1 Introduction

In synthetic biology based on the design-build-test-learn (DBTL) cycle, appropriate tuning of the promoter activity is essential to construct genetic systems [1, 2]. For example, artemisinic acid, the precursor of an anti-malarial drug, has been produced at a 100 mg/ml level in yeast containing an optimized gene network [3]. In a cell, however, the DBTL-based construction and optimization of a gene circuit is a quite time-consuming process. In fact, the construction of an artemisinic acid production pathway involving 8 genes has been estimated to require 150 person-years of work [3,4,5,6]. This is because the assessment of the gene circuits in the cell is dependent on the time for cell growth [7, 8]. In general, when using E. coli as the host, the validation for the promoter activities will require about 4 days because of multiple time-consuming processes, including plasmid construction using cloning strains and measurements for promoter activity using expression strains. Therefore, the amount of time required for such DBTL cycles hampers the rapid construction and evaluation of synthetic gene circuits in cells.

For applications of fine-tuned genetic circuits in cells, cell-free systems that correlate the promoter activity with the cellular systems are in high demand for rapid prototyping platforms, as in the case of breadboards in electronic engineering, because the turnover of DBTL cycles with cell-free systems is generally faster than that with live cells [9]. Recently, cell-free systems have been exploited to design a variety of applications (e.g., genetic code expansion [10], metabolic engineering [11], peptide engineering [12] and biosensors [13]). In the construction of these applications, using the cell-free system circumvents the limitation of waiting for cell growth and allows the rapid evaluation of the biological parts of interest. In the build and test steps in DBTL cycles, the cell-free system facilitates promoter activity evaluation two days earlier than traditional cell culture techniques, by circumventing the requirements for colony formation after transformation and overnight cultivation in liquid media. The DBTL cycle would be accelerated significantly if genetic circuits could be rapidly evaluated using a cell-free system and then transferred into the cell. Therefore, cell-free systems have the potential to function as molecular breadboards [8, 14], although the exploitation of these systems has had some challenges, as described in the following paragraphs.

Considering the capacity limitations derived from potential resource competition at each step of expression [15,16,17], a correlation between the in vitro and in vivo promoter activities can only be achieved under appropriate conditions, where any differences in the production efficiency of the transcription step are not masked by differences at later steps of gene expression, such as translation. The promoter activity, however, is often indirectly evaluated by measuring the protein encoded downstream of the promoter. As a result, only a few reports have succeeded in showing in vitro and in vivo correlations for promoter libraries [7, 8]. Therefore, to utilize cell-free systems as molecular breadboards in synthetic biology, we must validate whether a correlation between in vitro and in vivo systems for gene expression at each step is achieved.

From the viewpoint that measurements in multiple steps of gene expression are important to understand a correlation between in vitro and in vivo systems for a promoter library, we here measured both the in vivo mRNA and protein amounts produced from each promoter in the library used in our previous in vitro work [18]. To investigate this correlation, we measured the in vivo GFP fluorescence for comparison with the in vitro values [18]. Since protein production levels are affected by the limited capacity at each stage of gene expression, it was not clear whether these levels would correlate in vitro and in vivo [19,20,21,22] (Fig. 1). In this study, we found the existence of a translation bottleneck and showed that the in vitro and in vivo expression levels of the T7 promoter library correlate well when this limiting step is circumvented by regulating the transcription levels. Demonstrating this relationship using the T7 RNA promoter, a strong and popular inducible promoter, is critical to constructing cell-free systems as molecular breadboards that mimic a cellular environment and thus will contribute to the future rapid development of genetic circuits.

Fig. 1
figure 1

Scheme of translational bottleneck caused by excess mRNA. Gene expression consists of multiple steps, including transcriptional and translational processes. The capacity at each step can be a bottleneck that hinders the correct measurements of the activities of the genetic parts. A With high induction for RNA polymerase, a translational bottleneck state would be caused by the excess mRNA beyond the capacity of the translational machinery. Such bottlenecks can prohibit the correct measurement of promoter activity when using a reporter gene (e.g. GFP). This is because the differences in transcriptional intensities between variants cannot be reflected in the differences in protein amounts. B With low induction for RNA polymerase, the translational bottleneck is circumvented. Since the lower amounts of RNA polymerase would decrease the amount of mRNA to levels below the capacity of the translational machinery, the differences in transcriptional intensities between variants can be reflected by the differences in protein amounts

2 Results and Discussion

Before comparing the in vivo protein expression levels from T7 promoter variants, we defined the T7-RNA polymerase-induction conditions using the wild-type T7 promoter, which expresses a GFP variant (GFPuv) used in our previous in vitro study [18]. Among the three GFP variants, GFPuv showed clear signals from weak promoters, compared with the other two green fluorescent proteins also used in the study. Because the fluorescence signal of GFPuv is from the soluble protein, rather than the protein in inclusion bodies, we adjusted the culture conditions to ensure the adequate solubility of the reporter protein from the wild-type T7 promoter on the plasmid. Although rhamnose concentrations ranging from 6.1 µM to 6.1 mM induced GFPuv expression at 37 °C, most of the expressed protein was detected in the insoluble fraction (Supplemental Fig. 1A). When we lowered the expression temperature to 20 °C, soluble GFPuv was detected with all of the tested rhamnose concentrations (Supplemental Fig. 1B). GFPuv synthesis was drastically decreased with rhamnose concentrations below 6.1 µM. Although 6.1 µM rhamnose slightly induced GFP expression, the 610 µM conditions we adopted generated a strong GFP-band signal in SDS-PAGE nearly equal to that from the 6.1 mM conditions.

For comparisons of both in vitro and in vivo protein expression levels from a set of promoters, this in vivo study exploited the T7 promoter library that was also used in previous measurements of in vitro transcription [23] and our cell-free protein synthesis system [18]. Based on the cocrystal structure of T7 RNA polymerase with its promoter DNA, this promoter library was constructed by mutating a specific region in the consensus sequence (Fig. 2). In contrast to previous in vitro studies using the promoter library with a wide range of protein and RNA productivities [18, 23], our initial culture experiments with 610 µM rhamnose showed few differences in the protein expression levels among the promoter variants (Fig. 3, blue bars).

Fig. 2
figure 2

GFPuv expression system and T7 promoter library utilized in this study based on our previous in vitro study [18]. The promoter activities differed among the promoter variants. Underlined bases (from − 10 to − 6) play important roles in T7 RNA polymerase interactions

Fig. 3
figure 3

Comparison of cellular protein and mRNA amounts produced from 610 µM rhamnose induction of promoter variants. To determine the in vivo protein amount, we measured GFPuv fluorescence. After each culture harboring a promoter variant plasmid was induced with 610 µM rhamnose, the induced culture was analyzed by flow cytometry. To measure the cellular mRNA by flow cytometry, we modified the plasmid used for the protein quantitation. We fused the Romanesco sequence [24] to the deGFPuv 3'-UTR, which allows the quantification of RNA levels by fluorescence. We normalized both protein and mRNA signals to percentages for wild-type. These data represent the mean ± S.D. of three independent experiments. Among the data, the C−9T/T−8C variant is excluded because this variant shows no activity

Although one possibility for the minimal differences in GFP fluorescence among cells with various promoters using the 610 µM rhamnose induction would be derived from the saturation of GFP solubility in a cell, our analysis did not show this effect (Supplemental Fig. 2). Firstly, the amounts of soluble GFPuv were nearly equal among all variants except for the C−9T/T−8C promoter, which did not produce GFP. In addition to this result, the amounts of insoluble protein from all variants, except C−9T/T−8C, as well as the amounts of intact total protein in cells, were also nearly equal.

To determine whether the transcription or translation step was the bottleneck of GFP production, we quantified the in vivo amounts of GFPuv mRNA produced under the 610 µM rhamnose induction, among the promoter variants with few differences in the in vivo GFP fluorescence in the library. In this work, Romanesco, a Spinach-derived aptamer, was fused downstream of the GFPuv coding sequence. Romanesco (1120 nt) consists of 6 repeated 150 nt aptamers, which are each capable of binding to DFHBI-1T [24]. Although quantifying the mRNA from the cells expressing GFPuv is most desirable, Romanesco and GFP fluorescence cannot be discriminated due to the similar excitation-emission spectra between GFP and DFHBI-1T. We thus used a deactivated GFPuv (deGFPuv) produced by the replacement of Tyr66 with Ala, which disrupts chromophore formation [25].

Even under conditions where there were few differences in cellular protein synthesis between promoter variants in the library (Fig. 3, blue bars), there were clear differences in the levels of in vivo mRNA synthesis between the same variants (Fig. 3, red bars). In the figure, the WT promoter produced approximately three to four times more GFP mRNA than the low-group promoters (T−8C and C−7T), while showing only a 1.4-fold difference in the in vivo protein synthesis.

The different amounts of mRNA and the similar amounts of protein among the variants in the library suggest the existence of a translational bottleneck in the cell, where the translational machinery cannot handle the excess mRNA. The differences in the in vivo mRNA amounts among the promoter variants are consistent with the previous findings where the amounts of both the in vitro mRNA [23] and protein differed among the variants. In spite of the increase in the in vivo mRNA observed between a pair of promoters (e.g. from WT to T−8C), the translational products did not increase. This difference between transcription and translation suggests that the level of synthesized mRNA exceeded the capacity of the translational machinery. Taken together, the excess mRNA generated by a high concentration of T7 RNA polymerase would reveal the translational bottleneck.

To confirm that the bottleneck was at the translation step, we lowered the cellular mRNA levels by reducing the concentration of the inducer for T7 RNA polymerase, and the protein production levels were analyzed among the promoter variants. A qualitative difference in the protein amounts among promoter variants under the 6.1 µM rhamnose conditions was revealed by an SDS-PAGE analysis (Supplemental Fig. 2). To quantitatively measure the differences in the GFPuv fluorescence among the promoter variants under the 6.1 µM inducer conditions, we used flow cytometry. In contrast to the minimal differences in the GFP signals among the promoter variants under the 610 µM conditions (Fig. 4 blue plots), the 6.1 µM rhamnose induction divided the variants into two groups: (i) WT and A−6G, and (ii) T−8C, C−7T and A−10G (Fig. 4 green plots). In other words, there was a strong correlation (R2 = 0.97) between our previous in vitro (horizontal axis) [18] and these in vivo (vertical axis) protein expression levels among the promoter variants in the library when the intracellular mRNA levels were not excessive.

Fig. 4
figure 4

Comparison of in vitro and in vivo protein amounts produced from 610 and 6.1 µM rhamnose induction. Blue circle and green triangle plots show the sets of signals from the 610 and 6.1 µM rhamnose conditions, respectively. In contrast to the 610 µM rhamnose induction, the in vivo protein expression tendency for the T7 promoter library under the 6.1 µM rhamnose conditions was correlated with our previous in vitro study [18]. We normalized both protein and mRNA signals to percentages for wild-type. These data represent the mean ± S.D. of three independent experiments. The Pearson correlation coefficients for the 610 and 6.1 µM conditions were 0.89 and 0.97, respectively

The demonstrated results pave the way toward the validation of cell-free systems to molecular breadboards described in Introduction section. Only a few reports on promoter library profiles have revealed correlations between in vitro and in vivo systems, and even these reports were only for endogenous types of promoters [7, 8]. Because T7 RNA polymerase is widely used in genetic engineering, identifying the in vitro–in vivo correlation of a T7 promoter library is essential to enhance the utility of the breadboard. Although this correlation was demonstrated for parts of the biosystem in this study, such correlations at the whole system level have also been exploited. For example, genetic circuit prototypes in a cell-free system were introduced into cells in previous reports [6, 26]. In addition to the correlations found in whole system studies, the correlation revealed here between the in vitro and in vivo tendencies for the inducible promoter library is important to build a foundation for molecular breadboards.

In addition to our validated E. coli breadboard, achieved by modifying the transcription efficiency, we are also interested in developing cell-free breadboards for other strains and organisms. In this study, the clear correlation of the in vitro and in vivo promoter profiles was achieved because reducing the inducer concentration caused a decrease in the polymerase and mRNA amounts in vivo to circumvent the translational bottleneck. Taken together, although excess amounts of mRNA over the translational capacity had previously disrupted the relationship where the amount of expressed GFPuv is equal to the amount of synthesized mRNA, adjusting the mRNA within the translational capacity restored this relationship. In other words, we identified the conditions to circumvent the translational bottleneck, even when the GFPuv gene was transcribed from the strongest promoter in the library. As an alternative to reducing the inducer concentration, as exploited here, changing the high-copy replication origin on the plasmid to a low-copy origin would be an effective approach to decrease mRNA synthesis. In addition to our in vivo study, an in vitro study also revealed a translational bottleneck [27]. In that study, two cell extract preparations indeed showed different transcriptional efficiencies. However, the expressed protein amounts were equal. This report also suggested that excess mRNA levels beyond the capacity of the translational machinery are responsible for the translational bottleneck. As with other methods to circumvent the bottleneck, increasing the rate-limiting factors for translation in the cells might lead to a clear correlation between the in vitro and in vivo systems [28].

To advance molecular breadboard development in synthetic biology, the sensitivity of real-time RNA quantification must be improved. In this study, we used the Romanesco aptamer. However, direct mRNA measurements under low RNA conditions were hindered due to Spinach's low chromophore binding activity, from which Romanesco is derived. To overcome this issue, tandem Spinach aptamers can be used to enhance the fluorescence per mRNA molecule. Using 64 tandem Spinach increased fluorescence by 17-fold compared to a single one, the method is also effective with other aptamers [29, 30]. Boosting the number of aptamers from the current 6 per mRNA in Romanesco may allow the generation of variants with more efficient fluorescence emission.

3 Summary

In this work, a clear correlation between the in vitro and in vivo protein amounts for a T7 promoter library was achieved, representing an advancement in evaluating the feasibility of using cell-free systems as molecular breadboards. In addition to the E. coli system used here, cell-free systems for non-model organisms with a variety of potentials beyond E. coli have recently been reported [31]. To assess the cell-free systems derived from such non-model organisms for their functions as molecular breadboards, experiments to confirm the correlation between in vitro and in vivo systems for the genetic parts of interest will be extremely important. Similar to the transcription and translation investigations in this work, analyses of each step in gene expression will lead to the identification of bottlenecks, and will also be essential for the validation of breadboards derived from non-model organisms. Once the abilities of these cell-free systems to prototype the gene circuit are validated, these systems will accelerate the DBTL cycle and contribute to advancements in research toward valuable substance production.

4 Materials and Methods

4.1 Induction for Promoter Library

The pIVEX2.3d_GFPuv plasmid variant was introduced into competent KRX E. coli cells ([F´, traD36, ΔompP, proA + B + , lacIq, Δ(lacZ)M15] ΔompT, endA1, recA1, gyrA96 (Nalr), thi-1, hsdR17 (rk−, mk+), e14– (McrA–), relA1, supE44, Δ(lac-proAB), Δ(rhaBAD)::T7 RNA polymerase, Promega). Plasmid construction and mutagenesis methods are shown in the supplemental information. KRX cells harboring pIVEX2.3d_GFPuv were grown in 3 ml LB medium containing 50 µg/ml carbenicillin, at 37 °C and 180 rpm overnight. To express the protein in fresh medium, 30 µl of the overnight culture was diluted into 3 ml LB medium containing 50 µl/ml carbenicillin, and this culture was grown at 37 °C and 180 rpm until it achieved an OD590 = 0.4. Afterwards, the culture was induced with arbitrary concentrations of rhamnose, for 4 h at 37 °C or 20 h at 20 °C.

4.2 Quantification of Cellular GFPuv Using FACS

Overnight cultures of cells harboring each promoter variant plasmid were transferred into shaker flasks with 50 ml LB medium, containing 50 µg/ml carbenicillin, and grown at 37 °C and 140 rpm on an orbital shaker platform until the OD590 = 0.4. These cultures were induced with an arbitrary rhamnose concentration for 20 h at 20 °C. After induction (typical optical density = 1.7), 1 ml of culture was centrifuged for 1 min at 9000×g and the supernatant was discarded. The cell pellet was suspended in 1 ml 1 × PBS and suspension filtered into a 5 ml disposable tube, using a 40 µm Cell Strainer (Becton–Dickinson). The filtered sample was injected into a FACSCalibur flow cytometer (Becton–Dickinson) to measure the fluorescence from cellular GFPuv. FACS parameters for GFP analysis are shown in the supplemental information. In addition to GFP plasmid variants, we used a negative control plasmid, BBa_R0040 (Biobrick pSB1C3), which does not encode GFP and confers resistance to chloramphenicol.

4.3 Quantification of Cellular mRNA Levels Using Romanesco

After induction, as described above for the GFP measurement, 500 µl of the culture was centrifuged for 1 min at 9000×g and the supernatant was discarded. The cell pellet was washed with 1 × PBS and centrifuged for 1 min at 9000×g. The supernatant after centrifugation was discarded and the pellet was resuspended in 1 × PBS containing 200 nM DFHBI-1 T (Lucerna). The resuspended pellet was incubated for 30 min at 20 °C with a 1600 rpm mixing speed, using a Deep Well Maximizer (TAITEC). The incubated sample was filtered into disposable tubes using a 40 µm strainer and placed on ice for 30 min. The fluorescence from Romanesco was then quantified by a flow cytometer set to the same analysis parameters as for the GFPuv measurements, except for the band-pass filter gain. The filter gain was altered from 780 to 900, since the Romanesco fluorescence was weaker than the GFPuv fluorescence.