#### The Glycemic Response The glycemic response is the body's r...
### Background on Diabetes Diabetes is an umbrella term referrin...
### What is PPG? Postprandial blood glucose (PPG) levels are a m...
#### Measuring Glycemic Response Today, glycemic response is pre...
#### Continuous Glucose Monitors (CGMs) CGMs are commonly used b...
#### (Some) Microbes are your Friends! Humans have complex mutua...
### Measurements of Blood Glucose Blood glucose is most commonly...
#### The Kyoto Encyclopedia of Genes and Genomes (KEGG) Minoru K...
Article
Personalized Nutrition by Prediction of Glycemic
Responses
Graphical Abstract
Highlights
d High interpersonal variability in post-meal glucose observed
in an 800-person cohort
d Using personal and microbiome features enables accurate
glucose response prediction
d Prediction is accurate and superior to common practice in an
independent cohort
d Short-term personalized dietary interventions successfully
lower post-meal glucose
Authors
David Zeevi, Tal Korem, Niv Zmora, ...,
Zamir Halpern, Eran Elinav, Eran Segal
Correspondence
eran.elinav@weizman n.ac.il (E.E.),
eran.segal@weizmann.ac.il (E.S.)
In Brief
People eating identical meals present
high variability in post-meal blood
glucose response. Personalized diets
created with the help of an accurate
predictor of blood glucose response that
integrates parameters such as dietary
habits, physical activity, and gut
microbiota may successfully lower post-
meal blood glucose and its long-term
metabolic consequences.
Zeevi et al., 2015, Cell 163, 1079–1094
November 19, 2015 ª2015 Elsevier Inc.
http://dx.doi.org/10.1016/j.cell.2015.11.001
Article
Personalized Nutrition by Prediction
of Glycemic Responses
David Zeevi,
1,2,8
Tal Korem,
1,2,8
Niv Zmora,
3,4,5,8
David Israeli,
6,8
Daphna Rothschild,
1,2
Adina Weinberger,
1,2
Orly Ben-Yacov,
1,2
Dar Lador,
1,2
Tali Avnit-Sagi,
1,2
Maya Lotan-Pompan,
1,2
Jotham Suez,
3
Jemal Ali Mahdi,
3
Elad Matot,
1,2
Gal Malka,
1,2
Noa Kosower,
1,2
Michal Rein,
1,2
Gili Zilberman-Schapira,
3
Lenka Dohnalova
´
,
3
Meirav Pevsner-Fischer,
3
Rony Bikovsky,
1,2
Zamir Halpern,
5,7
Eran Elinav,
3,9,
*
and Eran Segal
1,2,9,
*
1
Department of Computer Science and Applied Mathematics, Weizmann Institute of Science, Rehovot 7610001, Israel
2
Department of Molecular Cell Biology, Weizmann Institute of Science, Rehovot 7610001, Israel
3
Immunology Department, Weizmann Institute of Science, Rehovot 7610001, Israel
4
Internal Medicine Department, Tel Aviv Sourasky Medical Center, Tel Aviv 6423906, Israel
5
Research Center for Digestive Tract and Liver Diseases, Tel Aviv Sourasky Medical Center, Sackler Faculty of Medicine, Tel Aviv University,
Tel Aviv 6423906, Israel
6
Day Care Unit and the Laboratory of Imaging and Brain Stimulation, Kfar Shaul Hospital, Jerusalem Center for Mental Health,
Jerusalem 9106000, Israel
7
Digestive Center, Tel Aviv Sourasky Medical Center, Tel Aviv 6423906, Israel
8
Co-first author
9
Co-senior author
*Correspondence: eran.elinav@weizmann.ac.il (E.E.), eran.segal@weizmann.ac.il (E.S.)
http://dx.doi.org/10.1016/j.cell.2015.11.001
SUMMARY
Elevated postprandial blood glucose levels consti-
tute a global epidemic and a major risk factor for pre-
diabetes and type II diabetes, but existing dietary
methods for controlling them have limited efficacy.
Here, we continuously monitored week-long glucose
levels in an 800-person cohort, measured responses
to 46,898 meals, and found high variability in the
response to identical meals, suggesting that univer-
sal dietary recommendations may have limited
utility. We devised a machine-learning algorithm
that integrates blood parameters, dietary habits, an-
thropometrics, physical activity, and gut microbiota
measured in this cohort and showed that it accu-
rately predicts personalized postprandial glycemic
response to real-life meals. We validated these
predictions in an independent 100-person cohort.
Finally, a blinded randomized controlled dietary
intervention based on this algorithm resulted in
significantly lower postprandial responses and
consistent alterations to gut microbiota configura-
tion. Together, our results suggest that personalized
diets may successfully modify elevated postprandial
blood glucose and its metabolic consequences.
INTRODUCTION
Blood glucose levels are rapidly increasing in the population, as
evident by the sharp incline in the prevalence of prediabetes and
impaired glucose tolerance estimated to affect, in the U.S. alone,
37% of the adult population (Bansal, 2015). Prediabetes, charac-
terized by chronically impaired blood glucose responses, is a sig-
nificant risk factor for type II diabetes mellitus (TIIDM), with up to
70% of prediabetics eventually developing the disease (Nathan
et al., 2007). It is also linked to other manifestations, collectively
termed the metabolic syndrome, including obesity, hypertension,
non-alcoholic fatty liver disease, hypertriglyceridemia, and cardio-
vascular disease (Grundy, 2012). Thus, maintaining normal blood
glucose levels is considered critical for preventing and controlling
the metabolic syndrome (Riccardi and Rivellese, 2000).
Dietary intake is a central determinant of blood glucose levels,
and thus, in order to achieve normal glucose levels it is impera-
tive to make food choices that induce normal postprandial (post-
meal) glycemic responses (PPGR; Gallwitz, 2009). Postprandial
hyperglycemia is an independent risk factor for the development
of TIIDM (American Diabetes Association., 2015a), cardiovascu-
lar disease (Gallwitz, 2009), and liver cirrhosis (Nishida et al.,
2006) and is associated with obesity (Blaak et al., 2012), and
enhanced all-cause mortality in both TIIDM (Cavalot et al.,
2011) and cancer (Lamkin et al., 2009).
Despite their importance, no method exists for predicting
PPGRs to food. The current practice is to use the meal carbohy-
drate content (American Diabetes Association., 2015b; Bao
et al., 2011), even though it is a poor predictor of the PPGR
(Conn and Newburgh, 1936). Other methods aimed at estimating
PPGRs are the glycemic index, which quantifies PPGR to con-
sumption of a single tested food type, and the derived glycemic
load (Jenkins et al., 1981). It thus has limited applicability in as-
sessing the PPGR to real-life meals consisting of arbitrary food
combinations and varying quantities (Dodd et al., 2011),
consumed at different times of the day and at different proximity
to physical activity and other meals. Indeed, studies examining
the effect of diets with a low glycemic index on TIIDM risk, weight
loss, and cardiovascular risk factors yielded mixed results
(Greenwood et al., 2013; Kristo et al., 2013; Schwingshackl
and Hoffmann, 2013).
Cell 163, 1079–1094, November 19, 2015 ª2015 Elsevier Inc. 1079
Nuts (456,000)
Beef (444,000)
Legumes (420,000)
Fruit (400,000)
Poultry (386,000)
Rice (331,000)
Other (4,010,000)
Baked goods (542,000)
Vegetables (548,000)
Sweets (639,000)
Dairy (730,000)
Bread (919,000)
Overall energy documented: 9,807,000 Calories
Glucose (mg/dl)
Time
Anthropometrics
Blood tests
Gut microbiome
16S rRNA
Metagenomics
Questionnaires
Food frequency
Lifestyle
Medical
Diary (food, sleep, physical activity)
Continuous glucose monitoring
Day 1 Day 2 Day 3 Day 4
Day 5 Day 6 Day 7
Standardized meals (50g available carbohydrates)
GGF
Bread Bread Bread &
butter
Bread &
butter
Glucose Glucose Fructose
Per person profiling
Computational analysis
Main
cohort
800 Participants
Validation
cohort
100 Participants
PPGR
prediction
26 Participants
Dietary
intervention
A
Glucose (mg/dl)
Day
BMI
1234567
Standardized meal
Lunch
Snack
Dinner
Postprandial glycemic response
(PPGR; 2-hour iAUC)
D
5,435 days, 46,898 meals, 9.8M Calories, 2,532 exercises
130K hours, 1.56M glucose measurements
BC
Frequency
Frequency
HbA1c%
45% 33% 22% 76% 21% 3%
% Protein
% Carbohydrate
% Fat
F
1000
2000
0
020406080100
Frequency
% of meal
Carbohydrate
Fat
Protein
E
Sleep
PCo1 (10.5%)
PCo2 (5.2%)
G
Study participants MetaHIT - stool
HMP - stool HMP - oral
PCo1 (27.9%)
PCo2 (2.2%)
Using smartphone-adjusted website
Using a subcutaneous sensor (iPro2)
Participant 141
HMP - urogenital
Figure 1. Profiling of Postprandial Glycemic Responses, Clinical Data, and Gut Microbiome
(A) Illustration of our experimental design.
(B and C) Distribution of BMI and glycated hemoglobin (HbA1c%) in our cohort. Thresholds for overweight (BMI R 25 kg/m
2
), obese (BMI R 30 kg/m
2
),
prediabetes (HbA1c% R 5.7%) and TIIDM (R6.5%) are shown.
(legend continued on next page)
1080 Cell 163, 1079–1094, November 19, 2015 ª2015 Elsevier Inc.
More broadly, ascribing a single PPGR to each food assumes
that the response is solely an intrinsic property of the consumed
food. However, the few small-scale (n = 23–40) studies that
examined interpersonal differences in PPGRs found high vari-
ability in the response of different people to the same food
(Vega-Lo
´
pez et al., 2007; Vrolix and Mensink, 2010), but the fac-
tors underlying this variability have not been systematically
studied.
Factors that may affect interpersonal differences in PPGRs
include genetics (Carpenter et al., 2015), lifestyle (Dunstan
et al., 2012), insulin sensitivity (Himsworth, 1934), and exocrine
pancreatic and glucose transporters activity levels (Gibbs
et al., 1995). Another factor that may be involved is the gut micro-
biota. Pioneering work by Jeffrey Gordon and colleagues previ-
ously showed that it associates with the propensity for obesity
and its complications, and later works also demonstrated asso-
ciations with glucose intolerance, TIIDM, hyperlipidemia, and in-
sulin resistance (Le Chatelier et al., 2013; Karlsson et al., 2013;
Qin et al., 2012; Suez et al., 2014; Turnbaugh et al., 2006; Zhang
et al., 2013). However, little is known about the association of gut
microbiota with PPGRs.
Here, we set out to quantitatively measure individualized
PPGRs, characterize their variability across people, and identify
factors associated with this variability. To this end, we continu-
ously monitored glucose levels during an entire week in a cohort
of 800 healthy and prediabetic individuals and also measured
blood parameters, anthropometrics, physical activity, and self-
reported lifestyle behaviors, as well as gut microbiota composi-
tion and function. Our results demonstrate high interpersonalvari-
ability in PPGRs to the same food. We devised a machine learning
algorithm that integrates these multi-dimensional data and accu-
rately predicts personalized PPGRs, which we further validated in
an independently collected 100-person cohort. Moreover, we
show that personally tailored dietary interventions based on these
predictions result in significantly improved PPGRs accompanied
by consistent alterations to the gut microbiota.
RESULTS
Measurements of Postprandial Responses, Clinical
Data, and Gut Microbiome
To comprehensively characterize PPGRs, we recruited 800
individuals aged 18–70 not previously diagnosed with TIIDM
(Figure 1A, Table 1). The cohort is representative of the adult
non-diabetic Israeli population (Israeli Center for Disease Con-
trol, 2014), with 54% overweight (BMI R 25 kg/m
2
) and 22%
obese (BMI R 30 kg/m
2
, Figures 1B, 1C, and S1). These proper-
ties are also characteristic of the Western adult non-diabetic
population (World Health Organization, 2008).
Each participant was connected to a continuous glucose
monitor (CGM), which measures interstitial fluid glucose every
5 min for 7 full days (the ‘connection week’’), using subcutane-
ous sensors (Figure 1D). CGMs estimate blood glucose levels
with high accuracy (Bailey et al., 2014) and previous studies
found no significant differences between PPGRs extracted
from CGMs and those obtained from either venous or capillary
blood (Vrolix and Mensink, 2010). We used blinded CGMs and
thus participants were unaware of their CGM levels during the
connection week. Together, we recorded over 1.5 million
glucose measurements from 5,435 days.
While connected to the CGM, participants were instructed to
log their activities in real-time, including food intake, exercise
and sleep, using a smartphone-adjusted website (www.
personalnutrition.org) that we developed (Figure S2A). Each
food item within every meal was logged along with its weight
by selecting it from a database of 6,401 foods with full nutritional
values based on the Israeli Ministry of Health database that we
further improved and expanded with additional items from certi-
fied sources. To increase compliance, participants were
informed that accurate logging is crucial for them to receive an
accurate analysis of their PPGRs to food (ultimately provided
to each of them). During the connection week, participants
were asked to follow their normal daily routine and dietary habits,
except for the first meal of every day, which we provided as one
of four different types of standardized meals, each consisting of
50 g of available carbohydrates. This resulted in a total of 46,898
real-life meals with close-to or full nutritional values (median of 54
(D) Example of continuous glucose monitoring (CGM) for one participant during an entire week. Colored area within zoom-in shows the incremental area under the
glucose curve (iAUC) which we use to quantify the meal’s PPGR.
(E) Major food components consumed by energy intake.
(F) Distribution of meals (dots) by macronutrient content. Inset shows histogram of meals per macronutrient.
(G) Bray-Curtis based PCoA of metagenome-based bacterial abundances of stool samples in our cohort and in the U.S. HMP and European MetaHIT cohorts.
Inset shows PCoA when samples from other HMP body sites are added. See also Figure S2.
Table 1. Cohorts Description
Main Cohort
Validation
Cohort
KS
p Value
Number of participants (n) 800 100
Sex (% female) 60% 60% 1
Age (y) Mean ± SD 43.3 ± 13.1 42.4 ± 12.6 0.972
BMI (kg/m^2) Mean ± SD 26.4 ± 5.1 26.5 ± 4.8 0.867
BMI R 25 428 (54%) 50 (50%)
BMI R 30 173 (22%) 18 (18%)
HbA1c% Mean ± SD 5.43 ± 0.45 5.50 ± 0.55 0.492
HbA1c% R 5.7 189 (24%) 31 (31%)
HbA1c% R 6.5 23 (3%) 3 (3%)
Total cholesterol
(non-fasting, mg/dl)
Mean ± SD
186.8 ± 37.5 182.7 ± 35.7 0.231
HDL cholesterol
(non-fasting, mg/dl)
Mean ± SD
59.0 ± 17.8 55.0 ± 16.1 0.371
Waist-to-hip
circumference
ratio Mean ± SD
0.83 ± 0.12 0.84 ± 0.07 0.818
KS - Kolmogorov-Smirnov test. See also Figure S1.
Cell 163, 1079–1094, November 19, 2015 ª2015 Elsevier Inc. 1081
A
B
D
E
H
G
F
C
Figure 2. High Interpersonal Variability in the Postprandial Glycemic Response to the Same Meal
(A) PPGRs associate with risk factors. Shown are PPGRs, BMI, HbA1c%, age, and wakeup glucose of all participants, sorted by median standardized meal PPGR
(top, red dots). Correlation of factors with the median PPGRs to standardized meals is shown along with a moving average line.
(legend continued on next page)
1082 Cell 163, 1079–1094, November 19, 2015 ª2015 Elsevier Inc.
meals per participant) and 5,107 standardized meals. The PPGR
of each meal was calculated by combining reported meal time
with CGM data and computing the incremental area under the
glucose curve in the 2 hr after the meal (iAUC; Wolever and Jen-
kins, 1986; Figure 1D).
Prior to CGM connection, a comprehensive profile was
collected from each participant, including: food frequency, life-
style, and medical background questionnaires; anthropometric
measures (e.g., height, hip circumference); a panel of blood
tests; and a single stool sample, used for microbiota profiling
by both 16S rRNA and metagenomic sequencing.
With a total of 10,000,000 Calories logged, our data provide a
global view into the cohort’s dietary habits, showing the fraction
that each food source contributes to the cohort’s overall energy
intake (e.g., dairy, 7%; sweets, 6%; Figure 1E), and macronutrient
intake (Figures S2B–S2D). Analysis of the caloric breakdown of
every meal by macronutrients revealed that protein intake varies
relatively little across meals (80% of meals have 5%–35% pro-
tein), while fat and carbohydrates have a wide and bimodal distri-
bution, whereone of the modes corresponds to fat-free meals and
constitutes 18% of all meals (Figure 1F).
Principal coordinates analysis (PCoA) on the Bray-Curtis
dissimilarity between metagenome-based relative abundances
(RA) revealed a similar degree of variability in the microbiomes
of our cohort and stool samples of the US HMP (Human Micro-
biome Project Consortium, 2012) and European MetaHIT (Niel-
sen et al., 2014) cohorts (Figure 1G). The first two principal coor-
dinates show some distinction between our cohort and the other
cohorts, but when HMP samples from other body sites are
added to the PCoA, stool samples from all three cohorts cluster
together and separate from the rest, indicative of overall similar-
ity in the gut microbiota composition of individuals from these
three distinct geographical regions (Figure 1G).
Postprandial Glycemic Responses Associate with
Multiple Risk Factors
Our data replicate known associations of PPGRs with risk fac-
tors, as the median standardized meal PPGR was significantly
correlated with several known risk factors including BMI (R =
0.24, p < 10
10
), glycated hemoglobin (HbA1c%, R = 0.49, p <
10
10
), wakeup glucose (R = 0.47, p < 10
10
), and age (R =
0.42, p < 10
10
, Figure 2A). These associations are not confined
to extreme values but persist along the entire range of PPGR
values, suggesting that the reduction in levels of risk factors is
continuous across all postprandial values, with lower values
associated with lower levels of risk factors even within the normal
value ranges (Figure 2A).
Utilizing the continuous nature of the CGMs, we also examined
the association between risk factors and the glucose level of
each participant at different percentiles (0–100) with respect
to all glucose measurements from the connection week. These
levels are affected by the PPGRs while also reflecting the general
glycemic control state of the participant. All percentiles signifi-
cantly associated with risk factors (wakeup glucose, BMI,
HbA1c%, and age; Figures S3A–S3D). The percentile at which
the glucose level correlation was highest varied across risk fac-
tors. For example, BMI had the highest correlation with the 40
th
glucose value percentile, whereas for HbA1c% percentile 95 had
the highest correlation (Figures S3A and S3C). These results
suggest that the entire range of glucose levels of an individual
may have clinical relevance, with different percentiles being
more relevant for particular risk factors.
High Interpersonal Variability in the Postprandial
Response to Identical Meals
Next, we examined intra- and interpersonal variability in the
PPGR to the same food. First, we assessed the extent to which
PPGRs to three types of standardized meals that were given
twice to every participant (Figure 1A), are reproducible within
the same person. Indeed, the two replicates showed high agree-
ment (R = 0.77 for glucose, R = 0.77 for bread with butter, R =
0.71 for bread, p < 10
10
in all cases), demonstrating that the
PPGR to identical meals is reproducible within the same person
and that our experimental system reliably measures this repro-
ducibility. However, when comparing the PPGRs of different
people to the same meal, we found high interpersonal variability,
with the PPGRs of every meal type (except fructose) spanning
the entire range of PPGRs measured in our cohort (Figures 2B,
2C, and S3E–S3H). For example, the average PPGR to bread
across 795 people was 44 ± 31 mg/dl*h (mean ± SD), with the
bottom 10% of participants exhibiting an average PPGR below
15 mg/dl*h and the top 10% of participants exhibiting an average
PPGR above 79 mg/dl*h. The large interpersonal differences in
PPGRs are also evident in that the type of meal that induced
the highest PPGR differs across participants and that different
participants might have opposite PPGRs to pairs of different
standardized meals (Figures 2D and 2E).
Interpersonal variability was not merely a result of participants
having high PPGRs to all meals, since high variability was also
observed when the PPGR of each participant was normalized to
his/her own PPGR to glucose (Figures S3I–S3K). For white bread
and fructose, for which such normalized PPGRs were previously
measured, the mode of the PPGR distribution in our cohort had
excellent agreement with published values (Foster-Powell et al.,
(B) Kernel density estimation (KDE) smoothed histogram of the PPGR to four types of standardized meals provided to participants (each with 50 g of available
carbohydrates). Dashed lines represent histogram modes (See also Figure S3).
(C) Example of high interpersonal variability and low intra-personal variability in the PPGR to bread across four participants (two replicates per participant
consumed on two different mornings).
(D) Heatmap of PPGR (average of two replicates) of participants (rows) to three types of standardized meals (columns) consumed in replicates. Clustering is by
each participant’s relative rankings of the three meal types.
(E) Example of two replicates of the PPGR to two standardized meals for two participants exhibiting reproducible yet opposite PPGRs.
(F) Box plot (box, IQR; whiskers, 10–90 percentiles) of the PPGR to different real-life meals along with amount of carbohydrates consumed (green; mean ± std).
(G) Same as (E), for a pair of real-life meals, each containing 20 g of carbohydrates.
(H) Heatmap (subset) of statistically significant associations (p < 0.05, FDR corrected) between participants’ standardized meals PPGRs and participants’ clinical
and microbiome data (See also Figure S4 for the full heatmap).
Cell 163, 1079–1094, November 19, 2015 ª2015 Elsevier Inc. 1083
2002), further validating the accuracy of our data (bread: 65 versus
71; fructose: 15 versus 19, Figures S3I and S3K).
Next, we examined variability in the PPGRs to the multiple
real-life meals reported by our participants. Since real-life meals
vary in amounts and may each contain several different food
components, we only examined meals that contained 20–40 g
of carbohydrates and had a single dominant food component
whose carbohydrate content exceeded 50% of the meal’s car-
bohydrate content. We then ranked the resulting dominant foods
that had at least 20 meal instances by their population-average
PPGR (Figure 2F). For foods with a published glycemic index,
our population-average PPGRs agreed with published values
(R = 0.69, p < 0.0005), further supporting our data (Table S1).
For example, the average PPGR to rice and potatoes was
relatively high, whereas that for ice cream, beer, and dark
chocolate was relatively low, in agreement with published data
(Atkinson et al., 2008; Foster-Powell et al., 2002). Similar to stan-
dardized meals, PPGRs to self-reported meals highly varied
across individuals, with both low and high responders noted
for each type of meal (Figures 2F and 2G).
Postprandial Variability Is Associated with Clinical and
Microbiome Profiles
We found multiple significant associations between the stan-
dardized meal PPGRs of participants and both their clinical and
gut microbiome data (Figures 2H and S4). Notably, the TIIDM
and metabolic syndrome risk factors HbA1c%, BMI, systolic
blood pressure, and alanine aminotransferase (ALT) activity are
all positively associated with PPGRs to all types of standardized
meals, reinforcing the medical relevance of PPGRs. In most stan-
dardized meals, PPGRs also exhibit a positive correlation with
CRP, whose levels rise in response to inflammation ( Figure 2H).
With respect to microbiome features, the phylogenetically
related Proteobacteria and Enterobacteriaceae both exhibit pos-
itive associations with a few of the standardized meals PPGR
(Figure 2H). These taxa have reported associations with poor
glycemic control, and with components of the metabolic syn-
drome including obesity, insulin resistance, and impaired lipid
profile (Xiao et al., 2014). RAs of Actinobacteria are positively
associated with the PPGR to both glucose and bread, which is
intriguing since high levels of this phylum were reported to asso-
ciate with a high-fat, low-fiber diet (Wu et al., 2011).
At the functional level, the KEGG pathways of bacterial
chemotaxis and of flagellar assembly, reported to increase
in mice fed high-fat diets and decrease upon prebiotics adminis-
tration ( Everard et al., 2014), exhibit positive associations
with several standardized meal PPGRs (Figure 2H). The KEGG
pathway of ABC transporters, reported to be positively associ-
ated with TIIDM (Karlsson et al., 2013) and with a Western
high-fat/high-sugar diet (Turnbaugh et al., 2009), also exhibits
positive association with several standardized meal PPGRs (Fig-
ure 2H). Several bacterial secretion systems, including both type
II and type III secretion systems that are instrumental in bacterial
infection and quorum sensing (Sandkvist, 2001) are positively
associated with most standardized meal PPGRs (Figure 2H).
Finally, KEGG modules for transport of the positively charged
amino acids lysine and arginine are associated with high PPGR
to standardized foods, while transport of the negatively charged
amino acid glutamate is associated with low PPGRs to these
foods.
Taken together, these results show that PPGRs vary greatly
across different people and associate with multiple person-spe-
cific clinical and microbiome factors.
Prediction of Personalized Postprandial Glycemic
Responses
We next asked whether clinical and microbiome factors could be
integrated into an algorithm that predicts individualized PPGRs.
To this end, we employed a two-phase approach. In the first,
discovery phase, the algorithm was developed on the main
cohort of 800 participants, and performance was evaluated us-
ing a standard leave-one-out cross validation scheme, whereby
PPGRs of each participant were predicted using a model trained
on the data of all other participants. In the second, validation
phase, an independent cohort of 100 participants was recruited
and profiled, and their PPGRs were predicted using the model
trained only on the main cohort (Figure 3A).
Given non-linear relationships between PPGRs and the
different factors, we devised a model based on gradient boost-
ing regression (
Friedman, 2001).
This model predicts PPGRs us-
ing the sum of thousands of different decision trees. Trees are
inferred sequentially, with each tree trained on the residual of
all previous trees and making a small contribution to the overall
prediction (Figure 3A). The features within each tree are selected
by an inference procedure from a pool of 137 features represent-
ing meal content (e.g., energy, macronutrients, micronutrients);
daily activity (e.g., meals, exercises, sleep times); blood param-
eters (e.g., HbA1c%, HDL cholesterol); CGM-derived features;
questionnaires; and microbiome features (16S rRNA and meta-
genomic RAs, KEGG pathway and module RAs and bacterial
growth dynamics - PTRs; Korem et al., 2015).
As a baseline reference, we used the ‘carbohydrate counting’’
model, as it is the current gold standard for predicting PPGRs
(American Diabetes Association., 2015b; Bao et al., 2011). On
our data, this model that consists of a single explanatory variable
representing the meal’s carbohydrate amount achieves a
modest yet statistically significant correlation with PPGRs (R =
0.38, p < 10
10
, Figure 3B). A model using only meal Caloric
content performs worse (R = 0.33, p < 10
10
, Figure 3C). Our pre-
dictor that integrates the above person-specific factors predicts
the held-out PPGRs of individuals with a significantly higher cor-
relation (R = 0.68, p < 10
10
, Figure 3D). This correlation ap-
proaches the presumed upper bound limit set by the 0.71–0.77
correlation that we observed between the PPGR of the same
person to two replicates of the same standardized meal.
Validation of Personalized Postprandial Glycemic
Response Predictions on an Independent Cohort
We further validated our model on an independent cohort of 100
individuals that we recruited separately. Data from this additional
cohort were not available to us while developing the algorithm.
Participants in this cohort underwent the same profiling as in
the main 800-person cohort. No significant differences were
found between the main and validation cohorts in key parame-
ters, including age, BMI, non-fasting total and HDL cholesterol,
and HbA1c% (Table 1, Figure S1).
1084 Cell 163, 1079–1094, November 19, 2015 ª2015 Elsevier Inc.
Notably, our algorithm, derived solely using the main 800 par-
ticipants cohort, achieved similar performance on the 100 partic-
ipants of the validation cohort (R = 0.68 and R = 0.70 on the main
and validation cohorts, respectively, Figures 3D and 3E). The
reference carbohydrate counting model achieved the same per-
formance as in the main cohort (R = 0.38). This result further sup-
ports the ability of our algorithm to provide personalized PPGR
predictions.
Factors Underlying Personalized Predictions
To gain insight into the contribution of the different features in
the algorithm’s predictions, we examined partial dependence
plots (PDP), commonly used to study functional relations be-
tween features used in predictors such as our gradient boosting
regressor and an outcome (PPGRs in our case; Hastie et al.,
2008). PDPs graphically visualize the marginal effect of a given
feature on prediction outcome after accounting for the average
effect of all other features. While this effect may be indicative of
feature importance, it may also be misleading due to higher-or-
der interactions (Hastie et al., 2008). Nonetheless, PDPs are
commonly used for knowledge discovery in large datasets
such as ours.
Predicted PPGR
(iAUC, mg/dl
.
h)
R=0.70
Validation cohort
prediction
Personal features
Meal features
Main
cohort
800 participants
Validation
cohort
100 participants
Time, nutrients,
prev. exercise
Meal response predictor
Meal
responses
Train predictor
Cross-validation
Leave-one-person-out
020 255 30
x4000
Use predictor to predict meal responses
Boosted decision trees
=
?
Meal response prediction
Predicted Measured
16S MG
Participant
Measured PPGR
(iAUC, mg/dl
.
h)
Meal Carbohydrates (g)
R=0.38
Carbohydrate-only
prediction
Predicted PPGR
(iAUC, mg/dl
.
h)
R=0.68
Main cohort prediction
(cross-validation)
A
BC
DE
Measured PPGR
(iAUC, mg/dl
.
h)
Calories-only
prediction
R=0.33
Meal Calories (g)
BQA
Figure 3. Accurate Predictions of Personal-
ized Postprandial Glycemic Responses
(A) Illustration of our machine-learning scheme for
predicting PPGRs.
(B–E) PPGR predictions. Dots represent predicted
(x axis) and CGM-measured PPGR (y axis) for
meals, for a model based: only on the meal’s car-
bohydrate content (B); only on the meal’s Caloric
content (C); our predictor evaluated in leave-one-
person-out cross validation on the main 800-per-
son cohort (D); and our predictor evaluated on the
independent 100-person validation cohort (E).
Pearson correlation of predicted and measured
PPGRs is indicated.
As expected, the PDP of carbohydrates
(Figure 4A) shows that as the meal carbo-
hydrate content increases, our algorithm
predicts, on average, a higher PPGR. We
term this relation, of higher predicted
PPGR with increasing feature value, as
non-beneficial (with respect to prediction),
and the opposite relation, of lower pre-
dicted PPGR with increasing feature
value, as beneficial (also with respect to
prediction; see PDP legend in Figure 4).
However, since PDPs display the overall
contribution of each feature across the
entire cohort, we asked whether the rela-
tionship between carbohydrate amount
and PPGRs varies across people. To this
end, for each participant we computed
the slope of the linear regression between
the PPGR and carbohydrate amount of all
his/her meals. As expected, this slope was
positive for nearly all (95.1%) participants, reflective of higher
PPGRs in meals richer in carbohydrates. However, the magni-
tude of this slope varies greatly across the cohort, with the
PPGR of some people correlating well with the carbohydrate
content (i.e., carbohydrates ‘sensitive’’) and that of others exhib-
iting equally high PPGRs but little relationship to the amount
of carbohydrates (carbohydrate ‘insensitive’’; Figure 4B). This
result suggests that carbohydrate sensitivity is also person
specific.
The PDP of fat exhibits a beneficial effect for fat since our al-
gorithm predicts, on average, lower PPGR as the meal’s ratio
of fat to carbohydrates (Figure 4C) or total fat content (Fig-
ure S5A) increases, consistent with studies showing that adding
fat to meals may reduce the PPGR (Cunningham and Read,
1989). However, here too, we found that the effect of fat varies
across people. We compared the explanatory power of a linear
regression between each participant’s PPGR and meal carbohy-
drates, with that of regression using both fat and carbohydrates.
We then used the difference in Pearson R between the two
models as a quantitative measure of the added contribution of
fat (Figure 4D). For some participants we observed a reduction
in PPGR with the addition of fat, while for others meal fat content
Cell 163, 1079–1094, November 19, 2015 ª2015 Elsevier Inc. 1085
Eubacterium rectale
PTR (59)
59
430
63
n.d. 1.1 1.2
PTR
223
0.75 0.8 0.85
Coprococcus catus
PTR (53)
158
455
103
n.d. 1.1
1.2
PTR
77
Time from
last sleep (12)
4971
10947
Time (min)
0 400 800 1200
Partial dependence
(a.u.)
0.6
0.3
0
-0.3
M00514
TtrS-TtrR TCS (27)
M00496
NblS-NblR TCS (28)
M00256
Cell div. trans. sys. (30)
Bacteroides
dorei (45)
Alistipes
putredinis (48)
Relative abundance
Alanine aminotransferase (ALT)
Age
BMI
Systolic blood pressure
Non-fasting total cholesterol
Glucose fluctuations (noise, σ/μ)
HbA1c%
Waist-to-hip ratio
Partial dependence
(a.u.)
0.3
0.1
-0.1
Partial dependence
(a.u.)
Partial dependence
(a.u.)
9
6
3
0
-3
-6
Meal
carbohydrates (2)
Weight (g)
PPGR (iAUC, mg/dl
.
h)
Meal carbohydrates (g)
Participant 49
Participant 145
F
G
2
1
0
-1
Meal sodium (5)
Meal
dietary fiber (14)
AB
CD
E
5
390
401
79
105
523
0
632
164
14
334
448
50
424
323
8400
7518
8623
7290
Meal water (21)
Amount (ml)
0 300 600
5588
103306968
8697
Weight (mg) Weight (g)
40
30
20
10
0
Meal carbohydrates (g)
Meal fat (g)
Participant 267 Participant 465
Color Scale
P<0.005 P<0.01 P<0.05 n.s. P<0.005P<0.01P<0.05
Positive associationNegative association
Frequency
Carbohydrate-PPGR slope
R-difference
Frequency
40
20
0
PPGR (iAUC, mg/dl
.
h)
0 40 80 120
0 1000 2000 0 3 6 9 12
24-hour
dietary fiber (25)
7575
8342
Weight (g)
0 20 40
n.d. 10
-6
10
-5
n.d. 10
-6
10
-5
n.d. 10
-3
n.d. 10
-4
10
-3
10
-2
n.d. 10
-4
10
-3
10
-2
Relative abundance Relative abundance Relative abundance Relative abundance
Parabacteroides
distasonis (63)
101
363
332
Relative abundance
n.d. 10
-4
10
-3
10
-2
Phylum
Bacteroidetes (95)
144
436
217
n.d. 10
-0.6
10
-0.2
Relative abundance
Ratio mapped to
gene-set (93)
320
448
24
Ratio mapped
PDP Legend
0.6
0.3
0
-0.3
-0.6
Feature name
(Feature rank)
Feature value
85 - # undetected
372- # with
above-zero
contribution
497 - #with below-
zero contribution
n.d 10
-6
10
-5
Feature
distribution
Negative trend:
Feature is beneficial
Partial dependence
(a.u.)
Meal
fat / carbohydrates (4)
log
2
(fat/carbs)
4
2
0
-2
-4
7611
8303
-2 -1 0 1 2
Above-zero contribution segment
Below-zero contribution segment
Slope > 0
95.1%
R difference:
0.21
R difference:
0.11
87
M00514 TtrS-TtrR TCS (27)
M00513 LuxQN/CqsS-LuxU-LuxO TCS (38)
M00472 NarQ-NarP TCS (41)
Alistipes putredinis (48)
M00664 Nodulation (49)
M00453 QseC-QseB TCS (50)
Sp. in genus Subdoligranulumun (54)
M00035 Methionine degradation (55)
Eubacterium rectale PTR (59)
M00112 Tocopherol biosynthesis (60)
Streptococcus salivarius PTR (65)
M00412 ESCRT-III complex (70)
Eubacterium eligens PTR (79)
M00066 Lactosylceramide biosynth. (80)
Akkermansia muciniphila PTR (82)
Alistipes finegoldii (83)
Bacteroides xylanisolvens (85)
Eubacterium rectale (87)
Akkermansia muciniphila (96)
M00156 Cytochrome c oxidase (98)
Phylum Euryarchaeota (99)
Phylum Cyanobacteria (107)
Figure 4. Factors Underlying the Prediction
of Postprandial Glycemic Responses
(A) Partial dependence plot (PDP) showing the
marginal contribution of the meal’s carbohydrate
content to the predicted PPGR (y axis, arbitrary
units) at each amount of meal carbohydrates
(x axis). Red and green indicate above and below
zero contributions, respectively (number indicate
meals). Boxplots (bottom) indicate the carbohy-
drates content at which different percentiles (10,
25, 50, 75, and 90) of the distribution of all meals
across the cohort are located. See PDP legend.
(B) Histogram of the slope (computed per partici-
pant) of a linear regression between the carbohy-
drate content and the PPGR of all meals. Also
shown is an example of one participant with a low
slope and another with a high slope.
(C) Meal fat/carbohydrate ratio PDP.
(D) Histogram of the difference (computed per
participant) between the Pearson R correlation of
two linear regression models, one between the
PPGR and the meal carbohydrate content and
another when adding fat and carbohydrate*fat
content. Also shown is an example of the carbo-
hydrate and fat content of all meals of one partici-
pant with a relatively low R difference (carb alone
correlates well with PPGR) and another with a
relatively high difference (meals with high fat
content have lower PPGRs). Dot color and size
correspond to the meal’s PPGR.
(E) Additional PDPs.
(F) Microbiome PDPs. The number of participants
in which the microbiome feature was not detected
is indicated (left, n.d.). Boxplots (box, IQR; whiskers
10–90 percentiles) based only on detected values.
(G) Heatmap of statistically significant correlations
(Pearson) between microbiome features termed
beneficial (green) or non-beneficial (red) and
several risk factors and glucose parameters .
See also Figure S5.
1086 Cell 163, 1079–1094, November 19, 2015 ª2015 Elsevier Inc.
did not add much to the explanatory power of the regressor
based only on the meal’s carbohydrates content (Figure 4D).
Interestingly, while dietary fibers in the meal increase the pre-
dicted PPGR, their long-term effect is beneficial as higher
amount of fibers consumed in the 24 hr prior to the meal reduces
the predicted PPGR (Figure 4E). The meal’s sodium content, the
time that passed since last sleeping, and a person’s cholesterol
levels or age all exhibit non-beneficial PDPs, while the PDPs of
the meal’s alcohol and water content display beneficial effects
(Figures 4E and S5A). As expected, the PDP of HbA1c% shows
a non-beneficial effect with increased PPGR at higher HbA1c%
values; intriguingly, higher PPGRs are predicted, on average, for
individuals with HbA1c% above 5.5%, which is very close to
the prediabetes threshold of 5.7% (Figure S5A).
The 72 PDPs of the microbiome-based features used in our
predictor were either beneficial (21 factors), non-beneficial (28),
or non-decisive (23) in that they mostly decreased, increased,
or neither, as a function of the microbiome feature. The resulting
PDPs had several intriguing trends. For example, growth of Eu-
bacterium rectale was mostly beneficial, as in 430 participants
with high inferred growth for E. rectale it associates with a lower
PPGR (Figure 4F). Notably, E. rectale can ferment dietary carbo-
hydrates and fibers to produce metabolites useful to the host
(Duncan et al., 2007), and was associated with improved post-
prandial glycemic and insulinemic responses (Martı
´nez
et al.,
2013), as well as negatively associated with TIIDM (Qin et al.,
2012). RAs of Parabacteroides distasonis were found non-bene-
ficial by our predictor (Figure 4F) and this species was also sug-
gested to have a positive association with obesity (Ridaura et al.,
2013). As another example, the KEGG module of cell-division
transport system (M00256) was non-beneficial, and in the 164
participants with the highest levels for it, it associates with a
higher PPGR (Figure 4F). Bacteroides thetaiotaomicron was
non-beneficial (Figure S5B), and it was associated with obesity
and was suggested to have increased capacity for energy har-
vest (Turnbaugh et al., 2006). In the case of Alistipes putredinis
and the Bacteroidetes phylum, the non-beneficial classification
that our predictor assigns to both of them is inconsistent with
previous studies that found them to be negatively associated
with obesity (Ridaura et al., 2013; Turnbaugh et al., 2006). This
may reflect limitations of the PDP analysis or result from a
more complex relationship between these features, obesity,
and PPGRs.
To assess the clinical relevance of the microbiome-based
PDPs, we computed the correlation between several risk factors
and overall glucose parameters, and the factors with beneficial
and non-beneficial PDPs across the entire 800-person cohort.
We found 20 statistically significant correlations (p < 0.05, FDR
corrected) where microbiome factors termed non-beneficial
correlated with risk factors, and those termed beneficial ex-
hibited an anti-correlation (Figure 4G). For example, higher levels
of the beneficial methionine degradation KEGG module
(M00035) resulted in lower PPGRs in our algorithm, and across
the cohort, this module anti-correlates with systolic blood pres-
sure and with BMI (Figure 4G). Similarly, fluctuations in glucose
levels across the connection week correlates with nitrate respira-
tion two-component regulatory system (M00472) and with lacto-
sylceramide biosynthesis (M00066), which were both termed
non-beneficial. Glucose fluctuations also anti-correlate with
levels of the tetrathionate respiration two-component regulatory
system (M00514) and with RAs of Alistipes finegoldii, both
termed beneficial (Figure 4G). In 14 other cases, factors with
beneficial or non-beneficial PDPs were correlated and anti-
correlated
with risk factors, respectively.
These results suggest that PPGRs are associated with
multiple and diverse factors, including factors unrelated to
meal content.
Personally Tailored Dietary Interventions Improve
Postprandial Responses
Next, we asked whether personally tailored dietary interventions
based on our algorithm could improve PPGRs. We designed a
two-arm blinded randomized controlled trial and recruited 26
new participants. A clinical dietitian met each participant and
compiled 4–6 distinct isocaloric options for each type of meal
(breakfast, lunch, dinner, and up to two intermediate meals), ac-
commodating the participant’s regular diet, eating preferences,
and dietary constraints. Participants then underwent the same
1-week profiling of our main 800-person cohort (except that
they consumed the meals compiled by the dietitian), thus
providing the inputs (microbiome, blood parameters, CGM,
etc.) that our algorithm needs for predicting their PPGRs.
Participants were then blindly assigned to one of two arms
(Figure 5A). In the first, ‘prediction arm,’ we applied our algo-
rithm in a leave-one-out scheme to rank every meal of each
participant in the profiling week (i.e., the PPGR to each predicted
meal was hidden from the predictor). We then used these rank-
ings to design two 1-week diets: (1) a diet composed of the
meals predicted by the algorithm to have low PPGRs (the
‘good’ diet); and (2) a diet composed of the meals with high pre-
dicted PPGRs (the ‘bad’’ diet). Every participant then followed
each of the two diets for a full week, during which they were con-
nected to a CGM and a daily stool sample was collected
(if available). The order of the 2 diet weeks was randomized
for each participant and the identity of the intervention weeks
(i.e., whether they are ‘good’ or ‘bad’’) was kept blinded from
CRAs, dietitians and participants.
The second, ‘expert arm,’ was used as a gold standard for
comparison. Participants in this arm underwent the same
process as the prediction arm except that instead of using our
predictor for selecting their ‘good’’ and ‘bad’’ diets a clinical
dietitian and a researcher experienced in analyzing CGM data
(collectively termed ‘expert’’) selected them based on their
measured PPGRs to all meals during the profiling week. Specif-
ically, meals that according to the expert’s analysis of their CGM
had low and high PPGRs in the profiling week were selected
for the ‘good’ and ‘bad’ diets, respectively. Thus, to the
extent that PPGRs are reproducible within the same person,
this expert-based arm should result in the largest differences
between the ‘good’ and ‘bad’ diets because the selection of
meals in the intervention weeks is based on their CGM data.
Notably, for 10 of the 12 participants of the predictor-based
arm, PPGRs in the ‘bad’’ diet were significantly higher than in
the ‘good’ diet (p < 0.05, Figure 5C). Differences between the
two diets are also evident in fewer glucose spikes and fewer
fluctuations in the raw week-long CGM data (Figure 5B). The
Cell 163, 1079–1094, November 19, 2015 ª2015 Elsevier Inc. 1087
success of the predictor was comparable to that of the expert-
based arm, in which significantly lower PPGRs in the ‘good’
versus the ‘bad’ diet were observed for 8 of its 14 participants
(p < 0.05, 11 of 14 participants with p < 0.1, Figure 5C).
When combining the data across all participants, the ‘good’
diet exhibited significantly lower PPGRs than the ‘bad’ diet
(p < 0.05, Figure 5D) as well as improvement in other measures
of blood glucose metabolism in both study arms, specifically,
Breakfast
Lunch
Snack
Dinner
123456
B
1 B2 B3 B4 B6B5
L1 L2 L3 L4
L6L5
S1 S2 S3 S4 S6S5
D1 D2 D3 D4 D6D5
Day
Dietitian prescribed meals
One week profiling
(26 participants)
16S MG
Personal features
Carbs > 10g?
HbA1c>5.7%?
BMI>25?
Firmicutes>5%?
YN
Y
Y
YN
N
N
020255 30
x4000
Predictor-based
Expert-based
‘Good’ diet
B4
B6
L2
L5
S5
S6
D2
D3
‘Bad’ diet
B1
B2
L3
L6
S1
S2
D1
D5
Choose meals for dietary intervention weeks
‘Good’ diet
B4
B5
L4
L5
S5
S6
D2
D4
‘Bad’ diet
B1
B3
L1
L6
S1
S2
D1
D6
Measure and analyze intervention weeks
Glucose (mg/dl) Glucose (mg/dl)
14 participants
(E1, E2, ..., E14)
165432
Day
L6
Text meal identifier
Color-coded response
‘Bad’ diet week
‘Good’ diet week
‘Bad’ diet week
‘Good’ diet week
(blue - low; yellow - high)
Find best
and worst meals
for each row
Predictor-based arm Expert-based arm
Predictor
Predictor
Predictor
Expert
Expert
Expert
****** ****** * *** * ** ***
n.s. n.s.
***** ******
n.s.
**
†††
n.s.
***
n.s.
*** *** ** ** ****
‘Bad’ diet week
‘Good’ diet week
PPGR (iAUC, mg/dl
.
h)
PPGR (iAUC, mg/dl
.
h)
Glucose fluctuations (noise, σ/μ)
Max PPGR (iAUC, mg/dl
.
h)
E
F
D
C
A
Profiling week measured
PPGR (iAUC, mg/dl
.
h)
Intervention predicted
PPGR (iAUC, mg/dl
.
h)
H
I
R=0.70 R=0.80
Participant P3
Participant E7
Intervention measured
PPGR (iAUC, mg/dl
.
h)
Intervention measured
PPGR (iAUC, mg/dl
.
h)
B
Pizza
Hummus
Potatoes
Chicken liver
Participants
G
Food consumed
during ‘good’
diet week
Food consumed
during ‘bad’
diet week
BQA
Schnitzel
P6
P10
P3
P8
P2
P5
P9
P4
P1
P11
P7
P12
E8
E7
E9
E4
E14
E11
E10
E12
E5
E3
E2
E1
E6
E13
E3
E4
P6
E8
E14
E6
E13
P8
P9
P10
P1
P2
P11
P12
12 participants
(P1, P2, ..., P12)
Figure 5. Personally Tailored Dietary Interventions Improve Postprandial Glycemic Responses
(A) Illustration of the experimental design of our two-arm blinded randomized controlled trial.
(B) Continuous glucose measurements of one participant from the expert arm (top) and another from the predictor arm (bottom) across their ‘good’’ (green) and
‘bad’ (red) diet weeks.
(C) Boxplot of meal PPGRs during the ‘bad’’ (red) and ‘good’ (green) diet weeks for participants in both the predictor (left) and expert (right) arms. Statistical
significance is marked (Mann-Whitney U-test, ***p < 0.001, **p < 0.01, *p < 0.05, y p < 0.1, n.s. not significant).
(D) As in (C), but for a grouping of all meals of all participants in each study arm (p, Wilcoxon signed-rank test).
(E) Boxplot of the blood glucose fluctuations (noise) of participants in both the ‘bad’’ (red) and ‘good’ (green) diet weeks for both study arms. Blood glucose
fluctuations per participant are defined as the ratio between the standard deviation and mean of his/her weeklong blood glucose levels (p, Wilcoxon signed-
rank test).
(F) As in (E), but for the maximum PPGR of each participant.
(G) Subset of dominant food components prescribed in the ‘good’’ (green) diet of some participants and in the ‘bad’ (red) diet of other participants. See also
Figure S6 for the full matrix.
(H) Dot plot between the CGM-measured PPGR of meals during the profiling week (x axis) and the average CGM-measured PPGR of the same meals during the
dietary intervention weeks (y axis). Meals of all participants in both study arms are shown.
(I) As in (H), but when PPGRs in the dietary intervention weeks are predicted by our predictor using only the first profiling week data of each participant.
Boxplots - box, IQR; whiskers 1.5*IQR.
1088 Cell 163, 1079–1094, November 19, 2015 ª2015 Elsevier Inc.
lower fluctuations in glucose levels across the CGM connection
week (p < 0.05, Figure 5E), and a lower maximal PPGR (p < 0.05,
Figure 5F) in the ‘good’ diet.
Both study arms constitute personalized nutritional interven-
tions and thus demonstrate the efficacy of this approach in
lowering PPGRs. However, the predictor-based approach has
broader applicability since it can predict PPGRs to arbitrary un-
seen meals, whereas the ‘expert’’-based approach will always
require CGM measurements of the meals it prescribes.
Post hoc examination of the prescribed diets revealed the
personalized aspect of the diets in both arms in that multiple
dominant food components (as in Figure 2F) prescribed in the
‘good’ diet of some participants were prescribed in the ‘bad’
diet of others ( Figures 5G and S6). This occurs when compo-
nents induced opposite CGM-measured PPGRs across partici-
pants (expert arm) or were predicted to have opposite PPGRs
(predictor arm).
The correlation between the measured PPGR of meals during
the profiling week and the average CGM-measured PPGR of the
same meals during the dietary intervention was 0.70 (Figure 5H),
which is similar to the reproducibility observed for standardized
meals (R = 0.71–0.77). Thus, as in the case of standardized meals,
a meal’s PPGR during the profiling week was not identical to its
PPGR in the dietary intervention week. Notably, using only the first
profiling week data of each participant, our algorithm predicted
the average PPGRs of meals in the dietary intervention weeks
with an even higher correlation (R = 0.80, Figure 5I). Since our pre-
dictor also incorporates context-specific factors (e.g., previous
meal content, time since sleep), this result also suggests that
such factors may be important determinants of PPGRs.
Taken together, these results show the utility of personally
tailored dietary interventions for improving PPGRs in a short-
term intervention period, and the ability of our algorithm to devise
such interventions.
Alterations in Gut Microbiota Following Personally
Tailored Dietary Interventions
Finally, we used the daily microbiome samples collected during
the intervention weeks to ask whether the interventions induced
significant changes in the gut microbiota. Previous studies
showed that even short-term dietary interventions of several
days may significantly alter the gut microbiota (David et al.,
2014; Korem et al., 2015).
We detected changes following the dietary interventions that
were significant relative to a null hypothesis of no change derived
from the first week, in which there was no intervention, across all
participants (Figures 6A and 6B). While many of these significant
changes were person-specific, several taxa changed consis-
tently in most participants (p < 0.05, FDR corrected, Figure 6C
and S7). Moreover, in most cases in which the consistently
changing taxa had reported associations in the literature, the
direction of change in RA following the ‘good’ diet was in agree-
ment with reported beneficial associations. For example, Bifido-
bacterium adolescentis, for which low levels were reported to be
associated with greater weight loss (Santacruz et al., 2009),
generally decrease in RA following the ‘good’ diet and increase
following the ‘bad’ diet (Figure 6C,D). Similarly, TIIDM has been
associated with low levels of Roseburia inulinivorans (Qin et al.,
2012; Figure 6E), Eubacterium eligens (Karlsson et al., 2013),
and Bacteroides vulgatus (Ridaura et al., 2013), and all these
bacteria increase following the ‘good’ diet and decrease
following the ‘bad’ diet (Figure 6C). The Bacteroidetes phylum,
for which low levels associate with obesity and high fasting
glucose (Turnbaugh et al., 2009), increases following the
‘good’ diet and decreases following the ‘bad’ diet (Figure 6
C).
Low
levels of Anaerostipes associate with improved glucose
tolerance and reduced plasma triglyceride levels in mice (Ever-
ard et al., 2011) and indeed these bacteria decrease following
the ‘good’ diet and increase following the ‘bad’ diet (Figure 6C).
Finally, low levels of Alistipes putredinis associate with obesity
(Ridaura et al., 2013) and this bacteria increased following the
‘good’ diet (Figure 6C).
These findings demonstrate that while both baseline micro-
biota composition and personalized dietary intervention vary be-
tween individuals, several consistent microbial changes may be
induced by dietary intervention with a consistent effect on PPGR.
DISCUSSION
In this work we measured 46,898 PPGRs to meals in a popula-
tion-based cohort of 800 participants. We demonstrate that
PPGRs are highly variable across individuals even when they
consume the same standardized meals. We further show that
an algorithm that integrates clinical and microbiome features
can accurately predict personalized PPGRs to complex, real-
life meals even in a second independently collected validation
cohort of 100 participants. Finally, personalized dietary interven-
tions based on this algorithm induced lower PPGRs and were
accompanied by consistent gut microbiota alterations.
Our study focused on PPGRs, as they were shown to be
important in achieving proper glycemic control, and when
disturbed are considered an independent disease risk factor
(American Diabetes Association., 2015a; Gallwitz, 2009). PPGRs
in our study also associated with several risk factors, including
BMI, HbA1c%, and wakeup glucose. In addition to its centrality
in glucose homeostasis, PPGRs serves as a convenient and ac-
curate endpoint, enabling continuous ‘point-of-care’ collection
of dozens of quantitative measurements per person during a
relatively short follow up period. Such continuous assessment
of PPGRs is complementary to other equally important clinical
parameters such as BMI and HbA1c%, for which changes typi-
cally occur over longer timescales and are thus difficult to corre-
late to nutritional responses in real time.
In line with few small-scale studies that previously examined
individual PPGRs (Vega-Lo
´
pez et al., 2007; Vrolix and Mensink,
2010), we demonstrate on 800 individuals that the PPGR of
different people to the same food can greatly vary. The most
compelling evidence for this observation is the controlled setting
of standardized meals, provided to all participants in replicates.
This high interpersonal variability suggests that at least with re-
gard to PPGRs, approaches that grade dietary ingredients as
universally ‘good’ or ‘bad’’ based on their average PPGR in
the population may have limited utility for an individual.
We report several associations between microbiome features
and variability in PPGRs across people. In some cases, such as
for Actinobacteria, Proteobacteria, and Enterobacteriaceae, the
Cell 163, 1079–1094, November 19, 2015 ª2015 Elsevier Inc. 1089
123456 123456
‘Bad’ diet week (day) ‘Good’ diet week (day)
Glucose (mg/dl)
Fold change
(w.r.t days 0-3)
Participant E3
123456 123456
‘Good’ diet week (day) ‘Bad’ diet week (day)
Glucose (mg/dl)
Participant P8
Bifidobacterium (G)
Bifidobacterium pseudocatenulatum (S)
Actinobacteria (C)
Alistipes putredinis (S)
Akkermansia muciniphila (S)
Parabacteroides merdae (S)
Streptococcus thermophilus (S)
Corpobacter fastidiosus (S)
Lactobacillus ruminis (S)
Bifidobacterium (G)
Bifidobacterium pseudocatenulatum (S)
AB
Actinobacteria (Phylum)
Actinobacteria (Class)
Bifidobacteriales (Order)
Coriobacteriales (Order)
Bifidobacteriaceae (Family)
Coriobacteriaceae (Family)
Bifidobacterium (Genus)
Collinsella (Genus)
Anaerostipes (Genus)
Dorea (Genus)
Bifidobacterium adolescentis (Species)
Collinsella aerofaciens (Species)
Anaerostipes hadrus (Species)
Eubacterium hallii (Species)
Dorea longicatena (Species)
Bacteroidetes (Phylum)
Viruses (Phylum)
Proteobacteria (Phylum)
Bacteroidia (Class)
Gammaproteobacteria (Class)
Deltaproteobacteria (Class)
Betaproteobacteria (Class)
Bacteroidales (Order)
Enterobacteriales (Order)
Burkholderiales (Order)
Viruses, noname (Order)
Desulfovibrionales (Order)
Prevotellaceae (Family)
Bacteroidaceae (Family)
Sutterellaceae (Family)
Prevotella (Genus)
Bacteroides (Genus)
Barnesiella (Genus)
Ruminococcus lactaris (Species)
Eubacterium eligens (Species)
Roseburia inulinivorans (Species)
Bacteroides vulgatus (Species)
Bacteroides stercoris (Species)
Alistipes putredinis (Species)
Bacteria decreasing in ‘good’ diet week Bacteria increasing in ‘good’ diet week
Paricipants - ‘good’ diet weekParicipants - ‘bad’ diet week
P9
E14
E6
P2
P8
E4
E12
P1
P10
E9
E2
E8
P6
E11
E5
E3
E7
P9
E14
E6
P2
P8
E4
P4
E12
P1
P10
E9
E2
E8
P6
E11
E5
E3
E1
C
Bifidobacterium adolescentis
Day
Fold change (with respect to days 0-3)Fold change (with respect to days 0-3)
Day
Roseburia inulinivorans
D
E
‘Good’ diet week
‘Bad’ diet week
‘Good’ diet week
‘Bad’ diet week
Fold change (days 4-7 vs. days 0-3)
-0.5 -0.25 0 0.25 0.5
Statistically significant
decrease (P<0.05)
Statistically significant
inecrease (P<0.05)
Fold change
(w.r.t days 0-3)
Figure 6. Dietary Interventions Induce Consistent Alterations to the Gut Microbiota Composition
(A) Top: Continuous glucose measurements of a participant from the expert arm for both the ‘bad’ diet (left) and ‘good’ diet (right) week. Bottom: Fold change
between the relative abundance (RA) of taxa in each day of the ‘bad’ (left) or ‘good’ (right) weeks and days 0–3 of the same week. Shown are only taxa that exhibit
statistically significant changes with respect to a null hypothesis of no change derived from changes in the first profiling week (no intervention) of all participants.
(B) As in (A) for a participant from the predictor arm. See also Figure S7 for changes in all participants.
(C) Heatmap of taxa with opposite trends of change in RA between ‘good’ and ‘bad’ intervention weeks that was consistent across participant and statistically
significant (Mann-Whitney U-test between changes in the ‘good’ and ‘bad’ weeks, p < 0.05, FDR corrected). Left and right column blocks shows bacteria
increasing and decreasing in their RA following the ‘good’ diet, respectively, and conversely for the ‘bad’ diet. Colored entries represent the (log) fold change
between the RA of a taxon (x axis) between days 4–7 and 0–3 within each participant (y axis). Asterisks indicate a statistically signifi cant fold change.
See also Figure S7 for all changes.
(legend continued on next page)
1090 Cell 163, 1079–1094, November 19, 2015 ª2015 Elsevier Inc.
direction of our associations are consistent with previous asso-
ciations reported between these taxa and higher-level pheno-
types such as dietary habits, obesity and overall glycemic
control (Wu et al., 2011; Xiao et al., 2014), raising testable hy-
potheses about how these taxa may mediate these host meta-
bolic effects. However, in most other cases we identify yet un-
known associations with particular biosynthesis pathways or
transport and secretion systems, which may be contributed by
different taxa in different individuals. These correlations thus pro-
vide concrete new pointers for further mechanistic research,
aimed at establishing causal roles for these bacterial taxa and
functional pathways in determining PPGRs.
Our study further attempts to analyze real-life meals that are
consumed in complex food combinations, at different times of
the day, and in varying proximity to previous meals, physical ac-
tivity, and sleep. While clearly of higher translational relevance,
the use of ‘‘real-life’ nutritional input also introduces noise into
the meal composition data. Despite that, our results show that
predictions for such meals can be made informative by inte-
grating data from a large cohort into a carefully structured pre-
dictor. Even better predictions can likely be achieved with further
research.
Our algorithm takes as input a comprehensive clinical and mi-
crobiome profile and employs a data-driven unbiased approach
to infer the major factors that are predictive of PPGRs. Intro-
specting the resulting algorithm shows that its predictions inte-
grate multiple diverse features that are unrelated to the content
of the meal itself. These include contents of previous meals,
time since sleep, proximity to exercise, and several micro-
biome-based factors. With respect to microbiome factors, our
algorithm identifies multiple functional pathways and bacterial
taxa as either beneficial or non-beneficial, such that in partici-
pants with increasing levels for these factors the algorithm pre-
dicts a lower or higher PPGR, respectively. In many such cases,
microbiome factors found to be beneficial with respect to PPGRs
are also negatively associated with risk factors such as HbA1c%
and cholesterol levels.
Dietary interventions based on our predictor showed signifi-
cant improvements in multiple aspects of glucose metabolism,
including lower PPGRs and lower fluctuations in blood glucose
levels within a short 1-week intervention period. It will be inter-
esting to evaluate the utility of such personalized intervention
over prolonged periods of several months and even years. If suc-
cessful, prolonged individualized dietary control of the PPGR
may be useful in controlling, ameliorating, or preventing a set
of disorders associated with chronically impaired glucose con-
trol, including obesity, prediabetes, TIIDM, and non-alcoholic
fatty liver disease (Grundy, 2012). These intriguing possibilities,
and the microbiome changes that accompany them, merit
further studies. Of equal interest and importance, our individual-
ized nutritional study protocols may be applicable to address
other clinically relevant issues involving nutritional modifications,
such as TIIDM and TIDM patient-specific determination of medi-
cation (e.g., insulin and oral hypoglycemics) dosing and timing.
Employing similar individualized prediction of nutritional ef-
fects on disease development and progression may also be
valuable in rationally designing nutritional interventions in a vari-
ety of inflammatory, metabolic, and neoplastic multi-factorial
disorders. More broadly, accurate personalized predictions of
nutritional effects in these scenarios may be of great practical
value, as they will integrate nutritional modifications more exten-
sively into the clinical decision-making scheme.
EXPERIMENTAL PROCEDURES
Human Cohorts
Approved by Tel Aviv Sourasky Medical Center Institutional Review Board
(IRB), approval numbers TLV-0658-12, TLV-0050-13 and TLV-0522-10; Kfar
Shaul Hospital IRB, approval number 0-73; and Weizmann Institute of Science
Bioethics and Embryonic Stem Cell Research oversight committee. Reported
to http://clinicaltrials.gov/, NCT: NCT01892956.
Study Design
Study participants were healthy individuals aged 18–70 able to provide
informed consent and operate a glucometer. Prior to the study, participants
filled medical, lifestyle, and nutritional questionnaires. At connection week
start, anthropometric, blood pressure and heart-rate measurements were
taken by a CRA or a certified nurse, as well as a blood test. Glucose was
measured for 7 days using the iPro2 CGM with Enlite sensors (Medtronic,
MN, USA), independently calibrated with the Contour BGM (Bayer AG, Lever-
kusen, Germany) as required. During that week participants were instructed to
record all daily activities, including standardized and real-life meals, in real-
time using their smartphones; meals were recorded with exact components
and weights. Full inclusion and exclusion criteria are detailed in Supplemental
Experimental Procedures. Questionnaires used can be found in Data S1.
Standardized Meals
Participants were given standardized meals (glucose, bread, bread and butter,
bread and chocolate, and fructose), calculated to have 50 g of available carbo-
hydrates. Participants were instructed to consume these meals immediately
after their night fast, not to modify the meal, and to refrain from eating or per-
forming strenuous physical activity before, and for 2 hr following consumption.
Stool Sample Collection
Participants sampled their stool following detailed printed instructions. Sam-
pling was done using a swab (n = 776) or both a swab and an OMNIgene-
GUT (OMR-200; DNA Genotek) stool collection kit (n = 413, relative abun-
dances (RA) for the same person are highly correlated (R = 0.99 p < 10
10
) be-
tween swabs and OMNIIgene-GUT collection methods). Collected samples
were immediately stored in a home freezer (20
C), and transferred in a pro-
vided cooler to our facilities where it was stored at 80
C(20
C for OMNII-
gene-GUT kits) until DNA extraction. All samples were taken within 3 days of
connection week start.
Genomic DNA Extraction and Filtering
Genomic DNA was purified using PowerMag Soil DNA isolation kit (MoBio)
optimized for Tecan automated platform. For shotgun sequencing, 100 ng of
purified DNA was sheared with a Covaris E220X sonicator. Illumina compatible
libraries were prepared as described (Suez et al., 2014). For 16S rRNA
sequencing, PCR amplification of the V3/4 region using the 515F/806R 16S
(D) For Bifidobacterium adolescentis, which decreased significantly following the ‘good’ diet interven tions (see panel C), sho wn is the average and standard
deviation of the (log) fold change of all participants in each day of the ‘good’ (top) diet week relative to days 0–3 of the ‘good’ week. Same for the ‘bad’ diet week
(bottom) in which B. adolescentis increases significantly (see panel C). Grey lines show fold changes (log) in individual participants.
(E) As in (D), for Roseburia inulinivorans.
Cell 163, 1079–1094, November 19, 2015 ª2015 Elsevier Inc. 1091
rRNA gene primers was performed followed by 500 bp paired-end sequencing
(Illumina MiSeq).
Microbial Analysis
We used USearch8.0 (Edgar, 2013) to obtain RA from 16S rRNA reads. We
filtered metagenomic reads containing Illumina adapters, filtered low quality
reads and trimmed low quality read edges. We detected host DNA by mapping
with GEM (Marco-Sola et al., 2012) to the Human genome with inclusive pa-
rameters, and removed those reads. We obtained RA from metagenomic
sequencing via MetaPhlAn 2 (Truong et al., 2015) with default parameters.
We assigned length-normalized RA of genes, obtained by similar mapping
with GEM to the reference catalog of (Li et al., 2014), to KEGG Orthology
(KO) entries (Kanehisa and Goto, 2000), and these were then normalized to a
sum of 1. We calculated RA of KEGG modules and pathways by summation.
We considered only samples with >10K reads of 16S rRNA, and >10 M meta-
genomic reads (>1.5 M for daily samples in diet intervention cohort).
Associating PPGRs with Risk Factors and Microbiome Profile
We calculated the median PPGR to standardized meals for each participant
who consumed at least four of the standardized meals and correlated it with
clinical parameters (Pearson). We also calculated the mean PPGR of replicates
of each standardized meal (if performed) and correlated (Pearson) these values
with (a) blood tests; (b) anthropometric measurements; (c) 16S rRNA RA at the
species to phylum levels; (d) MetaPhlAn tag-level RA; and (e) RA of KEGG
genes. We capped RA at a minimum of 1e-4 (16S rRNA), 1e-5 (MetaPhlAn)
and 2e-7 (KEGG gene). For 16S rRNA analysis we removed taxa present in
less than 20% of participants. Correlations on RAs were performed in logspace.
Enrichment analysis of higher phylogenetic levels (d) and KEGG pathways
and modules (e) was performed by Mann-Whitney U-test between log(p val-
ue)*sign(R) of above correlations (d, e) of tags or genes contai ned in the higher
order groups and log(p value)*sign(R) of the correlations of the rest of the tags
or gene s.
FDR Correction
FDR was employed at the rate of 0.15, per tested variable (e.g., glucose stan-
dardized PPGR) per association test (e.g., with blood tests) for analyses in Fig-
ure 2G and Figure S4; per phylogenetic level in Figure 6 and Figure S7; and on
the entire association matrix in Fig ure 4G.
Meal Preprocessing
We merged meals logged less than 30 min apart and removed meals logged
within 90 min of other meals. We also removed very small (<15 g and <70 Cal-
ories) meals and meals with very large (>1 kg) components, meals with incom-
plete logging and meals consumed at the first and last 12 hr of the connection
week.
PPGR Predictor
Microbiome derived features were selected according to number of estimators
using them in an additional predictor run on training data. For detailed feature
list see Supplemental Experimental Procedures. We predicted PPGRs using
stochastic gradient boosting regression, such that 80% of the samples and
40% of the features were randomly sampled for each estimator. The depth
of the tree at each estimator was not limited, but leaves were restricted to
have at least 60 instances (meals). We used 4000 estimators with a learning
rate of 0.002.
Microbiome Changes during Dietary Intervention
We determined the significantly changing taxa of each participant by a Z test of
fold-change in RA between the beginning and end of each intervention week
against a null hypothesis of no change and standard deviation calculated
from at least 25-fold changes across the first profiling week (no intervention)
of corresponding taxa from all participants with similar initial RA. We checked
whether a change was consistent across the cohort for each taxa by perform-
ing Mann-Whitney U-test between the Z statistics of the ‘good’’ intervention
weeks and those of the ‘bad’’ intervention weeks across all participants.
A detailed description of methods used in this paper can be found in the
Supplemental Experimental Procedures.
ACCESSION NUMBERS
The accession number for the data reported in this paper is ENA:
PRJEB11532.
SUPPLEMENTAL INFORMATION
Supplemental Information includes Supplemental Experimental Procedures,
seven figures, one table and one dataset and can be found with this article
online at http://dx.doi.org/10.1016/j.cell.2015.11.001.
AUTHOR CONTRIBUTIONS
T.K. and D.Z. conceived the project, designed and conducted all analyses, in-
terpreted the results, wrote the manuscript and are listed in random order. D.R.
conceived and directed the dietary intervention (DI) study and designed and
conducted analyses. T.K., D.Z., and A.W designed protocols and supervised
data collection. N.Z., D.I., Z.H., and E.E. coordinated and supervised clinical
aspects of data collection. T.K., D.Z., N.Z. and D.I. equally contributed to
this work. A.W. conceived the project, developed protocols, directed and per-
formed sample sequencing. M.R. and O.B.-Y. supervised the DI study. O.B.-Y.
conducted analyses and wrote the manuscript. D.L. conducted analyses, in-
terpreted results and advised nutritional decisions. T.A.-S. and M.L.-P. devel-
oped protocols and together with E.M. performed metagenomic extraction
and sequencing. N.Z., J.S., J.A.M., G.Z.-S., L.D., and M.P.-F. developed pro-
tocols and performed 16S sequencing. G.M., N.K. and R.B. coordinated and
designed data collection . Z.H. conceived the project and provided infrastruc-
ture. E.E. and E.S. conceived and directed the project and analyses, designed
data collection protocols, designed and conducted the analyses, interpreted
the results, and wrote the manuscript.
ACKNOWLEDGMENTS
We thank the Segal and Elinav group members for fruitful discussions; Keren
Segal, Yuval Dor, Tali Raveh-Sadka, Michal Levo, and Leeat Keren for fruitful
discussions and critical insights to the manuscript; Guy Raz and Ran Chen
for website development; Shira Zelber-Sagi for discussions; and Noya Horo-
witz for writing and submitting documents for review by IRBs. This research
was supported by the Weizmann Institute of Science. T.K., D.Z., and D.R. are
supported by the Ministry of Science, Technology, and Space, Israel. T.K. is
supported by the Foulkes Foundation. E.E. is supported by Yael and Rami Un-
gar, Israel; Leona M. and Harry B. Helmsley Charitable Trust; the Gurwin Family
Fund for Scientific Research; Crown Endowment Fund for Immunological
Research; estate of Jack Gitlitz; estate of Lydia Hershkovich; the Benoziyo
Endowment Fund for the Advancement of Science; John L. and Vera Schwartz,
Pacific Palisades; Alan Markovitz, Canada; Cynthia Adelson, Canada; estate of
Samuel and Alwyn J. Weber; Mr. and Mrs. Donald L. Schwarz, Sherman Oaks;
grants funded by the European Research Council (ERC); the Israel Science
Foundation (ISF); E.E. is the incumbent of the Rina Gudinski Career Develop-
ment Chair. E.S. is supported by a research grant from Jack N. Halpern, and
Mr. and Mrs. Donald L. Schwarz and by grants from the ERC and the ISF.
Received: October 5, 2015
Revised: October 29, 2015
Accepted: October 30, 2015
Published: November 19, 2015
REFERENCES
American Diabetes Association (2015a). (5) Prevention or delay of type 2 dia-
betes. Diabetes Care 38 (Suppl ), S31–S32.
American Diabetes Association (2015b). (4) Foundations of care: education,
nutrition, physical activity, smoking cessation, psychosocial care, and immu-
nization. Diabetes Care 38
(Suppl ),
S20–S30.
1092 Cell 163, 1079–1094, November 19, 2015 ª2015 Elsevier Inc.
Atkinson, F.S., Foster-Powell, K., and Brand-Miller, J.C. (2008). International
tables of glycemic index and glycemic load values: 2008. Diabetes Care 31,
2281–2283.
Bailey, T.S., Ahmann, A., Brazg, R., Christiansen, M., Garg, S., Watkins, E.,
Welsh, J.B., and Lee, S.W. (2014). Accuracy and acceptability of the 6-day
Enlite continuous subcutaneous glucose sensor. Diabetes Technol. Ther. 16,
277–283.
Bansal, N. (2015). Prediabetes diagnosis and treatment: A review. World J.
Diabetes 6, 296–303.
Bao, J., Gilbertson, H.R., Gray, R., Munns, D., Howard, G., Petocz, P., Cola-
giuri, S., and Brand-Miller, J.C. (2011). Improving the estimation of mealtime
insulin dose in adults with type 1 diabetes: the Normal Insulin Demand for
Dose Adjustment (NIDDA) study. Diabetes Care 34, 2146–2151.
Blaak, E.E., Antoine, J.-M., Benton, D., Bjo
¨
rck, I., Bozzetto, L., Brouns, F., Dia-
mant, M., Dye, L., Hulshof, T., Holst, J.J., et al. (2012). Impact of postprandial
glycaemia on health and prevention of disease. Obes. Rev. 13, 923–984.
Carpenter, D., Dhar, S., Mitchell, L.M., Fu, B., Tyson, J., Shwan, N.A.A., Yang,
F., Thomas, M.G., and Armour, J.A.L. (2015). Obesity, starch digestion and
amylase: association between copy number variants at human salivary
(AMY1) and pancreatic (AMY2) amylase genes. Hum. Mol. Genet. 24, 3472–
3480.
Cavalot, F., Pagliarino, A., Valle, M., Di Martino, L., Bonomo, K., Massucco, P.,
Anfossi, G., and Trovati, M. (2011). Postprandial blood glucose predicts car-
diovascular events and all-cause mortality in type 2 diabetes in a 14-year
follow-up: lessons from the San Luigi Gonzaga Diabetes Study. Diabetes
Care 34, 2237–2243.
Conn, J.W., and Newburgh, L.H. (1936). The glycemic response to isogluco-
genic quantities of protein and carbohydrate. J. Clin. Invest. 15, 665–671.
Cunningham, K.M., and Read, N.W. (1989). The effect of incorporating fat into
different components of a meal on gastric emptying and postprandial blood
glucose and insulin responses. Br. J. Nutr. 61 , 285–290.
David, L.A., Maurice, C.F., Carmody, R.N., Gootenberg, D.B., Button, J.E.,
Wolfe, B.E., Ling, A.V., Devlin, A.S., Varma, Y., Fischbach, M.A., et al.
(2014). Diet rapidly and reproducibly alters the human gut microbiome. Nature
505, 559–563.
Dodd, H., Williams, S., Brown, R., and Venn, B. (2011). Calculating meal glyce-
mic index by using measured and published food values compared with
directly measured meal glycemic index. Am. J. Clin. Nutr. 94, 992–996.
Duncan, S.H., Belenguer, A., Holtrop, G., Johnstone, A.M., Flint, H.J., and Lob-
ley, G.E. (2007). Reduced dietary intake of carbohydrates by obese subjects
results in decreased concentrations of butyrate and butyrate-producing bac-
teria in feces. Appl. Envir on. Microbiol. 73, 1073–1078.
Dunstan, D.W., Kingwell, B.A., Larsen, R., Healy, G.N., Cerin, E., Hamilton,
M.T., Shaw, J.E., Bertovic, D.A., Zimmet, P.Z., Salmon, J., and Owen, N.
(2012). Breaking up prolonge d sitting reduces postprandial glucose and insulin
responses. Diabetes Care 35, 976–983.
Edgar, R.C. (2013). UPARSE: highly accurate OTU sequences from microbial
amplicon reads. Nat. Methods 10, 996–998.
Everard, A., Lazarevic, V., Derrien, M., Girard, M., Muccioli, G.G., Neyrinck,
A.M., Possemiers, S., Van Holle, A., Franc¸ ois, P., de Vos, W.M., et al. (2011).
Responses of gut microbiota and glucose and lipid metabolism to prebiotics
in genetic obese and diet-induced leptin-resistant mice. Diabetes 60, 2775–
2786.
Everard, A., Lazarevic, V., Gaı
¨
a, N., Johansson, M., Sta
˚
hlman, M., Backhed, F.,
Delzenne, N.M., Schrenzel, J., Franc¸ ois, P., and Cani, P.D. (2014). Microbiome
of prebiotic-treated mice reveals novel targets involved in host response
during obesity. ISME J. 8, 2116–2130.
Foster-Powell, K., Holt, S.H.A., and Brand-Miller, J.C. (2002). International
table of glycemic index and glycemic load values: 2002. Am. J. Clin. Nutr.
76, 5–56.
Friedman, J.H. (2001). Greedy Function Approximation: A Gradient Boosting
Machine. Ann. Stat. 29,
1189–1232.
Gallwitz, B. (2009). Implications of postprandial glucose and weight control in
people with type 2 diabetes: understanding and implementing the International
Diabetes Federation guidelines. Diabetes Care 32 (Suppl 2 ), S322–S325.
Gibbs, E.M., Stock, J.L., McCoid, S.C., Stukenbrok, H.A., Pessin, J.E., Steven-
son, R.W., Milici, A.J., and McNeish, J.D. (1995). Glycemic improvement in dia-
betic db/db mice by overexpression of the human insulin-regulatable glucose
transporter (GLUT4). J. Clin. Invest. 95, 1512–1518.
Greenwood, D.C., Threapleton, D.E., Evans, C.E.L., Cleghorn, C.L., Nykjaer,
C., Woodhead, C., and Burley, V.J. (2013). Glycemic index, glycemic load, car-
bohydrates, and type 2 diabetes: systematic review and dose-response meta-
analysis of prospective studies. Diabetes Care 36, 4166–4171.
Grundy, S.M. (2012). Pre-diabetes, metabolic syndrome, and cardiovascular
risk. J. Am. Coll. Cardi ol. 59, 635–643.
Hastie, T., Tibshirani, R., and Friedman, J. (2008). The Elements of Statistical
Learning: Data Mining, Inference and Prediction (Stanford: Springer).
Himsworth, H.P. (1934). Dietetic factors influencing the glucose tolerance and
the activity of insulin. J. Physiol. 81, 29–48.
Human Microbiome Project Consortium (2012). Structure, function and diver-
sity of the healthy human microbiome. Nature 486, 207–214.
Israeli Center for Disease Control (2014). Health 2013 (Jerusalem: Israeli Min-
istry of Health).
Jenkins, D.J., Wolever, T.M., Taylor, R.H., Barker, H., Fielden, H., Baldwin,
J.M., Bowling, A.C., Newman, H.C., Jenkins, A.L., and Goff, D.V. (1981). Gly-
cemic index of foods: a physiological basis for carbohydrate exchange. Am. J.
Clin. Nutr. 34, 362–366.
Kanehisa, M., and Goto, S. (2000). KEGG: kyoto encyclopedia of genes and
genomes. Nucleic Acids Res. 28, 27–30.
Karlsson, F.H., Tremaroli, V., Nookaew, I., Bergstro
¨
m, G., Behre, C.J., Fager-
berg, B., Nielsen, J., and Ba
¨
ckhed, F. (2013). Gut metagenome in European
women with normal, impaired and diabetic glucose control. Nature 498,
99–103.
Korem, T., Zeevi, D., Suez, J., Weinberger, A., Avnit-Sagi, T., Pompan-Lotan,
M., Matot, E., Jona, G., Harmelin, A., Cohen, N., et al. (2 015). Growth dynamics
of gut microbiota in health and disease inferred from single metagenomic sam-
ples. Science 349, 1101–1106.
Kristo, A.S., Matthan, N.R., and Lichtenstein, A.H. (2013). Effect of diets
differing in glycemic index and glycemic load on cardiovascular risk factors:
review of randomized controlled-feeding trials. Nutrients 5, 1071–1080.
Lamkin, D.M., Spitz, D.R., Shahzad, M.M.K., Zimmerman, B., Lenihan, D.J.,
Degeest, K., Lubaroff, D.M., Shinn, E.H., Sood, A.K., and Lutgendorf, S.K.
(2009). Glucose as a prognostic factor in ovarian carcinoma. Cancer 115,
1021–1027.
Le Chatelier, E., Nielsen, T., Qin, J., Prifti, E., Hildebrand, F., Falony, G.,
Almeida, M., Arumugam, M., Batto, J.-M., Kennedy, S., et al.; MetaHIT con-
sortium (2013). Richness of human gut microbiome correlates with metabolic
markers. Nature 500, 541–546.
Li, J., Jia, H., Cai, X., Zhong, H., Feng, Q., Sunagawa, S., Arumugam, M.,
Kultima, J.R. , Prifti, E., Nielsen, T., et al.; MetaHIT Consortium; MetaHIT
Consortium (2014). An integrated catalog of reference genes in the human
gut microbiome. Nat. Biotechnol. 32, 834–841.
Marco-Sola, S., Sammeth, M., Guigo
´
, R., and Ribeca, P. (2012). The GEM
mapper: fast, accurate and versatile alignment by filtration. Nat. Methods 9,
1185–1188.
Martı
´
nez, I., Lattimer, J.M., Hubach, K.L., Case, J.A., Yang, J., Weber, C.G.,
Louk, J.A., Rose, D.J., Kyureghian, G., Peterson, D.A., et al. (2013). Gut micro-
biome composition is linked to whole grain-induced immunological improve-
ments. ISME J. 7, 269–280.
Nathan
, D.M., Davidson, M.B., DeFronzo, R.A., Heine, R.J., Henry, R.R.,
Pratley, R., and Zinman, B.; American Diabetes Association (2007). Impaired
fasting glucose and impaired glucose tolerance: implications for care. Dia-
betes Care 30, 753–759.
Nielsen, H.B., Almeida, M., Juncker, A.S., Rasmussen, S., Li, J., Sunagawa, S.,
Plichta, D.R., Gautier, L., Pedersen, A.G., Le Chatelier, E., et al.; MetaHIT
Cell 163, 1079–1094, November 19, 2015 ª2015 Elsevier Inc. 1093
Consortium; MetaHIT Consortium (2014). Identification and assembly of ge-
nomes and genetic elements in complex metagenomic samples without using
reference genomes. Nat. Biotechnol. 32, 822–828.
Nishida, T., Tsuji, S., Tsujii, M., Arimitsu, S., Haruna, Y., Imano, E., Suzuki, M.,
Kanda, T., Kawano, S., Hiramatsu, N., et al. (2006). Oral glucose tolerance test
predicts prognosis of patients with liver cirrhosis. Am. J. Gastroenterol. 101,
70–75.
Qin, J., Li, Y., Cai, Z., Li, S., Zhu, J., Zhang, F., Liang, S., Zhang, W., Guan, Y.,
Shen, D., et al. (2012). A metagenome-wide association study of gut micro-
biota in type 2 diabetes. Nature 490, 55–60.
Riccardi, G., and Rivellese, A.A. (2000). Dietary treatment of the metabolic syn-
drome–the optimal diet. Br. J. Nutr. 83 (Suppl 1 ), S143–S148.
Ridaura, V.K., Faith, J.J., Rey, F.E., Cheng, J., Duncan, A.E., Kau, A.L., Griffin,
N.W., Lombard, V., Henrissat, B., Bain, J.R., et al. (2013). Gut microbiota from
twins discordant for obesity modulate metabolism in mice. Science 341,
1241214.
Sandkvist, M. (2001). Type II secretion and pathogenesis. Infect. Immun. 69,
3523–3535.
Santacruz, A., Marcos, A., Wa
¨
rnberg, J., Martı
´,
A., Martin-Matillas, M., Cam-
poy, C., Moreno, L.A., Veiga, O., Redondo-Figuero, C., Garagorri, J.M.,
et al.; EVASYON Study Group (2009). Interplay between weight loss and gut
microbiota composition in overweight adolescents. Obesity (Silver Spring)
17, 1906–1915.
Schwingshackl, L., and Hoffmann, G. (2013). Long-term effects of low glyce-
mic index/load vs. high glycemic index/load diets on parameters of obesity
and obesity-associated risks: a systematic review and meta-analysis. Nutr.
Metab. Cardiovasc. Dis. 23, 699–706.
Suez, J., Korem, T., Zeevi, D., Zilberman-Schapira, G., Thaiss, C.A., Maza, O.,
Israeli, D., Zmora, N., Gilad, S., Weinberger, A., et al. (2014). Artificial sweet-
eners induce glucose intolerance by altering the gut microbiota. Nature 514,
181–186.
Truong, D.T., Franzosa, E.A., Tickle, T.L., Scholz, M., Weingart, G., Pasolli, E.,
Tett, A., Huttenhower, C., and Segata, N. (2015). MetaPhlAn2 for enhanced
metagenomic taxonomic profiling. Nat. Methods 12, 902–903.
Turnbaugh, P.J., Ley, R.E., Mahowald, M.A., Magrini, V., Mardis, E.R., and
Gordon, J.I. (2006). An obesity-associated gut microbiome with increased
capacity for energy harvest. Nature 444, 1027–1031.
Turnbaugh, P.J., Hamady, M., Yatsunenko, T., Cantarel, B.L., Duncan, A., Ley,
R.E., Sogin, M.L., Jones, W.J., Roe, B.A., Affourtit, J.P., et al. (2009). A core gut
microbiome in obese and lean twins. Nature 457, 480–484.
Vega-Lo
´
pez, S., Ausman, L.M., Griffith, J.L., and Lichtenstein, A.H. (2007).
Interindividual variability and intra-individual reproducibility of glycemic index
values for commercial white bread. Diabetes Care 30, 1412–1417.
Vrolix, R., and Mensink, R.P. (2010). Variability of the glycemic response to
single food products in healthy subjects. Contemp. Clin. Trials 31, 5–11.
Wolever, T.M., and Jenkins, D.J. (1986). The use of the glycemic index in
predicting the blood glucose response to mixed meals. Am. J. Clin. Nutr. 43,
167–172.
World Health Organization (2008). Global Health Observatory Data Repository
(Disease and Injury Country Estimates).
Wu, G.D., Chen, J., Hoffmann, C., Bittinger, K., Chen, Y.-Y., Keilbaugh, S.A.,
Bewtra, M., Knights, D., Walters, W.A., Knight, R., et al. (2011). Linking long-
term dietary patterns with gut microbial enterotypes. Science 334, 105–108.
Xiao, S., Fei, N., Pang, X., Shen, J., Wang, L., Zhang, B., Zhang, M., Zhang, X.,
Zhang, C., Li, M., et al. (2014). A gut microbiota-targeted dietary inter vention
for amelioration of chronic inflammation underlying metabolic syndrome.
FEMS Microbiol. Ecol. 87, 357–367.
Zhang, X., Shen, D., Fang, Z., Jie, Z., Qiu, X., Zhang, C., Chen, Y., and Ji, L.
(2013). Human gut microbiota changes reveal the progression of glucose intol-
erance.
PLoS ONE 8, e71108.
1094 Cell 163, 1079–1094, November 19, 2015 ª2015 Elsevier Inc.

Discussion

### What is PPG? Postprandial blood glucose (PPG) levels are a measurement of the body's glucose levels after a meal. In healthy individuals, glucose levels begin to rise about 10 minutes after eating. It will take about an hour to reach a peak glucose level (<140 mg/dl), and another 2-3 hours to return to baseline (70-110 mg/dl) in response to insulin secretion. Once blood glucose levels are low, insulin is no longer needed. Pre-diabetic and diabetic individuals have higher fasting glucose levels and higher baseline insulin levels. To return to baseline, insulin must be produced in higher quantities for a longer period of time. This results in a peak that is more gradual than in their healthy counterparts. Cells gradually become less responsive to insulin and blood glucose levels remain high. #### Continuous Glucose Monitors (CGMs) CGMs are commonly used by patients with Type I Diabetes. In recent years, there has been a push to develop an [artificial pancreas device](https://www.fda.gov/MedicalDevices/ProductsandMedicalProcedures/HomeHealthandConsumer/ConsumerProducts/ArtificialPancreas/ucm259548.htm) that monitors blood glucose to administer insulin and/or glucagon as necessary. Researchers are currently investigating ways to use CGMs to manage Type II Diabetes. #### Measuring Glycemic Response Today, glycemic response is predicted through one of the following nutritional metrics: * __Glycemic Index (GI)__: Ranks foods from 0-100 based on how they affect blood glucose levels. * __Glycemic Load (GL)__: Combines the quantity and quality of the carbohydrates to predict the effect. Calculated as [GI * Carbohydrates (g)]/100. Though widely accepted, GI and GL are not very effective metrics in predicting glycemic response. There are a number of other biological factors that play into metabolism, so different people likely have different responses to the same food. ### Measurements of Blood Glucose Blood glucose is most commonly measured using these metrics: #### HbA1c Level Glucose binds to the hemoglobin found in your red blood cells to create glycated hemoglobin. Because red blood cells have a life span of 2-3 months, measuring the amount of glycated hemoglobin in your blood is an indicator of a person's blood glucose over the past few months. #### Wakeup Glucose Level Also known as fasting plasma glucose (FPG). This is your blood glucose level after 8+ hours of fasting, so it's used as your baseline measurement. #### Oral Glucose Tolerance Test (OOGT) This test monitors an individual's blood glucose for three hours after they have a sugary drink. This measures a person's sensitivity to glucose and their glycemic response. #### NIH Reference Ranges |Diagnosis | HbA1c (%) | FPG (mg/dL) | OGTT (mg/dL)| | |-------------|:----------:|:----------------:|:---------------:|---| |Normal| < 5.7 | < 100| < 139 | | | Prediabetes | 5.7 - 6.4| 100 - 125| 140 - 199 | | | Diabetes | > 6.5| > 126 | > 200 | | #### The Glycemic Response The glycemic response is the body's reaction to consumption of carbohydrates. Monitoring one's glycemic response is an important factor in managing diabetes mellitus. Today, nutritional metrics use carbohydrate content and composition to predict glycemic response. However, an individual's response is dependent on a slew of biological factors, so an individualized approach may be more appropriate. This paper proposes the application of machine learning to personalized nutrition, suggesting that their algorithm served as a more accurate indicator of glycemic response than the current nutritional paradigm. ### Background on Diabetes Diabetes is an umbrella term referring to a group of conditions that are characterized by excessive thirst and urination. The less common *diabetes insipidus* is related to insufficient fluid regulation, while *diabetes mellitus* consists of high blood sugar. These are aptly named after the Latin *insipidus* and *mel*, translating to "tasteless" and "honey," respectively. Diabetes mellitus is further defined as Type I or Type II based on the cause of high glucose: * __Type I__ (TIDM): An autoimmune condition in which one's body attacks the insulin-producing beta cells of the pancreas. Patients with TIDM cannot produce insulin, so they use insulin injections after meals to facilitate glucose uptake. TIDM is often diagnosed at a young age. * __Type II__ (TIIDM): The pancreas outputs high volumes of insulin in response to high glucose levels, and after many years this can lead to *insulin resistance*. Fat, liver, and muscle cells no longer respond to insulin and uptake glucose as they should. TIIDM is treated with lifestyle changes and in some cases, medication. As insulin resistance develops over long periods of time, TIIDM is also termed "adult onset" diabetes. If you're thinking "insu-whaaaat?" here's a graphic of how glucose regulation works in your body: ![glucose regulation](https://i.pinimg.com/originals/e4/28/e9/e428e91483ab1168b6f67e8bd4cb6a09.jpg "Glucose Regulation") #### The Kyoto Encyclopedia of Genes and Genomes (KEGG) Minoru Kanehisa of Kyoto University started KEGG as an online resource for researchers to share cell and molecular findings from large genetic studies. These findings are useful in making higher level discoveries. KEGG includes 18 databases for a variety of systems and organisms. The KEGG Pathway database are a collection of hand-drawn pathway maps of molecular behavior in a variety of areas. You can learn more about KEGG [here](http://www.kegg.jp/kegg/kegg1a.html). #### (Some) Microbes are your Friends! Humans have complex mutualistic relationship with the estimated 100 trillion microbes that reside in their gut. Also known as the *gut microbiota*, these fellows are involved in a variety of vital physiological processes. Although the first "settlers" are transferred from mother to child during birth, the gut microbiota are quite dynamic. Diet, exercise, medication, and other factors can affect the diversity of one's gut. Imbalances can disrupt biological processes and, in extreme cases, cause disease. For a more information about the gut microbiota, check out [this article](https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5433529/).