Lactobacillus-vectored nanobodies improve broiler productivity under sub-clinical necrotic enteritis with associated microbiome and transcriptome changes

Abstract

Sub-Clinical Necrotic Enteritis (SCNE), caused by toxin-producing Clostridium perfringens, is a major challenge in poultry production. SCNE has traditionally been managed with in-feed antibiotics; however, increasing concerns about the spread of antimicrobial resistance call for antibiotic-free strategies for its control. We recently described an NE control strategy leveraging Limosilactobacillus reuteri probiotic strains genetically engineered to deliver nanobodies against alpha toxin and NetB from C. perfringens in the poultry gut. Here, in a controlled study under SCNE conditions, we found that the engineered strains significantly improved feed conversion ratios and weight gain of broilers, outperforming treatment with either a prophylactic antibiotic or the wild-type probiotic strains. To investigate the systemic factors contributing to these performance differences, we analyzed histomorphometrics of the small intestine, microbial metatranscriptomics of jejunal contents, and gene expression from the jejunum and liver tissues. Our results confirmed the in situ expression of the nanobodies and provided evidence that nanobody delivery mitigates SCNE-associated inflammation in the jejunum and toxin-induced damage in the liver, leading to a more quiescent immune state, lower oxidative stress, and improved growth performance. Our findings demonstrate the potential of probiotic-vectored nanobody delivery as an effective strategy for targeting gut antigens across a range of diseases.

Introduction

Antibiotic-free livestock production has increased in recent years, driven by concerns about the global rise of antibiotic resistance, shifting consumer preferences, and regulatory restrictions aimed at reducing the use of medically important human antibiotics^1,2. These changes have been particularly impactful in the poultry industry, with broilers raised without antibiotics increasing from ~5% to >50% in the U.S. between 2014 and 2021³. Antibiotics are commonly used to treat bacterial infections in diagnosed animals, to control the spread of illness from sick animals, and to prevent illness in healthy animals when exposure to bacterial disease is imminent. Therefore, when clinical or sub-clinical manifestations of livestock diseases are present, antibiotic-free production can pose several challenges for animal welfare. In addition to an increased prevalence and severity of diseases in broilers raised without antibiotics⁴, the loss of the growth-promoting effects of in-feed antibiotics⁵ translates to higher operational costs due to longer growth cycles, and increased feed and resource use per unit of animal protein produced.

Necrotic Enteritis (NE) is a multifactorial intestinal disease of poultry primarily driven by overgrowth and toxin production by the bacterium Clostridium perfringens^6,7. In its clinical form, NE is characterized by inflammation and necrosis of the intestinal lining, particularly the jejunum and ileum, leading to increased mortality and visible symptoms such as diarrhea, ruffled feathers, and depression. C. perfringens is a normal inhabitant of the chicken gut, with pre-disposing factors such as coccidiosis enabling C. perfringens overgrowth and development of the disease⁸. Notably, NE can also manifest in a sub-clinical form (SCNE), wherein there is no peak mortality and few, if any, external symptoms are visible. In SCNE, C. perfringens toxins such as NetB and alpha toxin cause damage to the intestinal epithelium, leading to poor digestion and absorption, and reduced feed conversion and weight gain⁹. It is estimated that the sub-clinical form of the disease has the biggest economic impact (>US$2 billion per year) because of this productivity loss^10,11, which is aggravated in the absence of prophylactic antibiotics.

To control NE in a post-antibiotic era, we recently developed a strategy consisting of two genetically engineered Limosilactobacillus reuteri probiotic strains – NE01 and NE06 – each designed to produce and secrete nanobodies against alpha toxin and NetB from C. perfringens¹², respectively. The neutralizing activity of the nanobodies was validated in vitro, and the efficacy of a 1:1 combination of the engineered strains in reducing NE-associated mortality by up to 48% was confirmed using a clinical NE challenge model¹². Given these encouraging results, we hypothesized that the engineered strains may improve bird performance under SCNE in a commercial-like production setting. Here, we present the results of a 43-day floor pen trial comparing the performance of birds receiving two doses of the engineered strains to control birds, birds treated with bacitracin methylene disalicylate (BMD, a prophylactic antibiotic) and birds receiving the wild-type probiotic strains: L. reuteri 3630 (the parental strain of NE01 hereafter referred to as Lr3630) and L. reuteri 3632 (the parental strain of NE06 hereafter referred to as Lr3632). In all cases, birds were challenged with C. perfringens¹³. We evaluate the treatment effects on weight gain and feed conversion efficiency and describe the systemic covariates of the observed differences, including the analysis of histomorphometrics, gut metatranscriptomics, and host gene expression data.

Results

L. reuteri-delivered nanobodies improve growth performance under a mild NE challenge

Ross 708 broilers were assigned to four treatment groups. All four groups were subject to a mild NE challenge¹³ consisting of a live coccidiosis vaccine on the day of hatch, followed by approximately 1 × 10⁹ CFU/bird of C. perfringens on days 15, 16, 17, and 28 as a top dressing on feed. Group 1, the challenge control, received no further treatment. Group 2 received 50 g per US ton of feed of BMD daily to evaluate the effect of conventional in-feed antimicrobial use. Group 3 received two doses of a 1:1 mixture of the NE01 and NE06 strains; the first dose was applied on the day of hatch, and the second on day 14. Group 4 received equivalent doses of a 1:1 mixture of the Lr3630 and Lr3632 parental probiotic strains. The mortality-adjusted feed conversion ratio (aFCR) and average weight gain per pen were measured on study days 15, 28, and 43 (Fig. 1a). We note that the day-15 measurements reflect performance differences prior to the C. perfringens challenge, which was introduced on that day.

Fig. 1. NE01 + NE06 improves broiler performance under a mild NE challenge.

a Schematic of the experimental C. perfringens challenge, timing of data and sample collection, dosage of NE01 + NE06, and analyses performed. b aFCR values per treatment normalized to the mean aFCR of the challenge control calculated on days 15, 28, and 43 of the study. c Weight gain in kg per treatment normalized to the mean weight gain of the challenge control on days 15, 28, and 43 of the study. In (b and c), the dashed black line indicates the mean of the challenge control, error bars represent the SEM, * indicates p-value < 0.05 compared to the challenge control. n = 20 pens per treatment. aFCR mortality-adjusted feed conversion ratio.

Compared to the challenge control, birds receiving NE01 + NE06 showed a 6-, 7-, and 4-point improvement in aFCR on days 15, 28, and 43 of the study, respectively (p-values = 0.024, 0.017, 0.054). Of the remaining treatment groups, only BMD-treated birds showed a trending improvement in aFCR on day 15 (4 points, p-value = 0.13) (Fig. 1b). On days 28 and 43, the aFCR was 8 and 5 points lower, respectively, in birds treated with NE01 + NE06 compared to birds receiving Lr3630 + Lr3632 (p-values = 0.016, 0.011). In addition to the aFCR results, the average weight gain was higher by 34 g and 62 g on days 28 and 43, respectively, in birds receiving NE01 + NE06 compared to the challenge control (p-values = 0.035, 0.067). Bird weights were comparable across treatments on day 15 (Fig. 1c, p-values > 0.05). On day 43, weight gain was higher in the NE01 + NE06 group compared to both the BMD and Lr360 + Lr3632-treated groups by 76 g and 81 g, respectively (p-values = 0.025, 0.017). These results demonstrate a substantial improvement in performance parameters associated with the in situ delivery of nanobodies against alpha toxin and NetB in the presence of a mild NE challenge.

Markers of sub-clinical NE show few differences across treatment groups

To model SCNE under commercial-like conditions, we used a mild NE challenge aiming for less than 5% mortality. Compared to a clinical NE model, birds were not exposed to coccidia or other diet interventions¹⁴ immediately prior to exposure to NetB- and alpha toxin-producing C. perfringens. As expected, the average NE-associated mortality observed was low (<2%). We did not detect significant differences in NE-associated mortality between treatment groups (Supplementary Table 1). Lesion scores for birds euthanized on days 18, 31, and 43 showed either no lesions or lesions in the lowest grade category (see “Methods” section), except for one bird from the challenge control group on day 18 showing grade 2 lesions. As expected, given the three consecutive days of C. perfringens challenge on days 15–17 compared to a single day on day 28, the average lesion score across treatments was higher on day 18 than day 31 (0.49 vs. 0.13, p-adj = 8 × 10⁻⁹). By day 43, virtually no lesions were detected. Similar to the NE-associated mortality result, there was no differentiation by treatment in the observed lesion scores (Supplementary Table 2).

On days 10, 18, 25, and 43, jejunal samples were collected for histomorphometrics analysis. Histological evaluation of hematoxylin and eosin-stained jejunal sections showed multiple pathological hallmarks and signs of mucosal inflammation across treatment groups, including epithelial hyperplasia, crypt dilation, expansion of gut-associated lymphoid tissue (GALT), and distorted villus tips (Fig. 2a). Heterophil infiltration in the lamina propria and intraepithelial lymphocytosis were also observed, evidencing mucosal inflammation (Fig. 2b). Consistent with the higher lesion scores observed on day 18, these features were less prevalent on days 31 and 43 samples (Fig. 2c). Among the histological characteristics quantified, only crypt hyperplasia showed a lower prevalence in birds receiving NE01 + NE06 compared to the challenge control (p-adj: 0.008) (Supplementary Table 3). This result was likely driven by differences on day 43 (Fig. 2c), when a higher villus-to-crypt depth ratio in the NE01 + NE06 group was also observed (Fig. 2d, p-value = 0.034, p-adj = 0.566).

Fig. 2. Histomorphometrics analysis of jejunal tissue and C. perfringens counts in fecal and cecal samples.

a Representative hematoxylin and eosin-stained jejunal sections showing key pathological features across treatment groups, including cystic crypt dilation (*), misshaped villus tips (arrowheads), shortened villi and deep hyperplastic crypts (double arrow lines), and increased gut-associated lymphoid tissue (boxed), indicative of enteric disruption and immune activation. b Histologic evidence of mucosal inflammation, including heterophils in the lamina propria (circled) and intraepithelial lymphocytes (boxed). c Cumulative pathology index and individual pathology scores across treatment groups on days 10, 18, 31, and 43 of the study. d Villus-to-crypt depth ratios across treatment groups as a function of time. Boxes represent the interquartile range. e, f C. perfringens counts in fecal and cecal samples as a function of study day. * p-value < 0.05, + < 0.10.

Plating cecal samples collected on days 10, 18, and 43 on solid growth media (data from day 31 were not generated), as well as fecal samples collected weekly starting on day 10, showed the presence of C. perfingens in about 5% of cecal samples and 50% of fecal samples. While there were no overall differences in C. perfringens abundance between treatments (Supplementary Table 4), we detected differences over time, with fecal samples collected on day 24 and cecal samples collected on day 18 showing higher C. perfringens detection rates than samples collected on day 10 (Fig. 2e, f, p-values = 0.007 and 0.059, respectively). Similar trends were observed for coccidia oocytes, peaking between days 18 and 31 of the study (Supplementary Fig. 1a, b). These results are consistent with the timing of the experimental challenge on days 15–17 and 28 and suggest that birds were, as expected, at a higher risk of NE on these days.

Taken together, the results demonstrate NE-associated mortality rates, morphological, histological, and microbiological changes consistent with SCNE, with only minor differences by treatment, despite the clear improvement in performance in birds that received NE01 + NE06.

Probiotic-vectored nanobodies mitigate microbial signs of inflammation in the jejunum

Given the mild differentiation of tissue histomorphometrics between treatment groups, we asked whether the intestinal microbiome would provide a higher resolution readout of the state of the gut. Previous reports have shown a relationship between growth performance, intestinal inflammation, and the composition of the microbiome in the small intestine of chickens¹⁵. Here, we carried out metatranscriptomics of the jejunal contents for birds in the challenge control, NE01 + NE06, and Lr3630 + Lr3632-treated groups on days 10, 18, 31 and 43 of the study (20 birds per treatment group –one bird per pen– were analyzed per study day). The data were used to simultaneously capture patterns of species abundances and their transcriptional activity.

Interestingly, we observed a high relative abundance of RNA from various virus species in the jejunum, especially at early time points, when Chicken astrovirus and Gallivirus A added up to 90% relative abundances at the species level (see “Methods” section, Supplementary Fig. 2a). These observations align with the cystic dilated crypts and villus/crypt dynamics observed at the tissue level (Fig. 2a), which are seen in commercial broilers with chicken astrovirus^16,17. While the overall abundance of viral RNA did not differ by treatment group (p-value > 0.78), there was a marked increase in diversity over time (Supplementary Fig. 2b), with only a few virus species decreasing (e.g. SARS-related coronavirus) or increasing (e.g., chicken picornavirus 1) in birds treated with NE01 + NE06 compared to the challenge control (Supplementary Table 5). Reads mapping exclusively to viral genomes were not considered in the remaining analyses.

Looking at the relative abundances of bacterial transcripts at the species level, we did not observe significant differences in alpha diversity between treatments, or across time (Supplementary Table 6). As expected, the data showed a strong dominance of several genera in the Lactobacillaceae family, which accounted for ~90% of transcript-level bacterial relative abundance. The next most frequent species included Helicobacter, Streptococcus, Enterococcus, and Ureaplasma species, accounting for 5% to 0.05% of species-level relative abundances, respectively (Supplementary Fig. 3). Comparison of species-level transcript abundance profiles using weighted UniFrac distances between samples¹⁸ showed a significant effect of day of sampling on transcriptome composition (Fig. 3a, PERMANOVA p-value = 0.001) but only a marginal contribution of treatment (Fig. 3b, p-value = 0.065). The major changes observed over time, regardless of treatment, involved differences in the proportions of Lactobacillaceae species, primarily on days 31 and 42, as well as an increase of Ureaplasma in samples collected on day 43 of the study (Supplementary Fig. 4).

Fig. 3. Diversity and composition of jejunal content metatranscriptomes across treatments.

a Principal Coordinates Analysis (PCoA) of jejunal content metatranscriptomics samples based on species-level weighted UniFrac distances. Colors indicate sample collection days. Numbers in parentheses indicate the variance explained by each principal coordinate. b PCoA for jejunal content samples collected at each time point colored by treatment group. c Relative abundance of L. reuteri transcripts as a function of treatment group and study day. d Number of samples with reads mapping to the anti-NetB or anti-alpha toxin nanobodies (VHHs) across sampling times and treatment groups. e Proportion of sequencing reads in the NE01 + NE06 samples mapping to either anti-alpha toxin or anti-NetB VHHs as a function of time. f Beta-dispersion between samples by treatment group and sampling day. Beta-dispersion is quantified by the weighted UniFrac distance between each sample and the centroid of the samples in a given group. ** p-value < 0.01. In (c and f), boxes represent the interquartile range.

As a function of treatment, we observed a significant increase in L. reuteri transcripts in birds treated with the engineered strains (Supplementary figure 4), which was apparent on days 10, 18, and 31 (Fig. 3c). We note that L. reuteri transcripts were detected in birds from the challenge control group as well as those treated with NE01 + NE06 and Lr3630 + Lr3632. This is not surprising given that L. reuteri is common in the jejunum of chickens¹⁹, and it suggests that only a fraction of L. reuteri reads were derived from the engineered or parental strains used in the present study. While the increased L. reuteri abundance could be a transcriptional response of naturally occurring strains to the treatment, the additional 1.5% and 3.7% relative abundance observed on days 10 and 18 compared to the challenge control (Fig. 3c) could also reflect the presence of the engineered strains in the jejunum. For comparison, the average relative abundance of C. perfringens in the jejunum estimated from the metatranscriptomics data was 0.004%, excluding a single sample from the challenge control group at day 18 where it was 81%. Nevertheless, given the high sequence similarity expected within L. reuteri lineages (98-99% average nucleotide identity)²⁰, data are insufficient to differentiate our strains from naturally occurring L. reuteri populations.

Along with the higher abundance of L. reuteri transcripts in birds receiving the engineered strains on days 10 and 18, we detected transcripts mapping to the anti-NetB and anti-alpha toxin nanobodies in more samples from this treatment group during the same periods (Fig. 3d). Only a few samples from the NE01 + NE06 group contained reads mapping to the nanobody sequences and in each case only 1 to 20 mapping read pairs were detected (Fig. 3e). Notably, sequencing reads mapping to the anti-NetB nanobody were detected in one sample from day 43, suggesting that the engineered strains can persist through the whole production cycle in at least some of the treated birds. The low frequency of reads mapping to the nanobody sequences in birds receiving NE01 + NE06 is expected given the obtained sequencing depth of ~33 million (host-filtered) read pairs per sample. Assuming a relative abundance of 1% of the engineered strains and that the nanobodies are likely to represent less than 1% of the proteome in NE01 and NE06, based on the promoter and signal peptide used¹², only tens of reads would be expected per sample. In addition, the short length of the nanobody coding sequences (~400 nucleotides) compared to a typical bacterial gene would further reduce the probability of detecting nanobody transcripts. Despite these limitations, the results demonstrate that the engineered strains can colonize and persist in the small intestine of treated birds, that both nanobodies are actively expressed in situ, and that even when the engineered strains may be only a minor fraction of the gut microbiome, the secreted nanobodies are sufficient to enhance performance under SCNE. We did not attempt to recover, quantify, or analyze nanobody particles from the jejunal contents.

While only marginal differences were observed in overall metatranscriptome compositions between treatments, we observed fewer outliers in birds treated with the engineered strains compared to birds in the challenge control or Lr3630 + Lr3632-treated groups (Fig. 3b). These results were confirmed by a significant effect of treatment on beta-dispersion between treatments quantified using weighted UniFrac, particularly on days 10 and 18 (Fig. 3e). The results indicate reduced microbiota variability in the jejunum of birds receiving the engineered strains, suggesting greater resilience to stochastic perturbations due to external stressors²¹, such as C. perfringens challenge.

Comparing the abundance of transcripts associated with specific gene functions using Gene Set Enrichment Analysis (GSEA)²² showed higher expression of ribosomal and fermentation genes in birds receiving NE01 + NE06 compared to the challenge control. In turn, we saw lower expression of genes associated with purine, cobalamin and glycogen biosynthesis (Fig. 4). Compared to birds receiving Lr3630 + Lr3632, we also saw higher expression of ribosomal genes and lower expression of genes involved in respiration, oxidative phosphorylation, nitrate reduction, motility and chemotaxis, and synthesis of fatty acids, biotin, purines, and glycogen, among others, in birds treated with NE01 + NE06 (Fig. 4). The higher relative expression of ribosomes vs. biosynthetic genes may suggest a lower metabolic independence of species in the microbiomes of birds receiving the engineered strains. Interestingly, metabolic independence has been recently associated with inflammation-related stress in human microbiomes²³. In addition, the lower expression of respiration genes, including genes involved in the reduction of nitrogen compounds, could also be a sign of reduced inflammation in birds treated with the engineered strains. Specifically, facultative anaerobes are favored by microaerophilic conditions or extracellular nitrate released during inflammation, which they can use as a terminal electron acceptor²⁴.

Fig. 4. Differentially expressed pathways in the jejunal microbiome from birds treated with NE01 + NE06.

a Significantly enriched pathways in birds treated with NE01 + NE06 compared to birds in the challenge control (pink) and the Lr3630 + Lr3632 (blue) groups. The x-axis shows the GSEA normalized enrichment score; positive values indicate higher expression in the NE01 + NE06 group. +p-adj < 0.1, *p-adj < 0.05. b Expression levels for the three enzymes in the 2,3-butanediol fermentation pathway. Error bars represent the SEM of expression values. ALS Acetolactate synthase, ALDC Acetolactate decarboxylase, BDH Butanediol dehydrogenase. TPM transcripts per million. c Microbial species expressing the 2,3-butanediol fermentation pathway ranked by their differential expression in the NE01 + NE06 group compared to the challenge control. *p < 0.05.

Among the fermentation-related transcripts that increased in abundance with NE01 + NE06, were enzymes involved in the conversion of pyruvate to 2,3-butanediol (i.e., acetolactate synthase, acetolactate decarboxylase, and butanediol dehydrogenase, Fig. 4b). Expression of these enzymes by the microbiome of birds receiving Lr3630 + Lr3632 was intermediate between those receiving NE01 + NE06 and the challenge control, though more closely resembling the latter (Fig. 4b). Notably, 2,3-butanediol has been reported to ameliorate acute lung injury and inflammatory responses induced by bacterial lipopolysaccharide (LPS), when provided to rats via gastric intubation²⁵. The metatranscriptomics data showed that among species expressing all three enzymes for the conversion of pyruvate to 2,3-butanediol, L. agrestis and L. reuteri accounted for the most transcripts (63–87%) and the strongest upregulation compared to the challenge control group (Fig. 4c).

Engineered strains modulate immune, signaling, and antioxidant function in host tissues

To investigate the systemic effects of the engineered strains under SCNE, we collected jejunum and liver tissue samples on days 10 and 18 of the study and used them for transcriptomics analysis. Similar to the metatranscriptomics analysis, we saw a large separation of expression profiles by time but little differentiation by treatment (Supplementary Fig. 5). The analysis of differentially expressed genes (DEGs) in the jejunum between day 10 and day 18 after controlling for treatment showed upregulation of genes associated with apoptosis and senescence, immune activation, oxidative stress, and energy metabolism. Conversely, we observed downregulation of genes associated with digestion and absorption, lipid metabolism, barrier function, and expression of gut hormones (Fig. 5a). Although we cannot disentangle the effect of time from the effect of the C. perfringens challenge, these changes are consistent with the observed lesions and histomorphology characteristics observed on day 18.

Fig. 5. Differential gene expression in the jejunum and liver tissue across time and treatment groups.

a Pathway activation scores for DEGs between days 10 and 18 in the jejunum. Positive values indicate upregulation on day 18. b Pathway activation scores for DEGs between days 10 and 18 in the liver. c Pathway activation scores on days 10 (left) and 18 (right) for jejunum gene expression in each treatment compared to the challenge control. d Pathway activation scores on days 10 (left) and 18 (right) for liver gene expression in each treatment compared to the challenge control. In (c and d), cells in color indicate p-adj < 0.2, * p-adj < 0.05. e Liver expression of genes in the insulin and mTORc signaling pathways was significantly downregulated with NE01 + NE06 compared to the challenge control. Values represent Z-scored normalized read counts across all samples. * p-value < 0.05, ** <0.01. Boxes represent the interquartile range.

In the liver, we observed an increase in expression of metabolic hormones and metabolic signaling, lipid metabolism, immune activation, and oxidative stress between day 10 and day 18. These changes were accompanied by a decrease over time for genes associated with cell replication, antioxidant systems, carbohydrate and energy metabolism (Fig. 5b), which would be consistent with the pro-inflammatory and disrupted barrier function gene expression observed in the gut.

As a function of treatment, birds treated with NE01 + NE06 showed the largest number of DEGs relative to the challenge control at both time points and organs compared to birds receiving the wild-type probiotic strains or BMD (Supplementary Fig. 6). In the jejunum, birds treated with NE01 + NE06 showed lower expression of genes involved in immune function and oxidative stress at both time points (Fig. 5c). Downregulated immune genes include pro-inflammatory and regulatory cytokines, interferons, Microbe-Associated Molecular Patterns (MAPMs) recognition, and components of the innate and adaptive immune system (Supplementary Fig. 7a). These differences were stronger and involved more DEGs and enriched pathways on day 10 samples than day 18 samples and may explain the reduced aFCR observed on day 10 compared to the challenge control (Fig. 1b). Also downregulated were protein kinases involved in oxidative stress induced senescence. In contrast, compared to the challenge control, these genes were often unchanged or upregulated in birds treated with BMD or Lr3630 + Lr3632. Immune activation genes were actually upregulated in birds receiving BMD on day 18 (Supplementary Fig. 7b). Across treatments, cell replication and apoptosis and senescence pathways were elevated relative to the challenge control on day 18, suggesting increased cellular turnover in the jejunum (Fig. 5c).

In the liver, the engineered strains showed the strongest downregulation of immune and oxidative stress genes on day 18, compared to the BMD and Lr3630 + Lr3632 groups (Fig. 5d). These changes were accompanied by strong upregulation of antioxidant systems and energy metabolism pathways, including oxidative phosphorylation, the citric acid cycle (TCA cycle) and the pentose phosphate pathway (Supplementary Fig. 8a). Interestingly, we also observed a strong downregulation of metabolic signaling in the liver in birds receiving the engineered strains, specifically of genes involved in insulin, glucagon, and mTORC1-mediated signaling (Supplementary Fig. 8b), which, at the pathway level, are in the opposite direction to the changes observed pre- and post-C. perfringens challenge (Fig. 5b). Specifically, compared to the challenge control at day 18, we observed lower expression of the insulin receptor (INSR) and PIK3 regulatory subunits (PIK3R2 and PIK3R4), as well as proteins in the mTORC1 signaling pathway (EIF4G1, MTOR, YWHAB, RPTOR) (Fig. 5e). Compared to the Lr3630 + Lr3632 group, these genes were also downregulated in the NE01 + NE06 group (Supplementary Fig. 9). Interestingly, higher mRNA levels of these signaling pathways in the challenge control and Lr36320 + Lr3632 groups may reflect toxin-induced liver injury, as previously reported in poultry²⁶.

Altogether, in both the jejunum and liver, the engineered strains partially reversed expression changes between day 10 and 18, presumed to be associated with C. perfringens challenge.

Discussion

The phase-out of antibiotics in livestock production poses major challenges for preserving animal health and productivity. While the beneficial effects of probiotics, prebiotics, phytochemicals, and other strategies as antibiotic alternatives have been widely reported, they often lack consistency across farms²⁷. This could be a direct consequence of their diverse mechanisms, including immune modulation, antimicrobial activity, competitive exclusion, and niche modification, among others^28,29, which are often context-specific^30–32. The present strategy for probiotic-vectored delivery of nanobodies¹² targets key molecular drivers of the disease to overcome this context-specificity, while still conferring the health benefits of the engineered probiotic strains. Although specificity towards specific antigens may also be achieved with vaccination or immunoglobulins⁶, microbial delivery offers several key advantages. The probiotic-vectored platform avoids the metabolic cost for the host to mount an immune response, eliminates the delay associated with immune activation, and does not require persistent dosing, relying instead on colonization by the engineered strains. In contrast to systemic vaccination, which often fails to elicit robust mucosal immunity^33,34, localized microbial delivery in the gut circumvents the need to induce local immune responses through direct neutralization of key toxins associated with disease pathology.

Interestingly, despite the performance improvement in birds receiving NE01 + NE06, there were very few visible signs of pathology that differentiated between treatments. Necrotic enteritis lesions evaluated at the time of necropsy received scores of 0: normal, or 1: slight mucus covering the small intestine³⁵, predominantly on day 18 of the study. This, along with the histopathology results (Fig. 2), indicates that birds were sub-clinically challenged, with no visible necrosis. It follows that the observed differences in performance were not driven by SCNE defects observed at the tissue level. Nevertheless, the lower cumulative pathology (driven by crypt hyperplasia) and higher villus-to-crypt ratios observed with NE01 + NE06 on day 43 could be an indication of better resolution post-challenge in birds receiving the engineered strains. As discussed below, the observed gene expression in both the jejunum and liver is indicative of a more quiescent immune and oxidation state in birds treated with NE01 + NE06 compared to the challenge control. Given the high metabolic costs associated with immune activation in broilers^36,37, it is possible that performance differences between treatments can be seen under SCNE even without major lesions or tissue differences. This would be consistent with the improved aFCR observed on day 15 (Fig. 1b) together with the stronger downregulation of immune pathways in the jejunum of NE01 + NE06 treated birds observed on day 10 (Fig. 5c).

We observed a reduction in aFCR in birds treated with NE01 + NE06 on day 15, prior to the C. perfingens challenge (Fig. 1b). This observation, along with the detection of C. perfringens in fecal samples on day 10 (Fig. 2e) suggest that naturally occurring C. perfringens may have been detrimental to performance, in particular following a pre-disposing live coccidiosis vaccine. Indeed, SCNE disease models relying on natural C. perfringens uptake have been proposed to better reflect the conditions of NE development in commercial operations³⁸. The very low C. perfringens transcript abundance observed in the jejunum suggests that even a small population of the engineered probiotics may be protective, which would be consistent with the performance results of our study and the low detection rate of nanobody transcripts, given the limited sequencing depth (Fig. 3d).

In contrast to the mild histomorphometric differentiation between treatments, we saw major differences in gene expression in both the jejunum and liver across treatment groups (Fig. 5). We observed birds in both the NE01 + NE06- and Lr3630 + Lr3632-treated groups showing evidence of lower immune activation in both organs, consistent with the immunomodulatory properties of the probiotic strains³⁹. However, NE01 + NE06 led to improved growth performance while Lr3630 + Lr3632 did not differentiate from the challenge control (Fig. 1). This suggests that immune modulation alone, or at the level observed with Lr3630 + Lr3632, is insufficient to overcome the negative effects of exposure to C. perfringens. It also suggests that probiotic properties of the strains complement toxin-neutralization in driving performance outcomes. Compared to Lr3630 + Lr3632, birds treated with NE01 + NE06 showed increased transcript levels for the microbial synthesis of 2,3-butanediol in the jejunum (Fig. 4b), with L. reuteri being among the species most contributing to this difference. Genomic analysis showed that all four strains (Lr3630, Lr3632, NE01, and NE06) have the genetic capacity to synthesize 2,3-butanediol, which was shown to alleviate LPS-induced inflammation in rats²⁵. Thus, it is possible that both the parental and engineered strains contributed to the differential abundance of the 2,3-butanediol synthesis pathway and its potential anti-inflammatory effects. Unfortunately, we cannot distinguish whether the L. reuteri transcripts detected originated from the strains provided as a treatment or from the native microbiome of the birds in the study. Given that the only genetic differences between the nanobody secreting strains and their parental probiotic strains are the corresponding expression cassettes¹², we hypothesize that neutralization of NetB and alpha toxin by the nanobodies led to a gut environment more conductive to microbial species, such as L. reuteri, expressing this pathway (Fig. 3c). If so, 2,3-butanediol synthesis could partially explain the stronger downregulation of inflammatory cytokines in the NE01 + NE06 treatment group.

By far, the biggest differences in gene expression were observed in the liver following exposure to C. perfringens (day 18). For NE01 + NE06 treated birds, these differences reverted many of the changes observed before and after C. perfringens exposure (Fig. 5b), such as upregulation of immune activation, oxidative stress, and metabolic signaling pathways. Notably, similar changes have been observed in the liver of broilers exposed to aflatoxin, LPS, ochratoxin A, and deoxynivalenol, among others^40–44. At the same time, NE01 + NE06 led to the upregulation of antioxidant systems and supporting energy metabolism pathways (e.g., via NADPH synthesis), which were depressed following C. perfringens exposure. A prior report of the upregulation of the PI3K/AKT/mTOR signaling pathway in the liver induced by aflatoxin²⁶ aligns with our results in birds from the challenge control (Fig. 5e) and Lr3630 + Lr3632 groups (Supplementary Fig. 9), which were exposed to toxin-producing C. perfringens. It is thus likely that neutralization of NetB and alpha toxin by the probiotic-delivered nanobodies is responsible for the lower expression of these pathways in birds treated with NE01 + NE06.

We conclude that under conditions consistent with SCNE in a commercial-like production setting, neutralization of alpha toxin and NetB by probiotic-delivered nanobodies significantly improves performance by avoiding the metabolic burden of sustained inflammation, oxidative stress, and tissue injury resulting from exposure to those toxins. While some of these effects could be ameliorated by metabolites (such as 2,3-butanediol) produced by the probiotic strains, it is the highly specific neutralizing activity against molecular drivers of the pathology that ultimately leads to improved productivity, even outperforming antimicrobial treatment with BMD. These encouraging results open the possibility of applying a similar strategy to treat multiple animal and human diseases mediated by bacterial toxins and other gut antigens.

Methods

The live animal experiment and procedures in this study conform to the Guide for the Care and Use of Agricultural Animals in Agricultural Research and Teaching⁴⁵. IACUC Approval Number: BE2024-018. All birds were euthanized by CO₂ inhalation following an approved facility standard procedure consistent with methods accepted by the American Veterinary Medical Association (AVMA). Chickens were placed at appropriate density in a sealed trailer and exposed to regulated CO₂ for 30 seconds with periodic re-flooding. After vocalization ceased, chickens remained undisturbed for 2 minutes to confirm death.

Study design and sampling schedule

2080 male Ross 708 broilers were randomly assigned to four treatment groups with 20 replicate pens per group and 26 birds per pen (Table 1). Pens were evenly distributed into two rows separated by a central aisle. On day 0, birds in the challenge control and BMD groups were sprayed with distilled water, whereas birds in the NE01 + NE06 group and the Lr3630 + Lr3632 group were sprayed with 1 × 10⁶ CFUs/bird of the corresponding strains. After birds were dry following this application, all birds were vaccinated for coccidiosis (Coccivac B52, Merck) through a commercial spray cabinet. On day 14, birds in the NE01 + NE06 and Lr3630 + Lr3632 groups were fed a target 1 × 10⁶ CFUs/bird dose via drinking water, with birds in the remaining groups receiving distilled water. On days 15, 16, 17, and 28, approximately 1 × 10⁹ CFU/bird C. perfringens CP6³⁵ was provided to all birds as a top dressing on the feed to approximate SCNE under commercial production conditions (NE mortality <5%)¹³. Birds were monitored until day 43. Feed formulations consisted of unmedicated commercial-type broiler starter, grower, and finisher diets compounded with commonly used United States feedstuffs representative of local formulations. The starter diet was fed from day 0 to day 15, the grower diet from day 15 to day 28, and the finisher diet from day 28 to day 43.

Table 1.

Study design

Treatment group	Treatment description	Chicks per pen × No. of replicates	C. perfringens challengea	Coarse Spray day 0 (CFU/bird)	Drinking water day 14 (CFU/bird)
Challenge control	Challenge control	26 chicks × 20 replicates	Yes	Water	Water
BMD	50 g BMD / US ton feedb	26 chicks × 20 replicates	Yes	Water	Water
NE01 + NE06	Nanobody producing strains	26 chicks × 20 replicates	Yes	1 × 106	1 × 106
Lr3630 + Lr3632	Parental strains	26 chicks × 20 replicates	Yes	1 × 106	1 × 106

^aCoccidiosis vaccine on day 0 and 1 × 10⁹ CFU/bird C. perfringens on days 15–17 and 28.

^bFed daily throughout the study.

On days 15, 28, and 43, body weights and feed intake were measured to determine aFCR and weight gain relative to day 0. At each time point d in days, aFCR was calculated as follows:

$$\mathrm{a}\mathit{FCR}\left(d\right)=\frac{\left[\left(FeedIssuethroughDayd\right)-\left(FeedWeighbackthroughDayd\right)\right]}{\left[\begin{array}{c}\left(PenWtthroughDayd\right)+\left(\mathrm{Sacrifice}WtthroughDayd\right)\hfill \\ +\left(MortWtthroughDayd\right)-\left(PenWtonDay0\right)\hfill \end{array}\right]}$$

On days 18, 31, and 43, one bird per pen was removed and euthanized for NE lesion scoring according to the following scale³⁵: 0 = Normal, 1 = Slight mucus covering small intestine, 2 = Necrotic small intestine mucosa, 3 = Sloughed and blood in small intestine mucosa and contents. Jejunum contents and tissue from the same birds used for lesion scoring, as well as one bird per pen euthanized on day 10, were used for metatranscriptomics and histologic analysis. Only samples in the challenge control, NE01 + NE06, and Lr3630 + Lr3630 groups were analyzed, and samples from only 5 birds per treatment were used for histology. Tissue samples from day 10 and day 18 from the jejunum and liver were used for transcriptomics analysis of the host tissues; only 16 randomly selected pens per treatment group were analyzed; all treatment groups were included in the transcriptomics analysis. Cecal contents from one bird per pen were collected on days 10, 18, and 43 for C. perfringens and Eimeria oocyst counts. Three composite fecal samples were collected from 3 pens per treatment group on days 10, 18, 24, 31, 38, and 43 for C. perfringens and Eimeria oocyst counts. Briefly, 3 distinct droppings per pen were collected into one 4 oz Whirl-Pak® and considered one composite sample. The same 3 pens per treatment were used for the collection of fecal samples throughout the study.

All birds that died were inspected for lesions. Birds confirmed to have succumbed to NE were used to calculate NE-associated mortality.

Histological processing and intestinal morphometry

A 3-cm section of jejunum was excised near the yolk stalk diverticulum within three minutes of euthanasia, ensuring a closed intestinal circumference. Each section was immersed in 10% neutral-buffered formalin and trimmed at the distal end to produce 2-mm cross-sections. Samples were processed using standard protocols and paraffin-embedded. The five jejunal samples from each treatment and time point were embedded together and sectioned at 5 μm thickness. Hematoxylin and eosin (H&E) staining was performed on an automated platform (Ventana HE600; Roche Diagnostics, Indianapolis, IN, USA), and sections were coverslipped and digitally scanned at 20X magnification using a Ventana DP600 scanner. Digital slide review and annotation were conducted using NAVIFY® Digital Pathology Software (Roche Diagnostics).

For morphometric analysis, full circular cross-sections were viewed at 1× magnification. Ten anatomically intact villi and corresponding crypts per jejunum were identified and measured following the criteria described by Wilson et al.⁴⁶. Measurements were conducted using Fiji open-source software⁴⁷, with pixel-to-micrometer calibration performed using an AmScope MR400 reference slide scanned at 1×. Villus height and crypt depth were recorded, and villus:crypt ratios were calculated individually (n = 10 per intestine).

Histologic lesions were evaluated semi-quantitatively using a scoring system adapted from Kinstler et al.⁴⁸. Lesions were scored on a six-point ordinal scale: 0 (absent), 1 (minimal), 2 (mild), 3 (moderate), 4 (marked), and 5 (severe). Cumulative inflammation scores were computed as the sum of lesion scores per jejunum, excluding coccidial lesions.

C. perfringens and Eimeria enumeration

To quantify C. perfringens load, CHROMagar C. perfringens plates were prepared from dry powder as recommended by the manufacturer (CHROMagar, Paris, France). Each fecal/cecal sample in the 4 oz Whirl-Pak bag was homogenized using hand massaging. Approximately 0.5–1 g of feces/cecal contents per sample was weighed into a 7-ounce, filtered, Whirl-Pak bag (Thomas Scientific #B01385, NJ, USA). 2× volume of 1× PBS (diluted with water from a 10× stock; ThermoFisher AM9625, MA, USA) was added, and the sample was again homogenized by massaging the bag gently. The filtrate was subsequently removed into a new microcentrifuge tube. After brief vortexing, the filtrate was further serially diluted 10-fold using 1× PBS up to 10-4 dilutions. 100 ml/dilution/sample was plated using the spread plate technique. For growth under anaerobic conditions, the plates were placed in an air-tight box (Rubbermaid Brilliance BPA Free Food Storage Containers with Lids) with an anaerobic gas pack (Millipore-Sigma #68061, St. Louis, MO) and anaerobic indicator strip (Hardy Diagnostics #BR0055B, Santa Maria, CA) and incubated for ~24 h at 37 °C. Colony-forming units (CFU) per gram were further determined from a dilution with countable colonies.

Eimeria oocyst quantification was performed using the PIPER™ Coccidia Assay Kit (Ancera, Inc., Branford, CT, USA) following the manufacturer’s instructions. Fecal or cecal samples were homogenized by manual massage within the collection bag. For fecal samples, 1.0 g of homogenized material (free of bedding) was transferred into a 7 oz Whirl-Pak® filter bag (Thomas Scientific #B01385, NJ, USA); when <1.0 g was available, a minimum of 0.5 g was used, and reagent volumes were adjusted accordingly. A 5× volume of 1 N NaOH (Cole-Palmer SK-80044-64, Vernon Hills, IL, USA) was added to the sample, mixed thoroughly, and incubated at room temperature for 15 min. Subsequently, an equal volume of Sample Additive (P/N ANC-EIM001-02, Ancera Inc.) was added and mixed. For cecal content samples, which were often <1.0 g and fluid in consistency, net weight was calculated by subtracting the tare weight of a pre-labeled empty sample bag. Reagents were scaled based on sample weight using the same 1:5:5 ratio for sample:NaOH:Sample additive. Following incubation and additive mixing, all samples were transferred to fresh 7 oz Whirl-Pak® filter bags for final processing. For each sample, a 1.75 mL or 2 mL microcentrifuge tube was prepared by adding 3 µL of Detection Reagent (P/N ANC-EIM001-04, Ancera Inc.). A 280 µL aliquot of sample slurry was transferred from the filtered portion of the bag into the tube, followed by 20 µL of Ferrofluid (P/N ANC-EIM001-03, Ancera Inc.). During incubation, the PIPER™ instrument was initialized and loaded with a disposable MagDrive™ coccidia cartridge (P/N 5.1.2.LT, Ancera Inc.). The prepared sample was loaded into one of the 12 cartridge lanes, and the assay was initiated through the user interface. Output data included total oocyst count per gram and oocyst size distribution (small, medium, large), automatically reported by the PIPER™ software.

Statistical analyses

Performance endpoints (aFCR and weight gain) were analyzed by study phase as separate responses using a linear model with fixed treatment and block effects. Lesion scores and histologic lesions were modeled using a linear model with fixed treatment, day, and interaction, with a block for lesion scores. Villus:crypt ratios were modeled using a linear mixed model with fixed treatment, day, and interaction as well as a random intercept for sample. C. perfringens and coccidia counts were modeled using a generalized linear or zero-inflated model with a log link, with candidate families including negative binomial (NB2 or NB1) and Poisson, with fixed effects for treatment, day, and interaction as supported. For linear models, fixed-effect inference used Type II (partial) F-tests; for the villus:crypt ratio mixed model, fixed-effect inference used joint fixed-effect tests with Kenward–Roger approximation; for count models, fixed-effect inference used asymptotic Wald joint tests. Pairwise estimated marginal mean contrasts were reported for each fixed effect (excluding block), including simple effects, with both unadjusted and Tukey HSD–adjusted p-values. Models were fitted using the lmerTest (v. 3.1.3)⁴⁹ and glmmTMB (v. 1.1.12)⁵⁰ packages in R.

Metatranscriptomics of jejunal contents

Total RNA was extracted from jejunal content samples using the RNeasy PowerFecal Pro Kit (Qiagen, Hilden, Germany). RNA quality was assessed using the High Sensitivity RNA Tapestation assay (Agilent Technologies Inc., California, USA), and concentration was determined using the AccuBlue® Broad Range RNA Quantitation assay (Biotium, California, USA). Ribosomal RNA was depleted using QIAseq® FastSelect HMR + 5S/16S/23S (Qiagen, Hilden, Germany) per manufacturer’s instructions to enrich for total mRNA. The library was prepared according to the NEBNext® Ultra™ II Directional RNA Library Prep Kit (New England BioLabs Inc., Massachusetts, USA) with Illumina® 8-nt dual-indices, generating libraries of approximately 300 bp. Library concentration was measured by a Qubit 2.0 fluorometer (ThermoFisher, Massachusetts, USA), and quality was assessed using the TapeStation D1000 ScreenTape assay (Agilent Technologies Inc., California, USA). Equimolar pooling of libraries was performed based on quality control values, and sequencing was performed on a DNBSEQ T7 platform, aiming for 40 M paired-end of 150 bp-reads per sample.

Raw reads were quality-trimmed, and DNBSeq adapters were removed using Trimmomatic (v 0.39)⁵¹ with options ILLUMINACLIP:<adapters > :2:30:10 LEADING:20 TRAILING:20 SLIDINGWINDOW:5:20 MINLEN:75. Reads mapping to the host genome (Gallus gallus, GenBank assembly GCA_000002315) were identified with Bowtie2 (v 2.3.5.1)⁵² and removed with SAMtools (v 1.11)⁵³.

Taxonomic and functional profiling was carried out by aligning the trimmed and host-filtered reads to a custom protein database using Kaiju (v 1.7.3)⁵⁴. The database included up to 10 representative proteomes per species for prokaryotic genomes available in the BV-BRC database⁵⁵, as well as fungal, viral, and protozoal genomes in RefSeq⁵⁶. The database comprised 68,974 proteomes and ~280 million protein sequences. The alignment was done in two stages. First, sequencing reads from two samples per treatment and time point were aligned to the full database. From the results, all species in the database that were detected in at least one sample with a relative abundance higher than 1e-5 were identified. Second, reads from all samples were mapped with Kaiju to a reduced database consisting of the proteomes associated with these species (566 species and 3,177,918 protein sequences). Only reads with an alignment score greater than 200 were considered. As described before⁵, when individual reads were mapped to proteins from more than one species in the reference database, the number of aligned base pairs assigned to each species was calculated based on the proportion of aligned base pairs to the corresponding species from unambiguously mapped reads. Relative abundances were estimated from the bases aligned to species in the reference database using metametamerge (v 1.1)⁵⁷ to correct for the proteome size (number of distinct proteins) of each species in the database. The resulting relative abundance tables represent the proportion of transcripts per protein per species. The number of reads mapping to proteins sharing the same RAST functional roles⁵⁸ was used to analyze differences at the gene-function level. Reads mapping to viral proteomes were not considered in the analysis of bacterial taxonomic or functional features.

Differential abundance analyses of taxa and functional roles were carried out with MaAslin2⁵⁹ with day and treatment group as fixed effects. Ordination and beta-dispersion analyses were performed using weighted UniFrac distances calculated at the species level¹⁸. A phylogenetic tree between species in the reduced protein database was constructed with FastTree (v 2.2)⁶⁰ based on a concatenated alignment of 81 core microbial genes generated with UBCG (v 2)⁶¹.

To identify reads mapping to the VHH sequences in NE01 and NE06, host-filtered reads were aligned to the genomes of the two strains with Bowtie2.

Gene set enrichment analysis (GSEA)²² was carried out starting from functional roles ranked by the MaAslin2-determined treatment group coefficients for the comparison of the NE01 + NE06 and Lr3630 + Lr3632 groups to the challenge control group. The grouping of functional roles into subsystems was obtained from the ModelSeed⁶² and BV-BRC databases. The fgsea (v1.32.4)⁶³ R package was used.

Transcriptomics and Pathway Activation Analysis

Total RNA was extracted from jejunum and liver samples collected on days 10 and 18 using the Zymo Direct-zol RNA kit with TRI reagent. Quality control on a Bioanalyzer (Agilent, California) showed a high average RNA integrity number (RIN) of 9.7. mRNA was sequenced using the NovaSeq X Plus Illumina platform with PolyA selection, and a 2 × 150 bp configuration to yield >20 M pair-end reads per sample.

FASTQ files were pre-processed using TrimGalore (RRID:SCR_011847) to remove low-quality sequences, adapter contamination, and low complexity reads. Cleaned reads were pseudo-aligned to the Gallus gallus reference transcriptome (Ensembl GRCg7b, 2022) using Kallisto⁶⁴, generating transcript-level abundance tables. Gene counts were normalized using the median-of-ratios method implemented in DESeq2⁶⁵ to correct for differences in sequencing depth and compositional bias. One outlier sample (JT163A from the BMD group at day 10) was excluded based on deviation in principal component analysis of both raw and normalized data. No additional outliers were detected.

Differential gene expression (DGE) analysis was performed using DESeq2 to evaluate transcriptional changes across treatments. DGE analyses were conducted independently for each tissue and age group combination using the following linear model: Log₂ normalized counts ~ TreatmentGroup + FloorSide, where FloorSide refers to the left or right side of the central aisle in the floor plan. To evaluate gene expression changes over time, we used the following linear model: Log₂ normalized counts ~ Day + TreatmentGroup + Day*TreatmentGroup + FloorSide. The model was fitted separately to data from each tissue type and each of the following pairs of treatments: Challenge control vs. NE01 + NE06, Challenge control vs. BMD, Challenge control vs. Lr3630 + Lr3632. The average Quantitative Pathway Activation (QPA) for the effect of Day across the three comparisons was reported.

Genes with an unadjusted p-value threshold of 0.05 were used to evaluate pathway-level responses using a topology-based QPA algorithm (Biofractal, Portugal). The algorithm integrates strategies from the pathway analysis methods evaluated by Ma et al.⁶⁶ as previously described⁶⁷. The method uses the expression levels of genes, their statistical significance, and their topological importance in the pathway to generate a QPA score for each sample. The QPA scores are tested for statistical significance using a Kruskal–Wallis t-test and adjusted for multiple testing using the Benjamini–Hochberg method⁶⁸, assuming significance at p-adj < 0.20. The QPA score represents the number of standard deviations a given data point lies above (positive score) or below (negative score) the reference mean. The pathway catalog for the QPA method was a customized catalog (Biofractal, Portugal) based on the Reactome database⁶⁹ with additions from the Gene Ontology database⁷⁰ and published literature.

Author contributions

S.M., J.S., C.H., D.G., and A.K. conceived and designed the study. S.M., C.H., L.M.P., L.F.R., D.S., and A.K. conducted the study and acquired the samples and data. A.N.H., J.S., L.M.P., L.F.R., and G.P. carried out bioinformatics analysis and statistics. F.J.H. performed the histology analysis. A.N.H. L.M.P., L.F.R., and G.P. carried out formal analysis. G.P. drafted the manuscript. All authors reviewed and edited the manuscript.

Data availability

Raw metatranscriptomics and transcriptomics reads are available from SRA under accession numbers PRJNA1330878 and PRJNA1330879, respectively. Processed data tables are available from 10.6084/m9.figshare.30275812.

Competing interests

BiomEdit Inc. has pending patent applications based on PCT/US2020/016522 and PCT/US2020/016668 related to the L. reuteri strains used in this work, on which D.G. and A.K. are listed as inventors. A.N.H., S.M., J.S., D.S., D.G., G.P., and A.K. are employed and own profit interest in BiomEdit Inc. The other authors do not have a competing interest.

Supplementary information

The online version contains supplementary material available at 10.1038/s41522-026-00916-w.