The bacterial strains used in this work are listed in Table 1. E. coli strains JM109 (Yanisch-Perron et al., 1985), and ET12567/pUZ8002 (MacNeil et al., 1992) were used for routine cloning and for conjugation or isolation of non-methylated DNA, respectively. E. coli strains were cultivated at 37°C on Luria-Bertani (LB) agar plates or in LB medium supplemented with the appropriate antibiotics. All media and routine Streptomyces techniques have been described previously (Kieser et al., 2000). S. peucetius G001 (Lambert and Ylihonko, 2008) was used as parental strain. Soy flour mannitol (SFM) agar plates were used to grow Streptomyces for phenotypical characterisation. Tryptone soy broth (TSB) was used for liquid cultivation of Streptomyces strains. The growth media were supplemented with 20 μg mL⁻¹ thiostrepton when required. Cultures were grown in a total volume of 20 mL of liquid medium in 100 mL Erlenmeyer flasks equipped with metal coils. Shake flaks were incubated in an orbital shaker with a 2-inch orbit at 200 rpm at 30°C. Due to poor sporulation of S. peucetius strains, mycelium stocks were prepared as an alternative to spore stocks. Strains were cultivated in TSB medium for 2 days, the biomass was washed with 10.3% (w/v) sucrose, resuspended in 20% (w/v) glycerol, and stored at −80°C.

All plasmids described in this work are listed in Supplementary Table S1 and primers in Supplementary Table S2. Plasmid maps were generated using SnapGene 6.0 (Supplementary Figure S1).

MAG304 was cultivated in E1 medium. Biomass was harvested after 2, 3 or 4 days of incubation (n = 3), snap-frozen in liquid nitrogen and stored at −80°C until analysis. The frozen biomass was lysed using a Bioruptor Plus (Diagenode SA) and proteins were extracted using lysis buffer [4% SDS, 100 mM Tris-HCl (pH 7.6), 50 mM EDTA]. Sample preparation for LC-MS/MS measurement was performed as described previously (Zhang et al., 2020). Briefly, total protein was precipitated using the chloroform-methanol method (Wessel and Flügge, 1984) and dissolved in 0.1% RapiGest SF surfactant (Waters Crop.) at 95 °C. The protein concentration was determined using the BCA method. Protein samples were reduced by adding 5 mM dithiothreitol (DTT) and incubated in the dark at 60°C for 30 min, followed by thiol group protection using 21.6 mM iodoacetamide and incubation in the dark at room temperature for 30 min. Subsequently, 0.1 µg of trypsin (recombinant, proteomics grade, Roche) per 10 µg of protein was added, and samples were digested overnight at 37°C. After digestion, trifluoroacetic acid was added to a concentration of 0.5%. The samples were incubated at 37°C for 30 min, followed by centrifugation to degrade and remove the RapiGest SF. The resulting peptide solution, containing 6 µg of peptides, was cleaned and desalted using StageTips (Rappsilber et al., 2007). Briefly, 6 µg of peptides was loaded on a conditioned StageTip with two 1 mm diameter C₁₈ disks (Empore, product number 2215), washed twice using a 0.5% formic acid solution, and eluted with elution solution (80% acetonitrile and 0.5% formic acid). Acetonitrile was evaporated using a SpeedVac. The final peptide concentration was adjusted to 40 ng μL⁻¹ using sample solution (3% acetonitrile and 0.5% formic acid) for analysis. Quantitative proteomics was performed as described previously (Zhang et al., 2020). Briefly, the desalted peptide solution was separated using an UltiMate 3,000 RSLCnano system (Thermo Scientific) set in a trap-elute configuration, coupled with a QExactive HF mass spectrometer (Thermo Scientific). The liquid chromatography system used a Waters nanoEase M/Z Symmetry C₁₈ trap column (5 μm, 100 Å, 180 μm × 20 mm) for peptide loading and retention, and a Waters nanoEase M/Z HSS T3 C₁₈ analytical column (1.8 µm, 100 Å, 75 μm × 250 mm) for peptide separation. The mass spectrometer was operated in positive mode with data-dependent acquisition. Raw LC-MS/MS files were analysed using MaxQuant software v2.2.0.0 (Cox and Mann, 2008) using the label-free quantification (LFQ) method.

The BGCs of doxorubicin/daunorubicin, aclarubicin and rhodomycin were visualised using clinker (Gilchrist and Chooi, 2021). DoxA homologs were identified using NCBI BLASTP search (http://blast.ncbi.nlm.nih.gov). Alignment of the obtained protein sequences was performed using Cluster Omega 1.2.4 (Madeira et al., 2022).

Enzymatic activity assays were conducted as described elsewhere in detail (Koroleva et al., 2024). The DoxA, DnrV, FDX4 and SFR proteins were produced as N-terminally 6×His-tagged recombinant proteins in E. coli TOP10 and purified by affinity chromatography using TALON Superflow resin (GE Healthcare). Proteins were concentrated using Amicon Ultra-4 10K centrifugal filters (Merck Millipore) and stored at −20°C in 40% glycerol. The proteins were analysed for purity and molecular weight using SDS-PAGE. Enzymatic activity measurements were carried out at room temperature overnight. Then, reactions were extracted with 4:1 mixture of chloroform and methanol. The extracts were evaporated using a vacuum concentrator and dissolved in methanol for HPLC analysis. HPLC analysis was performed using a Shimadzu Nexera X3 system with a PDA detector and a Phenomenex Kinetex C₁₈ column (2.6 µM, 100 Å, 4.6 × 100 mm). The column was run at a flow rate of 0.5 mL min⁻¹, using 0.1% formic acid, 15% acetonitrile and 85% H₂O as solvent A, and 100% acetonitrile as solvent B. A gradient was employed starting at 100% A for 2 min, then 0%–60% B for 18 min, 100% B for 4 min, and finally 100% A for 5 min. The absorbance of the samples was recorded at 490 nm. The reaction products were identified by comparison to reference compounds for 3–6, 9, 11 and 12. For enzymatic assays with compound 9 and 11, high resolution electrospray ionization mass spectra were recorded on a Waters Acquity RDa detector using a Waters XBridge BEH C₁₈ column (5 μm, 130 Å, 4.6 × 30 mm). The column was run at a flow rate of 0.8 mL min⁻¹, using 0.1% formic acid in H₂O as solvent A, and 0.1% formic acid in acetonitrile as solvent B. A gradient was employed starting at 2%–100% B for 132 s, then 100% B for 18 s, 100%–2% B for 18 s min, and finally 2% B for 12 s. The reaction products were identified by comparison to reference compounds for 11 (Supplementary Figure S19) and 12 (Supplementary Figure S20). Compound 10 was annotated based on calculated exact mass (Supplementary Figure S21).

To investigate the resistance of S. peucetius to anthracyclines, 5 µL of mycelium stock was spotted at a concentration of 1.0.10⁴ colony forming units (CFU) per spot on SFM agar plates supplemented with increasing concentrations of 13-deoxydaunorubicin 3) (van Gelder et al., 2023), doxorubicin 6) (Accord Healthcare Limited), N,N-dimethyl-13-deoxydaunorubicin 9) (van Gelder et al., 2023) or N,N-dimethyldoxorubicin (12) (Qiao et al., 2020). After 3 or 4 days of incubation at 30 °C, growth was examined visually.

The first challenge in the biosynthetic production of N,N-dimethyldaunorubicin (11) and N,N-dimethyldoxorubicin (12) is to provide the N,N-dimethylated amino sugar l-rhodosamine that cannot be naturally biosynthesised by S. peucetius. l-Rhodosamine occurs in various natural anthracyclines, such as aclarubicin and rhodomycin B (Figure 1A). The aclarubicin biosynthetic pathway of S. galilaeus features two N-methyltransferases, AclP and AknX2, which can catalyse the conversion of the activated amino sugar TDP-l-daunosamine to TDP-l-rhodosamine (Step 1, Figure 3C). Additionally, the aclarubicin biosynthetic pathway contains the glycosyltransferase/auxiliary protein pair AknS and AknT, which normally catalyse the glycosylation of aklavinone with rhodosamine to yield aclacinomycin T (14). However, the doxorubicin biosynthetic pathway differs from the aclarubicin pathway by the presence of DnrF, which catalyses the 11-hydroxylation of aklavinone to ɛ-rhodomycinone (16) before glycosylation (Figure 3C). If AknS/AknT would be able to glycosylate ɛ-rhodomycinone (16) with rhodosamine, it would result in the production of ɛ-rhodomycin T 7) (Step 2).

The construct pRDS (Han et al., 2011) harbours the methyltransferase genes aclP and aknX2 and the glycosyltransferase genes aknS and aknT from S. galilaeus, and genes for the biosynthesis of TDP-l-rhodosamine from S. peucetius and Streptomyces venezuelae. This construct is a derivative of pWHM3 (Vara et al., 1989), an unstable multicopy vector that harbours a thiostrepton cassette (Supplementary Figure S1). pRDS was introduced into G001 via protoplast transformation, resulting in strain MAG301. The recombinant strain harbours all the genes for the enzymes required for both Step 1 and Step 2 (Figure 3A). To evaluate the effect of introducing pRDS on the metabolite profile, both G001 and MAG301 were cultivated in E1 medium (Ylihonko et al., 1994) with added HP20 resin. The resin binds the anthracyclines, thereby preventing product inhibition and toxicity. After 4 days of incubation at 30 °C, metabolites were extracted using acetone, dried, re-dissolved in 80% acetonitrile and analysed using liquid chromatography-mass spectrometry (LC-MS). The LC-MS data were processed using MZmine, resulting in a list containing all the mass features and their peak areas detected in each crude extract. In cases where reference compounds were unavailable, annotation was based on the predicted m/z values of the [M + H]⁺ adduct ions and expected fragmentation patterns (Supplementary Table S3).

The LC-MS chromatograms of G001 and MAG301 (G001 pRDS) exhibited distinctly different mass peaks (Figure 3B). The main peak observed in the chromatogram of G001 corresponded to daunorubicin 5), and a smaller peak was annotated as doxorubicin 6). Furthermore, minor peaks were annotated as the precursors 13-deoxydaunorubicin 3) and 13-dihydrodaunorubicin 4), respectively (Figure 3B). The introduction of pRDS to G001 (MAG301) resulted in a shift in the metabolite profile (Figure 3B). The main mass peaks detected in the LC-MS chromatogram of MAG301 were annotated as to ɛ-rhodomycin T 7) and 4-methoxy-ɛ-rhodomycin T (13). These results indicate that ɛ-rhodomycinone (16) was successfully glycosylated with rhodosamine, yielding ɛ-rhodomycin T 7) (Figure 3C). The main peak corresponding to 4-methoxy-ɛ-rhodomycin T (13) indicated that 4-O-methylation activity was successful, but 10-decarboxylation and the final hydroxylation steps were unsuccessful. In the native doxorubicin pathway, DnrP catalyses 15-methylesterase activity of rhodomycin D 1), and subsequently DnrK catalyses 10-decarboxylation. The accumulation of 4-methoxy-ɛ-rhodomycin T (13) in the engineered pathway suggested that DnrP cannot catalyse 15-methylesterase activity of ɛ-rhodomycin T 7) or 4-methoxy-ɛ-rhodomycin T (13).

Furthermore, minor peaks were observed in the chromatogram of MAG301, which were annotated as aclacinomycin T (14) and 4-methoxy-aclacinomycin T (15), respectively (Figure 3C). The presence of these compounds suggests that 11-hydroxylation of aklavinone to ɛ-rhodomycinone (16) by DnrF was not complete. All in all, the expression of sugar N-methyltransferases and glycosyltransferases of the aclarubicin biosynthetic pathway resulted in the incorporation of rhodosamine onto ɛ-rhodomycinone (16) (Figure 3C). However, optimisation of the further tailoring reactions toward N,N-dimethyldoxorubicin (12) are required.

The chromatogram of MAG301 also contained a minor mass peak that corresponds to daunorubicin 5), indicating partial glycosylation of ɛ-rhodomycinone (16) with daunosamine rather than with rhodosamine (Figure 3B). The native doxorubicin glycosyltransferase DnrS may prefer TDP-l-daunosamine over TDP-l-rhodosamine, which could explain the observed by-products. To address this hypothesis, the native glycosyltransferase gene dnrS within the doxorubicin BGC was deleted (Figure 3A).

A deletion mutant of dnrS was created using a method published previously (Świątek et al., 2012), which is based on the unstable multicopy plasmid pWHM3 (Vara et al., 1989). A knock-out construct was generated that harbours the about 1 kb regions upstream and downstream of dnrS interspaced by the apramycin resistance gene aacC4 flanked by loxP recognition sites (pGWS1431, Supplementary Figure S1). The knock-out construct was introduced to G001 via protoplast transformation. The presence of the loxP recognition sites allowed the efficient removal of the apramycin resistance cassette by introduction of the pUWLCRE construct for expression of the Cre recombinase (Fedoryshyn et al., 2008). Consequently, we obtained a mutant where the entire coding region of dnrS was deleted, which is designated MAG302 (Figure 3A).

MAG302 (G001 ΔdnrS) was cultivated in E1 medium, and the metabolite profile was analysed in a similar manner as described above. As expected, no glycosylated anthracyclinones could be detected in the LC-MS chromatogram of MAG302 (Figure 3B). The main peak in the chromatogram of MAG302 corresponded to ɛ-rhodomycinone (16) (Supplementary Figure S3). Subsequently, pRDS was introduced into MAG302 via protoplast transformation to generate MAG303 (Figure 3A). In contrast to MAG301, MAG303 harbours only the aclarubicin glycosyltransferases, but not the doxorubicin glycosyltransferase. MAG303 (G001 ΔdnrS pRDS) was cultivated in E1 medium, and the metabolite profile was analysed in a similar manner as described above. Notably, the production of daunorubicin 5) was completely abolished in MAG303 (Figure 3B). When solely aclarubicin glycosyltransferases are present, ɛ-rhodomycinone (16) undergoes glycosylation exclusively with rhodosamine instead of daunosamine. These results indicate that the deletion of dnrS successfully directed the pathway toward glycosylation with rhodosamine.

After successful sugar N-methylation (Step 1) and glycosylation (Step 2), the next bottleneck in the N,N-dimethyldoxorubicin biosynthetic pathway lies in the downstream tailoring reactions (Step 3). The production of 4-methoxy-ɛ-rhodomycin T (13) indicated that DnrP could not catalyse 15-methylesterase activity of N,N-dimethylated substrates. Consequently, we searched for an alternative enzyme with 15-methylesterase activity of ɛ-rhodomycin T 7).

In S. purpurascens, the rhodomycin biosynthetic pathway harbours the enzyme RdmC, a homologue of DnrP, which accepts ɛ-rhodomycin T 7) as substrate (Grocholski et al., 2015). The coding region of rdmC was positioned behind the constitutive ermE* promoter (Bibb et al., 1985), and cloned into the integrative vector pSET152 (Bierman et al., 1992), resulting in pGWS1432 (Supplementary Figure S1). The construct was introduced to MAG303 via conjugation, resulting in strain MAG304. The resulting strain harbours the methyltransferases and glycosyltransferases genes from the aclarubicin BGC (on pRDS), the methylesterase gene rdmC from the rhodomycin BGC (on pGWS1432), and a deletion of the native glycosyltransferase (dnrS) (Figure 4A). MAG304 (G001 ΔdnrS pRDS + rdmC) was cultivated in E1 medium, and the metabolite profile was analysed in a similar manner as described above. The main peak observed in the chromatogram of MAG304 was annotated as N,N-dimethyl-13-deoxydaunorubicin 9), and a smaller peak as N,N-dimethyl-13-dihydrodaunorubicin (10). Remarkably, a minor peak was annotated as N,N-dimethyldaunorubicin (11), which is one of the targeted products (Figure 3B). Additionally, the production of ɛ-rhodomycin T 7), 4-methoxy-ɛ-rhodomycin T (13), aclacinomycin T (14) and 4-methoxy-aclacinomycin T (15) was abolished (Figure 4B). The metabolite profile of MAG304 indicated successful replacement of DnrP by RdmC. The 15-methylesterase activity of RdmC is essential for the 10-decarboxylation moonlighting activity of DnrK, thus enabling the production of N,N-dimethyl-13-deoxydaunorubicin 9). The final three steps from N,N-dimethyl-13-deoxydaunorubicin 9) to N,N-dimethyldoxorubicin (12) are catalysed by the cytochrome P450 monooxygenase DoxA (Figure 4C). For G001, the main peak in the chromatogram corresponds to daunorubicin 5), which indicates that the final 14-hydroxylation step toward doxorubicin 6) is inefficient, while the two 13-hydroxylation steps catalysed by DoxA exhibit high efficiency. In contrast, for MAG304, the main peak corresponds to N,N-dimethyl-13-deoxydaunorubicin 9), indicating that all steps catalysed by DoxA are inefficient when the compounds are N,N-dimethylated.

Taken together, introduction of rdmC into the engineered G001 strain proved to be a successful strategy to achieve 10-decarboxylation of ɛ-rhodomycin T 7). Combined activity of RdmC and DnrK resulted in the production of N,N-dimethyldaunorubicin (11). The accumulation of the precursor N,N-dimethyl-13-deoxydaunorubicin 9) indicates that the next bottleneck in the engineered pathway is the final enzyme DoxA.

To enable further genetic engineering, the genes located on pRDS and pGWS1432 were combined on a single integrative vector. For this, the coding region of rdmC with an engineered R15 ribosomal binding site (Bai et al., 2015) and L3S1P47 terminator (Chen et al., 2013) was amplified by PCR. The aclP coding region was amplified by PCR from pRDS. The two fragments were introduced into EcoRI-digested pSET152. The entire DNA fragment containing aknT-aknS-desIII-desIV-dpsG-dpsH-dnmT-dnmZ-dnmU was excised from pRDS and cloned into aclP-rdmC:pSET152 to generate pGWS1433-v1. Illumina sequencing of pRDS indicated the presence of an EcoRI site in the intergenic region between aclP and aknX2 (Supplementary Figure S1). Consequently, our cloning strategy resulting in the loss of a fragment of 70 bp from the construct, which we then re-introduced. For this, the dnmU-dnmV-dnmJ-aknX2 region was amplified from pRDS, including the missing sequence at the end of aknX2. The aclP-rdmC region was amplified by PCR from pGWS1433-v1. The two fragments were cloned into BamHI-linearised pGWS1433-v1 via Gibson assembly (Gibson et al., 2009), resulting in pGWS1433 (Supplementary Figure S1). The correct sequence of the whole construct was confirmed by Sanger sequencing. The resulting construct was introduced to MAG302 (G001 ΔdnrS) via conjugation to generate MAG305 (Figure 4A). MAG304 and MAG305 harbour the same heterologous genes. However, in the case of MAG305, all the heterologous genes are located on pSET152, whereas in the case of MAG304 all heterologous genes are located on pRDS, except for rdmC which is located on pSET152. MAG305 was cultivated in E1 medium, and the metabolite profile was analysed and annotated in a similar manner as described above. Similar as for MAG304, the main peak in the chromatogram of MAG305 corresponded to N,N-dimethyl-13-deoxydaunorubicin 9), while N,N-dimethyldoxorubicin (12) could not be detected (Figure 4B), indicating that this strain could be use for further engineering.

Heterologous expression of enzymes from the aclarubicin and rhodomycin pathways to G001 resulted in the biosynthesis of N,N-dimethyldaunorubicin (11). The results indicated that the bottleneck in the engineered biosynthetic pathway are the final three tailoring steps catalysed by DoxA. It is worth noting that the final 14-hydroxylation step catalysed by DoxA is notably inefficient in the doxorubicin biosynthetic pathway (Malla et al., 2010). In fact, enzyme kinetic experiments revealed that the catalytic constant V_max is 520-fold lower for the conversion of daunorubicin 5) to doxorubicin 6) compared to the preceding step (Walczak et al., 1999).

During the construction of the engineered G001 strains, we noticed that the development of the strains was blocked, most likely caused by the production of cytotoxic anthracyclines. On SFM agar plates, G001 and MAG301 (G001 pRDS) exhibited a distinctive ‘bald’ phenotype, characterised by the absence of aerial hyphae and spores (Figure 6A). Notably, the deletion of the glycosyltransferase gene dnrS (MAG302) led to in the production of white aerial hyphae. However, upon introduction of the aclarubicin glycosyltransferase and methyltransferase genes (MAG303) the strain reverted to the ‘bald’ phenotype without aerial hyphae production. The introduction of the methylesterase gene rdmC (MAG304) further crippled the strain, evident by the reduction in colony size. We hypothesised that product toxicity could inhibit the development of the strains and consequently impact productivity. Therefore, we conducted a microbial inhibition assay involving doxorubicin, N,N-dimethyldoxorubicin (12) and two precursors.

To investigate the resistance of G001, 5 µL of mycelium stock was spotted at a concentration of 1.0.10⁴ CFU per spot on SFM agar plates with increasing concentration of 13-deoxydaunorubicin 3), doxorubicin 6), N,N-dimethyl-13-deoxydaunorubicin 9) or N,N-dimethyldoxorubicin (12). After 3 days of incubation at 30 °C, growth was examined visually. Surprisingly, G001 exhibited no growth at 5 μg mL⁻¹ N,N-dimethyldoxorubicin (12), whereas its growth was uninhibited at 25 μg mL⁻¹ doxorubicin (Supplementary Figure S5). Furthermore, the main product of the engineered strain MAG304, N,N-dimethyl-13-deoxydaunorubicin 9), inhibited the growth of G001 at a concentration of 10 μg mL⁻¹, whereas no growth inhibition was observed when exposed to 25 μg mL⁻¹ 13-deoxydaunorubicin 3). These results suggest that the toxicity of the produced compounds may inhibit the biosynthesis of N,N-dimethylated anthracyclines.