Application of MALDI-TOF Protein Profiles for Rapid Detection of Streptococcus agalactiae Highly Virulent Strains: ST1

Abstract

Expanding the capacity of Matrix-Assisted Laser Desorption Ionization Time of Flight Mass Spectrometry (MALDI-TOF MS) beyond species identification to strain typing becomes a new challenge in clinical microbiology. This study demonstrated a specific identification of Streptococcus agalactiae sequence type 1 (ST1) by a manual decision tree and automatically ranking from the newly added MTPPs library, which has not been previously reported. The mass spectra of 25 STs (277 isolates) were generated. The presence and absence of specific peaks were combined to create a decision tree for manual identification. Three peaks at 3127, 5914, and 6252 in combination with m/z 3368 and 6281 were used for primary identification of ST1. However, to differentiate ST1 and ST314, five additional peaks were required. For the automatic system, the MTPP of all isolates was divided into three training–testing ratios of 40:60, 50:50, and 60:40. All categories revealed excellent accuracy rates of above 90% for ST1 identification. The 60:40 group showed the highest overall performance, in which sensitivity was observed at 83.9 to 96.8%, and specificity reached up to 100.0% for both the top two and the top three matches. In conclusion, we propose that the MTPP from MALDI-TOF is a potential model for speedy bacterial typing, crucial in epidemiology, prevention, and patient management.

1. Introduction

Streptococcus agalactiae or Group B Streptococcus (GBS) is globally known to be one of the most common bacteria, along with Streptococcus pneumoniae and Haemophilus influenzae, in causing invasive infection among newborn babies. The organism also causes various infections in adults and the elderly. Most of these strains are tetracycline-resistant, and many strains also develop resistance to erythromycin [1,2,3,4]. Sequence type 1 (ST1), a subtype within the GBS species, has emerged as a globally distributed lineage with clinical relevance in both neonates and adults. ST1 has increasingly been reported across various geographical regions, particularly in Europe, North America, and Asia, not only in neonatal infections but also in invasive infections among non-pregnant adults, including the elderly and those with comorbidities [5,6,7,8]. The clonal expansion of ST1 is attributed in part to its adaptability and the acquisition of antimicrobial resistance elements, particularly to macrolides and fluoroquinolones [9]. In certain regions, such as Southeast Asia and parts of China, ST1 has become one of the predominant lineages isolated from adult invasive disease cases, suggesting a shift in the epidemiological dynamics of GBS infections [2,10].

Clinically, ST1 strains of S. agalactiae are associated with a broad spectrum of infections. In neonates, ST1 can cause early-onset sepsis and late-onset meningitis, although it is less frequently implicated in neonatal disease compared to other serotypes like ST17 [11]. In contrast, among adults, ST1 is commonly linked to bacteremia, skin and soft tissue infections, urinary tract infections, and occasionally endocarditis. The virulence of ST1 is mediated by a combination of factors, including capsular polysaccharide type V (which is frequently but not exclusively associated with this lineage), surface proteins like Alpha C and Rib, and regulatory systems that enhance adherence, immune evasion, and invasion [12]. ST1 is also characterized by the presence of scpB (C5a peptidase) and hylB (hyaluronate lyase), which facilitate tissue invasion and immune modulation. While not considered as hypervirulent as the ST17 lineage, ST1 exhibits a high degree of invasiveness in adult populations, particularly in immunocompromised hosts. Its growing role in non-neonatal infections underscores the need for ongoing molecular surveillance, as shifts in dominant lineages may influence both vaccine design and public health strategies targeting GBS disease [11,13].

Multi-locus sequence typing (MLST) and whole-genome sequencing (WGS) are widely recognized as gold-standard methods for bacterial strain identification; however, their time-consuming nature limits their utility in urgent clinical and public health contexts. At the same time, effective infection prevention and control require microbiological typing methods that can accurately differentiate between bacterial isolates of the same species. The selection of an appropriate typing method depends on the specific epidemiological context, the required resolution, and the temporal and geographical scope of application. Ideally, a typing system should be highly typeable and possess strong discriminatory power to distinguish both unrelated and closely related isolates, particularly in outbreak investigations. Moreover, it should be rapid, cost-effective, reproducible, and easy to perform and interpret [14].

This study aims to develop a MALDI-TOF protein profile (MTPP)-based method for the rapid identification of specific STs, providing a potential framework for broader application to other clinically significant organisms in future research. MALDI–TOF is widely recognized for its speed and accuracy in genus and species identification, relying on a comprehensive spectral database. Beyond routine identification, its applications are expanding most notably in the early detection of antimicrobial resistance, which serves as a critical warning for severe pathogens such as methicillin-resistant Staphylococcus aureus (MRSA) and vancomycin-resistant Enterococcus (VRE) [15,16,17]. The use of MALDI-TOF MS in strain typing remains controversial, particularly due to variability in mass spectral peaks among organisms [18,19]. Nonetheless, several studies have demonstrated its utility in epidemiological typing of significant pathogens. For example, MALDI-TOF has been applied to distinguish Escherichia coli ST131, a globally dominant extraintestinal pathogenic lineage [20], as well as to track epidemic clones of S. aureus [21] and high-risk strains of Pseudomonas aeruginosa [22]. Distinct spectral biomarkers have also been reported for Acinetobacter baumannii, Klebsiella pneumoniae, Neisseria meningitidis, Listeria monocytogenes, Haemophilus influenzae, Streptococcus pneumoniae, and Streptococcus pyogenes [18,23,24,25,26,27,28,29,30].

Recently, specific mass spectral peaks were incorporated into statistical models to predict S. agalactiae STs in China, including ST10, ST12, ST17, and ST19, primarily identified through maternal-neonatal surveillance programs [19]. These findings highlight the potential of MALDI-TOF as a rapid, proteomic-based typing tool. Building upon the hypothesis that distinct MLST DNA profiles may correlate with unique protein signatures, the present study aims to further explore MALDI-TOF MS for the identification of S. agalactiae ST1 and other STs.

2. Materials and Methods

2.1. Protein Preparation for MALDI-TOF

Two hundred seventy-seven S. agalactiae clinical isolates from the previous study [2] representing 25 STs were used. Each isolate from the storage was sub-cultured on sheep blood agar at 35 °C for 18–24 h. A single colony was then sub-cultured one more time before being used. The protein extraction was processed according to the manufacturer’s ethanol/formic acid tube protocol. Briefly, a few bacterial colonies (approximately 1 µL standard loop full) were suspended in 300 µL of High-performance liquid chromatography (HPLC) grade sterile water (Sigma Aldrich, Merck KGaA, Darmstadt, Germany). Then, 900 µL of 95% ethanol, HPLC grade (Sigma Aldrich, Merck KGaA, Darmstadt, Germany), was added to the suspension and vortexed. The microtube was centrifuged at 13,000× g for 2 min, and the supernatant was removed. The pellet was air-dried in a Class II biosafety cabinet for approximately 10 min. Twenty-five microliters of 70% formic acid (HPLC grade) (Sigma Aldrich, Merck KGaA, Darmstadt, Germany) was added and mixed to break the bacterial cell wall. To enrich the bacterial protein, 25 µL of acetonitrile (HPLC grade) (Sigma Aldrich, Merck KGaA, Darmstadt, Germany) was added and mixed well before centrifuging at 13,000× g for 2 min. Finally, the supernatant was collected and used within 1 h.

2.2. MTPP Library

One microliter of extracted protein solution was spotted on each well of the MALDI target plate (30 spots/isolate), followed by overlaying 1 µL of the α-cyano-4-hydroxycinnamic acid (HCCA) matrix solution (10 mg HCCA in 1 mL of standard solvent, 50% acetonitrile + 47.5% water + 2.5% Trifluoroacetic acid) (Sigma Aldrich, Merck KGaA, Darmstadt, Germany). Notably, air-dry at room temperature for every step of all spots, and analyze within 30 min. Randomly, thirty mass spectra were recorded by MALDI-TOF MS (Sirius, Bruker Daltonics, Bremen, Germany) in positive linear mode, using the AutoXecute run editor version 3.4.150.0 program. In-run quality control was performed using the Burker bacterial test standard, with mass peaks ranging from 2000 to 20,000 Da. Only logarithm scores [log(s)] ≥ 2.3–3.00 were used to achieve acceptable mass peaks for the new MTPP library, and each mass peak’s inaccuracy was limited to 0.05%, and the mass peaks between run errors did not exceed ±0.5%. At least 20 spectra/isolates were used for creating new MTPP libraries. A summary of the feature MTPP data analysis to predict ST1 is shown in

Table 1. A summary of the feature MTPP data analysis to predict ST1.

Detail	Requirement
Bacterial growth conditions	18–24 h at 35 °C on sheep blood
Protocol guideline	CLSI M58-Ed1
Standard reference	MLST, whole genome sequencing (WGS)
Protein extraction	Standard tube ethanol/formic acid extraction
Matrix	HCCA
Target plate	Stainless
m/z range	2000–20,000
Internal mass control	Bacterial test standard (Bruker)
Instrument	MALDI-TOF MS Sirius
Program running	AutoXecute Run Editor. Version 3.4.150.0
Program analysis	With flexAnalysis software version 3.4
The acceptable mass peak of each ST for data analysis	Within in run not exceeding 0.05%
	Between runs not exceeding 0.5%
	s/n ratio > 5 is valuable
The acceptable minimum random mass PP/isolate	Concordance of at least 20 PPs/isolate
Selected the criteria for mass peaks for data analysis	Maximum mass peaks generated
Program automatic prediction	MBT compass explorer version 4.4.100
Interpretation criteria/decision criteria	The top three matches 3/3; top two matches 2/2; two of three matches 2/3
Training isolates the ST of interest for the new library	At least 30 isolates
Testing isolates ST	At least 30 isolates for the ST of interest and other STs up to 100 isolates for evaluation of sensitivity, specificity, and accuracy

Abbreviation: PP = Protein profile, Da = Dalton.

.

2.3. Manual Visualization of Mass Peaks

In order to investigate the MTPP manually, only the peaks that showed s/n above five were classified as positive, and s/n at five or below were classified as negative. The data was analyzed using the Excel program, and the candidate peaks were summarized as a decision tree.

2.4. MTPP Dendrogram

The MTPP from an individual isolate from any STs that contained five or more isolates was added into a dendrogram using MBT Compass Explorer version 4.4.100 software (Bruker Daltonics, Bremen, Germany) to create MTPP training (TR) groups at 40, 50, and 60%, respectively. However, those STs with 2–4 isolates, two or one of each, were used as the TR group. The remaining isolates were classified as the testing (TS) group. The STs containing only one isolate were all used as TR.

2.5. TR and TS Groups

The dendrogram of each ST was set to a maximum value of 1000 on the X-axis (Figure 1). According to the distance level, with cutoffs of 300, 200, and 100, the isolates of the TR group were at 40, 50, and 60%, respectively. An individual MTPP was selected to cover a maximum number of branches in each group.

2.6. Evaluation of the Automatic Identification of ST1

The MALDI-TOF result from each isolate was presented as the ten most matching MTPPs. Only the top three ranks were used for analysis. The data were evaluated for sensitivity (S), specificity (SP), and accuracy (AC) using the online statistical software calculator from MedCalc Software Ltd., https://www.medcalc.org/calc/diagnostic_test.php (Version 23.2.1; accessed on 23 March 2025). The system accepted 90% of AC agreements between sequence types of ST1 and non-ST1 [31].

3. Results

3.1. Manual Mass Peak Identification of ST1

All isolates of S. agalactiae displayed fifteen consistent mass peaks at m/z 2978, 3094, 3407, 3470, 3983, 4102, 4453, 5078, 5956, 6185, 6815, 6943, 7966, 8204, and 9092 ± 5. Among these, three peaks demonstrated high discriminatory potential. The peaks at m/z 3127 (84.0%) and 6252 (97.3%) were commonly found in ST1 isolates but rarely detected in non-ST1 strains (4.5 and 5%, respectively). In contrast, peaks at m/z 6893 were frequently observed in non-ST1 compared with ST1 isolates (94.6 versus 2.7%, respectively). It is worth mentioning here that the peak at 5914, although detected at a low frequency in non-ST1 isolates (22%), was completely absent in ST1. These peaks served as valuable indicators for ruling in or ruling out ST1, but alone are insufficient for definite identification. To improve classification accuracy, a hierarchical decision tree based on MTPP analysis was developed using a training set of 277 isolates, comprising 75 ST1 and 202 non-ST1 strains.

The initial classification step relied on the presence or absence of three key peaks: m/z 3127, 5914, and 6252. This generated four primary entry combinations that enabled early rule-out of non-ST1 isolates with high confidence. For example, isolates with peak combinations such as (+/+/+) or (−/+/−) were directly classified as non-ST1. Differentiation between ST1 and closely related sequence types, particularly ST314 (which shares a +/−/+ peak pattern), required further analysis using additional marker peaks, including m/z 2990, 9018, 7492, 3064, 3444, and 2313. Among these, the m/z 7492 peak was especially informative: its presence strongly supported ST1 classification, while its absence indicated ST314. Similarly, the combination of positive peaks at m/z 3064 and 3444 further reinforced ST1 identification, while their absence guided classification toward ST314. Other decision pathways within the hierarchical structure, such as (−/−/+) and (−/−/−), also effectively identified ST1 strains by incorporating unique discriminatory peaks at m/z 3368 and 6281, respectively (Figure 2).

3.2. Automatic Prediction of ST1

The predictive performance of ST1 identification using MALDI-TOF MS was assessed using three different training-to-testing data ratios: 40:60, 50:50, and 60:40. For each ratio, evaluation was conducted under three performance conditions: exact matches in all three replicates (3/3), correct identification within the top two ranked matches (Top 2), and at least two correct matches out of three (2/3). The distribution of isolates in each training and testing group is shown in

Table 2. Distributions of GBS isolates in Training: Testing groups for protein profile analysis.

Sequence Type	Number of Isolates	Proportions
		40:60		50:50		60:40
ST1	75	30	45	37	38	44	31
ST283	81	30	51	40	41	49	32
ST19	22	9	13	11	11	13	9
ST17	18	7	11	8	10	9	9
ST23	16	6	10	8	8	10	6
ST12	10	4	6	5	5	6	4
ST485	8	3	5	4	4	5	3
ST103	6	2	4	3	3	4	2
ST314	6	2	4	3	3	4	2
ST889	5	2	3	2	3	3	2
ST861	4	2	2	2	2	2	2
ST651	3	1	2	1	2	1	2
ST652	3	1	2	1	2	2	1
ST28	3	1	2	1	2	2	1
ST41	3	1	2	1	2	2	1
ST196	2	1	1	1	1	1	1
ST249	2	1	1	1	1	1	1
ST335	2	1	1	1	1	1	1
ST1626	2	1	1	1	1	1	1
ST14	1	1	0	1	0	1	0
ST509	1	1	0	1	0	1	0
ST739	1	1	0	1	0	1	0
ST751	1	1	0	1	0	1	0
ST1167	1	1	0	1	0	1	0
ST361	1	1	0	1	0	1	0
Total	277	111	166	137	140	166	111

. The results revealed that this new approach was excellent for identifying ST1, with a specificity above 95%, and the accuracy rates were above 90% in all categories. However, the sensitivity was slightly low in the 3/3. In contrast, the 2/3 groups showed the highest percentages of sensitivity, specificity and accuracy rate (

Table 3. The evaluation result of the automatic system prediction ST1 by MALDI-TOF.

Prediction ST1 Automatically Match	Training: Testing by MALDI-TOF MS
	40:60			50:50			60:40
	3/3	Top 2	2/3	3/3	Top 2	2/3	3/3	Top 2	2/3
Sensitivity (%)	73.3	88.9	97.8	71.1	76.3	94.7	83.9	90.3	96.8
Specificity (%)	100.0	97.5	96.7	100.0	100.0	99.0	100.0	100.0	98.8
Accuracy (%)	92.8	95.2	97.0	91.4	93.6	97.9	95.5	97.3	98.2

3/3, matched as ST1 in the top three ranks. Top 2, matched as ST1 in the top two ranks, and the 3rd rank was any ST (including ST1). 2/3, matched as ST1 at least two of the top three ranks. Sensitivity: True Positives (TP)/True Positives (TP)+ False Negatives (FN); Specificity: True Negatives (TN)/True Negatives (TN) False Positives (FP); % Accuracy: TP + TN/(TP + TN + FP + FN) × 100.

).

4. Discussion

This study highlights the utility of MALDI-TOF MS as a powerful tool for rapid, high-throughput typing of S. agalactiae strains, focusing on the identification and discrimination of ST1, a clone frequently associated with invasive infections in both neonates and adults [6,19,32]. In our previous study, we proposed the use of the Alpha C protein gene (bca) along with susceptibility profiles, particularly tetracycline and erythromycin, as a surrogate marker for ST283 identification in Thailand. However, a significant limitation of this approach is the overlap in phenotypic and genotypic profiles between ST283 and ST1 strains. For example, ST1 strains may also possess the bca gene and exhibit susceptibility to the same antibiotics, leading to potential misclassification [2]. To address these limitations, this study proposes a novel approach using MTPP to discriminate ST1 from ST283 and other STs based on specific spectral features.

Four mass peaks, m/z 3127, 6252, 5914, and 6893, demonstrated significant discriminatory value. Specifically, m/z 3127 and 6252 were consistently present in most ST1 isolates, while peaks m/z 5914 and 6893 were predominantly observed in non-ST1 isolates and absent in ST1. This differential pattern offers a means of rapid preliminary classification through rule-in/rule-out logic. Despite this, relying solely on these four peaks proved insufficient for definitive ST1 identification, particularly in the context of closely related sequence types with partial overlap in peak expression. These findings are consistent with prior observations that MALDI-TOF MS, although efficient for genus and species level identification, often requires refinement for subtyping and epidemiological discrimination [33,34,35].

To enhance precision, a hierarchical decision tree based on MTPP was constructed, offering a structured approach for ST1 prediction. This algorithm was trained on a set of 277 isolates (75 ST1 and 202 non-ST1) and initially stratified isolates based on the presence or absence of three critical peaks (m/z 3127, 5914, and 6252). The first tier of decision-making resulted in four possible combinations, enabling early exclusion of non-ST1 strains with high confidence. For instance, isolates presenting as (+/+/+) or (−/+/−) for these peaks were reliably classified as non-ST1. Such stratification is vital for early screening, especially in clinical microbiology settings where rapid decision-making can influence treatment choices and infection control measures [32]. However, to discriminate ST1 from closely related clones such as ST314, which shared an overlapping (+/−/+) peak pattern, additional decision nodes were required. These included secondary discriminatory markers at m/z 2990, 9018, 7492, 3064, 3444, 2313, 3368, and 6281. Notably, the presence of peak m/z 7492 proved especially useful; its detection favored ST1, whereas its absence indicated ST314. Similarly, positive peaks at m/z 3064 and 3444 supported ST1 classification, while negative outcomes shifted interpretation toward ST314. For isolates falling into less common pathways such as (−/−/+) and (−/−/−), distinct peaks at m/z 3368 and 6281 enabled accurate classification. It is worth mentioning that the peak at m/z 6893 has also shown strong potential as a marker for manual sequence type (ST) differentiation, and it was consistently reported in multiple studies [19,36,37]. With our proposed algorithm, using the m/z 6893 required more steps to distinguish ST1 from ST314. Therefore, the refined rules are required to maximize the discriminatory capacity of MALDI-TOF MS.

For automatic detection of ST1, the robustness of the classification system was further evaluated across three different training-to-testing data sets of 40:60, 50:50, and 60:40 ratios, respectively. Each category was assessed under three performance conditions: exact matches in all three replicates (3/3), correct identification within the top two matches (Top 2), and at least two correct matches out of three (2/3). The data revealed that the performance metrics improved according to the TR ratios. The TR:TS ratio at 60:40 yielded the highest overall performance, in which the accuracy reached 98.2% in the 2/3 group. These results suggest that as the training dataset becomes more comprehensive, the ability of the MALDI-TOF MS-based classification system to identify ST1 with high fidelity is markedly improved. The study encountered intra-ST1 heterogeneity, which hindered accurate model classification. To address this, the number of MALDI-TOF protein profiles (MTPPs) was increased to enhance robustness. Although the ideal design involved at least 30 ST1 isolates per training/testing set and up to 100 non-ST1 isolates for cross-validation, sample limitations prevented implementation of the planned 70:30 training-to-testing ratio [31,38]. The dendrogram was essential for guiding the selection of MTPPs in the training library, as randomly selected mass profiles often failed to provide comprehensive coverage of the protein expression diversity required for robust model development.

Taken together, the data illustrate the potential of MALDI-TOF MS, when coupled with well-structured algorithmic approaches like MTPP, to serve as a reliable and cost-effective method for subspecies or sequence type classification. Unlike whole-genome sequencing (WGS), which is labor-intensive, time-consuming, and not widely accessible in many healthcare settings, MALDI-TOF MS offers speed and affordability, with the results reporting time of less than an hour after the bacterial colony is processed. The ability to integrate discriminatory peak analysis with automated pattern recognition further enhances its utility. The hierarchical decision tree developed in this study represents a scalable model that can be adapted for other clinically significant STs or pathogens. Furthermore, the application of such a classification framework can significantly support antimicrobial stewardship and infection control. For example, rapid identification of ST1, a lineage associated with increased invasiveness and potentially distinct antimicrobial resistance profiles, can guide empiric therapy decisions and prompt surveillance interventions [39]. Given the growing emphasis on precision medicine and real-time diagnostics, the integration of MALDI-TOF MS strain typing into routine microbiology workflows could revolutionize pathogen characterization, particularly in resource-limited settings or during outbreak scenarios where rapid data turnaround is critical.

5. Conclusions

The results of this study provide strong evidence that MALDI-TOF MS, when enhanced through discriminatory mass peak analysis and hierarchical decision modeling, can accurately and efficiently classify S. agalactiae ST1. The system’s high sensitivity, specificity, and adaptability across various training set configurations affirm its clinical and epidemiological value. Future work should aim to validate these findings across independent datasets and diverse geographic regions, assess reproducibility across different MALDI-TOF MS platforms, and explore integration with machine learning models to optimize performance further. Additionally, expanding this framework to other sequence types and pathogens could further broaden its applicability, offering a transformative tool in microbial diagnostics and public health.