Molecular Investigation of HIV Transmission

In 1990, the U.S. Centers for Disease Control (CDC) and the Florida Department of Health and Rehabilitative Services announced their provisional finding that a Florida dentist with the acquired immunodeficiency syndrome (AIDS) had infected one of his patients in the course of dental surgery [1]. Not long after that, the dentist, David J. Acer, and later the patient, Kimberly Bergalis, died of complications of AIDS. By 1992, the CDC had completed the fieldwork and viral characterization, and the HIV Sequence Database and Analysis Unit at Los Alamos National Laboratory had completed the HIV genetic sequence analyses involved in the investigation of the case, which by this time included four additional infections that appeared to have originated from the dentist [2, 3]. Another dental patient with HIV strains matching those of Acer and the five patients was subsequently discovered. In 1994, an independent analysis of the dentist's and the six patients' viral sequences corroborated the CDC's original conclusion [4]. Nevertheless, questions persist [5], particularly about how the infections occurred and whether new cases of this kind are likely to be encountered.

These questions, for which no satisfactory answers are currently available, are understandable in light of several facts. First, apart from the Acer case, no HIV transmission from a health care worker to a patient in the United States has been documented. Furthermore, the rate of HIV infection of health care workers in health care settings in the U.S. remains relatively low: Fewer than 50 occupational transmissions had been documented as of December 1993 (there are at least twice as many possible transmissions under investigation), and most of these resulted from percutaneous exposures [6]. In addition, 22 000 patients are known to have been treated by HIV-infected dentists and surgeons and are known not to have been infected [6].

Although transmission of HIV in health care settings is relatively rare in developed countries, there is recent evidence in addition to the Acer case that nosocomial transmission of HIV can and does occur. In the last year, a case has been reported from Australia, where four women, none of whom had identifiable risks for HIV infection, were treated by the same surgeon on the same day as a fifth patient with AIDS. The four women became infected, apparently through a breakdown in infection control [7]. The surgeon himself was HIV negative. If the molecular analysis (viral DNA fingerprinting) shows the viruses to be similar—as is likely, given the extraordinary circumstances of the case—this will provide another example of multiple infection of patients in a health care setting.

Another case has been studied in southern Russia, where about 90 children became infected with the same viral strain; some were in a hospital in Elista, others were in a hospital in Rostov-on-Don, to which a group of children with HIV infection from Elista had been transported [8]. Because the viral strain infecting the children appears to be extremely rare not only in Europe but throughout the world, there is no doubt about the epidemiologic connection among these infections. We must conclude that nosocomial transmission of HIV can and does occur. In these two recently studied cases, as in the Acer case, the paths of infection are not and probably never will be precisely known.

At this time, it may be easier to prove that HIV transmission has taken place in a health care setting than to disprove that it has [9], as has been done in the case reported in this issue by Jaffe and colleagues [10]. This case involves a Miami dentist who practiced for 5 years while infected with HIV and who routinely failed to fully comply with recommended infection control procedures, but who does not appear to have infected any of his patients, 28 of whom have been found to be HIV-positive. To understand the difficulties of reaching a negative conclusion of this sort, it is necessary to review the basic premises underlying the molecular epidemiology of HIV. First, of course, is the fact that HIV has an unprecedentedly high mutation rate compared with other pathogens. Human immunodeficiency virus variation can be “tracked” by comparing either viral DNA sequences or viral protein sequences encoded by the DNA; in the case of DNA, the sequenced nucleotide strings typically consist of 300 or more characters (letters) that offer ample information for tracking purposes. We can, accordingly, determine a range of genetic relatedness for intrapatient variants (variants within a single person) and a decidedly different range of degree of relatedness for interpatient variants where there is no evidence of epidemiologic linkage.

Populations of viruses from the same infected person have DNA sequences that are typically 95% to 100% similar by an appropriate measure of sequence relatedness for the viral envelope gene [3, 9]. In contrast, the extent of dissimilarity of viruses from epidemiologically unlinked persons is greater, depending on the length of time a particular subtype of HIV has been circulating in a locale—the “viral diversification” factor—and on the array of highly distinct viral subtypes circulating in a locale—the “viral trafficking” factor. In Florida, for example, where there is currently only a single HIV-1 subtype, most HIV-1 sequences from ostensibly unlinked persons are 83% to 93% similar [3, 10]. In regions of the world where highly divergent viral subtypes are co-circulating, such as Russia, interpatient viral similarities can be as low as 70% or less [8].

Given these facts, it is to be expected that intrapatient sequence relationships, on the one hand, and sequence relationships for epidemiologically unlinked persons, on the other hand, bracket the sequence relationships of epidemiologically linked persons—mothers and their babies, sexual partners, blood donors and their recipients—whose viral sequences usually display sequence relationships of 94% to 98% similarity, regardless of the degree of viral subtype diversity in the population [3, 9, 11]. In the Acer case, the six dental patients had viruses that were all about 96% similar to those of the dentist and of each other (clearly within the range for linked persons); the average interpatient viral sequence similarity in the Miami area was approximately 89% at that time [3, 11].

Because interpatient viral genetic distances increase with time of infection, in a given locale (such as Miami) or between linked persons (a mother and her perinatally infected baby), it becomes increasingly difficult with passage of time to show a sequence relatedness that would signify epidemiologic linkage. For example, had Acer and Bergalis not been studied so soon after the transmission, it would have been more difficult for the CDC to establish linkage: Viral sequences differing by 4% in one year might differ by 6% or more in subsequent years, bringing the working window of linkage (say, 93% to 98% similarity) close to the distribution of sequence relationships at large (say, 83% to 94% in Miami). For this reason, investigators also undertake more complicated computational analyses—phylogenetic trees [3, 4, 10], signature patterns [3, 11], and statistical bootstrapping [3, 4, 10]—in an attempt to determine the HIV-1 genetic groupings. As the epidemic progresses, judgments in this field should be easier to make. This is because the background distribution of unlinked viral sequences will expand, due to genetic diversification and viral trafficking, toward reduced similarities, and epidemiologically linked persons will continue to display similar viral sequences, because the spectrums of viral sequences within persons do not build up over the course of the pandemic, due to constraints that are poorly understood at this time.

The interesting case of a second Miami dentist and his patients with HIV infection reported in this issue [10] illustrates these premises. Molecular epidemiology was essential to this CDC investigation because almost all of the dentist's 28 patients who were HIV positive had identifiable outside risks for infection. On the basis of similarity measurements and phylogenetic tree analysis, Jaffe and colleagues [10] found no evidence of HIV transmission from this dentist to his patients. Strictly speaking, the molecular evidence only rules out transmission in the later years of the dentist's practice. By the argument presented above, epidemiologic linkages blend into the background of unlinked transmissions as years (but not decades) elapse. What little we know about HIV transmission suggests that the potential for an accident would have been greatest in the later years of the dentist's practice, the period best covered by the CDC's molecular analyses.

In its investigations of the Acer and the more recent Florida dentist cases, the CDC has carefully kept the totality of evidence in mind; the molecular evidence is not being put forward on its own. In the Acer case, the lack of identifiable risks in the dental patients infected with HIV plus the molecular findings point to the conclusion that the closely linked viruses did not appear in Acer's patients by chance. This assessment is strengthened by our growing realization that chance groupings of viruses are not found in Miami or Amsterdam or New York or Brazil: Closely linked viral sequences are occasionally encountered in data sets from centers of the pandemic that have had HIV for a decade or longer, but not six or seven sequences at a time. The macroscopic parallel to this observation is that no web of epidemiologic linkage apart from shared dental exposure can be identified for Acer's six patients with HIV infection. Hence, difficult as it is to explain how Acer infected his patients, we must ask whether any alternative explanation would be less difficult to comprehend.

In the second Florida dentist case, the presence of identifiable risks plus the quantitative findings pertaining to the viral sequences point to an opposite conclusion. Uncertainties remain, but no web of transmission by any route, yielding viral sequence relationships similar to that seen for a pair of sexual partners, can be shown among the 28 persons with HIV who happen to have been patients of the same dentist. The fact that there is no chance viral grouping among the 29 persons involved in this second case makes the hypothesis of chance clustering of seven persons in the Acer case that much more unlikely.

• Copyright ©2004 by the American College of Physicians