E. coli TOP10 (Invitrogen, United States) was used for cloning purposes, while E. coli MG1655 (Blattner et al., 1997) was used as a chassis for testing the genetic circuits. The bacteria were grown in lysogeny broth (LB) (Sambrook and Russel, 2000) at 37°C, with shaking at 120 rpm (New Brunswick C25 Shaker), unless otherwise stated. The antibiotics were used in the following concentrations: ampicillin (250 μg ml⁻¹) and chloramphenicol (25 μg ml⁻¹).

The plasmids used in this study were constructed using the BioBrick assembly method (Shetty et al., 2008). The genetic parts were obtained from the Registry of Standard Biological Parts or designed to contain the prefix and suffix of BioBricks Standard Assembly (RFC 10). The transcriptional factor qdoR, with codons optimized for E. coli expression by JCat software (Grote et al., 2005), was synthesized on-demand (Genscript, United States) with the incorporation of the prefix and suffix of BioBricks Standard Assembly. The promoters of qdoI (P _qdoI ) and qdoR (P _qdoR ) were obtained by cloning of annealed oligos, with the incorporation of the prefix and suffix of BioBricks Standard Assembly (Table 1) into the EcoRI and PstI sites of pSB1C3. P _qdoR corresponds to the sequence between the nucleotides 4,107,952 and 4,107,995, and P _qdoI between the nucleotides 4,107,278 and 4,107,327 in the genome of B. subtilis 168 (Kunst et al., 1997). Seeking to make P _qdoI compatible with the Biobricks RFC10 assembly, an XbaI site in the wild-type sequence was mutated. Hence, P _qdoI was synthetized with the following substitutions: 41 G > C and 42A > T. Both substitutions lay between the -35 and -10 sites and outside the QdoR operators (Supplementary Table 1). From the Registry of Standard Biological Parts, we used synthetic constitutive promoters (BBa_J23114, BBa_J23115, BBa_J23116, BBa_J23105, and BBa_J23110), an RBS (BBa_B0034), a transcription terminator (BBa_B0015), and composite parts (BBa_I13504 and BBa_K1357010) formed by an RBS, gfp or rfp genes, and a transcription terminator. All the plasmids (Table 1) were constructed using pSB1C3 or pSB1A2 as backbone vectors (Registry of Standard Biological Parts), containing a pUC19-derived high copy replication origin and a chloramphenicol or ampicillin resistance marker. The DNA sequences of the genetic parts used in this work are provided in the Supplementary Material.

E. coli MG1655 cells transformed with a plasmid from Table 1 were inoculated and grown overnight at 37°C and 120 rpm. These cultures were diluted (1:100) in 200 µL of fresh LB medium and incubated in 96-well plates (Greiner Bio-One, 96 Flat Clear Bottom Black Polystyrene) to an optical density at 600 nm (OD₆₀₀) of approximately 0.7. Then, different concentrations of quercetin dissolved in dimethylsulfoxide (DMSO) were added. Fluorescence was followed during incubation at 37°C using two different methods: 1) culture directly in a fluorescence plate reader and 2) culture in 96-well plates with incubation in a shaker and cell fluorescence analyzed by flow cytometry.

The fluorescence plate reader was a Tecan Infinite 200 (Tecan, Switzerland). Cultures were performed directly in the reader with 5 mm orbital shaking for 6 h after induction with hourly measurements of fluorescence and OD ₆₀₀ . GFP fluorescence was measured with an excitation wavelength (λ_ex) of 485 nm and an emission wavelength (λ_em) of 535 nm with the gain set at 115 unless otherwise stated.

The flow cytometry measurements were done by first incubating the induced cultures for 4 h, centrifugation of 1 ml of culture (12,000 x g, 1 min, RT), and cell resuspension with TBAC buffer (PBS buffer containing 1 mm EDTA and 0.01% (v/v) Tween 20). The GFP fluorescence was measured in a BD Accuri™ C5 flow cytometer (San Jose, CA, United States) with a 488 nm longpass and a 533/30 nm bandpass filter set. Data were analyzed using FlowJo™ software to obtain FL median (the fluorescence median intensity) and coefficient of variation (CV) values. The CV was calculated as follows: C V   ( % )     =     ( s t a n d a r d   d e v i a t i o n   o f   t h e   s a m p l e ) / m e a n   ×   100 .

E. coli MG1655 cells transformed with plasmids carrying either gfp or rfp were separately inoculated and grown at 37°C and 120 rpm. The cultures were diluted 100-fold in LB medium (10 ml) and incubated in 60 ml flasks until OD₆₀₀ reached ∼0.8. Each culture was diluted to OD₆₀₀ of 0.6, and both cultures (one carrying gfp and another rfp) were mixed in equal proportion. 200 µL of the mixture was transferred to 96-well plates (Greiner 96 Flat Bottom Black Polystyrene) and increasing concentrations of quercetin in DMSO were added. Only DMSO was added to the uninduced control. The cultivation and fluorescence measurement were carried out as mentioned above. GFP fluorescence was measured with an excitation wavelength of 485 nm and an emission wavelength of 535 nm. RFP fluorescence was measured with an excitation wavelength at 540 nm and emission wavelength of 650 nm.

The fluorescence data for each concentration of quercetin were fitted with the Hill function, as follows: ( FL OD 600 ) = y 0 + β [ Q ] n ( [ Q ] n + K 0.5  n ) , where FL/OD₆₀₀ is the specific fluorescence, y₀ is the basal specific fluorescence, β is the relative maximum specific fluorescence, [Q] is the quercetin concentration in µM, n is the Hill coefficient, and K_0.5 is the Hill constant (half-maximal quercetin concentration, µM). The fitting was carried out using the Solver function in Microsoft Excel®. Statistical analyses were carried out using the independent two-sample t-test with the R package (R Core Team, 2020).

To construct genetic circuits responsive to quercetin, we first dissected the QdoR regulatory system of B. subtilis, isolating qdoR, qdoI, QdoR operators, and their promoters (Figure 1A). We reassembled them with synthetic parts (constitutive promoters, a ribosome binding site, and a transcription terminator) and cloned them in plasmids to transform E. coli. The sequence of each genetic part used is provided in Supplementary Table 1. Firstly, we cloned qdoR under the control of its promoter (P _qdoR ) and the reporter gfp under the control of the qdoI promoter (P _qdoI ). Both genes have a synthetic RBS (B0034) at their 5′ flank and a double terminator (B0015) at 3’ flank. Quercetin induced GFP expression in E. coli (Figure 1B), demonstrating that the P _qdoR -qdoR-P _qdoI -gfp circuit was responsive between 20 and 80 µM quercetin. The quercetin induction increased the specific fluorescence intensity (Fluorescence/OD₆₀₀) 33-fold, representing a 4.7-fold increase in the lowest to highest reporter expression compared to a previous quercetin biosensor in E. coli (Siedler et al., 2014). The Hill coefficient for P _qdoR -qdoR-P _qdoI -gfp circuit was 2.54 (R ² = 0.97), displaying a cooperative and ultrasensitive response (Bradley et al., 2016).

As reported by Hirooka (2007), the qdoI promoter has two QdoR operator sites, whereas only one operator was identified upstream qdoR. In turn, to check whether a more repressed state with two operators could amplify the output upon quercetin induction, we rearranged the circuit components putting qdoR under the control of the qdoI promoter (Figure 1C). For the sake of clarity, hereafter, we shall refer to input as the quercetin concentration added to the system and output as the fluorescence level generated by GFP or RFP expression. Although quercetin induced GFP expression, a high basal fluorescence was measured in uninduced E. coli, indicating leakage of the P _qdoI controlling the reporter. The basal fluorescence was 130 times higher than that in the negative regulated P _qdoR -qdoR-P _qdoI -gfp circuit. The specific fluorescence increased only 1.8-fold from 10 to 80 µM quercetin. The Hill coefficient to this plot was 2.61; however, the data from this circuit did not fit so well the sigmoidal function (R ² = 0.89).

The PqdoI controlling qdoR should be more repressed by QdoR than PqdoR, which decreased the QdoR concentration in E. coli and made the reporter module a little repressed. In contrast, P _qdoR was unable to control the reporter module, even when we put qdoR under the control of a strong constitutive promoter, J23110, as the GFP expression controlled by P _qdoR was completely derepressed and non-responsive to quercetin (Supplementary Figure S2). Even adding quercetin above 100 µM to E. coli with the P _qdoR -qdoR-P _qdoI -gfp circuit, we did not reach the maximum specific fluorescence produced by the P _qdoI -qdoR-P _qdoI -gfp circuit. Likely, the circuit P _qdoR -qdoR-P _qdoI -gfp forms negative feedback induced by quercetin, resisting to express more GFP as more quercetin is added. On the other hand, the P _qdoI -qdoR-P _qdoI -gfp circuit leaves the reporter module almost completely unrepressed due to low QdoR expression. In agreement, the basal specific fluorescence of P _qdoI -qdoR-P _qdoI -gfp is high ∼4×10³ AU/OD₆₀₀. Of note, quercetin uptake has already been studied in E. coli W3110, showing that in a solution with 69 µM quercetin, the intracellular quercetin concentration was stable at ∼2.5 µM from 2 to 6 h of incubation (Said et al., 2016). We did not notice the growth effect in E. coli MG1655 with quercetin from 10 to 100 (data not shown).

In summary, two operator boxes, as given by the qdoI promoter, are necessary to increase repression and control the reporter module. On the other hand, the weak repression in the qdoR promoter is necessary to maintain the QdoR levels to control both regulatory and reporter modules. Moreover, manipulating the QdoR expression would make it possible to tune circuits to give different outputs.

We then investigated whether we could get staggered outputs in circuits with the level of QdoR adjusted using constitutive promoters of medium and low transcription strength. By staggered outputs, we mean that the maximum GFP expression of these circuits will reach intermediate values, smaller than those for the unrepressed circuit (P _qdoI -gfp). The circuits were designed with non-feedback regulation since the QdoR expression does not depend on the quercetin concentration. Similar arranges expressing the TetR repressor constitutively were applied to evaluate the output levels and noise generated in E. coli (Dublanche et al., 2006) and Saccharomyces cerevisiae (Nevozhay et al., 2009). Although negative autoregulation has been used to adjust inducer-output response curves and reduce noise in expression (Becskel and Serrano, 2000; Dublanche et al., 2006; Nevozhay et al., 2009), if we apply a circuit where QdoR represses its own expression and the reporter simultaneously, we will obtain a closed loop giving a single input-output response upon induction. On the other hand, by analogy with Ohm’s law (V = I × R), if we design circuits with different resistances (QdoR expression, analogous to the resistance R), they should respond with different outputs (GFP expression, analogous to the current I) to the same input (quercetin concentration added, analogous to the applied voltage). To verify this hypothesis, we put qdoR under the control of five promoters with reference relative expression strengths ranging from 0.10 to 0.33 (Figure 2A, Supplementary Figure S1). The relative strengths were reported previously by J. Christopher Anderson and are available at the iGEM repository dataset (http://parts.igem.org/Promoters/Catalog/Anderson). However, as we cannot guarantee that our experimental conditions were the same as those used by J. Christopher Anderson, we determined the strengths of the synthetic promoters and the P _qdoI and P _qdoR promoters (Supplementary Figure S1). All the circuits detected a minimum concentration of 20 µM quercetin, and the highest fluorescence was reached with 80–100 µM (Figure 2B). The circuit with the lowest resistance (J23114-qdoR) had a 3.5-fold increase in the dynamic range with 80 µM of inducer and the highest cooperativity (n = 3.79, Table 2). The maximum output decreased with increasing resistance in J23115-qdoR and J23116-qdoR circuits, but they had lower reporter leakage in the uninduced state, comparing y₀ values in Table 2. Although the output gain between the uninduced and induced states was 15-fold, the cooperativity was reduced (J23115-qdoR, n = ∼2.1; J23116-qdoR, n = ∼2.4). However, in J23105-qdoR and J23110-qdoR, where the promoter controlling the circuit had in our experimental conditions 6 and 37% of the strength of a sigma 70 consensus promoter in E. coli (J23119), respectively (Supplementary Figure S1), although the circuit was still inducible, it tended to be locked into a non-activable state; 100 µM of inducer gave only 18% of the maximum output measured for the less resistive circuit with the lowest qdoR expression, J23114-qdoR (Figures 2B, C). The medium resistance circuits (J23115-and J23116-qdoR) gave maximum outputs that were around 3-fold higher than those obtained with the higher resistance circuits (J23110-and J23105-qdoR). The cooperativity was also severely reduced (n = ∼1.5). The J23119-qdoR circuit showed a low derepression even above 50 µM quercetin (the maximum output was 0.8 × 10³/OD₆₀₀, Supplementary Figure S3).

In summary, non-feedback circuits based on QdoR can exploit constitutive promoters with transcription forces less than 25% of the strongest consensus promoter, J23119. Above that, the amount of QdoR per cell would become excessive, blocking all QdoR operator sites in P _qdoI , even with a high inducer concentration.

Variation in transcription is a significant factor in generating gene expression noise (Raj and van Oudenaarden, 2008). Considering that the constitutive promoters controlling qdoR used in this work have a wide range of transcription strength, it is likely that some of the non-feedback circuits expressing QdoR would be noisy. We evaluated gene expression noise with the coefficient of variation (CV), given by the ratio of the standard deviation to the mean. The CV values were obtained by flow cytometry analysis of the fluorescence intensity of E. coli cells carrying each expression circuit. For the circuit with negative feedback by QdoR (P _qdoR -qdoR), the CV increased 2.5-fold between the uninduced and the fully induced state with 80 µM of quercetin (Figure 3A). The same increase in CV was also obtained for the circuit with higher resistance (J23110-qdoR) when quercetin was added to the system. In the same way, the CV in the induced state was twice the CV in the uninduced state. Interestingly, for the lower resistance circuit (J23114-qdoR), the CV decreased upon induction, reducing 2.3-fold with 50 µM of quercetin.

We extended the analysis to the other circuits and found that for the circuits with higher resistance to derepress GFP expression (J23110-qdoR and J23105-qdoR), the CV increased from the uninduced to the fully induced state. For the medium resistance circuits (J23115-qdoR and J23116-qdoR), induction reduced the CV by about 3-fold (Figure 3B, Supplementary Figure S4).

Under high repression (Figure 3C), when reporter expression is induced, many QdoR molecules can still bind many operators and block transcription of gfp. Therefore, a small number of cells within the clonal population can reach a derepression level sufficient to express some GFP. An initial highly repressed state makes the population more homogeneous but making it noisy when the input comes. Under low repression (Figure 3D), the scenario is the reverse; there are too few QdoR molecules to repress GFP expression entirely in the uninduced state, which leads to noise. Induction increases the homogeneity of the population since there are sufficient quercetin molecules to bind most of the QdoR molecules. This behavior is what Ozbudak et al. referred to as translational burst, which occurs when a cell population has a low transcriptional rate but a high translational rate (Ozbudak et al., 2002).

Controlling gene expression at different levels and in different cells simultaneously, adding only one inducer to the culture, is a tool with great biotechnological potential. For example, different gene expression levels resulting from the same inducer concentration are valuable for applications with mixed cultures in reactors (Jawed et al., 2019). However, in a mixture of bacteria expressing high and low levels of an inducer-binding repressor, it is necessary to rule out if cells expressing high levels of the repressor would cause an inducer titration effect. Therefore, to assess whether two populations would respond differently to the same input, we mixed two strains of E. coli transformed with different circuits expressing QdoR (Figure 4). Strain 1 carried the low resistance circuit (J23114-qdoR), controlling the expression of GFP, whereas strain 2 carried the high resistance circuit (J23110-qdoR), controlling the expression of RFP (Scheme in Figure 4A). The two cultures were mixed in the same proportion and induced with different concentrations of quercetin. Both circuits were induced, but as seen before, circuits with higher resistance gave lower output, regardless of the inducer concentration. To verify if the results would be affected by the intrinsic characteristics of each reporter, we exchanged them. The circuits kept the same expression pattern, ruling out any such bias due to the reporters GFP or RFP (Figure 4B).