Strain identity effects contribute more toPseudomonascommunity functioning than strain interactions

9 Microbial communities can shape key ecological services, but the determinants of their functioning often 10 remain little understood. While traditional research predominantly focuses on effects related to species 11 identity (community composition and species richness), recent work increasingly explores the impact of 12 species interactions on community functioning. Here, we conducted experiments with replicated small 13 communities of fluorescent Pseudomonas bacteria to quantify the relative importance of strain identity 14 versus interaction effects on two important functions, community productivity and siderophore 15 production. By combining supernatant and competition assays with an established linear model method, 16 we show that both factors have significant effects on functioning, but identity effects generally outweigh 17 strain interaction effects. These results hold irrespective of whether strains interactions are inferred 18 statistically or approximated experimentally. Our results have implications for microbiome engineering, 19 as the success of approaches aiming to induce beneficial (probiotic) strain interactions will be sensitive to 20 strain identity effects in many communities. 21

siderophores, pyoverdine, Pseudomonas, social interactions, community functioning 22 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted June 9, 2024. ; https://doi.org/10.1101/2024.06.07.597923doi: bioRxiv preprint

23 Microbial communities are present in almost all natural and anthropogenic environments on earth, and 24 shape important ecological services from primary production to the decomposition of organic matter and 25 the fixation of greenhouse gases (Stockmann et al. 2013; Whitman et al. 1998). Given that many of these 26 services can be harnessed for biotechnological applications such as bioremediation and biofuel 27 production (Alper & Stephanopoulos 2009; Minty et al. 2013; Piccardi et al. 2019; Wang & Chu 2016) , 28 recent years have seen a massive surge in studies investigating how contributions of specific microbial 29 players and their interactions with one another affect different functions ( collective properties ) of 30 microbial communities (Bell et al. 2005; Figueiredo et al. 2022; Jones et al. 2021; Li et al. 2019; Ratzke et 31 al. 2020; Venail & Vives 2013). While this research has elucidated independent effects of various factors 32 on functions from productivity to invasion resistance, understanding their combined influence is crucial 33 to predicting and manipulating how microbial communities work (Gorter et al. 2020; Konopka 2009). 34 Traditionally, studies on factors shap ing community functioning mainly focus on the role of 35 community composition and species richness (Bell et al. 2005; Loreau & Hector 2001; Smith & Knapp 36 2003; Wardle et al. 1998). Community composition typically matters because some species may 37 contribute more to functioning than others (Bell et al. 2005; Wardle et al. 1998). Conversely, species 38 richness typically matters because higher species numbers increase the likelihood of including species 39 with different niche requirements or strong effects on functioning (Bell et al. 2005; Loreau & Hector 2001). 40 In both cases, effects on community functioning are ul timately driven by characteristics inherent to 41 specific community members (identity effects ). More recently, research has increasingly focused on 42 effects on functioning driven by interactions among community members (interaction effects; Fiegna et 43 al. 2015; Figueiredo et al. 2022; Gorter et al. 2020; Gu et al. 2020b; Li et al. 2019, 2022; Ratzke et al. 2020). 44 In microbes, these interactions are often mediated by secreted products – such as toxins, antibiotic -45 degrading enzymes , and iron-scavenging siderophores – and can have strong effects on community 46 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted June 9, 2024. ; https://doi.org/10.1101/2024.06.07.597923doi: bioRxiv preprint dynamics (Granato et al. 2019; Kramer et al. 2020b; West et al. 2007). While such interactions are clearly 47 relevant, it remains unclear how their impact on functioning compares to the impact of identity effects 48 because studies typically focus on only one of the two factors (Bell et al. 2009; Connolly et al. 2013). 49 Here, we tackle this knowledge gap by quantifying and comparing the impact of strain identity 50 and strain interactions on the functioning of small communities of fluorescent pseudomonads. 51 Pseudomonas is a diverse and widespread genus of γ-proteobacteria, occurring in soil and freshwater 52 ecosystems as well as in animal hosts (Cornelis 2010). Fluorescent pseudomonads produce and interact 53 through a broad range of secreted compounds including proteases, biosurfactants and the eponymous 54 fluorescent siderophore pyoverdine (Butaitė et al. 2018; Kramer et al. 2020a). This versatility has made 55 representatives such as P. aeruginosa and P. fluorescens important models for studying microbial 56 interactions early on (Brockhurst et al. 2007; Griffin et al. 2004; Kümmerli et al. 2009; Rainey & Rainey 57 2003). More recently, fluorescent pseudomonads have increasingly been used to study how social traits 58 affect community functioning, including productivity, plant protection, and invasion resistance (Becker et 59 al. 2012; Figueiredo et al. 2022; Hodgson et al. 2002; Hu et al. 2016; Jones et al. 2021). 60 In our study, we used 64 diverse soil and freshwater Pseudomonas strains belonging to four 61 phenotype classes to examine how strain identity and strain interactions jointly shape two community 62 functions: productivity and the production of iron-scavenging pyoverdines. Pyoverdines are a diverse 63 group of siderophores, and each specific pyoverdine can either promote the growth of strains possessing 64 matching uptake receptors or inhibit the growth of strains without these receptors (Kümmerli 2023). To 65 quantify the positive and negative interactions through pyoverdines and other secreted compounds, we 66 conducted supernatant feeding assays under different conditions. These assays allowed us to calculate 67 community-level interaction metrics. Next, we grew all strains in monocultures and all combinations of 68 two, three, or all four strains per community, and measured their productivity and pyoverdine production 69 over time. This allowed us to leverage an established linear model method (Bell et al. 2009) to statistically 70 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted June 9, 2024. ; https://doi.org/10.1101/2024.06.07.597923doi: bioRxiv preprint disentangle the effects of strain identity and strain interactions, and to compare the numerically derived 71 measure of strain interactions with our experimentally measured interaction metrics. Finally, we used our 72 statistical results to identify strains with strong effects on functioning, and tested whether strain 73 interactions had a consistent impact on functioning across communities . Overall, our results show that 74 although strain interactions can significantly affect the functioning of Pseudomonas communities, their 75 effects are generally outweighed by those of strain identity. 76

77 Strain selection. We drew strains from an established collection of 3 15 pseudomonads, isolated from 78 eight soil and eight pond samples (18-20 isolates per sample). Sampling and identification of these strains 79 are described elsewhere (Butaitė et al. 2017, 2018). Here, we selected a subset of 64 strains comprising 80 four strains per sample [hereafter: community] based on their production of pyoverdine and exo -81 proteases. While our experiments contrasted conditions where pyoverdine is important for growth with 82 conditions where it is not (see below), we used protease production as a proxy for multidimensional 83 phenotype differences (Kramer et al. 2020a) and thereby managed to obtain a highly diverse set of strains 84 (supplementary material; Figure S1, Table S1). Per community, we chose the most divergent strains within 85 the observed phenotype space, aiming to select (i) one strain producing pyoverdine and proteases, (ii) 86 one strain producing only pyoverdine, (iii) one strain producing only proteases, and (iv) one strain 87 producing neither pyoverdine nor proteases (supplementary methods; Figure S2). Our communities thus 88 each featured two strains producing pyoverdine at high levels (the double and the pyoverdine producer; 89 hereafter producers PVDPRO and PVD) and two strains producing no to little pyoverdine (the protease and 90 the non-producer; hereafter non-producers NONPRO and NON). Hereafter, we use ‘strain type’ to refer to 91 those four phenotype classes and ‘strain ID’ to refer to specific representatives of our 64 strains. 92 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted June 9, 2024. ; https://doi.org/10.1101/2024.06.07.597923doi: bioRxiv preprint Growth and siderophore production measurements. We quantified growth and pyoverdine production 93 of all strains under iron-limited and iron-rich conditions. First, we grew precultures in 24-well plates with 94 1.5ml lysogeny broth (LB) per well under static conditions for 48h. Subsequently, we washed cells in 0.85% 95 NaCl and measured their growth ([OD600]; optical density measured at 600 nm) against a 0.85% NaCl blank 96 using an Infinite M200 PRO microplate reader (Tecan, Männedorf, Switzerland). Next , we adjusted 97 precultures to an OD600 = 0.4 and inoculated 2 µL of each adjusted culture into 96-well plates containing 98 200 µL medium per well in fourfold replication. We used two variants of CAA medium (5g casamino acids, 99 1.18g K2HPO4·3H2O and 0.25g MgSO 4·7H2O per liter), an iron-limited variant supplemented with 25 mM 100 HEPES buffer, 20 mM NaHCO 3 and 100 µg/mL apo -transferrin (a strong iron -chelator), and an iron -rich 101 variant supplemented with 25 mM HEPES buffer and 40 µM FeCl3. After 24h of static incubation at 28°C, 102 we quantified growth [OD 600] and pyoverdine production ( [RFUpvd]; relative fluorescence units ; 103 excitation|emission at 400|460 nm) after 120s of vigorous shaking using the same microplate reader. 104 Supernatant assay. We explored interactions through secreted compounds under iron-limited and iron-105 rich conditions by exposing each strain to its own supernatant and to each supernatant collected from its 106 community members . We harvested supernatants from cultures grown in the above -described 107 experiment by spinning them through 96-well filter plates with a 3 μm glass fiber/0.2 μm Supor membrane 108 (AcroPrep Advance; Pall Corporation, Port Washington, USA) and then collecting the sterile supernatants 109 in 96-well plates. These plates were sealed and stored at -20°C until further use. Next, we gr ew another 110 set of precultures, washed and adjusted them as before, and subjected them to three treatments: (i) 111 SNlimited: 180 μL of iron-limited CAA supplemented with 20 μL of supernatant generated under iron-limited 112 conditions; (ii) SNrich: 180 μL of iron -rich CAA supplemented with 20 μL of supernatant generated under 113 iron-rich conditions; and (iii) SNcontrol: 180 μL of iron-limited or iron-rich CAA supplemented with 20 μL of 114 0.85% NaCl (mimicking spent medium). Strains were grown in threefold (SNcontrol) or fourfold (SNlimited and 115 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted June 9, 2024. ; https://doi.org/10.1101/2024.06.07.597923doi: bioRxiv preprint SNrich) replication. We measured growth [OD600] and pyoverdine production [RFUpvd] of each replicate after 116 24h and 48h of incubation at 28°C under static conditions. We calculated the effects of each supernatant 117 on the producer and each of its community members as growth effect s: GEtreatment = (SNtreatment/SNcontrol), 118 where SN treatment = SN limited or SN rich, with growth values being calculated as the median growth across 119 replicates. Values smaller and greater than one indicate growth inhibition and stimulation, respectively. 120 We calculated three summary measures of supernatant -based interactions t o link supernatant 121 effects to community functioning in direct interactions of multiple strains. Specifically, we calculated for 122 each combination of two, three or all four strains per community, (i) the mean absolute effect and (ii) the 123 proportion of positive effects, thereby separately capturing the strength and sign of supernatant effects, 124 respectively. Additionally, we calculated (iii) an ‘interaction score’ to incorporate information on both 125 strength and sign. To this end, w e first tested for each donor-receiver pair whether their reciprocal 126 supernatant effect s were positive, negative , or neutral (i.e., whether SNtreatment values differed from 127 SNcontrol values; Table S2). Subsequently, we categorized all pairwise interactions based on the effects that 128 the strains had on each other , resulting in six interaction types: mutual stimulation [+/+], one-way 129 stimulation [+/0], no effect [0/0], contrasting effects [+/-], one-way inhibition [0/-], and mutual inhibition 130 [-/-]. Finally, we calculated interaction scores by valuing all effects on other strains [stimulation = 1; neutral 131 effect = 0; inhibition = -1] and then calculating an average score across all interactions for each 132 combination of two, three, or all four strains per community. Interaction scores smaller and greater than 133 zero indicate that inhibitory and stimulatory interactions prevail, respectively. 134 Competitions. To be able to assess the effects of strain interactions and strain identity on community 135 functioning, we competed each strain against combinations of its community members under iron-limited 136 and iron-rich conditions, and quantified community productivity and total pyoverdine production over 137 time. Specifically, w e set up competition experiments for each of our 16 communities involving all 138 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted June 9, 2024. ; https://doi.org/10.1101/2024.06.07.597923doi: bioRxiv preprint combinations of two, three or all four community members, and included monocultures as controls (15 139 treatment conditions per community: 4x1 strain + 6x2 strains + 4x3 strains + 1x4 strains). We grew 140 precultures from freezer stocks in 50 ml Falcon tubes containing 5mL LB at 28°C under shaking conditions 141 (170 rpm). After 48h of incubation, we washed cells in 0.85% NaCl, measured the OD600 of each culture 142 against a 0.85% NaCl blank, and then adjusted strains to OD 600 = 0.2. Next, we assembled the mixes and 143 inoculated them at a starting density of OD600 = 0.01 either in 6-fold (4-strain competitions) or 5-fold (other 144 treatment conditions) replication into 96-well plates containing 190 µL of iron-limited or iron-rich medium 145 per well. We used a substitutive design, whereby overall starting density is constant across different 146 mixes, while individual strain density decreases when strain number increases (Figueiredo et al. 2022; 147 O’Brien et al. 2023). We incubated plates in a plate reader at 28°C under static conditions and measured 148 the productivity [OD600] and total pyoverdine production [RFUpvd] of each culture every 15min over 48h. 149 We used these measurements to calculate integrals of productivity and total pyoverdine production as 150 our primary measures of community functioning. For some analyses, we additionally calculated deviations 151 from expected productivity and pyoverdine production as DEVtrait = TVmix – mean(TVmono), where TVmono = 152 trait values of monocultures of strains in the mix. The DEV trait values indicate whether the trait value 153 (productivity or pyoverdine production) of a specific strain mix was lower or higher than expected based 154 on the trait values of the monocultures of the constituent strains. 155 Linear model method. To compare the effects of strain identity and strain interactions on functioning at 156 the community level, we used an established linear model (LM) method that partitions the variance in a 157 community-level trait between different factors of interest (Bell et al. 2009). Briefly, this method uses a 158 series of three LMs to sequentially account for (i) the influence of strain number (entered as a continuous 159 variable), (ii) strain identity (entered as the presence [categorical] of each strain in a particular strain 160 combination), and (iii) strain interactions (strain number, entered as categorical variable). While entering 161 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted June 9, 2024. ; https://doi.org/10.1101/2024.06.07.597923doi: bioRxiv preprint strain number as continuous variable accounts for a linear increase of functioning with strain number, 162 entering it subsequently as categorical variable tests for additional, nonlinear (non -additive) effects, 163 thereby providing a measure of strain interactions among strains (Bell et al. 2009). The first LM always 164 uses the focal trait of interest as a response, whereas subsequent LMs are fitted on the residuals extracted 165 from the respective previous model . Note that t he effects obtained for strain identity and strain 166 interactions are orthogonal and thus independent of the order in which the corresponding LMs are fitted 167 (Bell et al. 2009). We ran LMs separately for each community and experimental condition on data from 168 all 11 combinations of two or more strains, focusing on community productivity and community 169 pyoverdine production as our primary traits of interest. When examining the contributions of different 170 strain types to community functioning, we included DEV productivity and DEV pyoverdine as additional traits to 171 identify strain types driving positive or negative deviations of functioning from its expected value. 172 To examine the relative importance of strain interactions and strain identity, we extracted the 173 mean squares from the relevant LMs [ii + iii] (Bell et al. 2009). Given that the non -linear richness term 174 provides a purely statistical proxy of strain interactions , we additionally considered our summary 175 measures of supernatant-based interactions. To this end, we replaced the strain richness term in the last 176 LMs [iii] by, respectively, the interaction score or the mean absolute supernatant effect as well as the ratio 177 of positive supernatant effects, then extracted the corresponding mean squares, and finally included them 178 together with the mean squares for strain identity and non-linear strain richness in an across-community 179 comparison (see below). To be able to examine whether specific strain types contributed disproportionally 180 to community functioning, we extracted the linear model coefficients obtained for each strain from the 181 LMs [ii] focusing on strain identity . These coefficients provide a measure of each strain’s effect on 182 functioning relative to that of the average strain in the community (Bell et al. 2009). 183 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted June 9, 2024. ; https://doi.org/10.1101/2024.06.07.597923doi: bioRxiv preprint Statistical analysis. To examine growth profiles and supernatant -based interactions, we tested whether 184 growth or pyoverdine production differed between strain types (PVD PRO, PVD, NON PRO, or NON) and 185 conditions (iron-rich or iron -limited). Next, we tested whether supernatant effects (GEtreatment) differed 186 between conditions or supernatant donor and receiver types. To examine the relative importance of strain 187 interactions and strain identity on productivity and pyoverdine production , we tested for differences 188 between the mean square values obtained from our linear model decomposition for strain identity, non-189 linear strain richness, the interaction score, and the combination of mean absolute supernatant effect and 190 the ratio of positive effects (see above), accounting for differences between conditions . Similarly, we 191 examined the contributions of different strain types to (deviations from expected) community functioning 192 by comparing the strain coefficients obtained through the linear model method across our communities. 193 To examine whether strain interactions consistently shape functioning across communities, we finally 194 tested whether deviations from expected productivity (DEVproductivity) were shaped by strain number (2, 3, 195 or 4; categorical), condition, or the interaction score summarizing supernatant -based interactions in the 196 different sets of strains. Note that we always included the strains’ habitat-of-origin (soil or pond) as a co-197 factor into our models, but do not report the corresponding results here because they do not affect our 198 main results and were rarely significant (all results are reported in the supplementary material). 199 We implemented our analyses in R 4.2.1 (www.r-project.org) using LMs, generalized least squares 200 (GLS) models and linear mixed models (LMMs). GLS models and LMMs were implemented using the gls 201 and lme functions (nlme package; Pinheiro et al., 2023). We obtained p-values of effects in these models 202 using the Anova function (car package; Fox & Weisberg, 2019). We used the emmeans package (Lenth, 203 2021) to perform post hoc analyses and adjusted p -values for multiple testing (n test > 2) using the false 204 discovery rate. Unless otherwise stated, models were initially fitted with all possible interaction terms. 205 Where required, we transformed response variables to obtain normally distributed residuals. To account 206 for the non-independence of strains from the same community and for multiple measurements of each 207 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted June 9, 2024. ; https://doi.org/10.1101/2024.06.07.597923doi: bioRxiv preprint strain under different conditions, we initially fitted all models as random intercept models using 208 community and, in case of repeated measurements, strain (mixture) identity nested within community as 209 random effect(s). Each final model was then selected in a two -step procedure. First, we used the Akaike 210 information criterion (AIC) to simplify the random effect structure and to select an appropriate variance 211 structure (using the weights -argument in the gls and lme function) where residual plots indicated a 212 deviation from homogeneity (Zuur et al. 2009). Second, we simplified the fixed component by dropping 213 non-significant interaction terms (p > 0.05). The structure of all final models is detailed in Table S3. 214

215 Growth and pyoverdine production profiles of Pseudomonas strains. We first confirmed that pyoverdine 216 producers (PVDPRO and PVD) and non-producers (NONPRO and NON) behaved as expected by quantifying 217 pyoverdine production and growth under iron-limited and iron-rich conditions. Indeed, we found that the 218 producer types featured higher pyoverdine production than the non-producer types and that pyoverdine 219 favors growth when iron is scarce (Table 1+S4, Figure 1). Specifically, PVDPRO and PVD grew better than 220 the non-producers under iron limitation, while NONPRO reached higher densities than NON . Somewhat 221 surprisingly, the growth differences between PVD and NON PRO were small despite marked differences in 222 pyoverdine production (Figure 1A+B), suggesting that NON PRO might produce secondary siderophores 223 such as pyochelin (Kümmerli 2023). Indeed, when measuring the total iron -chelating activity using the 224 cholorimetric CAS assay (Schwyn & Neilands 1987), we found that NON PRO featured the same activity as 225 PVD (supplementary material; Table S5, Figure S3). Under iron-rich conditions, where siderophores are 226 not required for sustained growth, all strain types reached similar densities (Table 1A, Figure 1A) and 227 produced little pyoverdine (iron-rich vs. iron-limited: PVDPRO: t70 = -14.20, p < 0.001; PVD: t70 = -12.98, p < 228 0.001; NONPRO: t70 = -4.35, p < 0.001; NON: t70 = -1.19, p = 0.239, Figure 1B). 229 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted June 9, 2024. ; https://doi.org/10.1101/2024.06.07.597923doi: bioRxiv preprint Supernatant effects are pronounced under iron -limited conditions. Next, we harvested supernatants 230 from each strain and measured their effects on the growth of community members and the producers 231 themselves (Table S6+S7, Figure 2) . We first focus on effects that the supernatants had on their own 232 producers. Under iron -limited conditions, ‘effects on self ’ were pronounced and overall positive for all 233 types, indicating that strains typically secrete pyoverdine or other compounds into the supernatant that 234 favor their own growth. This was not the case under iron -rich conditions, where effects on self were 235 typically neutral or negative and generally small (Table S7, Figure 2) . Focusing next on the impact of 236 supernatants on other community members, we likewise found that effects under iron -rich conditions 237 were neutral or negative and typically small, suggesting a low baseline production of toxic compounds. 238 Under iron-limited conditions, however, supernatant effects on others varied substantially, ranging from 239 strong inhibition to strong stimulation for specific strain combinations (Table S7, Figure 2) . To further 240 examine how interactions differ between iron conditions, we classified all pairwise interactions within a 241 community, ranging from mutual inhibition to mutual stimulation, and displayed them as individual 242 interaction heatmaps for each community and condition (Figure 3). This qualitative analysis revealed that 243 there are many more positive interactions among strains under iron-limited than iron-rich conditions. 244 Identity effects explain more variation in community functioning than strain interactions. We examined 245 the effects of strain identity and interactions on two metrics of community functioning, productivity and 246 pyoverdine production. We set up monocultures and all combinations of two, three, or all four strains per 247 community and used an established linear model method (Bell et al. 2009) to determine the variance 248 explained by strain identity and strain interactions. We considered three measures of strain interactions: 249 (i) non-linear strain richness, a statistical proxy for interactions (Bell et al. 2009); (ii) the interaction score, 250 a proxy reflecting the overall sign of supernatant effects within communit ies (Figure 3 ); and (iii) a 251 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted June 9, 2024. ; https://doi.org/10.1101/2024.06.07.597923doi: bioRxiv preprint combination of the mean absolute supernatant effect and the proportion of positive effects , which 252 accounts for both magnitude and sign of strain interactions. 253 We found that strain identity explained more variation in community productivity (Figure 4A) and 254 pyoverdine production (Figure 4B) than strain interactions, regardless of which measure of strain 255 interactions we considered , and independent of the iron condition (Table 2+S8). When comparing the 256 three measures of strain interactions, we found that the combination of mean absolute supernatant effect 257 and the proportion of positive supernatant effects explained more variation in functioning than both non-258 linear strain richness and the interaction score (Table 2, Figure 4). Independent of these effects , strain 259 identity explained more variation in productivity – and all predictors explained more variation in 260 pyoverdine production – under iron-limited than iron-rich conditions (Table S8, Figure 4). 261 Strain types vary in their impact on community productivity. Above, we have shown that strain identity 262 effects outweigh the effects of strain interactions on community functioning. However, this result does 263 not reveal whether strain types differ in their impact on functioning. To tackle this question, we again 264 leveraged the above -described linear model method, this time focusing on the ‘strain ID’ coefficients 265 extracted from models using either community productivity, pyoverdine production, or the deviation s 266 from their expected values as function s of interest. The strain ID coefficients provide a measure of each 267 strain’s effect on community functioning relative to an average strain’s contribution (Bell et al. 2009). 268 We found that PVDPRO made above-average contributions to community productivity across iron 269 conditions (main effect of type: χ 23 = 23.18, p < 0.001; PVD DUO: t 82.7 = 2.67, p = 0.0 09; Figure 5A), while 270 NON contributed less than the average (t82.7 = -3.89, p < 0.001; Figure 5A). Intriguingly, the presence of 271 NON strains was generally associated with higher -than-average deviations from expected productivity 272 (main effect of type: χ 23 = 7.89, p = 0.048; NON: t 124 = 2.36, p = 0.0 20; Figure 5B). When focusing on 273 pyoverdine production, we unsurprisingly observed that pyoverdine producers and non-producers made, 274 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted June 9, 2024. ; https://doi.org/10.1101/2024.06.07.597923doi: bioRxiv preprint respectively, above-average and below-average contributions across iron conditions (main effect of type: 275 χ23 = 138.85, p < 0.001; type-condition interaction: χ23 = 29.81, p < 0.001; Table S9; Figure 5C). In addition 276 to these strain type effects, we observed that individual strains of each strain type variously featured 277 above-average or below-average contributions to all measures of functioning (Table S10; Figure 5). 278 Interaction scores can predict productivity across communities . Our findings show that strain 279 interactions have a smaller impact on functioning than strain identity, but this does not mean that strain 280 interactions are insignificant. In a last step , we therefore tested whether strain interactions had a 281 consistent impact on functioning across communities. We focused on deviations from expected 282 productivity as function of interest and used the interaction score as a measure of interactions among 283 community members. We predicted that low (inhibitory) and high (stimulatory) interaction scores should 284 be associated with reduced and increased productivity deviations, respectively. In line with this idea, we 285 found a positive relationship between deviations from expected productivity and the interaction score 286 (Table 3; slope ± SE: 0.46 ± 0.18, t331 = 2.59, p = 0.010, Figure 6A). Independent of this effect, productivity 287 deviations were higher under iron-rich than under iron-limited conditions ( Table 3; soil: t331 = 8.13, p < 288 0.001; pond: t331 = 3.91, p < 0.001 ; Figure 6B) and increased with strain number (Table 3; two vs. three 289 strains: t331 = 3.67, p < 0.001; three vs. four strains: t331 = 2.36, p = 0.019; Figure 6C). 290

291 Although microbial communities shape critical ecosystem services, such as primary production and the 292 fixation of greenhouse gases, we are only beginning to understand how these communities function. Here, 293 we examined and compared the impact of strain interactions and strain identity on the functioning of 16 294 communities of Pseudomonas bacteria, each comprising four strains belonging to four distinct types 295 varying in their potential to secret the siderophore pyoverdine and proteases. We found that strain 296 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted June 9, 2024. ; https://doi.org/10.1101/2024.06.07.597923doi: bioRxiv preprint identity effects explained more variation in community productivity and p yoverdine production than 297 strain interactions, regardless of whether interactions were inferred statistically or measured more 298 directly using supernatant assays. While different strain types overall affected functioning in accordance 299 with their baseline (monoculture) growth and pyoverdine production, we also found that individual strains 300 deviated from this pattern in co-culture, suggesting that strain interactions can modulate the impact of 301 strain identity. In line with this idea , we found that deviations from expected productivity consistently 302 increased across communities as interactions through secreted compounds shifted from inhibitory to 303 stimulatory. Altogether, our findings suggest that although effects of strain identity may often outweigh 304 the effects of strain interactions, both factors will usually be required to gain a nuanced understanding of 305 how naturally diverse communities function. 306 Classical studies of sociomicrobiology focusing on closely related strains have consistently 307 reported that effects of strain interactions can outweigh the impact of strain identity on functioning (e.g., 308 (Gore et al. 2009; Griffin et al. 2004; Sandoz et al. 2007; Strassmann et al. 2000). For instance, (Griffin et 309 al. 2004) showed that a pyoverdine-producing P. aeruginosa strain grew to higher density than a closely 310 related non -producer in monoculture, but was exploited and eventually outcompeted by that non -311 producer in mixed culture. By contrast, we found that interactions between genetically more diverse 312 natural Pseudomonas strains have comparably smaller effects on functioning and are typically outweighed 313 by pronounced effects of strain identity. Identity effects often result from differences in individual -level 314 traits such as growth rate and metabolic capabilit ies (Lajoie & Kembel 2019) and such differences are 315 expected to increase as strains become genetically less similar (Bernhardsson et al. 2011; Oña et al. 2021). 316 Our findings thus suggest that the impact of strain interactions relative to strain identity declines when 317 moving from genetically homogenous to more diverse communities characterized by lower relatedness. 318 We have previously shown that the siderophores expressed by Pseudomonas isolates mediate 319 diverse and strong strain interactions (Figueiredo et al. 2022), and our supernatant assays indeed revealed 320 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted June 9, 2024. ; https://doi.org/10.1101/2024.06.07.597923doi: bioRxiv preprint a greater scope for such interactions under iron-limited than iron-rich conditions. We therefore initially 321 expected strain interactions to have a higher impact on community functioning under iron limitation. 322 Contrary to this expectation, we found little evidence that different iron conditions affected the relative 323 impact of social interactions on community functioning. A potential explanation for this pattern could be 324 social trait linkage. We have previously shown that the production of different secreted compounds , 325 including pyoverdines, proteases, biofilm and toxic compounds, are often positively correlated among our 326 natural isolates (Kramer et al. 2020a). This would mean that strains contributing a lot to pyoverdine 327 production under iron limitation may contribute a lot to other secreted compound s under iron rich 328 conditions, leading to consistent (large or small) effects of specific strains on functioning across conditions. 329 In support of this idea, we found that PVDPRO strains contributed more – and NON strains less – to 330 productivity than the average strain regardless of iron condition. Hence, positive trait linkage may stabilize 331 the impact of individual community members on functioning across environmental conditions. 332 We found that deviations from expected productivity increased when supernatant effects shifted 333 from inhibition to stimulation, indicating that mutually stimulatory effects of secreted compounds could 334 promote community functioning. One possible explanation is that such mutual stimulation entails mutual 335 benefits, which can occur if it increases the availability of a growth-limiting resource for strains with non-336 overlapping niche requirements (Finke & Snyder 2008; Oña et al. 2021). However, mutual stimulation 337 often results in mutual exploitation between competing strains, where each strain strives to increase its 338 own fitness at the expense of the other by using its secreted, publicly available compounds without paying 339 the associated costs (Oliveira et al. 2014). Such mutual exploitation (or cheating; (Ghoul et al. 2014)) could 340 promote higher than expected functioning at the community level if one strain derives net benefits from 341 the interaction that outweigh the net costs to the other strain(s) (MacLean et al. 2010; Mridha & Kümmerli 342 2022). Overall, these considerations highlight that the secretion of sharable ‘public goods’ might often 343 strongly affect the functioning of bacterial communities even when identity effects are strong. 344 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted June 9, 2024. ; https://doi.org/10.1101/2024.06.07.597923doi: bioRxiv preprint To quantify the effect of strain interactions on community functioning, we used non-linear strain 345 richness as a statistical proxy for strain interactions (Bell et al. 2009) and compared its impact with that of 346 more direct interaction measures based on supernatant effects. We found that non -linear richness 347 performed as well as our univariate interaction score, but captured less variation in functioning than a 348 combination of two variables separately capturing the magnitude and sign of supernatant effects. These 349 findings validate both non -linear richness and the interaction score as simple measures of strain 350 interactions. Moreover, they show that experimental data can provide additional benefits in estimating 351 the impact of strain interactions on functioning, but only if the strength and direction of interactions are 352 separately accounted for. We therefore recommend obtaining experimental data when the impact of 353 strain interactions is of particular interest, but to rely on non-linear richness as proxy when it is not. 354 In recent years, interest in approaches to manipulate microbial assemblies and microbiomes to 355 perform beneficial functions has skyrocketed (Gu et al. 2020a; Ibrahim et al. 2021; Inda et al. 2019; 356 Mueller & Sachs 2015) and the use of probiotics has become popular in agriculture (Menendez et al. 357 2017), aquaculture (Verschuere et al. 2000) and human health (Kerry et al. 2018). Many of the proposed 358 approaches make use of microbial interactions with the idea to introduce or promote probiotic strains in 359 communities that engage in resource or interference competition with pathogens (Hu et al. 2016) or 360 exploit their social traits (Brown et al. 2009). Such interactions are often driven by secreted secondary 361 metabolites such as toxins (interference competition ; Hu et al. 2016) and siderophores (siderophore 362 exploitation; González et al. 2021). While these approaches certainly hold great promise, our results 363 highlight that the introduction of cheating mutants might only be successful if their individual -level 364 characteristics allow the mutants to persist in their niche even when the pathogen they exploit declines 365 in frequency. More generally, our findings suggest that strain identity effects, in addition to social 366 interactions, are key to consider in order to develop powerful and sustainable probiotics. 367 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted June 9, 2024. ; https://doi.org/10.1101/2024.06.07.597923doi: bioRxiv preprint In conclusion, we showed that strain identity effects have a larger impact on the productivity and 368 pyoverdine production of Pseudomonas communities than strain interactions. Our results suggest that 369 strain interactions might often reinforce or diminish, but rarely overwrite, existing baseline differences in 370 how diverse microbial players affect the functioning of their community. While our study solely focused 371 on Pseudomonas bacteria, we anticipate even stronger identity effects on our functions of interest in 372 taxonomically more diverse microbial communities. At the same time, we note that the patterns observed 373 here might change for other functions. Productivity often resembles a zero-sum game where one strain 374 can only increase its contribution to functioning at the expense of another . By contrast, other functions 375 may leave more scope for net -positive complementation (e.g., community respiration; Bell et al. 2009) 376 and these functions should be determined to a greater extent by species interactions relative to identity 377 effects. Future work should theref ore consider multiple functions in addition to productivity to unravel 378 the importance of species interactions in natural microbial communities of varying taxonomical diversity. 379 Author contributions. Conceptualization: JK and RK; Experiments: JK, SM and ARTF; Data Analysis: JK and 380 ARTF; Writing (first draft): JK and RK; Writing (refinement): JK, SM, ARTF and RK. 381 Acknowledgements. We thank Elena Butaitė for collecting the natural isolates. 382 Funding. This research was supported by the German Science Foundation (DFG; KR 5017/2 -1 to JK), the 383 University of Zurich (Forschungskredit; FK -17-111 to JK), the Swiss National Science Foundation 384 (31003A_182499 and 310030_212266 to RK) and the Zurich University Research Priority Program (URPP) 385 ‘Evolution in Action’. 386 Conflict of interest. We have no conflict of interest. 387

388 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted June 9, 2024. ; https://doi.org/10.1101/2024.06.07.597923doi: bioRxiv preprint Alper, H. & Stephanopoulos, G. (2009). Engineering for biofuels: exploiting innate microbial capacity or 389 importing biosynthetic potential? Nat. Rev. Microbiol., 7, 715–723. 390 Becker, J., Eisenhauer, N., Scheu, S. & Jousset, A. (2012). Increasing antagonistic interactions cause 391 bacterial communities to collapse at high diversity. Ecol. Lett., 15, 468–474. 392 Bell, T., Lilley, A.K., Hector, A., Schmid, B., King, L. & Newman, J.A. (2009). A linear model method for 393 biodiversity–ecosystem functioning experiments. Am. Nat., 174, 836–849. 394 Bell, T., Newman, J.A., Silverman, B.W., Turner, S.L. & Lilley, A.K. (2005). The contribution of species 395 richness and composition to bacterial services. Nature, 436, 1157–1160. 396 Bernhardsson, S., Gerlee, P. & Lizana, L. (2011). Structural correlations in bacterial metabolic networks. 397 BMC Evol. Biol., 11, 20. 398 Brockhurst, M.A., Buckling, A. & Gardner, A. (2007). Cooperation Peaks at Intermediate Disturbance. 399 Curr. Biol., 17, 761–765. 400 Brown, S.P., West, S.A., Diggle, S.P. & Griffin, A.S. (2009). Social evolution in micro-organisms and a 401 Trojan horse approach to medical intervention strategies. Philos. Trans. R. Soc. B Biol. Sci., 364, 402 3157–3168. 403 Butaitė, E., Baumgartner, M., Wyder, S. & Kümmerli, R. (2017). Siderophore cheating and cheating 404 resistance shape competition for iron in soil and freshwater Pseudomonas communities. Nat. 405 Commun., 8, 414. 406 Butaitė, E., Kramer, J., Wyder, S. & Kümmerli, R. (2018). Environmental determinants of pyoverdine 407 production, exploitation and competition in natural Pseudomonas communities. Environ. 408 Microbiol., 20, 3629–3642. 409 Connolly, J., Bell, T., Bolger, T., Brophy, C., Carnus, T., Finn, J.A., et al. (2013). An improved model to 410 predict the effects of changing biodiversity levels on ecosystem function. J. Ecol., 101, 344–355. 411 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted June 9, 2024. ; https://doi.org/10.1101/2024.06.07.597923doi: bioRxiv preprint Cornelis, P. (2010). Iron uptake and metabolism in pseudomonads. Appl. Microbiol. Biotechnol., 86, 412 1637–1645. 413 Fiegna, F., Moreno-Letelier, A., Bell, T. & Barraclough, T.G. (2015). Evolution of species interactions 414 determines microbial community productivity in new environments. ISME J., 9, 1235–1245. 415 Figueiredo, A.R.T., Özkaya, Ö., Kümmerli, R. & Kramer, J. (2022). Siderophores drive invasion dynamics in 416 bacterial communities through their dual role as public good versus public bad. Ecol. Lett., 25, 417 138–150. 418 Finke, D.L. & Snyder, W.E. (2008). Niche Partitioning Increases Resource Exploitation by Diverse 419 Communities. Science, 321, 1488–1490. 420 Fox J, Weisberg S (2019). An R Companion to Applied Regression, Third edition. Sage, Thousand Oaks CA. 421 https://socialsciences.mcmaster.ca/jfox/Books/Companion/. 422 Ghoul, M., Griffin, A.S. & West, S.A. (2014). Toward an evolutionary definition of cheating. Evolution, 68, 423 318–31. 424 González, J., Salvador, M., Özkaya, Ö., Spick, M., Reid, K., Costa, C., et al. (2021). Loss of a pyoverdine 425 secondary receptor in Pseudomonas aeruginosa results in a fitter strain suitable for population 426 invasion. ISME J., 15, 1330–1343. 427 Gore, J., Youk, H. & Van Oudenaarden, A. (2009). Snowdrift game dynamics and facultative cheating in 428 yeast. Nature, 459, 253–256. 429 Gorter, F.A., Manhart, M. & Ackermann, M. (2020). Understanding the evolution of interspecies 430 interactions in microbial communities. Philos. Trans. R. Soc. B Biol. Sci., 375, 20190256. 431 Granato, E.T., Meiller-Legrand, T.A. & Foster, K.R. (2019). The evolution and ecology of bacterial warfare. 432 Curr. Biol., 29, R521–R537. 433 Griffin, A.S., West, S.A. & Buckling, A. (2004). Cooperation and competition in pathogenic bacteria. 434 Nature, 430, 1024–1027. 435 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted June 9, 2024. ; https://doi.org/10.1101/2024.06.07.597923doi: bioRxiv preprint Gu, S., Wei, Z., Shao, Z., Friman, V.-P., Cao, K., Yang, T., et al. (2020a). Competition for iron drives 436 phytopathogen control by natural rhizosphere microbiomes. Nat. Microbiol., 5, 1002–1010. 437 Gu, S., Yang, T., Shao, Z., Wang, T., Cao, K., Jousset, A., et al. (2020b). Siderophore-Mediated Interactions 438 Determine the Disease Suppressiveness of Microbial Consortia. mSystems, 5, e00811-19. 439 Hodgson, D.J., Rainey, P.B. & Buckling, A. (2002). Mechanisms linking diversity, productivity and 440 invasibility in experimental bacterial communities. Proc. R. Soc. Lond. B Biol. Sci., 269, 2277–441 2283. 442 Hu, J., Wei, Z., Friman, V.-P., Gu, S., Wang, X., Eisenhauer, N., et al. (2016). Probiotic Diversity Enhances 443 Rhizosphere Microbiome Function and Plant Disease Suppression. mBio, 7, e01790-16. 444 Ibrahim, M., Raajaraam, L. & Raman, K. (2021). Modelling microbial communities: Harnessing consortia 445 for biotechnological applications. Comput. Struct. Biotechnol. J., 19, 3892–3907. 446 Inda, M.E., Broset, E., Lu, T.K. & De La Fuente-Nunez, C. (2019). Emerging Frontiers in Microbiome 447 Engineering. Trends Immunol., 40, 952–973. 448 Jones, M.L., Rivett, D.W., Pascual-García, A. & Bell, T. (2021). Relationships between community 449 composition, productivity and invasion resistance in semi-natural bacterial microcosms. eLife, 450 10, e71811. 451 Kerry, R.G., Patra, J.K., Gouda, S., Park, Y., Shin, H.-S. & Das, G. (2018). Benefaction of probiotics for 452 human health: A review. J. Food Drug Anal., 26, 927–939. 453 Konopka, A. (2009). What is microbial community ecology? ISME J., 3, 1223–1230. 454 Kramer, J., López Carrasco, M.Á. & Kümmerli, R. (2020a). Positive linkage between bacterial social traits 455 reveals that homogeneous rather than specialised behavioral repertoires prevail in natural 456 Pseudomonas communities. FEMS Microbiol. Ecol., 96, fiz185. 457 Kramer, J., Özkaya, Ö. & Kümmerli, R. (2020b). Bacterial siderophores in community and host 458 interactions. Nat. Rev. Microbiol., 18, 152–163. 459 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted June 9, 2024. ; https://doi.org/10.1101/2024.06.07.597923doi: bioRxiv preprint Kümmerli, R. (2023). Iron acquisition strategies in pseudomonads: mechanisms, ecology, and evolution. 460 BioMetals, 36, 777–797. 461 Kümmerli, R., Griffin, A.S., West, S.A., Buckling, A. & Harrison, F. (2009). Viscous medium promotes 462 cooperation in the pathogenic bacterium Pseudomonas aeruginosa. Proc. R. Soc. B Biol. Sci., 276, 463 3531–3538. 464 Lajoie, G. & Kembel, S.W. (2019). Making the Most of Trait-Based Approaches for Microbial Ecology. 465 Trends Microbiol., 27, 814–823. 466 Lenth, R. (2021). emmeans: Estimated marginal means, aka least-squares means. R package version 467 1.6.2-1. 468 Li, M., Pommier, T., Yin, Y., Cao, W., Zhang, X., Hu, J., et al. (2022). Resource availability drives bacteria 469 community resistance to pathogen invasion via altering bacterial pairwise interactions. Environ. 470 Microbiol., 24, 5680–5689. 471 Li, M., Wei, Z., Wang, J., Jousset, A., Friman, V., Xu, Y., et al. (2019). Facilitation promotes invasions in 472 plant‐associated microbial communities. Ecol. Lett., 22, 149–158. 473 Loreau, M. & Hector, A. (2001). Partitioning selection and complementarity in biodiversity experiments. 474 Nature, 412, 72–76. 475 MacLean, R.C., Fuentes-Hernandez, A., Greig, D., Hurst, L.D. & Gudelj, I. (2010). A Mixture of “Cheats” 476 and “Co-Operators” Can Enable Maximal Group Benefit. PLoS Biol., 8, e1000486. 477 Menendez, E., Garcia-Fraile, P. (2017). Plant probiotic bacteria: solutions to feed the world. AIMS 478 Microbiol., 3, 502–524. 479 Minty, J.J., Singer, M.E., Scholz, S.A., Bae, C.-H., Ahn, J.-H., Foster, C.E., et al. (2013). Design and 480 characterization of synthetic fungal-bacterial consortia for direct production of isobutanol from 481 cellulosic biomass. Proc. Natl. Acad. Sci., 110, 14592–14597. 482 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted June 9, 2024. ; https://doi.org/10.1101/2024.06.07.597923doi: bioRxiv preprint Mridha, S. & Kümmerli, R. (2022). Enforced specialization fosters mutual cheating and not division of 483 labour in the bacterium Pseudomonas aeruginosa. J. Evol. Biol., 35, 719–730. 484 Mueller, U.G. & Sachs, J.L. (2015). Engineering Microbiomes to Improve Plant and Animal Health. Trends 485 Microbiol., 23, 606–617. 486 O’Brien, S., Culbert, C.T. & Barraclough, T.G. (2023). Community composition drives siderophore 487 dynamics in multispecies bacterial communities. BMC Ecol. Evol., 23, 45. 488 Oliveira, N.M., Niehus, R. & Foster, K.R. (2014). Evolutionary limits to cooperation in microbial 489 communities. Proc. Natl. Acad. Sci. U. S. A., 111, 201412673. 490 Oña, L., Giri, S., Avermann, N., Kreienbaum, M., Thormann, K.M. & Kost, C. (2021). Obligate cross-491 feeding expands the metabolic niche of bacteria. Nat. Ecol. Evol., 5, 1224–1232. 492 Piccardi, P., Vessman, B. & Mitri, S. (2019). Toxicity drives facilitation between 4 bacterial species. Proc. 493 Natl. Acad. Sci., 116, 15979–15984. 494 Pinheiro J, Bates D, R Core Team (2023). nlme: Linear and Nonlinear Mixed Effects Models. R package 495 version 3.1-163, https://CRAN.R-project.org/package=nlme. 496 Rainey, P.B. & Rainey, K. (2003). Evolution of cooperation and conflict in experimental bacterial 497 populations. Nature, 425, 72–74. 498 Ratzke, C., Barrere, J. & Gore, J. (2020). Strength of species interactions determines biodiversity and 499 stability in microbial communities. Nat. Ecol. Evol., 4, 376–383. 500 Sandoz, K.M., Mitzimberg, S.M. & Schuster, M. (2007). Social cheating in Pseudomonas aeruginosa 501 quorum sensing. Proc. Natl. Acad. Sci. U. S. A., 104, 15876–81. 502 Schwyn, B. & Neilands, J.B. (1987). Universal chemical assay for the detection and determination of 503 siderophores. Anal. Biochem., 160, 47–56. 504 Smith, M.D. & Knapp, A.K. (2003). Dominant species maintain ecosystem function with non‐random 505 species loss. Ecol. Lett., 6, 509–517. 506 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted June 9, 2024. ; https://doi.org/10.1101/2024.06.07.597923doi: bioRxiv preprint Stockmann, U., Adams, M.A., Crawford, J.W., Field, D.J., Henakaarchchi, N., Jenkins, M., et al. (2013). 507 The knowns, known unknowns and unknowns of sequestration of soil organic carbon. Agric. 508 Ecosyst. Environ., 164, 80–99. 509 Strassmann, J.E., Zhu, Y. & Queller, D.C. (2000). Altruism and social cheating in the social amoeba 510 Dictyostelium discoideum. Nature, 408, 965–967. 511 Venail, P.A. & Vives, M.J. (2013). Phylogenetic distance and species richness interactively affect the 512 productivity of bacterial communities. Ecology, 94, 2529–2536. 513 Verschuere, L., Rombaut, G., Sorgeloos, P. & Verstraete, W. (2000). Probiotic Bacteria as Biological 514 Control Agents in Aquaculture. Microbiol. Mol. Biol. Rev., 64, 655–671. 515 Wang, J. & Chu, L. (2016). Biological nitrate removal from water and wastewater by solid-phase 516 denitrification process. Biotechnol. Adv., 34, 1103–1112. 517 Wardle, D.A., Barker, G.M., Bonner, K.I. & Nicholson, K.S. (1998). Can comparative approaches based on 518 plant ecophysiological traits predict the nature of biotic interactions and individual plant species 519 effects in ecosystems? J. Ecol., 86, 405–420. 520 West, S.A., Diggle, S.P., Buckling, A., Gardner, A. & Griffin, A.S. (2007). The social lives of microbes. Annu. 521 Rev. Ecol. Evol. Syst., 38, 53–77. 522 Whitman, W.B., Coleman, D.C. & Wiebe, W.J. (1998). Prokaryotes: The unseen majority. Proc. Natl. 523 Acad. Sci. U. S. A., 95, 6578–6583. 524 Zuur, A.F., Ieno, E.N., Walker, N.J., Saveliev, A.A. & Smith, G.M. (2009). Mixed effects models and 525 extensions in ecology with R. Stat. Biol. Health. Springer, New York. 526 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted June 9, 2024. ; https://doi.org/10.1101/2024.06.07.597923doi: bioRxiv preprint Table 1 | Growth and siderophore production profiles. Determinants of (A) monoculture growth and (B) 527 pyoverdine production of soil and freshwater Pseudomonas strains belonging to four different strain types 528 varying in their production of proteases and the siderophore pyoverdine (PVDPRO, PVD, NONPRO, and NON). 529 (A) growth (B) pyoverdine production df χ2 p df χ2 p habitat 1 5.28 0.022 1 3.01 0.083 medium 1 217.93 < 0.001 1 267.68 < 0.001 type 3 65.90 < 0.001 3 115.62 < 0.001 habitat : type 3 9.75 0.021 - - - medium : type 3 9.79 0.021 3 122.89 < 0.001 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted June 9, 2024. ; https://doi.org/10.1101/2024.06.07.597923doi: bioRxiv preprint Table 2 | Impact of strain identity and interactions on community functioning. Post-hoc comparisons of 530 different determinants of (A) productivity and (B) pyoverdine production of 16 small Pseudomonas 531 communities. Significant p-values are in bold. 532 (A) productivity (B) pyoverdine production contrast ratio SE t108 p ratio SE t93 p strain ID - non-linear richness 317.56 213.91 8.552 < 0.001 1139.78 489.32 16.395 < 0.001 strain ID - interaction score 367.61 247.63 8.769 < 0.001 1104.74 474.28 16.322 < 0.001 strain ID - combined effects 24.50 16.50 4.749 < 0.001 82.22 35.30 10.271 < 0.001 non-linear richness - interaction score 1.16 0.78 0.217 0.828 0.97 0.42 -0.073 0.942 non-linear richness - combined effects 0.08 0.05 -3.803 < 0.001 0.07 0.03 -6.124 < 0.001 interaction score - combined effects 0.07 0.04 -4.021 < 0.001 0.07 0.03 -6.051 < 0.001 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted June 9, 2024. ; https://doi.org/10.1101/2024.06.07.597923doi: bioRxiv preprint Table 3 | Deviations from expected community productivity. Determinants of deviations from expected 533 community productivity. Significant p-values are in bold. 534 χ21 p habitat 0.371 0.542 medium 54.792 < 0.001 strain number 28.475 < 0.001 interaction score 6.725 0.010 habitat:medium 12.995 < 0.001 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted June 9, 2024. ; https://doi.org/10.1101/2024.06.07.597923doi: bioRxiv preprint Figure 1 | Strain types differ in their g rowth and pyoverdine production profiles. (A) Growth and (B) pyoverdine production of PVDPRO (red), PVD (green), NON PRO (orange), and NON (blue) strains isolated from eight soil (empty small circles) and eight freshwater (filled small circles) communities (one strain per type and community), measured in iron-limited and iron-rich medium. Small circles represent the median of four replicates obtained for each strain under each condition. Large circles and black lines show mean and standard error, respectively. Letters show significantly different types. All comparisons were performed within each medium (detailed statistical results are provided in Table S4). was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted June 9, 2024. ; https://doi.org/10.1101/2024.06.07.597923doi: bioRxiv preprint Figure 2 | Secreted compounds have pronounced effects under iron-limitation. Shown are effects that 535 PVDPRO, PVD, NON PRO, and NON strains isolated from soil (empty small circles) and pond (filled small 536 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted June 9, 2024. ; https://doi.org/10.1101/2024.06.07.597923doi: bioRxiv preprint circles) communities have on each other’s growth through compounds secreted into the supernatant 537 under iron-limited and iron-rich conditions. Small circles show the median of four replicates obtained for 538 each donor/receiver combination. Small grey circles show supernatant effects of specific donor/receiver 539 combinations that did not differ from neutrality, whereas small colored circles indicate significant effects 540 on receiver growth. Large circles and black lines show mean and standard error , respectively. Dashed 541 horizontal lines indicate the null line where compounds in the supernatant have no effect on receiver 542 growth. Colored rectangles highlight the effects that strains have on their own growth. Asterisks above 543 and below the null line indicate that the average supernatant effect of a specific combination of donor 544 and receiver types was significantly positive and negative, respectively (significance levels are indicated 545 as follows: * 0.05 ≥ p > 0.01; ** 0.01 ≥ p > 0.001; *** p ≤ 0.001; see Table S6+S7 for further details). 546 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted June 9, 2024. ; https://doi.org/10.1101/2024.06.07.597923doi: bioRxiv preprint Figure 3 | Interaction types illustrate the high potential for growth stimulation under iron limitation. 547 Interaction types between – and effects on self of – PVDPRO, PVD, NONPRO, and NON strains from eight soil 548 (s3a to s3h) and eight freshwater (3A to 3H) communities of Pseudomonas bacteria. Interaction types 549 (opaque colors) and effects on self (transparent colors) were assigned based on the positive, neutral, or 550 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted June 9, 2024. ; https://doi.org/10.1101/2024.06.07.597923doi: bioRxiv preprint negative effects that the strains had on each other and themselves through compounds secreted into the 551 supernatant under iron-limited and iron-rich conditions (see Figure 2 and the Methods for details). 552 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted June 9, 2024. ; https://doi.org/10.1101/2024.06.07.597923doi: bioRxiv preprint Figure 4 | Strain identity explains more variation in community productivity than strain interactions. 553 Shown are mean square values extracted from linear models fit for each of eight soil (empty small circles) 554 and eight pond (filled small circles ) communities to explain the impact of strain identity and three 555 measures of strain interactions, non-linear richness, the interaction score, and a combination of variables 556 on (A) community productivity and (B) pyoverdine production. The impact of linear richness , i.e. the 557 extent to which community functioning linearly increases with strain number, is shown for comparison 558 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted June 9, 2024. ; https://doi.org/10.1101/2024.06.07.597923doi: bioRxiv preprint (grey-shaded area) . While non -linear richness is a purely statistical proxy for strain interactions, the 559 interaction score and the combination of variables are based on supernatant effects. The interaction score 560 predominantly reflects the sign of supernatant effects, whereas the combination of variables includes the 561 mean absolute supernatant effect and the proportion of positive supernatant effects and thus separately 562 accounts for both sign and magnitude. Large circles and black lines show means and standard errors. Small 563 circles show mean square values obtained for specific communities from models fit separately to data 564 generated under iron -limited and iron -rich conditions (letters show significantly different impacts on 565 functioning based on the results of our statistical models; see the Methods and Table 2+S8 for details). 566 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted June 9, 2024. ; https://doi.org/10.1101/2024.06.07.597923doi: bioRxiv preprint was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted June 9, 2024. ; https://doi.org/10.1101/2024.06.07.597923doi: bioRxiv preprint Figure 5 | Strain types and individual strains differ in their impact on community productivity. Shown 567 are linear model coefficients indicating the effect that PVD PRO (red), PVD (green), NON PRO (orange), and 568 NON (blue) strains from soil (empty small circles) and pond (filled small circles) communities had, relative 569 to the average in their community , on (A) community productivity , (B) deviations from expected 570 community productivity , (C) pyoverdine production, and (D) deviations from expected pyoverdine 571 production. Large circles and black lines show means and standard errors. Dashed lines indicate average 572 strain effects. Small circles show linear model coefficients obtained for specific strains from models fit 573 separately to data generated for each community under iron -limited and iron -rich conditions, 574 respectively. Small grey circles indicate that linear model coefficients were not significant, whereas small 575 colored circles indicate significant coefficients. Asterisks above and below the average effect line indicate 576 that the average effect of a specific type was significantly positive and negative, respectively (significance 577 levels are based on the results of our statistical models and indicated as follows: * 0.05 ≥ p > 0.01; ** 0.01 578 ≥ p > 0.001; *** p ≤ 0.001; statistics for specific strains are given in Table S10). 579 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted June 9, 2024. ; https://doi.org/10.1101/2024.06.07.597923doi: bioRxiv preprint was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted June 9, 2024. ; https://doi.org/10.1101/2024.06.07.597923doi: bioRxiv preprint Figure 6 | Deviations from expected productivity increase as supernatant-based interactions shift from 580 inhibitory to stimulatory . Shown are relationships between deviations from expected community 581 productivity and (A) the supernatant-based interaction scores, (B) iron condition, and (C) strain number, 582 as measured under iron-limited (blue) and iron-rich (green) conditions in pond and soil strains [left and 583 right panels in (A) and (B), respectively] . High intera ction scores indicate that stimulatory effects of 584 secreted compounds prevail, whereas low interaction scores indicate a prevalence of inhibitory effects. 585 Solid lines and shaded areas in (A) are regression lines and 95% confidence intervals, respectively . Large 586 white circles and black lines in ( B) and (C) show means and standard errors, respectively. Significance 587 levels in (B) and (C) are indicated as follows: * 0.05 ≥ p > 0.01; ** 0.01 ≥ p > 0.001; *** p ≤ 0.001. 588 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (whichthis version posted June 9, 2024. ; https://doi.org/10.1101/2024.06.07.597923doi: bioRxiv preprint