1 Thermal and Mountain Medicine Division, United States Army Research Institute of Environmental Medicine, Natick 01760
2 Division of Pulmonary and Critical Care Medicine
3 Cardiology Division and the Gene Array Technology Center, Brigham and Womens Hospital/Harvard Medical School, Boston, Massachusetts 02115
![]() |
ABSTRACT |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
cDNA array; hypoxic stress; cellular hypoxia; heat shock; cell stress response
![]() |
INTRODUCTION |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
Changes in gene expression are an integral part of the human cellular response to hypoxia (30, 31). To date, at least 50 mammalian genes have been identified which show a change in expression during hypoxic exposure, including a number of genes that are thought to be part of a nonspecific cellular response to stress. Furthermore, at least three important mechanisms for altering gene expression during hypoxia have been identified, namely, changes in transcription [mediated by well-described transcription factors including hypoxia-inducible factor 1 (HIF-1); Refs. 30, 31], stabilization of hypoxia-sensitive RNA species [such as vascular endothelial growth factor (VEGF)] (7), and the existence of internal ribosomal entry sites (IRES) that permit cap-independent translation of molecules such as VEGF even under severely hypoxic conditions (7).
The advent of DNA microarray technology has greatly expanded the ability of investigators to identify novel hypoxia-responsive genes (9, 15, 29) and has provided a tool that allows simultaneous examination of the effects of hypoxia on expression of multiple functionally related genes. For example, a recent study by Fink et al. (9) used both DNA array technology and real-time PCR to identify genes affected by hypoxia in a variety of human hepatocyte cell lines. In addition to identifying several previously unrecognized hypoxia-responsive genes, it was found that hypoxic exposure without reoxygenation led to an overall decrease in the number of transcripts expressed by cells, and interestingly, that hypoxic exposure in these human cell lines (without reoxygenation) did not lead to an increase in expression of heat shock proteins. Since publication of these experiments, DNA arrays have increased substantially in power and complexity, and it seemed likely that application of late-generation arrays would enable identification of even more hypoxia-responsive genes. We therefore decided to examine the effects of hypoxic exposure on mRNA expression in a human cell line, using Affymetrix GeneChip arrays containing 12,600 sequences. A hepatocyte cell line was chosen because the liver is highly metabolically active and thus likely to be very sensitive to hypoxic exposures. To maximize the comparability of these experiments to the published literature, we chose a well-studied human cell line (HepG2 hepatocytes) and a hypoxic exposure (1% oxygen for 24 h) that is both conventional and known to produce somewhere between half-maximal and maximal activation of HIF-1 (14). Furthermore, as a first step toward assessing the extent to which the observed changes are part of a truly nonspecific cell stress response at the level of RNA expression, we have compared our results to published lists of mammalian genes that are known to be affected by heat shock.
![]() |
MATERIALS AND METHODS |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
Test conditions.
Cells were exposed to hypoxia by placing flasks of cultured cells into a custom-built airtight chamber designed to fit within a tissue culture incubator (5). It was superfused with a mixture of analyzed hypoxic gas consisting of 1.01.1% O2 and 5.3% CO2 (balance N2) (Liberty Supply, Leominster, MA) that had first been bubbled through an inline water trap contained within the incubator to provide heat and humidity. The concentration of CO2 used was chosen based on preliminary experiments, which revealed that a concentration of gas slightly greater than 5% was needed to compensate for line losses. The humidified gas fractions of oxygen and carbon dioxide within the chamber were measured by use of an inline oximeter/CO2 analyzer (AlphaOmega model 9500) connected to the outlet port of the chamber. At steady state, the gas concentration of CO2 in the five experiments we report here was 5.2 ± 0.1% (mean ± SD) and the concentration of O2 was 1.0 ± 0.05% (mean ± SD) (5). Cells exposed to control conditions were concurrently placed in a separate room air-5% CO2 incubator. Both sets of cultures were left undisturbed throughout the 24-h exposure period. At the end of the 24-h exposure, cell survival under control and hypoxic conditions was assessed by the trypan blue exclusion method.
For the heat shock control experiments, cells grown under conditions identical to those used for hypoxia were placed in a 43°C water bath (inside a 5% CO2 tissue culture incubator) for 30 min or maintained under control conditions (37°C, also in a 5% CO2 incubator) and then returned to 37°C. RNA was isolated immediately upon return to 37°C (time 0) and at 1, 2, and 4 h after heat shock or control exposure.
RNA purification,
RNA was extracted from cells after 24 h of control or hypoxic exposures with RNeasy kits (Qiagen, Valencia, CA), according to the directions of the manufacturer. At the conclusion of each hypoxic exposure, the culture flasks were removed from the incubator, and lysis of the cells was performed as quickly as possible to minimize the effects of reoxygenation. We estimate that the time required to do this was less than 2 min. The quality of the extracted RNA was assessed as recommended by Farrell (8), as described previously in detail (34). Samples were judged suitable for DNA array analysis only if the RNA was of a suitable yield (at least 2535 µg), exhibited intact bands corresponding to the 18S and 28S subunits, and displayed no spurious peaks on a UV absorption spectrum in the range 230320 nm. Furthermore, samples submitted for DNA array analysis were also analyzed by reverse transcription-PCR (RT-PCR) to ensure that a detectable hypoxic response had occurred at the level of mRNA expression, as judged by an increase in carbonic anhydrase IX (CA IX) under hypoxic conditions (Fig. 1), a gene which has been shown in several other tumor cell lines to be highly upregulated by hypoxia (40). Of the first eight experiments performed for purposes of oligonucleotide array analysis, five met our RNA quality criteria, and all of these showed an increase in CA IX expression as judged by RT-PCR (Fig. 1).
|
Data analysis.
Data analysis was performed using Microsoft Excel, Microsoft Access, and SigmaStat 2.0 for Windows. The fold changes in gene expression reported by the Affymetrix software (MAS 4.0) in the paired experiments were used to determine whether a statistically significant change in expression had occurred, by computing geometric means and 95% confidence intervals as described in detail previously (34).
Where noted in this paper, sequences that showed a statistically significant change in expression were filtered by two post hoc criteria. First, sequences were excluded if they were labeled as "absent" by the GeneChip-reading software in more than half of the control samples (for downregulated genes) or in more than half of the hypoxic samples (for upregulated genes). Second, sequences were excluded if the change in geometric mean expression was less than twofold.
Confirmatory RT-PCR.
Samples submitted for GeneChip array analysis were subjected to a separate poly-T primed RT-PCR using a commercially available kit (Retroscript First-Strand Synthesis Kit; Ambion, Austin, TX), following the manufacturers instructions. Each resulting mixture (consisting of cDNA and unreacted primers) was diluted to 50 ng/µl, and 100-ng aliquots were subjected to 31 cycles of PCR at a Tm of 60°C, using primers designed to recognize CA IX, MAX-interacting protein-1 (MXI-1), dual-specificity phosphatase-1 (DUSP-1), zinc finger protein 36 (ZFP-36), outer mitochondrial membrane translocator 34 (TOM-34), interferon-inducible protein 30 (IFI-30), or cyclophilin A. The ß-actin primers were obtained from a commercial source (Clontech, Palo Alto, CA) and the other primers were designed using PRIMER-3 software. The sequences (5'-3') of these primers were as follows: CA IX tatctgcactcctgccctct (forward primer), gctggcttctcacattctcc (reverse primer), designed to yield a 475-bp amplicon; MXI-1, tctcccatggagaagtggac (forward), agaggatggcatctccaatg (reverse), designed to yield a 450-bp amplicon; DUSP-1, aagaatgctggaggaagggt (forward), ttcagcaaatgtcttgacgc (reverse), designed to yield a 543-bp amplicon; ZFP-36, gtcaccctctgccttctctg (forward), ggtcaggggagtgggttaat (reverse), designed to yield a 454-bp amplicon; TOM-34, acatggttgttgcaccagaa (forward), atggacactgaccaaggagg (reverse), designed to yield a 546-bp amplicon; IFI-30, tgcaaattcaacaaggtgga (forward), ggtaagtagcaggtgccgag (reverse), designed to yield a 483-bp amplicon; and cyclophilin A, aggtcccaaagacagcagaa (forward), tgtccacagtcagcaatggt (reverse), designed to yield a 406-bp amplicon. The samples subjected to heat shock were analyzed using the following published (34) HSP 70B' primers (5'-3'): aggagatctcgtccatggtg (forward), ttccatgaagtggttcacga (reverse), designed to yield a 380-bp amplicon. The sequences of the HSP 70B' primers are complementary to both HSP 70B' (HSPA6) and to HSP 70B (HSPA7), a poorly expressed pseudogene of nearly identical sequence that encodes a truncated (and probably nonfunctional) protein (24).
![]() |
RESULTS |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
The expression of CA IX, a molecule known to be substantially upregulated by hypoxic stress in other tumor cell lines (40), was substantially increased in cells exposed to hypoxic conditions compared with those maintained under control conditions (Fig. 1). By contrast, there was no apparent effect on the expression of ß-actin.
GeneChip array results.
Under control conditions, HepG2 cells expressed a mean (±SE) of 5,493 ± 139 sequences as "present" or "marginal." Under hypoxic conditions, these cells expressed 5,182 ± 130 sequences. This decrease in the number of transcripts expressed showed a trend toward statistical significance (P = 0.12 by paired t-test, n = 5). The trend was also noted when the analysis was limited to only those sequences identified as "present" by the GeneChip-reading software. Under control conditions, 5,247 ± 137 sequences were identified as "present" compared with 4,930 ± 128 under hypoxic conditions (P = 0.11 by paired t-test, n = 5). Importantly, 5,393 sequences were expressed as "present" or "marginal" in at least 3 experiments under control conditions (3,944 in all 5), and 5,051 were expressed in at least 3 experiments under hypoxic conditions (3,677 in all 5).
Of the 12,600 sequences on the GeneChip, 2,908 showed a statistically significant change in expression between the control and hypoxic condition (as defined by geometric mean 95% confidence intervals that excluded unity). Of these, 1,255 showed an increase in expression, and 1,653 showed a decrease in expression. However, only 2,090 of these sequences (905 increased and 1,185 decreased) were expressed (i.e., identified as "present" or "marginal") in at least 3 experiments under control conditions (for the decreased genes) or under hypoxic conditions (for the increased genes). Furthermore, of these 2,090 sequences, only 387 (213 increased and 174 decreased) showed a mean change in expression of twofold or greater. Based on UniGene number assignments, we estimate that these 387 sequences represent about 352 different genes (185 increased, 167 decreased).
Effect of hypoxia on control gene expression.
Hypoxia had little or no effect on the expression of several control ("housekeeping") sequences (Table 1). Because glyceraldehyde-3-phosphate dehydrogenase (GAPDH) expression is known to be increased by hypoxia and is a target for HIF-1 (31), it was not included as a control. We found no significant changes in expression of cyclophilin A, the Alu-Sq family consensus sequence, ribosomal proteins L37a and L41, or the replication-independent histone H3FA3. We did observe a small but statistically significant change in ß-actin expression; however, the magnitude of this change in expression (1.3- to 1.5-fold) was smaller than what is typically detectable by RT-PCR and was not readily apparent in our RT-PCR experiments (Fig. 1)
|
Effect of hypoxia on known hypoxia-sensitive genes.
To verify that our experimental methods appropriately detected changes in expression that would be expected from a conventional hypoxic exposure, we examined the changes in expression that occurred in a list of genes that are known targets for HIF-1, as reported in two recent reviews by Semenza (30, 31) (Table 2). To this list, we added a number of other genes that have specifically been shown in HepG2 cells to be increased by hypoxia (3, 9, 10, 12, 19, 38). We did not apply post hoc filtering to the data in Table 2, as post hoc filtering eliminates genes from consideration based on criteria other than statistical significance. Our results were highly consistent with these previous findings. We identified a probe on the U95Av2 array for 36 of 38 target genes listed, and of these, 29 (81%) showed a statistically significant increase in expression as a result of hypoxic exposure. Only one (haptoglobin) showed a significant decrease in expression. Of the remaining sequences, three (heme oxygenase 1, -1 acid glycoprotein 1, and p21) were expressed in at least three of the five experiments but did not show a statistically significant change in expression as a result of hypoxia. None of the three other sequences (
1B-adrenergic receptor, nitric oxide synthase 2 (iNOS), and VEGF receptor FLT-1) were expressed in a majority of our experiments. Of these, the
1B-adrenergic receptor is known not to be expressed by HepG2 cells (2), and expression of nitric oxide synthase 2 (iNOS) by HepG2 cells is both very low under basal conditions (1) and shows a blunted increase in expression under interferon-stimulated conditions (1, 36).
|
Our results were also highly congruent with the effects of 1% oxygen exposure on HepG2 gene expression as reported on Northern analysis by Wenger et al. (38). We identified sequences on the GeneChip corresponding to 9 of 10 genes they reported to be upregulated by hypoxia, and of these, 7 (78%) showed a statistically significant increase in expression in our experiment (VEGF, erythropoeitin, aldolase A, transferrin, -1 antitrypsin,
-1 antichymotrypsin, and ß-actin, Tables 1 and 2).
-1-Acid glycoprotein, which was marginally elevated in their study, was not significantly affected in ours (Table 2). Haptoglobin, by contrast, was significantly downregulated in our experiment (Table 2). Among genes reported to be unaffected by hypoxia by Wenger et al. (38),
-fibrinogen and hemopexin were also unaffected in our study, and ß-fibrinogen showed only a small though significant decrease in expression (0.78-fold, 95% CI, 0.690.89).
We also compared our results to a list of 20 transcripts that were recently reported to be downregulated by hypoxia in HepG2 cells by real-time PCR methods (9). We were able to identify sequences on the GeneChip arrays corresponding to 17 of these 20 transcripts. Of these, five (HSC 70, HSP 70/90 organizing protein/transformation-sensitive protein-1, FLAP endonuclease 1, CDC25B, and chromatin assembly factor 1 p48 subunit) showed a statistically significant decrease in expression, consistent with previously reported results. Two others (ICAM-1 and ephrin receptor EphA2) showed a statistically significant increase in expression, in contradiction to previously reported results. The remaining 10 showed changes in expression that were not statistically significant (6 decreases and 4 increases).
Other genes affected by hypoxia.
Although 2,908 transcripts showed a statistically significant change in expression as a result of hypoxic exposure, only 387 also met our post hoc filter criteria. These post hoc filter criteria were intended to maximize the likelihood that the cDNA array findings will be reproducible by techniques such as RT-PCR, by requiring the presence of both a strong signal under at least one condition (control vs. hypoxia) in a majority of experiments and a large fold change in expression between the two conditions (at least 2-fold) on average over all five experiments. Of the 387 sequences that met our post hoc filter criteria, 59 were of unknown or unclear function. Of the remaining 328, about belonged to 1 of 9 functional categories: metabolic/biosynthetic enzymes and metabolic regulatory proteins (61 sequences); cell growth, proliferation, and differentiation (52 sequences, including at least 5 transcripts known to encode proteins that affect the myc system); transcription (30 sequences); signal transduction (28 sequences, including at least 4 transcripts known to encode proteins that affect MAP kinase pathways); immune function (19 sequences); growth factors and related proteins (19 sequences); RNA processing and conformation (20 sequences); membrane transport (15 sequences); and cytoskeleton and cell structure related proteins (10 sequences).
Additional sequences that were significantly and strongly affected by hypoxic exposure in our experiment are listed in Tables 3 and 4. Where possible, we have also listed examples of other systems in which the mRNA transcript in question is similarly affected by hypoxia. Many of the genes that were most strongly (4-fold) upregulated by hypoxia in this experiment (Table 3) have been reported to be similarly affected by hypoxia in other systems. By contrast, we found literature precedent for very few of the genes downregulated by hypoxia in this experiment (as noted above and as illustrated in Table 4).
|
|
|
|
|
A comparison of the 2,908 sequences that showed a statistically significant change in expression in this study to 2,903 sequences that were previously found to be significantly affected by heat (at 2 h and 40 min after a conventional heat shock) in normal human peripheral blood mononuclear cells (PBMCs) (34), revealed that only 12% (137 increased and 200 decreased) showed changes in expression that were both significant and in the same direction; another 16% (467 sequences) showed opposite changes in expression (significantly increased in one condition, decreased in the other). Likewise, of 37 sequences that were recently reported to be affected by sublethal heat shock in retinal pigment epithelial (RPE) cells (6), we were able to identify comparable sequences on the U95Av2 array for 36, and of these, only 5 (14%) were changed in the same direction by both hypoxia (at 24 h) and by heat shock (during at least one time point in the 24-h recovery period after thermal injury). Seven others (19%) showed opposite changes in expression.
We also compared our results to a published list of non-HSP genes that are affected by heat shock in mammalian systems (33). This list includes genes that exhibit changes in expression both during acute heat exposure as well as during recovery (i.e., after return to normothermic conditions). Of the 58 non-HSP genes on this list, we were able to identify corresponding probe sequences on the GeneChip array for 50, and of these, 11 (22%) showed changes in direction during hypoxia similar to those reported for heat shock. This list included several well-known stress response genes such as DUSP-1, mcl-1, fos, jun, IB
, VEGF (all increased), and C/EBP-
(decreased). Furthermore, another seven transcripts (14%) showed opposite changes in expression during the two conditions, including ICAM-1 and IL-8 (significantly increased during hypoxia but reported to be decreased by heat shock).
As noted previously, 21 of 26 assessable sequences identified by Semenza (30, 31) as targets of HIF-1 were also found to be upregulated by hypoxia in this experiment. This concordance (81%) was significantly greater than the concordance with non-HSP genes affected by heat shock (22%, P < 0.001 by chi-square analysis). This statistically significant difference persisted even when we included the list of downregulated genes reported by Fink et at. (9) in our analysis (P < 0.001 by chi-square).
![]() |
DISCUSSION |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
One of the greatest risks of an experiment that simultaneously assesses the expression of thousands of different sequences is the possibility of false-positive results. By random chance alone, one might expect 5% of sequences (about 630, in this case) to appear to be affected by any given stimulus. To address this issue, we subjected our findings to both external validation (comparison to lists of genes previously shown to be affected by hypoxia as well as to genes not expected to be affected by hypoxia) and to internal validation (replication of select findings by RT-PCR) and applied post hoc filters that we believe will minimize the number of false-positive reports in Tables 3 and 4. Indeed, the 95% confidence intervals reported in these Tables are generally well removed from onefold. Furthermore, the fact that our unfiltered analysis appropriately detected changes in expression where expected (Table 2) but not generally in housekeeping sequences (Table 1), coupled with the observation that a number of the sequences identified after post hoc filtering (in Table 3) have been found to be affected by hypoxia in other systems, increases our confidence that the novel findings reported here are likely to be true-positive results. This analytic approach has previously yielded highly satisfactory results in normal human cells subjected to heat shock (34).
Two important limiting issues must be kept in mind when interpreting these data. We studied an immortal tumor cell line which, though well characterized and conventionally studied, may in some ways respond differently to hypoxia than a terminally differentiated cell line. Furthermore, our control conditions (room air-5% CO2 incubator), although experimentally conventional, subject cells to a PO2 that is higher than the 1230 mmHg (the equivalent of 24% FiO2) typically reported in the interstitial fluid of normoxic organs (30). It is therefore possible that some of the apparent downregulation of gene expression by hypoxia in fact represents the effects of hyperoxic upregulation of genes in the control cells.
Although differing in some of the particulars, our data confirm and extend the findings of Fink et al. (9) in two very important respects. First, we found that hypoxia has a substantial downregulatory effect on transcript expression in HepG2 cells in addition to its well-known upregulatory effects. Prior to post hoc filtering, the number of transcripts that showed a statistically significant decrease in expression was greater than the number of transcripts that showed a statistically significant increase in expression. Furthermore, there was a statistical trend (P = 0.12) toward a decrease in the total number of transcripts expressed by the cells exposed to hypoxic conditions. In principle, such a downregulatory effect could be produced by decreasing expression uniformly across the entire population of cells or by causing a marked decrease in gene expression in a subset of cells within the population, as would occur if some of the cells were induced to undergo apoptosis. Although we were unable to distinguish between these two possibilities using trypan blue exclusion, it is noteworthy that Bae et al. (3) have found evidence that exposure of HepG2 cells to 1% hypoxia for 24 h induces apoptotic changes in about 3040% of cells in culture. It is therefore possible that some of the downregulatory effect of hypoxia that we and others have noted in this cell line is due to induction of apoptosis in a subset of the cells in culture.
The second important area of concordance between our results and those of Fink et al. (9) was the lack of a generalized induction of HSPs by exposure to 1% oxygen for 24 h. On the contrary, almost half of the HSPs and chaperonins examined that were expressed under control conditions showed a decrease in expression under hypoxic conditions. We would not have missed a generalized increase in HSP expression had one been present, as HepG2 cells are capable of producing a vigorous increase in HSP 70B' expression in response to a conventional heat shock (Fig. 3), and we have previously detected changes in HSP expression in PBMCs using an almost identical (U95A) Affymetrix array (34). However, it is also possible that the hypoxic stimulus, although conventional, was insufficiently severe to induce a heat shock response in this cell line or that the time point we chose to examine was inadequate to detect a heat shock response. In this regard, it is noteworthy that Patel et al. (25) have reported induction of HSP-70 RNA in HepG2 cells under conditions of anoxia (0% O2), with expression peaking at 6 h.
A comparison of our results to the heat shock literature supports the hypothesis that the human cellular response to stress involves both changes in expression that are stress specific (and perhaps even cell type specific) as well as changes that are truly nonspecific to the cell type and stressor applied. The existence of stress-specific changes in expression is not surprising, as the most important transcription factors activated by hypoxia and heat [HIF-1 (30, 31) and HSF-1 (20, 26), respectively] are distinct from each other. However, our findings also suggest that the truly nonspecific component of the human cell stress response is possibly quite small, amounting to somewhere between 10 and 15% of the total number of genes affected by a given environmental stressor. We believe that identifying and characterizing these nonspecific genes is important, as doing so may provide novel insights into how cells acquire cross-tolerance to multiple environmental stressors. Conversely, genes that are similarly affected by multiple different stressors are potentially poor targets for therapeutic manipulations that are aimed at adaptations to a particular stress [such as cytotoxic therapy directed at hypoxically adapted cells, as proposed by Wouters et al. (39)]. Experiments that are specifically designed to catalog and distinguish stress-specific and stress-nonspecific gene expression responses may be of considerable interest and potentially of practical value, particularly under conditions of compensable stress such as those studied in this experiment.
In conclusion, we have found that the response of human HepG2 cells to hypoxia at the level of mRNA expression is extensive and includes a significant component of downregulation. Interestingly, hypoxia did not induce a generalized heat shock response in HepG2 cells compared with control conditions, and the degree of overlap between the responses of HepG2 cells to 24 h of exposure to 1% oxygen and the known responses of mammalian cells to heat shock appears to be small. Our findings are consistent with the concept that the response of human cells to environmental stress includes both stress-specific and nonspecific components. In human cells, this nonspecific component may in fact be much smaller than generally recognized.
![]() |
ACKNOWLEDGMENTS |
---|
This work was supported in part by National Institutes of Health Grant RO1-HL/AI-64104.
Editor S. Gullans served as the review editor for this manuscript submitted by Editor R. E. Pratt.
![]() |
FOOTNOTES |
---|
Address for reprint requests and other correspondence: L. A. Sonna, Thermal and Mountain Medicine Division, United States Army Research Institute of Environmental Medicine, 42 Kansas St., Natick, MA 01760 (E-mail: larry.sonna{at}na.amedd.army.mil).
10.1152/physiolgenomics.00104.2002.
![]() |
References |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|