Unraveling the Independent Effects of Species Richness and Composition on Microbial Biofilm Growth Authors: So-Yeon Jeong, Ji Won Lee, Chi Won Lee, Tae Gwan Kim Published: 2026 ABSTRACT AbstractSpecies richness is often positively linked to ecosystem functioning. However, conventional approaches that manipulate richness frequently confound the richness effect with those of species composition. In this study, we quantitatively disentangled the independent effects of species richness and composition on biofilm productivity as a measure of ecosystem functioning. We constructed 300 independent richness gradients, each comprising 3–20 species, from a pool of 24 bacterial isolates spanning six taxonomic classes (Actinomycetia, Alphaproteobacteria, Bacilli, Betaproteobacteria, Gammaproteobacteria, and Sphingobacteriia). Our results revealed diverse richness–biofilm relationship forms, predominantly positive (68%), but also hump-shaped (16%), U-shaped (6%), null (6%), and negative (4%). When assessed individually, richness accounted for an average of 53.9% of the variation in biofilm growth, surpassing two measures of compositional variation—species-inherent ability (SIA) and species-dependent ability (SDA)—based on species’ biofilm-forming potential, which explained 9.6% and 14.4%, respectively. Collectively, richness, SIA, and SDA explained 73.4% of the variation. Notably, biofilm growth exceeded expectations in the mid-richness range (10–15 species). When richness and composition were assessed collectively across entire assemblages, richness explained 24.2% of the variation in biofilm growth, while SIA and SDA explained 12.7% and 28.7%, respectively. In contrast, species combination, assuming equal potential for all species, had only a marginal effect on biofilm growth. Our results demonstrate that species richness is a key, independent driver of biofilm growth, and that its effects are substantially underestimated when not properly separated from composition.Supplementary InformationThe online version contains supplementary material available at 10.1007/s00248-026-02789-0.Keywords: Species richness, Composition, Biofilm, Biodiversity, Richness- ecosystem functioning relationship Introduction Biofilms are ubiquitous and dominant microbial life forms on Earth, with 40–80% of all prokaryotic cells residing in this form [1]. Biofilms play an important role in ecosystem processes and biogeochemical cycles [1, 2]. Biofilms are defined as a structured aggregate of microbial cells embedded within a matrix of extracellular polymeric substances, produced by the organisms themselves [3]. They are complex communities with a huge variety of microbial species [2]. Microorganisms develop biofilms together, during which coinhabiting species interact with each other in a complex manner and constantly adapt to the surrounding environment [2–4]. Thus, biofilm growth plays a crucial role in determining microbial proliferation and activity. Biodiversity can remarkably influence dynamics and functioning of ecosystems [5, 6]. The relationship between biodiversity and ecosystem functioning (BEF), as well as the underlying ecological mechanisms, has received great attention in general ecology and more recently in microbial ecology [6, 7]. A positive BEF relationship is often observed in nature, but other patterns, such as curve-shaped or null relationships, also exist [6, 8, 9]. For example, trees and herbaceous plants exhibit positive associations between richness and functioning, such as productivity, stability, and nutrient dynamics [6, 8–10]. Bacteria and eukaryotic algae also display positive richness–functioning patterns for respiration, biomass growth, nitrogen uptake, and photosynthesis [7, 11, 12]. In general, research has focused on examining the overall response of functioning to a gradient of initial richness [5, 6, 8]. Most richness–functioning experiments have assembled communities without systematic control over composition across richness levels [6], although some studies have incorporated ecologically informed assembly rules, such as nested composition patterns [13]. In this conventional approach, assemblages at each richness level are constructed through random species selection with no systematic compositional overlap between levels, lack a nested subset structure, and are pooled across all richness levels into a single regression to estimate richness effects. This design may obscure a quantitative understanding of the richness–functioning relationship within a given ecosystem, as it captures only an average pattern across heterogeneous compositional trajectories, thereby overlooking the diversity of relationship forms—positive, negative, hump-shaped, or null—that may arise along individual compositional gradients. Moreover, it may underestimate the predictive power of richness due to variations in species composition within and between richness levels, thereby diminishing its ability to accurately predict ecosystem functioning. To overcome these challenges, we propose an alternative approach that examines the impact of richness on functioning within numerous individual richness gradients. Each gradient is constructed by sequentially adding species so that smaller assemblages are proper subsets of larger ones, ensuring that changes in biofilm along the gradient are attributable to the addition of specific species rather than to uncontrolled compositional turnover. Computer simulation confirmed that the richness effect on functioning was substantially underestimated using the conventional approach, compared to the suggested approach (Fig. 1). (a) Overview of collective and (b) individual experimental approaches for examining the impact of species richness on ecosystem functioning. (c) Simulated effects of richness on functioning in collective and individual assessments, assuming a positive, monotonic relationship between them Ecosystem functioning may exhibit diverse responses to a gradient of richness, influenced by factors such as richness and composition [6, 7, 14]. However, the frequencies of different richness–functioning relationship forms and the independent contributions of these two factors remain largely unresolved. In particular, it is unclear how much variation in functioning is attributable to increasing species richness along a controlled compositional trajectory, versus how much is driven by differences in species identity among assemblages at a given richness level. In this study, we hypothesized that species richness most frequently shows a positive association with biofilm growth and explains more variation than species composition. Such positive richness–biofilm relationships may arise through multiple mechanisms, including niche complementarity, the sampling effect, and the disproportionate prevalence of facilitative interspecific interactions characteristic of multispecies biofilm systems [9, 11, 15, 16]. We employed a new approach that deliberately examines the independent influences of richness and composition on functioning within numerous richness gradients, each consisting of assemblages with six richness levels (3, 5, 7, 10, 15, and 20 species). We assembled synthetic communities from naturally occurring bacterial isolates spanning distinct genera and diverse source environments, deliberately maximizing phylogenetic diversity while minimizing ecological redundancy and coevolutionary history that could obscure richness effects. This design provides a conservative test of whether species richness per se drives biofilm formation. We selected biofilm growth as a component of ecosystem functioning, as it is commonly measured as biomass or productivity turnover [6, 8, 10, 15]. We examined the probability of observing specific richness–biofilm patterns and quantified the independent influence of richness on biofilm growth. We then compared the effects of richness and composition on biofilm growth, assessed both within individual richness gradients and collectively across all assemblages. Materials and Methods Species Pools We used 29 bacterial isolates for assembling artificial communities in this study (Tables S1 and S2). They were isolated from a range of natural and engineered environments, including forest soils (temperate deciduous forests in Busan, South Korea), wetland soils (freshwater wetlands in Gyeongsangnam-do), intertidal mud flats (south coast of Korea), activated sludge (municipal wastewater treatment plants in Busan), air samples (Busan), and freshwater environments (rivers in Gyeongsangnam-do). All isolates were cultivated on R2A (MB cell, KisanBio, Seoul, Korea) agar plates as pure colonies. Two species pools were generated by randomly assigning isolates into two groups. Each pool was composed of 24 bacterial isolates, mostly from different genera (Tables S1 and S2), with 19 isolates shared between the two pools. The first pool was used for the main experiment, which included 300 richness gradients to ensure robust statistical power, while the second pool, consisting of 55 richness gradients, was used to validate the generality of observed patterns across species pools. Bacterial Cell Preparation Low-strength R2A (LS-R2A) medium was prepared at approximately one-third of the manufacturer’s recommended concentration (1 g/L instead of 3.15 g/L R2A powder, MB cell). LS-R2A was used to cultivate bacteria and biofilms in this study. Each pure colony was inoculated into a 250-mL flask containing 100 mL of LS-R2A medium. The flasks were incubated at 30 °C with continuous agitation of 200 rpm for 2 days. Cells were harvested by centrifugation at 3500 rpm for 20 min and suspended in sterile 0.9% NaCl solution. To measure cell biomass, cell suspensions were filtered through glass-fiber filter and dried at 105 °C for 2 h. Each cell suspension was adjusted to a final concentration of 1 g/L and stored at − 70 °C until use. Monoculture and Polyculture Biofilms Bacterial isolates were allowed to form biofilms in a 96-well plate (SPL Life Sciences, Pocheon, Korea) (diameter, 6.9 mm; and height, 10.8 mm). Each well was inoculated with 100 µL of bacterial cell suspension (1 g/L) and incubated for one day. The suspension was carefully discarded by pipetting and replaced with 100 µL of fresh LS-R2A medium daily for 6 days. The plates were incubated at 25 ℃ with agitation at 40 rpm for the entire period. There were three replicates for each isolate. To generate polyculture biofilms, 300 richness gradients were assembled based on the first species pool. Each gradient included 3-, 5-, 7-, 10-, 15-, and 20-species assemblages, generating 6 distinct richness levels (Tables S3–S14). They were constructed by sequentially adding species to an established assemblage in a predetermined order (smaller ones are subsets of larger ones). The first eight gradients were constructed as preliminary trials to verify the existence of a BEF pattern before proceeding with the fully randomized assembly of the remaining 292 gradients. Polyculture assemblages were constructed using a substitutive design [17], in which the total initial cell density was held constant (1 g/L) across all richness levels. Equal volumes of cell suspensions from each constituent species were combined, such that increasing species richness proportionally reduced the initial density of each individual species while maintaining a constant total community density. Another polyculture biofilm experiment was conducted with the 55 artificial community sets. Each artificial community had three replicates. Biofilm Quantification Biofilms formed in wells were quantified using a crystal violet (CV) staining method [18]. A plate was submerged in a 1-L beaker filled with 800 mL of distilled water to wash out unattached cells and medium. Then, 200 µL of 0.1% CV staining solution (Merck, Munich, Germany) was added to each well and incubated at 25 ℃ for 15 min. It was washed, turned upside down, and air-dried for 24 h. Subsequently, 200 µL of 30% acetic acid (Duksan, Ansan, Korea) was added and incubated at 25 ℃ for 15 min. The solubilized CV (125 µL) was transferred to a disposable cuvette (Kartell, Noviglio, Italy) filled up with 875 µL of 30% acetic acid. Absorbance was measured at 550 nm using a spectrophotometer (UV-1800, Shimadzu, Kyoto, Japan). Statistical Analysis All statistical analyses were conducted in R (v4.3.0). We evaluated richness effects on biofilm growth using two complementary perspectives: (i) individual assessments within each richness gradient (where smaller assemblages are subsets of larger ones), and (ii) a collective assessment across all assemblages. Because different gradients contain different species compositions, assemblages at the same richness level vary substantially in their species membership. This compositional heterogeneity introduces variation that is uncorrelated with richness, inflating residual variance and thereby attenuating the apparent explanatory power (R²) of richness in collective assessments. In contrast, within each individual gradient, composition changes systematically through sequential species addition, minimizing extraneous compositional variation and allowing richness effects to be estimated with greater precision. To illustrate how this distinction leads collective approaches to underestimate richness effects, we performed simulations comparing these two assessment modes (Supplementary Methods). Biofilm quantities were min–max normalized (0–1) within each gradient solely for visualizing the overall richness–biofilm relationship and classifying relationship forms; all analyses comparing the explanatory power of richness, C-SIA, and C-SDA were conducted using unnormalized data. For each gradient, we classified the richness–biofilm relationship into five types (positive, negative, null, hump-shaped, or U-shaped) [19, 20] (Fig. S1). Briefly, linear and quadratic models were fitted, and a comparison was used to test whether a quadratic model improved fit over a linear model. For relationships identified as curvilinear, the Mitchell–Olds and Shaw test (vegan) was applied to determine whether the quadratic extremum (maximum or minimum) occurred within the observed richness range, supporting hump- or U-shaped patterns. Gradients that were not supported as curvilinear were classified using the linear slope as positive or negative; gradients not significant in both linear and quadratic tests were classified as null. Full classification criteria are described in Supplementary Methods. To quantify compositional effects, we calculated two species-level metrics: species-inherent ability (SIA), defined as monoculture biofilm-forming ability, and species-dependent ability (SDA), defined as a species’ contribution to biofilm formation in polyculture. SDA was estimated by comparing biofilm formation between assemblages where the focal species was present versus absent, with uncertainty addressed via bootstrapping (Supplementary Methods). Community-level metrics (C-SIA and C-SDA) were computed as sums of the species SIA or SDA values present in each assemblage and were linearly detrended with respect to richness within each gradient to ensure independence from richness. We then used linear regression to assess the explanatory power of richness, C-SIA, and C-SDA on biofilm growth within each gradient and across all assemblages, including multi-predictor models (details, including coefficient constraints and additional diagnostics, are described in Supplementary Methods). Results Richness–biofilm Relationship Forms Biofilm formation, quantified as crystal violet absorbance, was plotted against richness for each gradient (the first experiment with 300 sets) (Figs. S2–S16). There were various richness–biofilm relationship forms across gradients, including positive (68.0%), hump-shaped (16.3%), U-shaped (6.3%), negative (3.7%), and null forms (5.7%) (Fig. 2a) (Fig. S17 for the representative forms). Within the positive relationship, linear, asymptotic, and exponential forms were observed, constituting 90%, 6%, and 4%, respectively (Fig. 2b). Relationships between richness and biofilm growth across individual richness gradients. (a) Frequencies of richness–biofilm relationship forms observed from 300 richness gradients and (b) frequencies of linear, accelerating, and decelerating forms observed from the 204 gradients showing a positive relationship A separate experiment using the 55 richness gradients assembled from the second species pool (which differed in species membership) yielded a similar distribution of relationship forms: positive (78.2%), hump-shaped (3.6%), U-shaped (3.6%), negative (1.8%), and null (12.7%) (data not shown). Linear, asymptotic, and exponential forms comprised 88%, 12%, and 0% of the total positive forms, respectively. The Chi-squared test showed that the ratio of positive to non-positive forms was consistent between the two experiments (P > 0.05). Subsequent analyses were conducted using the first experiment. Biofilm-forming Potentials Micrococcus showed the greatest SIA, followed by Rhodococcus, Agromyces, and Mycobacterium (Fig. 3a). Across all 24 species in the first species pool, SIA values followed a power-law distribution (Y = 0.00024‧x–3.007, P < 0.05, R2 = 0.983). All 24 species had positive SDA values, with bootstrapped SDA values greater than zero (P < 0.05; Fig. 3b). Across the 300 richness gradients, each of the 24 species appeared in 41.7% ± 2.5% of all assemblages, confirming that the random assembly procedure produced approximately equal representation of all species. Micrococcus (SDA, 0.128) and Novosphingobium (0.121) exhibited the highest SDA values, indicating the greatest positive contribution to polyculture biofilm formation, followed by Pseudomonas (0.076), Mycobacterium (0.074), and Agromyces (0.072). SDA values followed a normal distribution (a Gaussian function, Y = 0.540‧exp(–(x–0.060)2/(2‧0.0092)), P < 0.05, R2 = 0.948). There was no positive relationship between the SIA and SDA (R2 = 0.07) after outliers were removed using the boxplot method. (a) Species-inherent ability (SIA) and (b) species-dependent ability (SDA). SIA refers to the biofilm-forming ability of an individual species in monoculture, while SDA represents the contribution of a species to overall biofilm growth in polyculture. SDA was determined by comparing biofilm formation between assemblages with and without a species using bootstrapping with 10,000 resampling iterations. Error bars represent ± 1 standard deviation of the bootstrapped means. In addition, each figure contains an inset depicting the distribution pattern Effects of Richness and Composition on Biofilm Growth Within Each Gradient We did not evaluate the effect of species combination, defined as the grouping of individual species present within a community, on biofilm growth, because richness is synchronized with composition in each gradient. Correlations of biofilm formation with richness, C-SIA, and C-SDA were assessed within each gradient (Fig. S18). Positive, negative, and null correlations (P < 0.05) between richness and biofilm were found from 240 (80.0%), 13 (4.3%), and 47 (15.7%) gradients, respectively. The correlation coefficient was 0.65 ± 0.41 across gradients, exhibiting a pronounced left-skewed distribution. C-SIA had positive, negative, and null correlations with biofilm quantity in 58 (19.3%), 8 (2.7%), and 234 (78.0%) gradients, respectively. C-SDA had positive, negative, and null correlations with biofilm in 87 (29.0%), 5 (1.7%), and 208 (69.3%) gradients, respectively. The correlation coefficients were 0.19 ± 0.33 and 0.31 ± 0.31 for C-SIA and C-SDA, respectively, both exhibiting a bell-shaped distribution. To evaluate richness, C-SIA and/or C-SDA as predictors of biofilm growth, linear regression models were constrained to predict non-negative relationships (Fig. 4). Richness alone accounted for 53.9%±31.5% (261 gradients having a positive coefficient) of the biofilm variation across all gradients. C-SIA and C-SDA individually explained 9.6%±17.3% (142 gradients) and 14.4%±21.1% (178 gradients), respectively. Richness and C-SIA, richness and C-SDA, and C-SIA and C-SDA explained 65.5%±28.6%, 70.9%±25.7%, and 15.8%±22.3%, respectively. The full model, containing all predictors, explained 73.4%±25.1%. Effects of richness (R), community-level species-inherent ability (C-SIA), and community-level species-dependent ability (C-SDA) on biofilm growth across individual richness gradients. The linear regression models constrained coefficients for richness, C-SIA, and C-SDA to be non-negative. (a) Explanatory powers of richness, C-SIA, and/or C-SDA as predictors across all gradients. (b) Numbers of gradients where a predictor had a positive coefficient for the full model (Y = R + C-SIA + C-SDA). (c) Importances of predictors in the full model across the respecitve gradients. (d) Residual z-scores from the full model (box plot) and proportion of positive signs (line plot) at each richness level across the 279 significant gradients (model P < 0.05) The full model was significant in 279 gradients (P < 0.05). Richness, C-SIA, and C-SDA had positive coefficients in 273, 135, and 227 gradients, respectively. When assessing their relative importances in the full model, richness exhibited 9.5 ± 6.2, substantially higher than 2.8 ± 2.5 for C-SIA and 3.4 ± 3.1 for C-SDA across the respective gradients. The residual z-scores from the full model were 0.065 ± 0.882 (52.6% positive), − 0.184 ± 0.900 (38.9%), 0.004 ± 0.990 (50.9%), 0.192 ± 0.994 (61.3%), 0.133 ± 0.933 (55.9%), and − 0.209 ± 0.674 (39.7%) across the significant gradients at the 3-, 5-, 7-, 10-, 15-, and 20-richness levels, respectively. Overall Effects of Richness and Composition on Biofilm Growth There was a positive relationship between richness and normalized biofilm (0–1 scaling for each gradient) throughout all assemblages (Fig. 5a). The linear model (Y = 0.036·x + 0.036, P < 0.05) explained 40.9% of the total variation and indicated a 3.6%-point increase in biofilm formation per additional species. Difference in normalized biofilm between consecutive richness levels (biofilm at a given richness–biofilm at the next lower richness within a gradient) gradually increased from 0.055 ± 0.279 at the 3–5 richness range to 0.188 ± 0.347 at the 10–15 range and then declined to 0.093 ± 0.344 at the 15–20 range (Fig. 5b). Similarly, the proportion of positive differences peaked at around 73% between the 7–10 and 10–15 ranges. (a) Overall relationship between richness and normalized biofilm growth using all assemblages. The symbol * indicates statistical significance (P < 0.05). (b) Biofilm difference between consecutive richness levels (biofilm at a richness level – that at the next lower level within each gradient) and proportion of positive differences Without normalization, richness alone accounted for 24.2% of the variation, C-SIA for 12.7%, and C-SDA for 28.7% across all assemblages (Fig. 6). Richness and C-SIA, richness and C-SDA, and C-SIA and C-SDA explained 37.0%, 52.9%, and 28.7% respectively (Fig. 7b). All predictors together accounted for 52.9% (Fig. 7c). To assess the effect of species combination on biofilm growth, a linear regression was conducted between compositional distance (i.e., difference in composition among assemblages) and difference in biofilm growth (Fig. S19). Compositional distance explained 5.3% of the variation in biofilm difference. However, standardizing compositional distance by the richness difference reduced the explanatory power to 0.9% (Fig. S20). Furthermore, when assessed at each richness level, compositional distance accounted for less than 0.8%. Overall effects of richness, community-level species-inherent ability (C-SIA), and community-level species-dependent ability (C-SDA) on biofilm growth. Biofilm growth was plotted against (a) richness, (b) C-SIA, or (c) C-SDA. A linear regression analysis was performed for each predictor Comparison of the individual and overall assessments. Constrained regression models with (a) single, (b) two, and (c) all three predictors, including richness (R), community-level species-inherent ability (C-SIA), and community-level species-dependent ability (C-SDA), were constructed across all individual gradients and throughout all assemblages. The numbers represent the importance of richness relative to C-SIA or C-SDA Comparison Between the Individual and Collective Assessments The effects of richness, C-SIA, and/or C-SDA on biofilm growth were evaluated individually across all gradients and collectively throughout all assemblages (Fig. 7). Richness exhibited superior performance in explaining biofilm growth when assessed individually than when assessed collectively (R2, 0.539 vs. 0.242). Conversely, C-SIA or C-SDA had lower predictability when assessed individually (R2, 0.096 vs. 0.127 for C-SIA; and 0.144 vs. 0.287 for C-SDA). C-SDA had greater performance over C-SIA in both assessments. Accordingly, the model with richness and C-SDA performed better than that with richness and C-SIA. When evaluating the relative importance of predictors in two-predictor models, richness contributed over twice as much as C-SIA (2.3×) and C-SDA (2.1×) when assessed individually. When assessed collectively, contribution of richness was more comparable to them (1.4× for C-SIA; and 0.9× for C-SDA). Discussion We examined the responses of biofilm growth to increasing species richness within individual richness gradients, each comprising assemblages of 3, 5, 7, 10, 15, and 20 species. The two separate experiments (with 300 and 55 gradients) revealed diverse richness–biofilm relationship forms (Fig. 2). The positive relationship comprised 68.0% and 78.2% of the total forms in the first and second experiments, respectively, with similar odds of encountering positive forms between the experiments. Unfortunately, there are no previous studies that report the frequencies of different richness–functioning relationship forms within ecosystems, making direct comparison with our results impossible. Biofilm overyielding with increasing richness may be driven by different mechanisms, including niche complementarity (resource partitioning and spatial niche differentiation among species), sampling effect (the increased probability of including highly productive species in richer communities), and facilitation (one species enhancing conditions for another) or positive interspecific interactions [6, 8, 15, 21]. Additionally, hump-shaped, U-shaped, null, and negative forms were observed across both experiments (Fig. 2) (16.3%, 6.3%, 5.7%, and 3.7%, respectively in the first experiment). These non-positive forms indicate that adding species does not necessarily exert a linear, additive effect on biofilm growth, and that antagonistic interactions can counteract or even reverse the positive effects of richness in specific compositional contexts. Particularly, the hump- and U-shaped forms suggest that even a few species can alter the trajectory of biofilm expansion along a richness gradient. Therefore, a central finding of our study is that interspecific interactions during biofilm development are highly complex and context-dependent, generating a diverse array of richness–functioning relationship forms even within a single experimental system. This diversity of outcomes—impossible to detect in conventional collective analyses—underscores the importance of examining richness effects within individual compositional trajectories. In this study, composition effects were evaluated from two complementary perspectives: compositional variation, which captures the combined functional potential of the species in an assemblage, and species combination, which reflects the identity of the species present while assuming all species contribute equally to function. The compositional aspect, reflecting the functional potential of individual species within a community, is considered an important determinant of ecosystem properties, e.g., respiration in a microbial community and productivity and nitrogen retention in a plant community [7, 14]. SIA and SDA serve as quantitative proxies for compositional variation, representing each species’s potential contribution to biofilm formation in monoculture and polyculture conditions, respectively. Both SIA and SDA varied substantially among species (Fig. 3). For example, Micrococcus exhibited superior SIA and SDA, whereas Xanthomonas was among the least productive species. Interestingly, numerous species, including Novosphingobium, Pseudomonas, and Mucilaginibacter, exhibited higher SDA, relative to their SIA. Likewise, inherently poor biofilm-formers can display the improved growth fitness in polyculture or stimulate the biofilm formation of other species [16, 22]. Consequently, SIA did not correspond to the respective SDA and they exhibited different distribution patterns due to the presence of the species exhibiting better performance in polyculture (Fig. 3). SIA followed a power-law distribution, where few species showed high proficiency in biofilm formation while many others did not, whereas SDA displayed a normal distribution. Additionally, the bootstrapping analysis revealed that all species had a significant, positive effect on biofilm when grown together, indicating that the presence of each species, on average, increased biofilm formation in polyculture. This aligns with previous observations that positive interactions dominated biofilm development [16, 21, 23]. For example, Jeong and Kim [21] reported that positive interaction accounted for 91% of the total interactions in biofilm communities, whereas 49%, 59%, and 65% at random, lake, and soil microbial communities, respectively. A primary objective of this study was to decouple and assess the effects of richness and composition on biofilm growth. It was unable to evaluate the effect of species combination on biofilm growth for each gradient, because compositional difference is identical to the richness difference within a gradient in our experimental and analytical setups. C-SIA and C-SDA represent the combined influence of their respective abilities from all species present within an assemblage, which were independent of richness. In line with the richness–biofilm patterns, richness was positively correlated with biofilm in most gradients (80%) (Fig. S18). In contrast, C-SIA and C-SDA showed null correlations with biofilm in most gradients (78% and 69%, respectively). We constrained the linear regression models to predict only positive relationships to maintain the expectation that biodiversity-related factors consistently predict ecosystem functioning in a positive direction. Without this constraint, negative coefficients in a minority of gradients can counterintuitively inflate R² by capturing variance in a direction opposite to the ecological expectation, thereby overestimating the model’s predictive capacity under a positive BEF framework. Notably, richness alone explained 53.9% of the variation across gradients, substantially surpassing the variations explained by C-SIA and C-SDA (9.6% and 14.4%, respectively) (Fig. 4). C-SIA and C-SDA together explained 15.8% of the variation, which was less than the sum of their explanatory powers, implying covariation between them. The three predictors together explained 73.4% of the variation, with richness playing a more important role compared to C-SIA and C-SDA (relative importance, 9.5 vs. 2.8 and 3.4, respecitvely). The outperformance of richness over the effect of composition (53.9% vs. 15.8%) suggests richness as the predominant predictor of biofilm growth. The effect of richness on biofilm growth was collectively determined throughout all normalized assemblages (0–1 scaling for each gradient) (Fig. 5a). Richness had a positive relationship with the normalized biofilm, explaining 41% of the total variation. Biofilm growth increased by approximately 3.6% points per additional species, suggesting a potential plateau in biofilm growth in assemblages exceeding 28 species. Normalized biofilm growth appeared to accelerate between 7- and 15- richness levels and then decelerate between 15- and 20- richness levels, compared to the model expectations. Biofilm difference between consecutive richness levels and proportion of positive differences seemed to be greater in the 7–10 and 10–15 richness ranges (differences of 0.146 and 0.188; 73.0% and 72.3% positive, respectively), followed by a rapid decline in the 15–20 richness range (0.093 and 58.3% positive) (Fig. 5b). This pattern indicates that biofilm accumulation is not strictly linear across richness levels: intermediate richness promotes the strongest gains in biofilm growth, whereas higher richness levels yield diminishing returns, potentially reflecting niche saturation or intensified competitive interactions. Furthermore, we assessed the contribution of unknown factors, likely interspecific interactions, in the full model using residual z-scores (Fig. 4d). We postulated that the variability in biofilm within a gradient arises from interactions that the model predictors do not account for. Residual z-score and proportion of positive z-scores appeared to be greater between 10- and 15-richness levels (z-scores of 0.192 and 0.133; 61.3% and 55.9% positive, respectively), with the lowest values observed at the 20-richness level (–0.209 and 39.7% positive). Collectively, biofilm growth exceeded expectations between 10- and 15-richness levels, indicating the presence of a richness range where functioning is relatively superior. These findings suggest a hump-shaped relationship between richness and functioning relative to expectations. Richness accounted for 24.2% of the total variation when collectively assessed without the normalization (Fig. 6a), aligning with previous findings showing positive BEF relationships [7, 10, 11, 24]. Moreover, high-richness assemblages (15 and 20 species) appeared more often among the top 10% of biofilm-forming assemblages (OD ≥ 0.3574; n = 180 out of 1,800), with 50 and 48 occurrences, respectively, while low-richness assemblages (3, 5, and 7 species) were far less common (12, 10, and 20, respectively) (data not shown). These results suggest that communities with more species are more likely to achieve high biofilm growth. C-SDA, capturing 28.7%, was stronger in explaining biofilm growth than richness (24.2%) and C-SIA (12.7%) (Fig. 6). When evaluating C-SIA and C-SDA as predictors at each richness level, the explanatory power of C-SIA gradually decreased from 27.5% to 7.5% with richness. The power of C-SDA ranged from 35.8% to 43.7% between 3- and 15-richness levels before declining to 25.8% at the 20-richness level (data not shown). The superior performance of C-SDA over C-SIA was attributed to the fact that C-SIA was determined beforehand (a priori) in monoculture, while C-SDA was estimated afterward (a posteriori) in polyculture, taking interspecific interaction into account. Species combinations have been shown to influence the development of polyculture biofilms and the growth of individual species within biofilms [16, 22]. We found that assemblages with more dissimilar species compositions tended to produce more dissimilar biofilm quantities (R2 = 0.053), assuming no differences in the functional abilities of individual species (Fig. S19). However, when excluding the richness effect, compositional distance (reflecting species combination) only marginally reflected the biofilm difference (R2 = 0.009) (Fig. S20). Moreover, species combination did not have a strong impact on biofilm growth at each richness level (R2 < 0.008). Our findings indicate that the effect of species combination on biofilm growth cannot be easily predicted from species presence–absence patterns alone, reflecting the complex and context-dependent nature of interspecific interactions during biofilm development. This study evaluated the effects of richness, C-SIA, and C-SDA on biofilm growth individually in each gradient and collectively throughout all assemblages. Richness accounted 54% of the biofilm variation when assessed individually, markedly higher than 24% when assessed collectively (Fig. 7a), in accordance with our simulation results (Fig. 1c). This discrepancy may arise from variations in composition between assemblages within and between richness levels when collectively assessed. The normalization process of biofilm improved the explanatory power to 41%, confirming the influence of biofilm differences between assemblages within richness levels. Additionally, the presence of varied richness–biofilm relationship forms increased the discrepancy. These theoretical and experimental comparisons demonstrate that collective approaches can strongly entangle the effects of richness and composition on functioning. This entanglement can result in substantial variability in the overall richness–functioning relationship, underestimating the predictive capacity of richness. Moreover, richness contributed more than twice as much as the other predictors to biofilm growth when assessed individually, whereas its contribution was comparable to them when assessed collectively (Fig. 7b). The collective assessment undervalued the importance of richness in explaining biofilm growth when considering richness alone or comparing it with the other predictors. Conversely, the composition-based predictors captured a larger portion of the variation in biofilm growth when assessed collectively (R2 values of 0.127 vs. 0.096 for C-SIA; 0.287 vs. 0.144 for C-SDA). Thus, the collective assessment tends to amplify the influence of species in explaining biofilm growth, while diminishing the impact of richness. This study deliberately examined the effects of species richness and composition on biofilm growth within individual richness gradients. While species richness and composition are inherently linked because adding species inevitably changes composition, our approach explicitly quantified their independent effects. We found that all species had a positive effect on biofilm growth and the positive form predominated in the richness–biofilm relationship across gradients. Richness emerged as a superior predictor of biofilm growth, explaining more than 50% of the variation. Additionally, it exhibited a hump-shaped relationship with functioning relative to expectations. We discovered that the effect of richness was substantially underestimated when conventionally evaluated, due to its entanglement with the composition effect. These findings offer valuable insights into how species richness and composition shape biofilm development and provide a framework for understanding microbial community dynamics across diverse ecosystems. Supplementary Information Below is the link to the electronic supplementary material. Supplementary File 1 (DOCX 20.8 KB) Supplementary File 2 (DOCX 1.94 MB) Acknowledgements This research was supported by Basic Science Research Program through the National Research Foundation of Korea funded by the Ministry of Education (RS-2025-25400857). Author Contributions S.J. and T.G.K. conceived the ideas and designed the experiments. J.W.L. and C.W.L. executed the experiments with the supervision of S.J. S.J. analyzed the data and wrote a first draft. T.G.K. completed the analyses and finalized the draft. Data Availability The data underlying this article will be shared on request to the corresponding author. Declarations Competing interests The authors declare no competing interests. Footnotes Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Supplementary Materials Supplementary File 1 (DOCX 20.8 KB) Supplementary File 2 (DOCX 1.94 MB) Data Availability Statement The data underlying this article will be shared on request to the corresponding author.