Methods

Isolates. F. tularensis isolates submitted to or recovered at the CDC during 1964–2004 (from humans) and 1963–2005 (from animals) were analyzed. Isolates were confirmed to be F. tularensis, and identified as either type A or type B as described elsewhere [3, 7]. Clinical information was extracted from accompanying submission forms and included age, sex, onset date, site of isolate recovery, underlying illness, outcome, state and county of residence, state and county of exposure (if different than residence), and mode of transmission. Species and county of infection for isolates from animals was extracted from submission forms.

PFGE analysis. F. tularensis cells were embedded in agarose plugs and lysed. Genomic DNA was restricted with PmeI, and electrophoresis was performed [7]. Salmonella enterica serotype Braenderup strain H9812 restricted with XbaI was used for gel normalization [7]. PFGE patterns were analyzed using BioNumerics, version 3.5 (Applied Maths). Dendrograms were constructed using Dice similarity coefficients and unweighted pair group method with averages.

Statistical methods and maps. Epidemiologic analyses were performed with SAS, version 9.1 (SAS Institute). χ^>2 tests were used for categorical data, and Wilcoxon rank-sum tests were used for age distribution comparisons. P<.05 was considered to be significant in all analyses. Logistic regression analysis was used to identify factors associated with human mortality and to calculate adjusted ORs. Age, sex, immune status, and infecting genotype were assessed to determine their relationships with mortality. Age was converted into a categorical variable by quartiles. Because the isolation of an organism from an invasive site is indicative of disseminated infection and thus could be part of the biologic pathway to death, it was excluded from multivariable analysis. Characteristics of type B infections in humans, as described elsewhere, were included in multivariable analysis [7]. The multivariable model included variables with P<.1 in the crude analysis. Maps were generated by plotting isolates randomly within county of exposure with use of ArcGIS, version 9 (ESRI).

Results

PFGE genotyping. A total of 500 F. tularensis isolates were identified for analysis: 316 isolates from humans (208 type A and 108 type B) from 38 US states and 184 isolates from animals (97 type A and 87 type B) from 24 US states (table 1). Three type A isolates from humans were not viable; the remaining type A isolates (n=302; 205 from humans and 97 from animals) were genotyped using PFGE. Only a portion of the type B isolates (n=61; 22 from humans and 39 from animals) were genotyped using PFGE, because PFGE typing had previously indicated little to no diversity among type B isolates [7].

PFGE patterns grouped into primary clusters that corresponded with type A and type B (figure 1). Among type A isolates, patterns divided into the A1 and A2 genotypes, with 195 isolates (65%) classified as A1 (139 from humans and 56 from animals) and 107 isolates (35%) classified as A2 (66 from humans and 41 from animals) (figure 1). Among the A1 isolates, 2 distinct clusters were evident and were designated A1a and A1b. The A1a genotype consisted of 100 isolates (74 from humans and 26 from animals), and the A1b genotype consisted of 89 isolates (60 from humans and 29 from animals). Limited pattern diversity was observed among the A1a and A1b clusters, with 11 patterns comprising each genotype. Patterns for 6 A1 isolates (5 from humans and 1 from an animal) did not cluster with the A1a or A1b genotypes or with each other and were designated non-A1a and non-A1b (figure 1). PFGE patterns for A2 isolates also divided into 2 clusters, which were designated A2a and A2b (figure 1). The A2a genotype consisted of 65 isolates (45 human, 20 animal), and the A2b genotype consisted of 42 isolates (21 human, 21 animal). Pattern diversity was observed within the A2a and A2b clusters, with 28 and 17 patterns comprising these genotypes, respectively.

Human infections. Epidemiologic analyses were performed for all type A clusters identified by PFGE genotyping (A1 and A2, A1a and A1b, A2a and A2b). Characteristics of type B infections have been described elsewhere [7]. Human infections due to either A1 or A2 generally segregated east or west, respectively, of the 100th meridian (figure 2, top; table 1) [6, 7]. However, a small number (n=9; 6%) of infections due to A1 were identified west of the 100th meridian along the coast of California and Oregon and in areas of Idaho, Utah, and Colorado. A single infection due to A2 was identified east of the 100th meridian. Geographic distributions overlapped for infections due to A1a and A1b strains (figure 2, top) and for infections due to A2a and A2b strains (data not shown).

Age and sex were known for 191 (93%) and 205 (100%) patients, respectively. Age distributions differed among patients with infection due to type A strains; patients with infection due to A2 strains (mean age, 33 years; median age, 31 years) were younger than patients with infection due to A1 strains (mean age, 41 years; median age, 44 years) (P<.02). Patients with infection due to A1a strains (mean age, 37 years; median age, 40 years) were younger than patients with infection due to A1b strains (mean age, 44 years; median age, 44 years) (P<.01) (table 2). No difference was noted with regard to age distributions between patients with infection due to A2a and A2b. The majority (71%–80%) of type A infections across all genotypes occurred in men. An immunocompromising condition was reported for 5 (3.6%) of 139 patients with infection due to an A1 strain, 0 (0%) of 66 patients with infection due to an A2 strain, 1 (1%) of 74 patients with infection due to an A1a strain, and 3 (5%) of 60 patients with infection due to an A1b strain.

Most infections (72%) occurred between May and September. Month of onset did not differ among infections caused by different type A genotypes. Distributions of type A genotypes did not differ by decade of infection. Route of exposure was reported for 99 (48%) type A isolates. Arthropod bites and direct animal contact each accounted for roughly one-half of reported exposures. Contact with domestic cats or lagomorphs (i.e., rabbits and hares) occurred across all genotypes and accounted for 40 (85%) of 47 animal exposures. The type of cat exposure was recorded for 13 (72%) of 18 cat-associated cases: 7 exposures involved a bite, 5 involved a scratch, and 1 involved a hand wound sustained during a necropsy of an infected cat.

Information on the anatomic site of recovery was available for 188 isolates (92%). Overall, 69 (54%) of 127 A1 isolates were classified as invasive (i.e., were recovered from blood, lung, pleural fluid, or CSF; noninvasive isolates were recovered from ulcer, wound, lymph node, or eye), compared with only 6 (10%) of 61 A2 isolates (P<.001) (table 2). Only 30 (44%) of 68 A1a isolates were recovered from invasive sites, compared with 36 (67%) of 54 A1b isolates (P<.02) (table 2). Of the 6 invasive A2 isolates, 2 were genotyped as A2a and 4 as A2b.

Clinical outcome was reported for 108 (78%) of 139 infections due to A1 strains and 53 (80%) of 66 infections due to A2 strains. Consistent with previous findings, mortality differed significantly between infections caused by A1 versus A2 strains (13% vs. 0%; P<.01) [7]. Mortality also differed markedly among A1 genotypes; 12 of 49 infections due to A1b strains were fatal, compared with only 2 of 55 infections due to A1a strains (24% vs. 4%; P<.003) (table 2).

For logistic regression analysis of variables contributing to human mortality, the F. tularensis genotypes were categorized as A1a, A1b, A2, or type B. Because no deaths occurred among patients with infection due to genotypes A2, A2a and A2b isolates were not analyzed separately and exact logistic regression was used. In univariate analysis, genotype and age were each associated with mortality (P<.1) and were thus included in the multivariable model. In the multivariable model adjusted for age, F. tularensis genotype was independently associated with mortality. Specifically, only the A1b genotype was significantly associated with mortality when A2 was the referent group (adjusted OR, 13.5; 95% CI, 2.5–αnfin;) (table 3). When A1b was the referent group for F. tularensis genotype, all other genotypes differed significantly (A1a, P=.01; A2, P<.001; B, P=.01).

Animal infections. Epidemiologic analyses were performed for infections due to A1 and A2, A1a and A1b, and A2a and A2b strains and type B in animals. Geographic distribution of type A and type B isolates from animals corresponded with the distribution of type A and type B isolates from humans described elsewhere [7] (figure 2, bottom). Most A1 isolates from animals (95%) came from US states east of the 100th meridian (figure 2, top), whereas A2 isolates were from only US states west of the 100th meridian. A1a and A1b isolates (figure 2, top) and A2a and A2b isolates (not shown) overlapped in distribution. In all cases, the distribution of type A genotypes among animals paralleled the distribution of type A genotypes among humans (figure 2).

Most (97%) animal isolates originated from 4 source groups: rodents, lagomorphs, domestic cats, and captive primates. Rodents and lagomorphs are considered to be part of the enzootic cycle of F. tularensis, whereas domestic cats and primates are considered to be incidental hosts. The distribution of type A versus type B isolates among animal groups was not random (P<.001); the majority of isolates from cats and lagomorphs were type A, whereas nearly all isolates from primates and rodents were type B (table 4). Five type B isolates were recovered from other animal sources (i.e., a dog, birds, and weasels) (table 4).

Both cats and lagomorphs are reported sources of type A infections in humans and accounted for 85% of reported animal exposures in this study [2, 16–20]. Type A isolates from cats were genotyped as A1 and A2; 33 (80%) of 41 isolates were genotyped as A1. Among A1 isolates from cats, roughly one-half (48%) were A1b. Among A2 isolates from cats, 5 (63%) of 8 were A2b. Isolates from lagomorphs were also identified as both A1 and A2, with 21 (40%) of 52 isolates identified as A1. The majority (47 [90%] of 52 isolates) of lagomorph isolates originated from cottontail rabbits. Among isolates for which the rabbit species was known (n=31), all isolates from the eastern cottontail (Sylvilagus floridanus) were A1 (n=8), whereas all isolates from the desert cottontail (Sylvilagus audubonii) were A2 (n=23). Both A1a and A1b were isolated from lagomorphs; most were genotyped as A1b (13 [62%] of 21 isolates). A2a and A2b isolates were also both isolated from lagomorphs.

Discussion

Molecular subtyping of 363 F. tularensis isolates from humans and animals identified distinct type A genotypes A1a, A1b, A2a, and A2b, as well as type B. Significant differences between human infections caused by A1 and A2 and also between A1a and A1b were identified with respect to patient age, site of isolate recovery, and mortality. No differences were identified between infections caused by A2a and A2b genotypes. Thus, 3 type A genotypes—A1a, A1b, and A2—demonstrated clinical relevance in this study. These 3 genotypes were also observed in cats and lagomorphs, which provides evidence that A1a, A1b, and A2 genotypes exist among animals commonly linked to human type A infections (cats and lagomorphs) and those important for enzootic maintenance of type A (lagomorphs).

The mortality differences among type A infections were the most notable finding of this study. Although overall mortality due to type A was 9%, mortality associated with infection due to A2, A1a, and A1b differed significantly, at 0%, 4%, and 24%, respectively. Differences in mortality among infections caused by A1a, A1b, and A2 strains are not likely attributable to differential recognition and treatment of tularemia in varied geographic regions, because no difference in mortality was observed between type B infections in the eastern (8%) vs. western (6%) United States. In addition, most deaths attributable to A1b infections occurred in US states where both A1a and A1b caused infection.

Logistic regression analysis of human infections caused by A1a, A1b, A2, and type B strains indicated that some characteristic intrinsic to A1b strains such as a virulence factor, rather than host characteristics, is responsible for the higher mortality observed among individuals with infection due to A1b. In addition, although crude mortality was previously shown to differ between infections caused by type B (7%) and A2 (0%) strains and infections due to type B strains were suggested to be more severe than those caused by A2 strains [7], multivariable logistic regression analysis did not reveal a significant difference between infections due to type B and A2 strains when patient age and genotype were simultaneously assessed for their association with mortality.

Differences among type A strains have been described elsewhere [21]. Comparison of type A infections due to either Schu S4 or FSC033 strains (Georgia) in laboratory mice showed that FSC033 disseminated more rapidly and was associated with a shorter time to death than was Schu S4. Schu S4 is an A1a strain, and in our study, genotypes of isolates from Georgia were found to be both A1a and A1b, which raises the possibility that FSC033 may be an A1b strain. Future experiments will be important to determine whether pathogenicity differences among infections due to A1a, A1b, and A2 strains can be observed in a mouse model.

The geographic distribution of infections attributed to all genotypes was found to overlap between humans and animals. Because their movement is limited in comparison to humans, animals are more reliable indicators of exposure locations. Infections due to A1 strains predominate in the eastern United States, with a small number of cases identified in the west. Multiple lines of evidence support that A1 strains are endemic in the western United States. There was no evidence that the patients from the west who were infected with A1 strains had traveled to the eastern United States. In addition, identification of A1 isolates in animals (both domestic and wild) in the west provides evidence of enzootic transmission. Moreover, infections due to A1 strains were identified in both humans and lagomorphs in an outbreak of tularemia in Utah in 2007 [22]. In comparison, A2 strains appear to be restricted to the western United States. A single infection in a human and no infections in animals due to A2 strains were identified east of the 100th meridian. Although the single infection in a human was attributed to exposure while traveling in Oklahoma, the patient resided in New Mexico; our findings suggest that this patient may have acquired the infection in New Mexico.

Factors important to the geographic distribution of F. tularensis genotypes in the United States require further study. Geographic differences and strain variation could reflect adaptation to different animal hosts or vector species. Animal host differences for type A and type B strains have been described anecdotally in historical literature [2, 14, 15]. In our study, we found type B strains more frequently associated with rodents, whereas type A strains were more commonly associated with lagomorphs; both rodents and lagomorphs are considered to be important in enzootic maintenance of tularemia. Host species differences were also observed among type A isolates. The desert cottontail S. audubonii was associated with A2 isolates, whereas the eastern cottontail S. floridanus was associated with A1 isolates.

This study is subject to potential limitations. By analyzing only cases of human infection in which an isolate was recovered, our dataset may be biased toward severe cases. Isolate submission should not be biased by genotype, because it was unknown at the time of submission. Thus, relative mortality differences observed among genotypes should not be affected by an overall bias toward more severe cases. The site of organism recovery may be biased by differences in clinical practice. For example, it is unlikely that all noninvasive isolates were not present in blood, but it is more likely that a blood sample may not have been obtained. In addition, exposure information was unavailable for most infections due to type A strains, and where it was known, it was limited to the county level. Finally, these isolates represent a convenience sample. Most isolates were submitted as a result of proximity or established relationships with other laboratories, with the number of isolates analyzed in this study representing only 4% of the total reported cases among humans for this time period.

This is the first demonstration that A1 strains can be divided into 2 distinct genotypes. Preliminary whole genome single nucleotide polymorphisms analysis of F. tularensis strains also identified these 2 distinct A1 genotypes [12]. Identification of genotypes A1a and A1b by PFGE and single nucleotide polymorphism analysis, but not by multilocus variable number tandem repeat analysis, highlights the strengths and weaknesses of various subtyping methods. Multilocus variable number tandem repeat analysis is based on rapidly evolving tandem repeats, whereas both PFGE and single nucleotide polymorphism analysis are based on slowly occurring genome changes [23]. Although high-resolution methods that examine hypermutable regions of the genome are powerful for distinguishing individual strains in time and space, this level of differentiation may at times be too fine for the identification of epidemiologic clusters.

In conclusion, genetic diversity exists among type A strains of F. tularensis and correlates with significant differences in clinical outcome and geographic distribution. Although type A strains have generally been considered to be equivalent in terms of virulence, we demonstrate that a subset of type A strains (genotype A1b) are associated with significantly higher mortality in humans. This finding has important implications with regard to disease progression, disease prevention, and basic research projects, including diagnostic assay and vaccine development. Further studies are required to elucidate the epidemiologic, genetic, and ecologic differences between A1a, A1b, and A2 strains. Genotyping of isolates from animals may be a useful surveillance tool for predicting where human infections with specific type A genotypes may occur. Because cats are a source of human infection with A1b strains, veterinary education efforts to promote recognition of tularemia in domestic cats is warranted. In areas where infection with A1b strains occurs, physician education efforts to enhance early clinical recognition may help prevent deaths among humans.