Figures and data
Logic of the assembly protocol.
(A) We consider a set of m species (in this illustration, m = 8), and we identify each of the 2m consortia that can be formed with them by the binary number c = xmxm−1· · · x2x1 (with xk = 1, 0 representing the presence/absence of species k, respectively; see equation 1). (B) Our protocol is based on the idea that one could assemble each of the 23 = 8 combinations of species 1 to 3 in the first column of a 96-well plate. Iteratively duplicating these consortia and adding species 4 to m to them would eventually result in all 2m combinations of species, as described in the main text. (C) Each of the m species would be present in the indicated wells (see equation 2). (D) Positioning of the 2m consortia within the 96-well plates used in the protocol.
Full factorial construction of synthetic colorant combinations.
(A) We consider a set of eight synthetic colorants, each of which exhibits a different absorbance spectrum in the range 380-780 nm. (B) As an example, we show the empirically measured absorbance spectrum (solid black line) for the combination of colorants 1, 3, and 8 (10000101). The dashed black line represents the additive expectation, i.e., the sum of the spectra of the three constituent colorants (colored dashed lines). (C) Relative deviation between the empirical spectrum and the additive expectation (see equation 6) for the example combination 10000101 (left) and for all mixtures of 2 or more colorants (right). Relative errors were only computed when the true absorbance was above 0.1 A.U..
Co-culturing combinations of bacterial isolates.
(A) We consider eight P. aeruginosa strains obtained from a previous experiment. 33–35 Here we represent the absorbance spectra of the eight monocultures of these strains. Solid lines and shaded areas represent means and standard deviations across three independent biological replicates. (B) We followed our protocol to assemble different combinations of strains (in this example, the consortium formed by strains 2, 3, and 8, represented as 10000110). Each consortium was diluted into fresh medium by a factor of 1:400 and allowed to re-grow for 40 h, after which the absorbance spectrum of the culture was measured. Solid black line: empirical spectrum of consortium 10000110 (mean of three independent biological replicates, shaded region represents the standard deviation across them). Dashed black line: additive expectation if the three strains did not interact. Dashed colored lines: monoculture spectra of the three constituent strains.
Full factorial design of synthetic microbial consortia.
We represent the absorbance spectra of all consortia (of 2 or more strains) that can be formed with our library of eight P. aeruginosa isolates. As in Figure 3B, solid black lines and shaded areas represent the mean and standard deviation of the co-culture spectrum across three biological replicates; the dashed colored lines represent the individual monoculture spectra; and dashed black lines represent the sum of these individual spectra.
Full factorial design of microbial consortia enables the dissection of microbial interactions.
We focus on the absorbance at 600 nm as a proxy for the biomass of a consortium. (A) Consortium function (biomass quantified as Abs600) against the number of strains in the consortium. Dots and error bars represent the average and standard deviation across all consortia with a same number of strains. (B) Top three consortia exhibiting the highest biomass across all 28 combinations. (C) We represent the function (biomass; values shown are averages across three biological replicates) for the consortia assembled using our protocol. Nodes correspond to the different consortia, edges connect consortia that differ in the presence of a single particular strain. (D) Detail showing the effect of including strain 1, strain 2, and both, in a background consortium (in this example, the background corresponds to consortium 11001000). Red: the functional effect of strain 1 is quantified as the difference in function between the background consortium and the consortium resulting from the inclusion of that strain. Green: the functional interaction between strains 1 and 2 in this particular background is quantified as the difference between the additive expectation (dashed line) and the empirical value for the function of the consortium containing both strains, as explained in the main text. (E) Detail of the functional landscape corresponding to the three strains that are part of the overall highest-biomass consortium (00001101). Inset: functional interactions within this 3-strain consortium. Order 1 corresponds to the monoculture functions. Order 2 corresponds to pairwise species-by-species interactions as defined in panel D of this figure. Order 3 corresponds to the third-order interaction between these strains, defined as the deviation between the empirical function of the trio with respect to the null expectation from the additive and pairwise interactions. 19,29 19,29 (F) Distributions of functional effects of the eight strains used in our experiments across all of the 128 backgrounds where each may be included. (G) Distributions of functional interactions between all pairs of strains across all backgrounds where both may be included. (H) Fraction of functional variance due to interactions of different orders (order 1 corresponds to the additive effects of each strain). The fractions were computed as explained in the Supplementary Methods. 30 Dashed line is the cumulative sum. (I) We have recently shown that the functional effect of a strain is often predictable from the function of the background consortium through simple, linear regressions. 10 For the eight P. aeruginosa strains in our experiment, linear regressions do an excellent job at linking these functional effects with the function of the background consortia.
Plate layout at different stages of the assembly protocol for combinations of 8 colorants.
Absolute deviation, |Abs(c) − Abs(add)(c)|, between the spectra of the 8-colorant consortia and the additive expectations, for all consortia c of 2 or more colorants and for all wavelengths λ between 380 nm and 780 nm. Median absolute error is ∼ 0.025 A.U.
Relative deviation between the empirical spectra and the respective additive expectations (equation 6) versus the number of colorants in the consortium. Here we represent the deviations for all wavelengths in the 380-780 nm range.
Relative deviation between the empirical spectra and the respective additive expectations (equation 6) as a function of the wavelength. Wavelengths in the 380-780 nm range were split into ten quantiles. Within each quantile, we represent the relative deviations (equation 6) for all combinations of two or more colorants.
Previous work has expressed the community-level function of a consortium c as 
Correlations between the functional effect of a strain (Abs) and the function of the background consortium where it is included, such as those we report in Fig. 5I, can emerge when the mapping between community composition and function is entirely random. 10 (A) To test whether the empirical correlations we observe in Figure 5I could be attributed to this phenomenon, we randomized our empirical data 100 times. In each of these randomizations, we fit linear models linking the functional effect (ΔAbs) of each of the eight strains to the functions of their background consortia. (B) Here we compare the linear fits obtained in each of these 100 randomization controls (blue lines) with the true best fit to the empirical data (black lines). We find that the empirical fits are generally incompatible with the randomization controls.
