We have previously developed a melatonin production strain in E. coli (Luo et al., 2020a). In total, five heterologous enzymes were expressed in the E. coli BW25113 strain to produce melatonin from tryptophan (Figure 1). It is reported that low (1 mM or 232 mg/L) concentration, melatonin inhibits E. coli growth (Lopes and Sourjik, 2018), which agrees with our observation (Supplementary Material). We further tested the growth of an E. coli strain (HMP1741, a host strain for melatonin production) in the presence of various high concentrations (2–6 g/L) of melatonin in the growth medium. Melatonin was added into M9 medium supplemented with 0.2% glucose and the growth curves were monitored. As shown in Figure 2, compared to the control, the final biomass decreased ∼30% at 2 g/L melatonin, and further increasing melatonin concentration to 6 g/L decreases biomass yield by an additional 30%. Note that the control condition was M9 medium containing 4% ethanol, as ethanol was used as a solvent to dissolve melatonin. The data confirmed that melatonin severely inhibits the growth of the E. coli strain at concentrations above 2 g/L. This result allows efficient screening of transporter knockout libraries through differed growth profiles.

Since melatonin toxicity is one of the major bottlenecks during its fermentative production, we sought to mitigate melatonin inhibition by lowering the intracellular concentration of melatonin. We designed a workflow (Figure 3) to identify transporters capable of exporting melatonin in E. coli. It is anticipated that a melatonin importer deletion strain will grow better due to the reduced uptake of melatonin. Conversely, a melatonin exporter deletion strain can exhibit growth defects due to higher intracellular melatonin accumulation.

We have previously gathered an E. coli transporter knockout library (Jindal et al., 2019) predominantly selected from TransporterDB (Elbourne et al., 2017). Some knockout strains were missing from the Keio collection, resulting in a library of 522 transporter knockouts, of which 88% overlap with TransporterDB E. coli transporters. For proteins that are known to form protein complexes to be functional, we only included the knockout strain for a single subunit. Our final library consisted of a total 394 knockout strains from the Keio collection (Baba et al., 2006; Yamamoto et al., 2009). The list of these strains is given in the Supplementary Material. We analyzed the growth of these strains in the presence of 4 g/L of melatonin using plate-based high throughput growth screening (see section “Materials and Methods”). In such screening, higher growth rates and shorter lag phases compared to the wild-type strain can be expected in case a melatonin importer is knocked out, whereas a lower growth rate and a longer lag phase can be expected if a melatonin exporter is deleted (Figure 3).

From the results of the first screening round (in singlet), 30 candidate transporter targets were identified. In a second round, the growth test was repeated for those selected 30 candidates in triplicates. In parallel, we also performed colony PCR to confirm the knockout strain. Two of the 30 strains were excluded because they did not contain the expected gene knockouts. From the second round of screening, we identified five transporters responsible for melatonin tolerance; YhjV, GarP, ArgO, AcrB, and LysP (Table 1). However, we did not identify any importers. Despite being a xenobiotic, melatonin is likely imported by numerous promiscuous nutrient influx transporters, due to its structural similarity to tryptophan. Previous modeling has shown that in the presence of equilibrative transporters a concentrative uptake transporter and an efflux transporter for a given substrate, alterations of influx transporter function do not significantly alter the intracellular concentration of the substrate (Mendes et al., 2020). The existence of numerous unspecific importers that can uptake melatonin is the likely reason that we were unable to identify a single importer knockout in this assay.

Growth curves of the five selected transporter candidate deletion strains in the presence of 4 g/L of melatonin are shown in Figure 4. All five strains have decreased growth rates compared to the wild-type. The most prominent effect resulted from the acrB gene, whose deletion eliminates growth. Two transporter gene deletions, yhjV and garP, increased lag time and reduced growth rates. The lysP and argO deletion strains also showed reduced growth rates but no significant difference in lag-phase compared to the wild-type control strain (Figure 4). In summary, we observed that strains lacking any one of these five transporters are not able to maintain the same level of melatonin tolerance as compared to the control strain. This strongly suggests that these strains have a higher melatonin accumulation and these transporters are critical for the efflux of intracellular melatonin.

To implement the newly discovered transporters for melatonin production, we overexpressed these genes to increase melatonin tolerance in E. coli. Each transporter gene was cloned under the control of a weak constitutive promoter in a low copy number plasmid (Figure 5A). Furthermore, to ensure sufficient expression level without causing too much cellular stress, we fine-tuned the translation efficiency of each transporter. We designed degenerate sequences in ribosome binding sites (RBS) of each transporter gene, which normally give rise to a range of translation initiation rates of the target genes (Salis et al., 2009; Figure 5A, Materials and Methods). The transporter plasmid libraries were transformed into a melatonin production strain. The strains containing RBS libraries were cultivated in the 5 g/L of melatonin over 16 h to enrich the optimal RBS sequences. In the case of AcrA and AcrB that form protein complexes to function, we expressed both genes in an operon.

The enriched populations with RBS libraries were spread in LB-agar (Amp) to obtain single colonies. For each transporter, about 9–16 individual isolates were selected to test their ability melatonin production using a small-scale production assay (Materials and Methods). Strains with optimized expression of GarP, AcrAB, YhjV, or ArgO showed improved melatonin titers, whereas LysP did not show significant improvement with any of the selected RBS sequences (Supplementary Material). We further selected 2–3 colonies for each transporter GarP, AcrAB, YhjV, or ArgO. We confirmed their melatonin tolerance (Figure 5B) and repeated the production assay in triplicates (Figure 5C). One construct of AcrAB overexpression offered a small increase in melatonin titer. The orphan transporter, YhjV, increased melatonin titer by ∼27% when overexpressed. This strain can be selected as the lead strain for further process optimization.

We also transformed the YhjV overexpression plasmid pHM635 from HMP3337 into other melatonin-producing strains to evaluate its effect. Two strains HMP3396 with dihydromonapterin reductase FolM overexpression (Lin et al., 2014) and HMP3072 containing FolM overexpression and anthranilate synthase subunit TrpE mutant S40F (Caligiuri and Bauerle, 1991) were tested (Table 2). As shown in Figure 5D, YhjV overexpression resulted in 13 and 19% increase in melatonin titer for strain HMP3396 and HMP3072, respectively. This suggests that YhjV overexpression can be a general approach to improve melatonin production in E. coli.

Given that 5-Hydroxytryptophan (5-HTP) is also an important tryptophan derivative and an intermediate of melatonin biosynthesis (Figure 1), we also tested the effect of YhjV transporter overexpression in 5-HTP production. The 5-HTP production strain (HMP3403) was generated by overexpressing tryptophan dehydrogenase (TpH) and pterin-4-alpha-carbinolamine dehydratase (Pcd). The strain utilized O₂ and tetrahydromonapterin (MH₄) as co-factors for TpH as described in a previous publication (Luo et al., 2020b). We transformed the plasmid (pH635) with YhjV overexpression into HMP3403, and the resulting strain HMP3404 improved 5-HTP titers by 15% compared to the control HMP3403 (Figure 6) in a small-scale production assay (see section “Materials and Methods”).

Melatonin production strains were derived from BW25113 (Figure 1). To verify gene deletions in Keio collection strains, colony PCR was performed using a primer located in the kanamycin-resistant cassette (ATATTGCTGAAGAGCTTGGC) and a reverse primer located downstream of the targeted gene. A correct knockout strain generates a DNA band in electrophoresis. If no bands were generated, the knockout strains were excluded from the following procedures.

RBS sequence containing degenerate sequences tcttaatcatgcnnkggannkttaacttt (k = g/t) was cloned upstream of the coding sequence of each gene. The plasmid was constructed using USER cloning method (Geu-Flores et al., 2007). The plasmid libraries were transformed into strain HMP2993 (Table 2) together with pHM345 containing some pathway genes and cultured in the presence of 5 g/L melatonin.

For growth tests, E. coli strains were inoculated into 400 μl of M9 + 0.2% glucose media supplemented with required antibiotics in a 96 deep well-plate Then, the plate was incubated at 30°C at 300 rpm overnight (incubated for 18–24 h). The following day, the saturated pre-cultures were diluted 100 fold into 300 μl of M9 + 0.2% glucose media (and antibiotics as required) supplemented with desired concentrations of melatonin in a 96 well MTP plate. Melatonin was purchased from Sigma-Aldrich (>98%). All cultures were protected from light to avoid melatonin degradation (Moussaoui and Bendriss, 2014). Melatonin stock solution was prepared by dissolving 170 g/L melatonin in 75% ethanol. For testing growth inhibition of various melatonin, all cultures containing different concentrations of melatonin were adjusted to have 4% ethanol (Figure 2). Cells growing in M9 + 0.2% glucose media containing 4% ethanol were used as a control. For screening of Keio knockouts collection, 4 g/L melatonin was used. The wild-type strain E. coli BW25113 was used as a control. For testing the RBS libraries of all transporter genes, screening was performed at 5 g/L of melatonin. Culture plates were then incubated in Growth Profiler (Enzyscreen, Heemstede, the Netherlands) at 30°C and 250 RPM for 48 h with constant monitoring of the growth by taking pictures every 20 min. The image data was converted into digital OD600 nm value using the software GP960Viewer version 1.0.0.4 (Enzyscreen, Heemstede, the Netherlands). Data analysis was done using a CROISSANCE package (Schöning et al., 2016).

To measure the production of melatonin, bacterial cells were cultivated in deep-well plates in glucose slow-release (GSR) medium mimicking fed-batch fermentation at 30°C. 1 L of GSR medium contains 15 g of Maltodextrin (dextrose equivalent 4.0–7.0, Sigma-Aldrich), 200U of Amyloglucosidase (Sigma-Aldrich, ∼70 U/mg) for glucose release, 40 g of MES monohydrate (Sigma-Aldrich), 1.2 g of K₂HPO₄, 7 g of Ammonium sulfate, 120 mg of Sodium citrate, 8 mg of ZnCl₂, 12 mg of FeSO₄⋅7H₂O, 9 uM of CaCl₂, 12.5 mM of MgSO₄⋅7H₂O, as well as trace elements and vitamins. The media pH was adjusted to 6.4. The medium was supplemented with 500 mg/L tryptophan, 50 mg/L kanamycin, and 100 mg/L ampicillin. The bacterial cells were cultivated at 30°C in M9 medium supplemented with 0.5% glucose for 24 h. The precultures were diluted 100 times into the GSR medium and cultured at 30°C. After 24 h, the supernatant was collected by filtering the broth with 0.2 μm filters (Pall, New York, United States). The concentration of melatonin in the supernatant was determined by HPLC as previously described (Luo et al., 2019).