Abstract
Objective
The incidence of women entering into pregnancy with BMI indicating overweight or obesity is rising with concurrent increases in adverse complications such as gestational diabetes. Although several studies have examined the compositional changes to the microbiome across BMI classifications, there has been no investigation regarding changes in microbial function during pregnancy.
Methods
A total of 105 gastrointestinal microbiome biospecimens were used in this analysis. Biospecimens were sequenced by using the Illumina NovaSeq 6000 shotgun metagenomics platform.
Results
Findings indicate an enrichment in microbiota from the phylum Firmicutes across all pregravid BMI groups with a decrease in α diversity in groups with BMI indicating obesity or overweight compared with a group with BMI indicating normal weight (p = 0.02). More specifically, women with BMI indicating obesity or overweight had enrichment in Bifidobacterium bifidum and B. adolescentis. Women with BMI > 25 kg/m2 had a higher abundance of microbiota that support biotin synthesis and regulate epithelial cells in the lower gastrointestinal tract. These epithelial cells are responsible for host adaptability to dietary lipid variation and caloric absorption.
Conclusions
Our analysis suggests that there are differences in microbial composition and function between BMI groups. Future research should consider how these changes contribute to specific clinical outcomes during pregnancy.
Study Importance.
What is already known?
Several studies, using 16S ribosomal RNA gene sequencing, have already shown gastrointestinal microbial changes in the setting of pregravid obesity.
Little is known about the impact of pregravid obesity on changes to microbial function in the gastrointestinal tract during pregnancy.
What does this study add?
By using shotgun metagenomic sequencing, we were able to elucidate microbial composition and functional changes in the late second trimester associated with pregravid obesity.
Our functional analysis found biotin synthesis to be enriched in women with BMI indicating obesity or overweight, a finding yet to be published in relation to the gastrointestinal microbiome.
How might these results change the direction of research?
Our results indicate a clear need to move beyond microbial taxonomic classification to include microbial function as a variable of interest.
INTRODUCTION
Obesity, defined as body mass index (BMI) greater than 30 kg/m2, is a significant public health concern affecting almost 93 million adults in the United States [1]. When examining the age‐adjusted prevalence of obesity among adults over 20 years old, women are more affected than men, with Hispanic women and non‐Hispanic Black women experiencing significantly higher rates of obesity compared with non‐Hispanic White adults [1]. Women with obesity across the life course are at increased risk for metabolic syndrome, cardiovascular disease, type 2 diabetes mellitus, and cancer [2]. However, obesity in reproductive age is of great public health concern because of the increased risk of pregnancy complications such as preterm labor, gestational diabetes, and preeclampsia [2, 3]. Further, the critical period of pregnancy reflects a vulnerable state for both maternal and child health in which morbidity may instantiate adverse short‐ and long‐term outcomes.
Modifiable (e.g., diet, exercise) and nonmodifiable factors (e.g., genetics) influence the prevalence of maternal obesity and the health of a woman during her pregnancy. One recent area of exploration is the maternal gastrointestinal microbiome as an underlying mechanism for pregnancy complications with the exposure of increased BMI. The microbiome refers to all of the living organisms that live in and on the body (e.g., bacteria, fungi, viruses) that provide genes to support essential physiological processes [4]. Although the association between increased maternal BMI and pregnancy complications has been well documented, little is known about the microbial environment of the maternal gastrointestinal tract as a possible mechanism for increased risk.
Clinical research has revealed distinct changes in the maternal gastrointestinal microbiome as a result of maternal pregravid obesity when assessed by 16S ribosomal RNA (rRNA) gene sequencing [5, 6, 7, 8, 9]. A systematic review of maternal gastrointestinal microbiome composition found that there were substantial inconsistencies in 16S rRNA gene sequencing studies regarding reported significant microbiota between mothers with obesity and mothers with a normal BMI [10]. For example, the phylum Bacteriodetes, a highly abundant phylum in the gastrointestinal tract with a broad metabolic contribution, was increased in women with high BMI in two studies [5, 6] and was decreased in another [11]. An increased ratio of Bacteriodetes to Firmicutes was shown in mice and in humans with BMI > 25 [12]. Further, the review revealed the need for future studies to examine both the composition and function of microbiota present in the maternal gastrointestinal tract to identify key drivers for metabolic alterations during pregnancy [10]. As such, shotgun metagenomic sequencing is necessary to examine both these characteristics. The advantage of shotgun metagenomics in comparison to 16S rRNA gene sequencing is the ability to ascertain the functional relevance of microbiota within communities without relying on prediction. Microbial function can be defined in several ways, including as the molecular contribution determined by using the Kyoto Encyclopedia of Genes and Genomes (KEGG) database or as the genetic contribution determined by using Gene Ontology (GO) terms sourced from the GO knowledgebase, the largest database on the function of genes used in large‐scale biomedical research [13, 14, 15]. Examination of microbiome biospecimens during pregnancy using metagenomic sequencing is necessary to advance our biological knowledge and clinical translation.
The purpose of our study was to compare the composition, diversity, and function of maternal gastrointestinal microbiota collected at the time of glucose challenge testing (gestational diabetes screening) among women with normal weight (BMI < 24.9), overweight (BMI = 25.0–29.9), or obesity (BMI > 30) by using next‐generation shotgun metagenomic sequencing. Our central hypothesis was that the gastrointestinal microbiome biospecimens of women with BMI indicating overweight or obesity BMI will have reduced microbial diversity, as measured by α and β diversity, compared with the biospecimens of women with lower BMI. Further, we hypothesized that the functional gene pathways related to insulin resistance and glucose tolerance would be enriched, as measured by an increase in microbial gene abundance in women with BMI indicating overweight or obesity compared with women with BMI < 25.
METHODS
Data collection
Biospecimens used in this study were obtained from the Obstetric and Neonatal Outcomes Study (ONOS) at the University of Virginia Medical Center, Charlottesville, Virginia. ONOS is a biobank of maternal serum blood, maternal fecal swab, placenta, and cord blood biospecimens. Inclusion criteria for ONOS participation include (1) confirmed pregnancy at less than 20 weeks’ gestation at enrollment, (2) anticipation of an uncomplicated singleton pregnancy, (3) maternal age between 18 and 45 years old, (4) being either English or Spanish speaking, and (5) planning to remain in the Charlottesville area for 1 year following delivery with the willingness and ability to provide consent. Exclusion criteria for the parent study include (1) preexisting diabetes, (2) chronic hypertension, (3) multiple gestation pregnancy, (4) any diagnosis of a fetal anomaly requiring surgery, (5) previous Rh isoimmunization that required transfusion, or (6) a significant medical condition that requires long‐term medication, including, but not limited to, chronic thyroid disease, autoimmune disorders, or steroid use. Fecal biospecimens were self‐collected by the participant either at home or in the prenatal clinic between 26 and 34 weeks of gestation (suggested at the same visit as the glucose challenge test screening). The ONOS study coordinator gave participants a collection kit for the fecal swab (BD BBL CultureSwab) at the first study visit as a component of the introduction package outlining the goals of the study. Processing of the biospecimen swab included drying of the swab sample, the addition of a DNA Catch All Reagent (ThermoFisher) into a microcentrifuge tube, and a series of agitation steps for the swab and reagent combination. The resulting processed biospecimen included approximately 200 μL of liquid biospecimen combined with the lysis reagent in an aliquot. The processed biospecimen was placed in long‐term storage freezing at −70 °C to −80 °C.
All study procedures were approved by the University of Virginia Institutional Review Board for Health Sciences Research.
Sampling strategy and sequencing
As of April 2019, 109 participants completed collection of the fecal biospecimen for the ONOS biobank, and 105 (96.3%) were included in the current analysis. Biospecimens were transported to the National Cancer Institute (Bethesda, Maryland) for processing and DNA extraction. Automated DNA extraction was performed with the MagAttract PowerMicrobiome DNA/RNA kit (catalog number/identifier: 27500‐4‐EP) from Qiagen with Qubit quantification following the manufacturer's instructions. Four samples (3.7%) did not meet quality control standards, including having an adequate DNA concentration for sequencing. Library preparation was completed using the Illumina Nextera DNA Flex Library Prep kit. Biospecimens were sequenced using the Illumina NovaSeq 6000 sequencing platform at the National Cancer Institute (Frederick, Maryland).
Read processing pipeline and quality control
The paired‐end sequencing FASTQ files generated from the Illumina NovaSeq platform and the associated clinical metadata were inputted into the Just Another Microbiology System (JAMS) pipeline, version 1.39 [16]. The JAMS pipeline has two core phases, which include (1) JAMSalpha to determine the quantity and quality of both taxonomic and functional sequences within a biospecimen and (2) JAMSbeta to visualize and explore the differences between biospecimens, taking into account relevant clinical metadata [16].
In JAMSalpha, paired‐end sequencing reads were first quality trimmed by using Trimmomatic [17], a tool developed specifically for Illumina next‐generation‐sequencing files. Trimmed reads were then aligned to the human genome by using Bowtie2 to remove host sequences in the biospecimen [18]. After removal of host sequences, the resulting reads were assembled into contigs, an overlapping set of DNA segments, and mapped to the microbial genome by using Megahit [19]. Remaining reads that were not mapped to contigs were taxonomically classified by using k‐mer analysis performed by using Kraken [20]. All contigs and k‐mer–assigned reads were then assigned to a last known taxon, which represents the most specific taxonomic classification available. Each microbiota label represents a single microbe. For some species, the most specific classification may be at the phylum level. The JAMS pipeline also produced a mapping of microbial genes to functional terms by using the GO database. Microbial function, in this study, is defined as the relative abundance of genes contributing to a specific GO term. The GO term descriptions for the detailed tables were sourced from the Gene Ontology Consortium Resource website [13, 14]. Finally, we examined antibiotic resistance (AR) gene abundance in JAMS by using antibiograms, a percent calculation of individual bacterial pathogens that are known to be susceptible to antibiotic agents.
All biospecimens had at least 55% of assembled bases passing quality control, and assessment of the assembled contigs showed no bias between samples on the basis of maternal pregravid BMI, antibiotic administration, diagnosis of gestational diabetes, or the clinical value of impaired glucose tolerance at the glucose challenge test.
Clinical metadata
Clinical data related to past medical history, the index or current pregnancy, and the participant's labor and delivery admission were abstracted from the EPIC medical record (Epic Systems Corporation) by a trained research assistant. Data, with only the unique ONOS deidentified label, were entered into Qualtrics and then exported into a comma‐separated values (CSV) file for further analysis. The principal investigators for this study were blinded to personal identifiers and medical information. All clinical and metagenomic information input into the JAMS pipeline were deidentified. Clinical metadata in the CSV file were reviewed by two study team members to ensure accuracy.
Data analysis
For the main objective of this study, participant biospecimens were grouped by maternal pregravid BMI status as having normal weight (BMI < 24.9), overweight (BMI = 25.0–29.9), or obesity (BMI > 30). Three participants were classified as underweight with BMI < 18.5 (range: 18.20–18.43) and were grouped with the normal‐weight group. We calculated the pregravid BMI status by using the documented pregravid weight and maternal height in the electronic medical record system for which the clinical metadata were abstracted. All clinical metadata were compared between BMI groups by using either ANOVA and a follow‐up Tukey honestly significant difference test to determine pairwise differences for parametric continuous variables or the Kruskal‐Wallis rank sum test for nonparametric continuous variables. Further, we used χ2 testing and a post hoc Fisher exact test for categorical variables.
Between‐group comparisons for taxonomic composition and GO terms for microbial function were calculated by using linear discriminant analysis (LDA) with the Galaxy tool [21]. LDA identifies the microbiota that are differentially abundant between groups by using the nonparametric Kruskal‐Wallis sum rank test and estimates the effect size [22]. Per‐sample normalization and an α value of 0.5 for the Kruskal‐Wallis test were used within the Galaxy tool [21]. The logarithmic LDA score for discriminative features was set at an absolute value of 2.0‐fold change. The α and β diversity statistics were calculated by using the vegan package (version 2.5‐6) in RStudio (version 3.6.1). The α diversity comparisons were analyzed by using nonparametric Wilcoxon and Kruskal‐Wallis tests, depending on the number of BMI groups in the analysis. Dissimilarity coefficients using the Bray‐Curtis index were used for the calculation of β diversity. A comparison of β diversity among pregravid BMI groups was completed by using PERMANOVA analysis with the adonis function from the vegan package in RStudio. An LDA sensitivity analysis was completed by using antibiotic administration as the primary confounding variable. Antibiotic administration was assessed from the medication record in the electronic health record and defined as antibiotics prescribed between the time of ultrasound‐confirmed pregnancy and the collection of the biospecimen.
Power analysis
A post hoc power analysis was completed for comparison of α diversity among the three pregravid BMI groups by using the “pwr” package in R. After calculating the F statistic by using ANOVA at a significance level of 0.05 for the primary diversity metric, we had the power to detect a 0.99 effect size. Further, we calculated the power analysis across all taxa for samples from women with obesity or overweight compared with women with normal weight by using a Monte Carlo simulation to test the Dirichlet‐Multinomial parameter comparison in the “HMP” package.
RESULTS
Participant characteristics
A total of 105 participant biospecimens were included in this analysis. Among the participants, only gravida (i.e., the number of pregnancies) had a statistically significant difference by pregravid BMI status, as shown in Table 1. Women with BMI indicating obesity had a higher number of pregnancies than the women with overweight or normal weight. Maternal age and antibiotic administration were not significantly among between BMI groups (p = 0.79 and p = 0.56, respectively). In our study, 16.2% of women were categorized as having BMI in the obesity range. The most recent statistics from 2014 regarding the U.S. prevalence of pregravid obesity is 24.3% [23]. Over half of our participants self‐identified as White (55.2%), with others identifying as Hispanic/Latina (27.6%), Black (11.4%), Asian (0.9%), having multiple races or ethnicities (0.9%), and other (3.8%).
TABLE 1.
Participant characteristics by pregravid BMI group
| All | Normal weight a (BMI < 24.9 kg/m2) | Overweight a (BMI = 25–29.9 kg/m2) | Obesity a (BMI > 30 kg/m2) | p value | |
|---|---|---|---|---|---|
| Number of participants a | 105 | 49 (46.7%) | 39 (37.1%) | 17 (16.2%) | – |
| Race/ethnicity a , b | |||||
| White | 58 (55.2%) | 35 (71.4%) | 17 (43.6%) | 6 (35.3%) | 0.07 |
| Black | 12 (11.4%) | 2 (4.1%) | 7 (17.9%) | 3 (17.6%) | |
| Hispanic/Latina | 29 (27.6%) | 8 (16.4%) | 13 (33.3%) | 8 (47.1%) | |
| Multiple | 1 (0.9%) | 1 (2.0%) | 0 (0%) | 0 (0%) | |
| Asian | 1 (0.9%) | 1 (2.0%) | 0 (0%) | 0 (0%) | |
| Other | 4 (3.8%) | 2 (4.1%) | 2 (5.2%) | 0 (0%) | |
| Pregravid BMI (kg/m2) c , d | 26.4 ± 5.3 | 22.5 ± 1.7 | 27.2 ± 1.5 | 35.7 ± 4.1 | <0.00*** |
| Age (y) c , d | 29.5 ± 5.0 | 29.4 ± 4.9 | 29.5 ± 5.6 | 29.5 ± 3.6 | 0.97 |
| Antibiotic administration b , e | |||||
| Yes | 24 (22.9%) | 10 (20.4%) | 8 (20.5%) | 6 (35.3%) | 0.41 |
| No | 81 (77.1%) | 39 (79.6%) | 31 (79.5%) | 11 (64.7%) | |
| Gravida f | 2 (1–8) | 2 (1–6) | 2 (1–6) | 4 (2–8) | <0.00*** |
n (%).
χ2 testing and post hoc Fisher exact test for categorical variables.
Mean ± standard deviation.
Kruskal‐Wallis rank sum test for nonparametric continuous variables.
Antibiotic administration was defined as the prescription of antibiotics between the time of confirmed pregnancy and the time of the microbiome biospecimen collection.
Median (range).
p value significance at <0.00.
Composition
When comparing the microbiome biospecimens by maternal pregravid BMI group, we identified several taxa that were statistically significant according to an absolute LDA score above 2.0. Significant taxa between the normal‐weight group and the group with overweight or obesity are shown in Figure 1. Seven taxa were enriched in the overweight and obesity categories compared with the normal‐weight category, including Roseburia intestinalis and other Roseburia species. Other taxa enriched in the group with overweight or obesity include Faecalibacterium prausnitzii and Eubacterium rectale (LDA scores > 2.0), both of which belong to the phylum Firmicutes [24]. In the group with a normal‐weight BMI, there are 27 discriminant taxa, including Anaerococcus tetradius, A. obesiensis, and Prevotella bergensis, as well as many which were vaginally derived, such as Murdochiella vaginalis (LDA scores > 2.0).
FIGURE 1.
The LDA effect size for significant taxa between normal‐weight participants and participants with obesity or overweight. GO terms that are enriched in the group with obesity or overweight are labeled with LDA score in green. GO terms that are enriched in the normal‐weight group are marked in red. GO, Gene Ontology; LDA, linear discriminant analysis [Color figure can be viewed at wileyonlinelibrary.com]
Alpha and beta diversity
The α diversity, the within‐subject diversity, was calculated by using four measures, as shown in Table 2. The metrics include (1) Simpson (i.e., index accounting for the number of and abundance of each species in a biospecimen), (2) inverse Simpson (i.e., measure of richness or the number of distinct species in a community), (3) Shannon‐Weiner (i.e., index that accounts for both abundance and equitability, or evenness, of microbiota), and (4) Chao1 richness (i.e., an estimate of the total number of taxa within a biospecimen) metrics [25]. We found no statistically significant differences between BMI groups for α diversity, as measured by the Chao1 and Shannon‐Weiner indexes. Figures comparing α diversity between BMI groups are available in online Supporting Information. When participants with overweight and obesity BMI classes were grouped together, there was a substantial increase in statistical significance for inverse Simpson (p = 0.02) and Simpson (p = 0.02) measures when compared with the normal‐weight group (BMI < 25), with participants in the group with overweight or obesity having lower α diversity than the group with a normal BMI. There continued to be no statistical significance in the Shannon‐Weiner or Chao1 metrics.
TABLE 2.
Alpha diversity metrics by groups with obesity or overweight and normal weight
| Diversity metric | Normal weight (BMI < 24.9) | Obesity/overweight (BMI > 25) | p value a |
|---|---|---|---|
| Simpson | 0.94 (0.05) | 0.92 (0.07) | 0.02 |
| Inverse Simpson | 21.5 (8.5) | 17.7 (8.4) | 0.02 |
| Shannon‐Wiener | 3.96 (0.42) | 3.78 (0.53) | 0.07 |
| Chao1 | 8636.6 (1140.4) | 8840.4 (1045.8) | 0.28 |
Note: Data given as mean (standard deviation).
p value significance was set at 0.05 for a nonparametric Wilcoxon test between two BMI groups.
In comparison with α diversity, β diversity is the between‐subjects comparison of microbial biospecimens. We found no significant difference in β diversity between pregravid BMI groups (p = 0.26), and the significance levels were relatively unchanged after accounting for the interaction of maternal BMI and antibiotic use (p = 0.27).
Microbial function
A total of 11 GO terms were enriched in the group with overweight or obesity, and 8 terms were enriched in the normal‐weight group, implicating differences not only in microbial diversity but also in microbial function. Figure 2 shows the absolute LDA score for discriminative features. The terms were further organized in Table 3 by comparison groups, GO identification number, name, definition, type of function (i.e., cellular, biological, molecular), definition, and relevant genes and products in the pathway.
FIGURE 2.
The LDA effect size for significant GO terms between normal‐weight participants and participants with obesity or overweight (p < 0.05). GO terms that are enriched in the group with obesity or overweight are labeled with LDA score in green. GO terms that are enriched in the normal‐weight group are marked in red. GO, Gene Ontology; LDA, linear discriminant analysis [Color figure can be viewed at wileyonlinelibrary.com]
TABLE 3.
Definition of GO terms by enrichment group in between‐group analysis
| GO ID a by enrichment group | Name a | Type of function a | GO definition a | Relevant genes/products a |
|---|---|---|---|---|
| Obesity/overweight | ||||
| GO:0008152 | Metabolic process | Biological | Chemical reactions and pathways to transform chemical substances | RENBP, NAA10, A0A1D5WCG6 |
| GO:0007049 | Cell cycle | Biological | The combination of biochemical phases of cell replication | SEPT7, TSC2, CCN12, BUB3 |
| GO:0006810 | Transport process | Biological | Directed movement of molecules into or out of the cell | ITPR2, SLC31A1, GRID1 |
| GO:0004683 | Calmodulin‐dependent protein kinase activity | Molecular | Catalysis of reactions including ATP and proteins which require calcium present | CPK6, MKNK1, CDPK2 |
| GO:0035556 | Intracellular signal transduction | Biological | Cellular signaling to either propagate the signal or create a change in the cell itself | ITPR2, MGRN1, TSC2, ARHGEF3 |
| GO:0019239 | Deaminase activity | Molecular | Catalysis of an amino group resulting in the production of ammonia | ADA, AICDA, APOCEC1, AMPD1 |
| GO:0005516 | Calmodulin binding | Molecular | Binding of calmodulin in environments with and without calcium | KCNN4, CPK6, PLCB1, KCNQ5 |
| GO:0032955 | Regulation of division septum assembly | Biological | Process involved in septum creation during cytokinesis | minE, YALI0_E34650g |
| GO:0007017 | Microtubule‐based process | Biological | Cellular processes involved in creation of the microtubules and their associated proteins | CAMSAP2, MAPS1S, TMEM108 |
| GO:0017148 | Negative regulation of translation | Biological | Process that prevents or decreased the rate of reactions at the time of translation | TRIM71, FMR1, RPL13A |
| GO:0009102 | Biotin biosynthetic process | Biological | Chemical reactions associated with the creation of biotin | bioA, bioB, bioD |
| Normal weight | ||||
| GO:0004161 | Dimethylallyl transferase activity | Molecular | Catalysis of several molecules which includes the first step of forming farnesyl diphosphate | A0A1D5WH41, TcasGA2_TV009257 |
| GO:0018493 | Formylmethanofuran dehydrogenase activity | Molecular | Catalysis of molecules, including hydrogen, carbon dioxide, and methanofuran | MCA2319, MCA2859, MCA2860 |
| GO:0042891 | Obsolete antibiotic transport | Biological | Process by which an antibiotic is transported in or out of a cell | – |
| GO:0003779 | Actin binding | Molecular | Process associated with membrane actin binding | EPS8, MAP1S, FMNL2 |
| GO:0043817 | Phosphosulfolactate synthase activity | Molecular | Catalysis of the reaction resulting in phosphoenolpyruvate and sulfite | – |
| GO:0018423 | Protein C‐terminal leucine carboxyl O‐methyltransferase activity | Molecular | Catalytic modification at the C‐terminus of the protein | PPM1, LCMT1, W5MDM9 |
| GO:1900753 | Doxorubicin transport | Biological | Directed movement of doxorubicin in and out of a cell | RALBP1 |
| GO:0043215 | Doxorubicin transport | Biological | Directed movement of doxorubicin in and out of a cell | ABCB1A |
Influence of antibiotic administration as a covariate
In this study, 24 women (22.9%) received antibiotics between the time of confirmed pregnancy and collection of the fecal swab biospecimen. There was no significant difference in antibiotic administration between BMI groups (p = 0.06). The most frequent indications for antibiotic administration during this period were sexually transmitted infections (e.g., gonorrhea, chlamydia), urinary tract infections, and bacterial vaginosis. Common antibiotics prescribed were cefalexin (Keflex) (n = 12), azithromycin (n = 6), amoxicillin (n = 3), and ceftriaxone (Rocephin) (n = 3), depending on the indication. Between BMI groups, we found significant differences in AR genes for tetracycline (p < 0.00, FDR < 0.00), doxycycline (p < 0.00, FDR < 0.00), and minocycline (p < 0.00, FDR < 0.00). None of the significant antibiotics was prescribed to participants between the time of confirmed pregnancy and collection of the microbiome biospecimen.
Using an ordination plot, we found a significant difference in microbiota communities between women who received antibiotics and those who did not (p = 0.03). Because of this finding and the importance of antibiotic administration on the microbial environment, we completed a sensitivity analysis within the LDA to assess the significance of this variable to the influence of maternal obesity on microbial diversity and function. Figure 3 shows the compositional change in discriminative microbiota when antibiotic administration was used as a subclass in the analysis. Bifidobacterium adolescentis was enriched in the group with overweight or obesity, and Murdochiella vaginalis and A. lactolyticus were enriched in the normal‐weight group.
FIGURE 3.
LDA when including antibiotic administration as a covariate. LDA, linear discriminant analysis [Color figure can be viewed at wileyonlinelibrary.com]
DISCUSSION
This study highlighted several compositional and functional differences in the maternal gastrointestinal microbiome in the late second trimester by varying pregravid BMI groups. In summary, the microbiota of women who entered into pregnancy with BMI indicating obesity or overweight differed from women with BMI indicating normal weight. Results from our comparison of last known taxon showed a broad increase in bacteria from the phylum Firmicutes in pregravid BMI groups. Firmicutes, which is generally thought to have a negative influence on glucose and fat metabolism, at the phylum level is not enough to determine specific deviations in the gastrointestinal environment.
More specifically, women with BMI indicating obesity had enrichment in several taxa, including E. rectale and Bifidobacterium species. E. rectale was shown in a large study on energy harvest and obesity to encode for primary fermentation enzymes that specifically digest dietary polysaccharides [26]. Enrichment in B. bifidum and B. adolescentis may contribute to gluconeogenesis and the tricarboxylic acid cycle, the main series of chemical reactions to release stored energy from carbohydrates, fats, and proteins [24]. Enriched in the normal‐weight group was the bacteria family Peptostreptococcaceae. The Peptostreptococcaceae family regulates lipid absorption in the small intestine, and it was shown to decrease in the ileum and cecum when mice are fed a high‐fat diet in a mouse model [27]. Our interpretation of these microbiota, in combination with other enriched taxa such as A. tetradius, is that mothers with a normal‐weight pregravid BMI have microbiota that primarily regulate epithelial cells in the lower gastrointestinal tract and these epithelial cells are responsible for host adaptability to dietary lipid variation and caloric absorption. A review on the effect of the gastrointestinal microbiome on nutrient absorption and energy regulation supported this interpretation, as it highlighted the potential of an obesogenic microbiota to increase the abundance of Methanobrevibacter smithii, which was present but not significant in our analysis, and, therefore, increase energy extraction from ingested food [28].
Confirming our hypothesis, women who entered pregnancy with BMI indicating overweight or obesity had lower α diversity in their microbiome (inverse Simpson and Simpson diversity metrics) than women with normal weight. Both the Simpson and inverse Simpson metrics are weighted calculations that take into account the overall abundance of the microbiota in the biospecimen. The inverse Simpson metric is typically preferred because it accounts for richness in the community and it is more stable in respect to the number of biospecimens included in the analysis, whereas the Shannon index is not. Our study is an agreement with several previous research studies, which found that maternal pregravid overweight and obesity reduced α diversity of the maternal gastrointestinal microbiota [6, 9]. Three additional studies found no difference in α diversity by maternal pregravid weight category, but this may be because of differences in the diversity metrics that were reported (i.e., the use of the Shannon index in comparison with the Simpson metric) [11, 29, 30].
When we compared the functional contribution of the microbial genes, we found that women with obesity and overweight had enriched pathways related to cell replication (GO:0007049, GO:0007017), transport (GO:0007049), and signaling (GO:0035556); calmodulin binding (GO:0005516, GO:0004683); and biotin synthesis (GO:0009102) compared with women with normal weight. Functional terms related to gastrointestinal epithelial cell turnover and signaling reflected the potential for dysbiosis in nutrient absorption, hormone regulation, and downstream cell membrane barrier and transport functions. Research in a mouse model found that obesity, independent of diet, drove lasting effects on gastrointestinal epithelial cell turnover [31].
Functional terms related to biotin synthesis were also enriched in the group with obesity or overweight. A study published by Järvinen and colleagues examined the role of biotin in obesity [32]. The study found that biotin‐dependent functions, including the regulation of lipids and metabolism, were modified by adiposity level and influenced inflammatory processes [32]. Research from the same team also found that the biotin metabolism pathway was associated with metabolic dysregulation in adults with obesity through epigenetic mechanisms such as DNA methylation [33]. Our functional analysis, combined with previous research in mice, hints toward a deeper multi‐omics approach (i.e., microbiomics, epigenetics) to assess the influence of obesity on clinical outcomes.
Alternatively, women who entered into pregnancy with normal weight had genes that are enriched for catalytic reactions in the cell (GO:0004161, GO:0018493, GO:0043817, GO:0018423) and antibiotic regulation (GO:1900753, GO:0043215, GO:0042891) in comparison with those with BMI indicating overweight or obesity. A study conducted by Sugino et al. [34] examined the association of maternal prepregnancy BMI on both maternal and child microbiome composition in a cohort of women, in which only 2.6% (n = 1) were reported to be on antibiotics at the time of data collection. In our analysis, we found that it was critical to understand the role that antibiotics had in perturbing the microbial environment. There were fewer significantly differentially abundant taxa between BMI groups when antibiotics were included as a covariate. We believe that this attenuation is explained by the effect of antibiotics reducing the number of taxa present in each biospecimen in both pregnant and nonpregnant samples [35, 36, 37]. Not only were antibiotics a significant variable in our sensitivity analysis, but we also found differences in AR genes between pregravid BMI groups. AR genes are microbial contributors that have adapted to environmental changes shaped by antibiotics by developing resistance [38]. In this study, AR genes were expressed as an antibiogram, a definition of how susceptible specific pathogens are to antibiotic administration. Our research found significant differences in antibiogram clustering between pregravid BMI groups, suggesting that microbes present in the gastrointestinal tract of participants with BMI indicating overweight or obesity had different susceptibility to antibiotic‐resistant infections and ability for prescribed antibiotics to efficiently work. Specifically, we found A. lactolyticus as a differentiating microbe between maternal pregravid BMI groups. A. lactolyticus has been noted in polymicrobial urinary tract infections (a substantial indication for antibiotic administration in this cohort) [39], and it was shown to be highly susceptible to action by amoxicillin, doxycycline, and cefoxitin [40]. Doxycycline, in our study, was noted to be a significant antibiotic. Although antibiotics are critical during pregnancy to combat infections that have an opportunity to produce complications (i.e., group B streptococcus status, urinary tract infections, and nonreproductive infections), providers should consider the potential for downstream metabolic and immune changes.
The microbiome has the potential to be a modifiable factor in the development of obesity because of the close relationship between dietary changes and marked shifts in the compositional taxa and diversity of a biospecimen. During pregnancy, Ferrocino et al. [41] found significant relationships between microbial taxa and metabolic variables relating to clinical values of insulin resistance (homeostatic model assessment of insulin resistance) and fasting glucose and insulin after a dietary counseling intervention. Adherents to dietary counseling had improved metabolic and inflammatory profiles and apparent reductions in clinical values relating to impaired glucose tolerance and gestational diabetes. Targeted dietary counseling and personalized nutrition have been used in chronic illness [42] and inflammatory bowel disease [43] as a means of manipulating the gastrointestinal microbiome. Dietary interventions [44] and probiotic supplementation [45] are avenues of intervention research that could mitigate adverse pregnancy and childhood outcomes. Although this analysis was not examining the potential of the microbiome to be obesogenic or a modifiable factor for obesity, future studies should consider robust experimentation, for both maternal and infant outcomes, regarding the possibility.
Strengths of our study include leveraging metagenomic sequencing to validate previous findings about the influence of obesity on the gastrointestinal microbiome. The assessment of functional terms in this analysis helps to push the field toward a deeper and more comprehensive characterization of the microbiome during pregnancy. Future research should consider metagenomic sequencing the gold standard for sequencing of clinical biospecimens. Additionally, there were several limitations to this study. First, there are inherent issues with microbiome analysis in a cross‐sectional study. Little is known about the longitudinal changes in the maternal microbiome at a species level that would enable understanding how the discriminant microbiota identified in this study may be altered because of other unmeasured variables in our data set (i.e., maternal pH of the gastrointestinal tract, fecal quality, diet, environmental variables). Observational microbiome studies, especially those with heterogeneous samples, lack the ability for replication to overcome conflicting results of previously published work [46]. Second, there were three taxa that were enriched in the LDA analysis that likely contribute to vaginal dysbiosis, including A. mediterraneensis [47], A. vaginalis [48], and Murdochiella vaginalis [49]. This could be, in part, due to bacterial overgrowth and maternal anatomy or poor collection of the rectal fecal swab. More detailed metadata regarding sample collection (e.g., time of collection, time between collection and deposition to the biorepository, Bristol stool variables) should be ascertained in future studies. Third, only medications noted in the EPIC medical record were included in the current analysis. This is particularly salient for antibiotic administration in which participants may have received care at an outside clinic or urgent care facility, and, therefore, antibiotic usage would not be included in the medical record abstraction. Future studies should consider additional self‐report questionnaires regarding dietary differences, medication administration (i.e., over‐the‐counter medications), and socioeconomic variables, but this is a particularly challenging component considering the use of biobanked biospecimens.
CONCLUSION
The results of this study indicate microbial composition and function alternations in the women with BMI > 25 at the end of the second trimester. Our study provides an extension of previous work with a robust functional interpretation that has not, to our knowledge, been examined during pregnancy in relation to obesity. Future directions for this study integrate well with the technological advancement of microbial genetics, which enables sequencing of more biospecimens at a higher depth, and the emerging role that microbes play in complex disorders of pregnancy.
CONFLICT OF INTEREST
The authors declared no conflict of interest.
Supporting information
Appendix S1. Supporting Information
ACKNOWLEDGMENTS
The authors acknowledge Research Computing at the University of Virginia for providing computational resources and technical support that have contributed to the results reported within this publication (https://rc.virginia.edu). In addition, they would like to thank Briana Cortez Chronister for her role as the clinical research coordinator for the ONOS study, Wuxing Yuan for her extraordinary persistence in DNA extraction, Laura Habermeyer for her outstanding work as their research assistant, John McCulloch, PhD, for his creation and assistance in running the JAMS pipeline, and the University of Virginia Bioinformatics Core for code review.
Dreisbach C, Alhusen J, Prescott S, Dudley D, Trinchieri G, Siega‐Riz AM. Metagenomic characterization of the maternal prenatal gastrointestinal microbiome by pregravid BMI. Obesity (Silver Spring). 2023;31(2):412‐422. doi: 10.1002/oby.23659
Funding information Association for Women's Health, Obstetric, and Neonatal Nurses, Grant/Award Number: March of Dimes Margaret Comerford Freda Award; National Institute for Nursing Research, Grant/Award Number: F31NR017821
REFERENCES
- 1. Hales CM, Carroll MD, Fryar CD, Ogden CL. Prevalence of obesity among adults and youth: United States, 2015–2016. NCHS Data Brief, no. 288. National Center for Health Statistics; 2017. [PubMed] [Google Scholar]
- 2. Kulie T, Slattengren A, Redmer J, Counts H, Eglash A, Schrager S. Obesity and women's health: an evidence‐based review. J Am Board Fam Med. 2011;24(1):75‐85. doi: 10.3122/jabfm.2011.01.100076 [DOI] [PubMed] [Google Scholar]
- 3. Stubert J, Reister F, Hartmann S, Janni W. The risks associated with obesity in pregnancy. Dtsch Arztebl Int. 2018;115(16):276‐283. doi: 10.3238/arztebl.2018.0276 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Huttenhower C, Gevers D, Knight R, et al. Structure, function and diversity of the healthy human microbiome. Nature. 2012;486(7402):207‐214. doi: 10.1038/nature11234 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Gomez‐Arango LF, Barrett HL, McIntyre HD, Callaway LK, Morrison M, Dekker Nitert M. Connections between the gut microbiome and metabolic hormones in early pregnancy in overweight and obese women. Diabetes. 2016;65(8):2214‐2223. doi: 10.2337/db16-0278 [DOI] [PubMed] [Google Scholar]
- 6. Collado MC, Isolauri E, Laitinen K, Salminen S. Distinct composition of gut microbiota during pregnancy in overweight and normal‐weight women. Am J Clin Nutr. 2008;88(4):894‐899. doi: 10.1093/ajcn/88.4.894 [DOI] [PubMed] [Google Scholar]
- 7. Santacruz A, Collado MC, García‐Valdés L, et al. Gut microbiota composition is associated with body weight, weight gain and biochemical parameters in pregnant women. Br J Nutr. 2010;104(1):83‐92. doi: 10.1017/S0007114510000176 [DOI] [PubMed] [Google Scholar]
- 8. Chu DM, Antony KM, Ma J, et al. The early infant gut microbiome varies in association with a maternal high‐fat diet. Genome Med. 2016;8(1):77. doi: 10.1186/s13073-016-0330-z [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Stanislawski MA, Dabelea D, Wagner BD, Sontag MK, Lozupone CA, Eggesbø M. Pre‐pregnancy weight, gestational weight gain, and the gut microbiota of mothers and their infants. Microbiome. 2017;5(1):113. doi: 10.1186/s40168-017-0332-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Dreisbach C, Prescott S, Alhusen J. Influence of maternal prepregnancy obesity and excessive gestational weight gain on maternal and child gastrointestinal microbiome composition: a systematic review. Biol Res Nurs. 2020;22(1):114‐125. doi: 10.1177/1099800419880615 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Houttu N, Mokkala K, Laitinen K. Overweight and obesity status in pregnant women are related to intestinal microbiota and serum metabolic and inflammatory profiles. Clin Nutr. 2018;37:1955‐1966. doi: 10.1016/j.clnu.2017.12.013 [DOI] [PubMed] [Google Scholar]
- 12. Castaner O, Goday A, Park YM, et al. The gut microbiome profile in obesity: a systematic review. Int J Endocrinol. 2018;2018:4095789. doi: 10.1155/2018/4095789 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Gene Ontology Consortium . The Gene Ontology resource: 20 years and still GOing strong. Nucleic Acids Res. 2019;47(D1):D330‐D338. doi: 10.1093/nar/gky1055 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Ashburner M, Ball CA, Blake JA, et al. Gene Ontology: tool for the unification of biology. Nat Genet. 2000;25(1):25‐29. doi: 10.1038/75556 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Kanehisa M, Araki M, Goto S, et al. KEGG for linking genomes to life and the environment. Nucleic Acids Res. 2008;36(1):D480‐D484. doi: 10.1093/nar/gkm882 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. McCulloch JA. Johnmcculloch/JAMS_BW wiki. GitHub. Accessed May 17, 2022. https://github.com/johnmcculloch/JAMS_BW
- 17. Bolger AM, Lohse M, Usadel B. Trimmomatic: a flexible trimmer for Illumina sequence data. Bioinformatics. 2014;30(15):2114‐2120. doi: 10.1093/bioinformatics/btu170 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Langmead B, Salzberg SL. Fast gapped‐read alignment with bowtie 2. Nat Methods. 2012;9(4):357‐359. doi: 10.1038/nmeth.1923 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. Li D, Liu CM, Luo R, Sadakane K, Lam TW. Megahit: an ultra‐fast single‐node solution for large and complex metagenomics assembly via succinct de Bruijn graph. Bioinformatics. 2015;31(10):1674‐1676. doi: 10.1093/bioinformatics/btv033 [DOI] [PubMed] [Google Scholar]
- 20. Wood DE, Salzberg SL. Kraken: ultrafast metagenomic sequence classification using exact alignments. Genome Biol. 2014;15(3):R46. doi: 10.1186/gb-2014-15-3-r46 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21. Afgan E, Baker D, Batut B, et al. The Galaxy platform for accessible, reproducible and collaborative biomedical analyses: 2018 update. Nucleic Acids Res. 2018;46(W1):W537‐W544. doi: 10.1093/nar/gky379 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. Segata N, Izard J, Waldron L, et al. Metagenomic biomarker discovery and explanation. Genome Biol. 2011;12(6):R60. doi: 10.1186/gb-2011-12-6-r60 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23. Singh GK, DiBari JN. Marked disparities in pre‐pregnancy obesity and overweight prevalence among US women by race/ethnicity, nativity/immigrant status, and sociodemographic characteristics, 2012–2014. J Obes. 2019;2019:2419263. doi: 10.1155/2019/2419263 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24. Cruz J, Liu Y, Liang Y, et al. BacMap: an up‐to‐date electronic atlas of annotated bacterial genomes. Nucleic Acids Res. 2012;40(D1):D599‐D604. doi: 10.1093/nar/gkr1105 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25. Xia Y, Sun J, Chen D‐G. Statistical Analysis of Microbiome Data with R. SpringerLink; 2022. doi: 10.1007/978-981-13-1534-3 [DOI] [Google Scholar]
- 26. Turnbaugh PJ, Ley RE, Mahowald MA, Magrini V, Mardis ER, Gordon JI. An obesity‐associated gut microbiome with increased capacity for energy harvest. Nature. 2006;444(7122):1027‐1031. doi: 10.1038/nature05414 [DOI] [PubMed] [Google Scholar]
- 27. Martinez‐Guryn K, Hubert N, Frazier K, et al. Small intestine microbiota regulate host digestive and absorptive adaptive responses to dietary lipids. Cell Host Microbe. 2018;23(4):458‐469. doi: 10.1016/j.chom.2018.03.011 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28. Krajmalnik‐Brown R, Ilhan ZE, Kang DW, DiBaise JK. Effects of gut microbes on nutrient absorption and energy regulation. Nutr Clin Pract. 2012;27(2):201‐214. doi: 10.1177/0884533611436116 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29. Smid MC, Ricks NM, Panzer A, et al. Maternal gut microbiome biodiversity in pregnancy. Am J Perinatol. 2018;35(1):24‐30. doi: 10.1055/s-0037-1604412 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30. Gomez‐Arango LF, Barrett HL, McIntyre HD, Callaway LK, Morrison M, Dekker Nitert M. Increased systolic and diastolic blood pressure is associated with altered gut microbiota composition and butyrate production in early pregnancy. Hypertension. 2016;68(4):974‐981. doi: 10.1161/HYPERTENSIONAHA.116.07910 [DOI] [PubMed] [Google Scholar]
- 31. Zhou W, Davis EA, Dailey MJ. Obesity, independent of diet, drives lasting effects on intestinal epithelial stem cell proliferation in mice. Exp Biol Med. 2018;243(10):826‐835. doi: 10.1177/1535370218777762 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32. Järvinen E, Ismail K, Muniandy M, et al. Biotin‐dependent functions in adiposity: a study of monozygotic twin pairs. Int J Obes. 2016;40(5):788‐795. doi: 10.1038/ijo.2015.237 [DOI] [PubMed] [Google Scholar]
- 33. Ollikainen M, Ismail K, Gervin K, et al. Genome‐wide blood DNA methylation alterations at regulatory elements and heterochromatic regions in monozygotic twins discordant for obesity and liver fat. Clin Epigenetics. 2015;7:39. doi: 10.1186/s13148-015-0073-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34. Sugino KY, Paneth N, Comstock SS. Michigan cohorts to determine associations of maternal pre‐pregnancy body mass index with pregnancy and infant gastrointestinal microbial communities: late pregnancy and early infancy. PLoS One. 2019;14(3):e0213733. doi: 10.1371/journal.pone.0213733 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35. Blaser MJ. Antibiotic use and its consequences for the normal microbiome. Science. 2016;352(6285):544‐545. doi: 10.1126/science.aad9358 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36. Ramirez J, Guarner F, Bustos Fernandez L, Maruy A, Sdepanian VL, Cohen H. Antibiotics as major disruptors of gut microbiota. Front Cell Infect Microbiol. 2020;10:572912. doi: 10.3389/fcimb.2020.572912 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 37. Anthony WE, Wang B, Sukhum KV, et al. Acute and persistent effects of commonly used antibiotics on the gut microbiome and resistome in healthy adults. Cell Rep. 2022;39(2):110649. doi: 10.1016/j.celrep.2022.110649 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38. van Hoek AHAM, Mevius D, Guerra B, Mullany P, Roberts AP, Aarts HJM. Acquired antibiotic resistance genes: an overview. Front Microbiol. 2011;2:203. doi: 10.3389/fmicb.2011.00203 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39. Kline KA, Lewis AL. Gram‐positive uropathogens, polymicrobial urinary tract infection, and the emerging microbiota of the urinary tract. Microbiol Spectr. 2016;4(2): UTI‐0012‐2012. doi: 10.1128/microbiolspec.UTI-0012-2012 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 40. Veloo ACM, Welling GW, Degener JE. Antimicrobial susceptibility of clinically relevant gram‐positive anaerobic cocci collected over a three‐year period in the Netherlands. Antimicrob Agents Chemother. 2011;55(3):1199‐1203. doi: 10.1128/AAC.01771-09 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 41. Ferrocino I, Ponzo V, Gambino R, et al. Changes in the gut microbiota composition during pregnancy in patients with gestational diabetes mellitus (GDM). Sci Rep. 2018;8(1):12216. doi: 10.1038/s41598-018-30735-9 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 42. Schröder L, Kaiser S, Flemer B, et al. Nutritional targeting of the microbiome as potential therapy for malnutrition and chronic inflammation. Nutrients. 2020;12(10):3032. doi: 10.3390/nu12103032 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 43. Olendzki B, Bucci V, Cawley C, et al. Dietary manipulation of the gut microbiome in inflammatory bowel disease patients: pilot study. Gut Microbes. 2022;14(1):2046244. doi: 10.1080/19490976.2022.2046244 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 44. Keskitalo A, Munukka E, Aatsinki A, et al. An infancy‐onset 20‐year dietary counselling intervention and gut microbiota composition in adulthood. Nutrients. 2022;14(13):2667. doi: 10.3390/nu14132667 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 45. Jarde A, Lewis‐Mikhael AM, Moayyedi P, et al. Pregnancy outcomes in women taking probiotics or prebiotics: a systematic review and meta‐analysis. BMC Pregnancy Childbirth. 2018;18(1):14. doi: 10.1186/s12884-017-1629-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 46. Clausen DS, Willis AD. Evaluating replicability in microbiome data. Biostatistics. 2021;23(4):1099‐1114. doi: 10.1093/biostatistics/kxab048 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 47. Diop K, Bretelle F, Fournier PE, Fenollar F. “ Anaerococcus mediterraneensis” sp. nov., a new species isolated from human female genital tract. New Microbes New Infect. 2017;17:75‐76. doi: 10.1016/j.nmni.2017.02.007 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 48. Murphy EC, Frick IM. Gram‐positive anaerobic cocci–commensals and opportunistic pathogens. FEMS Microbiol Rev. 2013;37(4):520‐553. doi: 10.1111/1574-6976.12005 [DOI] [PubMed] [Google Scholar]
- 49. Diop K, Raoult D, Bretelle F, Fenollar F. “ Murdochiella vaginalis” sp. nov., a new bacterial species cultivated from the vaginal flora of a woman with bacterial vaginosis. Hum Microbiome J. 2016;2:15‐16. doi: 10.1016/j.humic.2016.11.002 [DOI] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Appendix S1. Supporting Information