Genetic Stability and Evolution of the sigB Allele, Used for Listeria Sensu Stricto Subtyping and Phylogenetic Inference

ABSTRACT

Sequencing of single genes remains an important tool that allows the rapid classification of bacteria. Sequencing of a portion of sigB, which encodes a stress-responsive alternative sigma factor, has emerged as a commonly used molecular tool for the initial characterization of diverse Listeria isolates. In this study, evolutionary approaches were used to assess the validity of sigB allelic typing for Listeria. For a data set of 4,280 isolates, sigB allelic typing showed a Simpson's index of diversity of 0.96. Analyses of 164 sigB allelic types (ATs) found among the 6 Listeria sensu stricto species, representing these 4,280 isolates, indicate that neither frequent homologous recombination nor positive selection significantly contributed to the evolution of sigB, confirming its genetic stability. The molecular clock test provided evidence for unequal evolution rates across clades; Listeria welshimeri displayed the lowest sigB diversity and was the only species in which sigB evolved in a clocklike manner, implying a unique natural history. Among the four L. monocytogenes lineages, sigB evolution followed a molecular clock only in lineage IV. Moreover, sigB displayed a significant negative Tajima D value in lineage II, suggesting a recent population bottleneck followed by lineage expansion. The absence of positive selection along with the violation of the molecular clock suggested a nearly neutral mechanism of Listeria sensu stricto sigB evolution. While comparison with a whole-genome sequence-based phylogeny revealed that the sigB phylogeny did not correctly reflect the ancestry of L. monocytogenes lineage IV, the availability of a large sigB AT database allowed accurate species classification.

IMPORTANCE sigB allelic typing has been widely used for species delineation and subtyping of Listeria. However, an informative evaluation of this method from an evolutionary perspective was missing. Our data indicate that the genetic stability of sigB is affected by neither frequent homologous recombination nor positive selection, which supports that sigB allelic typing provides reliable subtyping and classification of Listeria sensu stricto strains. However, multigene data are required for accurate phylogeny reconstruction of Listeria. This study thus contributes to a better understanding of the evolution of sigB and confirms the robustness of the sigB subtyping system for Listeria.

INTRODUCTION

The genus Listeria represents a group of Gram-positive, facultative anaerobic, non-spore-forming bacteria (1). It can be separated into Listeria sensu stricto and Listeria sensu lato, based on the relatedness of species to Listeria monocytogenes (2). Listeria sensu stricto consists of six closely related species, including two pathogenic species (L. monocytogenes and L. ivanovii) and four nonpathogenic species (L. innocua, L. welshimeri, L. seeligeri, and L. marthii). Listeria sensu stricto strains are globally distributed in diverse environments, except for L. marthii, which has been isolated only from very specific natural areas in New York State (USA), Connecticut Hill and Finger Lakes National Forest (2). In terms of public health, L. monocytogenes is arguably the most important species of this genus, as it can cause severe invasive disease in humans and animals (3, 4). Previous studies have shown that as a structured population, L. monocytogenes forms four lineages, designated as I, II, III, and IV (5, 6). Although the majority of human listeriosis outbreaks worldwide have been caused by lineage I strains, many studies found that most isolates from foods and food-associated environments group into lineage II (5, 7). Lineage III isolates are less common and have been isolated from animal clinical cases as well as some human clinical cases but have rarely been isolated from non-host-associated ecological habitats (7). Lineage IV isolates have predominantly been isolated from ruminants, including an unusual strain involved in a listeriosis outbreak in goats (8).

In order to accurately classify isolates into Listeria species, lineages, and subtypes, a robust subtyping system is required. sigB allelic typing has been widely used for species delineation and subtyping of Listeria (9, 10). The standard protocol for sigB allelic typing involves PCR amplification of a 780-bp fragment, followed by sequencing and analysis of a 660-bp sigB fragment (11). sigB encodes a stress-responsive alternative sigma factor, σ^B, which plays an important role in the survival of Listeria cells exposed to lethal conditions, e.g., heat, acid, ethanol, salt, oxidative agents, and/or carbon depletion (12), and contributes to the invasion of L. monocytogenes by controlling virulence gene expression (13). Compared with classical subtyping and characterization methods such as serotyping, biotyping, and phage typing, sigB allelic typing has been reported to be superior in its accuracy, reproducibility, sensitivity, and discriminatory power, hence allowing reliable identification and classification of Listeria spp. (14). Compared to other more comprehensive molecular characterization methods (e.g., multilocus sequence typing [MLST] and whole-genome sequencing [WGS]), sigB typing provides the advantage of low cost associated with the sequencing of only a single gene. However, despite the wide application of sigB allelic typing, there is still a lack of informative evaluation of this method from an evolutionary perspective. In particular, a robust candidate gene for genotyping should not be subject to frequent evolutionary forces such as horizontal gene transfer (HGT) and positive selection, since they will accelerate the molecular evolution of the specific gene, interfering with phylogeny reconstruction and resulting in less reliable subtyping. The validity of using a single gene for the reconstruction of a shared ancestry among lineages also needs to be evaluated by using multigene data sets (e.g., whole-genome sequence data), since evolutionary patterns vary among different genes.

Homologous recombination (HR) and positive selection are two important mechanisms of evolution that contribute to the speciation of many microbial lineages (15). HR, in which genetic material is acquired through the genetic exchange of similar DNA sequences (16), provides means of introducing genetic variation and can greatly increase the rate of adaptation to a changing environment (5). In the case of haploid bacteria, allelic HR typically occurs subsequent to HGT, which takes place through conjugation, transformation, or transduction (5, 17). In Listeria, conjugation and transduction are known to occur, whereas transformation is less likely, since natural competence has not been observed so far. Positive selection is a critical factor influencing allele frequency in a population, as it fixes beneficial variations in the population and increases the genetic diversity within species (18). A stringent and robust criterion indicating positive selection in a protein-coding gene is a ratio of nonsynonymous substitutions per nonsynonymous site (dN) to synonymous substitutions per synonymous site (dS) (ω) exceeding 1 (19). The contribution of homologous recombination and positive selection to sigB evolution has been assessed for L. monocytogenes (3, 5, 11). However, few studies have included a diverse set of Listeria sensu stricto isolates in analyses of these evolutionary processes at multitaxonomic levels.

The molecular clock hypothesis postulates that the rate of evolution of a gene or a protein is approximately constant through time and across clades (20–22). This hypothesis has helped in inferring evolutionary divergence dates (20) and fits well with the neutral theory that most evolutionary change at the molecular level arises from the fixation of neutral alleles through random genetic drift rather than natural selection (23). However, empirical studies have shown a great deal of variation in the rates of molecular evolution across clades and individual genomes as a consequence of dynamic interactions among evolutionary forces (i.e., mutation, selection, and drift) (11). Whether the evolution of sigB follows the molecular clock is still largely unexplored.

In this study, we assessed the discriminatory power of the sigB typing method using a set of 4,280 Listeria sensu stricto isolates. A subset of isolates representative of 164 unique sigB allelic types (ATs) was used in evolutionary analyses, including the detection of HR and positive selection and testing of the molecular clock. In order to assess the accuracy of a maximum likelihood (ML) phylogeny of Listeria sensu stricto constructed based on sigB, a sigB phylogeny was further compared with a core genome single nucleotide polymorphism (SNP) ML phylogeny of 21 isolates representative of different Listeria sensu stricto species and lineages. Hence, this study was carried out to (i) investigate the contribution of HR and positive selection to Listeria sensu stricto sigB evolution, (ii) test whether sigB evolution follows a molecular clock, and (iii) assess the advantages and disadvantages of using the sigB sequence for species classification, subtyping, and phylogenetic reconstruction of Listeria sensu stricto.

RESULTS

Descriptive analyses of sequence data.

Among 4,280 isolates, the internal fragment of sigB (660 bp, spanning positions 59 to 718 relative to a reference sigB sequence [GenBank accession number NC_003210.1] [930671.0.931450]) yielded 164 unique allelic types (see Data Set S1 in the supplemental material), with a nucleotide diversity per site (π) of 0.08 and a discriminatory power of 0.96 measured by Simpson's index of diversity (SID) (Table 1). These isolates represent Listeria sensu stricto, which includes L. monocytogenes, L. seeligeri, L. welshimeri, L. innocua, L. ivanovii, and L. marthii. A slight difference in the G+C content was observed across Listeria sensu stricto species (Table 1). L. monocytogenes, represented by four distinct lineages, was the most diverse species, with the highest nucleotide diversity (π = 0.02) and the most polymorphic sites (n = 94) (Table 1). L. welshimeri showed the lowest nucleotide diversity (π = 0.003), and L. marthii had the fewest polymorphic sites (n = 17). Within the phylogenetic lineages of L. monocytogenes, lineage II showed the highest level of polymorphism, while lineage III had the highest nucleotide diversity (Table 1). There were no apparent differences in sigB G+C contents among L. monocytogenes lineages. Across all Listeria sensu stricto strains, within each individual Listeria sensu stricto species, and within each L. monocytogenes lineage, there were more synonymous mutations than nonsynonymous mutations (Table 1).

TABLE 1.

Descriptive analysis of nucleotide sequence data for the 660-bp sigB fragment

Taxon (no. of sequences)	No. of:			G+C content (%)	πb	kc	Tajima's D valued	No. of mutationse		dN/dS ratiof
	Haplotypesa	Polymorphic sites	Mutations					Syn	Nonsyn
Listeria sensu stricto (4,280)	160	181	255	38.0	0.075	48.953	1.944	NA	NA	0.011
L. seeligeri (667)	23	49	51	38.3	0.017	11.239	1.524	49	2	0.013
L. welshimeri (565)	37	30	30	36.9	0.003	2.041	−1.374	24	6	0.027
L. innocua (1,226)	41	33	35	37.3	0.010	6.869	1.279	30	5	0.044
L. ivanovii (12)	4	25	25	39.6	0.015	9.879	0.861	23	2	0.017
L. marthii (14)	6	17	17	37.2	0.011	7.418	1.599	17	0	0.000
L. monocytogenes (1,796)	50	94	103	38.9	0.022	14.515	0.372	NA	NA	0.017
Lineage I (1,032)	10	13	13	38.9	0.004	2.452	−0.885	13	0	0.000
Lineage II (692)	12	34	35	38.8	0.001	0.757	−2.206	31	4	0.033
Lineage III (59)	24	26	26	38.7	0.005	3.571	−1.160	23	3	0.020
Lineage IV (13)	4	3	3	38.7	0.001	0.462	−1.652	3	0	0.000

^aSites with gaps are excluded in DnaSP, which was used to determine the number of haplotypes; as AT103, AT154, and AT165 have deletions in their sequences (2, 1, and 3 nucleotides, respectively), the number of Listeria sensu stricto haplotypes (n = 160) is less than the number of ATs (n = 164).

^bAverage pairwise nucleotide diversity per site.

^cAverage number of pairwise nucleotide differences per sequence.

^dTajima's D values significantly different from 0 are marked in boldface type.

^eSyn indicates the number of synonymous mutations; Nonsyn indicates the number of nonsynonymous mutations. If the number of synonymous or nonsynonymous mutations could not be unambiguously inferred, e.g., in highly divergent populations where there are multiple mutations per site, the result is reported as NA (not available).

^fThe ratio of nonsynonymous substitutions per nonsynonymous site to synonymous substitutions per synonymous site. The dN/dS ratio was calculated based on the average dN and dS values obtained by pairwise comparisons of sequences using approximate methods.

The data set containing all 4,280 isolates representing all Listeria sensu stricto isolates yielded a positive Tajima D value of 1.944. L. monocytogenes lineage II showed a significantly negative value (Tajima's D = −2.206; P < 0.01), indicating a recent lineage expansion after a population bottleneck or purifying selection (Table 1) (3, 24).

Recombination events.

One test–the pairwise homoplasy index (PHI) test–and two programs—recombination analysis using cost optimization (Recco) and RDP4, which incorporates nine different recombination detection methods (i.e., recombination detection program [RDP], Bootscan, GENECONV, Maxchi, CHIMAERA, SiScan, PhylPro, LARD, and 3Seq)—were used to detect HR events. PHI test found statistically significant evidence for intraspecific HR only among L. innocua sigB sequences (P = 0.002) (Table 2). Recco did not identify any significant recombination events. In RDP4, the Maxchi method detected two interspecific recombination events in L. monocytogenes, while the SiScan method detected one interspecific recombination in L. innocua when all Listeria sensu stricto sequences were analyzed. Within L. seeligeri sequences, the Maxchi method identified L. seeligeri AT 9 as being recombinant. Bootscan, SiScan, and 3Seq detected two intraspecific recombination events within L. innocua, one in L. innocua AT 124 and one in L. innocua AT 33. In both events, L. innocua AT 139 was identified as the minor parental sequence, and L. innocua AT 87 was identified as the major parental sequence. Within L. marthii sequences, the Bootscan, Maxchi, and LARD methods identified L. marthii AT 143 as a putative mosaic allele from L. marthii AT 95 and L. marthii AT 18. Intraspecific HR events might potentially occur in sigB genes among L. innocua sequences, as they were detected by both RDP4 and PHI test. However, since recombination events detected by only half of the nine methods in RDP4 were considered significant (25, 26), and the results obtained by three programs, PHI test, Recco, and RDP4, were not consistent, we concluded that only limited HR occurred in sigB of Listeria sensu stricto.

TABLE 2.

Putative recombination events inferred by RDP4 and PHI test^a

Taxon	Recombination determined by RDP4			P value for recombination determined by PHI testb
	Recombinant sequence	Putative parental sequence (minor × major donor)	Detection method(s)
Listeria sensu stricto	L. monocytogenes AT 98	L. marthii AT 143 × L. monocytogenes AT 83	Maxchi	0.930
	L. monocytogenes AT 151
	L. innocua AT 141	L. monocytogenes AT 75 × L. innocua AT 139	SiScan
L. seeligeri	L. seeligeri AT 9	L. seeligeri AT 68 × L. seeligeri AT 92	Maxchi	0.980
L. welshimeri				0.750
L. innocua	L. innocua AT 124	L. innocua AT 139 × L. innocua AT 87	Bootscan, SiScan, 3Seq	0.002
	L. innocua AT 33
L. ivanovii				1.000
L. marthii	L. marthii AT 143	L. marthii AT 95 × L. marthii AT 18	Bootscan, Maxchi, LARD	0.440
L. monocytogenes				0.297
Lineage I				0.045
Lineage II				1.000
Lineage III				0.440
Lineage IV				NA

^aRecco did not detect any significant recombination events.

^bP value of <0.01 is marked in boldface type; NA (not available) indicates that too few informative characters were present in a given taxon.

Positive selection and codon usage bias.

An initial assessment showed that dN/dS ratios were <1 across Listeria spp. and within species and lineages (Table 1), indicating no evidence for positive selection across all amino acid sites of sigB. However, even with an overall dN/dS ratio of <1, specific amino acid sites may still be under positive selection. Subsequent ML approaches that used the likelihood ratio test (LRT) to compare (i) M1a (the nearly neutral model) versus M2a (the positive selection model) and (ii) M7 (the beta model) versus M8 (the beta and ω models) showed no significant differences (Table 3). We therefore cannot reject the null hypothesis of no positive selection in sigB, which further supports that sigB did not evolve under positive selection in Listeria sensu stricto.

TABLE 3.

Results of positive selection analysis using PAML

Taxon	Ln L M1aa	Ln L M2ab	Significance value for comparison of Ln L M1a and Ln L M2ac	Ln L M7a	Ln L M8b	Significance value for comparison of Ln L M7 and Ln L M8c
Listeria sensu stricto	−3,692.729	−3,692.727	0.998	−3,685.556	−3,686.692	0.321
L. seeligeri	−1,198.649	−1,198.649	1.000	−1,198.655	−1,198.655	1.000
L. welshimeri	−1,138.146	−1,138.146	1.000	−1,138.166	−1,138.166	1.000
L. innocua	−1,354.596	−1,353.314	0.277	−1,354.272	−1,352.991	0.278
L. ivanovii	−956.510	−956.510	1.000	−956.511	−956.511	1.000
L. marthii	−958.010	−958.009	0.999	−958.009	−958.010	0.999
L. monocytogenes	−1,521.006	−1,521.006	0.999	−1,521.934	−1,521.047	0.412
Lineage I	−947.277	−947.276	1.000	−947.276	−947.277	1.000
Lineage II	−1,046.284	−1,046.283	1.000	−1,046.289	−1,045.390	0.407
Lineage III	−1,057.886	−1,058.357	0.625	−1,057.889	−1,057.889	1.000
Lineage IV	−872.673	−872.673	1.000	−872.673	−872.673	1.000

^aLn L M1a and Ln L M7, log-normal likelihood scores for the null hypothesis that the data set evolved following a neutral model.

^bLn L M2a and Ln L M8, log-normal likelihood scores for the alternative hypothesis that the data set evolved under positive selection.

^cThe test statistic was calculated as 2[(−Ln L M1a) − (−Ln L M2a)] and 2[(−Ln L M7) − (−Ln L M8)].

The N̂_c statistic (27, 28) was calculated for each AT to measure synonymous codon usage bias in sigB. N̂_c can take values from 20, in the case of extreme bias, where one codon is exclusively used for each amino acid, to 61, when the use of alternative synonymous codons is equally likely. The lowest N̂_c value, 43.64, was detected for L. innocua AT 10; the values for most other ATs (151/164) were >45 (see Table S1 in the supplemental material); these data suggest limited codon usage bias in sigB.

Models of sequence evolution.

The best DNA substitution models for sigB in Listeria sensu stricto, each Listeria sensu stricto species, and each L. monocytogenes lineage were determined by jModelTest (Table 4). sigB evolution in L. seeligeri, L. monocytogenes, and L. monocytogenes lineage III followed the Tamura-Nei substitution model, which includes variable base and transition frequencies but equal transversion frequencies (29). sigB evolution in other species, in other L. monocytogenes lineages, and across Listeria sensu stricto strains followed the Hasegawa-Kishino-Yano (HKY) substitution model, which allows for variable base frequencies and different substitution rates for transitions and transversions (30).

TABLE 4.

Molecular evolution parameters for sigB

Taxon	DNA substitution modela	ti/tv ratiob	Alphac	P-invd
Listeria sensu stricto	HKY+I+G	3.37	0.90	0.61
L. seeligeri	TrN+I			0.76
L. welshimeri	HKY+I	4.24		0.90
L. innocua	HKY+I+G	2.59	0.53	0.90
L. ivanovii	HKY	7.91
L. marthii	HKY+I	11.32		0.94
L. monocytogenes	TrN+G		0.22
Lineage I	HKY+I	2.35		0.93
Lineage II	HKY	2.31
Lineage III	TrN+I			0.81
Lineage IV	HKY	NA

^aHKY, Hasegawa-Kishino-Yano model; I, invariable sites; G, gamma distribution; TrN, Tamura-Nei model.

^bRatio of transitions (ti) to transversions (tv). If the ratio could not be unambiguously inferred, the result is reported as NA (not available); this is reported only for the HKY model since the TrN model uses variable transition rates.

^cThe alpha parameter defines the shape of the gamma distribution; this is reported only if the DNA substitution model includes a gamma distribution.

^dProportion of invariable sites; this is reported only if the DNA substitution model includes an invariable-site parameter.

Both PAUP* and phylogenetic analysis by maximum likelihood (PAML) were used to test the molecular clock hypothesis and yielded consistent results (Table 5). The null hypothesis that sequences evolved according to a molecular clock was rejected (P < 0.05) for sigB genes across Listeria sensu stricto strains as well as within L. seeligeri, L. innocua, L. ivanovii, L. marthii, L. monocytogenes, and L. monocytogenes lineages I, II, and III. On the other hand, in L. welshimeri and L. monocytogenes lineage IV, sigB was found to evolve following a molecular clock (P > 0.05). This indicates that the evolutionary rate of sigB varied in Listeria sensu stricto strains over time and across clades.

TABLE 5.

Test of the molecular clock

Taxon	Ln L clocka	Ln L no clockb	Significancec	Molecular clock conclusiond
Detected by PAUP*
Listeria sensu stricto	−2,911.235	−2,750.182	0.000	Reject
L. seeligeri	−1,345.168	−1,302.799	0.000	Reject
L. welshimeri	−1,232.536	−1,215.159	0.480	Fail to reject
L. innocua	−1,524.065	−1,474.704	0.000	Reject
L. ivanovii	−1,055.664	−1,026.337	0.000	Reject
L. marthii	−1,034.063	−1,025.077	0.001	Reject
L. monocytogenes	−1,837.100	−1,713.537	0.000	Reject
Lineage I	−1,026.237	−1,015.145	0.008	Reject
Lineage II	−1,150.381	−1,108.575	0.000	Reject
Lineage III	−1,154.203	−1,136.741	0.039	Reject
Lineage IV	−915.453	−914.652	0.449	Fail to reject
Detected by PAML
Listeria sensu stricto	−2,699.124	−2,544.663	0.000	Reject
L. seeligeri	−1,331.701	−1,289.192	0.000	Reject
L. welshimeri	−1,230.322	−1,213.313	0.515	Fail to reject
L. innocua	−1,450.986	−1,401.417	0.000	Reject
L. ivanovii	−1,042.710	−1,014.496	0.000	Reject
L. marthii	−1,031.025	−1,022.056	0.001	Reject
L. monocytogenes	−1,791.820	−1,666.501	0.000	Reject
Lineage I	−1,016.090	−1,005.958	0.016	Reject
Lineage II	−1,150.315	−1,108.882	0.000	Reject
Lineage III	−1,141.821	−1,123.196	0.022	Reject
Lineage IV	−914.253	−913.446	0.446	Fail to reject

^aLn L clock, log-normal likelihood score for the null hypotheses that the data set evolved following a molecular clock.

^bLn L clock, log-normal likelihood score for the alternative hypotheses that the data set did not evolve following a molecular clock.

^cThe test statistic for the molecular clock was calculated as 2[(−Ln L clock) − (−Ln L no clock)].

^dConclusion to reject or fail to reject the null hypothesis.

Phylogenetic analysis.

Maximum likelihood phylogenetic trees were constructed based on (i) sequence alignments of all unique sigB ATs identified to date in Listeria sensu stricto (n = 164) and (ii) the core genome SNPs from 21 publically available whole-genome sequences of Listeria sensu stricto spp. The results showed a robust separation of clades representing L. seeligeri, L. welshimeri, L. innocua, L. ivanovii, and L. marthii in both the sigB and WGS phylogenetic trees (Fig. 1a and 2). However, L. monocytogenes lineage IV isolates clustered with L. innocua in the phylogenetic tree based on sigB, with a high bootstrap value of 85% (Fig. 1a), while in a phylogram using core genome SNPs, they clustered with other L. monocytogenes lineages (Fig. 2). Hence, we manually placed L. monocytogenes lineage IV isolates with the other L. monocytogenes lineages in a phylogenetic tree file (see Fig. S1 in the supplemental material) and compared the maximum likelihood values obtained for the topology of this rearranged tree and those obtained for the topology of the optimal tree (Fig. S2) by running the Kishino-Hasegawa (KH) test, the Shimodaira-Hasegawa (SH) test, and the approximately unbiased (AU) test. The results of all three tests showed significantly different likelihood scores for the optimal tree compared to the rearranged tree (P < 0.001) (Table S2), confirming that the optimal tree represents the true history of the evolution of sigB. In addition, to achieve higher resolution, another phylogenetic tree using sigB was constructed for only L. monocytogenes lineage I, II, and III isolates (Fig. 1b). This phylogeny showed that each lineage formed a distinct clade, except for L. monocytogenes AT 86, which clustered with lineage III even though it represents a lineage II isolate.

FIG 1.

Phylogenetic trees inferred by the maximum likelihood method using sigB genes of 164 Listeria sensu stricto allelic types (a) and L. monocytogenes lineage I, II, and III isolates (b). Both trees were rooted by midpoint. For the phylogeny shown in panel a, the HKY substitution model with invariable sites and a gamma distribution (HKY+I+G substitution model) was used for constructing the tree with 1,000 bootstrap repetitions. Only bootstrap values of >70% are presented on the tree. L. monocytogenes is indicated in red, L. marthii is in brown, L. innocua is in green, L. welshimeri is in yellow, L. seeligeri is in blue, and L. ivanovii is in purple. For panel b, the GTR+G substitution model was used for constructing the tree with 1,000 bootstrap repetitions; only bootstrap values of >70% are presented in this tree. L. monocytogenes lineage I is indicated by triangles, lineage II is indicated by squares, and lineage III is indicated by circles.

FIG 2.

Phylogenetic tree inferred by the maximum likelihood method using the core genome SNPs of 21 Listeria sensu stricto isolates. kSNP2 with kmer size of 19 was used to identify core genome SNPs that were used for constructing the tree with the GTR+G substitution model and 1,000 bootstrap repetitions. The tree is rooted by the midpoint. Only bootstrap values of >70% are presented on the tree. L. monocytogenes is indicated in red, L. marthii is in brown, L. innocua is in green, L. welshimeri is in yellow, L. seeligeri is in blue, and L. ivanovii is in purple.

DISCUSSION

Subtype classification and species identification of isolates that are initially phenotypically classified into the genus Listeria represent important needs, particularly since monitoring of food processing plant environments often involves testing for Listeria spp. As traditional characterization methods (e.g., carbohydrate utilization patterns, which have been formatted into test kits such as API [31]) may not always be feasible or reliable and may be cost-prohibitive, there is a need for alternative rapid and reliable classification methods for Listeria species isolates, which are supported by sufficiently large curated reference databases. One such approach involves the sequencing of an internal fragment of the gene encoding σ^B (sigB). This approach has been widely used not only to classify Listeria isolates to the species and lineage levels but also to provide initial affordable subtype classification (9, 11, 32–34). In this study, we analyzed an initial reference set of >4,200 sigB sequences to more formally assess the strengths and limitations of the standardized Listeria sigB sequence-based typing approach using an evolutionary framework. Using multiple evolutionary and phylogenetic analysis approaches, we showed that (i) sigB of Listeria sensu stricto is a relatively stable gene evolving without a significant contribution of frequent HR and positive selection, (ii) the nearly neutral theory best explains the evolution of sigB in Listeria sensu stricto, and (iii) sigB sequence data allow discriminatory initial subtyping and reliable species and lineage classifications of all Listeria spp., even though they do not provide reliable reconstruction of the phylogenetic relationship of L. monocytogenes lineage IV with other Listeria sensu stricto spp.

sigB in Listeria sensu stricto is relatively stable and influenced by neither frequent HR nor positive selection.

Clonal bacteria can retain genetic variability via HR and positive selection (26), the effects of which have been detected in specific genes of many bacteria (35–39), including the genus Listeria (24, 40). Nevertheless, restricted by the expression and function of a gene, mechanistic barriers to HGT, such as surface exclusion and restriction and environmental and temporospatial limitations, HR and positive selection do not always occur. Our results indicate that these two evolutionary processes did not play a significant role in the evolution of sigB of Listeria sensu stricto, which is consistent with previous findings (24). These results indicate that sigB is a relatively stable gene, unlike fast-evolving genes, such as virulence genes, in which the occurrence of both positive selection and frequent HR was previously reported to significantly contribute to the evolution of Listeria (11, 41, 42). Previous studies indicated that informational genes of the central cellular machinery, such as those involved in DNA replication, transcription, translation, and related processes, are seldom horizontally transferred (43). The low frequency of HR in sigB, an informational gene in the core genome of Listeria sensu stricto (44, 45), may stem from the limited selective advantages that could be derived from acquiring sigB gene fragments from other species. The fact that sigB is part of complex and large sets of functional networks may also make horizontal transfer of sigB or sigB fragments less probable (46).

The absence of positive selection could be explained by the absence of frequent HR, since positive selection has been observed to be associated with recombination in many studies (e.g., see references 40 and 47). Recombination allows advantageous mutations present in different lineages to be combined and fixed in the same population. This function weakens the competition among advantageous mutations for fixation, speeds up their fixation, and thus prevents clonal interference (40). Recombination can also break down the linkage between advantageous and disadvantageous mutations and decrease the frequency or fixation in the population of linked disadvantageous mutations, thus reducing the genetic load (40). Thus, positive selection may be inefficient in sigB due to clonal interference and/or genetic load influenced by the absence of HR.

The nearly neutral theory best explains the evolution of sigB in Listeria sensu stricto.

The absence of evidence for positive selection suggests that sigB in Listeria sensu stricto evolved neutrally or under negative selection, which is a corollary of the neutral theory (23). The neutral theory, which predicts a constant rate of molecular evolution for a given mutation rate and proportion of neutral sites, is supported by the molecular clock hypothesis (22). In other words, under the assumption that the evolution of sigB in Listeria sensu stricto is driven largely by random chance, sigB should follow the molecular clock unless its mutation rate and proportion of neutral sites are frequently changing. The violation of a molecular clock indicated that the evolutionary rate of sigB in Listeria sensu stricto varied over evolutionary time and across clades, which does not meet the neutral-theory assumptions. The evolution of Listeria sensu stricto sigB can therefore be more reasonably explained by the nearly neutral theory (48, 49).

The nearly neutral theory acknowledges (i) deleterious mutations, the fate of which is determined by negative selection; (ii) neutral mutations, for which drift is the driving force; and (iii) nearly neutral mutations, which are governed by both selection and drift. According to the nearly neutral theory, the rate of molecular evolution can vary with changes in the mutation rate and the balance between selection and drift, which is influenced by the effective population size (22). For example, a marked reduction of the size of a population can be accompanied by a burst of fixation of nearly neutral alleles, resulting in an imbalance between selection and drift and finally leading to a relaxation of the molecular clock (22). The violation of a molecular clock in sigB of L. monocytogenes lineage II, an environmentally adapted group, may be caused by a recent population bottleneck followed by population expansion, as indicated by a significant negative Tajima D value. Consistent with our results, one previous study also found support for a recent population bottleneck followed by expansion in lineage II (3). L. monocytogenes lineage I might also have experienced a population bottleneck, based on data from a previous study (3), despite the lack of support by Tajima's D values calculated in the present study. The violation of a molecular clock in the sigB genes of L. monocytogenes and L. ivanovii detected in this study may result from their subspecies and well-supported subspecies-like clades (as detected by MLST) (3), which could exhibit different rates of molecular evolution. The reasons why the sigB genes in L. monocytogenes lineage III, L. seeligeri, L. innocua, and L. marthii did not follow a molecular clock are uncertain and warrant further investigation. Possible explanations include that the mutation rate of the sigB genes in these species might have changed over time due to genetic and environmental factors (50) or that a considerable proportion of their populations experienced dormancy during a certain time, which could result in a dramatic variation in evolutionary rates. With relatively low genetic diversity, and as the only species in Listeria sensu stricto whose sigB gene evolved according to a molecular clock, L. welshimeri likely has a unique natural history. It was previously shown that L. welshimeri appears to represent the most ancient clade of Listeria sensu stricto, arising from an ancestor without a prfA gene cluster (24). As our data showed that the evolutionary rate of sigB varies across species and lineages and over evolutionary time, it would not be appropriate to use sigB data to estimate evolutionary time scales in Listeria.

sigB sequence data allow discriminatory initial subtyping and reliable species and lineage classifications of all Listeria spp., even though they do not provide a reliable reconstruction of the phylogeny of L. monocytogenes lineage IV.

While 16S rRNA gene sequencing is widely used for the initial classification and characterization of bacteria, its limited discriminatory power represents a challenge, particularly for species classification within highly related genera such as Listeria (51). The absence of frequent HR, significant positive selection, and codon usage bias in sigB of Listeria sensu stricto and its high discriminatory power indicate that sigB is an appropriate target gene for initial characterization and subtyping of Listeria isolates. Our data showed clustering, based on partial sigB sequences, into clades that are consistent with species and L. monocytogenes lineages. This shows that sigB allelic typing can be used for the accurate classification of isolates into Listeria species and L. monocytogenes lineages. Reliable lineage classification of L. monocytogenes isolates is possible despite the fact that L. monocytogenes lineage IV phylogenetically clusters with L. innocua in a sigB genealogy, as all L. monocytogenes lineage IV isolates fall into a well-supported monophyletic clade that is assigned a lineage designation based on clustering with reference isolates. However, our analyses suggest that sigB sequence data do not appear to be appropriate for phylogenetic inference, which is more reliably achieved by using core genome SNP-based phylogenies.

The incongruence of the L. monocytogenes lineage IV placement in the sigB-based and SNP-based phylogenetic trees might result from tree topology bias, incomplete lineage sorting, or HR. The possibility of tree topology bias was rejected according to the results of the three-tree topology maximum likelihood tests, the KH, SH, and AU tests. Given the relatively short branch lengths leading to each L. monocytogenes lineage (Fig. 2), incomplete lineage sorting could possibly be a source of the sigB-based tree incongruence with the SNP-based tree (52, 53). Limited evidence for HR was found in the recombination analyses, even when only isolates of L. monocytogenes lineage IV and L. innocua were analyzed (see Table S3 in the supplemental material). However, it is possible that sigB was horizontally transferred to the ancestor of L. monocytogenes lineage IV, but the region of recombination spans the whole length of sigB and is therefore not detected when the allele is analyzed without flanking regions. Considering the low prevalence of L. monocytogenes lineage IV isolates in the whole population of Listeria sensu stricto and the low frequency of occurrence of HR in sigB, our results still support the validity of species classification and subtyping of Listeria based on sigB.

Conclusion.

Overall, our results indicate that sigB in Listeria sensu stricto is a relatively stably evolving gene without significant contributions by frequent homologous recombination, positive selection, or codon usage bias. The molecular evolution of sigB in Listeria sensu stricto was likely driven by nearly neutral mechanisms with variation in evolutionary rates across clades and over evolutionary time. Based on its genetic stability, high discriminatory power, and accuracy in species and lineage classifications, sigB allelic typing presents a valuable tool for Listeria sensu stricto species classification and subtyping. In addition to the advantages of an easily accessible reference collection of >4,200 sigB sequences (available at https://github.com/CCCeliaLiao/sigBdata), sigB allelic typing also has all the well-documented advantages of single-gene sequencing approaches, including a high level of inter- and intralaboratory reproducibility, cost-effectiveness, time-efficiency, and a high level of data portability (14). With the rapidly decreasing costs and shorter turnaround times associated with WGS and the rapidly growing number of Listeria genomes (2, 54–57), routine WGS will increasingly be used for Listeria subtyping and phylogenetic inference. Nevertheless, single-locus subtyping methods, such as sigB allelic typing, will remain highly useful and recommended for time- and cost-efficient primary classification and subtyping, including in resource-constrained situations and in locations that will only slowly gain access to affordable WGS resources.

MATERIALS AND METHODS

Isolate selection and sigB allelic typing.

Data for a total of 4,280 Listeria isolates available at the Food Microbe Tracker website (http://www.foodmicrobetracker.com/) (58) in November 2015 were used in this study. These isolates included 667 L. seeligeri, 565 L. welshimeri, 1,226 L. innocua, 12 L. ivanovii, 14 L. marthii, and 1,796 L. monocytogenes isolates (1,032 isolates for lineage I, 692 isolates for lineage II, 59 isolates for lineage III, and 13 isolates for lineage IV). All of these isolates had previously been characterized by PCR amplification and sequencing of the partial sigB gene. The alignment of the sigB AT sequences of the 4,280 isolates was deposited in GitHub (https://github.com/CCCeliaLiao/sigBdata). PCR amplification and DNA sequencing for these isolates were performed as detailed previously by Nightingale et al. (11) (see https://github.com/CCCeliaLiao/sigBdata/blob/master/sigB_PCR_Protocol.pdf for the sigB PCR protocol). sigB ATs, defined as a unique combination of polymorphisms (10), had been assigned to these 4,280 isolates (see Data Set S1 in the supplemental material). Among these 4,280 isolates, 164 isolates representing the 164 unique sigB ATs were selected for the evolutionary and phylogenetic analyses conducted here (Data Set S1). These 164 isolates represent 23 L. seeligeri ATs, 37 L. welshimeri ATs, 41 L. innocua ATs, 6 L. ivanovii ATs, 6 L. marthii ATs, and 51 L. monocytogenes ATs, including 11 ATs for lineage I, 12 ATs for lineage II, 24 ATs for lineage III, and 4 ATs for lineage IV. Three ATs (AT103, AT154, and AT165) had deletions, which were confirmed by repeated Sanger sequencing. The alignment of the sigB AT sequences of these 164 representative isolates is available (see https://github.com/CCCeliaLiao/sigBdata).

Whole-genome sequences.

A total of 21 whole-genome sequences of Listeria were downloaded from the NCBI database and used for phylogenetic analysis in this study. Nineteen of these genomes were closed, and two were drafts of high quality with a high N₅₀ value (WGS accession numbers are listed in Table S4 in the supplemental material). Genomes were selected to represent each Listeria species and L. monocytogenes lineage (L. monocytogenes lineage I strains SLCC2378, WSLC1042, and ATCC 19117; lineage II strains SLCC2479, FSL R2-0561, and WSLC1001; lineage III strains HCC23, L99, and SLCC2376; lineage IV strains FSL J1-0208; L. innocua strains Clip11262, FSL S4-0378, and FSL J1-0023; L. ivanovii strains WSLC3009, PAM 55, and FSL F6-0596; L. seeligeri strains FSL N1-0067, FSL S4-0171, and SLCC3954; L. marthii strain FSL S4-0120; and L. welshimeri strain SLCC5334).

Descriptive analyses.

The number of polymorphic sites and mutations, G+C content, nucleotide diversity per site (π), average pairwise nucleotide differences per sequence (k), Tajima's D, number of synonymous substitutions (S) and nonsynonymous substitutions (N), and the ratio of the nonsynonymous substitution rate to the synonymous substitution rate (dN/dS ratio) using approximate methods for all Listeria sensu stricto strains, each Listeria sensu stricto species, and each L. monocytogenes lineage were separately calculated by using DnaSP version 5.1 (59). SID was calculated according to methods described previously by Hunter and Gaston (60) to assess the discriminatory power of sigB allelic typing; higher SID values indicate better discriminatory power. All 4,280 isolates mentioned above were included in all descriptive analyses except for the dN/dS ratio calculation. The 164 representative isolates of sigB ATs were used for the dN/dS ratio calculation in DnaSP version 5.1 and in the following analyses.

Detection of homologous recombination.

PHI test, Recco, and RDP were used to detect HR within (i) Listeria sensu stricto, (ii) each Listeria sensu stricto species, and (iii) individual L. monocytogenes lineages. PHI test was conducted in SplitsTree4 V4.14.2 (61). Recco (62) was used for the prediction of recombination events, recombinant sequences, and recombination breakpoints, as it uses a fast and sensitive algorithm (25). Parameter α, which represents the cost of mutation relative to recombination, yields additional information as to which sequence might be recombinant. The third program used for detecting HR was RDP4, which incorporates nine different HR detection methods (i.e., RDP, Bootscan, GENECONV, Maxchi, CHIMAERA, SiScan, PhylPro, LARD, and 3Seq) (63). Consistent with a previously set criterion (25, 26), only recombination events detected by half of the methods used in RDP4 were considered to be significant.

Positive selection and codon usage bias.

Positive selection was assessed by using the LRT by comparing the M1a model (nearly neutral) with the M2a model (positive selection in a fraction of the sites [ω > 1]) and by comparing the M7 model (beta distribution) with M8 (beta distribution with positive selection in a fraction of the sites [ω > 1]), implemented in the PAML program version 4.8 (64). Parameters used in each model were described previously by Matute et al. (18) and Nielsen and Yang (19). The LRT statistic was calculated as 2 × ln(L₀/L₁), where L₀ is the likelihood estimate for the null model and L₁ is the likelihood estimate for the alternative model with more free parameters. The degree of freedom (df), which equals the difference in the numbers of free parameters between the null and alternative models, was set to 2. Statistical significance was determined by approximating the test statistic to a χ² distribution (11). L. monocytogenes AT 105 was excluded from positive selection analysis due to the presence of a premature stop codon within the sequence. Codon usage bias in the sigB gene of each AT was measured by Chips in EMBOSS using the N̂_c statistic (27, 28).

Evolutionary model and molecular clock hypothesis test.

The best evolutionary substitution model was identified based on the Bayesian information criterion determined with jModelTest version 2.1.10 (65). In order to separately test the null hypothesis that sigB evolved according to a molecular clock (i) among all Listeria sensu stricto strains, (ii) in each Listeria sensu stricto species, and (iii) in each of the four L. monocytogenes lineages, the no-clock model and the clock model were run in both PAUP* 4.0a149 (66) and PAML 4.8 (64) and compared by using the LRT. Due to intensive computational requirements, a randomly selected subset of ATs (n = 30) was used to test the molecular clock hypothesis for all Listeria sensu stricto strains. The phylogenetic tree of each bacterial group included in the molecular clock hypothesis test was generated by using MEGA software version 7.0.16 (67). In PAML 4.8, unrooted trees were used for phylogenies constructed without the clock model (clock = 0), while trees rooted by the midpoint were used for phylogenies constructed with the clock model (clock = 1). The criterion for rejection of the null hypothesis of the data set according to a molecular clock was set to a P value of <0.05.

Phylogenetic analysis and tree topology tests.

For the data set comprising 164 sigB ATs, a ML analysis inferring the phylogeny was performed with MEGA software version 7.0.16 using the best substitution model identified by jModelTest and 1,000 bootstrap replicates. A rearranged sigB-based phylogenetic tree was generated by manually placing L. monocytogenes lineage IV next to the other L. monocytogenes lineages in the optimal ML tree determined in the previous step. Three tree topology tests, (i) the two-tailed KH test (68) using normal approximation, (ii) the one-tailed SH test (69) using resampling estimated log likelihoods (RELL) bootstraps, and (iii) the Shimodaira AU test (70), were performed in PAUP* 4.0a149 (66) to compare the differences between the likelihood scores of the optimal tree and the rearranged tree. Significance was determined by permutation with 10,000 bootstrap replicates and an automatically generated starting seed. The null hypothesis of no difference between trees is rejected when P values for the test are <0.05. In order to achieve a higher resolution, a separate ML tree was constructed for the clade comprising L. monocytogenes lineage I, II, and III isolates using the general time-reversible (GTR) substitution model with a gamma distribution of substitution sites (GTR+G substitution model) and 1,000 bootstrap repetitions in RaxML version 8.2 (71). For the WGS data set, core SNPs were identified by kSNP2, using a kmer size of 19, determined by KChooser (72), and used to infer a ML phylogeny of Listeria using the GTR+G substitution model with ascertainment bias correction in RaxML version 8.2 (71). ML bootstrap values were calculated based on 1,000 bootstrap repetitions.

Access to isolate data.

All sigB sequence data are accessible on GitHub (https://github.com/CCCeliaLiao/sigBdata). A standard operating procedure (SOP) detailing the procedure for assignment of sigB allelic types is available on GitHub (https://github.com/CCCeliaLiao/sigBdata/blob/master/sigB_ATassignment_SOP.pdf).