Association between red biofluorescence and microbial dysbiosis in cariogenic biofilms: an in vitro study

Abstract

Dental caries is driven by microbial dysbiosis rather than single pathogens. Red biofluorescence (RF), emitted from porphyrin-producing bacteria, has been proposed as a non-invasive indicator of dysbiosis. This study aimed to evaluate whether RF intensity reflects caries-related dysbiosis during in vitro biofilm maturation. Saliva-derived microcosm biofilms were grown on hydroxyapatite discs under cariogenic conditions for seven days. RF intensity was measured using Quantitative Light-induced Fluorescence-Digital (QLF-D) and fluorescent hyperspectral imaging (FHSI). Microbial composition was analyzed by 16S rRNA sequencing. A caries-related dysbiosis index was defined from eight cariogenic species consistently reported in clinical lesions. Correlations between RF intensity and dysbiosis index were assessed, and partial least squares regression (PLSR) identified taxa contributing to RF. RF intensity increased markedly with biofilm maturation (13.6-fold at day 7, p < 0.05). The dysbiosis index also rose 5.8-fold, showing a strong correlation with RF intensity (ρ = 1.00, p < 0.01). Anaerobic species such as Veillonella dispar and Granulicatella adiacens group contributed most strongly to RF expression (VIP > 1.0). RF intensity reflects both microbial compositional shifts and dysbiosis in cariogenic biofilms. RF may serve as a comprehensive, non-invasive biomarker of pathogenic biofilm activity and caries risk. This study demonstrated that RF can reflect changes in the composition of caries-related microorganisms and the level of caries-related dysbiosis during in vitro biofilm maturation.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-50796-5.

Introduction

The human oral cavity harbors more than 700 bacterial species, which can adhere to tooth surfaces and form mature biofilms if not regularly removed¹. Pathogenic bacteria within mature biofilms metabolize dietary sugars to produce acids, which demineralize the enamel and serve as a primary cause of dental caries². For this reason, biofilms play a central role in the initiation and progression of dental caries. In the past, selective culture media were primarily used to identify bacteria associated with caries lesions, leading to the notion that specific bacterial species were the sole causative agents of caries. However, with the advent of advanced molecular techniques such as 16S rRNA sequencing, it has become possible to analyze the entire oral microbiota, drawing attention to the fact that interactions among microorganisms and the overall community structure have a greater impact on caries development than individual species. Indeed, biofilms function not as a simple aggregation of bacteria but as an organized ecosystem in which microorganisms interact synergistically and share metabolic activities³.

Recent studies have highlighted the perspective that dental caries is driven not merely by the presence of specific pathogenic bacteria, but by dysbiosis—an ecological shift in which certain bacterial species gain selective advantage due to environmental changes favoring acid production⁴. Such imbalances have been reported to be associated not only with oral but also with a range of systemic diseases⁵. Oral dysbiosis involves a relative increase in pathogenic bacteria alongside the disruption of the resident microbiota, and these ecological changes play a critical role in the pathophysiology of disease⁶. However, there remain limitations in clearly defining what constitutes dysbiosis, as well as in determining whether a given dysbiotic state is specific to dental caries or associated with periodontal disease. In clinical settings, disease states often create unique microenvironments that influence the composition and metabolic activity of the microbial community⁷. Therefore, understanding caries-related dysbiosis requires focusing on the overall composition and functional changes of the microbial community, rather than on the presence of individual bacterial species⁸. Against this background, there is growing interest in evaluating the pathogenic potential of biofilms through biofluorescence-based approaches^9–14.

During the formation and maturation of biofilms, certain pathogenic bacteria produce porphyrins as metabolic byproducts, which emit red biofluorescence (RF) when exposed to blue light at a wavelength of 405 nm^15–17. The biofluorescence can be used as a biological indicator reflecting the metabolic activity of bacteria, and can be noninvasively visualized and quantified using Quantitative Light-induced Fluorescence technology (QLF). In particular, RF intensity has been reported to be closely associated with the degree of biofilm maturation and pathogenicity^15,18,19. However, previous studies have predominantly focused on the mere presence or absence of biofluorescence or on quantitative changes in specific bacterial species, limiting the integrated interpretation of fluorescence responses in relation to the entire microbial ecosystem. More recent evidence suggests that RF is more closely associated with metabolic interactions among microorganisms or with ecological changes—namely, dysbiosis—rather than with single bacterial species²⁰. Accordingly, there is a need to interpret biofluorescence not merely as a standalone marker, but in relation to changes in the microbial community structure.

To clarify this relationship more precisely, fluorescent hyperspectral imaging (FHSI) has emerged as a valuable tool. FHSI captures biofluorescence responses across a broad range of wavelengths in the form of spectra, enabling the utilization of fluorescence patterns at different wavelengths as data, rather than simply detecting the presence or absence of biofluorescence²¹. In particular, analyzing biofluorescence spectra in relation to changes in the microbial community requires the application of multivariate statistical techniques that can integrate fluorescence data with microbiome composition information²². Spectrum-based analysis can sensitively reflect gradual ecological shifts, such as caries-related dysbiosis, and RF in this context holds potential as an ecological biomarker.

This study was undertaken to provide comprehensively analyze the relationship between changes in RF intensity and the level of microbial community dysbiosis during the maturation of an in vitro cariogenic microcosm biofilm. Accordingly, we sought to evaluate the potential of RF as an indicator of caries-related dysbiosis and to explore its applicability as a non-invasive approach for assessing biofilm pathogenicity.

Results

Changes in fluorescence properties and pH during biofilm maturation

The mean RF intensity of the biofilms increased progressively with maturation time (Fig. 1). Compared to the baseline value (10.59), the intensity showed a steep upward trend, increasing by approximately 2.6-fold on day 3 (27.51), 7.2-fold on day 5 (76.50), and 13.6-fold on day 7 (144.29). RF intensity was significantly higher from day 3 onward compared with day 1 (p < 0.05). Spectral analysis revealed generally higher fluorescence intensity in 7-day biofilms than in 3-day biofilms (Fig. 2A). While the 3-day biofilms exhibited a spectral peak in the 430–600 nm range (blue–green wavelengths), the 7-day biofilms showed a primary peak in the 620–720 nm range (orange–red wavelengths), indicating a distinct shift in fluorescence characteristics (Fig. 2B).

Fig. 1.

Changes in RF intensity and pH values of microcosm biofilm according to the maturation period, with representative biofluorescence images. *p < 0.05, One-way ANOVA post hoc Games-Howell. RF, Red biofluorescence.

Fig. 2.

Biofluorescence spectra plots of 3-day and 7-day biofilms. (A) The raw intensity plot normalized to the emission at 405 nm, showing biofluorescence intensity changes. (B) The color spectrum plot normalized to each spectrum’s maximum peak, showing biofluorescence color changes during biofilm maturation.

The pH value measured 4 h after inoculation was 8.26, which decreased to 4.76 on day 1 of biofilm maturation. From day 2 to day 4, the pH remained between 5.04 and 5.46, below the critical enamel demineralization threshold of 5.5, confirming the formation of a cariogenic biofilm (Fig. 1). According to the number of colony-forming units (CFUs) for 7-day biofilms (Mean ± SD), the total bacterial count was 7.24 ± 0.27 log₁₀ CFU/mL, and the aciduric bacteria count was 6.57 ± 0.55 log₁₀ CFU/mL.

Changes in microbial composition of the biofilms

For alpha diversity (Mean ± SD), the Shannon index was 1.69 ± 0.20 in 3-day biofilms and 1.93 ± 0.06 in 7-day biofilms. For the Simpson index, values were 0.29 ± 0.08 in 3-day biofilms and 0.22 ± 0.01 in 7-day biofilms, indicating a gradual decrease in microbial diversity with increased maturation time. However, no statistically significant differences were observed between samples. According to the number of operational taxonomic unit (OTU, mean OTUs), richness was 165 in 3-day biofilms and 140 in 7-day biofilms, but this difference was not statistically significant.

In terms of microbial composition, clear differences were observed between the inoculum and the biofilms, whereas the community composition was similar between 3-day and 7-day biofilms. At the phylum level, 3-day biofilms were dominated in order of Firmicutes (99.3%) and Proteobacteria (0.3%), whereas 7-day biofilms were dominated in order of Firmicutes (91.4%), Actinobacteria (8.5%), and Proteobacteria (0.1%). At the genus level, 3-day biofilms were dominated in order of Streptococcus (44.1%), Lactococcus (27.8%), and Veillonella (27.0%), whereas 7-day biofilms were dominated in order of Streptococcus (40.2%), Veillonella (35.2%), and Lactococcus (12.1%) (Fig. 3A).

Fig. 3.

Relative abundance of bacterial taxa in the inoculum and matured biofilm at the genus level (A) and Log₂ fold change (LFC) based on relative abundance between 3-day and 7-day biofilms (B). In (B), red bars indicate taxa with relatively higher abundance in the 7-day biofilms, whereas blue bars indicate taxa with relatively lower abundance, reflecting compositional shifts rather than absolute changes in bacterial quantity.

Log₂ fold change (LFC) analysis based on relative abundance data showed a compositional enrichment of Granulicatella (1.29), Atopobium (0.44), and Veillonella (0.30), and a relative decrease in Lactococcus (−1.42), in 7-day biofilms compared with 3-day biofilms (Fig. 3B).

Relationship between caries-related dysbiosis and RF

The caries-related dysbiosis index (%) increased approximately 5.8-fold, from 1.48% in 3-day biofilms to 8.57% in 7-day biofilms (Table 1). Among the eight selected species, four exhibited an increase in relative abundance, while the remaining species showed little or no change. Notably, Veillonella dispar increased by approximately 5.4-fold and the Granulicatella adiacens group by approximately 4.5-fold, showing the most pronounced relative abundance increases among all dysbiotic taxa. A very strong positive correlation was observed between caries-related dysbiosis and RF intensity (ρ = 1.00, p < 0.01).

Table 1.

Comparison of species-level relative abundance (percentage of total 16S rRNA reads) of taxa and dysbiotic status in the in vitro cariogenic biofilms at day 3 and day 7.

Taxon name	3-day biofilms (%)	7-day biofilms (%)
Veillonella dispar	0.81	4.34
Streptococcus constellatus group	0.00	1.64
Granulicatella adiacens group	0.44	1.97
Veillonella parvula group	0.09	0.58
Streptococcus mutans	0.00	0.01
Prevotella melaninogenica	0.01	0.00
Streptococcus sanguinis group	0.03	0.00
Streptococcus gordonii group	0.10	0.03
Caries-related dysbiosis index a	1.48	8.57

^aThe caries-related dysbiosis index was calculated as the sum of the relative abundance of these bacteria by converting it into percentages (%).

Identification of bacterial taxa associated with RF

Partial least squares regression (PLSR) analysis was performed at both the genus and species levels to identify taxa contributing to RF intensity. According to the PLSR biplot, samples were clearly separated along the coordinates based on the magnitude of RF intensity, and when considering the dysbiosis index values plotted on the same coordinates, an increase in RF values (intensity) was associated with a corresponding increase in the dysbiosis index. At the genus level, Granulicatella, Atopobium, Actinomyces, and Veillonella showed positive associations with RF intensity, whereas Streptococcus and Lactococcus exhibited negative associations and lower contributions (Fig. 4A). In the variable importance in projection score (VIP score) analysis, Granulicatella, Atopobium, Actinomyces, and Veillonella demonstrated the highest contributions to RF intensity (VIP score > 1.0, Fig. 4B). At the species level, Veillonella dispar, Granulicatella adiacens group, Veillonella parvula group, Atopobium parvulum, and Streptococcus anginosus group were positively associated with RF intensity, whereas Streptococcus salivarius group and Lactococcus lactis group were negatively associated and showed low contributions (Fig. 4C). VIP score analysis revealed that Veillonella dispar, Granulicatella adiacens group, Veillonella parvula group, Atopobium parvulum, and Streptococcus anginosus group had the highest contributions to strong RF intensity (VIP score > 1.0, Fig. 4D). Several of the key species that showed high contributions to RF intensity were also included among the eight caries-related species used to calculate the caries-related dysbiosis index in this study.

Fig. 4.

Partial least squares regression (PLSR) and variable importance in projection (VIP) score plots identifying taxa associated with red biofluorescence (RF) intensity in the dental biofilms. (A, C) PLSR biplots showing bacterial genera (A) and species (C) associated with RF intensity. Color scale indicates RF values (intensity) of each sample. Numbers next to the data points represent the caries-related dysbiosis index for each sample. Arrows indicate taxa, with the direction showing the association with increasing RF intensity and the length reflecting the relative contribution. Component 1 and Component 2 represent latent variables derived from PLSR, with the percentage in parentheses indicating the proportion of variance in the predictors explained by each component. (B, D) Corresponding VIP scores from the PLSR models in (A) and (C). Taxa with VIP > 1.0 (red dashed line) were considered influential.

Functional shifts in the microbial community during biofilm maturation

To investigate functional changes in the microbial community according to maturation period, the functions of biofilms matured for 3 and 7 days were analyzed using PICRUSt, and the LFC of each functional pathway was calculated (Fig. 5). With increased maturation time, fermentation pathways such as pyruvate fermentation to acetone (5.75), heterolactic fermentation (5.62), and pyruvate fermentation to propanoate I (5.12) showed substantial increases. In contrast, pathways involved in heme biosynthesis—specifically, the teichoic acid (poly-glycerol) biosynthesis (−5.93), hexitol fermentation to lactate, formate, ethanol and acetate (−5.85), superpathway of heme biosynthesis from uroporphyrinogen-III (−4.78)—decreased.

Fig. 5.

Log₁₀ fold changes (LFC) in functional pathways between 3-day and 7-day biofilms. Red bars indicate pathway increased in the 7-day biofilms compared to 3-day biofilms, while blue bars indicate those decreased.

Discussion

This study was conducted to determine whether the RF expressed in cariogenic microcosm biofilms can reflect the state of caries-related dysbiosis. The results showed that both RF intensity and the caries-related dysbiosis index increased significantly as the biofilms matured, with particularly high contributions from anaerobic species such as Veillonella dispar and the Granulicatella adiacens group. These findings suggest that RF may serve as an indicator that reflects not merely the presence of specific bacteria but increased metabolic activity of pathogenic species and community-level dysbiosis.

In the present study, the RF intensity of cariogenic biofilms formed using an in vitro microcosm biofilm model began to appear on day 2 after inoculation and increased significantly from day 3 of maturation. During this period, the supernatant pH remained below the critical threshold of 5.5, establishing a cariogenic environment. These changes indicate that the microcosm within the biofilm underwent dynamic shifts as maturation progressed, consistent with previous reports showing significant correlations between RF intensity, aciduric bacterial CFUs, and pH in cariogenic biofilms^15,18,19,23. Although the differences were not statistically significant, microbial diversity tended to decrease with increasing maturation time, and changes in the relative abundance of certain taxa were observed. This suggests that the microcosm biofilm shifted toward a dysbiotic state over time alongside an increase in RF intensity, indicating that RF may reflect both structural alterations in complex microbial communities and ecological imbalance.

When RF first became evident on day 3, biofilms were dominated by acidogenic and aciduric bacteria in the order of Streptococcus, Lactococcus, and Veillonella. As the biofilms matured to day 7, they exhibited stronger RF intensity and a shift toward an anaerobic environment, characterized by a decrease in aerobic bacteria such as Lactococcus and an increase in the relative abundance of anaerobic genera including Granulicatella, Atopobium, and Veillonella. Species of the genus Streptococcus and Lactococcus primarily ferment glucose to lactic acid, leading to acid accumulation within the biofilm⁷. This progressive increase in lactic acid likely favored the proliferation of anaerobic bacteria such as Granulicatella, Atopobium, and Veillonella, which can utilize lactate as an energy source and survive in acidic environments. Notably, Veillonella is known to participate in the heme metabolic pathway, resulting in porphyrin accumulation²⁴. Porphyrins emit RF when excited by violet light at 405 nm and are closely associated with bacterial metabolic activity and pathogenicity^15,25–27. Accordingly, the process of biofilm maturation facilitates pronounced shifts in microbial community structure and favors the selective proliferation of porphyrin-producing taxa, ultimately resulting in an elevation of RF intensity. These findings indicate that the manifestation of RF is attributable to the combined effects of shifts in the microbial community and metabolic alterations, including acid production, energy metabolism, and porphyrin accumulation. Importantly, porphyrins are not structural components of specific bacterial strains but rather metabolic byproducts produced by diverse microbial communities. Therefore, RF reflects the overall metabolic activity of the biofilm rather than the presence of specific taxa, indicating that it represents biofilm maturation and collective functional activity rather than individual pathogens. Accordingly, biofilm dysbiosis may be more appropriately interpreted in terms of altered metabolic activity rather than solely compositional changes⁸. These metabolic shifts are inherently linked to changes in community structure, and RF may thus serve as a useful tool for assessing functional transitions between symbiotic and dysbiotic states²⁸.

This study proposes a novel caries-related dysbiosis index based on the relative abundances of key species consistently associated with caries. The index is calculated as the sum of the relative abundances of caries-associated species within the total bacterial community of a biofilm, enabling quantitative assessment of the cariogenic dysbiosis level for each biofilm. In this study, the caries-related dysbiosis index increased approximately 5.8-fold as the biofilms matured from day 3 to day 7, and its strong correlation with RF intensity (ρ = 1.00, p < 0.01) confirmed that RF reflects the caries-related dysbiosis level. Traditionally, caries risk has been assessed primarily based on the presence or counts of individual species such as Streptococcus mutans or Lactobacilli^29–31. In contrast, the index proposed in this study quantitatively reflects functional changes and ecological imbalance at the microbial community level. This approach is analogous to that of a previous study in which the sum of the relative abundances of 11 subgingival plaque-specific bacteria (SUBP bacteria) was used to non-invasively screen for periodontitis through salivary microbiome analysis³². Using relative abundance as an index minimizes the variability among individual taxa while capturing overall changes in the microbial community. Importantly, the present index was constructed based on pathogenic species repeatedly identified in actual carious lesions, thereby ensuring both clinical reliability and biological validity. In this study, however, only four of the eight selected species exhibited an increase during biofilm maturation, while the others showed little or no change. This indicates that the progression of cariogenic dysbiosis may not involve a uniform expansion of all pathogenic species, but rather is driven by the dominant contribution of specific taxa such as Veillonella dispar and the Granulicatella adiacens group. In addition, the present study was based on biofilms derived from a single donor, which may have limited the representation of the full spectrum of cariogenic bacteria. Consequently, certain cariogenic taxa could have been missing from the index, and validation with biofilms established from multiple donors will be essential for further refinement. Moreover, no increase was observed for Streptococcus sanguinis and Streptococcus gordonii group up to 7 days in this model. These species are well recognized as early colonizers and are consistently reported in carious lesions, yet their abundance may remain stable in the early stages of biofilm maturation. Therefore, while their inclusion in the index enhances its biological relevance, further studies with longer incubation periods and diverse donor sources are warranted to determine whether the index should be refined to more accurately reflect cariogenic dysbiosis. Nonetheless, by summing across all species consistently associated with caries, the index captures the cumulative effect of both strongly and weakly responsive taxa, thereby reflecting the overall ecological imbalance at the community level. Recently, another study proposed a subgingival microbial dysbiosis index (SMDI) for periodontitis by applying machine learning to 16S microbiome data³³. However, this index is disease-specific for periodontitis, and to the best of our knowledge, the present study is the first to introduce an index that quantitatively reflects caries-related dysbiosis based on the microbial characteristics of carious lesions. Furthermore, rather than limiting the scope to the proposal of the index, this study applied it to a laboratory biofilm model and demonstrated its significant correlation with RF intensity, thereby providing evidence that RF may serve as a biological marker of cariogenic dysbiosis. This highlights the value of the index as a comprehensive measure of ecological imbalance within microbial communities, extending beyond the simple analysis of individual taxa.

In this study, PLSR analysis with VIP scoring was employed to identify bacterial species contributing to RF expression. Veillonella dispar (VIP score = 1.69) and the Granulicatella adiacens group (VIP score = 1.41) exhibited the highest contributions. Consistent with this result, both species showed marked increases in relative abundance in 7-day biofilms, and in the caries-related dysbiosis index, Veillonella dispar increased approximately 5.4-fold and the Granulicatella adiacens group approximately 4.5-fold, representing the most pronounced increases among caries-related dysbiosis index. The predominant contribution of Veillonella dispar is particularly notable, as this species can survive under acidic conditions and metabolizes lactic acid, thereby modulating biofilm pH toward a mildly acidic range³⁴. The observed increase in Veillonella dispar in this study suggests that aciduric species become selectively enriched as biofilms mature, driving the progression of dysbiosis, in agreement with previously reported microbial compositional shifts in actual carious lesions^35,36. The Granulicatella adiacens group, which showed the next highest contribution, has been frequently associated with dental plaque, endodontic infections, and dental abscesses^37,38. Granulicatella adiacens has been reported to adhere effectively to saliva-coated hydroxyapatite surfaces via ionic bridging between its anionic cell surface components and salivary Ca²⁺ ions³⁹. Following adhesion, it forms multilayered extracellular polymeric substance (EPS) matrices, thereby contributing to the stability of the biofilm structure. These features are likely related to biofilm maturation and may concomitantly increase RF intensity as observed in this study. In addition, the Veillonella parvula group (VIP score = 1.23), also identified as a contributor to RF in the PLSR analysis, has been described as playing a cofactor role in both mitigating acidification in cariogenic environments and promoting biofilm growth particularly in combination with Streptococcus mutans⁴⁰. Moreover, it has been reported as one of the major oral bacteria capable of accumulating metalloporphyrin-derived compounds that induce RF intensity⁴¹. Together, these findings demonstrate that the metabolic characteristics of cariogenic bacterial species and the progression of caries-related dysbiosis are closely associated with RF expression. They also support the notion that the in vitro cariogenic microcosm biofilm model employed in this study successfully reflects the ecological shifts occurring in actual carious environments. Furthermore, they suggest that RF functions not merely as an indicator of bacterial biomass but as a biological signal reflecting the dysbiotic state associated with dental caries.

In contrast, the PLSR analysis indicated that the Streptococcus salivarius group and the Lactococcus lactis group had low contributions to the increase in RF intensity. The Streptococcus salivarius group is a representative oral commensal species⁴². The Lactococcus lactis group has been reported to reside in the oral cavity, where it ferments sugars and produces bacteriocins such as lactic acid⁴³. In the present study, these taxa exhibited increased relative abundance in 3-day biofilms but declined sharply in 7-day biofilms. Considering that such commensal and early colonizing species showed minimal contribution to RF, it is plausible that low levels of RF intensity may represent an early stage in which pathogenic dysbiosis has not yet fully developed. These findings suggest that the interpretation of biofilm-associated RF should not focus solely on the presence or absence of specific pathogens but also take into account the ecological characteristics of the dominant microbial community. Therefore, RF should not be regarded exclusively as an indicator of pathogenicity but should be interpreted in parallel with comparisons to healthy microbial communities.

In this study, changes in metabolic pathways during the biofilm maturation process were evaluated using PICRUSt analysis, and it was confirmed that the pyruvate fermentation pathway and heterolactic fermentation pathway were activated. The activation of these pathways coincided with the increase in RF intensity and the progression of dysbiosis, suggesting a potential association among these phenomena. In particular, the activation of the heterolactic fermentation pathway plays an important role in cariogenic environments, generating lactic acid, ethanol, and carbon dioxide from glucose via the phosphoketolase pathway⁴⁴. These metabolites further accelerate acidification of the biofilm, thereby promoting the predominance of anaerobic bacteria under acidic conditions. Among the dominant taxa observed in this study, Veillonella is known to convert lactic acid supplied from outside into pyruvate, which is subsequently fermented into short-chain fatty acids (SCFAs) such as propionate and acetate, contributing to the maintenance of a mildly acidic pH within the biofilm³⁴. The activation of these fermentation pathways is therefore closely linked to the proliferation and increased metabolic activity of species such as Veillonella, which in turn paralleled both the increase in RF intensity and the progression of dysbiosis. Furthermore, the observed shift in community composition—from the predominance of the Lactococcus lactis group in 3-day biofilms to the relative increase of Veillonella dispar in 7-day biofilms—suggests that alterations in fermentation pathways are closely associated with changes in microbial interactions and ecological structure.

Meanwhile, biosynthetic pathways such as teichoic acid (poly-glycerol) biosynthesis, hexitol fermentation to lactate, formate, ethanol and acetate, and the superpathway of heme biosynthesis from uroporphyrinogen-III showed a trend of reduced predicted activity in 7-day biofilms. Generally, the heme biosynthetic pathway is associated with porphyrin accumulation and is known to contribute to the expression of RF. Heme is synthesized by the incorporation of Fe into protoporphyrin IX; thus, a reduction in heme biosynthesis implies intracellular accumulation of unmetabolized protoporphyrin IX⁴⁵. The accumulation of protoporphyrin IX, a known fluorophore responsible for RF, may thereby enhance biofluorescence expression. In this study, the predicted activity of the heme biosynthesis pathway decreased, whereas RF intensity increased, consistent with this established principle. Consequently, in 7-day biofilms, activation of fermentation pathways accompanied by reduced heme metabolism and increased protoporphyrin IX suggests a metabolically dysregulated state of the microbial community, leading to the manifestation of RF.

Unlike previous studies that focused primarily on direct correlations between RF and specific caries-associated species, the present study is distinguished by its comprehensive evaluation of ecological changes in the entire microbial community through 16S rRNA sequencing, combined with functional metabolic pathway analysis to elucidate functional imbalances in the biofilm. In particular, by applying a multivariate statistical approach (Partial Least Squares Regression; PLSR), this study quantitatively predicted the contributions of the overall taxa to RF expression, thereby enabling an integrated analysis that simultaneously considered microbial community composition and biofluorescence outcomes. Moreover, the caries-related dysbiosis index proposed in this study was defined on the basis of key species consistently detected in actual carious lesions, in contrast to traditional assessments that relied solely on species counts or the presence of individual pathogens such as Streptococcus mutans. By quantifying the relative abundance of cariogenic species at the community level, this index allows for the dysbiosis most relevant to caries development to be assessed as a quantitative score. This provides significant value, as it reflects microbial shifts more closely linked to disease onset. Importantly, this index also holds potential for application as a quantitative tool to evaluate caries risk in both diverse in vitro biofilm models and clinical samples.

This study has the inherent limitation of being based on an in vitro design. Nevertheless, unlike the commonly used single-species or limited multi-species (2–4 species) biofilm models, we employed a saliva-derived microcosm biofilm model, enables the development of a complex microbial community and thereby better reflects ecological interactions within oral biofilms. Importantly, this model was designed to reproduce cariogenic environmental conditions driven by sucrose exposure, rather than to preserve the original salivary microbiota. From this perspective, the rapid dysbiotic shift observed in vitro within a few days highlights the dynamic nature of oral biofilms and their susceptibility to environmental changes, particularly nutrient availability. In line with this, biofilm maturation in the present study was accompanied by pH reduction, shift in microbial composition, and a marked increase in RF intensity, supporting that this model effectively captures key ecological features associated with cariogenic dysbiosis. However, the microbial community analysis in this study was based on relative abundance data, which reflect compositional shifts within the biofilm community rather than absolute changes in bacterial load. Despite this limitation, saliva-derived microcosm biofilm model has been widely applied as a stable and versatile system for investigating biofilm behavior and evaluating interventions that cannot be directly tested in vivo^15,46–49. Furthermore, by systematically analyzing changes in the overall microbial community and metabolic pathways during biofilm maturation under controlled conditions, this study provides fundamental evidence for elucidating the patho-ecological mechanisms underlying RF. The findings confirm that RF reflects not merely the presence of specific bacterial species, but the dysbiotic state of the pathogenic microbiome.

However, to establish the clinical validity of the proposed caries-related dysbiosis index, further studies are required to investigate the correlation between RF expression patterns and the caries-related dysbiosis index using biofilm samples directly collected from the patients with multiple active caries lesions. Such validation should be conducted in clinically relevant setting rather than being limited to laboratory-based systems. While the present study employed a laboratory-based fluorescence imaging system (QLF-D), translation into clinical practice may be explored using commercially available QLF-based devices such as Qraycam Pro and Qraypen C (AIOBIO, Seoul, Republic of Korea). These systems have been reported to enable the detection of RF associated with pathogenic biofilms in clinical settings^{13,23,50–53}. Based on this technological foundation, a potential application of the present findings would be the non-invasive assessment of biofilm status in situ, where RF intensity and distribution could be used as surrogate indicators of dysbiotic changes in cariogenic biofilms.

Furthermore, with the accumulation of large-scale clinical fluorescence imaging data, integration with analytical approaches, including machine learning, may facilitate more objective interpretation of fluorescence patterns⁵⁴. However, such applications remain exploratory at this stage and require further validation. Overall, the findings of this study provide preliminary evidence supporting the potential of RF as an indicator associated with dysbiotic changes in cariogenic biofilms, warranting further investigation in clinical settings.

Conclusion

The new caries dysbiosis index increased with biofilm maturation under cariogenic conditions up to seven days in vitro. The increase in dysbiosis index correlated strongly with RF intensity as the biofilm matured. PLSR analysis identified bacterial taxa associated with RF expression, moving beyond the traditional single-species approach. These findings suggest that RF can be used as a simple method to detect increased metabolic activity of pathogenic species and dysbiosis.

Material and methods

Saliva collection

Stimulated saliva was collected from a single healthy female donor, aged 20–29 years, with no active dental caries or periodontal disease, after refraining from all oral hygiene practices for 24 h and from food intake for 4 h prior to collection. The study was approved by the institutional review board of Yonsei University Dental Hospital (Approval No. 2–2023-0054), and written informed consent was obtained. To stimulate salivary flow, the donor chewed wax during collection. Saliva secreted during the first 1 min was discarded, and the subsequently produced saliva was collected for approximately 10 min until a total volume of approximately 30 mL was obtained. The collected saliva was gently passed through sterile glass wool (Ducksan Chemicals, Ansan, Korea) to remove large particulate matter, such as food debris, epithelial cell aggregates, and mucin-rich clumps, thereby improving inoculum homogeneity and the reproducibility of biofilm formation. Then mixed with sterile glycerol to a final concentration of 30%. The saliva–glycerol mixture was stored at − 80 °C.

Microcosm biofilm formation

After thawing, saliva was inoculated onto hydroxyapatite (HA) discs (9.5 mm diameter; Clarkson Chromatography Products, South Williamsport, PA, USA), which were then placed into the wells of a 24-well plate (SPL Life sciences, Pocheon, Korea). A total of 13 HA discs were used in this experiment. The saliva-inoculated HA discs were incubated for 4 h to allow pellicle formation. The pellicle-coated HA discs were then transferred to new wells containing 1.5 mL of basal mucin medium (BMM) supplemented with 0.2% sucrose to promote biofilm formation and maturation in a cariogenic environment. The BMM consisted of the following components: 2.5 g/L porcine mucin (Type III, Sigma Chemicals, MO, USA), 10.0 g/L proteose peptone, 5.0 g/L trypticase peptone, 5.0 g/L yeast extract, 1 mmol/L urea, 1 mmol/L arginine, 2.5 g/L KCl and 1 mg/L menadione of pH 7.0^15,55. The HA discs were washed three times with 1 mL of phosphate-buffered saline (PBS) at 24-hour intervals to remove loosely bound biofilms and were then transferred to fresh BMM supplemented with 0.2% sucrose. All cultures were maintained for 7 days at 37 °C under anaerobic conditions (80% N₂, 10% CO₂, 10% H₂). The entire experimental procedure was repeated twice to ensure reproducibility.

Evaluation of cariogenic microcosm biofilm

Measurement of RF intensity

RF intensity of the biofilms was quantitatively measured at 24-hour intervals after inoculation using a Quantitative Light-induced Fluorescence-Digital system (QLF-D Biluminator™, Inspektor Research Systems BV, Amsterdam, The Netherlands). Standard white-light and biofluorescence images of all HA discs were captured using a digital SLR camera with a 22.3 × 14.9 mm sensor (EOS 550D, Canon, Tokyo, Japan) equipped with an EF-S 60 mm f/2.8 macro lens and an optical filters consisting of a 460 nm high-pass threshold and a 600–700 nm band-pass. Imaging was performed in “Live View” mode under the following conditions: for white-light images—shutter speed, 1/60 s; aperture, f/7.1; ISO, 1000; for biofluorescence images—shutter speed, 1/60 s; aperture, f/7.1; ISO, 1600. The illumination system consisted of twelve 405 nm blue LEDs for fluorescence excitation and four 6500 K white LEDs for standard imaging, all arranged to illuminate the sample surface perpendicularly. All images were automatically saved to a PC using dedicated software (C3 v1.24, Inspektor Research Systems BV). To minimize variability, all images were captured under identical conditions of illumination and camera–sample distance. Biofluorescence images were subsequently analyzed using ImageJ software. To quantify changes in RF intensity over time, an area of interest (AOI) was defined based on the contour of each HA disc and saved as a file. The mean RF intensity was calculated within this fixed AOI, with the relative intensity represented by the red component (R, 0–255) of the pixel’s RGB color. All image analyses were performed consistently by a single trained examiner.

Measurement of biofluorescence spectra

To obtain spectral information of the biofluorescence emitted from the biofilms, a fluorescent hyperspectral camera (Pika L, Resonon Inc., Bozeman, MT, USA) was used. Illumination was provided by twelve 405 nm blue LEDs arranged to illuminate the sample surface perpendicularly. The hyperspectral camera was equipped with an EF-S 60 mm f/2.8 macro lens and a > 460 nm cut-off filter. Biofluorescence spectra were acquired at 24-hour intervals after inoculation, and the data were processed using dedicated software (SpectrononPro v3.4.0; Resonon Inc., Bozeman, MT, USA). During acquisition, each HA disc was placed on a capture stage in an environment completely shielded from external light. The spectral range was scanned from 420 nm to 820 nm at 50 nm intervals. For data processing, fluorescence intensities were normalized according to the type of plot. In raw intensity plots, normalization was performed relative to the intensity at 405 nm, which could yield values exceeding 100 because the reference is a fixed wavelength rather than the peak. In contrast, in color spectrum plots, each spectrum was normalized to its maximum peak intensity, setting the highest peak to 100. Thus, although both approaches provide normalized intensity, but they differ in the choice of reference points.

Bacterial viability and acidogenicity of the biofilms

After 7 days of maturation, HA discs with established biofilms were washed three times with 1 mL of PBS to remove loosely attached bacteria and then transferred to conical tubes containing 2 mL of PBS. The biofilms were vortexed for 1 min, followed by sonication for an additional 1 min to prepare bacterial suspensions. The resulting suspensions were serially diluted (10⁻⁴ to 10⁻⁶) and plated onto two types of agar media: 5% tryptic soy blood agar for total bacterial counts and brain heart infusion agar adjusted to pH 4.8 for aciduric bacterial counts. After inoculation, all plates were incubated under anaerobic conditions at 37 °C for 72 h. Upon completion of incubation, CFUs of total and aciduric bacteria were enumerated by a single trained examiner. In addition, to evaluate the acidogenicity of the biofilms, the supernatant from each well was collected at 24-hour intervals, and pH was measured using a pH electrode (Orion 4-Star, Thermo Scientific, Waltham, USA).

16S rRNA sequencing and microbiome analysis

Preparation of biofilm suspension

To evaluate changes in bacterial community structure during biofilm maturation, 16S rRNA sequencing was performed on the inoculum and on biofilms matured for 3 and 7 days. For analysis, biofilms were collected from HA discs in suspension form. Each disc was washed three times with 1 mL of PBS to remove loosely attached bacteria and then transferred to a conical tube containing 2 mL of PBS. The discs were vortexed for 1 min, followed by sonication for an additional 1 min to detach the biofilm from the disc surface and produce a homogeneous biofilm suspension. The prepared suspensions were stored at − 80 °C until subsequent DNA extraction and analysis.

DNA Extraction, PCR amplification and sequencing

Total DNA was extracted using Maxwell RSC PureFood GMO and Authentication Kit (Promega), in accordance with the manufacturer’s instruction. PCR amplification was performed using fusion primers targeting from V3 to V4 regions of the 16S rRNA gene with the extracted DNA. For bacterial amplification, fusion primers of 341 F (5’-AATGATACGGCGACCACCGAGATCTACAC-XXXXXXXXTCGTCGGCAGCGTC-AGATGTGTATAAGAGACAG-CCTACGGGNGGCWGCAG−3’; underlining sequence indicates the target region primer) and 805R (5’-CAAGCAGAAGACGGCATACGAGAT-XXXXXXXX-GTCTCGTGGGCTCGG-AGATGTGTATAAGAGACAG-GACTACHVGGGTATCTAATCC−3’) were used. The Fusion primers are constructed in the following order which is P5 (P7) graft binding, i5 (i7) index, Nextera consensus, Sequencing adaptor, and Target region sequence. The amplifications were carried out under the following conditions: initial denaturation at 95 °C for 3 min, followed by 25 cycles of denaturation at 95 °C for 30 s, primer annealing at 55 °C for 30 s, and extension at 72 °C for 30 s, with a final elongation at 72 °C for 5 min. The PCR product was confirmed by using 1% agarose gel electrophoresis and visualized under a Gel Doc system (Biorad, Hercules, CA, USA). The amplified products were purified the CleanPCR Kit (CleanNA). Equal concentrations of purified products were pooled together and short fragments (non-target products) were removed with CleanPCR (CleanNA). The quality and product size were assessed on a Bioanalyzer 2100 (Agilent, Palo Alto, CA, USA) using a DNA 7500 chip. Mixed amplicons were pooled and the sequencing was carried out at CJ Bioscience, Inc. (Seoul, Korea), with Illumina MiSeq Sequencing system (Illumina, USA) according to the manufacturer’s instructions.

Data analysis pipeline

Raw read preprocessing began with quality assessment using Trimmomatic ver. 0.32, in which low-quality reads with a Q score below 25 were removed⁵⁶. After quality filtering, paired-end reads were merged using the fastq_mergepairs command in VSEARCH ver. 2.13.4⁵⁷. Primers were trimmed based on a similarity threshold of 0.8 using the Myers & Miller alignment algorithm⁵⁸. Non-specific amplicons that did not contain the 16S rRNA gene were detected using the nhmmer function of the HMMER software (ver. 3.2.1) in combination with the appropriate HMM profile⁵⁹. Duplicate reads were removed with the derep_fulllength command, and clustering was performed based on unique reads⁵⁷. Taxonomic assignment was carried out with reference to the EzBioCloud 16S rRNA database⁶⁰ using the usearch_global command in VSEARCH, followed by additional pairwise alignment for higher precision^57,58. Chimeric reads were removed by reference-based chimera detection using the UCHIME algorithm with the EzBioCloud non-chimeric 16S rRNA database, discarding reads with < 97% similarity⁶¹. Reads that remained unclassified at the species level (< 97% similarity) in the EzBioCloud database were clustered de novo using the cluster_fast command to generate new OTUs⁵⁷. Singleton OTUs (comprising only a single read) were excluded from further analysis. Secondary analyses were performed using an in-house bioinformatics pipeline provided by CJ Bioscience Inc. (Seoul, Korea). Alpha diversity was calculated based on the Shannon and Simpson indices⁶², and beta diversity was assessed using multiple distance metrics, including Jensen–Shannon distance⁶³, Bray–Curtis distance⁶⁴, generalized UniFrac distance⁶⁵, and fast UniFrac distance⁶⁶, to visualize inter-sample variation. All analyses were conducted using the EzBioCloud 16S-based MTP, a cloud-based 16S analysis platform. To evaluate changes in bacterial composition during biofilm maturation, Log₂ fold change (LFC) analysis was performed to compare the relative abundances of bacterial taxa between 3-day- and 7-day-matured biofilms. With functional profiles that are predicted by PICRUSt⁶⁷ and MinPath⁶⁸ algorithms, taxonomic biomarkers and functional biomarkers were discovered by statistical comparison algorithms (LDA Effect Size – LefSe⁶⁹ and Kruskal-Wallis H Test⁷⁰. All analytics mentioned above were performed in EzBioCloud 16S-based MTP, which is a CJ Bioscience’s bioinformatics cloud platform. Log₁₀ fold change (LFC) analysis was performed to compare the functional pathways between 3-day- and 7-day-matured biofilms, applying a threshold of 2.0.

Calculation of caries-related dysbiosis index

In this study, an operational definition of the caries-related dysbiosis index was established and applied to quantitatively assess the level of dysbiosis in the formed biofilms. To construct this index, previous studies reporting microbiome analyses of in vivo dental plaque associated with carious lesions were referenced^35,71,72. From these studies, 56 species with a relative abundance of ≥ 0.5% in carious lesions were selected to generate a caries-related community list (Supplementary Table 1). Among the bacterial species detected in the biofilms formed in the present study, those that were included in the caries-related community list and had a relative abundance of ≥ 0.01% at either the time point day 3 or the day 7 were further selected. A total of eight species (Veillonella dispar, Streptococcus constellatus group, Granulicatella adiacens group, Veillonella parvula group, Streptococcus mutans, Prevotella melaninogenica, Streptococcus sanguinis group, Streptotoccus gordonii group) met these criteria and were regarded as the caries-related species present in the biofilms of this study. The caries-related dysbiosis index was calculated as the sum of the relative abundances of these eight species within each biofilm sample.

Statistical analysis

To compare changes in RF intensity according to biofilm maturation period, a one-way ANOVA was performed. To assess the relationship between caries-related dysbiosis level and RF intensity, Spearman correlation analysis was conducted between the dysbiosis index and RF intensity for each biofilm sample.

PLSR analysis was performed to identify bacterial species contributing to RF intensity. In this analysis, the relative abundances of bacterial species detected in each biofilm sample were used as explanatory variables (X), and the corresponding RF intensity was set as the response variable (Y). Based on the PLSR results, VIP scores were calculated to evaluate the relative contribution of each bacterial species to RF intensity^73–75.

All statistical analyses, except for PLSR and VIP score calculations, were performed using the Statistical Package for the Social Sciences (SPSS) version 26.0 (SPSS Inc., Chicago, IL, USA), with the significance level (α) set at 0.05. PLSR and VIP score analyses were conducted using Python (version 3.11.11).

Electronic Supplementary Material

Below is the link to the electronic supplementary material.

Author contributions

Conceptualization and Supervision: E.S. L., B.I. K. Data Curation, Formal analysis, Investigation, Methodology and Writing—Original Draft Preparation: Y.J. C†, A.R. K†. Writing—Review and Editing: E.S. L, H.I. J, W. B, Á.M. L, B.I K.

Funding

This research was supported by the Bio & Medical Technology Development Program of the National Research Foundation (NRF) & funded by the Korean government (MSIT) (No. RS-2022-NR067350); under the framework of international cooperation program managed by the National Research Foundation of Korea (No. RS-2025-00437009); and by Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education (No. RS-2023-00237396).

Data availability

The datasets analyzed during the current study are available from the corresponding author upon reasonable request. The microbiome analysis datasets generated and analyzed the current study are available at [https://doi.org/10.6084/m9.figshare.30985459](https:/doi.org/10.6084/m9.figshare.30985459).

Declarations

Competing interests

The authors declare no competing interests.