Metabolic fingerprints of Mycobacterium bovis cluster with molecular type: implications for genotype-phenotype links

Mycobacterium bovis is the causative agent of tuberculosis in a range of animal species and man, with worldwide annual losses to agriculture estimated at US$3 billion. Many animals, including badgers, foxes, ferrets and deer, are believed to act as vectors in the transmission to cattle (Aranaz et al., 1996; Clifton-Hadley et al., 1995; Delahay et al., 2001; Griffin et al., 2005), with the badger (Meles meles) implicated as providing the most significant reservoir of infection in the UK. In the face of complex interactions between wildlife reservoirs and domesticated animals, it is essential that robust epidemiological techniques are applied to determine sources of infection and explore the extant M. bovis population. An M. bovis typing technique that is highly discriminatory, reproducible and high-throughput is therefore a major research goal.

The ideal microbial typing technique would be easy to use, automated, with intuitive nomenclature and have good inter-laboratory reproducibility (Roring et al., 2004). Restriction fragment length polymorphisms (RFLP) (Goyal et al., 1997) of the IS6110 element are generally used in discriminating members of the M. tuberculosis complex; however, this technique has limited use in M. bovis due to the low copy number of IS6110. Therefore, alternative genetic typing methods have been used in the analysis of M. bovis, with spacer oligonucleotide typing, spoligotyping (Kamerbeek et al., 1997), the current method of choice. Spoligotyping focuses on identifying repeat variation in the direct repeat (DR) locus, which is composed of multiple, virtually identical 36 bp repeats interspersed with unique DNA spacer sequences (direct variant repeats; DVRs) of similar size. Different spoligotype patterns are identified by the presence or absence of DVRs. In the UK ten major spoligotypes account for ∼90 % of all isolates, with the two dominant spoligotypes 9 and 17 representing 65 % of all isolates (Inwald et al., 2002). The dominance of two main types and the tendency of the spoligotype pattern to alter slowly with time (Smith et al., 2003) reduce the usefulness of spoligotyping as an epidemiological tool.

Alternative genotypic typing techniques identifying sets of tandem repeats, such as exact tandem repeats (ETRs), mycobacterial interspersed repetitive units (MIRUs) and variable number tandem repeats (VNTRs) have been described as independent tools for the discrimination of members of the M. tuberculosis complex (Roring et al., 2004). VNTRs have been successfully used in the discrimination of M. bovis isolates, and the combination of spoligotyping with VNTR analysis vastly improves the levels of discrimination (Gibson et al., 2004). Indeed, our previous work has shown that VNTR analysis can successfully subdivide spoligotype 9 into clusters of related strains; however, the VNTR technique requires optimization and the level of discrimination depends upon the loci applied and test panel chosen for the study. Different degrees of discrimination may be appropriate for different studies, with some loci useful for population-based studies whereas more discriminating loci may be required to monitor outbreak analysis (Roring et al., 2004).

In this study we investigated the ability of the vibrational spectroscopic technique Fourier-transform infrared spectroscopy (FT-IR) to produce robust biochemical fingerprints' of M. bovis isolates. Numerous advantages of using this method of discrimination over traditional microbiological and molecular techniques are documented; these include speed, minimal sample preparation, automation and relatively low expense. The use of multivariate statistical techniques to analyse the infrared spectra has proved successful in the discrimination of the bacteria to subspecies levels (Maquelin et al., 2003; Naumann et al., 1991; Timmins et al., 1998; Winder et al., 2004). To date, the discrimination of members of the Mycobacterium genus by FT-IR has not been reported in the literature. Moreover, we show that the FT-IR method successfully clustered strains of the same molecular type, suggesting that molecular types share distinct phenotypic characteristics that were discriminated by FT-IR. The implications of this result are discussed.

Cultivation of M. bovis.
Representative isolates of the ten major spoligotypes in the UK were selected. Details of the origins of the isolates are given in Table 1. Spectra were excluded from the study when the signal-to-noise ratio was below 80; this was calculated using the wavenumber range 19001800 cm¹ for noise and the 17001600 cm¹ region for signal. The strains were cultured in triplicate for 4 weeks at 37 °C on Middlebrook 7H10 medium [supplemented with 10 % oleic acid, albumin, glucose and catalase (OADC); Becton Dickinson USA] and with antifungal and selective agents fungizone (5 mg ml¹), polymixin B (100 000 IU ml¹), amoxil (200 mg ml¹) added. The biomass from the slopes was harvested in 100 µl sterile Millipure water. The cells were pasteurized at 80 °C for 1 h and the viability of the cells was checked to ensure no growth occurred.

Table 1. Details of M. bovis spoligotypes studied

FT-IR spectroscopy of mycobacteria.
Two methods of FT-IR were investigated, using two different sample presentation techniques transmission and reflection on two different spectrometers.

Reflectance-based FT-IR.
A 10x10 cm aluminium plate was rinsed with acetone and dried at 50 °C for 10 min prior to use. FT-IR analysis was initially performed on the plate without samples to provide a reference reading for each well. The plate was loaded onto the motorized stage of a reflectance TLC accessory (Winder et al., 2004) attached to a Bruker IFS28 FT-IR spectrometer (Bruker Spectrospin). This was equipped with a mercury-cadmium-telluride (MCT) detector, which was cooled with liquid N₂. A 5 µl aliquot of each sample was evenly applied in triplicate onto the 400-well plate (the so-called machine replicates). The plate was then dried at 50 °C for 30 min before FT-IR analysis was performed. Spectra were collected over the wavelength range of 4000600 cm¹ under the control of an IBM-compatible computer programmed with Opus 2.1 running under IBM OS/2 Warp, which was provided by the manufacturers. Spectra were acquired with a resolution of 4 cm¹ and 256 spectra were co-added and averaged. The collection time for the spectra was approximately 10 s per sample. The spectra are displayed in terms of absorbance, which was calculated from the reflectance-absorbance spectra using Opus software.

Transmission-based FT-IR.
Prior to use a 96-well zinc selenide plate was rinsed with 2-propanol and deionized water and allowed to dry at room temperature. Aliquots (15 µl) of each bacterial sample were evenly applied in triplicate onto the plate and dried at 50 °C for 30 min. The plate was loaded onto a motorized microplate module HTS-XT attached to an Equinox 55 module (Bruker Optics). The motorized module of this instrument introduces the plate into the airtight optics of the instrument, in which tubes of desiccant are contained to remove moisture (Harrigan et al., 2004). A deuterated triglycine sulphate (DTGS) detector was employed for transmission measurements of the samples to be acquired. Spectra were collected over the wavelength range 4000900 cm¹ (a reduced range is used in comparison to the reflectance measurements due to the zinc selenide focusing element) under the control of a computer programmed with Opus 4, operated under MS Windows 2000. Spectra were acquired at a resolution of 4 cm¹ and 64 spectra were co-added and averaged to improve the signal-to-noise ratio. The collection time for each spectrum was approximately 1 min and the spectra were displayed in terms of absorbance.

Preprocessing.
The ASCII data were imported into Matlab version 6 (The MathWorks). To minimize problems arising from baseline shifts, Matlab was used to correct for CO₂ vibrations [the CO₂ peaks at 24032272 cm¹ and 683656 cm¹ (when present) were removed and filled with a trend] and windows of the spectra likely containing H₂O vibrations were smoothed with a window of 55 cm¹ to reduce noise (40003615 cm¹ and 19981763 cm¹). The spectra were normalized such that the smallest recorded absorbance was set to 0 and the highest was set to 1 for each spectrum and then the second derivatives (Savitzky & Golay, 1964), with a window of 9), were used for cluster analysis.

Cluster analyses.
The unsupervised clustering method principal components analysis (PCA; Jolliffe, 1986) was performed on the spectra to reduce the dimensionality of the multivariate data whilst preserving the variance, prior to discriminant function analysis (DFA). DFA is a supervised technique that discriminates between groups on the basis of the retained principal components with a priori knowledge of which spectra were replicates. As this process uses information on the machine replicates for each isolate it does not bias the results in any way (MacFie et al., 1978; Windig et al., 1983). Hence the cluster analysis would not bias the clustering of the isolates according to their respective spoligotypes because the a priori knowledge is based on the isolate replicates. DFA was programmed to minimize within group variance and maximize between group variance. Finally, hierarchical cluster analysis (HCA) was used to construct a dendrogram from the a priori group centres in the PC-DFA space, using scaled Euclidean distances as described by Goodacre et al. (1998), and dendrograms were produced using the mean linkage clustering algorithm (Manly, 1994).

Typical transmission-generated FT-IR absorbance spectra of M. bovis illustrating examples of spoligotypes 9 and 17 are shown in Fig. 1; the baseline of spoligotype 9 is slightly offset to demonstrate the contours of each sample. Due to the complexity and qualitative similarity of these spectra, and indeed all data collected, multivariate statistical analysis was used to discriminate the strains of M. bovis.

(14K): Fig. 1. Typical example of FT-IR spectra of M. bovis (spoligotypes 9 and 17; the spectrum representing spoligotype 9 is offset).

Due to the different methods of FT-IR used, the data could not be analysed together. This is expected since the transmission-based sampling produces spectra from the absorbance of infrared light during single transmission (i.e. the sample is held on an IR-transparent window and light passes directly through it) and detected using a DTGS detector. By contrast, the reflection-based protocol involves the IR beam bouncing through the bacteria held on an aluminium plate and being reflected off the bottom of the wells back through the sample. This results in absorbance via reflection (two beam passes). In addition, because the sample is not infinitely thin, some diffuse-reflectance also occurs, which is recorded by a MCT detector together with the reflected light. Note that the DTGS and MCT detectors are different: the DTGS operates at room temperature whilst the MCT is liquid N₂ cooled; the MCT has a quicker response time so that spectra are collected in 10 s compared to 60 s for DTGS; finally, the DTGS tends to have a more stable baseline than MCT. While these data could not be analysed together, it is important to assess which produces more reproducible data and which is better at bacterial discrimination.

PC-DFA and HCA were used to produce ordination (scatter) plots and dendrograms respectively. Whilst both the ordination plots and dendrograms were used to investigate the discriminatory ability of FT-IR, for brevity and clarity the dendrograms are predominantly shown for reflection-based FT-IR (Fig. 2) and transmission-based measurements (see Fig. 4).

(30K): Fig. 2. Composite dendrogram illustrating the hierarchial cluster analysis of the data collected by diffuse reflectance FT-IR.

(25K): Fig. 4. Composite dendrogram illustrating the hierarchial cluster analysis of the data collected by transmission FT-IR.

It is clear from the two analyses that the discrimination of the various spoligotypes is different for reflection (Fig. 2) versus transmission (Fig. 4) measurements. These two figures reflect a series of analyses in which certain bacteria are sequentially removed because they cluster according to spoligotype and are clearly differentiated from the others. The reanalysis of the remaining spectra allows finer (more subtle) discrimination to be observed, and is commonplace when analysing spectroscopic data (Jarvis & Goodacre, 2004; Lopez-Diez & Goodacre, 2004). This process is briefly discussed below.

Reflectance-based FT-IR
The resulting dendrogram for the reflectance FT-IR illustrates the clear discrimination of spoligotypes 20, 25 and 35 (Fig. 2). The PC-DFA ordination plot of the first analysis of the composite dendrogram is shown in Fig. 3(a), to illustrate the clustering of spoligotypes 20, 25 and 35. Fig. 3(b) indicates the mean centres of the a priori groups and the circles indicate a 95 % χ² confidence region around the point. The overlapping circles are shown for spoligotypes 20, 25 and 35; the circles representing the other spoligotypes are not shown for clarity of the figure. The analysis shows the isolates representing the chosen spoligotypes cluster within the confidence regions. Removal of the three discriminated spoligotypes and the subsequent reanalysis of the data demonstrated the dispersal of spoligotype 9 throughout the HCA space (data not shown). The successive removal of spoligotype 9 and reanalysis of the data illustrates the recovery of spoligotypes 10 and 22. A fourth analysis was then performed on the remaining data following the removal of spoligotypes 10 and 22. In this final dendrogram, the representative samples of spoligotype 17 were recovered together; however, the isolates of spoligotypes 12 were recovered with M. bovis spoligotypes 11 and 13 isolates.

(26K): Fig. 3. (a) PC-DFA performed on the data illustrated in the first analysis in Fig. 2. PCs 1100 (total cumulative variance is 99.94 %) were used by the DFA algorithm with a priori knowledge of isolate biological replicates and DFs 16 used for HCA. (b) The points indicate the mean centre of each a priori group used in the DFA. The circles show the boundary of 95 % confidence region for that particular point. Each point would have the same area representing the 95 % χ2 confidence region; however, only the circles representing spoligotypes 35 (black), 25 (red) and 20 (blue) are shown, for clarity. The spoligotypes are represented as follows: square, 9; diamond, 10; down triangle, 11; up triangle, 12; left triangle, 13; right triangle, 17; pentagram, 20; hexagram, 22; circle, 25; x-mark, 35.

Transmission-based FT-IR
The initial analysis of all the data collected by transmission FT-IR clearly shows the discrimination of spoligotypes 10, 25 and 35 (Fig. 4). Following their removal and subsequent reanalysis of the data, it was again evident that the isolates representing spoligotype 9 were recovered throughout the HCA space (data not shown). Spoligotype 9 was therefore removed from the dataset and the resulting dendrogram (third analysis, Fig. 4) illustrated the clear discrimination of spoligotypes 17, 20 and 22. The samples representing spoligotypes 11, 12 and 13 were slightly intermixed, while isolates from 12 and 13 were recovered together and a second branch of spoligotype 13 was observed. Isolates of spoligotype 11 were recovered in two separate branches within the dendrogram. Further analysis of only spoligotypes 12 and 13 illustrates clustering of the isolates from spoligotype 12. Although the representatives of spoligotype 13 were grouped together on the dendrogram, the branching pattern illustrates diffuse clustering in the HCA space (fourth analysis, Fig. 4).

Comparison of the two FT-IR methods
In the initial analysis of both the reflectance and transmission data, three distinct branches were observed according to three spoligotypes. Spoligotypes 25 and 35 were recovered in both analyses but the third branch varied according to the different FT-IR techniques. The relationship between the spoligotypes analysed in this study illustrates spoligotypes 25 and 35 to be closer to the ancestor, in which all the spacers are present (Fig. 5). Spoligotype 10 was resolved for the transmission data whereas spoligotype 20 was initially resolved in the analysis of the reflectance data. The isolates representing spoligotype 9 were intermixed with the samples representing the other spoligotypes in the analysis for both the reflectance and transmission data. The evolutionary history indicates that a number of spoligotypes (10, 11, 12, 13, 17 and 22) evolved after spoligotype 9 (Fig. 5). The remaining spoligotypes (with the exceptions of 12 and 13) were resolved in the third analysis for the transmission data. A fourth analysis was necessary to recover the majority of the reflectance-collected samples according to spoligotype. In both analyses, the isolates representing spoligotypes 11, 12 and 13 were poorly resolved in the analysis. The representative samples of spoligotypes 12 and 13 (transmission data) were analysed separately; the isolates of spoligotype 12 were recovered in a cluster, but those from spoligotype 13 appeared diffuse in the analysis.

(3K): Fig. 5. Shows the relationships between the spoligotype patterns identified by a simple algorithm that assumes spacers are lost and never regained and that change involves the minimum number of deletion events at a time (VLA, unpublished data). A deletion event is any loss of spacers (one or more contiguous spacers). Ancestor refers to a spoligotype pattern with all the spacers present. The relationship of spoligotype pattern 20 is ambiguous and is therefore not shown. This is not a phylogenetic tree of the strains with these spoligotype patterns; however, for this set of strains further studies of informative single nucleotide mutations have shown that the relationships shown here do represent the evolutionary history of these strains.

Based on these observations and the complexity of the sequential cluster analysis, we believe that the transmission-generated data are more appropriate for this analysis. Indeed, the lack of appreciable baseline shifts and the simpler optical interrogation of the sample by transmission rather than a reflective measurement would indicate greater spectral reproducibility and support this conclusion.

Analysis of UK dominant M. bovis isolates
As reported above, spoligotypes 9 and 17 are the two dominant spoligotypes in the UK, accounting for 65 % of all isolates from cattle (Inwald et al., 2002). We therefore focused the final analysis on just type 9 and 17 isolates. The resulting PC-DFA scores plot is shown in Fig. 6, illustrating the differentiation of five clusters; this was confirmed by the 95 % χ² confidence region. It can be clearly seen that a tight cluster of the isolates representing spoligotype 17 was observed, which is perhaps not surprising as these are a single VNTR group (7555*33.1), whilst four of the five isolates representing spoligotype 9 in the analysis were dispersed throughout the PC-DFA space [note that each of the spoligotype 9 isolates was of a different VNTR group (Table 1)]. This clearly reflects what was observed in the dendrograms (Figs 2 and 4) discussed above. A similar pattern was observed in the analysis of the reflectance data (data not shown). The remaining representative of spoligotype 9 (9d, VNTR group 7554*33.1) clustered with the isolates of spoligotype 17 (VNTR group 7555*33.1). This may possibly result from the close phylogenetic relationship of these isolates according to the VNTR phylogeny as seen by us (to be reported elsewhere).

Table 1.

The phylogeny of UK spoligotype 9 strains is complex, with the identification of 22 VNTR types for strains with this spoligotype pattern (Smith et al., 2003). As can be seen in Table 1, the VNTR diversity in the spoligotype 9 strains is greater than for any other grouping. Indeed, the majority of distinct VNTR types of spoligotype 9 are representative of one geographical location, indicating recent clonal expansion of the strains in these locations (Smith et al., 2003). The wide distribution of spoligotype 9 through the PC-DFA space indicates phenotypic differences in these isolates, which reflects genotypic differences observed by VNTR analysis. In comparison, the representatives of spoligotypes 17 are tightly clustered in the PC-DFA space (Fig. 6), reflecting their limited VNTR variation (Table 1). The heterogeneity of spoligotype 9 is highlighted by the inclusion of nine other spoligotypes in this analysis (Figs 2 and 4).

Hence, the analysis of the strains using FT-IR allowed us to generate clusters that showed excellent agreement with clustering performed using genetic markers; this suggests that strains of the same molecular type have conserved phenotypic traits. This is intriguing when considered in the light of our knowledge of the population structure of M. bovis in Great Britain, which found that the frequency of recovery of spoligotypes 9 and 17 isolated in the UK cannot be explained by random mutation and drift (Smith et al., 2003). Instead these strains have undergone a clonal expansion, rising to a high frequency in the population either through a favourable mutation or by colonization of a new geographical location (Smith et al., 2003). The results reported here suggest that the success of M. bovis clonal groupings is due to favourable mutations in these strains that may affect virulence, transmissibility, or survival outside the host. Our FT-IR data therefore provide us with a unique opportunity to examine the relationship between clonal groups and phenotype.

Conclusions
The data collected by transmission FT-IR illustrated better differentiation of the M. bovis isolates according to spoligotypes than that collected by reflectance FT-IR. The phenotypic analysis in this study indicates homogeneity within spoligotypes 10, 17, 20, 22, 25 and 35, while spoligotypes 11, 12 and 13 appeared diffuse and difficult to cluster in the analysis. Tight clustering of molecular types detected by metabolic fingerprinting suggests that these clonal groupings share distinct phenotypic traits that may have a significant role in their success as pathogens. Isolates of spoligotype 9 appear very heterogeneous relative to other spoligotypes, an observation that was supported by VNTR analysis and reflects the wide variation within this group. Therefore this study describes the first use of FT-IR for the discrimination of M. bovis strains. As well as being a robust and high-throughput method that produces strain clusters reflecting those generated by genetic methods, FT-IR has the added benefit of providing the first insights into phenotypegenotype links in M. bovis clones. This study therefore sets the baseline for a more in-depth phenotypic analysis of M. bovis molecular types.

We acknowledge the Department of Environment Food and Rural Affairs (Defra) for funding this work. We are grateful to Noel Smith for kindly providing Fig. 5 and Roger Jarvis for his assistance and useful comments with the data analysis.