Dutta et al.

BMC Systems Biology 2019, 13(Suppl 2):36

https://doi.org/10.1186/s12918-019-0692-0

RESEARCH Open Access

Boolean network modeling of β-cell

apoptosis and insulin resistance in type 2
diabetes mellitus
Pritha Dutta1 , Lichun Ma2 , Yusuf Ali3 , Peter M.A. Sloot4 and Jie Zheng5*
From The 17th Asia Pacific Bioinformatics Conference (APBC 2019)
Wuhan, China. 14-16 January 2019

Abstract
Background: Major alteration in lifestyle of human population has promoted Type 2 diabetes mellitus (T2DM) to the
level of an epidemic. This metabolic disorder is characterized by insulin resistance and pancreatic β-cell dysfunction
and apoptosis, triggered by endoplasmic reticulum (ER) stress, oxidative stress and cytokines. Computational modeling
is necessary to consolidate information from various sources in order to obtain a comprehensive understanding of
the pathogenesis of T2DM and to investigate possible interventions by performing in silico simulations.
Results: In this paper, we propose a Boolean network model integrating the insulin resistance pathway with
pancreatic β-cell apoptosis pathway which are responsible for T2DM. The model has five input signals, i.e. ER stress,
oxidative stress, tumor necrosis factor α (TNFα), Fas ligand (FasL), and interleukin-6 (IL-6). We performed dynamical
simulations using random order asynchronous update and with different combinations of the input signals. From the
results, we observed that the proposed model made predictions that closely resemble the expression levels of genes
in T2DM as reported in the literature.
Conclusion: The proposed model can make predictions about expression levels of genes in T2DM that are in
concordance with literature. Although experimental validation of the model is beyond the scope of this study, the
model can be useful for understanding the aetiology of T2DM and discovery of therapeutic intervention for this
prevalent complex disease. The files of our model and results are available at https://github.com/JieZheng-
ShanghaiTech/boolean-t2dm.
Keywords: Boolean model, Type 2 diabetes mellitus, Insulin resistance, β-cell apoptosis

Background excessive nutrients could lead to hyperglycemia, elevated

Type 2 diabetes mellitus (T2DM) is characterized by free fatty acids (FFA), and inflammation, which severely
insulin resistance at its onset. Persistence of insulin impair β-cell functions, leading to insulin resistance and
resistance leads to pancreatic β-cell dysfunction and in β-cell apoptosis.
extreme cases to β-cell apoptosis [1–3]. Insulin resistance The ER in the β-cells is responsible for the production
increases the load on β-cells to produce more insulin in and secretion of insulin. The increased demand for insulin
order to maintain blood glucose at normal levels. This synthesis in the presence of high glucose and FFA lev-
homeostasis is maintained as long as β-cells can meet els triggers the accumulation of misfolded proteins in the
the increased insulin demand. However, persistence of ER, causing ER stress and the consequent activation of the
unfolded protein response (UPR). UPR initially attempts
*Correspondence: [email protected] to mitigate ER stress by degrading misfolded proteins and
5
School of Information Science and Technology, ShanghaiTech University, preventing their further accumulation. However, when ER
Shanghai, China
Full list of author information is available at the end of the article
stress is not mitigated, UPR activates the apoptosis signals
[4–6]. 78 kDa glucose regulated protein (GRP78) serves
© The Author(s). 2019 Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0
International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and
reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the
Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver
(http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.
Dutta et al. BMC Systems Biology 2019, 13(Suppl 2):36 Page 2 of 12

as a sensor of protein misfolding [7]. Under non-stressed ODE models require detailed kinetic knowledge and
conditions, GRP78 binds to three UPR initiator pro- time-series data for accurate parameter estimation. How-
teins, i.e. inositol requiring 1 (IRE1), PKR-like ER kinase ever, the size of our proposed network is relatively big
(PERK), and activating transcription factor 6 (ATF6), and (consisting of 72 nodes) and hence obtaining time-series
maintains them in the inactive state [8]. Under stressed expression data for all the genes would be expensive as
conditions, GRP78 dissociates from these three proteins, well as time-consuming. Also, estimating the parameters
causing their activation and initiation of UPR. of the ODE model with the time-series expression data
When ER stress can be resolved, the UPR assists β of only a small subset of genes would result in erroneous
cells in their survival. However, when ER stress cannot be parameter values. Furthermore, in a Boolean network
resolved the UPR activates the pro-apoptotic signals [9].
Hyperglycemia causes oxidative stress through the gener- Table 1 The gene interactions incorporated into the model with
ation of reactive oxygen species (ROS) [10]. In the absence reference to the existing literature
of an appropriate antioxidant response, the system expe- Gene interations Reference
riences redox imbalance, leading to the activation of
IRE1 ↑ → XBP1 ↑ → β-cell dysfunction [26]
oxidative stress-sensitive signaling pathways. Cytokines,
(IRE1 + TRAF2 + ASK1) ↑ → JNK ↑ → BCL2 [28–30]
including FasL, TNFα, and IL-6, play important roles in (anti-apoptotic gene)↓
the induction of β-cell apoptosis [11–15] as well as insulin
BCL2 ↓ → (BAX + BAK) (pro-apoptotic) ↑ [50, 51]
resistance [16, 17]. Caspases serve as the final mediators
of apoptosis. The upstream apoptosis initiator caspases PERK ↑ → EIF2S1 ↓ → ATF4 ↑ → CHOP (pro- [27]
apoptotic) ↑
8 and 9 are activated on receiving death signal from the
ATF6 ↑ → CHOP (pro-apoptotic) ↑ → BCL2 [51, 52]
death-inducing signaling complex (DISC) and apopto-
(anti-apoptotic gene)↓
some respectively, which in turn activate the downstream
Oxidative stress ↑ → ASK1 ↑, JNK ↑, p38 ↑ [31–33]
apoptosis effector caspases 3, 6 and 7, which ultimately
execute apoptosis [18]. p38 ↑ → CHOP (pro-apoptotic) ↑ [34]
Computational modeling is necessary to consolidate FasL ↑ → (FasR + FADD + pro-caspase-8) ↑ [53]
→ caspase-8 ↑ → caspase-3 ↑ → apoptosis
information from various sources, such as listed above,
in order to obtain a comprehensive understanding of the TNFα ↑ → (TNFR1 + TRADD) ↑ → RIPK1 ↑, [54]
FADD ↑, TRAF2 ↑
pathogenesis of T2DM and investigate possible interven-
tions by performing in silico simulations. A few dynamic FADD ↑ → caspase-8 ↑ [54]
models of insulin resistance in T2DM have been proposed RIPK1 ↑ → RAIDD ↑ → caspase-8 ↑ [54]
recently. For instance, Brannmark et al. [19] proposed an TNFα ↑ → TNFR2 TNFα ↑ → TRAF2 ↑ → [55–57]
ordinary differential equation (ODE) model of insulin sig- ... → JNK ↑, NF-kB ↑
naling in T2DM. Rajan et al. proposed an ODE model (BAX + BAK) (pro-apoptotic) ↑ → [6, 58]
to study the contribution of Forkhead box protein O1 Cytochrome c ↑ → (APAF1 + caspase-9) ↑
→ caspase-3 ↑
(FOXO1) to insulin resistance in T2DM [20]. Another
paper [21] presented an ODE model to simulate the devel- XIAP ↑ → caspase-3 ↓, caspase-7 ↓, caspase- [35, 36]
9↓
opment of insulin resistance by hyperglycemia, FFA, ROS,
DIABLO ↑, HtrA2 ↑ → XIAP ↓ [37]
and inhibition of glucose transporter type 1 (GLUT-1)
and glucose transporter type 4 (GLUT-4). However, there INSR ↑ → IRS ↑ → PI3K ↑ → ...→ AKT ↑ → [59–61]
FOXO1 ↓, GSK3β ↓, GLUT4 ↑
exists no model of β-cell apoptosis occurring in the T2DM
condition. Also, there is no existing work that attempts to GSK3β ↑ → GS ↓ → glycogen synthesis ↓ [42, 43]
integrate the insulin resistance and β-cell apoptosis path- FOXO1 ↑ → PEPCK ↑, G6PC ↑ → glucose [47, 47–49]
synthesis ↑
ways in order to obtain a comprehensive understanding
of the molecular mechanisms underlying T2DM. To dis- (mTORC1 + S6K) ↑ → IRS ↓ [44–46]
cover potential therapeutic interventions for T2DM, it is IKKβ ↑ → TSC1/2 ↓ → mTORC1 ↑ [62]
essential to have a more comprehensive model for the ER stress ↑ → ...→ IRE1 ↑ → ...→ JNK ↑ → [38, 39, 63]
mechanisms causing T2DM pathogenesis. IRS ↓
Therefore, we propose a Boolean network model inte- ER stress ↑ → ...→ IRE1 ↑ → XBP1 ↑ → [64]
grating the insulin resistance pathway and β-cell apoptosis FOXO1 ↓
pathway for the purpose of obtaining deeper insights PERK ↑ → FOXO1 ↑ [65]
into the mechanisms of development and progression ER stress ↑ → ...→ ATF4 ↑ → CHOP ↑ → [40, 41]
of T2DM. The aforementioned existing models are TRB3↑ → AKT ↓
ODE models, whereas we constructed a Boolean net- IL-6 ↑ → JAK ↑ → STAT3 ↑ → SOCS3 ↑ → [66–68]
work model. The reason behind this selection is that IRS ↓
Dutta et al. BMC Systems Biology 2019, 13(Suppl 2):36 Page 3 of 12

model, gene expression is represented by either TRUE (1) Methods

or FALSE (0). By simplifying the gene expression levels In this paper, we propose a Boolean network model of
into binary states, Boolean networks are feasible for sim- β-cell fate in T2DM. The model was constructed by
ulating the behaviour of large regulatory networks in a extracting information from the KEGG pathways [25] and
qualitative way. literature. The gene interactions incorporated into the
In a Boolean network model the state of each gene is model with reference to the existing literature are listed
represented by either 1 (TRUE), indicating the gene is in Table 1. In this model, we integrated the β-cell apopto-
highly expressed, or 0 (FALSE) when the gene is lowly sis pathway with the insulin resistance pathway, as shown
expressed. An edge in a Boolean network can be either in Fig. 1. The apoptosis pathway consists of the signaling
activating or inhibiting [22]. In this paper, we have used pathways triggered by ER stress (UPR pathway), oxida-
random asynchronous Boolean simulation [23, 24], which tive stress, and 3 cytokines, i.e. FasL, TNFα, and IL-6.
updates genes in a random order in each iteration. This The insulin resistance pathways consist of phosphatidyli-
random asynchronous update method is inspired by the nositide 3-kinase (PI3K)-protein kinase B (PKB or AKT)
stochastic nature of gene regulatory networks, where gene (KEGG ID: hsa04151), mammalian target of rapamycin
expression alteration occurs in a random order rather than (mTOR) (KEGG ID: hsa04150), janus kinase (JAK)- signal
simultaneously [24]. transducer and activator of transcription (STAT) (KEGG
Due to the lack of experimental gene expression data, ID: hsa04630), and insulin (KEGG ID: hsa04910) signaling
we validate our simulation results by comparing pre- pathways. T2DM first causes insulin resistance, i.e. insulin
dicted patterns of gene expression levels with experimen- fails to bind to insulin receptors in cells, thereby block-
tal observations reported in the literature. We also analyze ing the uptake of blood glucose by cells. Sustained insulin
the dynamical behaviors of the model by visualizing the resistance finally leads to β-cell failure and apoptosis.
state transition graphs under different combinations of The Boolean update functions, listed in Table 2, for
input signals. Our results show that the simple Boolean the target genes in the model are defined by combining
network model can capture some qualitative trends of the activating input genes using OR functions and inhibiting
genetic circuits regulating the cell fate decision of β-cells, input genes using AND functions. The reason behind this
and shed light on the causes and processes of dysfunc- combination strategy is that a target gene will be expressed
tional insulin metabolism and loss of β-cell homeostasis when at least one of its activating genes is expressed and
that occur in T2DM. all of its inhibiting genes are absent.

Fig. 1 Gene Regulatory Network. Insulin resistance and β-cell apoptosis pathways involved in the pathogenesis of Type 2 diabetes mellitus. The red
nodes denote the five input signals and the purple node represents β-cell apoptosis. A → B indicates activation of gene B by gene A, and A −| B
indicates inhibition of gene B by gene A
Dutta et al. BMC Systems Biology 2019, 13(Suppl 2):36 Page 4 of 12

Table 2 Boolean functions for the Boolean model The proposed Boolean network consists of 72 nodes, of
Node Boolean function Node Boolean function which five are input signals, one node represents Apop-
ER ER OS OS tosis, and the remaining 66 nodes represent genes. We
FasL FasL TNFα TNFα or NFKB employ the random asynchronous Boolean update [23, 24]
IL-6 IL-6 or NFKB method to perform the simulations. The random asyn-
GRP78 ATF6 or XBP1 or ATF4 ATF6 GRP78
chronous Boolean method first generates a random per-
and (not ER) mutation of the nodes at each time step and updates
PERK GRP78 and (not IRE1 BAX or BAK or GRP78 the states of the nodes in the order specified by the
DNAJC3) permutation. This allows us to capture the stochastic
EIF2S1 GADD34 and (not DNAJC3 ATF6 or XBP1 changes in gene expressions that occur in real gene reg-
PERK)
ulatory networks. The random asynchronous Boolean
ATF4 EIF2S1 CHOP ATF6 or ATF4
simulations were performed using the Python code pro-
XBP1 IRE1 GADD34 CHOP
vided in [23] which is available at https://gitlab.com/
TNFR1 TNFα TNFR2 TNFα
stemcellbioengineering/garuda-boolean.
TRAF2 IRE1 or TNFR2 or ASK1 OS or TRAF2 or DAXX
For example, suppose a gene regulatory network con-
TRADD
sists of 3 genes, {g1 , g2 , g3 }. The Boolean update functions
JNK OS or ASK1 or p38 OS or ASK1
GADD45 for the genes are as follows:
BCL2 (not JNK) and (not BID CASP8 and (not BCL2) g1 = g3
CHOP) and (not P53)
and (not BAD) g2 = g1 ∨ g3
BAX JNK or P53 and (not BAK BAX and (not BCL2)
BCL2) g3 = g2
DIABLO BAX or BAK or BID HtrA2 BAX or BAK or BID Suppose an iteration randomly generates a permuta-
FasR FasL TRADD TNFR1 tion of nodes as {3, 1, 2}. Then the asynchronous Boolean
DAXX FasR RIPK1 FasR or TRADD updates will be carried out as follows:
RAIDD RIPK1 FADD FasR or TRADD
g3 (t + 1) = g2 (t)
CASP8 RAIDD or FADD or CASP9 RAIDD or CASP8 or
CASP3 or CASP6 CASP3 or APAF1 or g1 (t + 1) = g3 (t + 1)
CASP12 and (not
XIAP) and (not AKT) g2 (t + 1) = g1 (t + 1) ∨ g3 (t + 1)
CASP3 CASP9 or CASP8 and CASP7 CASP9 or CASP8 or From the above equations, we see that the nodes are
(not XIAP) CASP3 or CASP6 and
(not XIAP) updated in a randomly generated order as specified by the
CASP6 CASP7 or CASP3 permutation, rather than simultaneously.
XIAP (not DIABLO) and (not CytochromeC BAX or BAK or BID After performing the simulations for a fixed number of
HtrA2) (not CASP3) iterations, a directed graph of states is obtained, where
APAF1 CytochromeC or P53 Apoptosis CASP3 or CASP6 or each state is a vector representing the expression lev-
CASP7 els of all genes at a particular time step. The strategy of
INS INS INSR INS strongly connected components (SCCs) is employed on
IRS INSR and (not SOCS3) PI3K IRS or JAK this directed graph to capture the dynamic nature of the
and (not JNK) (not
IKKβ) and (not S6K) states [23]. An SCC of a directed graph is a sub-graph
PIP3 PI3K PDK1 PIP3 that is strongly connected, i.e., each node is reachable
AKT PDK1 or mTORC2 and AS160 AKT from every other node in the sub-graph. An illustration
(not TRB3) of SCC is given in Fig. 2. Each node is a state with the
PKCα PDK1 GLUT4 AKT or AS160 or PKCα expression levels of all the genes in the network (for the
GSK3β not AKT GS not GSK3β example we assume a network with five genes) and there
FOXO1 PERK and (not AKT) PGC1α FOXO1 is a path between each pair of nodes in both directions.
and (not XBP1) Let us consider that an SCC consists of a set of N states
PEPCK FOXO1 G6PC FOXO1 {S1 , S2 , ..., SN }. The probability of state Si being one of the
PPARα PGC1α TRB3 PPARα or CHOP states of the SCC is given by:
TSC1/2 (not AKT) and (not Rheb not TSC1/2
IKKβ) number of occurrences of Si
P(Si ) = N .
mTORC1 Rheb S6K mTORC1
j=1 number of occurrences of Sj
mTORC2 not S6K BAD JNK and (not AKT)
JAK IL-6 and (not SOCS3) STAT3 JAK We calculate the gene expression level of each gene in a
SOCS3 STAT3 IKKβ TRAF2 particular SCC as the sum of probabilities of states where
NFκB not IKBα IKBα not IKKβ the gene is in the ON state. Therefore, the expression level
Dutta et al. BMC Systems Biology 2019, 13(Suppl 2):36 Page 5 of 12

Fig. 2 Strongly Connected Component. An example of a strongly connected component (SCC). Suppose the network consists of five genes. Then
each node is a state which contains the expression levels of the five genes. An arrow from state S1 to state S2 indicates an update step. In an SCC all
states can be reached from every other state

of a gene, gi , with respect to an SCC is determined as Due to the lack of experimental data, we validate our
follows: proposed Boolean network model using relevant literature
 (see Table 1). For each gene gi , we use the same symbol gi
Exp(gi ) = P(Sj ) to represent its binary expression level.
Sj ∈OnSt(gi )

where
1 if gi is reported as expressed in the literature
gi =
OnSt(gi ) = {Sj ∈ SCC | gi (Sj ) = 1}. 0 if gi is reported as not expressed in the literature

It is easy to see that

In our model, we determine the expression level of each

N
gene with respect to a particular SCC. Thus the gene
P(Sj ) = 1
expression levels are in the range [0, 1]. We assume that
j=1
if the expression value of a gene is greater than 0.50, then
We use ER stress, oxidative stress, TNFα, FasL, and IL- the gene is expressed, otherwise, it is not expressed.
6 as input signals. Also, based on the literature, some of For the purpose of validating our proposed model, we
the nodes are assigned specific values (Table 3) and the employ the performance metrics of precision, recall (sen-
rest are set to random values as initial conditions. We per- sitivity), specificity, and F1 score. The simulation result
formed simulations using different combinations of the of our proposed model is verified against the literature as
input signals, as shown in Table 4. We carried out 1000
simulation runs and 1000 Boolean update steps per simu-
lation for each input signal. The results of the simulations Table 4 Different combinations for the input signal nodes
are presented and discussed in the following section.
ER stress Oxidative stress TNFα FasL IL-6
Case 1 True False False False False
Table 3 Initial conditions Case 2 False True False False False
Node Initial value Reason
Case 3 True True False False False
Apoptosis False We set apoptosis to False to see Case 4 False False True False False
whether the input signals can
cause apoptosis Case 5 False False False True False
Caspases 3, 6, 7, 8, 9 False Caspases serve as the final Case 6 False False False False True
mediators of apoptosis. So, we set
Case 7 False False True True True
them to False to see whether the
input signals can activate them Case 8 True True True True True
Dutta et al. BMC Systems Biology 2019, 13(Suppl 2):36 Page 6 of 12

Table 5 Gene expressions of the significant genes in the model for input signal cases 1-5 and 7-8
Node Case 1 Case 2 Case 3 Case 4 Case 5 Case 7 Case 8
A1 A2 A1 A2 A1 A2 A1 A2 A1 A2 A1 A2 A1 A2
Apoptosis 1 1 1 1 1 1 1 1 1 1 1 1 1 1
AKT 0.50 0.50 0.49 0.49 0.49 0.49 0.49 0.49 0.50 0.49 0.49 0.50 0.50 0.49
APAF-1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
ASK1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
ATF4 1 0 1 0 1 0 1 0 1 0 1 0 1 0
ATF6 0 0 0 0 0 0 0 0 0 0 0 0 0 0
BAK 1 1 1 1 1 1 1 1 1 1 1 1 1 1
BAX 1 1 1 1 1 1 1 1 1 1 1 1 1 1
BCL2 0 0 0 0 0 0 0 0 0 0 0 0 0 0
Caspase-3 1 1 1 1 1 1 1 1 1 1 1 1 1 1
Caspase-6 1 1 1 1 1 1 1 1 1 1 1 1 1 1
Caspase-7 1 1 1 1 1 1 1 1 1 1 1 1 1 1
Caspase-8 1 1 1 1 1 1 1 1 1 1 1 1 1 1
Caspase-9 1 1 1 1 1 1 1 1 1 1 1 1 1 1
CHOP 1 0 1 0 1 0 1 0 1 0 1 0 1 0
DIABLO 1 1 1 1 1 1 1 1 1 1 1 1 1 1
EIF2S1 1 0 1 0 1 0 1 0 1 0 1 0 1 0
FADD 1 1 1 1 1 1 1 1 1 1 1 1 1 1
FASR 0 0 0 0 0 0 0 0 0 0 0 0 0 0
FOXO1 0 0 0 0 0 0 0 0 0 0 0 0 0 0
G6PC 0 0 0 0 0 0 0 0 0 0 0 0 0 0
GADD34 1 0 1 0 1 0 1 0 1 0 1 0 1 0
GLUT4 0.65 0.65 0.65 0.65 0.65 0.65 0.65 0.65 0.65 0.65 0.65 0.65 0.65 0.65
GRP78 1 1 1 1 1 1 1 1 1 1 1 1 1 1
GS 0.50 0.50 0.50 0.49 0.49 0.49 0.49 0.49 0.50 0.50 0.50 0.50 0.50 0.49
GSK3β 0.49 0.49 0.50 0.50 0.50 0.50 0.50 0.50 0.49 0.49 0.49 0.49 0.49 0.50
HtrA2 1 1 1 1 1 1 1 1 1 1 1 1 1 1
IKBα 0 0 0 0 0 0 0 0 0 0 0 0 0 0
IKKβ 1 1 1 1 1 1 1 1 1 1 1 1 1 1
INS 1 1 1 1 1 1 1 1 1 1 1 1 1 1
INSR 1 1 1 1 1 1 1 1 1 1 1 1 1 1
IRE1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
IRS 0 0 0 0 0 0 0 0 0 0 0 0 0 0
JAK 0.50 0.49 0.50 0.49 0.49 0.50 0.49 0.49 0.49 0.49 0.49 0.50 0.50 0.50
JNK 1 1 1 1 1 1 1 1 1 1 1 1 1 1
NFKB 1 1 1 1 1 1 1 1 1 1 1 1 1 1
PEPCK 0 0 0 0 0 0 0 0 0 0 0 0 0 0
PERK 0 0 0 0 0 0 0 0 0 0 0 0 0 0
PI3K 0.50 0.49 0.49 0.49 0.50 0.50 0.50 0.50 0.50 0.49 0.49 0.50 0.50 0.49
RAIDD 1 1 1 1 1 1 1 1 1 1 1 1 1 1
RIPK1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
S6K 1 1 1 1 1 1 1 1 1 1 1 1 1 1
SOCS3 0.49 0.49 0.50 0.49 0.50 0.50 0.49 0.49 0.50 0.50 0.49 0.49 0.49 0.50
STAT3 0.50 0.49 0.49 0.49 0.49 0.50 0.49 0.49 0.50 0.50 0.49 0.49 0.50 0.50
TNFR1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
TNFR2 1 1 1 1 1 1 1 1 1 1 1 1 1 1
Dutta et al. BMC Systems Biology 2019, 13(Suppl 2):36 Page 7 of 12

Table 5 Gene expressions of the significant genes in the model for input signal cases 1-5 and 7-8 (Continued)
Node Case 1 Case 2 Case 3 Case 4 Case 5 Case 7 Case 8
A1 A2 A1 A2 A1 A2 A1 A2 A1 A2 A1 A2 A1 A2
TRADD 1 1 1 1 1 1 1 1 1 1 1 1 1 1
TRAF2 1 1 1 1 1 1 1 1 1 1 1 1 1 1
TRB3 1 0 1 0 1 0 1 0 1 0 1 0 1 0
TSC2 0 0 0 0 0 0 0 0 0 0 0 0 0 0
XBP1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
XIAP 0 0 0 0 0 0 0 0 0 0 0 0 0 0
mTORC1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
p38 1 1 1 1 1 1 1 1 1 1 1 1 1 1
Here A1 and A2 denotes SCC1 and SCC2

follows. For each gene gi , ER stress sensor IRE1 and its downstream gene X-box
⎧ protein binding 1 (XBP1) are TRUE in some attrac-
⎪
⎪ True positive, if gi = 1 (simulation result) and gi = 1 (literature)
⎪
⎪ tors, and FALSE in others [26]. Another ER stress sen-
⎪
⎪
⎨True negative, if gi = 0 (simulation result) and gi = 0 (literature)
gi ∈ sor, PERK is observed to be FALSE in all the attrac-
⎪
⎪ if gi = 1 (simulation result) and gi = 0 (literature)
⎪
⎪ False positive, tors. Also, eukaryotic translation initiation factor 2 sub-
⎪
⎪
⎩ unit 1 (EIF2S1), activating transcription factor 4 (ATF4),
False negative, if gi = 0 (simulation result) and gi = 1 (literature)
and C/EBP homologous protein (CHOP) are TRUE in
The four evaluation metrics are calculated using the some attractors and FALSE in the others. PERK phos-
following formulae: phorylates and inactivates EIF2S1, which inhibits protein
True positive synthesis. Phosphorylated EIF2S1 increases the transla-
Precision = tion of ATF4 [8], which in turn activates pro-apoptotic
True positive + False positive
CHOP, causing β-cell dysfunction and death [27]. The
True positive attractors where IRE1, XBP1, EIF2S1, ATF4, and CHOP
Recallorsensitivity =
True positive + False negative have expression levels of 0 may denote the transi-
True negative tion states when these genes are not contributing to
Specificity =
True negative + False positive apoptosis.
2 × precision × recall While associating with TNF-receptor-associated fac-
F1 score = tor 2 (TRAF2) and apoptosis signal-regulating kinase 1
precision + recall
(ASK1), IRE1 activates jun N-terminal kinase (JNK) [28,
Results 29], which in turn inhibits the anti-apoptotic protein B-
Comparison with the literature cell lymphoma 2 (BCL2) [30]. Oxidative stress activates
The expression levels of genes in the SCCs obtained by ASK1 [31, 32], JNK and p38 [33]. Activated p38 phos-
performing simulations with our proposed Boolean model phorylates and elevates the expression of pro-apoptotic
are listed in Tables 5 and 6. Simulations performed using CHOP [34]. From the simulation results, we observe that
input signal cases 1, 2, 3, 4, 5, 7, and 8 (Table 4) result the pro-apoptotic genes, TRAF2, ASK1, JNK, p38, BAX,
in two attractors (SCCs). Apoptosis is ON in both of the and BAK are TRUE and the anti-apoptotic gene BCL2
attractors. Simulations performed using input signal case is FALSE in one attractor, while the reverse states are
6 (Table 4) result in six attractors (SCCs). Apoptosis is ON observed in the other. X-linked inhibitor of apoptosis pro-
in four attractors and OFF in the remaining two attrac- tein (XIAP), which inhibits Caspases 3, 7, and 9 [35, 36],
tors. These observations are consistent with the literature has an expression level of 0, whereas direct IAP-binding
where ER stress, oxidative stress, and cytokines have been protein with low pI (DIABLO) and high temperature
shown to cause apoptosis of β-cells individually as well as requirement protein A2 (HtrA2), which inhibit XIAP [37],
together [4–6]. have expression levels of 1.
From our simulation results, we observe that Caspases JNK phosphorylates and inhibits insulin receptor sub-
3, 6, 7, 8, and 9, which serve as the final mediators of strate (IRS) [38, 39]. IRS gene is FALSE in both of the
apoptosis [18] are TRUE in the attractors, even though attractors. PI3K has an expression level of around 0.50 in
in the initial condition they were set to FALSE. The all the attractors. Tribbles homolog 3 (TRB3) is induced by
Dutta et al. BMC Systems Biology 2019, 13(Suppl 2):36 Page 8 of 12

Table 6 Gene expressions of the significant genes in the model Table 6 Gene expressions of the significant genes in the model
for input signal case 6. Here A1-A6 denotes SCC1-SCC6 for input signal case 6. Here A1-A6 denotes SCC1-SCC6
(Continued)
Node Case 6
Node Case 6
A1 A2 A3 A4 A5 A6
A1 A2 A3 A4 A5 A6
Apoptosis 1 1 1 1 0 0
TRAF2 1 1 0 0 0 0
AKT 0.49 0.49 0.63 0.55 0.65 0.56
TRB3 1 0 0 1 0 1
APAF-1 1 1 0 1 0 0
ASK1 1 1 0 0 0 0 TSC2 0 0 0.37 0.45 0.36 0.44
ATF4 1 0 0 1 0 1 XBP1 1 1 0 0 0 0
ATF6 0 0 0 0 0 0 XIAP 0 0 0 0 1 1
BAK 1 1 0 0 0 0 mTORC1 1 1 0.63 0.55 0.63 0.57
BAX 1 1 0 0 0 0 p38 1 1 0 0 0 0
BCL2 0 0 1 0 1 0
Caspase-3 1 1 1 1 0 0 ER stress through the ATF4-CHOP pathway [40]. Over-
Caspase-6 1 1 1 1 0 0 expression of TRB3 inhibits AKT and decreases glucose
Caspase-7 1 1 1 1 0 0 uptake [41]. TRB3 is TRUE in one attractor and FALSE
Caspase-8 1 1 1 1 0 0 in the other. AKT has an expression level of 0.50 in both
Caspase-9 1 1 1 1 0 0 of the attractors. Thus, from the results, we observe that
CHOP 1 0 0 1 0 1 ER stress inhibits the PI3K-AKT signaling pathway and
DIABLO 1 1 0 1 0 0 promotes insulin resistance.
EIF2S1 1 0 0 1 0 1 Insulin promotes conversion of glucose to glycogen by
FADD 1 1 0 0 0 0 inhibiting glycogen synthase kinase-3β (GSK3β) through
FASR 0 0 0 0 0 0 the PI3K-AKT signaling pathway, which leads to the acti-
FOXO1 0 0 0 0 0 0 vation of glycogen synthase (GS) [42]. From the simulation
G6PC 0 0 0 0 0 0 results, we observe that the expression level of GSK3β,
GADD34 1 0 0 1 0 1
which inhibits glycogen synthesis through inhibition of
GLUT4 0.65 0.65 0.75 0.70 0.78 0.71
GS [42, 43] is approximately 0.49 and that of GS is approx-
imately 0.50. From these simulation results, we can infer
GRP78 1 1 0 1 0 1
that glycogen synthesis is reduced which contributes to
GS 0.49 0.49 0.62 0.54 0.65 0.55
insulin resistance.
GSK3β 0.50 0.50 0.38 0.45 0.36 0.45
In T2DM, the mammalian target of rapamycin com-
HtrA2 1 1 0 1 0 0
plex 1 (mTORC1)/ S6 kinase (S6K) signaling is activated
IKBα 0 0 1 1 1 1
[44] leading to the inhibition of IRS [45, 46]. We observe
IKKβ 1 1 0 0 0 0
from the simulation results that mTORC1 and S6K have
INS 1 1 1 1 1 1 expression levels of 1 thus inhibiting IRS which has an
INSR 1 1 1 1 1 1 expression of 0. These events cause PI3K and AKT to have
IRE1 1 1 0 0 0 0 low expression levels of approximately 0.50, which in turn
IRS 0 0 0.19 0.22 0.18 0.22 reduces glucose uptake through GLUT4 whose expression
JAK 0.49 0.49 0.49 0.49 0.49 0.50 level is around 0.65.
JNK 1 1 0 0 0 0 FOXO1 increases the expression of phosphoenolpyru-
NFKB 1 1 0 0 0 0 vate carboxykinase (PEPCK) and glucose-6-phosphatase
PEPCK 0 0 0 0 0 0 (G6PC) and thus promotes glucose synthesis [47]. Insulin
PERK 0 0 0 0 0 0 inhibits the expression of FOXO1 through the activa-
PI3K 0.49 0.49 0.54 0.55 0.54 0.56 tion of the PI3K/AKT signaling pathway, which in turn
RAIDD 1 1 0 0 0 0 suppresses PEPCK and G6PC, and thereby reduces glu-
RIPK1 1 1 0 0 0 0 cose synthesis [47–49]. From our simulation results, we
S6K 1 1 0.62 0.55 0.62 0.56 observe that FOXO1, PEPCK, and G6PC are FALSE. This
SOCS3 0.49 0.50 0.50 0.49 0.49 0.49 could be due to the fact that PI3K and AKT are not com-
STAT3 0.49 0.49 0.49 0.49 0.49 0.50 pletely inactive, though they may have low expression
TNFR1 1 1 0 0 0 0 levels, and hence is still able to inhibit the expressions of
TNFR2 1 1 0 0 0 0
FOXO1, PEPCK, and G6PC.
TRADD 1 1 0 0 0 0
In Case 6 where only signal IL6 is active, we observe
six attractors (Table 6), of which four indicate apoptosis
Dutta et al. BMC Systems Biology 2019, 13(Suppl 2):36 Page 9 of 12

and two do not. For the attractors where apoptosis is Figure 4 shows the state transition graph of the state
observed, the expression levels of the genes are similar to space generated by simulations conducted using input
those mentioned above for the other input signal cases. signal combination given in case 6 (Table 4). The four
When apoptosis is not observed, i.e. in the two remaining dense red regions represent the four SCCs where apopto-
attractors, the caspases, JNK, BAX, and BAK are FALSE. sis is ON. The two dense blue regions represent the two
In one of these two attractors, BCL2 is FALSE and CHOP SCCs where apoptosis is OFF. Thus from the state transi-
is TRUE. In the other attractor we observe the reverse tion graph, we observe that, in the presence of only IL-6,
expression pattern. Thus, in the presence of only IL-6, apoptosis may or may not be activated.
apoptosis may or may not be activated.
We further assessed the performance of our proposed Comparison with random Boolean networks
Boolean network model by comparing model predictions We also compared our Boolean network model with ran-
of gene expressions against the literature. Considering dom Boolean network models using the 8 input signal
the simulation results obtained using the 8 input signals combinations given in Table 4. For cases 1, 2, 3, 4, 5, 7,
listed in Table 4, the average precision, recall (sensitiv- and 8 we found that the number of attractors obtained
ity), specificity, and F1 score obtained for our model are by simulating the random Boolean networks ranges from
0.9524, 0.8, 0.875, and 0.8696, respectively. We observe 28 to 177, whereas for our Boolean network model the
that the validation scores for our model are not very high, number of attractors is 2. Similarly, for case 6, the number
maybe because our model is sensitive to some missing of attractors obtained by simulating the random Boolean
interactions. networks ranges from 25 to 180, whereas for our Boolean
network model the number of attractors is 6. Thus, from
State transition graphs the results we observe that the random Boolean networks
Figure 3 shows the state transition graph of the state space typically have large numbers of attractors.
generated by simulations conducted using input signal
combination given in case 8 (Table 4). The two dense red Conclusion
regions represent the two SCCs where apoptosis is ON. In this paper, we proposed a Boolean network model
The blue nodes represent states where apoptosis is OFF. of the integrated insulin resistance and β-cell apoptosis
Thus from the state transition graph, we observe that, in pathways. Such a model, which explores the combined
the presence of all input signals, apoptosis is eventually mechanism and interplay between insulin resistance and
activated, even though in the initial condition it is set to β-cell apoptosis in the pathogenesis of T2DM, has not
FALSE. been proposed before. We used the model to simulate the

Fig. 3 State Transition Graph 1. State transition graph obtained by simulating our proposed Boolean network model using input signal condition
given in Case 8 of Table 4. Simulations generate 2 attractors, both having the Apoptosis node activated. Apoptosis is ON in the red coloured states
and OFF in the blue colored states
Dutta et al. BMC Systems Biology 2019, 13(Suppl 2):36 Page 10 of 12

Fig. 4 State Transition Graph 2. State transition graph obtained by simulating our proposed Boolean network model using input signal condition
given in Case 6 of Table 4. Simulations generate 6 attractors. In four of the attractors Apoptosis is ON, denoted by red colour, and in the remaining
two attractors Apoptosis is OFF, denoted by blue colour

dynamics of gene expression induced by different com- play pivotal roles in insulin resistance and/or β-cell apop-
binations of the five input signals, i.e. ER stress, oxidativ tosis, and these genes could be further investigated for
e stress, and cytokines (TNFα, FasL, IL-6), which serve as possible disease interventions.
triggers for insulin resistance and β-cell apoptosis.
Abbreviations
The random order asynchronous update method was AKT (PKB): Protein kinase B; APAF1: Apoptotic protease-activating factor 1 ;
employed to perform the simulations, i.e. all nodes were ASK1: Apoptosis signal-regulating kinase 1; ATF4: Activating transcription
updated in a random order at each update step. We factor 4; ATF6: Activating Transcription Factor 6; BCL2: B-cell lymphoma 2;
CHOP: C/EBP homologous protein; DIABLO: Direct IAP-binding protein with
assessed the performance of our model using the met- low pI; EIF2S1: Eukaryotic translation initiation factor 2 subunit 1; ER:
rics of precision, recall (sensitivity), specificity, and F1 Endoplasmic reticulum; FADD: Fas-associated death domain-containing
score, when validating our model against the literature. protein; FasL: Fas ligand; FasR: Fas receptor; FFA: Free fatty acids; FOXO1:
Forkhead box protein O1; GADD34: Growth arrest and DNA damage-inducible
The precision score obtained is high, but sensitivity, speci- protein; G6PC: Glucose-6-phosphatase; GLUT-1: Glucose transporter type 1;
ficity, and F1 scores are not so. One possible reason may GLUT-4: Glucose transporter type 4; GRP78: 78 kDa glucose regulated protein;
be that some missing interactions affect the predictions of GS: Glycogen synthase; GSK3β: Glycogen synthase kinase-3β; HtrA2: High
temperature requirement protein A2; IL-6: Interleukin-6; INSR: Insulin receptor;
our model. We also compared our Boolean network model IRE1: Inositol Requiring 1; IRS: Insulin receptor substrate; JAK: Janus kinase; JNK:
with random Boolean network models and observed that Jun N-terminal kinase; mTORC1: Mammalian target of rapamycin complex 1;
random Boolean networks typically have large numbers mTORC2: Mammalian target of rapamycin complex 2; ODE: Ordinary
differential equation; PEPCK: Phosphoenolpyruvate carboxykinase; PERK:
of attractors ranging from around 25 to 180, whereas our PKR-like ER kinase; PI3K: Phosphatidylinositide 3-kinase; Rheb: Ras homolog
model shows small numbers of attractors ranging from enriched in brain; RIPK1: Receptor-interacting serine/threonine-protein kinase
2 to 6. 1; ROS: Reactive oxygen species; S6K: S6 kinase; SCC: Strongly connected
component; SOCS3: Suppressor of cytokine signaling 3; STAT3: Signal
As a future step, we can use this model to perform vir- transducer and activator of transcription 3; T2DM: Type 2 Diabetes Mellitus;
tual gene knockout experiments to determine genes that TNFα: Tumor necrosis factor α; TNFR1: Tumor necrosis factor receptor
Dutta et al. BMC Systems Biology 2019, 13(Suppl 2):36 Page 11 of 12

superfamily member 1A; TNFR2: Tumor necrosis factor receptor superfamily 6. Schröder M, Kaufman RJ. The mammalian unfolded protein response.
member 1B; TRADD: TNFR1-associated death domain; TRAF2: Annu Rev Biochem. 2005;74:739–89.
TNF-receptor-associated factor 2; TRB3: Tribbles homolog 3; TSC: Tuberous 7. Bertolotti A, Zhang Y, Hendershot LM, Harding HP, Ron D. Dynamic
sclerosis complex; UPR: Unfolded pro-tein respons; XBP1: X-box protein interaction of bip and er stress transducers in the unfolded-protein
binding 1; XIAP: X-linked inhibitor of apoptosis protein response. Nat Cell Biol. 2000;2(6):326.
8. Ron D, Walter P. Signal integration in the endoplasmic reticulum
Acknowledgements unfolded protein response. Nat Rev Mol Cell Biol. 2007;8(7):519.
We would like to thank Dr. Ket Hing Chong for his valuable discussions. 9. Erguler K, Pieri M, Deltas C. A mathematical model of the unfolded
protein stress response reveals the decision mechanism for recovery,
Funding adaptation and apoptosis. BMC Syst Biol. 2013;7(1):16.
This work was supported by MOE AcRF Tier 1 grant (2015-T1-002-094), Ministry 10. Ihara Y, Toyokuni S, Ichida K, Odaka H, et al. Hyperglycemia causes
of Education, Singapore and the Start-Up Grant of ShanghaiTech University, oxidative stress in pancreatic beta-cells of gk rats, a model of type 2
China. Publication of this article was sponsored by the Start-up Grant of diabetes. Diabetes. 1999;48(4):927.
ShanghaiTech University, China. 11. Donath MY, Shoelson SE. Type 2 diabetes as an inflammatory disease.
Nat Rev Immunol. 2011;11(2):98.
Availability of data and materials 12. Maedler K, Spinas GA, Lehmann R, Sergeev P, Weber M, Fontana A,
Data sharing is not applicable to this article as no datasets were generated or Kaiser N, Donath MY. Glucose induces β-cell apoptosis via upregulation
analysed during the current study. of the fas receptor in human islets. Diabetes. 2001;50(8):1683–90.
13. Spranger J, Kroke A, Möhlig M, Hoffmann K, Bergmann MM, Ristow M,
About this supplement Boeing H, Pfeiffer AF. Inflammatory cytokines and the risk to develop
This article has been published as part of BMC Systems Biology Volume 13 type 2 diabetes: results of the prospective population-based european
Supplement 2, 2019: Selected articles from the 17th Asia Pacific Bioinformatics prospective investigation into cancer and nutrition (epic)-potsdam study.
Conference (APBC 2019): systems biology. The full contents of the supplement Diabetes. 2003;52(3):812–7.
are available online at https://bmcsystbiol.biomedcentral.com/articles/ 14. Eizirik DL, Mandrup-Poulsen T. A choice of death–the signal-transduction
supplements/volume-13-supplement-2. of immune-mediated beta-cell apoptosis. Diabetologia. 2001;44(12):
2115–33.
Authors’ contributions 15. Donath MY. Targeting inflammation in the treatment of type 2 diabetes:
JZ initiated the project and idea, PD and LM constructed the model, and PD time to start. Nat Rev Drug Discov. 2014;13(6):465–76.
carried out the simulations and analysis; YA provided biological input, PS gave
16. Shoelson SE, Lee J, Goldfine AB. Inflammation and insulin resistance. J
suggestions on modeling; PD drafted the manuscript with critical input from
Clin Investig. 2006;116(7):1793–801.
JZ, and other authors provided comments on the manuscript. All authors have
17. Hameed I, Masoodi SR, Mir SA, Nabi M, Ghazanfar K, Ganai BA. Type 2
read and approved the final manuscript.
diabetes mellitus: from a metabolic disorder to an inflammatory
condition. World J Diabetes. 2015;6(4):598.
Ethics approval and consent to participate
18. Hui H, Dotta F, Mario UD, Perfetti R. Role of caspases in the regulation of
Not applicable.
apoptotic pancreatic islet beta-cells death. J Cell Physiol. 2004;200(2):
177–200.
Consent for publication
19. Brannmark C, Nyman E, Fagerholm S, Bergenholm L, Ekstrand E-M,
Not applicable.
Cedersund G, Stralfors P. Insulin signaling in type 2 diabetes-experimental
and modeling analyses reveal mechanisms of insulin resistance in human
Competing interests adipocytes. J Biol Chem. 2013;288(14):9867–80.
The authors declare that they have no competing interests.
20. Rajan MR, Nyman E, Brännmark C, Olofsson CS, Strålfors P. Inhibition of
foxo1 transcription factor in primary human adipocytes mimics the
Publisher’s Note insulin resistant state of type 2 diabetes. Biochem J. 2018;475:1807–20.
Springer Nature remains neutral with regard to jurisdictional claims in
21. Sarkar J, Dwivedi G, Chen Q, Sheu IE, Paich M, Chelini CM, D’Alessandro PM,
published maps and institutional affiliations.
Burns SP. A long-term mechanistic computational model of physiological
factors driving the onset of type 2 diabetes in an individual. PLoS ONE.
Author details 2018;13(2):0192472.
1 Interdisciplinary Graduate School, Nanyang Technogical University,
22. Albert I, Thakar J, Li S, Zhang R, Albert R. Boolean network simulations
Singapore, Republic of Singapore. 2 Biomedical Informatics Lab, School of for life scientists. Source Code Biol Med. 2008;3(1):16.
Computer Science and Engineering, Nanyang Technological University,
23. Yachie-Kinoshita A, Onishi K, Ostblom J, Langley MA, Posfai E, Rossant J,
Singapore, Republic of Singapore. 3 Lee Kong Chian School of Medicine,
Zandstra PW. Modeling signaling-dependent pluripotency with boolean
Nanyang Technogical University, Singapore, Republic of Singapore.
4 Complexity Institute, Nanyang Technogical University, Singapore, Republic of logic to predict cell fate transitions. Mol Syst Biol. 2018;14(1):7952.
24. Albert R, Thakar J. Boolean modeling: a logic-based dynamic approach
Singapore. 5 School of Information Science and Technology, ShanghaiTech
for understanding signaling and regulatory networks and for making
University, Shanghai, China.
useful predictions. Wiley Interdiscip Rev Syst Biol Med. 2014;6(5):353–69.
Published: 5 April 2019 25. Kanehisa M, Goto S. Kegg: kyoto encyclopedia of genes and genomes.
Nucleic Acids Res. 2000;28(1):27–30.
References 26. Allagnat F, Christulia F, Ortis F, Pirot P, Lortz S, Lenzen S, Eizirik DL,
1. Tomita T. Apoptosis in pancreatic β-islet cells in type 2 diabetes. Bosnian Cardozo AK. Sustained production of spliced x-box binding protein 1
J Basic Med Sci. 2016;16(3):162. (xbp1) induces pancreatic beta cell dysfunction and apoptosis.
2. Butler AE, Janson J, Bonner-Weir S, Ritzel R, Rizza RA, Butler PC. β-cell Diabetologia. 2010;53(6):1120–30.
deficit and increased β-cell apoptosis in humans with type 2 diabetes. 27. Oyadomari S, Mori M. Roles of chop/gadd153 in endoplasmic reticulum
Diabetes. 2003;52(1):102–10. stress. Cell Death Differ. 2004;11(4):381.
3. Cnop M, Welsh N, Jonas J-C, Jörns A, Lenzen S, Eizirik DL. Mechanisms 28. Urano F, Wang X, Bertolotti A, Zhang Y, Chung P, Harding HP, Ron D.
of pancreatic β-cell death in type 1 and type 2 diabetes: many Coupling of stress in the er to activation of jnk protein kinases by
differences, few similarities. Diabetes. 2005;54(suppl 2):97–107. transmembrane protein kinase ire1. Science. 2000;287(5453):664–6.
4. Harding HP, Ron D. Endoplasmic reticulum stress and the development 29. Lipson KL, Fonseca SG, Ishigaki S, Nguyen LX, Foss E, Bortell R, Rossini AA,
of diabetes: a review. Diabetes. 2002;51(suppl 3):455–61. Urano F. Regulation of insulin biosynthesis in pancreatic beta cells by an
5. Oyadomari S, Araki E, Mori M. Endoplasmic reticulum stress-mediated endoplasmic reticulum-resident protein kinase ire1. Cell Metab. 2006;4(3):
apoptosis in pancreatic β-cells. Apoptosis. 2002;7(4):335–45. 245–54.
Dutta et al. BMC Systems Biology 2019, 13(Suppl 2):36 Page 12 of 12

30. Xu C, Bailly-Maitre B, Reed JC. Endoplasmic reticulum stress: cell life and 53. Emamaullee JA, Shapiro AJ. Interventional strategies to prevent β-cell
death decisions. J Clin Investig. 2005;115(10):2656–64. apoptosis in islet transplantation. Diabetes. 2006;55(7):1907–14.
31. Ray PD, Huang B-W, Tsuji Y. Reactive oxygen species (ros) homeostasis 54. Baud V, Karin M. Signal transduction by tumor necrosis factor and its
and redox regulation in cellular signaling. Cell Signal. 2012;24(5):981–90. relatives. Trends Cell Biol. 2001;11(9):372–7.
32. Tobiume K, Matsuzawa A, Takahashi T, Nishitoh H, Morita K-i, Takeda K, 55. Akash H, Sajid M, Rehman K, Liaqat A. Tumor necrosis factor-alpha: Role
Minowa O, Miyazono K, Noda T, Ichijo H. Ask1 is required for sustained in development of insulin resistance and pathogenesis of type 2 diabetes
activations of jnk/p38 map kinases and apoptosis. EMBO Rep. 2001;2(3): mellitus. J Cell Biochem. 2018;119:105–10.
222–8. 56. Reinhard C, Shamoon B, Shyamala V, Williams LT. Tumor necrosis factor
33. Evans JL, Goldfine ID, Maddux BA, Grodsky GM. Oxidative stress and α-induced activation of c-jun n-terminal kinase is mediated by traf2.
stress-activated signaling pathways: a unifying hypothesis of type 2 EMBO J. 1997;16(5):1080–92.
diabetes. Endocr Rev. 2002;23(5):599–622. 57. Rothe M, Sarma V, Dixit VM, Goeddel DV. Traf2-mediated activation of
34. Wang X, Ron D. Stress-induced phosphorylation and activation of the nf-kappa b by tnf receptor 2 and cd40. Science. 1995;269(5229):1424–7.
transcription factor chop (gadd153) by p38 map kinase. Science. 58. Li P, Nijhawan D, Budihardjo I, Srinivasula SM, Ahmad M, Alnemri ES,
1996;272(5266):1347–9. Wang X. Cytochrome c and datp-dependent formation of
35. Tenev T, Zachariou A, Wilson R, Ditzel M, Meier P. Iaps are functionally apaf-1/caspase-9 complex initiates an apoptotic protease cascade. Cell.
non-equivalent and regulate effector caspases through distinct 1997;91(4):479–89.
mechanisms. Nat Cell Biol. 2005;7(1):70. 59. Perry RJ, Samuel VT, Petersen KF, Shulman GI. The role of hepatic lipids in
36. Holcik M, Gibson H, Korneluk RG. Xiap: apoptotic brake and promising hepatic insulin resistance and type 2 diabetes. Nature. 2014;510(7503):84.
therapeutic target. Apoptosis. 2001;6(4):253–61. 60. Nikoulina SE, Ciaraldi TP, Mudaliar S, Mohideen P, Carter L, Henry RR.
37. Saelens X, Festjens N, Walle LV, Van Gurp M, van Loo G, Vandenabeele P. Potential role of glycogen synthase kinase-3 in skeletal muscle insulin
Toxic proteins released from mitochondria in cell death. Oncogene. resistance of type 2 diabetes. Diabetes. 2000;49(2):263–71.
2004;23(16):2861. 61. Favaretto F, Milan G, Collin GB, Marshall JD, Stasi F, Maffei P, Vettor R,
38. Aguirre V, Uchida T, Yenush L, Davis R, White MF. The c-jun Naggert JK. Glut4 defects in adipose tissue are early signs of metabolic
nh2-terminal kinase promotes insulin resistance during association with alterations in alms1gt/gt, a mouse model for obesity and insulin
insulin receptor substrate-1 and phosphorylation of ser307. J Biol Chem. resistance. PloS ONE. 2014;9(10):109540.
2000;275(12):9047–54. 62. Lee D-F, Kuo H-P, Chen C-T, Hsu J-M, Chou C-K, Wei Y, Sun H-L, Li L-Y,
39. Hirosumi J, Tuncman G, Chang L, Görgün CZ, Uysal KT, Maeda K, Karin M, Ping B, Huang W-C, et al. Ikkβ suppression of tsc1 links inflammation and
Hotamisligil GS. A central role for jnk in obesity and insulin resistance. tumor angiogenesis via the mtor pathway. Cell. 2007;130(3):440–55.
Nature. 2002;420(6913):333. 63. Özcan U, Cao Q, Yilmaz E, Lee A-H, Iwakoshi NN, Özdelen E, Tuncman G,
40. Ohoka N, Yoshii S, Hattori T, Onozaki K, Hayashi H. Trb3, a novel er Görgün C, Glimcher LH, Hotamisligil GS. Endoplasmic reticulum stress
stress-inducible gene, is induced via atf4–chop pathway and is involved links obesity, insulin action, and type 2 diabetes. Science. 2004;306(5695):
in cell death. EMBO J. 2005;24(6):1243–55. 457–61.
41. Koh H-J, Toyoda T, Didesch MM, Lee M-Y, Sleeman MW, Kulkarni RN, 64. Zhou Y, Lee J, Reno CM, Sun C, Park SW, Chung J, Lee J, Fisher SJ,
Musi N, Hirshman MF, Goodyear LJ. Tribbles 3 mediates endoplasmic White MF, Biddinger SB, et al. Regulation of glucose homeostasis
reticulum stress-induced insulin resistance in skeletal muscle. Nat through a xbp-1–foxo1 interaction. Nat Med. 2011;17(3):356.
Commun. 2013;4:1871. 65. Zhang W, Hietakangas V, Wee S, Lim SC, Gunaratne J, Cohen SM. Er
42. Cohen P, Frame S. The renaissance of gsk3. Nat Rev Mol Cell Biol. stress potentiates insulin resistance through perk-mediated foxo
2001;2(10):769. phosphorylation. Genes Dev. 2013;27(4):441–9.
66. Klover PJ, Clementi AH, Mooney RA. Interleukin-6 depletion selectively
43. Lee J, Kim M-S. The role of gsk3 in glucose homeostasis and the
improves hepatic insulin action in obesity. Endocrinology. 2005;146(8):
development of insulin resistance. Diabetes Res Clin Pract. 2007;77(3):
3417–27.
49–57.
67. Klover PJ, Zimmers TA, Koniaris LG, Mooney RA. Chronic exposure to
44. Khamzina L, Veilleux A, Bergeron S, Marette A. Increased activation of
interleukin-6 causes hepatic insulin resistance in mice. Diabetes.
the mammalian target of rapamycin pathway in liver and skeletal muscle
2003;52(11):2784–9.
of obese rats: possible involvement in obesity-linked insulin resistance.
68. Howard JK, Flier JS. Attenuation of leptin and insulin signaling by socs
Endocrinology. 2005;146(3):1473–81.
proteins. Trends Endocrinol Metab. 2006;17(9):365–71.
45. Harrington LS, Findlay GM, Gray A, Tolkacheva T, Wigfield S, Rebholz H,
Barnett J, Leslie NR, Cheng S, Shepherd PR, et al. The tsc1-2 tumor
suppressor controls insulin–pi3k signaling via regulation of irs proteins. J
Cell Biol. 2004;166(2):213–23.
46. Shah OJ, Wang Z, Hunter T. Inappropriate activation of the
tsc/rheb/mtor/s6k cassette induces irs1/2 depletion, insulin resistance,
and cell survival deficiencies. Curr Biol. 2004;14(18):1650–6.
47. Cheng Z, White MF. Targeting forkhead box o1 from the concept to
metabolic diseases: lessons from mouse models. Antioxid Redox Signal.
2011;14(4):649–61.
48. Kousteni S. Foxo1, the transcriptional chief of staff of energy metabolism.
Bone. 2012;50(2):437–43.
49. Barthel A, Schmoll D. Novel concepts in insulin regulation of hepatic
gluconeogenesis. Am J Physiol Endocrinol Metab. 2003;285(4):685–92.
50. Wei MC, Zong W-X, Cheng EH-Y, Lindsten T, Panoutsakopoulou V, Ross AJ,
Roth KA, MacGregor GR, Thompson CB, Korsmeyer SJ. Proapoptotic bax
and bak: a requisite gateway to mitochondrial dysfunction and death.
Science. 2001;292(5517):727–30.
51. McCullough KD, Martindale JL, Klotz L-O, Aw T-Y, Holbrook NJ. Gadd153
sensitizes cells to endoplasmic reticulum stress by down-regulating bcl2
and perturbing the cellular redox state. Mol Cell Biol. 2001;21(4):1249–59.
52. Marciniak SJ, Yun CY, Oyadomari S, Novoa I, Zhang Y, Jungreis R,
Nagata K, Harding HP, Ron D. Chop induces death by promoting protein
synthesis and oxidation in the stressed endoplasmic reticulum. Gene Dev.
2004;18(24):3066–77.