The Journal of Nutrition 154 (2024) 1298–1308

journal homepage: https://jn.nutrition.org/

Nutrition and Disease

Comparison of a Daily Steviol Glycoside Beverage compared with a

Sucrose Beverage for Four Weeks on Gut Microbiome in Healthy Adults
David Kwok 1, Corey Scott 2, *, Noah Strom 3, Fei Au-Yeung 4, Caanan Lam 1,
Anirikh Chakrabarti 5, Thomas Hutton 2, Thomas MS Wolever 4
1
Frontage Laboratories (BRI), Vancouver, British Columbia, Canada; 2 Cargill R&D Center, Plymouth, Minnesota, USA; 3 Diversigen, New
Brighton, Minnesota, USA; 4 INQUIS Clinical Research, Toronto, Ontario, Canada; 5 Cargill R&D Centre Europe, Vilvoorde, Belgium

A B S T R A C T

Background: Recent studies suggest that some nonnutritive sweeteners (NNS) have deleterious effects on the human gut microbiome
(HGM). The effect of steviol glycosides on the HGM has not been well studied.
Objective: We aimed to evaluate the effects of stevia- compared with sucrose-sweetened beverages on the HGM and fecal short-chain fatty
acid (SCFA) profiles.
Methods: Using a randomized, double-blinded, parallel-design study, n ¼ 59 healthy adults [female/male, n ¼ 36/23, aged 319 y, body
mass index (BMI): 22.61.7 kg/m2] consumed 16 oz of a beverage containing either 25% of the acceptable daily intake (ADI) of stevia or 30
g of sucrose daily for 4 weeks followed by a 4-week washout. At weeks 0 (baseline), 4, and 8, the HGM was characterized via shotgun
sequencing, fecal SCFA concentrations were measured using ultra-high performance liquid chromatography-tandem mass spectrometry and
anthropometric measurements, fasting serum glucose, insulin and lipids, blood pressure, pulse, and 3-d diet records were obtained.
Results: There were no significant differences in the HGM or fecal SCFA between the stevia and sucrose groups at baseline (P > 0.05). At
week 4 (after intervention), there were no significant differences in the HGM at the phylum, family, genus, or species level between the
stevia and sucrose groups and no significant differences in fecal SCFA. At week 4, BMI had increased by 0.3 kg/m2 (P ¼ 0.013) in sucrose
compared with stevia, but all other anthropometric and cardiometabolic measures and food intake did not differ significantly (P > 0.05). At
week 8 (after washout), there were no significant differences in the HGM, fecal SFCA, or any anthropometric or cardiometabolic measure
between the stevia and sucrose groups (P > 0.05).
Conclusions: Daily consumption of a beverage sweetened with 25% of the ADI of stevia for 4 weeks had no significant effects on the HGM,
fecal SCFA, or fasting cardiometabolic measures, compared with daily consumption of a beverage sweetened with 30 g of sucrose.
Trial registration: clinicaltrials.gov as NCT05264636

Keywords: stevia, sucrose, gut microflora, short-chain fatty acids, non-nutritive sweeteners

Introduction factors, and diet [2–4]. The HGM performs essential roles,
including reducing potentially harmful microorganisms from
The human digestive tract consists of a diverse community of colonizing in the gut, aiding in digestion and metabolism, sup-
trillions of microorganisms that exist in a largely symbiotic porting the immune system, and producing SCFAs [5–7]. SCFAs,
relationship with the host and are collectively referred to as the notably butyric acid, have been shown to have multiple effects.
gut microbiome [1]. The human gut microbiome (HGM) is These include acting as signaling molecules and influencing the
typically dominated by 4 main phyla—Bacteriodetes, Firmicutes, function of adipose tissue, skeletal muscle, and liver tissue to
Actinobacteria, and Proteobacteria— however, the relative pro- enhance glucose homeostasis and insulin sensitvity [8,9].
portions and diversity of these microbiotas may vary widely even Perhaps the most significant factor influencing the composi-
between healthy individuals due to host genetics, environmental tion and function of gut microbiota, both positively and

Abbreviations: ADI, acceptable daily intake; AE, adverse events; ANOVA, analysis of variance; BMI, body mass index; EFSA, European Food Safety Authority; FDA,
United States Federal Food and Drug Administration; FDR, false discover rate; GRAS, generally recognized as safe; HGM, human gut microbiome; IVA, isovaleric acid;
LLOQ, lower limit of quantification; NNS, non-nutritive sweeteners; OUT, operational taxonomic unit; PCA, principle component analysis; PCoA, principle coordinate
analysis; SCFA, short-chain fatty acids; TG, triglyceride; UPLC-MS/MS, ultra-high performance liquid chromatography-tandem mass spectrometry.
* Corresponding author. E-mail address: corey_scott@cargill.com (C. Scott).

https://doi.org/10.1016/j.tjnut.2024.01.032
Received 27 September 2023; Received in revised form 12 December 2023; Accepted 5 January 2024; Available online 24 February 2024
0022-3166/© 2024 The Authors. Published by Elsevier Inc. on behalf of American Society for Nutrition. This is an open access article under the CC BY license (http://
creativecommons.org/licenses/by/4.0/).
D. Kwok et al. The Journal of Nutrition 154 (2024) 1298–1308

adversely, is diet. However, there is no consensus regarding the consent form. The method of obtaining informed consent is
influence of specific dietary components on gut microbiota, described in Supplementary Information.
mainly because of the confounding effects of large interindi- To be included, participants had to be nonpregnant, non-
vidual differences in response to an intervention [10]. Although lactating individuals (of whom n  25 were male, and n  25
the protein, fat, and sugar contents in the diet all contribute to were female) aged 18–50 y, with a BMI between 18.5 and 24.9
changes in the HGM profile [11,12], the largest impact is made kg/m2, including those who consumed  600 mL (20 oz) high-
by undigestible carbohydrates such as dietary fiber and resistant intensity sweetened beverages per week for 1 month before
starch that are fermented by the HGM with the production of screening and had a normal bowel frequency (3 per week).
SCFA [13]. Pharmaceuticals such as antibiotics and supplements, Details of the inclusion and exclusion criteria are provided in
including probiotics, synbiotics, and postbiotics, may also have Supplementary Information.
major effects on gut microflora growth and function.
Non-nutritive sweeteners (NNS) are ingredients used to replace
caloric sugars in foods and beverages and are typically several Investigation products
hundred- or thousand-fold sweeter than sucrose and yet have little The two investigational products (test product and compar-
to no calories. Depending on their metabolic fate, some NNS are ative product) were provided in 16 fl. oz (473 mL) bottles; each
known to travel to the distal gut, where they may interact with gut product was noncarbonated and contained water, citric acid,
microbiota [14,15]. Some research suggests that NNS may have potassium citrate, natural color, and natural flavor. The test
adverse effects on glucose sensitivity through processes mediated product was sweetened with 620 ppm steviol glycosides
by gut microbiota [15,16]. However, other studies suggest that (equivalent to 75.6 mg steviol), whereas the comparative prod-
the consumption of NNS has no significant impact on HGM uct was sweetened with 30 g of sucrose (Cargill, Inc.). Beverages
[17–20]. The NNS steviol glycosides (stevia) are extracted from were designed to contain specific doses of stevia and sucrose and
the leaves of the stevia rebaudiana bertoni plant and achieved were not matched for sweetness. Sensory data are available in
United States Food And Drug Administration Generally Recog- Supplementary Information.
nized as Safe (US FDA GRAS) status in 2008 as a general-purpose The test product contained a stevia sweetener (EverSweet™,
sweetener in food (GRN 000252 and 000253) and European Food produced by Cargill, Inc.) that is a mixture of 95% steviol gly-
Safety Authority (EFSA) approval as a food additive (E960) in cosides (largely rebaudiosides M and D) produced via yeast
2010. Steviol glycosides, as a class, represent over 40 molecules fermentation and is 300 to 400 times as sweet as sugar on a
consisting of a steviol backbone to which various sugar molecules weight basis. The ADI for steviol glycosides, as established by the
are attached. Steviol glycosides are known to travel to the distal Joint FAO/WHO Expert Committee of Food Additives, is 4 mg/
gut, where its sugar moieties are removed by gut microflora, and kg BW/day steviol equivalents. Thus, for an individual weighing
the steviol backbone is glucuronidated and excreted in the urine. 75 kg, the ADI is 300 mg/d steviol equivalents. The 75.6 mg
Overall, data from in vitro and animal studies suggest that steviol daily administration of steviol equivalents is well below the
glycosides have minimal or no effect on gut microbiota; however, assigned ADI for steviol glycosides. The beverages were
little is known regarding human feeding studies. The primary
objective of this study was to compare the effects of the daily TABLE 1
consumption of stevia compared with sucrose on the HGM Study visits and corresponding procedures
composition of healthy adults. Secondary objectives included Procedure Screening Baseline Week 2 Week 4 Week 8
determining the effects of daily stevia compared with sucrose on Consent X
fecal SCFA concentration, body weight, BMI, blood pressure, Medical/drug/ X X X X X
pulse, fasting serum glucose, fasting serum insulin, and fasting supplement
blood lipids. history
Height X
Weight, blood X X X X
Methods pressure,
pulse
The study had a randomized, controlled, double-blind, par- Adverse events X X X X
assessment
allel design and was conducted at INQUIS Clinical Research, a
Randomization X
contract research organization located in Toronto, Ontario, Fasting X X X
Canada. The study was registered on clinicaltrials.gov as glucose,
NCT05264636 and posted on March 3, 2022. insulin, lipids
Instruct how to X
keep a 3-
Participants d diet record
Generally healthy male and female from the greater Toronto Collect 3-d diet X X X
area were recruited from the pool of individuals who had pre- record
viously participated in studies at the study center and had given Dispense X
product
permission to be contacted for future studies and from online
Assess X X
advertisements, mostly via social media. The study protocol and compliance
associated documents were reviewed and approved by the Dispense stool X X X
Advarra institutional review board (IRB). Screening and clinical collection kit
visits are detailed in Table 1. Participants were not screened until Collect stool X X X
samples
they had provided informed consent by signing the approved

1299
D. Kwok et al. The Journal of Nutrition 154 (2024) 1298–1308

produced at 190  F and sealed in clear plastic bottles with a every reference sequence in Venti using fully gapped alignment
tamper-proof cap. Each bottle had a label containing a numeric with BURST [24]. Each input sequence was assigned the lowest
code corresponding to either the experimental or control sample, common ancestor consistent across at least 80% of all reference
allergen information, lot #, best before date, manufacturing sequences tied for best hit. The number of counts for each
location, and the statement “Non-commercial R&D sample Operational Taxonomic Unit (OTU) was normalized to the
exclusively for clinical study by registered participant.” average genome length. OTUs accounting for less than
one-millionth of all species-level markers and those with
<0.01% of their unique genome regions covered (and < 1% of
Fecal collection
the whole genome) were discarded. Samples with fewer than 10,
Participants were supplied with fecal collection kits contain-
000 sequences mapping to the database were also discarded.
ing disposable gloves, a disposable tray to place in the toilet to
Count data was then converted to relative abundance for each
hold the feces, 2 tubes, and a scoop to collect and store 2 small
sample. The genome length-normalized and filtered tables were
fecal samples. Participants were asked to collect a fecal sample
used for all downstream analyses.
the day before visits at week 0 (baseline), week 4 (end of treat-
ment period), and week 8 (end of washout period) or as close to
the visit as possible. The collection procedure consisted of
SCFA Analysis
defecating into the disposable tray and placing ~10 to 20 gm of
An ultra-high performance liquid chromatography-tandem
fecal material into each of the 2 tubes. The fecal sample for DNA
mass spectrometry (UPLC-MS/MS) assay was performed on
analysis was placed into a prelabeled DNA/RNA Shield Fecal
human stool samples to determine the SCFA metabolic profile of
Collection Tube (Zymo Research Corp.) containing a liquid
the gut microbiome. The concentrations of SCFAs, including
buffer to lyse and inactivate pathogens and preserve the DNA/
acetic (AA), butyric (BA), isobutyric (IBA), lactic (LA), propionic
RNA. The fecal sample for SCFA analysis was placed into the
(PA), valeric (VA), isovaleric (IVA), 2-methylbutyric (2-MBA),
prelabeled second tube, which was empty. After collection, the
maleic (MA), and succinic (SA) acids in each stool sample was
tubes were sealed in a plastic bag labeled biohazard and stored in
determined based on an established chemical UPLC-MS/MS
the volunteer’s home freezer at 20  C before being brought to
assay method. Details of the method are given in Supplemen-
the study center, where they were stored in a freezer at 80  C
tary Information. Chromatographic and mass spectra parameters
until shipped for analysis as described below.
are as available in the Supplementary Information.

DNA extraction, library preparation, and

sequencing Blood samples
Samples were extracted manually using a Quick-DNA Fecal/ Blood samples were collected from a forearm vein into a 5-mL
Soil Microbe 96 Magbead kit. (Zymo Research), with bead beating SST tube (BD Vacutainer, Beckton, Dickenson and Company,
in individual 0.1 mm glass bead plates. DNA quantification was Franklin Lakes, NJ). The tube was left for ~30 min at room
performed using a Quant-iT PicoGreen dsDNA assay kit (Thermo temperature to allow the blood to clot and then centrifuged for
Fisher Scientific), with fluorescence measured on a TECAN plate 10 min to separate the serum. After centrifugation, the tube was
reader (Tecan Group Ltd.). Samples were indexed using Integrated placed in a refrigerator and sent to LifeLabs within 24h for
DNA Technologies (IDT) Unique Dual Indexes. Libraries were analysis of glucose, insulin, and blood lipids [total- and HDL
pooled, followed by SPRI bead purification and concentration cholesterol, triglycerides (TGs), and calculated LDL cholesterol].
using SpeedBead Magnetic Carboxylate Modified Particles
(Cytiva). The resulting pooled libraries were diluted to a loading
3-Day diet record
concentration of 0.51 nM in 10mM Tris-HCL and spiked with 2%
Diet records were collected and analyzed using the Rx Food, a
PhiX. The pooled libraries were denatured using 2 N NaOH and
validated, image-based dietary assessment app (Inner Analytics)
neutralized using 400 mM Tris-HCl. Shotgun metagenomic
[25].
sequencing was performed on an Illumina NovaSeq instrument
using a NovaSeq S4 sequencing kit and a NovaSeq XP loading kit
(2  150 bp reads) to a target depth of 2 million reads per sample.
Randomization and concealment
Sequences were demultiplexed on the sequencer and then con-
Participants were randomly assigned to one of the two
verted to FASTQ files using bcl2fastq (Illumina). DNA sequences
treatments using blocks of various sizes to enhance allocation
were filtered for low quality (Q-Score < 20) and length (< 50 bp),
concealment. Eighty treatment assignments were sealed in
and adapter sequences were trimmed using cutadapt [21]. Finally,
sequentially numbered opaque envelopes kept by the study
FASTQ files were merged and converted into a single FASTA using
coordinator and assigned to participants in the order they
SHI727 [22]. All sequences were trimmed to a maximum length of
arrived for their baseline visit. Randomization and creation of
100 bp before alignment.
the sealed envelopes were performed at the study site by an in-
dividual who was not aware of which code number corresponded
Sequence alignment and annotation to the test product and which corresponded to the control
DNA sequences were aligned to a curated database containing product. The randomization code was kept in a sealed opaque
all representative genomes in RefSeq (release 81) for bacteria envelope in the regulatory binder to be opened only if needed for
[23], with additional manually curated strains (i.e., Diversigen’s the treatment of a serious medical condition thought to be
Venti database). Alignments were made at 97% identity against related to consumption of the study beverage. However, no such
all reference genomes. Every input sequence was compared with condition occurred during the study.

1300
D. Kwok et al. The Journal of Nutrition 154 (2024) 1298–1308

Calculations alpha diversity metrics observed OTUs, Chao1 Index, and Shan-
BMI was calculated as W/H2 where W was weight in kilo- non diversity were calculated from the rarefied taxa table. The
grams (to the nearest 0.1 kg) and H was height in meters (to the relationship between each response variable and baseline weight,
nearest 0.5 cm). LDL cholesterol (mmol/L) was calculated as TC treatment group, time point, and the interaction between treat-
– HDL – TG/2.2 where TC and HDL, respectively, are total- and ment group and time point was assessed using linear mixed effects
HDL cholesterol (mmol/L) and TG is triglycerides (mmol/L). models of the following form: with baseline weight, treatment,
LDL-C was not calculated if TG was > 4.5 mmol/L. and week as fixed effects and subject as a random effect:
Response_Variable ~ Baseline Weight þ Treatment*lspline(-
Week, knot ¼ 4) þ (1 | Subject). In the above model, response
Statistics variables included taxon abundance, SCFA concentration,
Only participants who completed the study per protocol were weight, BMI, SBP, DBP, pulse, glucose, insulin, TG, cholesterol,
included in the primary analysis. HDL, non-HDL, LDL, HOMA-IR, HOMA-B, observed OTUs,
Chao1 Index, and Shannon diversity.
SCFA data imputation Coefficients, standardized coefficients, and P values associated
For SCFA data that were below the lower limit of quantifi- with each model term were calculated for SCFAs, secondary end
cation (“<LLOQ”), values were imputed using the impute. QRILC points, and alpha diversity metrics. The coefficients of interest
function from the imputeLCMD R package. This function per- were those associated with the Treatment  Week interaction
forms missing values imputation based on quantile regression, a terms for each segment of the analysis (week 0 to week 4 and
method that has been shown to more accurately reflect the true week 4 to week 8). Coefficients for all fixed effects are given in the
properties of the data than replacing <LLOQ data with zeros supplementary information Phylum_coefficients.csv, Family_-
[26]. There was one data point above the upper limit of quan- coefficients.csv, Genus_coefficients.csv, OTU_coefficients.csv,
tification (“>ULOQ”). This >ULOQ data point was removed. SCFA_Coefficients, and Secondary_Endpoint_Coefficients.csv
SCFA concentration data were provided as wet and dry con- Differential abundance testing of taxa was performed at the
centrations in micromoles/g, and analysis was performed on level of OTUs, genera, families, and phyla. To analyze taxonomic
both data types. levels higher than OTUs, reads given the same taxonomic
assignment were summed. For example, all reads assigned to the
Beta diversity
genus Bifidobacterium were summed in the genus-level analysis.
For microbiome data, the filtered taxa table was rarefied to
Technical variation of taxa read counts was simulated and the
the depth of the sample with the fewest reads and then trans-
centered log ratio transformation was performed using the
formed to relative abundance. A dissimilarity matrix was created
aldex.clr function in the ALDEx2 R package. This resulted in 128
by calculating the Bray-Curtis dissimilarity between each sample
Monte Carlo instances, each of which was used for statistical
pair. For SCFA concentration data, after missing data imputation,
testing. Coefficients, standardized coefficients, and P values
the data were normalized by log transformation and scaled to
calculated for each term were averaged over all 128 Monte Carlo
unit variance. Euclidean distance between each sample pair was
instances for each taxon to give the final values. P values were
then calculated to create a distance matrix.
corrected for each term across all tested taxa, secondary end
The same statistical approaches were used to analyze the beta
points, and SCFAs using the false discover rate (FDR) method.
diversity of microbiome data and SCFA data, beginning with the
respective dissimilarity/distance matrices. PERMANOVA was
used to assess treatment group differences at each time point Summary stacked bar plots
using the adonis2 function in the vegan R package. To test dif- In order to create taxonomic summary plots, the taxa table
ferences between time points within each treatment group and to was transformed to relative abundance (proportions), and these
assess the significance of the interaction between the treatment proportions were summed at the genus level. Taxon proportions
group and time point, the permanovaFL function in the LDM R were averaged across all subjects within each treatment week.
package was used with permutations constrained within subject Excluding taxa identified as “Other”, the top 25 taxa were plotted
to account for repeated measurements. PERMANOVA effect sizes in stacked bar plots, with the remaining proportion of taxa rep-
(R2 and omega2) were also determined [27]. resented by a grey bar.
Principal coordinate analysis (PCoA) plots were generated for Reported concentrations of SCFAs (wet and dry) were used to
microbiome data for each comparison. For SCFA data, biplots create stacked bar plots showing profiles of the absolute con-
were generated for each comparison. First, a principal compo- centrations of SCFAs. Other than data imputation for <LLOQ
nents analysis (PCA) plot was generated using the Euclidean values, no data transformations were used before creating these
distance matrix to generate the biplots. Then, vectors corre- plots. SCFA concentrations were averaged across all subjects
sponding to the concentrations of each SCFA were overlaid on within each treatment week.
the plot using the PCA loadings data. This data are provided in
the Supplementary Information. Anthropometric and cardiometabolic data
Data were subjected to ANOVA using the linear model
Linear mixed effects model examining for the main effects of time and treatment and the
A linear mixed effects model was used to assess the differential timetreatment interaction. After demonstrating a significant
abundance of taxa, the differential concentration of SCFAs, dif- interaction the significance of the differences between individual
ferences in secondary end points values, and differences in alpha means were assess using Tukey’s test. The criterion for signifi-
diversity. SCFA concentration data were log-transformed. The cance was 2-tailed P < 0.05.

1301
D. Kwok et al. The Journal of Nutrition 154 (2024) 1298–1308

Power analysis more than n ¼ 1 subject occurred to an equivalent extent on

The study was powered to detect a difference in the primary stevia and sucrose, respectively; they were: bloating (n ¼ 3; 9%
outcome, which was the difference in alpha diversity, beta di- on each treatment), flatulence (n ¼ 1; 3% on each treatment),
versity, and relative abundance of taxa in the microbial popu- diarrhea [n ¼ 2 (6%) and n ¼ 1 (3%)] and constipation [n ¼ 1
lation between the steviol glycoside beverage and the sucrose (3%) on each treatment]. However, nausea was significantly
beverage via shotgun sequencing. Assuming a 10% dropout rate, more common on stevia (n ¼ 5, 15%) than sucrose (n ¼ 0, P ¼
a total of 66 volunteers were recruited with the aim of obtaining 0.027). Three other AE’s, chromhidrosis, light-headedness, and
a sample size of 30 finishers in each treatment group, a number mild burning sensation in the stomach, each occurred in n ¼ 1
sufficient to detect significant differences in microbial pop- participant on sucrose.
ulations based on prior research [11,15,28,29].
Body weight and vital signs
Results Subjects on stevia had, overall, significantly higher body
weight and BMI than those on sucrose (Table 3), but there was no
Participant details significant grouptime interaction by ANOVA. However, be-
tween weeks 0 and 4, subjects on sucrose tended to gain weight
Participant characteristics are listed in Table 2. One hundred
and increase their BMI (P ¼ 0.048). There was a significant in-
individuals were screened, of whom 66 (25 male, 41 female)
crease in BMI (0.3 kg/m2; P ¼ 0.013) in the sucrose group
were eligible and were randomly assigned to consume stevia or
relative to the stevia group after the 4-week intervention period.
sucrose beverages. One participant withdrew consent after the
baseline visit and one was withdrawn after the baseline visit
when it was discovered that they were consuming a product Fasting metabolic parameters
containing stevia. After completing all 3 visits, 5 subjects were There was no significant effect of treatment and no significant
withdrawn for low compliance to the protocol (n ¼ 1), pregnancy grouptime interaction for fasting glucose, insulin, HOMA-IR or
discovered at week 8 visit (n ¼ 1), a serious adverse event un- HOMA-B (Table 4). However, at week 8, both insulin and
related to the investigational product (n ¼ 1), an adverse event HOMA-IR tended to be less than that at week 4 (P ¼ 0.049 and P
unrelated to the investigational product resulting in week 8 visit ¼ 0.041, respectively, by t-test). There was no significant effect
occurring 4 weeks late (n ¼ 1) and consuming a product con- of treatment and no significant grouptime interaction for any
taining sucralose during the study (n ¼ 1) (Figure 1). Thus, n ¼ blood lipid fraction assessed (triglycerides and cholesterol; total,
59 completed the project without serious protocol violations and HDL, non-HDL, or LDL).
are included in the per-protocol analysis.
The medical history and medications used by the 66 Food intake and 3-day diet records
randomly assigned subjects was similar between the treatment Data from 26 subjects for the stevia beverage and 32 subjects
groups (details in Supplementary Information). for the sucrose beverage were evaluated for energy and macro-
Overall, both beverages were well tolerated throughout the nutrient intakes (Table 5). Based on the diet records, overall mean
duration of the study. Adverse events (AEs) were reported for all intakes were stable over time. The only significant main effects of
subjects who were exposed to at least 1 dose of their respective treatment were for total carbohydrate and sugars and were due to
treatment. A total of 90 AEs were recorded of which 68 (76%) the fact that the sucrose group consumed a sugar-sweetened
were considered to be not related or possibly related to the study beverage whereas the stevia group consumed a beverage con-
product (details in Supplementary information), and 22 (24%) taining no carbohydrate. The only significant timetreatment
which were considered to be probably or definitely related to the interaction was for sugars where the sucrose group consumed
study product. AEs related to the study product and occurring in significantly more sugar than stevia at week 4. Also, the changes

TABLE 2
Per protocol population on stevia compared with sucrose at screening
Variable Stevia (n ¼ 27) Sucrose (n ¼ 32) P
Sex (n F:M) [% F] 16:11 [59%] 20:12 [63%] 0.80
Ethnicity (n)1 A B Ca CB CF CL Ch A B Ca CB CF CL Ch P ¼ 0.602
0 0 8 1 0 0 5 1 1 13 0 2 1 6 P ¼ 0.383
F I K L O SA SE F I K L O SA SE
2 0 1 4 0 5 1 0 0 0 2 1 3 2
Age (y) 30.78.4 31.810.3 0.64
Height (cm) 169.48.8 166.28.9 0.16
Weight (kg) 66.47.6 61.27.9 0.014
Body mass index (kg/m2) 23.11.6 22.11.7 0.028
Systolic BP (mmHg) 11811 11511 0.31
Diastolic BP (mmHg) 738 717 0.29
Pulse (beats/min) 7611 7911 0.32

Values are meansSD.

1
n for A¼Arab, B¼Black, Ca¼Caucasian, CB¼Caucasian/Black, CF¼Caucasian/Filipino, CL¼Caucasian/Latin American, Ch¼Chinese,
F¼Filipino, I¼Indigenous, K¼Korean, L¼Latin American, O¼Other (Indocaribbean), SA¼South Asian, SE¼Southeast Asian
2
P value (χ2 test) comparing n for Ca, Ch, L, SA and others (8,5,4,5,5 for stevia compared with 13,6,2,3,8 for sucrose))
3
P value (χ2 test) comparing n for Caucasian and non-Caucasian (8,19 for stevia compared with 13,19 for sucrose).

1302
D. Kwok et al. The Journal of Nutrition 154 (2024) 1298–1308

FIGURE 1. Consort Flow Diagram

in sugars intake between weeks 0 and 4 and weeks 4 and 8 on significant difference when extrapolated to the general popula-
sucrose differed from those in the stevia group by t-test. tion. At week 0, the SCFA profiles differed between treatment
groups for wet (P ¼ 0.058, R2 ¼ 0.037, omega2 ¼ 0.02) and dry (P
¼ 0.051, R2 ¼ 0.039, omega2 ¼ 0.02) concentrations.
Microbiome analysis
The mean sequencing depth was 4,259,495 reads per sample.
Per-base median quality scores were all above 30 after quality Taxonomic differential abundance
control. Samples for microbiome analysis consisted of 26 sub- Other than differences in the intercept, there were no signif-
jects in the stevia treatment group and 32 subjects in sucrose icant findings after correcting for the false discovery rate within
treatment group at all 3 timepoints. each taxonomic level. The top twenty five most abundant taxa or
None of the alpha diversity metrics was found to differ taxa that could not be identified (listed as “other”) and listed in
significantly in the rate of change from week 0 to 4 or week 4 to 8 Figure 2A–D. There were, however, significant model terms
between the 2 treatment groups. For taxonomic and SCFA beta before FDR correction for several OTUs, genera, and families.
diversity, there were no significant treatment group: time point Most importantly, there were some differences in the change in
interactions (P < 0.05) observed, indicating that gut microbial OTU, genus, and family abundances during the treatment period
communities did not significantly diverge from each other over between the two treatment groups. For example, Bacteroides
time when exposed to stevia compared with sucrose. However, dorei decreased in abundance in the stevia group during the
for taxonomic beta diversity there was a significant effect of time treatment period but stayed relatively at level in the sucrose
observed for the sucrose treatment group from week 0 to week 4 group.
(P ¼ 0.047, R2 ¼ 0.056, omega2 ¼ 0). R2 ¼ 0.056 indicates that
time point explained 5.6% of the variation in microbial com- SCFA differential abundance
munities. However, omega2, which is a less biased measure of Nine SCFAs were detected: 2MBA, acetic acid, butyric acid,
effect size [27], was equal to 0, suggesting no biologically isobutyric acid, isovaleric acid (IVA), lactic acid, propionic acid,

Table 3
Body weight and vital sign measurements
End point Stevia Sucrose P1
Week 0 Week 4 Week 8 Week 0 Week 4 Week 8
2 3
Weight (kg) 66.11.5 66.01.5 66.01.5 61.11.4 61.51.4 61.31.4 0.60
BMI (kg/m2)4 23.00.3 22.90.3 22.90.3 22.00.3 22.20.33,5 22.10.3 0.34
SBP (mmHg) 114.92.4 116.22.4 114.92.4 113.61.9 114.82.3 113.32.1 0.99
DBP (mmHg) 72.11.4 73.11.4 73.21.4 72.41.5 73.81.5 71.31.36 0.50
Pulse (b/min) 70.51.5 74.61.5 72.21.5 78.92.1 78.01.7 79.6 1.9 0.28

Values are meansSEM in n ¼ 27 on stevia and n ¼ 32 on sucrose.

Abbreviations: BMI, body mass index; SBP, systolic blood pressure; DBP, diastolic blood pressure.
1
Significance of grouptime interaction by ANOVA.
2
Significant main effect of treatment group by ANOVA (P ¼ 0.014).
3
Significant difference from Week 0 by paired t-test (P < 0.05).
4
Significant main effect of treatment group by ANOVA (P ¼ 0.034).
5
Change from Week 0 to 4 on Sucrose (þ0.140.06) significantly different from that on Stevia (0.110.08) by t-test (P ¼ 0.013).
6
Significant difference from Week 4 by paired t-test (P ¼ 0.003)

1303
D. Kwok et al. The Journal of Nutrition 154 (2024) 1298–1308

TABLE 4
Fasting metabolic parameters
End point Stevia Control P1
Week 0 Week 4 Week 8 Week 0 Week 4 Week 8
Glucose (mmol/L) 4.850.09 4.850.09 4.750.09 4.860.07 4.840.06 4.830.07 0.77
Insulin (pmol/L)3 43.0 [36.0, 63.0] 45.0 [36.0, 74.5] 42.02 [32.0, 61.5] 47.0 [35.8, 59.5] 43.5 [33.0, 62.3] 44.0 [36.0, 61.5] 0.71
HOMA-IR4 1.49 [1.24, 2.36] 1.47 [1.27, 2.73] 1.382 [1.07, 2.25] 1.77 [1.35, 1.99] 1.48 [1.09, 2.36] 1.61 [1.27, 2.19] 0.76
HOMA-B4 126 [93, 181] 138 [101, 176] 117 [76, 168] 127 [85, 164] 129 [88, 158] 121 [89, 159] 0.97
TG (mmol/L)3 0.75 [0.62, 0.98] 0.89 [0.60, 1.16] 0.81 [0.61, 1.18] 0.96 [0.74, 1.25] 1.06 [0.73, 1.30] 0.98 [0.73, 1.42] 0.85
Chol (mmol/L) 4.870.17 4.820.17 4.730.17 4.790.18 4.770.18 4.680.17 0.98
HDL (mmol/L)5 1.670.07 1.670.07 1.620.07 1.520.05 1.520.05 1.470.05 1.00
Non-HDL (mmol/L) 3.200.19 3.150.19 3.110.19 3.270.18 3.250.18 3.210.17 0.98
LDL (mmol/L)6 2.810.17 2.740.17 2.660.17 2.780.16 2.740.16 2.700.15 0.92

Values are meansSEM or medians [interquartile range] in n ¼ 27 on stevia and n ¼ 32 on control.

Abbreviations: HOMA-IR, homeostatic model assessment of insulin resistance; HOMA-B, homeostatic model assessment of β-cell function; TG, tri-
glycerides; Chol, total cholesterol; HDL, high-density lipoprotein cholesterol; non-HDL, Chol-HDL; LDL, low-density lipoprotein cholesterol.
1
Significance of treatmenttime interaction by ANOVA.
2
Significant difference from Week 4 by paired t-test (insulin, P ¼ 0.049; HOMA-IR, P ¼ 0.041)
3
Normalized by log transformation.
4
Normalized by square-root transformation.
5
Significant main effect of treatment group by ANOVA (P ¼ 0.040).
6
LDL could not be calculated due to high triglycerides for one subject at Week 4 on Control; the missing value was imputed as the mean of weeks
0 and 8.

succinic acid, and valeric acid, with acetic acid being the most included a significant effect of treatment group at baseline for
abundant in all measurements. Differential concentration analysis 2MBA (wet concentration) and IVA (wet and dry concentration), a
of SCFAs revealed no significant findings after FDR correction significant non-zero change in IVA wet and dry concentration
(Table 6 and Figure 3). However, ignoring the intercept term, from week 0 to 4, and a significant difference in the change in IVA
there were six significant findings before FDR correction. These dry concentration between treatment groups from week 0 to 4.

TABLE 5
Energy, macronutrient, and cholesterol intakes
Nutrient Treatment Week 0 Week 4 Week 8 Mean weeks 0–8
Energy (kcal) Stevia 140365 143682 148083 144062
Control 163572 155371 155875 158262
Total carbohydrate (g) Stevia 16710 15710 16911 1648
Control 1908 1897 1718 18371
Sugars (g)2 Stevia 52.65.13 45.63.93 50.75.73 49.73.9
Control 60.55.13,4 75.24.54 49.24.43 61.64.15
Dietary fiber (g) Stevia 13.2 [9.6, 14.6] 12.4 [9.2, 18.2] 13.5 [10.1, 18.9] 13.9 [10.4, 15.4]
Control 14.8 [11.4, 20.2] 13.4 [10.0, 17.3] 15.0 [11.6, 17.6] 14.1 [12.0, 17.9]
Total fat (g) Stevia 55.22.9 60.74.2 60.83.9 58.92.9
Control 65.84.1 61.83.9 67.04.4 64.93.4
Trans fatty acids (g) Stevia 0.80 [0.53, 1.11] 0.90 [0.57, 1.21] 0.82 [0.42, 1.07] 0.90 [0.74, 1.21]
Control 0.88 [0.42, 1.47] 0.63 [0.40, 1.02] 0.76 [0.51, 0.87] 0.87 [0.56, 1.12]
Saturated fatty acids (g) Stevia 18.01.1 19.41.6 19.61.6 19.01.2
Control 20.61.5 19.41.4 21.01.6 20.41.3
Monounsaturated fatty acids (g) Stevia 13.20.9 14.11.1 15.81.3 14.40.8
Control 15.91.5 13.41.0 16.51.3 15.31.0
Polyunsaturated fatty acids (g) Stevia 7.6 [5.9, 10.5] 8.0 [5.2, 10.3] 6.9 [5.0, 10.1] 7.7 [6.7, 10.0]
Control 9.0 [5.8, 11.1] 6.7 [4.9, 11.8] 8.3 [6.3, 12.6] 8.6 [6.9, 11.1]
Cholesterol (mg) Stevia 24929 27032 28235 26724
Control 29431 21822 28330 26524
Protein (g) Stevia 58.02.5 64.64.2 64.83.7 62.42.8
Control 68.74.2 59.12.9 67.54.6 65.13.4

Values are meansSEM (if values were normally distributed) or medians [interquartile range] (if values were not normally distributed) in n ¼ 26 on
stevia and n ¼ 32 on control.
1
Significantly different from mean on Stevia (main effect of treatment by ANOVA), P ¼ 0.041.
2
Significant timetreatment group interaction, P ¼ 0.010. The changes from Weeks 0 to 4 and 4 to 8, respectively, on Control, þ14.73.5 and
26.03.4, differ from those on Stevia, 6.95.6 and þ5.15.1 by t-test (P ¼ 0.001 and P < 0.001).
3
Means not sharing the same superscript differ significantly by Tukey’s test (P < 0.05).
4
Means not sharing the same superscript differ significantly by Tukey’s test (P < 0.05).
5
Significantly different from mean on Stevia (main effect of treatment by ANOVA), P ¼ 0.023

1304
D. Kwok et al. The Journal of Nutrition 154 (2024) 1298–1308

FIGURE 2. Stacked bar plots showing relative abundance of genera averaged across all subjects within each treatment week. (A) Phylum. (B)
Family. (C) Genus. (D) Species. Stacked bar plots showing the relative abundance of the most abundant taxa averaged across all subjects within
each treatment week at the phylum (A), family (B), genus (C), and species (D) level. "Other" represents taxa that were not identified at a particular
taxonomic level or were not in the top 25 most abundant taxa.

Discussion largely influence the HGM. NNS are very popular and widely
used food ingredients that provide a sweet taste with little to no
The HGM is often referred to as the second human genome calories. To date, there are only a few human trials assessing the
and performs several essential tasks for the host’s health. Overall effects of steviol glycosides (stevia) consumption on the HGM.
diets, pharmaceuticals, and individual food ingredients can Steviol glycosides interact with colonic bacteria of the Bacter-
oidacea family, which can metabolize steviol glycosides by
removing the sugar residues conjugated to steviol [28,29]. There
TABLE 6 are in vitro and in vivo animal model studies [28–45] investi-
Weekly mean SCFA concentrations on a dry weight basis following
gating the effects of steviol glycoside supplementation on gut
treatments of stevia and sucrose
microbiome, showing conflicting effects largely due to con-
Analyte Treatment founding factors including steviol glycoside doses and type, diet,
Stevia Sucrose sample size, controls, and study design [46,47]. In vitro studies
Week Week have revealed that incubation of mixed fecal bacteria from
human volunteers with various steviol glycosides did not
0 4 8 0 4 8
significantly influence the microbial composition of fecal cul-
Mean Mean Mean Mean Mean Mean tures [29–31]. A study testing steviol glycosides and erythritol
 SD  SD  SD  SD  SD  SD
using six representatives of the gut microbiota in vitro found no
2MBA 3.36  2.77  3.12  3.18  3.11  3.50 
1.70 1.15 2.03 1.69 1.06 2.49 impact on bacterial growth, albeit treatment with erythritol
Acetic Acid 216  210  206  222  229  228  resulted in an enhancement of butyric and pentanoic acid pro-
151 122 136 148 138 172 duction when tested using a human gut microbial community
Butyric 55.0  43.7  54.8  52.2  42.8  50.4  [39]. Gerasimidis et al., [40] in an in vitro study, demonstrated
Acid 47.0 30.9 49.3 38.7 33.6 57.5
that fermentation of fecal samples from thirteen healthy volun-
Isobutyric 9.45  7.74  9.45  5.94  5.57  7.44 
Acid 13.5 9.91 12.4 4.35 4.71 13.7 teers in batch cultures with stevia resulted in increased Shannon
Isovaleric 4.19  3.38  3.61  4.34  3.84  4.47  diversity index; the researchers reported that this was due to an
Acid 2.11 1.55 2.56 2.49 1.57 3.50 effect on microbiome community evenness rather than an impact
Lactic Acid 3.45  4.11  8.15  8.35  10.7  4.42  on OTU richness and overall drew the conclusion that stevia
1.56 1.41 8.02 7.17 17.0 1.26
showed no or minimal effects on the broad composition and
Propionic 71.1  67.7  70.9  53.2  62.4  76.6 
Acid 71.3 80.9 59.8 33.8 54.0 80.0 fermentation capacity of the fecal microbiome. The effects of
Succinic 22.1  3.80  22.2  42.5  21.4  8.97  stevia on the gut microbiome in rodent studies have also been
Acid 39.6 1.11 39.3 66.5 34.8 4.90 evaluated. Nettleton et al. [41] examined the effects of stevia
Valeric 7.68  7.69  6.87  6.31  6.86  7.20  (rebaudioside A) consumption on gut microbiota in male
Acid 6.02 7.12 4.23 2.63 4.38 6.81
Sprague-Dawley rats that consumed either water or rebaudioside

1305
D. Kwok et al. The Journal of Nutrition 154 (2024) 1298–1308

FIGURE 3. Column plot showing concentration (wet weight) of short-chain fatty acids averaged across all subjects within each treatment week.
Reported concentration data has been log10 transformed to allow visual comparison of all short-chain fatty acids. Standard error bars are shown.

A for 9 weeks. Rebaudioside A consumption alone did not alter and stevia increased plasma insulin levels over 2 weeks, with
weight gain or glucose tolerance compared with the control diet only the stevia group returning to baseline levels after the
but increased SCFAs acetate and valerate, which were positively intervention period. The type of stevia is listed as commercial,
associated with fat mass and total weight. Another study [34] and the dose is reported to be 180 mg stevia or 75% of the ADI.
examined the effects of rebaudioside A administration at either a However, the ADI for stevia is defined based on steviol equiva-
low dose (5.5 mg/kg/d) or a high dose (139 mg/kg/d) in SPF lents and not fully intact steviol glycosides (as are present in
BALB/c mice over 4 weeks; the control group consumed distilled commercially available stevia), so it remains unclear if the dose
water. Although no difference was found in the total number of employed is representative of the ADI or not. Commercially
anaerobic bacteria, enterococci, enterobacteria, or lactobacilli, available stevia in packets contains both stevia and a bulking
there was an increase in the diversity of lactobacilli species for agent, which can influence changes in gut microflora, a fact that
the mice consuming high dose of rebaudioside A. A study of the the authors note in their study.
administration of steviol glycosides to Cebus apella (female In this study, steviol glycosides were fed at a relatively high
monkeys) for 2 weeks resulted in observable changes in the dose (25% ADI), reflecting the highest daily dose typically
microbial structure, alpha, and beta diversity (increased) of the consumed in North America, Europe, and Australia [47], but are
gut microbial community [39]. It is difficult to extrapolate these still well under stevia’s established ADI. In this study, we found
results to human studies, however, there are published studies, no significant effects of stevia on the relative abundance of gut
including one involving stevia in commercially available sachets, microflora over 4 weeks. This is consistent with recent studies on
describing the effects of NNS on gut microbiome in humans. A other NNS, which found no significant effects of NNS use on gut
cross-sectional study in the United States of human participants microbiome profile and function [18–20]. This study found ste-
showed that bacterial abundance does not differ between con- via to have no significant effects over 4 weeks on several
sumers and nonconsumers of NNS, and in particular aspartame anthropometric, glucoregulatory, and cardiometabolic factors,
and acesulfame-K consumers, but bacterial diversity was signif- consistent with other studies on stevia [48,49]. Compared with
icantly different across consumers and nonconsumers of aspar- sucrose, at 4 weeks, there was a significant difference in BMI by
tame and acesulfame-K [17]. A study [15] using saccharin -0.3 units (P ¼ 0.013) in the stevia group and a within-group
examined 7-day saccharin consumption effects on gut microflora increase in weight in the sucrose group. This modest difference
in healthy subjects and concluded that saccharin consumption in BMI and weight is most likely due to the theoretical caloric
induces glucose intolerance and changes in gut microbiota. A differential (120 kcal) between the two groups, as no differences
subsequent study [16] by the same group evaluated the effects of were observed in food or macronutrient intake other than sugar
four NNS- saccharine, sucralose, aspartame, and stevia in as noted in the diet records.
commercially available sachets over 2 weeks in healthy adults, SCFA are the primary metabolites produced in the colon by
along with glucose and no supplement control. The authors re- bacterial fermentation. Colonic SCFA has been recognized to
ported that each NNS could alter fecal microbiome, with only play a role in human health, acting as metabolic switches regu-
saccharin and sucralose also impairing glycemic responses in a lating metabolic diseases, neuro-immunoendocrine system, in-
process reported to be mediated through changes in the gut flammatory bowel diseases, and human health functions. Nine
microbiome. Changes in microbial members by stevia adminis- SCFAs were detected in this study, and stevia was found to have
tration were noted in this study for clostridium sp., bacteroides no significant effects on SCFA production after a 4-week inter-
glodsteinii, prevotella sp, and others, although relative abundances vention period and a 4-week postintervention period, compared
or fold changes were not listed. The study also found that glucose with sucrose at week 0.

1306
D. Kwok et al. The Journal of Nutrition 154 (2024) 1298–1308

Study strengths and limitations Funding

This study found no significant effects of steviol glycosides, This study was funded by Cargill, Inc, Wayzata, MN. CS, TH
a common zero-calorie sweetener, on human gut microflora and AC were employed by Cargill when this work was per-
profile and function. Strengths of this study include the use of a formed. Cargill is a manufacturer of stevia products.
relatively high dose of stevia (25% ADI), a larger number of
volunteers compared with previous similar studies using NNS, Data availability statement
a more precise shotgun sequencing DNA analysis, the detection Available data can be found in Appendix A Supplementary
of 9 SFCA compared with previous NNS studies and 3 fecal Data.
collection points (at baseline, after the intervention, and after a
washout period). Limitations of this study include a parallel Conflicts of interest
design rather than a crossover design. A water beverage (third The authors declare that they have no known conflicts of in-
arm) was considered in the design, however, a decision was terests including competing financial interests or personal re-
made to use a typical comparative beverage that contained lationships that could have appeared to influence the work
sucrose, as sucrose-sweetened beverages are often substituted reported in this paper.
with zero-calorie sweetener beverages. A similar study
compared two high-intensity sweetened beverages (aspartame Appendix A. Supplementary data
and sucralose) on human gut microflora but also did not Supplementary data to this article can be found online at
evaluate a placebo (water) control [19]. Although a parallel https://doi.org/10.1016/j.tjnut.2024.01.032.
design was chosen, fecal samples were collected at baseline, 4
weeks, and again at 8 weeks post-intervention, and no signif-
icant differences were observed due to time or treatment.
References
There was a significant difference in BMI and weight in the
[1] C.A. Lozupone, J.I. Stombaugh, J.I. Gordon, J.K. Jansson, R. Knight,
stevia group compared with the sucrose group after randomi- Diversity, stability and resilience of the human gut microbiota, Nature
zation. This was mainly due to a higher percentage of males in 489 (7415) (2012) 220–230, https://doi.org/10.1038/nature11550.
the stevia group (41%) compared with the sucrose group [2] F. B€ackhed, R.E. Ley, J.L. Sonnenburg, D.A. Peterson, J.I. Gordon, Host-
bacterial mutualism in the human intestine, Science. (New York, N.Y.).
(37%) after randomization. These differences in weight and 307 (5717) (2005) 1915–1920, https://doi.org/10.1126/
BMI were accounted for in our statistical models and did not science.1104816.
affect the results. [3] V. Iebba, V. Totino, A. Gagliardi, F. Santangelo, F. Cacciotti,
M. Trancassini, et al., Eubiosis and dysbiosis: the two sides of the
microbiota, New. Microbiol. 39 (2016) 1–12.
Conclusion [4] L.K. Ursell, J.L. Metcalf, L.W. Parfrey, R. Knight, Defining the human
microbiome, Nutr. Rev. 70 (Suppl 1) (2012), https://doi.org/10.1111/
j.1753-4887.2012.00493.x. S38–44.
To date, only a small number of human studies have evalu- [5] E.M. Quigley, Gut bacteria in health and disease, Gastroenterol.
ated the effects of high-intensity sweeteners on human gut Hepatol. (NY) 9 (9) (2013) 560–569.
microflora, and the results are conflicting [15–20]. This study [6] Y. Fan, O. Pedersen, Gut microbiota in human metabolic health and
disease, Nat. Rev. Microbiol. 19 (1) (2021) 55–71, https://doi.org/
has shown that a steviol glycoside-sweetened beverage at 25% of 10.1038/s41579-020-0433-9.
the ADI over 4 weeks was well tolerated and had no significant [7] H.J. Flint, K.P. Scott, P. Louis, S.H. Duncan, The role of the gut
effects on human gut microflora profile, SCFA production, and microbiota in nutrition and health, Nat. Rev. Gastroenterol. Hepatol. 9
(10) (2012) 577–589, https://doi.org/10.1038/nrgastro.2012.156.
several glycemic, cardiometabolic, and anthropometric func-
[8] E.S. Chambers, T. Preston, G. Frost, D.J. Morrison, Role of gut
tions. In addition, a slight improvement in BMI was observed in microbiota-generated short-chain fatty acids in metabolic and
the stevia group compared with the sucrose group. We report cardiovascular health, Curr. Nutr. Rep. 7 (4) (2018) 198–206, https://
that steviol glycosides can be a useful alternative to lower con- doi.org/10.1007/s13668-018-0248-8.
[9] E.E. Canfora, J.W. Jocken, E.E. Blaak, Short-chain fatty acids in control
sumption of added sugar while having no effects on human gut of body weight and insulin sensitivity, Nat. Rev. Endocrinol. 11 (10)
microflora. (2015) 577–591, https://doi.org/10.1038/nrendo.2015.128.
[10] A.R. Lobach, A. Roberts, I.R. Rowland, Assessing the in vivo data on
low/no-calorie sweeteners and the gut microbiota, Food. Chem.
Acknowledgments Toxicol. 124 (2019) 385–399, https://doi.org/10.1016/
j.fct.2018.12.005.
[11] L.A. David, C.F. Maurice, R.N. Carmody, D.B. Gootenberg, J.E. Button,
We thank all the participants for their participation in this
B.E. Wolfe, et al., Diet rapidly and reproducibly alters the human gut
clinical trial. We also thank Dr. Nikoleta Stamataki for compiling microbiome, Nature 505 (7484) (2014) 559–563, https://doi.org/
information on stevia and other NNS on human gut microflora 10.1038/nature12820.
and Melinda Montgomery for production assistance with the [12] K. Ray, Gut microbiota: filling up on fibre for a healthy gut, Nat. Rev.
Gastroenterol. Hepatol. 15 (2) (2018) 67, https://doi.org/10.1038/
investigational products. nrgastro.2018.2.
[13] H.J. Flint, S.H. Duncan, K.P. Scott, P. Louis, Links between diet, gut
microbiota composition and gut metabolism, Proc. Nutr. Soc. 74 (1)
Author contributions (2015) 13–22, https://doi.org/10.1017/S0029665114001463.
TMSW and FA-Y performed the study; NS performed the DNA [14] J.E. Nettleton, R.A. Reimer, J. Shearer, Reshaping the gut microbiota:
analysis; DK and CL performed the SCFA analysis; and CS impact of low calorie sweeteners and the link to insulin resistance?
Physiol. Behav. 164 (Pt B) (2016) 488–493, https://doi.org/10.1016/
designed and prepared the beverages. CS, TH, and AC assisted
j.physbeh.2016.04.029.
with the study design and all authors read and approved the final [15] J. Suez, T. Korem, D. Zeevi, G. Zilberman-Schapira, C.A. Thaiss,
manuscript. O. Maza, et al., Artificial sweeteners induce glucose intolerance by

1307
D. Kwok et al. The Journal of Nutrition 154 (2024) 1298–1308

altering the gut microbiota, Nature 514 (7521) (2014) 181–186, [33] I. Deniņa, P. Semjonovs, A. Fomina, R. Treimane, R. Linde, The
https://doi.org/10.1038/nature13793. influence of stevia glycosides on the growth of Lactobacillus reuteri
[16] J. Suez, Y. Cohen, R. Valdes-Mas, U. Mor, M. Dori-Bachash, S. Federici, strains, Lett. Appl. Microbiol. 58 (3) (2014) 278–284, https://doi.org/
et al., Personalized microbiome-driven effects of non-nutritive 10.1111/lam.12187.
sweeteners on human glucose tolerance, Cell 185 (18) (2022) [34] S. Li, T. Chen, S. Dong, Y. Xiong, H. Wei, F. Xu, The effects of
3307–3328, https://doi.org/10.1016/j.cell.2022.07.016. rebaudioside A on microbial diversity in mouse intestine, Food Sci.
[17] C.L. Frankenfeld, M. Sikaroodi, E. Lamb, S. Shoemaker, P.M. Gillevet, Technol. Res. 20 (2014) 459–467, https://doi.org/10.3136/fstr.20.459.
High-intensity sweetener consumption and gut microbiome content and [35] Q.P. Wang, D. Browman, H. Herzog, G.G. Neely, Non-nutritive
predicted gene function in a cross-sectional study of adults in the United sweeteners possess a bacteriostatic effect and alter gut microbiota in
States, Ann. Epidemiol. 25 (10) (2015) 736–742, https://doi.org/ mice, PLoS. One. 13 (2018) 7–20, https://doi.org/10.1371/
10.1016/j.annepidem.2015.06.083. journal.pone.0199080.
[18] P. Thomson, R. Santiba~ nez, C. Aguirre, J.E. Galgani, D. Garrido, Short- [36] V. Markus, O. Share, K. Teralı, N. Ozer, R.S. Marks, A. Kushmaro, et al.,
term impact of sucralose consumption on the metabolic response and Anti-quorum sensing activity of stevia extract, stevioside, rebaudioside
gut microbiome of healthy adults, Br. J. Nutr. 122 (8) (2019) 856–862, A and their aglycon steviol, Molecules 25 (22) (2020) 5480–5496,
https://doi.org/10.1017/s0007114519001570. https://doi.org/10.3390/molecules25225480.
[19] S.Y. Ahmad, J. Friel, D. Mackay, The effects of non-nutritive artificial [37] L. Boling, D.A. Cuevas, J.A. Grasis, H.S. Kang, B. Knowles, K. Levi, et al.,
sweeteners, aspartame and sucralose, on the gut microbiome in healthy Dietary prophage inducers and antimicrobials: toward landscaping the
adults: secondary outcomes of a randomized double-blinded crossover human gut microbiome, Gut. Microbes. 11 (4) (2020) 721–734, https://
clinical trial, Nutrients 12 (11) (2020) 3408–3424, https://doi.org/ doi.org/10.1080/19490976.2019.1701353.
10.3390/nu12113408. [38] J.E. Nettleton, N.A. Cho, T. Klancic, A.C. Nicolucci, J. Shearer,
[20] J. Serrano, K.R. Smith, A.L. Crouch, V. Sharma, F. Yi, V. Vargova, et al., S.L. Borgland, et al., Maternal low-dose aspartame and stevia
High-dose saccharin supplementation does not induce gut microbiota consumption with an obesogenic diet alters metabolism, gut microbiota
changes or glucose intolerance in healthy humans and mice, Microbiome 9 and mesolimbic reward system in rat dams and their offspring, Gut 69
(1) (2021) 11–29, https://doi.org/10.1186/s40168-020-00976-w. (10) (2020) 1807–1817, https://doi.org/10.1136/gutjnl-2018-317505.
[21] M. Martin, Cutadapt removes adapter sequences from high-throughput [39] KK Mahalak, J Firrman, PM Tomasula, A Nu~ nez, JJ Lee, K Bittinger,
sequencing reads, EMBnet. Journal. 17 (2011) 10–12, https://doi.org/ W Rinaldi, LS Liu, Impact of steviol glycosides and erythritol on the
10.14806/EJ.17.1.200. human and Cebus apella gut microbiome, J. Agric. Food. Chem. 68 (46)
[22] G.A. Al-Ghalith, B. Hillmann, K. Ang, R. Shields-Cutler, D. Knights, SHI7 (2020) 13093–13101, https://doi.org/10.1021/acs.jafc.9b06181.
is a self-learning pipeline for multipurpose short-read DNA quality [40] K. Gerasimidis, K. Bryden, X. Chen, E. Papachristou, A. Verney, M. Roig,
control, mSystems 3 (3) (2018) e00202–17, https://doi.org/10.1128/ et al., The impact of food additives, artificial sweeteners and domestic
msystems.00202-17. hygiene products on the human gut microbiome and its fibre
[23] N.A. O'Leary, M.W. Wright, J.R. Brister, S. Ciufo, D. Haddad, fermentation capacity, Eur. J. Nutr. 59 (7) (2020) 3213–3230, https://
R. McVeigh, et al., Reference sequence (RefSeq) database at NCBI: doi.org/10.1007/s00394-019-02161-8.
current status, taxonomic expansion, and functional annotation, [41] J.E. Nettleton, T. Klancic, A. Schick, A.C. Choo, J. Shearer,
Nucleic. Acids. Res. 44 (D1) (2016) D733–D745, https://doi.org/ S.L. Borgland, et al., Low-dose stevia (rebaudioside A) consumption
10.1093/nar/gkv1189. perturbs gut microbiota and the mesolimbic dopamine reward system,
[24] G.A. Al-ghalith, D. Knights, BURST enables optimal exhaustive DNA Nutrients 11 (6) (2019) 1248–1265, https://doi.org/10.3390/
alignment for big data, 2017. nu11061248.
[25] K. Jefferson, E. Choi, D. Lichti, J. Alfonsi, B. Patel, J. Hamilton, et al., Rx [42] M. Yu, T. Gao, Z. Liu, X. Diao, Effects of dietary supplementation with
food app: a proof-of-concept study of an image-based dietary high fiber (stevia residue) on the fecal flora of pregnant sows, Animals
assessment mobile application, Curr. Dev. Nutr. 5 (Suppl 2) (2021) (Basel) 10 (12) (2020) 2247–2266.
1000–1001, https://doi.org/10.1093/cdn/nzab052_004. [43] A.L. de la Garza, B. Romero-Delgado, A.M. Martínez-Tamez,
[26] R. Wei, J. Wang, M. Su, E. Jia, S. Chen, T. Chen, et al., Missing value M. Cardenas-Tueme, B.D. Camacho-Zamora, D. Matta-Yee-Chig, et al.,
imputation approach for mass spectrometry-based metabolomics data, Maternal sweeteners intake modulates gut microbiota and exacerbates
Sci. Rep. 8 (1) (2018) 663, https://doi.org/10.1038/s41598-017- learning and memory processes in adult male offspring, Front. Pediatr.
19120-0. 9 (2022) 746437, https://doi.org/10.3389/fped.2021.746437.
[27] B.J. Kelly, R. Gross, K. Bittinger, S. Sherrill-Mix, J.D. Lewis, [44] W. Wang, J.E. Nettleton, M.G. G€anzle, R.A. Reimer, A metagenomics
R.G. Collman, et al., Power and sample-size estimation for microbiome investigation of intergenerational effects of non-nutritive sweeteners on
studies using pairwise distances and PERMANOVA, Bioinformatics 31 gut microbiome, Front. Nutr. 8 (2022) 795–848, https://doi.org/
(15) (2015) 2461–2468, https://doi.org/10.1093/bioinformatics/ 10.3389/fnut.2021.795848.
btv183. [45] S.L. Becker, E. Chiang, A. Plantinga, H.V. Carey, G. Suen, S.J. Swoap,
[28] C. Gardana, P. Simonetti, E. Canzi, R. Zanchi, P. Pietta, Metabolism of Effect of stevia on the gut microbiota and glucose tolerance in a murine
stevioside and rebaudioside A from stevia rebaudiana extracts by model of diet-induced obesity, FEMS. Microbiol. Ecol. 96 (6) (2020),
human microflora, J. Agric. Food. Chem. 51 (22) (2003) 6618–6622, https://doi.org/10.1093/femsec/fiaa079 fiaa079.
https://doi.org/10.1021/jf0303619. [46] A.N. Kasti, M.D. Nikolaki, K.D. Synodinou, K.N. Katsas, K. Petsis,
[29] E. Koyama, K. Kitazawa, Y. Ohori, O. Izawa, K. Kakegawa, A. Fujino, et S. Lambrinou, et al., The effects of stevia consumption on gut bacteria:
al., In vitro metabolism of the glycosidic sweeteners, stevia mixture and friend or foe? Microorganisms 10 (4) (2022) 744, https://doi.org/
enzymatically modified stevia in human intestinal microflora, Food. 10.3390/microorganisms10040744.
Chem. Toxicol. 41 (3) (2003) 359–374, https://doi.org/10.1016/ [47] A.G. Renwick, The use of a sweetener substitution method to predict
s0278-6915(02)00235-1. dietary exposures for the intense sweetener rebaudioside A, Food.
[30] S. Purkayastha, S. Bhusari, G. Pugh Jr., X. Teng, D. Kwok, S.M. Tarka, In Chem. Toxicol. 46 (Suppl 7) (2008), https://doi.org/10.1016/
vitro metabolism of rebaudioside E under anaerobic conditions: j.fct.2008.05.009. S61–9.
comparison with rebaudioside A, Regul. Toxicol. Pharmacol. 72 (3) [48] P. Samuel, K.T. Ayoob, B.A. Magnuson, U. W€ olwer-Rieck,
(2015) 646–657, https://doi.org/10.1016/j.yrtph.2015.05.019. P.B. Jeppesen, P.J. Rogers, et al., Stevia leaf to stevia sweetener:
[31] S. Purkayastha, G. Pugh Jr., B. Lynch, A. Roberts, D. Kwok, exploring its science, benefits, and future potential, J. Nutr. 148 (7)
S.M. Tarka Jr., In vitro metabolism of rebaudioside B, D, and M under (2018) 1186S–1205S, https://doi.org/10.1093/jn/nxy102.
anaerobic conditions: comparison with rebaudioside A, Regul. Toxicol. [49] K.A. Higgins, R.D. Mattes, A randomized controlled trial contrasting the
Pharmacol. 68 (2) (2014) 259–268, https://doi.org/10.1016/ effects of 4 low-calorie sweeteners and sucrose on body weight in adults
j.yrtph.2013.12.004. with overweight or obesity, Am. J. Clin. Nutr. 109 (5) (2019)
[32] G. Kunova, V. Rada, A. Vidaillac, I. Lisova, Utilisation of 1288–1301, https://doi.org/10.1093/ajcn/nqy381.
steviol glycosides from stevia rebaudiana (Bertoni) by [50] A. Conz, M. Salmona, L. Diomede, Effect of non-nutritive sweeteners on
lactobacilli and bifidobacteria in in vitro conditions, Folia. Microbiol. the gut microbiota, Nutrients 15 (8) (2023) 1869–1898, https://
(Praha). 59 (3) (2014) 251–255, https://doi.org/10.1007/s12223-013- doi.org/10.3390/nu15081869.
0291-1.

1308