Infection profoundly affects the physiology of host cells, including the levels of mRNAs within each cell. The ability to measure changes in host mRNA levels during infection is greatly limited by the intrinsic variability of the tissue and the difficulty of accurately reproducing infection across multiple replicates. As result, typical studies report significant changes in only a few percentage of the genes being assayed [3
]. Similar difficulties limit the power of many other experiments that measure transcriptional responses to genetic and physiological perturbations. Here we have shown, by means of an experiment of unprecedented statistical power, that significant changes in the levels of nearly all host transcripts can be measured during infection of soybean with the eukaryotic pathogen Phytophthora sojae
. Up to 36.8% of the changes were greater than 2-fold in magnitude but the majority were of lower magnitude. Up to 15% of the changes were less than 1.2-fold in magnitude. Similar widespread reprogramming of the transcriptome was also detected in response to genotypic differences among soybean cultivars, and even in response to different times of the day at which plants were harvested.
An experiment such as this, combining 72 biological replicates, each comprised of 20–30 plants, is very successful at identifying overall trends, in this case identifying overall responses to pathogen infection. In many of the individual genes we examined, the overall measurement also was a very good representation of a consistent response to infection observed across many independent experiments. On the other hand, some genes showed significant fluctuations from experiment to experiment. A major challenge in the future will be to design experiments that can dissect these changes and ask whether the fluctuations simply result from technical variations in the assay, whether the fluctuations result from small uncontrolled variations in the conditions in which the biological material is grown, or the more interesting possibility that fluctuations in mRNA are a normal physiological event. It is possible for example that organisms have evolved the ability to tolerate significant changes in the levels of mRNAs or other cellular components, and there is no evolutionary advantage to imposing extremely ranges on mRNA levels.
By re-analyzing subsets of our data that represented the scale of replication more commonly found in microarray experiments (4 replicates), we showed that approximately half of the transcriptional changes that we detected could not be observed with the small-scale experiments. Taking the intersection of six independent subsets greatly reduced the number of detected changes further, showing that this common practice for combining the results of multiple experiments is extremely and unnecessarily conservative. On the other hand, taking the union of multiple gene lists more closely approximated the results from the full-scale experiment.
A recent expression QTL study in Saccharomyces cerevisiae
showed variations in the levels of 79%, 69% and 47% of the detectable transcripts in response to treatment, genotype and treatment × genotype interactions respectively [15
]. Those findings, together with ours, suggest that widespread, low-magnitude transcriptional remodeling may be a normal process during physiological adaptation in eukaryotes, but one that is missed by conventional experimental designs. The extensive treatment × genotype interactions observed in both these studies suggest that the transcriptional reprogramming is genetically controlled.
The overall physiological response of an organism or cell to a stimulus may require coordinated changes in a wide array of cellular components. Those changes in turn may require compensating or reinforcing changes in an even wider array of functionally-connected cellular components. As a result, understanding how a specific set of transcriptional changes relates causally to a physiological change at a whole system level, is a major challenge. Our analysis suggests that low magnitude expression changes may be of functional significance. The observation that the patterns of low-magnitude transcriptional remodeling were significantly different among most functional categories, in some cases even among genes with less than 20% perturbation, is consistent with the hypothesis that low amplitude remodeling has functional significance. However, we cannot rule out that some percentage of genes may show low magnitude transcriptional modulation that has no functional significance, that is, they represent a low uncontrollable level of transcriptional variation that soybean (and other organisms) may have evolved to tolerate. Interestingly, most of the transcriptional remodeling of defense and disease response genes was of high magnitude, while low-magnitude remodeling was widespread in the other functional categories. We speculate that strong transcriptional modulation of disease and defense-related genes is required for the host to directly engage the pathogen, while the numerous other genes that function in other categories are coordinately modulated to support or adjust to the direct response.
Biological gene regulatory networks are highly interconnected systems. Non-linear, synergistic interactions are common. Large numbers of genes with low magnitude transcriptional modulation could potentially be just as important in conferring phenotypes and mediating physiological adaptation as the small numbers of genes that show large magnitude modulations. However, understanding the role of pervasive low magnitude remodeling may require using computational modeling approaches at a systems level, as well as improved technologies for accurately and cheaply measuring those changes. Systems approaches will also be needed to develop a deeper understanding of how consistent small magnitude changes and stochastic fluctuations are integrated to produce phenotypes.