Fillable and unfillable gaps in plant transcriptome under field and controlled environments

Abstract The differences between plants grown in field and in controlled environments have long been recognized. However, few studies have addressed the underlying molecular mechanisms. To evaluate plant responses to fluctuating environments using laboratory equipment, we developed SmartGC, a high‐performance growth chamber that reproduces the fluctuating irradiance, temperature and humidity of field environments. We analysed massive transcriptome data of rice plants grown under field and SmartGC conditions to clarify the differences in plant responses to field and controlled environments. Rice transcriptome dynamics in SmartGC mimicked those in the field, particularly during the morning and evening but those in conventional growth chamber conditions did not. Further analysis revealed that fluctuation of irradiance affects transcriptome dynamics in the morning and evening, while fluctuation of temperature affects transcriptome dynamics only in the morning. We found upregulation of genes related to biotic and abiotic stress, and their expression was affected by environmental factors that cannot be mimicked by SmartGC. Our results reveal fillable and unfillable gaps in the transcriptomes of rice grown in field and controlled environments and can accelerate the understanding of plant responses to field environments for both basic biology and agricultural applications.


| INTRODUCTION
To optimize agricultural crop productivity and understand plant behaviour in natural environments, knowledge of plant responses to fluctuating field environments is essential. Numerous studies conducted in controlled environments, such as growth chambers and greenhouses, have facilitated the understanding of plant responses to environmental stimuli. However, such responses are sometimes different from those in controlled environments (Annunziata et al., 2017(Annunziata et al., , 2018Dantas et al., 2021;Matsubara, 2018;Matsuzaki et al., 2015;Nagano et al., 2012;Poorter et al., 2016;Song et al., 2018) due to differences between the two environments. Field environments experience daily fluctuations and gradual changes, particularly around dawn and dusk, whereas controlled environments usually fluctuate quickly and regularly between fixed (i.e., square-wave) conditions, which are constant during the day and night, and abruptly transition at dawn and dusk. Light quality, such as red light to far-red light ratio and the presence of ultraviolet-B light, also varies between field and controlled environments. Additionally, plants in the field experience abiotic and biotic stresses, such as wind, precipitation, and insect and pathogen attacks.
Such factors make it difficult to apply knowledge obtained from laboratory studies to the field studies in plant science.
To reveal plant responses to fluctuating field environments, field and laboratory studies have attempted to address the differences between the two settings. One approach involves transcriptome analysis of fieldgrown plants (Dantas et al., 2021;Iwayama et al., 2017;Kashima et al., 2021;Matsuzaki et al., 2015;Nagano et al., 2012Nagano et al., , 2019Takehisa & Sato, 2019;Zaidem et al., 2019). We previously developed a statistical model that predicts the transcriptome dynamics of rice leaves in the field using meteorological data (Nagano et al., 2012). The modelling approach provides valuable information about plant responses to the field environment, although the detailed mechanism still requires examination under laboratory conditions. Another approach is to mimic the field environment using laboratory equipment. Studies using this approach have clarified the characteristics of photosynthesis under fluctuating light (Alter et al., 2012;Kaiser et al., 2018;Matsubara, 2018;Niedermaier et al., 2020;Schneider et al., 2019;Tanaka et al., 2019;Vialet-Chabrand et al., 2017;Yamori, 2016), successfully mimicked primary metabolism of Arabidopsis leaves (Annunziata et al., 2017(Annunziata et al., , 2018, and determined the expression patterns of the Arabidopsis florigen gene, FLOWERING LOCUS T (FT) (Song et al., 2018) in field environments.
Previous studies have clarified the characteristics of plant responses to fluctuating field environments. However, a comprehensive understanding of the differences between plants grown in field and in controlled environments is still lacking. Therefore, we developed SmartGC, a high-performance growth chamber that can reproduce fluctuating field environments, to compare plant responses to field and controlled environments. By analysing massive transcriptome data of rice plants grown under field and SmartGC conditions, we revealed fillable and unfillable gaps in plant responses to field and controlled environments.

| Plant materials and growth conditions
In this study, we developed SmartGC, a high-performance growth chamber (Supporting Information: Figure S1a-d). SmartGC is composed of two parts: a growth chamber (LPH-240SP, Nippon Medical & Chemical Instruments Co., Ltd.) (for controlling temperature and relative humidity) and a Heliospectra L4A LED light source (Heliospectra) (for controlling light). Both parts have been customized to be controlled simultaneously by one computer, and they are scheduled to function for more than 24 h. The light source can independently control seven types of LEDs (violet to far-red) with a 1-s resolution, but it does not include UV-A and UV-B. The spectrum of the light source is shown in Supporting Information: Figure S2a,b. The output value of each LED can be set to 0 or 1 in increments from 15 to 1000 sv (set value). Temperature and relative humidity were set to 15-45°C and 50%-80%, respectively, at a 1-min resolution. SmartGC can record temperature and relative humidity every minute. Although the light source can control seven types of LEDs independently, we set the output of all LEDs to the same value for each setting.
A common japonica rice (Oryza sativa L.) cultivar, Nipponbare, was used in all the experiments in this study. Seeds were sterilized in a 2.5% (v/v) sodium hypochlorite solution for 30 min and then soaked in water at 30°C for 3 days. Germinated seeds were sown in a cell tray filled with nursery soil (N:P 2 O 5 :K 2 O = 0.6:1.2:1.0 g/kg). Plants were grown in Temperature and relative humidity were measured using THMchip thermo-hygrometers (THM10-TH, FUJIFILM Wako Pure Chemical Corporation) which were set in an aspirated radiation shield (Okada & Nakamura, 2010). Irradiance was measured using a quantum metre  Figure S2a-c, respectively. The R:FR ratio ranged from 1.1 to 1.6. The DLI was 30, 13 and 34 mol photons m −2 day −1 on the first, second and third days, respectively.
2. Fluctuating light, temperature and humidity (FL/FTH). Fluctuations in irradiance, temperature and relative humidity in the FIELD condition were simulated. Irradiance in the FIELD condition was simulated every minute by translating irradiance to the light source output using a calibration curve (Supporting Information: Figure S3). We measured the PFD of the output from 15, 100,200,300,400,500,600,700,800,900,950 and 1000 sv for each LED and constructed a calibration curve using linear regression.
Although plants can sense light below 1 µmol photon m −2 s −1 , particularly through phytochrome A, and low light can affect gene expression (Seaton et al., 2018;Shinomura et al., 1996), we regarded irradiance in the FIELD with PFD < 1 as darkness and set RICE TRANSCRIPTOME IN FIELD AND CHAMBER | 2411 F I G U R E 1 (See caption on next page) the output value of the light source to zero. Since the lowest output value for turning on the light source was 15 sv, the output value during the day was set to 15 sv, as the irradiance in the FIELD was lower than that in SmartGC, with an output value of 15 sv (Figure 1b, Supporting Information: Figure S4a,b). The photoperiod was 12 h and 37 min, 12 h and 31 min and 12 h and 31 min on the first, second and third days, respectively. In addition, the highest output value for the light source was 1000 sv, and the output value during the day was set to 1000 sv when the irradiance in the FIELD was higher than that at an output value of 1000 sv in SmartGC (Figure 1b) Table S1).

| RNA-Seq analysis
The leaf samples were ground under cryogenic conditions using a   Figure S7). The library was sequenced using HiSeq. 2500 (Illumina) at Macrogen or Takara with single-end sequencing lengths of 50bp or 100 bp, respectively.
All obtained reads were trimmed using Trimmomatic version 0.33 (Bolger et al., 2014)  The reads per million (rpm) were calculated using the nuclearencoded gene raw count data, excluding the genes encoding rRNA, as described by Kashima et al. (2021). In Experiments 1 and 2, 0.85-3.35 million and 1.11-4.27 million reads per sample were used for calculating rpm, respectively (Supporting Information: Figure S7a).
A total of 12,741 genes in which the average number of reads was >10 in all Experiment_1 samples was used for the statistical analysis (Supporting Information: Figure S7b).

| Inference of internal time using the molecular timetable method
We applied the molecular timetable method (Ueda et al., 2004) to the transcriptome data of Experiment_1 to infer the internal time of each sample, as described by Higashi et al. (2016). First, we selected timeindicating genes whose expression indicated periodicity and high amplitude. To evaluate the periodicity, we prepared 1440 cosine curves, which had different peaks (0-24 h) measured at 1-minute increments. We fitted the curves to the time-course transcriptome data of FIELD in Experiment_1 (52 total samples) and calculated the correlation coefficient (r) to identify the best-fitting cosine curve (Supporting Information: Figure S8a). The peak time of the bestfitting curve was estimated as the peak time for each gene and was defined as the molecular peak time. Thus, the molecular peak time  Table S2). The molecular peak time of the time-indicating genes was covered throughout the day, which ensured the accurate estimation of internal time (Supporting Information: Figure S8c).
We normalized the expression level of each time-indicating gene using the z-score, which is defined as the value of the individual expression level minus the average expression level, divided by the standard deviation. We then plotted expression profiles composed of the molecular peak time and the normalized expression level for each sampling time (Supporting Information: Figure S8d). Finally, the internal time was estimated using a plotted expression profile. We prepared 1440 cosine curves (with each having 1-min difference with respect to preceding one) and fitted them to the expression profiles.
We identified the best-fitting cosine curve, and the corresponding peak time was used to indicate the estimated internal time.
To validate the accuracy of inferring the internal time using the time-indicating genes, we calculated the measurement noise as the standard deviation of the difference between the real and estimated expression of each time-indicating gene. The measurement noise of each gene ranged from 91% to 100% (mean ± standard deviation: 99 ± 1%), indicating that 143 time-indicating genes were sufficient for accurately estimating the internal time (Ueda et al., 2004).

| Determination of starch and sucrose content
Starch and sucrose content were determined as described by Okamura et al. (2013).

| Analysis of public microarray data
We used the microarray data previously analysed by Nagano et al. (2012). This data was available on the GEO website (https://www.ncbi. nlm.nih.gov/geo/; accession numbers: GSE36777 and GSE36595) and Reads were then trimmed using Trimmomatic version 0.33 (Bolger et al., 2014) with the parameters described above and were then mapped to the rice reference genome using Bowtie2 with default parameters, except setting N = 1. After extracting the unmapped reads and removing the duplicated reads, de novo assembly was conducted using Trinity with default parameters. After removing redundant reads using CD-HIT (Fu et al., 2012), 19 contigs were identified. Each contig was annotated using BLASTn for nucleotides (Camacho et al., 2009) (Supporting Information: Figure S9, Supporting Information: Table S3).

| Statistical analysis
All statistical analyses were performed using R software version 3.5.3 (R core Team, 2019). Specifically, differentially expressed gene (DEG) analysis was conducted using R package TCC version 1.20.0 (Sun et al., 2013;Tang et al., 2015). Normalization was conducted using iDEGES/edgeR (Robinson et al., 2010) with a false discovery rate (FDR) of 0.1, and DEG detection was conducted using edgeR with FDR = 0.05. Gene enrichment tests for GO and Kyoto Encyclopaedia of Genes and Genomes (KEGG) (Kanehisa & Goto, 2000) pathways were conducted using the R package GO.db version 3.6.0 (Carlson, 2018) and KEGG.db version 3.2.3 (Carlson, 2016), respectively, as described by Nagano et al. (2019). The FDR was controlled using Benjamini and Hochberg's method (Benjamini & Hochberg, 1995) with FDR = 0.05. Log 2 (rpm) was calculated as log 2 (rpm + 1). Multiple comparison tests for starch and sucrose contents were conducted using the R package car version 3.0.10.

| Reproduction of environmental field conditions with SmartGC
SmartGC can control irradiance with a 1-second resolution and temperature and relative humidity with a 1-minute resolution (Supporting Information: Figure    These results suggest that the internal time progression of the samples was faster around 9:00 in CL/CTH and slower around 19:00_2 in CL/ CTH and CL/FTH compared with the other conditions, reflecting the differences in irradiance. This was also supported by the internal time inference of transcriptome samples using the molecular timetable method (Higashi et al., 2016;Ueda et al., 2004) (Supporting Information: Figure S8, Supporting Information: Table S2). These results are consistent with that of a previous study on statistical modelling with transcriptome data (Matsuzaki et al., 2015), which showed that the internal time progression in conventional growth chamber conditions was faster after lights-on and slower before lights-off than in the field.
Experiment_2 samples were also separated by time, excluding those in the morning, using PCA and hierarchical clustering (Figure 1h,j and Supporting Information: Figures S11b and S12). In the Experi- irradiance of the sampling time (Figure 1h and Supporting Information: Figure S12). Samples at 7:00 under CL/CTH were clustered with the 8:00 FL/CTH and CL/FTH samples and the 9:00 FL/FTH samples ( Figure 1j and Supporting Information: Figure S12). This time lag between conditions continued until 11:00, suggesting that the morning internal time progression was affected by temperature, humidity and irradiance.
We then evaluated transcriptome similarity between conditions using DEG analysis at each sampling time-point. We compared FIELD with the other conditions using DEG analysis in Experiment_1 (Supporting Information: Table S4). DEG analysis indicates the DEGs between FIELD and the other conditions. There tended to be fewer DEGs in FL/FTH than in other conditions (Figure 2a). This is consistent with the results that the rice transcriptome dynamics in FIELD were better reproduced under FL/FTH than in the other conditions (Figure 1d-f,i). The number of DEGs peaked at 7:00 in CL/CTH and FL/CTH, and at 19:00_2 in CL/CTH and CL/FTH (Figure 2a). Since temperature and humidity were equal in CL/CTH and FL/CTH, these results suggest that the difference between FIELD and CL/CTH in the morning was mainly due to temperature and/or humidity. In contrast, irradiance was equal for CL/CTH and CL/FTH, suggesting that the differences between FIELD and CL/CTH in the evening were mainly due to irradiance. Unlike at 19:00_2, no clear differences between FIELD and CL/CTH were observed at 19:00 ( Figure 2a). This may reflect the weather differences before sampling: the second day was cloudy, while the third day was sunny (Figure 1b,c).
We  Table S5). The overlap of DEGs in Experi-ment_2 showed that the number of LIGHT genes was high in the morning and evening, while that of TH genes was high only in the evening (Figure 2d). These results confirm the findings from PCA and hierarchical clustering (Figure 1g-j). Interestingly, >50% of the LIGHT genes at 8:00 and 19:00_2 overlapped (Figure 2h), suggesting that the effect of irradiance on transcriptome dynamics was different between the morning and evening. The number of TH genes was higher than that of LTH and LIGHT genes from 5:00 to 6:00 ( Figure 2d), indicating that temperature and humidity began affecting the transcriptome before dawn. In contrast, the number of LTH and LIGHT genes increased from 6:00 to 6:30, indicating that light began affecting the transcriptome 0.5-1 h after dawn. Since the start of dawn only differed by 10 min between FL/FTH and CL/CTH (Supporting Information: Figure S4), the gradual versus sudden increase of irradiance, and not the difference in the timing of dawn, caused the upregulation of LTH and LIGHT genes after dawn. This is consistent with previous studies on Arabidopsis, and it may have been caused by the regulation of gene expression by phytochrome A after dawn (Seaton et al., 2018). Likewise, the number of TH genes increased from 1 h before dusk (16:00-17:00) (Figure 2d Although we could not distinguish the effects of temperature and humidity on the transcriptome, the effect of temperature could be greater than that of humidity (Nagano et al., 2012). Therefore, we considered the effect of temperature or humidity as the effect of temperature in subsequent discussions.

| Diurnal fluctuation of irradiance affects rice leaf sugar metabolism
To characterize the environmentally-affected genes, we tested for enrichment of genes with annotations in the DEGs detected above harvesting, the expression of nine genes was affected by night temperature (Supporting Information: Figure S17, Supporting Information: Table S14). Information: Figure S16c,d).
These are examples of genes whose evening expression is affected by fluctuations in irradiance. In Arabidopsis, differences in irradiance between sinusoidal and square-wave conditions affect diurnal changes in carbohydrate content (Annunziata et al., 2017(Annunziata et al., , 2018 ,c and 5a,b, Supporting Information: Table S15). In CL/CTH and CL/FTH, starch and sucrose content decreased from dusk to dawn and then increased from dawn to dusk. For sucrose content, a delayed increase at dawn and an early decrease before dusk were observed in FIELD, FL/FTH and FL/CTH, which is consistent with previous results on Arabidopsis (Annunziata et al., 2017(Annunziata et al., , 2018. Delayed increase in Arabidopsis leaf starch at dawn was also observed under sinusoidal conditions with irradiance, but this trend was less prominent in rice. This reflects leaf carbohydrate composition; rice mainly stores sucrose, whereas Arabidopsis mainly stores starch (Okamura et al., 2017). Although Accordingly, with the differences in leaf carbohydrate content, sugar metabolism genes clearly differed between the conditions. Expression of AGPL3 and AGPS1, which encode adenosine diphosphate-glucose pyrophosphorylase (AGP), a key enzyme in starch synthesis (Okamura et al., 2017), tended to be higher in FIELD than in the other conditions (which were significantly enriched in DEGs in FIELD compared to other conditions), and LTH and UNREP genes (Figure 4c,d).
Phenylpropanoid biosynthesis-related gene expression (Figure 6b) was significantly upregulated in FIELD compared to the other conditions, mainly from 9:00 to 15:00, except for a gene encoding 4-coumarate:coenzyme A ligase (Os02g0177600), which was upregulated at night (Figure 6c). Since phenylpropanoid biosynthesis-related gene expression is induced by various biotic and abiotic stresses (Dixon & Paiva, 1995;Vogt, 2010), these results suggest that the upregulation of these genes was a response to field environment stresses.
To investigate field-specific gene expression, we focused on the transcriptome differences between FIELD and FL/FTH. We calculated the mean value of each gene's expression at all time points and extracted genes whose expression was 2× higher or lower in FIELD than in FL/FTH. We also extracted genes whose expression significantly differed between FIELD and FL/FTH at one or more time points. A total of 159 and 78 genes were identified as upregulated and downregulated, respectively, in FIELD ( Figure 6d, Supporting Information: Tables S16 and S17). Phenylpropanoid biosynthesis-related genes were observed among the highly expressed genes in FIELD (Supporting Information: Table S16). We also found several genes encoding pathogenesis-related (PR) proteins (Figure 6e, Supporting Information:  (Backer et al., 2019).

NONEXPRESSOR OF PATHOGENESIS-RELATED GENES 1 (NPR1) is
important for establishing SAR and indirectly activating PR gene expression (Backer et al., 2019). However, NPR1 and the other NPR genes were not upregulated in FIELD (Supporting Information: Figure S21), suggesting that PR gene upregulation was independent of NPR genes. Terpene synthesis-related genes, which defend against herbivore-and pathogen-caused tissue damage (Yoshitomi et al., 2016), were also upregulated in FIELD (Supporting Information: Figure S22, Supporting Information: Table S17). However, we did not observe any signs of herbivory or herbivorous insects. In addition, herbivoryinduced early defence signalling genes (Ye et al., 2019) were neither upregulated nor downregulated, except for WRKY30 downregulation in FIELD (Supporting Information: Figure S23). Therefore, PR and terpene synthesis-related gene upregulation was likely independent of the effect of insects.
Although no pathogen infection symptoms were observed, upregulation of PR and terpene synthesis-related genes may have resulted from pathogen infection. Therefore, we attempted to detect viral and fungal infections from RNA-Seq data using our previously reported pipeline (Kamitani et al., 2016) (Supporting Information: Figure S24) and de novo assembly of unmapped reads to the rice reference genome (Supporting Information: Figure S9).
The number of reads of viruses (Moriyama et al., 1995) and fungi with poly(A) tails was not specific for FIELD nor correlated with PR gene expression (Figure 6f and Supporting Information: Figures S24 and S25, Supporting Information: Table S3). Although we cannot exclude the possibility of infection by bacteria or viruses without poly(A) tails, these results suggest that upregulation of PR and terpene synthase genes in FIELD was a response to physical environmental factor(s) specific to the field. To determine whether field-specific gene expression is also present in paddy-field rice, we re-analysed the microarray data of rice leaves sampled from a paddy field and a growth chamber, which had been previously analysed by Nagano et al. (2012). Observations regarding up-or downregulation of genes in FIELD were consistent with the previous study (Nagano et al., 2012) (Figure 6d, Supporting Information: Tables S18 and S19), suggesting that the field-specific gene expression information we obtained is also applicable to paddyfield rice.

| DISCUSSION
Although differences between plants grown in the field and controlled environments are well known (Poorter et al., 2016), few studies have explored the underlying molecular mechanisms for these. Here, we established an experimental scheme for using laboratory equipment to evaluate plant responses to fluctuating environments. We revealed diurnal transcriptome dynamics in both environments and their fillable and unfillable gaps. Our results complement those that model plant transcriptome responses in field environments (Matsuzaki et al., 2015;Nagano et al., 2012). Gradual changes in irradiance affected transcriptome dynamics in the morning and evening, whereas temperature changes only had an effect in the morning (Figure 2). Accordingly, our statistical model suggested that the number of genes whose expression was affected by a time-specific temperature was the lowest from noon to dusk (Nagano et al., 2012). The number of genes whose expression was affected by time-specific daily irradiance was higher during daytime and the highest around noon. There was no difference in the number of genes expressed in the morning and evening. As the plant circadian clock is dawn-dominant (Edwards et al., 2010;Flis et al., 2016Flis et al., , 2019, the fact that both morning irradiance and temperature affect transcriptome dynamics is likely due to circadian entrainment by irradiance and temperature. Conversely, this suggests that the effect of evening irradiance is independent of circadian regulation. Accordingly, less than a half of the morning and evening LIGHT genes overlapped (Figure 3f). Since sugar metabolism gene expression corresponded to decreased sucrose content in the evening ( Figure 5 and Supporting Information: Figures S16 and S17) and diurnal changes in sugar status affect transcriptome dynamics independent of the circadian clock (Cookson et al., 2016;Flis et al., 2016), the effect of evening irradiance potentially depends on differences in carbon status between conditions. We cannot exclude the possibility that morning light and temperature cause additional effects independent of the circadian clock.
We found that the expression of RVE genes responded to gradual changes in irradiance and temperature (Figure 3b and Supporting Information: Figure S15a). Since TOC1/PRR1 and LUX/PCL1 are positively regulated by RVE in Arabidopsis (Grey et al., 2017) and the expression of two RVE genes (Os06g0728700 and Os02g0680700), TOC1/PRR1, and LUX/PCL1 were affected by a gradual change in irradiance (Figure 3a), the two RVE genes might play a role in the regulation of TOC1/PRR1 and LUX/PCL1 in rice. The expression pattern suggests that Os06g0728700 might serve as an activator of TOC1/PRR1. This is consistent with a previous study suggesting that the role of RVE genes as TOC1/PRR1 activators is conserved in rice (Toda et al., 2019). In contrast, Os02g0680700 might function as a repressor of LUX/PCL1 rather than an activator (Figure 3b). This would suggest that the positive regulation of LUX/PCL1 by RVE is not conserved in rice. A previous study suggested that RVE responded to low carbon status in Arabidopsis (Moraes et al., 2019). Thus, the expression of the two RVE genes might be affected by differences in carbon status derived from either gradual or sudden changes in irradiance. The expression patterns of the other two RVE genes (Os04g0538900 and Os02g0685200) were affected by temperature ( Figure 3b and Supporting Information: Figure S15a). However, the expression of core circadian clock genes including TOC1/PRR1 did not show clear difference between different temperature conditions ( Figure 3a and Supporting Information: Figure S13). As some RVE genes may be involved in the circadian clock output pathway of Arabidopsis (Rawat et al., 2009;Zhang et al., 2007), the impact of the temperature-affected RVE genes on circadian regulation might be different from that of the RVE genes affected by irradiance. Further studies are needed to clarify the role of RVE in circadian oscillator regulation through responses of rice to fluctuating environmental stimuli.
Field plants experience various biotic and abiotic stresses, such as insect and pathogen attack, wind and UV light, which were not simulated by SmartGC. In FIELD conditions, we found upregulated genes related to ribosomes, phenylpropanoid biosynthesis and F I G U R E 6 Field-specific expression of genes related to biotic and abiotic stress. (a) Expression of ribosome-related genes is upregulated in FIELD. Boxplot showing the normalized expression levels (z-score) of genes with annotations for ribosomes (GO:0005840) between FIELD and the other conditions at 13:00 in Experiment 1. Adjusted p-values of Wilcoxon rank-sum test between FIELD and the other conditions are shown. (b) Outline of the phenylpropanoid biosynthesis pathway. Enzymes catalysing each reaction are shown in red. CHI, chalcone isomerase; CHS, chalcone synthase; C4H, cinnamate 4-hydroxylase; F3H, flavanone 3-hydroxylase; F3'H, flavonoid 3-hydroxylase; 4CL, 4-coumarate:coenzyme A ligase; PAL, L-phenylalanine ammonia-lyase. (c) Expression of genes related to the phenylpropanoid biosynthesis pathway. Points indicate means, and error bars indicate standard deviations (n = 4). (d) Scatter plot showing the differences in the mean expression value between FIELD and FL/FTH in this study and between paddy field and the growth chamber in Nagano et al. (2012). Genes whose mean value of expression was more than 2.0 times higher or lower in both experiments and significantly different between FIELD and FL/FTH at one or more time points are shown as red and blue points, respectively. (e) PR genes whose expression was upregulated in FIELD. Points indicate means, and error bars indicate standard deviations (n = 4). (f) Scatter plot showing the relationship between the expression of Alphaendornavirus (Moriyama et al., 1995) and PR genes, which was upregulated in FIELD. [Color figure can be viewed at wileyonlinelibrary.com] RICE TRANSCRIPTOME IN FIELD AND CHAMBER | 2423 pathogen defence (Figure 6), all of which are responses to biotic and abiotic stress (Ali et al., 2018;Dixon & Paiva, 1995;Moin et al., 2016;Vogt, 2010). This indicates that plants cope with the field environment by upregulating stress-responsive genes, despite the less stressful field environment compared to the stress-treatment experiments. Alongside the finding that these genes are also upregulated in paddy fields (Figure 6f), our study suggests that these genes can be targets for rice productivity improvement in the field.
In addition to the presence or absence of UV light, factors related to light quality that were not simulated by SmartGC (Supporting Information: Figure S2), such as the R:FR ratio and the proportion of blue light, may have caused the differences between the plants grown in the field and with SmartGC. In this study, the R:FR ratio in SmartGC was higher and more fluctuating than that in FIELD (Supporting Information: Figure S2). This may have affected phytochrome signalling and phytochrome-dependent processes such as the circadian clock and leaf development (Soy et al., 2016). In Arabidopsis, a higher R:FR ratio between field and controlled environments has been shown to cause differences in the expression patterns of FT genes, and these differences involved phytochrome A (Song et al., 2018). Inactivation of phytochrome B by far-red light at the end of the day affects leaf development (Romanowski et al., 2021) and phytochrome B also acts as a temperature sensor (Jung et al., 2016;Legris et al., 2016). Therefore, a high R:FR ratio could affect transcriptome dynamics in a wide range of biological processes. Furthermore, the proportion of blue light also affects transcriptome dynamics (Pedmale et al., 2016). Since the proportion of blue light affects phenylpropanoid synthesis (Huché-Thélier et al., 2016), it is possible that higher expression levels of genes annotated for phenylpropanoid biosynthesis result from differences in the proportion of blue light, as well as other field-specific environmental factors ( Figure 6). It is necessary to further explore the effects of the R:FR ratio and the proportion of blue light on transcriptome dynamics in the future.
The differences between sunlight in the field and LED light in controlled conditions caused differences in the transcriptome and sugar metabolism between the treatment groups. The starch and sucrose contents in the leaves were higher in FIELD than in the other conditions ( Figure 5a,b). One reason for this is the inability of SmartGC to simulate the high irradiance that occurs during daytime (Figure 1b,c). It is also possible that the light source affected the irradiance received by the whole plant. Since the light source of SmartGC was located above the rice plants (Supporting Information: Figure S1), the irradiance received on the side of the rice plants was much less than that received at the top of the plants. In contrast, rice plants grown in the field received irradiance on the sides as well as the top because sunlight includes both diffused and direct light. Furthermore, the irradiance from LED light received by plants in SmartGC decreases with increasing distance from the light source, while that from sunlight does not (Niinemets & Keenan, 2012;Poorter et al., 2012). Furthermore, diffuse light reaches the lower part of the plant canopy more efficiently than does direct light (Li et al., 2014). Collectively, these factors mean that the total irradiance received by the plants in SmartGC is less than that received in the field.
Considering that the carbohydrate content in CL/CTH and CL/FTH was lower than that in FIELD (Figure 5a,b), the differences in irradiance received by the whole plant likely had a greater effect on carbohydrates than the differences in irradiance during daytime. This might also explain the differences between FIELD and FL/FTH at midday. The number of UNREP genes peaked at 13:00 in Experiment_1 (Figure 2c). Because the photoreceptors were completely saturated by the irradiance at this time of the day, this result might not be related to light signalling. However, the difference in irradiance received by the whole plant between FIELD and FL/FTH might explain the differences observed in the transcriptome, as these may have occurred due to differences in carbon metabolism. As GO terms and KEGG pathways were not significantly enriched in UNREP genes at 13:00 (Figure 4b,d), further studies are required to clarify the mechanisms underlying the differences between FIELD and FL/FTH at midday.
Although the starch and sucrose contents were higher in FIELD than in the other conditions, their trends in FIELD were simulated in FL/FTH (Figure 5a,b). These trends were consistent with previous results on Arabidopsis (Annunziata et al., 2017) We demonstrated the utility of SmartGC for understanding plant responses to fluctuating environments; however, SmartGC cannot completely reproduce field environments, and unfillable gaps remain between plants grown in the field and those grown in SmartGC. To overcome these gaps, it may be practical to control the light quality by increasing the output of far-red light to decrease the R:FR ratio in SmartGC, and by also including UV-A and UV-B light. Furthermore, it would be beneficial to clarify simpler conditions than those used in this study to simulate field environments, with the goal of mimicking field-grown plants more easily and cost-effectively than SmartGC.
Our results suggest that it is important to include gradual changes in the environmental factors of irradiance and temperature to accurately mimic the conditions of plants grown in the field, although this may depend on the target organs or the objectives of the experiment.
However, to clarify the most critical factors for simulating plants grown in the field, we need to further evaluate plant responses to environmental factors such as humidity, light quality, and light fluctuation, which are important yet thus far insufficiently examined features of the field environment. It will also be important to determine the extent to which conditions can be appropriately simplified by quantifying the effects of simplification on plants using SmartGC and transcriptome analysis.
SmartGC is especially useful for difficult-to-conduct field experiments, such as those using genetically modified plants, radioisotopes, or rare environmental conditions. Moreover, SmartGC can be used to predict the effects of future climate change on plants by allowing the evaluation of plants grown in a simulated environment.
In addition, SmartGC can contribute to improvements in modelling plant transcriptome dynamics. Since transcriptome models can be used to predict the environmental responses of plants, much effort has been made to improve the models. For example, Urquiza-García and Millar (2021) introduced absolute units of transcription (Flis et al., 2015) to the mathematical model of the circadian clock.
Models predicting plant transcriptomes in the field from training data obtained from controlled environments will be improved by using data obtained from the simulated field environment of SmartGC because the plant transcriptome training data will be more similar to field data than would data obtained from plants grown in a conventional growth chamber. Furthermore, statistical models that use data obtained from paddy fields as training data have low predictive power under environments that rarely occur in the field (Nagano et al., 2012). Since SmartGC can simulate rare environmental conditions in the field, incorporating transcriptome data obtained from plants grown in SmartGC will improve these models. Further studies utilizing SmartGC are needed to bridge the gap between field and laboratory studies and to facilitate a comprehensive understanding of plant responses to field environments.