Implications of Genome-Based Discrimination between Clostridium botulinum Group I and Clostridium sporogenes Strains for Bacterial Taxonomy

Abstract

Taxonomic classification of Clostridium botulinum is based on the production of botulinum neurotoxin (BoNT), while closely related, nontoxic organisms are classified as Clostridium sporogenes. However, this taxonomic organization does not accurately mirror phylogenetic relationships between these species. A phylogenetic reconstruction using 2,016 orthologous genes shared among strains of C. botulinum group I and C. sporogenes clearly separated these two species into discrete clades which showed ∼93% average nucleotide identity (ANI) between them. Clustering of strains based on the presence of variable orthologs revealed 143 C. sporogenes clade-specific genetic signatures, a subset of which were further evaluated for their ability to correctly classify a panel of presumptive C. sporogenes strains by PCR. Genome sequencing of several C. sporogenes strains lacking these signatures confirmed that they clustered with C. botulinum strains in a core genome phylogenetic tree. Our analysis also identified C. botulinum strains that contained C. sporogenes clade-specific signatures and phylogenetically clustered with C. sporogenes strains. The genome sequences of two bont/B2-containing strains belonging to the C. sporogenes clade contained regions with similarity to a bont-bearing plasmid (pCLD), while two different strains belonging to the C. botulinum clade carried bont/B2 on the chromosome. These results indicate that bont/B2 was likely acquired by C. sporogenes strains through horizontal gene transfer. The genome-based classification of these species used to identify candidate genes for the development of rapid assays for molecular identification may be applicable to additional bacterial species that are challenging with respect to their classification.

INTRODUCTION

Botulinum neurotoxins (BoNT) cause neuromuscular paralysis associated with botulism and are produced by various clostridia, most notably Clostridium botulinum. C. botulinum is a Gram-positive, anaerobic, endospore-forming bacillus that can be classified on the basis of metabolic properties into four separate groups (groups I to IV). Group I C. botulinum strains are proteolytic, saccharolytic, and capable of producing BoNT types A, B, and F. Group II C. botulinum strains are nonproteolytic and can produce BoNT types B, E, and F, while group III strains produce BoNT types C and D. Group IV strains, which are also identified as Clostridium argentinense, produce BoNT type G.

The nontoxic species Clostridium sporogenes shares nearly identical metabolic properties with group I C. botulinum, including the formation of lipase-positive colonies when grown on egg yolk agar. Because of these similarities and the comparable heat resistance of its spores, C. sporogenes has been used as a surrogate organism for C. botulinum in the study of thermal processing for foods (1). Previous studies have shown that some strains of C. sporogenes and group I C. botulinum can be differentiated phenotypically by soluble protein expression, measured by polyacrylamide gel electrophoresis (2) and gas-liquid chromatography of trimethylsilyl derivatives of whole-cell hydrolysates (3).

Genetic studies such as DNA-DNA hybridization and 16S rRNA gene sequence analysis have shown that these species are closely related (4–7). While acknowledging a genetic relationship at the species level, Olsen et al. (8) proposed that the species name C. botulinum be reserved for toxigenic strains and that C. sporogenes be conserved for nontoxigenic strains in order to prevent confusion between these organisms, and this proposal has met with wide acceptance.

As genome sequencing becomes more commonly used for microbial identification, it is important to evaluate this taxonomic organization in a phylogenomic context. Bacterial species are often determined on the basis of a limited number of diagnostic phenotypic traits, while the ability of bacteria to form actual discrete units (i.e., species) is a topic of intense research (9–12). Moreover, horizontal gene transfer (HGT) of genes encoding diagnostic traits may act to further reduce the boundaries between bacterial species (13). In this study, we used genome sequence analysis to resolve C. sporogenes and group I C. botulinum strains on the basis of genetic relatedness between, as well as within, these named species. In addition, this analysis demonstrated a previously unrecognized role for HGT of some BoNT/B-encoding genes between these organisms.

MATERIALS AND METHODS

Initial characterization of bacterial strains.

All strains examined in this study were grown in Trypticase-peptone-glucose-yeast extract (TPGY) medium (Remel, Lenexa, KS) at 35°C under anaerobic conditions. Single colonies were isolated from egg yolk agar (EYA). Strains presumptively identified as C. sporogenes (Table 1) based on microbiological characteristics and lack of toxicity using the mouse bioassay (14) were reexamined using the botulinum toxin gene real-time PCR assay (distributed by the Laboratory Response Network, CDC) to demonstrate absence of botulinum neurotoxin genes (A to G) and by botulinum toxin enzyme-linked immunosorbent assay (ELISA) (15) to demonstrate lack of toxin (types A, B, E, and F) production. Additional toxic and nontoxic C. botulinum and C. sporogenes strains examined in this study are also described in Table 1.

TABLE 1.

Characteristics of strains examined in this study^a

Strainb	Toxin type	Species	Location	Year	Source
CDC41370	Ab	C. botulinum	Mexico	1996	Food
CDC41370NT	NT	C. botulinum	NA	NA	Spontaneous mutant of CDC41370
CDC68016	B1	C. botulinum	West Virginia	2013	Stool
CDC68016NT	NT	C. botulinum	NA	NA	Spontaneous mutant of CDC68016
ATCC 7949	B2	C. botulinum	Unknown	Unknown	Unknown
Prevot25 NCASE	B2	C. botulinum	Unknown	Unknown	Unknown
Prevot1662	B2	C. botulinum	Unknown	Unknown	Unknown
ATCC 51387	B2	C. botulinum	Unknown	Unknown	Unknown
CDC66221	B6	C. botulinum	Colorado	2011	Stool
11579	NT	C. sporogenes	Unknown	1972	Unknown
85-3852	NT	C. sporogenes	Unknown	Unknown	Unknown
87-0535	NT	C. sporogenes	Unknown	Unknown	Unknown
88-0163	NT	C. sporogenes	Unknown	1988	Blood
CDC22719	NT	C. sporogenes	Colorado	1977	Chili sauce
CDC22720	NT	C. sporogenes	Colorado	1977	Cheese
CDC21678	NT	C. sporogenes	Hawaii	1976	Abdominal drainage
CDC21494	NT	C. sporogenes	Germany	1976	Canned ham
CDC22679	NT	C. sporogenes	Puerto Rico	1977	Canned meat
CDC22680	NT	C. sporogenes	Puerto Rico	1977	Canned meat
CDC22753	NT	C. sporogenes	South Carolina	1977	Wound (foot)
CDC22798	NT	C. sporogenes	South Carolina	1977	Blood
CDC23091	NT	C. sporogenes	Oklahoma	1978	Gastric fluid
CDC23284*	NT	C. sporogenes	Maryland	1978	Abdominal fluid
CDC23285*	NT	C. sporogenes	Maryland	1978	Peritoneal fluid
CDC24474	NT	C. sporogenes	Maryland	1979	Abdominal fluid
CDC24545c	NT	C. sporogenes	New Zealand	1979	Dotterel (bird)
CDC24442c	NT	C. sporogenes	New Zealand	1979	Dotterel (bird)
CDC24533	NT	C. sporogenes	Arizona	1979	Wound (leg)
CDC24968	NT	C. sporogenes	Missouri	1979	Stool
CDC24726	NT	C. sporogenes	North Carolina	1979	Liver paste
NCTC 534	NT	C. sporogenes	Unknown	∼1920	Human
CDC24614	NT	C. sporogenes	Florida	1979	Wound
CDC35120†	NT	C. sporogenes	California	1980	Blood
CDC35121†	NT	C. sporogenes	California	1980	Lung
CDC35566	NT	C. sporogenes	Missouri	1980	Unknown
CDC35197‡	NT	C. sporogenes	Alaska	1980	Wound drainage
CDC35196‡	NT	C. sporogenes	Alaska	1980	Wound drainage

^aNT, nontoxic; NA, not applicable.

^bIsolates from same individual share the same symbol.

^cIsolates from identical cultures.

DNA extraction and draft genome sequence assembly.

Genomic DNA from all isolates was extracted from TPGY cultures as previously described (16) and purified using the DNA Clean & Concentrator-5 kit (Zymo Research, Irvine, CA) prior to library construction. Genome sequencing was performed using either the Illumina MiSeq or Life Technologies Ion Torrent PGM instrument (Table 2). Illumina reads were filtered and trimmed at both the 5′ and 3′ ends based on a Phred score threshold (Q) of 20 using SolexaQA (17). For each isolate, quality-trimmed reads were assembled first in parallel runs of Velvet v1.0.13 (18) and SOAPdenovo2 (19) with a range of k-mer values, as described previously (20). The resulting precontigs were pooled and then assembled using the Genome Sequencer (GS) de novo assembler software v2.0.01.14 (Roche, Branford, CT). Ion Torrent reads were assembled using the GS de novo assembler. For both platforms, assembled contigs were ordered using Mauve Contig Mover (21) with C. botulinum strain Loch Maree as a reference, and assembly validation was performed with Mauve Assembly Metrics (22). The resulting draft genome sequences were annotated with the Rapid Annotation using Subsystems Technology (RAST) server (23). Despite a large number of contigs in some assemblies, more than 95% of the core genes were fully recovered, suggesting that no major sequencing gaps remained.

TABLE 2.

Statistics of the genome sequences generated in this study

Strain	Platforma	No. of assembled reads	No. of contigs	N50b	Estimated genome size (Mb)	Coverage	GenBank accession no.
CDC41370	PGM	536,228	980	7,449	4.1	26×	LAGI01000000
CDC41370NT	PGM	695,428	254	30,380	3.8	37×	LAGJ01000000
CDC68016	PGM	654,420	872	9,380	4.2	33×	LAGK01000000
CDC68016NT	PGM	392,523	1,714	3,800	4.3	18×	LAGL01000000
ATCC 7949	PGM	1,012,546	206	50,992	3.8	57×	LAGE01000000
Prevot25NCASE	PGM	1,088,680	192	42,026	4.1	56×	LAGN01000000
Prevot1662	PGM	775,204	593	19,719	4.9	30×	LAGM01000000
ATCC 51387	PGM	1,178,142	302	31,395	4.3	56×	LAGD01000000
11579	PGM	577,378	317	27,263	4.4	26×	JZJN01000000
85-3852	PGM	634,904	305	30,089	4.1	32×	JZJO01000000
87-0535	MiSeq	1,078,369	226	46,656	3.8	37×	JZJP01000000
88-0163	MiSeq	1,596,670	247	40,700	3.8	53×	JZJQ01000000
CDC66221	PGM	422,086	818	9,229	4.1	17×	LAGO01000000
CDC23284	PGM	580,480	506	17,646	4.1	28×	LAGF01000000
CDC24442	PGM	476,812	368	14,470	3.1	27×	LAGG01000000
CDC24533	PGM	793,767	344	24,303	4.1	41×	LAGH01000000

^aPGM, Life Technologies Ion Torrent personal genome machine. MiSeq was from Illumina.

^bN₅₀ is the contig length at which half of the total assembly is present in contigs of that length or longer.

The nucleotide sequences of bont/B2 and bont/B6 genes were determined by sequence read mapping using CLC bio Genomics Workbench version 7.5. For bont/B2-containing genomes, the ATCC 7929 bont/B sequence (GenBank accession number EF028395) was selected as the reference for read mapping, while the Okayama 2011 bont/B sequence (GenBank accession number AB665558) was used for bont/B6-containing genomes.

Additional comparative analysis included alignment of selected genome sequences with pCLD (from C. botulinum strain Okra) using the BLAST Ring Image Generator (BRIG) (24) and alignment of the chromosomally located toxin gene cluster region of specific genomes containing bont/B2 using the R package genoPlotR (25). The locations of specific genes within these toxin gene clusters were predicted using GeneMarkS (26).

Core genome phylogenetic reconstruction.

All predicted protein-coding genes from each of the sequenced genomes and available reference sequences were compared using an all-versus-all BLAST search as described previously (27). This analysis identified shared reciprocal best matches in all pairwise genome comparisons (core orthologs) of C. sporogenes and either all C. botulinum strains or only group I C. botulinum strains. These core orthologs were individually aligned using MUSCLE (28). The resulting alignments were concatenated to create a whole-genome alignment, and the phylogeny of genomes was reconstructed by computing a maximum-likelihood distance matrix with RAxML (29, 30).

Genetic signature identification.

The collection of variable genes in each genome, defined as genes absent in 1 or more genomes, was identified in the all-versus-all BLAST search described above. The presence or absence of these variable genes was used to hierarchically cluster genomes using complete linkage across a centered Pearson correlation similarity matrix using Cluster 3.0 (31). Variable genes were considered C. sporogenes specific if present in at least 7 out of 8 genomes and absent in all group I C. botulinum genomes. Conversely, genes present in 11 out of 13 group I C. botulinum genomes but absent in all C. sporogenes genomes were classified as C. botulinum specific. These ratios correspond to at least 80% of the population of strains examined for each species (representing values >1 standard deviation from the mean of the population) and are therefore expected to be robust to spurious matches.

Functional annotation of predicted proteins encoded by species-specific genes was bioinformatically inferred using a combination of three independent approaches. First, sequences were searched against the UniProtKB/Swiss-Prot database (32) using DELTA-BLAST (33) with a bit score cutoff of 200. Second, conserved domains were predicted by RPS-BLAST against the position-specific score matrices of the NCBI Conserved Domain Database (34) using an E-value threshold of 0.01. Finally, a BLASTX search against the NCBI nonredundant (nr) database was performed with a bit score cutoff of 200 and identity greater than 96%. The results of these independent searches were manually inspected, and the corresponding genes were functionally annotated based on the consensus of the three.

Genetic signature PCR.

A total of 143 putative C. sporogenes-specific orthologs were identified, from which 11 candidates were selected for primer design for PCR amplification. The annotated functions and corresponding NCBI locus ID for the candidate signatures are shown in Table S1 in the supplemental material. PCR amplification was performed on genomic DNA extracted from 24 C. sporogenes isolates using primers targeting each of the 11 biomarkers (shown in Table 3). PCR thermocycling conditions consisted of an initial denaturation at 95°C for 3 min followed by 35 cycles of 95°C for 30 s, 50°C for 30 s, and 72°C for 90 s with a final extension of 72°C for 10 min. The amplified products were examined on 1% ethidium bromide-stained agarose gels and visualized under UV transillumination.

TABLE 3.

C. sporogenes clade-specific gene signatures

Gene or locus	Annotated function	PCR primers (5′→3′)a
hsdMSR	Type I restriction modification	TGAATCGGAAACCGATGGAC
		TGCCACTTGGCTTCATTTCT
ssuABC	Alkane sulfonate transport	TAATCC CCTGGGCTTTACCT
		CCAGCTGTTATCTCATGCCA
isdAFD	Heme transport	CCAGGAAAGGGCAAAAACCA
		TGGAGCACCAGGACACTTTA
thiHG	Thiazole biosynthesis	RCGCTTGTGTCCAKCTAAAT
		GCCGGGTATGGAGYTAATTG
lipAS	Lipoate protein ligase and synthase	TGTGACGAAGCTAATTGTCCT
		CTTTTTCCCCAAGACCTACCA
ltaS	Lipoteichoic acid synthase	ATGGGAGAGGCCAGAAAGTA
		TGCCTCTCAGATACCATCC A
gadA	Glutamate decarboxylase	AGATATTGGAGCGCCTGAGA
		CTTGTCTCCCCAATCTGGTT
dacB	d-Alanyl-d-alanine carboxypeptidase	TCCAGAAGCCTGCTCTATCC
		AAAGTGCCTGGTGCTGCTAT
nifS	Cysteine desulfurase	ATAGACTCTGCTCAAACCGC
		GTGTAATCC GCATCTGGTCA
dapL	N-Acetyl-l,l-diaminopimelate deacetylase	ACGCCTTCTGTAGCATCAAA
		GCCTGGACTTTAGGTACTGC
spoVB	Stage V sporulation protein B	GTGTTTACGTCCTGCTTTTTCA
		TGCGCTCTAGAGGAAGATACT

^aThe pairs of primer sequences represent forward (top) and reverse (bottom) primers.

16S rRNA sequencing.

A ∼1.3-kb region of the 16S rRNA gene from presumptively identified C. sporogenes strains was amplified and sequenced using primers that were previously described (35). Sanger sequencing was performed using the GenomeLab dye termination cycle sequencing kit and a CEQ8000 genetic analysis system (Beckman Coulter, Brea, CA). Sequences were assembled and edited with Sequencher v 4.8 (Gene Codes, Ann Arbor, MI).

Nucleotide sequence accession numbers.

Newly determined sequence data have been deposited in GenBank under the accession numbers shown in Table 2.

RESULTS

Core genome phylogenetic reconstruction.

An all-versus-all BLAST comparison indicated that the pangenome for strains of C. sporogenes and C. botulinum (including groups I, II, and III) was comprised of 16,229 genes. Of these, 179 genes were shared by all strains (core orthologs), and a phylogenetic reconstruction based on their concatenated alignment revealed four distinct clades (Fig. 1). As expected, strains of C. botulinum groups I, II, and III were well separated. The C. sporogenes strains were most closely related to a clade of group I C. botulinum strains, consistent with previous genomic studies (36, 37), but formed their own discrete clade. The C. sporogenes clade contained C. sporogenes strains 88-0163, ATCC 15579, 87-0535, PA3679, 85-3852, and 11579 and two C. botulinum type B strains, Osaka05 and Okayama2011. Furthermore, the average pairwise genome aggregate average nucleotide identity (ANI) between genomes within the group I C. botulinum clade and the C. sporogenes clade was 93.4%, below the 95% ANI cutoff frequently used for species demarcation (38), supporting the classification of these strains into separate bacterial species. Further comparison of strains in the C. sporogenes and group I C. botulinum clades revealed a pangenome (for both clades combined) comprised of 9,306 orthologs, of which 2,016 were core and 7,290 were variable among these closely related genomes.

FIG 1.

Core gene phylogenetic trees of C. sporogenes and C. botulinum genomes. An initial tree was based on the maximum-likelihood distance calculated from the alignment of 179 shared orthologous genes of C. sporogenes and C. botulinum groups I, II, and III. The tree in the inset was based on the maximum-likelihood distance calculated from the alignment of 1,451 shared orthologous genes restricted to C. sporogenes and group I C. botulinum. Specific clades discussed in the text are labeled. The BoNT produced by each strain in indicated in brackets (NT, nontoxic). Genomes sequenced in this study are in red. Additional genome sequences were retrieved from GenBank.

Clade-specific gene identification.

Hierarchical clustering based on presence or absence of the variable genes (absent in one or more genomes) among the C. sporogenes and C. botulinum strains examined in this study again separated C. sporogenes strains from C. botulinum group I, group II, and group III mirroring the phylogenetic relationships inferred from the core genome phylogeny (data not shown). When the analysis was restricted to the 16,412 variable genes present in the C. sporogenes and group I C. botulinum genomes specifically, the C. sporogenes strains continued to form a discrete cluster, suggesting that there exist clade-specific gene signatures in C. sporogenes genomes (Fig. 2A).

FIG 2.

Hierarchical clustering based on the presence or absence of variable orthologs. (A) The presence (yellow) or absence (black) of variable orthologs (n = 16,412) in C. sporogenes and C. botulinum group I genomes was used to perform hierarchical clustering using a complete linkage across a centered Pearson correlation. (B) Identification of 143 orthologs genes that are present in the genomes of 7 out of 8 C. sporogenes strains and 91 genes that are present in 11 out of 13 group I C. botulinum strains.

Selective filtering of variable orthologs identified 143 clade-specific genes that were present in at least 7 out of 8 available C. sporogenes-clade genomes (strains 11579, 85-3852, 88-0163, ATCC 15579, 87-0535, PA3679, Osaka05, and Okyama2011) and absent in the genomes associated with the group I C. botulinum clade (Fig. 2B). Of these, 65 were annotated as encoding hypothetical proteins. Annotation of the remaining genes included various functions related to nutrient uptake (e.g., heme transport and alkane sulfonate uptake) and bacterial cell defense against invasion of foreign DNA (e.g., type 1 restriction modification system). Correspondingly, 91 genes were observed in at least 11 out of the 13 genomes associated with the group I C. botulinum clade that were absent from all C. sporogenes strains examined. From these, 44 genes were annotated as hypothetical proteins, while the remaining genes were functionally associated with metabolic pathways (e.g., V-type ATP synthase) and cell wall modifications.

Genetic signatures for clade discrimination.

Eleven signatures from the set of C. sporogenes clade-specific genomes, each representing a different operon or function, were further evaluated and annotated for potential use in PCR assays. The candidate signatures were selected from throughout the C. sporogenes genome in order to evenly sample the genome backbone (see Fig. S1 in the supplemental material). Manual curation of bioinformatic evidence was used to infer the putative function of each gene, and these annotation assignments are summarized in Table 3.

A total of 24 nontoxic lipase-positive strains presumptively identified as C. sporogenes were selected from the CDC culture collection to test the efficacy of C. sporogenes clade-specific PCR assays targeting the identified gene signatures. While PCR results were variable for some genes, a subset of five loci (hsdMSR, thiHG, lipAS, ltaS, and dapL) showed highly consistent results for the nontoxic strains examined (Fig. 3). These five conserved signatures distinguished the 24 nontoxic strains into two groups; those positive for the presence of all five signatures (referred to as “signature positive”; 18 strains) and those negative for all five signatures (“signature negative”; 6 strains).

FIG 3.

Gene signature PCR. The results of gene signature PCR assays for each locus are shown in the columns, which are labeled across the top. The presence of a PCR product for each assay among the strains examined (indicated to the left of the figure) is indicated in red. The dotted box indicates the loci that demonstrated invariant PCR results among the strains examined.

Three putative C. sporogenes strains were selected for genome sequencing to determine whether the genetic signatures detected by PCR were predictive of the clade assigned by core genome phylogeny. As expected, strain CDC23284, which was signature positive, clustered with the C. sporogenes clade accordingly (Fig. 1). Both CDC24533 and CDC24442 were signature negative, but only the CDC24533 clustered with the group I C. botulinum clade, while strain CDC24442 formed a distinct cluster (Fig. 1). Notably, the genome sequences of these three strains were also consistent with the 16S rRNA gene sequence analysis of all 24 strains (i.e., strains lacking the C. sporogenes clade-specific signatures based on PCR analysis were phylogenetically distinct from strains containing the signatures [see Fig. S2 in the supplemental material]). The 16S rRNA gene sequences of strains CDC23284 and CDC24533 differed by 0.3% compared to each other and by ∼1% compared to strain CDC24442. Strain CDC24442 shares 16S rRNA gene sequence identity with C. sporogenes subsp. tusciae biovar pennavorans (39) and clearly represents a distinct species, consistent with its distinct position in the core genome phylogeny.

Horizontal transfer of botulinum toxin genes.

Since neurotoxin gene loss should not affect core genome phylogenetic reconstruction, we also sequenced the genomes of two additional pairs of strains where a toxic and spontaneous nontoxic isolate were available. C. botulinum type B strain CDC68016 was isolated from the stool of an infant with botulism, and a nontoxic isolate (CDC68016NT) was also obtained during its laboratory cultivation. Both strains clustered with the C. sporogenes clade in the core genome phylogeny (Fig. 1). While strain CDC68016 was initially classified as C. botulinum based primarily on the production of botulinum neurotoxin, this strain and its nontoxigenic progeny (CDC68016NT) are more appropriately viewed as belonging to the C. sporogenes clade based on their phylogenetic relationships. These findings also suggest that the type B neurotoxin gene present in strain CDC68016 was horizontally acquired, presumably from a C. botulinum donor. Conversely, the C. botulinum type Ab strain CDC41370 (isolated from a food source) and its corresponding nontoxic derivative (CDC41370NT) clustered within the group I C. botulinum clade.

In order to examine whether strains containing the same toxin gene variant would belong to the same genomic clade, we sequenced strain CDC66221. This strain contains the bont/B6 variant (Fig. 4), similar to strains Osaka05 and Okyama2011, which belong to the C. sporogenes clade (Fig. 1). However, core genome phylogeny placed strain CDC66221 within the C. botulinum clade (Fig. 1). Hence, strain CDC66221 can be considered a member of C. botulinum according to a genome-based taxonomy, suggesting that strains with the same neurotoxin gene variant do not always group within the same clade.

FIG 4.

Neighbor-joining tree of bont/B sequences. bont/B2 sequences are in blue, and bont/B6 sequences are in red. Representative sequences of other bont/B variants were retrieved from GenBank as follows: B1, strain Okra (NC_010379); B3, strain CDC795 (EF028400); B4, strain 17B (EF051570); B5, strain CDC4013 (GU271943); B7, strain NCTC3807 (JN120760); B8, strain Chaiyaphum_2014 (KM067395).

To better appreciate the frequency of horizontal exchange of neurotoxin genes between strains belonging to either the C. botulinum or C. sporogenes clade, we analyzed four additional C. botulinum type B strains (Prevot1662, ATCC 51387, Prevot25 NCASE, and ATCC 7949) containing the neurotoxin gene variant bont/B2 (Fig. 4). Two strains (ATCC 51387 and Prevot 1662) were positive and two strains (ATCC 7949 and Prevot 25 NCASE) were negative for the C. sporogenes clade-specific locus hsdS by PCR (data not shown). Not surprisingly, the two hsdS-negative strains clustered with the C. botulinum clade and the two hsdS-positive strains clustered with the C. sporogenes clade in the core genome phylogeny (Fig. 1). Taken together, these data reveal that the transfer of type B neurotoxin genes between strains belonging to clades representing C. botulinum and C. sporogenes has occurred in multiple genomes and involved different bont/B sequence variants (e.g., strains ATCC 51387 and Prevot 1662, carrying bont/B2, and Osaka05 and Okyama2011, carrying bont/B6; all four strains group within the C. sporogenes clade).

Diversity of type B neurotoxin gene-bearing plasmids.

Comparison of the bont/B2-containing genomes of C. sporogenes clade strains ATCC 51387 and Prevot 1662 with the bont/B1-bearing plasmid (pCLD) of strain Okra (Fig. 5A) revealed a high level of nucleotide sequence similarity. These findings are consistent with the presence of a PCR product for the previously described PL-6 plasmid marker (40) in strains ATCC 51387 and Prevot 1662 (data not shown). In contrast to these strains, which may have acquired bont/B2 via the transfer of plasmid DNA, the C. botulinum clade strains ATCC 7949 and Prevot25 NCASE appear to contain bont/B2 within their chromosomes, colocated within the oppA-brnQ operon (Fig. 5B), similar to the location of the unexpressed bont/B found in C. botulinum type A(B) strain NCTC2916 (36).

FIG 5.

BLAST analysis of draft genome sequences. (A) Genomes were compared using BRIG with the bont/B-bearing plasmid (pCLD) in the C. botulinum type B strain Okra (GenBank accession number NC_010379). Genomes shown include (from innermost to outermost ring) ATCC 7949, Prevot 25 NCASE, ATCC 51387, Prevot 1662, Osaka05 (extrachromosomal element 2), Okayama2011, CDC66221, CDC41370NT, CDC41370, CDC68016NT, and CDC68016. Regions with >50% nucleotide identity compared to pCLD are shaded. Shading color corresponds to the bont/B sequence variant, as follows: blue, bont/B2; red, bont/B6; green, bont/B5; and purple, bont/B1. The positions of toxin gene cluster genes are indicated by black arrows. (B) The nucleotide sequence similarities of contigs containing the bont/B neurotoxin gene complex from assembled draft genome sequences ATCC 7949 and Prevot25 NCASE were compared. As a reference, the oppA-brnQ operon of C. botulinum type A(B) strain NCTC2916 (GenBank accession number NZ_ABDO02000001) was also compared to ATCC 7949. Regions sharing high nucleotide similarity are shaded.

Similarly, the genomes of bont/B6-containing strains Okayama2011 and CDC66221 also shared a high level of similarity with pCLD. The sequence of a 185-kb extrachromosomal element in strain Osaka05 (GenBank accession number NZ_BA000059) also contains the bont/B6 toxin gene cluster but shares less similarity with pCLD (Fig. 5A). While all three of these bont/B6-containing strains appear to carry the toxin gene on a plasmid, they differ with respect to their core genome phylogeny, suggesting that such a plasmid may be able to exist in strains of either the C. botulinum or C. sporogenes clade.

DISCUSSION

Previous studies have highlighted the challenging taxonomy of C. botulinum and C. sporogenes (7, 37). In one study, the 16S rRNA genes gene sequences belonging to 110 Clostridium species were compared to identify molecular markers capable of differentiating various Clostridium species (7). However, C. botulinum and C. sporogenes were found to cluster together in a single clade of the resulting phylogenetic tree. Additionally, comparative analysis of in silico restriction enzyme digestion showed that both species have exact profiles for the restriction enzymes AluI, BfaI, HaeIII, RsaI, Tru9I, and SmaI. The digestion profile for DpnII demonstrated that C. sporogenes was segregated into two separate populations, one of which showed exact homology with C. botulinum while the other was distinct from all other Clostridium species. These data, along with variations in DNA-DNA hybridization among different strains of C. sporogenes compared to C. botulinum (4, 5), suggest that C. sporogenes may be polyphyletic.

More recently, Bradbury et al. (37) reported the genome sequencing and annotation of Clostridium sporogenes strain PA3679, which has been used as a surrogate for group I C. botulinum in thermal processing of foods for over 70 years (41, 42). The percentage of shared k-mers (k = 25 bp) between the C. sporogenes PA3679 genome and several genomes of C. botulinum revealed that more than 85% of the C. sporogenes PA3679 genome matches group I C. botulinum strains (containing bont/A1). In addition, alignment of 16S rRNA gene sequences indicated a 99 to 100% nucleotide similarity between PA 3679 and several proteolytic C. botulinum strains, as well as other C. sporogenes strains.

In this study, we compared draft genome sequences to resolve C. sporogenes and C. botulinum group I strains by both core genome phylogeny and variable gene content analysis. Several bont-containing C. botulinum group I strains, such as Osaka05 and Okayama2011, were more closely related to C. sporogenes strains. Recent work by Kenri et al. (43) using whole-genome SNP analysis demonstrated that these two strains were distantly related to other C. botulinum group I strains examined. These results emphasize the conflictive nature of taxonomic designations based on the presence or absence of the botulinum neurotoxin gene, which does not reflect true phylogenetic relationships between strains. Kalia et al. (7) suggested, “It may not be too inappropriate to suggest that C. sporogenes is perhaps a sub-species of C. botulinum or it may find its appropriate place if C. botulinum can be reclassified as 4 different sub-species.” Our results support the latter, namely, that C. sporogenes is a distinct species of the genus, closely related to group I C. botulinum, because C. sporogenes strains are monophyletic on the core genome tree, show less than 95% ANI to their group I C. botulinum relatives but more than 95% ANI among themselves, and are characterized by specific genes and pathways that are rarely found in their relatives (Table 3). These genes may underlie important phenotypic differences between C. sporogenes and group I C. botulinum that remain to be elucidated. Identifying the associated phenotypes should be the subject of future work toward developing fast diagnostic tests for each of these two important bacterial taxa. While it is possible that some of the gene signatures for C. sporogenes may be found in other clostridial (or even nonclostridial) species, phylogenetic analysis would likely distinguish C. sporogenes from other species.

Among a panel of 24 nontoxic presumptive C. sporogenes isolated primarily from clinical samples, 18 were positive for C. sporogenes clade signatures. While the majority of BoNT-producing isolates examined in this study belonged to the C. botulinum group I clade, we observed some type B toxin producing isolates within the C. sporogenes clade suggesting that some toxin gene variants may be highly mobile among these two groups of organisms. Consistent with these findings, Carter et al. (44) identified only one named C. botulinum type B strain that grouped together with various C. sporogenes strains examined based on hybridization of a larger collection of strains against a ATCC 3502-specific (a group I C. botulinum type A strain) DNA microarray.

In a study of 63 C. botulinum type B strains, Franciosa et al. (45) found that more than half contained bont/B on a plasmid. Interestingly, neurotoxin gene-bearing plasmids were found among multiple bont/B sequence variants. At least some of these plasmids are likely to be mobile by conjugation, as shown by Marshall et al. (46). These findings are consistent with the results reported here that at least five C. sporogenes-like strains (CDC68016, ATCC 51387, Prevot 1662, Osaka05, and Okayama2011) appear to have acquired the neurotoxin gene via the horizontal transfer of plasmid DNA. Similar to another report (47), we also observed type B toxin gene loss in strains CDC41370 and CDC68016. The nontoxic strain CDC24533 belonging to the C. botulinum clade, specifically within a subclade containing strains that carry their toxin genes on plasmids (36, 48), presumably underwent plasmid loss. This strain was isolated from necrotic tissue of a leg wound and not associated with a case of wound botulism, indicating that loss of toxicity by this strain probably occurred prior to infection of the wound rather than during subsequent laboratory cultivation. Collectively, these results further support the conclusion that the current practice of classifying strains as C. botulinum or C. sporogenes based on the presence or absence of the botulinum toxin genes, respectively, cannot be highly reliable from a phylogenetic perspective, because toxin gene loss can result in confounding classification of the resulting nontoxic progeny.

In this study, two strains containing bont/B2 on a plasmid were associated with the C. sporogenes clade, while strains carrying this gene on the chromosome were located within the C. botulinum clade. Smith et al. (49) recently reported that the genome sequence of another bont/B2-containing strain (Prevot 594) with its neurotoxin gene on a plasmid was more related to strain Osaka05 than other toxigenic strains examined, suggesting that it was also a member of the C. sporogenes clade. Similarly, in silico multilocus sequencing typing of the draft genome sequence of C. botulinum strain 450 (isolated from a case of wound botulism and found to contain bont/B2) demonstrated a close phylogenetic relationship with C. sporogenes strain ATCC 15579 (50). Future work should include the study of a larger population of bont/B2-containing strains to determine the frequency at which bont/B2 is plasmid borne among strains belonging to the C. sporogenes clade. The results presented here, which are based on a limited number of strains, indicate that the transfer of neurotoxin genes among these clades may not be a rare event.

Distinguishing C. sporogenes strains from C. botulinum isolates that have lost toxicity has important implications. Since nontoxic isolates from distinct phylogenetic clades may have different phenotypic properties, it may be useful to consider a broader range of nontoxic organisms for C. botulinum surrogates (especially in food protection studies) than simply those identified as C. sporogenes.

Genomic and metagenomic approaches for microbial identification will likely lead to fewer laboratories attempting to detect botulinum toxin in culture. The results presented here clearly suggest that genomic sequencing technologies are highly reliable and robust in correctly classifying new strains and assessing their toxin production potential (i.e., identification of neurotoxin genes). While it is difficult to determine based on traditional techniques whether a nontoxic strain with microbiological properties resembling proteolytic C. botulinum or C. sporogenes may simply be the result of toxin gene loss, the genetic signatures developed in this study can help in determining the genomic background of these nontoxic and toxic strains and also correctly interpreting previous strain classifications. The presence of C. sporogenes clade-specific genes would provide investigators with a high-confidence genomic signature for “true” C. sporogenes strains. Nontoxic strains lacking such signatures may well be nontoxic isolates of C. botulinum and, in the context of a botulism case investigation, may necessitate screening larger numbers of isolates for toxicity. Collectively, the results presented here show that genome-derived data can offer robust resolution and classification of C. botulinum or C. sporogenes strains, even in the presence of HGT. These approaches provide a more phylogenetically accurate classification C. botulinum and C. sporogenes strains and promote a coherent bacterial species concept for these groups. Our results also underscore the limitations of the current bacterial classification system, which is frequently based on laboratory-assessed phenotypes that can be easily transferable between genetically distinct species, while providing an example of how to advance bacterial classification toward a genome-based taxonomy. This work should be applicable to additional bacterial taxa that are challenging with respect to classification.