Rational Design of DNA Sequence-Based Strategies for Subtyping Listeria monocytogenes

Abstract

The ability to differentiate bacteria beyond the species level is essential for identifying and tracking infectious disease outbreaks and to improve our knowledge of the population genetics, epidemiology, and ecology of bacterial pathogens. Commonly used subtyping methods, such as serotyping, phage typing, ribotyping, and pulsed-field gel electrophoresis, can yield ambiguous results that are difficult to standardize and share among laboratories. DNA sequence-based subtyping strategies can reduce interpretation ambiguity. We report the development of a rational approach for designing sequence-based subtyping methods. Listeria monocytogenes was selected as the model organism for testing the efficacy of this approach. Two housekeeping genes (recA and prs), one stress response gene (sigB), two virulence genes (actA and inlA), and two intergenic regions (hly-mpl and plcA-hly) were sequenced for 15 L. monocytogenes isolates. Isolates were chosen from a representative collection of more than 1,000 L. monocytogenes isolates to reflect the genetic diversity of this species. DNA sequences were aligned, and sliding window analyses were performed for each gene to define 600-bp-long regions that were (i) most polymorphic (using ProSeq) or (ii) most discriminatory (using a new algorithm implemented in WINDOWMIN). Complete gene sequences for actA (1,929 bp) and inlA (2,235 bp) provided the highest discrimination (identifying 15 and 14 allelic types, respectively). WINDOWMIN allowed identification of 600-bp regions within these genes that provided similar discriminatory power (yielding 15 and 13 allelic types, respectively). The most discriminatory 600-bp fragments identified in the housekeeping and stress response genes differentiated the isolates into 8 to 10 subtypes; intergenic region sequences yielded 8 and 12 allelic types based on 335- and 242-bp sequences for hly-mpl and plcA-hly, respectively. Regions identified as most polymorphic were not necessarily most discriminatory; therefore, application of the WINDOWMIN algorithm provided a powerful tool for determining the best target regions for DNA sequence-based subtyping. Our specific results also show that inclusion of virulence gene target sequences in a DNA sequence-based subtyping scheme for L. monocytogenes is necessary to achieve maximum subtype differentiation.

Listeria monocytogenes is a food-borne pathogen that causes approximately 2,500 cases of human illness and 500 deaths annually in the United States (21). Bacterial subtyping methods have improved our ability to detect and track human listeriosis outbreaks and have also provided tools for tracking sources of L. monocytogenes contamination throughout food systems. Application of subtyping methods also provides insight into the population genetics, epidemiology, ecology, and evolution of L. monocytogenes. A variety of conventional, phenotypic, and DNA-based subtyping methods have been described for differentiation of L. monocytogenes beyond the species and subspecies levels (14). While phenotype-based methods have been used for many years to subtype L. monocytogenes and other food-borne pathogens, DNA-based subtyping methods are generally more discriminatory and amenable to interlaboratory standardization and are thus increasingly replacing phenotype-based subtyping methods (29).

Commonly used phenotype-based subtyping methods for L. monocytogenes and other food-borne pathogens include serotyping, phage typing, and multilocus enzyme electrophoresis (MLEE) (14, 24). DNA-based subtyping methods include PCR-based approaches (e.g., random amplified polymorphic DNA and amplified fragment length polymorphism), ribotyping, and pulsed-field gel electrophoresis (2, 13, 16). These DNA-based methods define bacterial subtypes by using either PCR amplification or restriction digestion of bacterial DNA to generate DNA fragment banding patterns. While many of these methods have proven effective for differentiating L. monocytogenes subtypes, DNA fragment size-based subtyping methods have significant drawbacks. For example, despite the existence of software packages for data normalization and analyses (12), these subtyping methods are often difficult to standardize. As a consequence, the ease of exchanging and comparing subtype data among laboratories can be severely limited. While DNA fragment size-based subtyping methods have been used for cluster analyses, they generally do not provide information amenable to the inference of primary genetic characteristics (i.e., nucleotide sequences) for evolutionary analyses. As long-term studies on the epidemiology, ecology, and evolution of bacterial pathogens require subtyping data that can be used to infer and quantify the genetic relatedness of isolates, DNA fragment size-based subtyping methods have limited utility for these applications.

DNA sequencing-based methods are being developed and increasingly used for subtyping and characterizing bacterial isolates. In these methods, complete or partial nucleotide sequences are determined for one or more bacterial genes or chromosomal regions, thus providing unambiguous and discrete data. Sequencing can target a single gene (single locus approach) or multiple genes. The advantages of sequencing methods over DNA fragment size-based typing methods include their ability to generate unambiguous data that are portable through web-based databases and that can be used for phylogenetic analyses (3, 9). While a variety of DNA sequence-based subtyping strategies targeting virulence genes, housekeeping genes, or other chromosomal genes and regions are feasible, multilocus sequence typing (MLST), which is an extension of MLEE, represents a widely used strategy (26).

MLEE differentiates bacterial strains by detecting variations in the patterns of the electrophoretic mobilities of various constitutive enzymes. Cell extracts containing soluble enzymes are separated by size in nondenaturing starch gels, and enzyme activities are determined in the gels through application of color-generating substrates (14). While MLEE has been used to study the population genetics of many bacterial pathogens, including L. monocytogenes (23), this method is difficult to standardize among laboratories.

MLST directly determines the allelic variation of multiple housekeeping genes by using DNA sequencing instead of indirectly characterizing these alleles via measurement of the electrophoretic mobilities of the gene products through MLEE (7). MLST approaches have been developed for several organisms, including group A streptococci, Staphylococcus aureus, Neisseria meningitidis, and Campylobacter jejuni (4-6). MLST traditionally targets multiple loci that have slowly diversified from each other through accumulation of neutral or near neutral changes, thus providing reliable differentiation without the potentially confounding effects of positive selection that may particularly occur in certain categories of genes, such as bacterial virulence or surface genes (20, 26). On the other hand, direct sequencing of virulence genes (9) or intergenic regions may provide more-sensitive discrimination for cluster analyses and short-term epidemiological questions.

While some studies have explored the suitability of sequencing single genes to differentiate L. monocytogenes strains (10, 28), the discriminatory power of DNA sequencing strategies that target multiple distinct genes or regions has not yet been reported. Thus, we selected a well-characterized set of L. monocytogenes isolates to determine the suitability of DNA sequencing of housekeeping and stress response genes (sigB, prs, and recA), two virulence genes (actA and inlA), and two intergenic regions (plcA-hly and hly-mpl) to differentiate L. monocytogenes subtypes. The complete sequence information for these genes and regions was also used to define the most-discriminatory 600-bp fragments within these genes that could be used for rapid sequencing-based subtyping. This study also provides a general outline for a rational approach to the selection of target genes for DNA sequence-based subtyping of bacterial pathogens.

MATERIALS AND METHODS

Bacterial strains and lysate preparation.

A test set of 15 isolates, which includes 5 isolates from each of the three L. monocytogenes genetic lineages (30) (Table 1), was selected from a collection of more than 1,000 isolates from foods, environmental samples, and human and animal listeriosis cases. Within lineages I and II, isolates were selected to represent sets of two or three closely related strains to allow us to best determine the ability of different target genes to differentiate closely related genotypes. Lineage III isolates were chosen to represent one set of closely related strains and three diverse strains, as strains classified into lineage III appear more genetically diverse than strains in the other two lineages. Bacteria were grown on brain heart infusion (Difco, Sparks, Md.) agar and in brain heart infusion broth at 37°C. All stock cultures were stored at −80°C in 15% glycerol. Bacterial cell lysates for use in PCRs were prepared with lysozyme and proteinase K as previously described (11).

TABLE 1.

L. monocytogenes isolates used in this study

Isolate no.	Previous isolate no.	Sourcea	Origin	Serotype	hly typec	actA typec	EcoRI ribotypec	Lineagec
FSL J1-038	HO413	CDC	Human	1/2b	1	4	DUP-1042	I
FSL J2-039	NAb	CUVDL	Turkey	4b	1	3	DUP-1042	I
FSL J2-045	NA	CUVDL	Sheep	4b	1	4	DUP-1042	I
FSL J1-051	HO436	CDC	Human	4b	1	4	DUP-1052	I
FSL J2-064	NA	CUVDL	Cow	1/2b	1	3	DUP-1052	I
FSL J2-003	NA	CUVDL	Cow	1/2a	2	3	DUP-1030	II
FSL J2-017	NA	CUVDL	Cow	1/2a	2	4	DUP-1030	II
FSL J1-022	DD942, NCTC 4885	Qualicon, Inc.	Human	1/2c	2	4	DUP-1030	II
FSL J1-047	HO478	CDC	Human	1/2c	2	4	DUP-1039	II
FSL C1-117	2007202	CTDOH	Human	1/2a	2	4	DUP-1039	II
FSL J2-068	NA	CUVDL	Horse	4c	1b	4	DUP-1059	III
FSL X1-002	L99	S. Notermans	Food	4a	4a	3	DUP-1059	III
FSL W1-110	DD 3823	Qualicon, Inc.	NA	4c	4a	3	DUP-1055	III
FSL W1-111	DD 6821	Qualicon, Inc.	NA	4c	4b	3	DUP-1032	III
FSL J1-158	NA	CUFSL	Goat	4b	4b	3	DUP-10142	III

^aCDC, Centers for Disease Control and Prevention; CUVDL, Cornell University Veterinary Diagnostic Laboratory; CTDOH, Connecticut Department of Health; CUFSL, Cornell University Food Safety Laboratory.

^bNA, information not available.

^cDeterminations of hly types, actA types, EcoRI ribotypes, and lineages were performed as described in references 16 and 30.

Genes selected for sequence typing.

Two housekeeping genes (prs and recA), one stress response gene (sigB), two virulence genes (actA and inlA), and two intergenic regions in the prfA virulence gene cluster were sequenced (Table 2). The relative location of these genes on the L. monocytogenes chromosome is also shown in Table 2. The intergenic regions plcA-hly and hly-mpl were chosen because they constitute the largest intergenic regions found within the prfA virulence gene island.

TABLE 2.

Relevant characteristics of target genes and regions used for DNA sequencing

Gene or region	Functional property	Location on EGD chromosome (kb)a
actA	Virulence gene (protein involved in actin tail formation)	209.5
inlA	Virulence gene (internalin, involved in host cell invasion)	454.7
prs	Phosphoribosyl synthetase	202.6
sigB	Stress responsive alternative sigma factor B	930.7
recA	Recombinase A	1,425.5
plcA-hly	Intergenic region	206
hly-mpl	Intergenic region	208

^aLocations are based on the L. monocytogenes EGD genome (http://genolist.pasteur.fr/ListiList/index.html).

PCR amplification conditions.

All PCR amplifications were performed with Thermus aquaticus DNA polymerase (Perkin Elmer-Applied Biosystems, Foster City, Calif.), MgCl₂ at a final concentration of 1.5 mM, 1× PCR buffer, and deoxynucleoside triphosphates at a final concentration of 50 μM. PCR primers and conditions are summarized in Tables 3 and 4, respectively. PCR conditions were not optimized to achieve a single set of PCR conditions for all amplifications.

TABLE 3.

PCR primers

Primer name	Primer target and direction	Sequence (5′→3′)
LmsigB-15	sigB, forward	AATATATTAATGAAAAGCAGGTGGAG
LmsigB-16	sigB, reverse	ATAAATTATTTGATTCAACTGCCTT
Prs-F	prs, forward	GCGCGAAACAAGACAACAAAC
LMVI-prfA	prs, reverse	GC(T/A)GTTAG(C/T)AGAATT(C/A)TTTC
ActA-F176	actA, forward	TAGCGTATCACGAGGAGG
ActA-R2148	actA, reverse	TTTTGAATTTCATATCATTCACC
ActA-CF	actA, forward, lineage III	TTTATGCGTGCGATGATGGTAGTT
ActA-DR	actA, reverse, lineage III	CTTTTTCGTCTTCTGCACTTTTAG
2plcB-ecoR	actA, reverse, FSL W1-110 and FSL X1-002	GGAATTCTTTATGTGGTAATTTGCTGTC
recA-F23	recA, forward	GTTCGCTTTTTGCTTGCTTCTC
recA-R1368	recA, reverse	AAAAACTTCCCCCTTGTCTCC
inlAF1	inlA, forward	CAGGCAGCTACAATTACACA
inlAR1	inlA, reverse	ATATAGTCCGAAAACCACATCT
hlympl-F	hly-mpl, forward	TCTCCATCTGGGGCACTACAC
hlympl-R	hly-mpl, reverse	TTACCATGATGATACAAATA
plcAhly-F	plcA-hly forward	CCGCCTAATGGGAAAGTAAA
plcAhly-R	plcA-hly reverse	AGGCGGAGATGCTGGTG

TABLE 4.

PCR conditions

Gene or regions	Initial denaturation	Amplification (no. of cycles)a	Final hold
actA	94°C, 2 min 30 s	94°C, 3 min; 53°C, 1 min; 72°C, 2 min (40)	72°C, 5 min
inlA	94°C, 2 min 30 s	94°C, 3 min; 66°C, 1 min; 72°C, 2.5 min (35)	72°C, 5 min
sigB	94°C, 5 min	94°C, 30 s; TD 54-44°C, 30 s; 72°C, 1 min (40)	72°C, 7 min
prs	94°C, 5 min	94°C, 30 s; TD 60-50°C, 30 s; 72°C, 1 min (40)	72°C, 7 min
recA	94°C, 2 min 30 s	94°C, 30 s; 50°C, 30 s; 72°C, 1 min (40)	72°C, 7 min
hly-mpl and plcA-hly	94°C, 2 min 30 s	94°C, 30 s; 54°C, 30 s; 72°C, 1 min (40)	72°C, 7 min

^aTD, touchdown PCR. For sigB and prs, a touchdown PCR protocol was used with annealing temperatures decreasing at a rate of 0.5°C/cycle from an annealing temperature T for the first 20 cycles followed by another 20 cycles with the annealing temperature set at T − 10°C.

Purification and quantification of PCR products.

For sequencing, PCR products were purified by using the Qiaquick PCR purification kit (Qiagen, Inc., Valencia, Calif.). For eight L. monocytogenes isolates, nonspecific amplification products and/or primer dimers >60 bp were observed with the sigB PCR primers. For these PCRs, we used the Qiaquick gel extraction kit (Qiagen) to isolate the desired PCR product following electrophoresis in a 1.5% low-melting-point agarose gel and excision of the appropriate band from the ethidium bromide-stained gel. DNA concentrations of purified PCR products were estimated by comparison of amplicon band intensity with that of a DNA marker of known size and concentration (pGEM marker; Promega, Madison, Wis.) with LabImage software (Kapelan, Halle, Germany).

DNA sequencing and analyses.

DNA sequencing was performed at Cornell University's Bioresource Center with an ABI 3700 DNA Sequencer. With the exception of actA and inlA, PCR primers were also used for DNA sequencing. Following initial sequencing with PCR primers, primer walking with primers designed to match strain-specific internal sequences was used to complete the sequencing of actA and inlA. Nucleic acid sequences were proofread and aligned with Seqman (DNAStar) and MegAlign (Lasergene). Cluster analyses were conducted using MEGA, version 2.1 (19), and the unweighted pair group method with arithmetic mean (UPGMA) (number of nucleotide differences) model. The resulting clustering data were assessed together with the alignments to assign allelic types for each region or gene. Two sequences were assigned different allele numbers if the sequences differed by at least 1 nucleotide (nt). Sliding window analyses to determine the number of DNA polymorphisms per 600-bp window for each gene were performed with ProSeq software (http://helios.bto.ed.ac.uk/evolgen/filatov/proseq.html).

A new algorithm was developed and implemented for identifying the most-discriminatory 600-bp region within each gene from a set of aligned sequences (WINDOWMIN). In this algorithm, the most-discriminatory region for a sample of size n is defined as the region that maximizes the number of different DNA sequences in a sample. WINDOWMIN allows the user to define the criteria that will classify two DNA sequences as different. For example, if sequencing errors are common, two DNA sequences may be defined as different from each other if they differ at more than 2 nt positions. If sequencing errors can be ruled out, it may be more reasonable to define two DNA sequences as being different if they differ in at least 1 nt position. For this study, two DNA sequences were defined as different if they differed by at least 1 nt. In WINDOWMIN, the number of different DNA sequences distinguishable by at least k nucleotides in a window starting at position i of the sequences is defined as n_k,i. A window starting at position j in the sequence is most discriminatory if _nk,j = max_i(_nk,i) for a particular value of k. If many possible windows fulfill this criteria, the procedure can be iterated to find the set of windows fulfilling n_k+1,j = max_i(n_k+1,i) among the sequences fulfilling n_k,j = max_i(n_k,i). Window size must also be defined; for the analyses performed here, the window size was set to 600 bp. As a practical approach, we iterated the algorithm 10 times and then chose the window with the most segregating sites among the set of remaining windows. WINDOWMIN can be obtained by download at http://www.foodscience.cornell.edu/wiedmann/programs.html.

Nucleotide sequence accession number.

The DNA sequences determined in this study have been deposited in the GenBank database and given accession numbers AF497139 through AF497243.

RESULTS

Sequencing results.

The complete open reading frames (ORFs) of two virulence genes (actA and inlA), three housekeeping and stress response genes (prs, recA, and sigB), and two intergenic regions were sequenced for 15 L. monocytogenes isolates (Tables 1 and 2). Overall, approximately 7,500 bp of genomic DNA were sequenced for each strain. DNA sequences were deposited into GenBank and are also available through the Cornell PathogenTracker database (http://pathogentracker.net).

Analyses of DNA polymorphisms and allelic types.

DNA sequences for each of the seven selected genes and intergenic sequences were aligned and evaluated. The size of these seven target sequences varied from 242 bp (plcA-hly intergenic sequence) to 2,235 bp (inlA) (Table 5). Based on the total number of polymorphic sites in each sequence (Table 5), actA displayed the highest sequence variability (14.3% of the nucleotides were polymorphic) while prs displayed the lowest level of overall sequence variability (4.9%). Only one target sequence (actA) allowed discrimination of all 15 strains characterized. Two strains (FSL J1-022 and FSL J1-047) differed by only a single nucleotide in actA. The other target genes and regions differentiated between 8 and 14 allelic types for each gene (Table 5). Two sequences were assigned different allelic types (e.g., 1 to 15 for actA) if the sequences differed by at least 1 nt; the allelic types for all strains are summarized in Table 6. Table 5 also shows how many allelic subtypes were defined based on at least a single nucleotide polymorphism and how many subtypes could be defined based on a cutoff of at least a 2-nt difference. For example, for the full-length prs sequence 10 allelic subtypes were defined based on at least a 1-nt difference and 7 allelic subtypes were defined based on at least a 2-nt difference. Thus, three of the prs allelic types were defined based on only a 1-nt difference. In addition, allelic information for the housekeeping genes prs and recA and the stress response gene sigB was used to assign MLST types (12 MLST types, A through L) (Table 6) based on the allelic types for all three genes.

TABLE 5.

Summary of allelic subtypes and polymorphisms in actA, inlA, prs, sigB, recA, hly-mpl, and plcA-hly

Gene or intergenic region	Full sequence			Optimal window with PROSEQ (highest no. of polymorphisms)			Optimal window with WINDOWMIN (best descrimination)
	Length (bp)	No. of allelic subtypesa	No. of polymorphic sitesb	Location (nt)c	No. of allelic subtypesa	No. of polymorphic sitesb	Location (nt)c	No. of allelic subtypesa	No. of polymorphic sitesb
actA	1,929	15 (14)	276 (14.3)	1,330-1,929	13 (9)	104 (17.3)	316-915	15 (14)	86 (14.3)
inlA	2,235	14 (14)	186 (8.3)	964-1,563	13 (11)	66 (11.0)	1,090-1,690	13 (13)	65 (10.8)
sigB	780	10 (10)	79 (10.1)	13-612	9 (8)	74 (12.3)	150-749	10 (10)	58 (9.7)
recA	1,047	8 (7)	107 (10.2)	264-863	8 (7)	67 (11.1)	238-837	8 (7)	66 (11.0)
prs	957	10 (7)	47 (4.9)	316-915	10 (7)	42 (7.0)	313-912	10 (7)	41 (6.8)
hly-mpl	335	8 (7)	28 (8.4)	NAd	NA	NA	NA	NA	NA
plcA-hly	242	12 (8)	26 (10.7)	NA	NA	NA	NA	NA	NA

^aTotal number of allelic subtypes distinguished within the specified region by a difference of 1 nt. Numbers in parentheses denote the total number of alleles distinguished by using the criteria of 2 nt differences.

^bNumber of polymorphic sites within the specified region. Values in parentheses are the percentages of polymorphic sites out of all nucleotides for that specific region.

^cLocations are nucleotide (nt) positions (nucleotide 1 is the first nucleotide of the coding sequence).

^dNA, not available.

TABLE 6.

Allelic profiles of the virulence genes (actA and inlA), intergenic regions (hly-mpl and plcA-hly), and housekeeping genes (sigB, prs, and recA)

Strain	Allelic designation based on:							MLST type based on sigB, prs, and recA
	actA	inlA	hly-mpl	plcA-hly	sigB	prs	recA
FSL J1-038	1	1	1	1	1	1	1	A
FSL J2-039	2	2	1	2	1	2	2	B
FSL J2-045	3	3	1	3	2	1	2	C
FSL J1-051	4	4	1	3	2	3	2	D
FSL J2-064	5	5	1	2	3	4	2	E
FSL J2-003	6	6	2	4	4	1	3	F
FSL J2-017	7	7	3	5	5	5	3	G
FSL J1-022	8	8	2	6	5	5	3	G
FSL J1-047	9	8	2	6	5	5	3	G
FSL C1-117	10	9	2	7	5	5	3	G
FSL J2-068	11	10	4	8	6	6	4	H
FSL X1-002	12	11	5	9	7	7	5	I
FSL W1-110	13	12	6	10	8	8	6	J
FSL W1-111	14	13	7	11	9	9	7	K
FSL J1-158	15	14	8	12	10	10	8	L

Insertion or deletion polymorphisms in actA, inlA, and hly-mpl.

In addition to single nucleotide polymorphisms, sequence alignments revealed insertion-deletions in actA, inlA, and hly-mpl. Nucleotide sequence alignment for actA revealed a deletion of 105 bp in strains FSL W1-110, FSL W1-111, FSL J2-039, FSL J1-158, FSL J2-003, FSL J2-064, and FSL X1-002. This finding is consistent with a previous study that reported deletion of a proline-rich repeat region among isolates classified into each of the three genotypic lineages of L. monocytogenes (30). Moreover, isolates FSL W1-111 and FSL J1-158 showed one insertion or deletion at actA nt 323 (a 9-bp insertion compared to the other isolates). The inlA sequence alignment revealed an insertion or deletion at nt 2052 in three strains (a 9-bp deletion in FSL W1-110, FSL J2-068, and FSL J2-003) and another 3-bp insertion or deletion at nt 2227 in strains FSL W1-111 and FSL J1-158 (a 3-bp insertion in these two isolates). Furthermore, isolate FSL J1-158 had a single base pair deletion at nt 2219, which results in a frameshift mutation with 7 missense amino acid residues at the 3′ end of the inlA sequence.

Sequence analyses of the hly-mpl intergenic region also revealed lineage-specific insertions or deletions. The five lineage I strains show two additional nucleotides at position 49. All lineage II strains have an additional adenine at nt 52. Insertion or deletion polymorphisms at the nt 121 to 129 region also show distinct patterns that differentiate strains in lineages I, II, and III; lineage II isolates have a 1-bp deletion compared to lineage I isolates, whereas lineage III isolates show 9- and 8-bp deletions compared the lineage I and II isolates, respectively.

Definition of most-discriminatory gene fragments.

Since complete sequencing of large DNA fragments (e.g., the 1,929-bp actA ORF) is not practical, in terms of cost and time, for large-scale subtyping, we utilized two different approaches to define smaller (600 bp) gene fragments that will provide optimal strain differentiation, ideally with the same discriminatory power achieved by analysis of the complete ORF sequences. The software ProSeq (version 2.8) was used initially in a sliding window analysis to define the 600-bp region within each gene (actA, inlA, recA, sigB, and prs) that showed the highest number of nucleotide polymorphisms. ProSeq calculated the value п, which is the average number of nucleotide differences per site between two sequences (also termed nucleotide diversity) (22), for each gene from the nucleotide alignment of the 15 strains. The region with the maximum nucleotide diversity is shown as the highest peak on a plot of п determined at sliding windows of 600 bp (Fig. 1). The locations of the 600-bp regions with the maximum nucleotide diversities for each gene are shown in Table 5. For actA, inlA, and sigB, the 600-bp window with the maximum nucleotide diversity did not allow the same level of allelic subtype differentiation as was achieved with the full ORF sequences (Table 5).

FIG. 1.

Graphical representation of polymorphisms within 600-bp sliding windows in the genes recA (a) and actA (b). Pi denotes the average number of nucleotide differences per site between two sequences, or nucleotide diversity (22). Pi was calculated with the computer program ProSeq (see Materials and Methods).

DNA sequence alignments were also used to perform sliding window analyses for each gene to define 600-bp sections that were most discriminatory for differentiating the aligned sequences into allelic types. An algorithm (WINDOWMIN) was implemented to determine the most-discriminatory 600-bp section. For three (actA, inlA, and sigB) of the five gene alignments, the WINDOWMIN sliding window analysis defined 600-bp regions that showed superior discrimination of allelic subtypes compared to the regions with the highest nucleotide diversities defined by the ProSeq sliding window analysis (Table 5). Specifically, the 600-bp windows defined by WINDOWMIN allowed subtype discrimination equal to that achieved with the complete gene sequences for four of the genes sequenced (actA, recA, prs, and sigB), which ranged in length from 780 to 1,929 bp. WINDOWMIN was not able to define a 600-bp region that matched the allelic subtype discrimination of the full ORF for inlA, the largest gene sequenced (2,235 bp). The full inlA ORF allowed differentiation of 14 allelic subtypes, whereas the most-discriminatory 600-bp sequence identified by WINDOWMIN allowed differentiation of 13 allelic subtypes.

DISCUSSION

We used L. monocytogenes as a model organism to develop and apply a rational approach for designing a DNA sequence-based subtyping scheme. Two housekeeping genes, one stress response gene, two virulence genes, and two intergenic regions were sequenced for 15 L. monocytogenes isolates selected from a collection of more than 1,000 isolates. The DNA sequence data obtained (approximately 7,500 bp/isolate) were used to define both the most-polymorphic and the most-discriminatory 600-bp regions within each gene. This model for designing discriminatory DNA sequence-based subtyping strategies also should be applicable to other target organisms. The development of a new algorithm designed to determine the most discriminatory section within a DNA sequence alignment and the implementation of this algorithm in a new program (WINDOWMIN) will further facilitate the definition of target genes and gene fragments for subtyping applications.

Target genes for DNA sequence-based subtyping of L. monocytogenes.

Housekeeping genes are commonly used for conventional MLST as these genes are thought to diversify by neutral or near neutral nucleotide changes due to the vital roles of these gene products in contributing to an organism's survival (7, 9). While sequence analysis of housekeeping genes has been a valuable tool for studying the population genetics of bacterial pathogens, DNA sequencing of more-rapidly evolving genes may allow more-sensitive subtype discrimination. We thus selected the housekeeping genes recA and prs, the stress response gene sigB, the virulence genes actA and inlA, and the intergenic regions hly-mpl and plcA-hly to determine the relative abilities of these target genes to differentiate closely related L. monocytogenes strains. While prs is located close to actA on the L. monocytogenes chromosome, sigB, recA, and the virulence genes were chosen to represent distinct chromosomal locations. prs was chosen as a target gene that would allow comparison between the discriminatory power of virulence genes and housekeeping genes located in close proximity. actA and inlA were specifically selected as target virulence genes since they are located in two different L. monocytogenes virulence gene islands (18). The complete gene sequences for actA and inlA provided the highest discrimination among the 15 isolates (15 and 14 allelic types, respectively). In addition, preliminary analyses of our data showed that the actA virulence gene indeed may be under positive selection (d_N/d_S > 1.0; d_N, rate of nonsynonymous substitutions; d_S, rate of synonymous substitutions) (unpublished data). The complete gene sequences for the housekeeping and stress response genes provided discrimination into 8 to 10 subtypes, and the sequences for the intergenic regions provided 8 and 12 allelic types based on 335- and 242-bp sequences for hly-mpl and plcA-hly, respectively. When sequence information for the two housekeeping genes and the stress response gene was used to define MLST types, a total of 12 types (A to L) (Table 6) could be differentiated. These results indicate that sequencing of the virulence genes actA and inlA provides the most-discriminatory DNA sequence-based subtyping for L. monocytogenes. While limited sequencing and PCR-restriction fragment length polymorphism analysis of other L. monocytogenes virulence genes has been performed by various groups (10, 27, 28, 30), the results reported here provide the first comparative evaluation of different target genes for subtyping L. monocytogenes.

Our results support previous observations that DNA sequencing of virulence or surface protein-encoding genes that may have been subjected to positive selection pressures can allow more-sensitive strain discrimination than sequencing of housekeeping genes can. To illustrate, Enright et al. (9) showed that emm sequences differentiated more subtypes than single housekeeping gene sequences in Streptococcus pyogenes. However, interpretation of subtyping results based on highly variable sequences of virulence genes (such as actA) or surface proteins may be misleading, particularly if the data are used to probe the population genetics or long-term phylogenetic patterns of bacterial pathogens such as L. monocytogenes. Specifically, high rates of evolution and recombination among these genes may not reflect the true phylogenetic relationships among isolates (26). To overcome this obstacle, a subtyping scheme that includes sequencing of selected virulence genes in combination with sequencing of housekeeping and/or stress response genes and of regions with little or no selective pressure (such as intergenic regions) may provide the most appropriate approach for subtyping L. monocytogenes and other bacterial pathogens. Initial analyses for recombination within the two housekeeping genes and one stress response gene sequenced here indeed showed that these genes show no (sigB and recA) or weak (prs) indication for recombination (unpublished data). The sequencing of additional housekeeping genes as previously described for MLST approaches (4-6) for other bacteria may further improve the ability to study the phylogeny of L. monocytogenes. The inclusion of both positively selected genes, such as actA, and neutrally selected intergenic regions (such as the more-discriminatory plcA-hly region) allows maximization of the discriminatory power of a typing scheme though. Maximum discrimination is particularly important for bacterial pathogens such as L. monocytogenes, for which rapid and standardized cluster detection through molecular subtyping represents a critical public health need (29). The use of virulence gene targets in DNA sequence-based subtyping strategies also creates the opportunity to use pathogen-specific PCR primers to develop integrated PCR-based detection and subtyping strategies that do not require a culturing step (5).

Rational design of high-throughput DNA sequence-based subtyping schemes.

Sequencing of complete virulence, housekeeping, and stress response genes (with ORF lengths between 780 and 2,235 bp) (Table 5) does not provide a suitable approach for high-throughput subtyping. High-throughput sequencing for subtyping purposes generally targets gene fragments between 450 and 600 bp in length, since these fragment sizes can easily be amplified and sequenced with a single set of primers or one set of sequencing primers nested inside the PCR primers (7, 20). In the past, target regions have been selected without prior identification of specific desirable characteristics, such as an optimum number of polymorphic and discriminatory nucleotide sites. To identify target regions through a rational strategy, the complete ORFs for the five genes sequenced in this study were aligned to define the 600-bp section(s) for each gene that was (i) most polymorphic (with ProSeq) or (ii) most discriminatory (with WINDOWMIN). Not surprisingly, most genes, and particularly the larger virulence genes actA and inlA, displayed considerable differences in the numbers of polymorphic residues found in different regions. Thus, the allelic discrimination achieved with different 600-bp regions within a given gene also differed considerably (Table 5). Interestingly, the 600-bp regions with the highest proportion of polymorphic residues did not necessarily provide the highest level of allelic discrimination. For three of the five genes sequenced (actA, inlA, and sigB), WINDOWMIN was able to define more discriminatory 600-bp regions than a program (ProSeq) that only determined the most-polymorphic 600-bp region within a gene. We conclude that our newly developed algorithm provides an improved rational approach for the selection of target regions for DNA sequence-based subtyping of bacterial pathogens and other microorganisms.

DNA sequence-based subtyping in L. monocytogenes.

A variety of molecular subtyping approaches have been applied to L. monocytogenes, and the application of these techniques has allowed a better understanding of the biology, ecology, and epidemiology of L. monocytogenes and other bacterial pathogens (29). Studies using subtyping approaches also have suggested that L. monocytogenes subtypes may display heterogeneity in their potentials to cause disease in humans and animals (15, 16, 30). The data reported here provide the framework for the development and implementation of DNA sequence-based subtyping methods for L. monocytogenes. We have identified specific 600-bp regions that provide the most discriminatory targets for subtyping within different L. monocytogenes genes. The presence of insertion-deletions in some of these regions (e.g., at nt 852 to 1019 in actA) may complicate the interpretation of subtyping results for some isolates and may hamper design of PCR primers that allow reliable amplification of all L. monocytogenes subtypes. Thus, in addition to identifying the most discriminatory gene regions, the presence of insertion-deletions must also be carefully considered when selecting target regions for DNA sequence-based subtyping methods. Interestingly, our results also indicate that targeting specific insertion-deletions (e.g., in the hly-mpl intergenic region) by appropriate PCR assays may allow for sensitive differentiation of the three previously described L. monocytogenes lineages (17, 30). Phylogenetic analyses of housekeeping, stress response, and virulence gene sequences also confirmed that the 15 L. monocytogenes isolates tested fall into the previously determined three distinct lineages (unpublished data) (30). The existence of these lineages has previously been confirmed by a variety of subtyping methods and thus appears to be evolutionarily relevant (29).

While current DNA fragment size-based subtyping methods (such as pulsed-field gel electrophoresis and ribotyping) may provide good subtype differentiation, data obtained by these methods typically cannot be used to determine the evolutionary relatedness of isolates. The implementation of DNA sequence-based subtyping approaches for routine characterization of human, animal, and food L. monocytogenes isolates will not only allow for sensitive and standardized subtyping for outbreak detection, but will also provide an opportunity for using subtyping data to probe the evolution of this food-borne pathogen and to track the spread of clonal groups (25, 29). DNA sequence-based subtyping methods will also provide standardized data that can easily be shared electronically and through the World Wide Web (26), concomitantly providing public health professionals and laboratories around the world with direct access to the information needed to identify and monitor emerging pathogenic bacteria. The resolution power of DNA sequence-based subtyping methods is unmatched by any other subtyping method. For example, while it is estimated that MLEE requires approximately 26 nt changes in order to determine a new electrophoretic type (1), one single nucleotide change at a targeted locus will result in a new subtype classification for DNA sequence-based subtyping (8). The continued development of new technologies for automated high-throughput sequencing and the availability of these technologies at a reasonably low cost will further facilitate widespread implementation of DNA sequence-based subtyping methods. Further application of the DNA sequencing-based subtyping approaches described here on large sets of epidemiologically well-defined isolates will also provide critical validation of the subtyping scheme proposed here.