Ensuring food safety has become increasingly challenging due to the rapid industrialization of food production, the impact of global warming, and shifts in consumer preferences towards organic, minimally processed, and ready-to-eat (RTE) foods (Vlachos and Georgantzis, 2016; Wambogo et al., 2020; Duchenne-Moutien and Neetoo, 2021; Meijer et al., 2021). These challenges are further compounded by an increasing population of vulnerable individuals and the emergence of enhanced-virulent, antimicrobial-resistant foodborne pathogens (Kim and Ahn, 2022; Duma et al., 2024; Hu et al., 2024). Consequently, there is an urgent need for efficient and rapid methods to analyse food and ensure its safety.

Among foodborne pathogens, Listeria monocytogenes poses a significant public health threat, with infections often linked to contaminated RTE foods. Infections can lead to severe morbidity and mortality, particularly among high-risk populations such as pregnant women, neonates, and immunocompromised individuals (Buchanan et al., 2017). The persistence of L. monocytogenes in food-processing environments is facilitated by its stress tolerance and ability to form biofilms, which help it to withstand standard cleaning and disinfection protocols (Duze et al., 2021; Fagerlund et al., 2021; Osek et al., 2022; Finn et al., 2023). Additionally, its ubiquitous distribution makes it easy to be introduced into production facilities, where it can establish long-term contamination reservoirs.

Listeria monocytogenes isolates are traditionally differentiated into four serogroups (1/2, 3, 4, and 7), which are further subdivided into 14 serotypes, 1/2a, 1/2b, 1/2c, 3a, 3b, 3c, 4a, 4b, 4c, 4d, 4e, 4h, and 7, based on variations in somatic O and flagellar H antigens (Maury et al., 2016; Yin et al., 2019). Notably, serotypes 1/2a, 1/2b, and 4b account for 95% of human listeriosis cases (Orsi et al., 2011), with serotype 4b most frequently implicated in outbreaks and clinical cases probably due to its increased virulence (Maury et al., 2016; Muchaamba et al., 2021). Molecular typing techniques, including PCR and multi-locus sequence typing (MLST), have enabled the classification of isolates into PCR serogroups [e.g., IIa (1/2a, 3a), IIb (1/2b, 3b, 7), IIc (1/2c, 3c), IVb (4b, 4d, 4e), IVb var. (4b), and L (4a, 4c)], and the identification of clonal complexes (CCs) (Doumith et al., 2004; Ruppitsch et al., 2015). These CCs and serotypes have been grouped into four major lineages (I–IV), which are associated with distinct virulence, environmental distribution, and host range (Hyden et al., 2016). For instance, strains of serotype 4b and 4h, have been shown to be more virulent compared to other serotypes (Yin et al., 2019; Muchaamba et al., 2021). Timely and accurate detection and typing of such pathogens is crucial for preventing outbreaks, mitigating their impact, and controlling infections. It is also essential for monitoring L. monocytogenes in processing plants, enabling risk-based inspection programs, and ensuring effective decontamination to prevent contamination of end products. As the relationship between serotype, MLST, and pathogenic potential becomes clearer, the call for risk-based regulation of L. monocytogenes intensifies, underscoring the need for fast and accurate strain typing.

However, routine analysis of L. monocytogenes by traditional serotyping methods, such as agglutination assays, is limited by labour intensity, cost, and the need for technical expertise, with reproducibility at times being challenging (Doumith et al., 2004; Gorski, 2014). PCR-based techniques offer faster and more cost-effective alternatives; however, they still require trained personnel and may lack the resolution needed for detailed strain-level differentiation. These limitations can reduce their effectiveness for real-time surveillance and outbreak response. Advanced molecular approaches like MLST and whole genome sequencing (WGS) provide high-resolution insights but are constrained by higher costs, longer turnaround times, and the need for specialized bioinformatics expertise (Ruppitsch, 2016). Nevertheless, recent advances in sequencing technologies, such as Oxford Nanopore coupled with artificial intelligence (AI), are helping to reduce turnaround times and make WGS more accessible for routine use (Urban et al., 2023). Similarly, matrix-assisted laser desorption/ionization-time of flight (MALDI-TOF) mass spectrometry enables rapid species identification but lacks the discriminatory power required for accurate strain or serotype differentiation (Gant and Chamot-Rooke, 2024).

Fourier-transform infrared (FTIR) spectroscopy has emerged as a promising alternative for rapid, cost-effective pathogen typing (Al-Fraihat et al., 2025; Lurie-Weinberger et al., 2025). By measuring the absorption of infrared light at specific wavelengths, FTIR detects vibrational energy transitions of molecular bonds in cellular micro- and macromolecules, generating unique “spectral fingerprints” that reflect the biochemical composition of cells (Rakovitsky et al., 2020; Muchaamba and Stephan, 2024). This technique has demonstrated comparable performance to WGS in specific outbreak investigations, such as tracking multidrug-resistant Acinetobacter baumannii (Lombardo et al., 2021) and differentiating extended-spectrum β-lactamase-producing Klebsiella pneumoniae strains (Rakovitsky et al., 2020). Despite its potential, challenges such as inter-laboratory reproducibility and species-specific limitations (e.g., with Staphylococcus aureus) necessitate further optimization and validation of FTIR protocols.

The integration of FTIR with specialized analysis software, such as in the IR Biotyper® system, presents an opportunity to enhance pathogen classification processes. By leveraging advanced spectral analysis capabilities alongside AI and machine learning, FTIR’s discriminatory power is significantly enhanced, facilitating fast strain typing. The global nature of foodborne pathogen outbreaks highlights the need for widely accessible strain classification databases to standardize pathogen tracking and outbreak investigations, much like MALDI-TOF MS has done for species identification. Existing FTIR databases for L. monocytogenes (Oberreuter et al., 2023), Salmonella spp. (Cordovana et al., 2022; Oberreuter et al., 2025), Streptococcus pneumoniae (Passaris et al., 2022), and Legionella pneumophila (Pascale et al., 2022) demonstrate the potential of this approach.

Various FTIR systems, including those combined with multivariate analysis or machine learning algorithms like Artificial Neural Networks (ANN), have been successfully applied to differentiate Listeria species and L. monocytogenes serogroups, serotypes, or haplotypes, with reported accuracies of 80%–100% (Rebuffo et al., 2006; Rebuffo-Scheer et al., 2007; Janbu et al., 2008; Davis and Mauer, 2011; Romanolo et al., 2015; Oberreuter et al., 2023). However, despite these promising results, most previous studies were conducted under conditions that limit FTIRs direct applicability to routine food testing and outbreak response workflows. In particular, the use of non-selective growth media such as Blood agar (Oberreuter et al., 2023), Tryptone soy agar (TSA) (Rebuffo et al., 2006), and Tryptic soy broth (TSB) (Romanolo et al., 2015) does not align with ISO 11290-1:2017 recommendations, which specify selective chromogenic media for Listeria spp. isolation. As a result, additional subculturing steps are often required to meet FTIR spectral quality standards, delaying analytical workflows and reducing compatibility with routine diagnostic protocols. Moreover, key performance characteristics, including reproducibility over time, robustness to variation in growth conditions, discriminatory power at different typing levels, and database coverage across serogroups, serotypes, and clonal complexes, have not been systematically evaluated within a single, standardized framework. Many studies have also relied on custom-built data analysis pipelines, lacking integration into standardized, user-friendly platforms, further hindering broader adoption in diagnostic and regulatory settings. Collectively, these limitations highlight the absence of a clear benchmarking framework for evaluating FTIR performance under ISO-relevant culture conditions and in relation to operational needs, particularly when compared to WGS-based workflows. Consequently, the practical added value of FTIR in the current era of increasingly routine WGS remains insufficiently defined, particularly for early-stage triage, dereplication, and prioritization of isolates. This is especially relevant in settings where sequencing capacity, turnaround time, bioinformatic processing, and data management can still constrain the rapid analysis of large isolate collections.

In this study, we address these gaps by evaluating the discriminatory power of FTIR for distinguishing L. monocytogenes serogroups and strain clusters using the IR Biotyper® system, a commercially available platform. We benchmarked its performance against traditional typing methods and systematically assessed the impact of culture media and sample preparation on FTIR performance. We also evaluate a pre-developed FTIR-ANN-based serogrouping classifier using a diverse set of well-characterized field strains, testing its resolution and accuracy. In addition, we examine FTIR’s utility for outbreak or persistence investigation and dereplication (i.e., identification of redundant / duplicate isolates), with the aim of defining its role as a rapid first-line screening tool complementary to WGS.

This study demonstrates that FTIR spectroscopy is a robust and reproducible phenotypic approach for rapid L. monocytogenes strain typing. Across a diverse collection of well-characterized strains, FTIR showed high reproducibility and strong discriminatory power at the strain and serogroup levels, aligning with traditional serotyping methods and WGS-based assignments. With a turnaround time of approximately 3 hours per 30 isolates, these findings support the suitability of FTIR for high-throughput applications where rapid preliminary classification and exclusion of unrelated isolates are required, with WGS serving as a confirmatory high-resolution method.

As expected for a phenotypic technique, FTIR showed limited resolution for fine-scale differentiation based on MLST or for distinguishing closely related serotypes within the same serogroup, particularly among serotypes 1/2a, 1/2b, and 1/2c. This limitation likely reflects spectral overlap in polysaccharide-associated regions that dominate FTIR-based classification of L. monocytogenes. Similar constraints have been reported for other phenotypic typing methods, highlighting fundamental differences between biochemical fingerprinting and genome-based approaches. Consequently, FTIR should not be viewed as a replacement for WGS, but rather as a complementary tool designed to address different operational needs.

Some L. monocytogenes MLST CCs encompass multiple serotypes, complicating FTIR-based differentiation at this level. For example, CC9 includes isolates from serotypes 1/2a, 1/2c, and 3c (Muchaamba et al., 2021), which likely contributes to the observed clustering challenges. In contrast, FTIR studies on other species, such as Lactiplantibacillus plantarum and Bifidobacterium longum subsp. longum, demonstrated higher FTIR resolution than MLST and performance comparable to WGS or PFGE (Deidda et al., 2021; Li et al., 2022). In B. animalis subsp. lactis, FTIR provided superior differentiation of strains compared to MLST and PFGE. Thus FTIR’s ability to differentiate isolates by MLST might be species-specific. However, these studies analysed smaller strains sets (10–20) with fewer MLST CCs, whereas our study included a larger, more diverse strain collection, likely explaining the lower MLST level resolution.

The ANN-based serogroup classifier in the IR Biotyper® system performed consistently across selective and non-selective culture media, including ISO-relevant chromogenic agars, with high agreement to serological serotyping (ARI and AWC = 1.00) and WGS-based assignments (ARI = 0.948 and AWC = 0.987). Importantly, this study extends previous work by demonstrating that reliable serogroup assignment can be achieved without additional subculturing on non-selective media, thereby improving compatibility with routine food testing workflows. While classification confidence varied between media types, no systematic misclassification was observed, highlighting the robustness of the approach when growth conditions are standardized. These findings align with those of Oberreuter et al. (2023), who first evaluated the IR Biotyper® using the manufacturer-provided ANN classifier for L. monocytogenes, albeit with Blood agar only.

Rebuffo et al. (2006) and Rebuffo-Scheer et al. (2007) used ANN-based analysis of spectral regions “3,100 to 2,800 cm⁻¹, 1,800 to 1,500 cm⁻¹, and 1,200 to 700 cm⁻¹” and “1,200 to 900 cm⁻¹ and 1,800 to 1,400 cm⁻¹,” respectively, to differentiate Listeria species, and L. monocytogenes serogroups and serotypes. Our study expanded these approaches, incorporating the 1,300 to 800 cm⁻¹ and 1,200 to 900 cm⁻¹ windows for strain and serogroup differentiation. While serogroup agreement was comparable (98.8% vs. 100% with serological serotyping) using all four spectral windows, serotype resolution, which reached 91.6% in the other studies (Rebuffo-Scheer et al., 2007), was not assessed. According to the manufacturer, the current IR Biotyper® classifier is not yet effective for serotype assignment. Differences in classifier architecture, including training cycles and principal component selection, likely explain this resolution discrepancy. In the study by Rebuffo-Scheer et al. (2007) they used NeuroDeveloper™ software (Synthon GmbH), which employed a four-level hierarchical ANN, while our study classifier (IR Biotyper) relied on 30 PCs and 200 training cycles, focusing on the 1,300 to 800 cm⁻¹ window, preselected for serogroup differentiation. Therefore, refining spectral windows and classifier design may improve discrimination, particularly for closely clustered serotypes or MLST-specific profiles. Additionally, such refinements could enable FTIR-based analysis to provide risk assessment insights. For example, FTIR was able to cluster 89 L. monocytogenes isolates into two groups, which corresponded with their susceptibilities to sakacin P, a class IIa bacteriocin (Oust et al., 2006).

Growth conditions were identified as a major determinant of spectral variability. Media type affected clustering behavior more than incubation duration, with spectra from the same strain grown on different media often clustering separately. This emphasizes the importance of harmonized cultivation protocols and explains part of the inter-study variability reported in the literature (Hong et al., 2022; Hu et al., 2023). For example, Hu et al. (2023) demonstrated superior discriminatory power for P. aeruginosa grown on Mueller-Hinton agar compared to Blood agar, used in the study by Zendri et al. (2024). In our study, FTIR classification accuracy was maintained across all eight tested media types. However, classification confidence varied, with ALOA and Blood agar performing best, providing the most consistent, high-quality spectra. In contrast, RAPID’L.mono and BHI broth yielded lower classifications confidence scores, likely due to media-driven changes in cellular metabolic states and molecular composition.

Although liquid cultures require longer processing times due to multiple washing steps, they offer greater control over sample biomass and uniformity, making them a promising avenue for refining FTIR protocols. In our study, FTIR spectra of the same strain cultivated in BHI broth or BHI agar clustered closely, indicating retention of strain-specific spectral patterns, though some medium-induced separation was observed. This further highlights the need for consistent standardized cultivation protocols to minimize growth condition-induced spectral variability. Contrary to previous findings in other species, such as L. plantarum (Deidda et al., 2021), culturing L. monocytogenes in BHI broth at 37 °C did not improve discriminatory power at the serotype-level. In fact, classification scores decreased for some broth-grown strains compared to agar-grown counterparts, likely due to a mismatch between the sample preparation method and the ANN classifier, which was trained on spectra from agar-grown isolates. Therefore, aligning growth conditions with classifier training data is critical for optimal FTIR performance.

In a study by Davis and Mauer (2011), liquid cultures were chosen over solid media for FTIR analysis of L. monocytogenes. They hypothesized that faster bacterial growth in liquid media results in more uniform physiological states, whereas on solid media, colonies may exhibit heterogeneous physiological characteristics due to variations in colony size and age. This heterogenicity can negatively impact FTIR reproducibility (Choo-Smith et al., 2001; Sandt et al., 2006). Adding to the challenges of using agar cultures, previous studies (Janbu et al., 2008), including our own, have reported difficulties in harvesting or resuspending colony material from some types of agar media in ethanol. This is likely due to the hydrophilic surface properties of L. monocytogenes grown on these media. Additionally, the risk of media carryover is higher with agar plates than with broth, potentially interfering with the spectral profile and reducing classification confidence. To address these challenges, future research should pursue the optimization of liquid culture conditions to minimize spectral variation caused by colony heterogeneity, media carryover, and harvesting difficulties, ultimately improving FTIR’s reproducibility and discriminatory performance for L. monocytogenes strain typing.

While incubation duration did not significantly affect clustering patterns, it did influence classification confidence. Prolonged incubation (48 h) led to reduced confidence. This was expected, as the classifier was optimized for strains grown at 37 °C for 24 h. Conversely, a 20-h incubation maintained clustering accuracy for most strains, but insufficient colony growth resulted in suboptimal spectral quality for some. Although shortening the incubation period could accelerate results, it risks compromising spectra quality. Lowering the incubation temperature to 25 °C, where L. monocytogenes strongly expresses flagella, improved clustering, particularly for serotype differentiation within the 1/2 and 3 serogroups. This suggests that targeted growth conditions optimization may enhance FTIR’s resolution. However, it is important to note that these findings were based on an investigation of only 32 strains, representing 7 of the 14 serotypes of L. monocytogenes. Therefore, further studies with a broader range of strains and serotypes are needed to confirm the generalizability of these results.

In a blinded, outbreak-like cluster analysis, FTIR effectively grouped genetically related CC9 isolates and excluded most unrelated strains, despite limited resolution among very closely related cgMLST clusters. While FTIR could not fully resolve all genomic relationships, its ability to rapidly narrow the pool of candidate isolates demonstrates clear practical value. In outbreak and persistence investigations, this form of early-stage phenotypic triage can substantially reduce the number of isolates requiring immediate WGS (Jacobs et al., 2025), thereby improving turnaround times and optimizing sequencing resources. Establishing a cutoff based on cgMLST allelic differences would also aid in defining FTIR’s limit of detection in outbreak settings, ensuring that the tool is applied appropriately for various levels of relatedness. The most recent version of the IR Biotyper® software incorporate additional machine-learning approaches, including k-nearest neighbors (kNN), which may offer enhanced flexibility for FTIR-based classification. Preliminary exploratory application of kNN produced clustering trends broadly consistent with the outbreak cluster analysis, however, in the absence of validated performance benchmarks and standardized cut-off criteria for defining relatedness or clonality in L. monocytogenes, these results were not included. Systematic validation and standardization of kNN-based FTIR analyses, as well as other machine-learning models, should therefore be addressed in future work.

Compared to PCR serogrouping, FTIR offers a simpler workflow. Although its serogroup resolution is currently lower than PCR when using the existing IR Biotyper® serogroup classifier, FTIR demonstrates superior discriminatory power by enabling strain-level differentiation, which is not possible with PCR serogrouping. FTIR has also shown potential for differentiating serotypes, as demonstrated by our findings with cultures grown at 25 °C and in the work of Rebuffo-Scheer et al. (2007), who applied a four-level hierarchical ANN for serotype classification. Overall, FTIR’s ability to differentiate both strains and, in some cases, serotypes makes it a potentially more powerful tool for strain typing than PCR which only identifies serogroups (Doumith et al., 2004; Vitullo et al., 2013).

A key limitation of FTIR and similar phenotypic methods is susceptibility to phenotypic changes driven by small genetic mutations. In our study, strains N843_10 and N843_15 showed discordance in serotype classification, both were assigned to serogroup 3 by FTIR- and serological serotyping but classified as serogroup 1/2 by WGS. These strains, isolated from the same prosthetic hip joints 5 years apart, may have diverged genetically due to adaptation in the prosthetic joint environment. Notably, N843_15 underwent multiple genetic changes, likely driven by random mutations or selective pressure (Muchaamba et al., 2020). Such mutations can cause phenotypic changes that alter serotype classification, even in clonally related strains. This is consistent with other studies where clonally identical strains exhibited phenotypic differences, such as hemolysis, due to single SNPs (Charlier et al., 2012). This discordance highlights the limitations of phenotypic serotyping methods, as subtle genetic variations in key genes can significantly impact results. In contrast, WGS can detect and quantify these differences, making it more reliable for fine-scale strain differentiation, particularly when phenotypic changes arise from small genetic mutations. Furthermore, as with most typing methods, FTIR analysis depends on pure cultures for accurate strain identification. The spectral profiles of mixed cultures differ significantly from those of pure cultures, which can affect the accuracy of strain typing and differentiation (Nyarko and Donnelly, 2015).

Differentiation of serotype 4b from other serogroup 4 strains appears feasible, provided that enough non-serotype 4b isolates are included in the reference database. However, the current database contains only a limited number of non-serotype 4b serogroup 4 strains, which may affect classification robustness. A key limitation of our study is the underrepresentation, or complete absence, of some serotypes, including 3b, 3c, 4a, 4c, 4d, 4e, 4h, and 7, within our strain collection. This gap may hinder the reliability of FTIR-based classification, particularly for these serotypes. To address this, future studies should expand the strain database by incorporating a broader range of isolates from environmental, clinical, and food sources. Collaborative efforts with reference laboratories and strain repositories could facilitate access to rare serotypes, ultimately improving FTIR spectroscopy discriminatory power and reliability for L. monocytogenes serotype differentiation. Further enhancements could be achieved by refining the classifier with secondary models for serotype differentiation, which may improve both the accuracy and overall utility of FTIR systems for L. monocytogenes strain typing.

Overall, our findings confirm FTIR as a fast, accessible, and scalable first-line screening tool that complements WGS rather than competes with it. Although it may not resolve fine-scale distinctions like WGS, its primary strength lies in reproducibility, speed, and dereplication capacity, enabling efficient prioritization of isolates for downstream genomic analysis. However, the strong influence of growth conditions on spectral profiles highlights the need to apply optimized FTIR protocols to improve reproducibility and discriminatory power, particularly at the serotype level. Continued expansion of reference databases, optimization of cultivation protocols, and validation of emerging classification algorithms (e.g., kNN) are expected to further enhance the operational value of FTIR-based typing in food safety surveillance and outbreak response.