Get direct links to references, BibTeX extraction and comments on all arXiv papers with: Librarian
#### The Glycemic Response The glycemic response is the body's r...
### What is PPG? Postprandial blood glucose (PPG) levels are a m...
### Background on Diabetes Diabetes is an umbrella term referrin...
#### 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