Probiotic Product Enhances Susceptibility of Mice to Cryptosporidiosis

To test whether the probiotic product affects the severity of C. parvum infection, fecal oocyst output was measured by flow cytometry (FCM), as described in Materials and Methods. In experiment 1, a total of 92 oocyst concentration values were acquired from 16 mice at 6 time points over a 15-day period. In experiment 2, 79 data points were obtained from the same number of mice. In experiment 3, 12 time points for 4 groups generated 48 FCM values (Fig. 1 and Table 1). In experiment 1, mice that received probiotics excreted a significantly higher concentration of oocysts. A mean oocyst output of 76,463 oocysts/g feces (n = 44) was measured, compared with a mean of 26,732 oocysts/g for the control mice (n = 48). The difference between treatments was highly significant (Mann-Whitney U = 541, P < 0.001). An analogous significant probiotic effect was obtained for experiment 2 (mean of 378,736 oocysts/g [n = 37] for the probiotic-treated group versus mean of 68,778 oocysts/g [n = 42] for the control group; U = 269.5, P < 0.001). Similarly, the probiotic effect was significant in experiment 3 (Kruskal-Wallis one-way analysis of variance [ANOVA] on ranks, P = 0.001). Figure 1 shows the pattern of normalized oocyst outputs for the 3 experiments, plotted on a logarithmic scale.

To assess whether probiotic treatment affected the fecal microbiota, pairwise weighted UniFrac distances (24) between sequence data for 64 fecal samples from experiment 1 were visualized on a principal-coordinate analysis (PCoA) plot (Fig. 2). Samples collected from day 5 of probiotic treatment (day 4 postinfection [p.i.]) to day 16 of treatment (day 15 p.i.) were included. The number of samples in this analysis was smaller than that shown in Fig. 1 because not all fecal samples were sequenced. Consistent with an impact of probiotic consumption on the profile of the intestinal microbiota, this analysis revealed a nonoverlapping distribution of data points according to treatment. Analysis of similarity (ANOSIM) (25) was used to check the significance of the treatment effect. The test returned a highly significant R value of 0.305 (P < 10⁻⁵). Analogous results were obtained for experiment 2, based on 55 samples collected from day 10 after initiation of probiotic treatment (day 9 p.i.) to day 20 (day 19 p.i.). As for experiment 1, clustering by treatment was significant (R = 0.210, P < 10⁻⁵). In experiment 3, microbiota clustering by treatment was also significant (R = 0.09 [n = 34], P = 0.002).

Having detected effects of probiotic consumption on oocyst output and the fecal microbiota, the FCM and small-subunit 16S rRNA data were analyzed jointly to identify possible associations between the bacterial microbiota profile and the severity of cryptosporidiosis. These analyses included all samples for which 16S rRNA and FCM data were acquired. A total of 44 samples were included in experiment 1 and 53 in experiment 2. Because dysbiosis typically is characterized by low bacterial diversity and often is associated with increased susceptibility to enteric infections (26), the first global analysis examined the correlation between microbiota α-diversity (within-sample diversity) and total oocyst output (Fig. 3). Regardless of whether data were pooled by experiment or samples from probiotic-treated mice and control mice were analyzed separately, the correlation between α-diversity and oocyst output was very weak, explaining at most 9% of the oocyst count. These analyses indicate that the gut of mice excreting large numbers of oocysts generally is not populated with less diverse microbiota.

Redundancy analysis (RDA) was used to assess whether fecal oocyst concentrations significantly correlated with microbiota profiles and to identify bacterial OTUs correlating in relative abundance with the oocyst output. In experiment 1, a Monte Carlo test with 44 samples collected over the same time period as shown in Fig. 1 and with 1,000 permutations indicated a significant correlation between the oocyst concentration and the OTU profile (pseudo-F = 1.0, P = 0.014). This effect remained significant (F = 2.7, P = 0.024) after accounting for treatment, i.e., defining treatment (with or without probiotic) as a covariate, or accounting for “mouse,” i.e., defining mouse as a covariate (pseudo-F = 3.8, P = 6 × 10⁻⁵). The analogous test in experiment 2 also returned a significant pseudo-F ratio of 4.2 (n = 44; P = 0.0016). When the effect of treatment or mouse was removed by defining these variables as covariates, the association remained significant (pseudo-F values of 2.6 [P = 0.015] and 5.2 [P = 3 × 10⁻⁴], respectively). These and subsequent analyses were not performed for experiment 3 because samples from mice in the same group were pooled.

Reasoning that the microbiota at the time of infection could be important for subsequent parasite proliferation, we analyzed whether the microbiota in treated and control mice differed on the day of infection, i.e., 1 and 2 days after initiation of microbiota administration in experiments 1 and 2, respectively. In both experiments, a significant correlation between probiotic treatment and the composition of the bacterial fecal microbiota was observed (pseudo-F values of 5.3 [P = 0.028] and 3.5 [P = 0.055] in experiments 1 and 2, respectively). The permutation test is explained in Materials and Methods.

With a significant correlation between the fecal oocyst concentration and the OTU profile having been identified, the taxonomic composition of the fecal microbiota was examined in more detail. First, we used the program LEfSe (27) to identify OTUs that significantly defined the difference between samples with high versus low oocyst concentrations. This analysis was based on 20 samples for each experiment, namely, 10 samples with the highest oocyst concentrations and 10 samples with the lowest concentrations collected from the entire experiment. In experiment 1, 7/10 samples in the high-oocyst group originated from mice treated with probiotics and 7/10 samples in the low-oocyst group came from control mice (chi-square = 3.2, P = 0.07). In experiment 2, 9/10 samples in the high-oocyst group originated from treated mice and 8/10 samples in the low-oocyst group originated from control mice (chi-square = 9.9, P = 0.002). Feces from highly infected animals were characterized by a high Proteobacteria abundance, whereas feces from lightly infected animals were significantly enriched for Firmicutes. As observed for the effects of the probiotic on the severity of cryptosporidiosis (Fig. 1) and on the global microbiota profile (Fig. 2), a similar shift toward higher Proteobacteria abundance was observed in the 2 experiments (Fig. 4).

A second taxonomic analysis aimed at identifying OTUs enriched in highly infected animals was performed with RDA (28). For experiment 1, in the 10 OTUs that best correlated in relative abundance with the oocyst concentration, 61% of the sequences were classified as Lactobacillus, 24% as Proteus, and 14% as Enterococcus. In the 10 OTUs that most negatively correlated with the oocyst concentration, as seen with LEfSe, Blautia was the most abundant classification (61% of reads), followed by Clostridiaceae (27%), Lachnospiraceae (10%), Romboutsia (8%), and unclassified Firmicutes. For experiment 2, the same analysis of the 10 OTUs that best correlated in relative abundance with the oocyst concentration indicated that 92% of sequences originated from Enterobacteriaceae and 8% from Firmicutes. In the 10 OTUs with the lowest oocyst concentrations, 70% of sequences were classified as Lactobacillus, 21% as Turicibacter, 6% as unclassified Lactobacillales, and 1% as Stenotrophomonas (phylum Proteobacteria) (see Table S1 in the supplemental material). The results obtained with RDA are thus is close agreement with the LEfSe results shown in Fig. 4.

Figure 3 illustrates an important difference between the experiments; the mean fecal oocyst concentration, across all time points, treatments, and mice, in experiment 2 was 2.4 × 10⁵ oocysts/g (standard deviation [SD], 3.4 × 10⁵ oocysts/g [n = 79]), or about 5 times higher than that in experiment 1 (mean, 5.0 × 10⁴ oocysts/g; SD, 8.7 × 10⁴ oocysts/g [n = 92]) and >8 times higher than that in experiment 3 (mean, 2.8 × 10⁴ oocysts/g; SD, 4.6 × 10⁴ oocysts/g [n = 48]). Oocyst outputs were significantly different between experiments (Kruskal-Wallis ANOVA on ranks, H = 34.3 [2 degrees of freedom], P < 0.001). Pairwise comparisons between experiment 1 and experiment 2 and between experiment 3 and experiment 2 were significant, according to Dunn's multiple-comparison test, whereas that between experiment 1 and experiment 3 was not. It is likely that these differences were caused by the dexamethasone administration route (drinking water only in experiments 1 and 3 versus drinking water followed by subcutaneous injection in experiment 2) (Table 1). The more severe infection in experiment 2 represented an unplanned opportunity to further assess the impact of the infection on the gut environment. If heavier infections cause an increase in the relative abundance of Proteobacteria, as suggested by the LEfSe analysis (Fig. 4) and by the RDA described above, then a greater proportion of Proteobacteria in samples originating from severely infected mice in experiment 2 would be expected, which is exactly what was observed (Fig. 4 and RDA results). Furthermore, severe cryptosporidiosis can be expected, based on research involving human patients suffering from other enteric infections (29–31), to lead to changes in bacterial microbiota toward populations enriched for facultative anaerobes, possibly resulting in loss of diversity. To test this hypothesis, mean pairwise weighted UniFrac distances for microbiota from heavily infected samples were compared to distances for lightly infected samples. In experiment 1, the mean β-diversity (between-sample diversity) among the 10 high-oocyst samples was 0.536 (SD, 0.142 [45 pairwise distances]), whereas the mean β-diversity among the low-output samples was 0.590 (SD, 0.121 [45 pairwise distances]; Mann-Whitney T = 1819.0, P = 0.07). In experiment 2, the mean β-diversity values were 0.312 (SD, 0.117 [45 pairwise distances]) and 0.411 (SD, 0.137 [45 pairwise distances]) for the 10 samples with the highest and lowest oocyst concentrations, respectively (Mann-Whitney U = 590, P < 0.001). Although the effect was not significant for experiment 1, together these results are consistent with the model postulated above, i.e., that severe infection leads to a convergence of the microbiota toward a population rich in Proteobacteria and low in Firmicutes.

To extend the observed taxonomic differences between severely and lightly infected mice to the metagenome, the program PICRUSt (32) was used to infer microbiota function from OTU profiles. Given the more severe infection in experiment 2, metagenomic analyses are reported only for this experiment. PICRUSt identified 39 KEGG level 2 categories in the combined metagenome. A PCoA based on pairwise distances between KEGG abundance values normalized across KEGG categories revealed a tight clustering of samples with high oocyst concentrations, relative to the samples with lower oocyst concentrations (Fig. S1). This visual assessment was tested by comparing pairwise distances between KEGG profiles. The distances for the 10 samples with the lowest oocyst concentrations averaged 69.9 (SD, 47.6) and those for the same number of samples with the highest concentrations averaged 54.8 (SD, 41.41), which was not statistically significant (Mann-Whitney rank sum test, P = 0.121). If only 8 samples with the highest or lowest oocyst concentrations were tested (28 pairwise distance values for each group), then the distances between the high-concentration samples were significantly smaller (Mann-Whitney rank sum test, P = 0.015). As for the taxonomic analysis described above, we conclude from these results that proliferation of C. parvum leads to a convergence of the inferred bacterial metagenome.

As described above for the taxonomic analysis, LEfSe was used to identify KEGG pathways that differed significantly in abundance between the 10 samples from highly infected mice and the same number of samples from lightly infected mice. Underscoring the difference at the metagenomic level between fecal samples from severely and mildly infected mice, 22 of 39 KEGG level 2 pathways were significantly different between the 2 groups (Table S2). In comparison to the severely infected mice, the microbiota of mice with mild infections was characterized by a high abundance of pathways related to replication, such as carbohydrate, amino acid, and nucleic acid metabolism. These results extend the taxonomy presented in Fig. 4, suggesting that the mouse dysbiotic cryptosporidiosis metagenome is selected for functions other than bacterial replication.

Oocysts from C. parvum isolate MD (51) were used in experiments 1 and 3, whereas mice in experiment 2 were infected with C. parvum isolate TU114 (52). MD is a zoonotic isolate, and TU114 belongs to the anthroponotic subgroup characterized by an IIc GP60 genotype (53). Oocysts were purified from mouse feces on gradients of Nycodenz (Alere Technologies, Oslo, Norway), as described previously (54). The ages of the oocysts were 13, 110, and 62 days for experiments 1, 2, and 3, respectively.

To test the effect of a commercially available probiotic, 3 experiments were conducted in mice. In each experiment, 16 female CD-1 mice approximately 6 weeks of age were used. Upon delivery, each mouse was individually tagged and randomly assigned to 1 of 4 groups of 4 mice. Mice were immunosuppressed by adding dexamethasone 21-phosphate disodium (catalog no. D1169; Sigma) to the drinking water, at a concentration of 16 mg/liter (55), starting on the day of arrival, defined as day −7 p.i. (i.e., 7 days prior to infection). To deplete the native intestinal microbiota, vancomycin and streptomycin were added to the drinking water at concentrations of 500 mg/liter and 5 g/liter, respectively, starting on day −6 p.i. Metronidazole at a dose of 20 mg/kg was given daily by gavage, starting on day −6 p.i. The antibiotic treatment was discontinued on day −2 p.i. Starting on day −1 p.i., the drinking water was supplemented with 1.3 g of probiotic added to 500 ml of water. The product contained 15 bacterial strains, including strains belonging to the genera Bifidobacterium (4 species) and Lactobacillus (9 species), as well as Streptococcus thermophilus. In addition to bacteria, 1 g of product contained 8 mg acacia gum, 540 mg larch gum, 115 mg galacto-oligosaccharide, 212 mg l-glutamine, and 150 IU vitamin D₃; when the product was dissolved in 500 ml of water, the final concentrations of these ingredients were 20 μg/ml, 1.4 mg/ml, 0.3 mg/ml, 0.55 mg/ml, and 0.4 IU/ml, respectively. The product was flavorless. Drinking water with dexamethasone and probiotic was replaced every 3 days. Mice were infected orally with approximately 5 × 10⁴ C. parvum oocysts on day 0 p.i. In experiment 1, to compensate for increased water uptake in the 2 groups receiving probiotic, the concentration of dexamethasone phosphate in the water of the 2 treated groups was reduced to 10 mg/liter starting on day 10 p.i. In experiment 2, to avoid possible differences in dexamethasone uptake with drinking water, dexamethasone (catalog no. D1756; Sigma) was given only subcutaneously every second day starting on day 2 p.i., following 9 days of dexamethasone administration in the water (Table 1). A volume of 100 μl of a 10-mg/ml dexamethasone suspension was injected alternately into the right and left sides of the abdomen. In experiment 3, 16 mg/liter dexamethasone was added to the drinking water of all 4 groups of mice. To obtain fecal pellets for analysis of the microbiota, mice were individually transferred to a 1-liter plastic beaker and pellets were collected upon defecation. In experiment 3, feces were collected by group, not by mouse. Pellets were stored at −20°C. To collect feces for oocyst enumeration, mice were individually transferred overnight to collection cages fitted with a wire bottom. Feces collected from these cages were stored at 4°C. In the morning, the mice were returned to their respective group cages, such that they were housed individually for 14 to 16 h on the days when feces were collected for oocyst enumeration. While in conventional cages, the mice were always housed with the same cage mates. Animal experiments adhered to the NIH Guide for the Care and Use of Laboratory Animals and were approved by the Tufts University Animal Care and Use Committee.

Feces collected overnight were weighed, diluted 1:5 in distilled water, and homogenized with a vortex mixer. To remove debris, fecal homogenates were filtered through 100-μm cell strainers (catalog no. 431752; Corning) by centrifugation at 1,300 × g. The filtrates were homogenized, and volumes of 1 ml were centrifuged at 6,700 × g for 5 min. The supernatant was discarded, and the pellet was suspended in 500 μl of phosphate-buffered saline (PBS) supplemented with 10% fetal bovine serum (FBS). A volume of 20 μl of this suspension was transferred to a 1.5-ml microcentrifuge tube with 20 μl of a 1:5 dilution of monoclonal antibody 5F10 cell culture supernatant. This antibody reacts specifically with the Cryptosporidium oocyst wall without binding to other parasite antigens (A. S. Sheoran, unpublished observation). Samples were incubated for 30 min at room temperature. Following incubation with the primary antibody, the samples were centrifuged for 10 min at 6,700 × g, and the pellet was washed once in 500 μl of PBS with 10% FBS and incubated for 30 min at room temperature with 20 μl of secondary antibody (Alexa Fluor 488-conjugated goat anti-mouse IgG). After incubation, 500 μl of PBS was added and the samples was washed once in PBS. For each experiment, 5 samples were randomly selected for replication. Replication involved processing and labeling 5 separate aliquots originating from each strained and washed sample. Labeled samples were analyzed by FCM using a Becton Dickinson Accuri C6 cytometer. Distance matrices were calculated with GenAlEx 6.5 (56), based on the pairwise difference between oocyst concentrations. Specifically, the distance between sample A containing an oocyst concentration of x_A per gram of feces and sample B containing a concentration of x_B per gram of feces was calculated as (x_A − x_B)². The accuracy of the FCM oocyst enumeration method was evaluated by correlating the counts against the Cryptosporidium 18S rRNA gene copy number, as estimated using real-time PCR with published primers (57). This analysis based on 5 randomly chosen samples from experiment 1 generated a correlation coefficient of 0.69.

DNA was extracted from 10 mg of feces collected individually from each mouse. DNA was extracted in a QIACube instrument using the QIAamp PowerFecal DNA kit (catalog no. 12830-50; Qiagen), according to the manufacturer's protocol. DNA was eluted in 50 μl of elution buffer and stored at −20°C. A previously described PCR protocol to prepare 16S V1V2 amplicons libraries for high-throughput sequencing was used (5). The only deviation from this procedure was the downstream primer; instead of the canonical primer 338R, we used primer Bac R V2 short (GTTCAGACGTGTGCTCTTCCGATCtgctgcctcccgtaggagt, with the lowercase letters being equivalent to the conserved 338R sequence) (58). One microliter of primary PCR mixture was subjected to a secondary PCR to incorporate a 6-nucleotide barcode into each sample. The secondary amplification was as described previously (5), except that the downstream primer CAAGCAGAAGACGGCATACGAGATnnnnnnGTGACTGGAGTTCAGACGTGTGCTCTTCC was used; the italicized nucleotides represent the Illumina adaptor and the lowercase letters the barcode unique to each sample. To assess the quality of the PCRs, a portion of the amplification product was electrophoresed on a 1.5% agarose gel and visualized using GelRed (Biotium). The concentration of each final amplicon was measured in a Qbit spectrophotometer, and up to 80 amplicons were pooled at approximately equal concentrations. The pooled library was size selected with a Pippin HT library size selection system (Sage BioScience). Libraries were sequenced in an Illumina MiSeq sequencer at Tufts University Genomics Core Facility, using a single-end 300-nucleotide strategy. To control for technical variations introduced during library preparation and sequencing, each library included 2 replicates of 2 randomly chosen samples. Replication involved the process of duplicate fecal samples being processed, amplified, and barcoded individually.

FASTQ-formatted sequences were processed using programs found in mothur (59), essentially as described previously (5). Briefly, random subsamples of 5,000 sequences per sample were processed. This procedure is not expected to bias the analysis (60), since the numbers of sequences per barcode were relatively constant. The mean number of sequences per sample in experiment 1 was 1.03 × 10⁵ sequences/sample (SD, 2.44 × 10⁴ sequences/sample) and the average number of reads in experiment 2 was 1.01 × 10⁵ reads/sample (SD, 3.22 × 10⁴ reads/sample), representing coefficients of variation of 0.23 and 0.31, respectively. V1V2 sequences were trimmed to 200 nucleotides to eliminate 3′ sequences with mean Phred quality scores of <30. Sequences were aligned and sequences with the following properties were removed: sequences that did not align, sequences with ambiguous base calls, and sequences with homopolymers of >8 nucleotides. To further reduce the number of putative sequence errors, the program pre.cluster was used to merge unique sequences differing by 1 nucleotide position with the majority sequence (61). This sequence curation protocol removed 15,748 sequences from 400,000 sequences (3.9%) in experiment 1 and 21,546 sequences from 355,000 sequences (6.0%) in experiment 2. Pairwise UniFrac phylogenetic distances (24) between samples were calculated in mothur. ANOSIM (25) was used to test the significance of clustering by treatment. The program anosim was run in mothur using a weighted UniFrac distance matrix as input. OTUs were obtained using the program cluster, also found in mothur, using the OptiClust clustering method (62) and a distance cutoff value of 3%.

Linear discriminant analysis, as implemented in the program LEfSe (27), was used to identify statistically significant differences in OTU abundance profiles between 2 groups of samples, defined, independent of treatment, as heavily infected and lightly infected. Heavy and light infections were defined on the basis of fecal oocyst concentrations determined by FCM as described above. In the experiment-wide LefSe analyses, the 10 samples with the highest oocyst concentrations and the 10 samples with the lowest oocyst concentrations were selected. When subsamples of probiotic-treated and control mice were analyzed separately, the 5 samples with the highest oocyst concentrations and the same number of samples with the lowest oocyst concentrations were tested using LEfSe (27).

RDA was used to test the significance of associations between OTU profiles and oocyst concentrations or between KEGG function profiles and oocyst concentrations. The program was run in CANOCO (28). The pseudo-F statistic was calculated by Monte Carlo permutation of samples between treatment groups. OTU abundance values for the 100 most abundant OTUs or KEGG functions (n = 39) inferred with the metagenome prediction tool PICRUSt (32) served as dependent variables. Oocyst concentrations determined by FCM, as described above, served as the independent variable. Pairwise distances between KEGG profiles were calculated as described above for oocyst concentrations, except that the squared differences were summed over all KEGG categories; this metric is equivalent to the square of the Euclidean distance.

Raw sequence data were deposited in the ENA Sequence Read Archive under study accession numbers PRJEB25162 and PRJEB25164.