This study’s primers, plasmids, gBlocks and gRNA sequences are listed in Supplementary Tables S1−S4. Oligonucleotides and gBlocks were ordered from Integrated DNA Technologies (IDT). To generate the URA3 ^S81X (SbU⁻), HIS3 ^G26X (SbH⁻), and TRP1 ^P12X (SbT⁻) disrupted strains, gBlocks with respectively gRNA were assembled with pCfB2312 (Cas9-KanMX) and transformed with their respective donor primer (Supplementary Tables S1). To generate the thi2∆, thi6∆, snz1∆, sno1∆, rei1∆, and bts1∆ strains, gBlocks from respectively gRNA were assembled with pCfB3050 (pCas9-URA3) and transformed with respectively donor primer (Supplementary Table S1). The GFP, RFP and Exendin-4 integration plasmid (Supplementary Table S2) was generated by assembly of respective gBlocks (Supplementary Table S3) with pCfB2909 (marker free) and transformed together with pCfB6920.

All plasmid assemblies were conducted with Gibson Assembly (Gibson et al., 2009) and transformed into One Shot^® TOP10 Escherichia coli (Thermo Fisher Scientific). All E. coli were grown in lysogeny broth (LB) media containing 5 g/L yeast extract, 10 g/L tryptone and 10 g/L NaCl; (Sigma Aldrich) supplemented with 100 mg/L ampicillin sodium salt (Sigma Aldrich). LB agar plates contained 1% agar (Sigma Aldrich).

S. boulardii (S. cerevisiae ATCC^® MYA796™) was obtained from American Type Culture Collection (ATCC). The strains created in this study are listed in Supplementary Tables S5. S. boulardii transformations were performed via high-efficiency yeast transformation using the LiAc/SS carrier DNA/PEG method (Gietz and Woods, 2006). Genomic integrations cassettes were digested with the restriction enzyme NotI (FastDgiest Enzyme, Thermo Scientific™) prior to transformation and transformed together with various helper plasmids and pre-expressed Cas9 from pCfB2312. All transformations were incubated at 30°C for 30 min and then heat-shocked in a water bath at 42°C for 60 min. All transformations followed a recovery step. The transformation tubes were microcentrifuge for 2 min at 3,000 g. Pellets were resuspended in 500 µL of YPD media containing 10 g/L yeast extract, 20 g/L casein peptone and 20 g/L glucose (Sigma Aldrich) and incubated for 2–3 h at 30 °C before being plated. All yeast transformations were plated on synthetic complete (SC) plates containing 1.7 g/L yeast nitrogen base without amino acids and ammonium sulphate (Sigma Aldrich), 1 g/L monosodium glutamate (Sigma Aldrich), 1.92 g/L Yeast Synthetic Drop-out Medium Supplements without uracil (Sigma Aldrich) and 200 mg/L geneticin (G418; Sigma Aldrich) at 37°C. Colony-PCR using OneTaq (Thermo Scientific™) confirmed the genomic integration, gene disruption and knockout. Primers flanking the null mutant gene or integration site were used to confirm successful strain modification. Genomic DNA was extracted by boiling cells at 70°C for 30 min in 20 mM NaOH. The strains were cured for pCfB2312 and helper plasmids after genome integration. One single mutant from each generated strain was cryopreserved at −80°C and used for further characterisations. All strain replicates presented are technical replicates.

All cultivation was started from a pre-culture, inoculated with a recently streaked out colony from a −80°C cryostock, cultivated for 12–16 h. All pre-cultures were conducted in the same media as the following cultivation experiment unless other stated. Pre-culture to determine the auxotrophic strains was done in media supplemented with the required nutrition and washed three times in 1% PBS to ensure no traces of the required nutrition were transferred into the new cultures. Pre-culture to determine cold sensitivity was conducted at 37°C to ensure enough biomass.

Real-time OD₆₀₀ measurements were obtained every 10 min for 48–120 h with microplate reader Synergy™ H1 (BioTek) for aerobic and microaerobic conditions. Epoch 2 microplate reader (BioTek) was used for anaerobic real-timer OD₆₀₀ measurements. All cultures had an initial OD₆₀₀ of 0.05. The cultures were incubated into 200 µL minimal synthetic complete media (DELFT) containing 7.5 g/L (NH₄)₂SO₄, 14.4 g/L KH₂PO₄, 0.5 g/L MgSO₄ x 7H₂O, 20 g/L glucose, 2 ml/L trace metals solution, and 1 ml/L vitamins with or without thiamine and pyridoxine (Verduyn et al., 1992), and with or without 20 mg/L uracil, 20 mg/L histidine and 20 mg/L tryptophan supplemented. CELLSTAR^® 96 well cell culture plate (Greiner Bio-One) with an air-penetrable lid (Breathe-Easy, Diversified Biotech) was used for all cultivation. pH was adjusted with 1M HCl to 3, 4, 5 and 6 for the respective experiment. Cultivation was performed with continuous double orbital shaking of 548 cycles per minute (CPM) at 37 °C and 0%, 0.1%, 1% or 21% oxygen. Anaerobic conditions were obtained using a vinyl anaerobic chamber (Coy Laboratory Products Vinyl; gas mixture, 95% N₂ and 5% H₂), and microaerobic conditions were obtained using CO₂/O₂ Gas Controller (BioTek).

The cultures were incubated into 200 µL SC media without thiamine, containing 1.7 g/L yeast nitrogen base without amino acids and thiamine (FORMEDIUM™), 1.92 g/L Yeast Synthetic Drop-out Medium Supplements without uracil, for 48 h with an initial OD₆₀₀ of 0.05. The media was supplemented with either 0, 0.001, 0.01, 0.1, 1, or 400 μg/ml thiamine hydrochloride (Sigma Aldrich) and 20 mg/L uracil.

The SbU⁻ and SbU⁻ + thi6∆ strains were incubated in 25 ml SC with and without thiamine supplemented for 120 h. The cultures were incubated in a 250 ml shake flask with an initial OD₆₀₀ of 0.05. The culture was spun down and re-suspended in 500 µL PBS. A serial dilution was generated. 100 µL of the undiluted and 5 µL from each dilution were plated on SC plates with and without thiamine supplemented (Supplementary Figures S3B, C). The plates were incubated at 37°C for 72 h. Cell mass from the undiluted 5 µL spotting was spread out on SC plates with and without thiamine supplemented (Supplementary Figures S3B, C).

The SbU⁻, SbU⁻ + rei1∆, SbU⁻ + bts1∆ and SbU⁻ + bts1∆ + thi6∆ strains were incubated in 2.6 ml YPD in a 24-deep well plate (Axygen^®, VWR) with a sandwich cover (Enzyscreen) and with an initial OD₆₀₀ of 0.05. The plates were incubated at 15, 20, and 37°C for a maximum of 120 h. OD₆₀₀ was measured at 0, 8, 24, 32, 48, 72, 96, and 120 h.

Single and co-cultivation of (SbU⁻)-GFP and (SbU⁻ + bts1∆ + thi6∆)-mKate were cultured in 2 ml SC media with and without thiamine in a 24-deep well plate (Axygen^®, VWR) with a sandwich cover (Enzyscreen) and with an initial OD₆₀₀ of 0.1. The co-culture was started with a 1:1 ratio of each strain. 20 µL of single and co-cultures were transferred into fresh media every 48th hours for a total period of 96 h (two transfers).

Single and co-cultivation of (SbU⁻)-GFP and (SbU⁻ + bts1∆ + thi6∆)-mKate were cultured in 2 ml SC media YPD in a 24-deep well plate (Axygen^®, VWR) with a sandwich cover (Enzyscreen) and with an initial OD₆₀₀ of 0.1 at 15°C, 20°C, and 37°C. The co-culture was started with a 1:1 ratio of each strain. 20 µL of single and co-cultures were transferred into fresh media every 24th hours for a total period of 120 h (five transfers) for the 20°C and 37°C cultivation. Single and co-cultures at 15°C were transferred into fresh media every 48th hours for a total period of 144 h (three transfers). The culture was transferred every 48th hours to ensure that SbU⁻ reached similar cell counts at all temperatures (Supplementary Figure S6).

Serial dilution (1,000x and 10,000x) from the 37°C co-cultivation of (SbU⁻)-GFP and (SbU⁻ + bts1∆ + thi6∆)-mKate in YPD were plated on YPD plates and incubated at 15°C, 20°C and 37°C for 144, 96, and 48 h respectively. Red and green colony forming units (CFU) were counted under blue light.

All competition experiments were analysed with a NovoCyte Quanteon™ (Agilent) flow cytometry. 20 µL culture was diluted in 180 µL 1% PBS and run on the flow cytometry with a threshold of 5,000 yeast events or 60 µL sample injection. Yeast events were gated based on size events (FSC-A) < 6 × 10⁶ and complexity events (SSC-A) < 2 × 10⁵. Singlets were gate based on SSC (SSC-A vs. SSC-H). The (SbU-)-GFP population was quantified with the FITC channel, and (SbU⁻ + bts1∆ + thi6∆)-mKate was quantified with PE-TexasRed. The population distribution was quantified based on FITC-H vs. PE-TexasRed-H. Absolute events were calculated.

The Exendin-4 producing S. boulardii strains were incubated in 2 ml DELFT medium supplemented with 20 mg/L uracil in a 24-deep well plate (Axygen^®, VWR) with a sandwich cover (Enzyscreen). The cultures had an initial OD₆₀₀ of 0.05 and were performed with continuous shaking at 250 RPM at 37°C. All cultures were harvested after 24 and 48 h. Cell cultures were spun down at 10,000 g for 10 min at 4°C. Exendin-4 was quantified with Exendin-4 EIA (EK-070-94, Phoenix). The signals were detected by OD₄₅₀ using a microplate reader Synergy™ H1 (BioTek).

All animal experiments were conducted according to the Danish guidelines for experimental animal welfare, and the study protocols were approved by the Danish Animal Experiment Inspectorate (license number 2020-15-0201-00405). The study was carried out in accordance with the ARRIVE guidelines (du Sert et al., 2020). All in vivo experiments were conducted on male C57BL/6 mice (6−8 weeks old; Taconic Bioscience). Unless otherwise stated, all mice were housed at room temperature on a 12-h light/dark cycle and given ad libitum access to water and standard chow (Safe Diets, A30). All mice were randomised according to body weight and acclimated for at least 1 week prior to the first oral administration. Each animal study received a freshly prepared batch of S. boulardii. Body weight and food intake were recorded once per week. The researchers were blinded in all mouse experiments. The mice were euthanised by cervical dislocation at the end of the study.

The mice were divided into four groups (n = 4), either receiving the Sb wild-type, Sb bts1∆, Sb thi6∆ or Sb bts1∆ + thi6∆ strain. The mice were orally administered via intragastric gavage with ∼10⁸ CFU of the S. boulardii strain in 100 µL of 1x PBS and 20% glycerol. The mice were orally administered with S. boulardii for five consecutive days, followed by a 6-day washout. The drinking water was supplemented with an antibiotic cocktail containing 0.3 g/L ampicillin sodium salt, 0.3 g/L kanamycin sulfate, 0.3 g/L metronidazole, and 0.15 g/L vancomycin hydrochloride after the washout period. After 5 days of antibiotic treatment, the mice were orally administered with S. boulardii in 100 µL of 1x PBS (containing 1.0 g/L ampicillin sodium salt, 1.0 g/L kanamycin sulfate, 1.0 g/L metronidazole, and 1.0 g/L vancomycin hydrochloride) and 10% glycerol for five consecutive days. The washout for the antibiotic-treated mice was monitored for 33 days.

The faeces were collected in pre-weighed 1.5 ml or 2.0 ml Eppendorf tubes containing 1 ml of 1x PBS and 25% glycerol and weighed again to determine the faecal weight. All sample preparation for assessing CFU numbers was kept on ice and followed the same practice. The faecal samples were homogenised by vortexed at ∼2,400 rpm for 20 min. The samples were then spun down at 100 g for 30 s, followed by a dilution series, where 5 µL of each dilution was plated in duplicates or triplicates. Under the antibiotic-treated washout period, 100 µL was plated. The faecal samples were plated on SC supplemented with 20 mg/L uracil plates containing 100 mg/L ampicillin, 50 mg/L kanamycin, and 30 mg/L chloramphenicol (Sigma Aldrich).

All statistical analysing were performed in RStudio version 4.1.0 with the rstatix and DescTools packages. Unless otherwise stated, all data are presented as means +SEM. Statistical differences between groups of two were analysed with a dependent sample t-test. Bonferroni adjustments were used for multiple comparisons. Comparison of three or more groups was analysed by ANOVA with either Tukey’s HSD post hoc test or Dunnett’s post hoc test. The statistical significance level was set at p < 0.05.

We started by investigating the impact of constraining the probiotic yeast to become dependent on the exogenous supply of metabolites for growth and survival, as it is one of the most common biocontainment strategies for genetically modified microorganisms (Wook Lee et al., 2018). Here we generated a library of auxotrophic S. boulardii strains by disrupting several genes and evaluating the burden of the gene deletion on each strain. We created one nucleoside (uracil), two amino acids (histidine and tryptophan) and two vitamins (thiamine and pyridoxine) auxotrophic strains (Figure 2A). We started by introducing a stop codon in URA3 (uracil synthesis), HIS3 (histidine synthesis) and TRP1 (tryptophan synthesis), as a foundation for selecting a biocontainment that may also be used as an auxotrophic marker for further strain engineering. Auxotrophic markers are commonly used in microbial engineering to establish an antibiotic-free platform for plasmid-based heterologous gene expression (Durmusoglu et al., 2021). These auxotrophic strains have previously been reported in S. boulardii (Liu et al., 2016). Importantly, these gene disrupted strains are suitable hosts for further genetic manipulation using existing Saccharomyces cerevisiae tools (Hudson et al., 2014; Liu et al., 2016; Durmusoglu et al., 2021; Jin et al., 2021). The strains with URA3, HIS3, and TRP1 disrupted were unable to grow unless uracil, histidine or tryptophan was supplemented to the growth medium (Figures 2B, C). The TRP1 ^P12X (SbT⁻) strain also showed a metabolic burden when media was supplemented with tryptophan (Figure 2C; Supplementary Figure S1). Based on these observations, the HIS3 ^G26X (SbH⁻) and URA3 ^S81X (SbU⁻) strains showed to be the most promising nutritional auxotrophs compared to the SbT⁻ strain. Combination of the URA3, HIS3, and TRP1 gene disruptions maintained their auxotrophic phenotype and showed restricted growth in media lacking their respective nutrition (Figure 2; Supplementary Figure S2). However, the triple disruption strain showed to be unstable, with growth observed after 120 h in the absence of tryptophan. The SbU⁻ strain was selected for further gene knockouts based on a previous report showing higher gene expression from URA3-marker plasmids (Durmusoglu et al., 2021).

We further investigated the phenotype of disrupting two other biosynthetic pathways in strains with uracil synthesis deficiency to determine potential additive effects. Knocking out either the THI2 or the THI6 gene in the thiamine synthesis pathway resulted in strains that were unable to grow in media lacking thiamine (Figure 2B). Knocking out the SN O 1 and SNZ1 genes of the pyridoxine synthesis pathway individually resulted in no significant growth defect, contrary to previously reported (Rodríguez-Navarro et al., 2002). Non-etheless, a significant lower final OD₆₀₀ was observed when knocking out the SNO1 and SNZ1 genes in combination. To demonstrate that the growth defect resulted from the respective auxotrophies, we cultured the various strains in media containing the corresponding nutrient supplements and observed similar OD₆₀₀ reaching after 48 h (Figure 2C).

To examine the robustness of the auxotrophic strains towards conditions experienced in the human gastrointestinal tract, we assessed their growth performance in minimal media with pH and oxygen conditions more closely mimicking the gastrointestinal tract. The thi2∆ strain displayed a more considerable metabolic burden compared to thi6∆, demonstrating 25%–50% increased doubling time at pH 3, 6, and in microaerobic (0.1% and 1% oxygen) and anaerobic conditions compare to thi6∆ strain (Supplementary Figure S1). The thi6∆ strain showed a slight growth defect at pH 4 and 5, demonstrating 3%–6% higher doubling time than thi2∆ (Supplementary Figure S1).

We next evaluated the thiamine sensitivity in the strains to identify the minimum thiamine concentration required for growth by the thi2∆ and thi6∆ strains to circumvent the biocontainment. Both knockouts reached a lower final OD₆₀₀ at a concentration ≤0.01 μg/ml (Figure 2D). The thi6∆ strain appeared more sensitive to low thiamine concentrations than the thi2∆ strain. We also confirmed that from a 120-h culture with and without thiamine present in the media, zero escaper were identified for the thi6∆ strain (Supplementary Figure S3).

Another strategy to reduce the proliferation of genetically modified microorganisms in specific environments is to make them sensitive to temperature changes that may occur outside the targeted environment (Figure 3A). To assess this approach, we knocked out two genes (REI1 and BTS1) that were previously reported to exhibit a growth defect at a temperature lower than 25°C when knocked out in S. cerevisiae (Jiang et al., 1995; Hung and Johnson, 2006; Lebreton et al., 2006). The REI1 gene encodes a zinc finger protein part of the 60S complex, and the BTS1 gene encodes the yeast geranylgeranyl diphosphate synthase. We observed the expected growth defects at temperatures of 15°C and 20°C for both rei1∆ and bts1∆ strains (Figure 3B). While both bts1∆ and rei1∆ strains were hypersensitive at 15°C and showed limited growth after 120 h, the rei1∆ displayed a more pronounced growth defect; however, the rei1∆ strain also showed a slower growth at 37°C (Supplementary Figure S4).

Furthermore, evaluating the two cold-sensitive strains in the gut-like conditions, we demonstrate that the rei1∆ strain showed a more pronounced fitness cost in minimal media at different pH (Figure 3C). The rei1∆ strain was unable to grow at a pH lower than 4 after 72 h, and the strain displayed an approximate 25% slower doubling time at pH 4, 5, and 6 compared to the parental strain (SbU⁻). The rei1∆ strain also showed a significantly lower growth rate in anaerobic and microaerobic conditions than SbU⁻ (Supplementary Figure S5), while the bts1∆ demonstrated a slower growth rate in anaerobic conditions. The doubling time at anaerobic conditions was comparatively more pronounced for the bts1∆ than for the rei1∆ strain.

To generate a more robust biocontainment strain, we combined the nutritional auxotrophy and the cold-sensitive mutant (Figure 4A). Specifically, we chose thi6∆ and bts1∆, in the SbU⁻ background strain, based on functionality at different pH concentrations and oxygen gradients. Although the rei1∆ strain showed a more pronounced growth retardation at the lower temperature we hypothesis that additional fitness cost through genetic deletions or heterologous gene expressions would limit the strain to survive in the gastrointestinal tract. The double gene knockout of BTS1 and THI6 maintained the phenotypic effect of the individual knockouts, demonstrating the inability to grow in the absence of thiamine supplementation and slower growth in temperatures lower than 20°C (Supplementary Figure S4, S6).

Next, we evaluated the competitive fitness of the double knockout biocontainment strain in a co-culture with the parental control strain SbU⁻ in the presence and absence of thiamine and at different temperatures. The double knockout and control strain perform similarly in the presence of thiamine; however, in the absence of thiamine, the double knockout strain was undetected after two transfers (Figure 4A; Supplementary Figure S6). In the same manner, the double knockout strain and the control strain were able to grow simultaneously at the permissive temperature conditions of 37°C; however, at 15°C and 20°C, the percentage of double knockout strains was drastically reduced incrementally with the number of culture transfers (Figure 4B). The biocontainment double knockout strain maintained ∼1% of the total yeast population after two transfers of the co-culture at 15°C and ∼5% after five transfers at 20°C.

Furthermore, we tested whether the biocontainment strain would maintain its cold-sensitive phenotype after multiple generations in co-culture with the control strain at 37°C (120 h). Here, we observed that the double knockout strain maintained its cold-sensitive phenotype, demonstrating zero colonies after plating the culture, incubating the strain at 15°C and maintaining a slower growth phenotype at 20°C (Supplementary Figure S8). The double knockout biocontainment strain also showed no synergistic growth defect at different pH concentrations and oxygen percentages (Supplementary Figure S9). We also observed an increased thiamine sensitivity by the combined biocontainment strain, showing a significantly lower OD₆₀₀ in 1 μg/ml after 48 h (Supplementary Figure S10).

Finally, we sought to investigate if the double knockout biocontainment strain can produce a similar level of a recombinant protein as the parental strain SbU⁻. This is relevant since some gene knockouts might negatively impact protein synthesis (Puddu et al., 2019). Therefore, we engineered the biocontainment strains to produce and secrete a GLP-1 receptor agonist (Exendin-4) since in situ biosynthesis of Exendin-4 in the gastrointestinal tract with S. boulardii has previously been shown to reduce food intake and body weight in mice (Hedin et al., 2022a). We observed no significant difference in the production of Exendin-4 after 24 and 48 h (Figure 4C), suggesting that the biocontainment strain does not compromise the production of a heterologous protein.

While we have validated the biocontainment strategy to be functional in vitro, the generated knockout strains still need to be demonstrated to be safe and viable in the gastrointestinal tract to express a therapeutic activity. To assess this, we orally administered the biocontainment strains daily for 5 days in conventional mice with an intact microbiome (Figure 5A) to study if the biocontainment strains pose any fitness cost while interacting with other microbes. We demonstrated that the wild-type, single and combined knockouts colonise equally in the mice with intact microbiome (Figure 5B). In addition, all the S. boulardii strains were washed out after 6 days, except for one mouse receiving the bts1∆ strain (Figure 5C), indicating that the knockouts did not cause any unwanted engraftment in mice.

Furthermore, we evaluated the biocontainment strains in an antibiotic-treated mouse model designed for pre-clinical testing of yeast-based AMTs, which allows S. boulardii strains to colonise the murine gastrointestinal tract in higher numbers and for an extended period of time (Hedin et al., 2022b). The antibiotic-treated mice were orally administered with the biocontainment strains for 5 days. Wild-type, single and combined knockouts were demonstrated to also colonise equally in the mice with a reduced microbiota (Figure 5D). Although the S. boulardii was washout slower in the antibiotic-treated mice, no mice displayed any detectable levels 33 days after the last oral administration (Figure 5E).

To investigate if the newly generated strains posed any safety concerns to the health and wellbeing of the mice, body weight, food intake and behaviour were monitored throughout the whole study. We observed no change in body weight or food intake in the mice receiving the biocontainment strains (Figures 5F, G), both during low and high colonisation of the strains. In addition, no abnormal behaviour was observed in the mice.