Discordant 16S and 23S rRNA Gene Phylogenies for the Genus Helicobacter: Implications for Phylogenetic Inference and Systematics

Abstract

Analysis of 16S rRNA gene sequences has become the primary method for determining prokaryotic phylogeny. Phylogeny is currently the basis for prokaryotic systematics. Therefore, the validity of 16S rRNA gene-based phylogenetic analyses is of fundamental importance for prokaryotic systematics. Discrepancies between 16S rRNA gene analyses and DNA-DNA hybridization and phenotypic analyses have been noted in the genus Helicobacter. To clarify these discrepancies, we sequenced the 23S rRNA genes for 55 helicobacter strains representing 41 taxa (>2,700 bases per sequence). Phylogenetic-tree construction using neighbor-joining, parsimony, and maximum likelihood methods for 23S rRNA gene sequence data yielded stable trees which were consistent with other phenotypic and genotypic methods. The 16S rRNA gene sequence-derived trees were discordant with the 23S rRNA gene trees and other data. Discrepant 16S rRNA gene sequence data for the helicobacters are consistent with the horizontal transfer of 16S rRNA gene fragments and the creation of mosaic molecules with loss of phylogenetic information. These results suggest that taxonomic decisions must be supported by other phylogenetically informative macromolecules, such as the 23S rRNA gene, when 16S rRNA gene-derived phylogeny is discordant with other credible phenotypic and genotypic methods. This study found Wolinella succinogenes to branch with the unsheathed-flagellum cluster of helicobacters by 23S rRNA gene analyses and whole-genome comparisons. This study also found intervening sequences (IVSs) in the 23S rRNA genes of strains of 12 Helicobacter species. IVSs were found in helices 10, 25, and 45, as well as between helices 31′ and 27′. Simultaneous insertion of IVSs at three sites was found in H. mesocricetorum.

Helicobacter pylori was first identified in 1982 (39) as an infectious cause of chronic active gastritis and since then has been associated with peptic ulcers, gastric adenocarcinoma, and mucosa-associated lymphoid tissue lymphoma in humans (38). The genus Helicobacter now includes at least 26 formally named species (60). The tropism of the genus Helicobacter ranges from the stomach, cecum, and colon to the liver or genital tract of mammals and birds. Helicobacter spp. comprise a diverse group of potentially emerging pathogens in multiple species: humans, nonhuman primates, cats, dogs, ferrets, swine, sheep, cattle, wild birds, chickens, mice, rats, hamsters, gerbils, and a variety of wildlife, including dolphins, whales, and seals (60).

Over the past 15 years, microbiology has undergone a momentous shift from a determinative taxonomy to one based on phylogeny (61). Our current understanding of prokaryotic phylogeny rests largely on 16S rRNA gene sequence analysis, which has become the basis for reorganizing Bergey's Manual of Systematic Bacteriology and The Prokaryotes (3, 51). 16S rRNA gene analysis has proved useful for defining prokaryotic relationships from the species to phylum levels. Cases of very closely related strains or species have highlighted the need to use other, more rapidly changing molecules for some genera (36). Of concern for the validity of prokaryotic systematics are cases where 16S rRNA gene-derived phylogeny seems to conflict with other credible molecular or phenotypic data. Within the genus Helicobacter, there have been several reported conflicts. For example, Helicobacter nemestrinae (ATCC 49396^T), was once believed to be a novel gastric helicobacter isolated from a pigtailed macaque. However, it was later recognized to be a strain of H. pylori by correction of the 16S rRNA gene sequence and analysis of seven housekeeping and two flagellin genes, which clustered together with sequences from 20 or more H. pylori isolates from diverse sources (53). Vandamme et al. (58) have recently published data that indicated that “Helicobacter westmeadii” and Helicobacter sp. strain Mainz (27) are both strains of Helicobacter cinaedi. The 16S rRNA gene sequence for the Mainz strain differs by 4.3% from that of the type strain. The authors stated that similar within-species differences in 16S rRNA gene sequences have been reported in Campylobacter hyointestinalis, another species that belongs to the epsilon subdivision of the division Proteobacteria (25). The purpose of this study was to obtain extensive sequence data for a second phylogenetically informative molecule in order to understand the source of the discrepancy between 16S rRNA gene analyses and those from other phenotypic and genotypic methods. The nearly complete 23S rRNA gene sequences (>94%) for 55 helicobacter strains representing 41 helicobacter taxa were determined.

MATERIALS AND METHODS

Bacterial strains.

The strains sequenced and their sources of isolation are presented in Table 1. All strains were grown in an 80% N₂, 10% CO₂, 10% H₂ atmosphere at 37°C on Columbia blood agar plates.

TABLE 1.

Accession numbers for strains examined in this study

Species	Strain sequenceda	Culture collectionb	Host	GenBank accession no.		23S rRNA gene sequenced (bp)
				16S rRNA gene	23S rRNA gene
Campylobacter jejuni	NCTC 11168		Human	AL139074c	AL139074c	Genome
Helicobacter acinonychis	Eaton 90-119T	ATCC 51101T	Cheetah	M88148d	AY596214	2,715
Helicobacter aurati	MIT 97-5075T	ATCC BAA-1T	Hamster	AF297868	AY596241	2,711
Helicobacter bilis	MIT Hb1T	ATCC 51630T	Mouse	U18766d	AY596255	2,875
Helicobacter bizzozeronii	CCUG 35045T	ATCC 700030T	Dog	Y09404c	AY596215	2,715
Helicobacter canadensis	NLEP 16143T	ATCC 700968T	Human	AF262037	AY596220	2,713
Helicobacter canadensis	NLEP 16767	ATCC 700969	Human		AY596221	2,713
Helicobacter canis	23S, ATCC 51401T		Dog	AY631945e	AY596234	2,710
	16S, NCTC 12739T
Helicobacter canis	ATCC 51402	NCTC 12741	Dog	AY631946	AY596235	2,710
Helicobacter cetorum	MIT 99-5656T	ATCC BAA-540T	Dolphin	AF292378	AY596262	2,717
Helicobacter cholecystus	ATCC 700242T	ATCC 700242T	Hamster	AY686606e	AY596229	2,807
Helicobacter cinaedi	CCUG 18818T	CCUG 18818T	Human	M88150d	AY596254	2,718
Helicobacter cinaedi	MIT 01-5002		Monkey	AY631947	AY596261	2,718
Helicobacter felis	Lee CS1T	ATCC 49179	Cat	M57398d	AY596216	2,715
Helicobacter felis	Lee CS3		Cat	AY631948	AY596217	2,714
Helicobacter felis	Lee CS5		Cat	AY631949	AY596219	2,716
Helicobacter felis	Lee DS2		Dog	AY686607e	AY596218	2,717
Helicobacter fennelliae	CCUG 18820T	ATCC 35683T	Human	M88154d	AY596237	3,075
Helicobacter flexispira taxon 1	ATCC 43968		Pig	U96300	AY596249	2,712
Helicobacter flexispira taxon 2	ATCC49314		Sheep	AF225546	AY596257	3,052
Helicobacter flexispira taxon 3	ATCC 49320		Pig	AF225547	AY596256	2,876
Helicobacter flexispira taxon 4	ATCC 49310		Sheep	AF225548	AY596248	2,711
Helicobacter flexispira taxon 5	ATCC 43966		Sheep	M88137	AY596250	2,712
Helicobacter flexispira taxon 7	Eaton 1302		Dog	U51874d	AY596247	2,712
Helicobacter flexispira taxon 8	Schauer DBS59		Mouse	AY686608e	AY596253	2,808
Helicobacter flexispira taxon 8	ATCC 49309		Human	AY631950	AY686610	2,712
Helicobacter flexispira taxon 8	ATCC 43879		Human	M88138d	AY596252	2,713
Helicobacter ganmani	MIT 99-5102		Mouse	AY631951	AY596224	2,714
Helicobacter hepaticus	MIT Hh2T	ATCC 51448T	Mouse	U07574d	AY596243	2,887
Helicobacter hepaticus	MIT 96-1809		Mouse	AY631952	AY596244	2,807
Helicobacter hepaticus	MIT 96-284		Mouse	AY631953	AY596242	2,712
Helicobacter marmotae	MIT 98-6070T	ATCC BAA-546T	Woodchuck	AF333341	AY596245	2,716
Helicobacter mesocricetorum	ATCC 700932T		Hamster	AF072471	AY686611	3,641
Helicobacter muridarum	Lee ST1T	ATCC 49282T	Rat	M80205d	AY596238	3,084
Helicobacter mustelae	MIT 89-977		Ferret	AY631954	AY596233	2,795
Helicobacter mastomyrinus	MIT 94-022		Mouse	AF225550d	AY596240	2,810
Helicobacter mastomyrinus	MIT 97-5577T		Mastomys	AY631955	AY596265	2,683
Helicobacter pametensis	Seymour B9T	CCUG 29255T	Tern	M88147d	AY596230	2,745
		ATCC 51478T
Helicobacter pullorum	ATCC 51801T	NCTC 12824T	Chicken	AY631956e	AY596222	2,713
Helicobacter pullorum	ATCC 51864	NCTC 12827	Human	L36144c	AY596223	2,713
Helicobacter pylori	ATCC 26695		Human	AE000644c	AE000644c	2,843
Helicobacter pylori	J99		Human	AE001553c	AE001553c	2,886
Helicobacter rodentium	MIT 95-1707T	ATCC 700285T	Mouse	U96296	AY596225	3,086
Helicobacter rodentium	MIT 95-2160	ATCC 700286	Mouse	U96297	AY596227	3,086
Helicobacter rodentium	MIT 95-2178		Mouse	AY631957	AY596226	3,085
Helicobacter sp. strain Cotton-top tamarin	MIT 97-6194-5		Cotton-top tamarin	AF107494	AY596246	2,970
Helicobacter sp. strain Seal-1	MIT 01-5529-A		Seal	AY203898	AY596258	2,713
Helicobacter sp. strain Seal-2	MIT 01-5529-B		Seal	AY203899	AY596259	3,103
Helicobacter sp. strain Rhesus-1	MIT 99-5504		Rhesus monkey	AF333339	AY596263	2,714
Helicobacter sp. strain Rhesus-2	MIT 99-5507		Rhesus monkey	AF333340	AY596264	2,708
Helicobacter sp. strain Mainz	CCUG 33804		Human	AF207739c	AY596260	2,559
Helicobacter sp. strain CLO3	CCUG 14564		Human	M88151d	AY596236	2,722
Helicobacter sp. strain Bird-B	Seymour B10	ATCC 51480	Tern	M88139d	AY596232	2,707
Helicobacter sp. strain Bird-C	Seymour B52	ATCC 51482	House sparrow	M88144d	AY596231	2,707
Helicobacter trogontum	MIT 95-5368	ATCC 700117	Rat	AY686609	AY596251	2,712
Helicobacter typhlonius	MIT 97-6810		Mouse	AF127912	AY596239	2,712
Helicobacter winghamensis	ATCC BAA-430		Human	AF246984	AY686612	2,712
Wolinella succinogenes	Tanner 602WT	ATCC 29543T	Cow	M88159d	AY596228	2,881

^aStrains are from the following investigators: Eaton, Kathryn Eaton, University of Michigan; Lee, Adrian Lee, University of New South Wales, Australia; Seymour, Charles Seymour, deceased; Schauer, David Schauer, Massachusetts Institute of Technology; and Tanner, Anne Tanner, The Forsyth Institute.

^bATCC, American Type Culture Collection; CCUG, Culture Collection University Göteborg; MIT, Massachusetts Institute of Technology; NLEP, National Laboratory for Enteric Pathogens, Winnipeg, Canada.

^cThis sequence was obtained from GenBank and was determined by investigators listed in the GenBank entry.

^dGenbank entry from our laboratory has been revised, in some cases extensively.

^eNew, more accurate, or longer sequence for a strain previously deposited in GenBank by other investigators.

DNA extraction.

The High Pure PCR Template Preparation Kit (Roche Molecular Biochemicals, Indianapolis, IN) was used to extract DNA from bacterial pellets. Bacteria were scraped from plates and washed with 1 ml phosphate-buffered saline. After centrifugation, the cells were resuspended in 200 μl phosphate-buffered saline, and lysis was initiated by adding 15 μl of 10-mg/ml lysozyme. Following incubation at 37°C for 15 min, 40 μl of proteinase K (20 mg/ml) and 200 μl of binding buffer were added and incubated at 72°C for 10 min. The reaction mixture was loaded onto the filter tube and washed twice with wash buffer, and the DNA was eluted with 200 μl of Tris-EDTA.

PCR amplification of rRNA genes.

The 23S rRNA genes were amplified using PCR primers O68 and M89, complementary to highly conserved regions on 23S rRNA genes (Table 2). The reaction mixture (50 μl) contained 1× polymerase buffer, a 0.4 μM concentration of each of the two primers, a 200 μM concentration of each deoxyribonucleotide, and 2.5 U Platinum or AccuPrime Taq DNA polymerase (Invitrogen, Carlsbad, California). The following stepdown conditions were used for amplification: activation of polymerase at 94°C for 2 min; 8 cycles of denaturation at 94°C for 45 s, annealing at 62°C to 55°C for 45 s (dropping 1°C per cycle), and elongation at 72°C for 3 min; and 22 cycles of 94°C for 45 s, 55°C for 45 s, and 72°C for 3 min, followed by a final elongation step of 7 min at 72°C. A 15-μl portion of the sample was then electrophoresed through a 1% agarose gel to check the size and concentration of the amplicon.

TABLE 2.

23S rRNA gene PCR and sequencing primers used in this study

Primer designation	Position	Sequence (5′-3′)
Final seta
O68	44-60 Forward	AGG CGA TGA AGG ACG TA
V67	46-63 Forward	GCG ATG AAG GAC GTA CTA
M90	241-255 Forward	AGT AGY GGC GAG CGA A
V68	465-483 Forward	GAG GGA AAG GTG AAA AGA A
O20	571-587 Forward	CAT AAT GAT CCT GCG AGT T
M92	975-990 Forward	AGG GRA ACA RCC CAG A
M93	1090-1104 Forward	WGC GTA AYA GCT CAC
M94	1608-1623 Forward	AAA CCG WCA CAG GTR G
V63	2012-2032 Forward	GAA ATT GTA GTG GAG GTG AA
V65	2375-2393 Forward	GAA AGT CGG TCA TAG TGA T
M96	2498-2512 Forward	CCT CGA TGT CGR CTC
V66	2580-2596 Forward	TGG GTT CAG AAC GTC GT
M83	241-256 Reverse	KTT CGC TCG CCR CTA C
O21	571-587 Reverse	AAC TCG CAG GAT CAT TAT G
O22	681-698 Reverse	GGC CAT GGA TAG ATC ACT
M85	1091-1105 Reverse	AGT RAG CTR TTA CGC
V62	1331-1348 Reverse	CCC GAC TAA CCC TAC GAT
M86	1608-1623 Reverse	CYA CCT GTG WCR GTT T
V64	2012-2032 Reverse	TTC ACC TCC ACT ACA ATT TC
M88	2498-2512 Reverse	GAG YCG ACA TCG AGG
P46	2652-2668 Reverse	CGG TCC TCT CGT ACT AG
M89	2744-2659 Reverse	CTT AGA TGC YTT CAG C
Preliminary setb
ZSF4	940-958 Forward	GAG TCA GGC GGT GGG TGA T
ZSF3	1817-1836 Forward	GTA TAA GGT GTG ACG CCT GC
M95	2053-2069 Forward	GAC RGA AAG ACC CCR TG
ZSF5	2366-2384 Forward	AGC AGA GAC GAA AGT CGG T
ZSR7	454-473 Reverse	CTT TCC CTC ACG GTA CTT GT
ZSR6	799-818 Reverse	CGA AGA GAA CCA GCT ATC AC
M87	2053-2069 Reverse	CAY RGG GTC TTT CYG TC

^aFinal-set primers were designed based on alignment of approximately 20 nearly complete helicobacter 23S rRNA sequences. They contain some wobbles that are not necessary for helicobacters but which may make them useful for many epsilon Proteobacteria.

^bInitial-set primers were designed based on those described by Lane (31). These primers have mismatches with some or many helicobacter sequences and were not used once the final set of primers was designed.

16S rRNA genes were amplified using primers F24 and F25 as described previously (13).

Sequencing.

PCR-amplified fragments were purified with a QIAGEN PCR purification kit (QIAGEN Inc., Valencia, CA) for direct sequencing using a BigDye terminator cycle sequencing kit on an ABI 310 or ABI 3100 Genetic Analyser (Applied Biosystems, Foster City, CA). Quarter-dye chemistry was used following the manufacturer's instructions. 23S rRNA gene cycle sequencing was performed using 25 cycles as follows: denaturing at 94°C for 10 s and annealing and extension at 60°C for 4 min. Because intervening sequences (IVSs) of up to 415 bases occurred at four positions in helicobacter sequences, the full set of 20 primers was routinely used to ensure complete sequencing across any potential IVS. A map showing primers and potential IVS positions is given in Fig. 1.

FIG. 1.

Map of primer and IVS locations in helicobacter 23S rRNA gene. The positions of PCR and sequencing primers are shown by arrows with sequence designations above. The positions of four known intervening sequences are indicated by pairs of curly brackets.

Data analysis.

RNA sequences were aligned manually based on secondary structure in our database (43). Base positions (columns) which did not contain specific base information (ambiguous bases or indels) for at least 52 of the 56 sequences were removed. The aligned sequences were exported in Phylip format for analysis using the Phylip set of programs, version 3.6b (19). Similarity matrices were constructed using the program Dnadist. Distances were corrected for multiple base changes using the method of Jukes and Cantor (29). Neighbor-joining (NJ) trees were constructed using the program Neighbor (47). Parsimony trees were constructed using the program Dnapars with the order of sequence additions jumbled 25 times. Maximum likelihood (ML) trees were constructed using the program Dnaml with Speedier off and Global on. Bootstrapping for neighbor joining and parsimony analyses was performed using the program Seqboot to generate 100 resampled sets of sequences. Consensus trees from bootstrapping analyses were generated using the program Consense. Trees for publication were prepared from the Phylip tree files using TreeView version 1.6.6 (42).

Whole-genome comparisons.

The genome for Helicobacter hepaticus (NC_004917) was downloaded from GenBank. Each protein was subjected to a BLASTP search. The ranked order of similarity of H. hepaticus to H. pylori (either genome), Wolinella succinogenes, and Campylobacter jejuni was determined for those genes present in all four species. The mean similarity score (bits) for the set of shared genes was determined between H. hepaticus and the other three species.

Nucleotide sequence accession numbers.

The GenBank accession numbers and culture collection accession numbers for the species and strains examined in this study are listed in Table 1.

RESULTS

23S rRNA gene sequences.

Essentially complete 23S rRNA gene sequences (>2,700 bases) were determined for 55 Helicobacter strains representing 41 species or unnamed taxa. Each strain was sequenced on both strands using 16 to 20 primers listed in Table 2. For phylogenetic analysis, the sequences from the complete genome projects for H. pylori ATCC 26695 and J99 and Campylobacter jejuni were also included.

16S rRNA gene sequences.

Some of the early helicobacter 16S rRNA sequences from our laboratory had been determined using direct reverse transcriptase sequencing of rRNA and contained numerous “Ns” or were missing intervening sequences. Fifteen previously deposited sequences were redone, and the GenBank entries were updated (noted in Table 1). Five helicobacter sequences were deposited in GenBank to give longer or more accurate 16S rRNA gene sequences for entries previously submitted by other laboratories (noted in Table 1). We did not have the 16S rRNA gene sequence for H. canadensis NLEP 16767 in time to include it in the 16S rRNA gene phylogenetic analyses; thus, the 16S rRNA gene analyses are based on 57 rather than 58 taxa.

NJ analyses.

NJ trees based on 23S rRNA and 16S rRNA gene sequence analyses are shown in Fig. 2 and Fig. 3, respectively. The strain numbers, GenBank accession numbers, and species from which strains were isolated are included in the trees.

FIG. 2.

Neighbor-joining tree based on comparison of 23S rRNA gene sequences. The tree shown is the simple neighbor-joining tree. All sequences are labeled by species, strain number, GenBank accession number in curly brackets, and the species from which it was isolated. The consensus bootstrap neighbor-joining tree had identical topology. The numbers immediately to the left of branches indicate the number of times out of 100 the clade was recovered by bootstrap resampling. The tree is divided into eight numbered clusters.

FIG. 3.

Neighbor-joining tree based on comparison of 16S rRNA gene sequences. The tree shown is the simple neighbor-joining tree. All sequences are labeled as in Fig. 2. Nodes marked with Xs were not shared between the simple neighbor-joining tree and the consensus bootstrapping tree. The tree is divided into clusters. Clusters 1 to 3 correspond to those in the 23S rRNA gene tree, whereas clusters A to E differ from clusters 4 to 8 of the 23S rRNA gene tree.

In the 23S rRNA gene tree, there were eight major clusters. Cluster 1 included the gastric helicobacters H. pylori, H. acinonychis, H. cetorum, H. bizzozeronii, and H. felis. A partial sequence for H. salomonis (not deposited in GenBank or included in Fig. 1) placed it adjacent to H. bizzozeronii and H. felis. Cluster 2 included all helicobacters without sheathed flagella: H. rodentium, H. ganmani, H. winghamensis, H. canadensis, H. pullorum, and H. mesocricetorum. Also included in cluster 2 was the rumen organism Wolinella succinogenes. Cluster 3 contained H. pametensis, other bird strains, H. cholecystus, and H. mustelae. Clusters 4 to 7 were comprised of enterohepatic helicobacters. Cluster 8 was comprised of taxa with flexispira morphology (fusiform-shaped organisms with spiral periplasmic fibers and bipolar tufts of sheathed flagella), with the exception of H. cinaedi strains.

Eight major clusters were also identified in the 16S rRNA gene NJ tree. As in the 23S rRNA gene tree, cluster 1 was comprised of gastric strains, cluster 2 was comprised of helicobacters with unsheathed flagella (plus flexispira taxon 1), and cluster 3 was comprised of H. pametensis, H. cholecystus, H. mustelae, and helicobacter bird isolates. In the tree shown in Fig. 3, cluster 2 is broken into two adjacent paraphyletic clusters. Cluster C contained the same strains as cluster 6 in the 23S tree with the addition of the seal strain MIT 01 5529A. The remaining four clusters (B through E) are a random assortment of strains from 23S rRNA gene clusters 4, 5, 7, and 8.

The consensus NJ tree produced by bootstrap analysis of 100 resamplings of the 23S rRNA gene sequence data had branching identical to that of the simple neighbor-joining tree (bootstrapping values are shown at the nodes in Fig. 2). By contrast, the consensus NJ tree produced by bootstrap analysis of 100 resamplings of the 16S rRNA gene sequences produced a tree with several rearrangements of deep branching (Fig. 3). Nodes marked with Xs in Fig. 3 indicate differences between the simple neighbor-joining tree and the consensus bootstrap neighbor-joining tree. Thus, the 16S rRNA gene neighbor-joining tree was not stable to bootstrap resampling. The mean bootstrap value for all nodes in the 23S rRNA gene tree was 82.7, whereas it was only 66.9 for the 16S rRNA gene tree.

Parsimony analyses.

Consensus parsimony trees produced by bootstrap analysis of 100 resamplings of the 23S rRNA and 16S rRNA gene sequences are shown in Fig. 4. The 23S RNA consensus parsimony tree retains the eight clusters of the neighbor-joining consensus tree with minor rearrangements within clusters and the backbone branching of the clusters: cluster 3 was split into two adjacent paraphyletic clusters, and cluster 4 became a monophyletic cluster. The 16S rRNA gene consensus parsimony tree had substantial rearrangements of the branching of the eight clusters and the fragmentation of cluster A with a two-sequence fragment inserted in the middle of cluster B relative to the 16S rRNA gene NJ tree. The mean bootstrap value for the 23S rRNA gene parsimony tree was 71.9, whereas it was only 52.3 for the 16S rRNA gene tree. The 23S rRNA gene tree had only 1 node bootstrap value under 30, whereas the 16S rRNA gene tree had 11. Several parsimony analyses were run using 25 jumbles of the input sequence order with different random-number seeds. The tree with the fewest steps for the 23S rRNA gene had 3,086 steps with seed 13. This tree had minor rearrangements of the eight clusters relative to the consensus parsimony tree. Three trees with 1,265 steps were the smallest found for the 16S rRNA gene, also with a seed of 13. These trees were substantially rearranged compared to the bootstrap consensus parsimony tree.

FIG. 4.

Parsimony trees based on comparison of 23S rRNA gene and 16S rRNA gene sequences. The trees shown are consensus bootstrap trees based on 100 resampled parsimony trees. The numbers at the nodes are the number of times out of 100 resamplings that the node was present. Sequences are labeled as in Fig. 2 with GenBank numbers omitted. The clusters are labeled as in Fig. 2 and Fig. 3.

ML analyses.

ML trees for 23S rRNA and 16S rRNA gene sequence analysis were constructed (data not shown). The 23S rRNA gene ML tree was similar to the NJ tree and contained each of the eight clusters. The tree shown had a ln of −22,152.8. The 16S rRNA gene ML tree did not retain the major branching pattern shown in the NJ or parsimony tree. The ln was −97,337 for the 16S rRNA gene tree.

Within- and between-species sequence differences.

Several Helicobacter species had sequences determined for multiple strains. The within-species sequence variabilities for 23S rRNA gene sequences were as follows: H. canadensis, 0.1%, two strains; H. canis, 0.2%, two strains; H. cinaedi, 0.3%, three strains; H. felis, 2.0%, four strains; H. hepaticus, 1.8%, three strains; “H. mastomyrinus,” 1.2%, two strains; H. pullorum, 0.1%, two strains; H. pylori, 0.5%, two strains; H. rodentium, 0.1%, three strains; and Helicobacter sp. flexispira taxon 8, 0.2%, three strains. The interspecies variability was as low as 1.1% between the closely related species H. canadensis and H. pullorum and as high as 12% between H. felis and H. aurati. The helicobacter sequences differed by about 16% from that of Campylobacter jejuni. By 23S rRNA gene analysis, Wolinella succinogenes does not represent a neighboring genus but rather branches within the helicobacters. Its sequence differs from other helicobacter sequences by 8 to 12%.

Intervening sequences in 23S rRNA genes.

IVSs were found in 14 strains representing 12 species. The positions and lengths of these IVSs are shown in Table 3. IVSs were inserted into four distinct locations in the 23S rRNA gene. Ten of the sequences demonstrated single IVS insertions, but Helicobacter sp. flexispira taxon 2 contained IVSs in two locations, and H. mesocricetorum contained IVSs in three locations. The IVSs for flexispira taxa 2 and 3 are identical and differ from that of H. bilis by only 6 bases (3.4%). The IVSs for three H. rodentium strains differ from each other by 2 or fewer bases (or indels; 0.5%). While the IVS of H. muridarum is the same length as that from H. rodentium, it differs by approximately 240 bases and is only 37% similar (25% similarity is expected by chance).

TABLE 3.

IVSs in 23S rRNA gene sequences

Taxon	Straina	Length (bases)
		Helixb 10-10′ [150-176]c (9-11)d	Helix 25-25′ [545-549] (8)	Helices 31′-27′ [652-655] (4-5)	Helix 45-45′ [1171-1178] (2-8)
Helicobacter sp. strain Cotton-top tamarin	MIT 97-6194-5	269
Helicobacter bilis	MIT Hb1T	173e
Helicobacter sp. flexispira taxon 2	ATCC 49314	173e	185
Helicobacter sp. flexispira taxon 3	ATCC 49320	173e
Helicobacter mesocricetorum	ATCC 700932T	237	370	344
Helicobacter sp. strain Seal-2	MIT 01-5529-B		415
Helicobacter rodentium	MIT 95-1707T		381f
Helicobacter rodentium	MIT 95-2160		381f
Helicobacter rodentium	MIT 95-2178		380f
Helicobacter muridarum	Lee ST1T		380f
Helicobacter mustelae	MIT R85-13-6PT		93
Helicobacter fennelliae	CCUG 18820T			271
Helicobacter pametensis	Seymour B9T				159
Helicobacter cholecystus	ATCC 700242T				99

^aStrain sequenced. Abbreviations for sources are the same as in Table 1, notes a and b.

^bHelices are numbered as in reference 5.

^cIn square brackets the position of the insertion taken from the alignment numbered relative to Escherichia coli.

^dIn parentheses, the length of the region in Helicobacter species without an IVS.

^eThe IVSs for Helicobacter spp. flexispira taxa 2 and 3 are identical. The IVS for H. bilis differs from that of flexispira taxa 2 and 3 by 6 bases.

^fThe IVS for H. rodentium strains differ by 2 or fewer bases. The IVS for H. muridarum has only 37% similarity to that of H. rodentium strains.

Whole-genome comparisons.

BLASTP (1) comparison of all H. hepaticus proteins with those in H. pylori (genomes for two strains), W. succinogenes, and C. jejuni found that there were 870 proteins present in each of the four species with similarities for which E is less than e⁻⁵. E is the probability due to chance that there is another alignment with a similarity greater than the given S score. The raw score S is based on the number of matches a sequence has to a sequence in the database, with a penalty subtracted for substitutions or gaps. For proteins, substitution penalties are based on the PAM (percent of accepted mutations) matrix or BLOSUM (blocks substitution matrix). As shown in Fig. 5, H. hepaticus is more closely related to W. succinogenes than to H. pylori and most distantly related to C. jejuni based on mean S scores. For a four-taxon comparison, there are six possible tree topologies. Support for each of the six possible trees is shown in Fig. 6. Analyses of 870 protein trees support the 23S rRNA gene phylogeny and the branching of W. succinogenes within the helicobacters. Also, analyses with more stringent cutoffs of E less than e⁻¹⁵ or e⁻²⁵ give similar results with 769 or 657 proteins included in the four-way comparison.

FIG. 5.

Comparison of 870 H. hepaticus protein sequences to those from H. pylori (Hp), W. succinogenes (Ws), and C. jejuni (Cj). The sequences of 870 proteins shared by the four species were compared by BLASTP analysis. These 870 proteins were selected from all proteins found in these organisms based on reciprocal BLASTP matches with similarity values of E less than e⁻⁵ for all comparisons. The S value is a measure of similarity based on percent identity and length of comparison. The E value is the probability of obtaining an S value by chance based on comparison of the sequence to those in GenBank. The error bars represent standard deviations.

FIG. 6.

Phylogeny of H. hepaticus (Hh), H. pylori (Hp), W. succinogenes (Ws), and C. jejuni (Cj) based on comparison of 870 protein trees. There are six possible trees for a four-taxon comparison. The number of proteins supporting each topology is shown above the respective tree.

DISCUSSION

In this study, we found that 16S rRNA and 23S rRNA gene phylogenetic trees are discordant. The 23S rRNA gene analyses are in agreement with other molecular and phenotypic methods as described below, suggesting that the 16S rRNA gene sequence data analyses do not faithfully reflect phylogenetic relationships (22, 23, 40, 58). The findings support the majority of the criticisms by Vandamme et al. of the use of the 16S rRNA gene in helicobacter systematics (58). However, the implications for phylogenetic and taxonomic inference are not as straightforward as suggested in that article (58) or the study by Fox et al. (20). Discussion of our specific results will precede the discussion of the implications of our findings.

Helicobacter sp. strain Mainz (CCUG 33804) is the primary example cited for 16S rRNA gene sequence information giving discrepant phylogenetic information (58). The 16S rRNA gene sequence for the strain suggested that the Mainz strain is most closely related to H. fennelliae, whereas DNA-DNA hybridization, whole-cell protein profiles by polyacrylamide gel electrophoresis, and biochemical comparison suggested that it is a strain of H. cinaedi. As shown in Fig. 2, the 23S rRNA gene sequence for the Mainz strain is 99.8% similar to that of the type strain and to a monkey strain of H. cinaedi. Thus, in the Mainz strain example, the 23S rRNA gene sequence data, but not the 16S rRNA gene sequence data, are consistent with other phenotypic and genotypic analyses.

“Flexispira rappini” is a provisional name given by Bryner et al. (7, 8) to fusiform-shaped organisms with spiral periplasmic fibers and bipolar tufts of sheathed flagella that were first isolated by Kirkbride et al. (30). In a previous study, we demonstrated that strains with flexispira morphology fell into at least 10 Helicobacter taxa by 16S rRNA gene sequence comparison (14). Since that paper was published, Helicobacter aurati and strain MIT 01-5529A, isolated from a seal, representing two additional flexispira taxa, have been described (24, 44). As noted in Fig. 3, the 12 flexispira taxa are scattered throughout the 16S rRNA gene tree. Phylogenetic analysis of Helicobacter sp. flexispira taxa by 23S rRNA gene sequence data highlights a number of discrepancies with the previous 16S rRNA gene analyses. Most prominent is the position of Helicobacter sp. flexispira taxon 1, strain ATCC 43968, which falls in cluster 2 by 16S rRNA gene analysis (Fig. 3) but falls in cluster 8 and is essentially identical to Helicobacter sp. flexispira taxon 4, strain ATCC 49310, by 23S rRNA gene analysis (Fig. 2). Of note in the 23S rRNA gene tree is that strains with the flexispira morphology group together in clusters 7 and 8, with only a single strain, H. aurati, in cluster 6. Recently, Hänninen suggested that Helicobacter spp. flexispira taxa 1, 4, and 5 are members of a single species, H. trogontum (flexispira taxon 6) on the basis of sodium dodecyl sulfate-polyacrylamide gel electrophoresis protein pattern analysis, dot blot hybridization, PCR-restriction fragment length polymorphism patterns, and HSP60 gene analysis (23, 40). The 23S rRNA gene data support Hänninen's contention that Helicobacter spp. flexispira taxa 1, 4, and 5 and H. trogontum are closely related, but we would argue against placing these taxa into a single species at this time. Hänninen has also suggested that flexispira taxa 2, 3, and 8 be included in Helicobacter bilis on the basis of phenotypic characteristics, dot blot DNA-DNA hybridization, and phylogeny of ureB and HSP60 genes (22). While the 23S rRNA data are in general agreement with these results, we believe placement in a single species requires further analysis. There appears to be wide variability in the genetic diversity of helicobacter strains included in a species (lumping versus splitting). Several named species differ by only approximately 2% by both 16S and 23S rRNA gene sequence comparisons: H. canadensis and H. pullorum, H. rodentium and H. ganmani, H. cholecystus and H. pametensis, H. muridarum and H. typhlonius, and H. bilis and H. cinaedi. We believe that species with highly divergent 16S rRNA gene strain sequences, like H. felis, may be comprised of more than one species. Species boundaries need to be set by rigorous quantitative means, such as the renaturation rate DNA-DNA hybridization of De Ley et al. (12), or sequence comparisons examining within-species versus among-species comparisons (10, 37). Furthermore, we are not convinced that dot blot hybridization or other less strenuous DNA-DNA hybridization methods are either quantitative or reliable.

The discrepancy in the position of Wolinella succinogenes between 16S and 23S rRNA gene phylogenetic analyses is marked and has significant taxonomic implications. By 16S rRNA gene analysis, W. succinogenes branches outside the genus Helicobacter and appears to represent a genus (43). By 23S rRNA gene analysis, it branches within the genus Helicobacter, consistently branching with the unsheathed-flagellum helicobacters, cluster 2 (Fig. 2). Whole-genome comparison of 870 shared protein genes supports the 23S rRNA gene placement of W. succinogenes within the helicobacters. As shown in Fig. 5 and Fig. 6, H. hepaticus is more closely related to W. succinogenes than to H. pylori, supporting W. succinogenes' position within the nongastric subcluster of helicobacters. The original genome papers for H. hepaticus (52) and W. succinogenes (2) were published nearly simultaneously, so cross comparisons were included in neither paper. A recent paper by Eppinger et al. (17) comparing these four genomes also demonstrated that H. hepaticus shared more genes with W. succinogenes than with H. pylori or C. jejuni (see Fig. 1 in reference 17). Despite these finding, the authors concluded that the phylogenetic position of W. succinogenes was in agreement with published 16S rRNA gene analyses. We strongly disagree with their conclusion and believe their data support W. succinogenes branching within the helicobacters.

The 23S rRNA gene and whole-genome analyses indicate that W. succinogenes branches with, and is as much a helicobacter as, H. canadensis, H. ganmani, H. mesocricetorum, H. pullorum, H. rodentium, and H. winghamensis. W. succinogenes and these six Helicobacter spp. differ morphologically from other helicobacters in having unsheathed flagella. However, if W. succinogenes is recognized as a member of the genus Helicobacter, it presents an onerous issue for helicobacter nomenclature. Wolinella Tanner et al. 1981 (55) has priority over Helicobacter Goodwin et al. 1989 (21) as a genus name. Changing the name of the genus from Helicobacter to Wolinella would not be accepted by the medical or microbiological communities and is not suggested by the authors. Nevertheless, the 23S sequence and other phenotypic and genotypic analyses suggest that the genus Helicobacter has three major divisions, the gastric helicobacters (cluster 1), the unsheathed-flagellum helicobacters (cluster 2), and the enterohepatic helicobacters (clusters 3 to 8). The name Helicobacter could be preserved by restricting the name Wolinella to organisms in cluster 2. However, this would prompt the question of whether the enterohepatic helicobacters (clusters 3 to 8) should also be designated a novel genus, as the remaining helicobacters (minus cluster 2) would be members of two paraphyletic groups (cluster 1 and clusters 3 to 8). We make no taxonomic recommendations at this time but raise these issues for discussion, as they are implicit in our findings. The Subcommittee for the Taxonomy of Campylobacter, Helicobacter, and Related Organisms may be an appropriate body to consider these issues.

As described in the results section, the phylogenetic trees derived by neighbor-joining, parsimony, and maximum likelihood analyses of 23S rRNA gene sequence data support eight major clusters of helicobacters (Fig. 2 and 4). The trees derived by three different methods are generally congruent and have high bootstrapping values. The trees derived from the 16S rRNA gene sequence data differ significantly from one another depending on the treeing methodology (Fig. 3 and 4). The 16S rRNA gene sequence-based trees have significantly lower bootstrap values than the 23S rRNA gene-derived trees and are sensitive to change in outgroup or strain inclusion (unpublished data). Thus, while the phylogeny inferred from 23S rRNA gene sequence data is quite stable and consistent by different treeing methods and is consistent with other phenotypic and genotypic data, phylogenetic inferences made from 16S rRNA gene sequence data are discordant.

While small-subunit rRNA gene-based sequence analysis has become the foundation for prokaryotic and eukaryotic phylogenetic analysis, a number of issues are now recognized which qualify the meaning and use of the rRNA gene-derived phylogenetic trees. Early doubts about the validity of a universal tree of life based on the 16S rRNA gene analysis were in part dispelled by finding that the 23S rRNA gene trees and trees for certain conserved proteins, such as elongation factors, ATPase subunits, and RNA polymerases, were highly congruent (34, 35, 46). With the completion of numerous prokaryotic genomes, it became evident that horizontal gene transfer was a widespread phenomenon, raising new questions regarding the validity of 16S rRNA gene-based phylogenetic inference (9, 16). The realization that there is a distinction between gene phylogenies and organismal phylogeny added to the questioning of 16S rRNA gene-based phylogeny (16). These doubts were eased in part by studies such as those by Brochier et al. (4), who, using whole-genome data, generated a tree by concatenating 57 proteins from the translational apparatus and found it highly congruent with that obtained by concatenating 16S and 23S rRNA gene sequences from 45 bacterial species for which complete genome sequences were available. Jain et al. (28) articulated the “complexity hypothesis,” which states that genes whose proteins are components of complex interacting systems are less likely to be transferred than genes whose proteins are involved in few or no interactions. Several investigators have found that informational genes (transcription, translation, and related processes) are more phylogenetically informative than operational genes (metabolism, transporters, etc.). Lerat et al. (32) searched complete genomes of gamma Proteobacteria and identified genes for which all species had single orthologs. He found 203/205 genes gave the same tree, and gene concatenation gave a fully resolved phylogeny. Only two cases of lateral gene transfer were found. There appears to be a general (but not universal) consensus that an organismal phylogeny exists and that it can be deduced from a comparison of sets of conserved genes, but the best methods for such analyses are still being debated. However, the common theme emerging from this debate is that no individual molecule, including 16S rRNA, is completely reliable for reconstructing organismal phylogeny.

The use of 16S rRNA gene sequences for determining phylogeny was initially based on the assumption that sequence diversity was purely due to evolutionary change and that the 16S rRNA gene was not influenced by horizontal gene transfer due to “complexity hypothesis” considerations (28). However, as early as 1993, Sneath provided evidence for horizontal exchange of 16S rRNA gene fragments in Aeromonas species (50). A careful analysis of 16S rRNA genes in Streptomyces strains by Ueda et al. (56) demonstrated that there were two classes of substitution patterns. The first class included individual base changes, which were mainly transitions, and the second class involved groups of five or more base changes and involved mainly transversions, which suggested horizontal gene transfer. These findings were extended by Wang and Zhang (59), who examined actinomycetes. They found evidence consistent with lateral gene transfer of short gene segments corresponding to hairpins in the 16S rRNA gene. They described the exchange of 16S rRNA gene fragments as the “simplified complexity hypothesis.” They suggested that this mode of lateral gene transfer would create mosaic rRNA genes and gradually destroy the evolutionary history recorded in the sequence. The proposed transfer of 16S rRNA gene segments is supported by Schouls et al. (48), who examined the Streptococcus anginosus group, and van Berkum et al. (57), who examined rhizobia. In the case of helicobacters, neither the “complexity hypothesis” nor the “simplified complexity hypothesis” is supported by the sequence changes seen in the Mainz strain. Alignment of the Mainz strain with H. fennelliae and H. cinaedi suggests that the entire 5′ 1,260 bases were obtained from a helicobacter closely related to H. fennelliae.

Natural transformation in helicobacters has been studied most intensively in H. pylori. It has been found that recombination in H. pylori occurs at such a high rate that the population structure is panmictic and polymorphisms within many gene loci are at linkage equilibrium (54). DNA is thought to be taken up through a transmembrane channel system related to type IV secretion systems (49). Components of the transformation apparatus that have been recently characterized include a ComE3 homologue and a nuclease component (41, 62). Thus, it is not in the least surprising that helicobacter 16S rRNA gene sequences have been modified by horizontal exchange and their phylogenetic signal degraded. But as noted in the discussion above, it is still possible in the presence of genetic exchange to construct an organismal phylogeny. What is required for reliable phylogenetic analysis is inclusion of a sufficiently large number of informative bases to recover the true phylogenetic signal from the noise created by horizontal transfer of genetic information (whole genes and gene fragments). In analysis of helicobacter phylogeny, it would appear that the approximately 220 informative bases in the 16S rRNA gene sequence are insufficient, whereas the approximately 690 informative bases in the 23S rRNA gene sequence appear sufficient to recover the phylogenetic signal. Interestingly, neighbor-joining analysis of concatenated 16S and 23S rRNA gene sequences produces a tree essentially identical to the 23S rRNA gene tree (tree not shown). However, when neighbor-joining trees are generated from halves of the 23S rRNA gene data set yielding about 345 informative bases per half, the eight clusters start to be rearranged, and when data sets containing 220 bases are used, the trees are rearranged substantially. This suggests that phylogenetic robustness is dependent on the total number of informative bases rather than the particular gene, at least in the cases of 16S and 23S rRNA genes.

Despite the greater information content in the 23S rRNA molecule, there are sound reasons why 16S rRNA has become the primary molecule for phylogenetic inference. Obtaining a full 23S rRNA gene sequence requires substantially more than twice the effort to obtain twice the sequence length. Unlike the typical 16S rRNA operon, where highly conserved primer targets exist within 10 bases of the 3′ and 5′ ends, only semiconserved targets exist for 23S rRNA operons 250 and 150 bases from the 3′ and 5′ ends: Lane's primers 256f and 2447r, respectively (31). Lane's most 5′ 23S rRNA gene primer (130f) is not adequate for helicobacters. Primers closer to the ends, such as helicobacter 44-60 forward (O68) (Table 2), were designed specifically to amplify helicobacters and related taxa but were not intended to be universal. Sites suitable for sequencing primer targeting are also less conserved for the 23S rRNA gene than for the 16S rRNA gene. The initial sequencing primers used in this project were based on those described by Lane (31) and modified by examining the aligned sequences of H. pylori, E. coli, and other divergent taxa. After about 20 full helicobacter 23S rRNA gene sequences were obtained, their alignment was used for designing additional and more appropriate helicobacter-specific primers. The final primer set (Table 2) worked well for sequencing helicobacter 23S rRNA genes. For completeness, Table 2 also includes primers that were subsequently discarded. The sequences for the final primers contain some wobbles that are probably not required for the genus Helicobacter, but these may make them more likely to work with other epsilon Proteobacteria.

We conclude that the 23S rRNA gene sequence data are significantly more reliable than 16S rRNA gene sequence data for identification, classification, and phylogenetic analysis of helicobacters primarily because of the threefold-higher number of informative bases. This conclusion will best be supported or challenged by examining a third (or multiple) phylogenetically reliable molecule. This will not be a minor undertaking. The present study represented a major sequencing effort, as approximately 500,000 bp of raw sequence information were determined to generate the 150,000 bp of assembled 23S rRNA gene sequence information deposited in GenBank. To be useful for making a meaningful comparison with the 23S rRNA gene sequence data, the new molecule must have equal information content, about 700 informative bases for divergent Helicobacter species. Unfortunately, phylogenetic studies of helicobacters using the coding sequence for proteins, such as HSP60 (GroEL), examined a region of 590 bases with only approximately 120 informative bases (40). To obtain 700 informative bases will likely require sequencing 3 to 4 kb of finished sequence (a long single gene or multiple shorter genes) for each strain. Full double-strand coverage of a similar number of strains will require an effort comparable to the half million bases of raw sequence determined in the present study. Obtaining more reliable phylogenetic information than that provided by the 16S rRNA gene requires acquisition of greater sequencing information content, and unfortunately, there are no easy shortcuts.

The presence of IVSs in helicobacter 16S and 23S rRNA genes has been reported previously (26, 33), and they are known to occur in other prokaryotes, such as some Enterobacteriaceae (45). Where they have been examined, IVSs were cleaved without splicing by RNase III (18). Hurtado et al. (26) previously reported IVS insertions in helix 25. The present report documents IVS insertions in helices 10, 25, and 45 and between helices 31′ and 27. This is also the first report of multiple IVS insertions into a helicobacter 23S rRNA gene, which we noted in Helicobacter sp. flexispira taxon 2 and H. mesocricetorum. The three IVSs in H. mesocricetorum increase the size of the 23S rRNA gene by 930 bases. Primers O68 and V62, which produce a 1.3-kb amplicon for most helicobacters, produce a 2.1-kb amplicon in H. mesocricetorum. The functions of IVSs in both 23S and 16S rRNA genes remain unknown. There are examples, such as H. canis, where only some strains of a species contain an IVS (26). IVSs may originate from the insertion of a mobile DNA element into the rRNA gene. It is not known whether IVSs carry either a positive or negative selective advantage, but a slight positive pressure would favor horizontal transfer of rRNA gene fragments carrying an IVS, and a slight negative pressure would favor horizontal transfer of rRNA gene fragments without an IVS.

Bacterial taxonomy or systematics deals with classification, nomenclature, and identification (11). Classification deals with the consistent arrangement of like taxa into groups. Nomenclature is the naming of taxa and is regulated by the International Committee of Systematic Bacteriology and the Bacteriological Code. Identification is the process of characterizing an unknown and determining if it matches a named taxon. Strains from a majority of helicobacter species can be reliably identified by a 16S rRNA gene sequence match of >99% to a species type strain. However, Fox et al. (20) provocatively raised the question of “how close is close,” or how close 16S RNA sequences need to be to guarantee species identity. The answer is that a 99% 16S rRNA gene sequence similarity does not always guarantee species matching, as some closely related species have essentially identical 16S rRNA gene sequences. What was not addressed in the Fox paper is how often one would be led astray by assuming species identity for 99% 16S rRNA gene sequence similarity. Of the tens of thousands of bacterial strains that have 16S rRNA gene sequence information, less than a dozen cases have been cited where neighboring species have greater than 99% 16S rRNA gene sequence similarity. For routine identification, this level of error is substantially better than phenotypic analysis. A second important issue is how divergent 16S rRNA gene sequences can be and still represent a well-defined species (DNA-DNA hybridization of >70%). For the vast majority of species examined, <2% divergence, or >98% similarity, reliably defines a species. However, within the helicobacters, sequences for strains of H. felis differ from the type strain by up to 2% and could be confused with the closely related H. bizzozeronii and H. salomonis. The Mainz strain differs from other H. cinaedi strains by over 4% (58). Strains with sequences that are less than 99% similar to a type strain sequence fall into a poorly defined category where they cannot be reliably identified. Thus, while 16S rRNA gene analysis is generally more reliable than biochemical characterization for identification (because of the substantially greater information content), it is not perfect. However, the assertion that investigators have routinely misidentified helicobacters based on 16S rRNA gene sequence analyses is false. What is true is that investigators have misclassified helicobacter strains based on the assumption that a 16S rRNA gene sequence difference of 2% or 3% from named species indicated a novel species. As reported by Vandamme et al. (58), at least one species of helicobacter, H. cinaedi, has strains with highly divergent 16S rRNA gene sequences (4.3%). The divergent sequence of the Mainz strain led Husmann et al. (27) to incorrectly classify the Mainz strain as a new species. The Mainz strain was misclassified, not misidentified. The proper standards for classification of novel helicobacters are fully described by Dewhirst et al. (15). While 16S rRNA gene divergence of greater than 2% or 3% (98% or 97% similarity) was used previously as one of the criteria to classify helicobacter strains as belonging to novel species, it is clearly invalid to use such an argument in light of current knowledge. A final point regarding the Mainz strain and the divergence of 16S rRNA gene within the species H. cinaedi needs to be made. Sequences for H. cinaedi strains do not differ randomly by 0 to 4% from the sequence of the type strain, but rather, the vast majority of strains are 99% similar to the type strain. Most likely, through genetic exchange of 16S rRNA genes or gene fragments with a species closely related to H. fennelliae, the Mainz strain's 16S rRNA gene sequence came to differ markedly (>4%) from other H. cinaedi strains. There are probably other H. cinaedi strains closely related to the Mainz strain (>99% similarity). Thus, H. cinaedi strains can be reliably identified if their 16S rRNA gene sequences are 99% similar to that of either the type strain or the Mainz strain. It is misleading for investigators to cite the H. cinaedi 4.3% within-species 16S rRNA gene sequence divergence without appropriate caveats.

The findings in this study have implications for prokaryotic phylogenetic inference and systematics. Where 16S rRNA gene-based phylogeny conflicts with other credible phenotypic and genotypic methods, as in the cases of helicobacters and the Rhizobiaceae (6), more robust molecular data should be obtained prior to making classification or taxonomic proposals. However, more robust data require determining more informative bases than are present in the 16S rRNA gene. Whether this is done by sequencing the 23S rRNA gene or sequencing other phylogenetically informative gene targets, it will likely require 3 to 4 kb of sequence information for each strain to be truly robust. Finally, identifying strains is always a trade-off between certainty and expense. Ideally, information on a plethora of phenotypic and genotypic traits would be obtained for this purpose. Despite the caveats presented in this paper, 16S rRNA gene sequence information is still among the most cost-effective and useful information to have for strain identification.