Massive Parallel Sequencing Provides New Perspectives on Bacterial Brain Abscesses

Our understanding of polymicrobial infections has been hindered by our limited possibilities for describing them. Recent investigations of bacterial brain abscesses using universal amplification of the bacterial 16S rRNA gene, followed by Sanger sequencing of cloned amplicons, have revealed that only a fraction of the bacteria present are identified by culture (1, 2). Nevertheless, this approach has limitations when it comes to detecting smaller subpopulations in a multispecies community, unless very high numbers of clones are sequenced (3). This is problematic, since the species structure of an abscess may change over time and pathogens important for establishing the infection potentially remain at only low concentrations in the more mature abscesses. Furthermore, the species that are important for maintaining and expanding the abscess might primarily exist close to the abscess wall and do not necessarily dominate in the pus obtained by aspiration. Rapid development within the field of massive parallel sequencing technologies (MPS) is about to provide the diagnostic laboratories with tools that can characterize even the most complex microbial communities. The aim of the present study was to use recent advances within this field to conduct the most thorough characterization of bacterial brain abscesses to date. By combining a high number of reads per sample with a nationwide collection and systematic classification of specimens, we sought to recognize microbial patterns and start delineating a pathogenesis for spontaneous polymicrobial brain abscesses. We further address general matters, like the depth of analysis needed for reliable characterization of polymicrobial infections, and define a new sample-specific cutoff for differentiating sample bacterial DNA from background reagent bacterial DNA in MPS protocols.

During the study period, we received sample material from 56 patients. Four samples were rejected due to fungal etiologies. The remaining 52 samples were divided into three categories: spontaneous brain abscesses (n = 37), spontaneous subdural infections (n = 3), and postoperative infections (n = 12). Six additional culture-positive specimens (five spontaneous brain abscesses and one postoperative infection) were registered by the participating laboratories but not included either because of too-small remaining specimen volumes or because of administrative failures.

Forty-three specimens were available for Ion Torrent analysis (31 spontaneous abscesses, three spontaneous subdural infections, and nine postoperative infections). The detailed Ion Torrent results for the 31 available spontaneous brain abscesses, including sample-specific cutoffs, number of reads per species, and rejected species are provided in Table S1 in the supplemental material. The average number of reads per sample was 535,635. After removal of the short reads (<200 bp) and chimeras, an average of 245,600 valid reads (q) remained (range, 81,836 to 736,566; median, 229,938).

An overview of all 37 spontaneous brain abscesses, including a comparison of the identifications made by the three available methods, is given in Table 2. A summary of all the detected bacterial species is provided in Table 3. Among the 160 species identifications from the 31 spontaneous abscesses available for MPS, 98 (61%) were detected by the modified Sanger approach, and 49 (31%) were successfully cultured. Two identifications were made exclusively by culture (one Mycobacterium tuberculosis and one Campylobacter gracilis). Characterizations and results for the postoperative infections and the subdural infections are summarized in Table S2 in the supplemental material.

The MPS part of the study included batch processing of amplified DNA. Although standard measures were taken to prevent the transfer of DNA from one sample to another, we did see evidence of low-level cross-contamination in some samples. This was suspected when a dominant species in one sample was seen as a minor constituent of another sample processed in the same batch. All 10 identifications considered to represent cross-contamination are marked in Table S1 in the supplemental material and were excluded from the results. In all but three samples (P12, P31, and P53), suspected cross-contamination also fell below the respective sample-specific cutoffs. Of the 160 accepted identifications made by MPS, 109 were confirmed by Sanger sequencing and/or culture performed sequentially following sample inclusion.

Internal organ abscesses are illustrative of the shortcomings of bacterial culture. They often contain anaerobic or fastidious bacteria vulnerable to sample collection procedures and transportation, and they can also contain atypical or slow-growing organisms that will not necessarily form colonies on routine media or during a standard incubation time. Also, antibiotic therapy is frequently initiated prior to sample collection due to delayed recognition and complicated access.

The amplification of bacterial 16S rRNA genes directly from clinical samples, followed by Sanger DNA sequencing, provided, for the first time, a universal culture-independent alternative for detecting and identifying bacteria (9). Although its usefulness has been well documented (5, 10–12), a major limitation has been its relative unsuitability for samples containing more than one species.

We hypothesize that polymicrobial infections undergo an ecological succession in which the relative concentration of a species varies at different stages of formation, maturation, and treatment (13, 14). The ability to detect only living or dominant bacteria at the moment of sample collection can make it difficult to recognize patterns or groups of core pathogens. The occasional isolation of organisms that are more difficult to culture can be confusing and result in concerns of a special infection that requires modifications to antimicrobial therapy. MPS does not have limitations related to sample complexity or concentration differences if used with a sufficiently high number of reads per sample. Although in the present study MPS generally increased the complexity of the results on the individual sample level, it revealed surprisingly homogeneous overall patterns for spontaneous brain abscesses. The 10 most prevalent genera and the 10 most prevalent species represented 92% and 70% of the identifications, respectively. The number of reads per species varied with a factor of ≥1,000 in several samples, confirming large concentration differences (see Table S1 in the supplemental material). The average 250,000 valid sequences per sample did not give a high enough resolution to detect background bacterial DNA in all specimens, indicating that an even higher number of reads could ideally have been used.

We find it unlikely that these complex infections are the result of a single bacterial seeding and believe that some species are pioneers, able to survive the oxygenated environment in the brain and prepare conditions for later entrance of strictly anaerobic bacteria (13–15). All 25 spontaneous polymicrobial abscesses were found to contain Aggregatibacter aphrophilus (n = 1), Fusobacterium nucleatum (n = 2), Streptococcus intermedius (n = 3), or combinations of these (n = 19). S. intermedius and A. aphrophilus were the only bacteria found in both polybacterial and monobacterial infections. F. nucleatum is a moderate anaerobe that tolerates low oxygen concentrations (16). Evidence exists that it possesses capacity for further oxygen adaptation (17, 18) and even that oxidative stress can increase Fusobacterium pathogenicity (19). All three species have repeatedly been reported from suppurative infections, including pulmonary empyema, liver abscesses, and brain abscesses (5, 20–23). We therefore suggest that A. aphrophilus, F. nucleatum, and S. intermedius are pioneer pathogens, permissive for the formation of polymicrobial brain abscesses.

It has been demonstrated that oral anaerobic bacteria can survive in an oxygenated environment by interacting with facultative or aerobic bacteria (15). In oral microbiology, F. nucleatum is considered a key organism in both biofilms and planktonic phases for bridging the transition between early aerobic/facultative colonizers and later obligate anaerobes by aggregating them into metabolically organized units in which the anaerobic bacteria are protected by the metabolism of aerobic and aerotolerant species (24, 25). We identified F. nucleatum group bacteria in 19 out of 20 brain abscesses containing strictly anaerobic species, strongly indicating an equally essential role in these infections.

Actinomyces species are infrequently cultured from brain abscesses and have historically been considered complicated to treat, with prolonged treatment recommendations of up to 6 months (26, 27). Our study shows that Actinomyces is among the most common genera in polymicrobial brain abscesses (14 out of 25 patients), although it was successfully cultured from only three patients in this study. These observations lead us to question the role of Actinomyces as an indicator of a special infection necessitating prolonged therapy. This is in support of a case series in which three brain abscesses with Actinomyces were successfully treated with burr hole aspiration and short-course antimicrobial therapy (3 to 4 weeks) (28) and in concordance with treatment experiences from other body localizations (29). In the present study, the median duration of antimicrobial treatment for the actinomycotic abscesses was 9 weeks (range, 6 to 16 weeks), and no recurrences were reported after a minimum observation time of 1 year.

Campylobacter gracilis (previously Bacteroides gracilis) and Campylobacter rectus are the most dominant Campylobacter species in the oral cavity and represent another long-acknowledged genus in intracranial infections rarely recovered by culture (1, 4, 30). In this study, C. gracilis (n = 8) was found exclusively in samples positive for Actinomyces spp. (n = 14), and Eikenella corrodens (n = 6) was found exclusively in samples with C. gracilis, indicating that Actinomyces might be important for the introduction of these two fastidious organisms into the abscess. Coaggregation between C. gracilis and Actinomyces spp. has also been demonstrated experimentally (31). In general, if some species already exist as coaggregates in the oral cavity or other origins, it is reasonable to consider that bacterial seeding can also take place in the form of aggregates. Also A. aphrophilus was found to be associated with Actinomyces spp. (8 out of 11 isolations).

The samples from intracranial postoperative infections contained Propionibacterium acnes, Staphylococcus aureus, and coagulase-negative staphylococci. This is in concordance with previous investigations. For the three subdural empyemas, no particular patterns were observed, and as a group, they did not share the characteristics discussed for spontaneous brain abscesses.

The presence of bacterial DNA in everything from sample collection devices to lab reagents is a major concern when performing universal 16S rRNA gene amplification directly from clinical samples (32, 33). For handling this in MPS reactions, in which the high number of wells can result in the cosequencing of background bacterial DNA even in strongly positive samples (e.g., if a clinical specimen contains 1,000 times more bacterial DNA than the negative control, in the result, 1:1,000 reads is likely to represent background DNA), we introduce a new sample-specific cutoff grounded on the relative quantifications of bacterial DNA obtained in the real-time universal 16S PCR. Incorporated in the cutoff is a three-cycle (8-fold) safety margin. It could be argued from our data (see Table S1 in the supplemental material) that this is a conservative approach, but the very high DNA concentrations found in purulent specimens can reduce PCR efficacy and consequently the ΔC_T value in some samples (thereby giving a higher cutoff for a valid species). Regardless of the number of cycles chosen as a safety margin, the principle of estimating a sample-specific cutoff remains valid.

Because of the strict criteria needed to reliably define a true positive sample, a universal 16S PCR will have a significantly lower sensitivity than a species-specific PCR. This is not altered by the introduction of MPS, as exemplified by the two isolations made exclusively by culture in this study. In sample P36, M. tuberculosis was both cultured and detected by a M. tuberculosis complex-specific PCR, whereas the universal 16S PCR reached C_T after the negative control. Upon Ion Torrent sequencing, a few reads from M. tuberculosis were actually detected but at lower levels than a range of typically environmental species (see Table S1 in the supplemental material). In the polymicrobial sample P21, which was clearly positive by the universal PCR (ΔC_T = 15.7) one of the culture-proven species (C. gracilis) is still present in a low concentration comparable to that of the bacterial background and drops below cutoff (see Table S1).

In Norway, ceftriaxone or cefotaxime combined with metronidazole is the first-line therapy for brain abscesses, with a suspected primary focus in the oral cavity, sinuses, ears, or lungs, or with an unknown primary focus. None of the species detected exclusively by MPS in this study would be considered to have reduced susceptibility to this regimen. However, atypical organisms, like Mycoplasma spp., have been reported by others (1, 2) in polymicrobial brain abscesses, and the ability to detect unexpected pathogens remains among the most important tasks for diagnostic microbiology. Reliable diagnostics will also encourage clinicians to customize treatment once the results are available and can lead to more targeted and narrow-spectrum treatment.

To the best of our knowledge, this is the first systematic investigation of a human bacterial infection using MPS. No relevant gold standard is available for comparison. Although we were able to confirm 68% of the identifications with culture or Sanger sequencing, it will be important that both our findings and methodological approach are challenged by other research groups.

This work was supported by the Department of Microbiology, Haukeland University Hospital, Norway, and the Department of Microbiology and Infection Control, Akershus University Hospital, Norway. The reagents for Ion Torrent sequencing were provided by Life Technologies (California, USA).

Ø.K. and Ø.S. are cofounders and shareholders of iSentio AS (Bergen, Norway).