The document discusses the development and utilization of a phenotype database (dbnp) for conducting meta-analyses of plasma glucose data from various studies, enabling research into health assessments through challenge tests like the oral glucose tolerance test (OGTT). By integrating data from multiple studies, the project aims to analyze metabolic responses and identify subgroups based on clinical characteristics such as diabetes and BMI. Future research will focus on understanding differences in metabolic responses among subgroups and developing metabolomics-based fingerprints for health evaluation.

1. • Combined studies in the Phenotype Database can be used as virtual cohort
• We show that meta-analysis on plasma glucose data of studies in which an OGTT is
performed is possible, although we have to look into some detail of the covariates
• The Phenotype Database contains studies (OGTT with time resolved data for glucose,
insulin and metabolomics) and the relevant subjects to answer our biological research
question (see future work)
NuGO has initiated the development of the Phenotype Database (dbNP). This database is developed together with several other consortia (e.g. Netherlands Metabolomics Centre)
and is currently used within several European projects, such as Food4me, NU-AGE, Bioclaims and Nutritech.
The Phenotype Database (www.dbnp.org) is a web-based application/database that can store any biological study. We used this application to perform an analysis on a
combination of several studies with the objective to test if it is possible to answer new research questions using a ‘virtual cohort’.
Study comparison:
The assessment of the health status of an individual is an important but challenging issue. Nowadays, challenge tests are proposed as a method to assess and quantify health
status. We would like to find mechanistic explanations for differences in clinical subgroups and to develop a metabolomics platform based fingerprint at baseline that represents
important parameters of the challenge test. Currently, there is not one single study available that includes enough subjects from specific clinical subgroups to develop such a
fingerprint or study the biological processes specific for those subgroups. Therefore, we developed a toolbox that facilitates the combined analysis of multiples studies.
A real life example to show the added value
of the Phenotype Database (dbNP)
Bas Kremer, Robert Ernst, Thomas Kelder, Jeroen Wesbeek, Kees van Bochove, Carina
Rubingh, Ruud Boessen, Chris Evelo, Ben van Ommen, Margriet Hendriks, Suzan Wopereis,
Jildau Bouwman and many others of the Phenotype Foundation
Introduction
TNO Quality of Life
Microbiology &
Systems Biology
Zeist, The Netherlands
jildau.bouwman@tno.nl
Studies stored in a structured way
The concept: Challenge tests
Complex data analysis: suitable for virtual cohort?
Conclusions & Discussion
Glucose
3
4
5
6
7
8
Admit Study 3 Study 4
Study
Value(mmol/l)
Gender
Female
Male
Glucose
0.0
0.5
1.0
3 4 5 6 7 8
Value (mmol/l)
Density
Study − Gender
Admit − Female
Admit − Male
Study 3 − Female
Study 3 − Male
Study 4 − Female
Study 4 − Male
Base model: value = Study_1 * time0 + healthy + BMI : Normal
Value Std.Error DF t-value p-value
Study_1
study_2 -0.06 0.42 156 -0.14 0.89
study_3 -0.77 0.57 156 -1.35 0.18
Time: 0
Time: 30 2.14 0.60 470 3.57 0.00 **
Time: 60 1.98 0.54 470 3.67 0.00 **
Time: 120 -0.19 0.44 470 -0.44 0.66
Healthy
Diabetes 3.04 0.60 156 5.09 0.00 **
Prediabetes 1.21 0.21 156 5.71 0.00 **
BMI: Normal
BMI: Morbid obese 1.02 0.49 156 2.07 0.04 *
BMI: Obese 0.78 0.29 156 2.71 0.01 *
BMI: Overweight 0.22 0.20 156 1.12 0.26
BMI: Underweight -0.23 0.35 156 -0.67 0.51
Study1:Time30
Study_2:time30 0.15 0.62 470 0.24 0.81
Study_3:time30 1.28 0.72 470 1.77 0.08 #
Study1:Time60
Study_2:time60 -1.08 0.56 470 -1.94 0.05 #
Study_3:time60 1.64 0.67 470 2.44 0.02 *
Study1:Time120
Study_2:time120 0.00 0.47 470 0.01 0.99
Study_3:time120 2.00 0.60 470 3.34 0.00 **
To test if it is possible to use data from the virtual cohort for complex data-analysis we evaluated the plasma glucose response to
the OGTT and looked at different covariates where differences were to be expected. A linear mixed model was used to model the
plasma glucose response. The model included Study (1, 2 and 3), Time (0, 30, 60 and 120 minutes), Health status (healthy,
diabetic and prediabetic), BMI group (normal, underweight, overweight, obese and morbid obese) and the Study*Time interaction
as fixed factors and a random Subject intercept. Gender and Age were evaluated as well, but were excluded from the final model
since both did not significantly improve the model’s performance. Significant differences are indicated with stars. As expected
significant differences (α = 0.05) were found between:
• Subjects with a normal BMI level and subjects with an obese or morbid obese BMI level
• Healthy subjects and subjects classified with prediabetes and diabetes
• Baseline and time points 30 and 60 minutes
The Study*Time interaction effect was significant for Study_3 and time points 60 and 120 minutes. This means that subjects of
this study differed significantly in their plasma glucose response at these time points from the subjects of the other studies. We
should analyze whether meta-data can explain these differences in Study 3.
The complex data analysis shows that we can find health related differences in response to OGTT (based on the single
parameter glucose) by combining studies. The next step is to include metabolomics data and define subgroups.
*) P < 0.05
**) P < 0.01
#) P <0.1 (trend)
The Phenotype Database stores data in a structured
way. The wizard guides you through all the essential
parts of the study upload. Parts of the study design
can also be uploaded from Excel sheets. This
system stores individual data, which facilitates it’s
use as a virtual cohort.
Templates in the Phenotype
Database structure the data in all
steps. Template administrators
can adjust the information
included in the study by adding a
‘field’ to the templates. A link to
BioPortal makes it possible to use
ontologies as input in the system.
Studies can be made accessible to specific
people/research groups or to the whole world.
Individual subject data can be made available
through this system
Statistical analysis of the combined data can be
performed in R which is connected to the
database via a web service (API).
Currently, 40 (nutritional) studies are included in
the database of which 7 include a challenge
test.
Health can be defined as: ‘the ability to adapt and self-manage in the face of social, physical, and emotional challenges’ (Huber et
al. 2011). To test health in this perspective, challenge tests are widely used in research and health care nowadays. The amplitude
or time to get to normal values are examples of parameters that can be used as health parameter and can be markers for the
functioning of a certain health related process. The Oral Glucose Tolerance Test (OGTT) is the most well-known challenge test.
Baseline: studies can be combined
Density and box plots from 3 different studies from the Phenotype Database (dbNP) that show plasma glucose
concentrations at baseline of the OGTT. Baseline differences in the glucose concentration ranges are observed between the
studies at baseline, but the ranges show an overlap between the studies. Similar trends in gender differences are seen in all
three studies: male subjects in general have a slightly higher glucose level.
As an overlap in the baseline levels between the studies is observed, a further study comparison can be performed.
Description Fasting glucose (mmol/l) Glucose 120 min after OGTT (mmol/l)
Diabetes (T2DM) 7,0 > 11.0
IGT (Impaired glucose tolerance) < 5.6 (normal) 7.8 – 11.0
IFG (impaired fasting glucose) 5.6 – 6.9 < 7.8
IGT&IFG 5.6 – 6.9 7.8 – 11.0
Normal < 5.6 < 7.8
Study 1
Study 3 All studies
Study 2
Analysis of the distribution of (pre)diabetic subgroups in 3 different studies.
We would like to use the virtual cohort to look into the mechanism underlying different clinical (pre)diabetic subgroups by
analyzing metabolic responses to the challenge. In addition, we would like to use these data to try to find metabolomics
platform based fingerprints at baseline that represents important phenotype related parameters of the clinical subgroups.
The 3 studies represent all 5 (pre)diabetic subgroups and therefore can be used to answer our biological question.
Future work
• How different are the subgroups metabolically in the available studies at baseline?
• Are the subgroups different in metabolic response to the OGTT in the available
studies?
• Can we find differences in biological mechanisms in different subgroups?
• Can we predict some of the OGTT health parameters with a metabolomics platform
based fingerprint at baseline?
Defining clinical (pre)diabetic subgroups
Study_1 Study_2 Study_3
Study_1-Female
Study_1-Male
Study_2-Female
Study_2-Male
Study_3-Female
Study_3-Male