Three-dimensional biofilm growth supports a mutualism involving matrix and nutrient sharing

Life in a three-dimensional structure such as a biofilm is typical for many bacteria, yet little is known about how strains with different genotypes interact in this context. Here, to systematically explore gene knockdowns across various three-dimensional contexts, we created arrayed libraries of essential-gene CRISPRi knockdowns in the model biofilm-forming bacterium Bacillus subtilis and measured competitive fitness during colony co-culture with a wild type-like parent on different media and at different knockdown levels. Partial knockdown led to a wide range of fitness phenotypes, with targeting of translation-related genes often leading to lower growth rates and rapid out-competition by the parent. Several knockdowns competed differentially in biofilms versus non-biofilm colonies, in some cases due to lack of a particular nutrient in one medium. Cells depleted for the alanine racemase AlrA died in monoculture, but co-cultures survived via nutrient sharing in a biofilm but not in liquid. This rescue was enhanced in biofilm co-culture with a parent unable to produce extracellular matrix, due to a mutualism involving nutrient and matrix sharing. Including alrA, we identified several examples of mutualism involving matrix sharing that occurred in a three-dimensional biofilm colony but not when growth was constrained to two dimensions. These findings demonstrate that growth in a three-dimensional biofilm can promote genetic diversity through sharing of secreted factors, and illustrate the role of matrix production in determining trajectories for biofilm evolution that may be relevant to pathogens and other environmental bacteria.


51
In natural environments, many bacteria grow in dense, three-dimensional multicellular were functionally exploited by multiple species to drive emergent structural and 57 mechanical properties of the biofilm that affected viability [4]. Additionally, the rate at 58 which mutations fix in a population of a given size is higher in microbial colonies 59 compared to well-shaken, liquid cultures [5], suggesting that spatial confinement 60 supports an increase in genetic variation. Spatial confinement dramatically increases 61 the frequency of interactions between nearby cells and thus the potential for coupled 62 evolutionary outcomes, enhancing random genetic drift [6]. However, mechanisms that 63 support genetic diversity in the context of a three-dimensional bacterial colony or biofilm 64 remain underexplored [7].

151
Construction of a knockdown library for probing gene essentiality in B. subtilis 152 3610 153 To study genetic interactions involving critical cellular processes within a biofilm, we 154 constructed a CRISPRi knockdown library in the biofilm-forming B. subtilis strain 3610 155 (Methods). The library contains 302 strains: the 252 known essential genes in B. subtilis 156 strain 168, 47 genes that were initially classified as essential in 168 [39] but later 157 revealed to be non-essential or conditionally essential [29], and three internal controls 158 expressing dCas9 without any guide RNAs (Table S1) [30]. Each strain in the library 159 contains a xylose-inducible copy of dcas9 and an sgRNA targeting the gene of interest 160 (Fig. 1A). In addition, gfp is incorporated at the sacA locus to allow visualization and 161 quantification of the knockdown strain (Fig. 1A). The sacA::gfp strain exhibited similar 162 growth and biofilm wrinkling as a parental unlabeled control on both non-biofilm LB agar 163 and biofilm-promoting MSgg [10] agar (Fig. 1B). We refer to colonies on LB and MSgg 164 as "non-biofilm" colonies and "biofilm" colonies, respectively, although it is important to 165 note that the biofilm definition is nuanced and colonies on LB may have some biofilm 166 characteristics [9].

168
To determine whether CRISPRi can be used to knock down gene expression in non-169 biofilm and biofilm colonies, we engineered a parent strain containing rfp under a 170 constitutive promoter and used CRISPRi to target rfp. In this strain, RFP levels in non-171 biofilm colonies on LB agar plates ranged from 40% (basal knockdown) to ~0% (full 172 knockdown) (Fig. S1A), a comparable range to knockdown of the domesticated strain 173 168 in liquid LB [30]. RFP levels in biofilm colonies on MSgg agar ranged from ~90% 174 (basal knockdown) to ~0% (full knockdown) (Fig. S1A). Thus, CRISPRi is an effective 175 tool to knock down gene expression in non-biofilm and biofilm colonies.

177
High-throughput screening of competitive fitness in a colony 178 We compared the colony-growth phenotypes of GFP-labeled knockdown strains grown 179 either alone or mixed with a control strain modified with xylose-inducible dCas9 (without 180 an sgRNA) and constitutive expression of RFP (henceforth referred to as parent-RFP) 181 that exhibits wild-type-like biofilm formation (Fig. 1A,B). After growing each strain 182 individually in liquid LB, GFP-labeled knockdown strains were spotted either alone or 183 mixed with parent-RFP onto agar plates (Fig. 1C, Fig. S2). Colony phenotypes were 184 quantified using a custom image-analysis pipeline that segmented plates into colonies 185 and computed the ratio of GFP:RFP for each colony; colony size was measured 186 manually ( Fig. 1D, S1B,C; Methods). Each plate included a titration row of colonies 187 grown from mixtures of the parent-GFP strain with the parent-RFP strain at known 188 concentrations from 0% to 100% parent-GFP (Fig. 1D, Fig. S2). Quantification of the 189 titration row closely agreed with the predicted ratio of GFP:RFP at each time point (16,190 24, and 48 h) (Fig. 1D, S1D), indicating that the relative fraction of  in co-culture with the parent-RFP strain can be accurately quantified through 192 comparison of the GFP:RFP ratio with the titration row (Methods). Thus, our screen 193 allows us to quantitatively compare growth as a monoculture to growth in co-culture 194 through this competitive fitness value (Fig. 1C).  Table S2). As expected, when transcription was fully 212 knocked down, fitness was even more compromised: 168 and 143 strains had fitness 213 <0.08 after 16 h of growth on LB and MSgg agar, respectively (Fig. 2B, Table S2). 214 Together, these data demonstrate that even though phenotypes were generally subtle 215 for monocultures grown on non-biofilm-promoting LB agar, screening the library on 216 biofilm-promoting MSgg agar or in competition with a parent strain uncovered 217 phenotypes even under basal knockdown.

219
Several strains competed poorly with the parent even with basal knockdown in both 220 non-biofilm and biofilm colonies (Fig. 2B). Interestingly, some non-essential genes had 221 low competitive fitness. For instance, mapA, which encodes a methionine 222 aminopeptidase, competed poorly in both LB and MSgg colonies, and CRISPRi 223 induction further reduced fitness (Table S2, Fig. S3B). Analysis of DAVID functional 224 annotations of strains with competitive fitness >2 standard deviations below the mean of 225 controls revealed significant enrichment of structural constituents of ribosomes 226 (p=9.8×10 -4 and p=2.1×10 -2 on LB and MSgg agar, respectively). Some of the 227 ribosomal-protein strains that competed most poorly exhibited ~20% lower maximum 228 growth rate than wild type in liquid cultures (Fig. 2C), suggesting that the reduced 229 competitive fitness of these strains is due to their reduced growth rate. Indeed, a 230 reaction-diffusion model of colony growth of a co-culture indicated that a strain's 231 maximum growth rate is a major determinant of competitive fitness, and that the 20% 232 decrease of maximum growth rate in certain ribosomal protein knockdowns is consistent 233 with our experimental measurements of their competitive fitness (Fig. S3C, Methods).

235
By contrast, several strains had fitness in co-culture with the wild-type-like parent similar 236 to controls for basal and/or full induction (Fig. 2B), suggesting that the targeted gene 237 was rendered less essential by co-culture with the wild-type-like parent. DAVID (menH and cytC on LB and aroF and rny on MSgg) in which monoculture growth was 243 clearly compromised by induction but competitive fitness remained high (Fig. S3B). As 244 menH and aroF are involved in synthesis of menaquinone (vitamin K2) and aromatic 245 amino acids, respectively, the high competitive fitness may result from nutrient sharing 246 within the colony. In addition, aromatic amino acid biosynthesis genes were enriched in 247 basal knockdowns with high competitive fitness on LB agar (p=1.6×10 -2 ), potentially due 248 to the presence of aromatic amino acids in rich LB medium but not in MSgg. These data 249 underscore the medium-dependence of gene essentiality in co-culture colonies.

251
To validate these findings, we replicated fitness measurements over time on a subset of 252 strains with the highest or lowest competitive fitness values during basal knockdown on 253 LB or MSgg agar. We found that the competitive fitness phenotype was highly 254 reproducible and relatively stable over two days of colony growth (Fig. 2D,E, S3D,E), 255 highlighting the utility of our CRISPRi library for probing the fitness of essential gene 256 knock down in co-cultures.  Table S2). Nonetheless, we 267 identified 36 strains competed with the parent-RFP strain better on MSgg than on LB, 268 and/or vice versa, under basal or full knockdown (Fig. 3A,B, Table S3). Strains that 269 competed better in one medium compared to the other were statistically enriched for 270 genes involved in amino-acid biosynthesis (p=4.7x10 -6 and p=8.3x10 -5 for basal and full 271 knockdown, respectively), suggesting that some strains benefit from nutritional 272 components specific to one medium (Fig. 3A, Table S3).

274
Despite the undefined nature of LB, it was still possible for many strains to identify 275 candidate components whose addition to the medium with lower competitive fitness 276 might rescue the deficit. We selected 20 knockdowns with medium-dependent fitness to 277 pursue further, many of which displayed fitness differences for both basal and full 278 knockdown (Table S3). The glyA knockdown had higher competitive fitness on LB agar 279 than on MSgg agar, and as hypothesized, addition of glycine to MSgg significantly 280 improved fitness in basal and induced conditions, to levels closely matching fitness on 281 LB (Fig. 3C). Similarly, adding Mg 2+ or Mn 2+ to LB agar at the same levels as in MSgg 282 restored the competitive fitness of mgtE and mntA full knockdowns, respectively, to 283 levels on MSgg (Fig. 3C, S4A). Somewhat surprisingly, even though many of the 284 remaining 17 knockdowns with medium-dependent fitness naturally suggested 285 candidates for a missing nutrient, they did not exhibit increased fitness when the 286 hypothesized nutrient was added to the medium with reduced fitness (Fig. S4B, Table   287 S3), showing that at least exogenous provision of those nutrients is insufficient to 288 complement the medium-specific fitness defect. Together, these results indicate that 289 medium-dependent competition ratios can arise due to both nutrient compositional 290 differences between media and other mechanisms that remain unindentified but 291 highlight potentially important factors in selection during colony growth.

293
Wild-type cells rescue alrA knockdown in a biofilm colony by sharing D-alanine 294 Since growth in a structured community provides opportunities for nutrient sharing and 295 cellular differentiation, we hypothesized that some essential-gene knockdowns would be 296 unable to grow as a colony in monoculture but would fare better in co-culture with wild-297 type-like parent-RFP cells. Across our sacA::gfp essential-gene knockdown library, we 298 did not identify any knockdowns that exhibited robust growth in a colony co-culture but 299 died as a monoculture (Fig. S2, S3, Table S2). (To conform to our inoculation protocol 300 for biofilm cultures in which we used 1 µL of a liquid culture of OD ~1.0 (Methods), our 301 high-throughput screen involved inoculation of each ~2 mm-diameter spot with ~2×10 5 302 cells, likely facilitating the partial growth of some knockdowns that would be hampered 303 in growth from a single cell on plates with xylose.) We found that disruption of the thrC 304 locus prevents wrinkling formation on MSgg, presumably reflecting a growth defect, so 305 we hypothesized that insertion of gfp at the thrC locus might exacerbate growth 306 inhibition due to knockdown of certain essential genes. Thus, we constructed a second,  In the thrC::gfp library, alrA was the only knockdown that failed to form a colony as a 310 monoculture but survived with the parent-RFP strain in biofilms on MSgg agar with 311 xylose (Fig. 4A, S5A,B). AlrA is a racemase that converts L-alanine to D-alanine and is 312 required for cross-linking of the peptidoglycan cell wall (Fig. 4B). Full knockdown of alrA 313 expression during liquid growth in a strain without gfp disruption of thrC led to bulging 314 indicative of cell-wall defects (Fig. 4C, S5C). The alrA strain from the thrC::gfp library 315 managed to grow as a monoculture into a colony similar in size to the inoculation spot 316 on LB-xylose agar but not beyond, as did the alrA strain from the sacA::gfp library on 317 LB-xylose and MSgg-xylose plates. As discussed above, the absence of complete lysis 318 is likely due to the high initial density driving growth of a visible colony (    To test this idea, we deleted the genes encoding both of the main extracellular matrix Our finding that the competitive fitness of alrA decreased over time in a co-culture with 376 parent-RFP in which both strains were matrix-deficient (and hence did not form a 377 canonical biofim) (Fig. 5A) indicates that the mutualism between alrA and the parent-378 RFP strain requires growth in a matrix-capable biofilm. A recent study showed that 379 biofilm colony expansion in three dimensions depends heavily on extracellular matrix, 380 while two-dimensional growth relies more on cell growth and division [43]. To test 381 whether the mutualism between alrA and the matrix-deficient parent was dependent on 382 three-dimensional growth, we grew co-cultures between an agar pad and a coverslip. In  Table S2). We measured competitive fitness at 16, 24, and 400 48 h to identify strains that had increased and relatively stable fitness when they were 401 the sole provider of extracellular matrix relative to competition with the wild-type-like 402 parent-RFP. For each strain, we calculated a mutualism score defined as the fitness 403 increase due to deletion of matrix components from the parent at 48 h compared with 404 the fitness increase at 16 h; a positive score reflects a growing benefit of being the sole 405 matrix provider, and hence implies relatively stable fitness (Fig. 5D, Fig. S6D). We 406 focused on all strains with a mutualism score 2 standard deviations above the mean 407 across controls (>0.22, Fig. S6D).  Since the alrA full knockdown exhibited mutualism in a three-dimensional biofilm but not 425 when growth was confined to two dimensions, we tested whether the other mutualisms 426 were maintained during two-dimensional growth. We grew racE, menE, rnz, and hbs full 427 depletions in individual co-cultures with the matrix-deficient parent-RFP between a glass 428 slide and a coverslip. In each case, the knockdown was out-competed by the parent-429 RFP strain at the growing edge of the colony (Fig. 5F). In sum, these data suggest that 430 matrix-dependent mutualisms generally require growth in a three-dimensions.

432
Here, we created two new libraries of essential gene knockdowns in a biofilm-capable 433 B. subtilis strain, and developed a powerful high-throughput screen of competition in 434 bacterial co-cultures to reveal genetic interactions specific to growth in three-435 dimensional colonies. First, we showed that basal knockdown of some ribosomal 436 proteins reduces competitive fitness with a wild-type-like strain in co-culture colonies 437 (Fig. 2), suggesting a high degree of selection on these genes during colony growth.

438
Second, we found that medium composition can dramatically alter competition (Fig. 3), 439 highlighting the role of the extracellular environment during evolution in a multicellular 440 context. Third, we discovered that knockdown of alrA can be rescued through sharing of 441 D-alanine in a three-dimensional biofilm, a context in which the gene is "less essential," 442 but not in liquid or when growth is confined two dimensions between an agar surface  The stable mutualism that we discovered between the alrA knockdown and a matrix-   Colonies were grown on 1.5% agar plates. For nutrient addition assays (Fig. 3, S4), we 516 supplemented LB with one of the following: 0.5% (w/v) monosodium glutamate, 2 mM 517 MgCl2•6H2O, 50 μM MnCl2•4H2O, 2 mM MgCl2•6H2O, 0.5% (w/v) L-asparagine, 0.5% 518 (w/v) L-aspartic acid, 0.5% (w/v) L-lysine, or 0.5% (w/v) D-glutamic acid, and we 519 supplemented MSgg with one of the following: 0.5% (w/v) L-cysteine, 0.5% (w/v) L-520 glutamine, 0.5% (w/v) L-glycine, 0.5% (w/v) L-serine, or 0.5% (w/v) L-tryptophan. Where  Antibiotics for selection of mutant strains were used as follows: kanamycin (kan, 5 527 μg/mL), MLS (a combination of erythromycin at 0.5 μg/mL and lincomycin at 12.5 528 μg/mL), chloramphenicol (cm, 5 μg/mL), tetracycline (tet, 12.5 μg/mL) and 529 spectinomycin (spc, 100 μg/mL).  Table S1. For library construction, we used 533 SPP1 phage transduction [47]. We used a 168 strain containing Pxyl-dCas9 at the lacA 534 locus (CAG74399) as a donor and wild-type strain 3610 (a gift from Dan Kearns) as the 535 recipient to create the 3610-dCas9 parent strain (CAG74331) using MLS for selection. 536 We then used this 3610-dCas9 parent as the recipient and a strain with Pspachy-gfp at   were used as the inoculum for the monoculture screen. To quantify competitive fitness, 610 a titration row of parent-RFP and parent-GFP mixtures in 10% increments (100% 611 parent-RFP+0% parent-GFP; 90% parent-RFP+10% parent-GFP, etc.) was added to 612 each plate (Fig. 1D) plates (with 35 mL of medium poured on a level surface for co-cultures, or 50 mL of 617 medium for monocultures), without and with xylose. We used the "spot many" protocol of the Singer ROTOR HDA to mix the wells before spotting and transferred 12 times 619 from the source liquid plate to the target agar plate. For some assays, a RAININ 620 Benchsmart 96-well pipetting robot was used rather than the Singer ROTOR HDA to 621 pipet 1 µL onto the agar plates. Agar plates were incubated at 30 °C and placed in a 622 box or were loosely covered in plastic to reduce drying.

624
When screen outliers (Fig. 2D, 3C) were replicated, strains were streaked for single were maintained for the duration of the experiment, using the "manual" mode on the 638 camera. The EOS Utility software was used to run the camera. Plates were imaged 639 colony side up to avoid imaging through the agar. Images were analyzed using FIJI and 640 scored as "grew outside original spot", "did not grow outside original spot", or "died and/or threw off suppressors." The ones classified "died and/or threw off suppressors" 642 were assigned a colony size of 0. Suppressors were identified based on off-center 643 colonies, often in flower petal-like arrangements in which one or a few cells within the 644 original spot eventually grew but the majority of the cells did not. Colony size was 645 measured manually in FIJI by drawing a diagonal line across the diameter of the colony. where I is the GFP/RFP ratio, G is the fraction GFP, and a and b are fit parameters (Fig.   660   1D). The fit parameters from the titration row were used to map the library data and 661 assign assign GFP fractions; the data were normalized so that the average of the 662 internal parent-RFP and parent-GFP co-culture controls was 1.

663
CRISPRi rfp knockdown 665 Wild-type 3610, the parent-RFP, and the CRISPRi-RFP strains were cultured in 5 mL 666 test tubes at 37 °C to an OD600~1 in liquid LB. The parent-RFP strain was spotted onto 667 LB and MSgg agar plates without xylose, while the CRISPRi-RFP strain was spotted 668 onto LB and MSgg agar in 12-well plates containing 0.0005% to 1% xylose. The RFP 669 fluorescence of the colonies was imaged as described above, and FIJI was used to 670 quantify the fluorescence intensity of each colony, using wild-type 3610 as a blank. Wild-type 3610, the parent-GFP, and the parent-RFP strains were cultured in liquid LB 675 to an OD600~1 at 37 °C. One microliter of each culture was spotted onto LB or MSgg 676 agar in a 6-well plate. Colonies were cultured for 48 h at 30 °C in a plastic bag with a 677 wet paper towel to increase humidity. Colonies were imaged using the DSLR setup as  To determine how competitive fitness in a co-culture colony is affected by differences in 688 growth rate, we simulated a reaction-diffusion model in which two cell types with 689 densities ( ! , = 1,2) are inoculated in a circular spot from which they spread randomly 690 in two dimensions to compete for fresh nutrients ( ) and grow with distinct maximal 691 growth rates ( ! ), according to the following equations: 694 " is the cell diffusivity, $ is the diffusivity of nutrients, is a conversion factor dictating 695 how nutrients lead to cell growth, and cell growth is limited by nutrients when ~ or 696 less. Initially = % everywhere and, to represent the initial pinning of cells to the agar 697 surface, = % within a disc of radius % and outside of this disc = 0.
where ! is the ratio of ! compared with its value if cells of type i consume all available 704 nutrients, and is the ratio of nutrient compared with its initial value (so % = 1). So, the 705 governing dimensionless parameters are knockdowns compared with wild type (Fig. 2C). We estimated the initial areal density of 712 cells compared with their saturation density to be Cultures of the alrA knockdown strain (HA420) and the parent strain (HA2)  The mutualism screen (Fig. 5D) was performed as described above, except two screens 773 were performed side by side: one in which each strain in the sacA::gfp library was co-774 cultured with the parent-RFP strain (HA12), and one in which each strain was co-cultured with the ∆epsH ∆tasA parent-RFP strain (HA825). These screens were carried 776 out on MSgg+1% xylose plates with a titration row of HA773 (parent-GFP) in 777 combination with either HA12 or HA825, as described above.

779
Qualitative fitness determination via dilution streaking 780 Strains were inoculated from a fresh colony into 5 mL LB and incubated at 37 °C for ~5 781 h on a roller drum. Cultures were streaked onto agar plates using sterile wooden sticks.