Arbuscular mycorrhizal fungi increase grain yields: a meta-analysis
Summary
- Increasing grain yields of food cereal crops is a major goal in future sustainable agriculture. We quantitatively analyzed the potential role of arbuscular mycorrhizal (AM) fungi in enhancing grain yields of seven cereal crops with exceptional importance for human nutrition across the globe: corn, wheat, rice, barley, sorghum, millet and oat.
- We conducted a meta-analysis for three datasets including both English and Chinese language publications: the ‘whole’ dataset including both laboratory and field studies (168 articles); the ‘field’ dataset comprising only field studies (97 studies); and the ‘field-inoculation’ dataset including only AM fungal inoculation studies conducted in field conditions (70 articles).
- We found that the AM fungal effect on grain yield was less pronounced in field and noninoculation studies. AM fungal inoculation in field led to a 16% increase (overall effect) based on the ‘field-inoculation’ dataset; this effect was variable (77% trials had positive values), crop-specific, lower for new cultivars released after 1950 and further modulated by soil pH.
- Although there are neutral and negative effects of AM fungi on grain yields, we emphasize the importance of integrating AM fungi in sustainable agriculture to increase grain yields of cereal crops.
Introduction
Feeding people depends heavily on c. 100 crop plants (Prescottallen & Prescottallen, 1990) of several thousand edible plant species worldwide (FAO, 1995). Among these, cereal crops are of exceptional importance for human nutrition across the globe due to their high content of starch but also fat and protein (Sarwar, 2013). Therefore it is alarming that grain yields of many of these cereal crops plateaued in the 21st century in major production regions as East Asia, Northwest Europe and India (Grassini et al., 2013), while we are faced with the need to feed an ever-increasing size of human population. Reasons for this are manifold and include land degradation, impoverishment (e.g. nutrient depletion, decrease in soil biodiversity and erosion) and contamination (e.g. by fertilizers, pesticides and heavy metals) of soils, and agricultural mismanagement.
It is urgent not only to improve food security but also to protect and promote soil biodiversity and functionality by implementing sustainable management practices. Arbuscular mycorrhizal (AM) fungi are a promising option in terms of sustainable agriculture and food security (Rillig et al., 2016; Thirkell et al., 2017). These fungi are integral components of soil and plant roots forming a symbiosis with many cereal crops (Smith & Smith, 2011). In this symbiotic relationship, the crops supply lipids and/or sugars to their symbiotic fungi, thus providing the fungi with a source of carbon for their metabolic needs (Bago et al., 2000; Jiang et al., 2017); in return, the fungi provide benefits to their associated hosts (Parniske, 2008). The most prominent feature of AM fungi is the acquisition of immobile nutrients beyond the range of plant roots via their hyphae (Smith & Read, 2008) including macro- and micronutrients (Marschner & Dell, 1994; Govindarajulu et al., 2005). Plants with arbuscular mycorrhiza often show increased nutrient concentrations (Lehmann et al., 2014; Lehmann & Rillig, 2015) and improved growth and yield production (Hoeksema et al., 2010; Pellegrino et al., 2015). Additional potential functions and services provided by AM fungi to their hosts are stress alleviation (Chandrasekaran et al., 2014; Jayne & Quigley, 2014), pathogen protection (Veresoglou & Rillig, 2012) and weed control (Meng et al., 2001; Veiga et al., 2011).
Outcomes of the AM fungal symbiosis are strongly context dependent (Hoeksema et al., 2010). Various biotic and abiotic factors modulate the AM fungal effect on the yield of their associated hosts. Among the biotic factors, the traits of the host and the fungi themselves are most critical. Crop plants vary in their responsiveness to AM fungi due to their morphology (e.g. root characteristics: Yang et al., 2015) while AM fungi differ in benefits provided to the plant (Powell et al., 2009; Werner & Kiers, 2015). Additionally, the high yielding crops widely used today are the product of artificial selection typically performed under high nutrient conditions. These conditions are not only less favourable for diverse AM fungal communities but also alter the AM fungal and crop plant interactions (Lehmann et al., 2012; Martín-Robles et al., 2018). In addition to the fungal partner, there is an immense diversity of soil microbes potentially influencing the crop plant–AM fungi association (Zhang et al., 2016). Microbes also affect nutrient availability by mobilization and mineralization processes and can hence modulate AM fungi-mediated effects (Hetrick et al., 1990; Khadge et al., 1990).
Among the abiotic factors, clearly the nutritional status of the soil and plant plays a key role in the modulation of the AM fungal effects. The major function of AM fungi is suggested to be nutrient acquisition; hence under high nutrient conditions AM fungi can shift from a net benefit to a cost for the host (Johnson et al., 2015; Liu et al., 2015; Werner & Kiers, 2015). Additionally, the nutritional status of the soil and plant is strongly influenced by soil parameters including texture and pH determining, for example, nutrient availability and solubility (Karagiannidis & Hadjisavva-Zinoviadi, 1998). Hence, soil parameters also have a strong potential to influence AM fungal effects on crop plants and their yield (Lehmann et al., 2014; Lehmann & Rillig, 2015). In addition to soil and nutrients, the type and management of the agricultural system further influences the AM fungal effect on their host (Ryan et al., 2008; Bilalis et al., 2012). The degree of environmental control of the study system or the introduction or management of AM fungi in the soil strongly affects AM fungal performance (Lehmann et al., 2014; Thirkell et al., 2017). Crop plants experience a multitude of stressors during their growth, for example pests, heavy metals and drought. The resulting challenges to plant performance are another critical modulator of AM fungal effects (Chandrasekaran et al., 2014; Jayne & Quigley, 2014; Yang et al., 2014; Cabral et al., 2016).
To evaluate the importance of AM fungi for cereal grain yield given certain abiotic and biotic factors, we need a profound quantitative understanding. However, only one quantitative synthesis has so far targeted the AM fungal effects on wheat in field studies, investigating the effect of AM fungal symbiosis on vegetative and edible plant portions and nutritional parameters (Pellegrino et al., 2015). For other cereal crops, including rice or maize, such syntheses are unavailable. Here, we conducted a global meta-analysis on published articles reporting AM fungal effects on grain yields of seven cereal crops: corn, wheat, rice, barley, sorghum, millet and oat. We used articles in Chinese as well as English to increase coverage significantly. We constructed three datasets (Fig. 1a): the ‘whole’ dataset including all studies irrespective of type of study (i.e. laboratory and field); the ‘field’ dataset comprising only field studies and the ‘field-inoculation’ dataset including only AM fungal inoculation studies conducted in field. Using these studies carried out on five continents (Fig. 1b), we aimed at addressing the following questions:
- What is the overall and crop plant specific AM fungal effect on grain yields?
- Is the AM fungal effect on grain yields related to improved plant nutrition?
- How do biotic and abiotic factors and their interaction modulate the AM fungal effect on grain yields?
Materials and Methods
Literature search
This meta-analysis followed the PRISMA guidelines (Moher, 2009) (Supporting Information Fig. S1). We focused on cereal crops with exceptional importance for human nutrition that are grown worldwide (Awika, 2011): corn, wheat, rice, barley, sorghum, millet and oat. We searched both Web of Knowledge (WOK, http://apps.webofknowledge.com/) and China National Knowledge infrastructure database (CNKI, http://www.cnki.net/) accessed 4 July 2017 and updated on 12 November 2017 to collect any newly published articles. Our search terms were: ‘Mycorrhiza* AND (barley OR hordeum OR maize OR corn OR zea OR wheat OR triticum OR durum OR rye OR secale OR oat OR avena OR rice OR oryza OR sorghum OR millet) AND (yield* OR grain* OR seed* OR kernel* OR panicle*)’. Studies matching our criteria were included in the compilation of our dataset. The articles had to be original research articles; report grain yields data and involve an AM fungi treatment and a corresponding control (Table S1) while other agricultural practices did not differ between the two groups. We found no significant differences in the effect sizes between Chinese and English publications (Table S2) so that we combined all studies to enhance statistical power. In total, 168 studies complied with the inclusion criteria and formed the ‘whole’ dataset (Notes S1, S2); 97 studies formed the ‘field’ dataset and 70 studies comprised the ‘field-inoculation’ dataset (Fig. 1a).
From 168 articles, we extracted information on yield (Table S3), variance, sample size (N) and 23 moderator variables covering plant, microorganisms, nutrition, soil and experiment-related factors of the studies. If data were only available via graphs, we used free digitizing software for data extraction (GetData Graph Digitizer v.2.20).
Effect size
We used the natural log response ratio (rr) as effect size in our analyses to represent the AM-mediated effect on grain yields (rrY). For the calculations, we followed the function: rrY = log(YM/YC), with YM denoting yield of arbuscular mycorrhizal plants and YC indicating yield of the corresponding control plants. A positive rrY indicated a beneficial AM fungal effect on yield, while negative values represented a detrimental effect.
The effect size representing the AM fungal effect on plant nutrition was calculated in the same way and effect sizes of plant nutrition (pn) nitrogen (N), phosphorus (P) and potassium (K), were denoted as pnN, pnP and pnK. We calculated the effect sizes in R v.3.4.1 (R Core Team, 2017) with the package ‘metafor’ (Viechtbauer, 2010) by integrating control and treatment means, the corresponding variances (SD) and the sample size (N). When only standard errors (SE) were presented, we calculated SD following the function: SD = SE × sqrt (N). When both SD and SE were unavailable, the median of the effect size variance calculated based on the ‘whole’ dataset was used as the surrogate for the missing variation (Shuster, 2011).
Moderator variables
For a more detailed investigation, we focused on five groups of moderator variables covering study system-, plant-, microbe-, nutrient- and experiment-related parameters for their potential to modulate the AM fungal effect on crop yield.
To define the study system for which our results apply, we screened the studies for: type of study (laboratory or field) where the experiments took place; and how the AM fungal treatment was established. Subsequently, we defined the two moderator variables: type of study and intervention.
Type of study had two levels lab and field. Experiments conducted in pots under controlled environmental conditions (e.g. glasshouse) were allocated to lab. Those carried out in field conditions were denoted field. In one study, potted test plants were grown in the open. Due to the higher degree of control on this growth system compared with the field situation, we included this special case in the level lab.
Intervention had three levels: inoculation, rotation and tillage. Inoculation denoted studies where AM fungal inoculum was applied or not to form an ‘AM fungi treatment’ or the ‘corresponding control’. This type of intervention allows the highest level of attribution of causality of effects. At rotation, the tested crops were rotated or intercropped with AM-host plants or nonhost plants (or a fallow stage as extreme form) for ‘AM fungi treatment’ or the ‘corresponding control’. At tillage, studies were included applying no-tillage or reduced tillage to create an ‘AM fungi treatment’ as the extraradical hyphae of AM fungi were not or only slightly damaged by ploughing. Conventional tillage was considered for the ‘corresponding control’ where the extraradical hyphae were potentially severely damaged. These last two types of interventions, while including AM fungal effects, also come with concurrent changes in other parameters, and thus attribution of causality is not as strong as in the inoculation case.
We chose these two moderators to examine if and how these fundamental aspects affect analysis outcomes, and to include the broadest possible suite of factors affecting AM fungal abundance (at a range of different levels of causal attribution to AM fungal effects). Studies conducted under controlled environmental conditions tend to have more pronounced effects compared with the field, for example (Lehmann et al., 2017).
To test the contribution of the plant partner to the AM fungal effect on crop yield, we chose the moderators carbon fixation type, crop plant and year of release.
Carbon fixation type had two levels C3 and C4. Wheat (Triticum aestivum L.), rice (Oryza sativa L.), oat (Avena sativa L.) and barley (Hordeum vulgare L.) were grouped in C3 and maize (Zea mays L.), sorghum (Sorghum bicolor L.) and millet (Setaria italica L.) in C4. For oat, we only found one article presenting data on two cultivars investigated under laboratory conditions; hence, data on oat were not available in the ‘field’ dataset and the ‘field-inoculation’. Similarly, data on millet was not available in ‘field-inoculation’ dataset.
Crop plant had seven levels: wheat, rice, barley, maize, sorghum, millet and oat. Data on oat were not available in the ‘field’ and ‘field-inoculation’ dataset, while data on millet were not available in the ‘field-inoculation’ dataset.
Year of release included three levels new, old and ancestor. The new group contained cultivars released after 1950; the old were released between 1900 and 1950. The ancestor included all varieties released before 1900 as well as the wild crop relatives and landraces, for which no year of release exists. This followed the rationales stated by Hetrick et al. (1992). Before 1900, plant breeding was more likely to be the products of anthropogenic selection events with the criteria to improve traits such as panicle size and taste. After 1900, cultivars were mainly from hybridization of inbred lines, while after 1950, cultivars bred were based on the high yielding varieties, such as Norin-10-based semi-dwarfs. Our ‘field’ and ‘field-inoculation’ datasets lost the old level.
Among the microbe-related moderators, we tested for the effect of the AM fungal inoculum and the application of additional microbes on the AM fungal effect on crop yields.
AM fungal inoculum had two levels: single and mixed (i.e. artificially composed assemblages). There were insufficient data available to investigate species richness in AM fungal inoculum. Single species inoculum was dominated by the species Funneliformis mosseae (synonym Glomus mosseae).
Microbial supplement had two levels: yes and no. Yes denoted that besides AM fungi other soil microorganisms (e.g. bacteria) were present, while in studies with no microbial supplement AM fungi were the only alive soil microorganisms in the experimental system (at least initially). Trials without substrate sterilization were allocated to yes due to the existence of indigenous microorganisms. Addition of a microbial wash or filtrate or inoculation of plant growth promoting rhizobacteria, phosphate-solubilizing bacteria and nitrogen-fixing bacteria were also allocated to yes.
Nutrient status of the soil and the plant is one main driver determining the symbiosis outcome between AM fungi and associated plants. There, we investigated here the impact of the macronutrients nitrogen, phosphorus and potassium on AM fungal effects on crop yields. We evaluate this effect on the level of fertilizer application, plant-available nutrients in the soil and nutrient concentration in vegetative plant tissue. Additionally, we tested for the importance of micronutrient applications.
Nitrogen, phosphorus and potassium fertilization (fN, fP and fK) had two levels: no and yes. Typical forms of additional nutrient application were, for example, addition of Hoagland solution or compound fertilizers including chemical (e.g. nitrogen–phosphorus–potassium compound fertilizer) and organic fertilizers (e.g. manure, compost, fly ash and crop residue). Specific nutrient fertilizer compounds for fN were ammonium, urea, nitrate and nitrochalk, for fP were phosphogypsum, superphosphate, phosphate, superphosphate and rock-phosphate, and for fK were potash, potassium chloride, potassium sulfate and potassium nitrate.
Plant available (pa) soil nutrient concentrations (in mg kg−1) were established for nitrogen, phosphorus and potassium, hereafter called paN, paP and paK. They all had two levels: low and high. The ranges of paN, paP and paK were 0.07–430, 0.26–258, 0.11–3420 mg kg−1, respectively. We grouped the data based on the median value of the available data. The medians of paN, paP and paK were 26.64, 11.60 and 143 mg kg−1, respectively. Studies with available nutrient concentration below the median value formed low; those above formed high. We pooled the data indicating ‘available nitrogen’ or ‘hydrolysable nitrogen’ or ‘mineralizable nitrogen’ or ‘mineral nitrogen’. When only NH4+-N and/or NO3−-N concentration of soil were measured, we extracted these data as paN (we use the sum if both NH4+-N and NO3−-N concentration were available). When only total N was presented, paN was estimated via the linear relationship between paN and total N (Guinto & Catto, 2016). In terms of paP, we only extracted Olsen phosphorus. For paK, we extracted data indicating ‘available potassium’ or ‘exchangeable potassium’. For missing paK data, we estimated it based on the linear relationship between total potassium and paK in agricultural fields (Rayment, 2013).
Plant nutrient (pn) concentrations (% of biomass) were continuous variables. The resulting effect sizes of plant nitrogen, phosphorus and potassium concentration (pnN, pnP and pnK) represented the AM fungal effect on nutrient status of vegetative tissues, such as leaves and tillers, rather than reproductive tissues (e.g. grains, spikes, panicles, cobs). The rationale for this decision was that the nutrient status of vegetative tissues indicates plant health state while that of the reproductive tissues suggested food quality. In addition, here, we wanted to focus on plant health status. If data for multiple growth stages were available, only the data close to or at maturity were collected, as the most data were only available for this growth stage.
Micronutrient application had two levels: yes and no. Added micronutrients were magnesium, zinc, sulfur, calcium, boron, iron and manganese. Studies in which at least one of these micronutrients was added were grouped into the level yes. Micronutrients were applied by nutrient solution (e.g. Hoagland) while some were added separately. Additionally, studies with organic fertilizers (e.g. manure, compost, fly ash and crop residue) formed the level of yes as these organic fertilizers also contained or increase the availability of soil micronutrients (Provin et al., 2008; Channabasava et al., 2015).
Experiment-related variables analyzed here comprised soil parameters but also the influence of stressors, growing season and substrate sterilization on AM fungal effects on crop yield.
Soil texture had two levels: sandy and not sandy. The data were classified following the USDA Natural Resources Conservation Service soil taxonomy (http://soils.usda.gov). All soils with a sand content > 50% (sand, loamy sand, sandy loam, sandy clay loam and sandy clay) were allocated to the level sandy, soils with other textures formed the level not sandy.
Soil pH had the three levels acidic, neutral and alkaline. The classification was conducted based on the USDA criteria (http://soils.usda.gov) where acidic ≤ 6.5, neutral 6.6–7.3 and alkaline ≥ 7.4. There were three reagent used for soil pH measurement: water, CaCl2 and KCl. They were converted following Brennan & Bolland (1998) and Kabala et al. (2016). We assumed water was used if there was no information on the reagent reported (Lehmann & Rillig, 2015) as this method is commonly used in agriculture.
Soil organic matter (SOM) had three levels low (SOM: < 1.38%), medium (SOM: 1.38–2.59%) and high (SOM: > 2.59%). Where the content of organic matter was not available, we estimated it by converting organic carbon via the van Bemmelen coefficient (organic matter = organic carbon × 1.724) (Perie & Ouimet, 2008).
Stress had two levels yes and no. Studies with soil compaction, heavy metal pollution, high level of soil salinity, drought, heat and cold application, long-term flooding and addition of nematodes were allocated to yes.
Growing season had two levels: one and more. For experiments that lasted longer than one growing seasons, we collected data only for the last season.
Substrate sterilization had two levels yes and no. When in studies the soil was pretreated to eliminate indigenous AM fungi, for example by fumigation, autoclaving or pasteurization, they were grouped in level yes.
Datasets
The data collection and data table construction were conducted by two separate investigators and checked by a third independent investigator. Following the analyses on the moderator type of study in ‘whole’ dataset, we constructed the ‘field’ dataset excluding all studies conducted in the laboratory. After the analyses on the moderator intervention, we built the ‘field-inoculation’ dataset from the ‘field’ dataset excluding all studies allocated to rotation and tillage of the intervention moderator.
The majority of the studies included in our datasets comprised more than one trial. This violated the traditional assumption of study independence in meta-analysis (Stevens & Taylor, 2009). To deal with this nonindependence issue, we conducted two types of corrections as applied in previous publications (Lehmann et al., 2014): (1) For studies presenting trials with multiple treatment data and one common control, we followed the approach by Lajeunesse (2016) implemented in the ‘metagear’ package. For this common control correction, a variance-covariance matrix was aligned with the corresponding effect size values to generate a corrected and merged single effect size value and effect size variance. (2) For studies presenting multiple trials, we combined trials into one effect size value following a fixed effects meta-analytical approach. This latter approach was not applicable for trials differing in factors assigned to moderator variables (Lehmann et al., 2017).
Statistics
All the analyses were conducted in R v.3.4.1 (R Core Team, 2017) with the ‘metafor’ package (Viechtbauer, 2010). The overall effect of AM fungi on grain yields was estimated with the rma.uni() function by fitting a random effects model with a restricted maximum-likelihood method. We assessed the heterogeneity in our effect sizes by performing I2 statistics (Cooper et al., 2009). For the categorical variables, we investigated whether the between-level differences of each moderator were significant (only levels including at least two studies were presented). For the continuous variables, the random probability values of slopes were estimated. For each moderator level, rrY (or slope) was considered different from zero based on P < 0.05. For each categorical moderator, the P-values were used to indicate the differences between/among the cut-off levels of this moderator. Since the data were not normally distributed we applied permutation and bootstrapping approaches (3999 iterations) implemented in the ‘metafor’ and ‘boot’ packages (Davison & Hinkley, 1997; Knapp & Hartung, 2003; Canty & Ripley, 2017) to estimate P-values and 95% bias-corrected confidence intervals (CIs). To avoid misleading graphs, we did not depict extreme values in the graphs but we included them in the analysis.
Sensitivity analysis
To verify our analysis outcomes, we investigated the presence of potential confounding factors by using subset analyses (Fig. S2), publication bias (Fig. S3) (Sterne & Egger, 2001) and any disproportional impact of studies (Copas & Shi, 2000) on our analysis outcomes (Figs S4–S7). All the graphs were created based on the outcomes of sensitivity analysis (we removed the studies with potential confounding effects and/or disproportional impacts).
Results
Considering all 168 studies, we found an overall positive AM fungal effect on grain yield (21% increase, CI = (16–26%), Fig. 2a) in ‘whole’ dataset. When differentiating field and lab studies, we found higher effect size estimates for lab studies (Fig. 2a), which was true for all five investigated crops. Including ‘lab’ studies in the analyses would lead to an overestimation bias; hence, we excluded lab studies from the following analyses.
In the next step, we tested in ‘field’ dataset the importance of the AM fungal intervention (i.e. inoculation, rotation and tillage). In inoculation studies, the AM fungal treatment and control were directly manipulated by applying AM fungal inoculum while in the other two intervention types, AM fungi were indirectly manipulated via management practices. We found that inoculation studies had clearly positive rrY while the other two had neutral rrY (Fig. 2b; Table S4, except corn under rotation).
To estimate the AM fungal effects on grain yield with greatest degree of confidence in causality, we focused on field studies using inoculation interventions; hereafter, called the ‘field-inoculation’ dataset. For this final dataset, we found that the effect size rrY was not uniform in either direction or size and ranged from 17% decrease to eight times increase (the extreme value was due to an exceptionally small yield of a control in arsenic contaminated soils, Fig. 3a). With 77% of all trials having positive rrY values (Fig. S8), the overall positive effect of AM fungal effect on grain yields in this dataset was 0.15 (CI = (0.09 - 0.19); Fig. 3b), translating to a 16% increase (CI = (9–21%)).
Importance of plant and AM fungi
To investigate the context of this positive overall effect of AM fungi on crop yield, we first focused on both plant and fungal symbiosis partners. We tested carbon fixation type and found that both C3 and C4 plants yielded positive and comparable effect sizes (Fig. 3b). For crop plant, we found significant differences: positive for corn (13% increase, CI = (2.5–25%)), rice (17% increase, CI = (11–23%)), wheat (17% increase, CI = (13–22%)) and sorghum (37% increase, CI = (5.3–81%)), but only neutral for barley (0.94% increase, CI = (−5.2% to 7%), Fig. 3b). For the two latter crops, only a limited number of studies and trials was available resulting in a large variability.
The moderator year of release had a significant impact on the effect size rrY: the AM fungal effect was less pronounced in new cultivars than in ancestor varieties (Fig. 3c). This was also evident for corn and wheat, for which the most data were available even though the difference was not significant.
On the microbial side, we did not detect any significant impact of either AM fungal inoculum or microbial supplement (Fig. 3d,e). In spite of this, the application of mixed species AM fungal inoculum tended to have higher rrY than their single species counterparts, especially where substrate was not sterilized (Table S5). Furthermore, the addition of soil microbes (e.g. Rhizobia) did not alter rrY even in studies with sterilized substrate (Table S5).
Impact of nutritional moderators
Having evaluated the importance of the biotic components, we next examined the impact of macronutrients on the AM fungal effect on crop yield. For this, we first tested how fertilization with the three macronutrients N, P and K affected the effect size rrY and found no significant differences (Fig. 4a); the application of fertilizer did not alter the AM fungal contribution to crop yield. The same holds true on crop plant level (Table S6). In the next step, we investigated if the concentrations of plant available N, P and K in the soil affect the symbiosis. No significant effect on rrY was detected (Fig. 4b; Table S6). Additionally, when analyzing the impact of fertilizer application on low and high plant available nutrient content soils, respectively, we did not find any links between these two moderators (Fig. 3c). However, the highest rrY was detected in low N and P soils, which were also not fertilized with N or P. In the following step, we tested how the nutrient concentration in the vegetative plant tissue modulates the AM fungal effects on crop yield but we did not find a significant effect for any of the three macronutrients (Fig. 4d–f). In the last step, we tested for the importance of micronutrient supplements. The application of micronutrients did not significantly change the AM fungal effect on crop yields (Table S7).
Influence of experiment-related moderators
The experiment conditions including soil parameters and treatments are known for their potential to modulate AM fungal – plant interactions. Therefore, we investigated six additional moderators to evaluate their importance. We found that soil texture and soil organic matter had no significant effect on rrY while rrY in alkaline soils was lower than those from acidic and neutral soils (Fig. 5a–c). The analyses on the impact of substrate sterilization, stressors and the number of growing seasons did not reveal any significant effects (Fig. 5d–f).
Discussion
We present a global-scale meta-analysis investigating the AM fungal effect on grain yield for field studies where AM fungal inoculum was applied with an associated control. We found that AM fungi increased grain yields of major cereal crops: wheat, maize, rice and sorghum. This is in correspondence with other meta-analyses (Lekberg & Koide, 2005; Pellegrino et al., 2015) focusing on crop yield. Our meta-analysis gives unprecedented evidence for the potential of AM fungi to improve yield production of four of the major staple food crops feeding the world. By combining data from English and Chinese language articles, we extend the statistical power of our analyses compared to previously published meta-analyses. Additionally, we included only field data with clearly defined AM fungal inoculation treatments in our detailed analyses, allowing us to make unbiased inferences for the situation that is most agronomically relevant.
Influence of experimental systems
Our analysis clearly shows that laboratory data need to be backed up with field trials to estimate true effect sizes, as discussed in Ryan & Graham (2018). In their review, Ryan & Graham (2018) reached the conclusion that AM fungi were not worth managing in terms of crop yield, which clearly contrasts with our results. This is mainly due to the difference in the intervention types (i.e. how to establish AM fungal treatment and the corresponding control) between the two studies. Their review mainly focused on manipulating indigenous AM fungi by crop management, such as rotation with mycorrhizal and nonmycorrhizal host (including plant-free fallow as the extreme form); we considered not only rotation and fallow but also two other intervention types: inoculation with AM fungi or not and reduced tillage or not. When only rotation or tillage was considered, we arrived at the same conclusion as Ryan & Graham (2018); however, we support the opposite conclusion when only including inoculation studies. In addition to the difference in intervention types, the targeted crops were not the same between their review and our study: the studies providing evidence in their review considered the yield of wheat, flax and field peas while our study demonstrated the positive effect of AM fungi on yields of wheat, rice, corn and sorghum, drawing on the Chinese and English language literature. We also emphasize that inoculation studies, while not necessarily of direct agronomic relevance, as mentioned in Ryan & Graham (2018), serve to establish a solid baseline for the causal link between AM fungal presence and any yield effects. It is not helpful to ignore these studies when studying effects of AM fungi on crop yield. Considering all these points, we feel that our meta-analysis directly invalidates the major conclusion of Ryan & Graham (2018), namely that it is not worth managing for AM fungi.
As inoculation with AM fungal inoculum led to higher increase in AM fungal colonization (29%) than other agriculture practices (20% by shortened fallow and 7% by reduced tillage 7%) (Lekberg & Koide, 2005), the data using inoculation might carry higher likelihood to attribute causality to AM fungal effects, compared with rotation, intercropping or reduced tillage. Taking the spread of nonnative AM fungal genotypes into consideration, inoculation with nonnative AM fungal inocula is not recommended unless there is robust evidence showing that they are noninvasive fungi. As economic constraints also come into play, inoculation with native AM fungal inoculum requires more data on the cost of AM fungal inoculum.
Importance of biological moderators
We did not find detectable differences in rrY between C3 and C4 plants while we found that crops behaved differently within each group, and we found this difference in C3 plants and cross the two groups (Fig. 3a). Specifically, we found that corn, rice, sorghum and wheat had pronounced positive rrY while barley had neutral rrY (Fig. 3a). Differences in rrY among the crops suggested that novel approaches supporting AM fungi should be encouraged in corn, wheat and rice production and highlight the need for focused data collection for barley, oat and millet.
It is important to know whether agricultural and breeding practices during domestication impact on the effect of AM fungi on grain yield of cereal plants. In one study in which only wheat was involved, Hetrick et al. (1992) found that in terms of biomass, new cultivars (released after 1950) had lower reliance on AM fungi than their ancestors (released before 1900) and old wheat (released between 1900 and 1950). However, a meta-analysis covering cereals, vegetables and legumes found no difference among these groups (Lehmann et al., 2012). We found that new crops had lower rrY than their ancestor relatives (Fig. 3b), indicating the negative impact of domestication on rrY. Therefore, modern crop breeding should take grain yield response to AM fungi into consideration to achieve higher grain yields (Sawers et al., 2008).
We hypothesized that the mixed species AM fungal inoculum might have higher rrY. However, we did not find such influence no matter whether substrate was sterilized or not (Table S5). Numerous explanations are possible for this pattern, including failure of different AM fungal species to establish in sterilized substrates, or increased competition among AM fungi for carbon (Liu et al., 2015). Alternatively, species of AM fungi were not always functionally complementary with respect to grain yield. We also hypothesized other soil microbes would significantly increase rrY. Our results did not support this hypothesis (Fig. 3) even when we analyzed a subset only including studies with substrate sterilized. Reasons for this are unclear, and more data on the composition of other soil microbes would be needed to shed light on this pattern.
Impact of nutritional moderators
Fertilization is common in crop production and availability of nitrogen, phosphorus and potassium in soil varied widely. In our study, none of these factors had an impact on rrY (Fig. 4). We hypothesized the positive effect of AM fungi on grain yields resulted from the AM-mediated improvement in plant nutrient concentration (% of biomass). Surprisingly, we did not find any correlations between rrY and pnN, pnP and pnK (Fig. 4), providing no evidence for this hypothesis. One likely reason might be that AM fungi increased yields by allocating more nutrients to the reproductive tissues (Zhang et al., 2014), rather than increasing overall nutrient concentrations. It also might be that the AM fungal effect on yield results from their nonnutritional roles: such as stress alleviation (Chandrasekaran et al., 2014; Jayne & Quigley, 2014), pathogen protection (Veresoglou & Rillig, 2012) and weed control (Meng et al., 2001; Veiga et al., 2011).
Influence of experiment-related moderators
In our study, soil texture and organic matter content, stress, growing season and substrate sterilization (Fig. 5) did not alter rrY, indicating that studies conducted in soils with different texture and/or organic matter content, under stress, for one growing season and with sterilized substrate also provide valuable information. In addition, we expected that in the nonneutral pH range, AM fungi would perform better as they could reduce the stress of immobilized P caused by a strong acidic or alkaline pH (Cartmill et al., 2007, 2008; Cardarelli et al., 2010). We found that alkaline soil had lower, rather than higher rrY than the other two soils (Fig. 5). This may be due to that the way in which AM fungi improved grain yields was not to enhance plant P nutrition, supported by the nonsignificant correlation between the effect size rrY and pnP (Fig. 4).
In conclusion, we found 168 studies that quantitatively addressed whether AM fungi enhanced grain yields of cereal crops with exceptional importance for human food and grown worldwide (Awika, 2011): corn, wheat, rice, barley, sorghum, millet and oat. Our results show that AM fungal effect on grain yield was less pronounced in field and noninoculation studies. AM fungi increased grain yields by 16% based on field studies where AM fungal inoculation was conducted to establish AM treatment. This effect of corn, rice, sorghum and wheat was clearly positive; that of barley was neutral (more data are required to validate these patterns for millet and oat). Additionally, this effect was lower for crops released after 1950 compared with those released before 1900 and also lower in alkaline soils. Other biotic and abiotic factors were not likely to alter this effect and there was no clear evidence showing that this AM fungal effect on grain yield was due to plant nutrient improvement.
Given that AM fungal effects in our dataset with unprecedented coverage were overwhelmingly positive (77% of trials in our ‘field-inoculation’ dataset), our analysis suggests that managing for AM fungi in agroecosystems is worthwhile even for cereal crops, contrary to conclusions of Ryan & Graham (2018). This is in addition to other potential benefits of AM fungi on agricultural soils, including soil aggregation (Lehmann et al., 2017). We suggest renewed efforts into finding the causes behind negative or neutral effects of the symbionts, which we also identified here.
Acknowledgements
This work was supported by Natural Science Foundation of Jiangsu Province (grant no. BK20160689) and the Green Talents Research Stay of the German Federal Ministry of Education and Research. MCR acknowledges funding from the German Federal Ministry of Education and Research for the BonaRes project INPLAMINT. We would like to thank Wenfei Yun, Haiyang Xu, Li Zhang, Sijie Wang for help with data collection.
Author contributions
SZ and MCR planned and designed the research. SZ, AL and ZY conducted literature search and data collection; SZ, AL and WZ analyzed data; SZ wrote the manuscript.
