Do inoculated microbial consortia perform better than single strains in living soil? A meta-analysis

Microbial consortium inoculation has been proposed as a natural-based strategy to safeguard multiple ecosystem services. Still, its empirical effects and comparisons to single-species inoculation have yet to be systematically quantified. In this global meta-analysis of 51 live-soil studies, we compared the impact (mean and variability) of single-species and consortium inoculations on biofertilization and bioremediation. Our results showed that both single-species and consortium inoculations increased plant growth by 29% and 48%, respectively, and pollution remediation by 48% and 80%, respectively, compared with non-inoculated treatments. We revealed the potential mechanisms contributing to the effectiveness of consortium inoculation, which are associated with the diversity of inoculants and the synergistic effect between frequently used inoculums (e.g., Bacillus and Pseudomonas). Despite a reduction in efficacy in field settings compared to greenhouse results, consortium inoculation had a more significant overall advantage under various conditions. We recommend increasing original soil organic matter, N, and P content and regulating soil pH to 6-7 to achieve a better inoculation effect. Overall, these findings support the use of microbial consortia for improved biofertilization and bioremediation in living soil and suggest perspectives for constructing and inoculating beneficial microbial consortia.

inoculations on the relative variability of different beneficial functions, we calculated 169 the natural logarithm of the ratio between the coefficients of variation (lnCVR) using 170 , X is the mean value, SE is the standard 171 error, t is the treatment applied with single inoculant or consortium, and c is the 172 treatment without inoculum, respectively (Nakagawa et al., 2015). We obtained 796 173 effect sizes for both lnRR and lnCVR. The data set focusing on biofertilization 174 consisted of 586 effect sizes from 45 studies, and on bioremediation consisted of 210 175 effect sizes from 15 studies. 176

Meta-analyses 177
The meta-analytic (intercept) model was fitted to assess the overall effect of 178 inoculation in the absence of moderators. We then used inoculant type (i.e., single 179 inoculant and consortium) as the main moderator to compare the difference between 180 single inoculant and consortium in effect sizes. A meta-regression model was 181 performed to investigate whether the inoculant species of consortium and inoculation 182 duration affect the inoculant functioning. Pairwise comparisons of subgroups were 183 conducted to assess the moderating effects of experiment types, inoculation methods, 184 and plant/pollutant types. Statistical significance compared to the reference was 185 assumed when 95% confidence intervals (95% CIs) did not span zero. The differences 186 between subgroups were estimated by Tukey's HSD comparisons using the ghlt 187 function in "multcomp" package (Hothorn et al., 2008). The heterogeneity in each 188 model was assessed using I 2 (Higgins et al., 2003)

and provided in Supplementary 189
Tables. The index R 2 , describing the ratio of explained variance to the total variance, 190 was calculated for each model using the r2_ml from the "orchard" package 191 (Nakagawa et al., 2021) and provided in Supplementary Tables. 192 All models were run with the chosen random effect structure using the R 193 package "metafor" (Version 3.8.1) for biofertilization and bioremediation data 194 separately. The "orchard" package (Nakagawa et al., 2021) was used for visualization . 195 The following sources of non-independence were identified and considered: species 196 effects (e.g., inoculants within experimental cases and studies), case effects (e.g., 197 experimental cases within studies), study effects (effect sizes calculated from the 198 same survey), pairwise comparisons for effect size calculations (effect size ID;199 variability in the true effects within studies). The final random effect structure was set 200 as study ID, case ID, species ID, and effect size ID after comparing AICc values 201 among models. 202 For testing publication bias in our intercept models, we used the funnel plot 203 based on the meta-regression residuals (Doleman et al., 2020)

and the conventional 204
Egger's regression test (Sterne et al., 2001). The former was performed by fitting 205 standard error as a unique moderator (Nakagawa et al., 2022). The results of 206 publication bias are provided in Supplementary Fig. S3 and Table S1. We also tested 207 the time-lag bias using meta-regression analysis to assess the relationship between 208 publication year and effect size (Supplementary Fig. S4 and Table S2). All analyses 209 were conducted using R (version 4.2.1). 210 211

Results 212
Our dataset includes globally distributed empirical studies conducted in living soil. 213 These studies were conducted in Asia, Europe, North Africa, North America, and 214 South America (Fig. 1, 51 studies and 796 effect sizes). Most studies focused on 215 11 bioremediation (N = 210, 26.4%). By using the non-inoculated treatment as the 217 reference, the counts of effect size were 358 (45.0%) and 228 (28.6%) for 218 biofertilization and 149 (18.7%) and 61 (7.7%) for bioremediation under the single-219 species inoculant and consortium treatments, respectively. 220

Fig. 2. The beneficial effect of bacterial inoculants on biofertilization and 240
bioremediation. The overall effects (A, B) and comparisons between single-species 241 and consortium inoculants (C, D) were estimated. Blue points in the central represent 242 estimated means, thick bars represent 95% confidence intervals, and thin bars 243 represent 95% prediction intervals. Each background point is an effect size (lnRR), 244 and its size is scaled by the precision of that estimate (1/SE). Blue asterisks denote the 245 rejection of the hypothesis that the estimated mean equals 0 (p < 0.001), indicating a 246 significant effect compared to non-inoculant treatments. Black asterisks suggest a 247 significant difference in estimated means between the two groups (***, Tukey's HSD, 248 p < 0.001). 249 250

Moderators influencing the inoculation effect 260
For biofertilization, the application of bacterial consortium had a higher positive 261 impact on plant growth, whether tested in pot/greenhouse (by 51.7%) or field (by 262 37.7%) conditions, compared to treatments applied with single-species inoculum (by 263 30.4% and 21.1% for pot and field settings, respectively) ( Fig. 3A and Table S7). The 264 estimated lnRR under pot/greenhouse and field conditions inoculants showed no 265 statistical difference for single species (p > 0.05). Still, they were significantly 266 different for the consortium, and the lnRR was lower under field conditions (lnRR = 267 0.42 and 0.32 for pot and field settings, respectively, p < 0.05) ( Fig. 3A and Table S7). 268 In contrast, no significant difference was observed for bioremediation between 269 experiment types (i.e., microcosm and pot/greenhouse) (p > 0.05, Fig. 3B and Table  270 S8). We found the lnCVR was significantly less than 0 only in pot/greenhouse 271 conditions despite using single-species and consortium for biofertilization (p < 0.05, 272 The inoculation method (i.e., soil inoculation and seed/seedling inoculation) 288 had no significant influence on the estimated lnRR applying either single-species or 289 consortium inoculum for biofertilization and bioremediation (p > 0.05, Fig. 4A and 290 4B, Table S9 and S10). However, a larger mean effect size was observed in 291 consortium application treatment than single species treatment for biofertilization and 292 bioremediation, regardless of inoculation methods ( Fig. 4A and 4B). 293 Plant type was evaluated as an essential moderator of the inoculum's function 294 for biofertilization (Fig. 4C). Specifically, the lower estimated lnRR of cereal plants 295 (lnRR = 0.19 and 0.32) was found compared to non-cereal plants (lnRR = 0.33 and 296 0.51) for single-species inoculations and consortium inoculations, respectively (p < 297 0.05, Fig. 4C and Table S11). For bioremediation, the effect sizes showed no 298 difference between pollutant types (i.e., metal and organic pollutants) for both single-299 species inoculations and consortium inoculations (p > 0.05, Fig. 4D and Table S12). 300 Furthermore, the advantage of applying consortium was observed for both 301 biofertilization and bioremediation, regardless of plant or pollutant types ( Fig. 4 and 302 Table S12). was grouped into cereal and non-cereal (Non-C) plants, and the pollutant type was 307 divided into heavy metal and organic pollutants. Letters S and C for subgroups 308 represent single-species and consortium inoculations, respectively. Asterisks denote 309 the rejection of the hypothesis that the estimated mean equals 0 (p < 0.001), indicating 310 a significant effect compared to non-inoculant treatments. Different letters on the right 311 suggest significant differences between subgroups (p < 0.05, Tukey's HSD). 312

313
The impact of inoculation time (i.e., experimental duration) on effect sizes was 314 further investigated. Our result showed that experimental duration had no significant 315 correlation with the effect size for biofertilization, although the estimate was higher 316 after consortium applications (p > 0.05, Fig. 5 and Table S13). For bioremediation, a 317 weak impact on inoculation effects after the one-off inoculation was observed for both 318 single-species (estimate = 0.0005, 95% CI = -0.0003-0.0014, t value = 2.11, p = 0.027) 319 and consortium (estimate = 0.0017, 95% CI = 0.0003-0.0032, t value = 2.23, p = 320 0.021) treatments ( Fig. 5 and Table S13). 321 The properties of the soil used for inoculation experiments affected inoculation 331 effectiveness (Fig. 6). The results showed that the relationship between soil pH and 332 lnRR was significantly fitted with a polynomial curve (p = 0.0031) in the consortium 333 treatment for biofertilization, but not in the single-strains treatment (p = 0.35). For 334 bioremediation, original soil pH was negatively correlated with lnRR for both single-335 strain and consortium treatments (p = 0.00055 and 0.05, respectively). The content of 336 organic matter in soil was significantly and positively correlated with lnRR in two 337 treatments for biofertilization (p < 0.01). However, these relationships were not 338 significant for bioremediation (p > 0.05). 339 The impacts of soil texture and available nutrients, such as nitrogen (N), 340 phosphorus (P), and potassium (K), on effect sizes, were assessed only for 341 biofertilization, since limited data was obtained from bioremediation cases (Fig. 6). To explore the potential synergistic effect of inoculants, we first summarize 380 the type of inoculants (genus level) used in constructing consortia. The two frequently 381 used inoculants were Bacillus and Pseudomonas, used in 26 and 19 of 51 studies (Fig.  382   7B). Compared to these genera, other species were relatively less studied, such as 383 Microbacterium (8)

Inoculation effect influenced by different regulators 429
Our second hypothesis that environmental conditions influence the inoculation effect 430 was supported by our results. We found that the experimental condition is an essential 431 factor determining the success of the inoculation, where a significantly larger effect 432 size the decrease in the plant growth variation were observed in the pot/greenhouse in both pot/greenhouse and field conditions. Therefore, consortium inoculation can 443 perform better than single strains, although its efficacy may be diminished in the field 444 compared to greenhouse results. 445 In terms of inoculation methods, we found no significant differences in effect 446 sizes between soil inoculation and seed/seedling inoculation either for single-species 447 or consortium treatments, which was consistent with the results reported by meta-448 analysis focusing on single-species inoculations (Rubin et al., 2017). However, this 449 does not mean that the inoculation method can be freely selected. On the contrary, it 450 is highly possible that the appropriate protocol had been determined when each study 451 24 was carried out, as the protocol for soil inoculation depends on the inoculation 452 purpose and the type of inoculants (Fukami et al., 2016;O'Callaghan, 2016). Besides, 453 another recent meta-analysis suggested that combining multiple inoculation 454 approaches (e.g., soil, seed, and foliar) and biological effectors, such as amino acids 455 and microorganisms, can achieve a better effect (Herrmann et al., 2022). Therefore, 456 developing microbial consortia that can successfully colonize various ecological 457 niches, such as bulk soil, rhizosphere, and even endosphere, may increase the scope 458 and effect of inoculants.