Host-aware RNA-based control of synthetic microbial consortia

Microbial consortia have been utilised for centuries to produce fermented foods and have great potential in applications such as therapeutics, biomaterials, fertilisers, and biobased production. Working together, microbes become specialized and perform complex tasks more efficiently, strengthening both cooperation and stability of the microbial community. However, imbalanced proportions of microbial community members can lead to unoptimized and diminished yields in biotechnology. To address this, we developed a burden-aware RNA-based multicellular feedback control system that stabilises and tunes coculture compositions. The system consists of three modules: a quorum sensing-based communication module to provide information about the densities of cocultured strains, an RNA-based comparator module to compare the ratio of densities of both strains to a pre-set desired ratio, and a customisable growth module that relies either on heterologous gene expression or on CRISPRi knockdowns to tune growth rates. We demonstrated that heterologous expression burden could be used to stabilise composition in a two-member E. coli coculture. This is the first coculture composition controller that does not rely on toxins or syntrophy for growth regulation and uses RNA sequestration to stabilise and control coculture composition. This work provides a fundamental basis to explore burden-aware multicellular feedback control strategies for robust stabilisation of synthetic community compositions.


173
showed that the LLL quorum sensing system, i.e. the system in which C6-HSL binds to the regulator 174 LuxR and activates the lux promoter (pLux), exhibits similar properties to the LRR

227
Workflow for the characterisation of the HSL sender/receiver pairs. Sender strains producing either C6-HSL or 228 pC-HSL, respectively, are incubated for a period of 1 to 6 hours before being centrifuged for their supernatants 229 to be collected every hour and mixed with the appropriate C6-HSL and pC-HSL receiver strains. The response of 230 the receiver strains to the HSL produced by the sender strains is monitored using a plate-reader assay detecting

331
anti-STAR is maximally expressed and GFP capacity is recovered up to 60% of its original value, while 332 growth rate is recovered up to 90% of the original value observed when eforRed is not expressed 333 ( Figure 3B). We note that GFP capacity is not fully recovered when anti-STAR is expressed and that 334 capacity when anti-STAR alone is expressed is lower than when anti-STAR is not expressed. This reflects 335 the cost of expressing the controller species, STAR and anti-STAR (Supplementary Figure 10). To 336 further explore if burden could be used to regulate growth rate, we expressed VioB-mCherry, a large fusion protein previously shown to impose burden on E. coli ( Figure 3C). Cellular growth rate was 338 reduced by 53% when expressing VioB-mCherry. Sequestration by anti-STAR led to restoring the host 339 growth rate up to 90% of the original value measured when STAR is not expressed, i.e. in the absence 340 of the STAR inducer, L-arabinose ( Figure 3C, Supplementary Figure 11). We note that the comparator 341 could only regulate growth rate by tuning VioB-mCherry expression if enough VioB-mCherry was 342 expressed by the system (Supplementary Figure 11) Figures 15D-F, 16). By regulating the expression level of the gRNA, the comparator 358 could reduce cellular growth rate by 64%, and STAR sequestration by anti-STAR was able to recover 359 growth rate up to 80% of its original value when no gRNA was expressed, i.e. in the absence of L-360 arabinose.

361
Taken together, these approaches demonstrate that the STAR and anti-STAR sequestration system 362 can successfully up-and down-regulate growth rate following a quorum sensing input. We envision that the controller could be tuned further via the addition of external molecules, such as L-leucine,

364
which inhibits E. coli K-12 strains growth in the absence of L-isoleucine (Supplementary Figure 17).

401
In Figure 4B, we showed that for the OL cocultures, as we externally supply neither, either or both L-  through down-regulation of the expression of VioB in the slowest growing strain -here the YFP Strain 406 as, in the CL, VioB-CFP is not downregulated, while VioB-YFP is always downregulated (Figure 4C). The

407
YFP strain growth rate is more affected by the expression of VioB than the CFP Strain for two reasons.

408
First, the araBAD promoter is stronger than the rhaBAD promoter (Supplementary Figure 18) and, 409 second, VioB-YFP expression is more burdensome than VioB-CFP expression (Supplementary Figure   410   19). The C6-HSL sender strain Lux5 and the pC-HSL sender strain Rpa5 were chosen as host for the CL 411 circuit as weaker production of quorum sensing molecules did not lead to the stabilisation of the 412 coculture composition around a 1:1 ratio (Supplementary Figures 20 & 21). We note however that

429
The Rpa2 strain is designated as a weak pC-HSL sender, Rpa4 as a medium-strength pC-HSL sender and

430
Rpa5 as a strong pC-HSL sender. As predicted, increasing the production of pC-HSL by the CFP sender   cytometry. OD and fluorescence data were collected using a microplate reader.

467
As a result, coculture composition was brought closer to a 1:1 ratio as pC-HSL production increased,

468
upregulating anti-STAR production in the YFP strains, and thus sequestrating more STAR to prevent 469 the production of VioB-YFP that destabilises the coculture composition. Another parameter that is 470 interesting to tune is the maximal output of the STAR-based comparator by using different toehold 471 domains. To this end, we used the LLL1 and LLL2 comparator designs from Figure 2D to increase the 472 output of the STAR comparator. As the output of the LLL1 comparator is 2-fold higher than that of LLL2,

473
we used LLL2 for medium-strength STAR expression and LLL1 for high STAR expression. Doing so, we 474 observed that as STAR expression increases, the YFP strain is more rapidly outcompeted by the CFP 475 strain as VioB-YFP expression increases ( Figure 5B, Figure 5C). If anti-STAR is not present in high 476 enough concentrations, STAR is not fully sequestered, leading to stabilisation of the coculture at a 2:1 477 ratio, otherwise the composition stabilises around a 1:1 ratio, that is the same as the initial seeding 478 ratio. The results confirm that tuning the input-output properties of the RNA-based comparator by 479 changing the binding affinity of STAR and anti-STAR or by tuning anti-STAR expression, we can modify 480 the composition of a two-member E. coli coculture.

481
Our multicellular controller has the potential to balance burden and production, thus we investigated 482 how our system influences coculture biomass accumulation and product yields when using VioB as a 483 proxy protein as a proof-of-concept. Looking at case 4 of Figure 5B for which both VioB-YFP and VioB-outcompete the YFP strain as no protein of interest is being produced (Figure 5D). Interestingly, the 486 CL coculture diverges from its equilibrium composition after 6 hours, which can be explained by the 487 difference in burden caused by the different anti-STAR designs from the CFP and YFP strains 488 (Supplementary Figure 8). Next, in the presence of L-arabinose only, we observed that the OL 489 coculture ratio is driven out of its equilibrium as the CFP strain quickly outcompetes the YFP strain. The

490
CL coculture however can remain around a 1:1 ratio, keeping a stable coculture composition over time.

491
When the system is induced with both L-arabinose and HSLs, the OL and CL cocultures both stabilise 492 around a 1:1 ratio, demonstrating that the comparator can compensate for the difference in density 493 of the two strains when quorum sensing molecules are present in sufficiently large amounts. When 494 looking at the final density of the cocultures, we observe that when externally inducing the system 495 with L-arabinose, the CL coculture achieves a final density that is 2-fold higher than that of the OL 496 coculture and reaches a similar density (~0.6) than the non-induced OL ( Figure 4C). This increase of 497 biomass accumulation for the CL coculture translates to an 81% increase in total VioB-YFP produced 498 and a 35% increase in total VioB-CFP produced after 24 hours compared to the performance of the OL 499 coculture ( Figure 5E). By repressing VioB-YFP and VioB-CFP expression to balance the coculture 500 composition, the controller effectively allowed both strains to grow better, thus improving biomass 501 accumulation, which in turn led to higher production yields of the protein of interest. When both L-502 arabinose and HSLs were externally added into the medium, VioB-CFP and VioB-YFP expression were 503 inhibited, resulting in little VioB-CFP and VioB-YFP being produced. Finally, we tested whether the 504 STAR-based controller could stabilise population composition when the initial starting ratio was 505 different from 1:1 (Figure 5F). For this, we decided to inoculate our OL and CL cocultures over a range 506 of seeding ratios. After 12 hours, YFP Strain fraction in all OL cocultures had dropped by at least 25% 507 and as much as 55%. For the CL coculture however, the YFP Strain fractions appear to stabilise around 508 their initial seeding ratio, not deviating from it more than 12%. This highlights the ability of our CL 509 system to robustly stabilise coculture composition around the initial seeding ratio of the coculture.

Discussion 511
Here, we present a multicellular control strategy using molecular sequestration to stabilise the 512 composition of an engineered microbial consortia. E. coli was engineered to express three modules for 513 the bottom-up assembly of microbial consortia: (a) a quorum-sensing-based communication module

514
to obtain information about the cocultured strains densities, (b) an RNA-based comparator module to 515 compare the population density of two strains grown in a coculture, and (c) a growth module, which 516 modulates expression of a growth regulator to tune cellular growth and thereby the desired coculture 517 composition. The RNA-based comparator, the first of its kind, can modulate growth rates via burden 518 regulation of either a single protein or a metabolic pathway, but also through essential gene 519 knockdown using CRISPRi. As a result, the genetic circuit, split across the two microbial species, is able 520 to stabilise population composition when their respective protein productions impose a different 521 burden on each host. We used the burden caused by the expression of heterologous genes of interest 522 (GOI) to control the growth of the two cocultured bacterial strains. As such, the growth control 523 mechanism does not rely on consuming additional cellular resources to produce mutagenic toxins that 524 kill the hosts rather that slowing down their growth. Our multicellular gene circuit could stabilise 525 population composition over time compared to a coculture deprived on the controller circuit. In 526 addition, by modulating the burdensome expression of the GOIs, we found that the multicellular 527 controller improved the total production yields by 81% in the slowest growing strain and by 35% in the 528 fastest growing strain. We identified several parameters that can be used to tune community 529 composition: quorum sensing production rate, transcription rate of the GOI, the transcription rate of 530 the reference signal promoter and the binding affinity of the STAR to its target and to its antisense 531 specie, anti-STAR. As such, the platform we developed could provide a means to balance heterologous 532 expression burden and production, leading to better biomass accumulation and production yield in

616
Quorum sensing sender-receiver assay 617 Sender strains and receiver strains were inoculated in 1 mL of rich M9 medium supplemented with the 618 appropriate antibiotics in a 2 mL deep-well 96-well plate (VWR), covered with a "Breathe Easier" 619 membrane (Sigma), and incubated overnight at 30°C in a plate shaker incubator (Infors HT Multitron) 620 shaking at 700 rpm overnight. In the morning, sender strains were diluted to 1:4 in fresh M9 medium 621 in a 1 cm cuvette and OD700 was measured in the spectrophotometer. Cells were diluted to OD700 622 0.05 in 1 mL of fresh M9 medium with antibiotics in a new 2 mL deep-well 96-well plate, covered with 623 a "Breathe Easier" membrane (Sigma), and incubated at 700 rpm, 30°C in the plate-shaker incubator.

624
Every hour for 6 hours, the deep-well plate was taken out of the incubator and new wells were 625 inoculated with 1 mL of OD700 0.05 of each culture. Receiver strains were diluted 1:200 in fresh M9 626 with antibiotics in a new deep-well plate and incubated at 700 rpm, 30°C in the plate-shaker incubator.

627
At 6 hours, 250 μL of each sample from the sender strains deep-well plate was transferred to 1 cm 628 cuvette to measure OD700 of each sender strain culture. The sender strains deep-well plate was then 629 centrifuged at 4000 rpm for 5 minutes (Centrifuge Eppendorf 5810R) and the supernatants were 630 transferred by pipetting in a new 2 mL deep-well plate, diluted 1:1 with fresh M9 with antibiotics and 631 the pellets were discarded. The OD700 of the receiver strains were measured in the 632 spectrophotometer by diluting 1:2 with fresh M9 in 1 cm cuvettes. Receiver strains were then diluted 633 to OD700 0.01 in 1 mL of the diluted sender strains' supernatants. 200 μL of each sample was 634 transferred to a clear flat-bottom 96-well plate and inducers (and dye if dBroccoli was expressed) were 635 immediately added to the appropriate wells. The microplate was covered with a "Breathe-Easy" 636 membrane (Sigma) and incubated in the Tecan Spark for 12 hours. OD and fluorescence were 637 measured every 15 minutes as previously described in the above "plate-reader assay" section.