The Princeton Guide to Evolution

VIII.4

Evolution and Microbial Forensics

Paul Keim and Talima Pearson

OUTLINE

1. Evolutionary thinking, molecular epidemiology, and microbial forensics

2. The uses of DNA in human and microbial forensics

3. Genetic technology and the significance of a “match”

4. The Kameido Aum Shinrikyo anthrax release

5. The Ames strain and the 2001 anthrax letters

6. From molecular epidemiology to microbial forensics and back

The tools of molecular biology coupled with the evolutionary methods of phylogenetics have found powerful applications in tracking the origins and spread of infectious diseases. Microbial forensics is a new discipline focused on identifying the source of the infective material involved in a biological crime and it, too, increasingly depends on evolutionary analysis and molecular genetic tools.

GLOSSARY

Clonal Populations. Populations in which members, called clones, have diverged without exchanging any genetic material across lineages. Members of such populations (e.g., many recently emerged pathogens) are genetically identical with the exception of variation generated by subsequent mutations.

Homoplasy. A shared genetic (or phenotypic) characteristic produced by convergent evolution or horizontal genetic exchange between lineages, rather than by descent from a common ancestor that shared the same characteristic.

Match. An identical genotypic profile (often called a DNA fingerprint) based on a particular technology.

Membership. A phylogenetic concept more useful than a “match” for describing relationships among bacterial isolates. Two isolates can be members of the same phylogenetic group without being absolutely identical in their genome sequences.

Monophyletic. A phylogenetic term referring to all descendants of a common ancestor.

Multiple Loci VNTR Analysis (MLVA). A DNA fingerprinting method widely used to differentiate bacterial types. Here, VNTR stands for variable number of tandem repeats.

Single Nucleotide Polymorphism (SNP). A single base-pair difference between the DNA sequences of two individuals including, for example, two closely related bacterial strains.

The investigation of infectious disease outbreaks has a long history and even predates our understanding of the germ theory of disease, which was formulated by Louis Pasteur in the early 1860s. The classic example, a seminal event in epidemiology, occurred in 1854 when John Snow implicated London’s Broad Street water pump as the focus of a cholera outbreak. The correlative association of disease occurrence, potential causative infectious agents, and their sources has grown increasingly sophisticated over the years. Today it is common to examine the genomes of bacteria and viruses to precisely define the pathogen subtype, with the aim of identifying specific case clusters that can reasonably be presumed to be a part of the same outbreak. This approach strengthens any correlative study that aims to identify the disease source by eliminating similar disease cases that did not emanate from the same focus.

These same genomic methods became important after the bioterrorism events of October 2001, when letters laden with Bacillus anthracis spores were sent through the US Postal Service, and the investigation that followed sought to identify the source of the letters. Evolutionary theory concerning bacterial populations, mutational processes, and phylogenetic reconstruction were essential for this science-based forensic investigation. The development of the field of microbial forensics was greatly accelerated by the anthrax-letter investigation and it now provides a paradigm for both forensic cases and other public health investigations that involve infectious agents.

1. EVOLUTIONARY THINKING, MOLECULAR EPIDEMIOLOGY, AND MICROBIAL FORENSICS

The fields of molecular epidemiology and microbial forensics are populated by well-educated individuals. Nonetheless, the failure of these fields to employ evolutionary thinking sometimes limits the quality of the evidence and resulting inferences. For example, public health investigations of bacterial diseases have, in recent years, become highly dependent on one particular DNA-based technology called pulsed-field gel electrophoresis (PFGE). PFGE has the advantage that it can be applied to any bacteria, but its drawback is that the resulting data preclude more thorough evolutionary analyses. In particular, PFGE generates restriction fragment patterns—often called DNA fingerprints—that are analyzed using simple matching algorithms that produce yes/no outcomes, without allowing more sophisticated evolutionary analyses to identify the similarities and differences among the samples of interest. A “match” between fragment patterns is inferred by the analysts based on their experience and the rarity of a particular pattern in large databases. Unfortunately, little effort has been made to understand the evolutionary paths that may connect and explain the varying degrees of similarity among these patterns, and probabilistic models to place confidence estimates on relationships (e.g., a match) are rarely used. Most PFGE practitioners appreciate the validity of evolution, but their use of rigorous evolutionary analysis has been stymied by the difficulty in applying theory to such data and by resistance to making the changes necessary to improve on a widely used method.

In contrast with DNA studies of bacterial diseases, no established uniform technology exists in public health investigations of viral diseases; instead, each pathogen is typically analyzed by sequencing a particular, unique target gene. These sequence data are almost always analyzed using phylogenetic methods, and the analyses frequently include probabilistic models to test alternative hypotheses about the sources of the viruses. These DNA sequence data are in a universal digital format, and evolutionary models of sequence evolution are well developed, allowing for the rapid adoption and application of methodologies from other fields. By contrast, DNA fingerprints are poor substitutes for phylogenetic analyses, and the blind application of phylogenetic algorithms is inappropriate without a better understanding of underlying character state changes. The PFGE-based fragment patterns that constitute the DNA fingerprint can be thought of as complex phenotypes determined by the genotype—but following ill-defined rules—which illustrates the weakness of this approach. However, the lack of evolution-driven approaches in bacterial molecular epidemiology is starting to be overcome as sequence-based methods begin to dominate this discipline, and the costs of sequencing genomes keep dropping. The golden age for the molecular epidemiology of bacterial infectious diseases is arriving with the widespread adoption of whole-genome analysis.

2. THE USES OF DNA IN HUMAN AND MICROBIAL FORENSICS

The utility of DNA fingerprinting for human identification in forensic analysis has had a major impact on society and the legal system: it has led to the exoneration of falsely accused individuals and to the conviction of guilty criminals. The primary methodology is similar in some regards to the PFGE method described for bacteria in the previous section. However, in the case of human forensics, after several years of scientific discussion and debate, the statistical methods used to evaluate matches are firmly grounded on population genetic models and the scientific understanding of human biology, inheritance, and population subdivisions.

But these same statistical models have little utility in microbial forensics owing to the profound differences between bacteria and humans in terms of reproductive biology and modes of genetic inheritance. DNA is the genetic material of both bacteria and humans, of course, but that fact does not mitigate these differences. While DNA analysis in humans and in bacteria may be similar in terms of the molecular methods used, the inferences that can be drawn must reflect their different modes of inheritance and population structures.

It is equally important to realize that in addition to these differences between bacteria and humans, bacterial species—and even populations of the same nominal species—also differ from one another in ways that can influence the interpretation of genetic relationships. One important variable is the relative extent of vertical and horizontal modes of inheritance. Bacteria reproduce asexually, so their inheritance is primarily vertical (mother cell to daughter cell). However, horizontal gene transfer (HGT) between bacterial cells also sometimes occurs, and when it does so, it can move genes not only within but also between different species. HGT can have important consequences, such as the movement of antibiotic-resistance genes between species, and can leave conspicuous genetic evidence when it occurs between distantly related species. However, at a finer scale, many bacterial populations, including many recently emerged pathogens, show little or no detectable HGT. Thus, in many epidemiological and forensic situations, the relevant models and hypotheses are for lineages that are strictly clonal (asexual) in their derivation. In these cases, evolutionary analyses are focused on phylogenetic relationships and mutation rates.

3. GENETIC TECHNOLOGY AND THE SIGNIFICANCE OF A “MATCH”

The idea of a genetic match between two DNA fingerprints is jargon that has entered the scientific lexicon via the fields of human identification and forensics. Because individual humans are almost always the unique product of two unique gametes (identical twins being the exceptions), almost every person can be uniquely identified based on his or her alleles at a relatively few hyperdiverse regions of the human genome characterized by short tandemly repeated sequences. An exact allelic match between DNA samples from two individuals is so unlikely that a “match” has been used as the only physical evidence needed to link an individual to the scene of some crime. Likewise, a “nonmatch” can be used to exonerate a suspect. The idea of unambiguous matches and nonmatches has thus proven to be very powerful in the justice system. Unfortunately, this same terminology is often applied to scenarios in microbial epidemiology and forensics; however, the interpretations may be very different as a consequence of biological differences between humans and microbes.

With microbes, the technological context is also critical to understanding the significance of a “match.” A perfect genetic “match” can be lost using methods with greater resolution and discriminatory power. Low-resolution methods, including PFGE and multiple-locus sequence typing, would show that many bacterial isolates have identical alleles, but these methods see only a small portion of the genome. Greater discrimination can be achieved using multiple-locus VNTR analysis (MLVA), a technique that involves screening multiple loci with variable numbers of tandem repeats (VNTR), or by sequencing the entire genome of a bacterial isolate. Such whole-genome sequences may seem to be the ultimate standard, but bacterial geneticists have long realized that mutations will generate variation even within a colony of cells separated by only a few generations. Whole-genome sequencing does not detect these mutations because most applications generate a consensus DNA sequence that ignores rare variants; in fact, the accuracy of current technologies is such that rare sequencing errors obscure such rare mutations. In the future, however, new sequencing methods might detect rare variants directly from their individual DNA molecules. Therefore, a seemingly perfect match between two samples can be broken either by increasing the extent of genomic sampling or by searching more thoroughly for variants within the population. When it becomes possible to discriminate even between two colonies derived from the same progenitor strain, the ideal of seeking a perfect genetic match becomes more problematic than useful.

Rather than a match, a microbe’s “membership” in a phylogenetic group or clade is a more meaningful concept for epidemiological and forensics work. In a clonal lineage with little or no horizontal gene transfer, one can define phylogenetic relationships based on informative characters with membership in a particular clade based on shared derived states. Single-nucleotide polymorphisms, or SNPs, are now commonly employed in this way because they are produced by rare mutation events and thus are usually stable over appropriately long periods. With sufficient data, such as obtained by sequencing whole genomes, this stability can easily be tested by discriminating between convergent (e.g., homoplastic) and vertically inherited matches at the level of each SNP. Moreover, additional SNPs elsewhere in the genome are not problematic for inferring membership, because diversity is hierarchically nested within clades. Thus, additional SNPs produce novel genotypes that are still members of the clade. Even a reversion—a mutation to a prior state—does not change clade membership per se, although it can complicate inferences about membership.

In most cases, multiple point mutations will have occurred along most or all evolutionary branches. However, a single canonical SNP can be used to represent each branch, which can simplify phylogenetic analyses. This paring down of the number of characters is not essential, and it may result in less phylogenetic precision if a sample belongs to some subclade that has not been extensively characterized. In such cases, the failure to include all SNPs along a particular branch may lead to the assignment of that sample to the wrong subclade. This mistake may be caught, of course, by including more SNPs. Thus, the hierarchical redundancy of phylogenetically ordered SNPs creates a safeguard against incorrect assignments of samples to subclades.

4. THE KAMEIDO AUM SHINRIKYO ANTHRAX RELEASE

In the summer of 1993, an attack using a biological weapon was carried out in Kameido, a highly populated suburb of Tokyo, by the Aum Shinrikyo religious cult (currently called Aleph). Although the cult was large, well financed, and had well-educated scientists involved in the planning, the anthrax attack failed to kill or even sicken the targeted population. In fact, it was many years later before scientists realized there had been a failed attack.

In late June 1993, public health officials were notified by Kameido residents of a highly unusual and odiferous mist emanating from the roof of the Aum’s facility. Unsure of what was occurring, government health officials collected samples of the spray and submitted them for chemical analysis. The analyses evidently provided no evidence of toxic chemicals, and the cult discontinued the spraying, so no further actions were taken. Two years later, however, the cult carried out a chemical weapons attack by releasing sarin gas in the Tokyo subways. Ten people were killed and hundreds seriously injured. It was only after the arrest of cult members and during their subsequent questioning that the Kameido anthrax attack was discovered for what it was. The mist coming off their building was, they stated, from a culture of Bacillus anthracis—the causative agent of anthrax.

Hiroshi Takahashi was the investigating epidemiologist, and in 1997, he discovered a small tube of liquid that had been collected from the Kameido building at the time of the 1993 attack. He transferred this material to the United States, where B. anthracis cells were cultured and then genetically analyzed using MLVA, which was the best available technology at that time. Eight variable loci were analyzed, including six on the chromosome and one on each of two extra-chromosomal plasmids that carry virulence factors. Seven of the loci matched a well-known strain of B. anthracis, called Sterne, that is used in the production of a vaccine against anthrax. The assay for the eighth locus failed, a result that was also consistent with the Sterne strain because it is missing the pXO2 plasmid that carries this locus. Indeed, the absence of that plasmid is the reason that the Sterne strain is not virulent. Thus, the anthrax attack had failed to kill anyone because the Aum Shinrikyo cult had used a harmless strain of bacteria. The evidence of a vaccine strain raised the question, Why had the cult used a harmless strain? Was it a mistake on the part of the cult? Was it a practice run for a possible later attack? This question remains unanswered today.

B. anthracis is a pathogen with very low genetic diversity, reflecting its recent origin. In the pregenomics era, MLVA was one of the only available methods for distinguishing one B. anthracis strain from another. The database at that time contained only 89 distinct genotypes, or fingerprints. Even so, the results of the assays supported several important conclusions: (1) the cult had indeed used B. anthracis; (2) several commonly studied and virulent strains (e.g., Ames, Vollum) were excluded as the attack material; (3) the failure of one assay was consistent with the strain’s lack of one of two virulence plasmids; and (4) that failure, as well as results from the other seven loci, matched the fingerprint of the widely available vaccine strain Sterne. The first two conclusions were robust. The match to Sterne, however, was less so because other strains share the same seven-locus genotype; the null allele for the plasmid-encoded locus produced additional ambiguity. For the reasons discussed earlier, DNA fingerprinting methods such as MLVA are not well suited for evolutionary inferences. Nonetheless, an important forensic principle is evident—one we will revisit in the next section—in terms of the strength of exclusionary versus inclusionary findings.

5. THE AMES STRAIN AND THE 2001 ANTHRAX LETTERS

Only a few weeks after the September 11, 2001, terrorist attacks had killed thousands of people, the United States faced another shocking incident in October, one that employed a deadly biological weapon. The attacker(s) used the US Postal Service to send at least seven letters containing B. anthracis spores. These letters were sent to specific targets, but their routing through the postal system resulted in widespread contamination by spores, which disrupted several mail centers and other government facilities including congressional buildings. Molecular genetics and evolutionary approaches were central to the forensic investigation.

Although whole-genome sequencing methods were eventually brought to bear on this case, the investigation began at a time when that technology was not sufficiently developed to allow it to be used with the immediacy that the circumstances demanded. Public health as well as national security considerations meant that it was critical to identify the likely source—or at least to exclude certain sources—as quickly as possible.

To that end, Paul Keim and colleagues were able to quickly perform an initial analysis of the DNA from the spores in the letters using the same MLVA system used to analyze the B. anthracis from the Kameido event, and with an expanded reference database. In 1991, the United Nations Special Commission had discovered weaponized anthrax spores during inspections following the Gulf War. Bacteria were recovered from ordnance, and they were identified as B. anthracis, but little other characterization was done at that time. After the 2001 anthrax letter attacks, there was renewed interest in the Iraqi weapon strain, given suspicions of foreign involvement from some quarters. Identifying the Iraqi weapons strain and its relationship to the strain in the anthrax letters was therefore critical.

Within just days of the hospitalization of the first victim in Florida, MLVA showed that the B. anthracis isolated from that victim matched the Ames strain at all eight loci. Analyses of samples from the letters also matched the Ames strain. The Ames strain is a virulent one, unlike the Sterne strain that was deployed in the failed attack in Japan. The Ames strain was known to be used in several US government laboratories and, despite its name, it was originally isolated from Texas. The search was then on for the source of the attack strain, with three critical issues at hand. First, was the Iraqi strain also an Ames strain? Second, were other B. anthracis strains that could be isolated from nature similar enough to the Ames strain that they would produce a match at all eight MLVA loci? And third, what higher-resolution techniques could be employed to distinguish among sublineages within the clade that contains the Ames strain and its close relatives to trace the attack strain to a specific source?

In December 2001, the Iraqi strain was characterized using the MLVA method, and a match at all eight loci was established to another strain called Vollum. In fact, an Iraqi scientist had purchased the Vollum strain from a culture collection in 1986, indicating the likely source of that strain. Importantly, the Ames and Vollum strains differ at multiple MLVA loci. These differences meant that the B. anthracis strain discovered at the Iraqi bioweapons facility could be excluded, with a high degree of confidence, as the source of the spores in the letter attacks.

So, what was the source of the Ames-related material in the letters? Was the material derived from the Ames strain, which had been distributed to various laboratories? Or could it be a different isolate that just happened to match the Ames strain at all eight loci used in the MLVA testing? In fact, the database showed that an isolate obtained in 1997 from a goat in Texas also matched the Ames strain at all eight loci. The circumstances of the attacks made it clear that these anthrax cases were not a natural outbreak. In principle, someone might have reisolated a B. anthracis strain from nature that happened to be a close relative of the Ames strain. In any case, it became imperative to employ genetic methods that would allow maximum resolution to determine the source of the attack material.

To that end, whole-genome sequencing was employed to find genetic differences that could be analyzed using phylogenetic methods. SNPs are ideal for this purpose because reversion mutations should be rare in such young lineages as B. anthracis and especially the clade containing the Ames strain. Indeed, the extent of homoplasy in species-wide SNP data is only about 0.1 percent across the entire species. This approach identified four SNPs specific to the laboratory Ames strain, which could be used to differentiate it from natural isolates. By screening for these four SNPs, it was possible to exclude other strains including the isolate from the Texas goat (which had matched the Ames strain at all eight loci used in the MLVA test) as well as additional isolates from the same geographic region. Thus, it became possible to determine that a strain was a member of the Ames group of lab-derived isolates with much more confidence than with the fragment-matching approach of MLVA.

Whole-genome sequencing was also employed in other lines of the investigation. With the increasingly strong evidence that the attacks had used spores derived from the Ames strain present in several laboratories, the key genetic issue became one of searching for mutations in the attack materials that might match mutations found in some laboratories but not others. Thus, one line of the investigation involved comparing the genome sequences of the B. anthracis isolated from the Florida victim and another Ames-derived strain, called Porton, whose virulence plasmids had been “cured” (eliminated). The Porton strain was used because it was already in the process of being sequenced and analyzed prior to the attacks, thus expediting the investigation. In fact, several mutational differences were discovered between the Florida and Porton derivatives of the Ames strain. However, these differences turned out to be useless for the investigation because all of them were unique to the Porton strain; the mutations probably arose during the mutagenic procedures employed to eliminate the plasmids.

The other line of investigation using genomic sequences proved to be more useful but also quite complicated. In the early stages of the investigation, microbiologists had allowed some of the B. anthracis spores taken from the letters to germinate and produce colonies. They observed subtle variation in the appearance of colonies, with one predominant type and several variants at lower frequencies. Thus the differences were heritable, which implied that the differences in colony “morphology” had resulted from mutations. If confirmed, the mutant subpopulations might then provide a signature to distinguish possible sources of the spores used in the attacks. In summary, several clones with variant morphologies were sequenced, and mutations were identified. These mutants had not been seen in previous sequencing attempts because they were rare in the population of spores, and the resulting sequence represented a consensus sequence from the sampled cells. Next, the sampled cells were selected specifically to include these morphological variants. Molecular assays could then be developed to screen for four of these mutations.

In the meantime, the Federal Bureau of Investigation (FBI) had created a repository of more than 1000 samples, all derived from the Ames strain, from about 20 laboratories known to have worked with that strain. These samples were then screened for the four mutations. None of the four mutations were detected in most of the samples, but eight of them gave positive results for all four mutations. (There are many complications related to the sensitivity and specificity of the assays used to detect the mutations, as well as other issues that in the interest of brevity, are not presented here but are discussed in a 2011 report prepared by a committee of experts convened by the National Research Council.) The eight samples were all apparently derived from the same source—a flask of spores identified as RMR-1029—based on information obtained by the FBI. The contents of the flask had been generated by pooling several separately grown batches of spores, to produce a single large stock of material for experiments that would be performed at different times. This manner of preparing the flask of spores might account for the diversity of variant colony types that led to this line of investigation. In any case, these results pointed toward a particular flask and samples taken from that flask as a possible source of the spores placed in the attack letters. The criminal investigation was thus also focused on those individuals who had access to the RMR-1029 flask and its derivatives.

This chapter is focused on the role of evolutionary thinking in microbial forensics; it is not the place to discuss other aspects of the criminal investigation. But for those readers who want to know, very briefly, the outcome of this investigation, the FBI identified a government scientist as the lone suspect of the anthrax letter attacks. Before the US Department of Justice could bring formal charges, that individual committed suicide.

Genomic technologies continue to advance at a rapid pace, and it is possible that spores from the attack letters and from the RMR-1029 flask could be examined even more fully by so-called deep sequencing. That approach could, in principle, expand the analysis of diversity in those samples well beyond the four mutations that were discovered based on the variation in colony morphologies.

6. FROM MOLECULAR EPIDEMIOLOGY TO MICROBIAL FORENSICS AND BACK

Over the course of several decades, increasingly powerful molecular-based methods have been used to identify the source and track the spread of infectious diseases. These methods also served as the starting point for the forensic investigation of the anthrax attacks. Nonetheless, that investigation pointed to the limitation of these methods. The urgency and resulting high levels of funding to investigate the anthrax letters enabled the application of whole-genome sequencing—an approach that molecular epidemiologists had not been able to employ previously owing to its high costs. The genomic methodologies and analytical approaches have now become much less expensive, and so they should be applied much more broadly in molecular epidemiological studies motivated by public health concerns. Both forensic and epidemiological investigations are also well served by using phylogenetic approaches to analyze genomic data for determining relationships among samples, especially as the number of key samples becomes progressively smaller as one hones in on a probable source.

Thus, we predict with confidence that the use of whole-genome sequencing to understand evolutionary relationships will become common in public health. This technological change will bring with it changes in data analysis such that the full power of evolutionary theory, models, and methods can be used to determine infectious sources during natural disease outbreaks.