The valine-leucine-Ehrlich (VLE) pathway converts pyruvate to 3MB through the valine and leucine biosynthetic reactions, followed by α-ketoacid decarboxylation and alcohol dehydrogenation characteristic of the Ehrlich pathway (Figure 1). It is currently the most extensively engineered pathway for 3MB production, especially in Escherichia coli, Corynebacterium glutamicum, and S. cerevisiae (Table 2).

Pyruvate is first converted to α-acetolactate by acetolactate synthase (ilvIH, ILV2/ILV6). Keto-acid reductoisomerase (ilvC, ILV5) reduces this intermediate to α,β-dihydroxyisovalerate using NADPH, which is subsequently dehydrated by dihydroxyacid dehydratase (ilvD, ILV3) to form α-ketoisovalerate (KIV). KIV partitions into two possible reactions: transamination to valine via branched-chain amino acid aminotransferases (ilvE, BAT1/BAT2), or elongation toward leucine. The first committed step in leucine biosynthesis is catalyzed by α-isopropylmalate synthase (leuA, LEU4/LEU9), forming α-IPM from KIV and acetyl-CoA. Dehydration and oxidative decarboxylation yield α-ketoisocaproate (KIC), which is subsequently decarboxylated by α-ketoacid decarboxylases (PDC1/PDC5/PDC6/ARO10) and reduced by alcohol dehydrogenases (ADH1/ADH2/ADH5/ADH6/ADH7) to produce 3MB. In yeast, Adh2 is glucose-repressed, and Adh6/7 are NADPH-dependent with strong preference for branched alcohols. Importantly, these promiscuous enzymes also produce other fusel alcohols like isobutanol and propanol, which implies that the end product is often a complex mixture of structurally related molecules.

While the pathway is native in yeast, bacteria lack the Ehrlich reactions and therefore require expression of the α-ketoacid decarboxylases KivD from Lactococcus lactis or Aro10 from S. cerevisiae, and the alcohol dehydrogenases Adh2 or Adh6 from S. cerevisiae. LlKivD has higher specificity toward α-ketoisovalerate, the precursor of isobutanol. ScAro10 has higher selectivity toward KIC, increasing the 3MB:isobutanol ratio (Atsumi et al., 2008).

All engineering strategies discussed in this subsection are summarized in Table 2 to facilitate visual comparison.

The mevalonate (MVA) pathway has been widely engineered in bacteria to produce products derived from isoprenoids and isoprenol (Wang C. et al., 2018, 2023; Kang et al., 2024; Yin et al., 2025). Introducing the heterologous eukaryotic MVA pathway enables the accumulation of isopentenyl pyrophosphate (IPP) and dimethylallyl pyrophosphate (DMAPP), which can be converted to isopentenols (3 M-3-butenol and 3 M-2-butenol) and subsequently reduced to 3MB (Figure 2).

To date, 3MB production using the MVA pathway has only been tested in E. coli (Table 3). MVA-based production of 3MB starts with the condensation of two molecules of acetyl-CoA to acetoacetyl-CoA, catalyzed by the endogenous acetoacetyl-CoA thiolase AtoB. Subsequently, the reactions from acetoacetyl-CoA to DMAPP are performed by heterologous proteins. Acetoacetyl-CoA is converted to hydroxymethylglutaryl-CoA (HMG-CoA) and reduced to MVA by HMG-CoA synthase MvaS and HMG-CoA reductase MvaA from Staphylococcus aureus (George et al., 2015; Eiben et al., 2020). MVA is then phosphorylated twice by ATP-dependent mevalonate kinase (ERG12) and phosphomevalonate kinase (ERG8) from S. cerevisiae, yielding mevalonate diphosphate (MVAPP). This intermediate undergoes ATP-dependent decarboxylation by mevalonate diphosphate decarboxylase from S. cerevisiae (ERG19) to generate IPP. Then, IPP is isomerized to DMAPP by IPP isomerase, either by the endogenous enzyme Idi or the exogenous Idi1 from S. cerevisiae. IPP and DMAPP are then dephosphorylated by the phosphatase NudB to form the isopentenols 3-methyl-3-butenol (3 M-3-butenol) and 3-methyl-2-butenol (3 M-2-butenol), respectively. Finally, 3MB is produced from the reduction of 3 M-2butenol by the reductase nemA (Chou and Keasling, 2012; George et al., 2015).

In E.coli, all genes required to convert acetyl-CoA to 3MB are typically overexpressed from plasmids under inducible promoters (e.g., IPTG-inducible systems). To compete with other native pathways utilizing IPP and DMAPP and to improve 3 M-2butenol production, ScIdi1 was fused with NudB, resulting in higher titers of isopentenols and 3MB (Chou and Keasling, 2012). Adding an extra copy of nudB, however, increased 3 M-3butenol but not 3 M-2butenol. With these engineering strategies, the highest 3MB titer and yield reported via the MVA pathway were 0.3 g/L and 0.03 g/g glucose in a two-phase fermentation using an oleyl alcohol overlay and rich defined medium (George et al., 2015). There is likely still room for further increases in production efficiency. For example, although never studied specifically for 3MB production, mvaS and mvaA from Enterococcus faecalis or Lactobacillus casei have yielded the highest mevalonate titers in E. coli (Wang et al., 2023), suggesting that using these enzyme variants might help increase 3MB titers.

The MVA pathway naturally occurs in yeast but exhibits inherently lower flux due to the tight regulation from sterol metabolism (Wang et al., 2023), and there are no public reports of engineering the MVA pathway in yeast specifically for 3MB production. Instead, engineering the MVA pathway in yeast primarily focuses on isoprenol production (Kim et al., 2021; Wang et al., 2022; Banerjee et al., 2024).

Production of 3MB is also possible by combining the isovaleryl-CoA pathway with the non-oxidative glycolysis (NOG), a synthetic glycolytic route designed to maximize carbon conservation (Bogorad et al., 2013; Eiben et al., 2020). This pathway branches from the MVA pathway at HMG-CoA and uses a heterologous set of Myxococcus xanthus enzymes to convert HMG-CoA to isovaleryl-CoA. The resulting isovaleryl-CoA is then reduced to 3MB using the butyryl-CoA reductase complex from Clostridium acetobutylicum.

The NOG pathway converts one glucose into three acetyl-CoAs, precisely the three required to synthesize one molecule of 3MB through the MVA or isovaleryl-CoA pathways. As a result, the NOG–isovaleryl-CoA system loses only one carbon as CO₂. However, this carbon advantage comes at the cost of a redox imbalance because the isovaleryl-CoA pathway requires three NADH molecules per 3MB synthesized, whereas NOG generates no net ATP or NADH (Figure 2, Table 4).

The only published study evaluating this strategy used a two-phase fermentation with an oleyl alcohol overlay and achieved a 3MB titer of 80 mg/L and a yield of 0.04 g/g glucose, with 1.46 g/L acetate as a major byproduct (Eiben et al., 2020) (Table 3). The high acetate levels may result from overflow metabolism driven by the accumulation of acetyl-CoA produced via NOG. Unlike the MVA pathway that typically accumulates significant amounts of isopentenols, the NOG-isovaleryl-CoA pathway produces 3MB as the sole fusel alcohol product.

Each pathway differs substantially in carbon yield, ATP requirements, and cofactor balance (Table 4). The VLE and MVA pathways are carbon-intensive when coupled to glycolysis, losing four carbons as CO₂ per 3MB produced, corresponding to a theoretical yield of 0.33 g 3MB per g glucose. The glycolysis-MVA pathway uses three NADPH per 3MB, making NADPH regeneration, such as coupling to oxidative pentose phosphate pathways or heterologous NADPH-regenerating enzymes, essential. In contrast, NOG-coupled isovaleryl-CoA pathways are more carbon-efficient (theoretical yield of 0.49 g 3MB per g glucose) but suffer from high NADH demand, which likely would produce NADH-generating byproducts like glycerol. Instead, the isovaleryl-CoA pathway could be coupled with glycolysis that would supply the required NADH, but would also suffer more carbon losses in the form of CO₂.

The VLE pathway implemented in E. coli has the best reported titers and yields (up to 9.5 g/L 3MB and 0.11 g/g glucose) (Connor et al., 2010), reaching approximately 33% of the theoretical yield of 0.33 g/g glucose. This production strain overexpresses a leucine-resistant LeuA allele along with other VLE pathway genes, and was cultivated in a two-phase fermentation setup. When implemented in C. glutamicum, the VLE pathway also enabled relatively high titers (2.76 g/L), albeit with lower yields (0.07 g/g glucose) and increased isobutanol byproduct formation (Vogt et al., 2016). Expressing the VLE pathway in S. cerevisiae achieved substantially lower titers and yields (approximately 1.5 g/L 3MB and 0.01 g/g glucose) (Hammer et al., 2020; Xiao et al., 2025), likely due to significant carbon loss to ethanol production. Nevertheless, high 3MB selectivity and reduced isobutanol byproduct formation were observed in S. cerevisiae.

Strains engineered with the MVA pathway also achieve substantially lower performance, with maximum titers of only 0.3 g/L 3MB and yields of 0.03 g/g glucose (George et al., 2015), far below the theoretical yield of 0.33 g/g glucose. This reduced efficiency is likely attributable to the MVA pathway’s heavy energetic burden—requiring more ATP and NADPH per unit of 3MB than the VLE pathway—as well as its greater genetic complexity, which involves introducing at least six heterologous genes. These factors make metabolic flux balancing far more challenging than in the VLE pathway, which is native to yeast and in bacteria requires expression of only two to three heterologous genes.

The NOG-isovaleryl-CoA pathway, although theoretically promising due to its superior carbon efficiency, has so far yielded only a single study reporting titers of 0.08 g/L and a yield of 0.04 g/g glucose, likely due to significant carbon loss to acetate and formate overflow (Eiben et al., 2020).

Although the VLE pathway implemented in bacteria currently represent the most effective strategy, they rely on plasmid-based expression systems with IPTG-inducible promoters (Connor et al., 2010; Vogt et al., 2016). Plasmid stability may therefore pose challenges during scale-up, and the use of costly inducers such as IPTG is unlikely to be economically viable in large scale. In yeast, the amino acid auxotrophy complementation system was often used (Hammer et al., 2020; Xiao et al., 2025), complicating cultivation in industrial media that often contains the amino acid as selective marker (Pronk, 2022). However, re-engineering these pathways into industrial hosts, using markerless systems and stable genome integrations should not be an issue, especially when some of the newer technologies, such as CRISPR-Cas, are used. Selection of an optimal host and pathway will ultimately also depend on substrate type, desired product profile, byproduct tolerance, oxygen requirements, and fermentation scale.

Isobutanol formation remains the principal competing flux in the VLE pathway. Future progress could rely on protein engineering of α-ketoacid decarboxylases, such as generating LlkivD variants with higher KIC selectivity for a more selective 3MB synthesis (Mak et al., 2015; Chen et al., 2017; Miao et al., 2018). Fine-tuning ILV, LEU, PDC, and ADH gene expression using genome-scale engineering tools such as GEMbLeR could also be a strategy to optimize the VLE pathway (Cautereels et al., 2024). Moreover, biosensor-guided adaptive evolution using α-IPM or 3MB-responsive sensors could be an efficient approach to enrich for strains with improved 3MB production (Yu et al., 2019; Zhang et al., 2022).

One of the major challenge of producing 3MB through the MVA pathway is the formation of high levels of isopentenols, especially 3 M-3butenol, which often accumulate to concentrations 5–7 fold higher than 3MB (Chou and Keasling, 2012; George et al., 2015). One strategy to address this is to improve the substrate specificity of NudB so it prefers DMAPP over IPP. For example, the E. coli ADP-ribose pyrophosphate gene nudF has higher specificity to 3 M-2butenol. Overexpressing nudF increased 3 M-2butenol to five times the level of 3 M-3butenol (Zheng et al., 2013). Coexpressing nudF with nemA could potentially further improve 3MB production.

Extensive research on isobutanol biosynthesis could provide a strong foundation for engineering 3MB production via the VLE pathway. Many isobutanol strains naturally generate 3MB, making them attractive chassis for redirection toward 3MB with minimal redesign. Introducing leucine-resistant LeuA or Leu4 variants, increasing leucine pathway flux, and improving α-ketoacid decarboxylase specificity could be key strategies to shift carbon flow toward α-KIC and 3MB. Given the significant progress made in diverse hosts, feedstocks, and fermentation strategies for isobutanol (De Lima et al., 2024; Nawab et al., 2024), these insights can be readily transferred to 3MB engineering.

Cell-free biosynthesis offers a promising alternative to circumvent limitations imposed by living cells, including 3MB toxicity, pathway byproduct formation, and constraints on NAD(P)H availability. Recent work on isobutanol demonstrated that cell-free enzyme cascades can achieve exceptionally high titers and yields (275 g/L, 0.95 g/g glucose) (Sherkhanov et al., 2020), suggesting considerable potential for 3MB if enzyme sets are tailored for the α-IPM - α-KIC - 3MB conversion. Interestingly, early work in 1983 already demonstrated 3MB formation from leucine using Streptococcus lactis extracts (Braun et al., 1983), yet this concept has not been revisited.

Process optimization for 3MB production, from media, growth conditions, reactor design, and downstream processing, has received relatively little exploration compared to strain engineering. Parameters such as temperature, pH, dissolved oxygen, and nutrient-feeding strategies greatly influence metabolic fluxes (Stanbury et al., 2016) and represent untapped opportunities to improve productivity.

Current evidence reveals markedly different aeration requirements depending on the host organism and pathway. Insights from metabolic engineering studies suggest that enhanced aerobicity may benefit 3MB production in yeast via the VLE pathway. A genome-wide perturbation study found upregulated respiratory genes in high-3MB-producing strains (Xiao et al., 2025), and increased oxygenation has been linked to higher alcohol formation during wine fermentation (Valero et al., 2002). For 3MB production via the MVA pathway, even higher aeration levels may be required to meet the substantial ATP demands of the pathway. In contrast, microaerobic conditions favor 3MB production in E. coli via the isovaleryl-CoA pathway by suppressing competing prenol formation (Eiben et al., 2020).

Most studies to date rely on small-scale batch fermentations (≤0.1 L), underscoring that process development and industrial implementation remains in its infancy. Established industrial approaches such as fed-batch cultivation, where carbon feeding is controlled to achieve stable optimal carbon concentrations in the medium, could further increase titers by limiting overflow metabolism and catabolite repression (Yu et al., 2019; Zhang et al., 2019). For the production of fermented beverages, higher alcohol formation has been shown to be significantly enhanced under fed-batch conditions (Visinoni et al., 2022).

Conversely, the selection and engineering of chassis strains with high stress tolerance and fitness under industrial conditions represent another important route for process optimization (Smets et al., 2025; Yogiswara et al., 2025; Zhu et al., 2025). Further engineering efforts aimed at improving tolerance to high sugar or osmotic stress, elevated temperatures, and reduced nutrient or oxygen availability could further enhance productivity, as comprehensively summarized in recent review articles (Deparis et al., 2017; Mohedano et al., 2022).

Beyond yield optimization and strain robustness, economic considerations play a critical role in process feasibility. Factors such as inducer cost, strain stability, and long-term viability can significantly influence overall process economics. Many promising strains discussed in this review rely on plasmid-based and IPTG-based inducible expression systems. However, plasmid instability remains a challenge during scale-up because plasmids are sometimes lost during cell division or undergo structural rearrangements that result in reduced yields. The use of minimal plasmids or genomic integration of pathway genes could substantially improve genetic stability over extended cultivation periods (Oliveira et al., 2009). Moreover, costly gene expression inducers such as IPTG may significantly increase production costs at industrial scale. Alternative strategies, including self-inducible expression systems or autoinduction media, could help mitigate these costs (Rosano et al., 2019).

Downstream purification is another critical consideration, particularly for 3MB production via the VLE pathway, where structurally similar fusel alcohol byproducts complicate separation. Several studies have explored separation strategies for fusel alcohol mixtures and may provide a foundation for the development of 3MB-specific purification processes (Ferreira et al., 2013; Mendoza-Pedroza et al., 2021; Missyurin et al., 2024).

In situ product removal (ISPR) remains another effective yet underexplored optimization strategy. To date, 3MB removal has primarily relied on liquid–liquid extraction using solvent overlays. However, ISPR approaches developed for butanol and isobutanol, such as gas stripping, pervaporation, liquid-membrane extraction, and vacuum stripping (Lakshmi et al., 2021; Huang and Ma, 2023), are potential candidates for adaptation to 3MB. Commercial-scale isobutanol production already employs low-pressure evaporation to continuously flash product from the broth (Gevo, 2023). Given that 3MB has a higher vapor pressure than isobutanol, evaporation-based technologies may prove particularly efficient for mitigating toxicity and increasing achievable titers.

Despite substantial progress in engineering yeast for 3MB production, current titers and yields remain far below what is needed for cost-effective standalone fermentation. An alternative opportunity comes from the fact that large-scale bioethanol plants already generate 3MB naturally as part of the fusel alcohol byproduct stream (dos Santos et al., 2024). Leveraging this existing stream bypasses several challenges associated with de novo pathway optimization, provided that 3MB can be recovered at sufficiently high purity. However, fusel alcohol mixtures are difficult and energy-intensive to fractionate, due to the heterogeneous azeotropes that occur between 3MB, water, ethanol and other fusel alcohols (Mendoza-Pedroza et al., 2021). Increasing both the yield and the purity of 3MB within this stream may substantially reduce downstream separation costs. A recent work illustrates this potential: an engineered strain expressing Leu4^{S542F A551V} combined with ALD6, EAT1, and MET17 deletions has an increased 3MB titers and purity within the fusel alcohol stream from 42 to 71%, without compromising ethanol productivity (Yogiswara et al., 2025). Implementing such a strain would enable bioethanol producers to maintain their primary ethanol output while generating additional revenue from 3MB co-production.