Actinomycetes: what more can they offer in an era of metabolic engineering and artificial intelligence?

Understanding actinomycetes’ metabolism

Complex biosynthesis of secondary metabolites^1–4 in wild types of streptomycetes usually yield low titres for metabolites. Biosynthetic steps in primary and secondary metabolism of actinomycetes as well as their regulation, however, can be suitable targets for metabolic engineering for overproduction of compounds.^5,6 Streptomycetes’ secondary metabolism is dependent upon precursor supplies derive from primary metabolism and such dependency can be optimised for industrial needs, and if coupled with available genetic manipulation techniques superior results can be achieved.^5,6 One example is cellular nitrogen metabolism where relevant nitrogen-containing compounds are supplied as precursors into secondary metabolism, or this metabolic cycle can be induced by compounds like ammonium, nitrate, amino acids and polyamines.^7,8 A combination of metabolic engineering strategies targeting relevant genes for nitrogen supply combined with feeding strategies can thus be a very effective strategy for the optimisation of the production of compounds in streptomycete strains.^7,8 Comparative metabologenomics can also be a powerful approach to expose genomic features that differentiate strong, antibiotic producers from weaker ones. Rational discovery efforts might thus prove useful if they are coupled with currently used standard methods such as engineering and modification of molecular signalling.^9,10

Biosynthetic gene clusters and heterologous expressions

Genome analysis of bioactive compound producing actinomycetes revealed that this cluster of bacteria contained more biosynthetic gene clusters (BGCs) compared to other bacteria.⁵ These analyses exposed the potential for modification of their already known biosynthetic state to a new superior level.¹¹ Moreover, the Next-Generation Sequencing (NGS) facilitated (1) exploration of untapped biosynthetic pathways from uncultured actinomycetes hidden in diverse under-explored and extreme habitats; and (2) revealed uncharacterised biosynthetic pathways embedded in the genome of cultured actinomycetes.¹¹ Regulator-based engineering methods such as activator overexpression, repressor deletion, promoter replacement and cluster refactoring and amplification have also been used to promote overproduction of known antibiotics and activation of cryptic antibiotic BGCs for the discovery of new antibiotics.¹¹ Examples include heterologously expressed cryptic biosynthetic gene clusters like the one from Streptomyces prunicolor yielding the novel bicyclic peptide prunipeptin.¹² Production of a novel macrolactam compound by the heterologous expression of a large cryptic biosynthetic gene cluster of Streptomyces rochei (IFO12908)¹³ or the heterologous expression of a cryptic gene cluster from Streptomyces leeuwenhoekii (C34T) yielding a novel lasso peptide, Leepeptin.¹⁴ In another example expression of a Streptomyces albus subsp. chlorinus NRRL B-24108 genomic library in the modified S. lividans ΔYA9 and S. albus Del14 strains resulted in the production of new representatives of the Pyrrolobenzodiazepines family of natural products.¹⁵

Activation of the silent mutaxanthene pathway and yield enhancement by single cell mutant selection (SCMS) used by Akhgari et al.¹⁶ resulted in 5-fold yield increase in Amycolatopsis orientalis [NRRL F3213]. Furthermore, Akhgari et al.¹⁶ developed a fluorescence-activated cell sorting (FACS) methodology that was capable of reproducibly identifying high-performing individual cells from a mutant population directly from liquid cultures. Their method resulted in 5-fold yield improvement for industrial cholesterol oxidase ChoD from the producer strain.

Past investigations focussing on the existing regulator-based construction process of high-yield strains to enhance antibiotic biosynthetic pathways often did not include optimisation of other key factors (precursors or cofactors) that influence antibiotic production.¹⁷ However, recent studies for development of effective regulatory strategies to achieve antibiotic overproduction, focused on elucidation of regulatory mechanism of antibiotic biosynthesis.¹⁷ A significant number of recent studies have been conducted on the regulatory mechanisms of cluster-situated regulators (CSRs) or pleiotropic regulators in the control of antibiotic production by influencing the expression levels of BGCs.^17,18 An example study involved Milbemycin production in streptomycetes that revealed that the production was tightly controlled by hierarchical regulators (CSRs, pleiotropic or global regulators, environmental and physiological cues).¹⁸ Using transcriptome-guided identification methods, Ye et al.¹⁸ identified a four-component system that modulates milbemycin biosynthesis by influencing gene cluster expression, precursor supply and antibiotic efflux. Accordingly, understanding of the regulation of milbemycin biosynthesis from the use of metabolic engineering of the native host led to improved milbemycin production.¹⁸

Metabolic engineering

Multilevel metabolic engineering strategies were used for Daptomycin yield improvement.^19,20 Examples include Lyu et al.,¹⁹ who used Streptomyces roseosporus to reconstruct high-quality daptomycin overproducing strain L2797-VHb by (i) precursor engineering (i.e. refactoring kynurenine pathway), (ii) regulatory pathway reconstruction (i.e. knocking out negative regulatory genes arpA and phaR), (iii) byproduct engineering (i.e. removing pigment), (iv) multicopy BGC, and (v) fermentation process engineering (i.e. enhancing O₂ supply). They were able to achieve 565% higher titre for Daptomycin. Other examples of yield increases are given in Table 1.

Heng et al.²² noted that large-scale, high-throughput engineering efforts were encountering barriers due to the difficulties in manipulating the often large and GC-rich BGCs of actinomycetes, however, advances powered by synthetic biology simultaneously entered the scene and accelerated DNA synthesis and assembly. These advances were: (i) Gibson assembly, (ii) Golden Gate assembly and (iii) yeast recombination. Moreover, complementing these technologies several types of nucleases became available for precise genome editing, examples include (i) meganuclease I-SceI, (ii) zinc-finger nucleases, and (iii) transcription activator-like effector nucleases.

Deletion of the pathway-specific repressor papR3 in Streptomyces pristinaespiralis yielded sole production of pristinamycin I, facilitating downstream purification while avoiding the undesirable synergistic cytotoxic effects of pristinamycin I and II.^23,24 Similarly, deleting an in-cluster acyltransferase improved dalbavancin yields by 25% in Nonomuraea gerenzanensis sp. ATCC 39727 by eliminating an inhibitory acetylated product.²⁵

CRISPR-Cas systems

An adaptive defence mechanism in bacteria called CRISPR-Cas (CRISPR-associated proteins) systems has been proven to be easily programmable and rapidly adapted for genome editing.²² CRISPR-Cas technology as a synergistic combination of advances in genomics, synthetic biology and bioinformatics has also yielded opportunities to improve the production of natural products from actinomycetes. CRISPR-Cas, which is a versatile, programmable DNA targeting tool has been increasingly employed to accelerate host and pathway engineering of actinomycetes in particular for genome editing of non-Streptomyces actinomycetes.²²

Fazal et al.²⁶ coupled conventional marker exchange mutagenesis and polymerase chain reaction (PCR)-targeted recombineering techniques with CRISPR-Cas genome editing and disrupted five endogenous BGCs producing antimycin, candicidin, albaflavenone, surugamide and fredericamycin in Streptomyces albus S4. As a result, they removed the antifungal and antibacterial properties of this organism and subsequently used this genome-reduced ‘chassis’ host to produce targeted natural products such as actinorhodin, cinnamycin and prunustatin.

Rare actinomycetes

Approximately two-thirds of all known antibiotics are produced by actinomycetes, predominantly by the members of the genus Streptomyces.^1,4 It is believed that the actinomycetes are the main source of all microorganism-derived bioactive substances so far discovered, with a small percentage of the total originating from the rare actinomycetes whose isolation frequencies are much lower than the streptomycetes. Examples include Micromonosporaceae derived antibiotics (e.g. Gentamicin), followed by contributions from the Pseudonocardiaceae (e.g. Vancomycin) and Thermomonosporaceae (e.g. Kijanimicin). Rare actinomycetes however, can be a valuable source of novel compounds, if improved selective isolation strategies are implemented to increase the frequency with which they are currently isolated.^3,27

Heng et al.²² noted that the identification of suitable vectors, promoters and selection markers is a key first step for successful genome editing on rare actinomycetes. In addition, CRISPR-Cas-based genome or transcriptome engineering methods require the simultaneous introduction of multiple elements, such as Cas-effector proteins, guide RNAs, repair templates and sometimes recombinases or reporter genes, into target cells. Adoption of the original Streptomyces CRISPR-Cas systems without additional modifications by the rare actinomycetes has been achieved. Examples include successful direct adoption of the original Streptomyces CRISPR-Cas systems without additional modifications in Actinoplanes and Micromonospora strains, proving that these genetic parts can also work in some rare actinomycetes.²⁸

Artificial intelligence-driven drug discovery

Although successful discovery of new natural products with biological activity using conventional methods termed the ‘Waksman Platform’ has been achieved²⁹ it has been a time-consuming process, with high costs and low success rates. Accordingly, artificial intelligence (AI) and its subdiscipline machine learning (ML) can now bypass the limitations and challenges of these traditional approaches and hence can accelerate the drug discovery processes in an efficient, cost- and time-effective ways.²⁹ Interest will be greater again on actinomycetes due to AI-driven biological activity predictions capitalising on actinomycete bioactive compound databases.

Recent notable examples include discovery of Halicin by group of researchers at the Massachusetts Institute of Technology (MIT) in 2020, which was the first groundbreaking drug discovery where ML was used.^29,30 Although a known compound, its targets got expanded with the use of ML and covered activity against multidrug-resistant pathogens like Clostridiodes difficile, Acinetobacter baumannii and Mycobacterium tuberculosis.³¹ In their approach, the MIT researchers employed a deep-learning algorithm with a diverse dataset of ~2500 US Food and Drug Administration-approved drugs and natural products and the model employed recognised different features of compounds that were associated with various chemical structures.³⁰ Subsequently they correlated chemical structure data with compounds’ antibacterial properties and target directed selection was made.^29,30 Outcomes highlighted the power of ML for novel drug discovery.

AI-based drug discovery delivered another breakthrough and facilitated the discovery of Abaucin with potency against Acinetobacter baumannii – a serious pathogen causing blood, urinary tract and lung infections.³² A paradigm shift has now begun in drug discovery as these advanced technologies empowered researchers with tools so that the analyses of large number of datasets can be achieved and molecular properties of a compound can easily be predicted, and potential drug candidates can be identified with greater precision and speed. As a result, AI and ML have emerged as invaluable tools in the quest for novel therapeutics from natural sources.^29,33,34

Conclusions

The expression of heterologous genes in microbial hosts has now become a common practice. To find streptomycetes that produce unknown antibiotics and secondary metabolites first there is a need to better understand the morphology, physiology, metabolic changes and signalling that occur during the switch between primary and secondary metabolisms, as well as the supply of precursors from primary metabolism.^30–33,35 Moreover, other approaches, such as mutagenesis and genomics can aid modification of biosynthetic gene clusters in streptomycetes leading to production of high-value antibiotics and other secondary metabolites.^30–33,35

Natural products have been key contributors to the discovery of potent drugs; however, due to the challenges faced when conventional systems were used, their discovery rate has declined. However, thanks to ML, these limitations can now be overcome as complex datasets can be modelled.³² Moreover, ML can be applied to the interpretation of metabolomic data, which can be utilised for the efficient dereplication of previously characterised microbial secondary metabolites to prevent wasting of resources and time. However, ML will also have limitations if high quality and Open Access natural product (NP) datasets are not available to further increase the utility of ML in NP discovery.^29,35