Results

Synthetic Clostridium promoters which are too strong in E. coli prevent standard combinatorial assembly

Initially, we set out to use the standard Start-Stop Assembly system [12] to generate a plasmid-encoded combinatorial metabolic pathway library for C. acetobutylicum. This would entail hierarchical library assembly in E. coli, using standard Start-Stop Assembly vectors for most steps, then for the final assembly step using a shuttle vector suitable both for E. coli and for transfer to the target organism, in this case C. acetobutylicum.

We began by identifying sets of promoters spanning a range of strengths that would be needed. Libraries of synthetic promoters for Clostridium were previously generated [19] based on the approach pioneered by Jensen and Hammer [20] Bases were randomized at 39 core positions in the promoter of the C. acetobutylicum thiolase gene (P_thl_), excluding the -35 (TTGATA) and -10 (TATAAT) elements, which were preserved (note these are nearly identical to the E. coli consensus -35 (TTGACA) and -10 (TATAAT) elements). Synthetic promoters from the library obtained had a wide range of transcription strengths (260-fold range, from 0.7% to 178% activity of P_thl_) [19]. Some Clostridium promoters are very strong in E. coli, which can cause deleterious effects in E. coli cells, making DNA constructs containing such promoters difficult to construct and maintain in E. coli, and negatively affecting the diversity and quality of combinatorial libraries assembled in E. coli. For example, during our previous promoter library study [19] it was difficult to construct pPM4 in E. coli, in which the gusA glucuronidase reporter gene was under the control of P_thl_, and many colonies obtained contained plasmids with mutations. The Nottingham 2018 iGEM team described unsuccessful efforts to construct a similar P_thl_-gusA plasmid [21] Therefore, we tested whether fifteen promoters from two libraries previously characterized in C. acetobutylicum [19] pPM36-79% and pPM36-58% (where the % number is the chance of each position varied in the library matching the parent promoter in each variant) showed activity in E. coli by performing glucuronidase reporter assays (Fig. 1). All tested promoters were active in E. coli and the set displayed a much narrower distribution of strengths (2-fold change, with the weakest and strongest promoters showing, respectively, 72% and 139% activity of P_thl_) than the same promoters in C. acetobutylicum. The same reporter vector was used previously during generation of synthetic promoter libraries, which showed that the system can measure a wide range of promoter strengths in E. coli, including completely inactive promoters [19].

Figure 1: Activity of selected synthetic promoters from pPM36-79% and pPM36-58% plasmid libraries in C. acetobutylicum ATCC 824 and E. coli NEB5-alpha. Reporter glucuronidase activity of mid-exponential phase cells was normalized to the activity of a positive control strain transformed with plasmid pPM36-Pthl (gusA under Pthl). A strain containing empty vector pMTL83122 was used as a negative control. Data is presented using a logarithmic scale. Error bars represent standard deviations of three independent experiments.

Next, we tested the feasibility of using these Clostridium synthetic promoters for combinatorial assembly. Six synthetic promoters from the 79% library (79%–26, −25, −33, −29, 52, and − 20) showing 70-fold change between the strongest and the weakest promoters were chosen and cloned in standard Start-Stop Assembly Level 0 format, meaning as α-β fusion site parts in the pStA0 storage plasmid (Table S1) in which the cloning site is flanked by E. coli terminators to prevent potential read-in or read-out. These promoter parts were then used in Level 1 assembly reactions with previously reported synthetic RBSs R1-R6 (plasmids pGT330-335), terminators (plasmids pGT337-340) [12] and a variety of different CDSs (eyfp, adhE2, bktb, and ter; plasmids pGT431, pPM921, pPM924, and pPM934, respectively). After assembly reactions, the assembly reaction products were used to transform E. coli NEB5-alpha, and transformed cells were plated on LB agar supplemented with X-gal. Blue-white screening of the transformants showed that the apparent efficiency of Level 1 plasmid assembly was very low (0%–2% of white colonies per plate), whereas non-expression controls in which the promoter parts were replaced with a short spacer sequence (Spacer 1 α-β [12]) showed much higher assembly efficiency (>98% white colonies) typical for Start-Stop Assembly [12]. This suggested that assembly was occurring normally, but constructs with burdensome strong synthetic Clostridium promoters were not viable in E. coli, so transformants containing those constructs did not grow to form white colonies.

The above observations indicate that Clostridium promoters which are too strong in E. coli, including the synthetic promoters studied here, are unsuitable for standard combinatorial assembly, because the expression units formed at intermediate steps in E. coli would be too burdensome. This problem is not inherently limited to the promoters tested here, and may represent a more general barrier to combinatorial assembly for Clostridium and potentially other organisms.

A new approach: integration-coupled activation of promoterless sequences

To address the potentially general barrier to combinatorial assembly arising from excessive promoter strength in E. coli, we set out to develop an alternative approach.

When using standard Start-Stop Assembly to construct operons, only the Level 1 expression unit for the first CDS in each operon requires a promoter, and only the Level 1 expression unit for the last CDS in each operon requires a terminator. Other (mid-operon) CDSs do not receive a promoter (or terminator) at Level 1 assembly. As a side effect, such mid-operon CDSs are not transcribed and therefore do not cause metabolic burden until a later assembly step (Level 2 or 3, depending upon the design), when they are placed downstream of a promoter. For our new approach, we generalized this effect by using a design in which every CDS, crucially even the first CDS, lacks a promoter or terminator, so the complete hierarchical assembly results in an entirely promoterless operon. This approach avoids transcription and thus associated metabolic burden throughout. Each CDS still receives an RBS, and mixtures of RBSs of varying strengths can be used to generate combinatorial libraries in which expression levels are varied. To be functional, these promoterless operon constructs and libraries must ultimately be transcribed. To achieve this, our new approach combines promoterless operon assembly with allele-coupled exchange (ACE) [8], a two-step allele exchange method for chromosomal integration of DNA. We used a version of ACE in which the second (final) homologous recombination event activates a promoterless antibiotic resistance marker gene (ermB) by placing it downstream of a promoter located on the chromosome (thl), ensuring cells only acquire resistance to the corresponding antibiotic (erythromycin) if the desired recombination has occurred. By constructing combinatorial libraries as promoterless operons via Start-Stop Assembly in an ACE vector, we avoid expression and burden throughout the entire assembly process. The entire operon is then activated simultaneously by ACE integration into the chromosome of the target organism, downstream of a promoter (Fig. 2). We call this approach integration-coupled activation of promoterless sequences (ICAPS).

Figure 2: Integration-coupled activation of promoterless sequences (ICAPS) approach and ICAPS vector for C. acetobutylicum. RBS (β-γ) and CDS (γ-δ) parts are stored in standard Start-Stop Assembly Level 0 plasmids (resistant to ampicillin/carbenicillin). Level 1 plasmids (resistant to tetracycline) are generated by assembling RBS and CDS parts into modified Level 1 acceptor plasmids (β-δ, plasmids pPM902-pPM910). Up to five Level 1 plasmids and the new ICAPS Level 2 E. coli-Clostridium shuttle plasmids (pPM64, pPM65, chloramphenicol resistant) are used to generate Level 2 plasmids encoding promoterless operons encoding metabolic pathways. C. acetobutylicum is transformed with the Level 2 plasmid (selection with thiamphenicol). The pathway is activated by integration of the assembled promoterless operon into the bacterial chromosome downstream of a promoter (selection with erythromycin). See main text for details.

The existing Start-Stop Assembly system provides short spacer parts for use in place of promoters and/or terminators for construction of operons. Except for the special case of ICAPS in the present study, monocistronic designs are typically preferred, because varying both promoters and RBSs provides a wider range of expression levels, whereas operons provide options for less common cases such as linking the regulation of several CDSs using an operon design with a single regulated promoter. However, in ICAPS, every CDS will always lack a promoter and terminator, so if using standard Level 1 Start-Stop Assembly, spacers would be required instead of a promoter and terminator for every CDS. Therefore, to simplify, save effort and increase assembly efficiency, for ICAPS we constructed a set of nine alternative Level 1 Start-Stop Assembly vectors omitting the standard promoter and terminator positions, thus with β-δ acceptor fusion sites instead of the standard α-ε acceptor fusion sites (pPM902-pPM910, Fig. 2, Table S1). Level 1 assembly using these vectors generates Level 1 expression units without promoters or terminators, without requiring spacer parts, but is otherwise the same as standard Start-Stop Assembly, using standard Level 0 UTR/RBS (β-γ) and CDS (γ-δ) parts, and the same assembly procedure.

In order to implement ICAPS, combining promoterless operon assembly with ACE, we constructed vector pPM64 (Fig. 2), which contains a Level 2 Start-Stop Assembly cassette (the same as pStA212) with the relevant Level 2 (A-Z) acceptor fusion sites [12], in place of the previous multiple cloning site of C. acetobutylicum ACE vector pMTL-JH16 [8]. The key functional features (see Fig. 2) in 5′-3′ order are: the T1 terminator from CD0164 of Clostridium difficile 630 preventing transcriptional read-in from the backbone, a 300 bp fragment of the 3′ end of the thl gene serving as the left (second) homology arm (LHA) for chromosomal integration at the C. acetobutylicum ATCC 824 thl locus, a promoterless ermB gene conferring resistance to erythromycin only after integration, the Level 2 Start-Stop Assembly cassette from pStA212 and a 1200 bp right (first) homology arm (RHA). This arrangement means insertions will be made in the same genomic and transcriptional context as previously used successfully for expression of heterologous sequences [8].

Level 2 assembly is used to insert the promoterless and terminatorless Level 1 expression units into pPM64 in the designed order, resulting in a plasmid (or library) with a promoterless operon ready for integration and coupled activation. C. acetobutylicum is transformed with the assembled Level 2 plasmid or library by electroporation and transformants are selected on thiamphenicol. Initially, thiamphenicol-selected cells contain autonomous plasmids, but single-crossover integrants soon arise by homologous recombination at the large RHA, and grow more rapidly under thiamphenicol selection than the initial transformants which are limited by the plasmid’s pIM13 replicon [8]. Colonies are then re-streaked onto agar plates supplemented with erythromycin to select for cells in which a second recombination event at the short LHA has completed the double-crossover integration, placing the promoterless operon including ermB downstream of the strong chromosomal thl promoter, which is active in both acid and solvent production growth phases [8].

Initial experiments using pPM64 in Level 2 assembly showed frequencies of plasmid transformation into C. acetobutylicum, which were significantly lower than frequencies previously reported for other plasmids using the same pIM13 replicon, such as pMTL85151 [4]. Therefore, we tested if the low frequency could be a result of transcription of the CAC2872 gene located in the RHA affecting the function of the Gram+ replication origin located downstream. We constructed pPM65 by inserting the TT2 terminator from the fdx gene of C. pasteurianum into pPM64 between the RHA and pIM13 replicon. The transformation frequency of the new plasmid pPM65 was improved 3 times when compared with pPM64 and was similar to pMTL85151 (Fig. S1).

Construction and testing of synthetic RBSs using a context-resistant design

Synthetic promoters are often used to vary expression levels in combinatorial designs, but ICAPS reserves transcriptional control to ensure assembled sequences are not expressed during assembly, and are activated upon integration, so promoters cannot be used to vary expression levels of CDSs in ICAPS. Instead, translational control can be used, which requires RBSs with a range of translation initiation rates.

Previously, a library of Clostridium synthetic parts allowing 20-fold change of the translation initiation rates was generated by changing the length of the spacer located between the Shine-Dalgarno (SD) sequence and the start codon in the reporter construct [22]. Another way of generating such libraries is randomization of sequences flanking the consensus SD sequence, which has been successfully used in other organisms [12, 23]. However, these strategies result in RBS parts with highly variable lengths or sequences, and the relative translation efficiency of these RBS parts may be context-dependent, meaning significantly affected by the CDS [24]. Therefore, we set out to generate a library of highly similar, context-resistant, synthetic RBSs showing a wider range of translation initiation ranges than the library reported before [22].

As a starting point for generation of synthetic RBSs, we used a 16 bp fragment of the expression plasmid pMTL83122 encoding the RBS of the thl gene with an NdeI restriction site at the start codon (Fig. 3a). This RBS was previously shown to allow significant expression of a number of genes [8, 19, 25] In order to generate a set of RBSs with different translation initiation ranges we decided to randomize only two consecutive bases, the −1 and +1 positions of the Shine-Dalgarno AGGAGGT sequence, assuming that these changes will impact on the base-pairing interaction between the mRNA and the 16S rRNA and thus the translation initiation process, but the small number of differences will limit context-dependence of the RBS variants (Fig. 3a). Synthetic RBSs were generated as Start-Stop Assembly Level 0 parts (with β-γ sites, standard for RBSs) by inverse PCR using empty pStA0 vector as a template and primers which introduce RBS sequences, giving a set of seven plasmids (pPM926-932) with different mutations in the RBS. These synthetic RBSs were assembled with the FLAG-gusA reporter CDS (from pPM911) in Level 1 assemblies, which were each used to transform E. coli, then plasmid DNA was prepared for each and used in Level 2 assembly with Level 2 ICAPS vector pPM64, resulting in the pPM71 plasmid library of promoterless RBS-FLAG-gusA reporter constructs ready for integration. Constructs were methylated in vitro, transformed into C. acetobutylicum and then integrated into the chromosome by selection on erythromycin and activation of the ermB marker (Fig. 3b). Plasmid integrations were confirmed by PCR and sequencing. Glucuronidase activity assays of mid-exponential phase cultures of recombinant strains showed 65-fold change in the gusA expression between the strongest and the weakest RBS, O2, and O11, respectively (Fig. 3c). The strongest RBS, O2, improved glucuronidase expression 3.7 times when compared with the Start-Stop Assembly formatted RBS from pMTL83122 plasmid (O20). Therefore, we successfully generated a library of highly-similar RBS parts (which are intended to be context-resistant) showing a wide range of translation initiation ranges in C. acetobutylicum tested using a reporter system.

Figure 3: Context-resistant synthetic RBS library in C. acetobutylicum. (a) Nucleotide sequences of RBSs in the thl gene, the pMTL83122 expression plasmid and the synthetic RBS library. SD sequence is underlined. Start codon (ATG, Level 1 β-fusion site) and Level 1 γ-fusion site (CCA) are marked in bold. CAT resulting from the introduction of the NdeI restriction site is marked in orange. The context-resistant synthetic RBS library was generated by randomization of −1 and +1 bases of SD (NN marked in orange). (b) Locus of the thl gene after integration of the pPM71 plasmid library encoding context resistant synthetic RBSs and FLAG-gusA reporter gene. (c) Relative glucuronidase activity of mid-exponential phase cultures with integrated context resistant synthetic RBSs (O1, O2, O4, O11, O20, O28) and FLAG-gusA reporter gene. Activities were normalized to the activity of the strain encoding RBS O20 (Start-Stop adapted RBS from pMTL83122 plasmid). Strain transformed with the empty pPM64 integration plasmid was used as a negative control (Ctrl). Error bars represent standard deviations of three independent experiments.

We also tested the strengths and the range of synthetic RBSs in the E. coli expression system. Plasmid library pPM71 contained promoterless synthetic RBS-FLAG_-gusA_ expression units which could therefore not be used to determine translation initiation rates in E. coli. An additional set of Level 1 plasmids (plasmid library pPM87) was assembled in which the synthetic RBS-eyfp expression unit was transcribed from the inducible rhaBAD promoter. All seven synthetic RBSs were strong when compared with the E. coli RBSs R1-R6 characterized previously [12] (Fig. S2). The set of RBSs showed a much narrower range of strengths in E. coli than in C. acetobutylicum, only 1.7 fold change between the strongest and the weakest RBSs (O28 and O1, respectively).

Assembly and integration-coupled activation of promoterless hexanol pathway operon

To test the ICAPS approach and the new RBSs, we attempted to engineer the CoA-dependant alcohol pathway in C. acetobutylicum, to extend it to produce 1-hexanol. Previously, production of 1-butanol in E. coli was achieved by overexpression of C. acetobutylicum CoA-dependent pathway genes (thl, hbd, crt, bcd-etfAB, and adhE2) [26, 27] and further improved by creating NADH and acetyl-CoA driving forces and replacing the acyl-CoA dehydrogenase (Bcd) with NADH-dependent trans-enoyl-CoA reductase (Ter) from Treponema denticola [28]. Subsequently, the 1-butanol pathway was extended to 1-hexanol, a six-carbon product, by overexpression of the β-ketothiolase (Bktb) from Cupriavidus necator H16 showing specificity for condensation of longer carbon chain substrates [29].

It is not known whether the native acyl-CoA dehydrogenase Bcd is able to reduce hexenoyl-CoA to hexanoyl-CoA. It has been shown that this step can be catalyzed in E. coli by heterologous NADH-dependent trans-enoyl-CoA reductase Ter from T. denticola [29]. However, overexpression of the latter enzyme in C. acetobutylicum from a strong ptb (phosphotransbutyrylase) gene promoter resulted in a decreased butanol/acetate ratio [30]. Moreover, to improve alcohol titers, overexpression of the bifunctional aldehyde/alcohol dehydrogenase may be required. Bifunctional alcohol/aldehyde dehydrogenase AdhE2, used previously in a recombinant pathway in the E. coli, was shown to reduce hexanoyl-CoA to 1-hexanol, but it also showed activity towards acetyl-CoA and butyryl-CoA [29], which are intermediates in the pathway. Therefore, fine tuning of Bktb, Ter, and AdhE2 expression levels is required to generate a strain with a functional 1-hexanol pathway with flux towards the final product rather than ethanol or butanol, making this an interesting and challenging case for the design or combinatorial development of a metabolic pathway-encoding construct (Fig. 4). Previous work on hexanol production in Clostridium spp. has focused on fermentation conditions of species already capable of producing hexanol natively, such as Clostridium carboxidivorans, including in co-culture with Clostridium kluyveri [31–33]. Pathways from known hexanol producers have also been integrated into other Clostridium spp. [34].

Figure 4: Recombinant 1-hexanol pathway in C. acetobutylicum. The native butanol pathway is extended to produce a six carbon chain alcohol by recombinant expression of β-ketothiolase (BktB) from C. necator H16, trans-enoyl CoA reductase from Treponema denticola (Ter) and overexpression of C. acetobutylicum bifunctional aldehyde-alcohol dehydrogenase (AdhE2). Other enzymes involved are annotated as follows: lactate dehydrogenase (Ldh), pyruvate:ferredoxin oxidoreductase (Pfor), phosphotransacetylase (Pta), acetate kinase (Ack), bifunctional aldehyde-alcohol dehydrogenase (AdhE1), acetoacetate decarboxylase (Adc), coenzyme A transferase (CtfAB), 3-hydroxybutyryl-CoA dehydrogenase (Hbd), short-chain-enoyl-CoA-hydratase (Crt), butyryl-CoA dehydrogenase (Bcd), phosphotransbutyryalse (Ptb), butyrate kinase (Buk).

The CDSs of bktb and ter were codon-optimized for expression in C. acetobutylicum, and the CDS of adhE2 was recoded (the sequence was similarly regenerated with suitable codon usage) to avoid undesired recombination between the introduced copy of the CDS and the native adhE2 gene on the C. acetobutylicum native pSOL1 plasmid. All three CDSs were synthesized as linear DNA fragments (gBlocks, Integrated DNA Technologies) and inserted into Start-Stop Assembly Level 0 plasmid resulting in plasmids pPM921, pPM924, and pPM934. CDSs were combinatorially assembled with six synthetic RBSs (O2, O4, O11, O20, O24, and O28) in Level 1 plasmids pPM902 (AB), pPM904 (BC), and pPM907 (CZ) and used to transform E. coli to generate Level 1 plasmid libraries. White colonies from transformation plates were pooled and plasmid DNA from each Level 1 library was purified. Randomization of the RBS sequences was confirmed by DNA sequencing (Fig. S3a). Level 1 plasmid libraries were then assembled together in the Level 2 ICAPS vector pPM65 to generate Level 2 plasmid library pPM76, encoding the full-length promoterless combinatorial 1-hexanol pathway. The fidelity of the pPM76 assembly was assessed by picking 10 random white colonies from the transformation plate and restriction analysis of the purified plasmid DNA (Fig. S3b). All tested plasmids showed the expected band pattern showing that the pPM76 plasmid library contains correctly assembled expression units. The remaining white colonies from pPM76 were pooled and used for plasmid DNA purification. The plasmid DNA was methylated and used to transform C. acetobutylicum. The empty pPM65 vector was also transformed to generate control strains not containing the recombinant pathway. Transformants were selected by incubation on agar plates containing thiamphenicol, then colonies were re-streaked onto agar plates containing erythromycin in order to select for integrants in which integration-coupled activation of ermB and the assembled promoterless hexanol operon had occurred (Fig. 5a). Erythromycin-resistant colonies were used to inoculate 10 mL liquid cultures in CBMS medium supplemented with erythromycin and grown for 72 h. The concentration of ethanol, butanol, and 1-hexanol was quantified by GC–MS (Fig. S4). A total of 33 out of 40 strains showed visible growth and alcohol production. Six strains (76-16, -17, -18, -20, -25 and -31) produced detectable amounts of 1-hexanol (from 0.026 mM to 0.048 mM). Interestingly, two other strains (76-2 and -6) showed increased titers of ethanol when compared to the pPM65 empty vector control strains (65-1, -2, and -3). Genomic DNA from selected strains was purified and used as a template for PCR to identify RBSs for each coding sequence in the combinatorial pathway (Fig. 5b). Genotyping showed that the top three 1-hexanol producers had the strongest synthetic RBS O2 in the bktb expression unit, whereas distribution of synthetic RBSs for other genes was more diverse. These results may suggest that condensation of acetyl-CoA and butyryl-CoA to 3-hydroxyhexanoyl-CoA is the rate limiting step in the 1-hexanol pathway. Analysis of RBSs in the ethanol hyper-producing strains showed that in strain 76-2 all CDSs were expressed from weak RBSs (O4 and O24), whereas strain 76-6 had strong RBSs for all three CDSs (O2 and O28). However, the bktb coding sequence in this strain 76-6 was interrupted with a 2 kb fragment of the plasmid backbone, which could explain lack of 1-hexanol production.

Figure 5: Recombinant 1-hexanol pathway in C. acetobutylicum. (a) Locus of the thl gene after integration of the pPM76 plasmid library encoding the combinatorially assembled pathway consisting of adhE2, bktb, and ter genes. Six RBSs of different strengths (O2, O4, O11, O20, O24, and O28) were used for expression of each gene. The intensity of the colour represents the strength of the RBS (based on the assay with the FLAG-gusA reporter). (b) Fermentation experiment using the recombinant strains. Forty erythromycin-resistant strains were grown in CBMS for 72 h. Integrants of the empty pPM65 plasmid were used as controls. Concentrations of ethanol, butanol, and hexanol in the culture medium were quantified using GC–MS. Results are presented for strains producing detectable amounts of hexanol (the limit of detection for hexanol was 0.002–0.005 mM) or showing increased production of ethanol or butanol. Error bars represent standard deviations of three independent experiments. RBSs for each gene were determined by PCR and DNA sequencing. Strain marked with * (76-6) showed unexpected length of the bktb PCR product and large rearrangements in the coding sequence.