SYTO 9 fluorescent dye was acquired from Thermo Fisher Scientific (Waltham, MA, United States). Streptococcus thermophilus BH16 (S. thermophilus), Lactobacillus casei ATCC344 (L. casei), and Lactobacillus bulgaricus Y7135 (L. bulgaricus) cultures were maintained in our laboratory stock collections. Calcium chloride (CaCl₂), sodium alginate, and De Man, Rogosa, and Sharpe (MRS) broth were procured from Baisi Biotechnology Co., Ltd. (Hangzhou, Zhejiang, China) and Aladdin Bio-Chem Technology Co., Ltd. (Shanghai, China), respectively. Additional chemical reagents were sourced from Sinopharm Chemical Reagent Co., Ltd. (Shanghai, China). Microbiological-grade water, prepared using a Milli-Q RG purification system and sterilized by filtration through a 0.22 μm membrane filter, was employed for all buffer solutions.

For all the three different live LAB strains, a 100 μL aliquot of LAB suspension (about 1 × 10⁸ CFU/mL in PBS) was pipetted into a 1.5 mL Eppendorf tube, followed by adding 100 μL SYTO 9 solution (1 μM/mL). The solution was incubated at room temperature for 5 min, which was then subjected to fluorescence spectral scanning by a fluorescence spectrophotometer. The excitation wavelength was set to be 450 nm and integration time was 3 s. Fluorescence intensity at 520 nm was recorded and used for calculating the mean fluorescence intensity. To cause LAB cells death, the LAB suspensions were heated at 80 °C for 10 min. The dead LAB cells were stained and analyzed as the steps above.

The stained LAB samples were also analyzed by fluorescence microscope (DMi8, Leica, Germany) and flow cytometry (FCM, Accui C5, BD, United States) as described previously (He et al., 2024). For FCM analysis, specifically, a laser of 20 mW at 488 nm was used for excitation. The optical configuration of the flow cytometer included a 530/30 nm bandpass filter to detect green fluorescence from SYTO 9. Data were acquired in log-scale, with the acquisition threshold set on the forward scatter (FSC) signal. Photomultiplier tube voltages were configured at 350 V for FSC and 200 V for the green fluorescence channel (FL1). A dual-parameter gate based on FSC and FL1 intensities was applied to distinguish bacterial cells from background noise. For statistical reliability, a minimum of 5,000 cellular events were recorded per sample. Subsequent data analysis was performed using FlowJo software (BD, United States) to determine precise fluorescence intensity.

Streptococcus thermophilus, L. casei, and L. bulgaricus were cultivated in MRS broth using established protocols described by He et al. (2017) and Chen et al. (2024a). Upon reaching mid-exponential phase (OD₆₀₀ value reached about 1.0), cellular biomass was harvested via centrifugation (5,000 × g, 5 min). The pellets underwent three successive washes with sterile phosphate-buffered saline (PBS, containing 137 mM NaCl, 2.7 mM KCl, 10 mM Na₂HPO₄ and 1.8 mM KH₂PO₄, at pH 7.4). Then the LAB cells were resuspended in fresh PBS to achieve a standardized optical density of OD₆₀₀ = 0.5. Cell viability exceeding 99% was confirmed for all processed suspensions using the SYTO 9/propidium iodide (PI) dual-staining assay as described previously (Chen et al., 2024a).

Bacterial suspensions (1 × 10⁷ to 1 × 10⁸ CFU/mL) were allocated to eight experimental cohorts. Each cohort underwent thermal exposure in a water bath at distinct temperatures (35, 40, 45, 50, 55, 60, 65, and 70 °C) for 60 s to assess thermotolerance. Immediately following heat treatment, aliquots were aseptically collected for viability quantification via standardized plate counts as described previously (Wang et al., 2023; Cao et al., 2025). Serial decimal dilutions were prepared, and 100 μL aliquots were plated in triplicate onto sterile MRS agar. After 72 h aerobic incubation at optimal growth temperatures (37 °C), colonies were enumerated on plates exhibiting 30–300 colonies per plate.

To classify different LAB strains, a 15 μL bacterial suspension (1 × 10⁶–1 × 10⁸ CFU/mL) was mixed with an equal volume of 1 μM SYTO 9 solution in PCR tubes. Samples were analyzed using a thermoregulated fluorometer (StepOnePlus™, Thermo Fisher Scientific, United States) with the following thermal profile: 5 s at 37 °C, ramp to 98 °C at 0.1 °C/s, and 5 s at 98 °C. The fluorescence intensity and temperature were used to plot melting curves, which were subsequently processed through multifactor dimensionality reduction via the software SPSSPRO analytics platform.¹

For the simultaneous classification and quantification of LAB in yogurt, the yogurt sampled from the grocery store was directly subjected to 10 times gradient dilution by PBS, with a final dilution of 0 to 1 × 10² CFU/mL. A 5 μL aliquot of bacterial suspension was pipetted into a PCR tube. Another 5 μL aliquot of SYTO 9 solution (2 μM) and a 10 μL aliquot of MRS medium (2×) were added to the tube. Such protocol ensured that each tube contained no more than one LAB cell. The LAB containing tubes were incubated at 37 °C for 12 h to amplify the LAB. After amplification, the tubes were placed in a thermoregulated fluorometer for testing as described above.

A 500 μL aliquot of LAB solution (1 × 10⁸ CFU/mL) was mixed with 500 μL calcium chloride (CaCl₂, with final concentrations ranging from 0 to 50 mg/mL), or sodium alginate (with final concentrations ranging from 0 to 200 mg/mL). The mixtures were incubated 10 min, and then a 15 μL LAB suspension was mixed with an equal volume of 1 μM SYTO 9 solution in PCR tubes. The tubes were placed in a thermoregulated fluorometer for melting curve analysis as described above. To prepare alginate gel embedded LAB, 1 mL LAB suspension (1 × 10⁸ CFU/mL) was mixed homogenously with 1 mL alginate solution (20 mg/mL), as described previously (Cabrita et al., 2013; He et al., 2014). The mixture was added into 1% calcium solution by peristaltic pump and left to harden for 2 h. The alginate beads containing LAB were directly subjected to melting curve analysis as described Section 2.5.

For all the experiments, the mean values and standard deviations (SD) were calculated by Origin software. All the difference analyses were performed using the Student’s t-test.

Throughout the course of evolutionary biology, microbial lifeforms like bacteria have developed extensive genetic variation within their DNA sequences as an adaptive strategy against dynamic environmental challenges. Such genomic variability fundamentally drives metabolic and morphological differentiation, enabling the emergence of diverse bacterial variants characterized by distinct thermal tolerance profiles, antimicrobial resistance patterns, and cellular membrane properties (Chen et al., 2024b; Letizia et al., 2025). This biological principle provides a potential approach for distinguishing LAB based on their differences in thermotolerance and DNA composition. Scheme 1 describes the work principle of melting curve for thermotolerance and DNA composition analysis of LAB. The membrane semipermeable and dsDNA-specific dye, SYTO 9, was incubated with LAB cells. As the temperature rose from 40–70 °C, progressive membrane disruption led to LAB cell death, facilitating the influx of SYTO 9 into the cells and enhancing fluorescence intensity. However, further heating caused gradual denaturation of bacterial genomic dsDNA, leading to a subsequent decline in fluorescence signal. Variations in thermostability and dsDNA composition among different LAB strains enabled their differentiation by analyzing the melting profiles using multidimensional factor reduction techniques. Meanwhile, the three characteristic values, semilethal temperature (Ts), total lethal temperature (Td), and DNA melting temperature (Tm) can be captured for assessing the thermotolerance of a LAB strain. The selection of SYTO 9 was based on its dual specificity: an affinity for dsDNA and a selective staining of dead bacterial cells. This dual specificity enables the generation of distinctive bacterial melting curves. In contrast, other nucleic acid stains, such as PI and ethidium bromide (EB), do not possess these combined properties.

Although SYTO 9 is known for its good permeability to live cells and bacteria, here we found by chance that SYTO 9 exhibited greater permeability to dead S. thermophilus than to live S. thermophilus. To further extend this phenomenon to other LAB strains, we used S. thermophilus, L. bulgaricus, and L. casei as LAB models to systematacially investigate the selectivity of SYTO 9 against live and dead LAB. These three LAB strains were chosen because they are the most widely used in dairy industry in China. Figures 1a–c shows the fluorescence (FL) spectra of live and dead cells for the three LAB strains. The LAB were incubated with SYTO 9 at a final concentration of 1 μM/mL and with free SYTO 9 molecules remaining in the LAB solutions. Similar to the previously reporters, the SYTO 9 could enter live LAB and showed significant FL enhancement at 520 nm once binding with dsDNA of the LAB. However, we found that for all the three LAB strains, a further pronounced FL enhancement occurs in dead LAB cells. The degree of this enhancement varied among strains; among them, L. casei exhibited the greatest difference in FL intensity at 520 nm between dead and live cells. The results of quantitative FCM analysis and semi-quantitative fluorescence microscope observation agreed well with that of FL spectra (Figures 1d–g). Therefore, it is supposed that although SYTO 9 exhibits non-discriminatory staining across LAB viability states, the pronounced fluorescence enhancement occurs exclusively in dead LAB cells due to the compromised membrane barrier function.

SYTO 9 is also a selective dye for dsDNA. It generates green emission upon intercalation into the dsDNA helix and this signal diminishes when thermal denaturation separates the DNA strands, releasing unbound dye molecules. Based on this mechanism, the dye has become a staple tool in quantitative PCR (qPCR), serving both to monitor real-time DNA amplification and to characterize products via melting curve profiling (Horakova et al., 2011; Wan et al., 2016; Shen et al., 2023). Given that thermotolerance and genomic dsDNA exhibit variations across LAB species and strains, we explored the possibility of using SYTO 9 combined with a thermoregulated fluorometer to classify LAB through melting curve profiling.

Figure 2a displays the dynamic FL (λ = 520 nm) changes for S. thermophilus of different concentrations observed during the LAB cells heating from 37 °C to 98 °C. The FL intensity correlated positively with LAB concentration. When the LAB concentration exceeded 1 × 10⁶ CFU/mL, the curve profiles remained consistent across varying concentrations. Normalization of FL signals resulted in nearly superimposable curves (Figure 2b), suggesting that the observed trends were independent of LAB concentration but potentially dependent on species or strain specific thermotolerance and genomic differences. Interestingly, the temperature-responsive FL profile of LAB samples diverged markedly from conventional dsDNA melting curves. While standard dsDNA analysis shows monotonic FL reduction with rising temperature (Li et al., 2024; Sun et al., 2024), our LAB model demonstrated biphasic behavior: an initial FL enhancement phase (37–68 °C) followed by signal attenuation. We attributed this distinctive pattern to two sequential thermal transitions: biomembrane permeabilization phase and dsDNA denaturation phase. In biomembrane permeabilization phase (37–68 °C), progressive biomembrane disruption (Figure 2a, gray region) facilitated SYTO 9 penetration into LAB cells, amplifying fluorescence. In dsDNA denaturation phase (>68 °C), the subsequent heating melted chromosomal DNA in nonviable cells (Figure 2a, blue region), causing FL decline as dsDNA transitioned to ssDNA.

Figure 2c exhibits the derivative FL reporter of melting curves, which identified three characteristic temperatures named semilethal temperature (Ts), total lethal temperature (Td), and DNA melting temperature (Tm). We believed that at Ts point, about 50% LAB cells would die, and at Td point all the LAB cells were dead. When the temperature continued to rise to Tm point, about 50% dsDNA would melt. When the dead LAB was subjected to heating again, no biomembrane permeabilization phases were observed as shown in Figure 2d. Therefore, the two sequential thermal transitions together shaped the melting curves of the live LAB. Good reproducibility of these patterns was demonstrated through quintuplicate independent experiments (Figures 2e–h). However, when the LAB concentrations were lower than 10⁶ CFU/mL, the FL signal was not enough for drawing the melting curves (Figures 2e,f). For the quintuplicate independent experiments of S. thermophilus, significant differences between Ts, Td, and Tm were observed, and no intra-experiment variations were observed (Figure 2i). Therefore the three characteristic temperatures had the potential to be used for LAB classification. Plate cultivation was carried out to further confirmed the two sequential thermal transitions, as shown in Figures 2j,k. About half of the LAB died when the temperature reached 60 °C (about Ts point of Figures 2c,g); all the LAB died when the temperature reached 70 °C (about Td point of Figures 2c,g), where the FL reporter in Figures 2a,b increased to a maximum value. The results of L. bulgaricus and L. casei were similar to that of S. thermophilus, as shown in Supplementary Figures S1, S2. However, the three LAB strains exhibited differences in detailed melting curve and characteristic temperatures. Therefore the composite “melting curve” simultaneously reports membrane integrity loss (LAB died) and nucleic acid denaturation, providing a multiple-parameter thermal signature for LAB classification.

To visualize the features of different LAB strains, the Ts, Td, and Tm were extracted from the melting curves and plotted to a three dimensional scatter plot as shown in Supplementary Figure S3a. Obviously, the three LAB strains S. thermophilus, L. bulgaricus, and L. casei were clearly differentiated from each other. However, the absence of any one of the three parameters precludes the classification of these LAB (Supplementary Figures S3b–S3d), indicating the absolute necessity of both the biomembrane permeabilization phase and dsDNA denaturation phase for LAB classification. More simply, the melting curves could be directly converted into two-dimensional scatter plots through multivariate dimensionality reduction, as shown in Figure 3. The algorithms, principal component analysis (PCA), kernel principal component analysis (KPCA, Cosin and RBF kernels), linear discriminant analysis (LDA), and t-distributed stochastic neighbor embedding (T-SNE) were applied in the dimensionality reduction. For every LAB strain under investigation, over 200 individual melting curves were analyzed, with each curve corresponding to a unique LAB colony (Figure 3A). The visualization results (Figures 5b–f) demonstrated clear spatial segregation among data points representing different LAB strains following dimensionality reduction. Comparative evaluation revealed t-SNE’s superior performance in achieving complete discrimination between the three LAB strains. Consequently, this algorithm was selected for the subsequent analysis of melting curve patterns and LAB classification purposes.

To simulate yogurt composition, Ultra-high temperature sterilized (UHT) milk was inoculated with three representative LAB strains: S. thermophilus, L. bulgaricus, and L. casei. This tri-strain combination was selected as the model system due to its prevalence in Chinese yogurt products. We firstly evaluated the method’s capability for concurrent LAB quantification and differentiation. Prior to milk inoculation, LAB cultures were grown to log phase in MRS medium to ensure viability and culturability. The single-cell derived LAB suspensions were then analyzed via BMCA. Post-dimensionality reduction, the scatter plots (Figures 4a–c) revealed four distinct clusters: one control group and three strain-specific populations. A linear relationship was observed between LAB-positive signals and spiked concentrations within a fixed volume, enabling direct quantification by scaling cell counts to sample volume. Method validation against plate counting demonstrated strong agreement (R² = 0.9914, Figure 4d). The as-developed BMCA-based method was further tested for the analysis of targeted LAB ratio in UHT milk. The UHT milk samples spiked with three LAB strains of different L. bulgaricus ratios were subjected to BMCA, as shown in Figures 4e–h. An excellent correlation between the BMCA measured L. bulgaricus ratios and theoretical ratios was obtained with R² of 9,878. Hence, the BMCA could achieve the simultaneous quantification and classification of the LAB.

In Walmart and other supermarkets in China, there are variety of yogurts that claim containing active LAB of multi-species and with related health benefits. According to the national standard (GB 19302-2010, China), the claimed live LAB count should be higher than 1 × 10⁶ CFU/mL. To assess these claims, the study applied the BMCA technique to quantify and identify LAB in store-bought samples. Initial processing involved serial dilution of yogurt to achieve LAB concentrations <10² CFU/mL. Subsequently, 5 μL aliquots of the diluted suspension were transferred to PCR tubes preloaded with SYTO 9 and MRS broth for culturing and subsequent melting curve analysis (Figure 4i). This workflow ensured each reaction vessel contained ≤1 viable LAB cell for precise detection. For sterilized yogurt, no live LAB cells were detected (Figure 4j). The three LAB strains were indeed identified by BMCA in the yogurt claimed with live L. casei, L. bulgaricus, and S. thermophilus (Figure 4k). We further compared yogurts of four different bands named NO1, NO2, NO3, and NO4. All of them claimed to have the above three live LAB strains. As shown in Figure 4l, although the total LAB of the four yogurts varied from 8 × 10⁷ to 20 × 10⁷ CFU/mL, all meeting the standard of ≥1 × 10⁶ CFU/mL. Significant variations in inter-species ratio also occurred among different brands. For instance, in NO3 yogurt, S. thermophilus was the most abundant while in other yogurt L. casei showed relatively high abundance. Additionally, it was found that the live L. bulgaricus (2 × 10⁵) in NO3 yogurt was lower than the standard, and thus its claim to have functional live L. bulgaricus was arguable.

Food products distributed across regions inevitably undergo environmental stresses such as high temperatures, elevated pressure, and UV radiation, thereby accelerating spoilage (Shima et al., 2010; Mbye et al., 2020; Wu Z. et al., 2025). Most LAB strains are sensitive to thermal change, necessitating refrigerated storage to maintain LAB viability throughout the storage period for conventional live-probiotic foods and drugs. Such stringent storage requirements substantially increase logistical and preservation costs. Recent efforts have sought to enhance probiotic stress resistance through encapsulation or surface modification strategies (Thi-Tho et al., 2022; Kwon et al., 2025; Xie et al., 2025). Here, we comparatively evaluated the thermotolerance of LAB in different formulations by using BMCA to capture their characteristic temperatures. The edible calcium alginate was used to prepared encapsulated LAB. Figures 5a,b show the melting curves (exhibited as derivative reporter) of the LAB in the absence and presence of alginate (Figure 5a) and calcium chloride (Figure 5b). Neither alginate nor calcium chloride alone could changed the melting curve of the tested LAB. All the characteristic temperatures were consistent among alginate and calcium chloride of different concentrations (Figures 5c,d), indicating that they had no influence on the thermotolerance of LAB. However, when the alginate and calcium chloride were combined to encapsulated the LAB into the calcium alginate gel, the encapsulated LAB exhibited migration at characteristic temperatures (Figures 5e,f). Both Ts and Td values for the LAB embedded in calcium alginate were higher than that for free LAB. Therefore, calcium alginate gel may serve as a heat barrier to improve the thermotolerance of LAB.