| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
|
First published online October 16, 2003; 10.1104/pp.103.028191 Plant Physiology 133:1190-1197 (2003) © 2003 American Society of Plant Biologists What Affects mRNA Levels in Leaves of Field-Grown Aspen? A Study of Developmental and Environmental Influences1Umeå Plant Science Centre, Department of Plant Physiology (K.W., S.J.) and Research Group for Chemometrics, Department of Chemistry (F.P., A.B.), University of Umeå, S–901 87 Umeå, Sweden
We have analyzed the abundance of mRNAs expressed from 11 nuclear genes in leaves of a free-growing aspen (Populus tremula) tree throughout the growing season. We used multivariate statistics to determine the influence of environmental factors (i.e. the weather before sampling) and developmental responses to seasonal changes at the mRNA level for each of these genes. The gene encoding a germin-like protein was only expressed early in the season, whereas the other tested genes were expressed throughout the season and showed mRNA variations on a day-to-day basis. For six of the genes, reliable models were found that described the mRNA level as a function of weather, but the leaf age was also important for all genes except one encoding an early light-inducible protein (which appeared to be regulated purely by environmental factors under these conditions). The results confirmed the importance of several environmental factors previously shown to regulate the genes, but we also detected a number of less obvious factors (such as the variation in weather parameters and the weather of the previous day) that correlated with the mRNA levels of individual genes. The study shows the power of multivariate statistical methods in analyzing gene regulation under field conditions.
All genes are regulated by complex, gene-specific combinations of factors, including nutrient status, physicochemical conditions, developmental events, and a host of environmental parameters. In the cell, the information received through several signaling pathways is integrated and used to regulate gene expression by modulating critical variables such as the amount and DNA-binding activity of specific transcription factors. The classical method to elucidate factors that influence expression levels of a gene is to grow the organism under controlled conditions while varying one, or a few, factors at a time. However, in natural environments, organisms are exposed to conditions that are highly variable, and fluctuations are often both irregular and stochastic (for example, attacks by pathogens). As a consequence, plants have been obliged to evolve strategies to cope with such fluctuations. This may be particularly important for plants that are unable to move to avoid extreme, potentially lethal, variations in these factors. We decided to take an alternative approach, namely to exploit recent advances in statistics to elucidate factors influencing gene expression in aspen (Populus tremula) leaves grown under natural conditions. We describe here the methods used and results of this analysis.
Gene regulation is a complex process: Not only may a wide range of input signals be involved, but regulatory processes affecting expression may also operate at several different levels. For instance, transcription initiation, mRNA half-life, and translation efficiency may all be affected. The ultimate measure of gene expression, the amount of biologically active protein in the cellular compartment where the gene is functionally active, is often not easy to quantify. So, we decided to measure mRNA levels that are not only relatively easy to quantify but are also indicative of wider adjustments in the plant's polypeptide composition in response to changes in the environment. Thus, we measured the mRNA levels of 11 different genes in leaves of a field-grown aspen tree throughout the whole growing season, i.e. from bud burst to leaf abscission, and used multivariate statistics to correlate gene expression with meteorological data to see whether the genes were under the control of environmental factors, developmental processes, or both. Here, "environmental factors" refers to weather parameters on the day of sampling or the days before, whereas "developmental processes" refers to mechanisms responsible for transitions between phases, such as leaf flush and growth cessation, induced by longer term environmental cues during the growth season. Leaf flush in aspens native to Umeå typically occurs around June 1, and leaf expansion continues up to about June 20. No apparent changes to the foliage take place after this point until mid-September, when autumn coloration develops and the leaves are finally shed at the end of September. It is worth noting that the foliage develops synchronously: All leaves on a single tree are of the same age and most likely in the same developmental stage. In the interpretation of the results, we use the term development to refer to day dependence. We selected genes on the basis of expressed sequence tag sequencing data (Bhalerao et al., 2003
By applying multivariate methods, such as partial least squares (PLS; Wold et al., 1984
We used the 11 probes to follow changes in the amounts of homologous mRNAs over the growing season. For simplicity, in the following text we refer to the mRNAs hybridizing to each of the probes under most stringent conditions as mRNA from a single gene. This can, as in other studies dealing with species where the full genome sequence has not been determined, not easily be verified, but two genes so similar that they will be cross-hybridizing under these conditions are likely to have identical functions, so in a study like this, this is a reasonable simplification. The expression patterns of the 11 genes over the growing season all differed according to the mRNA measurements. Some, like Deh (Fig. 1), showed large day-to-day variations but no apparent longer-term trends, indicating that they are regulated primarily by environmental factors. Others appeared to be regulated mainly by developmental processes. For example, Glp mRNA was only found in June, whereas Mt mRNA increased steadily as the leaves aged (Fig. 1). Yet others showed trends in mRNA abundance, combining features of both environmental and developmental patterns. RbcS, for example, had a consistently high mRNA level in June, but later in the growing season, the levels appeared to be regulated by environmental factors. The mRNA blots for all the 11 genes are presented in the supplementary material. An overview of the weather conditions during the whole period is found in Figure 2.
To get high-quality expression data allowing meaningful statistical analysis, the raw data were subjected to an elaborate normalization and quantification process. All blots were hybridized with an rRNA probe to correct for loading differences, and a standard mixture was applied to each gel, allowing comparison of hybridization signals for the same gene on different gels. The resulting expression data were then calculated for each sample from corrected hybridization signals measured with a phosphor imager. The expression profiles for all the genes are given in Figures 3, 4, 5. Figure 3 also presents results from the PLS models (see below).
A PCA analysis was done on the normalized mRNA levels for all the genes throughout the season. An initial PCA analysis showed that the June 6 sample was a strong outlier; therefore, it was excluded from this analysis. This day and the day preceding it were unusually rainy and cold. Seasonal changes in gene expression were detected in the score plot for the two first principal components, explaining 26% and 23% of the variance, respectively (Fig. 6). The different phases of the season, identified as separate regions of the scatter plot, related to different stages of leaf development. Phase 1 (May 22–June 1) corresponded to the "pregrowth" period before the start of rapid leaf expansion. Phase 2 (June 9–June 22) coincided with the "growth" period when rapid leaf expansion occurred. Phase 3 (June 26–September 1) was designated the "productive phase." Phase 4 (September 5–September 29) corresponded to leaf senescence. The only ambiguity in these assignments occurred in the transition between phases 3 and 4 because the September 8 scores appeared in the phase 3 region. Apart from this, the different developmental stages were well separated by the expression patterns.
It was not possible to derive a model that was valid for the whole season for any of the genes. Presumably, mRNA levels depend on different mechanisms in the different developmental phases, resulting in different variables that are important for different phases. So, instead, local models (covering shorter time periods) had to be developed, and for each gene, we present here the model that successfully described its expression pattern for the longest period (Fig. 3). In each case, we used as simple a PLS model as possible, i.e. developmental and environmental factors that did not appear to influence the gene's mRNA levels were excluded from the final PLS model. In the following sections, we discuss the data related to each of the 11 genes. For Cox, only the later part of the season (11 observations; August 25–September 29) could be successfully modeled. The most important factors were Day (the number of days since the 1st d of harvesting) and relative humidity, both of which had a negative effect on expression, whereas Sun had a positive contribution. In other words, expression was positively correlated to sunlight and low humidity and was weaker later in the season. The similarity in size of the two bars for the two Sun parameters (for the sampling day and the day before) in Figure 2 shows that the amount of sunlight the day before the sampling day was as important as the amount of sunlight on the sampling day. The expression of Deh was modeled for the whole season except for the period covered by the first five sampling occasions (May 22–June 6). As expected, its expression was strongly positively correlated to low temperatures on the sampling day (as shown by the large negative orange bars in Fig. 3) and, to a lesser extent, on the previous day. Interestingly, large variations in temperature, wind, and relative humidity also had a positive effect on its mRNA levels. Furthermore, Deh showed a fairly low Day dependency, even though the PLS model was based on a large interval. This shows that Deh is mainly environmentally regulated in aspen leaves. Elip could only be modeled in the "productive phase" or, more precisely, between June 12 and August 22. During this period, the expression of Elip was also positively correlated to high light and negatively correlated to rain and high humidity. These factors also influence its expression in the "senescence phase" according to C. Külheim, J. Keskitalo, K. Wissel, M. Sjöström, and S. Jansson (unpublished data), but our data provided no evidence to support this assertion. Descriptors from the sampling day for all weather parameters are most important for the model. Large deviations from 20°C also make a positive contribution to its expression. This model shows the smallest Day dependency of all models; thus, the mRNA levels are only environmentally regulated. Mt mRNA levels were modeled from June 6 to September 15. The slope of the seasonal expression profile can be modeled using Day as single variable, indicating that the Mt gene was mainly developmentally regulated and gradually induced as the leaf aged. By including four more descriptors describing Sun, Wind, and Temp (all affecting expression negatively), seasonal peaks can be roughly described. The model for RbcS included all phases except the initial pregrowth phase. Only a few descriptors were included in the model, in which Day was the most important factor. High light was correlated with high expression, whereas deviation from 20°C and high humidity was correlated with low mRNA levels. Thus, developmental processes were most important (the highest mRNA levels occurred during leaf expansion), but environmental factors modulated the RbcS mRNA levels. Ubq was modeled for 31 observations ranging from June 6 to September 26. The expression of Ubq has a flat profile with a slight increase over time. High Temp and low Sun have positive correlation with its expression. Thus, Ubq mRNA levels were mainly determined by weather factors during most of the season, but late in the "senescence phase," the gene was induced by developmental factors. For the remaining five genes, no valid models were obtained, although all showed variations in mRNA levels over the season (Figs. 4 and 5). For Glp, the expression pattern is clear: The gene is only expressed during the "growth phase" and is turned off for the rest of the growth season. H2B had high mRNA levels on June 6, during the transition from the "pregrowth" to the "growth phase," coinciding with the expected peak of cell division activity in the leaves. Surprisingly, the second highest mRNA level was recorded on September 29, when the leaves were soon to be abscised. Between these two peaks, the mRNA levels fluctuated considerably without any obvious trend. For PrP, Gs, and PsbS mRNA, fluctuations were large but did not appear to be correlated strongly enough to weather or developmental phases to be modeled using this approach.
We have used multivariate statistics to determine the main factors influencing mRNA levels of 11 different genes expressed in aspen leaves. Although environmental influence of plant metabolism is a central issue in ecophysiology and has been studied with advanced statistics (see e.g. Ekblad et al., 1995  ; Luwe, 1996), this is, to our knowledge, the first study where multivariate statistics have been used to study gene expression in plants grown under natural conditions in the field. All 11 genes showed considerable variations in mRNA levels, and for seven of them, we were able to identify the factors most correlated with the expression level, thus probably regulating the gene expression. For one, Glp, the developmental influence was clearly dominant because it was only expressed in the "growth phase." For two genes (Deh and Elip), only weather parameters were of clear importance, and for the remaining four (Cox, Mt, RbcS, and Ubq), both developmental processes and environmental factors were important. Regulatory factors have been identified previously for seven of the tested genes in aspen: Deh, Elip, Glp, Mt, Ubq, and PsbS. The importance of the previously identified factors was confirmed for four of these genes in this study: Deh was induced by low temperatures, Elip by high light and low temperatures, Glp was expressed early in the season, and both Mt and Ubq mainly late in the season. The only exception, for which the importance of previously identified factors was not confirmed, was PsbS. An earlier, more detailed analysis has shown that the PsbS mRNA level could be totally predicted by weather factors under some circumstances but developmental factors in others (Külheim et al., 2002). This pattern was either not presented in the growth season studied, or our analytical methods were insufficiently sensitive to elucidate the important factors. The result for RbcS is reasonable considering what is known about the photosynthetic reaction: High light has been shown to induce Rubisco levels in many other species (Anderson et al., 1995); also, more dark reaction, as compared with light reaction, components are expected to be needed in low temperatures, which slows down enzymatic activity, and drought, which leads to stomatal closure. These observations indicate that PLS could correctly extract important parameters from the very complex data set. For dissecting the importance of highly correlated variables (such as Temp, Sun and Day), standard multiple linear regression methods have less analytical power than PLS. However, although multivariate methods can dissect complex phenomena, there are also cases where the eye is able to see patterns that the models miss. For example, Mt and Ubq were prominently induced in the late stages of senescence, but the number of data points (days) was too low to generate a statistically reliable model showing this induction. In addition to these known factors that induced the selected genes, many unexpected factors were detected, e.g. the induction of Cox by high light, drought, and low temperatures in the autumn, the induction of Mt by low temperatures and low light and its repression by wind, and the repression of Ubq by high light and low temperatures. It is tempting to speculate on the biological implications of these findings. It is, for example, possible that stress factors that promote leaf senescence and chlorophyll breakdown (high light, drought, and cold) induce mitochondrial activity because respiration has to be responsible for a larger proportion of the energy generation in the leaf when photosynthesis is impaired. However, such hypotheses have to be addressed in more specific studies. It was also interesting to see that the weather factors of the previous day could be as important (as in the case of Cox) or even more important (as in the case of Ubq) than those of the current day. We have found recently that the PsbS gene is also regulated primarily by the previous day's weather (C. Külheim, J. Keskitalo, K. Wissel, M. Sjöström, and S. Jansson, unpublished data). Although this is not surprising from a biological perspective, it is a feature that probably would not have been detected in "classical" molecular biological experiments. This again shows the strength of the multivariate methods.
Another factor that could easily be overlooked in traditional gene expression studies is the importance of variation, here illustrated by Deh, which was induced by variable weather. Although variation is highly relevant to an organism in its natural environment and a factor that ecologists often consider, it is traditionally neglected by molecular biologists. Variation may be particularly stressful for plants (that are sessile), and we hope that this work could inspire other researchers to address the importance of variation for gene expression more carefully. For example, increasing the ability to cope with rapid fluctuations in light was shown recently to be the probable functional significance of a process regulating photosynthetic light harvesting (Külheim et al., 2002 As stated in the introduction, gene regulation is a very complex phenomenon. It is possible that our results would have been different, at least in part, if we had performed the study in a different year. For example, the summer in the year 2000 was unusually wet, and the influence of short-term fluctuations in rainfall on gene expression, which were minor this year, may be greater in summers when low soil water content is a severe stress factor for the tree. Nevertheless, we believe that the results demonstrate that the multivariate techniques and genomic tools (DNA microarrays) that we have developed have the potential to address sophisticated questions about the interactions between plants and their environment.
Plant Material and RNA Preparation
Leaves (about 20 per sampling) were collected twice a week from a free-growing aspen (Populus tremula) tree on the Umeå University campus (63° 50'N, 20° 20'E) from May 22 to October 3, 2000, on each occasion at 11 to 12 AM. They were snap frozen in liquid nitrogen and stored at –70°C until RNA was extracted and prepared according to Bhalerao et al. (2003
Clones encoding the various proteins were selected from the PopulusDB (http://www.upsc.nu): Mt, UA31BPD11; Cox, UA17CPG08; H2B, I004P48; RbcS, I001P90; Prp, UL56P.G09; Gs, G065P67Y; Glp, C043P24U; Deh, UA14CPF09; Elip, I001P19; Ubq, I037P11; PsbS, I001P27; and an rRNA probe. The selected expressed sequence tag clones were amplified by PCR in 50-µL reaction volumes containing 10x reaction buffer, 4 µL of 25 mM MgCl2, 2 µL of 10 mM dNTP, 8 pmol plasmid-specific 5' and 3' primers, and 2 units of Taq-DNA-Polymerase (Perkin-Elmer Applied Biosystems, Foster City, CA). The PCR products were excised from polyacrylamide gels and purified using a QIAquick Gel Extraction Kit (Qiagen, Hilden, Germany) as described by the manufacturer. Twenty-five nanograms (10 ng for the rRNA probe) of the probes was labeled with
RNA was separated, blotted, and hybridized according to Ganeteg et al. (2001
The signal intensities of the hybridization products were quantified by counting the
Relevant weather descriptors were calculated from data collected at a weather station located on the University of Umeå campus about 50 m from the tree from which the leaves were harvested. The weather station is supervised by the department of applied physics and electronics at the University of Umeå, and data are collected hourly throughout the year. For the modeling procedure, 24 descriptors were calculated describing temperature, sun, rain, humidity, and wind. Most variables were calculated for both the day of sampling and the day preceding the sampling. Weather descriptors were based on data for the 10, 36, or 48 preceding hours, depending on the interval of interest. Descriptors with the suffix 36 are based on the 36-h interval preceding harvesting. Rain_tot-2 refers to rainfall in the 2 d before harvesting. All other descriptors are based on the 10-h interval preceding harvesting. All weather descriptors are further explained in Table I.
Two methods of multivariate data analysis were used in this work. PCA (Jackson, 1991
The central idea of PCA is to extract a few so called principal components which describes as much as possible of the variation present in the data. The principal components are linear combinations of the original variables and uncorrelated to each other.
The principal components can be determined using the NIPALS algorithm (Wold, 1966
PLS is a multivariate regression method that relates the data matrix X, here the weather conditions, to a y response that can be either single (y) or multiple (Y). PLS has proved to be a powerful tool for finding relationships between descriptor matrices and responses when there are more variables than observations, the variables are collinear to each other, and they are noisy. In the article, the response is the mRNA levels of the different genes. The PLS theory and methods discussed in this report concern single y responses. As in PCA, principal components are constructed to reduce the dimensions of X. To obtain the principal components, PLS maximizes the covariance between the response variable (y) and a linear combination of the original variables (t = Xw), where t is the score vector, X is the data matrix, and w is the weight vector (for a more in-depth description of PLS, see (Burnham et al., 1996
It has been shown recently that the first weight vector, w1, for the PLS model is the best estimate for how important a variable is for describing the response (Trygg and Wold, 2002 Received June 6, 2003; returned for revision August 11, 2003; accepted August 11, 2003.
Article, publication date, and citation information can be found at www.plantphysiol.org/cgi/doi/10.1104/pp.103.028191.
1 This work was supported by the Swedish Research Council and Kempestiftelserna.
2 Present address: Universität Rostock, Fachbereich Biowissenschaften Biochemie, Albert-Einstein-Strasse 3, 18051 Rostock, Germany. * Corresponding author; e-mail stefan.jansson{at}plantphys.umu. se; fax 46–90–786–66–76.
Anderson JM, Chow WS, Park Y-I (1995) The grand design of photosynthesis: acclimation of the photosynthetic apparatus to environmental cues. Photosynth Res 46: 129–139[CrossRef]
Bhalerao R, Keskitalo J, Sterky F, Erlandsson R, Björkbacka H, Jonsson Birve S, Karlsson J, Gardeström P, Gustafsson P, Lundeberg J et al. (2003) Gene expression in autumn leaves. Plant Physiol 131: 430–442 Burnham AJ, MacGregor JF, Viveros R (1999) A statistical framework for multivariate latent variable regression methods based on maximum likelihood. J Chemometrics 13: 49–65 Burnham AJ, Viveros R, MacGregor JF (1996) Frameworks for latent variable multivariate regression. J Chemometrics 10: 31–45[CrossRef] Ekblad A, Lundquist P-O, Sjöström M, Huss-Danell K (1995) Day-to-day variation in nitrogenase activity of Alnus incana explained by weather variables: a multivariate time series analysis. Plant Cell Environ 17: 319–325
Ganeteg U, Strand Å, Gustafsson P, Jansson S (2001) The properties of the chlorophyll a/b-binding proteins Lhca2 and Lhca3 studied in vivo using antisense inhibition. Plant Physiol 127: 150–158 Garthwaite PH (1994) An interpretation of partial least-squares. J Am Stat Assoc 89: 122–127[CrossRef] Jackson JE (1991) A User's Guide to Principal Components. Wiley, New York Joliffe IT (1986) Principal Component Analysis. Springer, New York
Külheim C, Agren J, Jansson S (2002) Rapid regulation of light harvesting and plant fitness in the field. Science 297: 91–93 Luwe M (1996) Antioxidants in the apoplast and symplast of beech (Fagus sylvatica L.) leaves: seasonal variations and responses to changing ozone concentrations in air. Plant Cell Environ 19: 321–328[CrossRef] Trygg J, Wold S (2002) Orthogonal projections to latent structures (O-PLS). J Chemometrics 16: 119–128[CrossRef]
Welling A, Moritz T, Palva ET, Junttila O (2002) Independent activation of cold acclimation by low temperature and short photoperiod in hybrid aspen. Plant Physiol 129: 1633–1641 Wold H (1966) Estimation of principal components and related models by iterative least squares. In PaR Krishnaiah, ed, Multivariate Analysis. Academic Press, New York Wold S, Ruhe A, Wold H, Dunn IIIWJ (1984) The collinearity problem in linear regression: the partial least squares approach to generalized inverses. SIAM J Sci Stat Comput 5: 735–743[CrossRef] This article has been cited by other articles:
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| ASPB Publications | PLANT PHYSIOLOGY® | THE PLANT CELL | |
|---|---|---|---|
TOP
ABSTRACT
RESULTS
-32P-dCTP, using the RadPrime DNA Labeling System (Invitrogen, Stockholm, Sweden), again as described by the manufacturer. 
-decay in a PhosphorImager (Molecular Imager System GS-505/525, Bio-Rad Laboratories, Hercules, CA) and processing the acquired data with Molecular Analyst software (version 2.1.1, Bio-Rad). rRNA hybridization intensity was used to quantify the relative loading of the sample and standard RNA on the gels. All northern blots were hybridized once with a 32P-labeled rRNA probe and quantified as described above. Sample and ST gene expression intensities were normalized after background subtraction to their respective rRNA intensities using correction factors determined by relating each rRNA signal to the strongest signal in the same lane of transferred RNA. The sample and ST intensities were multiplied by the correction factors thus derived, adjusted in accordance with the respective total amount of RNA loaded onto the agarose gel, and transferred to nylon membranes. The corrected signal intensities were then normalized to those of the ST by calculating the ratio of the sample to the ST expression intensities. For each probe, data were normalized to the maximum expression (set to 100%). The normalized data set was used for further analysis. 
