Bifidobacterium and Lactobacillus Probiotics and Gut Dysbiosis in Preterm Infants: The PRIMAL Randomized Clinical Trial

Abstract

Importance: The effects of probiotic interventions on colonization with resistant bacteria and early microbiome development in preterm infants remain to be clarified.

Objective: To examine the efficacy of Bifidobacterium longum subsp infantis, Bifidobacterium animalis subsp lactis (BB-12), and Lactobacillus acidophilus (La-5) probiotics to prevent colonization with multidrug-resistant organisms or highly epidemic bacteria (MDRO+) and to shape the microbiome of preterm infants toward the eubiotic state of healthy full-term infants.

Design, setting, and participants: The multicenter, double-blinded, placebo-controlled, group sequential, phase 3 Priming Immunity at the Beginning of Life (PRIMAL) randomized clinical trial, conducted from April 2018 to June 2020, included infants with gestational age of 28 to 32 weeks at 18 German neonatal units. Data analyses were conducted from March 2020 to August 2023.

Intervention: A total of 28 days of multistrain probiotics diluted in human milk/formula starting within the first 72 hours of life.

Main outcomes and measures: Colonization with MDRO+ at day 30 of life (primary end point), late-onset sepsis and severe gastrointestinal complication (safety end points), and gut dysbiosis, ie, deviations from the microbiome of healthy, term infants (eubiosis score) based on 16-subunit ribosomal RNA and metagenomic sequencing.

Results: Among the 643 infants randomized until the stop of recruitment based on interim results, 618 (median [IQR] gestational age, 31.0 [29.7-32.1] weeks; 333 male [53.9%]; mean [SD] birth weight, 1502 [369] g) had follow-up at day 30. The interim analysis with all available data from 219 infants revealed MDRO+ colonization in 43 of 115 infants (37.4%) in the probiotics group and in 39 of 104 infants (37.5%) in the control group (adjusted risk ratio, 0.99; 95% CI, 0.54-1.81; P = .97). Safety outcomes were similar in both groups, ie, late-onset sepsis (probiotics group: 8 of 316 infants [2.5%]; control group: 12 of 322 infants [3.7%]) and severe gastrointestinal complications (probiotics group: 6 of 316 infants [1.9%]; control group: 7 of 322 infants [2.2%]). The probiotics group had higher eubiosis scores than the control group at the genus level (254 vs 258 infants; median scores, 0.47 vs 0.41; odds ratio [OR], 1.07; 95% CI, 1.02-1.13) and species level (96 vs 83 infants; median scores, 0.87 vs 0.59; OR, 1.28; 95% CI, 1.19-1.38). Environmental uptake of the B infantis probiotic strain in the control group was common (41 of 84 [49%]), which was highly variable across sites and particularly occurred in infants with a sibling who was treated with probiotics.

Conclusions and relevance: Multistrain probiotics did not reduce the incidence of MDRO+ colonization at day 30 of life in preterm infants but modulated their microbiome toward eubiosis.

Trial registration: German Clinical Trials Register: DRKS00013197.

Figure 1.. Enrollment, Randomization, and Follow-Up in the Priming Immunity at the Beginning of Life (PRIMAL) Randomized Clinical Trial

The study flow diagram describes the study design. The lack of adherence to block randomization toward the end of the enrollment period was mainly related to shortage of boxes for intervention for specific strata. The boxes had been prepacked by the study pharmacy, shipped to the sites in 1 load, and stored on site until use. Site investigators had chosen remaining boxes from different gestational age or sex strata than set by the study protocol in 110 infants (18%). One sibling pair was mixed and did not receive the allocated intervention. GCP indicates Good Clinical Practice.

Figure 2.. Probiotics, the Preterm Microbiome, and the Eubiotic State Typical of Full-Term Infants

Eubiosis modeling was created for the first time by using published metagenomes to distinguish between preterm microbiomes (339 metagenomes of probiotic-naive infants from 5 studies) and healthy full-term microbiomes (153 metagenomes from 7 studies). The model produces a eubiosis score from 0 to 1, ie, higher values reflect the probability of a eubiotic (healthy) state and, therefore, less likely to be dysbiotic. At both species and genus level, this model distinguished preterm infant microbiomes from healthy, full-term microbiomes, with an accuracy of greater than 90% and an area under the curve of greater than 0.93 under cross-validation grouped by study. To assess whether differences in eubiosis scores were due to the presence of the probiotic bacteria themselves or due to other community composition differences, a second set of models was trained. Here, the modeling was repeated, but the probiotic taxa (Bifidobacterium longum, Bifidobacterium animalis, and Lactobacillus acidophilus) were excluded from model building. The resulting species and genus models distinguish appropriately between the preterm and full-term groups (accuracy >88%; area under the curve >0.92). The most important taxa for these models are the same as the previous models, without Bifidobacterium. These models were then also applied to the PRIMAL day 30 fecal microbiomes. A, Eubiosis score at genus resolution in the 16-subunit (S) dataset, compared among verum, control, and the healthy term EMMA (Impact of Mother’s Own Milk on the Development of Allergy and Airway Infections) trial cohort samples on day 30 of life. Infants in the verum group had higher eubiosis scores than infants in the control group but did not reach the maturation level of exclusively breastfed term infants at day 30. B, Eubiosis score at species resolution in the metagenomic (metaG) dataset, compared among verum, control, and the healthy term EMMA cohort samples on day 30 of life. C, Feature importance analysis of the genus resolution random forest model. Bar chart displaying the 12 most important features used by the random forest model to calculate the eubiosis score. The model was trained on published metagenomes from preterm and healthy term infants. The x-axis represents the feature importance. The color shades indicate whether the genus reported was more abundant in term infants or preterm infants in the published metagenomic datasets. Shannon diversity of those samples also contributed to the prediction score, with higher Shannon index found in healthy term infants. D, Alpha diversity, measured by the Shannon Index in the 16S dataset, compared among the verum, control, and the healthy groups, exclusively breastfed term infants (EMMA cohort samples on day 30 of life). ^aWilcoxon test significance denoted as (false discovery rate = 0.001). ^bWilcoxon test significance denoted as (false discovery rate = 2.2 × 10⁻¹⁶), explorative values.

Figure 3.. Analysis of Probiotic Uptake via Microbiome Profiling Demonstrating Environmental Acquisition of Probiotic Bifidobacterium infantis Strain

Environmental uptake of probiotic B infantis strain in placebo infants mainly related to multiple birth (verum-treated sibling) and hospital site. Cross-colonization of control infants with Bifidobacteria has been suggested in the PIPS (Probiotics in Preterm Infants) trial without provision of firm evidence for this. We analyzed a subset sample undergoing metagenomic sequencing (n = 184) and found probiotic bacteria more prevalent in the verum group than in the placebo group. The abundance of the B infantis probiotic strain (PS), when present, was not significantly different between the verum and control group (Wilcoxon P = .20). B animalis and Lactobacillus acidophilus were observed in 65% and 49% of the verum group samples, respectively, and were rarely observed in the control group (<5%) (eTable 13 in Supplement 3). All 3 probiotic strains were observed in 43 of 100 infants in the verum group and 2 of 84 infants in the control group. We next addressed potential causes of B infantis PS environmental uptake in control group infants, which varied greatly across hospitals (10%-100%). A verum-treated sibling (twin or triplet) caused environmental uptake in 21 of 23 control siblings (90%). Exposure units were 3 times higher for B infantis PS-positive control infants than for PS-negative control infants (mean [SD], 25.5 [11.4] vs 7.9 [10.5] units). A, Percentage of infants per hospital wherein the PS were detected, compared between treatment groups. Each hospital is indicated by a circle, where green represents the control group, purple represents the verum group, and the size of the circle indicates the number of infants per hospital in each group. B, Abundance of B infantis PS between the verum and control groups. We performed high-resolution (single-nucleotide variant–level) detection of PS and estimated differences in the abundance between the groups using the Wilcoxon rank sum test. The control group is further divided into infants where the B infantis PS was detected (control B infantis PS+) and infants where it was not detected (control B infantis PS−). Abundance is measured as the mean depth of coverage of reads mapped to the reference genome divided by the total number of reads per sample normalized by z score. C, Each horizontal bar represents the percentage of control-group infants with B infantis PS detected in each hospital. Only the 8 hospital sites with infants having metagenomic sequenced samples are shown. Colors indicate whether B infantis PS was detected or not. The yellow bar displays the percentage of B infantis PS+ infants who had a sibling in the verum group. D, Comparison of probiotic exposure units between the control group infants with B infantis PS detected (control B infantis PS+) and those where it was not detected (control B infantis PS−). ^aWilcoxon test P < .001, explorative values.