Response of steady-state photosynthesis to elevated temperature
Maximum Asat was 27 and 14% greater for soybean relative to Arabidopsis and poplar, respectively (Fig. 1a). The thermal optimum of Asat was determined by fitting a modified four-parameter Gaussian function to the response curves. The temperature optimum for Asat was greatest for soybean (34.3 °C, SE = 0.32), followed by poplar (32.6 °C, SE = 0.57) and Arabidopsis (30.9 °C, SE = 0.24). A linear regression was fitted to the data from optimum to 42 °C to determine the rate of Amax decline. Poplar had the greatest decrease in slope (−1.34, SE = 0.042, r2 = 0.95) followed by Arabidopsis (−1.22, SE = 0.172, r2 = 0.95) and soybean (−0.95, SE = 0.62, r2 = 0.91).
Figure 1. Temperature-dependent species-specifc optima of: (a) light-saturated rate of photosynthesis (Asat). Arrows point to calculated optimum for each speices; (b) photosystem II (PSII) electron transport rate (ETR); (c) the ratio of internal CO2 (Ci) to ambient CO2 (Ca); and (d) the dark respiration of Arabidopsis thaliana, Populus and Glycine max. Values are means of two separate growth chamber experiments (±SE), n = 2.
Download figure to PowerPoint
The estimated proportion of electrons passing through PSII (ETR) as measured by chlorophyll a fluorescence showed a similar, yet slightly more robust response to temperature than that observed for Asat. The optimum for ETR was again greatest for soybean (37.2 °C, SE = 0.43), followed by poplar (34.8 °C, SE = 0.67) and Arabidopsis (33.2 °C, SE = 0.89; Fig. 1b). The ratio of intercellular to ambient CO2 (Ci/Ca) was plotted against temperature and showed little evidence for stomatal limitation (Fig. 1c). Dark respiration showed the typical rise in respiration as leaf temperature increased (Fig. 1d). Increasing and decreasing leaf temperature from 22 to 42 °C indicated no hysteresis in the response of gas exchange or chlorophyll fluorescence parameters (data not shown).
Comparison of inferred metabolic pathway alterations underlying heat shock
The PageMan (Usadel et al. 2006) and MapMan (Thimm et al. 2004; Usadel et al. 2009b) bioinformatics software packages were used to infer metabolic pathways, cellular processes and hormonal regulation underlying distinct phases of photosynthetic temperature response from transcriptome profiles. Under- and over-represented functional groups are determined based on Fisher's exact test and Wilcoxon rank summary test statistics, and displayed using false colours (Abarca et al. 2001; Usadel et al. 2005, 2006, 2009b). As temperature increased from baseline to photosynthetic optimum, Arabidopsis drastically suppressed transcription of genes participating in photosynthesis, particularly those contributing to PSII (Fig. 3). This is in contrast to soybean and poplar, both of which displayed a clear induction of PSII to optimum. Furthermore, this trend continued through inhibition 20% for poplar. Induction of the abiotic stress pathway was apparent for Arabidopsis and soybean from baseline to inhibition 20%. Poplar exhibited a cyclical trend in the abiotic stress pathway with induction from baseline to optimum followed by pathway suppression at inhibition 20% and induction again at inhibition 30%. Similar trends for protein synthesis pathways were observed for all species with induction from baseline to optimum, followed by repression to inhibition 20% and induction once again to inhibition 30%. Major carbohydrate synthesis pathways were relatively unaltered for Arabidopsis according to this analysis (although one synthesis category was suppressed; Fig. 3), while soybean carbohydrate synthesis pathways were suppressed from optimum to inhibition 30%. Poplar exhibited an unexpected induction in carbohydrate synthesis from optimum to inhibition 20%, followed by repression at inhibition 30%. In soybean, the minor carbohydrate raffinose pathway was induced from baseline to optimum (Fig. 3), and this was mirrored with clear induction of galactinol synthase 1 (GolS1) and raffinose synthase homologs (Supporting Information Fig. S1). Poplar exhibited a similar trend with induction of GolS1 and 2 homologs and raffinose synthase homologs from baseline to optimum. In Arabidopsis, the raffinose/galactinol pathway was induced from optimum to inhibition 20% with elevated expression of GolS1 and 2.
Figure 3. PageMan display of selected gene categories for stress-related, and secondary and primary metabolism pathways. An unpaired Wilcoxon rank sum test was used to determine if the median fold change within a particular ontological group is the same as the median fold change of all genes not in that group. Multiple testing was corrected with Bengermani Hochberg. Resultant P values were transformed to z values with P = 0.05 set to 0. False colours are used to distinguish among over-(yellow) and under-(blue) represented categories.
Download figure to PowerPoint
Co-expression network modules at distinct states of photosynthetic decline
The above analysis is based on organizing genes into functional bins representative of cellular pathways determined from model organisms such as Arabidopsis. Extending these analytical approaches to other species requires genes to be categorized into bins based primarily on sequence homology. Although useful, such an approach may not accurately characterize novel species-specific pathways and could falsely categorize genes that are not yet well assembled and propagate annotation errors. The recent genome duplication of poplar and soybean relative to Arabidopsis, and the potential effects of redundancy on understanding cellular pathways from inferred sequence homology alone illustrate the need to complement this approach with additional analyses.
To address this, we constructed a weighted gene co-expression network for the most differentially expressed (most significant F-test among all three contrasts) 4000 genes along the distinct physiological states of the response curves (see Fig. 2a for sampling schematic). These networks are composed of modules that contain genes with high topological overlap, which is representative of the relationship similarity between the expression of two genes relative to all other genes within the network. Thus, modules contain genes sharing highly correlated expression patterns and are often involved in the same biological function (Barabasi & Oltvai 2004; Subramanian et al. 2005). This has recently been termed the ‘guilt by association’ paradigm, allowing one to infer biological function of genes based on network neighbourhood (Usadel et al. 2009a). R-scripts and input network data for this analysis are available as Supporting Information Tables S4–S7.
The weighted gene co-expression network algorithms generated five modules for Arabidopsis and eight modules for soybean that grouped into two large meta-modules (correlated gene expression patterns) for each species (Fig. 4a,c). The poplar transcriptome generated 10 smaller yet distinct modules (Fig. 4e). Entire network results are reported in Supporting Information Tables S10–S12. The eigenvalue for each module was correlated to treatment to identify patterns of module expression, and thus potential underlying pathway expression to physiological states.
Figure 4. Weighted gene co-expression network construction and module correlation to physiological state. (a) (Arabidopsis), (c) (soybean) and (e) (poplar) are multi-dimensional scaling plots of the gene co-expression network. Each circle represents a single gene and the colour of the circle corresponds to module designation. The distance between circles is a function of topological overlap and provides a visual representation of gene and module relationships within network. (b) (Arabidopsis), (d) (soybean) and (f) (poplar) report the positive (+) and negative (−) Spearman correlation of the module eigenvalue (y-axis) with the physiological state (x-axis). Symbols indicate ratio was significantly different than zero at P < 0.001 (***), P < 0.01 (**), P < 0.05 (*).
Download figure to PowerPoint
We expected that modules negatively correlated to baseline and optimum, and positively correlated to photosynthetic inhibition states would contain genes underlying the heat shock response (Fig. 4b,d,f). Significant trends were observed for the Arabidopsis black (permutation t-test, P = 0.027) and blue (permutation t-test, P = 0.0005) modules during 30% inhibition of photosynthesis. Gene characterization analysis using the PageMan over-representation software (Usadel et al. 2006) identified the Arabidopsis black module to be most significantly enriched with genes involved in carbohydrate metabolism (P = 5.8 × 10−3) and lipid metabolism (P = 8.8 × 10−3). The Arabidopsis blue module was most significantly enriched with genes involved in amino acid metabolism (P = 1.5 × 10−20) and protein degradation (P = 2.9 × 10−11).
Contrary to our expectations, the Arabidopsis module eigenvalue most significantly correlated with optimum photosynthesis (brown module; Fig. 4b, permutation t-test, P = 0.0007) was significantly enriched with genes responsive to abiotic stress and heat (P = 1.5 × 10−19). This module contained 15 sHSPs including seven HSP17, two HSP18, four HSP20s, one HSP60s, seven HSP70s, three HSP80s, one HSP90, one HSP101, two HSFs and 12 DNAJs (see Table 1; Supporting Information Table S10). In addition, the eigenvalue of the Arabidopsis brown module significantly correlated with the expression of the known heat shock responsive marker gene, for example, AtHSP18.1-Cl (At5g59720; Pearson cor = 0.932, P = 0.0234) as previously identified (Takahashi, Naito & Komeda 1992; Prandl et al. 1998; Swindell 2006).
The soybean network greenyellow (permutation t-test, P = 0.002) and black (permutation t-test, P = 0.005) modules have significant eigenvalue correlations to 30% inhibition of photosynthesis (Fig. 4d). According to the over-representation analysis, both modules were enriched with genes participating in abiotic stress and heat in particular (greenyellow, P = 5.6 × 10−48; black, P = 4.3 × 10−11). The greenyellow module contains numerous heat shock members including 11 HSP17 family members, three HSP18s, seven HSP20s, three HSP70s and two HSFs (Table 1; Supporting Information Table S11), thereby identifying a clear function for this module in the heat shock response.
The poplar network has a significant correlation between the brown module eigenvalue and 30% inhibition of photosynthesis (permutation t-test, P = 0.024) and was significantly enriched with genes responsive to abiotic stress and heat (P = 1.2 × 10−8). As seen in Table 1 and Supporting Information Table S12, the poplar brown module contains numerous genes with strong homology to Arabidopsis HSPs, including two DNAJ domain containing proteins, three HSP17s, two HSP18s, two HSP70 and one HSP101. In addition, the turquoise module had a significant correlation to 30% inhibition and was enriched with genes participating in abiotic stress (P = 5.0 × 10−5).
Putative expansion of soybean HSP17s
The large number of soybean HSP17s within the heat shock module relative to Arabidopsis and poplar motivated us to perform a more inclusive comparative phylogenetic analysis of this gene family (Fig. 5). It appears that the HSP17s in soybean are greatly expanded relative to Arabidopsis and poplar. We identified 25 genes exhibiting both high identity to Arabidopsis HSP17s and molecular weights in the specified range to include in our analysis (Supporting Information Table S8). This expansion appears to be related to extensive segmental as well as tandem duplication occurring in the soybean genome (http://www.phytozome.net/soybean).
All of the Arabidopsis HSP17s were components of the heat shock network; only four of the putative poplar HSP17s (following Waters et al. 2008) were in the network, although all were represented on the NimbleGen arrays used for expression analysis. Despite our identification of 72 JGI soybean gene models containing the HSP20 protein domain, only 34 of these had corresponding probes on the Affymetrix array, and only 22 sHSP genes were represented in the network. However, the soybean chip is based on EST data available prior to the release of reference genome sequence, and the assembly and annotation of the reference soybean genome are ongoing. All of the soybean HSP17 genes that had corresponding probes on the Affymetrix microarray were evident in the network modules for heat stress.
Specialization of ROS production and scavenging
Within the identified heat shock modules for each of the respective species, there were several genes annotated for ROS scavenging. In the Arabidopsis brown module, for example, ascorbate peroxidase 2 (APX2; At3g09640), APX1 (At1g07890), copper/zinc SOD 3 (At5g18100) and glutathione binding/transferases (At1g78380, At2g29420, At2g47730) were present (Supporting Information Table S10). Genes encoding antioxidant enzymes were also featured in the poplar brown and soybean black heat shock modules (Supporting Information Tables S11 & S12), suggesting potential co-regulation of heat shock with the antioxidant system.
To investigate this notion further, total H2O2 was measured for all species along the temperature response curve. In Arabidopsis and poplar, H2O2 levels increased with increasing temperature after a slight decline from baseline to optimum (Supporting Information Fig. S3). Maximum H2O2 values were observed at 30% photosynthetic inhibition for both species. In soybean, maximum H2O2 accumulation spiked at optimum photosynthesis (40% over the baseline), and then declined to 10 and ∼20% over baseline at 20 and 30% photosynthetic inhibition, respectively.
The total antioxidant capacity, including enzymatic and non-enzymatic processes, was determined by estimating the rate of oxidation of a fluorescent probe (fluorescein) using the ORAC assay as described by Gillespie et al. (2007). In all species, ORAC increased with an increase in temperature from baseline to optimum (Fig. 6a). Arabidopsis showed the greatest increase in ORAC over baseline (103%) relative to soybean and poplar at 30% inhibition. Unlike poplar and Arabidopsis, there was a slight decrease in ORAC from 20 to 30% inhibition for soybean, suggesting that the antioxidant system was not entirely compromised.
Figure 6. Effect of heat on reactive oxygen species (ROS) scavenging in leaves of Arabidopsis, soybean and poplar. (a) Oxygen radical absorbance capacity (ORAC), an indicator of antioxidant status of the samples, was measured as Trolox equivalents and expressed as the % of the baseline sample per gram of fresh weight. (b) Total soluble peroxidase activity per milligram of protein measured from the leaf extracts. (c) Total glutathione reductase (GR) activity per milligram of protein measured from the leaf extracts. (d) Total Cu–Zn superoxide dismutase (SOD) activity per milligram of protein measured from the leaf extract. All the values are expressed as percent of the baseline activity derived from means of two separate growth chamber experiments (±SE), n = 2.
Download figure to PowerPoint
The enzymatic contribution to the plant antioxidant system was determined by measuring the activities of three key ROS-sequestering enzymes: PEROX, GR and SOD. The GR activities for all three species were greatest at 30% inhibition, although trends throughout the curves did vary (Fig. 6b). Poplar and Arabidopsis gradually increased the GR activity, while soybean GR activity displayed a steep increase at optimum followed by a sharp decline at 20% inhibition. The soybean PEROX and GR activities were slightly different, as the highest GR activity was observed at 30% photosynthetic inhibition rather than optimum (Fig. 6b,c). In Arabidopsis and poplar, total PEROX activity decreased from baseline to optimum by ∼12.2 and 8.9%, respectively, while soybean displayed its greatest measured peroxidase activity at optimum (60% relative to baseline; Fig. 6c). Peroxidase activity in soybean declined to near baseline levels and remained relatively unaltered throughout the curve. Arabidopsis and poplar peroxidase activity continued to rise throughout the temperature response curve. Like GR, the SOD activity increased with the increase in temperature in both Arabidopsis and poplar. Alternatively, soybean SOD activity remained relatively unaltered throughout the temperature response curves (Fig. 6d).