. Cells were thawed, expanded, and transferred into duplicate sets of 24-well plates. One set of plated cells were exposed to 0.1% O₂ for 24 h, whereas the other remained in normoxic conditions. FACS was use to evaluated the eGFP signal in the hypoxic cells vis-à-vis that of the control. Based on these evaluations, two clones (designated as #C53 and #C5) were chosen and characterized in terms of growth kinetics, morphology, hypoxia-inducible TK activity, and xenograft formation in nude mice. Both clones showed similar characteristics, and clone #C53 was used for this study.

Induction of hypoxia in cell cultures. To induce hypoxia, monolayers of parental HT29 and #C53 were exposed to 0.1% O₂ in an In vivo 2-400 hypoxic workstation for different time intervals (29, 40).

Radiotracer assay for TKeGFP expression in vitro. In vitro radiotracer accumulation studies with [¹⁴C] 2′-Deoxy-2′-fluoro-β-d-arabinofuranosyl-5-iodouracil (¹⁴C-FIAU) were performed to assess the hypoxia-induced TKeGFP expression (29). Briefly, #C53 cells (2 × 10⁵ cells per well in 6-well plates) were exposed to 0.1% O₂ for different time intervals. After this treatment, the culture medium was replaced with medium containing ¹⁴C-FIAU (Moravek Biochemicals; 0.025 μCi/mL; chemical purity, 97.7%) and incubated for 1 h at 37°C. The cells were then rinsed thrice with cold PBS and lysed with 1 mL of 0.3N NaOH/1% SDS. The lysates were neutralized (0.1 mL 3 N HCl) and assayed for protein level and ¹⁴C-FIAU activity using a Wallac 1410 liquid scintillation counter. The ¹⁴C-FIAU levels were normalized to the protein content of each sample. Under our experimental condition, no apparent quenching was observed.

Animal studies. Animal protocols were approved by the Institutional Animal Care and Use Committee at Memorial Sloan-Kettering Cancer Center (MSKCC). HT29 (parental) and #C53 tumors were formed by injecting 5 × 10⁶ cells (in 50 μL) s.c. into the limbs of 6- to 8-wk-old female nude mice maintained under pathogen-free conditions (NCr athymic nu/nu; National Cancer Institute Frederick Cancer Research Institute; except that the data of Fig. 6B was obtained with male mice of the same strain as the females were not available from the vendor). The tumors grew to a diameter of ∼6 and ∼10 mm in 8 to 9 and 10 to 13 d, respectively.

In each experiment, 5 to 6 animals were used with three tumors per mouse, the parental on the shoulder, and the #C53 on the two hind limbs. All experiments were performed at least twice for reproducibility. The data from all the animals were compared, and representative results, consistent with that from the others, are presented here.

When the xenografts reached ∼10 mm in diameter, serial microPET imaging was performed. A schematic of the in vivo imaging protocol using ¹⁸F-FMISO and ¹²⁴I-FIAU tracers is given in Fig. 2A. Briefly, on day 1 of the experiment, nude mice each bearing 3 xenografts were injected i.v. with 1.5 mCi of ¹⁸F-FMISO, and imaged with the microPET 2 h later. Immediately afterwards, the mice were injected with 630 μCi of ¹²⁴I-FIAU, and microPET imaging again performed 24 h later (day 2). Given the physical half-life of ¹⁸F (1.83 h), the image on day 2 was due entirely to ¹²⁴I-FIAU (physical half-life of ¹²⁴I is 4.2 d). The thyroids of the animals were blocked by adding SSKI to their drinking water 1 d before ¹²⁴I-FIAU injection.

Immediately after the acquisition of the ¹²⁴I-FIAU images on day 2 (Fig. 2A), the tumor-bearing mice were injected i.v. with 2.2 mCi ¹⁸F-FMISO and pimonidazole (80 mg/kg), so as to compare the intratumoral distributions of the tracers (¹²⁴I-FIAU and ¹⁸F-FMISO) using autoradiographs (phosphor plate imaging). Two hours later, the mice were sacrificed. Thirty minutes before sacrifice, lectin-TRITC conjugate (for functional vessels; 100 mg/kg; Sigma-Aldrich), and 1 min before sacrifice, Hoechst 33342 (for perfusion; 25 mg/kg; Sigma-Aldrich) were injected i.v., respectively. After animal sacrifice, tumors were excised, snap frozen, and embedded in ornithine carbamyl transferase. Sections of 8-μm thickness were prepared for dual-tracer digital autoradiographs, IHC analysis, and fluorescence microscopy studies of the same and/or adjacent sections.

For in vivo studies involving reduced PO₂ environment, mice with #C53 xenografts (6–8 mm in diameter) were kept for 30 h in a custom-made, sealed-environmental chamber (VetEquip) that was infused with 10% oxygen and 90% nitrogen. In one experiment, EF5 (24 mg/kg) was coinjected i.v. with pimonidazole 2 h before sacrifice. Control mice with matched tumor volumes were maintained in room air.

Radioisotope production and radiochemical synthesis. ¹⁸F and ¹²⁴I were produced on the MSKCC cyclotron (TR19/9; EBCO Technologies, Inc.) using previously described techniques (29, 35, 43). ¹⁸F-FMISO and ¹²⁴I-FIAU were synthesized as previously described (35, 43). Unlabeled precursors were purchased from ABX GmbH.

Noninvasive PET imaging and dual-label phosphor plate imaging. For noninvasive imaging, the animals were anesthetized with isoflurane and imaged with the Focus 120 microPET, a small-animal PET scanner (Concorde Microsystems). PET images were acquired and reconstructed as previously described (30). In general, a minimum of ∼10 million events were acquired in 5 to 40 min, depending on the radiotracer(s) injected, the administered activity, and the time postinjection (p.i.). The image data were corrected for nonuniformity of response of the microPET, dead time count losses, and physical decay. An empirically determined system calibration factor was used to convert voxel count rates (adjusted for the ¹²⁴I branching ratio) to activity concentrations, and the resulting data were then normalized to the administered activity to yield percent of the injected dose per gram of tissue (%ID/g or %ID/cc). Attenuation and scatter corrections were not applied because they are the same for the serial images and therefore unnecessary in image comparison.

To acquire the distributions of ¹²⁴I-FIAU and ¹⁸F-FMISO with improved spatial resolution (of ∼100 μm), dual exposure phosphor plate imaging was used (FUJI FILM). With this technique, the same specimen is used in two separate exposures. The first exposure was performed at the earliest possible time after tumor sectioning, for a duration of ∼4 h. The resulting image was a composite of the signals from both ¹²⁴I-FIAU and ¹⁸F-FMISO. The second exposure, initiated ∼48 h later after the complete decay of ¹⁸F, consists of the signal from longer-lived isotopic ¹²⁴I-FIAU only (half-life, 4.2 d). The duration of this second exposure was 48 h, to allow sufficient signal accrual. To obtain the ¹⁸F-FMISO image, we back decayed the second image (¹²⁴I-FIAU only) to obtain the contribution of ¹²⁴I-FIAU, and subtracted it from the first image (¹⁸F-FMISO + ¹²⁴I-FIAU). Digital autoradiographs were processed using image analysis software (FUJI FILM Science Lab 2003). Image colocalization study was performed using the software package ImageJv1.38s (44, 45).

IHC staining and fluorescence microscopy. Tumor sections were air dried for 30 min and fixed in ice-cold acetone for 20 min. Sections were incubated in SuperBlock (Pierce) for 30 min, and exposed to primary antibody diluted in blocking solution for 1 h at room temperature. Primary antibodies used were as follows: FITC-conjugated murine antipimonidazole monoclonal antibody (Chemicon) diluted 1:25; humanized anti-CA9 monoclonal antibody (cG250; a gift of Dr. Divgi, Memorial-Sloan Kettering Cancer Center, New York, NY; ref. 46), 25 μg/mL; and Cy3-conjugated anti-EF5 monoclonal antibodies (ELK3-51; a gift of Dr. Cameron J. Koch, University of Pennsylvania, Philadelphia, PA; ref. 47), 75 μg/mL. Sections were washed thrice with PBS and exposed to the appropriate secondary antibody for 1 h. CA9 was detected with an Alexa Fluor 568–conjugated goat anti-human antibody (Molecular Probes), 20 μg/mL. EF5 was detected with sheep anti-cyanine (5 μg/mL; USBiological) diluted in SuperBlock/mouse serum (1:1), and Alexa Fluor 568–conjugated anti-sheep antibody (20 μg/mL; Molecular Probes).

The sections were scanned on an image analysis system consisting of a Carl Zeiss Axiovert 200 mol/L fluorescence microscope using a computer-controlled motorized stage with a digital camera, and Metamorph software 7.0. All images were scanned at ×50 magnification. Composite images of sections were generated by the software from individual microscopic images. Sections were first imaged for eGFP, lectin-TRITC, and Hoechst, then stained for pimonidazole, EF5, and CA9, and imaged again with the appropriate markers visualized by green and red fluorescence, respectively. Finally the sections were stained with H&E according to standard protocol.

In vitro characterization of human colorectal tumor HT29-9HRE-TKeGFP cells. As was described, the #C53 cells, containing the dual reporter HSV1-TKeGFP under the control of 9HRE, were selected in medium containing 400 μg/mL G418, and enriched by three cycles of FACS sorting of live cells gated for hypoxia-induced eGFP expression.

FACS analysis of hypoxia-induced eGFP expression, after incubating #C53 cells in air or at 0.1% O₂ for 24 hours, is shown in Fig. 1B and C. The scatter plots in Fig. 1B, and of the two histograms in Fig. 1C, clearly shows a significant increase in the expression of eGFP in hypoxic cells relative to that of the oxic cells. A comparison of the fluorescent microscopic images of hypoxic and aerated cells (Fig. 1D) showed that green fluorescence was observed only in the hypoxic cells, confirming the results of flow cytometric analysis. In Fig. 1E, the kinetics of hypoxia-induced expression of eGFP in #C53 cells are shown. Upon exposure to 0.1% O₂, the cellular eGFP expression assessed by FACS (Fig. 1E) increased with time, reaching half-maximum after ∼8 to 12 hours, and a maximum (8–10 times that of control) after ∼16 to 20 hours. No changes in eGFP expression were observed in the HT29 parental cells under hypoxic condition, or in #C53 cells under aerobic condition (data not shown).

Figure 1F shows the hypoxia-induced expression/function of TK to phosphorated FIAU in #C53 cells after exposure to 0.1% O₂ for various times, as assessed by ¹⁴C-FIAU accumulation. Similar to the kinetics of hypoxia-induced eGFP expression (Fig. 1E), ¹⁴C-FIAU uptake increased with time under hypoxia (Fig. 1F). ¹⁴C-FIAU accumulation reached half maximum after ∼10 to 15 hours of hypoxia treatment, and the highest level (∼20 times of control) after 20 to 25 hours.

Characterization of hypoxia-inducible HT29-9HRE-TKeGFP tumor model in vivo. To verify our in vitro findings, we performed experiments using transplanted tumors (∼10 mm diameter) as was described in Fig. 2A (also see Materials and Methods). In Fig. 2B, the picture on the right illustrates the three tumors implanted in a nude mouse: parental (fore limb), and HT29-9HRE-TKeGFP (#C53) in both hind limbs. Serial microPET images are shown in the other two columns, with transverse images on the left and coronal images on the right. The top right panel (coronal) shows similar level of uptake of ¹⁸F-FMISO in the HRE (red arrows) and the parental (yellow arrow) tumors, with heterogeneous distributions of activity within each of the xenografts. In contrast, the 24 h p.i. image of ¹²⁴I-FIAU uptake shows significant difference among the xenografts. There was no perceptible ¹²⁴I-FIAU accumulation in the parental HT29 tumor but significantly detectable and heterogeneous distribution of ¹²⁴I-FIAU in the #C53 xenografts (bottom). The patterns of the tracer uptake in the #C53 tumor are similar for ¹²⁴I-FIAU and ¹⁸F-MISO (compare top and bottom), indicating that both are trapped in the same regions within the #C53 tumor. The middle panels are merged images of ¹⁸F-FMISO and ¹²⁴I-FIAU. Similar serial microPET images were acquired from at least 10 animals, and yielded consistent results.

Phosphor plate imaging of ¹⁸F-FMISO and ¹²⁴I-FIAU distribution. To compare the respective radioactivity distributions with improved spatial resolution, autoradiographs were acquired using phosphor plate imaging (see Materials and Methods; Fig. 2A). The middle panel of Fig. 3A were the second images, obtained at 48 hours p.i. of 2.2 mCi ¹⁸F-FMISO. Given the 1.83 hours physical half-life of ¹⁸F, these images were due entirely to ¹²⁴I-FIAU. We back-decayed these images to obtain the contribution of ¹²⁴I-FIAU, and subtracted it from the first image (obtained beginning ∼6 hours p.i. of 2.2 mCi ¹⁸F-FMISO and exposed for 4 h) to obtain the ¹⁸F-FMISO image (Fig. 3A,, top). Adjacent sections stained with H&E are shown in the bottom panels for comparison (Fig. 3A).

For the parental HT29 xenograft (Fig. 3A,, left), the ¹²⁴I-FIAU image (middle left) is nearly blank, consistent with the expectation for this negative control. The first image shows a heterogeneous distribution of radioactivity due to ¹⁸F-FMISO, indicative of tumor hypoxia. This first image is also consistent with the H&E image, in that there is negligible ¹⁸F-FMISO accumulation in the lightly stained necrotic area. For the #C53 (HRE-TKeGFP) tumor, the ¹⁸F-FMISO and ¹²⁴I-FIAU images are very similar and both indicate significant heterogeneity in tracer uptakes (comparing top right and middle right; Fig. 3A).

Digital autoradiograph images of ¹⁸F-FMISO and ¹²⁴I-FIAU were processed using the image analysis software FUJI FILM Science Lab 2003. The ¹⁸F-FMISO and ¹²⁴I-FIAU images were then coregistered and colocalization study performed using the software package Image J 1.38s (44, 45). In Fig. 3B, the images of ¹⁸F-FMISO (red), ¹²⁴I-FIAU (green), the overlay (yellow resulting from red plus green), and the correlation scatter plot are shown. The spatial distributions of ¹²⁴I-FIAU and ¹⁸F-FMISO are very similar, as indicated by the Pearson's coefficient of 0.88.

Comparison of endogenous and exogenous hypoxic markers. Given the experimental protocol of Fig. 2A, we were able to study and compare the spatial relationship of various biomarkers, including Hoechst (perfusion), ¹²⁴I-FIAU, pimonidazole, eGFP, and CA9, in either the same or adjacent tumor sections. Figure 4 shows images obtained in one experiment: (A) H&E, (B) ¹²⁴I-FIAU autoradiogram, (C) eGFP, (D) Hoechst, (E) composite of eGFP and Hoechst, (F) magnified view of a region in E, (G) CA9, and (H) pimonidazole. There was consistency in the observed distributions of the different biomarkers. In the necrotic zones, as identified in the H&E images, there was an absence of accumulation for all the tracers. The distributions of the endogenous markers eGFP and the exogenous reporter substrate ¹²⁴I-FIAU were identical as they should be because both stemmed from hypoxia-induced expression of the TKeGFP fusion protein. In addition, eGFP and ¹²⁴I-FIAU were most intense where there was low Hoechst staining, i.e., hypoxic regions with little perfusion. Similarly, the endogenous CA9 colocalizes with the exogenous pimonidazole but not with Hoechst. In general, there are good correlations in the spatial distributions of all four hypoxia markers, eGFP, ¹²⁴I-FIAU, CA9, and pimonidazole. The magnified view (F) of the fused image clearly shows that the hypoxia reporter eGFP (green) is in regions of low perfusion as indicated by low Hoechst staining.

The clear and unequivocal visualization of eGFP in #C53 tumor sections is a significant improvement relative to the previous R3327-AT and C6 models from this institution. The very high level of eGFP expression is likely the result of the 9HRE promoter and the repeated FACS sorting for selecting and isolating the #C53 cells. A careful examination of Fig. 4 shows low eGFP in necrotic regions (compare H&E section of A and the eGFP of C), high eGFP in viable but nonperfused regions (F), and concordance between eGFP pattern and those of the hypoxia-surrogates CA9 and pimonidazole. The white arrows and boxes in Fig. 4 (C, E, G, and H) highlight distinct features that illustrate the similarity in the distribution patterns of eGFP, CA9, and pimonidazole. These observations underline the usefulness of eGFP and the importance of the present tumor model for optical imaging of tumor hypoxia.

Figure 5 provides a more detailed examination and comparison of the different biomarkers in magnified views of a region from a HT29-9HRE-TKeGFP tumor (#C53). The top panels, from left to right, represent the following: (A) lectin-TRITC (functional blood vessels), (B) Hoechst (perfusion), (C) eGFP, (D) pimonidazole, and (E) CA9. The bottom row images are composites of two or three markers. The merged image of lectin-TRITC and Hoechst (Fig. 5F) shows functional blood vessels (red) surrounded by perfusion (blue), except for the vessel in the top right corner, which seems to be dysfunctional. The second from the left merged image (Fig. 5G) shows that eGFP (green) are located away from the blood vessels (red) and perfusion (blue), consistent with hypoxia-induced expression of the eGFP reporter gene. Also, in Fig. 5H, the exogenous hypoxic marker pimonidazole (green) is accumulated in regions of low perfusion, as is the endogenous hypoxic marker CA9 (red) in Fig. 5I. Lastly, the image in Fig. 5J shows colocalization of CA9 and pimonidazole (yellow, resulting from red plus green) in regions of low Hoechst.

As a further demonstration of the hypoxia-inducibility of the reporter system in vivo, we compared eGFP and pimonidazole in tumors of animals that breathed air vis-à-vis 10% O₂. In these experiments, we used ∼6-mm diameter tumors that under normal conditions have low hypoxic fractions and low eGFP expression. Figure 6A (a and b) shows that there is little eGFP or pimonidazole in tumors of an air-breathing mouse, whereas there is high eGFP and pimonidazole signals in tumors of an animal breathing 10% O₂ (Fig. 6A,, c and d). In another experiment, we compared the detailed spatial distributions of eGFP, pimonidazole, CA9, and EF5 in animals breathing air vis-à-vis 10% O₂. Whereas there was hardly any signal of these hypoxic markers in air-breathing animals (data not shown), there was significant accumulation of all 4 markers in the tumors of animals breathing 10% O₂ (Fig. 6B). These data clearly showed that the system is responsive to manipulation of tumor oxygenation.

In this study, we established and evaluated a human xenograft model (#C53 or HT29-9HRE-TKeGFP) that links hypoxia-mediated molecular pathway to hypoxia imaging, specifically optical imaging with IHC and noninvasive PET imaging. Because the expression of TKeGFP is a direct consequence of HIF-1α stabilization/HIF-1 transactivation, the first step of a major hypoxia-regulated molecular pathway, the level of the TKeGFP in the tumor is a surrogate of hypoxia-mediated HIF-1α stabilization/HIF-1 transactivation and, thus, can serve as a reference for downstream events and other tumor hypoxia markers.

A very important element of the present study is the extremely high level of hypoxia-induced eGFP expression such that the intense green fluorescence signal can be easily visualized in freshly frozen tumor sections. This permitted, for the first time, eGFP optical imaging of hypoxia in tumor sections and provided a valuable tool for the study of endogenous and exogenous hypoxia markers in vivo. Optical imaging is easier to perform than nuclear imaging and, in this case, involves endogenous eGFP expression, thereby obviating the need of marker substrate (e.g., in bioluminescence imaging).

In this study, we showed the usefulness of this approach by comparing the distribution of eGFP with those of CA9, pimonidazole, EF5, and Hoechst in tumor sections. CA9, known to be regulated by the hypoxia-mediated signaling pathway in many cell types, has often been used as an endogenous surrogate for tumor hypoxia (23, 24). In Figs. 4 and 5, it is shown that the distributions of eGFP and CA9 are well-correlated. Pimonidazole, FMISO, and EF5, members of the 2 nitro-imidazole family of compounds, form adducts that become trapped within hypoxic cells and are commonly used in hypoxia studies (25). As shown in Figs. 4, 5, and 6, the spatial distributions of eGFP and pimonidazole are extremely similar under all experimental conditions. We believe that this provides evidence for the first time of the linkage between the initial hypoxia-induced molecular events, i.e., HIF-1α stabilization and HIF-1 transactivation, and the utility of 2 nitro-imidazoles in marking tumor hypoxia in vivo.

To show the effect of tumor oxygenation manipulation on our model system, we compared the accumulation of eGFP, CA9, pimonidazole, and EF5 in animals breathing either 10% O₂ or air. Whereas there is little accumulation in air-breathing mice, there is high eGFP, CA9, EF5, and pimonidazole signals in mice breathing 10% O₂ (Fig. 6). In addition, the spatial distribution patterns of the four markers were similar.

Another important aspect of this tumor model is that the HSV1-TK gene is also expressed as a consequence of tumor hypoxia. This permits the use of noninvasive PET scanning with the marker substrate ¹²⁴I-FIAU to image hypoxia-induced gene expression and the associated spatial distribution of tumor hypoxia. Thus, the transactivation of the reporter gene by tumor hypoxia can be imaged in three-dimensional, generating a map of the molecular events associated with tumor hypoxia, i.e., hypoxia-induced HIF-1α stabilization leading to the reporter expression. This is shown by the PET images of Fig. 2. In addition, by comparing the respective PET images (Fig. 2), one can compare and spatially correlate the distributions of ¹²⁴I-FIAU (and the associated hypoxia-induced molecular events) and the exogenous hypoxia marker ¹⁸F-FMISO, and validate the use of ¹⁸F-FMISO in imaging tumor hypoxia.

To compare the distributions of ¹²⁴I-FIAU and ¹⁸F-FMISO with better spatial resolution (relative to PET images), we used the dual-tracer autoradiographic technique. By sequentially exposing the same tumor section to the phosphor plate at different times p.i. of the two tracers, and exploiting the different physical half-lives of ¹²⁴I (4.2 h) and ¹⁸F (1.83 h), we derived the respective distributions of ¹²⁴I-FIAU and ¹⁸F-FMISO. As shown in Fig. 3A, the detailed spatial distributions of ¹²⁴I-FIAU and ¹⁸F-FMISO seem identical in a #C53 tumor section. In contrast, they are totally disparate in the parental tumor because the HSV1-TK protein was absent, ¹²⁴I-FIAU cannot be phosphorylated and trapped inside the hypoxic cells. Statistical analysis showed that the spatial distributions of the two tracers are well-correlated in the #C53 tumor section (Fig. 3B).

As shown in Figs. 4 and 5, the multiparametric analysis of the different manifestations of tumor hypoxia generated a coherent picture. All the exogenous (pimonidazole and ¹⁸F-FMISO) and endogenous markers (CA9) colocalize with the hypoxia-induced expression of eGFP, and with HSV1-TK as assessed by the trapping of the marker substrate ¹²⁴I-FIAU. In addition, they are located away from the perfused regions as marked by Hoechst staining.

In summary, we have developed and characterized a novel human xenograft model in which the hypoxia-induced molecular events can be visualized by either optical (fluorescent microscopy) or nuclear imaging (PET or autoradiography) techniques. Tumor hypoxia distributions determined by these methods have been compared with those measured by other surrogates (such as pimonidazole, EF5, ¹⁸F-FMISO, and CA9), yielding excellent agreement. This system is responsive to manipulation of tumor oxygenation and provides a powerful tool for hypoxia research in vitro and in vivo, including those directed at therapeutic intervention targeting the hypoxia pathways.

No potential conflicts of interest were disclosed.

Grant support: NIH grants CA56909, CA109772, and CA115675. This publication also acknowledges NCI grant P30-CA 08748, which provides partial support for the Research Animal Resource Center (RARC), Animal Imaging, Molecular Cytology and Flow Cytometry core facilities at MSKCC used in conducting this investigation.

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.

We thank Dr. Dennis F. Deen for providing the 9HRE DNA fragment, Drs. Ronald Blasberg and Juri Gelovani for the plasmid containing tkeGFP fusion gene, and Maayan Korenblit for her assistance in the preparation of the manuscript.