Exploring the potential of Rhodosporidium toruloides in biomanufacturing: a comprehensive review

1 Introduction

Microbial hosts lie at the heart of modern synthetic biology, enabling the sustainable production of fuels, chemicals, and pharmaceuticals from renewable feedstocks. While Escherichia coli and Saccharomyces cerevisiae have long served as the workhorses of metabolic engineering, growing interest in complex and hydrophobic molecules—such as lipids and isoprenoids including carotenoids—has exposed critical limitations in these model organisms (Worland et al., 2020). This has sparked the exploration of alternative, non-model microbes with innate biosynthetic capabilities that can support industrial-scale biomanufacturing (Goswami et al., 2024).

Rhodosporidium toruloides (also known as Rhodotorula toruloides) is a basidiomycetous oleaginous yeast and has emerged as a promising non-conventional host for synthetic biology as recently explored for ergothioneine production (Liu et al., 2024). Notable for its ability to accumulate lipids and carotenoid pigments, R. toruloides can utilize a wide range of carbon sources, including lignocellulosic hydrolysates, and displays natural tolerance to environmental stressors (Priyadarshini and Kataria, 2025) (Figure 1). These features make this microbial host attractive for the production of various bioproducts including biofuels, nutraceuticals, and other value-added compounds as listed in Table 1 (Adamczyk et al., 2025b). Importantly, R. toruloides possesses a native isoprenoid biosynthesis capability via endogenous mevalonate pathway and routinely channels metabolic flux toward acetyl-CoA-derived products, offering a favorable metabolic background for isoprenoid biosynthesis (Kirby et al., 2021).

TABLE 1. List of biochemicals produced by R. toruloides.

Despite these advantages, the genetic manipulation of R. toruloides presents considerable challenges. Compared to model systems such as E. coli and S. cerevisiae, it suffers from limited transformation efficiencies, a scarcity of characterized genetic parts, poor homologous recombination rates, and the absence of stable plasmid systems (Brink et al., 2023; Xu et al., 2025). While Agrobacterium tumefaciens-mediated transformation (ATMT) has enabled functional engineering, its random integration and low throughput limit a predictable engineering and iterative Design-Build-Test-Learn (DBTL) workflows (Szarzanowicz et al., 2025). CRISPR/Cas-based genome editing tools have recently been explored and reported in R. toruloides, but their integration-dependent nature and variable efficiency underscore the need for further refinement (Otoupal et al., 2019; Jiao et al., 2019; Schultz et al., 2019).

This review aims to provide a comprehensive summary and perspective of current advances and remaining challenges in the genetic engineering of R. toruloides, with a focus on enabling its broader adoption in synthetic biology. We examine the limitations of available transformation methods, compare genetic tractability with established hosts, and highlight emerging tools such as promoter libraries, CRISPR technologies, and counter-selection strategies. A particular attention is given to R. toruloides as a chassis for isoprenoid biosynthesis, a key application area that exemplifies both its potential and engineering hurdles. We also discuss the use of automation for engineering R. toruloides. By integrating recent findings and identifying current bottlenecks, we aim to inform future efforts to unlock the full synthetic biology potential of this unique yeast.

2 Gene transformation methods in Rhodosporidium toruloides

R. toruloides has emerged as a promising non-model host for biotechnological applications, converting diverse substrates into fuels and chemicals (Wen et al., 2020; Lyu et al., 2023). Omic analyses and genome-scale modeling have illuminated its metabolic capabilities (Wen et al., 2020). However, engineering R. toruloides poses unique challenges compared to well-established hosts such as E. coli or S. cerevisiae and the limitation of the available gene transformation method is one of the drawbacks in engineering R. toruloides. Here we provide a comprehensive overview of the available transformation methods in R. toruloides (summarized in Figure 2).

2.1 Agrobacterium tumefaciens-mediated transformation (ATMT)

One of the primary genetic tools for transforming genes into R. toruloides is Agrobacterium tumefaciens-mediated transformation (ATMT), which is a broadly applicable tool across fungi and some non-model yeasts. ATMT is the most widely used genetic engineering approach for R. toruloides as it enables integration of foreign DNA into the genome—an outcome that earlier methods failed to achieve reliably (Figure 2). ATMT replaced inefficient techniques such as PEG-mediated protoplast transformation by using A. tumefaciens T-DNA to generate stable transformants. The method supports sequential integration of multiple genes, making it well suited for complex metabolic engineering of lipid and carotenoid pathways (Lin et al., 2014). Moreover, the establishment of robust ATMT protocols has driven the development of essential genetic tools, including well-characterized promoters and selectable markers (Nora et al., 2019). ATMT, however, comes with notable limitations in the context of engineering R. toruloides when compared to the engineering tools for other conventional microorganisms. Here, we will review the limitations of the ATMT-based approach in R. toruloides engineering.

First, ATMT is a time-consuming process. ATMT requires a co-cultivation period with A. tumefaciens, making a single transformation cycle take up to 2 weeks (Wen et al., 2020). This is markedly slower than standard chemical or electroporation methods used in conventional hosts. As a result, iterative engineering (i.e., multiple rounds of gene insertion or deletion) in R. toruloides via ATMT can be labor- and time-intensive.

Second, ATMT shows a low and variable efficiency. The transformation efficiency in R. toruloides using ATMT is relatively low – typically on the order of 70–1,000 transformants per plate depending on the strain and marker (Lin et al., 2014). Overall, yeast transformations by any method tend to yield only ∼10³ transformants per µg of DNA (Kumar, 2016). These are several orders of magnitude lower than the efficiencies achieved in E. coli or S. cerevisiae, making it challenging to obtain sufficient mutants or clones for screening. It has been noted that when the frequency of desired recombinants is below 10^–3, obtaining a “favorable” mutant becomes difficult without screening extremely large numbers of colonies (Kumar, 2016). In practice, substantial manpower must be devoted to increasing transformant numbers and filtering out background (ectopic) insertions (Wen et al., 2020).

Third, ATMT delivers DNA into the host genome via T-DNA transfer, which predominantly integrates at random sites through non-homologous end-joining (Sun et al., 2017; Wang et al., 2017). This random insertion means each transformant can have the transgene at different loci and with different copy numbers. Consequently, researchers must screen many colonies to find those with optimal expression or with the correct integration, especially if targeting a specific gene disruption. Random insertion also carries the risk of disrupting endogenous genes or regulatory elements, potentially causing unintended phenotypes (Hooykaas et al., 2018; Koh et al., 2014). In R. toruloides, the proportion of mutants arising from true targeted homologous recombination via ATMT is very low, as most integration is ectopic (Wen et al., 2020). This necessitates careful confirmation of clones and often the use of counter-selection strategies to eliminate random insertions (Wang et al., 2017). For example, placing a toxic or counter-selectable marker outside the homology arms on the T-DNA can help “poison” random integrants, a technique used in fungi to enrich for true targeted events (Khang et al., 2005).

Recently there were few studies that have been that increases the efficiency of the ATMT. Using the top-performing origin, pVS1, higher-copy-number variants isolated that enhanced stable transformation efficiency by 390% in the oleaginous yeast R. toruloides (Szarzanowicz et al., 2025). It has been shown that removing the filtration step in AMT fungal transformation reduces hands-on time from >4 h to 15 min and boosts efficiency by 69% in Rhodotorula toruloides. Further optimization of the A. tumefaciens:yeast ratio and resuspension volume achieved >210,000 CFU per transformation, a 2–3 times improvement over existing protocols (Incha et al., 2026). In other study, authors optimized ATMT in R. toruloides by tuning the T-DNA promoter, A. tumefaciens:yeast ratio, acetosyringone level, and cocultivation conditions, achieving 2,369 transformants per 10⁵ cells – 24 times higher than previous reports (Sun et al., 2021).

In addition, each ATMT vector typically carries a single selection marker (e.g., antibiotic resistance). Given random integration, multiple genes must often be integrated via separate transformations or a polycistronic construct. R. toruloides cannot maintain standard plasmids extra chromosomally, which is discussed more below, so every gene addition usually means a new genomic insertion (Lin et al., 2014). Moreover, if multiple T-DNA copies are inserted, they may recombine or induce gene silencing. These factors complicate pathway assembly via ATMT.

2.2 Lithium acetate/polyethylene glycol and electroporation method

In the presence of polyethylene glycol (PEG), Rhodosporidium toruloides protoplasts are transformed at frequencies of approximately 1 × 10³ transformants per μg of DNA (Tully and Gilbert, 1985) (Figure 2). In addition, lithium acetate–PEG protocol was employed for R. toruloides transformation of plasmid DNA and has been used in a lot of recent works with advantages of low costs, no special equipment, and adaptability (Nora et al., 2019). In one of the studies, a straightforward and time-efficient method for transforming TK16 using linear DNA fragments in a lithium acetate–PEG mixture was established (Tsai et al., 2017) (Figure 2). Although this method provides flexibility in regular lab protocols, methods have low efficiency, are labor-intensive and poor for large DNA sequences (Tsai et al., 2017).

Recently, linear DNA fragment-based electroporation method was developed as a simple method for gene integration in R. toruloides (Liu et al., 2017). This method achieved a transformation efficiency of 2 × 10³ CFU/μg DNA, exceeding the efficiencies reported for LiAc-based methods (25–40 CFU/μg DNA) (Liu et al., 2017; Tsai et al., 2017) (Figure 2).

2.3 Strain and protocol sensitivity

Transformation outcomes can vary with R. toruloides strain background and subtle protocol differences. Variables such as the A. tumefaciens strain, co-cultivation conditions, or selection regimen can impact transformation efficiency significantly (Lin et al., 2014). Researchers have explored alternatives such as protoplast-PEG transformation or electroporation, but they have yielded low efficiencies as low as 25 transformants/µg DNA by PEG, or 40–1000 CFU/μg by electroporation (Tsai et al., 2017). Therefore, as of recent studies, ATMT remains the most reliable method for R. toruloides gene transformation, despite its drawbacks (Wen et al., 2020).

In summary, ATMT enables genetic modification of R. toruloides while a few other options are also available. However, its slow processing time, low efficiency, and random-integration nature pose significant limitations for synthetic biology and metabolic engineering applications of ATMT to R. toruloides in general. Improving transformation efficiency and throughput (e.g., via transposon mutagenesis or optimized protocols) is an active area of development (Szarzanowicz et al., 2025).

2.4 Transformation efficiency

E. coli is renowned for extremely high transformation efficiencies (often ≥10⁷ CFU (colony-forming units) per µg DNA for plasmid uptake), and S. cerevisiae typically achieves ∼10⁴–10⁵ transformants/µg with plasmid DNA under optimized conditions. In stark contrast, R. toruloides yields on the order of 10²–10³ transformants per µg DNA (Liu et al., 2017), even with its best methods. This means experiments that are routine in E. coli (e.g., constructing mutant libraries or multi-gene variant libraries) become far more laborious in R. toruloides. Indeed, low efficiency has hindered approaches like genome-wide mutagenesis or barcoded libraries in this yeast, which rely on generating thousands of independent transformants (Coradetti et al., 2018).

3 Genetic accessibility of R. toruloides

Compared to model hosts such as E. coli and S. cerevisiae, R. toruloides has a significantly lower genetic tractability and transformation efficiency. Some key differences are listed and reviewed in this section.

3.1 DNA uptake and integration

E. coli readily takes up plasmids, and S. cerevisiae can either maintain plasmids or integrate DNA via homologous recombination. R. toruloides, however, lacks any stable episomal plasmid system (Lin et al., 2014). Consequently, any gene introduced into R. toruloides must be integrated into the genome for stable maintenance (Wen et al., 2020). This reliance on chromosomal integration, which is often random, makes genetic modifications permanent and harder to either prototype or tune. While we can easily shuffle plasmids or use CRISPR for scarless edits in S. cerevisiae, in R. toruloides, even CRISPR/Cas9 constructs should be delivered into the chromosome by genome integration in most cases (Schultz et al., 2022).

3.2 Homologous recombination vs. NHEJ

S. cerevisiae is highly proficient at homologous recombination (HR), allowing precise gene knockouts or insertions by simply providing ∼40 bp flanking sequences. R. toruloides, however, has a much lower native HR frequency and non-homologous end joining (NHEJ) dominates DNA repair (Sun et al., 2017). For example, one study achieved only ∼10.5% targeted integration even with designed HR in wild-type R. toruloides, whereas deleting the Ku70 NHEJ gene raised targeting to 75% – but at the cost of far fewer total transformants (Koh et al., 2014). Thus, wild-type R. toruloides is heavily biased toward random integration. In practical terms, this means that unlike S. cerevisiae, precise gene edits such as knockouts and insertions in R. toruloides require either laborious screening or specialized strategies like Ku70-deficient strains (Wen et al., 2020) or incorporating selection schemes as noted above.

E. coli does not perform homologous recombination on its own at high frequency either, but recombineering tools or selection-linked strategies such as λ-Red or CRISPR-assisted methods exist to achieve precise edits in that host. Developing analogous techniques and strategies in R. toruloides has been challenging but it is highly desired tasks to make R. toruloides more favorable host for biotechnology.

3.3 Genetic tools parts

The molecular biology “toolkit” for E. coli and S. cerevisiae is vast as numerous well-characterized promoters, terminators, selection markers, reporter genes, and regulatory elements are available. In R. toruloides, however, these parts have historically been limited or less characterized. In a related study, authors screened for controllable promoters and identified inducible elements from NAR1, ICL1, CTR3, and MET16. These promoters respond to distinct induction/repression cues and show varied expression strengths and kinetics in R. toruloides, enabling flexible control of transgene expression across experimental contexts (Johns et al., 2016). In another study, five constitutive promoters evaluated from R. toruloides and found their strengths varied within a five-fold range, ranking PPGI > PPGK > PFBA > PTPI > PGPD (Wang et al., 2016). RNA-seq guided the discovery of >20 native promoters (including strong bidirectional promoters) in R. toruloides (Nora et al., 2019). One native promoter (LDP1, a lipid droplet protein gene promoter) was found to be up to 10 times stronger than the traditional GPD1 promoter (Wen et al., 2020) and a D-amino acid inducible promoter system (DAO1) was also developed (Liu et al., 2015). These are valuable additions, but the depth of available tools still lags those in S. cerevisiae (which benefits from decades of research). Furthermore, R. toruloides is a basidiomycetous yeast, evolutionarily distant from S. cerevisiae (an ascomycete), so not all yeast tools are cross-compatible. Many heterologous parts must be validated or adapted to be used in R. toruloides. Until recently, R. toruloides lacked standard synthetic biology tools, relying on native promoters (e.g., glycolytic genes, elongation factors) and selection markers such as hygromycin B, G418, nourseothricin, zeocine, and blasticidin resistance. Multi-gene pathways were assembled on large T-DNA vectors or integrated in parts, and the absence of well-characterized inducible systems limited dynamic gene control—unlike S. cerevisiae, which has extensive promoter libraries. Recent advances have identified strong constitutive and at least five inducible promoters, including nitrogen starvation–responsive ones relevant to lipid production (Nora et al., 2019). Reporter systems (e.g., luciferase, fluorescent proteins) have further enabled part characterization (Guo et al., 2023). However, constructing complex synthetic circuits in R. toruloides remains more challenging due to its still-developing toolkit.

3.4 Cell biology and lifecycle

R. toruloides has some biological differences that can affect engineering. It is often haploid and can undergo a sexual cycle to form basidiospores, but laboratory work typically keeps it in a yeast form. Its optimal growth temperature (∼30 °C) and doubling time are comparable to other yeasts, though it can exhibit slower growth in a minimal medium or during lipid accumulation phases. E. coli (a bacterium) outpaces yeasts in growth rate and simplicity (i.e., no organelles, no splicing of mRNA, etc.), which partly explains why iterative design-build-test cycles are faster in E. coli. S. cerevisiae has the advantage of a well-characterized cell cycle and the ability to sporulate or use stable diploids/haploids as needed, but little is published about manipulating the ploidy or mating of R. toruloides for genetic purposes (though strains like R. toruloides NP11) are haploid and amenable to transformation (Lyu et al., 2023).

In summary, R. toruloides is less tractable than E. coli or S. cerevisiae due to its low transformation efficiencies, lack of plasmid maintenance, and poor homologous recombination. While E. coli and S. cerevisiae allow rapid, modular genetic edits (often with multiple genes at once), R. toruloides engineering has required more sequential, trial-and-error approaches. This gap is gradually narrowing down as new tools (such as promoters, selection markers, and CRISPR systems) are developed for R. toruloides, but it remains a key consideration when choosing it as a production host.

4 Engineering constraints and current developments in R. toruloides as a next synthetic biology host

Given the above limitations, several specific challenges confront researchers working to engineer R. toruloides, and we will list some of these challenges in this section.

4.1 Pathway assembly difficulties

Engineering a metabolic pathway often requires inserting multiple genes for enzymes and regulators, and sometimes, it requires tuning their ratios. In E. coli, one can assemble a multigene plasmid or use operons, and in S. cerevisiae, one can leverage in vivo homologous recombination to stitch fragments together during transformation. R. toruloides has low HR efficiency, and this means in vivo assembly is inefficient too. Typically, researchers must assemble multi-gene constructs in E. coli first, then introduce them via ATMT into R. toruloides. Each assembly and integration is a substantial effort. One case study described a functional integration of multiple genes in R. toruloides by constructing a multigene binary vector and obtaining stable integrants (Wen et al., 2020), but such successes are non-trivial. Moreover, because each integration is random, getting all pathway genes expressed at adequate levels can require screening many clones or re-integrating genes to achieve higher copy number (Liu et al., 2021; Geiselman et al., 2020b). Some groups have improved copy number by targeting the ribosomal DNA locus or other multi-copy loci, but this is still experimental. The lack of autonomously replicating plasmids also means one cannot simply transform a library of pathway variants and select as every variant must be a separate integrant line. This dramatically reduces strain optimization throughput.

4.2 CRISPR/cas and genome editing constraints

The advent of CRISPR/Cas9 technology promised easier genome editing in non-model organisms (Figure 3). Indeed, three independent groups reported CRISPR/Cas9 systems for R. toruloides in 2019 (Otoupal et al., 2019; Schultz et al., 2019; Jiao et al., 2019). These methods achieved gene disruptions or edits by expressing Cas9 and a guide RNA in the yeast. However, because no plasmids can be used in R. toruloides, the Cas9 and sgRNA expression cassettes themselves had to be integrated via ATMT or other transformation method (Wen et al., 2020). This raised the following concerns. First, the random integration of a constitutively expressed Cas9 and sgRNA could potentially cause off-target mutations or toxicity (Wen et al., 2020). Furthermore, unless Cas9 and the guide are removed later, they remain in strains, which is undesirable for industrial application of the strain. Some studies have used URA3 blaster or Cre/lox systems to remove integrated Cas9 after gene editing, but each additional step adds time. The CRISPR editing efficiency in R. toruloides is also not yet as high or predictable as what is shown in S. cerevisiae. In S. cerevisiae, a plasmid-borne CRISPR can yield >90% correct edits in one transformation, while in R. toruloides, achieving similar rates might require enriching for correctly edited cells by using a co-transformed marker or counter-selection. In summary, CRISPR is a powerful addition for R. toruloides, but its use is more convoluted than in model organisms due to the need for chromosomal integration and the host’s DNA repair bias. Recently, a CRISPR/Cas-based cytidine base editor (CBE) was developed for the first time in R. toruloides and used to disrupt four lipid metabolism genes, enabling production of 512.3 mg/L free fatty acids (Yu et al., 2025b).

4.3 Host-specific regulation and physiology

R. toruloides’ metabolism is geared towards lipid and carotenoid production as a stress response. For example, it can funnel intracellular lipid content > 67% under nitrogen limitation condition (Li et al., 2007). This means that engineered pathways may compete against or be constrained by the host’s native regulation. For example, when trying to produce a non-native compound, the availability of precursors/cofactors such as acetyl-CoA, NADPH, and ATP might be strongly influenced by R. toruloides’ inclination to produce storage lipids. A clear case is the production of isoprenoids. R. toruloides naturally makes carotenoids at ∼0.1% of cell dry weight using resources such as acetyl-CoA, NADPH, and ATP (Lyu et al., 2023), but if one attempts to push more flux into isoprenoids, the cell’s lipid synthesis may steal those resources. A recent study found that knocking out a lipid droplet protein (perilipin, LDP1) in R. toruloides led to 13% lower lipid content but 93% higher carotenoid accumulation (Jiao et al., 2021). This demonstrates a competition between lipid and terpene pathways and provides a strategy to favor one over the other. For metabolic engineers, such a host-specific phenomenon means that pathway optimization must include host metabolic rewiring. It may be necessary to delete or downregulate native genes (e.g., those involved in fatty acid synthesis or storage) to free up carbon for the desired product.

Likewise, R. toruloides has a robust stress tolerance mechanism. If an engineered pathway produces a toxic intermediate, the yeast might activate stress responses that slow down the growth or reduce the flux toward this intermediate. Various Omics such as transcriptomics or metabolomics can facilitate the understanding of these regulatory networks and will be important to identify metabolic bottlenecks.

Genomic sequencing and multi-omics (transcriptomic and proteomic) analyses of R. toruloides under different culture conditions revealed that nitrogen-limited lipid accumulation is closely linked to lipogenesis, macromolecule metabolism, and autophagy. The study also identified novel genes and pathways associated with fatty acid synthesis and lipid storage, improving understanding of oleaginicity and guiding metabolic engineering (Zhu et al., 2012).

4.4 Pathway bottlenecks and unknowns

Non-model hosts often harbor unknown enzymatic activities or regulatory circuits that can interfere with heterologous pathways. For instance, R. toruloides’ genome encodes various proteases and transporters, and the introduced heterologous proteins might be degraded faster by the promiscuous activity of proteases, or the product could be pumped out or modified in unforeseen ways. For example, expression of a plant monoterpene synthase in R. toruloides led to a mixture of products rather than a single product possibly due to side activities or substrate pooling in the host (Zhuang et al., 2019). In addition, inefficient homologous recombination of R. toruloides makes it harder to do fine-tuned edits like promoter replacements or point mutations in enzymes to relieve feedback inhibition. Thus, alleviating pathway bottlenecks (e.g., by modifying allosteric sites or introducing competing pathway knockouts) could be more cumbersome in R. toruloides. Researchers often compensate by overexpressing upstream enzymes to brute-force more flux. For example, to boost terpenoid production, R. toruloides strains have been engineered to overexpress rate-limiting mevalonate pathway enzymes (like HMG-CoA reductase) and to increase gene copy number of terpene synthases (Geiselman et al., 2020b; Liu et al., 2021). These strategies have been moderately successful, but more elegant genetic manipulations (common in E. coli or S. cerevisiae) remain difficult.

4.5 Recent developments in genetic toolbox availability

R. toruloides genetic engineering is challenging due to high GC content and the absence of replicating plasmids, requiring genome integration. To overcome these limitations, recently, the RT-EZ (R. toruloides Efficient Zipper) toolkit was developed using Golden Gate Assembly (Koh et al., 2025). The toolkit enables rapid and flexible vector construction with features such as bidirectional promoters, 2A peptides, RFP-based screening, and compatibility with Agrobacterium tumefaciens-mediated transformation. It allows assembly of up to four genes in a single reaction, greatly improving engineering efficiency.

Current expression systems rely mainly on Agrobacterium-mediated transformation, which restricts efficient multi-gene expression. This study shows that viral 2A peptides—especially Porcine teschovirus-1 2A (P2A) peptide—enable effective co-expression of multiple genes, significantly enhancing β-carotene production in engineered strains (Jiao et al., 2018). A study on constitutive promoters, a total of 31 promoters were identified, exhibiting strengths ranging from 0.1 to 19.0 times that of the commonly used strong promoter of the glyceraldehyde-3-phosphate dehydrogenase gene (P_GPD1) (Guo et al., 2023). The following study characterizes four type I and one novel type II peroxisomal targeting signals (PTS1/PTS2) and demonstrates their first application for limonene production in R. toruloides (Gao et al., 2024).

5 Case studies and standard protocols

While protocols have been established in E. coli and S. cerevisiae for almost everything from transformation to fermentation optimization through decades of research, R. toruloides is relatively new host in synthetic biology field and it requires development of methods from scratch or troubleshooting basic issues such as finding a suitable growth medium. Especially several ‘desired’ R. toruloides genetic tools such as replicative plasmids and high-efficiency mating systems for constructing hybrids are simply not available yet. Even though the field has been moving fast with reviews highlighting emerging techniques (Johns et al., 2016; Wang et al., 2016), there is still a need for broader community sharing of standard parts and strains. Despite these challenges, R. toruloides offers compelling advantages such as robust growth on cheap feedstocks, high tolerance to inhibitors, and the ability to accumulate and presumably secrete hydrophobic products, and researchers are actively expanding its synthetic biology toolkit to steadily improve the strain engineering workflows (Wen et al., 2020). The unique hurdles posed by R. toruloides have driven creative solutions that may also be applicable to other non-model organisms.

5.1 Potential and bottlenecks of isoprenoid biosynthesis in R. toruloides

Isoprenoids (terpenoids) are a broad class of valuable natural products derived from the five-carbon building blocks isopentenyl diphosphate (IPP) and dimethylallyl diphosphate (DMAPP). R. toruloides naturally possesses the mevalonate (MVA) pathway for IPP/DMAPP synthesis (Christianson, 2007; Daletos et al., 2020) which it uses to produce sterols and carotenoid pigments such as β-carotene, torulene, and torularhodin (Figure 4). This inherent pathway makes R. toruloides an intriguing host for isoprenoid production – it already has substantial flux through acetyl-CoA to IPP, and it accumulates isoprenoid end-products (carotenoids) in lipid droplets. Below, we focus on R. toruloides’ capacity for isoprene and other isoprenoids and the challenges in engineering these pathways.

5.2 Monoterpenes (C₁₀) and sesquiterpenes (C₁₅)

Beyond isoprene, R. toruloides has been actively engineered to produce larger isoprenoids. Monoterpenes (C₁₀, such as limonene, α-pinene, linalool, α-terpineol) and sesquiterpenes (C₁₅, such as farnesene, bisabolene, patchoulol, prespatane) are valuable as flavorings, fragrances, and fuel precursors (Table 1). R. toruloides’ innate carotenoid pathway naturally forms geranylgeranyl pyrophosphate (GGPP, C₂₀) from IPP via GPP (C₁₀) and farnesyl diphosphate (FPP, C₁₅) intermediates (Zhuang et al., 2019; Yaegashi et al., 2017). By inserting heterologous terpene synthase (TS) genes, researchers have diverted these intermediates to produce different terpenes. In a broad screening, Zhuang et al. expressed 16 terpene synthases from various sources in R. toruloides. 8 of them were functional, yielding nine different monoterpenes including 1,8-cineole (eucalyptol), sabinene, ocimene, α-pinene, and limonene (Wen et al., 2020). Intriguingly, some engineered strains produced a mixture of monoterpenes rather than a single product (Wen et al., 2020), indicating that R. toruloides can support multiple branching reactions or that some TS enzymes had broad product profiles in this host. For sesquiterpenes, R. toruloides has achieved significantly high titers in certain cases. Yaegashi et al. engineered a strain to produce bisabolene, a C₁₅ sesquiterpene considered as a potential bio-jet fuel, at ∼521 mg/L on glucose, and >680 mg/L on lignocellulosic hydrolysate (Yaegashi et al., 2017). Further optimization through process engineering in a 2 L bioreactor raised bisabolene titers to 2.2 g/L (Kirby et al., 2021) (Figure 5). This was achieved in high cell density cultivation using “separation-free” biomass hydrolysate (i.e., in one-pot without separating the pretreatment residuals) (Sundstrom et al., 2018; Perez-;Pimienta et al., 2019). Recently, R. toruloides was further engineered to produce 20.8 g/L bisabolene using 300 g/L corn stover hydrolysate which is the highest reported titer for this sesquiterpene (Adamczyk et al., 2025a). These results highlight R. toruloides’ robustness in industrially relevant conditions and its ability to funnel significant carbon into isoprenoids when appropriately engineered. Similarly, an engineered R. toruloides strain produced linalool (an important fragrance and a precursor of bio-jet fuel) by introducing a plant-derived linalool synthase and overexpressing heterologous MVA pathway genes (Zheng et al., 2025). Also, Lyu et al. demonstrated α-terpineol production in R. toruloides by employing a modular engineering approach (Lyu et al., 2023). They enhanced precursor supply by upregulating MVA pathway, blocked competing pathways such as carotenoid biosynthesis or fatty acid synthesis and knocked out specific genes to redirect flux. One combination–disrupting lipid accumulation while boosting MVA flux – yielded the best result, with ∼1.5 mg/L α-terpineol. While 1.5 mg/L is modest, the study clearly showed that reducing carbon flux to lipids dramatically increased terpenoid output (Lyu et al., 2023). This reinforces a key point: R. toruloides’ metabolic network can be rebalanced away from storage compounds (i.e., lipids) towards isoprenoids, but it may require knocking out native genes that would otherwise consume the precursors.

6 Perspectives – a benefit of automation tool development for R. toruloides research

Protocols in academic life science laboratories predominantly depend on the manual handling of tools, reagents, and instruments by research staff and students. Unlike industrial and clinical laboratories, the adoption of automation to enhance the throughput or substitute a less efficient manual processes remains minimal (Holland and Davies, 2020). Especially in R. toruloides engineering research, automation would play a pivotal role in accelerating research as the strain diversity generated because of random integration via ATMT is huge. In addition, increasing precision and enhancing reproducibility will be another benefit of applying automation strategy. In this section, we describe critical thoughts on the role of automation in context to R. toruloides being used as laboratory host.

6.1 High-throughput colony screening from ATMT

Automation enables rapid screening of genetic variants or growth conditions. Using automated platforms such as liquid handlers, colony pickers, and robotic workstations, researchers can test hundreds or thousands of strains without repetitive human labor, speeding up the discovery of optimized phenotypes (Figure 6). A predominant method of genetic engineering in R. toruloides relies on the genomic integration of genes of interest using ATMT. Transformation by ATMT generates many strains with various copy numbers of integrated genes at random sites on the genome. In other words, each colony on the agar plate from one ATMT process is likely to be a different strain from another colony in the same plate, and the screening of the desired phenotypes from several different colonies is a very critical but laborious process. Therefore, automation-mediated colony pick-up followed by screening or selection of preferred phenotype on microtiter plates would be pivotal when using this strain as a host.

6.2 Data generation and analysis

Machine learning (ML) has emerged as a powerful tool to predict the behavior of biological systems and enhance the Learn phase of the design, build, test and learn (DBTL) cycle, driven by advancements in high-throughput phenotyping technologies (Radivojević et al., 2020) (Figure 6). Producing the volume and quality of data required for effective ML application in biological system is quite unattainable without automation-based approaches. The current state of biology is comparable to relying on mailed paper photographs: most assays remain low-throughput and in manual way with phenotypic data predominantly generated and analyzed within individual laboratories (Carbonell et al., 2019). Access to large volume of high-quality data would empower computational biologists to develop robust theories without relying on their own experimental facilities. These data-driven theories would, in turn, enable experimentalists to design more effective experiments and address questions of broader significance (Carbonell et al., 2019). Especially in R. toruloides research, automated systems including data generating instruments such as spectrophotometers, flow cytometers, and sequencing platforms, can generate vast datasets, and integration of these datasets with machine learning models can identify patterns and predict optimal engineering strategies. This will significantly reduce the number of strains required to achieve the best performing strain, and as a result, reduce the time required to generate many differently engineered R. toruloides strains to achieve similar results in traditional engineering strategy.

6.3 Reproducibility

Over 70% of researchers have attempted and failed to replicate another scientist’s experiments, while more than half have struggled to reproduce their own. This striking statistic stems from Nature journal’s survey of 1,576 researchers who completed a brief online questionnaire on research reproducibility (Baker, 2016). Manual methods often introduce variability. Laboratory automation devices, such as liquid-handling robotics, can enhance the interlaboratory generation of more reliable and more reproducible experimental data by minimizing the influence of human variability (Linshiz et al., 2014). Automation ensures standardized protocols, which enhance reproducibility across experiments, and this is especially critical for scaling-up processes for industrial applications. Developing uniform protocols in scripted language (machine learning) for R. toruloides would be pathbreaking in the metabolic engineering of this host for preferred criteria (e.g., specific titer, rate, yield, or product).

6.4 Accelerating biomanufacturing and potential challenges

Accelerating the deployment of R. toruloides in biomanufacturing hinges on reducing bottlenecks across the Design–Build–Test–Learn (DBTL) cycle. Key priorities include expanding well-characterized genetic parts (e.g., strong/inducible promoters, terminators, integration loci), improving genome editing efficiency (e.g., optimized CRISPR/Cas systems and multiplex editing), and developing high-throughput transformation and screening pipelines. Integration of multi-omics datasets with machine learning can further refine pathway design and predict flux distributions, while adaptive laboratory evolution (ALE) can complement rational engineering to enhance robustness, substrate utilization, and product tolerance.

However, several challenges still constrain DBTL efficiency in R. toruloides. The Build phase is slowed by relatively low transformation efficiencies and limited homologous recombination, leading to reliance on random integration. In the Test phase, lack of standardized and high-throughput assays as well as genetic instability can complicate phenotype evaluation. The Learn phase is hindered by incomplete genome annotation and limited systems-level understanding, which reduce the prediction accuracy. Additionally, lipid accumulation pathways are tightly regulated and condition-dependent, which introduces variability across cycles. Together, these factors elongate iteration times and increase uncertainty, underscoring the need for more standardized tools, robust chassis engineering, and data-driven design frameworks.

7 Conclusion

Rhodosporidium toruloides represents a compelling addition to the synthetic biology toolbox, particularly for applications requiring robust lipid and isoprenoid biosynthesis from renewable and heterogeneous feedstocks. Its innate oleaginous phenotype, functional mevalonate pathway, broad substrate utilization, and stress tolerance collectively position it as a strong candidate for the sustainable production of carotenoids and other acetyl CoA–derived value-added compounds. These intrinsic advantages directly address several metabolic and physiological limitations encountered in traditional model hosts.

However, the full industrial and synthetic biology potential of R. toruloides remains constrained by significant gaps in genetic tractability. Low transformation efficiencies, limited availability of standardized genetic parts, poor homologous recombination, and reliance on random integration strategies continue to impede predictable strain engineering and high-throughput DBTL cycles. Although ATMT and emerging CRISPR/Cas-based approaches have laid important groundwork, their current limitations underscore the need for more efficient, precise, and modular genetic tools specifically tailored for this organism.

Future progress will depend on coordinated efforts to expand well-characterized promoter and terminator libraries, improve genome editing precision and efficiency, develop stable episomal or site-specific integration systems, and establish standardized engineering workflows comparable to those available in E. coli and S. cerevisiae. Addressing these challenges will not only accelerate isoprenoid and lipid-focused applications but also broaden the scope of bioproducts accessible through R. toruloides.

Automation is also essential for advancing R. toruloides as a laboratory and industrial host by enabling high-throughput strain screening, generating high-quality data for machine learning applications, and improving experimental reproducibility, thereby accelerating metabolic engineering and strain optimization.

In summary, R. toruloides stands at a transitional stage—from a promising non-model yeast to a viable industrial chassis. Continued methodological innovation and community-driven tool development will be essential to unlock its full potential and enable its routine use in a next-generation synthetic biology and biomanufacturing platform.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Nam Ngoc Pham, Joint Genome Institute, United States

Reviewed by: Zongbao K. Zhao, Dalian University of Technology, China

Hyungi Koh, Hongik University, Republic of Korea

Eric Young, Worcester Polytechnic Institute, United States