Tuberculosis Recurrences: Reinfection Plays a Role in a Population Whose Clinical/Epidemiological Characteristics Do Not Favor Reinfection

Background  Tuberculosis (TB) recurrences can be due to either reactivation by the same strain (standard assumption) or reinfection by a new strain. Reinfection has mainly been studied in selected populations with a high risk of reexposure to TB. Our aim was to analyze the role of reinfection in TB recurrences in unselected populations, without the clinical/epidemiological circumstances that favor the involvement of a new different strain of Mycobacterium tuberculosis in the recurrence.

Methods  A molecular typing analysis was performed with 92 sequential isolates of M tuberculosis from 43 patients with recurrent TB, during a 12-year period. The subjects were both positive and negative for the human immunodeficiency virus, most did not adhere to anti-TB therapy, and they lived in an area with a moderate incidence of TB. Recurrence was considered as being caused by reinfection when the molecular fingerprints for the strains involved in the sequential episodes of TB were different.

Results  In 14 (33%) of the 43 patients, different M tuberculosis strains were involved in the first and in subsequent episodes of TB. Reinfection was found for patients who were both positive and negative for the human immunodeficiency virus, and most patients did not adhere to anti-TB therapy. Differences between the reinfection and reactivation groups were not significant (P = .77) according to the time interval between episodes.

Conclusions  Reinfection plays an important role in recurrent TB in a population without the clinical/epidemiological circumstances that are usually assumed to favor it. Reinfection should, thus, be considered as a cause of TB recurrences in a wider context than before.

THE PROPORTION of patients with a well-documented first episode of tuberculosis (TB) who have a second recurrent episode is not well-known for unselected populations, and the proportion depends on different socioeconomic conditions. Tuberculosis recurrences are assumed to be mainly due to mismanagement of the disease, resulting from either poor adherence to correct therapy or the administration of inadequate treatment.

Recurrences have traditionally been considered as endogenous reactivations of the strain that caused the primary episode. Different studies have failed to find reinfections^1-3 or have only described anecdotal cases in which different strains are isolated from the primary and postprimary episodes.^4-6 Some researchers^7,8 described a high rate of recent infections in studies of TB transmission dynamics. A few studies^9-11 have found a role for reinfection in TB recurrences, always in selected high-risk groups of patients in whom reinfection is favored by specific epidemiological circumstances.

Our study searches for the rate of recurrent episodes of TB occurring in a large unselected population, not particularly prone to reinfection, during a 12-year period, and tries to assess the role of a new strain (reinfection) or the same strain (reactivation) in these recurrences—a key issue with clinical, therapeutic, and epidemiological repercussions.

Our institution is a 1700-bed hospital that serves a population of 630 000. The percentage of the working population is 77%; 74% have completed high school studies, and 12% are educated to the university level. The percentage of immigrants is 6%. The incidence of TB in this area from January 1, 1994, through December 31, 1999, has decreased, mainly due to the reduction of acquired immunodeficiency syndrome–associated cases after the introduction of highly active antiretroviral treatment, from 66 to 29 cases per 100 000 inhabitants per year.

We reviewed the records (databases) of the mycobacteriology laboratory of our institution from January 1, 1988, to December 31, 1999. All patients with 1 or more isolates of M tuberculosis in different respiratory clinical samples were considered. We considered patients with recurrent episodes, ie, all those with new isolates of M tuberculosis separated by more than 100 days (median, 430 days) from the primary episode (day 0). Nonadherence to treatment was not a criterion of exclusion, because our aim was to explore the role of reinfection in circumstances in which it is not usually expected, including in those patients in whom reactivation is usually assumed to be due to treatment failure. In the patients selected with recurrent episodes, we collected information regarding sex, age, human immunodeficiency virus (HIV) status, other risks for immunosuppression, anti-TB therapy, duration of treatment, time between episodes, and physicians' opinion regarding individual adherence to therapy. Adherence to treatment was defined as completion of at least 6 months of combination therapy. Nonadherence was considered exclusively when it was confirmed by the patient and the clinician. Directly observed therapy was not available in our patients. All these data are compiled in Table 1.

The strains cultured from the first and subsequent episodes for all patients were frozen at −70°C. For study purposes, only cultures from respiratory specimens were considered for analysis.

Clinical specimens were processed according to standard methods and inoculated in Löwenstein-Jensen slants and, since January 1, 1995, also in mycobacteria growth indicator tube media (Becton, Dickinson and Company, Sparks, Md). Susceptibility testing for isoniazid, rifampin, streptomycin sulfate, and ethambutol hydrochloride was performed for all the strains involved in reinfections. For strains before January 1, 1992, the proportions method was applied, and from this date, it was performed by another system (MB/BacT; Organon Teknika Diagnostics, Durham, NC). Correlations between these 2 approaches have been proved.¹² The susceptibility patterns were confirmed by reassaying all cases in which resistance was detected.

The spoligotyping assay, which explores polymorphisms in the directed repeats locus of Mycobacterium tuberculosis, was performed as follows. For chromosome extractions, 1 mL of the bacteria cultured in mycobacteria growth indicator tubes (Becton, Dickinson and Company) was centrifuged. Cells were resuspended in lysis reagent from Gene-Probe (Accuprobe, San Diego, Calif) and boiled for 5 minutes. Lysis beads (Gene-Probe) were added, and the suspension was sonicated for 5 minutes. The polymerase chain reaction (PCR) for amplifying the directed repeats region was performed using primers of the spoligotyping kit (Isogen Bioscience, Maarssen, the Netherlands), following the manufacturer's instructions. All spoligotypes showing differences for the strains from each patient were reassayed to confirm reinfection. When a spoligotype showed ambiguities in any of the boxes, the assay was repeated to obtain definite hybridization signals.

The DRE-PCR assay was performed as a secondary typing method on all the isolates that were considered to be indistinguishable by spoligotyping. The chromosome of M tuberculosis was prepared as described for spoligotyping. The DRE-PCR assay was performed as described elsewhere.^13,14

Amplified products were loaded in precast polyacrylamide gels (GeneGel Excel 12.5; Amersham-Pharmacia Biotech, Uppsala, Sweden) and run in an electrophoresis instrument (GenePhore; Amersham-Pharmacia Biotech). Gels were silver stained using a kit (Amersham-Pharmacia Biotech). These electrophoresis conditions highly improved the reproducibility of the assay and the sensitivity of detection of minor bands, thus eliminating the limitations of this technique when it is performed in agarose gels and with ethidium bromide staining.¹⁵ All DRE-PCR assays showing differences between strains from each patient were reassayed to confirm reinfection. The DRE-PCR types were considered different when differences in more than 3 major bands were observed between strains. This is an acceptable degree of difference because 3 bands constitute a high proportion of the total number of bands obtained for the strains analyzed.

Reactivation was considered the cause of the recurrence when the strains isolated in the sequential episodes were identical, as determined by spoligotyping and DRE-PCR. Reinfection was considered the cause of recurrence when different typing patterns were obtained for the strains isolated from the sequential episodes in spoligotyping or DRE-PCR assays.

To rule out cross contamination as a cause of misassignment of reinfections, we followed 2 approaches that were applied alternatively, depending on the availability of samples: (1) typing of the strains isolated from all the specimens that were processed in the laboratory on the same day as the strains in the analysis or (2) typing independent isolates for the same patient belonging to the same episode (<7 days apart). Spoligotypes for the strains processed on the same day from different patients were always different. For all control samples within the same episode of each patient, the same spoligotype was obtained. Both approaches ruled out cross contamination, indicating that the strains assigned to each of the recurrent episodes caused by reinfections were truly cultured from the patient's specimens.

The statistical package used for the analysis was Intercooled Stata 7.0 for Windows (Stata Corp, College Station, Tex). Because the number of patients was small, the Fisher exact test was preferred for the comparisons between the distribution of risk factors included in Table 2. The equality of medians was tested using the median test.

In our institution, from 1988 to 1999, 2567 patients were diagnosed as having TB based on the isolation of M tuberculosis. Overall, 172 patients had at least a second episode of TB more than 100 days apart (7% of the patients with a confirmed first episode). Ninety-two M tuberculosis sequential isolates from 43 patients with more than 1 episode (range between episodes, 106-3373 days; median, 430 days) were available for analysis.

All the 92 strains were typed by spoligotyping. Forty-six different spoligotypes were obtained for the whole analysis group (Figure 1 A and B).

When considering the sequential isolates for each of the 43 patients, 10 (23%) showed different strains for the first and second episodes, meaning reinfection was the cause of their recurrences. The spoligotypes for the sequential strains considered as different showed differences in a minimum of 4 spacers of the directed repeats locus (indicated by differences in 4 boxes in the spoligotype) and a maximum of 20, which means that the strains were clearly different (Figure 1 A). In the other 33 patients (77%), the isolates cultured from their sequential episodes were indistinguishable by spoligotyping (Figure 1 B).

For 2 patients with 3 episodes analyzed (patients 10 and 43), their recurrences showed a combined pattern: 2 episodes caused by reactivation, and the other caused by reinfection (Figure 1 A).

To check the typing data and to increase the discriminatory power of the spoligotyping assay, a second molecular typing method (DRE-PCR) was performed for the 33 patients whose sequential isolates were considered indistinguishable. With this second-line typing assay, the presence of identical strains in the first and postprimary episodes was confirmed for 29 of these 33 patients (Figure 2 A). In the remaining 4 patients, DRE-PCR found major differences between the patterns of the sequential isolates (Figure 2 B). Therefore, after the second-line typing assay, an additional 4 (9%) of the 43 patients had different strains for their first and second episodes, thus indicating reinfection.

If the data for the double-typing approach are taken together, reinfection was found in 14 (33%) of the 43 patients analyzed. Reactivations involving the same strain were found in the remaining 29 (67%) of the 43 patients.

For the reinfection and reactivation groups, there were no significant differences according to HIV status or to other risk factors, such as adherence to anti-TB therapy, intravenous drug abuse, alcoholism, homelessness, or prison stay (Table 2). The differences between the median (95% confidence interval) time between episodes were not significant (reinfection vs reactivation group, 479.5 [254.0-1020.0] vs 303.0 [228.9-578.1] days; median test, continuity-corrected Pearson χ²₁ = 0.09, P = .77). The mean (SD) times between episodes were as follows: reinfection group, 800.1 (837.3) days; and reactivation group, 593.7 (744.0) days. The interquartile range was 831.7 days for the reinfection group and 528.0 days for the reactivation group.

The antimicrobial susceptibility of the strains involved in the cases of reinfection remained unchanged (drug susceptible) in all patients but 3 (Table 1). Two of these acquired resistance (patients 10 and 28), and in the other (patient 1), reinfection was caused by a more susceptible strain than the one from the primary episode (Table 1).

Of all the patients diagnosed as having TB in our institution during the past 12 years, 7% had a recurrent episode. Data available for comparison are scarce, generally from old series or from selected groups of the population, and depend on anti-TB therapeutic regimens.Figures for recurrence range from 1% to 11%,^16-18 and for one of the classic studies,¹⁷ the recurrence proportion (6%) is practically equivalent to ours, despite the availability of anti-TB drugs, the low level of primary resistance, and the close follow-up of patients with TB in our country.

Our study presents an analysis of 92 strains involved in sequential episodes of TB from 43 patients, during a 12-year period. To our knowledge, this is the longest follow-up and the largest group of patients studied in the context of TB recurrences.

An unexpectedly high frequency of reinfection (33%) was found in our population. A previous study by van Rie et al¹¹ found a large proportion of reinfection in a selected population of HIV-negative patients after curative treatment; these patients lived in an extremely high incidence area (1000 cases per 100 000 population per year). On the contrary, the population in our study was unselected, there were both HIV-positive and HIV-negative patients, the patients lived in an area with a much lower incidence of TB (mean incidence for the last 6 years of the study period, 44 per 100 000), and the patients were not in high-exposure situations (except for 1 patient). Nevertheless, reinfection was the cause of a high percentage of recurrences, which suggests that reinfection should not only be considered in circumstances that favor reexposure. In a recent report by Caminero et al,¹⁹ a high percentage of reinfection was also found among patients with recurrent TB who lived in a geographic setting with a moderate incidence of the disease.

One feature in our study could account for the frequent finding of reinfection—the high variability found among the strains circulating in our area. Of the 46 spoligotypes obtained, 41 were unique. In this sense, there have been reports¹¹ on the inability to detect reinfections when 2 independent strains of the same majority endemic clone are involved in 2 sequential episodes. Molecular epidemiological studies^1,20 in patients with TB frequently find the presence of majority clones. This could be the cause of an underestimation of reinfection in other studies. It is possible that the wide variation of clones in our population provided the optimal conditions to reveal the real proportion of reinfection in patients with TB.

Most of the patients in our study with recurrences due to reinfection did not adhere to anti-TB treatment. Recurrences in patients without curative therapy of their primary TB are expected to be due to reactivations of the same strain, which cannot be assumed to have been eliminated from the organism. Surprisingly, all but 3 of our 14 reinfected patients did not adhere to treatment, and this may imply a role for reinfection for more cases than previously expected. Initially, our study could be criticized for having selected patients with a second isolate of M tuberculosis who did not adhere to therapy during their first episode. These patients are not usually considered to have recurrences but are considered to be patients in whom treatment has failed. Nevertheless, our data indicate that in many of these cases, a different M tuberculosis strain is responsible for the second isolation, which suggests that the concepts of recurrence/reinfection could not a priori be defined by adherence to therapy during the first episode. It is difficult to explain reinfection in patients in whom primary TB has not been efficiently treated. The strain involved in the first episode may not have been eliminated by therapy, but could have been displaced after competition with the new strain, if the new strain showed higher biological fitness or more efficient interactions with host factors. Another explanation for reinfection in patients who did not adhere to anti-TB therapy could be the occurrence of coinfection with more than 1 strain and the selective growth of different strains for the first and second episodes. We have preliminary data from a selection of these patients in whom clonally homogeneous populations of M tuberculosis are found after typing multiple independent colonies for each of their episodes. Thus, coinfection with more than 1 strain could reasonably be excluded.

In our study, we did not find significant differences in the distribution of the main risk factors for TB between the reinfection and reactivation groups, although we cannot rule out that it might be due to a lack of statistical power to detect differences, given the small number of patients. Both HIV-positive and HIV-negative patients were found in the reinfection and reactivation groups. An assumption in the analysis of recurrences in patients with TB is the immune protection that the primary episode is supposed to confer. Therefore, reinfection is usually not considered for HIV-negative patients not living in high-exposure situations. In our analysis, reinfection is also found for HIV-negative patients, and differences in HIV status are not significant for the reinfection and reactivation groups. This suggests a role of factors other than immune status in modulating reactivation and recurrence dynamics.

Furthermore, the assumption that reinfection is more likely in episodes farther apart in time, whereas reactivations are assumed for episodes closer in time, should be approached with caution. Some researchers^11,21 have found cases of reinfection at the end of therapy and even during the therapy period. In our analysis, the differences in time intervals between reinfection and reactivation episodes were not significant. Reinfections were found for episodes close in time, and reactivations occurred for episodes far apart in time.

Some studies^5,10,22,23 have found reinfection to be associated with the acquisition of strains with higher resistance to anti-TB drugs. In our case, strains involved in reinfection are rarely associated with a higher resistance, and in 1 case, the strain in the second episode was more susceptible than that which caused the first episode. This is consistent with the nonadherence to therapy generally found in our patients. Thus, resistance confers no advantages on new strains if treatment has not been adhered to.

In our study, the typing design is different from that of previous reports,^2,9,11,21 in which restriction fragment length polymorphism was the method selected. We performed a double-line typing assay, following previously recommended procedures.^24-27 In our case, a highly reproducible method, such as spoligotyping, was first performed^15,28 to search for differences between strains involved in reinfection. A secondary typing method, DRE-PCR,^13,14 was applied to increase the discriminatory power of the assay to guarantee cases sharing typing patterns and, therefore, confirm reactivation. This approach lacks the limitations that are found for IS6110 restriction fragment length polymorphism,^24,29-31 and has been proved to have an equivalent discriminatory power as this reference method.²⁵

To test the validity of our molecular data, we performed 2 alternative approaches to rule out the potential role of laboratory cross contamination in misassigning some cases of reinfection.^11,32 We checked that (1) the strains from specimens cultured in the laboratory on the same day as those in analysis did not share spoligotypes or (2) the strains cultured from other samples close in time for the same patient showed an identical typing pattern. Both observations lead us to be confident about the validity of our data.

In conclusion, our study shows a high proportion of TB recurrences caused by reinfection with a new strain. Reinfection in our analysis was found in a group of unselected patients and, therefore, they were not as homogeneous as others in previous reports. It was detected for HIV-positive and HIV-negative patients, in conditions in which high exposure was not expected, and in patients who did not generally adhere to anti-TB treatment. Our data suggest that even when clinical/epidemiological characteristics do not particularly favor reinfection, it should not be ruled out.

Accepted for publication January 3, 2002.

We thank Beatriz Pérez Gómez for her help with the statistical analysis; Oscar Cuevas for his help with the typing assays; and Thomas O'Boyle for his revision of the English in the manuscript.