Genetic diversity and evolutionary adaptation of Clostridium paraputrificum, a rare pathogen in humans Authors: Jiao Liu, Yizhu Chen, Hangxiang Du, Lechen Yang, Zhuofei Xu, ZhenLiang Wen, Weifeng Shang, Xiaojun Pan, Lidi Zhang, Yunliang Cui, Dechang Chen, Sheng Zhang Published: 2025 ABSTRACT Clostridium paraputrificum infection in humans is extremely rare, and its pathogenesis and evolutionary mechanisms remain poorly understood. We sequenced the genome of strain CP_SH01 isolated from the blood culture of an ICU patient with acute abdominal injury and bloodstream infection. The genome harbored an intact prophage and ten insertion-sequence elements. Genome-wide metabolic reconstruction uncovered 42 fully complete modules linked to the utilization of different sugars and fermentation. Phylogenomic analysis of 42 global strains revealed two clades with substantial genetic diversity. A comparative genomic analysis was carried out to decode the pangenomic structure of C. paraputrificum, and functional differences between 2,434 core genes and 5,939 accessory genes. Meanwhile, 52 genes linked to 16 virulence factors and 15 antimicrobial resistance genes were identified. The majority of these clinically relevant genes, especially drug-resistant genes, were accessory and likely acquired via possible horizontal transfer. Additionally, genes under positive Darwinian selection were identified, potentially crucial for bacterial adaptation and immune interaction. This study presents the first blood-derived C. paraputrificum genome, revealing the global distribution of virulence and resistance genes. The accessory genome serves as a reservoir for clinically relevant traits, and positively selected genes provide insights into adaptive evolution and pathogenicity. Introduction Clostridium paraputrificum is a gram-positive bacterium of firmicute in the family Clostridiaceae. It is an obligate anaerobic, rod-shaped, spore-forming bacterium1. The Clostridiaceae family comprises common pathogens in humans and animals, for instance, C. perfringens, C. botulinum, C. chauvoei, and C. tetani2. Unlike these well-known Clostridium spp., C. paraputrificum is a rare opportunistic pathogen that can normally colonize the gastrointestinal tract in adults and infants, and in some cases cause a diverse range of infections. Species identification of C. paraputrificum isolated from human feces was first reported by Simunek et al.3. A large-scale microbiome project has also uncovered C. paraputrificum and many other oxygen-sensitive intestinal bacteria from healthy adult human donors through targeted phenotypic culturing and whole-genome sequencing4. Ciuro et al. have pointed out that C. paraputrificum may lead to septic arthritis in sickle cell disease as an emerging pathogen5. A recent report has presented a case of C. paraputrificum bacteremia in the setting of presumptive appendicitis with abscess, suggesting the potential of this gut species invading other body sites6. Although the links between C. paraputrificum and various diseases have been uncovered in the last decade, so far, few epidemiological surveys have been introduced for this pathogen. The virulence properties of pathogenic Clostridium species are generally multifactorial, including copious toxins, degradative enzymes, adhesins, biofilms, the capsule, and sporulation7. With the popularization of whole genome sequencing, comparative genomics provides systematic approaches to uncover the genetic repertoire of virulence factors. A recent genomic project of C. perfringens has assessed the distribution of virulence genes among phylogroups of 372 strains, revealing conserved virulence factors encoded by most genomes, e.g., alpha-toxin Plc, sialidase NanH, and collagenase ColA8. Toxin genes can be exchanged by mobile genetic elements between toxigenic species like C. botulinum and C. perfringens9. For C. paraputrificum, the cytotoxic effects have been experimentally verified on Vero cells, and the distribution of some virulence determinants has been further characterized across six isolate genomes, such as Cholera toxin secretion proteins, toxin A, transcriptional regulators, and phage-related holin10. Nevertheless, previous research on profiling genes relevant to virulence and antibiotic resistance in C. paraputrificum has been limited to a small number of genomes. Accumulated genomic sequences of bacterial populations have offered unprecedented opportunities for testing the Darwinian evolution of adaptive mutations via the dN/dS ratios in protein-coding genes: >1 denotes positive (Darwinian) selection and <1 for negative selection11. The genes that have experienced positive selection are usually vital for bacterial defense and immune evasion in pathogen-host co-evolutionary “arms race”12. Prospecting of this genetic evidence is a valuable approach in the era of population genomics, which enables researchers to uncover novel genes implicated in the viability and pathogenesis of many pathogenic microbes, especially for difficult-to-cultivated bacteria like Mycobacterium tuberculosis13 and Chlamydia psittaci14. Besides, the genes under negative selection are fair loci for the development of molecular typing methods, like multi-locus sequence typing (MLST)15. In the present study, we determined a nearly complete genome of C. paraputrificum from the blood culture of an inpatient with abdominal pain and bloodstream infection. Genome biology of this emerging pathogen was deciphered through the reconstruction of metabolic pathways, and analyses of phylogeny and MLST. To provide a genetic landscape of C. paraputrificum, a comparative genomic analysis was further performed using the newly sequenced genome and publicly available genomes. Based on a pangenome framework, we investigated the diversity of genetic determinants coding for virulence factors and antimicrobial resistance, as well as the genes showing evidence for positive selection. Results Assembly of C. paraputrificum CP_SH01 from blood culture The positive blood culture for C. paraputrificum was initially detected by MALDI-based mass spectrometry from an ICU-admitted patient with bloodstream infection. Isolated colonies were subject to whole-genome shotgun sequencing for further analysis. We found ~81.5% of total reads (40.13 million paired reads) aligned to reference genomes of C. paraputrificum. Iterative assembly by Megahit and Spades generated a draft genome consisting of 18 contigs more than 500 bp (Supplementary Data 1). Genomic N50 and completeness of the resulting assembly were 588 kb and 99%, respectively, indicating it was a qualified genome assembly. According to genome-wide nucleotide identity analysis, CP_SH01 shared high ANI values (>99%) with both C. paraputrificum strains NCTC11833 (GCF_900447045) and AGR2156 (GCF_000424025) (Fig. 1A). The lower ANI values between CP_SH01 and the other Clostridium species were observed, ranging from 68% to 77%. It again confirmed that the strain CP_SH01 isolated in this study was derived from C. paraputrificum.Fig. 1Genome characteristics of C. paraputrificum CP_SH01.A Heatmap of pairwise ANI values between paired genomes from the species within the genus Clostridium. B Circular representation of the CP_SH01 genome. The rings from innermost to outermost denote GC content, genes of forward and reverse strands, mobile orthologous genes, prophage and insertion sequences. The donut chart displays the frequencies of five classes of mobile orthologous genes. C Whole-genome phylogenetic tree based on 388 conserved gene markers present in representative genomes of 88 Clostridium species. D Genetic organization of the prophage region in CP_SH01. The genic compositions are color-coded by different classes. Genome characteristics of C. paraputrificum CP_SH01. A Heatmap of pairwise ANI values between paired genomes from the species within the genus Clostridium. B Circular representation of the CP_SH01 genome. The rings from innermost to outermost denote GC content, genes of forward and reverse strands, mobile orthologous genes, prophage and insertion sequences. The donut chart displays the frequencies of five classes of mobile orthologous genes. C Whole-genome phylogenetic tree based on 388 conserved gene markers present in representative genomes of 88 Clostridium species. D Genetic organization of the prophage region in CP_SH01. The genic compositions are color-coded by different classes. Genome biology of C. paraputrificum The draft genome of C. paraputrificum CP_SH01 consisted of 3,504,994 bp in length, which encoded 3334 protein-coding genes, 7 rRNA, and 73 tRNA genes (Table 1 and Fig. 1B). The GC content of the CP_SH01 genome was 29.8%, which was identical to the mean GC content of the other C. paraputrificum genomes (Supplementary Data 2). To get a glimpse of the phylogenetic placement of CP_SH01, Fig. 1C displays the evolutionary relationships of 88 species representative genomes affiliated to the genus Clostridium proposed by NCBI Taxonomy. Additional taxonomic classification using GTDB-Tk determined that 68 out of all the genomes (77%) were Clostridium species within the family Clostridiaceae, while the remaining ones were derived from other families within the class Clostridia (Supplementary Data 2). C. paraputrificum CP_SH01 was adjacent to a clade containing 11 species with similar GC composition, e.g., C. butyricum (28.7%), C. neonatale (28.6%), C. beijerinckii (29.9%), and C. diolis (29.9%). Interestingly, the genome sizes of these organisms varied greatly, ranging from ~3.5 Mbp for C. taeniosporum to 6.7 Mbp for C. saccharoperbutylacetonicum. It indicated that speciation would be restricted by clade-specific nucleotide compositions, but horizontal gene transfer may facilitate significant variations in genome sizes of closely related species. As shown in Fig. 1B, 83 genes mediating essential functions of mobile genetic elements assigned to five categories were identified in C. paraputrificum CP_SH01. Among these mobile genes, eight code for products involved in integration and excision from one genetic locus to another, and 21 genes for phage-specific biological processes. Besides, 10 copies of five IS elements were found, including IS21, IS110, IS3, IS1182, and IS1595. It was worth noting that a ~48-kb genomic region flanked by two pairs (12-bp and 25-bp direct repeats) of attachment sites attL/R was predicted to be a complete prophage. The CP_SH01 prophage encodes 40 phage-related proteins, 22 hypothetical proteins, and nine bacterial proteins (Fig. 1D). This prophage was the most common to a tailed dsDNA phage ct2h02 (GenBank accession BK032370) from the class Caudoviricetes, sharing 99% identity over ~28 kb alignment sequences. Two head proteins (304 aa and 517 aa) of the CP_SH01 prophage shared 97% and 83% identity with the capsid proteins (304 aa and 517 aa) of the phage ct2h02, which has been discovered in human microbiomes associated with chronic diseases16. The gene product PdaA, adjacent to an IS3 family transposase, was homologous to the Listeria monocytogenes endo-1,4-beta-xylanase PgdA, both of which harbored a domain (PF01522) of polysaccharide deacetylase catalyzing peptidoglycan N-deacetylation for immune evasion17. The abundant mobilizable elements mentioned above were likely drivers to facilitate gene loss and/or gain for bacterial diversity and the pathogenic capacity of C. paraputrificum.Table 1Summary of comparison for selected species representative genomes in ClostridiumAccessionGWHEUUW00000000GCF_000007625GCF_000063585GCF_001244495GCF_002896855GCF_018223745GCF_020138775SpeciesC. paraputrificumC. tetaniC. botulinumC. phoceensisC. thermosuccinogenesC. beijerinckiiC. perfringensStrainCP_SH01E88ATCC 3502GD3DSM 5807DSM 791CPI 18-6Size (bp)3,504,9942,873,3333,903,2603,446,5474,731,2165,950,2473,275,424GC (%)29.7528.6428.2459.3241.5429.8728.44N50587,7232,799,2513,886,916192,1614,731,2165,876,9023,260,088L503116111Completeness99.1998.2899.3185.9697.9999.19100Contamination00.14003.710.810Gene3414286736693393408053502980CDS3334279535633338400852062855rRNA718275124930tRNA73547950599494 Table 1Summary of comparison for selected species representative genomes in ClostridiumAccessionGWHEUUW00000000GCF_000007625GCF_000063585GCF_001244495GCF_002896855GCF_018223745GCF_020138775SpeciesC. paraputrificumC. tetaniC. botulinumC. phoceensisC. thermosuccinogenesC. beijerinckiiC. perfringensStrainCP_SH01E88ATCC 3502GD3DSM 5807DSM 791CPI 18-6Size (bp)3,504,9942,873,3333,903,2603,446,5474,731,2165,950,2473,275,424GC (%)29.7528.6428.2459.3241.5429.8728.44N50587,7232,799,2513,886,916192,1614,731,2165,876,9023,260,088L503116111Completeness99.1998.2899.3185.9697.9999.19100Contamination00.14003.710.810Gene3414286736693393408053502980CDS3334279535633338400852062855rRNA718275124930tRNA73547950599494 Summary of comparison for selected species representative genomes in Clostridium Since C. paraputrificum is a strictly anaerobic bacterium lacking genetic manipulation systems, we carried out a genome-wide metabolic reconstruction to characterize physiological functions. Herein, 1711 CDSs of C. paraputrificum CP_SH01 were assigned with 1,307 KOs, which were distributed in 218 KEGG pathways and 153 modules (Supplementary Data 3). Among the modular functional units, 42 were found to be complete, carrying all biochemical blocks required for the relevant metabolic reactions (Supplementary Data 3). For instance, five intact modules consisting of genes involved in central carbohydrate metabolism were identified in the CP_SH01 genome, including glycolysis (M00001/M00002), pyruvate oxidation (M00307), non-oxidative pentose phosphate pathway (M00007), and PRPP biosynthesis (M00005). In contrast, the modules mostly used by aerobic oxidation of glucose derivatives for energy production were incomplete in CP_SH01, including the tricarboxylic acid (TCA) pathway (M00009-11), oxidative pentose phosphate pathway (M0004), and Entner-Doudoroff pathway (M0008). Meanwhile, all genes of the glyoxylate cycle were absent in the CP_SH01 genome, implying C. paraputrificum would fail to grow if acetate was used as a sole carbon source. Additionally, The CP_SH01 genome comprised five genes constituting a complete module (M00793) responsible for the biosynthesis of the nucleotide sugar precursor dTDP-L-rhamnose, which is necessary to produce rhamnose polysaccharides in the cell wall of many gram-positive pathogens18. For the uptake of essential nutrients into the cell, we identified many complete sets of genes coding for ATP-binding cassette (ABC) transporters, such as methyl-galactoside (mglABC), phosphate (pstABCS), zinc (znuABC), spermidine (potABCD), and lipoprotein (lolCDE). C. paraputrificum CP_SH01 also possessed multiple phosphotransferase systems for the utilization of carbon sources, including glucose (ptsG-crr), N-acetyl-D-glucosamine (nagE), mannose (manXYZ and manP), cellobiose (celABC), and D-glucosaminate (dgaABCD). Ten genes encoding chitinases (K01183, EC 3.2.1.14) were also identified in the CP_SH01 genome, which can hydrolyze chitin into oligomers of N-acetyl-D-glucosamine. The presence and absence of these metabolic units were consistent with the physiological features that C. paraputrificum grows anaerobically but not aerobically in media with chitin as the main carbon source3. Genetic relationship of CP_SH01 and other C. paraputrificum strains To study the genome epidemiology of the newly sequenced strain CP_SH01, we reconstructed a maximum likelihood phylogenetic tree based on 126,334 core SNPs across 42 C. paraputrificum genomes. As displayed in Fig. 2A, the phylogenetic topology of C. paraputrificum was divided into two evolutionary clades. CP_SH01 was placed in the Clade I comprising 23 strains, and it was neighboring to both strains NCTC11833 (99.37% ANI; 11,669 SNPs) and 28-X28 (99.40% ANI; 8,995 SNPs). The former was a type strain of C. paraputrificum without any isolation information, and the latter was isolated from mice in 2016. Among all the strains with known isolation sources, CP_SH01 was the sole strain isolated from blood, while the others were isolated from fecal samples. The distribution of pairwise SNP distances also showed that the inter-group distances (median distance of 88,852 SNPs) between Clade I (median: 10,065 SNPs) and Clade II (median: 15,240 SNPs) were higher than the intra-group distances (median: 11,779 SNPs) (Supplementary Data 4). For instance, CP_SH01 possessed 89,950 SNPs and 80,598 SNPs with the Clade II strains Gcol.A11 isolated in Colombia and LH058 in the United Kingdom, respectively. Moreover, CP_SH01 exhibited relatively low ANI values (95.37% and 95.88%) with the two strains, hinting at considerable genetic divergence among the C. paraputrificum populations worldwide.Fig. 2Phylogeny and molecular typing of CP_SH01 and other publicly available C. paraputrificum strains.A The maximum likelihood phylogenetic tree reconstructed based on core SNPs present in all the genomes. The metadata adjacent to the tree includes host, isolation year, country of origin, and strain identifiers of individual genomes. The bottom panel illustrates minimum spanning trees of all C. paraputrificum strains based on an allelic profile of seven housekeeping genes. The nodes are colored by different STs B and countries C. The node size is proportional to the number of strains. Phylogeny and molecular typing of CP_SH01 and other publicly available C. paraputrificum strains. A The maximum likelihood phylogenetic tree reconstructed based on core SNPs present in all the genomes. The metadata adjacent to the tree includes host, isolation year, country of origin, and strain identifiers of individual genomes. The bottom panel illustrates minimum spanning trees of all C. paraputrificum strains based on an allelic profile of seven housekeeping genes. The nodes are colored by different STs B and countries C. The node size is proportional to the number of strains. Besides, we developed an initial MLST scheme for molecular typing of C. paraputrificum. In this scheme, seven housekeeping genes (aroE, galU, lysA, pfp, rlmN, trpS, and ycdT) with known functions and subject to negative selection were chosen from the set of core genes detected by the pangenomic analysis. The genes encode products involved in amino acid metabolism (aroE and lysA), translation and ribosomal biogenesis (rlmN and trpS), carbohydrate metabolism (pfp), signal transduction mechanism (ycdT), and nucleotide sugar biosynthesis (galU). These genes were characterized by gene length, GC content, pairwise identity between the entries in the alignment, evolutionary pressures (dN/dS) acting on the entire coding region, and number of alleles (Table 2). The average dN/dS ratios for the genes were estimated to be less than 0.1, indicating that their evolution was largely driven by negative selection. Based on the MLST scheme, 28 STs were predicted among all the C. paraputrificum strains (Fig. 2B). The strain CP_ SH01 was assigned to ST28 and closely linked to the strain 1001285H_161024_H4 belonging to the ST21, both of which showed the differences in the two alleles, aroE and ycdT (Supplementary Data 5). The majority (21 out of all) of STs comprised single isolates, which further suggested the high genetic diversity within the C. paraputrificum species.Table 2Genetic properties of seven housekeeping genes selected for the MLST scheme of C. paraputrificumSymbolIDFunction of coding productsLength (bp)GC (%)Identity (%)dN/dS ratio (ω)No. of allelesaroEOG0636Shikimate dehydrogenase80431.296.400.097610ycdTOG0932Diguanylate cyclase107727.896.790.056417lysAOG0941Diaminopimelate decarboxylase129331.597.550.058513galUOG1920UTP-glucose-1-phosphate uridylyltransferase86432.897.040.017213pfpOG20616-phosphofructokinase121531.498.590.049815rlmNOG2130Ribosomal RNA large subunit methyltransferase N104129.898.520.079711trpSOG2312Tryptophan--tRNA ligase100231.597.430.042416 Table 2Genetic properties of seven housekeeping genes selected for the MLST scheme of C. paraputrificumSymbolIDFunction of coding productsLength (bp)GC (%)Identity (%)dN/dS ratio (ω)No. of allelesaroEOG0636Shikimate dehydrogenase80431.296.400.097610ycdTOG0932Diguanylate cyclase107727.896.790.056417lysAOG0941Diaminopimelate decarboxylase129331.597.550.058513galUOG1920UTP-glucose-1-phosphate uridylyltransferase86432.897.040.017213pfpOG20616-phosphofructokinase121531.498.590.049815rlmNOG2130Ribosomal RNA large subunit methyltransferase N104129.898.520.079711trpSOG2312Tryptophan--tRNA ligase100231.597.430.042416 Genetic properties of seven housekeeping genes selected for the MLST scheme of C. paraputrificum Characteristics of C. paraputrificum pangenome To investigate bacterial pangenomic structure, a clustering analysis of 145,989 coding sequences in 42 C. paraputrificum genomes was carried out. In total, 8,373 clusters of OGs were detected and divided into three groups: 2,434 core genes present in all strains, 3,786 accessory genes present in at least two strains but not all strains, and 2,153 strain-specific genes. Of these, 92% were single-copy genes. Fig. 3A shows the frequency of genes present in the k (1 to 42) genomes. Based on bacterial pangenome profile analysis, the pangenome of C. paraputrificum was open and would expand with the addition of newly sequenced genomes (Fig. 3B).Fig. 3Pangenome structure and function of C. paraputrificum.A Distribution of gene clusters detected in the pangenome. B Fitted curves showing estimated numbers of genes in the pangenome and core genome. C COG functional classification of highly conserved genes. The minimum alignment identity for the detection of these genes is 99.5%. D Histogram showing proportions of core genes, dispensable genes, and strain-specific genes in individual COG categories. Pangenome structure and function of C. paraputrificum. A Distribution of gene clusters detected in the pangenome. B Fitted curves showing estimated numbers of genes in the pangenome and core genome. C COG functional classification of highly conserved genes. The minimum alignment identity for the detection of these genes is 99.5%. D Histogram showing proportions of core genes, dispensable genes, and strain-specific genes in individual COG categories. On the other hand, the functional distribution of core genes, accessory genes, and strain-specific genes was analyzed according to COG functional classification. The proportion of genes per functional category is displayed in Fig. 3D (see details in Supplementary Data 6). It was clear that the functional profile of core genes was distinctive, with 14 categories exhibiting significantly differential abundance in comparison to those of accessory and strain-specific genes (FDR < 0.01). Of these, 12 functional categories were enriched in the set of core genes, e.g., “Translation and ribosomal structure”, “Signal transduction mechanisms”, “Posttranslational modification, protein turnover, chaperones”, “Energy production and conversion”, “Amino acid transport and metabolism”, “Nucleotide transport and metabolism”, “Coenzyme transport and metabolism”, “Lipid transport and metabolism”, and “Inorganic ion transport and metabolism”. Conversely, genes assigned to the category “Mobilome: prophages, transposons”, which generally encode proteins associated with mobile genetic elements, were abundant in the sets of accessory and strain-specific genes. Bacterial essential genes are supposed to be more evolutionarily conserved due to rigorous functional constraints under strong purifying selection19. We found 37 highly conserved genes with alignment identities greater than 99.5% in the C. paraputrificum genomes (Fig. 3C). It was apparent that 25 genes participating in “Translation and ribosomal structure” were significantly abundant. Of these, 12 genes encoded fully identical sequences in all genomes tested, including 50S ribosomal proteins L14/L28/L29/L30/L31/L32/L35, and 30S ribosomal proteins S10/S17/S19/S21. Gene patterns of virulence factors and antimicrobial resistance In the genic repertoire of the C. paraputrificum pangenome, 52 virulence-associated genes (VAGs) were detected to play potential roles in the biosynthesis and transport of 16 virulence factors (Supplementary Data 7). The majority of the VAGs code for the products involved in the assembly of capsule (30 genes) and hyaluronic acid capsule (5 genes). Herein, 16 VAGs were core genes present in all strains tested, some of which may encode essential proteins for bacterial pathogenicity. For instance, the products of four VAGs associated with adherence were identified in all strains, including chaperonin (OG1842, 539 aa), ABC transporter substrate-binding lipoprotein (OG0861, 333 aa), fibronectin-binding protein (OG0771, 580 aa), and bifunctional acetaldehyde-CoA/alcohol dehydrogenase (OG0963, 875 aa). The former three genes (OG1842, OG0861, and OG0771) shared 74%, 47%, and 43% amino acid sequence identity with Clostridioides difficile GroEL (WP_003435012, 542 aa), CD0873 (WP_011860976, 340 aa), FbpA (WP_011861612, 591 aa), respectively. Both CD0873 and FbpA are surface-exposed proteins contributing to adherence and colonization of C. difficile20,21. Besides, C. paraputrificum may harbor a functional Clp protease complex as the genes encoding ATP-consuming hexameric Clp-ATPases (OG1116, 814 aa; OG1394, 867 aa) and the barrel-shaped proteolytic core ClpP (OG1553, 199 aa) were detected in all strains. The Clp proteins of C. paraputrificum were homologous to the L. monocytogenes ClpC ATPase (820 aa) and ClpP (198 aa) required for in vivo survival and the expression of virulence factors22. Besides, more than half of all VAGs were accessory genes (26) and strain-specific genes (10), which may be responsible for the production of diverse virulence factors specific to individual strains (Fig. 4).Fig. 4Distribution of clinically relevant genes in the genomes of C. paraputrificum.The heatmap shows the presence and absence patterns of accessory virulence-associated genes and AMR genes. The upper dendrogram is the same phylogeny shown in Fig. 2. Functional classification, protein subcellular localization, and COG category for each gene are illustrated in the right panel. Distribution of clinically relevant genes in the genomes of C. paraputrificum. The heatmap shows the presence and absence patterns of accessory virulence-associated genes and AMR genes. The upper dendrogram is the same phylogeny shown in Fig. 2. Functional classification, protein subcellular localization, and COG category for each gene are illustrated in the right panel. On the other hand, 15 genes encoding determinants of antimicrobial resistance were detected (Supplementary Data 8) and the pattern of their sequence conservation across all strains is illustrated in Fig. 4. As expected, all acquired AMR genes were accessory or strain-specific genes, which were assigned to three COG categories “Translation and ribosomal structure” (7 genes), “Carbohydrate transport and metabolism” (2), and “General function prediction only” (4). The detection rate of genes conferring resistance to aminoglycosides was the highest, and the next was tetracycline resistance genes. Two resistance genes were found in the genome of CP_SH01, which encode a ribosomal-protection protein TetB(P) (652 aa) and a transmembrane efflux protein TetA(P) (420 aa), conferring resistance to tetracyclines. Both proteins shared 98% and 100% identity with the C. perfringens TetB(P) (WP_317320786) and TetA(P) (WP_006440994) that are linked in an operon for tetracycline resistance determinants23. Evidence for positive Darwinian selection on protein-coding genome of C. paraputrificum To mine functionally important genes from the perspective of pathogen-host co-evolution, we conducted a positive selection analysis for 2783 single-copy genes of C. paraputrificum. Based on the likelihood values of nested codon models M1a/M2, a total of 103 genes were found to be under positive selection with p values < 0.05 (Supplementary Data 9). Of these, more significant selection pressures were likely acting on 29 genes with corrected q values < 0.2 (Table 3). Nearly half of these positively selected genes encode proteins associated with cell wall and membrane biogenesis, for instance, nedA, tee6, ygfU, outO, ppkA, msrR, cotH, and ftsW. Besides, 17 genes subject to positive selection encode putative secretory proteins carrying N-terminal signal peptides. Topological and structure models of three surface proteins and their amino acid sites under strong positive selection are illustrated in Fig. 5. Both OutO (OG2316, 253 aa) and YgfU (OG2242, 455 aa) were located on the cytoplasmic membrane with 8 and 14 transmembrane helices (TMHs). The four positively selected sites (residues 2, 3, 4, 8) of OutO were located in the N-terminal extracellular loop; whereas the selected site (residue 408) of YfgU was located in the C-terminal extracellular loop. The other protein encoded by gene tee6 (OG3309, 511 aa), carrying an N-terminal signal peptide and a GramPos_pilinBB domain (PF16569), seemed to be a major subunit forming gram-positive pili. All three sites subject to positive selection were exposed on the cell surface.Fig. 5Structural models of surface proteins subject to positive selection.The topology of transmembrane segments in the protein sequence is color-coded with the predictions of DeepTMHMM: red for alpha helical membrane, green for outside cell, cyan for inside cell, and yellow for signal peptide. Protein models and residues under significant positive selection (posterior probability > 95%) are shown in the bottom panel.Table 3Genes undergoing positive Darwinian selection in C. paraputrificumIDSymbolProtein functionCOGLocalizationa2 Δℓωbq valuePositively selected sitesOG0010nagJGlycosyl hydrolaseGEC25.292.440.0008225, 774, 939, 1181, 1215, 1523OG0042ftsWCell division proteinDCM12.984.140.151946, 211OG0167glxKGlycerate kinaseGCP31.7617.276.00E-05221OG0202bmrUDiacylglycerol kinaseIRNA18.8254.330.0109288OG0322msrRCell envelope-related transcriptional attenuatorMCM24.943.040.0008376, 381, 391, 400, 406, 407, 410, 411, 412, 422, 425, 427, 432OG1174-Uncharacterised protein-CP15.64120.540.0438OG1325rumA23S rRNA (uracil-5-)-methyltransferaseJCP52.5031.501.13E-08448, 453, 455, 456OG1344-HAD-superfamily hydrolaseHRCP31.3028.956.16E-05159, 175OG1708yheSABC transporterRCP26.5115.800.0005515, 518OG1877hptHypoxanthine phosphoribosyltransferaseFCP18.75220.500.010928OG2242ygfUXanthine/uracil permease family proteinFCM37.1161.236.92E-06408OG2316outOType IV leader peptidase/N-methyltransferaseNUCM36.449.286.92E-062, 3, 4, 8OG2526cotHInner spore coat protein HMCM12.933.690.1519142OG2616pduLPropanediol utilization protein PduL-like proteinQCP17.5812.720.018728OG2697nedASialidase-like proteinGCW18.998.160.0106226, 261, 269, 273, 286, 289, 1362, 1495OG2714recQHRDC domain-containing proteinLCP14.573.790.0719118OG2720-Uncharacterised proteinRCP19.569.180.008499, 100OG2772-Uncharacterised protein-CP30.3814.247.17E-05197, 198, 199OG2809-Uncharacterised protein-CM30.9227.326.16E-05267, 269, 270, 272OG2840-Mutator mutT-like protein-CP19.8163.260.0083123, 127, 128OG2845-Uncharacterised protein-CM25.01999.000.000862, 64, 65, 66, 67, 68OG2963-Uncharacterised protein-NA37.38999.006.92E-0679, 80, 81, 85, 86, 87, 92, 93, 96, 97OG3036-4Fe-4S ferredoxinRNA21.72999.000.0034125, 127, 128, 130OG3158-Hypothetical protein-CM16.333.200.032537, 136, 151OG3309tee6Surface proteinSCW22.478.260.0025281, 283, 370OG3323-Hypothetical protein-NA19.659.290.008429OG3395ynaDN-acetyltransferase GCN5JONA16.3197.250.0325158OG3561-Hypothetical proteinUCM30.894.176.16E-05672, 678, 689, 701, 703, 711, 712, 736, 738, 742OG3607ppkAProtein kinaseMCM36.7622.166.92E-06583, 592, 606, 609, 611, 617, 621, 632, 635, 641, 646aThe abbreviations of protein subcellular localization: CP cytoplasmic, CM cytoplasmic membrane, CW cell wall, EC extracellular, NA unknown.bThe dN/dS ratio for the sites under positive selection. Structural models of surface proteins subject to positive selection. The topology of transmembrane segments in the protein sequence is color-coded with the predictions of DeepTMHMM: red for alpha helical membrane, green for outside cell, cyan for inside cell, and yellow for signal peptide. Protein models and residues under significant positive selection (posterior probability > 95%) are shown in the bottom panel. Table 3Genes undergoing positive Darwinian selection in C. paraputrificumIDSymbolProtein functionCOGLocalizationa2 Δℓωbq valuePositively selected sitesOG0010nagJGlycosyl hydrolaseGEC25.292.440.0008225, 774, 939, 1181, 1215, 1523OG0042ftsWCell division proteinDCM12.984.140.151946, 211OG0167glxKGlycerate kinaseGCP31.7617.276.00E-05221OG0202bmrUDiacylglycerol kinaseIRNA18.8254.330.0109288OG0322msrRCell envelope-related transcriptional attenuatorMCM24.943.040.0008376, 381, 391, 400, 406, 407, 410, 411, 412, 422, 425, 427, 432OG1174-Uncharacterised protein-CP15.64120.540.0438OG1325rumA23S rRNA (uracil-5-)-methyltransferaseJCP52.5031.501.13E-08448, 453, 455, 456OG1344-HAD-superfamily hydrolaseHRCP31.3028.956.16E-05159, 175OG1708yheSABC transporterRCP26.5115.800.0005515, 518OG1877hptHypoxanthine phosphoribosyltransferaseFCP18.75220.500.010928OG2242ygfUXanthine/uracil permease family proteinFCM37.1161.236.92E-06408OG2316outOType IV leader peptidase/N-methyltransferaseNUCM36.449.286.92E-062, 3, 4, 8OG2526cotHInner spore coat protein HMCM12.933.690.1519142OG2616pduLPropanediol utilization protein PduL-like proteinQCP17.5812.720.018728OG2697nedASialidase-like proteinGCW18.998.160.0106226, 261, 269, 273, 286, 289, 1362, 1495OG2714recQHRDC domain-containing proteinLCP14.573.790.0719118OG2720-Uncharacterised proteinRCP19.569.180.008499, 100OG2772-Uncharacterised protein-CP30.3814.247.17E-05197, 198, 199OG2809-Uncharacterised protein-CM30.9227.326.16E-05267, 269, 270, 272OG2840-Mutator mutT-like protein-CP19.8163.260.0083123, 127, 128OG2845-Uncharacterised protein-CM25.01999.000.000862, 64, 65, 66, 67, 68OG2963-Uncharacterised protein-NA37.38999.006.92E-0679, 80, 81, 85, 86, 87, 92, 93, 96, 97OG3036-4Fe-4S ferredoxinRNA21.72999.000.0034125, 127, 128, 130OG3158-Hypothetical protein-CM16.333.200.032537, 136, 151OG3309tee6Surface proteinSCW22.478.260.0025281, 283, 370OG3323-Hypothetical protein-NA19.659.290.008429OG3395ynaDN-acetyltransferase GCN5JONA16.3197.250.0325158OG3561-Hypothetical proteinUCM30.894.176.16E-05672, 678, 689, 701, 703, 711, 712, 736, 738, 742OG3607ppkAProtein kinaseMCM36.7622.166.92E-06583, 592, 606, 609, 611, 617, 621, 632, 635, 641, 646aThe abbreviations of protein subcellular localization: CP cytoplasmic, CM cytoplasmic membrane, CW cell wall, EC extracellular, NA unknown.bThe dN/dS ratio for the sites under positive selection. Genes undergoing positive Darwinian selection in C. paraputrificum aThe abbreviations of protein subcellular localization: CP cytoplasmic, CM cytoplasmic membrane, CW cell wall, EC extracellular, NA unknown. bThe dN/dS ratio for the sites under positive selection. Discussion Diseases and clinical impact caused by the emerging pathogen C. paraputrificum have been recently described by medical communities5,6. Unlike the ESKAPE pathogens (Enterococcus faecium, Staphylococcus aureus, Klebsiella pneumoniae, Acinetobacter baumannii, Pseudomonas aeruginosa, and Enterobacter sp.) as critical multi-drug resistance bacteria24,25, molecular epidemiology and population genomics of C. paraputrificum have yet to be thoroughly investigated, perhaps due to challenging to culture in the clinical setting for this strictly anaerobic bacterium. In this report, we characterized genome biology of bacterial phylogeny and physiological functions for a C. paraputrificum strain isolated from blood culture. Together with dozens of publicly available genomes of C. paraputrificum, we further revealed pangenomic structure, genetic diversity of virulence and antimicrobial resistance genes, as well as the evidence for natural selection acting on the protein-coding genome for this uncommon etiological agent. The proportion of core genes in the genomes of C. paraputrificum was approximately 70%, which was comparable to that (~78%) in another Gram-positive bacterium S. aureus but was significantly higher than that (~44%) in the Gram-negative bacterium Escherichia coli26,27. In comparison to an early study on 11 genes encoding virulence factors in C. paraputrificum10, more genes related to bacterial pathogenesis were deciphered herein, some of which were discussed below. Diversity of clinically relevant genes in C. paraputrificum Identification of genetic determinants of virulence factors and antibiotic resistance generally provides meaningful information for phenotypic inference and treatment of infectious diseases. Horizontal gene transfer can facilitate the spread of pathogenic and drug-resistant genes whose occurrences at the strain level are often random or group-specific. Herein, many VAGs of C. paraputrificum were clade-specific or serotype-specific. For instance, two genes coding for pantothenate synthetase PanC (OG2898, 281 aa) and aspartate-1-decarboxylase PanD (OG2899, 129 aa) were identified in the strains of Clade I only. Both enzymes of C. paraputrificum exhibit 55% and 76% identity with the corresponding homologs in M. tuberculosis, which can synthesize pantothenate, an essential and rate-limiting nutrient for the survival and growth of numerous bacteria, fungi, and protozoa. Analogs of the natural compound pantetheine could inhibit bacterial coenzyme A biosynthetic pathway and have been developed as novel antibiotics for gram-positive bacteria28. In addition, encapsulated gram-positive bacteria often have structurally different polysaccharide molecules, which are synthesized and assembled by highly diverse capsule gene clusters29. It was worth noting that, among all virulence factors detected in C. paraputrificum, the number of VAGs associated with capsule synthesis was highest, including four core genes and 31 accessory genes (Fig. 4). Two genes (OG3999, 400 aa; OG3881, 389 aa) encoding enzymes that catalyze the conversion of UDP-4-keto-6-deoxy-L-AltNAc to UDP-L-FucNAc were shared by CP_SH01 and the other seven strains within Clade I. Both enzymes shared 71% and 81% identity with Cps4K (WP_000594468, 409 aa) and Cps4L (WP_000695335, 394 aa) of Streptococcus pneumonia, in which 98 different capsular types have been reported30. Similar to the above two genes, another gene (OG3931, 456 aa) coding for polyprenyl glycosylphosphotransferase involved in exopolysaccharide production was also detected in the eight C. paraputrificum strains of Clade I. Furthermore, some strain-specific genes affiliated to the biosynthetic loci of capsules were identified in individual strains, such as five genes (OG7861/OG7864/OG7865/OG7866/OG7875) specific to the strain LH025, OG6722 to the strain D33t1_170424_A1, and OG8349 to the strain MGYG-HGUT-00151. Such abundant capsular genotypes in C. paraputrificum may produce different serotypes for adaptation to various host and environmental niches. Besides, none of the AMR genes was detected in all C. paraputrificum strains, and the gene with the most occurrences was tet(M), present in 11 strains. AMR genes were absent in 23 strains. The pathogenic strain 107 (GCF_036689015) isolated from a patient with pseudomembranous colitis in Mexico was found to carry the most AMR genes, including aad9, aac(6’)-Ie/aph(2”), aph(3’)-IIIa, ant(6)-Ia, lnu(C), and tet(M). Two genes erm(B) and erm(Q) encoding 23S rRNA methylases, which confer combined resistance to macrolide, lincosamide and streptogramin B (MLS), were identified in five and four C. paraputrificum strains, respectively (Fig. 4). Erm(B) is the most common emr gene that has been horizontally transferred across 36 genera via MGEs31. It suggested that horizontal gene transfer might play a vital role in the transmission of drug-resistant genes across C. paraputrificum strains. However, more investigations are still needed to study the acquisition mechanism of antimicrobial resistance in C. paraputrificum. Genome-scale evidence for adaptive changes in function Surface molecules on gram-positive cell walls can exert multifunctional roles in adherence, invasion of host cells, and persistence in infected tissues32. To overcome host immune pressures and antibiotics, base mutations usually continually occur on surface proteins, which is beneficial for bacterial viability and pathogenesis. Herein, a dozen genes predicted to localize on cell surface/membrane were found to be actively evolving in C. paraputrificum (Supplementary Data 9). For instance, the gene ppkA (OG3607) encoding a protein kinase (655 aa) involved in peptidoglycan metabolism was subject to significantly positive selection (q value < 0.001). The kinase possessed a Pkinase domain (PF00069) at the N-terminus and multiple PASTA domains (PF03793) at the C-terminus. All 11 positively selected sites were found to be adjacent to PASTA domains that can bind β-lactam antibiotics and have been characterized in penicillin-binding proteins and serine/threonine kinases of Gram-positive Firmicutes and Actinobacteria33. Another gene ftsW (OG0042, 373 aa) codes for a cell division protein encompassing ten TMHs, which has an identical topology with E. coli FtsW, an essential protein involved in peptidoglycan assembly during cell elongation, division, and sporulation34. The two positively selected sites (residues 46 and 211) were just located in the first and fourth extracellular loops outside the membrane. Similar evidence for diversifying selection on FtsW has been proposed by a previous study on the evolutionary biology of Pasteurella multocida, a Gram-negative pathogen35. In addition, positive selection pressures also acted on three C. paraputrificum genes (tee6, surface protein; nedA, sialidase-like protein; OG0223, phosphohydrolase) located on the cell wall, all of which encode a signal peptide (SP) at their N-terminals. Interestingly, a longer SP (56 aa) was observed in NedA (OG2697, 1569 aa) that shared 30% identity and 48% similarity with C. perfringens NanJ (WP_011590241, 1173 aa), a secreted exosialidase that can generate breakdown sialoglycoconjugates found on host cell surfaces or mucus7. These findings indicated that selective forces may prompt genetic variations in certain genes encoding surface proteins interacted with host factors or antibiotics, which were promising candidates for vaccine design and drug development. Extracellular degradative enzymes are a class of secreted virulence factors critical for Gram-positive pathogens7,36. In C. paraputrificum, we identified four positively selected genes encoding extracellular proteins, all of which were annotated as uncharacterized proteins except the gene nagJ (OG0010, 1687 aa). NagJ is a signal peptide-bearing glycosyl hydrolase harboring an N-terminal discoidin domain (PF00754) followed by a catalytic domain (PF00728) of Glycosyl hydrolase family 20, and four continuous FIVAR domains (PF07554) mostly occurring in cell wall-associated proteins. Notably, C. paraputrificum NagJ shared 53% similarity with a polysaccharide lyase (1372 aa) of E. faecalis, which is not only a carbohydrate-degrading enzyme but also a pathogenic exoenzyme for cleaving hyaluronan in the extracellular matrix of connective tissues36. For those function-unknown proteins secreted by C. paraputrificum to extracellular space, more research on their virulence contribution is desirable. Except for cell wall/membrane biogenesis, we observed that the C. paraputrificum genes evolving under positive selection were distributed mainly in the COG functional categories below: “Transcription” (8 genes), “Carbohydrate transport and metabolism” (8), “Amino acid transport and metabolism” (6), “Replication, recombination and repair” (5), “Defense mechanisms” (5), “Signal transduction mechanisms” (5), and “Translation and ribosomal structure” (4) (Supplementary Data 9). For instance, rumA (OG1325,459 aa) coding for a 23S rRNA (uracil-5-)-methyltransferase involved in posttranscriptional modifications of nucleotides in RNA was undergoing intensive selection pressure (q-value < 0.001). The other two genes (frlR, 122 aa; gmuR, 227 aa) encoding transcriptional regulators of GntR family were also positively selected in C. paraputrificum. Similar phenomena for the genes related to translation and transcription have been pointed out for rumA in Actinobacillus pleuropneumoniae12 and GntR family transcriptional regulators in Streptococcus37. In addition, a sensor histidine kinase encoded by the positively selected gene baeS comprised a dimerization and phospho-acceptor domain (PF00512), which has been deciphered in PhoR, a sensor kinase important for the virulence of M. tuberculosis38. BaeS (OG1420, 454 aa) of C. paraputrificum shared 51% similarity with M. tuberculosis PhoP (NP_215271, 485 aa), which could regulate the expression of genes related to surface remodeling and adaptation to intracellular growth39. It suggested that natural selection would operate on the genes that enable bacteria to better survive, colonize, invade, and proliferate in the complicated niches. In summary, genome biology and metabolic pathways of C. paraputrificum isolated from blood culture were linked to the physiological and pathological knowledge of this uncommon anaerobe. The pangenomic structure of 42 C. paraputrificum strains were deciphered to reveal the functional properties of core and accessory genes. We further characterized genetic diversity of 52 virulence-associated genes, 15 antimicrobial resistance genes, and 104 genes favored by positive Darwinian selection in C. paraputrificum. These findings provide meaningful gene markers for future research into the pathogenesis mechanisms of clostridial infection. Methods Sample collection In May of 2024, a 55-year-old male patient was admitted to Shanghai Jiao Tong University affiliated Ruijin Hospital. He experienced persistent abdominal pain for three days together with nausea and vomiting. The study protocol was approved by the Ethics Committee of Ruijin Hospital, and the permit number was No.2021213. The patient provided written informed consent. All ethical regulations relevant to human research participants were followed. During admission, routine clinical indicators were measured, including white blood cells, neutrophils, C-reactive protein, procalcitonin, and IL-6 levels. Microbiological testing was conducted using samples of the patient’s blood, sputum, and drainage fluid. Bloodstream infection was diagnosed, and anti-infection therapy was adopted by combinations of antibiotics: tienam, fluconazole, tigecycline, and morinidazole. Colonies from blood agar medium under anaerobic conditions were identified as C. paraputrificum by Matrix-Assisted Laser Desorption/Ionization Time-of-Flight VITEK® Mass Spectrometry (VITEK MS, bioMérieux). C. paraputrificum was positive in blood culture two days apart. Swabs of bacterial colonies were stored at -20˚C for sequencing experiments. Genomic sequencing Total genomic DNA was extracted using a Blood & Cell Culture DNA Mini Kit (Qiagen, Germany). A library with the insertion fragments of ~350 bp was constructed with Covaris M220 (Covaris, USA) and the TruSeqTM DNA Sample Prep Kit (Illumina Inc., USA). The quantity and quality of DNA libraries were examined by Qubit 4.0 (Invitrogen). Paired-end 150-bp read sequencing was carried out using Illumina NovaSeq at Personalbio Co., Ltd (Nanjing, China). Raw sequencing reads were trimmed and filtered by fastp v0.23.440 with default options to remove adapter sequences and duplicated reads. Genome assembly and annotation To enrich reads derived from the bacterial target, reference genomes of C. paraputrificum were retrieved from NCBI Assembly database and further filtered to keep the genomes with the number of contigs < 100. High-quality reads were aligned to reference genomes by BBmap v39.06. Matched reads were used for de novo assembly by Megahit v1.2.941. The resulting contigs were then used for graph construction and gap closure by Spades v3.14.1 with the option –trusted-contigs42. Quality metrics of genome assemblies were estimated with the lineage_wf module in CheckM v1.2.243. Average nucleotide identity (ANI) of pairwise genomes was calculated using the package PYANI v0.2.1144. Prokka v1.14.6 was employed to annotate protein-coding genes and non-coding genes (rRNAs and tRNAs)45. For metabolic reconstruction, genes were assigned with KEGG Ortholog (KO) identifiers by the server BlastKOALA46. The resulting KOs were then assigned to metabolic pathways and functional modules using KEGG Mapper (https://www.genome.jp/kegg/mapper/). Genetic elements associated with horizontal gene transfer were searched by mobileOG-db47 and ISEScan48. The prophage region was predicted by PHASTEST49. Phylogenomic and pangenomic analysis To infer the inter-species phylogenetic placement of the newly sequenced strain, the representative genomes for individual species within the genus Clostridium were downloaded from NCBI RefSeq database. Taxonomic lineages of the genomes were further prospected by the Genome Taxonomy Database (GTDB) release r214 using GTDB-Tk v2.3.2 with the “classify_wf” workflow and default options50. PhyloPhlAn v3.1.1 was adopted to detect and align sequences against a set of 400 universal prokaryotic markers51. The concatenated sequence alignments were generated to reconstruct an approximate maximum-likelihood tree using FastTree v2.1.10. On the other hand, an intra-species phylogenomic tree was built based on a core SNP alignment generated by Snippy v4.6.0 (https://github.com/tseemann/snippy). The distribution of pairwise SNP distances was assessed by GraphSNP52. For the pangenome analysis, Roary v3.13.053 was used to cluster orthologous genes (OGs) across the strains with a minimum percentage identity set to 90%. Fitting curves of bacterial pan-genome and core-genome genes were calculated by PanGP v 1.0.154. Development of the MLST scheme Due to the unavailability of applications for molecular typing of C. paraputrificum, we created an MLST scheme to decode clonal diversity. Seven single-copy core genes were identified and selected from gene clusters generated by the pangenome analysis. Duplicated nucleotide sequences for each gene were removed from the alignments, and unique alleles were then counted. Selective pressure on the coding region was estimated by the ratio (ω) of non-synonymous to synonymous substitutions (dN/dS), which was measured using the M0 model (one ω for all sites and branches) implemented by CODEML in the PAML package v4.955. Referring to most of the schemes in pubMLST56, 450-bp internal fragments of seven conserved genes under purifying selection (ω < 1) were extracted to define the sequences for every allele locus. The new MLST scheme was applied to the C. paraputrificum genomes by mlst v2.23.0 (https://github.com/tseemann/mlst). The resulting allelic profile was used to build the minimum spanning tree and to visualize the relatedness of sequence types (STs) together with strain metadata using GrapeTree v1.5.057. Analysis of gene function Functional categories and domains of the OGs were annotated according to the databases COG58 and Pfam v35.059. Virulence-associated genes were searched based on a manually curated set of reference proteins derived from Gram-positive bacteria in the virulence factor database VFDB60. Using Blast v2.12.0+61, the best hit for each query was retained according to the thresholds below: E-value of 1e-20, minimum sequence identity of 40%, and subject coverage greater than 60%. Antimicrobial resistance genes were predicted using AMRFinderPlus v3.12.8 with database version 2024-07-22.1 and default options62. Protein subcellular localization was predicted using PSORTb v3.03 web server with the option: gram stain positive63. The topology of transmembrane proteins was detected by DeepTMHMM v1.064. Detection of positive Darwinian selection Evidence for positive selection driving adaptive protein evolution was detected according to a modified protocol described previously65. Briefly, both nucleotide and amino acid sequences of single-copy OGs present in at least 12 genomes were extracted based on the pangenome matrix. Multiple sequence alignments of protein clusters were performed using MUSCLE v5.166 and then converted to codon alignments using PAL2NAL67. Badly aligned regions and gapped sites were masked from the alignments using GBLOCKS 0.91b68. The short alignments with less than 50 codons were also filtered out. The remaining alignments were used to build maximum-likelihood gene trees using IQ-TREE v2.1.469, which assessed the best-fit substitution models and the SH-like approximate likelihood ratio test. The tests for positive selection were carried out using site-specific models implemented by CODEML: M1a (Nearly Neutral) vs. M2a (Positive Selection)55. The posterior probability of individual codon sites from the class of positive selection with ω > 1 was estimated by Bayes empirical Bayes. Positively selected genes and residues were visualized in three-dimensional structural models using Phyre270. Statistics and reproducibility Fisher-exact test was carried out to determine significance of enrichments between different gene-sets in individual COG functional categories. To control Type I erros, p values were then corrected using the Bonferroni method. The significance of positively selected genes was inferred by twice the log-likelihood difference (2 Δℓ) of the M1a-M2a comparison and a chi-square distribution with two degrees of freedom. Multiple testing correction was conducted to calculate q-values using the Benjamini & Hochberg method71. The significance threshold was set to 0.05 in this study. The results of statistical analysis were shown in the Supplementary Tables S6 and S9. Supplementary information Description of additional supplementary file Supplementary Data 1-9 Description of additional supplementary file Supplementary Data 1-9