In Vivo Profiling of Hypoxic Gene Expression in Gliomas Using the Hypoxia Marker EF5 and Laser-capture Microdissection

Hypoxia/anoxia is a well-characterized component of the solid tumor microenvironment and has profound consequences for patient outcome. Hypoxic tumors are more resistant to radiotherapy and chemotherapy and patients with more hypoxic solid tumors fare worse compared with patients with better oxygenated tumors, irrespective of therapeutic modality (1, 2). At the cellular/molecular level, hypoxia promotes selection and clonal expansion of cells with inactivated p53 (3), increases metastatic potential (4, 5), and promotes genomic instability (6). These laboratory and clinical findings strongly suggest that hypoxia selects for a more aggressive tumor phenotype (7) and have stimulated efforts to develop accurate and tractable ways to image and quantify tumor hypoxia in human patients.

Several invasive and noninvasive methods currently exist for measuring hypoxia in solid tumors. Although there is no consensus among hypoxia researchers on which method is considered the “gold standard” of hypoxia measurements in vivo, the two most widely used and referenced methods are polarographic electrodes, which are invasively inserted into accessible tumors in patients under anesthesia, and bioreductive dyes including EF5 and pimonidazole, which are injected into patients prior to surgery (2). EF5 is a 2-nitroimidazole that has been used for more than 15 years in both in vitro and in vivo studies to measure hypoxia (8). The drug is reduced in regions of hypoxia, allowing for covalent binding to cellular proteins. Binding is stable so that after the tumor is removed and sections are made, they can be stained using anti-EF5 monoclonal antibodies conjugated to fluorochromes such as Cy3. By measuring EF5 immunofluorescence and then comparing it to a calibrated fluorescence scale that is generated using cells grown under specified O₂ concentrations in vitro, the absolute po₂ present in tissues can be estimated (9).

During the last two decades, several molecular “hypoxia markers” have been identified (e.g., CA9, GLUT-1, VEGF) and are currently being investigated as qualitative and quantitative correlates of tumor hypoxia, but with mixed results (10). The lack of reliable hypoxia markers seems to stem from a) the great intratumoral and intertumoral heterogeneity of hypoxia, b) the coregulation of the expression of such genes by factors other than hypoxia (e.g., oncogenes), and c) the disparity in regulation of expression of specific genes in vitro versus in vivo (10).

Gene expression profiling of tumors has shown promise in the detection of patient populations, with markedly distinct prognostic outcomes in retrospective and prospective studies (11, 12). Specific tumor gene expression “signatures” that can be used to predict patient outcome with considerable reliability are expected to revolutionize medicine in the next few decades (13). The regulation of mRNA and miRNA expression by hypoxia has been studied extensively in vitro (14, 15). Gene expression analyses using oligonucleotide-based microarrays following exposure of cells to hypoxia and comparison to normoxic controls have confirmed the central role of the hypoxia-inducible factors HIF-1 and HIF-2 in mediating a large portion of hypoxic mRNA regulation (15, 16). Despite this progress, there have been no studies reporting a comprehensive analysis of hypoxia's influence on gene expression in vivo due to problems with mRNA integrity following immunohistochemical staining of frozen tissue. As hypoxia in tumors exists in a background of additional physiologic stresses (e.g., low nutrients, low pH, infiltrating immune cells), it is likely that the intricate interplay of these stresses on gene expression profiles in vivo will not be recapitulated faithfully by in vitro experiments utilizing these stresses individually (17). These confounding factors likely contribute to the lack of correlation between expression of known hypoxia-regulated genes, with immunohistochemical detection of these markers in areas of hypoxia in vivo (17, 18).

Here, we report the identification of the hypoxic gene expression profile in a solid tumor, using a prototypical hypoxia marker EF5, laser-capture microdissection (LCM), and microarray gene expression profiling (EF5-LCM-MGEP). We have utilized LCM to isolate total RNA from normoxic and hypoxic regions in the rat 9L glioma tumor model grown in its isogenic host (Fischer rat). Microarray analysis of the isolated RNA generated an in vivo hypoxia expression profile. For comparison, we also analyzed RNA samples isolated from rat 9L glioma cells in culture exposed to hypoxia or normoxia. Microarray results for select endogenous hypoxia-related markers were validated through quantitative PCR (qPCR). Our analysis reveals some major differences between in vitro and in vivo gene expression signatures, including a pronounced repression of mRNAs coding for immune system and DNA repair–related proteins in hypoxic areas of these solid tumors.

The 9L gliosarcoma cells were grown in MEM media with 12% newborn calf serum with twice-weekly transfer. To initiate the experiment, cells were either incubated at approximately 20% O₂ (“normoxia”) or 0.2% O₂ (“hypoxia”) for 6 or 16 hours. Cells were subsequently washed with HBSS (Invitrogen), detached with 0.25% trypsin-EDTA, and resuspended in RNA lysis solution (Ambion) at a concentration of 50,000 cells/100 μL lysis solution. The lysate was stored at −80°C prior to RNA isolation.

All animal experiments were conducted in accordance with NIH guidelines and with the approval of the University of Pennsylvania Animal Use Committees (IACUC). Six-week-old male Fischer 344 rats (F344/Ncr; National Cancer Institute, Frederick, MD) were inoculated via subcutaneous injection on the hind flank with 9L cells (10⁶) to generate passage “zero” tumors (p0). The tumor was excised, cut into 1- to 2-mm³ pieces and used to generate subsequent passages. The epigastric tumors were initiated by tying p5 or p6 tumor pieces onto the epigastric artery–vein pair as previously described (19).

Once 9L epigastric tumors attained a size of 1–1.5 g, animals were anesthetized with 2.5% isoflurane and maintained at 35°C–37°C with a heated pad. The EF5 bolus was administered via tail vein catheter at a dose of 30 mg/kg. Three hours after EF5 administration, 1.0 mL Hoecsht (10 mmol/L) was injected via tail catheter. After 5 minutes, the tumor was excised under 4% isoflurane in air and embedded in 3% low-melt agarose (Biorad Laboratories). The embedded tumor was then frozen on a dry ice-cold aluminum block and stored at −80°C.

See Supplementary Information

LCM sections were fixed for 1.5, 1.5, and 5 minutes with ice-cold 70%, 85%, and 100% EtOH, respectively, followed by a quick rinse in 100% EtOH at room temperature and then dried. Tissue sections were incubated in a humidified chamber for 30 minutes at 4°C with 150 μL antibody solution consisting of 540 μL Rinse solution, 30 μL Cy3-ELK3-51 antibody (EF5 antibody, 4 mg/mL), and 30 μL RNasin (40 U/μL; Promega Corp.). Rinse solution contained 3 mL RNase-free H₂O, 32 μL 1 mol/L HEPES, 160 μL 20 × Earle's NaCl, and KCl salt and 40 μL RNasin. Slides were washed 6× with 100 μL rinse solution and with increasing ice-cold EtOH concentrations (70%, 85%, and 100%). Slides were dipped into 100% EtOH maintained at room temperature, allowed to dry, and stored in sterile 50 mL tubes with desiccant (4A molecular sieve; Aldrich) prior to LCM. If kept more than 1.5 hours before LCM, the tubes were stored in dry ice, before returning them to room temperature, when the slides were dipped into 100% ethanol at room temperature and dried. For codetection of hypoxia and HSP27, Rad51 and CD8⁺ cells, frozen tumor sections were incubated with the corresponding primary antibodies following incubation (anti-HSP27; Cell Signaling) or together (anti-Rad51, Thermo Scientific; HIS52 and anti-Rat CD8; Biolegend) with the Cy3-ELK3-51.

LCM was carried out using the PALM system from Zeiss (LCM Facility, Children's Hospital of Philadelphia). Stained 9L rat gliosarcoma tumor sections mounted onto Zeiss membrane slides were kept on dry ice and dehydrated in 100% EtOH just prior to use in order to optimize excision and limit RNA degradation. EF5 fluorescence was detected with the Texas Red filter and indicated areas of hypoxia, whereas the antibody absence corresponded to normoxic regions. Tissue displaying necrotic morphology based on fragmented appearance of nuclei (Hoechst 33342) was avoided. Once sufficient normoxic sample had been excised (∼100 ng), the tissue was dissolved in 100 μL lysis solution and stored at −80°C pending RNA isolation. Hypoxic tissue was collected and processed similarly.

See Supplementary Information.

Total RNA was linearly amplified using the NuGEN Ovation WT-Pico kit. Briefly, total RNA (2 ng/sample) was reverse transcribed to generate cDNA that incorporated an RNA primer binding sequence. After second-strand synthesis, cDNA templates were amplified by ribo-SPIA. The resulting amplified cDNA was fragmented, biotinylated, and hybridized to Rat Gene 1.0ST Arrays (Affymetrix). Microarrays were stained with streptavadin-phycoerythrin and incubated with biotinylated antistreptavadin to enhance fluorescence signal detection. Comparison of signal intensity yielded relative gene expression under hypoxia/normoxia.

See Supplementary Information.

All analyzed microarray data are in accordance with the MIAME guidelines and microarray data has been deposited at the GEne Expression Omnibus (GEO) site https://www.ncbi.nlm.nih.gov/geo/ (Samples GSM637895-GSM 637912). Affymetrix.cel files were imported into Partek Genomics Suite (v6.5b; Partek Inc.). Robust Multichip Average normalization was applied. Global sample variability was assessed by principal components analysis (PCA). To identify differentially expressed genes, a 3-way mixed-model ANOVA was calculated, with the factors oxy state (H or N) sample type (vivo, vitro 6, vitro16) and batch/tumor. An interaction term between oxygenation state and sample type was also included, as were pairwise contrasts of H versus N for each of the sample types. P values resulting from the ANOVA but not the pairwise comparisons were corrected for multiple testing using the stepup method of Benjamini and Hochberg as implemented in Partek (20). For each of the sample types, the appropriate pairwise contrast was filtered for significance (P < 0.05) and then by fold change to yield 2 gene lists, one for those genes that were more than 2× higher in hypoxia and for those genes that were less than −2× lower in hypoxia. This yielded 6 gene lists, which were then compared using Venn diagrams, one for all of the lists with genes up in hypoxia and a second for the lists containing genes down in hypoxia. For the gene list used in the hierarchical clustering, the most significant genes (false discovery rate 20%) for the interaction term between sample type and oxy state were selected. This yielded 49 genes. To generate Heat Maps for the 3 experimental conditions, separately, for up and down changes, the significant gene lists described for each of the 3 comparisons (cutoffs 2-fold, P < 0.05) were combined. The log₂ ratios for each tumor pair were used to hierarchically cluster genes (188 genes from the union of up lists, 549 from the union of down lists) and samples.

See Supplementary Information

In seeking an appropriate animal model in which to study gene expression under defined conditions of tissue hypoxia, we considered the following parameters: the tumor should exhibit properties similar to human tumors, have a well-defined hypoxic response, and be of sufficient size to allow large tissue areas to be collected by LCM. The 9L gliosarcoma grown in its isogenic host (Fischer rat) as an epigastric pedicle has been well characterized by us (19) and fulfilled several of these parameters (Fig. 1A). This tumor (a) is isogenic to the Fischer rat, allowing for interrogation of the influence of the immune system in hypoxic gene expression, (b) has a well-developed stromal support and has been described as pathophysiologically similar to human glial and sarcoma tumors, and (c) is one of a handful of rodent tumors with intertumor variation in hypoxia-mediated radiation response similar to human tumors, as measured by EF5-binding and radiation response (21). The 9L tumors develop relatively normal stroma with diffuse, scattered, rather than central, necrosis (ref. 22; Fig. 1B).

Initially, we attempted to isolate total RNA from tumors following EF5 staining protocols long established by our group (21). However, we found that previously established fixation methods and the extensive time of incubation with blocking solution and anti-EF5 antibody led to significant RNA degradation that was not acceptable for subsequent microarray gene expression analysis. Therefore, considerable effort was devoted to optimize the protocol for EF5 staining. We adopted ethanol fixation and a very fast “quick-stain” (45 minutes for the whole procedure), using highly purified, Cy3-labeled anti-EF5 antibodies and excess of RNase inhibitor to overcome this limitation.

EF5 antibody binding was detected using the Texas Red filter. Regions of high fluorescence corresponded to regions of hypoxia, whereas areas with minimal fluorescence corresponded to regions of normoxia. Areas without fluorescence corresponded to regions of normoxia. To ensure that EF5-negative regions consisted of viable, normoxic tumor cells and not necrotic areas (which do not metabolize EF5 to the active nitroimidazole), we stained serial sections placed adjacent to the EF5-stained regions with Hoechst 33342 dye. Regions with necrotic tissue have a distinct morphology with a more sparse arrangement of nuclei that also exhibit a smaller and fragmented morphology. We avoided such areas in the corresponding EF5 slide and collected normoxic tissue from areas devoid of these characteristics. Using this approach, we collected separately approximately 100 ng of normoxic and hypoxic tissue from 2 sections each and repeated the process for 3 different 9L tumors (Fig. 1C). The quality of the isolated RNA was analyzed using the Agilent Bioanalyzer (Penn Microarray Facility). A RNA Integrity Number (RIN) of 7 was deemed of sufficient quality for further amplification and array analysis. As shown in Figure 1C (right and bottom panels), the RNA isolated in this manner had clearly visible bands representing the large ribosomal RNA (28s and 18s) as well as small RNA (5s, tRNA).

Because of the small amount of RNA isolated from the in vivo samples, a high-efficiency linear amplification step was necessary prior to hybridization to the gene microarray (for technical consistency, the same amplification method was carried out using the same amount of input RNA from the in vitro samples). For this, 100 ng of RNA was amplified using the NuGEN Ovation WT-Pico kit. Following the amplification step, the cDNA was hybridized to the Affymetrix Rat Gene 1.0ST Array for 16 hours. Analysis of the expression signal was carried out by the Penn Microarray Facility, using established methods (Affymetrix Expression Console), and statistical tests for significant differences between conditions were carried out using Partek Genomics Suite. An aliquot from each amplification reaction was preserved for subsequent verification of select upregulated or downregulated genes with real-time qPCR.

To compare the gene expression profile of in vivo isolated RNA from hypoxic and normoxic areas to that obtained after in vitro exposure to hypoxia or normoxia, we exposed 9L rat cells to hypoxic conditions, using an established hypoxia control system (23). In vivo, tumor cells can be exposed to a wide range of oxygen tensions, ranging from normoxia (5%–8% O₂ for most tissues) to complete anoxia for different periods of time. As analysis of all possible combinations of po₂ and time of exposure would be prohibitively time-consuming and costly, we chose to use stringent hypoxic conditions (0.2% O₂) for 6 and 16 hours. These conditions are routinely used for in vitro analysis of hypoxic gene expression (24). Furthermore, clinical studies using EF5 in human patients with glioblastomas indicate that po₂ levels in this range exist in these tumors (25). Total RNA isolation, analysis of its quality, and hybridization to oligonucleotide arrays were done right after treatment as described above for in vivo–isolated RNA.

As a first step in the analysis of the in vivo and 6 and 16 hours in vitro samples (thereby referred to as IVV, IVR6, and IVR16, respectively), we conducted PCA using all 3 of the independent replicate samples for each condition. PCA is a mathematical algorithm that allows a 3-dimensional graphical representation of variation within a data set by reducing dimensionality of the data (26). As shown in Figure 2A, the 3 biological replicates of each experimental group largely clustered together in 3-dimensional space, with the exception of one of the IVV replicates that seemed to diverge from the other two IVV samples. This analysis indicates that there may be higher variability of gene expression pattern in vivo compared with that in vitro and that the different duration of hypoxic exposure results in distinct clustering of expressed genes.

Further analysis of gene expression between each group was done using supervised clustering of the top upregulated and downregulated mRNAs, as described in the “Methods” section. As shown in Figure 2B, the 2 IVR samples showed similar clustering of these specific mRNAs. In contrast, in the IVV sample, there was a more pronounced increase in the expression of several upregulated mRNAs and there was a distinct cluster of mRNAs whose expression was significantly downregulated in vivo (blue) but upregulated or not changed in vitro. A similar trend was observed when the results were analyzed without filtering for level of induction but only for statistical significance (P < 0.05; Supplementary Fig. S1A). Indeed, analysis of the top 20 downregulated and upregulated mRNAs (i.e., the 20 genes demonstrating the highest fold-increases or decreases with a P < 0.05; (Fig. 3) shows that whereas in vitro, mRNAs displayed a maximum of 2-fold (6 hours) or 2.4-fold (16 hours) decreases under hypoxia, in vivo, the decreases in mRNA levels ranged from 7.0-fold to 3.6-fold (Fig. 3C). Among the upregulated mRNAs with >2-fold changes, 17 were found to be upregulated by all 3 conditions whereas 13 were upregulated only in vitro but not in vivo (Fig. 2C). Remarkably, 115 genes were significantly upregulated in vivo but not by either in vitro condition. Also surprising was the finding that there were no commonly repressed genes among the 3 conditions. In stark contrast, 539 genes were significantly downregulated (>2-fold decreases) in hypoxic compared with normoxic regions in vivo (Fig. 2C). Similar patterns were observed in the analysis of common and unique mRNAs among the 3 data sets filtered only for statistical significance (P <0.05, no fold change restriction). Here too, the number of significantly upregulated and downregulated mRNAs in hypoxic versus normoxic areas in vivo was substantially larger than those in the 2 in vitro samples (Supplementary Fig. S1B).

Both in vitro conditions generated results that are in agreement with those previously reported by several studies. For example, well-known hypoxia and HIF-regulated mRNAs such as Pgf, CA9, PDK1, Ndrg1, etc., were significantly upregulated by in vitro hypoxia at both time points (Fig. 3A and B; ref. 27). Interestingly, from all the repressed genes, under both conditions, only one (Cyp1a1) has been previously shown to be downregulated by hypoxia (ref. 28; Fig. 3).

Of the 20 top in vivo upregulated genes, 14 are known hypoxia-regulated genes, a finding that validates our EF5-LCM-MGEP approach. Notably, Bnip3, a gene shown in several studies to be upregulated by HIF-1 and HIF-2 and to mediate hypoxia-induced apoptosis and autophagy (29, 30) was the mRNA with the highest hypoxic upregulation. This was followed by heme oxygenase I (decycling), which mediates hypoxia-induced metabolic changes in cells, the transcription factor ATF3, which is activated in response to endoplasmic reticulum stress, and hypoxia and HSP27 (Hsbp1), a small heat shock protein, known to be a direct HIF transcriptional target and to confer antiapoptotic properties to tumor cells under hypoxia (refs. 31, 32; Fig. 3C). However, our analysis also revealed many mRNAs upregulated in hypoxic areas including cyr61, Jmjd1a, and Znf609 that have not been previously directly implicated in hypoxic gene regulation. The in vivo analysis has revealed for the first time a number of mRNAs whose expression was markedly lower in hypoxic compared with normoxic areas. Intriguingly, a majority of these mRNAs seem to play a role in immune system functions, specifically in lymphocyte- and leukocyte-mediated immunity and antigen presentation. These gene products include Ly6C, CD3D, and CXCL9 (33–35). A recent study found that increased expression of CD3D mRNA associated positively with increased survival in patients with metastatic melanoma. Another interesting finding was the downregulation of mRNA levels of several proapoptotic such as Ddx60 and DNA repair proteins such as Rad51 (36).

To verify the results obtained from the gene microarray analysis, we conducted qPCR on total mRNA isolated in vitro and in vivo. As shown in Figure 4A and B, the levels of Vegf, Pdk1, Glut1, and CA9 were all significantly upregulated by both 6 and 16 hours of hypoxia in vitro. The magnitude of the increases was larger based on qPCR and is within the range of published studies. Expression of Vegf and Glut1 was also found to be significantly upregulated in the in vivo samples (Fig. 4C).

To further analyze the effects of hypoxia on cellular processes and pathways in the solid tumors, we retrieved gene ontology (GO) annotations using mRNAs with >2.5-fold changes in normoxic versus hypoxia areas using the GOrilla application (Technion University; ref. 37). As expected from the analysis of individual genes and from previous published in vitro studies, the top pathways whose genes were upregulated by hypoxia-included glycolysis and glycolysis-related processes, hypoxic response, stress response, and angiogenesis (Supplementary Table S1). In contrast, pathways that included genes with decreased expression included those involved in immune system processes such as antigen recognition and chemotaxis, cell-cycle progression, DNA synthesis, cell killing, and DNA repair (Supplementary Table SII).

To test whether the expression of select proteins correlates with that of their corresponding mRNAs, we analyzed the expression patterns of 2 proteins whose mRNAs were found to be significantly upregulated (HSP27-Hsbp1; 4.7-fold increase, P = 0.002) or downregulated (Rad51; 3.0-fold decrease, P = 0.01) in hypoxic versus normoxic areas, respectively (Fig. 3 and Supplementary Data). As shown in Figure 5A and Supplementary Figure S1, expression of HSP27 was rather heterogeneous throughout the tumor but primarily localized at or near areas of hypoxia (red and orange in combined images). In contrast, the expression of Rad51 was found to be primarily absent from hypoxic areas and was highest in areas of normoxia (Fig. 5B). Interestingly, the anti-Rad51 antibody also seemed to stain blood vessels, which in their majority are normoxic. Although this pattern of expression of Rad51 has not been previously reported, interrogation of a publicly available gene expression database (Novartis BioGPS) showed that expression of Rad51 mRNA is highest in endothelial cells, reaching levels 9 times higher than the median expression in other cell lines or tissues (Supplementary Figure S3).

As several mRNAs that were found to be downregulated in hypoxic areas code for proteins with roles in T-cell recruitment (e.g., CXCL9, Hcst/DAP10), we analyzed the relationship between CD8⁺ cells and hypoxic areas in the rat 9L gliomas. Immunohistochemical staining with an Alexa 488–labeled anti-CD8 antibody revealed the presence of several CD8⁺ cells with strong membrane staining in the tumor (Fig. 6A–C), although the strongest staining was in the tumor periphery (not shown). Costaining for EF5 adducts revealed that the vast majority of CD8⁺ staining was found outside hypoxic areas (19.7% of cells in normoxic areas vs. 5% in hypoxic areas were CD8⁺) and was concentrated primarily in the vicinity of Hoechst-perfused blood vessels (Fig. 6A and C).

Direct, in vivo interrogation of the effects of components of the tumor microenvironment on gene expression offers significant advantages over similar in vitro approaches. First, EF5-LCM-MGEP offers the opportunity to analyze the effects of one stress (in this case, low oxygen tension) in a physiologic setting, which invariably includes additional parameters such as nutrient availability, low pH, and high interstitial pressure, all of which are known to influence gene expression in tumor cells (38, 39). Second, as our results indicate, this type of analysis has the potential to uncover stress responses that might be unanticipated or simply could not be interrogated under in vitro conditions. In our study, we observed a pronounced downregulation of a large number of mRNAs associated with immune system regulation and function in hypoxic areas, raising the intriguing possibility that hypoxic tumor areas represent an immunoprivileged tumor niche. If confirmed in other types of tumors, this finding would predict a new resistance factor for hypoxic tumors.

This in vivo approach of gene expression analysis is not without some limitations. For example, without further analysis, it cannot be determined whether the observed decreases in gene expression of immune response gene products under hypoxia are the result of direct downregulation of transcriptional programs or the consequence of over/underrepresentation of one type of lymphocyte or macrophage over another in hypoxic regions. Tumor-associated macrophages (TAM) have been shown to segregate into 2 broadly defined classes: the prooxidant, cytotoxic, and antitumorigenic type M1 and the proangiogenic, protumorigenic type M2 (40, 41). Our findings indicate that expression of genes associated with the M1 phenotype (Ly6C, Cxcl9, Cxcl10; refs. 42, 43) are significantly underrepresented in hypoxic versus normoxic areas. This could arise from either (a) an active repression of M1-associated genes under hypoxic stress or (b) exclusion of M1 or attraction of M2 in hypoxic compared with normoxic tumor areas. Our findings that hypoxic areas are largely devoid of CD8⁺ cells (cytotoxic T lymphocytes) argue for the latter mechanism. However, it is formally possible that CD8 expression may also be downregulated by hypoxia, even though it did not seen in our list of downregulated mRNAs. Intriguingly, in a recently published study, Doedens and colleagues reported that TAMs under hypoxic conditions negatively regulate T-cell function through HIF-1α (44). Although in this study the localization of T cells in solid tumors with relation to hypoxic areas was not analyzed, these results suggest that hypoxia may have additional repressive effects on T-cell activity, thereby promoting tumor aggressiveness. Additional studies with specific markers and immunohistochemical localization of the 2 types of macrophages and T cells with respect to hypoxic regions in vivo would help elucidate the precise mechanism, and these studies are currently under way in our laboratory.

Our analysis has revealed a significantly higher number of mRNAs affected by hypoxia in vivo compared with in vitro. The increased number of upregulated mRNAs in vivo versus in vitro is in agreement with other studies showing a more robust gene upregulation in response to stresses in vivo. For example, Khodarev and colleagues showed that ionizing radiation induced significantly higher upregulation of genes in xenografts than in in vitro cultures (45) and Camphausen and colleagues have shown that the tumor microenvironment could exert substantial influence on gene expression in gliomas in response to ionizing radiation (46). Although several mechanisms could underlie this difference for hypoxia, it is noteworthy that there seems to be a trend toward higher number of upregulated mRNAs from 6-hour hypoxia to 16-hour hypoxia (from 37 to 66, respectively). It is possible that in vivo, where moderate hypoxic conditions can persist for more than 16 hours without significantly affecting cell survival, the prolonged exposure to low oxygen levels can result in a higher number of upregulated genes.

Identification of an mRNA “hypoxic signature” in solid tumors has important implications with clinical significance. For example, several proteins such as GLUT-1, VEGF, and osteopontin have been shown to correlate with and predict a poorer patient prognosis (47). More recently, hypoxia-induced miRNAs such as mir210 have been associated with increased metastasis and poorer patient outcome (48, 49). These studies are almost invariably initiated by in vitro analyses of gene expression patterns and subsequent immunohistochemical analysis (in the case of proteins) or PCR analysis (in the case of miRNA) in patient samples. The EF5-LCM-MGEP method offers a more systematic approach in which multiple targets are identified and validated in vivo and then a select few targets of interest are moved to the clinic. Our approach is readily amenable to further analysis of gene expression to include miRNA and protein expression using miRNA microarrays and proteomics, respectively. Preliminary studies in our laboratory have shown feasibility for the analysis of miRNAs by using a similar approach, and we are in the process of obtaining the hypoxic miRNA expression profile in human solid tumors. It is therefore conceivable that in the near future, a systems-level analysis of the hypoxic response in solid tumors will be feasible that could uncover novel targets for therapeutic intervention.

No potential conflicts of interest were disclosed.

We thank Luca Bernabei and Xiangyang Yang for expert technical help. We also thank Andy Minn (University of Pennsylvania), Rob Bristow (University of Toronto), and Edith M. Lord (University of Rochester Cancer Center) for helpful discussions. Donald and Andrea Marotta are especially appreciated for their continued support.

This work was supported by an ORISE fellowship from the Oak Ridge National Laboratory to D. Marotta and by pilot funds from the Department of Radiation Oncology, UPenn, to C. Koumenis.

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.