Selection and Development of Nontoxic Nonproteolytic Clostridium botulinum Surrogate Strains for Food Challenge Testing

Clostridium botulinum causes severe foodborne intoxications by producing a potent neurotoxin. Challenge studies with this pathogen are an important tool to ensure the safety of new processing techniques and newly designed or modified foods, but they are hazardous and complicated by the lack of an effective selective counting medium. Therefore, this study aimed to develop selectable nontoxic surrogate strains for group II, or nonproteolytic, C. botulinum, which are psychotropic and hence of particular concern in mildly treated, refrigerated foods. Thirty-one natural nontoxic nonproteolytic strains, 16 of which were isolated in this work, were characterized in detail, revealing that 28 strains were genomically and phenotypically indistinguishable from toxic strains. Five strains, representing the genomic and phenotypic diversity of group II C. botulinum, were selected and successfully equipped with an erythromycin (Em) resistance marker in a defective structural phage gene without altering phenotypic features. Finally, a selective medium containing Em, cycloserine (Cs), gentamicin (Gm), and lysozyme (Ly) was developed, which inhibited the background microbiota of commercial cooked ham, chicken filet, and salami, but supported spore germination and growth of the Em-resistant surrogate strains. The surrogates developed in this work are expected to facilitate food challenge studies with nonproteolytic C. botulinum for the food industry and can also provide a safe alternative for basic C. botulinum research.


Introduction
Refrigerated processed foods with extended durability (REPFEDs) are an important and rapidly growing segment in the commercial food market. Many innovations have also been adopted in this segment, such as the use of novel protein sources, salt and sugar reduction, mild and nonthermal processing, and omission or replacement of chemical preservatives by natural alternatives [1,2]. REPFEDs are commonly processed by mild heat treatment in a temperature range of 70-95 • C, eliminating vegetative bacterial cells but not all spores. Moreover, they are often packaged under vacuum or modified atmosphere, which raises concern about the growth of anaerobic pathogens such as the sporeformer Clostridium botulinum [1,3]. This pathogen is notorious for producing botulinum neurotoxins (BoNTs), which, with an oral lethal dose as low as 1 ng/kg for humans, are the most potent natural toxin known [4]. BoNTs causes flaccid paralysis by blocking the release of the neurotransmitter acetylcholine at the peripheral cholinergic nerve terminals [5].
The BoNT producing Clostridia are divided into seven genomically and physiologically distinct species (C. botulinum groups I-IV, C. sporogenes, C. baratii, and C. butyricum), and eight major BoNT serotypes (A-G, X) have been reported [6][7][8][9]. C. botulinum groups I and II are primarily associated with human foodborne botulism, an intoxication caused by the consumption of food containing preformed BoNT [10,11]. Group II comprises nonproteolytic, saccharolytic strains that can produce toxin serotypes B, E, or F, and whose growth is usually prevented by a w < 0.97 (corresponding to 5% NaCl), pH < 5.0, T < 3 • C [11,12]. The heat resistance of group II spores is moderate and strongly influenced by the presence of lysozyme. For example, one study reported D-values at 82.2 • C of 2.4 or 231 min in the absence or in the presence of lysozyme in the recovery medium, respectively [11]. This effect is explained by the fact that exogenous lysozyme can replace the endogenous cortex hydrolases, which are crucial for spore germination but are quite heat-sensitive [13]. Spore germination and BoNT production of group II C. botulinum in refrigerated foods can be controlled either by limiting the product shelf-life to <10 days [3,14], by imposing a 6-log spore reduction by heating for 10 min at 90 • C (or an equivalent time-temperature combination), or by preventing growth and toxin formation by adding preservatives, acidifying the product to pH ≤ 5, reducing the a w to < 0.97, or by a combination of heat and preservative factors [15,16].
Predictive microbiology models are useful tools to assist in the design of safe food products and in the documentation of the safety of existing foods, and several models predicting the growth and toxin production of group II C. botulinum are currently available [17][18][19]. However, these models usually do not account for the-sometimes unexpected and pronounced-effects of the food matrix, and may thus over-or underestimate the safety risk. Therefore, unless food safety can be guaranteed by design, microbiological challenge tests should be performed to validate the safety of refrigerated foods with regard to group II C. botulinum. Those tests serve to assess whether new food formulations can control growth of pathogenic bacteria throughout the product shelf life, or to validate production processes intended to kill the pathogens [20]. Briefly, products are inoculated with the pathogen, incubated, and sampled over time to monitor its growth and/or inactivation. Detailed protocols for conducting challenge testing in a reproducible way have been described, also for C. botulinum [21]. A particular recommendation is the use of a cocktail of three to five representative strains to account for the genetic and physiological strain variability existing in the species [21].
Given that large-scale challenge testing with a highly toxic pathogen like C. botulinum represents a significant safety risk, the use of so-called surrogate organisms can provide an alternative [16]. The properties of C. botulinum surrogates should ideally be identical to those of the pathogen except for toxin production [22]. The use of such surrogates allows to identify conditions that prevent outgrowth of the pathogen and will therefore strongly reduce or eliminate the need for challenge testing with toxic strains. One possible source of nontoxic surrogate strains are natural strains of C. botulinum or a closely related species lacking the BoNT genes. For group I C. botulinum, C. sporogenes PA3679 has been widely used for several decades as a surrogate for the validation of thermal processes, although recent data indicate that it has a higher spore heat resistance than the (most) toxic group I C. botulinum strains, which may lead to overprocessing [23,24]. Additionally, recent evidence indicates that C. sporogenes is genetically distinct from group I C. botulinum based on average nucleotide identity (ANI) values, multilocus sequence typing (MLST) and core genome reconstruction [7,25,26]. For group II C. botulinum, in contrast, a widely used and validated nontoxic surrogate is not available [22,23,27]. One study has evaluated the potential of three naturally nontoxic isolates to serve as group II surrogates [27]. Compared to their toxic counterparts, the strains showed similar or slightly faster growth at 10 and 7 • C and over a range of different a w and pH values and had similar or somewhat higher spore heat resistance. From these results, the authors concluded that these strains could serve as nontoxic surrogates for group II C. botulinum. However, they were poorly characterized, lacking an identification to the species level and thus having an unclear relatedness to group II C. botulinum. Consequently, even with the properties of the strains investigated in the paper being similar to those of toxic group II strains, it cannot be excluded that they are genetically unrelated and may show different growth, survival, or inactivation behavior in other conditions and in real foods. The study also did not investigate mutual antagonism between the strains-an undesired property for strains to be used in a cocktail. Finally, the strains had been identified as nontoxic based on the absence of detectable toxin in culture supernatant, but this does not formally exclude the possible presence of BoNT genes and the potential to produce toxin under other conditions. Besides natural nontoxic strains, an alternative source of surrogate strains are toxic strains that have been disarmed by inactivation or deletion of their toxin genes. Bradshaw et al. (2010) insertionally inactivated the BoNT/A gene in group I C. botulinum strain 62A with the ClosTron technology and confirmed that no phenotypic features were altered apart from toxin production [28]. Likewise, we inactivated the BoNT/E gene of the group II strain NCTC11219 with the ClosTron system and also successfully deleted the gene by allelic exchange, and the resulting nontoxic mutants had unaltered spore heat resistance and growth in unstressed and stressed conditions (NaCl, acid, and low-temperature stress) compared to the toxic parental strain [29]. Moreover, CRISPR/Cas9 technology was recently used to introduce inactivating point mutations in the bontE gene of group II strain Beluga, and the researchers were able to confirm the unchanged phenotype except for the production of biologically inactive BoNT/E [30].
Another hurdle that complicates challenge testing with C. botulinum is the lack of an effective selective medium. Botulinum selective medium (BSM), a medium containing egg yolk, cycloserine, sulfamethoxazole, and trimethoprim, was developed decades ago but does not suppress or allow to distinguish several other clostridial species and, importantly, inhibits some nonproteolytic C. botulinum strains [31,32]. The inclusion of antitoxin antibodies in the growth medium was shown to improve the detection of BoNT-producing strains by the formation of a precipitation zone around the colonies, but no data is available on the performance of these plates with complex samples containing background microbiota [33,34]. Finally, fluorescent antibodies against vegetative cell walls of C. botulinum have been used in culture, but this technique has never been adopted for use in solid media to determine C. botulinum cell counts [35].
The objective of this study was to overcome aforementioned problems by developing surrogate group II C. botulinum strains that (i) are nontoxic, (ii) have properties mimicking toxic strains, and (iii) can be easily counted on a selective plating medium. To this end, we isolated natural nontoxic group II strains, phenotypically and genotypically characterized them to select strains resembling toxic strains, inserted an erythromycin (Em) resistance gene in a defective structural prophage gene of five selected strains, and optimized a suitable selective medium for counting the surrogate strains in challenge studies. The availability of these strains and an accompanying selective medium will greatly facilitate challenge testing with group II C. botulinum in foods and thus stimulate innovation in the food industry.

Bacterial Strains, Media and Growth Conditions
Naturally nontoxic C. botulinum strains studied in this work include 16 own isolates (for isolation method, see Section 2.2), 12 isolates from the Robert Koch Institute (Berlin, Germany), three isolates previously described by Parker et al. (2015), and one isolate obtained from the German Collection of Microorganisms and Cell Cultures (DSM, Brunswick, Germany). Further, nontoxic bontE deletion mutants of C. botulinum NCTC8266 and NCTC11219 that our research group constructed previously were included for comparison in the physiological characterization of the naturally nontoxic isolates [29]. All these strains are listed in Supplementary Table S1, together with the E. coli strains used for plasmid maintenance and as conjugation donor. Media and media supplements used for growing C. botulinum and E. coli are listed in Supplementary Table S2.
All experiments described in this study were performed in a Biosafety level (BSL) 2 environment, but using BSL 3 practices when the possible presence of toxic C. botulinum strains (for environmental samples and enrichment cultures) or of botulinum toxin genes (for pure cultures) had not been excluded. Once the absence of the toxin genes had been demonstrated by whole genome sequencing, strains were handled using BSL2 procedures.
This approach was formally approved by the Service for Biosafety and Biotechnology of the Scientific Institute for Public Health (Sciensano, document number SBB 219 2014/0018).

Isolation, Identification, and Whole Genome Sequencing of Nontoxic Nonproteolytic C. botulinum
Eighty-three environmental and food samples were subjected to an in-house optimized isolation protocol for group II C. botulinum ( Figure 1). Briefly, samples were suspended in reinforced clostridial medium (RCM), subjected to a heat treatment (1 h at 67 • C) to select for sporeformers, and enriched for three days at 30 • C in an anaerobic workstation (Don Whitley A35, Don Whitley Scientific, Bingley, UK), using an atmosphere of 80% N 2 , 10% CO 2 , and 10% H 2 . Subsequently, samples were diluted and plated on RCM agar and incubated anaerobically at 30 • C for up to three days, after which morphologically different colonies were analyzed under the microscope to identify rod-shaped sporeformers. The 16S rRNA gene from selected colonies was amplified by PCR using universal primers (B27F, AGAGTTTGATCMTGGCTCAG; U1492R, GGTTACCTTGTTACGACTT) [36], the amplicons were sequenced (Macrogen Europe, Amsterdam, The Netherlands), and the results were analyzed with BLAST (https://blast.ncbi.nlm.nih.gov/Blast.cgi, accessed on 1 September 2018) and EzBiocloud (https://www.ezbiocloud.net/, accessed on 1 September 2018 [37]). Next, genomic DNA (gDNA) was isolated from all the isolates belonging to group II C. botulinum based on their 16S rRNA sequence (Table S1). In short, the cell pellet of an overnight culture in liquid RCM was treated for 1 h with PIV buffer (10 mM Tris, 1 M NaCl, pH 7.5) containing 4% formaldehyde to inhibit extracellular DNases. Afterwards, the cells were washed twice with PIV buffer to remove formaldehyde and gDNA was then extracted using a genomic DNA extraction kit (Qiagen, Hilden, Germany) and the quantity and quality of the extracted DNA was verified spectrophotometrically (mySPEC, VWR, Haasrode, Belgium; Qubit, ThermoFisher Scientific, Waltham, MA, USA) and by agarose gel electrophoresis. The gDNA was then processed through the Illumina Flex Library Prep kit (Illumina, San Diego, CA, USA) and the Nextera DNA CD Index kit (Nextera, Juno Beach, FL, USA) to prepare sequencing libraries that were run on an Illumina MiniSeq platform (Illumina, 150 bp paired-end sequencing). The reads were checked for quality, trimmed, and assembled de novo into contigs using CLC Workbench 12.0 (Qiagen). Annotation of the whole genome sequences (WGS) was performed by Rapid Annotation using Subsystems Technology (RAST Seedviewer, https://rast.nmpdr.org/, accessed on 3 January 2019) [38].

Genomic Characterization and Phylogenetic Analysis of the Isolates
Isolates whose WGS did not contain (parts of) a BoNT gene cluster were compared to the well-known toxic strains Alaska E43 (NC_010723.1) and Eklund 17B (NC_018648.1) by genome size and GC content (RAST Seedviewer), and by average nucleotide identity (ANI) (EzBiocloud). WGS from nine toxic strains available in GenBank (Table S1) were additionally included in this analysis. The phylogenetic relationship between the nontoxic and toxic strains was established by multilocus sequence analysis (MLSA) with 12 housekeeping genes (Table 1) as described previously [39]. Alleles were retrieved using a BLAST search in RAST Seedviewer with the 12 genes from strain Eklund 17B (NC_018648.1), aligned using MUSCLE (https://www.ebi.ac.uk/Tools/msa/muscle/, accessed on 1 February 2019 [40]) and concatenated per strain via FASTA alignment joiner (https://birc.au.dk/~palle/php/fabox/alignment_joiner.php, accessed on 1 February 2019). A maximum likelihood phylogenetic tree was built using MEGA7. Additionally, a phylogenetic tree was established by comparing the core genome of the isolates and a set of toxic strains via Realphy (https://realphy.unibas.ch/realphy/, accessed on 1 February 2019 [41]). Multiple sequence alignments are built from these one-to-one alignments and from these a phylogenetic tree is constructed by applying the maximum likelihood method PhyML [41]. Trees were managed using the iTol online tool (https://itol.embl.de/, accessed on 1 March 2019 [42]).

Phenotypic Characterization
(i) Stressed growth. Salt tolerance of the isolates was determined by inoculating 0.1% of an overnight RCM broth culture in 4 mL fresh RCM broth with varying total NaCl concentrations (2.5-4.5%) and incubating anaerobically at 30 • C for up to 21 days. The optical density (OD 620 nm ) was monitored daily (Ultrospec 10, Biochrom, Cambridge, UK), and cultures were considered to show growth when OD 620 exceeded 0.15. In a similar way, acid tolerance was investigated by examining growth (OD 620 > 0,15) for 21 days in RCM broth acidified with 1 M HCl (pH 4.94-5.25; measured after autoclaving). Growth at low temperature was evaluated by restreaking fresh 24-h colonies on RCM agar, incubating the plates at 12 • C, 7 • C, and 4 • C, and recording the time when single colonies > 1 mm diameter first appeared.
(ii) Heat resistance of spores. Spore crops were produced in a two-phase sporulation medium as described before [43]. In short, a culture grown at 30 • C from a single colony in 1 mL RCM broth was layered onto solid sporulation medium (30 mL deionized water, 3 g cooked meat medium (Thermofisher Scientific), 0.45 g agar (Neogen) and 0.03 g glucose (Thermofisher Scientific)) in a flask, together with 4 mL deoxygenized deionized water (liquid phase). After six days of anaerobic incubation, the spores were collected and washed by four cycles of centrifugation (18,000× g, 4 • C, 15 min) and resuspension with saline (8.5 g/L NaCl), and stored in 1 mL of saline at 4 • C. Heat resistance was measured by placing 50 µL of the suspensions in a microcentrifuge tube in a heating block at 75 • C or 85 • C for different times (1-40 min), immediately cooling them on ice, and determining plate counts on RCM agar. For the spores heated at 85 • C, the plating medium was supplemented with lysozyme (10 µg/mL) to rescue the germination mechanism of heatdamaged spores [13]. Since the first part of the inactivation curves often deviated from log-linear kinetics, decimal reduction values (D-value) were calculated from the log-linear second part of the curves, which was manually determined and included minimally four data points.
(iii) Mutual antagonism. Pairwise antagonism between the isolates was assessed using a spot halo assay. Four hundred µL of an overnight culture of every strain in RCM broth was spread on RCM agar and 5 µL of every other strain was spotted on the indicator lawn.
After incubation for 24 h at 30 • C, plates were checked for inhibition halos, which reflect the production of an antimicrobial compound by the spotted strain against the indicator strain.

Construction of Erythromycin-Resistant Strains
The ClosTron plasmid constructs -pMTL007C-E2:XkdK-471s, pMTL007C-E2:SPP1-105s and pMTL007C-E2:TTMP523s-, harboring an intron targeted to specific genes of defective prophages (XkdK and SPP1 of phage phiCT453A, and TTMP of phage phiCT19406A) were designed via the ClosTron design tool (http://clostron.com/clostron2.php, accessed on 1 September 2019 [44]), transformed via electroporation into E. coli DH5α for maintenance, and into E. coli S17 λpir for conjugation to C. botulinum. The sequence of the target genes and of the retargeted intron are provided in Table S3. E. coli S17 λpir carrying a ClosTron plasmid construct (pMTL007C-E2:XkdK-471s, pMTL007C-E2:SPP1-105s or pMTL007C-E2:TTMP-523s) was grown overnight in lysogeny broth (LB) with chloramphenicol (Cm), washed twice by centrifugation (6700× g, 5 min, 4 • C). Subsequently, the pellet was brought into the anaerobic cabinet and resuspended in 200 µL of an overnight culture of the recipient nontoxic C. botulinum isolate in RCM broth. This mixture was spread onto a 0.45 µm filter (Sigma Aldrich, Saint-Louis, MO, USA) on RCM agar. After 24 h of anaerobic incubation at 30 • C, the cells were recovered in 1 mL saline, and the mixture was plated on RCM agar with cycloserine (Cs) and thiamphenicol (Tm) to counterselect the E. coli donor and to select for conjugants bearing the ClosTron plasmid, respectively. Colonies were purified and the presence of the plasmid was confirmed by PCR with primers for the catP gene (Table 2). Purified conjugants were subsequently streaked onto RCM agar with erythromycin (Em, 5 µg/mL) to select for colonies in which the ClosTron intron and the embedded retrotransposition-activated marker (RAM) based on the ErmB gene, had been inserted into the genome. After purification of the resulting colonies, correct insertion of the intron was confirmed by colony PCR and sequencing of the Xkdk, SPP1 or TTMP target genes (Table 2). Finally, selected colonies were verified to have lost the ClosTron plasmid by loss of Tm resistance and by PCR targeting the catP gene. WGS analysis of the mutants was conducted to confirm the absence of adventitious mutations. To study the stability of the introduced Em marker, overnight cultures of the Emresistant mutants, in triplicate, were re-inoculated (1:1000) into fresh RCM broth every 24 h for 7 days, after which a plate count on RCM agar with and without Em (5 µg/mL) was conducted.

Selective Medium for C. botulinum Surrogate Strains
A selective medium for food challenge studies with the constructed Em-resistant C. botulinum surrogates was composed by supplementing RCM with different combinations of antibiotics, and with lysozyme to stimulate spore germination. The effectiveness of the media to suppress background microbiota was tested with three commercial meat products (Table 3). From the 83 collected food and environmental samples, 16 strictly anaerobic sporeforming isolates clustering with toxic group II C. botulinum strains based on 16S rRNA sequence similarity were isolated. The whole genome sequences (WGS) of these isolates and of 15 isolates from other sources (Table S1) were determined and annotated, and all were found to lack a BoNT toxin gene cluster. The WGS were then compared to 11 publicly available WGS from toxic group II strains (Table 4).  ANI analysis shows that 26 nontoxic strains share more similarity (>97%) with toxin type B strain Eklund 17B than with toxin type E strain Alaska E43 (<94%), while the opposite is true for two strains (ZBS3 and DSM1985). Eklund 17B and Alaska E43 were chosen because they are representative for the previously identified type BEF and type E clusters of group II C. botulinum, respectively [7,39,45]. Furthermore, three strains (ZBS2, ZBS15, and CMCC3677) show low similarity (<93%) with either of the toxic strains, but also with other C. botulinum groups or closely related species such as C. perfringens, C. butyricum, and C. baratii (data not shown). The genome size and GC content of all 31 nontoxic strains are in the same range.
Phylogenetic analysis of the 31 nontoxic strains and 11 toxic strains (WGS from NCBI), based on 12 household genes ( Figure 2A) and on the core genomes ( Figure 2B), confirmed the existence of the previously identified type E and type BEF cluster. The BEF cluster itself consists of two subclusters, one containing toxin type B4 and F6 strains as well as nontoxic strains, and the other lineage containing toxin type B4 and E and nontoxic strains [45]. The presence of nontoxic and toxic strains in all these clusters and subclusters indicates that nontoxic and toxic group II C. botulinum are phylogenetically indistinguishable (Figure 2A,B). The phylograms based on housekeeping gene analysis and core genome analysis also both show two BEF subclusters, similar to the findings of Brunt et al. (2020b), but the strains are not always consistently mapped to these subclusters. For example, strain ZS1 is in subcluster one based on the housekeeping gene phylogram, and in subcluster two based on the core genome phylogram. The phylograms also confirm the finding from ANI analysis that strains ZBS2, ZBS15, and CMCC3677 constitute a separate lineage that is unrelated to group II C. botulinum. CMCC3677, together with CMCC3676 and CMCC3678, was previously proposed to be a suitable nontoxic surrogate strain for group II C. botulinum based on phenotypic analysis, but our genomic analysis clearly isolates it from the two other strains. This makes it conceivable that this strain may possess phenotypic traits that have not been revealed but may render it unsuitable as a surrogate for food challenge studies, thus underscoring the importance of genomic analysis for identifying nontoxic surrogate strains for food challenge studies.

Growth in High Salt, at Low pH and at Low Temperature
Growth of the nontoxic group II C. botulinum strains under salt, acid, and cold stress was determined in comparison to two toxic strains in which the toxin gene had been deleted and replaced by an Em resistance marker (NCTC8266∆bont::ermB and NCTC11219∆bont::ermB).
The maximum NaCl concentration allowing growth under the applied test conditions ranged from 2.5 to 4.0%, depending on the strain, but also with large differences in the time to growth (OD 620 > 0.15) ( Table 5). The most salt-tolerant strain (ZBS3) was able to grow at 4% NaCl in 2 days but was nevertheless not able to grow at 4.5% NaCl. In acidified RCM broth, growth of all the strains except NCTC11219∆bont::ermB was inhibited at pH 4.94, and the minimum pH supporting growth was 5.05 or 5.12 depending on the strain (Table 6). Finally, the temperature limiting growth was 4 • C or 7 • C depending on the strain (Table 7). Growth at 4 • C was always slow, with 20-21 days required to observe 1 mm colonies. The observed pH and temperature limits of the nontoxic strains generally correspond to the range described in the literature for toxic group II strains. On the other hand, the maximum salt tolerance found in our work appears to be lower than the 5% maximum found in some reports [1,11]. However, some studies also report lower experimental maximum values, and these differences may relate to the different incubation times used [16,46]. Possibly, extending the incubation time could result in higher salt tolerance values for our strains as well.   Strains from the two group II genomic subclusters showed no notable differences in acid, salt and cold tolerance. In contrast, strains from the non-group II cluster-ZBS2, ZBS15, and CMCC3677-showed higher salt and lower cold tolerance than most genuine group II strains, supporting the hypothesis that this group is not only genetically but also phenotypically distinct from group II C. botulinum. In a previous genomic and physiological study of nonproteolytic strains, the fermentation of different carbohydrates was found to be cluster-specific. Strains from the BEF cluster produced acid from amylopectin, amylose, and glycogen, but not from melezitose, while cluster E strains only fermented the latter [12]. Therefore, it is recommended to select strains from both the type E cluster and the type BEF cluster for challenge studies, since different characteristics may result in different growth patterns in food products [10].
Besides strain-to-strain variation, considerable variation between replicate cultures of the same strain was sometimes observed in these tests ( Figures S1-S3). This variation may be due to the large variation in lag phase between individual cells that exists particularly in stressed growth conditions, and that is believed to be advantageous to ensure population survival [47].

Spore Heat Resistance
It has been well documented that the apparent heat resistance of group II C. botulinum spores is strongly increased when lysozyme is present in the recovery medium [13,43]. This is because the endogenous spore cortex hydrolases, which are essential for spore germination, are very heat-sensitive, but can be replaced by exogenous hydrolases. Therefore, spore heat resistance was analyzed at 75 • C in the absence, and at 85 • C in the presence of lysozyme.
At 75 • C, many inactivation curves showed a shoulder or even an activation phase that suggests the presence of a superdormant spore fraction, as has been previously observed for spores of several bacilli and clostridia ( Figure S4) [48]. Superdormant fractions comprised up to 99% of the spore population in some strains (KI2, ZBS17, RO132). The D 75 values, calculated from the log-linear part of the inactivation curves, showed a broad variation, ranging from 1.8 to 21.4 min (Figure 3), but the average D 75 (7.4 min) corresponded well to the value reported in a meta-analysis of literature data for toxic group II strains (5.1 min) [49]. Furthermore, the four strains of the type E cluster included in our experiment (ZBS3, DSM1985, NCTC8266 ∆bont::ermB and NCTC11219 ∆bont::ermB) had the lowest spore heat resistance of all the strains in the absence of lysozyme (D 75 < 3 min), confirming a similar finding from the same meta-analysis. At 85 • C, the addition of lysozyme resulted in biphasic reduction curves with a rapid inactivation phase followed by a slower phase, probably because the spore coat in a fraction of the spore population is impermeable to lysozyme and this fraction can therefore not be rescued [50]. D-values at 85 • C were therefore determined from the heat-resistant (lysozyme-rescued) spore fraction only. The average D-value of the nontoxic isolates at 85 • C in the presence of lysozyme was 19.8 min (range from 9.1 to 40.7 min), which is somewhat lower than the corresponding value obtained for toxic strains under similar conditions (33.1 min) obtained in the meta-analysis of Wachnicka et al. (2016). The reason for this difference is not clear but may be related to the wide variation in experimental conditions underlying the data in the meta-analysis, which included studies conducted in food matrices and using different concentrations of lysozyme in the recovery medium. Interestingly, while the spores of the type E cluster strains were the most heat sensitive in the absence of lysozyme, their average heat resistance in the presence of lysozyme (average D 85 = 27.8) was higher than that of the type BEF cluster strains (average D 85 = 18.7). This may indicate that type E cluster strains have more heat-sensitive cortex hydrolases.
Finally, two of the three strains from the separate non-group II phylogenetic cluster (ZBS15, CMCC3677, but not ZBS2) showed high spore heat resistance both at 75 • C and with lysozyme at 85 • C.

Mutual Antagonism between Nontoxic Isolates
Since it is the intention to combine selected nontoxic C. botulinum strains into multistrain cocktails for conducting food challenge tests, it is important that they do not inhibit each other by the production of antimicrobial compounds. Therefore, the pairwise mutual inhibition of all the strains was systematically analyzed in a plate halo assay ( Table 8). The majority of strains (23) produced no or only very small halos against all the other strains. Very small halos were neglected because they were considered more likely to be the result of nutrient competition than from the production of specific antimicrobial substances in this assay. Eight strains produced a halo of moderate size against at least one other strain. Here, production of an antimicrobial compound cannot be excluded but should be further confirmed. Finally, two strains produced a large halo against at least one other strain. ZBS20 produced a large halo against one and a moderate halo against seven other strains, and DSM1985 produced a large halo against 24 and a moderate halo against eight strains. The strong antagonism of DSM1985 (also designated as strain S5 in other studies) against other nonproteolytic C. botulinum strains has been reported before, and production of inhibitory substances has been documented also in other strains of both nonproteolytic and proteolytic C. botulinum [51][52][53]. It should be noted that the absence of detectable antimicrobial activity in the halo assay does not exclude the possibility that a strain may still produce such activity in other conditions, but this was not further investigated in our work.

Selection of Strains for Challenge Testing and Introduction of an Erythromycin Resistance Marker
The data obtained from the genomic and phenotypical characterization were used to select a set of five strains for inclusion in a cocktail deemed to be suitable for food challenge studies. Since challenge studies are aimed at identifying potential microbiological risks, it is important that the selection includes at least one strain that has a high tolerance to growth inhibition by low pH, low temperature, and high salt concentration, respectively, and one strain with highly heat-resistant spores. Furthermore, both phylogenetic lineages should be represented because these may differ in some growth or survival-related properties that were not tested in this work but that may be relevant in specific foods. Finally, the selected strains should not show cross-inhibition. Based on these considerations, five strains were selected (Table 9). ZBS3 was selected to represent the type E cluster, since the only other nontoxic strain from this cluster (DSM1985) was antagonistic to most other strains. ME22 and VAP51 were included for their good growth at low temperature, ZBS4 for its salt tolerance, and CH2 for its tolerance to low pH and the heat resistance of its spores. To facilitate selective counting of the selected strains in food challenge tests, an Em resistance gene was inserted in their chromosome using the ClosTron system [44]. Insertions were targeted to structural genes of defective prophages, since it was anticipated that this would be unlikely to modify any relevant phenotype. Since the five strains did not share a common defective prophage structural gene, three different targets were used: gene XkdK from the non-intact phage phiCT453A for strains ME22, VAP51 and ZBS4, gene SPP1 from the same phage for strain ZBS3, and gene TTMP from prophage phiCT19406A for strain CH2. After insertion of the ClosTron intron in these phage genes, Em-resistant mutants that had lost the Clostron delivery plasmid were identified by PCR. Loss of this plasmid is important since this precludes the LtrA-mediated splicing of the intron from the target sequence and is thus expected to enhance the stability of Em resistance. In addition, WGS analysis confirmed that the Em-resistant mutants had not picked up any adventitious mutations. Finally, the growth under salt, acid, and cold stress and the spore heat resistance of the Em-resistant mutants was re-evaluated and found to be indistinguishable from their respective parental strains (data not shown). Based on these data, it was concluded that well-targeted ClosTron insertion can be used for generating antibiotic-resistant C. botulinum strains without altering important phenotypical features.

Stability of Erythromycin Resistance
Intron integrants such as ClosTron mutants are intrinsically stable [44]. However, the prophage regions into which the introns were targeted in this work, despite being predicted to be defective, could potentially be unstable. The stability of the introduced Em resistance marker was therefore determined by growing three independent cultures of the five constructed ClosTron mutants for approximately 70 generations without Em, and subsequent plate counting on medium with and without Em. No significant differences in the plate counts on the two media were found for any of the cultures, indicating that the introns carrying the Em resistance marker are stable in all the mutants (data not shown).

Selective Medium for Challenge Studies
The insertion of an Em resistance marker is expected to facilitate the counting of the modified nontoxic C. botulinum strains in food challenge experiments because it allows to include Em in the plating medium to increase its selectiveness, in particular against other clostridial species which are naturally susceptible to Em [54][55][56]. Here, a selective medium for use with the Em-resistant challenge strains is composed and its effectiveness is tested.
In a first experiment, different selective media containing Em were evaluated to compare their effectiveness to suppress the background microbiota of cooked ham, chicken filet, and salami, which amounted to 6.5 ± 0.1 log cfu/g, 3.3 ± 0.9 log cfu/g and 7.0 ± 0.04 log cfu/g, respectively, on RCM agar upon anaerobic incubation. Em was added at 3 µg/mL, a concentration at which the Em-resistant ClosTron mutants still showed 100% plating efficiency. Since this concentration of Em in the RCM plating medium only partially inhibited the endogenous microbiota (Figure 4), Cs and Gm were used as additional antibiotics to identify a combination that was more effective against the background (Figure 4). Use of these antibiotics to select and isolate C. botulinum has been previously proposed, based on the natural levels of resistance of this organism against Cs (> 256 µg/mL) and Gm (87% of the tested strains was resistant to 16 µg/mL, 95% to 8 µg/mL) [55][56][57]. Suitable concentrations of Cs and Gm that did not reduce the plating efficiency of the five ClosTron mutants were determined to be 100 µg/mL and 15 µg/mL, respectively. When used alone, Cs and Gm were generally less effective than Em to suppress the background microbiota of the three tested meat products. Pairwise combinations of antibiotics were more suppressive, but none of the combinations was sufficiently effective on all food products, and particularly the background of the cooked ham was insufficiently inhibited. However, the combination of all three antibiotics completely eliminated the background of all three products, except for one colony in one of the three replicates of cooked ham. In conclusion, RCM + Em + Cs + Gm is very effective to suppress the endogenous microbiota of these cooked meat products, even at high levels ( Figure 4).
In a second experiment, the plating efficiency of the five Clostron mutant strains on RCM + Em + Cs + Gm compared to RCM without antibiotics was evaluated. This was done for each strain separately and with the five-strain cocktail, and both with spore suspensions and suspensions of vegetative cells ( Figure 5). The plating efficiency of the vegetative cells on the non-selective and selective medium was indistinguishable for each of the strains and the cocktail. In contrast, counts of the spore suspensions were 0.5-1.7 log lower on the selective medium, depending on the strain, although the differences were not always significant at the 5% level. Prolonged incubation did not increase the counts on the selective medium. The lower plating efficiency specifically of the spores suggests that the antibiotic cocktail somehow interferes with spore germination, and therefore the selective medium was additionally supplemented with lysozyme in an attempt to stimulate spore germination. The results in Figure 5 show that lysozyme indeed increased the plating efficiency of the spore suspensions for all the strains. For three strains, the plating efficiency was restored to the level of the non-selective medium, for one strain it was still slightly lower (0.4 log), and for one strain it was even higher, which might be due to a superdormant fraction that does not germinate even on RCM without the help of exogenous lysozyme   ClosTron mutants, separately and together in a cocktail, on RCM agar, RCM agar + Em (3 µg/mL) + Cs (100 µg/mL) + Gm (15 µg/mL), and RCM agar with the three antibiotics and lysozyme (lys, 50 µg/mL). Error bars represent standard deviations of three independent overnight cultures/spore suspensions. Significant differences (*, p < 0.05) between counts on the selective media and RCM agar were calculated using a one-way ANOVA and a paired t-test.

Conclusions
This work reports a detailed phenotypic and genomic analysis of 31 nontoxic nonproteolytic C. botulinum strains with the aim of developing a set of surrogate strains for challenge testing in foods. The genomic analysis based on the strains' WGS revealed three lineages; the previously described type E and type BEF clusters of toxic nonproteolytic strains, and a novel, quite distinct and yet uncharacterized cluster. The strains' capacity to grow under salt, acid, and low temperature stress, and their spore heat resistance were determined and shown to fall in the same range as described for toxic nonproteolytic strains. Together, these results show that, apart from the presence of a BoNT toxin gene cluster, nontoxic, and toxic nonproteolytic Group II C. botulinum strains are genetically and phenotypically indistinguishable. Five nontoxic strains, encompassing the natural genomic and phenotypic diversity present in group II C. botulinum, were equipped with a genomic Em resistance marker, and a selective medium that suppresses the background microbiota of three commercial meat products and allows quantitative recovery of the Em-resistant surrogate strains, was developed. Although they still need to be validated in full-scale challenge studies, the surrogate strains and the accompanying selective medium developed in this work can be anticipated to make food challenge testing with nonproteolytic C. botulinum safer, easier, and more reliable, and thus more accessible to the food industry.
It is important to note that toxic strains may still need to be used, because guidelines in some countries require neurotoxin-based challenge testing for C. botulinum. However, even in such cases, prior testing with the surrogates will allow to reduce the design and cost of challenge testing with toxic strains. Finally, the availability of a collection of nontoxic strains that is otherwise indistinguishable from toxic strains also paves the way for use of these strains as a safe alternative in basic research on nonproteolytic C. botulinum, and this will hopefully promote a better understanding of the genetics, physiology, and ecology of these interesting bacteria.
Supplementary Materials: The following supporting information can be downloaded at: https: //www.mdpi.com/article/10.3390/foods11111577/s1, Table S1: C. botulinum and E. coli strains and whole genome sequences used in this work. * Strains harboring the toxin cluster, Table S2: Growth media and medium supplements used in this work, Table S3: Sequences of the target genes for ClosTron intron insertion, and sequences of the retargeted introns. The introns insert between the red nucleotides indicated in bold, Figure S1: Salt tolerance of different nontoxic isolates, NCTC8266∆bont::ermB and NCTC11219∆bont::ermB in liquid RCM medium supplemented with different amounts of NaCl. Growth at 30 • C in 2.5%, 3.0%, 3.5%, 4.0% and 4.5% NaCl was studied for 21 days for three replicates per strain. Growth of all replicates is indicated. None of the strains grew at 4.5% NaCl, Figure S2: Acid tolerance of different nontoxic isolates, NCTC8266∆bont::ermB and NCTC11219∆bont::ermB in liquid RCM medium supplemented with 1 M HCl (pH values; 5.25, 5.12, 5.05 and 4.95). Growth was studied for 21 days for three replicates per strain. Growth of all replicates is indicated, Figure S3: Low temperature tolerance of different nontoxic isolates, NCTC8266∆bont::ermB and NCTC11219∆bont::ermB on solid RCM medium incubated at different temperatures (7 • C, 10 • C and 12 • C). Growth was studied for 21 days for three replicates per strain. Growth of all replicates is indicated, Figure S4: Reduction curves for heat inactivation of all nontoxic isolates, NCTC8266∆bont::ermB and NCTC11219∆bont::ermB. The recovery medium for spores treated at 85 • C was supplemented with lysozyme (10 µg/mL). D-values of biphasic curves were calculated from the second part of the curve, taking the circled data point as the start. Average values ± STDEV from three replicates are shown.