Inter-bacterial correlations in subgingival biofilms: a largescale

Transcription

1 J Clin Periodontol 2014; 41: 1 10 doi: /jcpe Inter-bacterial correlations in subgingival biofilms: a largescale survey Loozen G, Ozcelik O, Boon N, De Mol A, Schoen C, Quirynen M, Teughels W. Inter-bacterial correlations in subgingival biofilms: a large-scale survey. J Clin Periodontol 2014; 41: doi: /jcpe Abstract Aim: Although the complexity of the oral ecology and the ecological differences between health and disease are well accepted, a clear view on the dynamics in relation to disease is lacking. In this study, the prevalence and abundance of 20 key oral bacteria was assessed in health and disease and more importantly a closer look was given to the inter-bacterial relationships. Materials and methods: A blinded microbiological database was analysed in this cross-sectional, retrospective study. The database was constructed based on microbiological analyses of samples from 6308 patients, with gradations of periodontitis (healthy to periodontitis). Data concerning the abundance of 20 oral bacteria and probing pocket depth were provided. Results: Porphyromonas gingivalis, Tannerella forsythia, Treponema denticola, Eubacterium nodatum, Porphyromonas micra and Porphyromonas intermedia showed a clear increase in abundance and prevalence with increasing pocket depth. Correlation matrices illustrated that almost all microorganisms were in one way correlated to other species and most of these correlations were significant. Several beneficial bacteria showed strong correlations with other beneficial bacteria. Conclusion: Knowledge on bacterial correlations can pave the way for new treatment options focusing on restoring the shifted balance. Gitte Loozen 1, Onur Ozcelik 2, Nico Boon 3, Antony De Mol 1, Christian Schoen 1, Marc Quirynen 1 and Wim Teughels 1,4 1 Department of Oral Health Sciences, KU Leuven & Dentistry, University Hospitals Leuven, Leuven, Belgium; 2 Department of Periodontology, Cukurova University, Adana, Turkey; 3 Laboratory of Microbial Ecology and Technology (LabMET), Ghent University, Gent, Belgium; 4 Fund for Scientific Research Flanders, FWO, Brussels, Belgium Equally contributing authors. Key words: bacteria; ecology; microbiology; periodontitis Accepted for publication 8 September 2013 Currently, periodontal diseases are considered as polymicrobial infections but for a long time the presence of several key pathogenic bacteria was considered crucial in Conflict of interest and source of funding statement The authors declare that there are no conflicts of interest in this study. This study was supported by grants of the KU Leuven (OT 07/057) and by grants of the Research Foundation Flanders (G and N). the onset of periodontitis. Furthermore, it was assumed that the start of periodontitis was accompanied by a shift from a healthy oral microbiota, consisting primarily of Gram-positive, saccharolytic, and facultative anaerobic bacteria to one consisting primarily of Gram-negative and obligate anaerobic species (Slots 1977, Paster et al. 2001). The studies of Socransky et al. (1998) have strengthened this assumption, with the introduction of colour-coded complexes to group oral bacteria into distinguishable clusters associated with health or disease. Two complexes were clearly associated with periodontal disease, the red complex consisting of Porphyromonas gingivalis, Tannerella forsythia and Treponema denticola and the orange complex harbouring Prevotella intermedia, Fusobacterium nucleatum, Prevotella nigrescens and Parvimonas micra. Bacteria in these complexes showed a strong correlation with pocket depth, with increases in prevalence and numbers when pocket depth increased. Furthermore, red complex bacteria were also strongly associated with bleeding 1

2 2 Loozen et al. upon probing. Other complexes, harbouring Gram-positive bacteria like Streptococcal species and Actinomyces species, were correlated with health. Although Gram-negative bacteria could play a role in the onset of disease, the presence of these species does not imply a certainty that disease will occur. Healthy periodontal tissues have been shown to harbour these pathogenic microorganisms, albeit in low numbers (Marsh 2003). Consequently, other microorganisms could also be implicated in disease. These observations have led to new ideas concerning the aetiology of periodontal infections. New theories consider the oral cavity as a particular ecosystem, in which the host, the environment and microorganisms all contribute to the onset of disease (Marsh 2003, Darveau 2010). According to this ecological plaque hypothesis in healthy circumstances a symbiotic balance exists between the host and all microorganisms. However, a change in the number of pathogenic or beneficial bacteria and their balance, or a change/activation in/of the host response could lead to the onset of periodontitis. For this two-way interaction, it is unclear if it is the microbial dysbiosis which leads to the inflammatory destruction of the periodontal tissues or whether it is the inflammatory response which leads to a dysbiosis. Modern molecular techniques allow focusing on this ecological concept via advanced DNA sequencing approaches. These community profiling techniques attempt to capture/analyse the whole community instead of focusing on specific taxa. They can discover new microorganisms implicated in periodontitis since the oral ecosystem is not formed by just a few players. Griffen et al. (2012) used pyrosequencing of 16S rrna genes to map microbial community differences between periodontitis and health. Data confirmed the association of specific Gram-negative species with periodontitis, like P. gingivalis, T. denticola and T. forsythia However, a large number of additional species also showed significant associations with the disease. Especially, Filifactors alocis, a Gram-positive bacterium and some other yet uncultivated species seemed to be associated with periodontal disease. Several microorganisms showed to be correlated with health under which the Streptococcal, Acinetobacter, Actinomyces and Moraxella species were the most important ones. Although the complexity of the oral ecology and the ecological differences between health and disease are well accepted, a clear view on the dynamics in relation to disease severity or progression is lacking. The ecological changes in relation to disease progression are difficult to study longitudinally since it is impossible to determine in advance which subjects will develop disease and once disease starts to develop, it is unethical not to intervene. Despite these problems, a few studies have attempted to follow the microbiological changes occurring during the transition from health to disease (MacFarlane et al. 1988, Machtei et al. 1997, Papapanou et al. 1997, Tanner et al. 1998). Machtei et al. (1997) established that presence of T. forsythia, P. intermedia and P. gingivalis was a prognostic factor for further periodontal breakdown. Whereas data from Tanner et al. (1998) suggested that T. forsythia, Campylobacter rectus and Selenomonas noxia were major species characterizing sites converting from periodontal health to disease. However, because of above mentioned reasons most data concerning ecological changes in relation to disease progression are derived from crosssectional studies (Socransky et al. 1998, Haffajee et al. 2008). In these studies, microbiological differences between healthy, gingivitis, moderate or severe periodontitis patients or between shallow, moderate and deep pockets are used to describe the microbial dynamics from health to disease. In order to understand the ecological dynamics within periodontal pockets in relation to disease progression, it is necessary to study microbial changes in relation to increasing periodontal pocket depths (PD). Since longitudinal studies are considered unethical and seeing the wide variety of possible probing PD s, cross-sectional studies need to incorporate large numbers of patients to provide sufficient power. The associated costs with microbial analyses often hamper analysing large numbers of patients. The limited number of patients that can be analysed also hampers the possibility to study the effect of the outgrowth of one member of the ecology on the other members of this ecology. Such analysis could potentially provide valuable information regarding inter-species interactions and associations within the dynamic microbial ecology. Therefore, the aim of this study was to analyse the microbial changes within the subgingival microbial ecology in relation to probing pocket depth and in relation to individual species numbers based on a set of 6308 unique subgingival plaque samples which were analysed with a DNA chip for 20 representative oral species. Material and Methods Sample population A blinded microbiological database for the years 2007 through 2009, provided by Advanced Dental Diagnostics B.V. (ADD) (Malden, the Netherlands), was screened and analysed in this cross-sectional, retrospective study. The study received approval from the Ethical Committee of the University Hospital of the Catholic University of Leuven (ML 9333). The database was constructed based on microbiological analyses of patient samples provided by oral healthcare workers in the Netherlands, Belgium and Germany. Next to the pooled sample, clinical patient and sample related information based on standard questionnaire forms was provided. These forms contained information regarding probing pocket depth, gender, date of birth, smoking behaviour, oral hygiene and treatment stage. The sample population consisted of patients from Western Europe, predominantly the Netherlands and Germany. If the questionnaire did not provide the primarily required patient/sample information, the patient/sample was excluded from further analysis. Based on the provided information, the average PD, maximum PD, number of pooled sites and patient age at the moment of sample taking were calculated. Since samples in this database were derived from pooled samples, the

3 Microbiological shift in periodontitis 3 PD and correlating standard deviation (SD) of each participating individual was determined as an intra-subject average over the one to five registered PD at the sampling sites. Patients with PD-values over 12 mm were excluded from analysis. As such the population consisted of 12,884 patients. For further analysis, the sample population was divided into two main groups: patients/samples before initial therapy (abbreviated as BIT; n = 6308) and patients/samples after initial therapy but before surgery (abbreviated as AIT; n = 6576). In this study, only the group of patients/samples before initial therapy were considered (BIT; n = 6308). The population was arbitrarily divided into 8 PD-categories: <3(equal or smaller than 3); >3 (larger than 3); >4; >5; >6; >7; >8; >9. Microbiological assessment Subgingival plaque samples were taken by dental clinicians with access to case history, radiographs and clinical data. Clinicians had a choice to take between 1 and 5 samples using sterile paper points. These paperpoints were deposited into marked eppendorf tubes and sent to the company (ADD) at room temperature for molecular microbiological assessment. DNA from the samples was obtained using the Quickextract DNA extraction (Epicentre Biotechnologies, Madison, WI, USA) according to the manufacturer s instructions. The polymerase chain reaction (PCR) was performed in a total volume of 20 ll using 0.2 ml PCR reaction tubes (Greiner bio-one, Frickenhausen, Germany) as described in Vienna et al. (2005). In short, the PCR reaction mixture contained 18.8 ll of PCR MasterMix (Parocheck20 â kit Greiner, Frickenhausen, Germany), 1.25 units of Taqpolymerase (Promega, Madison, WI, USA) and 1 ll of bacterial DNA (2 10 ng/l). The negative control contained ultrapure water in place of bacterial DNA. Amplification was performed on a GeneAmp 9700 (Perkin Elmer, Fremont, CA, USA). Assay conditions consisted of an initial minute at 94 C, followed by 45 cycles of 95 C for 20 s, 60 C for 20 s and an elongation step at 72 C for 1 min. Afterwards, samples were kept at 22 C. Heating ramp was set to 1 C/s and the temperature of the lid was 105 C. This assay allowed for the amplification of parts of the specific 16S rrna gene in the presence of fluorescence-labelled primers. Hybridization was performed according to the instructions of the ParoCheck â DNA chip manufacturer (Greiner Bio-One) (Vienna et al. 2005). This chip contained specific probes for the 20 different periodontal taxa as outlined in Table 1. Chips were analysed on an Axon Genepix4000 scanner (Molecular Devices, Sunnyvale, CA, USA). PMT-gain was adjusted to 600 (Cy3) and 700 (Cy5) respectively. The start resolution was set to 10 lm. Analysis was possible if controls (hybridization control, PCR control and orientation control) had been passed. Values were plotted as signal to noise ratios (SNR). The average of 5 spots represented one microbe. These average SNR values were then translated into bacterial amounts with the help of a standard curve. Such curves (SNR values versus bacterial amounts) were available for all 20 strains. The limit of detection of the ParoCheck â DNA chip was determined by multiple analysis of serial dilutions of bacterial DNA under presence of sample matrix. The Table 1. limit of detection for each bacterium was calculated for a cut off value of the signal to noise ratio equal 12. Statistical analysis Multiple tables were created from the data set to be able to allow for constrained ordination and correlations. Microbial data were available in terms of counts of the 20 test species. PD s were considered as a variable discontinuous factor and all PD higher than 10 mm were set as 11 mm. Bacterial numbers were log transformed. Before correlation assessment, data were displayed by use of cumulative histograms. The relationship between the number of microorganisms and the PD was assessed with the spearman correlation coefficient. Significance of the correlation was assessed with Chi-square test and correction for multiple comparisons was carried out. The relation between the PD and the absence or presence of a certain bacterium was evaluated by means of a generalized linear model, from which a coefficient was calculated. In both cases, a positive value of the correlation coefficient represents a positive relation, a negative value depicts a negative relation. Afterwards, the Twenty periodontal bacteria detectable with Parocheck â kit Abbreviation (as used in figures and tables) Aggregatibacter actinomycetemcomitans Actinomyces odontolyticus Actinomyces oris* Tannerella forsythia Campylobacter concisus Campylobacter gracilis Campylobacter rectus Capnocytphaga (gingivalis/sputigena/ochracea) species Eikenella corrodens Eubacterium nodatum Fusobacterium nucleatum Parvimonas micra Porphyromonas gingivalis Prevotella intermedia Prevotella nigrescens Streptococcus constellatus group (constellatus, intermedius, anginosus) Streptococcus gordonii group or mutans group (gordonii, salivarius, cricetus, mutans, uberis) Streptococcus mitis group (mitis, oralis, sanguinis, parasanguinis) Treponema denticola Veillonella parvula A.a A.o A.or T.f C.c C.g C.r C.s E.c E.n F.n P.m P.g P.i P.n Scg Sgg Smg T.d V.p *Contains strains previously identified as Actinomyces naeslundii serotypes II, III and NV and Actinomyces viscosus serotype II (Henssge et al. 2009).

4 4 Loozen et al. relation between the different microorganisms was assessed. Spearman rank correlations were calculated and significance of the correlation was evaluated with Chi-square test and correction for multiple comparisons was carried out. All assessments were done based on the whole data set and based on the data without zero values (data not shown). Finally, an evaluation was made based on the degree on which microorganisms appear together. Phi was calculated as a measure of association of presence (0: no association; 1: strong association). p- value was calculated with the Chisquare test. Results Correlation between bacterial species and pocket depth Figure 1 showed that for P. gingivalis, T. forsythia, T. denticola, P. intermedia, P. micra and Eubacterium nodatum the prevalence increased with increasing pocket depth, whereas the opposite was true for A. actinomycetcomitans, Prevotella nigrescens, Campylobacter gracilis, Eikenella corrodens, Capnocytophaga species, Campylobacter concisus, Streptococcus mitis group, Streptococcus gordonii group, Streptococcus constellatus group, Actinomyces odontolyticus and Actinomyces oris. These microorganisms showed a higher prevalence in shallow (<3 mm) and intermediate pockets (3 6 mm). Prevalence of F. nucleatum, C. rectus and Veillonella parvula remained relatively stable between the various PD. Instead of looking qualitatively (Fig. 1), data were also quantitatively approached (Fig. 2). P. gingivalis, T. forsythia, T. denticola, E. nodatum, P. micra and P. intermedia showed a clear increase in abundance with increasing PD (Fig. 2), However, the majority of the investigated microorganisms (A. actinomycetemcomitans, P. nigrescens, E. corrodens, Capnocytophaga species, C. gracilis, C. concisus, S. mitis group, S. constellatus group, S. gordonii group, A. odontolyticus, A. oris and V. parvula) showed a decrease in abundance with increasing PD. For two microorganisms, F. nucleatum and C. rectus, the abundance remained stable when going from a healthy to a diseased pocket. Table 2 demonstrates the correlation coefficient between the bacterial species and the PD based on the abundance of the microorganisms as well as the prevalence of the microorganisms in relation to PD. Values in bold represent all the significant correlations (p < 0.05). Correlations were divided into categories based on their strength. Between 0 and 0.1: none to very weak correlation; : weak correlation; : moderate correlation; 0.7 1: strong correlation (Dancey & Reidy 2004). The strongest correlation was observed for T. forsythia and T. denticola. Looking at prevalence, these microorganisms showed a strong positive correlation with PD and a weak positive correlation for abundance. This clearly indicates that the prevalence of these two microorganisms, as well as their abundance to a lesser extent, increases with increasing pocket depth. A. oris on the other hand displayed a moderate negative correlation with PD for both prevalence and abundance. V. parvula also showed a moderate negative association with PD when looking at prevalence; however, this was only a weak negative association when looking at abundance. For most other microorganisms, weak correlations were observed for abundance as well as prevalence. For A. actinomycetemcomitans, P. intermedia, C. rectus, P. micra and P. nigrescens only very weak correlations were demonstrated. Inter-bacterial relationships between tested oral microorganisms based on abundance data Figure 3 indicates the different positive or negative correlations between the different microorganisms. Several correlations could be observed and most correlations were significant (p < 0.05). Strong positive correlations exist between T. forsythia and T. denticola and between all three streptococcal groups. Furthermore, A. oris showed a strong positive correlation with the S. mitis group and S. gordonii group. All significant (p < 0.05) weak, moderate and strong correlations were selected and displayed in Figures S1 S5 and Table S1. These graphs show a sub analyses of the data set where the microbiolog- 120% <3 >3 >4 >5 >6 >7 >8 >9 100% 80% 60% 40% 20% 0% Aa Pg Tf Td Pi Fn Pm Pn Cg Cr En Ec Cs Cc Smg Sgg Scg Ao Aor Vp Fig. 1. Prevalence of bacterial species in relation to pocket depth (PD). Percentages are displayed for each bacterium in function of the specific pocket depth category. PDs are divided into eight categories (mm).

5 Microbiological shift in periodontitis <3 >3 >4 >5 >6 >7 >8 >9 5 log (bacteria/ml) Aa Pg Tf Td Pi Fn Pm Pn Cg Cr En Ec Cs Cc Smg Sgg Scg Ao Aor Vp Fig. 2. Log CFU of bacterial species in relation to pocket depth (PD). Bacterial numbers (log CFU) were displayed for each bacterium in function of the specific pocket depth category. PD s were divided into eight categories. Error bars represent standard errors of the mean (n = 6308). CFU, colony forming units. Table 2. Correlation coefficient bacterial species and PD. Coefficients are displayed in two categories: correlation between abundance of species and PD and correlation between detection frequency and PD. Correlations were divided into categories based on their strength. Between 0 and 0.1: none to very weak correlation (no colour); : weak positive correlation (yellow); : moderate correlation (orange); 0.7 1: strong correlation (red); 0.1 to 0.3: weak negative correlation (green) (Dancey & Reidy 2004). All values in bold represent significant correlations (p < 0.05) Abbreviations as in Table 1. ical changes were not expressed in relation to PPD but in relation to the concentration of the bacterial species that had a significant correlation with each other. The histograms showed that T. forsythia and T. denticola decreased when the concentration of A. oris, S. constellatus group, S. gordonii group, S. mitis group, Capnocytophaga species or E. corrodens increased and vice versa. Furthermore, an increase in A. oris also led to decreases in C. rectus, whereas an increase in the S. mitis group was accompanied with a decrease in P. gingivalis. Higher concentrations of F. nucleatum, A. odontolyticus and C. concisus led to higher concentrations of all microorganisms shown in the histogram. P. nigrescens was positively correlated with P. intermedia, but its increase also led to decreases in E. nodatum. C. gracilis and C. concisus were positively correlated with all Streptococci and A. oris among others. However, they also showed a positive correlation with C. rectus. Inter-bacterial relationships between tested oral microorganisms based on prevalence Table 3 gives an overview of the significant correlations between microorganisms based on prevalence. It was observed that several microorganisms increased in prevalence when another bacterium was present. P. gingivalis showed an increase in prevalence when T. forsythia and T. denticola were present, whereas no effect of the presence of P. gingivalis on detection percentages of T. forsythia and T. denticola was observed. It was also clear that T. forsythia and T. denticola were dependent of each other, percentages increased when they were present together. The dependence on the other microorganisms was slightly more pronounced for T. forsythia. Presence of T. denticola also caused an increase in detection of P. micra, P. intermedia and C. rectus, vice versa no effect was shown. Such a onesided influence was also seen for the S. constellatus group and A. odontolyticus, and the S. constellatus group and V. parvula. The prevalence of this Streptococcus group clearly increased when A. odontolyticus and V. parvula were present. Additionally, A. oris seemed to be dependent

6 6 Loozen et al. Aa Tf Pi Pg Fn Cc Ec Sgg Ao Vp Cs Smg Scg Cr Aor Cg Pm Pn En Td Fig. 3. Overview of inter-bacterial correlations based on abundance data. Figure shows all significant (p < 0.05) bacterial correlations with a correlation coefficient above 0.1. Correlations were divided into categories based on their strength. Between 0.1 and 03: weak correlation (thin line); : moderate correlation (semi-thick line); 0.7 1: strong correlation (very thick line). Positive correlations are displayed in green, negative correlations in red. on presence of V. parvula. For the rest of the correlations in Table 3, dependence was observed in two directions. Detection of the S. mitis group for instance increased when microorganisms from the S. constellatus group were present and similarly the latter group increased when microorganisms from the S. mitis group were present. Discussion As the oral microbiota can contain over 700 organisms, different microorganisms could play important Aa Tf Pi Pg Fn Cc Ec Sgg Ao Vp Cs Smg Scg Cr Aor Cg Pm Pn En Td roles. It is possible that the whole oral microbiota displays a polymicrobial synergy as suggested in current aetiological hypotheses (Marsh 2003, Darveau 2010). Periodontitis could then be caused by a dysbiotic change in the complex oral microbiota, for instance a change in the relative abundance of certain members relative to their abundance in health. Such a change could disturb the host-microbe relationship leading to inflammation. In this study, the prevalence and abundance of 20 representative oral bacteria was assessed in health and disease and more importantly a closer look was given to the inter-bacterial relationships. First, the prevalence and relative abundance of the microorganisms was examined in 6308 patients, presented with different gradations of periodontitis from healthy patients to patients with severe periodontitis. Furthermore, the correlation of prevalence and abundance of bacterial species with PD was assessed. When comparing our results with the colour-coded complexes described by Socransky et al. (1998), it was shown that the red complex bacteria also showed a clear association with disease in our study. T. forsythia, T. denticola and P. gingivalis showed a clear increase in prevalence and abundance with increasing pocket and they displayed an apparent positive correlation with PD. All these bacteria are strictly anaerobic bacteria this in part explains the higher frequency and numbers in deeper pockets, an extra anaerobic environment. Their correlation with disease has also been shown in other studies (Socransky et al. 1998, Ximenez-Fyvie et al. 2000, Haffajee et al. 2008, Griffen et al. 2012). Additionally, our study also illustrated a diseaseassociation for P. intermedia, P. micra and E. nodatum, based on their numbers and prevalence in deeper pockets. Except for P. micra they also showed a positive correlation with PD, although this was only a weak correlation. These bacteria were members of the orange complex bacteria in the study by Socransky et al. (1998), that complex was also shown to be strongly correlated with the red complex and with disease in that study. Different bacteria displayed a decrease in prevalence and abundance with an increase in PD and a clear negative correlation with PD in our study under which Streptococcal, Capnocytophaga and Actinomyces species, V. parvula, E. corrodens and C. concisus. These species groups have already been associated with a healthy periodontium in earlier studies (Griffen et al. 2012, Bik et al. 2010, Roberts & Darveau, 2002). In the study of Socransky et al. (1998), they were also defined as non-disease associated, situated in the green, yellow or purple complexes. However, some discrepancies were also shown between our study and the analysis by Socransky et al.

7 Microbiological shift in periodontitis 7 Table 3. Correlation based on detection: Phi value (value between 0 and 1 that depicts the association of two bacteria based on appearance) and Chi-square (p-value) was calculated for all bacterial combinations. Those bacterial combinations that showed a significant p-value (<0.05) and a difference of factor 0.2 between the percentages in A + B and A + B + or B A + and B + A + were displayed in the table. Four detection percentages are shown: B A + : detection percentage of A when B is absent; A B + : detection percentage of B when A is absent, A + B + : detection percentage of B when A is present, B + A + : detection percentage of A when B is present Detection percentage (%) A B Phi value A + B + A B + B + A + B A + A.a C.s P.g T.f P.g T.d T.f T.d T.f E.n T.d P.m T.d P.i T.d C.r P.m E.n P.n C.s C.g C.c C.g E.c C.r C.c E.c C.s E.c C.c E.c Smg E.c A.or C.s V.p Smg Scg Smg A.or Sgg Scg Sgg A.or Scg A.o Scg V.p A.o A.or A.or V.p Abbreviations as in Table 1. (1998). In our study, a decrease in prevalence and abundance was observed for both P. nigrescens and C. gracilis and a negative correlation with PD, whereas they were positioned in the orange disease-associated complex by Socransky et al. (1998). Furthermore, F. nucleatum and C. rectus, both defined as orange-complex bacteria, showed a decrease in numbers and detection frequency with disease and a very weak correlation with PD was observed. Our results showed that F. nucleatum had a high prevalence and abundance in all PD categories. This might be explained by the co-aggregation capacities of this bacterium, which is able to form bridges between early and late colonizers (Kolenbrander et al. 2002). A. actinomycetemcomitans in other studies often mentioned as an important player in aggressive periodontitis also displayed a very weak negative correlation with PD. This bacterium was also positioned in a distinct complex in the study by Socransky et al. (1998). After establishing which microorganisms were associated with disease or health, the inter-bacterial relationships were assessed. Correlation matrices illustrated that almost all microorganisms were in one way correlated to other species and most of these correlations were significant. Histograms displayed the weak, strong and moderate correlations and showed clear microbial shifts. When observing the correlation between the streptococcal groups and other bacteria, it has to be kept in mind that these groups consist of several streptococcal species. Especially the S. gordonii group is important here as it consist of earlier defined beneficial species such as S. salivarius as well as the cariogenic bacterium S. mutans. The most obvious negative correlations were observed between the disease-associated bacteria T. forsythia and T. denticola and some health-associated bacteria (Streptococcal groups, A. oris, V. parvula, E. corrodens and Capnocytophaga species). Studies concerning T. forsythenis and T. denticola are scarse as these bacteria are difficult to culture. However, it is known that the Streptococci often produce bacteriocins, inhibiting pathogenic bacteria (Santagati et al. 2012). Hillman et al. (1985) already described a negative correlation between the viridians streptococci (S. sanguinis, S. uberis) and T. forsythia. A study by Apol^onio et al. (2007) also demonstrated that E. corrodens was able to produce antagonistic substances effective against several oral bacteria, such as P. gingivalis. Most correlations in the matrices were positive and several of the bacterial interactions displayed in the histograms have already been described in earlier studies. A mutual symbiotic relationship between T. denticola and P. gingivalis has been described, succinic acid a growth factor produced by T. denticola appears to be incorporated in the cell envelop of P. gingivalis (Grenier 1992). The strong correlation bet-ween T. forsythia and T. denticola has already been described in several studies (Bolstad et al. 1996, Ikegami et al. 2004). As F. nucleatum is considered a bridging species it displays a positive correlation with several oral bacteria. Aggregations have already been described between F. nucleatum and P. gingivalis (Kolenbrander & London 1993, Kolenbrander et al. 2002, Periasamy & Kolenbrander 2009), Gram-positive species (Kolenbrander et al. 1989, Kang et al. 2005), Prevotella species (Okuda et al. 2012a), E. nodatum (George & Falkler 1992), P. micra (Kremer & Van Steenbergen 2000), T. forsythia (Sharma et al. 2005), T. denticola (Kolenbrander et al. 1995) and Capnocytophaga species (Okuda et al. 2012b) among others. The strongest positive correlations were described between the Streptococcal species, V. parvula, A. oris and E. corrodens. Streptococcus species and Veillonella species have already been found to occur often at the same site in the oral cavity, most probably because Veillonella utilizes short-

8 8 Loozen et al. chain acids such as lactate secreted by Streptococci (Kreth et al. 2008). Ebisu et al. (1988) demonstrated coaggregation between E. corrodens and A. oris and S. sanguinis by means of a bacterial lectin-like substance. However, it was also observed that three bacteria, whose increase led to increases in the concentration of several health-associated bacteria such as the Streptococcal species, and from which the concentration and prevalence decreased with increasing PD, also had a strong correlation with one or more disease-associated bacteria. P. nigrescens showed a positive correlation with P. intermedia and C. gracilis was closely related to C. rectus which may also explain why these bacteria were earlier defined as a member of the disease-associated orange complex. Furthermore, A. odontolyticus showed a positive correlation with both P. gingivalis and P. intermedia. The correlation between A. odontolyticus and P. intermedia was already described by Nesbitt et al. (1993). Besides correlations based on abundance, correlations were also presented based on detection frequencies, which illustrated that several bacteria are dependent on the presence of other bacteria, as was shown for P. gingivalis whose detection percentage increased when T. forsythenis or T. denticola were present, the relation between P. gingivalis and T. denticola was already mentioned before (Grenier 1992). This large-scale retrospective study was able to correlate several bacteria with disease and illustrated several bacterial interactions. However, a critical view on the used methodology is required. It has to be kept in mind that the method of data collection lent itself to variability due to participation of an unknown number of non-calibrated dental practitioners. However, the extent of the sample population should compensate for this suspected variability. Furthermore, a downside of the study is that a variable number of sites were sampled from each subject and that these sites could be of a different pocket depth. Therefore, the analysis was based on the mean pocket depth of the samples sites. By this, some effects might have been overlooked. However, the average SD of these mean PD was limited to 0,95 mm. Often clinical and microbiological data are grouped for analysis in shallow ( 3 mm), moderate (4 6 mm) and deep pockets ( 7 mm). This approach groups data with a variation of 2 or 3 mm and more. For 69.5% of the samples, the difference between the largest and the smallest sampled pocket was 2 mm, for 87.2% of the samples, this difference was 3 mm. In this view, however, taking into account that average PD was used, and seeing the limited possibilities one has to look at these events using large population based samples, the authors feel that the error induced by this approach is limited. Another limitation of the study is that only 20 oral bacteria are examined. As the oral cavity can harbour over 700 microorganisms, other bacteria can also play a crucial role. However, the 20 bacteria investigated in the current study are all key bacteria in the oral environment and have proven to play a crucial role in the oral cavity (Socransky et al. 1998). In conclusion, many significant inter-bacterial correlations were observed. In general, it appeared that beneficial bacteria have a bigger network, correlating strongly with other beneficial bacteria. It has to be investigated in the future whether a combination of several of these beneficial bacteria can lead to a greater effect, then a single beneficial bacterium. Furthermore, it has to be kept in mind that although a species seems to have no correlation with clinical factors its increase could indirectly lead to disease. The results of this study give a suggestion of a microbial profile correlated with periodontal disease or health. Whereas higher prevalence and abundance of T. forsythia, T. denticola, P. gingivalis, P. intermedia, P. micra and E. nodatum are correlated with disease, several Gram-positive bacteria and bacterial groups (Streptococcal species, V. parvula, A. oris, E. corrodens) are correlated with periodontal health. In addition and maybe more importantly, a framework for microbial interactions based on correlations between 20 species has been given. This might help to understand and give a basis to further investigate community changes and inter-microbial interactions. In this study only a small, but relevant array of oral bacteria was tested, in the future analysis could be performed working with the human microbe identification microarray (HOMIM). A study by Ahn et al. (2011) already compared hybri -dization based on HOMIM with pyrosequencing and demonstrated that both techniques were highly correlated on both phylum and genus level. However, current knowledge on bacterial correlations and relations already paves the way for possible new treatment options not focusing on removal of the whole oral microbiota, but focusing on these bacterial interactions thereby restoring the shifted balance. Acknowledgements We are grateful to Advanced Dental Diagnostics B.V. for providing the microbiological data set on which analysis was performed. Particular thanks go to Peter Camp from Advanced Dental Diagnostics B.V. who passed away unexpectedly during the preparation of this manuscript. References Ahn, J., Yang, L., Paster, B. J., Ganly, I., Morris, L., Pei, Z. & Hayes, R. B. (2011) Oral microbiome profiles: 16S rrna pyrosequencing and microarray assay comparison. PLoS ONE 6, e Apol^onio, A. C. M., Carvalho, M. A. R., Ribas, R. N. R., Sousa-Gaia, L. G., Santos, K. V., Lana, M. A., Nicoli, J. R. & Farias, L. M. (2007) Production of antagonistic substance by Eikenella corrodens isolated from the oral cavity of human beings with and without periodontal disease. Journal of Applied Microbiology 103, Bik, E. M., Long, C. D., Armitage, G. C., Loomer, P. M., Emerson, J., Mongodin, E. F., Nelson, K. E., Gill, S. R., Fraser-Liggett, C. M. & Relman, D. A. (2010) Bacterial diversity in the oral cavity of 10 healthy individuals. The ISME Journal 4, Bolstad, A. I., Jensen, H. B. & Bakken, V. (1996) Taxonomy, biology, and periodontal aspects of Fusobacterium nucleatum. Clinical Microbiology Reviews 9, Dancey, C. & Reidy, J. (2004) Statistics without maths for psychology: using SPSS for windows. London: Prentice Hall. Darveau, R. P. (2010) Periodontitis: a polymicrobial disruption of host homeostasis. Nature Reviews Microbiology 8, Ebisu, S., Nakae, H. & Okada, H. (1988) Coaggregation of Eikenella corrodens with oral

9 Microbiological shift in periodontitis 9 bacteria mediated by bacterial lectin-like substance. Advances in Dental Research 2, George, K. S. & Falkler, W. A., Jr (1992) Coaggregation studies of the Eubacterium species. Oral Microbiology and Immunology 7, Grenier, D. (1992) Nutritional interactions between two suspected periodontopathogens, Treponema denticola and Porphyromonas gingivalis. Infection and Immunity 60, Griffen, A. L., Beall, C. J., Campbell, J. H., Firestone, N. D., Kumar, P. S., Yang, Z. K., Podar, M. & Leys, E. J. (2012) Distinct and complex bacterial profiles in human periodontitis and health revealed by 16S pyrosequencing. The ISME Journal 6, Haffajee, A. D., Socransky, S. S., Patel, M. R. & Song, X. (2008) Microbial complexes in supragingival plaque. Oral Microbiology and Immunology 23, Henssge, U., Do, T., Radford, D. R., Gilbert, S. C., Clark, D. & Beighton, D. (2009) Emended description of Actinomyces naeslundii and descriptions of Actinomyces oris sp. Nov. and Actinomyces johnsonii sp. nov., previously identified as Actinomyces naeslundii genospecies 1, 2 and WVA 963. International Journal of Systemic and Evolutionary Microbiology 59, Hillman, J. D., Socransky, S. S. & Shivers, M. (1985) The relationships between Streptococcal species and periodontopathic bacteria in human dental plaque. Archives of Oral Biology 30, Ikegami, A., Honma, K., Sharma, A. & Kuramitsu, H. K. (2004) Multiple functions of the leucine-rich repeat protein LrrA of Treponema denticola. Infection and Immunity 72, Kang, M. S., Na, H. S. & Oh, J. S. (2005) Coaggregation ability of Weissella cibaria isolates with Fusobacterium nucleatum and their adhesiveness to epithelial cells. FEMS Microbiology Letters 253, Kolenbrander, P. E., Andersen, R. N., Blehert, D. S., Egland, P. G., Foster, J. S. & Palmer, R. J., Jr (2002) Communication among oral bacteria. Microbiology and Molecular Biology Reviews 66, Kolenbrander, P. E., Andersen, R. N. & Moore, L. V. (1989) Coaggregation of Fusobacterium nucleatum, Selenomonas flueggei, Selenomonas infelix, Selenomonas noxia, and Selenomonas sputigena with strains from 11 genera of oral bacteria. Infection and Immunity 57, Kolenbrander, P. E. & London, J. (1993) Adhere today, here tomorrow: oral bacterial adherence. Journal of Bacteriology 175, Kolenbrander, P. E., Parrish, K. D., Andersen, R. N. & Greenberg, E. P. (1995) Intergeneric coaggregation of oral Treponema spp. with Fusobacterium spp. and intrageneric coaggregation among Fusobacterium spp. Infection and Immunity 63, Kremer, B. H. & Van Steenbergen, T. J. (2000) Peptostreptococcus micros coaggregates with Fusobacterium nucleatum and non-encapsulated Porphyromonas gingivalis. FEMS Microbiology Letters 182, Kreth, J., Zhang, Y. & Herzberg, M. C. (2008) Streptococcal antagonism in oral biofilms: Streptococcus sanguinis and Streptococcus gordonii interference with Streptococcus mutans. Journal of Bacteriology 190, MacFarlane, T. W., Jenkins, W. M., Gilmour, W. H., McCourtie, J. & McKenzie, D. (1988) Longitudinal study of untreated periodontitis (II). Microbiological findings. Journal of Clinical Periodontology 15, Machtei, E. E., Dunford, R., Hausmann, E., Grossi, S. G., Powell, J., Cummins, D., Zambon, J. J. & Genco, R. J. (1997) Longitudinal study of prognostic factors in established periodontitis patients. Journal of Clinical of Periodontology 24, Marsh, P.D. (2003) Are dental diseases examples of ecological catastrophes? Microbiology 149, Nesbitt, W. E., Fukushima, H., Leung, K. P. & Clark, W. B. (1993) Coaggregation of Prevotella intermedia with oral Actinomyces species. Infection and Immunity 61, Okuda, T., Kokubu, E., Kawana, T., Saito, A., Okuda, K. & Ishihara, K. (2012a) Synergy in biofilm formation between Fusobacterium nucleatum and Prevotella species. Anaerobe 18, Okuda, T., Okuda, K., Kokubu, E., Kawana, T., Saito, A. & Ishihara, K. (2012b) Synergistic effect on biofilm formation between Fusobacterium nucleatum and Capnocytophaga ochracea. Anaerobe 18, Papapanou, P. N., Baelum, V., Luan, W. M., Madianos, P. N., Chen, X., Fejerskov, O. & Dahlen, G. (1997) Subgingival microbiota in adult Chinese: prevalence and relation to periodontal disease progression. Journal of Periodontology 68, Paster, B. J., Boches, S. K., Galvin, J. L., Ericson, R. E., Lau, C. N., Levanos, V. A., Sahasrabudhe, A. & Dewhirst, F. E. (2001) Bacterial diversity in human subgingival plaque. Journal of Bacteriology 183, Periasamy, S. & Kolenbrander, P. E. (2009) Mutualistic biofilm communities develop with Porphyromonas gingivalis and initial, early, and late colonizers of enamel. Journal of Bacteriology 191, Roberts, F. A. & Darveau, R. P. (2002) Beneficial bacteria of the periodontium. Periodontology , Santagati, M., Scillato, M., Patane, F., Aiello, C. & Stefani, S. (2012) Bacteriocin-producing oral Streptococci and inhibition of respiratory pathogens. FEMS Immunology & Medical Microbiology 65, Sharma, A., Inagaki, S., Sigurdson, W. & Kuramitsu, H. K. (2005) Synergy between Tannerella forsythia and Fusobacterium nucleatum in biofilm formation. Oral Microbiology and Immunology 20, Slots, J. (1977) Microflora in the healthy gingival sulcus in man. Scandinavian Journal of Dental Research 85, Socransky, S. S., Haffajee, A. D., Cugini, M. A., Smith, C. & Kent, R. L., Jr (1998) Microbial complexes in subgingival plaque. Journal of Clinical Periodontology 25, Tanner, A., Maiden, M. F., Macuch, P. J., Murray, L. L. & Kent, R. L., Jr (1998) Microbiota of health, gingivitis, and initial periodontitis. Journal of Clinical Periodontology 25, Vienna, M. E., Horz, B. G. & Conrads, G. (2005) Microarrays complement culture methods for identification of bacteria in endodontic infections. Oral Microbiology and Immunology 20, 1 6. Ximenez-Fyvie, L. A., Haffajee, A. D. & Socransky, S. S. (2000) Comparison of the microbiota of supra- and subgingival plaque in health and periodontitis. Journal of Clinical Periodontology 27, Supporting Information Additional Supporting Information may be found in the online version of this article: Figure S1. Inter-bacterial relationships: Effect of categorical increase in A. actinomycetemcomitans, E. corrodens, Capnocytphaga species and F. nucleatum on other bacteria. Bars indicate the correlation degree: weak correlation, moderate correlation. Error bars represent the standard error of the mean (n = 6308). Figure S2. Inter-bacterial relationships: Effect of categorical increase in P. nigrescens, V. parvula, P. ghigivalis and T. forsythensis on other bacteria. Bars indicate the correlation degree: weak correlation, moderate correlation, strong correlation. Error bars represent the standard error of the mean (n = 6308). Figure S3. Inter-bacterial relationships: Effect of categorical increase in T. denticola, P. intermedia, C. gracilis and C. rectus on other bacteria. Bars indicate the correlation degree: weak correlation, moderate correlation, strong correlation. Error bars represent the standard error of the mean (n = 6308). Figure S4. Inter-bacterial relationships: Effect of categorical increase in C. concisus, P. micros, E. nodatum and A. oris on other bacteria. Bars indicate the correlation degree: weak correlation, moderate correlation, strong correlation. Error bars represent the standard error of the mean (n = 6308). Figure S5. Inter-bacterial relationships: Effect of categorical increase in S. mitis, S. gordonii and S. constellatus group and A. odontolyticus on other bacteria. Bars indicate the correlation degree: weak correlation, moderate correlation, strong correlation. Error bars represent the standard error of the mean (n = 6308). Table S1. Inter-bacterial correlation table based on spearman correlations: correlations between 20 oral bacteria are displayed. Address: Wim Teughels Department of Oral Health Sciences, Oral Microbial Ecology Research Cluster, Periodontology Section, KU Leuven, Kapucijnenvoer 7, 3000 Leuven, Belgium wim.teughels@med.kuleuven.be

10 10 Loozen et al. Clinical Relevance Scientific rationale for the study: The aim of this study was to analyse the microbial changes within the subgingival microbial ecology in relation to probing pocket depth and in relation to individual species numbers. Principal findings: This large-scale retrospective study was able to correlate several bacteria with disease and illustrated several bacterial interactions. Practical implications: Practical implications remain difficult at the moment. However, current knowledge on bacterial correlations and relations already paves the way for possible new treatment options not focusing on removal of the whole oral microbiota, but focusing on restoring the shifted balance.