In Cambodia, spontaneous fermentation or back-slopping is commonly used to produce fermented foods for preservation and storage as well as maintenance of the food supply (Ly D. et al., 2018). However, the growth of the involved and uncharacterized microbiota, both safe and unsafe microorganisms is completely uncontrolled and unpredictable, resulting in less uniform sensory food characteristics and compositions (Radulović et al., 2011). With rapid urbanization and increasing exports, the demand for fermented foods of good quality is increasing in Cambodia. To make safe, acceptable and consistent products with a long shelf life, controlled and accelerated fermentation processes by the application of starter cultures are needed (Holzapfel, 2002; Ammor and Mayo, 2007; Corsetti et al., 2012). Research and optimization in the development of starter cultures that produce the unique taste, aroma and texture of Cambodian foods is important to meet consumers’ expectations, preserve the variability of fermented foods, and increase competitiveness in local and international markets (Ly et al., 2019). Until now, however, only few studies have been performed on the role of microbiota in Cambodian food fermentation (Ly S. et al., 2018).

Lactic acid bacteria (LAB) produce lactic acid, acetic acid, ethanol, aroma compounds, exopolysaccharides, bacteriocins and enzymes. Resultingly, they improve the texture, increase the shelf-life and microbial safety, and contribute to a pleasant sensory profile of the final products (Leroy et al., 2006). To assure high metabolic performance as well as microbial dominance against possible undesirable microorganisms, starter cultures are added in large quantities to the initial food product (10⁶–10⁷ bacterial cells per gram) (Ramesh, 2007). Some LAB species belong to those mentioned on the qualified presumption of safety (QPS) list of the European Food Safety Authority (EFSA) (Adams, 1995; EFSA, 2018a). Strains of these species that fall within a QPS group do not require a full safety assessment. Regardless of the QPS status, the presence of transferable antibiotic resistance (ABR) must be tested strain by strain (Laulund et al., 2017). The absence of phenotypic ABR in strains intended for use as starters is of course preferred, but strains with phenotypic resistance may still be suitable if the transferability of the observed ABR can be excluded by analyzing the underlying genetic mechanism of the ABR in combination with additional tests (Laulund et al., 2017). Silent, pseudo-, partial or unexpressed genes may cause false positive results, leading to discordance between phenotypic and genotypic susceptibility data. Thus, uninvestigated resistance genes and mutations; novel, unknown or unusual resistance determinants; and non-specific resistance mechanisms such as multidrug resistance transporters or genes associated with oxidative stress can also lead to inconsistencies (Campedelli et al., 2019; El Issaoui et al., 2021; Flórez et al., 2021; Stefańska et al., 2021). Another challenge in resistance determination is the unavailability of standardized cutoff values for most LAB and commonly used antibiotics (ABs) in susceptibility testing (de Castilho et al., 2019; Das et al., 2020). These standardized cutoffs should also account for intrinsic resistance (IR) in cases of inherent structural and functional features (Campedelli et al., 2019; El Issaoui et al., 2021; Stefańska et al., 2021). Since it can affect the decision to consider a bacterium susceptible or resistant, the correct determination of minimal inhibition concentration (MIC) cutoff values is important (de Castilho et al., 2019; Colautti et al., 2022). Due to the great complexity of lactobacilli, comprising most food-fermenting LAB, it has been shown that susceptibility to ABs is species dependent and varies highly among species (Anisimova and Yarullina, 2019; Stefańska et al., 2021). Thus, establishing species-specific cutoffs for lactobacilli is recommended (Ma et al., 2021; Stefańska et al., 2021).

This complexity and extreme heterogeneity of lactobacilli is also the reason for a recent significant change in the taxonomy of this genus, which consisted of 261 species at the time of change (Zheng et al., 2020). Based on a polyphasic approach, Zheng et al. (2020) reclassified the former genus Lactobacillus into 25 genera, including the genus Paralactobacillus and 23 new ones. The amended genus Lactobacillus itself currently consists of species that have been assigned to the Lactobacillus delbrueckii group (Zheng et al., 2020). The updated nomenclature of the genera lactobacilli proposed by Zheng et al. is given in the parentheses, namely Lb. acidipiscis (Ligilactobacillus acidipiscis), Lb. fermentum (Limosilactobacillus fermentum), Lb. pentosus (Lactiplantibacillus pentosus), Lb. plantarum (Lactiplantibacillus plantarum), Lb. namurensis (Levilactobacillus namurensis), Lb. zymae (Levilactobacillus zymae), Lb. futsaii (Companilactobacillus futsaii). From now on, the updated nomenclature of this study is used in reference to microbiological cutoff values from EFSA guidance.

The autochthonous microbial population of Cambodian fermented foods, a valuable source of starter cultures, was examined for the presence of LAB (Ly et al., 2020). In order to exploit strains as potential starter cultures, the assessment of 46 LAB strains must be performed to guarantee safety and avoid harmful questions. Therefore, the aim of the present study is to critically assess the safety of lactobacilli and pediococci based on the phenotypic antibiotic susceptibility to clinically relevant ABs and those used in agriculture and aquaculture.

Lactobacilli are usually susceptible to cell-wall-targeting ß-lactams (AMP) and inhibitors of protein synthesis like phenicols (CHL), macrolides (ERY), lincosamides (CLI), and tetracyclines (TET) (Anisimova and Yarullina, 2019). Except for CLI, mainly unimodal distributions of MICs below the corresponding breakpoint were found for these ABs. Therefore, most lactobacilli in this study were also susceptible to AMP, CHL, ERY and TET. The sole bimodal MIC distributions in the case of the AB–species combinations AMP– Lactiplantibacillus pentosus and TET–Levilactobacillus namurensis resulted from rare strains possessing MICs outside the unimodal MIC distribution. This could be due to the small number of strains tested per species, which should be at least 10 in order to obtain acceptable MIC ranges for the definition of WT populations (Klare et al., 2007). According to Klare et al. (2007), such “outsiders” can be ignored if their MICs are ± one dilution step away from the values of the other strains, as this is the normal standard deviation for MIC determinations. This was the case for both AB–species combinations. It is therefore questionable whether the two TET-resistant Levilactobacillus namurensis or two AMP-resistant Lactiplantibacillus pentosus strains are indeed resistant. The MICs of these strains were only one dilution step higher than the respective breakpoint. Furthermore, the WT distribution of the AMP–Lactiplantibacillus pentosus combination is split by the breakpoint, which shouldn’t be (EUCAST, 2019). However, a splitting of WT populations through breakpoints cannot be completely avoided. Due to differences in the applied susceptibility testing method as well as intra- and inter-laboratory variations when using the same method, there will never be a perfect correlation between phenotypic data and breakpoints (Turnidge, 2016). For example, although there is a standard for non-enterococcal LAB for determining MICs to ABs that uses the widely applied microdilution broth method and the specially defined LSM medium (ISO/IDF, 2010), the MRS medium is still used (Colautti et al., 2022), leading to inconsistent MIC determination in LAB (Álvarez-Cisneros and Ponce-Alquicira, 2018). Bimodal MIC distributions were also observed for CLI and Lactiplantibacillus plantarum, Lactiplantibacillus pentosus and Ligilactobacillus acidipiscis. In the case of Lactiplantibacillus plantarum and Ligilactobacillus acidipiscis, nearby outsiders may in turn be responsible for the observed profile. However, several dilution steps lie between the respective three Lactiplantibacillus pentosus strains with low and higher MICs. In addition to all three Levilactobacillus namurensis strains, two of the three Lactiplantibacillus pentosus strains with higher MICs were also CLI resistant according to the corresponding breakpoint. This suggests an AR.

All tested lactobacilli were susceptible to LNZ and RIF, which corresponds to the literature (Sirichoat et al., 2020). Even the MIC values obtained for LNZ and RIF in this study fall within the range of the four or five most frequently received LNZ (1–8 µg ml⁻¹) or RIF (0.12–2 µg ml⁻¹) MICs of Campedelli et al. (2019). This may be due to the use of the same method for antimicrobial susceptibility testing with the microdilution broth method, LSM medium, ISO 10932/IDF 233, and 48 h incubation, which may largely avoid variations (Turnidge, 2016). The applied LSM medium supports the growth of some Lactobacillus spp. only weakly or not at all (Mayrhofer et al., 2014; Campedelli et al., 2019). As the MICs of weakly growing strains might be falsely low, the outcomes of these strains are even more confusing than those of strains that fail to grow (Mayrhofer et al., 2014). The bi- and multimodal MIC distributions of the Ligilactobacillus acidipiscis species for more than half of the ABs tested (KAN, NEO, STR, CLI, LNZ, VAN, NIT, SMX, TMP) could be attributed to the poor growth of this species. The subpopulations of this species with higher MIC values, which are mostly in the same concentration ranges as those of other Lactobacillus strains classified as susceptible, could be formed by better growing Ligilactobacillus acidipiscis strains. In particular, the VAN-resistant population with three susceptible outsiders is confusing because the type-species of Ligilactobacillus genus is Ligilactobacillus salivarius, which is normally intrinsically resistant to VAN (Zhang et al., 2018). In addition to its weak growth, incorrect species identification could be another reason for the bi- and multimodal MIC distributions of the Ligilactobacillus acidipiscis species. Proper species identification is a prerequisite for appropriate data evaluation (Laulund et al., 2017). Only representatives of the Lb. acidophilus group, which belong to the OHO lactobacilli, are described as susceptible to VAN (Hamilton and Shah, 1998; Adimpong et al., 2012; Abriouel et al., 2015). However, the three allegedly VAN-susceptible strains were assigned to the species Ligilactobacillus acidipiscis by both methods applied with a high level of reliability (Ly et al., 2019). This supports the assumption of bi- and multimodal distributions due to weakly growing strains. Since the two Companilactobacillus futsaii strains are OHO lactobacilli, it is assumed that they are also VAN susceptible (EFSA, 2018b). However, both strains had MICs > 128 µg ml⁻¹, greatly exceeding the microbiological VAN cutoff value of EFSA for OHO Lactobacillus (2 µg ml⁻¹) (EFSA, 2018b). Relevant susceptibility data for other OHO species of the phylogenetic Lb. alimentarius group (Companilactobacillus genus), which contains the Companilactobacillus futsaii species (Pot et al., 2014), are rare, but the related species Lb. crustorum (Companilactobacillus crustorum) and Lb. farciminis (Companilactobacillus farciminis) were also described as VAN resistant (Scheirlinck et al., 2007; Sandes et al., 2017). Moreover, Zhang et al. (2018) predicted VAN resistance for the Lb. alimentarius group (Companilactobacillus genus) members by investigating the sequences of the active site of the D-alanyl-D-alanine dipeptide ligase, which determines VAN resistance or susceptibility.

Resistance to aminoglycoside ABs is also considered to be intrinsic in Lactobacillus. The reduced susceptibility of lactobacilli to aminoglycosides originates from membrane impermeability due to the lack of cytochrome-mediated drug transport (Elkins and Mullis, 2004; Wolupeck et al., 2017). Campedelli et al. (2019) determined a higher resistance toward KAN and STR than to GEN and NEO. Within this study, the MIC values of GEN and NEO also tended towards the low end of the concentration range examined, while those of STR and KAN were distributed almost over the entire concentration range. The higher susceptibility to GEN is most likely linked to the superior ability of this AB to cross the membrane (Campedelli et al., 2019). Most bimodal distributions for these ABs may again be explained by the small number of strains tested per species and the poor growth of Ligilactobacillus acidipiscis strains. Interestingly, one Limosilactobacillus fermentum strain had higher MIC values for GEN and NEO than the other strains of the same species. Nevertheless, this strain was not classified as resistant because its MICs were lower than the corresponding breakpoints. In contrast, both Companilactobacillus futsaii strains should be KAN resistant based on the breakpoint. However, only the second Companilactobacillus futsaii strain could in fact possess AR because it had a higher KAN MIC in addition to GEN and NEO and was also resistant to STR. AR to aminoglycosides is explained by impaired uptake, efflux, target modification or enzymatic inactivation. Of these, enzymatic inactivation is generally associated with gram-positive bacteria, especially LAB (Aslangul et al., 2006; Jaimee and Halami, 2016).

Unimodal MIC distributions in the concentration range of 2–32 µg ml⁻¹ were determined for CIP. The higher MICs of lactobacilli for quinolones such as CIP are considered to arise from IR due to cell-wall impermeability or efflux mechanisms (Anisimova and Yarullina, 2019). Broad bimodal or multimodal profiles were observed for NIT. The basic action mechanism of nitrofuran ABs such as NIT is unclear (Vardanyan and Hruby, 2016). Multiple mechanisms like the inhibition of many microbial enzyme systems or damage to DNA, RNA and proteins are described (Blass, 2015; Vardanyan and Hruby, 2016). These diverse mechanisms could explain the obtained profiles, since accumulations of different numbers of adequate resistance mechanisms could lead to a broad MIC distribution (Phillips, 1998). Therefore, further studies are required to generally elucidate the mechanisms of this AB and the associated resistances. Next to NIT, MIC profiles that were difficult to evaluate were also observed for TMP and SMX. This may be due to the presence of antagonists (TMP: thymidine; SMX: p-aminobenzoic acid) in the LSM medium (Klare et al., 2007) and reading the MIC at a concentration with a growth inhibition ≥80% (ISO/IDF, 2010). Both ABs act by targeting successive steps in the folic-acid metabolism necessary for DNA synthesis (Minato et al., 2018). Lactobacilli have been described to be intrinsically resistant to both ABs (Rojo-Bezares et al., 2006; Danielsen et al., 2007; Abriouel et al., 2015). TMP resistance is supposed to be due to the presence of a TMP-insensitive target enzyme or the lack of a metabolic pathway of folic-acid synthesis (Guo et al., 2017; Wolupeck et al., 2017; Rozos et al., 2018). Because of a wide MIC range (0.125–64 µg ml⁻¹), which was also obtained by Guo et al. (2017) (0.25–64 µg ml⁻¹), resistance to TMP was not easy to recognize. Contrarily, all SMX MIC values were at the higher concentration range (64–>1,024 µg ml⁻¹). Resistance to this AB is probably due to the cell-wall structure and membrane impermeability, sometimes complemented by efflux mechanisms (Rojo-Bezares et al., 2006).

Most species of Pediococcus are described to be intrinsically resistant to aminoglycoside ABs (GEN, KAN, STR, NEO), VAN, TET, CIP and SMX with or without TMP (Rojo-Bezares et al., 2006; Danielsen et al., 2007; Hummel et al., 2007; Franz et al., 2014; Singla et al., 2018; Zommiti et al., 2018). In contrast, pediococci are mainly found to be susceptible to AMP, CHL and ERY (Singla et al., 2018). Compared to lactobacilli, however, susceptibility data are rare, meaning fewer breakpoints have been defined. Frequently, these are only given at genus level and may even be unsuitable (Ludin et al., 2018), leading to a lack of correlation between phenotypic and genotypic analyses of ABR in pediococci (Akpınar Kankaya and Tuncer, 2020). According to these breakpoints, all seven Pediococcus strains would have AR to KAN and five strains to TET, regardless of their supposed IR. As with the AMP–Lactiplantibacillus pentosus combination, the MIC distribution of the seven Pd. pentosaceus strains tested, which was revealed as unimodal, mostly over two dilutions, was split by the indicated TET breakpoint. Based on the LNZ breakpoint, all strains would also possess AR to this AB. Although another medium and incubation time were used, comparable MIC ranges (4–8 µg ml⁻¹) were obtained by Danielsen et al. (2007) (2–4 µg ml⁻¹) and Klare et al. (2007) (0.5–2 µg ml⁻¹) for this AB. It is therefore questionable whether all these strains have indeed AR to KAN, LNZ and TET. All seven Pediococcus pentosaceus strains seem to be susceptible to CHL, ERY, CLI and AMP as well as IR to GEN, STR and VAN. For ABs for which no categorization of these strains was possible due to missing breakpoints (NEO, CIP, RIF, SMX, TMP), susceptibilities similar to those of the tested lactobacilli and those mentioned in the literature were found. As with the tested lactobacilli, it was not possible to classify the Pediococcus pentosaceus strains for NIT.

Already Laulund et al. (2017) reported a lack of knowledge about normal ABR levels for individual species, which can make a proper assessment of these species difficult. Furthermore, according to EFSA, the microbiological cutoff values, which were last updated in 2018, are only a pragmatic response, that is, reviewed regularly and adjusted if necessary (EFSA, 2012). Thus, many of the species examined in the studies of Campedelli et al. (2019) and Ma et al. (2021) had MICs above the EFSA-recommended cutoffs, suggesting that these should be reviewed with consideration for the genetic basis of resistance (Campedelli et al., 2019) to better distinguish between susceptible strains and those with AR (Ma et al., 2021). The current study also indicates that with the exception of the Lb. acidophilus group, a re-evaluation would be needed for the microbiological VAN cutoffs of OHO lactobacilli as well as for the microbiological KAN, LNZ and TET cutoffs of pediococci. Compared to VAN resistance in lactobacilli, however, further investigations are necessary to determine the underlying genetic mechanism of KAN and TET resistance in Pediococcus pentosaceus as well as exclude AR for LNZ in this species.

Although the LAB investigated in this study were isolated from eight fermented fish products, two fermented vegetables and five fermented mixed products made from fish or shrimp and vegetables (Table 1), all possibly resistant lactobacilli originated from fermented vegetables or mixed products. Interestingly, the aminoglycoside AB STR is one of the primary ABs used in vegetable farming. GEN is also used in crops but is not as common (Williams-Nguyen et al., 2016). Environmental pollution could also be the reason for the emergence of lactobacilli with AR (Founou et al., 2016). Since no guidelines address animal waste disposal and irrigation, untreated or insufficiently treated AB-rich wastewater is common in developing countries (Nadimpalli et al., 2018). The susceptibility of LAB strains isolated from fermented fishery products can be explained by the use of freshwater fish or shrimp as ingredients. Captured freshwater fish make up the largest part of Cambodia’s fish supply (Baran and Gallego, 2015). Compared to aquaculture, which is in its infant stage in this country (Lang, 2015), the AB selection pressure in freshwater would be much lower. However, it is difficult to verify these possible explanations for AR in lactobacilli based on general information, as surveillance of AB use and ABR in non-humans has been relatively neglected in Cambodia and relevant data are lacking (Om and McLaws, 2016; Zellweger et al., 2017; Reed et al., 2019).

The results demonstrate that two Lactiplantibacillus pentosus, three Companilactobacillus futsaii, three Levilactobacillus namurensis and seven Pediococcus pentosaceus strains possess AR characteristics when applying only breakpoint assessment. Other results were found with one Companilactobacillus futsaii, one Limosilactobacillus fermentum, three Lactiplantibacillus pentosus and three Levilactobacillus namurensis strains that would possess AR characteristics if taking into account further available scientific knowledge. In particular, it must be considered that microbiological cutoff values published by EFSA are only a pragmatic response. They are under review regularly and modified when necessary. Since phenotypic MIC testing alone is only useful as a preliminary screening, a subsequent molecular analysis is usually required to identify the possible presence and nature of genes associated with phenotypic resistance and to differentiate between IR and AR. Therefore, detailed analysis by whole-genome sequencing is the next step in determining the suitability of these LAB strains as potential autochthonous starter cultures for the production of fermented foods that meet the dietary habits of Cambodians.