|Home | About | Journals | Submit | Contact Us | Français|
Numbers of Clostridium difficile infections have increased worldwide in the past decade. While infection with C. difficile remains predominantly a health care-associated infection, there may also be an increased incidence of community-associated infections. C. difficile strains of public health significance continue to emerge, and reliable genotyping methods for epidemiological investigations and global surveillance of C. difficile are required. In this study, multilocus sequence typing (MLST) and multilocus variable-number tandem-repeat analysis (MLVA) were performed on a set of 157 spatially and temporally diverse C. difficile isolates that had been previously genotyped by restriction endonuclease analysis (REA) to determine the concordance among these genotyping methods. In addition, sequence analysis of the tcdC genotype was performed to investigate the association of allelic variants with epidemic C. difficile isolates. Overall, the MLST and MLVA data were concordant with REA genotyping data. MLST was less discriminatory than either MLVA or REA, yet this method established C. difficile genetic lineage. MLVA was highly discriminatory and demonstrated relationships among the MLST genetic lineages and REA genotypes that were previously unrecognized. Several tcdC genotypes were specific to epidemic clones, highlighting the possible importance of toxin misregulation in C. difficile disease pathogenesis. This study demonstrates that a combination of MLST and MLVA may prove useful for the investigation and surveillance of emergent C. difficile clones of global public health concern.
Clostridium difficile is a Gram-positive, spore-forming anaerobe and the causative agent of most hospital-acquired, antibiotic-associated diarrhea. The number of incidences of severe C. difficile infection (CDI) resulting in colectomy and death has increased dramatically worldwide over the past decade (2, 5, 18, 22, 24, 28). In addition, severe community-associated disease may be more frequent and widespread (1, 6, 7, 14, 39). Reliable genotyping methods for epidemiological investigations and global surveillance are required, as C. difficile strains of public health significance continue to emerge. Several of the most commonly used methods including pulsed-field gel electrophoresis (PFGE), restriction endonuclease analysis (REA), and PCR ribotyping generate subjective data, and the low discriminatory power of PFGE and PCR ribotyping limits their utility in epidemiological investigations (20).
Multilocus variable-number tandem-repeat analysis (MLVA) and restriction endonuclease analysis are both highly discriminatory C. difficile genotyping tools (20, 26, 36). Because MLVA provides objective and highly discriminatory results, it is particularly useful for tracking C. difficile transmission at the local level (10, 26). In contrast, multilocus sequence typing (MLST) lacks discriminatory power and is therefore better suited for investigations of C. difficile population structure and global epidemiology (20, 23). In a previous study, MLST performed on a global collection of 72 isolates demonstrated that C. difficile has a predominantly clonal population structure consisting of stable subpopulations that are globally disseminated (23). In this study, MLVA and MLST genotypes from 157 REA-typed C. difficile isolates collected over a 25-year period were compared to determine the congruence among these methods and to examine the relationships among genetic lineages. A combination of MLST and MLVA genotyping may provide insights into the origins and evolutionary relationships among C. difficile genetic lineages of clinical and public health importance. In addition, allelic variants of tcdC, which encodes a negative regulator of C. difficile toxin production, were associated with epidemic clones in previous studies (3, 9, 35). Therefore, the correlation of tcdC genotypes with genetic lineage was also investigated.
A total of 157 C. difficile strains that had previously been typed by restriction enzyme analysis (REA) were obtained from the Hines Veterans Affairs Hospital (HVA) C. difficile research laboratory (8). A detailed description of the isolates will be provided upon request (see Table S1 in the supplemental material). The isolates comprise 12 different REA groups and 92 different REA types and represent the most common epidemic and endemic REA groups in the HVA collection. Isolates representing multiple types within specific REA groups were selected to examine the concordance between REA, MLVA, and MLST. In addition, multiple isolates of a specific REA type were selected to evaluate the stability of MLVA genotyping over time. Finally, C. difficile isolates of human and animal origins were included to examine the genetic relatedness of isolates defined by toxinotyping (32). Isolates belonging to the CF REA group are toxA negative/toxB+ and toxinotype VIII, while REA group AA isolates are toxA−/toxB− and toxinotype XI. REA group BK isolates are toxA+/toxB+, toxinotype V, and binary toxin positive and bear the tcdC-A genotype, characterized by a 39-bp deletion and a nonsense mutation at nucleotide position 184 that truncates the wild-type 232-amino-acid protein to 61 amino acids (35). REA group BK isolates are frequently recovered from animals as well as humans (13, 19). REA group BI isolates are toxA+/toxB+, toxinotype III, and binary toxin positive and bear the tcdC-1 genotype, characterized by an 18-bp deletion and a deletion at nucleotide 117 that results in a 64-amino-acid truncation of the protein (9). BI REA group isolates have been responsible for multiple recent hospital outbreaks across North America and Europe (21, 27, 29). Isolates within each REA group were collected over a period of time ranging from 3 to 24 years and from diverse locations including the United States, the United Kingdom, Europe, and South America. Thus, the study collection was spatially, temporally, and genetically diverse.
The isolates were cultured from meat broth onto sheep blood agar (SBA) at 37°C under anaerobic conditions in a Coy anaerobic chamber (Coy Laboratories, Grass Lake, MI). Genomic DNA from each strain was harvested from a plate after 48 h of growth by using the Qiagen DNeasy blood and tissue kit according to the manufacturer's instructions for gram-positive organisms (Qiagen, Valencia, CA).
Automated genotyping was performed by using 6 of the 7 previously described MLVA loci (26). CDR59 was omitted from the protocol, as this locus generated few alleles and contains 2 tandem-repeat loci that cannot be differentiated by fragment analysis. PCR amplification of 1 μl of purified genomic DNA (~20 ng) was performed in 2 separate multiplex amplification reactions (multiplex 1 and multiplex 2). Genoplex 1 consisted of CDR5, CDR48, CDR49, and CDR60. Primer sequences were the same as those described previously, with the exception of the added fluorochrome and the addition of a 5′ 7-nucleotide “pigtail” to the reverse primers to improve automated allele calling (4) (Table (Table1).1). Multiplex 2 consisted of CDR4 and CDR9. The primer sequences with corresponding fluorochromes and final reaction mixture concentrations are listed in Table Table1.1. Primers for the amplification of CDR4 and CDR9 were described previously, with the exception of CdG8R2, the choice of fluorochrome, and the 7-nucleotide “pigtail” on reverse primers (36). All reactions were carried out in 50-μl reaction mixture volumes with 1.5 units of AmpliTaq Gold and 2.5 mM MgCl2 (Applied Biosystems, Foster City, CA). Cycling conditions for both multiplex 1 and multiplex 2 were an initial denaturation step at 95°C for 5 min followed by 35 cycles of 95°C for 1 min, 51°C for 45 s, and 72°C for 1 min, followed by a final extension step for 7 min at 72°C. Capillary electrophoresis was performed with 1 μl of each multiplex reaction mixture on an Applied Biosystems 3730xl DNA analyzer. Products were sized against the 6-carboxyfluorescein (FAM)-labeled MapMarker 1000 ladder (BioVentures, Murfreesboro, TN). Raw allele data were acquired for each locus by using GeneMapper software v4.0 (Applied Biosystems), and final allele calls were generated by user-defined equations for each locus to account for sequence-flanking tandem repeats and platform-dependent allele variation. Alleles at each of the 6 loci were concatenated to generate an MLVA type. Minimum-spanning-tree analysis of MLVA was performed by using BioNumerics software v5.10 (Applied Maths, Austin, TX). The summed tandem-repeat difference (STRD) was used as the coefficient for calculating the minimum-spanning tree as previously described (26). This method of analysis was validated with serial patient isolates and a collection of isolates known to be related by REA (26). Clusters containing 6 or more isolates whose MLVA types generated a summed tandem-repeat difference of ≤10 defined a clonal complex.
MLST was performed by using 6 of the 7 original housekeeping genes described previously (23). The ddl housekeeping gene was excluded from the analysis, as this locus proved to be either absent or unstable in a subset of isolates. The exclusion of ddl did not affect sequence type (ST) assignments. ST and allele assignments were generated through the Clostridium difficile MLST database maintained by the Institut Pasteur (www.pasteur.fr/recherche/genopole/PF8/mlst/). Minimum-spanning-tree analysis of MLST data was performed by using BioNumerics software v5.10 (Applied Maths). Priority rules within the BioNumerics software were set to assign the primary founder as the ST with the most single-locus variants (SLVs), as was previously described for the eBURST algorithm for inferring patterns of evolutionary descent from MLST data (11). Clonal complexes were defined as a cluster of STs in the minimum-spanning tree in which all STs were linked as SLVs to at least one other ST (11).
Restriction endonuclease analysis of isolate DNA was performed at the HVA as previously described (8). REA groups consist of REA types whose restriction patterns share ≥90% band similarity. REA groups are designated by letters, and the REA types within a group are represented numerically by numbers (8).
The tcdC genotype for each isolate was determined by PCR amplification and sequence analysis of the entire TcdC coding region. Briefly, purified genomic DNA was PCR amplified with 1.5 units of AmpliTaq Gold (Applied Biosystems) in a 50-μl volume with 200 nM (each) primers tcdCprF (5′-TATCAATTTATTTATGCTCTTTC-3′) and tcdCprR (5′-TTGCAATTATAAAAACATCTT-3′) under the following cycling conditions: 95°C for 5 min followed by 40 cycles at 95°C for 1 min, 47°C for 1 min, and 72°C for 1.5 min, ending with a final extension step at 72°C for 7 min. The PCR product was sequenced as previously described by using primer sets tcdCprF/tcdCprR and C1/C2 (9, 35). Alleles were assigned as previously described, and novel sequences were designated new tcdC genotype numbers accordingly (9).
MLVA, MLST, and tcdC genotyping methods were compared by using Simpson's index of diversity (D) to measure the probability that 2 unrelated strains will be differentiated by various typing methods (15). The concordance between the typing methods was determined by using the Wallace coefficient (W), which calculates the probability of 2 isolates being typed together by one method knowing that they were typed together by another method (30). Diversity index and concordance calculations were performed by using an online tool for comparing microbial typing methods (www.comparingpartitions.info).
MLVA had the greatest discriminatory power, generating a D value of 0.998 for the 157 C. difficile study isolates (Table (Table2).2). The majority of the 143 MLVA types identified were defined by a single isolate. There were 10 MLVA types identified that had more than one isolate. REA identified 92 types generating a D value of 0.979, while MLST identified only 17 different STs with an index of diversity of 0.879. The majority of the study isolates comprised 5 STs: ST1, ST2, ST3, ST6, or ST41 (Fig. (Fig.1).1). There were 12 different tcdC genotypes observed in the study collection. One isolate (G131), a toxinotype XI variant, was tcdC negative (see Table S1 in the supplemental material). The most common tcdC genotype, tcdC-0 corresponded to reference strain VPI 10463 (35).
Minimum-spanning-tree analysis of the MLST data from the 157 C. difficile study isolates revealed 3 clonal complexes (Fig. (Fig.1).1). The largest clonal complex was comprised of 5 STs (Fig. (Fig.1A)1A) and included 65 isolates belonging to the B, R, K, and Y REA groups. The ST clonal complex including ST19 and ST9 (Fig. (Fig.1B)1B) comprised 11 isolates belonging to the G, Y, and W REA groups. The third clonal complex, comprised of ST52 and ST53 (Fig. (Fig.1C),1C), is a 5-locus variant of ST49 and is therefore considered to be distantly related to the remainder of the isolate collection. Similarly, isolates belonging to ST41 were distantly related to the rest of the isolate collection, as these isolates differ from ST49 at all 6 MLST loci (ST41) (Fig. (Fig.1).1). Interestingly, most of the isolates comprising ST41, ST52, ST53, ST2, and ST54 are of toxinotype VIII (toxA negative/toxB+) and produce only toxin B (17, 33). A subset of the ST41 isolates (REA group AA) is toxinotype XI (toxA negative/toxB negative). The toxin phenotypes of these isolates are different from those of the isolates belonging to the other STs, which produce both toxins A and B.
In general, there was good concordance between the REA groups and STs (Table (Table3).3). The probability of predicting the correct ST based on knowledge of the REA group was 77%. This concordance is due to the observation that a majority of isolates within a given REA group belong to a single ST. There were 4 REA groups, however, whose isolates comprised multiple STs. For instance, REA group Y isolates belong to ST1, ST36, ST49, and ST9, and isolates belonging to REA group G, R, and CF belong to multiple STs.
The probability of predicting the correct REA group from the ST was 64%. There were 3 STs that represented multiple REA groups. ST6 contained the largest number of isolates from multiple REA groups, including B, K, and R. In addition, MLST could not discriminate the BK and AA REA groups comprising ST41.
The probability of predicting the correct ST based on the REA type was 92% (Table (Table3).3). This high concordance is due to the fact that the majority of REA types are represented by a single ST. Only REA type Y4 isolates belong to multiple STs, either ST36 or ST49 (Fig. (Fig.1;1; see also Table S1 in the supplemental material).
Minimum-spanning-tree analysis of the MLVA data from the 157 study isolates identified 6 clonal complexes containing 6 or more isolates (Fig. (Fig.2).2). The largest clonal complex is comprised of 46 isolates of diverse genetic lineages and REA groups (Fig. (Fig.2B).2B). The majority of the isolates belong to either REA group B and ST6 or REA group Y and ST1. This pattern of clustering is similar to that of clonal complex A in the MST of the MLST data (Fig. (Fig.1A),1A), suggesting that the REA group B ST6 and REA group Y ST1 isolates are genetically related. Unlike the MLST data, the K and R REA groups appear as distinct populations by MLVA (Fig. 2A and C). In addition, the MLVA data suggest that isolates belonging to the R and J REA groups are related (Fig. (Fig.2A).2A). This relationship was not observed for the MST of the MLST data. Similarly to the MLST data, MLVA demonstrated that isolates belonging to variant toxinotypes (ST41 and ST2) are distantly related not only to each other but also to the rest of the isolate collection (Fig. 2E and F). In addition, MLVA was able to discriminate animal from human toxinotype V isolates within the BK REA group (Fig. (Fig.2E2E).
The concordance between MLVA type, REA group, and ST was excellent. The probability of predicting the REA group or ST of a particular isolate based on the MLVA type was 100% (Table (Table3).3). The majority of REA types could be predicted from the MLVA type. There were only 3 MLVA types (types 657, 674, and 688) that represented more than one REA type, all within the B REA group (see Table S1 in the supplemental material). Conversely, the concordance between REA typing and MLVA was only 6% (Table (Table3).3). This discrepancy was due to the higher discriminatory power of MLVA. There were 16 REA types with multiple isolates that could be further discriminated by MLVA. For instance, the 11 isolates belonging to REA type B1 were differentiated into 6 different MLVA types. The 10 isolates belonging to REA type BI6 generated 8 MLVA types (Fig. (Fig.2D)2D) Thus, the MLVA type could not be consistently predicted from the REA type.
In order to assess the utility of MLVA for C. difficile genotyping over time, the stability of the tandem-repeat loci and MLVA genotypes was examined for REA types containing 9 to 10 isolates collected over 2 to 14 years. The most stable locus was CDR5, which was generally invariant across the 5 REA types examined: B1, BI6, J9, Y1, and Y4 (see Table S1 in the supplemental material). CDR4 was the least stable locus, generating a unique allele for the majority of isolates in 3 of the 5 REA type collections investigated. The B1 collection of 11 isolates displayed the greatest genotype stability over time, generating 6 different genotypes over a 14-year period (Fig. (Fig.3A).3A). Most of the genotype variation observed in the B1 isolates was due to single-tandem-repeat differences at either CDR4, CDR9, CDR49, or CDR60. There were 2 isolates, G006 and G053, that were distantly related to the rest of the B1 isolates (Fig. (Fig.3A).3A). This genetic distance was due primarily to large tandem-repeat differences at CDR4. These data suggest that CDR4 may undergo genetic recombination to generate large changes in copy numbers over a short period of time similar to large tandem-repeat copy number mutations observed previously for Escherichia coli O157 (38). For instance, isolates G011 and G006, which were isolated within a 3-month period in 1982, have an STRD of 7 at CDR4 (see Table S1 in the supplemental material). Alternatively, the tandem-repeat copy number variation observed at CDR4 could be due to slipped-strand mispairing during DNA replication (37).
The 10 BI6 isolates collected over a 2-year period from 2003 to 2005 generated 8 different MLVA genotypes. CDR5 was the most stable locus for these isolates, consistently generating allele 3. CDR48 was also relatively stable for this population, generating 3 alleles, the majority of which were allele 9. Copy numbers of 10 and 12 were also observed in subsequent years, suggesting that the number of CDR48 repeats may increase by one tandem repeat over time. CDR4 and CDR49 were relatively unstable for this group of isolates, generating 7 and 5 alleles, respectively. This instability, however, permitted the detection of regional clones among the BI6 isolates. For instance, the Oregon isolates can be distinguished from the Chicago-A isolates, which are distinct from the New Jersey, Ohio, Pennsylvania, and Chicago-B isolates (Fig. (Fig.3B3B).
The J9 isolates collected over 14 years demonstrated relative stability at CDR5, CDR9, CDR48, and CDR60. However, genetic instability was evident for CDR4 and CDR49. CDR4 displayed a large copy number range, generating 9 different alleles with no temporal pattern of variance evident. CDR49 generated 6 alleles with a small copy number range. Like the BI6 isolates, differences in copy numbers at CDR4 and CDR49 may reveal regional diversity among the J9 isolates.
The Y1p isolates generated stable alleles for CDR5, CDR48, and CDR60 over the 12-year period examined. Similar to the J9 isolates, instability at CDR4 and CDR49 was evident. Stability at CDR5 and CDR48 was seen for the Y4 isolates collected over 13 years. However, variability at CDR4, CDR9, CDR49, and CDR60 obscured the detection of significant genetic relationships (STRD ≤ 10) among this group of isolates.
Sequence analysis of the tcdC gene identified 12 genotypes, 7 of which were previously described (9). One toxinotype VIII REA group CF isolate belonging to ST53 did not generate a PCR product. There were 8 tcdC genotypes that each represented a single ST, and the probability of predicting the ST based on the tcdC genotype was 63% (Table (Table3).3). The toxinotype V isolates belonging to REA groups AA and BK comprising ST41 consistently generated the tcdC-A genotype, and tcdC-1 was consistently associated with ST44, REA type BI6, toxinotype III isolates. In addition, tcdC-3 was consistently associated with ST3, J REA group isolates. The tcdC-0 genotype corresponding to the VPI 10463 tcdC gene sequence originally described by Spigaglia and Mastrantonio was identified in multiple lineages including ST6, ST9, and ST48 (35). The tcdC-9 genotype was also associated with multiple genetic lineages including ST1, ST5, ST6, ST36, and ST49 and appeared to be specific to REA groups R and Y.
The tcdC-7 genotype was identified only in the REA group CF and CG toxinotype VIII isolates belonging to ST2, ST52, and ST54. These data suggest that tcdC allelic variants may be lineage dependent.
In this study, MLST genetic lineages that were single-locus variants of one another defined a clonal complex in the MST analysis. There were 17 C. difficile genetic lineages defined by MLST, and a large clonal complex comprising 5 genetic lineages and 41% of the study isolates was identified. Of note, the majority of MLST lineages demonstrated relative concordance with REA groups representing the most common epidemic C. difficile clones in the HVA collection. However, MLST could not discriminate all epidemic REA groups. For instance, the ST6 lineage was comprised of REA groups B, K, and R, and ST41 included isolates belonging to both the BK (toxinotype V) and AA (toxinotype XI) REA groups. In some instances, MLST provided greater discrimination than REA. MLST identified 4 REA group G genetic lineages that were unrelated to one another. This finding is supported by the observation that REA group G isolates tend to be sporadic, representing an endemic C. difficile population that is rarely associated with epidemic disease. REA group Y isolates were also further discriminated by MLST but were single-locus variants, with the majority of Y isolates clustering together on the MST. These data demonstrate the validity of MLST for the investigation of C. difficile phylogenetics and evolution. The MLST data also revealed that the lineages comprising isolates of variant toxinotypes were distantly related to the rest of the isolate collection. These results suggest that REA groups BK (toxinotype V), CF (toxinotype VIII), and AA (toxinotype XI) diverged from a common progenitor at an early stage in C. difficile evolution and are unrelated to the most common C. difficile epidemic clones.
Like MLST, MLVA is an objective genotyping method that provides greater discriminatory power than any other method tested to date. Used in combination, these 2 typing methods can provide details regarding C. difficile population structure both globally and locally. The fine-typing capabilities of MLVA are demonstrated in this study. Where MLST assigned isolates belonging to the B, K, and R REA groups to the same genetic lineage, MLVA revealed the genetic diversity of ST6. Furthermore, the MLVA data suggest that a clonal population of REA group B isolates gives rise to multiple C. difficile subpopulations. MLVA also demonstrates a genetic relationship between the ST3 and ST6 genetic lineages, suggesting that the REA group R and J epidemic clones are related.
The MLVA and MLST data both indicate that the isolates of variant toxinotypes belonging to REA groups BK, CF, and AA are genetically distinct from each other and from the rest of the isolate collection. Of note is the finding that the nontoxigenic REA group AA isolates are genetically related to REA group BK by both MLVA and MLST. This observation is consistent with previous characterizations of these binary-toxin-positive clones (12). One possibility is that REA group AA may have evolved from an REA group BK progenitor through a recombination event that resulted in the deletion of all of tcdB and part of tcdA. Isolates belonging to the BK REA group are of toxinotype V and have been linked to community-associated C. difficile infections (16). Significant public health concern exists due to the association of these isolates with food animals (16, 19, 34). In this study, the toxinotype V REA group BK isolates of human and animal origins were highly related but differentiated by MLVA. These results are similar to data from a recent investigation of ribotype 078 prevalence using a different MLVA scheme (13). Moreover, the discovery of C. difficile in retail ground meat emphasizes the concern regarding animal-to-human transmission and food safety, although no human CDI cases have been documented to have been acquired from food (31). Epidemiological investigations of animal, food, and human isolates by MLVA may help elucidate any causal relationships among food products and C. difficile-associated human disease, assuming that MLVA discrimination is not too sensitive.
Several of the tcdC genotypes described in this study were associated exclusively with epidemic clones. For instance, the tcdC-A genotype was associated exclusively with the ST41 REA group BK and AA isolates. This allele, originally described by Spigaglia and Mastrantonio, encodes a 61-amino-acid truncation of TcdC (35). The ST44 BI6 isolates bear the tcdC-1 genotype, characterized by a deletion of nucleotide A117, which results in a 64-amino-acid truncation of TcdC (9, 25). These truncating tcdC genotypes presumably generate nonfunctional TcdC and are therefore thought to contribute to an increased virulence of the strains bearing these particular mutations (9, 25; S. Matamouros, R. Govind, and B. Dupuy, presented at ClostPath 2006: 5th International Meeting on Molecular Biology and Pathogenesis of Clostridia, Nottingham, United Kingdom, 21 to 25 June 2006). While the significance of these allelic variants to C. difficile pathogenesis remains unclear, these characteristic genotypes help define and differentiate epidemic C. difficile clones.
The isolates examined in this study were collected from diverse geographic locations over as many as 24 years. However, the MLVA results demonstrate that the tandem-repeat loci used in this analysis are sufficiently stable to reveal phylogenetic associations that could not be discerned by either MLST or REA. The STRD metric used to define MLVA clonal complexes in this study can be sensitive to spatial and temporal variations as well as data set completeness. Results presented here indicate that for most C. difficile clonal lineages, CDR4 evolves rapidly, is highly variable, and can achieve high copy numbers. Therefore, investigations of C. difficile phylogeny by MLVA should be performed either on large, all-encompassing isolate collections or in combination with MLST to first establish genetic lineage. The use of MLST as a backbone to support the more-discriminating MLVA data will be useful in inferring the genetic relationships of clonal populations of C. difficile.
The high concordance between MLVA typing and both REA and MLST provides further validation of these conclusions.
MLST is a reliable, objective, and accurate method to characterize relationships among C. difficile genetic lineages. The high discriminatory power of MLVA provides the fine-typing required to discern genetic relationships among a clonal population of C. difficile. In combination, MLST and MLVA provide phylogenetic information that will be valuable for investigations of C. difficile population structure and clonal emergence.
While this paper was under review for publication, a new C. difficile MLST scheme and database were developed by Lemée and colleagues at the Institut Pasteur (http://www.pasteur.fr/recherche/genopole/PF8/mlst/). Readers should be aware that the data presented in this manuscript are based on the “old MLST scheme” available at http://www.pasteur.fr/recherche/genopole/PF8/mlst/Cdifficile.html.
This work was supported by a research grant from ViroPharma Inc. and grants from the U.S. Department of Veterans Affairs Research Service (D.N.G. and S.J.).
D.N.G. has served as a consultant for ViroPharma.
†Supplemental material for this article may be found at http://jcm.asm.org/.
Published ahead of print on 2 December 2009.