Essential genetic elements for a shuttle vector include a replication origin, expression elements, and a selectable marker. A replication origin is the critical component as it guarantees stable replication of plasmid. Identifying a replication origin in a non-model bacterium typically relies upon two approaches. First, utilize known replication origins (E. coli). A variety of E. coli replication origins that have different functionalities (copy numbers) have been studied and characterized (Baker and Wickner, 1992). Second, test a suspected replication origin from the non-model organism. One strategy is to construct a shuttle vector containing a native cryptic replication origin in conjugation with an E. coli low-copy number replication origin. In this method, shuttle vectors are typically built and propagated in E. coli, purified, and transformed into a heterologous host to generate an engineered strain (Chung et al., 2013). One significant concern of such a dual replication origin shuttle vector may be instability either in E. coli or in the target hosts. A difference in GC-content between E. coli and a Gram-positive host tends to be structurally and segregationally instable in both organisms (Savijoki et al., 1997).

At this point, shuttle vectors containing only one replication origin seem to be superior. Spath et al. (2012) demonstrated that a direct transformation of recombinant expression vectors from PCR/ligation reactions into Lactobacillus plantarum yield a compatible transformation efficiency compared to an approach of propagation in E. coli. Such an approach has three advantages: (1) It does not pose a requirement of an E. coli replication origin; (2) It reduces the amount of bench work; (3) It develops a potential platform for high-throughput screening of a heterogenic pool of mutants. However, the authors also indicated that the transformation efficiency might largely decrease in the presence of restriction modification (RM) systems.

After identifying a replication origin, an expression vector can still be ineffective if it is degraded by intracellular nucleases, especially endonucleases. The active endonucleases in a non-model strain can truncate circular dsDNA at a specific sequence, known as a defensive mechanism, which accounts for the primary barrier for efficient and successful transformation. In nature, methylation at the cutting site is a means for preventing native DNA from host endonuclease. Understanding restriction/methylation (RM) systems on a molecular level (base-to-base) can help develop and improve transformation efficiency. Among E. coli RM systems, Dam (G^mATC) and Dcm [C^mC(A/T)GG] methylation systems are two well-characterized methods. For instance, Clostridium thermocellum exhibits sequence-specific restriction endonuclease activity (GATC) with little non-specific exonuclease activity. Protection from digestion was observed when plasmids were prepared from a Dam⁺ (G^mATC) E. coli strain, while plasmids prepared from a Dam^- E. coli strain were digested by the cell extract cocktail (Klapatch et al., 1996). However, studies also found that additional methylation of plasmid DNA in E. coli can inhibit transformation into the target hosts (Kieser and Hopwood, 1990; Macaluso and Mettus, 1991). Recently, Guss et al. (2012) demonstrated that plasmid transformation efficiency was enhanced by 500-fold with the transformation of Dam⁺Dcm^- methylated plasmids, rather than Dam⁺Dcm⁺ methylated plasmids, in C. thermocellum DSM1313 (Guss et al., 2012). Although the mechanism is still unclear, understanding such RM systems are necessary as they can improve transformation efficiency and shuttle vector stability, especially for Gram-positive bacteria. However, using shuttle vectors still remains challenging in some Gram-positive strains due to the presence of various organism-specific nucleases.

The second essential component for a shuttle vector design is a robust selection system. Antibiotics have been used for selection markers for decades. However, some non-model microorganisms are naturally resistant to common laboratory use of antibiotics. One strategy for selection, based upon antibiotics resistance (ampicillin, kanamycin and chloramphenicol) in a new organism, is to conduct an initial screen to test native resistance by increasing antibiotic concentrations. However, this approach sometimes requires intensive manual labor especially for the expression of genes that hinder the host growth. Less viable approaches include identifying or developing new antibiotic resistance genes or discovery of new antibiotics based upon design-screen-test approaches (Heap et al., 2009). In some circumstances, antibiotics are not a favorable selectable marker due to a specific feature (i.e., thermophiles grown above 70°C might inactivate temperature sensitive antibiotics) or specific considerations for industrial applications (i.e., high cost or not applicable for food-grade application). Two alternative classes of selection markers are dominant markers and complementation markers, shown in Table 1. Dominant markers usually confer a new ability to the host strain such as bacteriocin immunity/resistance (Takala and Saris, 2002), heat-shock resistance (El Demerdash et al., 2003), or sugar utilization abilities (Boucher et al., 2002). On the other hand, a complementation selection marker is based on an auxotrophic strain for an essential metabolite (amino acid, DNA/RNA precursor, and sugar), obtained by mutating or deleting the corresponding chromosomal gene, which can be complemented with the plasmid-borne selection gene. Examples of such marker genes include the thymidylate synthase gene (thyA) (Sasaki et al., 2004), lactose phosphotransferase gene (lacF) (MacCormick et al., 1995) or phosphor-β-galactosidase gene (lacG) (Platteeuw et al., 1996), and alanine racemase gene (alr) (Nguyen et al., 2011).

Promoters are key tools in controlling gene expression since they initiate gene transcription. Promoter regions can be hard to predict in a non-model organism. Therefore, libraries of well-characterized promoters are often only available in well-established organisms. Thus, transcription of foreign genes or activation of cryptic endogenous clusters is one main obstacle for heterologous gene expression or for understanding cryptic novel endogenous cluster functions, respectively. With the advent of high-throughput sequencing technology (mRNA-seq), gene expression levels can be determined with low cost, reliable results, and high efficiency according to the intracellular transcriptomics data. Accordingly, promoters can be rationally designed based on transcription level and tested by using reporter proteins (i.e., GFP).

Recently, systematic identification of constitutive promoters in different microorganisms has been demonstrated as a useful, efficient, and robust platform in Saccharomyces cerevisiae, C. thermocellum, as well as Streptomyces sp. (Sun et al., 2012; Li et al., 2015; Luo et al., 2015; Olson et al., 2015). In order to obtain a library of constitutive promoters, genome-scale transcriptomics data is ranked and compared under various culture conditions or under different time series. Promoter strength can be measured by both protein expression level (GFP) and mRNA level (real-time PCR). Such a panel of constitutive promoters can facilitate precise and quantitative control over gene expression that can potentially be applied to improve microbial chemical production by promoter engineering (Sun et al., 2012; Olson et al., 2015). A strong constitutive promoter is capable of triggering expression of some cryptic clusters, resulting in the discovery of novel natural products and the study of a genetic regulation system (Li et al., 2015; Luo et al., 2015).

Gene downregulation is an effective approach to study gene/protein function or inactivate certain metabolic pathways in non-model hosts. It is sometimes superior to direct gene knock-out mainly because it can avoid the pitfalls of lethal mutations or complete inhibition of protein production. Gene downregulation can be achieved by introducing an exogenous complementary single strand RNA (asRNA) or CRISPR/dCas9 that inactivates at the translational or transcriptional level, respectively.

Antisense RNA has been widely used for downregulating gene expression in many microorganisms. Although the mechanism of asRNA action is not fully understood, asRNA have been successfully applied in non-model organisms, such as Clostridium sp. (Desai and Papoutsakis, 1999; Tummala et al., 2003) and Lactococcus lactis (Sybesma et al., 2004). Such a method has obvious advantages: First, it can be an alternative approach especially for genes that are essential for cell growth. Second, asRNA can be induced conditionally to reduce gene expression without competing for resources with cell growth (metabolic burden). Third, asRNA can be designed for multiplex genome engineering to repress multiple genes simultaneously (Desai and Papoutsakis, 1999; Yang Y. et al., 2015). However, this approach still requires careful design for the asRNA structure and length before experimental implementation and optimization of the asRNA concentration since it is correlated with the inhibition efficiency. Furthermore, the majority of various intracellular RNA pools have the potential to hinder the efficiency of asRNA (Song et al., 2015).

Recently, a catalytically inactive dead-Cas9 (dCas9) was constructed by introducing the D10A and H840A mutations into each nuclease domain, which removed the nuclease activity while maintaining its ability to bind to the target site (Jinek et al., 2012). This modified and simplified CRISPR/dCas9-mediated transcription repression system was successfully used for gene expression regulation in model strains (Qi et al., 2013; Cress et al., 2016). Compared to the asRNA-based system, which regulates gene expression at the translation level, the CRISPR/dCas9 system controls gene expression at the transcription level. Moreover, the CRISPR/dCas9 system uses exogenous parts that are orthogonal to those in host organism and has less chance of competing with the native regulatory systems. Details regarding the expression and efficiency concerns when applying the CRISPR/Cas9 system in non-model hosts are discussed in the genome engineering section.

Upregulation of a gene expression from a single copy number level can be beneficial to the host without introducing an exogenous part. Upregulation can be accomplished either by increasing transcriptional rate through addition of an UP element or by accelerating translational initiation rate through reprogramming a stronger RBS sequence.

Upstream elements, a consensus AT-rich sequences located upstream of the -35 promoter region, have been reported as a gene enhancer of transcription since they can alter the interaction between RNAP and promoters by additional binding with the RNAP alpha subunit C-terminal domain (αCTD). UP elements have been reported to enhance bacterial promoter strength in E. coli and Bacillus subtilis (Frisby and Zuber, 1991; Meng et al., 2001), and recently have been applied in metabolic engineering applications with various successes (Elmore et al., 2017; Yan and Fong, 2017). A novel feature of UP elements was observed to decrease gene expression noise by applying them to E. coli constitutive promoters (Yan and Fong, 2017). In the study, a further mechanistic insight between gene expression and in vitro RNAP and promoter interaction was found that a decent expression level only occurs at a moderate binding affinity. Thus, UP elements can potentially be used as a simple module to enhance gene expression, with several merits. First, it can facilitate promoter engineering by inserting an UP element to each core promoter by a short ssDNA; Second, multiplex genome engineering can be applied by designing site-specific ssDNAs. Third, UP elements can, potentially, stringently control gene expression by reducing basal leaky expression for inducible promoters.

Besides the transcriptional level gene upregulation, a gene expression level can also be elevated by carefully tuning its TIRs, namely by design of RBS sequence. The RBS Calculator is a computational program that enables both a de novo design of the RBS sequence at a certain TIR (from 0.001 to 100,000) and prediction of an RBS sequence and calculation of the TIR in a bacterial genome (Salis et al., 2009). In the model, the hybridization between an mRNA and 16S rRNA and the interactions between the 30S complex and an mRNA were taken into account to quantify a TIR. Such a model has been applied to non-model hosts for many applications, including manipulating proteins expression level (Pogrebnyakov et al., 2017), optimizing synthetic metabolic pathways (Temme et al., 2012), and predicting TIRs across a genome (Salis, 2011). Temme et al. (2012) demonstrated the capability of utilizing synthetic parts (T7 promoter and RBS) to control 20 nitrogen fixation genes expression in Klebsiella oxytoca. Elmore et al. (2017) developed a chromosomal insertion method for functional gene expression in Pseudomonas putida KT2440, which allowed a panel of promoter library (-35/-10 variants) characterization. The author also showed that the expression was further tuned by 2-fold with redesign of RBS and insertion of UP-elements (Elmore et al., 2017).

Homologous recombination is one of most commonly used site-specific genome engineering approaches. HR can facilitate an in-frame deletion by an endogenous exonuclease (e.g., RecA in E. coli) in conjugation with an exogenous DNA into non-model hosts (Radding et al., 1983; Deng and Fong, 2010). However, the native HR system efficiency is relatively low, and successful operations require a careful consideration of conditions (i.e., transformation efficiency, competent cell, selection) (Deng and Fong, 2010). The HR efficiency can also be improved by introducing a heterologous recombination system, including a Cre recombinase (Khodakaramian et al., 2006), a Cre-like tyrosine recombinase (Dre) (Herrmann et al., 2012), a Flp recombinase (Fedoryshyn et al., 2008) and a bacteriophage lambda recombinase (Kim et al., 2008; Lindenkamp et al., 2012). Reviews regarding these approaches can be found recently (Fogg et al., 2014; Deng and Zhang, 2017).

Recently, HR was applied for multiplex genome engineering using ssDNA, an approach named MAGE (Wang et al., 2009, 2012a,b; Isaacs et al., 2011). In this method, a mutS mutated E. coli harbors lambda-Red genes. It is believed that ssDNAs prefer to target the lagging strand rather than the leading strand due to the much higher efficiency (more than 50-fold). The lambda-Red HR facilitates the ssDNA’s targeting at the lagging strand of the replicating genome to introduce mutations. In the MAGE application paper, the authors demonstrated engineering 20 gene RBSs for optimizing the 1-deoxy-D-xylose-5-phosphate (DXP) pathway to produce lycopene (Wang et al., 2009). In order to overcome a common barrier for fast and efficient selection of a genomic mutant after MAGE, development of methods for co-selection markers along with target sites proved to be able to enhance the recombination efficiency of MAGE to greater than 70% (Wang et al., 2012b). Nevertheless, some improvements still need to be considered to carry on: (1) MAGE is limited to integrate large genes (>1 kb) since it depends on ssDNAs; (2) it is still a challenge to widely use this methodology in other non-model bacteria.

A CRISPR/Cas9 system is composed of a clustered set of Cas genes and the signature CRISPR array (Deveau et al., 2010; Horvath and Barrangou, 2010; Terns and Terns, 2011). The Cas9 gene is translated into an endonuclease, whereas the repeated spacer array is transcribed into a long precursor and subsequently processed to generate small crRNAs that direct the endonuclease to cleave dsDNA at specific target sequences (protospacers) (Hale et al., 2009; Makarova et al., 2011). The PAMs that are located immediately downstream of the protospacer is another essential element responsible for directing cleavage of DNA (Deltcheva et al., 2011; Hatoum-Asian et al., 2011).

Such CRISPR/Cas9 systems genome editing tools have been programmed and developed in a wide range of organisms (DiCarlo et al., 2013; Feng et al., 2013; Jiang et al., 2013; Wang H. et al., 2013). In bacteria, the CRISPR/Cas9-mediated genome editing tools rely on introducing two foreign components: Cas9 and a crRNA-trancrRNA duplex or a single guide RNA (sgRNA). Table 2 summarizes current strategies of developing CRISPR/Cas9 for genome editing in non-model bacteria. Excellent reviews can also be found describing general consideration and design when harnessing the CRISPR/Cas9 system in bacteria (Hsu et al., 2014; Selle and Barrangou, 2015; Luo et al., 2016; Mougiakos et al., 2016).

The high efficiency of CRISPR/Cas9 genome editing tool can reach an efficiency up to 100% due to its ability to select out edited cells from non-edited background cells. The first successful CRISPR/Cas9 systems genome editing in bacteria was published in Jiang et al. (2013). The authors combined Cas9 with a CRISPR array and a trancrRNA to generate targeted genome editing in both Streptococcus pneumonia and in E. coli. A separate oligonucleotide whose sequence contained a mutated PAM sequence that can prevent recognition and cleavage from the endonuclease after a repair from HR. Thus, the CRISPR/Cas9 system can not only make defined site mutation but also served as ‘cleanup’ role by eliminating cells that does not undergo recombination. After the initial publication in S. pneumonia and E. coli, applications of CRISPR/Cas9-mediated genome editing have been demonstrated in non-model bacteria, such as Lactobacillus reuteri (Oh and van Pijkeren, 2014), C. beijerinckii (Wang et al., 2015), C. cellulolyticum (Xu et al., 2015), Streptomyces sp. (Cobb et al., 2015; Tong et al., 2015). Chimeric sgRNA designs were subsequently investigated instead of the dual-RNA expression of a tracrRNA and crRNA system. CRISPR/Cas9-based multiplex genome targets were also investigated (Cobb et al., 2015).

Applying this technology for non-model bacteria faces challenges and requires careful consideration of several factors. First, the wide range of microbial diversity poses challenges for basic genetic manipulation. For example, heterologous expression of Cas9 in non-model bacteria may be hindered due to a lack of efficient approaches such as transformation, plasmid replication, and an ability for gene expression. From this point of view, exploring the native machinery system seems to be a superior strategy by precluding heterologous expression of Cas9. And experimental evidence was reported in some Clostridium species that the native CRISPR/Cas9 system (Type I-B) performed higher editing efficiency (100% versus 25%) than the heterologous Type II CRISPR/Cas9 system (Pyne et al., 2016). Second, understanding the PAM sequence is useful for design as PAM and seed sequences are essential for recognition and activity for the gene target. Third, although a 20-nt guide sequence of the sgRNA is believed to guide the Cas9 to the target sequence, potential off-target cleavage activity could still occur on DNA sequences with three to five base pair mismatches. Many computational softwares can be readily found for designing sgRNA with a goal of improving accuracy (Doench et al., 2014, 2016; Moreno-Meteos et al., 2015). Fourth, since the CRISPR/Cas9 system is sometimes lethal to bacteria due to introduced dsDNA breaks, the reparation of DSB is not possible. Thus, a dsDNA or an ssDNA was supplied to facilitate HR or NHEJ as a selection tool against the non-edited background cells (Wang et al., 2015, 2016). HR can be achieved either using its native HR system (i.e., RecBCD in E. coli) (Wang et al., 2015; Cobb et al., 2015) or overexpression of the lambda-Red recombination system (Jiang et al., 2015). Another strategy is to use mutated Cas9 (D10A or H840A), which are believed to reduce the lethality of cleaving by nicking the Cas9 target site without introducing a double-stranded break. For example, Xu et al. (2015) introduced a mutated Cas9 (D10A) and sgRNA as well as an dsDNA template in the same vector resulted in efficiency up to 95%, while the normal Cas9 did not yield any colonies. To date, there are few studies implementing CRISPR/Cas9 genome editing using the NHEJ repair strategy. Tong et al. (2015) expressed LigD along with sgRNA and Cas9 to complete the NHEJ repair system in S. coelicolor, and demonstrated the feasibility of introducing NHEJ in bacteria for CRISPR/Cas9 genome editing (Tong et al., 2015).

Genetic engineering approaches can improve microbial cellulolytic degradation through strategies such as increasing secretion enzyme quantity, improving an enzymatic property, investigating cellulosome regulatory mechanism and enhancing cellular growth on specific biomass (Liambertz et al., 2014; Book et al., 2016). Recently, the roles, activities, and regulatory mechanisms controlling expression of different cellulases were studied, and found to be an effective approach to facilitate cellulose degradation or increase cellulase production (Deng and Fong, 2010; Wang S. et al., 2013; Aghcheh et al., 2014; Häkkinen et al., 2014). The improved cellulase activity can reach from a 2- to 16.69-fold increase after either relieving repression or activating production. An example of combining these approaches was accomplished by carefully blocking intracellular inducer hydrolysis (deletion bgl2), increasing the activator (overexpression clrB) and relieving the repression (deletion creA) simultaneously in Penicillium oxalicum. The optimized strain enabled an increase of cellulase activities from 10- to 50-fold (Yao et al., 2015).

The other main challenge relies upon reinforcing flux toward desired chemicals to increase titer, yield, and productivity. In terms of various pathways for biofuel production as a CBP scheme (Liao et al., 2016), every opportunity to increase product titer, yield or productivity was explored. It always starts with creating/overexpression of desired pathway genes, followed by deletion of competitive pathways to drain precursors, products and cofactors. Production titers can be further improved through beneficial metabolic strategies, including eliminating toxic intermediates, enhancing driving force for pathway flux, and prolonging the cellular production phase. Specifically, these metabolic engineering strategies can be technically achieved by approaches such as promoter engineering, pathway modification by global mutation and selection strategies, cofactor regeneration, knock-out competitive pathway and co-culture. A summary of recent publications on metabolic engineering strategies for native CBP microorganisms is shown in Table 4.

Promoter engineering can facilitate to enhance production titer by adjusting the metabolic burden found during cloning and by reinforcing the rate-limiting step through pathway balancing. For example, in order to get a stable and robust construct for propagating in E. coli and for expression in C. acetobutylicum, Lin et al. (2015) identified and characterized 21 C. acetobutylicum native promoters to drive isobutanol synthetic operons and tested isobutanol production. This study provided a solution of stable and robust expression for a heterologous pathway in non-model microorganisms (Lin et al., 2015). In order to prolong cellular ABE production phase, an exogenous four-step biotin synthetic operon was introduced in C. acetobutylicum ATCC 824 and the gene cluster expression level was tuned by engineering the RBS region. The optimized engineered strain can be boosted to produce solvent titer 21.9 g/L and productivity 0.30 g/L h (Yang et al., 2016).

Pathway modification is another useful strategy for increasing alcohol production by identifying the rate-limiting step and pulling the flux by switching a higher performance enzyme. In the CoA-dependent 1-butanol pathway in C. acetobutylicum, NADH and acetyl-CoA pool serve as driving forces. Shen et al. (2011) demonstrated that modification of the native pathway by switching enzymes (use NADH without coupling ferredoxin/flavoproteins as reducing source) or higher specific activity enzyme can improve 1-butanol production. In order to increase the ratio of butanol to acetone without the formation of additional ethanol, Bormann et al. (2014) demonstrated that utilizing a more butanol-specific aldehyde and alcohol dehydrogenase enabled selectively increase butanol titers while maintaining acetone production. The molar ratio of butanol to acetone after optimization improved from 2.04 to 2.45. Pathway modification can also be extended to introducing a non-native pathway to facilitate more flux toward a desired product. Due to a thermodynamically unfavorable formation of crotonyl-CoA through acetyl-CoA in the first step of 1-butanol formation, an alternative route through malonyl-CoA synthesis was introduced in Synechococcus elongatus PCC7942. Although the resulting pathway resulted in more ATP consumption, the optimized strain enabled direct 1-butanol production (29.9 mg/L) after photosynthesis while the acetyl-CoA dependent pathway barely detected 1-butanol (6.5 mg/L) (Lan and Liao, 2012).

Cofactor engineering is an important strategy for alcohol production in the CoA-dependent pathway where NADH/NADPH is the main supply of reduced cofactors. Disrupting competitive pathways (i.e., formate, hydrogen) can contribute to increasing alcohol production by increasing electron supply (Biswas et al., 2015). Papanek et al. (2015) reported that deletion of genes involved in acetate, lactate, formate, and most of the hydrogen production in a single strain can yield a maximum ethanol titer of 73.4 mM. Understanding electron transfer and cofactor supply can also favor improving alcohol production since NADH is the main electron donor for alcohol formation. Deng et al. (2013) demonstrated an endogenous three-step pathway (called malate shunt) converting phosphoenolpyruvate to pyruvate, which contains an NADH-linked malate dehydrogenase. By disrupting the malate shunt pathway genes coupled with expressing a pyruvate kinase, ethanol production can increase 3 to 3.25-fold (Deng et al., 2013). Lo et al. (2017) investigated different electron supply resource’s effects on fuel production and identified that an NAD⁺ oxidoreductase is the main supply for NADH regeneration. The results showed that overexpression of rnf genes resulted in an increase in ethanol production of about 30% (Lo et al., 2017).

Recently, co-culture has been developed as an effective metabolic engineering strategy that can construct and modulate each organism’s expression system and pathway in parallel so that the time required for making the product would be substantially reduced (Zhou et al., 2015). In CBP schematic fuel production, an artificial consortium can create a symbiotic relationship to produce fuel in a synergistic manner. Nakayama et al. (2011) developed a cooperator-cooperator consortium by a cellulolytic strain and a butanol-producing strain. The cellulolytic strain secreted butyric acid that can induce butanol production in the butanol-producing strain. The co-cultured system produced 7.9 g/L butanol from 4% Avicel after 9 days of incubation (Nakayama et al., 2011). In order to ferment a variety of sugars (i.e., hexose, pentose, xylose) other than cellulose, the co-cultured system can be developed based on a strain’s metabolic capability. Wen et al. (2014) constructed a co-culture of C. cellulovorans and C. beijerinckii. Under an optimized condition, the co-culture produced 11.9 g/L of solvents from 68.6 g/L alkali pretreated corn cobs (Wen et al., 2014). In the following study, the authors genetically modified the cellulolytic strain to pull more flux toward butyrate production and the solvent-producing strain to enhance organic acids reassimilation and pentose utilization. The engineered consortium was shown to produce 22.1 g/L of solvents from 83.2 g/L lignocelluloses hydrolysate (Wen et al., 2016).