Complex multicellularity is linked with expanded specialized metabolite production in microorganisms

Abstract

Single-celled microorganisms evolving to form cooperative multicellular units represent a key innovation in the history of life. Multicellularity enables cellular differentiation and, as such, is critical to the emergence of biological complexity. Here we investigate the hypothesis that multicellularity facilitated massive expansions in specialized metabolite production, including polyketides, non-ribosomal peptides and terpenes, across bacteria and fungi. Systematic investigations reveal that unicellular taxa are largely limited in biosynthetic potential for secondary metabolite production. By contrast, profound biosynthetic expansions in Actinomycetota, Cyanobacteriota, Myxococcota, Pezizomycotina and Agaricomycetes coincide with independent origins of multicellular development, such as the formation of mycelia and fruiting bodies. These multicellular lineages are concomitantly enriched in carbohydrate-active enzymes, suggesting that ancient chemical innovations were further shaped by catabolic processes. By linking intraspecific cooperation to the evolution of specialized metabolism, this study advances our understanding of major evolutionary transitions and provides a framework for discovering chemical compounds to counter antimicrobial resistance.

The evolutionary emergence of complex multicellularity in bacteria and fungi, such as the ability for mycelial growth, is strongly associated with increased genomic carriage of biosynthetic machinery for producing specialized chemical metabolites.

Main

Evolving the ability to produce and secrete specialized metabolites, including antibiotics, represents a key innovation in microbial life^1,2. Unlike primary metabolites essential for basic cellular functions, such as amino acids and vitamins, specialized metabolites are leveraged by microorganisms to directly mediate complex interspecies interactions, compete for resources and establish ecological niches^3,4. Biosynthetic gene clusters (BGCs) encode the critical molecular machinery that enable microorganisms to produce this vast diversity of specialized chemicals. By encoding complex enzymatic pathways for synthesizing bioactive compounds, BGCs have allowed microorganisms to evolve potent, diverse and often highly specific metabolites that dramatically expand their functional capabilities to interact with and adapt to complex environmental challenges. The natural products derived from microbial specialized metabolites represent a critical source of bioactive compounds with immense potential for drug discovery, having yielded numerous life-saving pharmaceuticals such as antibiotics, immunosuppressants and anticancer agents⁵. Understanding the evolution of natural product synthesis by microorganisms is crucial for deciphering the role secreted metabolites have in mediating ecological interactions and shaping microbial diversity, and has transformative potential in drug discovery.

Multicellularity represents another major transition in the history of microbial life, enhancing biological complexity and biodiversity⁶. The advantages conferred by multicellularity include improved resource acquisition, environmental resilience and division of labour⁷. Multicellularity has emerged independently multiple times across the tree of life, including within the bacterial phyla Cyanobacteriota, Myxococcota and Actinomycetota. Cyanobacteriota represent one of the earliest origins of multicellularity, with some species forming filamentous chains and differentiated cells that carry out specialized functions such as nitrogen fixation in heterocysts⁷. The emergence of multicellularity in cyanobacteria has also been linked with taxonomic diversification and the Great Oxidation Event⁸. Some members of the bacterial phylum Myxococcota show a form of collective behaviour in which individual cells coordinate to form complex multicellular structures called fruiting bodies in response to starvation conditions, with species such as Myxococcus xanthus showing sophisticated cell-to-cell communication and cooperative movement through social motility^7,9.

The Actinomycetota represent a particularly fascinating example of bacterial multicellularity, with some taxa forming complex filamentous structures through hyphal growth that enables them to colonize and penetrate solid substrates such as soil particles and plant material¹⁰. The multicellular lifestyle, as particularly evident in the genus Streptomyces, supports sophisticated developmental programmes regulated by complex signalling networks, including the formation of aerial hyphae and spores in response to environmental cues¹¹. Although other bacterial taxa can also show allorecognition¹² or form structured biofilms¹³, these traits often pale in complexity to the multicellular structures and life cycles of Actinomycetota, Myxococcota and Cyanobacteriota or have emerged more recently and, as such, did not shape the genomic content of the deeper ancestral lineage from which it evolved.

Historically, filamentous bacteria and fungi within the phyla Actinomycetota and Ascomycota, respectively, have served as the two primary microbial sources of natural products, contributing many lifesaving drugs, such as streptomycin, tetracycline and erythromycin from bacteria, alongside penicillin, cyclosporine and lovastatin from fungi¹⁰. Filamentous Ascomycota and Actinomycetota show complex life cycles during which they can form mycelia and intricate multicellular structures with distinct cell types as well as generate spores to survive stressful conditions and aid geographic dispersal to colonize new niches^10,14. More recently, species belonging to the bacterial phyla Cyanobacteriota and Myxococcota have emerged as valuable and rich sources of novel natural products with biomedical potential^15–17. In addition to containing taxa enriched in their ability to produce natural products, these microbial lineages also contain taxa that form cooperative multicellular units⁷. The concomitant occurrence of multicellularity and enriched biosynthetic repertoires in such microbial lineages^4,18,19 suggests that the emergence of multicellularity is closely associated with the expansion of genomes and carriage of biosynthetic machinery. This is an intriguing idea that until now has not been systematically investigated across all major domains of microbial life.

Building on the intriguing pattern of key sources of natural products being multicellular, we systematically investigate biosynthetic capacities across the bacterial domain and fungal kingdom to test our hypothesis that multicellularity facilitated expansions in specialized metabolite production. To evaluate the breadth of known microbial specialized metabolite classes—including polyketides, non-ribosomal peptides and terpenes—we comprehensively annotated for over 80 types of BGCs using antiSMASH²⁰. We highlight that the majority of bacterial and archaeal taxonomic orders are on average very limited in the number of specialized metabolites they encode per genome. In stark contrast, orders within the phyla Cyanobacteriota, Myxococcota and Actinomycetota are broadly composed of extant genera rich in machinery for specialized metabolite production. Within all three phyla, the ancestral enrichment in biosynthetic repertoire coincides with early, independent origins of multicellularity. For Actinomycetota, we show that the emergence of multicellularity and biosynthetic trait expansions align with the origin of the class Actinomycetes, the most sourced reservoir of bacterial natural products. We next investigate biosynthetic potential across the fungal kingdom and show that most lineages are substantially limited in canonical specialized metabolism relative to the Dikarya, which are renowned for forming complex multicellular structures such as fruiting bodies. Through comparative and association analyses, we show that increased biosynthetic capacity across microbial lineages is strongly associated with enrichments in carbohydrate-active enzymes as well as proteins mediating self-recognition in fungi. Finally, we discuss the association between multicellularity and genomic expansions of machinery for specialized metabolism. We frame these genomic expansions through a public goods context, wherein extracellular molecules are subject to tragedy of the commons dynamics that shape their production and regulation.

Results

Ancestral expansions in biosynthetic capacity are limited across the bacterial tree of life

To begin evaluating the association of multicellularity with enrichments in specialized metabolism, we first determined distributions of per-genome biosynthetic carriage for 74 well-sequenced bacterial and archaeal orders. Both the median number of BGCs as well as the median BGC-ome size, the summed length of predicted BGC regions in a single genome, were assessed across 4,732 genus-representative genomes from the orders (Fig. 1a and Supplementary Table 1). A strong correlation was observed between the two metrics, and while complete genomes were preferentially used where available, BGC-ome size was largely assessed throughout this study to alleviate overassessment of biosynthetic capacity arising from assembly fragmentation issues. In addition, although BGC-ome sizes reported might be inclusive of flanking genomic regions unrelated to specialized metabolism, trends in BGC-ome size were correlated with other measures of biosynthetic capacity, including those applying stricter criteria for defining BGCs and their proximities (Extended Data Fig. 1).

Fig. 1. Ancestral expansions in biosynthetic content coincide with developments of multicellularity.

a, Median count of BGC regions and median BGC-ome size across 74 bacterial and archaeal orders. b, Number of distinct Pfam domains from key biosynthetic genes used for BGC detection relative to the order’s phylogenetic branch sum, visually standardizing for core genome evolutionary breadth. c, Timetree of 74 bacterial and archaeal orders. Bars depict the number of distinct metabolite types (NPAtlas nodes) per order. Box plots depict total BGC counts for representative genomes (n = 4,732). Box plot boundaries correspond to the 5th, 25th, 50th, 75th and 95th percentiles. The branch width corresponds to the standard deviation of time estimates, with greater variability shown as bolder branches. Evolutionary expansions in median BGC-ome size and shift magnitudes (${\hat{\beta }}_b$) are indicated to the right. d, Relationship between distinct multicellularity genes (defined previously³¹) and BGC-ome size for species-representative Cyanobacteriota genomes (n = 484). The point colour indicates the number of akinete-related genes. e, Comparisons of BGC-ome sizes for Actinomycetes genomes with (n = 1,615) and without (n = 2,383) SsgB, a genetic marker for complex morphology in the class, alongside BGC-ome sizes for other classes from Actinomycetota (n = 357). f, Relationship between distinct social lifestyle-related genes (defined previously²⁹) and BGC-ome size for species-representative Myxococcota genomes (n = 84). The point colour indicates the number of sporulation-specific proteins. For d–f, box plots are shown in Tukey’s style, in which they extend to the largest value that is less than 1.5 times the IQR from the top hinge of the box and the smallest value that is greater than 1.5 times the IQR from the bottom hinge of the box. GOE, Great Oxidation Event; Pfam, a database of protein families and domains; Mb, megabase pairs.

Source data

Extended Data Fig. 1. Different metrics for assessing biosynthetic potential are highly correlated.

The relationships between BGC-ome size run using relaxed configurations in antiSMASH for 4,732 diverse bacteria and archaea with a, the count of BGC, b, the summed length of high-confidence BGC regions featuring NRPS or PKS protoclusters, c, the proportion of entire genomes corresponding to BGCs, and d, the summed length of only the protocores of BGC regions.

Source data

The distribution of biosynthetic genomic content was heavily skewed, with the vast majority (65 of 74) of bacterial and archaeal orders having a median count of 5 or fewer BGC regions or a maximum median BGC-ome size of less than 150 kilobase pairs (kb) (Fig. 1a). None of the eight archaeal orders had a median count of five or more BGCs, confirming their known lack of biosynthetic traits²¹. Large-scale annotation efforts have also previously suggested that BGC carriage is limited in many bacterial species²², and databases cataloguing the taxonomic origins of known natural products have shown that they are disproportionately sourced from just six microbial phyla^23,24.

To establish which taxonomic orders show substantial increases in their median BGC-ome size relative to their phylogenetic contexts, we next constructed a timetree of the 74 bacterial and archaeal orders and applied a phylogeny-informed approach leveraging Ornstein–Uhlenbeck processes (Fig. 1b,c and Supplementary Table 2). This analysis highlighted five distinct orders belonging to three phyla: Cyanobacteriota, Myxococcota and Actinomycetota. Because each of the three phyla are enriched in biosynthetic content across multiple genera, we posit that they correspond to cases of ancestral enrichments in biosynthetic traits. While it is difficult to resolve whether these orders experienced large-scale acquisition of biosynthetic machinery via ancestral horizontal gene transfer (HGT) events, they do not appear to have elevated rates of recent BGC-specific transfer relative to other orders (Extended Data Fig. 2a–c).

Extended Data Fig. 2. Systematic assessments of the impact of recent horizontal gene transfer on BGC evolution across bacteria, archaea, and fungi.

a, A 2D histogram shows the overall distribution of BGCs as a factor of length and codoff-based codon usage discordance percentile to their genomic contexts across 74 bacterial and archaeal orders. b, For the same 74 orders, codoff-based codon usage discordance percentile distributions for BGCs (n = 21,843) are shown as boxplots. c, Histograms of BGC lengths for species-representative genomes across the five bacterial phyla from with the most characterized BGCs. The coloring of BGCs shows whether BGCs were found to overlap or reside on temperate phages or plasmids. The piecharts to the right side show the overall percentage of BGCs found on phages or plasmids irrespective of length. d, codoff-based codon usage discordance percentile distributions for BGCs (n = 6,647) from fungal genomes are shown as boxplots. e, For the same fungal taxa, distributions of the count of predicted tyrosine site-specific recombinases per genome (n = 256) are shown. Counts per genome were capped at 10. For e and d, the fill of boxplots indicate whether they exhibit complex multicellularity and the outline color corresponds to their life cycle. Boxplots are shown in Tukey’s style; they extend to the largest value that is less than 1.5 times the interquartile range (IQR) from the top hinge of the box and the smallest value that is greater than 1.5 times the IQR from the bottom hinge of the box.

Source data

In alignment with recent research highlighting the immense diversity of BGCs that exists within Cyanobacteriota, Myxococcota and Actinomycetota¹⁷, we observed that specific orders within these phyla also featured more distinct domains within key biosynthetic proteins than would be expected based on the extent of their ribosomal diversification (Fig. 1b). In addition, as described in the introduction, the three phyla highlighted feature some of the most well-characterized model bacterial species used for studying complex multicellularity and life strategies. Four other bacterial orders have median counts of six or more BGCs per genome and also include species that show multicellularity-like or cooperative behaviour. They include the orders Desulfobacterales, which feature multicellular magnetotactic bacteria²⁵, and Paenibacillales, which encompasses species that have complex physiologies and can perform organized colony swarming²⁶.

Ancestral enrichments in bacterial biosynthetic capacity are associated with multicellularity and other factors

We next investigated whether a variety of genomic, functional and ecological traits correlate with increased biosynthetic capacity across the three phyla highlighted as having undergone ancestral biosynthetic expansions, in addition to the phyla Pseudomonadota and Bacillota (Supplementary Notes, Extended Data Fig. 3 and Supplementary Table 3). Across 17,844 species-level representative genomes from the phyla, multiple genomics-inferred features were positively correlated with BGC-ome size, including (1) carriage of oxidative phosphorylation-related enzymes, (2) increased coding usage of specific amino acids and (3) an inferred preference for mesophilic growth. After genome size, increased BGC-ome size was most strongly associated with enrichment for carbohydrate-processing enzymes, including in Pseudomonadota and Bacillota. Furthermore, multiple independent genome-wide association studies (GWAS) across phyla revealed that several plant-related carbohydrate usage enzyme families were significantly associated with biosynthetic capacity. The only family found to be associated with BGC-ome size in four or more of the five phyla was glycosyl hydrolase family 19, chitinases that are rare in microorganisms. Within Streptomyces, homologues of this enzyme family are hypothesized to have been acquired via HGT from plants²⁷. These analyses support a pronounced increase in biosynthetic capacity that occurred across diverse bacteria coinciding with the increased availability of biopolymers from land plants²⁸.

Extended Data Fig. 3. Factors associated with increased biosynthetic capacity for the five bacteria phyla contributing most natural products.

a, Systematic relationships of various quantitative factors to BGC-ome size using two-sided Spearman’s ranked correlation across all phyla (General) and for each of the five phyla individually. b, A lower-triangle heatmap showing the Spearman correlation between 38 quantitative variables. c, The BGC-ome size distributions of species-level representative genomes (n = 17,844) with and without glycosyl hydrolase family 19. For Myxococcota, the association was not significant after multiple testing correction. Boxplots are shown in Tukey’s style; they extend to the largest value that is less than 1.5 times the interquartile range (IQR) from the top hinge of the box and the smallest value that is greater than 1.5 times the IQR from the bottom hinge of the box. d, The relationship of BGC-ome size to genomic predictions for ideal temperature for growth of species in Celsius using the amino-acid frequency-based formula by Kurokawa et al. 2022. The yellow shaded area represents the temperature range for mesophiles (optimal growth temperature is between 20-45 C°). e, The relationship between median BGC-ome size and the number of MIBiG characterized BGC records for bacterial genera with ten or more characterized BGCs. Coloring corresponds to phyla, the size of the dots to the number of metagenomes the genus was inferred to be present in from the SandPiper database, and the shape corresponding to whether metagenomes genera were found in corresponded primarily to host-associated or environment, also based on the SandPiper database.

Source data

Within the three phyla determined to have had ancestral expansions in biosynthetic content, positive associations were observed between counts for known protein markers of multicellularity and sporulation^29–31 with BGC-ome size (Fig. 1d–f). For Cyanobacteriota, carriage of eight or more multicellularity-associated genes and three or more akinete-associated genes was significantly associated with increased BGC-ome size (P = 2.59 × 10⁻⁵⁹, Wilcoxon rank-sum statistic = 16.19). Similarly, for Myxococcota, genomes with greater than 200 social lifestyle-associated genes and 2 or more sporulation-associated genes had larger BGC-ome sizes (P = 4.65 × 10⁻¹³, Wilcoxon rank-sum statistic = 7.14). The substantial overlap in social lifestyle-associated orthologues among genomes spanning diverse taxonomic orders further suggests that sociality was ancestral to the phylum Myxococcota or, at least, multiple classes within it (Extended Data Fig. 4). Among Actinomycetota, the carriage of ssgB homologues, which are involved in the process of sporulation and associated with increased genome size and complex physiology³⁰, was also a clear indicator of increased biosynthetic potential (P < 1 × 10⁻³⁰⁰, Wilcoxon rank-sum statistic = 39.96). Furthermore, many species within the class Actinomycetes, even those without ssgB homologues, had larger BGC-omes compared with other classes in the phylum (P = 4.50 × 10⁻¹²⁶, Wilcoxon rank-sum statistic = 23.85; Fig. 1e).

Extended Data Fig. 4. Many genes contributing to sociality in Myxococcota are likely ancestral to the phylum.

a, A phylogeny of genus representatives from the Myxococcota phylum, with two major clades, is shown alongside whether corresponding genera have been experimentally validated as displaying a social lifestyle (fill circle = social lifestyle present, empty circle = social lifestyle absent; data from Murphy et al. 2021), distributions of their BGC-ome size and the count of distinct social lifestyle-associated genes as described by Murphy et al. 2021 across species-level representative genomes (n = 70). The presence or absence of 622 individual social lifestyle associated from the study as identified in the genus representative genomes is depicted as a heatmap. Boxplots boundaries correspond to 5th, 25th, 50th, 75th, and 95th percentiles. Venn diagrams show the overlap of distinct social lifestyle associated genes observed between b, Sorangium and at least one of four genera from a separate clade of the phylum, all of which are known to form fruiting bodies, as well as between c, their respective clades.

Source data

The last common ancestor of Actinomycetes was multicellular and possessed a large reservoir of biosynthetic traits

To further understand the enrichment of biosynthetic content and emergence of multicellularity in the class Actinomycetes, we first used phylogenomics to determine the evolutionary relationship between 133 genera with high-quality genomes from across the phylum Actinomycetota (Fig. 2a and Supplementary Table 4). A maximum-likelihood phylogeny was constructed based on single-copy-core orthologue groups revealing two clades within the Actinomycetes class, which we refer to as clade 1 and clade 2 (Fig. 2a,b and Supplementary Table 4). On the basis of literature^10,32, both clade 1 and clade 2 were found to feature prominent filamentous and mycelium-forming genera and also to genomically encode proteins associated with multicellularity, such as SsgB, SepX, RamA and RamB^30,33,34.

Fig. 2. The LCA of Actinomycetes was morphologically complex and possessed a rich biosynthetic repertoire.

a, Maximum-likelihood phylogeny of representative genera from the Actinomycetota phylum based on conserved single-copy orthologue groups. The Actinomycetes class is highlighted in pink (clade 1) and violet (clade 2), indicating two major ancestral partitions observed for the class. The estimated age interval for the LCA of Actinomycetes corresponds to the 89% high-density interval predicted using MCMCTree. The presence of orthologue groups corresponding to genetic markers associated with multicellularity, sporulation or mycelium formation in Actinomycetota, primarily characterized in Streptomyces, is shown across the phylogeny of the phylum. Box plots farthest to the right showcase the BGC-ome size of species-representative genomes (n = 4,021) for each genus. Box plot boundaries correspond to the 5th, 25th, 50th, 75th and 95th percentiles. b–d, A schematic of the proposed evolutionary model of Actinomycetota (b), formula (c) and overview of the D.stat metric used to assess support for the evolutionary model (d) to infer whether smaller-genome Actinomycetes in clade 2 share a more recent common ancestor with Streptomycetaceae than with families belonging to clade 1. e, Distributions of the D.stat computed for the single-copy core alignment with different combinations of families from clade 2 (box plots) and clade 1 (grids). Clade 1 families included Micromonosporaceae (n = 639), Pseudonocardiaceae (n = 2,130) and Mycobacteriaceae (n = 1,917). Box plots are shown in Tukey’s style, where they extend to the largest value that is less than 1.5 times the IQR from the top hinge of the box and the smallest value that is greater than 1.5 times the IQR from the bottom hinge of the box. f, Maximum-likelihood phylogeny of SsgB proteins with sequences from smaller genomes (<6 Mb) from clade 2 marked. g, Phylogenetic distance of Actinomycetes genomes to the LCA node in the phylogeny correlates with estimated cumulative gene loss from ancestral state reconstruction inference. A linear model was fitted; the grey band corresponds to a 95% confidence interval.

Source data

The phylogenetic positioning of filamentous genera suggests that multicellularity was ancestral to the Actinomycetes class. This is supported by multicellularity being present across most clade 1 lineages as well as clade 2 featuring the early-diverging and filamentous family of Streptomycetaceae. This suggests that morphologically simpler genera from clade 2, such as Micrococcus, have experienced large-scale gene loss and a loss of physiological complexity. While ancestral loss of morphological complexity has previously been suggested within the order of Mycobacteriales in clade 1 (ref. ¹¹), to our knowledge, it has not been hypothesized to have occurred at large scale, beyond one order³⁵, within clade 2 Actinomycetes. Due to low bootstrap values at certain ancestral nodes in our phylogenomic model, we further assessed the proposed topology for Actinomycetes evolution using several additional phylogenomic and evolutionary approaches (Supplementary Notes, Fig. 2c–g and Extended Data Fig. 5). This process included developing a statistical framework to assess highly conserved sites along conserved and single-copy orthologues, which revealed consensus support that unicellular taxa from clade 2 share a more recent last common ancestor (LCA) with Streptomycetaceae than with morphologically complex families from clade 1.

Extended Data Fig. 5. Assessment of support for different topological models of Actinomycetota.

a, A phylogenetic network constructed from individual gene trees using SplitsTree4. Select genera are marked along with information on whether literature review indicates complex multicellularity exists for each such genus. b, An alternate topological model, Model B, for the evolution of the Actinomycetota phylum as proposed in Lewin et al. 2016 and by Gao and Gupta 2012. c, Under Model B, the ratio of BBAA & AABA sites to BABA and ABAA sites should be equivalent. d, A maximum-likelihood phylogeny constructed on alignments from 84 near (>80%) single-copy core is concordant with model B, placing Streptomyces into clade-1 with other morphologically complex Actinomycetes. e, A neighbor-joining tree based on Hamming distances of shared genes using hamtree also places Streptomyces with morphological complex Actinomycetes from clade-1. f, A consensus phylogeny based on individual gene trees for 186 core ortholog groups (include multi-copy genes) supports Model A.

Source data

A multicellular LCA of Actinomycetes coinciding with the broad enrichment of biosynthetic traits in its extant genera further supports that a large biosynthetic repertoire was ancestral to this class. Genome streamlining in less morphologically complex taxa across Actinomycetes also probably explains why BGCs are elevated across the class but natural product discovery has still been largely limited to filamentous actinomycetes²⁴ (Extended Data Fig. 6a). Some BGCs were also probably lost in these genera together with the ability to filament and form mycelia. To investigate the conservation of specialized metabolism across the phylum Actinomycetota, we assessed orthologue groups found to be key components of BGCs in two or more Actinomycetes. Of the 428 biosynthetic orthologue groups identified, 366 (85.5%) were found within BGC contexts in both major clades of the class (Fig. 2g, Extended Data Fig. 6b and Supplementary Table 5). Furthermore, of those 366 orthologue groups, 266 (72.7%) were either exclusive to Actinomycetes or exclusively found within a BGC context within Actinomycetes. This includes the orthologue group corresponding to 3-dehydroquinate (DHQ) and aminoDHQ synthases, involved in shikimate and 3-amino-5-hydroxybenzoic acid (AHBA) biosynthesis, respectively. Phylogenetic investigations further revealed that key enzymes involved in AHBA biosynthesis, AHBA synthase and aminoDAHP synthase, emerged within Actinomycetes (Supplementary Notes and Extended Data Fig. 7). AHBA is a key precursor of the functionally diverse ansamycin class of specialized metabolites³⁶, which the antibiotic rifamycin belongs to.

Extended Data Fig. 6. Additional genetic, ecological, and biosynthetic information for Actinomycetota.

a, A collapsed version of the phylogeny presented in Fig. 2a where each leaf corresponds to a distinct family. b, Distribution of families across metagenomes from different habitats based on SandPiper derived data shown as a dotplot. c, The number of genomes within GTDB-R214 belonging to the family shown as a barplot (black). d, The number of distinct characterized BGCs from MIBiG where at least 50% of key biosynthetic genes were detected at >95% amino acid identity being shown. e, The median copy-count for each Resfams resistance element is shown for families. f, Conservation of 366 ortholog groups found as core biosynthesis genes in both clade-1 and clade-2 Actinomycetes genomes is shown across a phylogeny of the Actinomycetota phylum. g, Details on the counts of genomes/genera with each of the 366 ortholog groups identified as corresponding to key protocore genes of BGCs in genomes from both major partitions of Actinomycetes. Upwards bars correspond to the number of Actinomycetes genomes with each ortholog group whereas downward bars indicate the number of non-Actinomycetes genomes with the ortholog group.

Source data

Extended Data Fig. 7. Ansamycin BGCs and rifamycin resistance elements are conserved across both major clades of Actinomycetes.

a, Schematic of the BGC underlying rifamycin biosynthesis in Amycolatopsis from MIBiG. Polyketide synthase and AHBA biosynthesis genes are shown in dark red and light red, respectively. b, Phylogeny of AHBA synthase, RifK, and homologs. A distinct, Actinomycetes-specific clade of RifK was identified. Within this clade a subclade corresponded to RifK from primarily BGCs involved in biosynthesis of naphthalenic ansamycins. Tracks from left to right show: (i) whether RifK homolog from a genome belonging to the orders of Mycobacteriales [clade-1, pink] or Streptomycetales / Streptosporangiales [clade-2, violet], (ii) whether +/- 50 kb of genomic context was able to be extracted on both sides of the RifK homolog, (iii) the count of IS-associated proteins in the flanking context of RifK homologs, and (iv) the count of predicted PKS proteins in the flanking context of RifK homologs. c, A schematic of a BGC predicted to encode for synthesis of a naphthalenic ansamycin found within a genomic context with signatures of being a mobile-genetic element. d, Average amino-acid identity plot from detected homologous regions to the rifamycin BGC [MIBiG entry BGC0000136] as a query against all Actinomycetota genomes in GTDB R214 using fai. e, A heterotachy-aware, maximum-likelihood, multi-gene-based phylogeny across homologous BGCs identified by fai to rifamycin exhibiting ≥ 60% AAI and ≥ 25% shared genes with the query BGC. Select characterized BGCs from MIBiG were also included. f, A heatmap depicting the presence of domain-resolution ortholog groups across the core-subset of BGCs predicted or shown to encode for naphthalenic ansamycins. g, When considering this set of BGCs, the sharing of domain-resolution ortholog groups was assessed based on taxonomic origins between pairs of BGCs (n = 378) and depicted as boxplots in Tukey’s style. That is, boxplots extend to the largest value that is less than 1.5 times the interquartile range (IQR) from the top hinge of the box and the smallest value that is greater than 1.5 times the IQR from the bottom hinge of the box. h, A clinker-based visualization of sequence similarity between different characterized BGCs from MIBiG. i, A phylogeny of DAHP synthases from Actinomycetota with RifH homologs from MIBiG ansamycin(-like) BGCs marked in purple. j, Percentage of Actinomycetes or other Actinomycetota genomes with different types of DAHP synthases and inferred RifH homologs. k, The percentage of genomes within families across the Actinomycetota phylum carrying a rifamycin-associated element (RAE) sequence motif, often co-located with and involved in the regulation of rifamycin resistance enzymes, are shown. l, A phylogeny of homologous proteins of the rifamycin resistance contributing HelR enzyme from Streptomyces venezuelae, with the clade to which the query belongs to highlighted in red. The colored track indicates the taxonomic grouping of the genome from which the homolog was identified.

Source data

While rifamycin resistance is observed across multiple bacterial phyla³⁷, within Actinomycetota, genetic and protein markers associated with rifamycin resistance were largely specific to Actinomycetes and common across clade 1 and clade 2 of the class, including many genera lacking morphological complexity (Supplementary Table 6). Additional investigations into known and uncharacterized ansamycin-related BGCs revealed that although horizontal transfer has probably impacted their evolution, sequence divergence due to evolution by descent has also expanded the breadth of chemicals they synthesize. Specifically, when comparing closely related BGCs for the production of naphthomycin ansamycins, BGCs from diverse clade 1 orders were found to show greater overlap in genetic content with each other than with BGCs from clade 2 genomes (Supplementary Notes and Extended Data Fig. 7). Together, these results suggest that the ansamycin and AHBA biosynthesis machinery were encoded by early Actinomycetes and build on past evolutionary studies that estimate the origins of other biosynthetic enzymes to overlap with the emergence of Actinomycetes³⁸. They also show how some genes in specialized metabolism can arise from those involved in primary metabolism^1,39.

Fungal biosynthetic enrichments associate with developments in complex multicellularity and haplontic life cycles

In addition to the domain Bacteria, the eukaryotic kingdom Fungi has historically served as another key reservoir of natural products and the source of several important drugs including the antibiotic penicillin, the immunosuppressant cyclosporine and the cholesterol-lowering statins. The kingdom also contains diverse species capable of forming complex cooperative units, such as some Ascomycota, which can form mycelia resembling structures produced by some bacterial Actinomycetes. Similar to filamentous Actinomycetes, filamentous Ascomycota are also renowned for their diverse chemical repertoire, suggesting that enrichment in biosynthetic potential in Fungi is also associated with multicellularity. To systematically investigate BGC enrichment across Fungi, we first constructed a high-resolution phylogeny of the kingdom using highly conserved genes to assess the relationship of 312 genomes, including several recently sequenced early-diverging taxa⁴⁰ (Fig. 3, Extended Data Fig. 8a and Supplementary Tables 7 and 8).

Fig. 3. Phylogenetic overview for enrichments in BGC-ome size across kingdom Fungi.

a, Maximum-likelihood phylogeny of Fungi rooted with outgroups using UFCG marker genes. The innermost bars (black) indicate genome size; the asterisk indicates that Gigaspora rosea (522.9 Mb) is capped at 250 Mb. The middle bars (green) depict the total count of carbohydrate-active enzyme (CAZy) homologues. The outer bars (grey shades) show the BGC-ome size using more relaxed and stringent criteria for defining BGCs. Clades enriched in BGCs are highlighted: Neocallimastigomycota (gold), Ascomycota (blue) and Basidiomycota (red). Classes within the clade labelled as BGC-enriched Pezizomycotina include Sordariomycetes, Dothideomycetes, Eurotiomycetes, Leotiomycetes, Lecanoromycetes, Xylonomycetes and Arthoniomycetes. b, Occurrence of NRPS, different PKS types and select terpene synthesis domains across fungal clades. ‘Other early-diverging Fungi’ corresponds to the taxa Rozella allomycis, Paramicrosporidium saccamoebae and Mitosporidium daphniae.

Source data

Extended Data Fig. 8. Evolutionary and biosynthetic trends across the kingdom of fungi.

a, The phylogeny for the fungal kingdom pruned for 34 representative fungal taxa is shown alongside a heatmap depicting the species: BGC-ome size (from smallest to largest: blue [0], gold [1,811,838], red [3,623,676]), multicellularity type (complex multicellularity according to Nagy et al. 2018; yeasts (which can be dimorphic) = gold; simple multicellularity (for example sporangium / rhizoid formation) = blue), dominant ploidy status (haploid-dominant based on Amses et al. 2022 analysis or literature = red; diploid+ based on Amses et al. 2022 = blue; unknown = grey), zoosporic status (non-zoosporic = red; zoosporic = blue), lifestyle (free-living = red; host-associated / free-living = gold; endoparastic = blue), total CAZyme count (from smallest to largest: blue [0], gold [624.5], red [1,249]). Phylogenies for b, ketosynthase domains from PK synthases and c, condensation domains from NRP synthetases are shown with inner tracks depicting whether domains are from characterized BGCs in MIBiGv3.1 (dark blue) or early-diverging fungal genomes (non-Dikarya & non-Zoopagomycota/Mucoromycota; grey). Outer tracks show the domain class based on NaPDoS2 annotation. Comparisons of BGC-ome size are shown between d, Neocallimastigomycota (n = 3) and other non-dikaryon fungi (n = 71), e, BGC-enriched Pezizomycotina (n = 89)and other fungi (n = 187), and f, Agaricomycetes (n = 47), non-Agaricomycetes Basidiomycota (n = 37), and non-dikaryon fungi (n = 74). For def, boxplots are shown in Tukey’s style, where they extend to the largest value that is less than 1.5 times the interquartile range (IQR) from the top hinge of the box and the smallest value that is greater than 1.5 times the IQR from the bottom hinge of the box. The association of g, genome size and h, protein count with BGC-ome size. Genomes are colored in accordance with clade coloring in Fig. 3a. R2 values were computed using linear regressions adjusted for phylogenetic relationships between genomes.

Source data

Next, we conducted comparative genomics exploring the biosynthetic repertoire across the fungal tree of life. As with bacteria, we show that most fungal lineages are composed of taxa with limited biosynthetic potential, with ~77% of distinct non-dikaryotic genera having ten or fewer BGC regions or BGC-ome sizes of 500 kb or shorter (Fig. 3a). Among Fungi, two lineages (Rozella, Paramicrosporidium and Mitosporidium and the Blastocladiomycota) have the smallest BGC-ome sizes and lack polyketide synthatases (PKSs) and non-ribosomal peptide synthetases (NRPSs) (Fig. 3b). These trends and the lack of BGCs in Fonticula, a close outgroup for Fungi, support the conclusion that limited biosynthetic potential is symplesiomorphic in Fungi—that is, limited biosynthetic potential is probably an ancestral characteristic of the kingdom. Fungal genomes from the Chytridiomycota, Mucoromycota and Zoopagomycota all have relatively small BGC-ome sizes but contain a diverse set of terpene domains as well as some PKS and NRPS containing BGCs, including those substantially diverged from sequences underlying the synthesis of known specialized metabolites (Fig. 3b and Extended Data Fig. 8b,c). Interestingly, the phylum Neocallimastigomycota showed enrichment of BGCs relative to chytrids and other non-dikaryon Fungi (P = 2.48 × 10⁻³, Wilcoxon rank-sum statistic = 2.81; Extended Data Fig. 8d). These anaerobic fungi commonly inhabit the guts of herbivores, can form morphologically complex rhizoids⁴¹ and have a high biosynthetic capacity that recent research has suggested was mostly acquired from bacteria via ancient lateral transfer events⁴². Although we found limited genomic indications of recent HGT of BGCs in fungi (Extended Data Fig. 2de), over the course of fungal evolution, HGT has probably contributed extensively to the biosynthetic diversity of the fungal kingdom^43,44.

Next, in our evaluation of the evolution of biosynthesis in Fungi, we find support for two independent origins of BGC enrichment occurring separately within subkingdom Dikarya, which are able to form dikaryons, cells with two nuclei containing separate genomes that are physically associated and co-replicate^18,40. Many members of this subkingdom show physiological complexity and form macroscopic sexual fruiting bodies, such as cup-shaped ascocarps and mushroom basidiocarp structures. Specifically, phylogenetic analyses indicated that minimal BGC enrichment is symplesiomorphic in the Dikarya, with the sister lineage to Dikarya, the Mucoromycota, and early-diverging lineages of both the Ascomycota and Basidiomycota having relatively small BGC-ome sizes. Among Ascomycota, a monophyletic subdivision within the Pezizomycotina was determined to be highly enriched in BGCs relative to all other Fungi (P = 2.34 × 10⁻²⁹, Wilcoxon rank-sum statistic = 11.19; Fig. 3a and Extended Data Fig. 8e), a result previously identified by others^45,46. While hyphal multicellularity evolved early in fungal evolution⁴⁷, consistent with our observations in bacteria, the ancestral expansion of biosynthetic content in the Pezizomycotina coincides with developments in complex multicellularity, such as the formation of macroscopic fruiting bodies. Within Basidiomycota, the class Agaricomycetes represents a second origin of BGC enrichment with taxa generally featuring larger BGC-ome sizes than other Basidiomycota species as well as non-dikaryotic fungal species (P = 3.87 × 10⁻¹², Wilcoxon rank-sum statistic = 6.84 and P = 6.68 × 10⁻¹⁴, Wilcoxon rank-sum statistic = 7.40, respectively; Fig. 3a and Extended Data Fig. 8f). As with the Pezizomycotina, the Agaricomycetes represent the other major fungal group well known for possessing complex multicellularity⁴⁸.

We next investigated which orthologous families of proteins associate with increased biosynthetic capacity in Fungi. To do so, we performed comparative genomics and GWAS (Fig. 4a). One trait revealed as enriched in BGC-rich dikaryotic fungal clades are proteins implicated in functioning as heterokaryon incompatibility (HI) factors. Heterokaryon or vegetative incompatibility (HI) proteins enable self-identification among filamentous fungi^49,50 and mechanistically involve autophagy, which are key properties of many multicellular organisms. A search for proteins with domain architectures associated with HI revealed that many species within BGC-enriched Pezizomycotina and Agaricomycetes feature hundreds of proteins with these domains in their genomes, in particular the heterokaryon incompatibility protein (HET) domain (Fig. 4bc, Extended Data Fig. 9a,b and Supplementary Table 9). The correlation between BGC-ome size and the number of proteins with HI-associated domains was further supported at the genus resolution within Aspergillus (Extended Data Fig. 9c–h and Supplementary Table 10). Outside these two clades, other fungi dedicate a lower proportion of their proteins for mediating HI (P = 2.81 × 10⁻¹⁶, Wilcoxon rank-sum statistic = 8.09).

Fig. 4. Genetic correlates of biosynthetic potential across Fungi.

a, Phylogenetic distribution of proteins predicted to function in carbohydrate utilization, virulence, and heterokaryon or vegetative incompatibility (HI) across Dikarya. Estimated age intervals for the LCA of BGC-enriched Pezizomycotina and Agaricomycetes represent 89% high-density intervals (MCMCTree). GH, glycosyl hydrolase. b,c, Distribution of BGCs and HI-associated proteins across large scaffolds (>1 Mb) in Hyaloscypha hepaticicola (b) and Pisolithus tinctorius. (c). The total counts of suspected HI-associated proteins per genome are shown in parentheses. d, Association between carbohydrate utilization protein count and BGC-ome size. Points are coloured according to clades in Fig. 3a. R² values were calculated using phylogenetic linear regression. e, Box plots of pairwise genome fluidity values (n = 6,342) standardized by phylogenetic distance (capped at 10.0). The box frame colour indicates the life cycle category; the fill colour indicates the presence of complex multicellularity. Box plots are shown in Tukey’s style, in which they extend to the largest value that is less than 1.5 times the IQR from the top hinge of the box and the smallest value that is greater than 1.5 times the IQR from the bottom hinge of the box. Mb, megabase pairs.

Source data

Extended Data Fig. 9. Overview of HI proteins across fungi and associations with BGC-ome size and genome fluidity across the genus of Aspergillus. from fungal genomes.

a, Overview of characterized proteins involved in HI from Neurospora crassa, Podospora anserina, and Cryphonectria parasitica. Domain architecture schematics are not to scale. Conservation of domain architectures across different fungal taxonomic groups is shown as the right-three columns. b, An approximate maximum-likelihood phylogeny of the HET domain from proteins across the fungal kingdom. From top to bottom, the first track indicates whether the domain is from BGC-enriched Pezizomycotina species (light blue), Agaricomycetes (salmon); the track in different shades of purple indicates the strength of homology between a HET domain and HET domains from characterized HI proteins; the following track indicates whether the HET | NACHT | WD40 domain architecture, similar to het-E from P. anserina; next the first bargraph indicates the length of the protein the HET domain was extracted from; finally the bottom bargraph shows the number of distinct Pfam domains co-found with HET in proteins. Co-occurrence information of the ten domains most commonly found co-occurring with HET in proteins are shown as the outer-tracks. The relationship between the number of suspected HI-associated protein with c, BGC-ome size, d, the proportion of genome size corresponding to BGCs, e, the genome size, and f, the number of total CAZy homologs is shown across 78 genomes representative of different species belonging to the genus Aspergillus. g, The core-genome phylogenetic relationship of the 78 Aspergillus species is shown alongside barcharts depicting the total number of suspected HI-associated proteins they have and their BGC-ome sizes. Clades were visually determined based on similar HI count distributions. h, The relationship between the number of distinct auxiliary ortholog groups (not belonging to the 80% loose core) with the summed branch length of clades from the core genome phylogeny is shown. Two clades with high counts of HI-associated proteins, the A. flavus and A. luchuensis clades, had more expansive pangenomes than some clades with lower HI counts, such as the A. fumigatus clade.

Source data

Given their function, the enrichment of HI-associated proteins across Pezizomycotina is in alignment with perspectives that they primarily exist in a vegetative state as haplontic homokaryons^18,40. To assess the genomic impacts of life cycles, we next developed a phylogenetically standardized measure of genome fluidity and used it to determine that clades with greater genomic fluidity tended to be haplontic (Fig. 4e and Extended Data Fig. 10a). Similar trends were observed for BGC-ome size distributions, with haplontic clades generally showing greater biosynthetic potential (Extended Data Fig. 10b). Despite having fewer BGCs, Agaricomycetes might simply appear to have more expansive pangenomes than Pezizomycotina because they are more likely to be in a dikaryotic form and harbour two distinct genomes at the time of sampling for genome sequencing⁴⁰.

Extended Data Fig. 10. Comparisons of unstandardized genome fluidity and BGC-ome size across fungal clades.

The distributions of a, unstandardized genome fluidity (not accounting for phylogenetic proximity between genomes) for pairs of genomes (n = 6,342) and b, the BGC-ome size per genome (n = 256) is shown for different fungal clades. Boxplots are shown in Tukey’s style, where they extend to the largest value that is less than 1.5 times the interquartile range (IQR) from the top hinge of the box and the smallest value that is greater than 1.5 times the IQR from the bottom hinge of the box.

Source data

Our investigations further revealed that several specific orthologue groups predicted to function in carbohydrate processing or plant virulence are either associated with increased biosynthetic capacity across Fungi or enriched in BGC-rich Pezizomycotina and Agaricomycetes relative to other Dikarya (Fig. 4a and Supplementary Tables 11 and 12). Carbohydrate-processing enzyme enrichment was more strongly associated with biosynthetic capacity than even genome size in fungi (Fig. 4d and Extended Data Fig. 8g,h). Thus, provided that both Pezizomycotina and Agaricomycetes probably originated around the same time as the emergence of land plants^51,52, intricate relationships with plants and the development of certain life strategies are probably associated with biosynthetic enrichments in the two dikaryotic clades in addition to developments in multicellularity and mechanisms for self-recognition^53–55.

Discussion

The production of bioactive specialized metabolites is a hallmark of microbial diversity and drug discovery. In this Article, we systematically confirm that most lineages of microbial life carry a limited number of BGCs, as inferred by genomic annotation of canonical biosynthetic content. We then show that in three bacterial phyla and two fungal subphyla, signatures of ancestral expansions in biosynthetic traits coincide with early developments of multicellularity. Furthermore, our evolutionary analyses infer that the LCA of the class Actinomycetes, formerly Actinomycetia, which features many biosynthetically gifted genera including Streptomyces, was multicellular and potentially able to form advanced structures such as aerial hyphae. While hyphal formation is probably ancestral to the fungal subkingdom Dikarya⁴⁷, complex multicellular fruiting bodies likely arose through separate, independent evolutionary events in the biosynthetically rich lineages of Pezizomycotina and Agaricomycetes. The broad co-occurrence of complex multicellularity with BGC enrichment in extant lineages of bacterial and fungal taxa suggests that intraspecific cooperation helped spur ancestral expansions in biosynthetic capacity to produce specialized metabolites, including natural products that serve as key drugs in human health.

Other multicellular organisms, especially plants, are known to produce various specialized metabolites⁵⁶. Interestingly, a reanalysis of recent biosynthetic characterizations across the plant phylum Streptophyta⁵⁷ suggests that vascular land plants also experienced an expansion in BGCs that parallels increased morphological complexity (Supplementary Notes). Given this, as well as reports that other multicellular Eukarya, such as slime moulds and animals, also contain machinery for the biosynthesis of specialized metabolites^58,59, future systematic studies in the Eukarya might reveal that our findings apply across the tree of life.

The benefits microorganisms derive from producing natural products are obvious. These potent bioactive molecules mediate complex intraspecies and interspecies interactions within free-living and host-associated microbial communities, including in the human microbiome. However, these specialized metabolites require complex enzymatic machinery and are energetically expensive compounds to produce. Multicellular growth and cellular differentiation probably help mitigate the fitness costs associated with parallel production of specialized metabolites. For example, multicellular organization in Streptomyces has recently been shown to mitigate the fitness costs associated with the production of multiple specialized metabolites through division of labour owing to genetic heterogeneity⁶⁰. Such dynamics probably facilitated the evolution of expanded genomes that encode more BGCs.

The higher biomass obtained from multicellular growth would also facilitate the use of secreted carbohydrate-active enzymes to deconstruct the more complex biopolymers that would otherwise be unavailable, such as chitin and cellulose, which would concomitantly support the biosynthesis of bioactive molecules to suppress competitors for these energy-rich substrates. Indeed, our finding that BGC expansions in both fungi and bacteria coincide with substantial enrichments in carbohydrate utilization enzymes suggests that catabolic processes help these microorganisms obtain the energy required to produce specialized metabolites at ecologically relevant scales. Furthermore, microorganisms and their specialized metabolites have been shown to participate in intricate relationships with terrestrial and marine eukaryotic hosts, such as defensive symbiosis^61,62. Taken together, our study supports a model in which several biosynthetic expansions have occurred over geological time, with an earlier expansion coinciding with early developments in multicellularity and a more recent expansion owing to emergent relationships with algae, invertebrates and plants. The subsequent expanded access to energy through the use of carbohydrate-active enzymes probably further fuelled developments in more complex multicellular structures, offset costs of carriage of a greater arsenal of specialized metabolites and promoted the transition from diffusion-feeding life histories to invasive growth strategies.

In addition to carbohydrate utilization machinery, we highlight the high prevalence of proteins suspected to function in HI in the two BGC-rich fungal clades of Pezizomycotina and Agaricomycotina. HI proteins allow filamentous dikaryons to self-recognize and initiate autophagy to prevent the formation of heterokaryons. The ability to self-distinguish is probably important for fungi when competing against other fungal species for access to potentially limited carbon sources within their natural habitats; however, HI could have other roles in fungal biology, including viral defence⁶³ and promoting haplontic life cycles.

Many specialized metabolites must be secreted to mediate interactions within an organism’s ecological niche and, as such, operate as ‘public goods’^64,65. The production of costly public goods makes cells vulnerable to exploitation by ‘cheater’ cells that benefit from the specialized metabolite without investing metabolic resources in their production, setting up ‘tragedy of the commons’ dynamics. Some unicellular bacteria use quorum-sensing systems to help resolve tragedy of the commons dynamics; these chemical communication networks enable synchronized group behaviours only when sufficient population density is reached, thereby preventing premature and inefficient production of costly public goods. In multicellular organisms, genetic relatedness among cells reduces the incentive to ‘cheat’. Kin selection conferred by multicellularity thus avoids the tragedy of the commons dynamics, which cause breakdowns in cooperative behaviour when individual cells pursue self-interest rather than invest in costly specialized metabolite production and secretion.

The evolution of multicellularity presents a fundamental paradox in evolutionary biology, as it requires individual cells to sacrifice their direct reproductive fitness to benefit the collective organism. Our findings suggest that a major benefit conferred by the emergence of multicellularity is the concomitant ability to evolve a high capacity to produce potent chemical mediators of intra- and interspecific interactions. These chemical compounds may serve yet unrecognized roles in mediating within-species dynamics, enforcing cooperative dynamics within multicellular entities, as hinted at by our finding of abundances of HI-associated proteins within BGC-rich and fruiting body-forming fungi. Indeed, such metabolites could even have played important roles in the origin of multicellularity itself. Ultimately, our findings advance understanding around the evolution of multicellularity and suggest that multicellular taxa, especially those that are understudied or from less commonly investigated habitats, might be prime sources for the discovery of truly novel classes of specialized metabolites.

Methods

Construction of a bacterial and archaeal timetree and determination of evolutionary expansions in biosynthetic traits

A bacterial and archaeal timetree was constructed for 74 representative bacterial and archaeal orders that have been well sequenced. To construct the phylogeny, a two-tiered approach was used. First, high-resolution phylogenies were constructed for each of the five bacterial phyla with the most number of characterized BGCs based on predetermined gene sets shown to be largely single copy and highly conserved for bacteria^66,67 using IQ-TREE (v2.2.0.3)⁶⁷. These were assessed to determine more reliable topologies for individual phyla than what could be achieved with a more limited number of single-copy core genes that are more broadly conserved across both bacteria and archaea. To construct the final phylogeny of the 74 representative bacterial and archaeal orders, a guiding tree constraining intra-phylum topologies and specifying a partition between Terrabacteria and Gracilicutes⁶⁸ was provided to IQ-TREE (v1.6.12) together with trimmed protein alignments of eight conserved ribosomal proteins⁶⁹, which featured a combined total of 1,095 amino acid sites and were conserved and non-ambiguous in ≥90% of genomes. IQ-TREE was run requesting an edge-unlinked partition model with ModelFinder limited to the Le and Gascuel (LG) and Whelan and Goldman (WAG) substitution matrices.

Next, to establish a timetree, we followed a previously described protocol with some adaptations to parameter configurations⁷⁰. Fossil and previous-inference-based calibration information for node ages were gathered from literature^71,72 (Supplementary Table 2). These calibrations were encoded as upper and lower bounds for select nodes in the phylogeny provided as input for MCMCtree (v4.10.0)⁷³. The software antiSMASH (v7.0.0)²⁰ was used for BGC annotation of genus-level representative genomes. Finally, to determine evolutionary shifts in the median BGC-ome size across the phylogeny, we used an Ornstein–Uhlenbeck process-based phylogenetic lasso approach via the l1ou⁷⁴ (v1.43) R library (Fig. 1c). A set of 4,732 genus-representative genomes across the 74 orders were used to assess their functional profile and biosynthetic potential (Supplementary Table 1).

For additional details on the construction of the phylum-specific phylogenies, including construction with and without filtering for compositional heterogeneity, construction of the overall bacteria–archaea phylogeny and details on parameter configurations for MCMCTree, please see the section with the same name in Supplementary Methods.

Genome selection and annotation for systematic investigations of genetic traits and ecological trends correlated with bacterial biosynthetic potential

Species-resolution representative genomes were selected for each of the five bacterial phyla with the greatest number of characterized BGCs. Only species belonging to genera with either ≥5 (for Myxococcota), ≥10 (for Actinomycetota and Cyanobacteriota) or ≥20 (Bacillota and Pseudomonadota) genomes in GTDB R214 were considered (Supplementary Table 3)⁷⁵. Different thresholds for different phyla were used to limit the number of genomes yet maintain an adequate sampling breadth for proper phylogenetic investigation of the focal phylum. For Bacillota, genomes belonging to related but distinct phyla in GTDB R214 (for example, Bacillota_A, which includes the genus Clostridioides) were also included. A total of 17,844 species-representative genomes were assessed.

Functional annotation of coding genes for BGCs was performed using antiSMASH. Large-scale annotations for known phylum-specific protein markers associated with multicellularity and sporulation^{29–31,76–79} were primarily performed using HMMER (v3.3.2)⁸⁰ with profile HMMs gathered or constructed following de novo orthologue group inference across a more limited set of representative genomes⁸¹. SandPiper (v0.2.0)⁸² was used to extract ecological context based on metagenomic profiling for various GTDB taxa. For additional information on functional analysis of bacterial species-representative genomes, see the section with the same name in Supplementary Methods.

In-depth phylogenomic and evolutionary investigations of the phylum Actinomycetota

Orthologue groups were identified across a set of 133 genus-level representative genomes from the phylum of Actinomycetota using OrthoFinder (v2.5.4)⁸¹. The 133 representative genomes were a subset of those used for phylum-specific phylogeny (Fig. 2a and Supplementary Tables 3 and 4) consisting of only assemblies that were designated ‘Complete’ or ‘Chromosome’ level for completeness status in the Genome database maintained by the National Center for Biotechnology Information (NCBI). Coarse orthologue groups, rather than refined hierarchical orthologue groups, were used to prevent possible inappropriate paralogy splitting by OrthoFinder’s algorithm when applied to bacterial genomes (Supplementary Table 6).

A total of 34 orthologue groups were identified as being single-copy core, mostly corresponding to ribosomal proteins (26; 76.5%). Protein alignments for the 34 orthologue groups were generated using MUSCLE (v5.1), filtered using trimAl (v1.4.1) in strict mode, and subsequently used as input for IQ-TREE for phylogenetic modelling of the phylum. IQ-TREE was run using an edge-unlinked partition analysis with ModelFinder limited to the LG and WAG substitution models and 1,000 bootstraps requested. The resulting phylogeny was largely topologically concordant with those based on a subset of the 74 universal bacterial genes by GToTree detection. The phylogeny was rooted by the genome for Rubrobacter_D, which was observed to be an early-diverging genus in previously constructed phylum-specific phylogenies that included genomes from other phyla as outgroups.

Detailed methods for analyses pertaining to investigations of Actinomycetes evolution can be found in Supplementary Methods. This includes sections describing the D.stat metric for assessing topological models based on highly conserved sites of core proteins, the annotation of multicellularity and sporulation-associated proteins, and comprehensive assessment of ansamycin BGCs and AHBA biosynthesis enzymes in 33,804 Actinomycetota genomes.

Phylogenomic investigations of biosynthetic potential across the fungal kingdom

A total of 312 genomes were selected from across the fungal kingdom featuring representatives of different dikaryon families, representatives of different classes from other fungal subkingdoms and phyla, and genomes included in a recent study⁴⁰, in which several non-dikaryotic lineages of fungi, including many diverse chytrids, were sequenced (Supplementary Table 7). Representative genomes for dikaryon families and non-dikaryotic classes were selected from species-level representative genomes that did not feature ‘fam_Incertae_sedis’ or ‘unidentified’ in their taxonomic designations in the Universal Fungal Core Genome (UFCG) database⁸³ based on assembly N50 measured using abyss-fac (v2.3.9). To properly assess biosynthetic potential, we required genomes to have a minimum N50 of 10 kb. Four non-fungal eukaryotic genomes were included in analyses as outgroups: GCF_000151315.2, Capsaspora owczarzaki; GCF_000001215.4, Drosophila melanogaster; GCA_000388065.2, Fonticula alba; and GCA_000002865.1, Monosiga brevicollis. Genomes were downloaded using ncbi-genome-download, and taxonomic names were systematically gathered from NCBI using taxonkit⁸⁴.

Coding sequence (CDS) coordinates for assemblies were gathered in GFF format from the RefSeq database of NCBI or, if not available, from the GenBank database of NCBI. If CDS coordinate information was not available in either database or there were difficulties with using them to run antiSMASH for BGC predictions downstream (for example, lack of CDS features or errors with overlapping CDS coordinates), then funannotate (v1.8.15; https://github.com/nextgenusfs/funannotate) was used to first mask repeat sequences, clean and sort assemblies before performing gene prediction⁸⁵.

The UFCG profile (v1.0.5)⁸³ was systematically applied to identify universal core protein sequences directly within genomic assemblies. Protein sequences for each of the 51 core universal gene markers found in ≥90% of genomes were extracted, aligned using MUSCLE and trimmed using trimAl in strict mode. Afterwards, these alignments were provided as input to IQ-TREE to construct a maximum-likelihood phylogeny using an edge-unlinked partition analysis with ModelFinder, allowing only for LG and WAG substitution models, and requesting 1,000 bootstraps. A monophyletic clade within Pezizomycotina enriched for BGCs was identified through visual examination and validated to include taxonomic subphyla described as BGC enriched in previous studies^45,46. This clade encompasses the taxonomic classes Sordariomycetes, Dothideomycetes, Eurotiomycetes, Leotiomycetes, Lecanoromycetes, Xylonomycetes and Arthoniomycetes. For comparative analyses of biosynthetic genes or gene clusters across taxonomic groups, only a single representative genome per genus was regarded, the one with the largest BGC-ome size. Most of the genera with multiple genomes were non-Dikarya (18 of 20), and this approach would allow us to more conservatively assess whether they are depleted in BGCs relative to the higher Fungi.

For additional details on methods pertaining to further typing and assessing the novelty of NRP synthetases or PK synthases^86,87, annotation of HI-associated domain architectures and CAZymes^88,89, and comparative genomics and GWAS to understand enrichments in genetic factors potentially contributing to biosynthetic enrichment in Agaricomycetes and Pezizomycotina, please see the corresponding section in Supplementary Methods.

Visualization and select statistical tests

Visualizations were primarily created using libraries in R and iTol⁹⁰. Limits of box plots shown alongside phylogenies correspond to the 5th, 25th, 50th, 75th and 95th percentiles, whereas stand-alone box plots are always shown in Tukey’s style, in which they extend to the largest value that is less than 1.5 times the interquartile range (IQR) from the top hinge of the box and the smallest value that is greater than 1.5 times the IQR from the bottom hinge of the box.

Spearman correlation coefficients and significance P values of genetic and ecological quantitative measures with BGC-ome size were obtained using the spearmanr function in scipy with a two-sided alternative model. One-sided rank sum tests were used to assess the statistical significance of differences in BGC-ome size between Actinomycetes with ssgB, Actinomycetes without ssgB and other Actinomycetota without ssgB.

BGC-ome sizes and the proportion of total proteins with HI-associated domain architectures were compared for genus-representative genomes between different taxonomic groupings using one-sided Wilcoxon rank-sum tests. Taxonomic partitions within early-diverging fungal lineages were largely defined following highlighted taxonomic groupings⁴⁰.

A phylogenetic linear regression model was fit to assess the relationship between fungal BGC-ome size with genome size, the count of total proteins and the count of predicted carbohydrate-processing proteins using the phylolm package in R. The BGC-ome and genome size were provided in Mb units. Pagel’s lambda model was used to model the covariance.

Additional details on other statistical models and tests can be found in Supplementary Methods.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

C.R.C. conceptualized the origin of the study. R.A.S. curated the data, performed formal analysis and developed genomic methodologies. R.A.S., L.R.K. and C.R.C. designed the study, wrote the original draft of the paper, developed visualizations, and reviewed and approved of the final paper.

Peer review

Peer review information

Nature Microbiology thanks Marnix Medema and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Peer reviewer reports are available.

Funding

This work was supported by grants from the National Institutes of Health (NIAID U19AI142720 (C.R.C. and L.R.K.) and NIGMS R35GM137828 (L.R.K.)). C.R.C. received partial support from the Stephen A. Jarislowsky Foundation and the Canadian Institute for Advanced Research. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Data availability

NCBI accession codes for genomic assemblies used for different analyses presented in this study are provided in Supplementary Tables 1, 3, 4, 6, 7 and 10. Larger datasets, including select files from antiSMASH BGC predictions, and fungal gene calling data are available via Zenodo at 10.5281/zenodo.15168076 (ref. ⁹¹). Source data are provided with this paper.

Code availability

Programs, scripts, commands and configuration files for the study are available via GitHub at https://github.com/Kalan-Lab/Ancestral_BGC_Exansions_Study, for which releases are version tracked and stored in Zenodo at 10.5281/zenodo.19632677 (ref. ⁹²). The codoff program, used to assess recent horizontal transfer of BGCs, is available via GitHub at https://github.com/Kalan-Lab/codoff. The paiof and hamtree programs, used to profile the amino acid identity distributions between paired genomes and construct neighbour-joining trees based on orthologue group carriage information following OrthoFinder analysis, are available via GitHub at https://github.com/raufs/paiof. The psaps program to assess pangenome expansion rate differences between fungal clades is available via GitHub at https://github.com/raufs/psaps. Releases of all GitHub repositories are also tracked in immutable forms on Zenodo.

Competing interests

The authors declare no competing interests.

Extended data

is available for this paper at 10.1038/s41564-026-02385-5.

Supplementary information

The online version contains supplementary material available at 10.1038/s41564-026-02385-5.