Effects of plant growth-promoting rhizobacteria on co-inoculation with Bradyrhizobium in soybean crop: a meta-analysis of studies from 1987 to 2018

Bibliographic research and data collection

Figure 1 shows the search strategy for the review presented according to the PRISMA reporting guidelines (Liberati et al., 2009). Data were collected from articles published in scientific journals, which were obtained by a systematic literature review using the Web of Science® and Google Scholar® databases. The search strategy “soybean AND (co-inoculation OR PGPR)” was applied in both databases in February 2018 by two independent reviewers (DMZ and LHF). Discussion between the two reviewers resolved any differences. If no consensus could be reached, another reviewer (LSAG) resolved the conflict. After screening relevant titles and filtering out duplicates, 79 articles were reviewed. The final article number was then reduced to 42 based on the following criteria: (i) articles written in English, Spanish, or Portuguese; (ii) studies that presented a measure of variance: coefficient of variation (CV), mean square residual (MSR), standard error of the mean (SE), or standard deviation of the mean (SD); (iii) studies showing the number of nodules, nodule biomass, shoot biomass, root biomass, shoot N content, and/or grain yield traits; and (iv) studies comparing inoculated treatments (Bradyrhizobium) × co-inoculated (Bradyrhizobium + PGPR). Interaction data with biotic or abiotic stresses were not extracted from articles.

Nodule, root, and shoot biomass were generally presented as dry biomass; however, in some cases, the values of fresh biomass were used when they were the only type of measure available. For the variable shoot N content, protein content was also used as an indirect source. The means and the measures of variance were extracted from the article tables, when provided. For figures, we extracted data using the ImageJ 1.5 software (Pérez & Pascau, 2013). Bar graphs that contained variance without specification were considered as SD.

Effect size and moderator variables

Estimates of the effects of the PGPR on the evaluated traits were obtained using the natural logarithmic response ratio (ln R) as effect size (Hedges, Gurevitch & Curtis, 1999): ${\displaystyle lnR=ln\left(\frac{Ti}{Tc}\right)}$

in which Ti is the mean of the co-inoculated treatment (Bradyrhizobium + PGPR) and Tc is the mean of the control treatment (Bradyrhizobium). The rate of the response is useful when different units are reported in the studies, while logarithmic transformation is necessary to properly balance the treatments of positive and negative effects to maintain symmetry within the analysis (Cooper, Hedges & Valentine, 2009). Thus, values above zero indicate an increase in the variable induced by PGPR, while values below zero reflect a reduction, and a value that equals zero means absence of the effect of PGPR. In addition, the ln R can be easily transformed into a percentage response (%R), using the following formula: ${\displaystyle \text{\%}R=100\times \left[exp.\left(lnR\right)-1\right]}$

Experimental conditions (field or pot) and PGPR genera used in co-inoculation were used as moderator variables in the present study, since they may influence the response of soybean to the effects of co-inoculation. Moderator variables were selected based on the criterion of a minimum of 15 observations in at least two scientific articles. The moderator variables were tested even when the evaluated trait presented no significant value, since the positive results may have been diluted in the general effect.

Meta-analysis

Prior to the construction of the meta-analysis models, data heterogeneity was verified by the Q (Cochran, 1954) and I² (Higgins & Thompson, 2002) tests to determine the use of fixed or random/mixed-effects model approaches. The synthesis produced by the meta-analysis is balanced according to the weight of each of the studies, so that they can contribute individually to the meta-analytic result. In this study, the inverse variance method (Hedges, Gurevitch & Curtis, 1999) was used to assign the weights: ${\displaystyle Wi=\frac{1}{Vi}}$

in which Wi represents the weight assigned to the i-th study and Vi is the variance of the i-th study. Thus, the lower the study variance, the greater its contribution to the synthesis generated.

The estimates produced by the meta-analysis and their respective 95% confidence intervals (95% CI) were presented in forest plot graphs. Therefore, the mean effect size was considered significant when its 95% CI did not overlap with zero. Statistical analyses were performed in the software R (https://r-project.org), using the meta (Schwarzer, 2007), metafor (Viechtbauer, 2010), and ggplot2 (Wickham, 2016) packages.

Metadata

Metadata was obtained from 42 published articles from 13 countries between 1987 and 2018 (Fig. 2A; Table S1). A total of 976 observations (n) were obtained from an aggregate of 74 trials, where each observation included a co-inoculated treatment (PGPR + Bradyrhizobium) and a control treatment (Bradyrhizobium) for the number of nodules (n = 278), nodule biomass (n = 228), shoot N content (n = 88), and grain yield (n = 78). Among the observations, 53% (n = 525) were obtained in pots and 47% (n = 451) under field conditions (Fig. 2B). Except for grain yield, reported only under field conditions, all other traits were observed under pot and field conditions. A total of 16 different genera of PGPR were used as co-inoculants (Fig. 2C).

Heterogeneity on the full dataset was highly significant by the Cochran test (Q = 29822.77, df = 975, p < 0.0001). The I² statistic also indicated high heterogeneity, which showed a magnitude of 96.40%. Due to the great heterogeneity of the observations, the meta-analysis was performed using random-effects models. Likewise, significant heterogeneity (p < 0.0001) was observed for the six evaluated traits grouped by the moderator variables, suggesting the use of mixed-effects models, in which we evaluated the moderator variables as random effect covariates and the observations as fixed effects (Cooper, Hedges & Valentine, 2009).

General effect of co-inoculation

The co-inoculation of soybean with PGPR showed a positive and significant effect on the number of nodules (11.40%, 95% CI [7.06 –15.93%]), nodule biomass (6.47%, 95% CI [0.59–12.70%]), root biomass (12.84%, 95% CI [3.64–22.85%]), and shoot biomass (6.53%, 95% CI [3.34–9.82%]) (Fig. 3). However, there was no increase in grain yield and shoot N content associated with co-inoculation, since their 95% CI overlapped with zero.

Effects of the moderator variables

The effects of the moderator variables on the number of nodules are shown in Fig. 4. Regarding the experimental conditions, the tests conducted under field and pot conditions showed significant effects of 8.55% (95% CI [3.09–14.29%]) and 12.84% (95% CI [7.38–20.12%]), respectively, on the evaluated traits (Fig. 4A). Both effect sizes can be considered similar, since the 95% CI overlapped considerably. Regarding the PGPR, the genera Azospirillum, Bacillus, and Pseudomonas showed positive effects for this moderator variable, increasing the number of nodules in 11.05% (95% CI [1.90–19.48%]), 26.05% (95% CI [14.71–36.59%]), and 10.41% (95% CI [3.43–17.41]), respectively (Fig. 4B). In relation to PGPR, only the genus Bacillus presented significant effects, leading to average increments of 33.12% (95% CI [22.27–44.93%]) (Fig. 4C). In contrast, in the pot experiments, the genera Azospirillum, Bacillus, and Pseudomonas presented significant effects of 26.77% (95% CI [8.26–48.44]), 22.09% (95% CI [6.67–39.72%]), and 9.81% (95% CI [2.13–26.30%]) on the number of nodules, respectively (Fig. 4D).

As shown in Fig. 5A, only the experiments conducted in pots showed significant effects on nodule biomass, with an average increase of 9.50% (95% CI [1.40–18.40%]). As for PGPR, the genera Azospirillum and Pseudomonas presented positive effects on this trait, showing increases of 14.65% (95% CI [6.76–23.13%]) and 17.34% (95% CI [7.17–29.49]), respectively (Fig. 5B). Although no significant effect of co-inoculation on nodule biomass was observed in the experiments conducted under field conditions, the partitioning of this effect in relation to the PGPR genera indicated a positive and significant effect of the genus Azospirillum, increasing the value of the trait in 10.69% (95% CI [3.70–18.16]) (Fig. 5C). In contrast, different PGPR in the pot studies revealed that only the genus Pseudomonas showed significant improvements in nodule biomass, presenting an increase of 16.80% (95% CI [6.58–27.90]) (Fig. 5D). On the other hand, a reduction of −18.32% in the average nodule biomass (95% CI [−32.08–1.74]) was observed by co-inoculation of other PGPR genera (Actinomadura, Aeromonas, Bacillus, Enterobacter, Herbaspirillum, Nocardia, Nonomuraea, Pseudonocardia, Rhizobium, and Streptomyces).

The effects of the moderator variables on root biomass are presented in Fig. 6. For the experimental conditions, only the experiments conducted in pots showed significant values, with an increase of 15.79% (95% CI [4.33–28.49%]) in root biomass (Fig. 6A). Regarding PGPR, the genus Pseudomonas was the only one with a positive effect on this trait, presenting an increment of 28.89% (95% CI [10.93–49.77%]) (Fig. 6B). Furthermore, according to the results, only the genus Pseudomonas resulted in a significantly increased root biomass (28.96%) (95% CI [10.68–50.25%]) (Fig. 6C).

Figure 7 shows the effects of the moderator variables on the shoot biomass. When the experimental conditions were analyzed, it was possible to verify that the trials carried out under field and pot conditions presented significant values of 5.44% (95% CI [3.14–7.80%]) and 8.27% (95% CI [3.06–13.76%]), respectively (Fig. 7A). Both effect sizes can be considered similar, since the IC overlapped considerably. For this moderate variable, the genera Azospirillum, Bacillus, and others (Actinomadura, Aeromonas, Enterobacter, Herbaspirillum, Methylobacterium, Nocardia, Nonomurae, Pseudocardia, Rhizobium, Stenotrophomonas, and Streptomyces) were the only ones that presented positive effects on shoot biomass, leading to increases of 6.39% (95% CI [3.12–9.76%]), 4.92% (95% CI [1.82–8.12%]), and 31.46% (95% CI [22.07–41.58]), respectively (Fig. 7B). The partitioning of PGPR genera under field conditions indicated that co-inoculation with bacteria of the genus Azospirillum increased plant biomass in 5.42% (95% CI [2.95–7.95%]) (Fig. 7C). In the pot trials, an extra 28.39% (95% CI [17.50–40.27%]) in the average shoot biomass (Fig. 7D) was promoted by the grouped genera (Actinomadura, Aerobonas, Enterobacter, Herbaspirillum, Methylobacterium, Nocardia, Nonomurae, Pseudocardia, Rhizobium, Stenotrophomonas, and Streptomyces).

For the traits shoot N content and grain yield, none of the differences were statistically significant, since the 95% CI of the moderator variables overlapped with zero (Figs. 8 and 9).