Diet Potentially Drives the Differentiation of Eating Behaviours via Alterations to the Gut Microbiome in Infants

Certain infant eating behaviours are associated with adverse health outcomes such as obesity. While a diet consisting of infant formula has been linked to higher-risk eating behaviours and changes in the gut microbiome, little is known about what role the gut microbiome plays in mediating eating behaviours. Using 16S rRNA sequences extracted from 96 fecal samples collected from 58 infants, we identified a subset of bacterial taxa that were more abundant in formula-fed infants, primarily composed of the phylum Firmicutes. The presence of these taxa correlated with a lower drive to eat (i.e., lower food responsiveness). Furthermore, short-chain fatty acid production pathways were significantly more abundant in formula-fed infants, negatively correlated with food responsiveness, and positively associated with relative abundance of the Firmicutes subset. Our results suggest that higher abundances of Firmicutes in formula-fed infants may decrease their food responsiveness through short-chain fatty acid production in the first four months of life. Taken together, these findings suggest a potential role for the infant’s diet in impacting eating behaviour via changes to the gut microbiome, which may lead to the development of novel interventions for the prevention of childhood obesity.

for the infant's diet in impacting eating behaviour via changes to the gut microbiome, which may 23 lead to the development of novel interventions for the prevention of childhood obesity. 24

INTRODUCTION 26
The human gut microbiome comprises trillions of different bacteria that interact to 27 influence an individual's physiology and mental health via immunologic, endocrine, and neural 28 pathways (1). It can protect an individual by barricading pathogenic organisms from colonizing 29 the body, aids in metabolism through processes that promote the breakdown of toxins or vitamin 30 synthesis, and serves a trophic role by maintaining tolerance to antigens in food (1). For infants, 31 their gut microbiota is similarly crucial for health and development, and is impacted by factors 32 such as mode of delivery, exposure to antibiotics or probiotics, and diet (2). 33 Regarding diet specifically, previous literature has suggested that breastfeeding shapes 34 the gut microbiota in neonates through direct introduction of the mother's milk microbiota and 35 prebiotics such as human milk oligosaccharides (HMO) (3). The consensus is that breastfed 36 infants have gut microbiota with lower diversity and lower levels of the phylum Firmicutes 37 compared to formula-fed infants (1,4). Despite the impact of infants' diet on specific taxa in 38 their gut microbiota being explored, research regarding the effects of breastfeeding or formula-39 feeding on inter-microbial communities is lacking. As bacteria exist in complex networks rather 40 than independently, this level of understanding is critical. 41 Additionally, breastfeeding may be associated with lower rates of childhood obesity 42 through the theorized mechanisms of developing healthier food preferences and eating 43 behaviours (5, 6). In infants, eating behaviours can be evaluated through the Baby Eating 44 Behaviour Questionnaire, which measures several different eating behavior profiles: food responsiveness -the extent to which a child indicates an interest in and desires to spend time to eat; satiety responsiveness -the extent to which a child becomes full easily and leaves food 48 when finished eating; slowness of eating -the pace at which the child consumes their food; and 49 general appetite, which correlates with the other metrics (7, 8). 50 An growing body of research indicates that an adult's gut microbial profile may play a 51 key role in their eating behaviours (9). A recent clinical study conducted by Sanmiguel et al. 52 showed that interventions shaping the microbiomes of obese patients led to a reduction in their 53 cravings (9). Other studies have shown that taking probiotic supplements decreases food intake 54 in mice (10), and leads to weight loss in humans (11, 12). This may be because gastrointestinal 55 microbes are incentivized to manipulate their hosts' eating behaviour in order to minimize 56 selective pressures, either by inducing intake of foods that "suppress their competitors", or that 57 "enhance their own fitness" (13). Furthermore, although previous studies support the gut-brain 58 axis model where nervous stimulation by gut bacterial peptides results in activating the vagus 59 nerve to regulate eating behaviours and body weight (14, 15), there remains a lack of research on 60 how bacterial populations are associated with eating behaviours in infants. 61 In this study, we explored relationships between the infant's diet, gut microbiome, and 62 eating behaviours using the "eating behaviour development in infants'' data repository by Rhee et 63 al. Our aim was to examine how different diets influence the diversity and community 64 composition of infants' gut microbiomes, and explore how these microbiomes may relate to 65 infant eating behaviors. Overall, we seek to propose a possible pathway for how the gut 66 microbiota may influence eating behaviours, which could have implications for infants' health.
affects their eating behaviours. More precisely, we predict that formula-feeding is associated and mean scores for each subscale were then calculated (range: 1-5). It generates 4 subscales (an 92 example of a question is provided in brackets): Enjoyment of food (4 items; "My baby seemed 93 contented while feeding"), Food responsiveness (6 items; Even when my baby had just eaten 94 well s/he was happy to feed again if offered"), Slowness in eating (4 items; "My baby took more 95 than 30 minutes to finish feeding"), Satiety responsiveness (3 items; "My baby got full before 96 taking all the milk I thought s/he should have"), and General appetite (1 item). Internal reliability Responsiveness subscale in this sample, analyses of this subscale were omitted. 102 To assess infant dietary intake, we used selected questions from age-appropriate 103 questionnaires developed by the U.S. Center for Disease Control (CDC); at each age point, 104 mothers reported, in the last 7 days, the number of feedings per day of formula or breastmilk. 105 From these data, infants were classified as exclusively breastfed, exclusively formula fed, or 106 mixed. Only infants from the first two groups were included in subsequent analyses. At each age 107 point, the mother reported, from a list of possible signs and symptoms (e.g., diarrhea, fever, 108 vomiting), whether the infant had any health issues in the preceding two weeks. Mothers also 109 reported whether they or their infant had taken any probiotics or antibiotics in the last two weeks. 110

Mothers reported mode of delivery (Caesarean versus vaginal). 111
Fecal samples were collected from mothers and infants at each age point. Fecal samples 112 from the initial cohort were collected using BD Swube TM dual headed. DNA was extracted and 113 the 16S rRNA region was sequenced on an Illumina MiSeq platform using the 515F/806R primer set. Sequencing data was deposited in the European Nucleotide Archive (ENA) at EMBL-EBI under the accession number PRJEB39437 by the University of California San Diego Microbiome 116 Initiative, with all other data recorded in the metadata. 117

Identification of Confounding Variables 118
Potential factors that could influence the infant gut microbiome independent of diet were 119 assessed based on prior knowledge and included maternal and infant intake of probiotics and/or 120 antibiotics in the 2 weeks prior to sample collection, and mode of delivery (vaginal vs 121 Caesarean-section). Associations between these factors, and diet and eating behaviours were 122 evaluated using Fisher's test and Mann-Whitney U test. Their impact on the gut microbiome was 123 assessed based on weighted UniFrac distance and permutational analysis of variance 124 (PERMANOVA) using the vegan package (16). 125

Microbiome Sequences Analysis 126
Unless stated otherwise, the following analyses were performed using QIIME2 127 (v2020.11) and its plugins (17). Exact commands can be found in the supplementary command 128 line script. After demultiplexing, 16S rRNA sequences underwent quality control using DADA2 129 (18). Next, a phylogenetic tree was generated and used to plot an alpha-rarefaction curve to 130 identify the sampling depth at which richness has been fully observed. Taxonomy was assigned 131 using the Greengenes 99% OTU database (19). 132

Alpha and Beta Diversity Calculations 133
Metadata, along with the phylogenetic tree and taxonomy-annotated feature table 134 exported from QIIME2, were imported into R. Ape (20) was used to convert QIIME2's 135 multichotomous tree into a dichotomous one for downstream analyses. Phyloseq (21) and btools 136 were used to calculate alpha and beta diversity metrics for comparing breastfed and formula-fed Inverse Simpson, and Fisher alpha diversity metrics, and Bray-Curtis, Jaccard, weighted 139 UniFrac, and unweighted UniFrac distances for beta diversity. Btools was used to calculate 140 Faith's phylogenetic diversity. Statistical significance was evaluated using the Mann-Whitney U 141 test for alpha diversity and PERMANOVA for beta diversity. 142

Random Forest Classifier 143
Using caret (22) and randomForest (23), a random forest classifier was optimized, 144 trained, and used to predict diet based on genus-level relative abundance. Receiver operating 145 characteristic curves and feature importance were also calculated using these two packages. 146

Co-abundant Clusters Identification 147
Microbial co-abundance at the genus level was calculated for genera that were present in 148 at least twenty percent of the infants. Spearman correlation distance and Ward's linkage were 149 calculated for the centre log ratio-transformed relative abundance values and used to cluster 150 microbes, as previously described by Cirstea et. al (24). The Mann-Whitney U test was used to 151 compare relative cluster abundance between breastfed and formula-fed infants. Spearman 152 correlation was calculated to assess the correlation between cluster relative abundance and eating 153 behaviours. 154

Metabolic Pathways Analysis 155
Inferred functional microbiota profiling was done using PICRUSt2 (v2.3.0b) (25). 156 Differences in the relative abundance of metabolic pathways present in at least five percent of 157 the infants were assessed using ALDeX2 (26). Pathways were deemed to be statistically 158 significantly differentially present when the Benjamini-Hochberg corrected P values for Welch's

Participant Characteristics 163
This study uses data collected from 58 infants at ages 2 weeks, 2 months, and 4 months 164 for a total of 96 samples. These infants were exclusively breastfed or formula-fed in at least the 7 165 days leading up to data collection, and had no vomiting, diarrhea, or fever. Sample 166 characteristics are shown in Table 1. Because the sample size did not provide sufficient power 167 for a linear mixed effects (LME) model with infant ID as a random effect and age as a nested 168 random effect, all samples were treated as independent even if they came from the same infant at 169 different timepoints. Some infants provided only one sample, making a random-slope and 170 random-intercept LME model infeasible. An LME model with only infant ID as the random 171 effect indicated no significant associations between the gut microbiome and eating behaviours. 172 Consequently, we assessed the effect of each factor within our study cohort. As the PCoA plot 180 for weighted UniFrac distance accounted for the most variance compared to those based on 181 Jaccard, Bray-Curtis, and unweighted UniFrac distances (Fig. 1a, Supplementary Fig. 1), we 182 used weighted UniFrac as our metric for evaluating the impact of confounders on the gut 183 microbiome. Additionally, we also ensured that no other confounding variable was associated with the infant's diet. We found that infant gut microbiomes did not cluster differently based on using QIIME2, the sampling depth was set at 14,000 reads. This sampling depth led to 17 infants 207 being excluded from diversity analyses. 208 Next, weighted UniFrac was again used as the beta diversity metric for how infants 209 clustered based on diet (Fig. 1a). Although our PCoA displayed no obvious visible clustering, the 210 gut microbiomes of breastfed and formula-fed infants were significantly different 211 Faith's phylogenetic diversity than formula-fed infants (P = 0.046; Fig. 1b). 216 The fact that breastfed and formula-fed infants differed in terms of weighted UniFrac and 217 multiple alpha diversity metrics suggests differences in taxonomic composition between the two 218 diets. However, instead of merely identifying differentially abundant taxa, we attempted to 219 distinguish between breastfed and formula-fed infants based on relative abundance at the genus 220 separate healthy dogs from those with irritable bowel syndrome (28). With our model, we 222 achieved an area under the curve (AUC) of 0.87 (Figure 1c). These data demonstrate that 223 breastfed and formula-fed infants' gut microbiomes are distinct in genus-level composition, 224 particularly for a phylum reported to be more abundant in formula-fed infants (29). Since most of the genera that best distinguish between infants with different diets 237 belonged to the phylum Firmicutes, we tested if those genera are related functionally. Within the 238 gut, microbes are part of networks that cooperate and compete (30). We inferred the presence 239 and composition of these types of communities based on genus-level coabundance. Spearman's 240 correlations between genera present in at least twenty percent of the infants were calculated and 241 used for clustering into a dendrogram. Covariance was then visualized using a heatmap, and 242 three clusters of high covariance composed of at least three genera were identified 243 (Supplementary Figure 4a). Out of these, only two were found to be differentially abundant 244 between breastfed and formula-fed infants (Supplementary Figure 4b; Figure 2a). 245 The first, Cluster I, is composed of the genera Dorea, Eubacterium, Blautia, and 246 Oscillospira, and is significantly more abundant in formula-fed infants (P < 0.001; Figure 2b). 247

This is in concordance with previous studies reporting decreases in levels of the phylum 248
Firmicutes and order Clostridiales in breastfed infants (31). Blautia regulates G-protein coupled 249 receptors through butyric and acetic acid production, decreasing obesity and visceral fat 250 accumulation (32). Oscillospira is also associated with leanness as it degrades animal-derived 251 glycans from the host (33). Eubacterium is a prolific butyrate producer, breaking down complex 252 carbohydrates from dietary fibers (34). 253 Cluster II, composed of the genera Bifidobacterium, Lactobacillus, Haemophilus, Rothia, Figure 2c). Bifidobacterium is well-known to dominate the microbiota of breastfed infants, with 256 some studies reporting as much as double the relative abundance in breastfed infants compared 257 to formula-fed infants (35). The same study also noted higher levels of Lactobacillus and 258 Streptococcus. Bifidobacterium and Lactobacillus digest dietary fibers and produce acetate (36), 259 while Streptococcus and Veillonella induce cytokine production to modulate the gut immune 260 system (37). Rothia degrades gluten (38). Additionally, research has shown that genera in Cluster 261 II make up a large proportion of the breast milk microbiota (39). The composition of the gut microbiome has been found to impact eating behaviours, but 275 most research has involved adult rather than infant cohorts, and focused on individual taxa rather 276 than microbial communities (13). Therefore, we sought to uncover relationships between the two 277 identified clusters and eating behaviours assessed using the BEBQ. Spearman correlations were 278 calculated between the relative abundances of Clusters I and II, and the five eating behaviours, 279 and then visualized with a heatmap. Cluster I relative abundance was significantly correlated increasing insulin secretion following food intake (44). As such, our results seem to suggest that higher abundances of Cluster I in formula-fed infants may decrease the food responsiveness of 322 these infants through SCFA production. 323 One study suggests that breastfeeding protects against childhood obesity (45), which 324 contradicts our suggested model. An explanation for this contradiction is that high abundances of 325 SCFA-producers, such as Firmicutes, in the stool of formula-fed infants do not always indicate 326 proper SCFA absorption by the infant. SCFA and metabolites may be excreted instead of 327 absorbed, reducing satiety and increasing the risk of obesity (43). Furthermore, prior studies are 328 discrepant with regard to whether breastfed or formula-fed infants show greater SCFA 329 production and absorption, and report that SCFA distributions in infants vary by infant age (43). 330 There is also no literature regarding how SCFA correlates with infant weight and whether 331 breastfeeding protects against childhood obesity remains contested. Therefore, our proposed 332 model, rather than being a dogma, should be treated as a call to further research regarding the 333 relationship between the infant's diet, gut microbiome, and eating behaviours. Our study investigated how the infant's diet impacts their gut microbiota and eating 337 behaviours during the first 4 months of life. We initially hypothesized that formula-feeding 338 would be associated with a higher abundance of the phylum Firmicutes and exhibition of 339 obesity-prone eating behaviours. While we did find that formula-fed infants hosted greater levels 340 of a cluster rich in Firmicutes, these microbes were associated with lower levels of food 341 responsiveness, which would theoretically correspond to a lower obesity risk. Furthermore, we 342 identified the production of SCFAs by the Firmicutes-rich cluster as a mechanism for decreasing 343 food responsiveness. Our model postulates that formula impacts eating behaviours by altering future microbial colonization, and affect longer-term eating behaviors and growth trajectories 346 remain to be seen. However, these results provide a new understanding of the 347 psychophysiological impacts of gut microbial communities in the first 4 months of life, and calls 348 for additional research to be done to better understand how infant diet impacts the development 349 of adult microbial communities and subseqnet growth and development of eating behaviors.