cGMP-independent nitric oxide signaling and regulation of the cell cycle

Background Regulatory functions of nitric oxide (NO•) that bypass the second messenger cGMP are incompletely understood. Here, cGMP-independent effects of NO• on gene expression were globally examined in U937 cells, a human monoblastoid line that constitutively lacks soluble guanylate cyclase. Differentiated U937 cells (>80% in G0/G1) were exposed to S-nitrosoglutathione, a NO• donor, or glutathione alone (control) for 6 h without or with dibutyryl-cAMP (Bt2cAMP), and then harvested to extract total RNA for microarray analysis. Bt2cAMP was used to block signaling attributable to NO•-induced decreases in cAMP. Results NO• regulated 110 transcripts that annotated disproportionately to the cell cycle and cell proliferation (47/110, 43%) and more frequently than expected contained AU-rich, post-transcriptional regulatory elements (ARE). Bt2cAMP regulated 106 genes; cell cycle gene enrichment did not reach significance. Like NO•, Bt2cAMP was associated with ARE-containing transcripts. A comparison of NO• and Bt2cAMP effects showed that NO• regulation of cell cycle genes was independent of its ability to interfere with cAMP signaling. Cell cycle genes induced by NO• annotated to G1/S (7/8) and included E2F1 and p21/Waf1/Cip1; 6 of these 7 were E2F target genes involved in G1/S transition. Repressed genes were G2/M associated (24/27); 8 of 27 were known targets of p21. E2F1 mRNA and protein were increased by NO•, as was E2F1 binding to E2F promoter elements. NO• activated p38 MAPK, stabilizing p21 mRNA (an ARE-containing transcript) and increasing p21 protein; this increased protein binding to CDE/CHR promoter sites of p21 target genes, repressing key G2/M phase genes, and increasing the proportion of cells in G2/M. Conclusion NO• coordinates a highly integrated program of cell cycle arrest that regulates a large number of genes, but does not require signaling through cGMP. In humans, antiproliferative effects of NO• may rely substantially on cGMP-independent mechanisms. Stress kinase signaling and alterations in mRNA stability appear to be major pathways by which NO• regulates the transcriptome.


Background
Nitric oxide (NO • ) plays a pivotal role in vascular biology through both cGMP-dependent and -independent mechanisms. In health, NO • regulates vascular tone by activating soluble guanylate cyclase [1][2][3]. However, other important effects of NO • in the vasculature such as cytoprotection and anti-adhesion appear to occur independent of cGMP signaling [4][5][6]. Likewise, NO • regulation of inflammation has frequently been associated with signal transduction events that do not involve cGMP [7,8]. NO • induces TNFα in human cells by decreasing intracellular levels of cAMP, thereby removing cAMP-mediated repression of the TNFα promoter through a proximal Sp element [9,10]. Analogs of cAMP and Sp site mutation both block, while antagonists of cAMP-dependent protein kinase simulate the effect of NO • on TNFα. [9,11]. In contrast to TNFα, NO • induces interleukin-8 (IL-8) [12] through a distinct post-transcriptional mechanism that is both cGMP-and cAMP-independent. IL-8 mRNA is stabilized by NO • activation of p38 MAPK, increasing its halflife and translation [13]. These and other reports [14][15][16]. suggest that cGMP-independent gene regulation by NO • occurs through multiple pathways.
Similar to the regulation of blood pressure and inflammatory responses, NO • regulation of cell proliferation is of central importance to circulatory health. Failure of this regulatory pathway has been linked to atherosclerosis and other forms of vascular dysfunction [17][18][19]. Despite extensive investigation, the relative contribution of cGMP-independent NO • signaling in the regulation of cell cycle genes remains controversial. In rats, NO • has been shown to activate transcription through cGMP-dependent Distribution of NO • -regulated genes Figure 1 Distribution of NO • -regulated genes. Of 110 differentially regulated genes, 71 were up-regulated (red) and 39 (green) were down-regulated. Genes were classified into functional categories using NIH-DAVID [83] and PubMed [52]. Data are from seven independent microarray experiments. effects on AP-1 promoter sites [20]. Also in rodents, a NO • -cGMP-PKA-ERK1/2 signal transduction pathway has been described that inhibits cell proliferation [21,22] and increases expression of p21/Waf1/Cip1 [23,24]. A master regulatory gene, p21 directly inhibits Cdk complexes [25,26] and represses the transcription of many cell cycle genes through CDE/CHR (cell cycle dependent element/ cell cycle gene homology region) promoter elements [27,28]. In contrast to rodents, NO • regulation of cell cycle genes in humans, including regulation of p21, appears to occur, at least in part, independent of cGMP [19,29]. However, a global examination of cGMP-independent NO • effects on the transcriptome in general or on cell cycle genes specifically has not been undertaken in either rodents or humans.
Here, oligonucleotide microarrays and human U937 cells that lack soluble guanylate cyclase [9,30] were used to globally characterize the cGMP-independent effects of NO • on gene expression. Differentiation with PMA was  * * * * employed to render cells capable of cytokine production [9]. This treatment also forced >80% of cells into the G0/ G1 phase of the cell cycle, which facilitated the analysis of cell cycle gene regulation. Since NO • lowers cAMP levels in U937 cells [9] and cAMP is known to affect cell proliferation, NO • effects were also tested in the absence and presence of a cell permeable cAMP analog. For genes affected by NO • -induced decreases in cAMP, cAMP analog would be expected to produce an apposite effect. Hypotheses generated from microarray results were further investigated by examining downstream changes in protein expression and signal transduction pathways.

Functional distribution of NO • -regulated genes and hypothesis generation
Of 110 NO • -responsive genes, 71 were induced, and 39 were repressed; the majority were not previously known to be NO • -responsive. Both naïve and differentiated U937 cells lack NO • -sensitive soluble guanylate cyclase [9,30], and therefore gene regulation by NO • in these cells can be attributed to cGMP-independent mechanisms. Genes were annotated into functional categories ( Fig. 1 [13]. Therefore, the ARE database http:// rc.kfshrc.edu.sa/ared was used to identify ARE-containing genes among those regulated by NO•. Twenty-two of 110 genes contained ARE (20%) compared to 540 ARE genes of 5086 on the microarray (11%; P = 0.008). An additional 11 ARE-containing genes were identified in PubMed for a total of 33 ( Table 2). Nearly half of these genes (14/33; 42%) have been reported to be p38 MAPK regulated (Table 2). Importantly, for these 14 genes, p38 MAPK activation produces responses that are in the same direction as those observed here for NO•. The broad influence of NO• on cell cycle-related genes and ARE-containing transcripts independent of cGMP was unexpected, as was the strong association of these effects with p38 MAPK. Therefore, further experiments were performed to confirm these results and to define underlying regulatory mechanisms that might link NO• effects on the cell cycle with post-transcriptional gene regulation through ARE sites. Real-time RT-PCR (TaqMan ® ) was used to validate NO •mediated changes in mRNA levels ( Fig. 2A and 2B). Of 18 selected genes, 13 were NO • up-regulated, and 5 were down-regulated. Fold changes from microarray experiments strongly correlated with results from RT-PCR (R = 0.95, P < 0.0001).

Validation of NO • -regulated genes
Western blotting of key cell cycle genes regulated by NO • was performed to test whether microarray results accurately predicted changes in protein expression ( Fig. 2C and 2D). Three induced genes, E2F transcription factor 1 (E2F1), p21/Waf1/Cip1 (Cdk inhibitor; CDKN1A), and cell division cycle 6 (CDC6) were examined. E2F1 and p21 are well-characterized master regulatory proteins that control the cell cycle. Four repressed genes, cyclin A1 (CCNA1), cyclin B1 (CCNB1), polo-like kinase (PLK) and cyclin F (CCNF) were also measured by Western blotting. In all cases, directional changes in protein expression were consistent with the differential effect of NO • on corresponding transcripts as determined by microarray analysis.  To further compare the effects of NO • and cAMP, a hierarchical cluster analysis was performed using the 35 cell cycle genes regulated by NO • (Fig. 3). For each of the 5 cell cycle genes significantly affected by both NO • and Bt 2 cAMP [c-Myb, B-cell translocation gene 1 (BTG1), dual specificity phosphatase 4 (DUSP4), growth factor independent 1 (GFI1), and cyclin A1 (CCNA1)], the direction of regulation was the same. Further, for cell cycle genes regulated by NO • , cAMP analog either had no effect on or produced expression changes that were similar to and additive with those observed for NO • (Fig. 3). These results suggest that NO • effects on cell cycle genes are independent of its interference with cAMP signaling, since cAMP analog (the opposite signal) was not antagonistic to the actions of NO • .

Analysis of NO • effects on the cell cycle
NO • causes arrest in either the G1 or G2/M phase of the cell cycle [19,[35][36][37]. However, the mechanisms underlying this effect are not well characterized. Annotation of NO • -regulated genes to their respective phase of the cell cycle revealed that expression changes were not random (Table 1). Most NO • up-regulated genes (7/8) were G1/S associated, while down-regulated genes were strikingly G2 and G2/M phase associated (24/27). The latter included topoisomerase II alpha (TOP2A), cyclin B1, PLK, and CDC25B, genes that are necessary factors for mitosis. These results show that NO • suppresses the cell cycle in early G2/M by triggering a highly integrated program of gene regulation that does not require soluble guanylate cyclase or cGMP.
To further test this hypothesis, cell cycle analysis was performed on U937 cells using flow cytometry. PMA-differentiation significantly increased the portion of cells in G0/ G1 (P < 0.0001), while decreasing cells in S (P = 0.0007) and G2/M (P = 0.003) compared to a naïve, undifferentiated cell population (   6C). NO • stabilized p21 mRNA in the absence of SB202190 ( Fig. 6D;P = 0.004), but had no effect in the presence of SB202190 (P = 0.5).

NO • regulation of the cell cycle through E2F1 and p21
The E2F family of transcription factors and p21 act as master regulatory switches that control the cell cycle. E2F1 regulates target genes through E2F-binding sites and thereby plays an essential role in DNA synthesis and the G1/S Next, electrophoretic mobility shift assays (EMSA) were performed to test whether NO • altered protein binding to E2F and CDE/CHR consensus sequences. PMA-differentiated U937 cells were treated with PBS, GSH, or GSNO followed by preparation of nuclear extract. NO • increased binding to both E2F ( Fig. 7A and 7B) and CDE/CHR probes ( Fig. 7C and 7D). Site-directed mutagenesis of each consensus sequence abolished competition (Fig. 7) and E2F1-directed antibody blocked complex formation with labeled E2F probe ( Fig. 7A and 7B).

Summary
NO • , independent of cGMP, regulated a diverse subset of genes involved in inflammation, metabolism, apoptosis, the cell cycle, proliferation, signal transduction, and transport. Notably, genes associated with the cell cycle and pro-Cell cycle analysis of U937 cells liferation, including the master cell cycle regulatory genes E2F1 and p21, were over-represented. Further, NO • -regulated transcripts had ARE (post-transcriptional regulatory sites) in their 3' UTR and were p38 MAPK responsive more frequently than expected. E2F1 induction by NO • was associated with up-regulation of several genes involved in G1/S transition that contain E2F-binding sites. NO • also induced p21, an ARE-containing gene, through p38 MAPK activation and mRNA stabilization. This was associated with the down-regulation of G2/M phase genes, at least in part, through changes in protein binding to CDE/ CHR promoter sites. Collectively, these results demonstrate that NO • , independent of cGMP and cAMP, triggers a specific and highly coordinated genetic program that alters the G1/S transition and induces arrest in early G2/ M (Fig. 8). MAPK pathways and mRNA stability are major mechanisms by which NO • regulates the transcriptome.

Discussion
NO • has potent anti-tumor and anti-atherosclerotic effects that are closely associated with its ability to block cell proliferation [18,53]. This activity of NO • has been ascribed to both cGMP-dependent and -independent mechanisms. Experiments in rodents have found, with a few notable exceptions [54,55], that NO • controls the cell cycle through cGMP. These studies have focused on the importance of a NO • -cGMP-PKA-ERK 1/2 signal transduction pathway [22][23][24]. Accordingly, cAMP itself has also been reported to inhibit cell proliferation through activation of PKA and/or ERK 1/2 with the up-regulation of p27 or p21 in a cell-specific manner [31][32][33]56]. In contrast, the antiproliferation effects of NO • in human cells have been frequently associated with cGMP-independent signaling [19,29]. Here, a transcriptome-wide approach revealed that NO • exerts broad control over the cell cycle through p38 MAPK activation and mRNA stabilization.
In a previous study, we found that NO • up-regulates TNFα by decreasing cAMP, an effect antagonized by cAMP analogs. Therefore Bt 2 cAMP was used in this investigation to explore whether some effects of NO • on the transcriptome could be attributed to its interaction with cAMP signaling. However, our results indicate that NO • -cAMP signaling appears to be a minor pathway, regulating less than 6 of the affected transcripts in U937 cells (data not shown). These included TNFα, as well as pim-1 oncogene (PIM1), TNFα-induced protein 2 (TNFAIP2), and glutathione reductase (GSR). Importantly, for cell cycle genes, NO • and Bt 2 cAMP consistently had the same directional effect on transcripts, although NO • regulated more genes overall. Thus, decreases in intracellular cAMP appear unrelated to NO • effects on the cell cycle. Furthermore, treatment with both NO • and Bt 2 cAMP synergistically provoked cell cycle arrest in G2/M, suggesting that NO • -induced decreases in cAMP may attenuate some of its effects on the cell cycle. Although this experiment also provides useful intormation on gene regulation by cAMP in U937 cells, the reader should keep in mind that Bt 2 cAMP was the only analog studied and some effects may have been caused by its butyryl component.
U937 cells were PMA-differentiated in the current experiments to render them capable of producing TNFα and IL-8, two cytokines previously identified as NO • -responsive [7][8][9][10][11][12][13]. Further, this treatment also reduced cell proliferation and forced >80% of the cells into the G0/G1 phase of the cell cycle, allowing for a more coherent analysis of cell cycle regulation (Fig. 4). However, PMA itself had large effects on NO • -regulated genes such as p21 (Fig. 6D) and the findings here cannot be extrapolated directly to naïve U937 cells. Fortunately, Turpev and colleagues have recently reported selected microarray results from NO • exposure of undifferentiated U937 and Mono Mac 6 cells [57]. Of interest, a number of key genes identified by this group were also found to be similarly regulated by NO • in PMA-differentiated cells including HMOX1, IL-8, activating transcription factor 4 (ATF4), BCL2/Adenovirus E1B 19 kD-interacting protein 3 (BNIP3), and importantly p21/Waf1/Cip1. were incubated with increasing concentrations of the p38 inhibitor SB202190 (0 nM to 25 nM) for 30 min, then exposed to PBS, glutathione (GSH; 400 μM) or GSNO (400 μM) for 12 h. Western blotting was performed to detect p21 expression. Results were quantified with laser densitometry. Data, presented as fold change relative to PBS control values, are means ± SE of three independent experiments. Next, TaqMan ® RT-PCR was used to quantify p21 mRNA levels normalized to GAPDH mRNA. (C) Changes in p21 mRNA levels during differentiation of U937 cells (1 × 10 7 ) with PMA. Data, presented as fold change relative to mean mRNA level in naïve cells, are means ± SE of three independent experiments. (D) NO • stabilization of p21 mRNA is dependent on p38 MAPK. U937 cells (1 × 10 7 ) were differentiated with PMA for 8 h. After 30 min pretreatment with actinomycin D (2.5 μg/ml) without and with SB202190 (0.1 μM), cells were further incubated with GSH (400 μM) or GSNO (400 μM) for 2 to 4 h. At the specific time points, cells were harvested for total RNA extraction. Data, presented relative to mRNA level at 0 h (arbitrarily set to 100%), are means ± SE of three independent experiments.  Others have found that NO • increases p21 mRNA and protein expression in human vascular smooth muscle independent of cGMP [58]. In addition, p38 MAPK activation has been shown to increase p21 expression by both transcriptional activation and protein stabilization [42]. E2F1 is also known to induce p21 transcription [59] providing another mechanism by which NO • may have increased p21 expression in the current experiment. Conversely, as already discussed, NO • decreases cAMP, reducing the ability of Sp1 to bind to GC box elements and thereby repressing the transcription of Sp1-dependent genes such as eNOS [11]. Interestingly, p21 is highly dependent on Sp1 for transcription [60] and is induced by cAMP [56], findings consistent with the possibility that p21 may be transcriptionally repressed by NO • -cAMP-Sp1 signal transduction. Nonetheless, NO • induction of p21 demonstrates that other mechanisms dominate over any negative effects of NO • on Sp1 binding to the p21 promoter. Here, we focused on mRNA stabilization, because     [66]. Further, cyclin A1 was down-regulated by NO • and has been shown to turn off E2F1 target genes by decreasing E2F1 DNA binding [47,67].

A D e g r a d e d G S N O D e g r a d e d N O N O a t e D e g r a d e d S N A P S N A P N O N O a t e G S N
Notably, c-Myb and a number of G2 or G2/M phase genes that contain E2F sites were down-regulated by NO • . E2F sites can function as repressors in some genes and their disruption by mutation leads to promoter activation [46,61,68]. Further, NO • -responsive genes with both E2F elements and CDE/CHR repressor sites were uniformly down-regulated. Promoters with CDE/CHR motifs are repressed by p21 [28,39], which was also induced by NO • . Therefore, even for promoters activated by E2F1, repression through CDE/CHR sites appears to be the dominant action of NO • in this cellular context. Moreover, E2F and CHR sites may cooperate as co-repressors [69]. Although CDC25B lacks an identifiable CDE/CHR site, it does have a proximal repressor and its regulation is similar to CDE/ CHR-containing genes [70]. indicating that NO • might alter the stability of these transcripts. Notably, cyclin A1 and cyclin B1 are down-regulated by p38 MAPK, a signal transduction pathway that was activated by NO • in the current experiments. Importantly, p38 MAPK has been shown to induce cell cycle arrest at the G2 checkpoint through mechanisms that were only partially dependent on p21 [74].
ARE in 3' UTR have been implicated in the control of transcript stability and have an important post-transcriptional impact on transcriptome content [11,[75][76][77]. We previously demonstrated that independent of cGMP, NO • upregulates IL-8, but not TNFα post-transcriptionally through p38 MAPK activation [13]. In the current investigation, ARE-containing genes including IL-8 and p21 were over-represented among NO • -regulated genes. Nearly half of these ARE genes have been reported to be regulated by p38 MAPK (Table 2). Notably, NO • responses were all in the same direction as those reported for p38 MAPK activation. Previous microarray experiments that globally tested mRNA stability found that 10% of transcripts were associated with p38 MAPK-dependent regulation [77]. The over-representation of p38 MAPKregulated genes in our experiments indicates that this stress kinase is an important target of NO • .

Conclusion
The present investigation was focused on understanding cGMP-independent gene regulation by NO • . Major themes within the identified gene list were the predominance of cell cycle-related genes and ARE-containing transcripts. NO • was found to trigger a specific and coordinated cell cycle arrest independent of both cGMP and cAMP. E2F1 induction up-regulated target genes involved in G1/S transition through E2F sites. NO • stabilization of p21 mRNA was p38 MAPK dependent and led to increased protein binding to CDE/CHR promoter sites and the down-regulation of G2/M phase genes. The cell cycle is a major target of NO • -mediated gene regulation. Importantly, p38 MAPK and mRNA stability are major intermediary mechanisms through which NO • affects the human transcriptome.
Microarray experiments NO • donor, GSNO (400 μM), or its precursor GSH (400 μM) was added into differentiated U937 cells in the absence or presence of Bt 2 cAMP (100 μM) followed by incubation at 37°C for 6 h (N = 7). Cells were then washed three times with ice cold PBS. Total RNA was extracted using RNeasy Mini kits (Qiagen, Valencia, CA) and reverse transcribed (10 μg) using the SuperScript II ® custom kit (Invitrogen, Carlsbad, CA). Resulting cDNA (1 μg) was in vitro transcribed into biotin-labeled cRNA using the BioArray high yield RNA transcript labeling kit (Enzo Life Sciences, Farmingdale, NY). After fragmentation, biotin-labeled cRNA (20 μg) was hybridized to Affymetrix HuGeneFL 6800 ® microarrays [> 5,000 unique transcripts after masking uninformative probe sets [Affymetrix Website, #106] following the Affymetrix protocol [78]. After staining with streptavidin phycoerythrin (Molecular Probes) and enhancing with anti-streptavidin (0.5 mg/ml, Vector Laboratories, Burlingame, CA), microarrays were scanned using Agilent GeneArray Scanner. The TaqMan ® Real time RT-PCR system (Applied Biosystems, ABI, Rockville, MD) was employed to quantify mRNA levels. Gene-specific TaqMan ® probes and PCR primers were designed using Primer Express 1.0 (ABI) and their sequences are provided in supplemental data [see Additional file 5]. The High-capacity cDNA Archive kit (ABI, Foster City, CA) was employed to prepare cDNA from 2 μg of total RNA. The resulting cDNA was used for RT-PCR in triplicate according to the standard ABI protocol. The average quantities of target gene mRNA relative to GAPDH mRNA was determined for each sample. The target gene/GAPDH ratio in GSH treated cells was arbitrarily set at 1 and results from all other samples were expressed relative to that standard.

Cell cycle analysis
Cells were harvested and stained with propidium iodide and the cell cycle distribution of stained cells was determined by flow cytometry (FACS Calibur, Becton Dickinson). The percentage of cells in G0/G1, S, and G2/M was determined using ModFit (Verify Software House Inc., Topsham, ME) and expressed as relative change compared to PMA-differentiation alone. Naïve U937 cells were compared to cells (1 × 10 6 ) incubated with PMA for 48 h to examine the effects of differentiation on the cell cycle. As expected, PMA differentiation pushed cells into G0/G1 arrest (>80% of cells). These cells were then treated with GSH (400 μM) or GSNO (400 μM) without or with Bt 2 cAMP for 24 h and processed for cell cycle analysis as described above.

Microarray data analysis, gene annotation, and statistics
Images were analyzed using Microarray Suite 4.0 (Affymetrix). Global scaling was set at 100. Data were transformed and analyzed using the MSCL Analyst's Toolbox http:// affylims.cit.nih.gov written in the JMP scripting language (SAS Institute, Cary, NC). Average difference (AD) values were standardized and transformed using the Symmetric Adaptive Transform [79][80][81] yielding quantile-normalized, homogenous variance scaled results. Differentially regulated genes were identified from 7 independent experiments using a combination of consistency tests set at a 4% false discovery rate (FDR) and an average AD above 20 for at least one condition. One of 7 experiments was an outlier for some genes, but was not allowed to eliminate genes found significant in the other six. Fold change in gene expression was calculated directly from AD results after raising negative values to 10, and likewise adding 10 to all positive values.
Genes were annotated by searching NIH-DAVID [82,83] and PubMed [52]. Over-representation of gene categories among differentially expressed transcripts was tested using Expression Analysis Systematic Explorer [84,85]. EASE scores (penalized Fisher exact test), corrected for multiple comparisons using bootstrap resampling with 10,000 iterations, are reported as P-values. These analyses and tests of significance relied on databases within EASE and therefore did not include additional genes that were annotated to particular functional categories using PubMed.
All data not derived from microarrays are presented as mean ± standard error (SE) of at least three independent experiments. All P-values are two-sided unless noted otherwise, and considered significant if less than 0.05. To compare treatment effects on cytokine secretion, a twoway ANOVA with blocking for experiment was carried out on the logarithm of the measured concentrations for TNFα, IL-8 and IL-1β (supplemental Fig. 1A). A linear regression of RT-PCR log fold change versus microarray log fold change was generated to evaluate the validity of the microarray data (Fig. 2B). To determine whether NO • affected the protein expression of various cell cycle genes, paired t-tests, unadjusted for multiple comparisons, were performed for GSH versus GSNO, after log normalization to PBS (Fig. 2D). Log percentages of naïve and PMA-differentiated U937 cells in each phase of the cell cycle were compared by paired t-tests, unadjusted for multiple comparisons (Fig. 4A). Two-way ANOVAs with blocking were performed on log percentage of cells in each phase of the cell cycle to assess the significance of the NO • effect, cAMP effect, and their interaction (Fig. 4B). Effects of NO • on p38 MAPK phosphorylation (Fig. 5) were investigated with a one-way ANOVA comparing the log fold change of laser densitometry intensity (pp38/p38) over different concentrations of GSNO. A post-hoc Dunnett's test was carried out to determine the lowest concentration at which the effect became significant compared to control. The expression of p21 in the presence of GSNO, SNAP, or DETA-NONOate was compared to that in the presence of their respective degraded controls with paired t-tests, unadjusted for multiple comparisons (Fig. 6A). The dose effect of SB202190 on NO • -induced p21 protein expression normalized to PBS was analyzed using a one-way ANOVA (Fig. 6B). NO • stabilization of p21 mRNA over time (with and without SB202190) was assessed using constrained one-way analysis of covariance, after natural log transformation of relative mRNA amounts (Fig. 6D).