Meta-analysis of the effect of expression of MYB transcription factor genes on abiotic stress

Data collection

To collect studies involving MYB overexpression and its effect on the abiotic stress response, a systematic search of the scientific literature was conducted using six electronic databases, namely KJD, MEDLINE, BCI, SCIELO, WOS, and RSCI, with ISI Web of Science (http://apps.webofknowledge.com/) and Endnote X9 (Thomson Scientific Company). The literature search was conducted on October 9–11, 2018 using a combination of terms such as (‘MYB gene’/‘MYB’) and (‘cold’/‘salt’/‘drought’/‘overexpress’/‘transgenic’/‘stress’) (see Supplemental Information 1—search term). The initial list contained 10,217 papers, out of which 7,836 were removed as they were determined to be duplicates. The remaining papers were examined, and another 2,355 were excluded due to the following reasons: non-original research articles, 104 (comprised of 37 reviews, 57 books, nine patents, and one conference abstract); data unrelated to plants, 606; data unrelated to abiotic stress, 1,162; MYB gene(s) not overexpressed, 472; and the abiotic stress was not the focus of the meta-analysis, 11. In total, 26 articles, written in English, spanning 18 years were selected for the meta-analysis (see Supplemental Information 1—paper). A total of 308 independent studies were extracted from 26 papers, and if there were multiple treatments in the paper, each treatment was treated as an independent study and appeared as an individual unit in the meta-analysis—an approach commonly used for plant biology meta-analysis (Kluemper & Qaim, 2014; Lehmann & Rillig, 2015; Wujeska, Bossinger & Tausz, 2013). The breakdown of the 308 independent studies was as follows: low temperature, 69; drought stress, 71; salt stress, 83; and normal conditions, 85. The means and sample sizes of the treatments for each individual study were obtained from the original study (Fig. 1). The data from the figures in the articles were extracted using GetData Graph Digitizer v2.21 (http://getdata-graph-digitizer.com). (The authors Zhaolan Han and Yuanchun Ma performed the search strategy and data collection. When any dispute occurs, Yuanchun Ma is our manuscript referee).

Effect size and moderators

Meta-analysis was conducted on several key response characters in cold-, drought-, and salt-stressed plants, and the data were presented as the natural logarithm of the response ratio (ln R). The response ratio in the meta-analysis results is an arbitrary unit of standardized expression, which is commonly used in the field of plant responses (Dong et al., 2018; Schaeffer, Manson & Irwin, 2013; Worchel, Giauque & Kivlin, 2013), it reflects effect size, which is often used to evaluate the overall and comprehensive effects of the study. ln R is used in the meta-analysis to indicate the log-transformed appropriate balance of positive and negative treatment effects between different response ratios (Borenstein et al., 2010; Hedges, Gurevitch & Curtis, 1999). ${\displaystyle lnR=ln{Y}_{\mathrm{TC}}{∕Y}_{\mathrm{NC}}}$

where Y_TC is the mean of transgenic plants (TC) with MYB overexpression and Y_NC is the mean of non-transgenic plants (NC) or the mean of plants with an empty vector. ln R-values above 0 indicate that the TC-induced condition caused a positive effect on the parameter, whereas values below 0 indicate that the TC-induced condition caused a negative effect on the parameter, and a value of 0 indicates no effect of MYB overexpression on plants (Dong et al., 2018; Ma et al., 2017a; Ma et al., 2017b).

The summary effect size reflects the weighted mean of the effect sizes from the primary studies. It was computed for each plant characteristic based on at least three studies from more than one article (see Supplemental Information 1).

Moderator variables were used to determine whether the effects of MYB overexpression appeared to be more pronounced with specific experimental conditions compared with others. Details on the experimental design or experimental variables that might influence the response of plants to abiotic stress (low temperature, drought or salt exposure) were collected from each study. These moderators were of two types: 1) experimental conditions: treatment medium, stress severity, and stress duration; 2) experimental materials: taxonomic class (monocot or dicot), genus of the gene donor and recipient, and whether the donor and recipient were from the same genus. A graded analysis was considered if the moderator included at least two categorical levels and the data of each level was sourced from at least three studies from ≥2 articles. If the data of the categorical level did not meet the criteria, they were grouped into one level designated as “other”. This group also required at least three studies from ≥1 article. When an experiment contained more than one level of severity, we classified each level as one of four types, “low”, “chilling”, “freezing”, and “severe freezing” to evaluate whether the summary effect size was affected by different levels of stress severity. For example, when plants were subjected to cold stress, 10–25 °C was defined as “low”, 0–10 °C as “chilling”, −10–1 °C as “freezing”, and −20–11 °C as “severe freezing” (see Supplemental Information 1—moderator level).

Meta-analysis

Comprehensive Meta-Analysis (CMA) software (v.2.2.023; Biostat, Englewood, NJ, USA; 2018) and GraphPad Prism software (v.7.00) created the forest plots. A random effect model was employed in all analyses. Some plant biology studies lacked the measurements of variation (Auge, Toler & Saxton, 2014; Dong et al., 2018; Mayerhofer, Kernaghan & Harper, 2013); therefore, the weight on the sample size (non-parametric variance) was calculated. Each primary study was weighted using nonparametric variance: ${\displaystyle V\ln R=\left(n_{\mathrm{TC}}+n_{\mathrm{NC}}\right)∕\left(n_{\mathrm{TC}}\ast n_{\mathrm{NC}}\right)}$ where V ln R reflects the variance of the natural log of the response ratio, n_TC is the number of transgenic samples, and n_NC is the number of non-transgenic samples or transgenic samples with an empty vector (Borenstein et al., 2010). The summary effect size was considered significant if p <  0.05, and the effect of a categorical variable was considered significant when the confidence intervals (95%) did not cross the dotted line. The sample size was used directly if it was stated explicitly in the article; otherwise, the following criteria were used to determine sample size: (1) if the sample size was not reported, n = 1 was used; (2) when only the LSD or standard error was given, n = 3 was used; and (3) if the sample size was reported as a range, the smallest value was used.

Heterogeneity, which is a measure of the real or true variation in effects, was evaluated with the Qt statistic and I². The Qt statistic is based on weighted squared deviations of a descriptive index that estimates the ratio of true heterogeneity to total heterogeneity across the observed effect sizes. It is a measure of weighted squared deviations (Borenstein et al., 2010; Higgins & Thompson, 2002). The Q-test was considered significant if p <  0.1, which indicated significant heterogeneity in the summary effect sizes. By contrast, I² ranges from 0 to 100%, and it reflects the proportion of true heterogeneity in the data set. The larger the I² value, the higher the probability of heterogeneity among the studies (Iacovelli et al., 2014).

It is recognized that publication bias exists because the studies that show relatively high effect sizes are more likely to be published (Borenstein et al., 2010). Potential publication bias was examined using three methods, namely the funnel plot (Viechtbauer, 2007), Begg and Mazumdar rank (Kendall) correlation analysis (Begg & Mazumdar, 1994; Borenstein et al., 2010), and Egger’s linear regression (Harbord, Egger & Sterne, 2006). Details of these methods are provided in our previously published paper (Ma et al., 2017a; Ma et al., 2017b).

Summary effects

The natural logs of the summary effect sizes of MYB overexpression in transgenic plants, relative to non-transgenic plants, are shown in forest plots (Figs. 2–10), including the plants in cold, drought, and salt stress conditions, respectively, and each figure contains the effect sizes of plants under non-stress conditions. The ‘forest plot’ showed the response ratio of the transformed/non-transformed treatments and the CIs of the transformed and non-transformed treatments. ln R-values greater than 0 indicated that MYB overexpression had a positive effect on the effect size, and negative values indicated that MYB overexpression inhibited the effect size. The summary effect size reflected the relative magnitude of the effect of foreign MYB overexpression, and its confidence intervals (CIs) indicated its precision. The summary effects were considered significant if their 95% CIs did not overlap with 0 (p ≤ 0.05). The raw percentage changes induced by foreign MYB overexpression are listed in Tables 1–3. Based on this criterion, four summary effect sizes were computed for studies addressing cold stress, six summary effect sizes were computed for studies addressing drought stress, and 13 effect sizes were computed for studies addressing salt stress.

Four summary effect sizes in plants under cold stress and non-stress conditions are shown in Fig. 2. The results are based on five species within 82 studies (see Supplemental Information 2). The majority of the MYB genes were from monocots (68 studies), among which 67 were from Oryza sativa studies. The remaining MYB genes were from other studies as follows: three from Malus domesica, 11 from Lycopersicum esculentum, and one from Triticum aestivum. Arabidopsis thaliana was the most represented recipient species (68 of the 82 studies). Under cold stress, two of the four measured plant parameters were significantly impacted by foreign MYB overexpression (Fig. 2A, p ≤ 0.05, Table 1). Specifically, the summary effects of foreign MYB genes on the survival rate of transgenic plants were significantly higher (75%) than those on NC plants. However, the overexpression of MYB genes in transgenic plants resulted in a significant reduction in relative electrolyte leakage by 32% compared with NC plants under cold stress. In contrast to cold stress, the plant parameters in non-stress conditions (Fig. 2B, p >0.05) were not affected by MYB overexpression.

Figure 3 indicates the six summary effect sizes of the TC/NC response ratio in plants under drought stress vs. non-stress conditions. Meta-analysis was performed on 12 species within the 93 studies contained in 16 publications (see Supplemental Information 3). Of the 93 studies of MYB overexpression, eleven different species served as gene donors. Oryza sativa and Arabidopsis thaliana were the main sources of MYB genes (27 and 25 studies, respectively), and dicots were the most common type of donor (50 studies). In terms of recipient plants, four species were studied, with Arabidopsis thaliana being the most representative (51 out of 93 studies). Dicots were more common as receptors than monocots (76 out of 93 studies). Only two of the six plant parameters were significantly affected in transgenic plants overexpressing MYB genes when subjected to drought stress (Fig. 3A, p ≤ 0.05). No significant difference in the summary effect size was detected under non-stress conditions (Fig. 3B). The decrease in water loss in drought-stressed TC plants at -22% was the most prominent among the other parameters compared to NC plants at 1% after MYB overexpression. In addition, the survival rate of TC plants increased dramatically by 3.6-fold compared with NC plants (Table 2).

Figure 4 shows the thirteen summary effect sizes of the TC/NC response ratio in plants under salt stress conditions compared to non-stress conditions. The results of the meta-analysis were based on 143 studies involving 13 species (see Supplemental Information 4). Of the 143 studies, 12 different species served as MYB donors, with Malus domesica and Scutellaria baicalensis (each with 24 studies) supplying the MYB genes. Dicots accounted for most of the gene donors with 88 studies. Four recipient species were studied, with Arabidopsis thaliana (68 studies) and Nicotiana tabacum (69 studies) being the most representative. Dicots were more common than monocots as gene recipients (139 vs. four). Only four of the 13 plant parameters were significantly impacted in MYB- overexpressing plants under salt stress (Fig. 4A, p ≤ 0.05). By contrast, only two of the 13 plant parameters were significantly impacted by MYB overexpression in non-stressed plants (Fig. 4B). The survival rate increased the most among all of the parameters under salt stress as a result of MYB overexpression. The survival rate of TC plants was 174% that of NC plants. The fresh weight and superoxide dismutase (SOD) activity of TC plants increased by 51% and 60%, respectively, compared with NC plants under salt stress. In addition, the germination rate of TC plants was markedly improved by MYB overexpression under both salt stress (52%) and non-stress (12%) conditions compared with NC plants (Table 3).

Heterogeneity

Heterogeneity, which represents the variation between studies, was determined by moderator analysis. This analysis evaluated whether the variation in observed treatment effect was over what was expected from the imprecision of the results within each study. p-_hetero for the Q-test < 0.1 or I² >50% (Borenstein et al., 2010) indicated significant heterogeneity in the size of the summary effect, meaning that the influence of the moderator on summary effect size was inconsistent among studies (Iacovelli et al., 2014). In this case, the differences in summary effect size among moderator groups were examined using the random-effects model to determine p-_hetero, which represents the heterogeneity among moderator classes.

The results suggested that none of the summary effects exhibited significant heterogeneity difference under cold, drought and salt stress.

Caution should be taken during interpretation due to the low statistical power caused by the small sample sizes from available studies for certain parameters, or large differences among moderators, both of which can affect heterogeneity. Likewise, a p-_hetero value of < 0.1 was insufficient to demonstrate that the true effects were consistent with the summary effect. Therefore, a random effect model was used to conduct a moderator analysis of the different variables according to the summary effects significantly affected by foreign MYB expression to investigate the influence of different moderators (Figs. 5–10). This situation is often encountered in meta-analyses in plant biology (Ma et al., 2017a).

Of the ten summary effects that were significantly affected by foreign MYB gene expression, six were subjected to moderator analysis because the moderators of these six effects could be grouped into at least two categorical levels (Figs. 5–10). A total of 12 moderators were analyzed in cold, drought, or salt stress studies.

Moderator analysis under cold stress

Moderator analysis under drought stress

Moderator analysis under NaCl stress