Examination of the test set of hydrated biofilm samples showed that overall, 93% of voxels were properly identified using the transferred model. Our previous work saw 96% correct identification of species (Beier et al.
). The move to the new sample type involved several changes relative to the original training set, including a replacement of the laser with a corresponding 2 nm wavelength difference in excitation, the use of a water-immersion objective rather than air-immersion, the addition of depth scanning, and a decrease in signal level associated with decreased concentrations within hydrated biofilms. The high performance on this single species confirmation set given these system changes indicates the robustness of the prediction model.
With the ability to properly identify the species within hydrated samples thus confirmed, biofilms from a mixed culture were analyzed. As mentioned above, the spatial arrangement of clusters of like-classification was a reasonable result, though it was not directly verifiable. What can be asserted with confidence, however, is that in mixed-species biofilms the “minority” species was typically assigned frequently enough to exceed the baseline error rate of 7% seen in single-species studies. The fact that four of the five mixed-species biofilms had Raman-assigned minority-species levels that significantly exceeded 7% (Figure
) indicates that the presence of both species has been detected and quantified in the mixed biofilm samples. The fifth biofilm, which did not see a statistically significant level of the minority species, may simply have been scanned in a region occupied by a single species.
There is another interesting element to the depth scan in Figure
. Although S. mutans was introduced days after S. sanguinis, the locations classified as S. mutans tended to be closer to the substrate, as if the S. mutans had migrated beneath the layer of S. sanguinis. This observation was consistent across all four of the biofilms in which significant levels of both species were observed. For the particular species in this study, this could potentially have significance as it relates to the cariogenic properties of dental plaque. Further experiments are needed to explore this initial observation.
In summary, Raman spectroscopy has been implemented through a confocal microscope and used to successfully classify oral bacteria in hydrated biofilms of one or two species. A model constructed with spectra from dehydrated single-species biofilms has been transferred for the species prediction of hydrated samples. In a test set of single-species hydrated biofilms, voxel-by-voxel species assignments were 93% accurate.
When two-species biofilms from mixed culture were examined, the presences of both S. sanguinis and S. mutans were detected in four out of five biofilm regions. We cannot rule out the possibility that the remaining biofilm was simply scanned over a single-species region. The spatial arrangement of species observed in these mixed biofilms has potential implications for the study of dental plaque cariogenicity, though this aspect of our study would require further investigation. To the best of our knowledge, this is the first time two bacterial species of the same genus and sub-genus group have been mapped in a biofilm using Raman spectroscopy.
While the specific experiments described above have shown the discrimination between S. sanguinis and S. mutans, the method presented here could be applied to the study of other microbes. Although the formation of Raman-based species maps is not fast, the technique offers the ability to use intrinsic chemical differences between cells to create multidimensional maps of microbial structures without extensive knowledge of the cells’ genomes and without requiring any invasive sample preparation that could potentially alter the sample under study.