Legionella genus includes bacteria that commonly inhabit fresh water environments where they are able to parasitize various protozoa species (Rowbotham, 1980; Fields, 1996; Fields et al., 2002). Legionella becomes a risk for human health when it spreads into man-made water systems, which act as amplifiers and disseminators of the microorganism representing the main sources of infection (Newton et al., 2010; Mercante and Winchell, 2015).

Although there has been only one report of infection transmitted from person to person (Borges et al., 2016; Correia et al., 2016), the inhalation of contaminated aerosol is the typical Legionella transmission route. Upon aerosol formation via man-made water systems, Legionella can infect the human lung and cause mainly two forms of disease: Pontiac fever, a non-pulmonary flu-like symptoms syndrome, and Legionnaires' disease, a severe pneumonia that can lead to death if specific antibiotic therapy is not promptly taken (Cunha et al., 2016).

Legionella virulence is due to its ability to infect and grow in human macrophages, while its environmental persistence largely depends on its ability to replicate inside protozoa and biofilms growth (Newton et al., 2010; Abu Khweek and Amer, 2018), and also to resist shock or continuous disinfection treatment applied to the water distribution systems. Host susceptibility is a crucial point for human infection: people with risk factors such as smoking, immune or chronic diseases, and age more than 50 years are more exposed to Legionella infections (Cunha et al., 2016).

Although 66 Legionella species have been documented (Parte et al., 2020), and about half of them are linked to human infections, Legionella pneumophila (Lp) is the most virulent and studied species. There are 15 distinct serogroups of Lp based on their reactions with antibodies that recognize the bacterial lipopylisaccharide (LPS) molecule (Ciesielski et al., 1986). In particular, according to epidemiological data, Lp serogroup 1 (Lp Sg1) is responsible for the majority of recognized Legionnaires' disease cases in Italy and in other European countries (European Centre for Disease Prevention Control, 2021; Rota et al., 2021a).

In Italy, the incidence of legionellosis in 2020 was 34.8 cases per million inhabitants, with a significant decrease compared to 2019, mostly due to the COVID-19 pandemic. Among the 2,074 surveillance notification forms sent to the National Register of legionellosis Surveillance relating to as many cases of legionellosis, 53 are reported as probable cases and 2,021 cases have been subsequently identified (Rota et al., 2021a).

Considering the European data, in 2019, 11,298 cases were reported, of which over 80% have been classified as confirmed and all ascribable to Lp Sg1 (European Centre for Disease Prevention Control, 2021). However, Legionnaire's disease remains widely underestimated because of the predominant usage of diagnostic methods that do not detect all Legionella species and serogroups.

According to the EU case definition, laboratory criteria for case confirmation must include at least one of the following three: (1) isolation of Legionella spp. from respiratory secretions or any normally sterile site; (2) detection of Lp antigen in urine; (3) significant rise in specific antibody level to Lp Sg1 in paired serum samples (European Commission, 2018).

The gold standard method to isolate Legionella both from clinical and environmental samples is the classical culture technique. Nevertheless, it is important to note that it is seldom performed in clinical laboratory practice not only due to the need of specific and expensive media but also to the unavailability of adequate specimens from the lower respiratory tract (Fields et al., 2002; Murdoch, 2003; Mercante and Winchell, 2015; Whiley and Taylor, 2016). Furthermore, after the isolation, serological tests such as the direct fluorescent antibody (DFA), indirect immunofluorescent assay (IFA), latex agglutination, immunochromatographic tests, etc, must be used. However, due to the frequent use of polyclonal antibodies, none of these tests provide highly specific and sensitive results for each of the Lp serogroups and Legionella non-pneumophila species (non-Lp).

Moreover, the 15 known Lp serogroups can be subdivided further into monoclonal subgroups. The specific monoclonal antibodies (MAbs) and /or Legionella serogroup rabbit antisera were introduced for serotyping. The most famous Dresden panel, based on monoclonal antibodies (MAbs) described by Helbig et al. (1997) and validated by European Study Group for Legionella Infections (ESGLI) (Helbig et al., 2002), represents a recognized approach for isolates serotyping. Unfortunately, the most precious monoclonal antibodies produced by the Dresden University are at a limited availability to be provided to all the laboratories (Helbig et al., 1997, 2002).

A novel method based on Real Time PCR to distinguish Lp and Lp Sg1in clinical specimens has been described in Mentasti et al. (2015), suggesting the possibility to introduce the detection of Lp nucleic acid among the criteria to define a LD confirmed case. This approach has been demonstrated to be very useful in outbreak events to discriminate Sg1 and the cluster of Sg 2–15 in recent Italian epidemic events (Scaturro et al., 2021). Unfortunately, these techniques are expensive and scarcely utilized by routine laboratories. Therefore, there is the need to find other strategies to discriminate Lp serogroups that overcome the limits of the current diagnostic tests.

The urinary antigen test (UAT) is a fast and non-invasive diagnostic method, but most of the commercial tests are highly specific only for Lp Sg1, and few of these tests show a cross-reactivity with the other serogroups (Helbig et al., 2001; Svarrer et al., 2012; Wong et al., 2021).

It is noteworthy that several severe cases of pneumonia due to other serogroups have been reported (Grottola et al., 2012; Nishizuka et al., 2014). Recently, an outbreak of community-acquired Legionnaires' disease caused by Lp Sg2 has been described in Italy (Scaturro et al., 2021). During this outbreak, the adoption of multiple diagnostic tests (culture, real-time PCR, and serology) made it possible to detect an otherwise unidentified epidemic event. Furthermore, the UAT poses a limitation into epidemiological investigation because it does not allow the correlation between clinical and environmental strains.

Molecular methods, such as PCR and real-time PCR, have improved the diagnosis enabling rapid and reliable identification of all Legionella species. Nevertheless, these techniques require a subsequent sequencing step to determine the serogroup, and they have a limited potential for routine application (Blyth et al., 2009). However, to date, a molecular standardized method for the discrimination of all different Lp serogroups has not been established (Mentasti et al., 2014).

In conclusion, for routine purpose, conventional methods are time-consuming and elaborate, often with a limited availability in routine laboratories (e.g., monoclonal antibodies, PCR or sequencing), therefore some handy and common tests (e.g., agglutination test) have reported low sensitivity and the occurrence of false negative results, leading to inaccurate Legionella identification (Fields et al., 2002; Pascale et al., 2020).

Recently, different rapid methods with high sensitivity and specificity have been developed to overcome the limitations of traditional methods for the detection and identification of several pathogens (Grenga et al., 2019).

Mass spectrometry is nowadays a widely used method to identify many microorganisms at species level, replacing traditional biochemical methods, and it is the reference method in clinical routine diagnostics (Rychert, 2019; Oviaño and Rodríguez-Sánchez, 2021). The MALDI Biotyper system (Bruker Daltonics GmbH & Co. KG) proved to be very effective for the identification of Legionella spp., although it requires the expansion of its library and fails to discriminate between Lp serogroups (Fujinami et al., 2011; Gaia et al., 2011; Svarrer and Uldum, 2012; Pascale et al., 2020).

Fourier Transform Infrared Spectroscopy (FTIR) has been proven to be an innovative and reliable approach for microbial typing. The technique enables the analysis of the biochemical composition of the microbial cells using the infrared light absorption, which is the result of the interaction of all the different (bio)molecules present in the cells with the infrared light (Martak et al., 2019). Several studies, performed with many different microbial genera and species, showed a high discriminatory power, enabling a differentiation at the species or subspecies level (Preisner et al., 2010).

The IR Biotyper® (Bruker Daltonics GmbH & Co. KG) is a FTIR-based system suitable for microbial typing in routine practice. Since its introduction, several studies investigated the use of IR Biotyper® to type different bacteria, especially enterobacteria and non-fermenting Gram-negative bacilli other than Gram-positive, such as Streptococcus pneumoniae and Staphylococcus aureus (Burckhardt et al., 2019; Martak et al., 2019; Rakovitsky et al., 2020).

On the contrary, to the best of our knowledge, only few and not recent studies have been performed for Legionella (in particular for Lp) identification using FTIR, and no studies have been carried out yet to develop a classifier based on an artificial neural networks (Horbach et al., 1988; Helm et al., 1991a,b).

In this study, different Lp serogroups, including American Typing Culture Collection (ATCC) reference strains and environmental isolates, were tested by using the IR Biotyper® classifier discriminatory power to establish the ability of this system to differentiate the single Lp serogroups. The results were compared with the traditional latex agglutination test, and the limitations and potential of this methodology to lay the foundation for Legionella classifier improvement of an artificial neuronal network development were discussed.

The recent Italian outbreaks and the increment of Lp infections incidence in Italy and in other countries have demonstrated that a correct and fast characterization of isolates is important and necessary (Szewzyk et al., 2000; Faccini et al., 2020; Brunello et al., 2022). The diagnostic techniques used by clinical and environmental laboratories often lead to an absence or mis-classification of isolates, and as a consequence to the underestimation of the risk of infection, as documented by epidemiological data (Rota et al., 2021a). These issues require more relevant attention, also in the context of the pandemic event linked to Sars-Cov-2. Several epidemiological data (Rota et al., 2021b) other than recent literature (Dey and Ashbolt, 2021; Liang et al., 2021; Rhoads and Hammes, 2021) have already documented how the lockdown period and the absence of the preventive measures have led to an increase of Lp infection risk. Moreover, a mis-classification of cases was reported, which is attributable to the similarity of pneumonia symptoms with COVID-19 (Rota et al., 2021a). Therefore, the improvement or the introduction of techniques able to better identify the Legionella isolates could contribute to overcome the limit of traditional methods, increasing the discriminatory power and more accurately focusing on the public health choices.

The culture technique with a preliminary discrimination by the subculture in BCYE cys+ and cys- media is the first and essential step required for Legionella identification. In laboratory practice, Lp serogrouping is performed using the latex agglutination test, which consists of specific polyclonal antibodies available to identify the two major groups (Sg1 and Sg2–14), as well as individual serogroups. Although this technique gives rapid results and is relatively inexpensive, it nevertheless presents problems due to cross-reactions between the antibodies.

Recently, several studies demonstrated the advantages obtained by mass-spectroscopy techniques on bacteria classifications as well as the discriminatory power at species or strain levels (Li et al., 2019; Cordovana et al., 2021).

Among them, other than the widely recognized application of MALDI-TOF MS, the FTIR spectroscopy is being developed in routine species identification, but in the field of Legionella surveillance only few studies have been reported (Helm and Naumann, 1995).

Scientific literature have already shown how FTIR technique was able to discriminate and identify both Gram-positive and Gram-negative bacteria. The authors demonstrated its application on Listeria monocytogenes, Streptococcus pneumoniae, Salmonella spp, Bifidobacterium spp., and characterization both in clinical and food fields showing the potentiality of the technique as well as its limits (Helm et al., 1991b; Rebuffo-Scheer et al., 2007; Maity et al., 2013; Lasch et al., 2018; Deidda et al., 2021). Although Horbach et al. (1988) already demonstrated the application of FT-IR on Legionella, characterizing clinical strains involved in three cases of simultaneous infections, the application of FTIR on Legionella classification is poorly documented.

In this study, for the first time, the results of FT-IR technique using reference and environmental strains were compared to obtain reliable and standardized methods to quickly classify and correlate the isolates. Moreover, a new dataset with a large number of isolates was built to provide a new scheme of Lp serogroups classification that overcame the limit of software classifier provided by IR Biotyper®.

The results of this study show that FTIR technology can be used to correctly discriminate the Sg1, the most clinically relevant one, from the other Lp serogroups with high accuracy, showing advantages in terms of time and costs. The entire analysis requires about 2 hfor a 96-well plate, considering the time from beginning of sample preparation to results delivery. The interpretation of results is simple, quick, and non-operator dependent, and the storage of all data in the IR Biotyper® software allows the traceability of results that permits to reanalyze them in different time or match old results with new ones acquired over time. Moreover, the IR Biotyper® sensibility and specify reduce the laboratory costs compared to the traditional agglutination test.

Regarding the identification of other serogroups, the main isolates were classified inside the range 2–15 with moderate reliability (yellow score), and only four of them were classified as not reliable (red score). Despite the red score, the results obtained could be considered correct because our dataset was represented by isolates that had been already serogrouped with agglutination tests, and the single serogroup was already established. Also only one exception concerning two Sg7 isolates was observed, including the Lp ATCC strain, that was miss-classified as Sg1. Nevertheless, the Sg7 isolates were clearly separated from the other serogroups, when analyzed by PCA/LDA, and this apparent discrepancy could be explained with the assumption that the training set used to develop the classifier included no Lp Sg7 isolates, therefore the ANN could not possibly to learn to recognize the distinctive features of this serogroup, and consequently it was not able to recognize them when they were exhibited by unknown isolates.

Moreover, the cut-off value fixed by manufacturer classifier has been built with a limited dataset of isolates identified only between the main two groups: Sg1 vs. Sg2–15, without a specific serogrouping identification for the second one. This limit could explain the low score obtained during some identification, and the mis-classification returned for Sg7. Further and more focused investigation would be necessary to evaluate the potential of this technique for a discrimination of the other serogroups, especially for the more pathogenic ones like Lp Sg3 and Sg6. An extension of dataset with biological replicates grown on different culture conditions and coming from different geographical area could contribute to amplify the classifier with more features to set up a new cut-off, increasing the discrimination power of the instrument.

Similarity analysis by LDA delivered results coherent with results of the classifier, showing for the environmental strains a clear separation of Sg1, Sg7, and Sg11 from the other serogroups. By contrast, the ATCC strains appear well separated from each other, without any overlapping, while among the environmental isolates only 3 serogroups could be distinguished.

It is known that the ATCC strains are reference isolates that have been deposited in a collection and remain preserved. Therefore, compared to environmental strains, they do not undergo environmental stresses like the presence of disinfectants, microbial competition, and water temperature and flow that could affect their biochemical membrane components. Moreover, regarding the analysis performed, the different separations could be explained with the fact that each serogroup of the ATCC dataset is represented by one single isolate, tested as biological triplicate, while in the environmental dataset, each environmental serogroup is represented by more isolates. Therefore, the inter-serogroup variance is higher (as each isolate brings to the group its own variance) and the spectral area of the different serogroups may overlap, making them indistinguishable. Further investigations, using more isolates for each serogroup, a higher number of discriminative features (PCs and LDs), and a multi-step classification approach (e.g., a first differentiation in 1–7–11 vs. the rest, and a second step focused on “the rest”) could help in the separation of the other Sgs.

The HCA delivered promising preliminary results regarding correlation between the serogrouping already separated (e.g., Sg1, 7, and 11). The results clearly showed how isolates that come from several facilities and collected in different years were correlated with each other and regrouped in the same clade. These results should be confirmed and compared by molecular reference methods.

Our results support the proposal to use the IR Biotyper® system as an initial screening to distinguish the Lp serogroups isolates during clinical and environmental surveillance. Moreover, its association with MALDI-TOF MS, due to easy sample preparation and a rapid returned result, could improve the microbial identification and epidemiology investigation.

Currently the IR Biotyper® is able to distinguish only Lp Sg1 from all the other serogroups but, as demonstrated by our results, with greater accuracy. This is in accordance with the National and International Guidelines that recommended during the Legionella routine monitoring only the discrimination between Lp Sg1 and Lp Sg 2–15 (Italian National Institute of Health, 2015).

It is important to underline that the IR Biotyper® classifier is built with Lp Sgs isolates coming from Austria and Germany, while all our Lp Sgs have been isolated from facilities distributed on the Italian territory. Despite the different origin and year of isolation of the strains, it seems that the isolate's geographical origin did not affect the instrument's discriminatory power since our isolates were well identified and clustered.

This study opens up the possibility to use this technique to recognize and eventually correlate also isolates coming from different countries, underlying the role of its application also during epidemic events or travel-associated Legionnaires' disease cases. Moreover, in a first future step it will be essential to train the instrument (e.g., biological and technical replicates) with isolates that come from different geographical areas to overcome the limit of Sgs prevalence found in our study. When the classifier instrument will be trained with several Sgs', it could be possible to build a new classifier also for other Legionella species.

A second step will be the development of a new classifier starting from other isolates' features. Considering that the differentiation into 15 serogroups is given by different fatty acid components, it will be interesting to focus new studies on the evaluation of lipid profile adsorption. The IR Biotyper® is already able to perform it on other microorganisms, characterizing several differences on fatty acid and lipidic profile that are able to differentiate them also at species level (e.g., for Mycobacteria; Solntceva et al., 2021).

The limitation of the study could be attributed to the low numbers of isolates of some serogroups (e.g., Lp Sg7) with respect to the one most abundant (e.g., Lp Sg1), although this reflects the real prevalence of the isolates in the sampled area.

Moreover, an expertise is required to develop a new classifier, as we have tried to do in our study, to separate the isolates within the two main groups already defined by the classifier. This limit is exceeded when the user applied only the classifier present in the IR Biotyper® system that discriminates between Lp Sg1 and Lp 2–15.

In conclusion, the evaluation of applicability of the IR Biotyper® system is ongoing, and a prospective could be a multicentric study to test the method of reproducibility across different countries.

The findings of this study represent a starting point for the introduction of this technology for the rapid identification of Lp strains not only for diagnostic purposes and environmental surveillance, but mainly during outbreaks and clusters. By this promising and innovative tool both the implementation of control measures, risk assessment analysis, and the management of water distribution systems could be improved.

The original contributions presented in the study are included in the article/supplementary materials, further inquiries can be directed to the corresponding author/s.

SC and MRP conceived and designed the experiments, wrote the article, and performed data interpretation. MC performed data interpretation and provided all the information about the instrument classifier. MM, LG, and FM performed sample collection and the culture experiments. MRP, FB, and PDM performed the IR-Biotyper® analysis. MLR and MS supplied the communities of ATCC Lp strains other than to contribute to data interpretation and critical review of the manuscript. All the authors have read and agreed to the published version of the manuscript.

MC was employed by Bruker Daltonic GmbH (the manufacturer of IR-Biotyper®). The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.