FOODBORNE PATHOGENS AND DISEASEVolume 3, Number 1, 2006© Mary Ann Liebert, Inc.

Interpretation of Pulsed-Field Gel ElectrophoresisPatterns in Foodborne Disease Investigations

and Surveillance

TIMOTHY J. BARRETT, PETER GERNER-SMIDT, and BALA SWAMINATHAN

ABSTRACT

Since the establishment of the well-known Tenover criteria in 1995 (Tenover et al., 1995), relatively few papershave been published about the interpretation of subtyping data generated by pulsed-field gel electrophoresis(PFGE). This paper describes the approach that has been used in the PulseNet network during the past 10 years.PFGE data must always be interpreted in the proper epidemiological context and PFGE data can not alone provean epidemiological connection. The Tenover criteria are not generally applicable to the interpretation of PFGEsubtyping data of foodborne pathogens. The reproducibility of the method with a particular organism, the qual-ity of the PFGE gel, the variability of the organism being subtyped, and the prevalence of the pattern in questionmust always be considered. Only isolates displaying indistinguishable patterns should be included in the detec-tion of clusters of infections or the initial case definition in a point-source outbreak. More variability (patternsdiffering from each other in two to three band positions) may be accepted if the outbreak has been going on forsome time or if person-person spread is a prominent feature. If epidemiological information is sufficiently strong,isolates with markedly different PFGE patterns may be included in an outbreak.

20

INTRODUCTION

SUBTYPING is the characterization of bacteriabelow the species and subspecies level. It

may be performed by a multitude of phenotypicand genotypic techniques. Subtyping may bedone for taxonomic purposes, study of pop-ulation structure of a species, phylogeneticanalyses directed at determining relatedness of the target organism to other organisms or delineating steps in the evolution of the tar-get organism, or for molecular epidemiology. Herein, we focus our attention solely on molec-ular epidemiologic application of subtyping.

Levin et al. (1999) defined molecular epidemi-ology as the identification of microparasites re-sponsible for infectious disease and determiningtheir physical sources, their biological relation-ships, and their route of transmission and those

of the genes responsible for their virulence, vaccine-related antigens, and drug resistance.From a public health standpoint, the determi-nation of the “physical source of the micropar-asite” that caused human illness, particularlyin an outbreak setting, and understanding itsroute of transmission from the source to the af-fected person are of paramount importance.However, another aspect of the statement ofLevin et al. (1999) needs to be emphasized. Inconsidering the subtype of an organism, onemust consider all available data such as bio-chemical reaction profiles, serotype, bacterio-phage type, antimicrobial susceptibility profiles,presence and/or type of surface appendagesand virulence factors; reliance on a single para-meter for characterization should be avoidedwherever possible.

An infectious disease outbreak is defined as

Foodborne and Diarrheal Diseases Branch, Centers for Disease Control and Prevention, Atlanta, Georgia.

a cluster of acute illnesses caused by a pathogenthat are geographically and temporally associ-ated, and occur in excess of what is usually ex-pected for that time and place. The smallestoutbreak consists of two cases. In a few in-stances, where a pathogen or its toxin causesextremely severe disease that may result indeath (e.g., botulism), even a single case is con-sidered and investigated as an outbreak. Onlya small proportion of enteric illnesses manifestas common source outbreaks; nevertheless,these outbreaks contribute greatly to the un-derstanding of the transmission of the patho-gens causing foodborne and diarrheal diseases,because they offer the best opportunities toidentify the source of the outbreak by provid-ing multiple case histories to compare for com-mon exposure to a specific source or vehicle(Keene, 1999). Further, the identification of thefood, water or environmental source of infec-tion and the institution of appropriate publichealth remedial measures (e.g., recall of food,closing of the water source, closure of an im-plicated petting zoo) will result in preventionof additional exposure and disease. When thesource of contamination of an outbreak is de-termined to be a widely distributed food, theremedial measures may have long-term impli-cations for an entire segment of the food in-dustry and may lead to significant changes inspecific food production and processing proto-cols for that segment of the food industry (e.g.,requirements for more thorough cooking ofhamburger patties in fast-food restaurants af-ter the 1993 Western states Escherichia coliO157:H7 outbreak). Findings from some out-break investigations also stimulate further re-search on the prevalence and persistence of thepathogen in the vehicle or source of the out-break that may lead to the development of pre-vention strategies that may reduce or eliminatethe recurrence of similar contamination eventsin the future.

BENEFITS TO PUBLIC HEALTH

Strain identification by subtyping has manypublic health benefits. While patient diagnosisand treatment are seldom impacted by the ad-ditional information provided by subtyping,

subtyping has a profound impact on publichealth surveillance, disease cluster identifica-tion, and outbreak investigations. For example,the identification of a diarrheal pathogen as Sal-monella and determining its antimicrobial sus-ceptibility profile provide adequate informa-tion to a physician for the treatment of thepatient. However, further characterization ofthe clinical Salmonella isolate by determinationof its serotype, bacteriophage type, its resis-tance to antimicrobial agents, and the determi-nation of its DNA “fingerprint” are necessaryand essential for public health officials to studythe emergence and disappearance of specificserotypes, and for timely recognition of newlyevolving strains, particularly those that may beresistant to treatment by multiple antimicrobialtherapies. Further, serotype and DNA “finger-print” data enable the detection of disease clus-ters, particularly those that are geographicallydispersed. Subtyping also assists epidemiolo-gists in separating outbreak-associated casesfrom temporally associated sporadic cases ofdisease caused by the same pathogen, therebyincreasing the power of epidemiologic analy-ses. Once the epidemiologists complete theiranalysis of an outbreak and implicate a specificfood as the cause of the outbreak, isolation ofthe pathogen from the implicated food and itscharacterization by subtyping provide inde-pendent confirmation of epidemiologic find-ings.

Subtyping methods may be phenotypic orgenotypic and may vary in discriminatingpower from low to very high. Examples of thephenotyping methods are biotyping, serotyp-ing, antimicrobial susceptibility profiles, andbacteriophage typing. Except for bacteriophagetyping, the other phenotyping methods gener-ally have low discriminating power and areused primarily as first level subtyping meth-ods. Among the plethora of DNA-based sub-typing methods, macrorestriction analysis bypulsed-field gel electrophoresis (PFGE) has be-come the gold standard for bacterial pathogensubtyping during the past ten years because of its broad applicability, high discriminatingpower, and epidemiologic concordance. ThePulseNet network has demonstrated that PFGEpatterns generated using highly standardizedPFGE protocols and their analysis by com-

INTERPRETATION OF PFGE IN FOODBORNE DISEASE INVESTIGATIONS 21

puter-assisted pattern normalization techniquesresults in very high levels of reproducibility ofPFGE patterns within and between laboratories(Swaminathan et al., 2001).

Despite its many advantages, the interpreta-tion of PFGE patterns in the context of foodbornedisease epidemiology is not simple and there iscritical need for guidelines for interpretation ofthese data. The widely cited Tenover criteria(Tenover et al., 1995) were originally developedas guidelines for the investigation of nosocomialoutbreaks but have been universally applied.These criteria may not be appropriate for appli-cation to highly clonal enteric pathogens such asSalmonella and E. coli. Further, point mutationswere thought to be one of the major contribut-ing factors to PFGE pattern diversity at the timethe Tenover criteria were developed. However,Kudva et al. (2002) have demonstrated that thePFGE patterns diversity in E. coli O157:H7 is pri-marily attributable to insertions and deletionsand not to point mutations. Based on our ex-tensive experience with PFGE pattern analysisof E. coli O157:H7, Salmonella, Shigella and Liste-ria monocytogenes, we have developed guidelinesfor the interpretation of PFGE patterns of theseenteric pathogens in the context of foodbornedisease surveillance, disease cluster recognitionand outbreak investigations. We hope that theseguidelines will serve to stimulate discussion ofthis topic.

INTERPRETATION OF PFGE RESULTS

First consideration: are observed differences real?

Even when a carefully standardized proce-dure for PFGE is followed, artifacts may occurwhich could lead to erroneous conclusionsabout the relationship between profiles. It istherefore important to know the nature of theseartifacts in order to recognize and correct them.Artifacts in this context are banding patternsthat can not be reproduced on subsequent test-ing in the same or in a different laboratory.When the results of the subtyping of strains gen-erated by a particular method are not repro-ducible, there are two general explanations, onerelating to the subtyping procedure and theother relating to the strain in question. The lat-

ter depends on the stability of the typingmethod, and is simply a reflection of the factthat bacterial strains change over time. If strainschange quickly enough, different subcultures ofthe same strain may show different subtypes(Iguchi et al., 2002). On the other hand, differ-ences due to the subtyping procedure usuallymanifest as artifacts introduced by small, oftenunrecognized differences in the subtyping pro-cedure. Sometimes these small differences inprocedure manifest as subtle differences in bandseparation (e.g., one thick band versus two thin-ner bands), but with PFGE the most common ar-tifact is probably incomplete restriction.

When DNA is restricted, the restriction en-zyme will ideally cleave the DNA every timeit meets its recognition sequence, cutting theDNA in a well-defined number of pieces ofequally well defined size. If the restriction is in-complete, some DNA molecules will be com-pletely restricted whereas others will not becleaved at all the recognition sequences. Thiswill result in the generation of more fragments,some with the expected size, and some largerfragments containing one or more restrictionsites that were not cleaved. In a PFGE gel, thiswill show up as additional bands, usually atthe top of the gel. However, if the incompleterestriction is little pronounced, as may be thecase if a restriction site is methylated (McClel-land et al., 1994), this may result in the addi-tion of a single or a few extra bands. Bands pre-sent due to incomplete restriction are typicallyof a weaker intensity than their neighbors(“ghost bands”). The presence of such ghostbands should always trigger a repeated re-striction and analysis of the isolate in question.

Another common error causing lack of re-peatability is working with cultures that are notpure. If a culture contains more than one strain,there is obviously a risk of picking differentstrains when selecting single colonies for re-peating subtyping of the same culture. In caseof swarming bacteria such as Campylobacter, itmay be difficult to obtain a pure culture or picka single colony. In this case, the subtyping resultswill be completely unpredictable and repeatedsubtyping of such cultures will almost invariablylead to different results. Media or culture condi-tions preventing swarming should be used forobtaining a pure culture.

BARRETT ET AL.22

If an organism causing an outbreak is part ofthe normal flora of humans, animals, or food,there is always a risk of overlooking the out-break strain in a particular specimen if multi-ple strains are present but only one colony isinvestigated. This may be the case with Listeriamonocytogenes, which is widely distributed inthe environment and in food (Tham et al., 2000,2002); the solution to this problem is to pickmultiple colonies from the primary cultures forsubtyping. Organisms such as Clostridium per-fringens, which may be either a normal com-mensal or cause disease if the strain producesenterotoxin, present another problem. For suchorganisms, only colonies producing entero-toxin should be selected for further subtypingin an outbreak setting (Lukinmaa et al., 2002).

Tenover criteria and sources of variability inPFGE patterns

In 1995, Tenover et al. published a paper onthe interpretation of PFGE profile differences.According to the authors, profiles differingfrom each other in the position of up to threebands should be considered closely related andprofiles differing by up to six bands should beconsidered possibly related. The reasoning be-hind this recommendation is that a single ge-netic event (a point mutation in a restrictionsite, a deletion or an insertion) would result inup to three-band differences. Thus, a three-band difference would be the result of one ge-netic event and a six-band difference the resultof two genetic events. A point mutation in a re-striction site will lead to the loss of that site,and the two original fragments with the re-striction site will merge to one larger fragment,leading to a three-band difference (loss of twofragments, gain of one). Conversely, if a re-striction site is gained by a point mutation, alarge fragment is split into two smaller frag-ments, also resulting in a three-band difference.Deletions and insertions will show as a changein the size of one band leading to a two-banddifference (loss of a band of one size, gain of aband of another). If the insertion/deletion con-tains a restriction site, additional fragmentsmay be gained or lost relative to the pattern ofthe strain without the insertion/deletion.

Over the past 10 years, the Tenover criteria

have proven to be useful, especially when com-paring profiles of nosocomial pathogens and insuspected settings of ongoing transmission. Forexample, Aucken et al. (2000) compared PFGEpatterns obtained with XbaI digestion of ge-nomic DNA from 28 unrelated isolates of Ser-ratia marcescens and 29 isolates from two out-breaks. Twenty-six of the 28 unrelated isolateshad unique profiles with more than 10-banddifferences. Twenty-seven of the 29 outbreakisolates could be assigned to the correct out-break based on PFGE results, with most iso-lates differing by less than three bands. Threeisolates differed by five to seven bands.

As useful as the Tenover criteria have beenin many circumstances, they are clearly not ap-plicable to all situations. The criteria are verygeneral and assume that all genomic fragmentsare visible in the gel and that the plasmid con-tent in the strains is stable, at least for the plas-mids visible in the gel. If these assumptionswere always correct, all profiles would eitherbe identical or differ from each other in the po-sition of at least two bands (a change in bandsize being interpreted as loss of one band andgain of another). It is clear that the plasmid con-tent of bacterial cells is not stable; plasmids areoften gained or lost. PFGE profiles differingfrom each other by a single band are common,at least in foodborne pathogens. Thus, these as-sumptions are not always sound.

One potential cause of a one band differenceis the presence of an extra plasmid in one ofthe strains. Plasmids may be visible in a PFGEgel, often as intense bands. This intensity maybe due to the presence of multiple copies of theplasmid versus the single copy of the chromo-some. If plasmids are digested by the restric-tion enzyme used, the resulting fragments willmigrate as a function of their size in a PFGEgel, just like the chromosomal fragments. Forexample, in XbaI digested preparations of Sal-monella Enteritidis, the 57-kb virulence plasmidis visible in the gel and strains lacking this plas-mid may differ from strains harboring thisplasmid by just this single band (Buchrieser etal., 1994; Powell et al., 1994). However, if theplasmids are not in linear conformation, theirmigration is unpredictable. Large undigestedplasmids will typically move very slowly andmay not be visible in a PFGE gel, but they may

INTERPRETATION OF PFGE IN FOODBORNE DISEASE INVESTIGATIONS 23

also be visible almost anywhere in the gel(Buchrieser et al., 1994; Barton et al. 1995).

Because of the variability in how plasmidsmay migrate, it has been suggested that frag-ments of a size less than 125 kb should be ex-cluded from analysis (Olsen et al., 1994). Eventhis drastic measure may not solve the prob-lem, however, as plasmids may appear in ahigher size range than 125 kb, either becausethey actually are that large (“megaplasmids”)or because they have not been digested by therestriction enzyme used and are not in a linearconformation (Barton et al., 1995). It is well-known that plasmids, including megaplas-mids, are common in foodborne pathogens(Bradbury et al., 1983; Wachsmuth et al., 1991;Olsen et al., 1994; Barton et al., 1995; Iteman etal., 1996; McLauchlin et al., 1997), but the in-fluence of the plasmid content on the PFGEprofiles of foodborne pathogens has not beenwell characterized.

A second cause of one-band differences inPFGE patterns is the possibility that one ormore of the fragments generated by deletion,insertion, or point mutation are so small thatthey migrate off the gel. On the other hand, itshould also be kept in mind that even changesaffecting larger fragments will not always bereadily apparent. Only comparatively largedeletions or insertions will be detectable byPFGE. Band size differences of 1–2% (0.5–1 kbat the bottom and �10–15 kb at the top of thegel) are not detectable under most conditions.It has been shown that changes in PFGE pro-files of Shiga toxin–producing E. coli O157 arecaused primarily by deletions and insertionsand not by point mutations in the restrictionsites (Kudva et al., 2002). Little is known aboutthe cause of changes in the PFGE profiles ofother foodborne pathogens. Even if point mu-tations are common, the rareness of restrictionsites for enzymes used in PFGE would dictatethat only a tiny fraction of mutations would beexpected to occur within a specific restrictionsite and thus change the PFGE pattern. It seemslikely that in many, if not most cases, severalgenetic events have likely occurred before thePFGE pattern is visibly altered, and the Ten-over criteria represent a conservative estimateof the number of genetic events behind changesin PFGE profiles.

Interpreting differences in the context oforganism diversity

Optimal interpretation of the differences (orlack thereof) between the PFGE patterns of twoisolates depends largely on the variability of the organism being typed. Some organisms arehighly clonal while others demonstrate extremevariability. Helicobacter pylori, for example, showsso much diversity that molecular subtyping isused primarily to determine if an infection is newor recurring, and contributes little to the under-standing of H. pylori epidemiology (Taylor et al.,1995). Even within a genus, species or serotypes

BARRETT ET AL.24

FIG. 1. Pulsed-field gel electrophoresis (PFGE) patternsof S. Baildon (A) and S. Bareilly (B) in the PulseNet USAuntil 2000.

may vary greatly in genetic diversity. Dendro-grams of PFGE patterns of two Salmonellaserotypes from the PulseNet database are shownin Figure 1. As of 2000, S. Baildon isolates wereall highly similar (Fig. 1a), while S. Bareilly iso-lates were highly diverse (Fig. 1b). MatchingPFGE patterns for isolates of S. Bareilly wouldthus be more likely to indicate epidemiologic re-latedness than matching patterns for S. Baildon,while minor differences between S. Baildon pat-terns would be more significant than the samedifferences in S. Bareilly patterns. Recent S. Bail-don isolates have begun to show more diversity(Fig. 2), but the primary pattern still constitutesmost of the database. S. Bareilly patterns con-tinue to be highly diverse.

In addition to the diversity of subtypes of theorganism being tested, one must also considerthe prevalence of the particular subtype or pat-tern. A recent study in our laboratory foundconsiderable diversity in S. Heidelberg PFGEpatterns, but 33/60 (55%) apparently unrelatedisolates had the same PFGE pattern (Kubota etal., 2000). This pattern represented about halfof all S. Heidelberg isolates in our collectionfrom each time period sampled dating back to1985. The remaining 27 isolates were dividedamong 21 PFGE patterns. In the absence of epi-demiologic information, one would have toconclude that isolates sharing one of the lesscommon patterns are more likely have comefrom a common source than isolates sharing thepredominant pattern.

Ongoing transmission versus a point sourceoutbreak: interpreting differences in different settings

Time is also a critical factor in interpretingtyping results. The shorter the duration of an

outbreak, the less time the outbreak strain hasto undergo mutations that change the PFGEpattern. Foodborne disease outbreaks often re-sult from a single contamination event whereall of the patients are exposed to the same con-taminated food at the same time, and variabil-ity among isolates from patients is expected tobe minimal. Hospital or community outbreaks,on the other hand, are frequently prolonged,with strains being passed from person to per-son with much more opportunity for muta-tional changes in PFGE patterns. An outbreakof Shigella flexneri serotype 2a in Taiwan town-ship in 1996 illustrates both situations (Chen etal., 2003).

During the month of August, when the out-break was identified, all outbreak-related iso-lates had the same PFGE pattern. By the end ofDecember, eight variants with differences ofthree or fewer bands were detected. By the endof 2000, a total of 50 variations of the originalpattern had been seen. A similar situation wasreported during a large person-to-person out-break of S. sonnei in Massachusetts in 1999(Hackbarth et al., 2000). The outbreak was rec-ognized in June, when isolates from three caseswere found to have indistinguishable PFGEpatterns. A second pattern had emerged byJuly, and by December of that year, 21 differ-ent PFGE patterns had been seen in isolatesfrom persons who could be epidemiologicallyconnected.

In a multistate outbreak of listeriosis in theUnited States in 1998, outbreak-associated iso-lates included variants of those with the majorPFGE pattern isolated from patients and theepidemiologically implicated food product. Of108 cases identified as part of this outbreak, 77had L. monocytogenes with the major outbreakpattern, 29 were infected with a variant that dif-fered from the major outbreak strain by the ab-sence of an 80-kb ApaI restriction fragment, andtwo were infected with a variant that differedby the absence of the 80-kb fragment and thepresence of a 190-kb fragment. Epidemiologicdata strongly associated all three types with theoutbreak. All three types and an additionalvariant (containing the 190-kb fragment) werefound in foods obtained from the implicatedfood processing facility. The investigators hy-pothesized that the major strain may have been

INTERPRETATION OF PFGE IN FOODBORNE DISEASE INVESTIGATIONS 25

FIG. 2. Pulsed-field gel electrophoresis (PFGE) patternsof S. Baildon in the PulseNet database as of 2005.

a resident of the processing plant for a dura-tion long enough to produce the genetic vari-ants (Graves, 2005).

In contrast to these descriptions of ongoingtransmission, outbreaks with a point sourceand little secondary transmission are typicallyyield isolates with a more limited range of pat-terns. Two other Shigella outbreaks illustratethis point. The first outbreak occurred in Janu-ary, 2000 and involved a contaminated foodproduct (bean dip) (CDC, 2000). The subtypingresults were typical of a foodborne (pointsource) outbreak. S. sonnei isolates from 25 per-sons and two foods were available for subtyp-ing. Isolates from 21 patients and both foodswere indistinguishable by PFGE, and the otherisolates had highly similar PFGE patterns. Asecond outbreak occurred in July and Augustof 1998, and was due to consumption of cont-aminated parsley at restaurants in the UnitedStates and Canada (Naimi et al., 2003). Seventy-seven percent of the isolates had one of twoclosely related PFGE patterns. These outbreaksillustrate the impact of time and transmissionon PFGE patterns, and emphasize the impor-tance of the outbreak setting in interpretingPFGE results.

Lab results do not tell the whole story

Perhaps the most important factor to con-sider in interpreting PFGE results is context:how the laboratory results fit together with epi-demiologic and environmental investigations.The simple fact that common PFGE patterns ex-ist for many organisms demonstrates that in-distinguishable PFGE patterns alone do notunequivocally demonstrate an epidemiologicconnection. Even if PFGE patterns could provethat two isolates were the same strain, it is al-ways possible that two people could have beeninfected with the same strain by differentroutes.

It is perhaps less obvious, but equally truethat clearly differing patterns do not prove thatisolates are epidemiologically unrelated. Atleast four distinct PFGE patterns (differing bythree or more bands) were found among E. coliO157:H7 isolates from persons who ate at thesame restaurant in Seattle over the course of a

few days in 1993 (Jackson et al., 2000). Acase/control study implicated salad bar items,apparently due to cross-contamination fromraw beef in the restaurant kitchen. If epidemi-ologists had relied solely on molecular subtyp-ing data rather than patient interviews and en-vironmental observations of the restaurant, thesource of this outbreak would likely have beenunidentified.

WATERBORNE OUTBREAKS OF PATHOGENS USUALLY ASSOCIATED WITH FOOD

Contaminated drinking water has thepropensity to cause large outbreaks of diar-rheal disease (Hrudy et al., 2003). Molecularsubtyping may have less utility in the early de-tection of some outbreaks caused by contami-nation of potable water supplies with diarrhealbacterial pathogens if the outbreaks have a very sharp epidemic curve resulting from largenumbers of people in a geographically local-ized area being exposed to the infectingpathogens over a very short window of time.These outbreaks are easily detected by sharpincreases in the number of patients seekingmedical attention or even by the rapid deple-tion of over-the-counter anti-diarrheal medica-tions in stores in a specific area (Hrudy et al.,2003; CDC, 1999; Mac Kenzie et al., 1994). Nev-ertheless, molecular subtyping has providedcritical information for the investigation of waterborne outbreaks and for tracking theirsources.

In an outbreak in Walkerton, Canada, a sin-gle E. coli O157:H7 strain was isolated from thecase-patients and from the farm that was iden-tified as the source of contamination. In con-trast, four SmaI PFGE types of C. jejuni wereisolated from outbreak-associated case-patients(Clark et al., 2003). The four outbreak-associ-ated PFGE patterns shared a common back-bone and differed in the presence or absence offour DNA fragments in the middle portion ofthe DNA “fingerprint.” Another large dual-pathogen outbreak of waterborne disease oc-curred in upstate New York following a countyfair in 1999. Once again, E. coli O157 and C. je-

BARRETT ET AL.26

juni were the two pathogens that were in-volved. However, in this outbreak, PFGEanalysis of E. coli O157:H7 showed that 65% ofthe patient isolates, and 63% of the isolatesfrom the implicated water source shared thesame PFGE profile. An additional 27% of thepatient isolates were represented by a PFGEprofile that differed from the major pattern bytwo bands. The remaining isolates were alsovariants of the major PFGE profile with 1–3-band differences from the major pattern (Boppet al., 2003). These data suggest that the prog-enitor of the outbreak strain may have per-sisted on the farm (from which cow manurewas suspected to have leached into the wellthat served as the source of water for the countyfair) long enough to diverge into variants ob-served by PFGE typing. Changes in E. coliO157:H7 PFGE patterns over time have beendemonstrated in vitro (Iguchi et al., 2002).

Interestingly, there was much less genetic di-versity among the C. jejuni isolates (83% of theisolates had the same PFGE profile) associatedwith this outbreak when compared with theWalkerton, Canada outbreak. It is possible thatthe C. jejuni may have contaminated the watersource from a source other than the farm be-cause no C. jejuni was recovered from any ofthe samples taken from that farm.

A waterborne salmonellosis outbreak of S.Bareilly infections in 95 persons spread across10 states highlights another aspect of inter-pretation of PFGE data in the context of epi-demiologic information. PFGE subtyping ofthe S. Bareilly isolates from case-patients re-vealed numerous PFGE patterns. However,the epidemiologic investigation revealed thatthe infections were associated with drinkingwater from private wells and springs in thesoutheastern United States and from drinkingbottled water from a bottling operation thatused water from a spring in that area (Epi-AidReport 00-61; CDC, unpublished data). Be-cause S. Bareilly is an amphibian-associatedserotype, investigators hypothesized that mul-tiple S. Bareilly strains may have persisted inamphibian populations in and around the dif-ferent water sources to confound the inter-pretation of the PFGE patterns associated withthis outbreak.

HOW DIFFERENT SHOULD PFGE PATTERNS BE

CONSIDERED “DIFFERENT”?

This simplest and probably most useful wayto answer this question is to consider patternsthat show any discernable differences to be dif-ferent patterns. This is the approach thatPulseNet has chosen. In point source outbreaks(typical of most foodborne outbreaks) the vastmajority of outbreak related isolates display thesame profile (Mølbak et al., 1999; Naimi et al.,2003). In investigating recognized outbreaks, itis fairly easy and often useful to designate pat-terns that differ slightly from the primary out-break pattern as “subpatterns” or “related pat-terns.” In surveillance for cluster detection,such an approach quickly becomes unwieldyand could easily be misleading. A recent multi-state outbreak of E. coli O157 illustrates the ra-tionale of this approach (Gerner-Smidt et al.,2005).

In the summer of 2002, a cluster of cases ofE. coli O157 infections was detected in Col-orado. It soon became evident that cases wereoccurring in other states. In accordance with es-tablished PulseNet protocol, only patients withisolates displaying patterns indistinguishablefrom the original outbreak pattern were in-cluded in the initial case interviews. Thirty-seven isolates of the outbreak pattern were up-loaded to the PulseNet server during theoutbreak period. During the same time period,forty-one isolates of a frequently encounteredpattern that differed from the outbreak patternby a single band were also uploaded. The case-control study pointed to ground beef from oneparticular manufacturer as the source of theoutbreak. If the interviews had included casesinfected with the related, frequently encoun-tered strain in the case definition, the results ofthe case-control study might have been lessconclusive. While some of the patients infectedwith isolates having the common, non-out-break pattern may have actually acquired theirinfection from eating meat from the implicatedproducer, the fact that this pattern was com-monly seen before the outbreak argues thatmost probably did not. Including these patientsin the case interviews would have made iden-

INTERPRETATION OF PFGE IN FOODBORNE DISEASE INVESTIGATIONS 27

tifying the source of the outbreak much moredifficult. At the same time, epidemiologic evi-dence supported the inclusion of two siblingsas part of the outbreak, even though their iso-lates had the frequently encountered (non-out-break) pattern. While it has been well docu-mented that a strain may undergo changesduring an outbreak causing minor differencesin the number or position of the DNA frag-ments in a PFGE gel (Proctor et al., 2002), including variants in the case definition may complicate the interview studies of theoutbreak, especially if one of the variants represents a common pattern. This example illustrates the complexities involved in inter-pretation of outbreak investigation data andunderscores the importance of considering allthe available evidence in assessing strain relat-edness.

AVOIDING “FALSE ALARMS” INCLUSTER DETECTION

If subtyping is to be useful for cluster detec-tion and alerting public health authorities topotential outbreaks, there must be a balance be-tween reporting too many clusters, many ofwhich will prove to be “false alarms,” and fail-ing to identify real outbreak possibilities. Oneof the advantages of PulseNet is that the data-bases are large and contain data many years ofdata. These data provides a historical baselineagainst which possible clustered cases may becompared. The cluster detection level for acommon profile should be higher than for arare or unique profile to avoid reacting to ac-cumulations of common profiles that are to beexpected to occur by chance only (“falsealarms”). It also helps to identify organismswhere PFGE is not sufficiently discriminatoryand where alternative subtyping methodsshould be applied in case clusters are detected.

While cluster profiles should be comparedwith the full database to determine the fre-quency of the observed pattern, the time factorshould also be considered. The dynamics ofsubtypes are constantly changing, and a pat-tern that is common in the database as a wholemay not have been common for several years.

Such a pattern would suggest reacting to alower number of reports than if the pattern oc-curred evenly over the past several years.

RECOMMENDATIONS FORINTERPRETING PFGE RESULTS

When any differences in PFGE patterns areobservable, the patterns should be reported asdifferent. Likewise, patterns that are indistin-guishable should be reported as such. Indistin-guishable is a more accurate term than “identi-cal,” since it implies only that any differencesare not observable under the conditions used.Interpretation of PFGE results, however, shouldinclude each of the following steps.

1. The gel should be of sufficient quality to beproperly interpreted. A gel that includespartial digestions or other artifacts, or inwhich the bands are not sharp and clear,should be re-run without attempting to in-terpret the results. Interpreting a poor qual-ity gel can be misleading, and can actuallybe worse than having no results at all.

2. Consider the diversity of the organism be-ing tested. Large databases such as thosemaintained by PulseNet should provide suf-ficient data to make reasonable determina-tions of diversity. If the organism displayslittle diversity, one should be cautious in as-suming that closely related patterns, or evenindistinguishable patterns, indicate a highlikelihood of a common source. PFGE usinga second enzyme may be useful. Other avail-able laboratory data such as phage type, bio-chemical profile, antimicrobial susceptibilityprofile, or multilocus sequence type shouldalso be considered. If the organism beingtested shows substantial diversity, one muststill consider whether there are clonal pop-ulations within a non-clonal organism, as isthe case with S. Heidelberg. If the PFGE pat-tern in question matches the pattern of alarge clonal group, the overall diversity ofthe serotype is of little relevance and addi-tional data is needed to interpret the PFGEresults. On the other hand, when an organ-ism demonstrates extreme variability, such

BARRETT ET AL.28

as that seen with H. pylori, any patternmatches are probably significant.

3. After establishing the diversity of the or-ganism and whether or not there are clonalpopulations, one should then consider theoutbreak setting. If it appears to be a pointsource outbreak without continued trans-mission, only very minor differences arelikely to be observed and isolates that differby several bands are probably not part of theoutbreak (unless it is a multi-strain out-break). When there is ongoing transmission,such as a community outbreak of shigellosis,more variability should be expected. Isolatesthat are temporally or geographically re-lated, and have PFGE patterns that are moresimilar to patterns known to be outbreak-as-sociated than to other patterns in the data-base, may be considered potentially out-break-related and worth further laboratorystudy or epidemiologic investigation.

CONCLUSION

Bacterial subtyping has proven useful foroutbreak investigations and cluster detectionin so many settings over the years that its util-ity is seldom questioned. However, the inter-pretation of subtyping results can still proveproblematic. It can not be overemphasizedthat proper interpretation can only be made incontext. The reproducibility of the subtypingmethod, quality of the PFGE gel, variability ofthe organism being subtyped, and prevalenceof a particular pattern are all critical factors.Perhaps the most important factor, however,is how the laboratory data fit with the epi-demiologic and environmental information.PFGE results alone can not establish an epi-demiologic connection between isolates. All ofthe available information must be consideredin order to interpret subtyping results appro-priately.

REFERENCES

Aucken, H.M., T. Boquete, M.E. Kaufmann, and T.L. Pitt.2000. Interpretation of band differences to distinguish

strains of Serratia marcescens by pulsed-field gel elec-trophoresis of XbaI DNA digests. Epidemiol. Infect.125:63–70.

Barton, B.M., G.P. Harding, and A.J. Zuccarelli. 1995. Ageneral method for detecting and sizing large plasmids.Anal. Biochem. 226:235–240.

Bopp, D.J., B.D. Sauders, A.L. Waring, J. Ackelsberg, N.Dumas, E. Braun-Howland, D. Dziewulski, B.J. Wallace,M. Kelly, T. Halse, K.A. Musser, P.F. Smith, D.L. Morse,and R.J. Limberger. 2003. Detection, isolation, and mole-cular subtyping of Escherichia coli O157:H7 and Campy-lobacter jejuni associated with a large waterborne out-break. J. Clin. Microbiol. 41:174–180.

Bradbury, W.C., M.A. Marko, J.N. Hennessy, and J.L. Pen-ner. 1983. Occurrence of plasmid DNA in serologicallydefined strains of Campylobacter jejuni and Campylobactercoli. Infect. Immun. 40:460–463.

Buchrieser, C., S.D. Weagant, and C.W. Kaspar. 1994. Mol-ecular characterization of Yersinia enterocolitica by pulsed-field gel electrophoresis and hybridization of DNA frag-ments to ail and pYV probes. Appl. Environ. Microbiol.60:4371–4379.

CDC (Centers for Disease Control and Prevention). 1999.Public health dispatch: outbreak of Escherichia coliO157:H7 and Campylobacter among attendees of the Wash-ington County Fair—New York, 1999. Morbid. Mortal.Wkly. Rep. 48:803–805.

CDC (Centers for Disease Control and Prevention). 2000.Outbreak of Shigella sonnei infections associated with eat-ing a nationally distributed dip—California, Oregon, andWashington. Morbid. Mortal. Wkly. Rep. 49:60–61.

Chen, J.-H., C.-S. Chiou, P.-C. Chen, T.-L. Liao, T.-L. Liao,J.-M. Li, and W.-B. Hsu. 2003. Molecular epidemiology ofShigella in a Taiwan township during 1996 to 2000. J. Clin.Microbiol. 41:3078–3088.

Clark, C.G., L. Price, R. Ahmed, D.L. Woodward, P.L.Melito, F.G. Rogers, F. Jamieson, B. Ciebin, A. Li, and A.Ellis. 2003. Characterization of waterborne outbreak–as-sociated Campylobacter jejuni, Walkerton, Ontario. Emerg.Infect. Dis. 9:1232–1241.

Gerner-Smidt P., J. Kincaid, K. Kubota, K. Hise, S.B.Hunter, M.-A. Lambert-Fair, D. Norton, A. Woo-Ming, T.Kurzynski, K. Holt, and B. Swaminathan. 2005. Molecu-lar surveillance of Shiga-toxigenic Escherichia coli O157 byPulseNet USA in 2002. J. Food Protect. 68:1926–1931.

Graves, L.M., S.B. Hunter, A.R. Ong, D. Schoonmaker-Bopp, K. Hise, L. Kornstein, W.E. DeWitt, P.S. Hayes, E.Dunne, P. Mead, and B. Swaminathan. 2005. Microbio-logical aspects of the investigation that traced the 1998outbreak of listeriosis in the United States to contami-nated hot dogs and establishment of a molecular subtyp-ing-based surveillance for Listeria monocytogenes in thePulseNet network. J. Clin. Microbiol. 43:2350–2355.

INTERPRETATION OF PFGE IN FOODBORNE DISEASE INVESTIGATIONS 29

Hackbarth, A., A. Goddard, P. Kludt, B. Bolstroff, C.Rauch, J. Jaques, and K. Liptak. 2000. Abstr. Int. Conf.Emerg. Inf. Dis.

Hrudey, S.E., P. Payment, P.M. Huck, R.W. Gillham, andE.J. Hrudey. 2003. A fatal waterborne disease epidemic inWalkerton, Ontario: comparison with other waterborneoutbreaks in the developed world. Water Sci. Technol.47:7–14.

Hutwagner, L.C., E.K. Maloney, N.H. Bean, L. Slutsker,and S.M. Martin. 1997. Using laboratory-based surveil-lance data for prevention: an algorithm for detecting Sal-monella outbreaks. Emerg. Infect. Dis. 3:395–400.

Iguchi, A., R. Osawa, J. Kawano, A. Shimizu, J. Tera-jima, and H. Watanabe. 2002. Effects of repeated sub-culturing and prolonged storage at room temperatureof enterohemorrhagic Escherichia coli O157:H7 onpulsed-field gel electrophoresis profiles. J. Clin. Micro-biol. 40:3079–3081.

Iteman, I., A. Guiyoule, and E. Carniel. 1996. Comparisonof three molecular methods for typing and subtypingpathogenic Yersinia enterocolitica strains. J. Med. Micro-biol. 45:48–56.

Jackson, L.A.W.E. Keene, J.M. McAnulty, E.R. Alexander,M. Diermayer, M.A. Davis, K. Hedberg, J. Boase, T.J. Bar-rett, M. Samadpour, and D.W. Fleming. 2000. Where’s thebeef? The role of cross-contamination in four chain restau-rant-associated outbreaks of Escherichia coli O157:H7 inthe Pacific Northwest. Arch. Intern. Med. 160:2380–2385.

Keene, W.E. 1999. Lessons from investigations of food-borne disease outbreaks. JAMA 281:1845–1847.

Kubota, K., E.M. Ribot, N. Marano, and T.J. Barrett. 2000.Abstr. 100th Gen. Meet. Am. Soc. Microbiol. 2000 A-144.

Kudva, I.T., P.S. Evans, N.T. Perna, T.J. Barrett, F.M.Ausubel, F.R. Blattner, and S.B. Calderwood. 2002. Strainsof Escherichia coli O157:H7 differ primarily by insertionsor deletions, not single-nucleotide polymorphisms. J. Bac-teriol. 184:1873–1879.

Levin, B.R., M. Lipsitch, and S. Bonhoeffer. 1999. Popula-tion biology, evolution, and infectious disease: conver-gence and synthesis. Science 283:806–809.

Lukinma, S., E. Takkunen, and A. Siitonen. 2002. Molec-ular epidemiology of Clostridium perfringens related tofood-borne outbreaks of disease in Finland from 1984 to1999. Appl. Environ. Microbiol. 68:3744–3749.

MacKenzie, W.R., N.J. Hoxie, M.E. Proctor, M.S. Gradus,K.A. Blair, D.E. Peterson, J.J. Kazmierczak, D.G. Addiss,K.R. Fox, J.B. Rose, and J.P. Davis. 1994. A massive out-break in Milwaukee of Cryptosporidium infection trans-mitted through the public water supply. N. Engl. J. Med.331:161–167.

McClelland, M., M. Nelson, and E. Raschke. 1994. Effectof site-specific modification on restriction endonucleasesand DNA modification methyltransferases. Nucleic AcidsRes. 22:3640–3659.

McLauchlin, J., M.D. Hampton, S. Shah, E.J. Threlfall, A.A. Wieneke, and G. D. Curtis. 1997. Subtyping of Listeriamonocytogenes on the basis of plasmid profiles and arsenicand cadmium susceptibility. J. Appl. Microbiol. 83:381–388.

Mølbak, K., D.L. Baggesen, F.M. Aarestrup, J.M. Ebbesen,J. Engberg, K. Frydendahl, P. Gerner-Smidt, A.M. Petersen,and H.C. Wegener. 1999. An outbreak of multidrug-resis-tant, quinolone-resistant Salmonella enterica serotype Ty-phimurium DT104. N. Engl. J. Med. 341:1420–1425.

Naimi, T.S., J.H. Wicklund, S.J. Olsen, G. Krause, J.G.Wells, J.M. Bartkus, D.J. Boxrud, M. Sullivan, H. Kassen-borg, J.M. Besser, E.D. Mintz, M.T. Osterholm, and C.W.Hedberg. 2003. Concurrent outbreaks of Shigella sonneiand enterotoxigenic Escherichia coli infections associatedwith parsley: implications for surveillance and control offoodborne illness. J. Food. Prot. 66:535–541.

Olsen, J.E., M.N. Skov, E.J. Threlfall, and D.J. Brown. 1994.Clonal lines of Salmonella enterica serotype Enteritidis doc-umented by IS200-, ribo-, pulsed-field gel electrophoresisand RFLP typing. J. Med. Microbiol. 40:15–22.

Powell, N.G., E.J. Threlfall, H. Chart, and B. Rowe. 1994.Subdivision of Salmonella enteritidis PT 4 by pulsed-fieldgel electrophoresis: potential for epidemiological surveil-lance. FEMS Microbiol. Lett. 119:193–198.

Proctor, M.E., T. Kurzynski, C. Koschmann, J.R. Archer,and J.P. Davis. 2002. Four strains of Escherichia coliO157:H7 isolated from patients during an outbreak of dis-ease associated with ground beef: Importance of evaluat-ing multiple colonies from outbreak-associated product.J. Clin. Microbiol. 40:1530–1533.

Swaminathan, B., T.J. Barrett, S.B. Hunter, and R.V.Tauxe. 2001. PulseNet: the molecular subtyping networkfor foodborne disease surveillance, United States. Emerg.Infect. Dis. 7:382–389.

Taylor, N.S., J.G. Fox, N.S. Akopyants, D.E. Berg, N.Thompson, B. Shames, L. Yan, E. Fontham, F. Janney, andF.M. Hunter. 1995. Long-term colonization with singleand multiple strains of Helicobacter pylori assessed byDNA fingerprinting. J. Clin. Microbiol. 33:918–923

Tenover, F.C., R.D. Arbeit R.V. Goering, P.A. Mickelsen,B.E. Murray, D.H. Persing, and B. Swaminathan. 1995. In-terpreting chromosomal DNA restriction patterns pro-duced by pulsed-field gel electrophoresis: criteria for bac-terial strain typing. J. Clin. Microbiol. 33:2233–2239.

Tham, W., J. Alden, H. Ericsson, S. Helmersson, B.Malmodin, O. Nyberg, A. Pettersson, H. Unnerstad, andM.L. Danielsson-Tham. 2002. A listeriosis patient infectedwith two different Listeria monocytogenes strains. Epi-demiol. Infect. 128:105–106.

Tham, W., H. Ericsson, S. Loncarevic, H. Unnerstad, andM.L. Danielsson-Tham. 2000. Lessons from an outbreakof listeriosis related to vacuum-packed gravad and cold-smoked fish. Int. J. Food Microbiol. 62:173–175.

Wachsmuth, I.K., J.A. Kiehlbauch, C.A. Bopp, D.N.Cameron, N.A. Strockbine, J.G. Wells, and P.A. Blake.

BARRETT ET AL.30

1991. The use of plasmid profiles and nucleic acid probesin epidemiologic investigations of foodborne, diarrhealdiseases. Int. J. Food Microbiol. 12:77–89.

Wegener, H.C., T. Hald, D.L. Wong, M. Madsen, H. Ko-rsgaard, F. Bager, P. Gerner-Smidt, and K. Molbak. 2003.Salmonella control programs in Denmark. Emerg. Infect.Dis. 9:774–780.

Address reprint requests to:Dr. Timothy J. Barrett

Centers for Disease Control and Prevention1600 Clifton Rd., Mailstop C03

Atlanta, GA 30333

E-mail: tbarrett@cdc.gov

INTERPRETATION OF PFGE IN FOODBORNE DISEASE INVESTIGATIONS 31