The effect of modulating the quantity of enzymes in a model ethanol pathway on metabolic flux in Synechocystis sp. PCC 6803

Culture conditions

The glucose-tolerant wild-type strain of Synechocystis sp. PCC 6803 (provided by Dr. Patricia Armshaw, UL, Limerick, Ireland) and transformed strains thereof were were cultured in BG-11 medium (Stanier et al., 1971). Cultures were grown in 100 mL Erlenmeyer flasks containing 20 mL of BG-11 at 30 °C in a Photon Systems Instruments AlgaeTron230 growth chamber having 1% (v/v) CO₂ (using Ecotechnics (UK) CO₂ sensor and controller) and a light intensity setting of 60 µmol photons/m²/s (cool-white LED with added far-red LEDs 735 nM). The orbital Unimax 1010 (Heidolph Instruments) shaker was set to shake at a speed of 210 rpm. Cultures were supplemented with spectinomycin (50 µg/mL final concentration) and IPTG (one mM final concentration) when appropriate. Growth was monitored by following optical density (OD) via absorbance at 730 nm (Infinite 200 Pro spectrophotometer; Tecan Group Ltd., Zurich, Switzerland). E.coli strains were grown at 37 °C in Luria-Broth medium supplemented with antibiotic when necessary at the following final concentrations: spectinomycin 50 µg/mL, kanamycin 50 µg/mL or ampicillin 100 µg/mL.

DNA constructs

All DNA manipulations were performed using BASIC DNA assembly (Storch et al., 2015) and plasmids transformed and propagated in E. coli DH5 alpha. Briefly, DNA parts used for the assembly were amplified using primers containing BASIC Prefix and Suffix and cloned into pJET1.2 vector. The primers used are listed in Table S1. The DNA constructs were assembled using BsaI-digested pJET-storage vectors and appropriate linkers as described in (Storch et al., 2015). BASIC linkers that encoded five different RBS sequences (Table S2) were used to assemble both single gene and operon constructs of varying translational strength. Table S3 contains a list of plasmids constructed in this study.

Synechocystis sp. PCC 6803 transformation

Transformation of Synechocystis sp. PCC 6803 with plasmid DNA constructs was performed as described in (Armshaw et al., 2015) except for the use of glucose. Spectinomycin at a final concentration of 50 µg/mL was used as a selection marker. Chromosomal integration and full segregation were verified by PCR and Sanger sequencing of the obtained amplification products (Source BioScience, UK).

Fluorescence measurements

For eYFP measurements cells were grown up as described previously to OD_730 nm of 0.5 and IPTG was added to a final concentration of one mM. After 7 days, the cultures were diluted 10 times and absorbance and fluorescence readings were taken using the Infinite 200 Pro spectrophotometer (absorption: 730 nm; excitation 503 nm, emission 540 nm). The fluorescence was normalized against wild-type strain control and cell density.

Ethanol measurements

Cells were grown up to OD_730nm of 0.5 and IPTG was added to final concentration of one mM. After 7 days, one mL of every culture was taken, the cells were spun down (4,000 g, 10 min) and the ethanol content of the supernatant quantified by HPLC (1,200 series; Agilent Technologies, Böblingen, Germany). A prepacked HPLC carbohydrate analysis AMINEX HPX-87H column (300 × 7.8 mm, hydrogen form, nine µm particle size, 8% cross-linkage) with industrial grade guard cartridge (30 × 4.6 mm, hydrogen form) (Bio-Rad, Hercules, CA, USA) was heated up to 60 °C. The mobile phase was 5 mM sulphuric acid in water (HPLC grade) with a flow rate of 0.6 mL/min. The signals were acquired with a refractive index detector. A standard curve made up of serial dilutions of absolute ethanol (VWR) was used to determine the quantity of ethanol in each sample.

Sample preparation for SRM Mass spectrometry analysis

Cells were grown up to OD_730 nm of 0.5 and IPTG was added to final concentration of one mM. After 7 days, 15 mL of cultures were harvested by centrifugation (4,000 g, 30 min). The pellets were re-suspended in one mL of extraction buffer (20 mM Tris-HCl pH 8.0, one mM EDTA, two mM DTT). Five hundred microliters of cell suspension were mixed with 500 µL of extraction buffer and washed glass beads (Sigma) in a two mL lysis tube. The cells were disrupted for 10 min at 30 Hz using the TissueLyser II (Qiagen, Valencia, CA, USA). To remove unbroken cells from the lysate the samples were spun down at 12,000 g for 10 min. Protein concentration in the supernatant was estimated using the DC Protein Assay (Biorad, USA). Next, 100 µg of protein was reduced with 10 mM DTT in 50 mM ammonium bicarbonate (final concentrations) for 1 hr at 56 °C, shaking followed by alkylation with 50 mM iodoacetamide (final concentration) for 30 mins at 37 °C in the dark, 500 rpm (Eppendorf thermomixer). Proteomics-grade trypsin (Promega, USA) was added to the protein sample in a 1:50 mass ratio and the samples were incubated for about 4 h at 37 °C (700 rpm, Eppendorf thermomixer). Subsequently, trypsin was added to the samples again in the same ratio and the samples were incubated at 37 °C (700 rpm, Eppendorf thermomixer, Germany) for a further 16 hrs to ensure full digestion of the proteins. The digestion was then stopped by lowering the pH to less than 2 by adding formic acid and incubating for 30 min at 37 °C, 500 rpm. Sample clarification and removal of water-immiscible products were achieved by centrifugation for 10 min at 12,000 g. Supernatants were stored at −80 °C until further SRM analysis.

Analysis of protein by LC-MS/MS

The LC-MS/MS comprised of an Agilent 1100 LC system and an ABSciex 6500 Qtrap MS. Chromatography was performed on a Phenomenex Luna C18(2) column (100 mm × two mm × three µm) at a temperature of 50 °C utilising a gradient solvent system of A (94.9% H₂O, 5% CH₃CN, 0.1% formic acid,) and B (94.9% CH₃CN, 5% H₂O, 0.1% formic acid). A gradient from 0% B to 35% B over 30 min at a flow rate of 250 mL/min was used. The column was then washed with 100% B for 3 min and then re-equilibrated with 100% A for 6 min. Typically 40 µL injections were used for the analysis.

The MS was configured with an Ion Drive Turbo V source, Gas 1 and 2 were set to 40 and 60 respectively, the source temperature to 500 °C and the ion spray voltage to 5500 V.

The MS, configured with high mass enabled, was used in “Trap” mode to acquire Enhanced Product Ion (EPI) scans for peptide sequencing and “Triple Quadruple” mode for Multiple Reaction Monitoring. Data were acquired and analysed using SCIEX software Analyst 1.6.1 and MultiQuant 3.0.

Statistical analysis

For all statistical analysis with samples taken from microbial cultures, a normal distribution was assumed, as discussed in Fay & Gerow (2013). Three biological replicates were employed for every treatment and/or condition. One-way ANOVA was carried out using the one-way ANOVA with post-hoc Tukey HSD (Honestly Significant Difference) test calculator prepared by Navendu Vasavada (http://astatsa.com/OneWay_Anova_with_TukeyHSD/). For the analysis where two factors (RBS in front of PDC, RBS in front of ADH) were evaluated in a combinatorial design we employed the anova2 and multcompare functions of Matlab R2017b. An Excel template provided by Weissgerber et al. (2015) was modified and used for all scatterplots.

Selected RBS variants are functional in vivo and display a wide range of protein expression strength

A total of 5 RBS variants (Table S2) were selected for characterisation via assays of fluorescence emitted by RBS-eyfp constructs. Four RBS sequences have been characterised previously in E. coli, while the fifth is specific to the anti-Shine-Dalgarno sequence of the Synechocystis sp. PCC 6803 ribosome (Heidorn et al., 2011). These sequence variants were contained within linker sequences (“RBS linkers”) used in the BASIC DNA construct assembly process (Storch et al., 2015).

Five constructs were generated for PA1lacO1 promoter-driven expression of eYFP containing one of five 5′-proximal RBS variants (Table S3). The PA1lacO1 promoter was previously shown to be inducible with IPTG in Synechocystis sp. PCC 6803 (Guerrero et al., 2012). All ethanol pathway encoding constructs were designed for chromosomal integration at the well-studied slr0168 neutral site (Angermayr, Paszota & Hellingwerf, 2012).

All five constructs were introduced into Synechocystis sp. PCC 6803 and fully segregated mutants were isolated. The fluorescence of the cultures was measured following a 7-day induction period in the presence of IPTG (Fig. 1). RBS A, which has been reported as weak in E. coli (Salis, Mirsky & Voigt, 2009), turned out to also show minimal expression in the cyanobacterial host strain. The commonly used RBS elements in iGEM projects, RBS B (BBa_B0064) and RBS C (BBa_B0034), displayed intermediate strength. RBS E, which has been chosen as a cyanobacteria-specific RBS, enabled the highest expression of all tested sequences. Based on the eYFP measurements, the tested RBS sequences could be placed in the following relative strength order in cyanobacteria: RBSA <RBS B <RBS D <RBS C <RBS E. This set of RBS sequences satisfied the main objectives with this initial screening, to be capable of modulating protein abundance across a broad range in cyanobacteria.

Creation of a combinatorial library of ethanologenic strains

We wanted to evaluate how the expression and organisation of the pdc and adh genes in an operon would impact both flux and product formation through the synthetic ethanol pathway. Further, we wanted to correlate the output to protein levels and understand the impact on growth. To achieve this we designed an operon matrix with varying levels of expression of the two genetic components governed by the characterised RBS linker sequences used in the construction. In total, 25 constructs were created with all combinations of RBS sequences upstream of the pdc and adh genes. Promoter, terminator and genomic integration locus were the same as for the eYFP constructs described above. Twenty-four out of the 25 constructs were successfully used to transform Synechocystis followed by segregation through serial passage and confirmation by PCR. These mutants were then used to test ethanol production efficiency and quantify the abundance of the pathway components for each combination of RBS variants. The mutant containing RBS D upstream of pdc and RBS A upstream of adh could not be segregated despite repeated attempts.

RBS variation results in substantial differences in ethanol production

The accumulation of ethanol after 7 days of cultivation was assessed by HPLC. As expected, variation in ethanol production was observed among the tested strains (Fig. 2). The strains exhibited minor differences in growth curves and final optical density values with the exception of strains harboring RBS B for PDC and RBS C for ADH (Data S2). Ethanol production was strongly dependent on the RBS sequences for the pdc gene as significant variation in ethanol production was observed between strains harboring different RBS upstream of pdc while having the same RBS upstream of adh. No such trend was observed for variation of the RBS upstream of adh, when the pdc RBS was held constant. Interestingly, RBS A placed upstream of pdc led to higher ethanol production than RBS B upstream of pdc, even though RBS A was the weakest RBS in the context of eyfp. RBS D, which was considered weaker than RBS C when tested with the eyfp gene, provided a higher level of ethanol when placed upstream of pdc than RBS C. However, such differences in RBS activity are not surprising as it is well known that the strength of a chosen RBS sequence depends not only on its own sequence but also on the local DNA sequence context (Salis, Mirsky & Voigt, 2009). Therefore, we cannot assume that changes in RBS will lead to the same variation in protein abundance when coupled with pdc or adh as compared to eyfp.

RBS variants modulate protein abundance

To understand the cause-and-effect relationships better, we analysed the abundance of Pdc and Adh in our ethanologenic strains via selected reaction monitoring (SRM) on an LC-MS/MS (Fig. 3), assuming that each internal enzyme activity at least to some degree correlated with the measured quantity of each protein. Protein abundances were normalised to the abundance of the ATP synthase beta subunit (AtpB, slr1329). From these data, it is clear that RBS strengths can be ranked in the following order: RBS A ≤ RBS B <RBS C <RBS D <RBS E. This relationship holds for both Pdc (Fig. 3A) and Adh (Fig. 3B). When a weak RBS (RBS A or RBS B) was placed upstream of pdc, regardless of the RBS present upstream of adh, both the ethanol yield and relative abundance of Pdc were low. With increasing pdc RBS strength the production of ethanol and the accumulation of both Pdc and Adh all increased. The strength of the RBS upstream of adh, while capable of modulating Adh accumulation, had negligible influence on the abundance of Pdc.

Next, we calculated the correlation between enzyme abundance and ethanol yield and also compared how the abundance of one enzyme correlated with the other. The correlation between protein abundance and ethanol yield was strongest for Pdc (R² = 0.89; Fig. 4A), while for Adh this correlation was much weaker (R² = 0.44; Fig. 4B). Comparing protein levels, Adh abundance was, although weak, slightly more strongly correlated with Pdc abundance (R² = 0.51) (Data S2, Fig. 4 sheet). Together, these observations indicate the operation of “translational coupling” (Levin-Karp et al., 2013) between high-level Pdc expression and Adh.

Enhanced Pdc expression is sufficient for yield-improvement

The weak correlation between Adh and ethanol titre suggested that Adh over-expression was not necessary. In order to test this hypothesis, five new strains were constructed that were similar to the previous 24 strains except that they lacked the adh gene. Once again, five different RBS sequences were placed upstream of pdc. The constructs were transformed into Synechocystis 6803 and fully segregated mutants were obtained for all combinations. Ethanol production and protein abundance were tested, with five of the previously constructed ethanol-producing strains used as controls (AA, BB, CC, DD, EE). Ethanol yields were, greater for the monocistronic (Pdc-only) strains compared with the equivalent RBS bicistronic (Pdc-Adh) strains, when comparing C, D and E with CC, DD and EE, respectively (Fig. 5A). As expected, the Pdc protein abundance was also greater in three out of five mono- vs. double-cistronic comparisons (i.e., A, D, E all had increased Pdc content compared to AA, DD and EE, respectively; Fig. 5B). These observations are consistent with either expression from a shorter transcript being more efficient and/or that there is a reduced metabolic burden placed on the cell as fewer resources are required to express the single protein. Alternatively, the new transcripts may have differed in their secondary structure. Lim, Lee & Hussein (2011) have reported that, in E. coli, proteins encoded on long operons are more highly expressed. This is in contrast to what we observed in cyanobacteria with proteins encoded on longer operons being poorly expressed. Pdc abundance was once again linearly correlated with ethanol yield (Fig. 5C) and can be improved further as there is no evidence of a plateau in ethanol productivity at the protein abundances achieved in these strains.

Further modulation of host metabolism does not improve product yield

Sustainable commercial production of ethanol using cyanobacteria will require increased productivities and product titres. The strong relationship between ethanol productivity and quantity of pyruvate decarboxylase suggested this was the key limiting factor. However, previous studies have found at least some impact from optimization of native metabolism even when the activity of a single pyruvate-dependent enzyme (lactate dehydrogenase) was clearly rate-limiting (Angermayr et al., 2014). We therefore evaluated the effect of overexpressing pyruvate kinase (Pyk) from E. coli, which should improve the supply of the precursor pyruvate as was previously reported (Angermayr et al., 2014). Another possibility is to over-express BiBP, a bifunctional fructose-1,6-bisphosphatase/sedoheptulose-1,7-bisphosphatase from Synechocystis sp. PCC 6803 (Jiang, Wang & Wen, 2012), which has been reported to improve growth and photosynthetic rate in the cyanobacterium Synechococcus sp. PCC 7002 and Synechocystis sp. PCC 6803 presumably by enhancing flux through central carbon metabolism (De Porcellinis et al., 2018; Liang et al., 2018).

To strains already harbouring the pdc-adh cassette with RBS E in front of both genes, we introduced Pyk and BiBP as a second operon at a different genomic locus (slr1395). The new cassette harbouring Pyk and BiBP was combined with four different RBS elements (A, B, D, E) located upstream of each coding sequence and placed under the control of the constitutive P_trc promoter. Fully segregated mutants were obtained for eight out of sixteen designs. However, analysis of these strains revealed no improvement in ethanol production (Fig. 6A) compared to the original strain, despite detectable increases in both Pyk and BiBP abundance (Fig. 6B). In an earlier study with Synechococcus sp. PCC 7002, over-expresison of BiBP resulted in a growth benefit dependent on both high light and high carbon availability (De Porcellinis et al., 2018). In the present study, the over-expression of BiBP had no or a negative impact on growth (Fig. 6C). However, the environmental conditions, the engineering designs and the host strain were different in the two studies, thereby ruling out any possibility to compare between them. Nevertheless, the increase in Pyk and BiBP abundance correlated well with the strength of the RBS used. Again we observed that the increased expression level of the protein encoded first in the operon (pyk) influenced the expression of the protein encoded last (Bibp). For example, this can be seen if we compare the expression of Pyk and BiBP in strains EE_AD & EE_DD.

New insights were uncovered and several challenges remain to be addressed

All DNA constructs used in this study were generated via BASIC methodology (Storch et al., 2015). This method enabled quick and reliable generation of a series of DNA constructs. The main advantage of BASIC is its modularity, where a DNA part once cloned into a storage vector can be used in various assemblies without any further need for amplification. It proved extremely useful in our work involving the preparation of a large number of molecular constructs with different genetic parts, including small RBS elements. In overlap assembly techniques like Gibson (Gibson et al., 2009) such work would involve amplification of each gene with different primers for each construct in order to introduce the RBS sequence and compatible overhangs. Whereas when using BASIC the order of assembly is directed by the employed linkers which may also contain the RBS sequence, for example. Furthermore, in contrast to Golden Gate based approaches, the single-tier BASIC formatted parts can be assembled in any order and in varying numbers without reformatting. The approach we took significantly reduced the workload and enabled us to generate a large number of the required molecular constructs in a very short time. The bottleneck in strain construction was instead shifted to cyanobacteria transformation and segregation. For example, we were unable to isolate a strain containing RBS D upstream of pdc and RBS A upstream of adh. Frequently mooted hypotheses for such observations include metabolic burden of protein over-production and product or intermediate substrate toxicity. We infer from our data that this strain would not have produced higher levels of protein, ethanol or the intermediate acetaldehyde than other strains in the set. It would appear therefore, that some other underlying phenomenon may be at play.

We could also not isolate any strains for the four-cistron operon designs and several attempts were required to generate strains containing tri-cistronic operons. This suggests that there exists a trade-off between integration/segregation efficiency and insert length. To the best of our knowledge there have been no detailed studies performed on the impact of insert length on transformation efficiency in cyanobacteria. Three tri-cistronic operon strains remained only partially segregated, which cannot be confidently attributed to genomic integration locus, burden of protein over-production or substract/product toxicity. Operon lengths in these cases were identical. The presence of specific sequence features within these constructs and their impact on genome stability cannot be discounted (Jones, 2014).

In terms of lessons learned, one of the more striking observations in our data is the evidence for translational coupling. Translational coupling in general is not well characterised, particularly in cyanobacteria. The extent of translational coupling is related to inter-cistronic distance and has been at least partially characterised in other species (Levin-Karp et al., 2013). Despite this being highly relevant for strain design, the distances required to either maximise or abolish translational coupling in cyanobacterial species remains entirely undescribed. This also highlights that, while assembly methods such as BASIC are extremely useful tools for rapid prototyping, other bottlenecks may limit throughput and delivery of fully optimized bioproduction strains.

It has been reported by others that longer operons result in higher protein abundance for proximally-encoded cistrons (Lim, Lee & Hussein, 2011). Their results, demonstrated using E. coli, are not consistent with the data we present here on Pdc abundance: longer operons result in lower protein abundance for proximally-encoded cistrons in Synechocystis sp. PCC 6803. Whether our observations are specific for the ethanol-cassette design or reflect more general differences between the cyanobacteria and other model bacterial species remains to be investigated.

The work also illustrates the importance of limiting factors for metabolic engineering, as also highlighted in the work of Angermayr and colleagues with lactate (Angermayr & Hellingwerf, 2013). At the enzyme level, the first enzyme of the introduced pathway (Pdc) was clearly limiting flux through the pathway and changes to the second enzyme (Adh) did not have much of an impact. At the whole-cell metabolism level, despite successful over-expression of two enzymes previously demonstrated to stimulate either growth or pyruvate-dependent product yield, no or minimal impact on the ethanol pathway was observed. If possible, identification of the key limiting factor(s) that control most of the metabolic flux through an introduced pathway, should first be pursued. With respect to the ethanol pathway, Pdc was that factor. Future studies should investigate how this activity could be enhanced.