Effect of Rhizobium Seed Inoculation on Grain Legume Yield and Protein Content – a Systematic Review and Meta-Analysis

1. Introduction

Grain legumes are important food and feed crops and account for ~13% of the grain crops (cereals, pulses and oilseeds) produced globally [1,2]. Legumes play an important role in human nutrition, particularly in developing countries and have been described as “poor man’s meat” due to their high protein content and also being a good source of slow release carbohydrates, minerals and vitamins [3]. Grain legumes also account for a substantial proportion of total protein intake in individuals following a vegetarian or vegan diet and are an important component of traditional Mediterranean diets, which have been linked to a range of health benefits including a lower incidence of chronic and degenerative diseases including type-2 diabetes, certain cancers, cardiovascular diseases, dementia and obesity [3,4,5].

Grain legume crops have the capacity to fix atmospheric nitrogen via a symbiotic relationship with rhizobia (species of the soil bacterial genera Rhizobium and Bradyrhizobium) in the root nodules of legume plants. Root nodules are specialist root organs of legume plants, which are formed in response to infection of roots by rhizobia and provide favourable environmental conditions (e.g., maintenance of low levels of free oxygen via the O₂-buffering activity of leghaemoglobin proteins) required for optimum N-fixation by the bacterial symbionts [6,7,8]. N-fixation by the bacterial symbionts is an energy-requiring process, which is fuelled by a substantial proportion of the carbohydrate produced by the legume plants photosynthetic capacity and this partially explains the lower grain yields per unit area in grain legumes compared to cereal crops [7,8,9]. However, the capacity of symbiotic N-fixation also (i) explains the higher protein content in legume compared with cereal grain and (ii) reduces the requirement for N-fertilizer inputs and thereby greenhouse gas emissions/carbon footprint of arable crop rotations [7,8,9]. Additional benefits of including grain legumes into arable rotations can be increases in above- and below-ground biodiversity resulting in a reduction in weed, pest and disease pressure in subsequent cereal crops [7,8,9]. The use of grain legumes as break crops in arable rotations is therefore particularly beneficial in organic farming systems, which prohibit the use of synthetic, chemical crop protection products and mineral N-fertilizers [9,10,11,12].

Different groups of rhizobia are known to form functioning root nodules (=nodulate) with the main grain legume species and a recent review described that that there are now 14, 11, 6, 5, 5, 4, 3 and 2 rhizobia species known to nodulate common bean (Phaseolus vulgaris L.), soybean (Glycine max L.), cowpea (Vigna unguiculata L.), chickpea (Cicer arietinum L.), peanut (Arachis hypogaea), lentil (Vicia lens L.), faba bean (Vicia faba L.) and pea (Pisum sativum), respectively [13]. For example, species that efficiently nodulate common bean and faba bean are Rhizobium leguminosarum sv. phaseoli and Rhizobium leguminosarum sv. viciae respectively, while Bradyrhizobium japonicum and Rhizobium fredii are known to efficiently nodulate soybean [13]. The establishment of N-fixing root nodules therefore relies on high enough populations of legume species-specific rhizobium species being present in soil and the number of active root nodules and N-fixation capacity in a legume crop is correlated to the population density of rhizobia in the soil, which is known to decrease if suitable legume host plants are not grown for many years in the rotation [6,12,14,15].

It is also important to note that, both, (i) the establishment and subsequent N-fixing capacity of root nodules and (ii) legume crop performance, are known to be affected by environmental (including soil type, pH, available N-concentrations and organic matter content, precipitation, irradiation and temperature) and agronomic factors (including N-fertilizer inputs, irrigation) [12,14,15,16,17,18,19,20]. Also, legume plants have efficient mechanisms to regulate the level of nodulation in response to environmental factors (e.g., N-availability in soils) to prevent symbiotic N-fixation from becoming a metabolic burden and having a negative effect on plant growth and grain yield of legume crops [18,19,20].

One strategy to increase the performance (e.g., N-fixation, grain yield and protein content) of legume crops has been to inoculate soils or seeds with artificial Rhizobium/Bradyrhizobium inocula to compensate for (i) the absence of suitable rhizobia in soil (e.g., when non-indigenous legume species are grown for the first time) or (ii) a rapid decline in Rhizobium population density (e.g., where climatic conditions and/or infrequent growing of grain legume crops creates unfavorable conditions for persistence of rhizobia in soils) [21,22,23,24,25]. Commercial seed inoculum products, based on Rhizobium/Bradyrhizobium strains isolated from soils or root nodules, are now available for all major grain legume species [21]. There are also now efficient protocols to (i) remove bacteriophages from Rhizobium/Bradyrhizobium strains, (ii) produce inoculum by liquid fermentation technology and (iii) formulate rhizobia (e.g., with peat and/or clay carrier materials), and commercial seed inoculum products are often described to have superior nodulation and N-fixation capacity compared with strains naturally found in agricultural soils [21,22,23].

However, controlled experiments which compared grain yields and/or protein contents in grain legume crops grown with and without artificial inoculation have reported variable and sometimes contrasting effects. For example, studies by Elsheikh et al. [24], Sanora & Mallik [25] and Provorov et al. [26] reported significantly higher, similar and significantly lower grain yield in grain legume crops produced from inoculated compared with non-inoculated seed respectively. Also, a range of studies reported significant differences in nodulation/N-fixation capacity between strains of the same rhizobia species and a recent review concluded that rhizobia strains isolated in the target environments have superior nodulation/N-fixation capacity under field conditions, compared with strains selected for high N-fixation capacity in bio-assays and/or pot trials carried out under controlled growth chamber/greenhouse conditions [22,27,28].

As a result, there is still considerable controversy on whether and to what extent the use of artificial rhizobium inocula will significantly improve the yield and protein content and/or the economic performance of grain legume crops. Also, recent studies in Northern Europe have contributed to this uncertainty, because they showed that (i) even fields with no recent legume cultivation have sufficiently high rhizobia populations in soil for optimum nodulation/N-fixation of grain legume crops like faba bean and (ii) grain yield of non-inoculated faba beans, peas and lupins in long term experiments were as stable as those recorded for other spring crops (e.g., cereals) [28,29,30].

The objectives of this systematic review and meta-analysis were therefore to (i) estimate the overall effect of seed inoculation on grain yields and protein contents and (ii) identify potential confounding factors (fertilization, grain legume type/species, study type, country is which studies were carried out) that may explain the observed variation and/or contrasting effects reported in the scientific literature.

2. Materials and Methods

2.1. Criteria for Selecting Studies for the Review

All studies comparing grain yield and protein content of legumes inoculated with Rhizobia versus non-inoculated plants were included. The study included both field and glasshouse/controlled environment experiments which was included as a factor in the analysis. In addition, all species of legumes and Rhizobium or Bradyrhizobium bacteria were initially included in the review. The main outcomes were seed yield and protein content to reflect the effectiveness of Rhizobium inoculation. Yield was expressed as unit weight per given area, i.e., kg/ha, t/ha and g/plant, while protein was expressed as a percentage.

2.2. Literature Search Strategy

The literature search strategy was based on previously published protocols by Brandt et al. [31] and Baranski et al. [32]. Relevant publications were identified through an initial search of the literature with Web of Science using two search terms: “*rhizobium” and “inocula*”. The search terms were combined with Boolean logic (AND) and with truncation (*) in order to find all contrasting interventions and participants for this study. There were no language restrictions, all papers in different languages were included, and translation of papers was done by colleagues and/or using Google translate.

2.3. Screening Procedure

An EndNote library was constructed, and all duplicates were removed before the results were screened according to the eligibility criteria. Papers found in the initial search were filtered in a two-stage process:

(i)

Title and abstract screening. In order to minimize the risk of error two independent researchers read the title and abstract as recommended by Higgins & Thompson [32]. All studies relevant to the topic were included and if there was any doubt about suitability then the papers were included for stage two. In case the abstract was not available at this stage then the paper was also included for stage two.

(ii)

Full text screening. The full text of all relevant studies identified in Stage 1 was read and assessed for relevance.

2.4. Data Management and Extraction

All data relevant to primary objectives of the review were extracted from the included studies i.e., number of participants, mean value, standard deviation or standard error of the outcome measurement in each intervention group. Results reported as numerical values in the text or tables were copied directly into a spreadsheet, while data published in graphical form was enlarged, printed, measured (using a ruler) and then entered as described previously [32,33].

Authors of publications for which only the abstract was available were contacted by email and requested for full text together with any unpublished data from experiments which were within the scope of this study. Additionally, authors of all collected papers were asked for any other publications or unpublished results that could be added to this review. In order to obtain possible missing data (including measures of variation i.e., SD, SEM) the authors of the study were contacted, but very few responded and provided the additional information requested.

2.5. Data Synthesis and the Strength of Evidence Tests

The standardized mean difference (SMD) was used as an effect size to summarize treatment effects in each trial as previously recommended [32,34,35,36]. Results of the meta-analysis were visualized on forest plots which show SMD values together with 95% confidence intervals (CI) as a measure of variability (Supplementary Figures S3 and S4). The importance of each study was also expressed as a weight equal to the inverse variance which is used to calculate study weights. The SMD was calculated in relation to non-inoculated (control) plants, therefore a negative value indicates a higher performance (yield or protein content) of non-inoculated plants. Weighted meta-analysis was employed according to Baranski et al. [32] using the ‘metafor’ package of the R statistical environment [37]. The SMD from a single study was calculated using standard formulas within ‘metafor’ as follows

$$\mathrm{SMD}=\frac{{\overline{X}}_o-{\overline{X}}_c}{S_{within}}\times J$$

where ${\overline{X}}_o$ is the mean value for inoculated samples, and ${\overline{X}}_c$ is the mean value for non-inoculated samples. S_within is the pooled standard deviation of the two groups, and J is a factor used to correct for small sample size_.

J was calculated as:

$$J=1-\frac{3}{4\times \left(n_c+n_o-2\right)-1}$$

where $n_c$ and $n_o$ are the sample sizes in the two groups

and $S_{within}$ is calculated as:

$$S_{within}=\sqrt{\frac{\left(n_o-1\right)S_1^2+\left(n_c-1\right)S_c^2}{n_o+n_c-2}}$$

where $S_o$ and $S_c$ are the standard deviation in the two groups (organic and conventional).

The pooled SMD (SMDtot) across all studies was computed as:

$${SMD}_{tot}=\frac{{{\sum }_i^n}_1\left(\frac{1}{vi}\times {SMD}_I\right)}{{\sum }_i^{n}1\left(\frac{1}{vi}\right)}$$

where $v_i$ is a sampling variance estimated as:

$$v_i=\frac{n_c+n_o}{n_c{n}_o}+\frac{{SMD}^2}{2\times \left(n_c+n_o\right)}$$

The pooled or summary effect (${SMD}_{tot}$) was calculated for all parameters reported in a minimum of three studies, using the methods described by Lipsey and Wilson [38].

A positive SMD value indicates that the mean trait i.e., protein and yield are higher in inoculated plants, while a negative SMD indicates that the mean effect is higher in non-inoculated plants. Confidence intervals for each SMD were estimated based on standard methods using the ‘metafor’ package [37]

Unweighted meta-analyses were carried out to allow inclusion of studies that have only mean values and when measures of variability and/or sample size were unavailable. The effect size was computed as an ln-transformed ratio of inoculated means: non-inoculated means ${\overline{(X}}_o/{\overline{X}}_c)$ and was expressed as a percentage as shown below. The resampling method was used to evaluate comparison of the arithmetic average of samples with $ln\left(100\right)$. P values were derived from Fisher’s one-sample randomization test and p<0.05 was considered statistically significant [32].

$$lnRatio=ln\left(\frac{{\overline{X}}_o}{{\overline{X}}_c}\times 100\%\right)$$

The mean percentage difference (MPD) was calculated to facilitate value judgements regarding the nutritional importance of the relative effect magnitudes, and to compare these between weighted and unweighted meta-analysis. The MPD was expressed as ‘% higher ‘in inoculated or non-inoculated plants, and provide estimates for the magnitude of differences. For each data-pair used for SMD calculation the mean percentage difference was computed as:

$$\begin{gathered} +[{(\overline{X}}_o\times 100/{\overline{X}}_c)-100] \mathrm{f}\mathrm{o}\mathrm{r} \mathrm{d}\mathrm{a}\mathrm{t}\mathrm{a} \mathrm{s}\mathrm{e}\mathrm{t} \mathrm{w}\mathrm{h}\mathrm{e}\mathrm{r}\mathrm{e} {\overline{X}}_O>{\overline{X}}_c \\ \mathrm{or} \\ -[{(\overline{X}}_o\times 100/{\overline{X}}_c)-100] \mathrm{f}\mathrm{o}\mathrm{r} \mathrm{d}\mathrm{a}\mathrm{t}\mathrm{a} \mathrm{s}\mathrm{e}\mathrm{t} \mathrm{w}\mathrm{h}\mathrm{e}\mathrm{r}\mathrm{e} {\overline{X}}_O<{\overline{X}}_c \end{gathered}$$

2.5. Assessment of Heterogeneity

Heterogeneity between studies resulting from contrasting methodology used in each trial was assessed using Q statistics and I² statistics [39]. The results of these assessments were reported and discussed. A significant heterogeneity was assumed if the I² was less than 25% and the P value for the Q statistics was greater than 0.01. In order to explore causes of between-study variability (heterogeneity) the effects of moderators were analyzed using a group meta-analytical approach.

2.6. Assessment of the Risk of Bias

The article quality and risk of bias was assessed using a critical appraisal of each paper that provided suitable data for the weighted meta-analyses. It helped to monitor a number of quality criteria that have the potential to impact on the results of the study. Study quality was evaluated using the method described by Yin [40]. The assessment was done using a form of statements for which the following responses were possible: ‘Yes’ when the statement reflected the content of the paper, ‘No’ when there was no information in the paper described by the statement or ‘Unclear’ when information provided was inconsistent with the statement (see Supplementary Table S1. tor details of the results of the quality assessments of studies included in the weighted meta-analyses).

The final rating of the overall methodological quality was carried out using the method described by Guyatt et al. [41]. Only data/means from studies which also reported the number of replicates (n) used and measures of variation were included in the weighted meta-analyses, while data/means from all studies were included in the unweighted meta-analyses.

An overall assessment of the strength of evidence was performed using an adaption of GRADE (Grading of Recommendations, Assessment, Development and Evaluation) system, which included information about risk of bias for all studies that provided date suitable for the weighted meta-analyses, as well as inconsistency, indirectness and imprecision of the results, and publication bias [40]. Publication bias was assessed by funnel plot inspection as recommended by Mlinaric et al. [42].

3.4. Study Limitations

The main limitation of the evidence is that a large number of publications did not report measures of variation (e.g., standard errors, standard deviation, 95% confidence intervals) and could therefore not be included in the more scientifically sound weighted meta-analyses. This reduced the statistical power of the weighted meta-analyses, especially for protein content.

Another limitation was that the limited background information provided in most articles that contributed to the evidence base and as a result, we could not investigate interactions between soil, climate and agronomic explanatory variables, that may have explained the variation between legume species, fertilization regime and country/pedoclimatic background conditions identified in the multi-level model based weighted meta-analysis. Specifically, limitations included (i) insufficient information on soil, climatic and agronomic background conditions, (ii) different rhizobium strains and/or grain legume species being used in countries where experiments being carried out, (iii) the relatively small number of studies available from most countries, and (iv) the variation in study design and agronomic protocols used.

5. Conclusions

The study provided strong evidence that overall, the use of rhizobium inocula will result in a substantial increase in grain yields in bean, chickpea, lentil and soybean crops.

However, results also confirm previous studies which reported that (i) the yield gap between inoculated and non-inoculated crops decreases with increasing fertilization intensity (especially mineral N-fertilizer inputs) and (ii) under certain pedoclimatic and/or agronomic background conditions, inoculation may have no significant effect or result in lower yields compared with non-inoculated crops.

The finding that rhizobium inoculation resulted in similar or higher protein content suggests that N-fixation and residual-N available to crops grown after grain legumes will also increase substantially if inocula are used. This confirms that the use of rhizobium inoculum has a significant positive impact not only on (i) grain yields and associated economic gains for farmers, but also (ii) substantially increases environmental benefits (reduced need for mineral N-fertilizers and associated reduction in greenhouse gas emissions) of including grain legumes in arable rotations.

There is considerably scope to further improve the efficacy of rhizobium inocula through (i) research that improves the understanding of the complex interactions between legumes, rhizobium, other plant growth promoting microbes including VA mycorrhizal fungi, soil physical, chemical and biological and climatic factors and (ii) the development of improved strain isolation, efficacy screening and formulation methods for Rhizobium inocula.