THE DYNAMICS OF GLUCOSE-INSULIN ENDOCRINE METABOLIC

REGULATORY SYSTEM
by
Jiaxu Li
A Dissertation Presented in Partial Fulllment
of the Requirements for the Degree
Doctor of Philosophy
ARIZONA STATE UNIVERSITY
December 2004
THE DYNAMICS OF GLUCOSE-INSULIN ENDOCRINE METABOLIC
REGULATORY SYSTEM
by
Jiaxu Li
has been approved
December 2004
APPROVED:
, Chair
Supervisory Committee
ACCEPTED:
Department Chair
Dean, Division of Graduate Studies
ABSTRACT
A model with two time delays is presented for modeling the insulin secretion ul-
tradian oscillations in the glucose-insulin metabolic system. One delay is for the insulin
response time delay (around 6 minutes) to the glucose concentration level increase, and
the other is for the hepatic glucose production time delay (around 36 minutes). The
results of the analysis of this model are in agreement with the experimental observations
and exhibit intrinsic insulin secretion ultradian oscillations. The results show that both
these time delays are necessary for the insulin secretion ultradian oscillation sustain-
ment and only the relative moderate glucose infusion rate and insulin degradation rate
can sustain the oscillations. The numerical simulations demonstrate that the insulin
concentration level peaks after the glucose concentration level. These results also indi-
cate that the hepatic glucose production and its time delay are insignicant in modeling
intravenous glucose tolerance tests (IVGTT).
A generic dynamic IVGTT model and two models for special cases are devel-
oped to simulate the short time (30-120 minutes) dynamics. As expected, such models
frequently produce globally asymptotically stable steady state dynamics. The easy-to-
check conditions, which guarantee the steady state to be stable, are provided.
In the last model, we take the active -cell mass into consideration and study the
eects of the -cells in the glucose-insulin regulatory system. The numerical simulations
show that the insulin concentration peaks after the active -cell mass peaks, which peaks
after the glucose concentration peaks. Other results are also in agreement with reported
data.
iii
In Memory of My Mother
To My Father
To My Wife and Daughters
To My Sisters
iv
ACKNOWLEDGMENTS
I would rst of all like to thank Dr. Yang Kuang for his guidance during my doc-
toral study at Arizona State University. I am forever indebted to his advise, suggestions,
support, encouragement, understanding and, in particular, patience. I would like to
thank Dr. Steven Baer, Dr. Carlos Castillo-Chaves, Dr. Hal Smith and Dr. Horst Thieme
for their interest, carefully reading the manuscript, valuable input and suggestions for
improving this dissertation. It is my great pleasure to work with them and I feel so
lucky and proud that I have such a wonderful supervisory committee, one of the best
in the world. I would also like to thank the external reviewer for the valuable input.
My special thanks go to my master thesis advisor Prof. Xiudong Chen.
I would also like to extend my gratitude to Dr. Bingtuan Li for the various
broad discussions, to Dr. Athena Makroglou for her initiating the collaborate paper
[59] and providing references (for example, [65] and [64]), to Prof. Edoardo Beretta for
his providing the manuscript of [23], to Mr. Clint Mason for his providing reference [9]
and [85], to Ms. Debbie Olson and Ms. Joan Person for their administrative support,
to Dr. Jialong He for his IT support, and to Mr. Rafael Mendez for the proof-reading
of the most of this dissertation.
Last, but not the least, I would like to thank my wife, Dr. Guihua Li, for her
long lasting love and support.
Jiaxu Li
December 11, 2004
Arizona State University, Tempe, Arizona USA
v
TABLE OF CONTENTS
Page
LIST OF TABLES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix
LIST OF FIGURES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . x
CHAPTER 1 Introduction and Physiological Background . . . . . . . . . . . . . 1
1. Diabetes Mellitus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
2. Glucose-Insulin Endocrine Metabolic Regulatory System . . . . . . . . . 4
3. The pancreas and Its Endocrine Hormones . . . . . . . . . . . . . . . . . 6
3.1. The pancreas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
3.2. Glucose Transporters . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.3. Secretion and Actions of Insulin . . . . . . . . . . . . . . . . . . . 10
3.4. Insulin Receptors . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.5. Insulin Resistance . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.6. Insulin Degradation and Clearance . . . . . . . . . . . . . . . . . 16
3.7. Production and Consumption of Glucose . . . . . . . . . . . . . . 17
4. Glucose Tolerance Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
5. The Organization of This Dissertation . . . . . . . . . . . . . . . . . . . 21
CHAPTER 2 The Ultradian Oscillations of Insulin Secretion . . . . . . . . . . . 23
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2. Sturis-Tolic ODE Model and Current Research Status . . . . . . . . . . . 25
3. Two Time Delay DDE Model . . . . . . . . . . . . . . . . . . . . . . . . 34
4. Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
vi
5. Global Stability of Steady State . . . . . . . . . . . . . . . . . . . . . . . 43
6. Linearization and Local Analysis . . . . . . . . . . . . . . . . . . . . . . 44
7. Numerical Analysis of Stability Switches and Bifurcations . . . . . . . . . 56
7.1. Insulin Response Time Delay
1
. . . . . . . . . . . . . . . . . . . 59
7.2. Glucose Infusion Rate G
in
. . . . . . . . . . . . . . . . . . . . . . 60
7.3. Insulin Degradation Rate d
i
. . . . . . . . . . . . . . . . . . . . . 63
7.4. Hepatic Glucose Production
2
. . . . . . . . . . . . . . . . . . . . 63
7.5. Parameter
1
vs. G
in
. . . . . . . . . . . . . . . . . . . . . . . . . 64
7.6. Parameter
1
vs. d
i
. . . . . . . . . . . . . . . . . . . . . . . . . . 66
7.7. Parameter G
in
vs. d
i
. . . . . . . . . . . . . . . . . . . . . . . . . 67
7.8. Insulin Concentration Peaks after Glucose Concentration Peaks . 68
8. Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
CHAPTER 3 Modeling Intra-Venus Glucose Tolerance Test . . . . . . . . . . . 75
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
2. Current Research Status . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
3. More Generic IVGTT Model . . . . . . . . . . . . . . . . . . . . . . . . . 80
4. Preliminary Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
5. Global Stability of Steady State . . . . . . . . . . . . . . . . . . . . . . . 87
6. Local Stability of Steady State and Stability Switch . . . . . . . . . . . . 93
7. Delay Independent Stability Results for Discrete Delay Model . . . . . . 95
8. Delay Dependent Stability Conditions . . . . . . . . . . . . . . . . . . . . 99
8.1. The case of discrete delay . . . . . . . . . . . . . . . . . . . . . . 100
8.2. The case of distributed delay . . . . . . . . . . . . . . . . . . . . . 101
8.3. Expression of H() . . . . . . . . . . . . . . . . . . . . . . . . . . 102
vii
9. Numerical Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
10. Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
CHAPTER 4 The Eects of Active -Cells: A Preliminary Study . . . . . . . . 108
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
2. Current Research Status . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
3. Active -Cell Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
4. Numerical Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
4.1. Insulin Response Delay and Hepatic Glucose Production Are Crit-
ical for Sustain Insulin Secretion Oscillations . . . . . . . . . . . . 115
4.2. Insulin Response Time Delay
1
as a Bifurcation Parameter . . . 115
4.3. Glucose Infusion Rate G
in
as a Bifurcation Parameter . . . . . . . 116
4.4. Peaks of Oscillations in One Cycle . . . . . . . . . . . . . . . . . 118
4.5. -cell Deactivation Rate k [0.01, 2] as a Bifurcation Parameter . 119
4.6. Parameter as a Bifurcation Parameter . . . . . . . . . . . . . . 120
4.7. The Changes of Insulin Degradation Rate d
i
[0.025, 0.1] Do Not
Aect the Oscillations . . . . . . . . . . . . . . . . . . . . . . . . 122
5. Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
viii
LIST OF TABLES
Table Page
1.4.1. Fasting Glucose Tolerance Test . . . . . . . . . . . . . . . . . . . . . 20
1.4.2. Oral Glucose Tolerance Test . . . . . . . . . . . . . . . . . . . . . . . 20
1.4.3. Gestational Diabetes Glucose Tolerance Test . . . . . . . . . . . . . . 21
2.2.1. Parameters in the Sturis-Tolic Model (2.2.1). . . . . . . . . . . . . . . 28
2.2.2. Parameters of the functions in the Sturis-Tolic Model (2.2.1). . . . . . 28
2.7.1. Parameters of the functions in Two Time Delay Model (2.3.1). . . . . 57
3.9.1. Parameters for subjects 6 and 7 in IVGTT Models (b
5
= 23min.) . . . 104
4.2.1. Parameters of the Model 4.2.1 . . . . . . . . . . . . . . . . . . . . . . 111
ix
LIST OF FIGURES
Figure Page
1.2.1. Glucose-Insulin Regulatory System . . . . . . . . . . . . . . . . . . . . . . . 7
1.3.1. Langerhans islets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.3.2. The cells secrete insulin when glucose concentration level elevated . . . . . . . 12
1.3.3. Insulin signals cells to utilize glucose . . . . . . . . . . . . . . . . . . . . . . 15
2.1.1. Insulin Secretion Ultradian Oscillations . . . . . . . . . . . . . . . . . . . . . 24
2.2.1. Physiological Glucose-Insulin Regulatory System . . . . . . . . . . . . . . . . 26
2.2.2. Functions f
i
(I), i = 1, 2, 4, 5. . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.3.1. Two Time Delay Glucose-Insulin Regulatory Model . . . . . . . . . . . . . . . 35
2.7.1. Bifurcation diagram of
1
. . . . . . . . . . . . . . . . . . . . . . . . . . . 59
2.7.2. Periods of periodic solutions when
1
[0, 20] and bifurcation diagram of d
i
. . . 60
2.7.3. Bifurcation diagram of G
in
[0, 2.16] . . . . . . . . . . . . . . . . . . . . . 61
2.7.4. Limit cycles in (G
in
, G, I)-space when G
in
[0, 2.16] . . . . . . . . . . . . . . 62
2.7.5. Periods of periodic solutions when G
in
[0, 2.16] . . . . . . . . . . . . . . . . 62
2.7.6. Periods and peak time dierences when d
i
changes in [0.001, 0.7] . . . . . . . . 63
2.7.7. Hepatic production delay has no impact to sustained oscillations . . . . . . . . . 64
2.7.8. Stability Region in (
1
, G
in
)-plane . . . . . . . . . . . . . . . . . . . . . . . 65
2.7.9. Bifurcation diagrams and stability regions in (
1
, G
in
)-space . . . . . . . . . . . 66
2.7.10. Stability Regions in (
1
, d
i
)-plane and (G
in
, d
i
)-plane . . . . . . . . . . . . . . 67
2.7.11. Glucose concentrations peak before insulin does . . . . . . . . . . . . . . . . . 69
3.9.1. Periodic solutions for the discrete delay model (3.3.2) for subject 6 and 7 . . . . . 105
4.3.1. Glucose-Insulin with Active -cell Interaction Diagram . . . . . . . . . . . . . 113
4.3.2. Function g(G) in GI-Model . . . . . . . . . . . . . . . . . . . . . . . . . 115
x
4.4.1. Orbits of (G, I, ) of GI model . . . . . . . . . . . . . . . . . . . . . . . . 116
4.4.2. Bifurcation diagram of
1
[0, 20] . . . . . . . . . . . . . . . . . . . . . . . 117
4.4.3. Bifurcation diagram of G
in
[0, 3.0] . . . . . . . . . . . . . . . . . . . . . . 118
4.4.4. Periodic solutions and periods when G
in
[0, 3.0] . . . . . . . . . . . . . . . . 119
4.4.5. Peaks of Oscillations in One Cycle . . . . . . . . . . . . . . . . . . . . . . . 120
4.4.6. Bifurcation diagram of k [0.01, 2] . . . . . . . . . . . . . . . . . . . . . . . 121
4.4.7. Bifurcation diagram of [0.0001, 0.1] . . . . . . . . . . . . . . . . . . . . . 121
4.4.8. Limit Cycles when [0.0001, 0.1] . . . . . . . . . . . . . . . . . . . . . . . 122
4.4.9. There is no bifurcation when d
i
[0.005, 0.01] . . . . . . . . . . . . . . . . . 122
4.5.1. Possible -cell pulsatile oscillation? . . . . . . . . . . . . . . . . . . . . . . 124
xi
CHAPTER 1
Introduction and Physiological Background
1. Diabetes Mellitus
Human bodies need to maintain a glucose concentration level in a narrow range
(70 - 109 ml/dl or 3.9 - 6.04 mmol/l). If ones glucose concentration level is signicantly
out of the normal range (70 - 110 ml/dl), this person is considered to have a the plasma
glucose problem: hyperglycemia (140 mg/dl or 7.8 mmol/l after an Oral Glucose
Tolerance Test, or 100 mg/dl or 5.5 mmol/l after a Fasting Glucose Tolerance Test)
or hypoglycemia (less than 40 mg/dl or 2.2 mmol/l) ([89], [96]).
Diabetes mellitus is a disease in the glucose-insulin endocrine metabolic regula-
tory system, in which the pancreas either does not release insulin or does not properly
use insulin to uptake glucose in the plasma ([9], [85]), which is referred as hyperglycemia.
The consequences are that the body does not metabolize the glucose and builds
up hyperglycemia which eventually damages the regulatory system. Complications of
diabetes mellitus include retinopathy, nephropathy, peripheral neuropathy and blind-
ness ([25]).
Diabetes mellitus is one of the worst diseases with respect to size of the aected
population. According to the data published by American Diabetes Association ([94]),
in the United States in 2002, 18.2 million people - 6.3% of the total population - had
2
diabetes. The direct and indirect cost of diabetes in 2002 was $132 billions. The world
wide diabetics population is much higher, especially in underdeveloped countries.
Diabetes mellitus is currently classied as type 1 diabetes or type 2 diabetes
([9], [85]). Type 1 diabetes was previously called insulin-dependent diabetes mellitus
(IDDM) or juvenile-onset diabetes. It develops when the bodys immune system de-
stroys pancreatic beta cells, the only cells in the body that make the hormone insulin,
which regulates blood glucose. This form of diabetes usually strikes children and young
adults, although disease onset can occur at any age. Type 1 diabetes may account
for 5% to 10% of all diagnosed cases of diabetes. Risk factors for type 1 diabetes in-
clude autoimmune, genetic, and environmental factors. Type 2 diabetes is adult onset
or non-insulin-dependent diabetes mellitus (NIDDM) as this is due to a decit in the
mass of cells, reduced insulin secretion [53], and resistance to the action of insulin
[32]. The relative contribution and interaction of these defects in the pathogenesis of
this disease remains to be claried [17]. About 90% to 95% of all diabetics diagnose
type 2 diabetes. Type 2 diabetes is associated with older age, obesity, family history
of diabetes, prior history of gestational diabetes, impaired glucose tolerance, physical
inactivity, and race/ethnicity. African Americans, Hispanic/Latino Americans, Native
Americans, some Asian Americans, Native Hawaiian, and other Pacic Islanders are
at particularly high risk for type 2 diabetes. Type 2 diabetes is increasingly being
diagnosed in children and adolescents ([93]).
In addition to Type 1 and Type 2 diabetes, gestational diabetes is a form of
glucose intolerance that is diagnosed in some women during pregnancy ([9], [85], [97]).
Gestational diabetes occurs more frequently among African Americans, Hispanic/Latino
Americans, and Native Americans. It is also more common among obese women and
3
women with a family history of diabetes. During pregnancy, gestational diabetes re-
quires treatment to normalize maternal blood glucose levels to avoid complications in
the infant. After pregnancy, 5% to 10% of women with gestational diabetes are found
to have type 2 diabetes. Women who have had gestational diabetes have a 20% to
50% chance of developing diabetes in the next 5-10 years. Other specic types of dia-
betes result from specic genetic conditions (such as maturity-onset diabetes of youth),
surgery, drugs, malnutrition, infections, and other illnesses. Such types of diabetes may
account for 1% to 5% of all diagnosed cases of diabetes ([97]).
The relative contribution and interaction of these defects in the pathogenesis of
this disease remains to be claried ([17]).
Due to the large population of diabetes patients in the world and the big health
expenses, many researchers are motivated to study the glucose-insulin endocrine metabolic
regulatory system so that we can better understand how the mechanism functions ([79],
[84], [85], [67], [74], [31], [85], [4] and their references), what cause the dysfunctions of
the system ([9] and its rich references), how to detect the onset of the either type of
diabetes including the so called prediabetes ([10], [83], [8], [97], [23], [57], [6], [63] and
their references), and eventually provide more reasonable, more eective, more ecient
and more economic treatments to diabetics. For example, according to Bergman ([6],
2002), there are now approximately 50 major studies published per year and more than
500 can be found in literature related to the so called minimal model ([10], [83], [8]) for
modeling the intra-venous glucose tolerance test.
4
2. Glucose-Insulin Endocrine Metabolic Regulatory System
Metabolism is the process of extracting useful energy from chemical bounds. A
metabolic pathway is a sequence of enzymatic reactions that take place in order to
transfer chemical energy from one form to another. The chemical adenosine triphos-
phate (ATP) is a common carrier of energy in a cell. There are two dierent ways to
form ATP:
1. adding one inorganic phosphate group (HPO
2
4
) to the adenosine diphosphate
(ADP), or
2. adding two inorganic phosphate groups to the adenosine monophosphate (AMP).
The process of inorganic phosphate group addition is referred to phosphorylation. Due
to the fact that the three phosphate groups in ATP carry negative charges, it requires
lots of energy to overcome the natural repulsion of like-charged phosphates when addi-
tional groups are added to AMP. So considerable amount of energy is released during
the hydrolysis of ATP to ADP ([51], [89] and [91]).
In the glucose-insulin endocrine metabolic regulatory system, the two pancre-
atic endocrine hormones, insulin and glucagon, are the primary dynamic factors that
regulate the system.
When the plasma glucose concentration rises, the elevation in the ratio of ATP/ADP
in a cell in the pancreas causes ATP-sensitive K
+
channels (K
ATP
channels) in the
plasma membrane to close. The decreased K
+
permeability leads to membrane depo-
larization, opening of voltage-dependent Ca
2+
channels, Ca
2+
inux, and eventual rise
of the cytosolic Ca
2+
concentration ([Ca
2+
]
c
) that triggers exocytosis ([91]).
5
When the serum insulin concentration increases, more insulin receptors of cells
are bound by insulin. The binding of insulin to its receptors on the surfaces of cell
membranes leads to an increase in glucose transporter (GLUT4) molecules in the outer
membrane of muscle cells and adipocytes, and therefore to an increase in the uptake
of glucose from blood into muscle and adipose tissue. Thus, the intracellular glucose is
consumed and energy is released ([91]).
After some amount of the plasma glucose is utilized by the cells and the concen-
tration level is low, the cells are signaled not to release insulin. Then the amount of
extracellular glucose transported into intracellular by the glucose transporters is signi-
cantly reduced or even stopped due to the decreased number of insulin receptors bound
by insulin. Therefore, the consumption of glucose is tremendously decreased.
When the glucose concentration level is low, the cells in the pancreas will
release glucagon to the liver and the liver will convert glucagon into glucose. The liver
also converts glycogen into glucose.
In short, when humans the plasma glucose concentration level is high, the fol-
lowing processes will occur:
1. the pancreas is signaled to release insulin from cells;
2. serum insulin (including newly secreted insulin) binds to the cells insulin recep-
tors,
3. the insulin receptors bound by insulin cause the glucose transporters (GLUT4)
transport glucose molecules into the cells;
4. the cells consume the glucose and convert to energy.
6
These processes decrease the glucose concentrations in the plasma. Almost all the cells
in human body have insulin receptors, including fat cells and muscle cells. Glucose
is also utilized by other cells without insulin involvement. The brain cell is a typical
example.
When a humans the plasma glucose concentration level is low, a dierent series
of processes will occur:
1. the pancreas is signaled to release glucagon from cells;
2. glucagon is transported to the liver;
3. the liver converts the glucagon to glucose.
These processes increase the glucose concentration level in human plasma.
Exogenous glucose infusion also increases glucose concentration. The typical ex-
ogenous glucose infusions include meal ingestion, oral glucose intake, continuous enteral
nutrition, and constant glucose infusion.
The liver plays a key role in keeping the glucose and insulin amount in human
plasma oscillating smoothly ([96]). Figure 1.2.1, which is adapted from [96], illustrates
the plasma glucose-insulin endocrine metabolic regulatory system.
3. The pancreas and Its Endocrine Hormones
3.1. The pancreas. The pancreas lies interior to a humans stomach, in the
size of a humans st and is in the bend of the duodenum. Scattered through out
inside of the pancreas, there are about a million Langerhans islets. Each Langerhans
islet contains about three hundred cells and each cell contains about one thousand
granules. Approximately 5% of the total pancreatic mass is comprised of endocrine
7
Glucose Infusion,
meal, enteral,
oral intake
and others
Exercises,
fasting
and others
-cells release
glucagon
-cells release insulin
Liver converts partial
glucagon released
from -cells and partial
glycogen stored in liver
to glucose
Low
Plasma Glucose
Level
High
Plasma Glucose
Level
Normal
Plasma Glucose
Level
Pancreas
Insulin
Liver
Glucagon
Insulin helps
to consume
plasma glucose
Figure 1.2.1. Glucose-Insulin Regulatory System
The dashed lines indicate that exercises and fasting consume glucose and lower the glucose concentra-
tion, which signals the pancreas to release glucagon and the liver converts the glucagon and glycogen
to glucose. The solid lines indicate that the glucose infusion elevate the plasma glucose concentration
level which signals the pancreas to secrete insulin and consume the glucose. (This gure is adapted
from [96].)
cells. These endocrine cells are clustered in groups within the pancreas, which look
like little islands of cells when examined under a microscope. The pancreas is both an
endocrine and an exocrine gland. The exocrine functions are concerned with digestion.
The endocrine function consists primarily for the secretion of the two major hormones,
insulin and glucagon, which participate in the regulation of carbohydrate metabolism.
Five types of cells in a Langerhans islet are identied: cells, which occupy
65-80% of the islet and make insulin; cells, which occupy 15-20% and make glucagon;
cells, which occupy 3-10% and make somatostatin ([87]); and pancreatic polypeptide-
containing PP cells and D
1
cells comprise 1% ([2]), about which little is known. Figure
8
3.1 shows a Langerhans islet and the cells it contains. The cells in the pancreas are
the only cells in which the growth hormone insulin is synthesized and secreted. The
insulin is synthesized in cells as part of a larger preprohormone - preproinsulin - which
includes a 23 amino acid leader sequence attached to proinsulin; this leader sequence
is lost upon entrance of the molecule into the endoplasmic reticulum leaving the pro-
insulin molecule. Kallikrein, an enzyme present in the islets, aids in the conversion
of proinsulin to insulin. In this conversion, a C-peptide chain is removed from the
proinsulin molecule producing the disulde-connected and chains that are insulin.
Insulin is an anabolic hormone, that is, it increases the storage of glucose, fatty acids
and amino acids in cells and tissues.
Figure 1.3.1. Langerhans islets
A Langerhans islet contains -cells, -cells, -cells and others. This image contains three islets of a
house. (The image is from [89].)
It has been believed that the cells do not replicate and neogenerate after ones
birth. If ones cells are damaged in a large amount, he/she might have to suer
diabetes due to cell dysfunction. However, this hypothesis was recently challenged
by S. Bonner-Weir ([12], 2000). The new perspective assumes that the -cells can be
replicated and neogenerated. Several invivo and invitra experiments support this
9
new perspective ([13], [14], [34], [82] and [49]).
The cells release glucagon, a protein hormone that has important eects in the
regulation of carbohydrate metabolism. Glucagon is a catabolic hormone, that is, it
mobilizes glucose, fatty acids and amino acids from storage into the blood. When the
glucose concentration level in the plasma is low, the liver will convert the glucagon to
glucose.
Both insulin and glucagon are important in the regulation of carbohydrate, pro-
tein and lipid metabolism.
Somatostatin is secreted from the cells in the Langerhans islets in the pan-
creas and is a hormone inhibiting the secretion of many other hormones. Somatostatin
acts through both endocrine and paracrine pathways to aect its target cells. In the
pancreas, somatostatin appears to act primarily in a paracrine manner to inhibit the se-
cretion of both insulin and glucagon. In the brain (hypothalamus) and the spinal cord
it may act as a neurohormone and neurotransmitter. The eects of somatostatin to
glucose-insulin regulatory system is small, indirect and negligible. Its paracrine manner
makes the secretion of insulin and glucagon smoother.
3.2. Glucose Transporters. Glucose is transported by its transporters. There
are total ve transporters in the family, that is, GLUT1 to GLUT5 ([91]).
GLUT1 is ubiquitously distributed in various tissues.
GLUT2 is found primarily in intestine, kidney and liver.
GLUT3 is found in the intestine.
GLUT4 is primarily contained in insulin-sensitive tissues such as skeletal muscle
and adipose tissue.
10
GLUT5 is found in the brain and testis. GLUT5 is also the major glucose trans-
porter present in the membrane of the endoplasmic reticulum (ER) and serves the
function of transporting glucose to the cytosol following its dephosphorylation by
the ER enzyme glucose-6-phosphatase.
When the concentration of blood glucose increases in response to food intake,
pancreatic GLUT2 molecules mediate an increase in glucose uptake which leads to
increased insulin secretion. Recent evidence has shown that the cell surface receptor
for the human T cell leukemia virus (HTLV) is the ubiquitous GLUT1. ([91])
3.3. Secretion and Actions of Insulin. Insulin secretion is pulsatile and is
regulated primarily by the glucose metabolism ([67], [74]). Numerous in-vivo and in-
vitro experiments have shown that insulin concentration oscillates in two dierent time
scales: rapid oscillation with a period of 5-15 minutes and ultradian oscillation with a
range of 50-140 minutes ([67], [74] and their cited references). The rapid oscillations
are caused by coordinating periodic secretory bursting of insulin from cells contained
in millions of the Langerhans islets in the pancreas. These bursts are the dominant
mechanism of insulin release at basal level ([67]). Ultradian oscillations of insulin con-
centration are believed to be mainly due to glucose interaction in the plasma ([79], [84],
[74]). These ultradian oscillations are best seen after meal ingestion, oral glucose intake,
continuous enteral nutrition or intravenous glucose infusion ([79]). In addition, muscle,
the brain, nerve and others utilize the plasma glucose to complete the regulatory system
feedback loop. So, insulin production, glucose infusion and production (for example,
meal and continuous enteral nutrition in daily life) and glucose utilization (for example,
in daily life, exercise) are the three major variables of this intricate regulatory system
([74], [79]).
11
P. Gilon, M. A. Ravier, J.-C. Jonas, and J.-C. Henquin summarized the mech-
anism of insulin secretion control in 2002 ([39]). Glucose stimulates insulin secretion
from -cells by activating two pathways that require metabolism of the sugar as follows
([47]).
Triggering Pathway The GLUT2 transports the glucose into the cell. It
causes the rise in the ratio of ATP/ADP which causes ATP-sensitive K
+
channels
(K
ATP
channels) in the plasma membrane to close. The decreased K
+
permeability
leads to membrane depolarization, opening of voltage-dependent Ca
2+
channels,
Ca
2+
inux, and the eventual rise of the cytosolic Ca
2+
concentration ([Ca
2+
]
c
)
that triggers exocytosis. This pathway is also called K
ATP
channel-dependent
pathway. See Figure 1.3.2 for an illustration.
Amplifying Pathway The K
ATP
channel-independent pathway simply increases
the eciency of the Ca
2+
on exocytosis when the concentration of Ca
2+
has been
elevated.
The pulsatility of insulin secretion might result from oscillations in either of these
transduction pathways. Because metabolism and [Ca
2+
]
c
play key roles in the control
of insulin secretion and have been reported to oscillate, many eorts have been spent
to investigate which of these two mechanisms is the primary factor of pulsatile insulin
secretion ([39]). The essential role of Ca
2+
inux in the generation of [Ca
2+
]
c
oscillations
by glucose, in either whole islets or single -cells, is demonstrated by their abrogation
upon omission of extracellular Ca
2+
([44], [38]) or blockade of voltage-dependent Ca
2+
channels ([26]). [Ca
2+
]
c
oscillations are linked to oscillations of the membrane potential
in -cells ([72], [38]), and it is assumed that mixed [Ca
2+
]
c
oscillations result from an
irregular (so-called periodic) electrical activity ([3], [46], [20]). Synchronization of the
12
-cell electrical activity ([62]) by gap junctions is likely to underlie the synchronization
of [Ca
2+
]
c
oscillations between -cells within the islet ([44], [50] and [43]). See Figure
1.3.2 for an illustration.
Elevate
K+
Close K+
channels
Open Ca2+
channeles
Ca2+
influx
G
r
a
n
u
le
s
I
n
s
u
lin
Elevated Ca2+
NAD(P)H
H+
Glucokinase
G
l
u
c
o
s
e
Cell Depolarization
Protein
Phosphorylations
Glucose-6-phosphate
ATP
ADP
G
L
U
T
2
Glucose Metabolism
Figure 1.3.2. The cells secrete insulin when glucose concentration level elevated
The facilitated GLUT2 transport the glucose into the cell and the glucose is phosphorylated by
glucokinase. The ratio of ATP:ADP is elevated. The glucose metabolism causes ATP-sensitive K
+
channels to close, the membrane to depolarize and the Ca
2+
channels to open. This triggers a cascade
of protein phosphorylations and leads to insulin exocytosis [68]. (The gure is partially adapted from
[68].)
The insulin has ve major actions. These include:
facilitation of glucose transport through certain membranes (e.g. adipose and
muscle cells);
stimulation of the enzyme system for conversion of glucose to glycogen (liver and
muscle cells);
slow-down of gluconeogenesis (liver and muscle cells);
13
regulation of lipogenesis (liver and adipose cells); and
promotion of protein synthesis and growth (general eect).
These actions of insulin are mediated by the binding of the hormone to membrane re-
ceptors to trigger several simultaneous actions. A major eect of insulin is to promote
the entrance of glucose and amino acids in cells of muscle tissues, adipose tissue and
connective tissue. Glucose enters the cell by facilitated diusion along an inward gradi-
ent created by low intracellular free glucose and by the availability of a specic carrier
called transporter. In the presence of insulin, the rate of movement of glucose into the
cell is greatly stimulated in a selective fashion. ([89].)
In the liver, insulin does not aect the movement of glucose across membranes
directly but facilitates glycogen deposition and decreases glucose output. Consequently,
there is a net increase in glucose uptake. Insulin induces or represses the activity of
many enzymes; however, it is not known whether these actions are direct or indirect. For
example, insulin suppresses the synthesis of key gluconeogenic enzymes and induces the
synthesis of key glycolytic enzymes such as glucokinase. Glycogen synthetase activity is
also increased. Insulin likewise increases the activity of enzymes involved in lipogenesis
.
3.4. Insulin Receptors. In molecular biology, the insulin receptor is a trans-
membrane glycoprotein that is activated by insulin. It belongs to the large class of
tyrosine kinase receptors. Two subunits and two subunits make up the insulin re-
ceptor. The subunits pass through the cellular membrane and are linked by disulde
bonds ([90]).
The insulin receptors are embedded in the plasma membrane of hepatocytes
and myocytes. The binding of insulin to the receptors is the initial step in a signal
14
transduction pathway, triggering the consumption and metabolism of glucose ([89], [86]).
Bound by insulin, the insulin receptor phosphorylates from ATP to several proteins
in the cytoplasm, including insulin receptor substrates (IRS-1 and IRS-2) containing
signaling molecules, activates Phosphatidylinositol 3-kinase (PI-3-K) and leads to an
increase in glucose transporter (GLUT4) molecules ([98]) in the outer membrane of
muscle cells and adipocytes, and therefore to an increase in the uptake of glucose from
blood into muscle and adipose tissue ([89]). GLUT4 will transport the glucose to the
cells eciently. Figure 1.3.3 elucidates this signaling pathway.
Intracellular phosphorylation of glucose is rapid and ecient and therefore the
glucose concentration is low. Thus, a certain amount of glucose moves into the cell
regardless of the existence of insulin. With insulin, however, the rate of glucose entry is
much increased due to the facilitated diusion as mediated by the glucose transporters
([89]). Refer to Figure 1.3.3.
However, the kinetics of insulin receptor binding are complex. The number of
insulin receptors of each cell changes opposite to the circulating insulin concentration
level. Increased insulin circulating level reduces the number of insulin receptors per cell
and the decreased circulating level of insulin triggers the number of receptors to increase.
The number of receptors is increased during starvation and decreased in obesity and
acromegaly. But, the receptor anity is decreased by excess glucocorticoids. The
anity of the receptor for the second insulin molecule is signicantly lower than for the
rst bound molecule. This may explain the negative cooperative interactions observed
at high insulin concentrations. That is, as the concentration of insulin increases and
more receptors become occupied, the anity of the receptors for insulin decreases.
Conversely, at low insulin concentrations, positive cooperation has been recorded. That
15
ATP
Phosphorylations
IRS-1
IRS-2
PI-3 Kinase
GLUT4
Insulin
Cell membrane
G
Other activities
Glucose
G
G
G
G
G
G
I
G
I
I
G
G
G
G
Insulin receptor
G
-
Unit
-
Unit
-
Unit
-
Unit
-S-S-
-S-S-
-S-S-
G
G
Figure 1.3.3. Insulin signals cells to utilize glucose
Insulin binds to its receptors on the membrane of the cells and phosphorylates several proteins in the
cytoplasm, including insulin receptor substrates (IRS-1 and IRS-2) containing signaling molecules, ac-
tivates Phosphatidylinositol 3-kinase (PI-3-K) and leads to an increase in glucose transporter (GLUT4)
molecules. This leads to an increase in glucose transporter (GLUT4) molecules. GLUT4 will transport
the glucose to the cells eciently.
is, the binding of insulin to its receptor at low insulin concentrations seems to enhance
further binding (([89]), [86]).
3.5. Insulin Resistance. Insulin resistance is dened as when insulin is inef-
cient in causing the plasma glucose to enter the cells of a body and to be utilized by
the cells for energy, even if there is enough insulin in serum. That is, the cells resist
the insulin. In addition, the liver may continue to secrete glucose into the bloodstream
even when the glucose is not needed.
The reasons for insulin resistance occurring are still uncertain. Certain genes
predispose certain people to develop insulin resistance. Some factors are, for example,
lack of exercise, obesity, and chronically high blood sugar levels may cause insulin
resistance in susceptible individuals. [95]
16
Previously, the perspective was that the abnormal binding to the insulin receptors
of the cells was the major reason of insulin resistance. This is no longer believed to be
the case. [95]
Currently, many researchers are active in determining the cause of insulin resis-
tance at the cellular and molecular levels. Postbinding abnormalities, believed by
most researchers, is the cause of insulin resistance. Several chemical pathways and
genes causing the abnormalities have been identied. A typical example is that the
glucose transporter GLUT4 is decient in some individuals showing insulin resistance.
The activity of GLUT4 is to transport the glucose into the body cells after the insulin
is bound to the insulin receptors. [95]
3.6. Insulin Degradation and Clearance. Insulin degradation is a broad
and rich research area and this is not the major focus of this dissertation. We will only
discuss this briey. (For more information, refer to [5], [27], [33], [42] and their cited
references.)
Insulin is cleared mainly by the liver and kidney, but most other tissues also
degrade the hormone ([33]). Insulin-degrading enzyme (IDE) is the major enzyme in
the proteolysis of insulin in addition to several peptides ([27]). It resides in a region
of chromosome 10q that is linked to Type 2 diabetes ([42]). IDE is the major enzyme
responsible for insulin degradation in vitro, but the extent to which it mediates insulin
catabolism in vivo has been controversial, with doubts expressed that IDE has any
physiological role in insulin catabolism ([33] and cited references). Insulin is degraded
by enzymes in the subcutaneous tissue ([64]) and interstitial uid as well ([7]). The
insulin is degraded by insulin receptors as well as when the insulin is bound to its
receptors ([85]).
17
3.7. Production and Consumption of Glucose. Glucose is liberated from
dietary carbohydrates such as starch or sucrose by hydrolysis within the small intestine,
and then is absorbed into the blood. The most often ways of glucose infusion are through
meal ingestion; oral glucose intake; continuous enteral nutrition; and constant glucose
infusion ([79] and [84]).
Insulin controls the hepatic glucose production (conversion from glucagon) and
release rate by the liver ([89]). When the blood glucose level drops, the liver converts
glycogen to glucose and releases it into the bloodstream. When there is enough glucose
in the bloodstream, insulin secreted by the pancreas signals the liver to shut down
glucose production. In healthy people, the pancreas continually measures blood glucose
levels and responds by secreting just the right amount of insulin. The liver converts the
glycogen to glucose as well as when the plasma glucose concentration level is low.
The insulin receptor leads that the glucose molecules go into the muscle cells,
fat cells and others. These cells utilize the glucose. Elevated concentrations of glucose
in the blood stimulate the release of insulin. Insulin acts on cells throughout the body
to stimulate uptake, utilization and storage of glucose. Within seconds to minutes the
rate of glucose entry into tissue cells increases 15 to 20 times. Once glucose enters
the tissue cells, insulin enhances its oxidation, stimulates its conversion to glycogen,
activates transport of amino acids into cells, promotes protein synthesis and inhibits
virtually all liver enzymes that promote gluconeogenesis. The eects of insulin on
glucose metabolism vary depending on the target tissue. Two important eects are
([89]) (see also Figure 1.3.3 for an illustration.):
Higher Insulin Concentration Leads to More Glucose Uptake Insulin
facilitates entry of glucose into muscle, adipose and several other tissues. The only
18
mechanism by which cells can take up glucose is by facilitated diusion through a
family of hexose transporters. In many tissues, e.g., muscle, the major transporter
used for uptake of glucose (GLUT4) is made available in the plasma membrane
through the action of insulin.
Lower Insulin Concentration Leads to Less Glucose Uptake In the
absence of insulin, GLUT4 glucose transporters are present in cytoplasmic vesicles,
where they are useless for transporting glucose. Binding of insulin to receptors
on such cells leads rapidly to fusion of those vesicles with the plasma membrane
and insertion of the glucose transporters, thereby giving the cell the ability to
eciently take up glucose. When blood levels of insulin decrease and insulin
receptors are no longer occupied, the glucose transporters are recycled back into
the cytoplasm. Therefore, the glucose uptake is signicantly decreased.
Insulin stimulates the liver to store glucose in the form of glycogen. A large
fraction (50%) of glucose absorbed from the small intestine is immediately taken up by
hepatocytes, which convert it into the storage polymer glycogen ([89]).
Insulin has several eects in the liver that stimulate glycogen synthesis. First, it
activates the enzyme hexokinase, which phosphorylates glucose, trapping it within the
cell. Coincidentally, insulin acts to inhibit the activity of glucose-6-phosphatase. Insulin
also activates several of the enzymes that are directly involved in glycogen synthesis,
including phosphofructokinase and glycogen synthase. The net eect is clear: when the
supply of glucose is abundant, insulin signals the liver to store as much of it as possible
for use later ([89]).
Many cells consume the glucose without involvement of the insulin receptor eect.
The brain and the liver do not use GLUT4 to transport glucose. Instead, a type of
19
insulin-independent transport is used. This constitutes the insulin-independent glucose
utilizations ([89]).
4. Glucose Tolerance Test
A series of glucose tolerance tests have been developed over the year and applied
in clinics and experiments ([93], [10], [8], [41], [76], [16] and [61]). Each of the glucose
tolerance tests is to diagnose if an individual has diabetes or has potential to have
diabetes. The basic idea is to test ones glucose-insulin endocrine metabolic system
after a large amount of glucose infusion.
The glucose tolerance tests include Fasting Glucose Tolerance Test (FGTT),
Oral Glucose Tolerance Test (OGTT), Intra Venous Glucose Tolerance Test (IVGTT),
frequently sampled Intra Venous Glucose Tolerance Test (fsIVGTT) ([93], [60] and [59]).
The Fasting Glucose Tolerance Test (FGTT) needs the individual to fast for 8-10 hours
before his/her the plasma glucose is sampled. The meanings of the test results are
summarized in Table 1.4.1. The Oral Glucose Tolerance Test (OGTT) is another type
of glucose tolerance test. The individual is given a glass of glucose liquid (75mg) to
intake and his/her the plasma glucose level will be sampled. The test result meanings
are dened in Table 1.4.2. To diagnose gestational diabetes, a pregnant woman is
required to drink a glass of glucose water containing 50mg glucose. Her the plasma
glucose is sampled one hour later. The meanings of the test results are listed in Table
1.4.3. The American Diabetes Association suggests two tests need to be performed to
determine if an individual has diabetes or pre-diabetes ([93]).
The Intra-venous Glucose Tolerance Test (IVGTT) and the frequently sampled
Intra-venous Glucose Tolerance Test (fsIVGTT) are to test the insulin sensitivity or
20
Table 1.4.1. Fasting Glucose Tolerance Test
The plasma Glucose Meaning
70-99 mg/dl (3.9-5.4 mmol/l) normal glucose tolerance
100-125 mg/dl (5.5-6.9 mmol/l) impaired fasting glucose (pre-diabetes)
Over 126 mg/dl (7.0 mmol/l) and above probable diabetes
Table 1.4.2. Oral Glucose Tolerance Test
The plasma Glucose Meaning
Below 140 mg/dl (7.8 mmol/l) normal glucose tolerance
140-200 mg/dl (7.8-11.1 mmol/l) impaired fasting glucose (pre-diabetes)
Over 200 mg/dl (11.1 mmol/l) probable diabetes
response to high the plasma glucose concentration. The procedure of IVGTT is similar
to other glucose tolerance tests but the plasma glucose and serum insulin are sampled
more frequently. In the test, the individual to be tested needs to fast 8-10 hours and
is then given a bolus of glucose infusion, for example, 0.33 g/kg body weight [23]
or 0.5 g/kg body weight of a 50% solution and is administered into an antecubital
vein in approximately 2.5 minutes. Within the next 180 minutes, the individuals
the plasma glucose and serum insulin are sampled frequently. According to the rich
information in the sampled data, the insulin sensitivity can be accurately determined.
Many models study the Intravenous Glucose Tolerance Test (IVGTT), which focuses
on the metabolism of glucose in a short time period starting from the infusion of big
bolus (0.33 g/kg) of glucose at time t = 0. As pointed out in Chapter 2, due to the
large amount of intravenous glucose infusion, the insulin response time delay of the
small amount of hepatic glucose production is insignicant and thus negligible and
furthermore is assumed at a small constant infusion rate in the models ([10], [8], [23],
[57] and [63]). The most noticeable model is the so called Minimal Model which
21
Table 1.4.3. Gestational Diabetes Glucose Tolerance Test
The plasma Glucose Meaning
Below 140 mg/dl (7.8 mmol/l) normal glucose tolerance
Over 140 mg/dl (7.8 mmol/l) abnormal, needs oral glucose tolerance test
contains minimal number of parameters ([10], [8]) and it is widely used in physiological
research work to estimate metabolic indices of glucose eectiveness (S
G
) and insulin
sensitivity (S
I
) from the intravenous glucose tolerance test (IVGTT) data by sampling
over certain periods (usually 180 minutes) ([41]). Also a few are on the control through
meals and exercise ([25]). See also a review paper by Mari ([60]) for a classication of
models.
5. The Organization of This Dissertation
In this dissertation, we propose a more realistic DDE model for the insulin secre-
tion ultradian oscillations in Chapter 2. This model (Model (2.3.1)) contains two time
delays: the rst mimic the hepatic glucose production time delay and the other reects
the insulin response time delay to increased glucose concentration. Both analytical and
numerical analysis are performed. The results obtained include global and local sta-
bility analysis of steady state, persistence of solutions and numerical simulation with
insightful results.
In Chapter 3, we propose three models (Model (3.3.1), (3.3.2) and (3.3.3)) for
modeling the eective and powerful intravenous glucose tolerance test. We performed
global and local stability analysis of the steady state and numerical simulations based
on clinic data from diabetics.
22
In Chapter 4, we present another DDE model to investigate the eects of the
mass of the active cells. Our numerical analysis shows that we simulated the glucose-
insulin endocrine metabolic system taking active cell mass into account. Due to the
fact that this area is relatively new, our study is still preliminary. More thorough studies
are needed.
CHAPTER 2
The Ultradian Oscillations of Insulin Secretion
1. Introduction
Endocrine systems often secrete hormones in pulses [21] [56]. Examples include
the release of growth hormone and gonadotropins, and also the secretion of insulin from
the pancreas, which are secreted over intervals of 1-3 hours and 80-150 minutes, respec-
tively. It has been suggested that relative to constant or stochastic signals, oscillatory
signals are more eective at producing a sustained response in the target cells [40] [58].
Numerous in-vivo and in-vitro experiments have shown that insulin concentration
oscillates in two dierent time scales: rapid oscillation with a period of 5-15 minutes
and ultradian oscillation with a range of 80-150 minutes ([79], [67], [74] and [73]).
The mechanisms underlying both types of oscillations are not fully understood.
The rapid oscillations may arise from an intra-pancreatic pacemaker mechanism [77]
and caused by coordinating periodic secretory bursting of insulin from cells contained
in the millions of the Langerhans islets in the pancreas. These bursts are the domi-
nant mechanism of insulin release at basal level ([67]). Often, the rapid oscillation is
superimposed on the slow (ultradian) oscillation ([79]).
Ultradian oscillations of insulin concentration are believed to be mainly due to
glucose interaction in the plasma and an instability in the insulin-glucose feedback sys-
24
G
lu
c
o
s
e

(
g
m
/
d
l)
I
n
s
u
lin

(

U
/
m
l)
80
100
120
140
240 480 720 960 1200 1440
30
A
10
20
40
0
50
G
lu
c
o
s
e

(
g
m
/
d
l)
60
100
120
160
60 120 160 240
140
80
0
40
60
100
80
20
I
n
s
u
lin

(

U
/
m
l)
B
G
lu
c
o
s
e

(
g
m
/
d
l)
I
n
s
u
lin

(

U
/
m
l)
240 480 720 960 1200 1440
30
C
10
20
40
0
50
G
lu
c
o
s
e

(
g
m
/
d
l)
100
120
140
180
240 480 720 1200
160
10
20
30
40
I
n
s
u
lin

(

U
/
m
l)
D
40
80
100
140
120
60
840
Figure 2.1.1. Insulin Secretion Ultradian Oscillations
These gures illustrate the insulin secretion ultradian oscillations. The glucose infusion rate are A.
meal ingestion; B. oral glucose intake; C. continuous enteral nutrition; D. constant glucose infusion,
respectively. (The gures are adapted from [79].)
tem ([79], [84], [74] and [60]). These ultradian oscillations are best seen after meal
ingestion, oral glucose intake, continuous enteral nutrition or intravenous glucose in-
fusion (Figure 2.1.1). In addition, muscles, the brain, nerves and others utilize the
plasma glucose to complete the regulatory system feedback loop ([79], [84]). So, insulin
production, glucose infusion and production (for example, meal and continuous enteral
nutrition in daily life) and glucose utilization (for example, in daily life, exercise) are
the three major factors of this intricate regulatory system ([74], [79] and [59]).
The hypothesis that the ultradian insulin secretion is an instability in the insulin-
glucose feedback system has been the subject of a number of studies, including some
which have developed a mathematical model of the insulin-glucose feedback system
([51], [79], [84], [31] and [4]).
This chapter is organized as follows. Section 2 summarizes the current study
status with focus on the Sturis-Tolic Model. Section 3 presents our two time delay
25
model for the insulin secretion ultradian oscillations. Preliminary results are presented
in Section 4. Two global stability results of the steady state are given in Section 5 and
followed by Section 6 for the local stability study. Intensive numerical simulations are
presented in Section 7 and then Section 8 is for discussion.
2. Sturis-Tolic ODE Model and Current Research Status
To determine whether the ultradian oscillations could result from the interaction
between insulin and glucose, a parsimonious nonlinear mathematical model consist-
ing the six ordinary dierential equations including the major mechanisms involved
in glucose regulation was developed by J. Sturis, K. S. Polonsky, E. Mosekilde and
E. Van Cauter ([79]) in 1991 and recently simplied by I. M. Tolic, E. Mosekilde and
J. Sturis ([84]) in 2000. The purpose of these two models was to provide a possible
mechanism for the origin of the ultradian insulin secretion oscillations. Included in this
model is the feedback loop (refer to Figure 2.2.1): glucose stimulates pancreatic insulin
secretion, insulin stimulates glucose uptake and inhibits hepatic glucose production,
and glucose enhances its own uptake. The model takes following form.
26
Insulin
Glucose
Glucose
Production
Glucose
Utilization
Insulin
Secretion
(-) (-)
(-) (-)
Figure 2.2.1. Physiological Glucose-Insulin Regulatory System
These four negative feedback loop show the glucose stimulating pancreatic beta cells to secrete insulin,
insulin stimulating glucose uptake and inhibiting hepatic glucose production, and also positive feedback
as glucose enhances its own uptake ([79]). (This gure is adapted from [79].)
_

_
dG(t)
dt
= G

= G
in
f
2
(G(t)) f
3
(G(t))f
4
(I
i
(t)) + f
5
(x
3
),
dI
p
(t)
dt
= I

p
= f
1
(G(t)) E(
I
p
(t)
V
p

I
i
(t)
V
i
)
I
p
(t)
t
p
,
dI
i
(t)
dt
= I

i
= E(
I
p
(t)
V
p

I
i
(t)
V
i
)
I
i
(t)
t
i
,
dx
1
(t)
dt
= x

1
=
3
t
d
(I
p
x
1
),
dx
2
(t)
dt
= x

2
=
3
t
d
(x
1
x
2
),
dx
3
(t)
dt
= x

3
=
3
t
d
(x
2
x
3
),
(2.2.1)
27
where G(t) is the amount of glucose, I
p
(t) and I
i
(t) are the amount of insulin in the
plasma and the intercellular space, respectively, V
p
is the plasma insulin distribution
volume, V
i
is the eective volume of the intercellular space, E is the diusion transfer
rate, t
p
and t
i
are insulin degradation time constants in the plasma and intercellu-
lar space, respectively, G
in
indicates (exogenous) glucose supply rate to plasma, and
x
1
(t), x
2
(t) and x
3
(t) are three auxiliary variables associated with certain delays of the
insulin eect on the hepatic glucose production with total time t
d
. f
1
(G) is a function
modeling the pancreatic insulin production as controlled by the glucose concentration,
f
2
(G) and f
3
(G)f
4
(I
i
) are functions, respectively, for insulin-independent and insulin-
dependent glucose utilization by various body parts (for example, brain and nerves (f
2
),
and muscle and fat cells (f
3
f
4
)) and f
5
(x
3
) is a function modeling hepatic glucose pro-
duction with time delay t
d
collaborated with auxiliary variables x
1
, x
2
and x
3
. Based
on experimental results ([79], [84]), all the parameters in the model are given in Table
(2.2.1) and f
i
, i = 1, 2, 3, 4, 5, take following forms and the parameters listed in Table
2.2.2.
f
1
(G) =
R
m
1 + exp((C
1
G/V
g
)/a
1
)
, (2.2.2)
f
2
(G) = U
b
(1 exp(G/(C
2
V
g
))), (2.2.3)
f
3
(G) =
G
C
3
V
g
, (2.2.4)
f
4
(I
i
) = U
0
+
0.1(U
m
U
0
)
1 + exp( ln(I
i
/C
4
(1/V
i
+ 1/Et
i
)))
, (2.2.5)
28
Table 2.2.1. Parameters in the Sturis-Tolic Model (2.2.1).
Parameters Values Units
V
p
3 l
V
i
11 l
E 0.2 l min
1
t
p
6 min
t
i
100 min
Table 2.2.2. Parameters of the functions in the Sturis-Tolic Model (2.2.1).
Parameters Units Values
V
g
l 10
R
m
Umin
1
210
a
1
mg l
1
300
C
1
mg l
1
2000
U
b
mg min
1
72
C
2
mg l
1
144
C
3
mg l
1
1000
Parameters Units Values
U
0
mgmin
1
40
U
m
mgmin
1
940
1.77
C
4
Ul
1
80
R
g
mgmin
1
180
lU
1
0.29
a
1
Ul
1
26
f
5
(x) =
R
g
1 + exp((x/V
p
C
5
))
, (2.2.6)
Figure (2.2.2) display the graphs of the above functions, f
i
, i = 1, 2, 3, 4, 5. The
importance of these functions is their shapes rather than their forms [51].
This model comprised of two major negative feedback loops describing the eects
of insulin on glucose utilization and glucose production, respectively, and both loops
include the stimulatory eect of glucose on insulin secretion. The authors of [84] hoped
to identify a possible mechanism behind the eciency of oscillatory insulin secretions.
Analysis of the original model revealed that the slow oscillations of insulin secretion
could arise from a Hopf bifurcation in the insulin-glucose feedback mechanism. The
model included several feedback loops (see Figure 2.2.1), including: glucose stimulating
29
G
40000 30000 20000 10000 0
70
60
50
40
30
20
10
0
800
600
400
200
I
400 300 200 100 0
f
2
(G) f
4
(I)
160
120
80
40
0
x
200 150 100 50 0
200
150
100
50
0
G
40000 30000 20000 10000 0
f
5
(I) f
1
(G)
Figure 2.2.2. Functions f
i
(I), i = 1, 2, 4, 5.
pancreatic beta cells to secrete insulin, insulin stimulating glucose uptake and inhibiting
hepatic glucose production, and also positive feedback as glucose enhances its own
uptake.
The model includes two signicant delays. One, 5-15 min., is sluggish eect of
insulin on glucose utilization, reecting that the eect is dependent on the concentra-
tion of insulin in a slowly equilibrating intercellular compartment as opposed to the
concentration of the plasma insulin. The other delay, 25-50 min., is due to the time
lag between the appearance of insulin in the plasma and its inhibitory eect on hepatic
30
glucose production. This delay is simulated by introducing three auxiliary variables
x
1
, x
2
and x
3
, which is called the third order delay. We demonstrate how the auxiliary
variables simulate time delay as follows. For simplicity, assume the rst order delay,
that is, x

1
(t) = (I
p
(t) x
1
(t))/t
d
, where t
d
> 0 is the time delay. Then
I
p
(t t
d
) = x
1
(t t
d
) + x

1
(t t
d
)t
d
Observe the Taylors expansion of x
1
(t) at t t
d
,
x
1
(t) = x
1
(t t
d
) + x

1
(t t
d
)t
d
+ o(t
d
).
So x
1
(t) I
p
(t t
d
). The occurrence of sustained insulin and glucose oscillations was
found numerically to be dependent on these two time delays.
Model simulations suggested that the interaction of the oscillatory insulin supply
with the glucose receptors of the glucose utilizing cells was of minimal importance. This
was because the oscillations in the concentration of the intercellular insulin were small,
and changes in the average glucose utilization only depend weakly on amplitude. How-
ever, with their model they were able to resolve conicting results from clinical studies.
Dierent experimental conditions will inuence hepatic glucose release. If hepatic glu-
cose release is occurring near its maximum limit, an oscillatory insulin supply will be
more eective at lowering the blood glucose level than a constant supply. However, if
the insulin level is suciently high to cause the hepatic release of glucose to virtually
disappear, the opposite is observed. For insulin concentrations close to the point of
inection of the insulin-glucose curves (f
1
and f
5
), an oscillatory and a constant insulin
secretion produce similar eects. Under the assumption of constant glucose infusion,
the authors observed following numerical observations.
ST1 The ultradian insulin secretion oscillation is critically dependent on hepatic glu-
cose production, that is, if there is no hepatic glucose production, then there is
31
no insulin secretion oscillation.
ST2 When the hepatic glucose production time delay
2
(25, 50), the period of the
periodic solutions of both insulin and glucose is in interval (95, 140) (min.), that
is, (95, 140).
ST3 To obtain the ultradian oscillation (periodic solutions), it is necessary to break
the insulin into two separate compartments, the plasma and interstitial tissues.
ST4 The ultradian oscillation is sensitive to both the speed of insulin reaction to the
increased plasma glucose concentration level and the speed of the hepatic glucose
production triggered by insulin. Specically, if the slope in the reexive points of
function f
1
and f
5
is reduced by 10 20%, the oscillation becomes damped.
K. Engelborghs, V. Lemaire, J. Belair and D. Roose ([31], 2001) introduced a
single time delay in the Negative Feedback Loop Model and proposed following DDE
model.
_

_
G

(t) = E
g
f
2
(G(t)) f
3
(G(t))f
4
(I(t)) +f
5
(I(t )),
I

(t) = f
1
(G(t))
I(t)
t
1
,
(2.2.7)
where the functions, f
i
, i = 1, 2, 3, 4, 5, and their parameters are assumed to be the
same as those in the Model (2.2.1). E
g
stands for the glucose infusion rate and the
term 1/t
1
is the insulin degradation rate. The positive constant delay mimics the
hepatic glucose production delay (5-15 min.). This model ignores the glucose stimulat-
ing insulin secretion time delay. Due to the complex chemical reactions on the cells,
32
the insulin secretion occurs a few minutes after the plasma glucose concentration rises.
This signicant time delay (5-15 min.) is not negligible in physiology.
The other DDE model proposed by K. Engelborghs, V. Lemaire, J. Belair and
D. Roose ([31], 2001) is trying to model the exogenous insulin infusion. The authors
assumed that the exogenous insulin infusion function takes the same form as internal
insulin production, which is, as the authors admitted, too articial.
_

_
G

(t) = E
g
f
2
(G(t)) f
3
(G(t))f
4
(I(t)) + f
5
(I(t
2
)),
I

(t) = f
1
(G(t))
I(t)
t
1
+ (1 )f
1
(G(t
1
)).
(2.2.8)
Nevertheless, a noticeable addition to the work of [31] is the usage of DDE-
BifTool software package ([30]) to analyze and simulate the bifurcation diagram and
other numerical analysis.
Due to the lack of physiological meanings, we would not summarize the analytical
and numerical results presented in [31].
In 2004, D. L. Bennett and S. A. Gourley ([4]) modied the Sturis-Tolic ODE
Model ([79] and [84]) by removing the three auxiliary linear chain equations and their
associated articial parameters and introducing a time delay into the model explicitly.
This time delay stands for the hepatic glucose production, which is the same as
proposed in [31]. Unlike [31] in which the sluggish eect of glucose on insulin is ignored,
D. L. Bennett and S. A. Gourley ([4]) kept the idea in [79] and [84] of breaking the
insulin in two compartments to simulate the time delay of insulin secretion triggered
by rising glucose concentration level. The DDE model takes following form. All the
33
parameters and functions are the same as that in model (2.2.1) given in (2.2.2) to (2.2.6)
and Table 2.2.1 and 2.2.2.
_

_
G

(t) = G
in
f
2
(G(t)) f
3
(G(t))f
4
(I
i
(t)) + f
5
(I
p
(t )),
I

p
(t) = f
1
(G(t)) E(
I
p
(t)
V
p

I
i
(t)
V
i
)
I
p
(t)
t
p
,
I

i
(t) = E(
I
p
(t)
V
p

I
i
(t)
V
i
)
I
i
(t)
t
i
,
(2.2.9)
Their major analytical results are a sucient condition of global asymptotical
stability induced by a Liapunov function for the case that the hepatic glucose production
time delay = 0 and one for the case > 0. This analytical result shows that if the
hepatic glucose production time delay and the insulin degradation time delay between
the plasma and interstitial compartments t
i
and t
d
are suciently small, then solutions
converge globally to the steady state or the basel levels of glucose and insulin. In other
words, there are no sustained oscillations. For larger delay, whose range is not given in
[4], oscillatory solutions become possible and under these circumstances it seems that
likely candidates for having sustainable oscillatory insulin and glucose levels are those
subjects with low degradation rates of the two insulin compartments.
Two other observations in [4] are that large glucose infusion rate could cause
insulin secretion oscillations, and the insulin oscillations are sensitive to the values of
|f

1
(C
1
V
g
)| = R
m
/(4a
1
V
g
) or |f

5
(C
5
V
p
)| = R
g
/(4V
p
). This means if the cells do not
release enough insulin into the bloodstream, or glucose production is not sensitive to
insulin and keeps at a constant moderate rate (R
g
/2), then the insulin oscillation will
34
not sustain. Similarly, if the hepatic glucose production rate R
g
is too small, regardless
of sensitivity to insulin, the oscillations of insulin and glucose disappear.
3. Two Time Delay DDE Model
Glucose molecules are in the bloodstream or the plasma. When the concentration
level rises, electronic signals are sent to the pancreas and the cells secrete insulin.
The liver delivers the insulin into the plasma. This process takes about 5-15 minutes
depending dierent individuals. So, to more intuitively and precisely model the glucose-
insulin ultradian oscillations, we introduce two time delay parameters in to the glucose
and insulin regulatory system. The model diagram is shown in Figure 2.3.1. We remove
the insulin compartment split in the Sturis-Tolic Model ([79], [84]). The two time
delays are the hepatic glucose production time delay
2
as in [4] and [31] and the eect
of glucose concentration level on insulin secretion time delay
1
due to the complex
electro-chemical reactions when the rising glucose concentration level triggers the
cells to release insulin. The delay
1
can be referred as insulin response time delay. The
two time delay DDE model we propose is as follows.
_

_
dG(t)
dt
= G
in
f
2
(G(t)) f
3
(G(t))f
4
(I(t)) + f
5
(I(t
2
)),
dI(t)
dt
= f
1
(G(t
1
)) d
i
I(t),
(2.3.1)
where the initial condition I(0) = I
0
> 0, G(0) = G
0
> 0, G(t) G
0
for all t [
1
, 0]
and I(t) I
0
for t [
2
, 0] with
1
,
2
> 0. In addition,
35
G
l
u
c
o
s
e

u
t
i
l
i
z
a
t
i
o
n
G
l
u
c
o
s
e

p
r
o
d
u
c
t
i
o
n
Liver converts
glucagon and
glycogen to
glucose
I
n
s
u
l
i
n

p
r
o
d
u
c
t
i
o
n
I
n
s
u
l
i
n

c
l
e
a
r
a
n
c
e
Insulin independent:
brain cells, and
others
Insulin dependent:
fat cells, and
others
Insulin degradation:
receptor, enzyme, and
others
Delay
Delay
Glucose Infusion:
meal ingenstion,
oral intake,
enteral nutrition,
constant infusion
Glucagon
secrete
Glucose Controls
insulin secretion
Glucose Controls
glucagon secretion
Insulin helps cells consume glucose
Insulin secretion
Insulin Controls
Hepatic
glucose production
Glucose
Insulin
Pancreas
Liver
cells -cells
Figure 2.3.1. Two Time Delay Glucose-Insulin Regulatory Model
The divide lines (dash-dot-dot) indicate insulin controlled hepatic glucose production with time delay;
the dash-dot lines indicate the insulin secretion from the -cells stimulated by elevated glucose concen-
tration level with time delay; the dashed lines indicate low glucose concentration level triggers -cells
in pancreas to release glucagon; and the dot line indicates the insulin accelerates glucose utilization in
cells.
36
(i) G
in
is due to glucose infusion, e.g., by meal ingestion, oral glucose intake, contin-
uous enteral nutrition or intravenous glucose infusion;
(ii) f
2
(G(t)) stands for insulin independent glucose consumption by the brain, nerve
cells and others. f
2
(0) = 0, f
2
(x) > 0 and f

2
(x) > 0 are bounded for x > 0.
Denote M
2
:= sup{f
2
(x) : x 0} < and M

2
:= sup{f

2
(x) : x > 0} < .
(iii) f
3
(G(t))f
4
(I(t)) stands for insulin dependent utilization/uptake by muscle, fat
cells and others. f
3
(x) = k
3
x, where k
3
> 0 is a constant. f
4
(0) > 0, for
x > 0, f
4
(x) > 0 and f

4
(x) > 0 are bounded above. f
4
(I(t)) is in sigmoidal
shape. Denote M

3
:= sup{f

3
(x) : x > 0} < , m
4
:= inf{f
4
(x) : x 0} > 0,
M
4
:= sup{f
4
(x) : x 0} < , and M

4
:= sup{f

4
(x) : x > 0} < .
(iv) f
5
(I(t
2
)) indicates hepatic glucose production that is dependent on insulin in
the plasma with time delay
2
> 0. The time delay
2
> 0 reects that the liver
does not convert the stored glucose and glycogen into glucose immediate when the
insulin concentration level decreases. When insulin concentration level increases,
the liver converts glucagon and glycogen to glucose decreasingly. f
5
(0) > 0 and,
for x > 0, f
5
(x) > 0 and f

5
(x) < 0. f
5
(x) and |f

5
(x)| are bounded above for x > 0.
Denote M
5
:= sup{f
5
(x) : x 0} < and M

5
:= sup{|f

5
(x)| : x > 0} < .
f
5
(x) is in an inverse sigmoidal shape.
(v) f
1
(G(t
1
)) stands for insulin secretion from the pancreas. Insulin is stored in
-cell granules. Glucose is the primary stimuli of insulin secretion from cells.
The delay is due to the complex electric processes inside of a islet. These processes
include that glucose molecules enter islets through GLUT2, elevate ATP and then
close the K
+
channels. When K
+
channels are closed, Ca
2+
channels are open.
37
The inux of Ca
2+
ions causes cell granules to secrete insulin. f
1
(0) > 0 and, for
x > 0, f
1
(x) > 0, f

1
(x) > 0, f

1
(x) > 0 and bounded. Denote M
1
:= sup{f
1
(x) :
x 0} < and M

1
:= sup{f

1
(x) : x > 0} < . f
1
(x) is in sigmoidal shape.
(vi) d
i
I(t) stands for insulin degradation and constant d
i
is the degradation rate.
Insulin is cleared mainly by the liver and kidney ([33]). Insulin is degraded by
enzymes in the subcutaneous tissue ([64]) and interstitial uid as well ([7]).
We will study this model analytically and numerically.
4. Preliminaries
We will give some preliminary analysis in this section. First we illustrate the
uniqueness of the steady state of the model (2.3.1).
Proposition 2.4.1 The Model (2.3.1) has unique positive steady state (G

, I

), where
G

is the unique solution of equation

H(x) = G
in
f
2
(x) f
3
(x)f
4
(d
1
i
f
1
(x)) + f
5
(d
1
i
f
1
(x)) = 0, x > 0, (2.4.1)
and
I

= d
1
i
f
1
(G

). (2.4.2)
Proof All we have to show is that equation (2.4.1) has a unique root in (0, ). In fact,
observe that f

1
(x) > 0, f

2
(x) > 0, f

4
(x) > 0, f

3
(x) > 0, and f

5
(x) < 0, then H

(x) < 0.
Notice that
H(0) = G
in
f
2
(0) f
3
(0)f
4
(d
1
i
f
1
(0)) +f
5
(d
1
i
f
1
(0))
= G
in
+ f
5
(d
1
i
f
1
(0)) > 0,
38
and
lim
x
H(x) = G
in
lim
x
f
2
(x) lim
x
f
3
(x)f
4
(d
1
i
lim
x
f
1
(x))
+f
5
(d
1
i
lim
x
f
1
(x))
= G
in
M
2
f
4
(d
1
i
M
1
) lim
x
(k
3
x) + f
5
(d
1
i
M
1
)
< 0.
In addition, f
1
(x) is strictly monotone increasing, so the proof is completed.
We show the positiveness and boundedness of the solutions of the model (2.3.1).
Proposition 2.4.2 All solutions of model (2.3.1) exist for all t > 0, are positive and
bounded. Furthermore,
limsup
t
G(t) M
G
:=
G
in
+ M
5
m
4
k
3
(2.4.3)
and
limsup
t
I(t) M
I
:= d
1
i
f
1
(M
G
). (2.4.4)
Proof. Observe that the |f

i
(x)|, i = 1, 2, 3, 4, 5, are bounded, thus f
i
(x), i = 2, 3, 4,
and f
j
(x
t
), j = 1, 5, are Lipschitz and completely continuous in x 0 and x
t

C[max{
1
,
2
}, 0], respectively. Then by Theorem 2.1, 2.2 and 2.4 on page 19 and 20
in [54], the solution of equation (2.3.1) with given initial condition exists and unique
for all t 0. If there exists a t
0
> 0 such that G(t
0
) = 0 and G(t) > 0, for 0 < t < t
0
,
then G

(t
0
) 0. So
0 G

(t
0
)
= G
in
f
2
(G(t
0
)) f
3
(G(t
0
))f
4
(I(t
0
)) + f
5
(I(t
0

2
))
= G
in
f
2
(0) f
3
(0)f
4
(I(t
0
)) + f
5
(I(t
0

2
))
= G
in
+ f
5
(I(t
2
)) > 0
39
This implies that G(t) > 0, for all t > 0. If t

0
> 0 such that I(t

0
) = 0 and I(t) > 0
for all 0 < t < t

0
, then I(t

0
) < 0. Therefore, 0 > I(t

0
) = f
1
(G(t

0
) d
i
I(t

0

1
)
f
1
(G(t

0
)) > 0 implies that I(t) > 0 for all t > 0.
Notice that m
4
f
4
(x) M
4
and f
5
(x) M
5
and f
3
(x) = k
3
x, for x > 0. Thus
G

(t) = G
in
f
2
(G(t)) f
3
(G(t))f
4
(I(t)) + f
5
(I(t
2
))
G
in
m
4
k
3
G(t) + M
5
.
Therefore, for any given

t > 0, if t >

t, we have
d
dt
(e
m
4
k
3
t
G(t)) (G
in
+ M
5
)e
m
4
k
3
t
e
m
4
k
3
t
G(t) G(

t) +
_
t

t
(G
in
+ M
5
)e
m
4
k
3
s
ds
G(t) G(

t)e
m
4
k
3
t
+
_
t

t
e
m
4
k
3
s
ds
= G(

t)e
m
4
k
3
t
+
G
in
+ M
5
m
4
k
3
(e
m
4
k
3

t
e
m
4
k
3
t
)
Thus
limsup
t
G(t)
G
in
+ M
5
m
4
k
3
:= M
G
Since |f
1
(x)| M
1
, given > 0, I

(t) f
1
(M
G
+) d
i
I(t) for suciently large t > 0.
Then we have
limsup
t
I(t) d
1
i
f
1
(M
G
+).
Notice that > 0 is arbitrary, so
limsup
t
I(t) d
1
i
f
1
(M
G
) := M
I
.
The following lemma is elementary. See [48] for a proof.
40
Lemma A Let f : R R be a dierentiable function. If l = liminf
t
f(t) <
limsup
t
f(t) = L, then there are sequences {t
k
} , {s
k
} such that for all
k, f

(t
k
) = f

(s
k
) = 0, lim
k
f(t
k
) = L and lim
k
f(s
k
) = l.
We will apply Lemma A in follows and prove a few preliminary results. Let
(G(t), I(t)) be a solution of (2.3.1). Throughout this paper, we dene
G = limsup
t
G(t), G = liminf
t
G(t)
and
I = limsup
t
I(t), I = liminf
t
I(t).
Due to the Proposition 2.4.1 and 2.4.2, we see that these limits are nite.
The following lemma regarding the upper limits and lower limits of a solution of
the model (2.3.1).
Lemma 2.4.1 If (G(t), I(t)) is a solution of (2.3.1), then
f
1
(G) d
i
I d
i

I f
1
(

G), (2.4.5)
f
2
(

G) + f
3
(

G)f
4
(I) G
in
+ f
5
(I), (2.4.6)
G
in
+ f
5
(

I) f
2
(G) + f
3
(G)f
4
(

I), (2.4.7)
Proof. First we show (2.4.5) holds. Due to Fluctuation Lemma and Proposition 2.4.2,
there exist sequences {t
k
} , such that I

(t
k
) = 0, lim
k
I(t
k
) = I. Thus,
0 = I

(t
k
) = f
1
(G(t
k

1
)) d
i
I(t
k
) for all k.
Therefore,
f
1
(G) d
i
I(t
k
) f
1
(G(t
k

1
)) d
i
I(t
k
) for k = 1, 2, 3, ...
41
Thus,
f
1
(G) d
i
I 0.
On the other hand side, there exists a sequence {s
k
} such that lim
k
I(s
k
) =
I and I

(s
k
) = 0 for all k > 0. So,
f
1
(G) d
i
I(s
k
) f
1
(G(s
k

1
)) d
i
I(s
k
) for k = 1, 2, 3, ...
Thus,
f
1
(G) d
i
I 0.
Now we show (2.4.6) holds. Again, due to Proposition 2.4.2 and Fluctuation
Lemma, there exists a sequence {t

k
} as k such that lim
k
G(t

k
) = G and
0 = G

(t

k
)
= G
in
f
2
(G(t

k
)) f
3
(G(t

k
))f
4
(I(t

k
)) + f
5
(I(t

k

2
)), k = 1, 2, 3, ....
Then, notice that f
4
and f
5
0,
0 = G
in
f
2
(G(t

k
)) f
3
(G(t

k
))f
4
(I(t

k
)) + f
5
(I(t

k

2
))
G
in
f
2
(G(t

k
)) f
3
(G(t

k
))f
4
(I) + f
5
(I), k = 1, 2, 3, ...
and therefore
G
in
f
2
(G) f
3
(G)f
4
(I) + f
5
(I) 0.
Similarly we can show (2.4.7) is true. According to Proposition 2.4.2 and Fluctu-
ation Lemma, there exists a sequence {s

k
} as k such that lim
k
G(s

k
) = G
and
0 = G

(s

k
)
= G
in
f
2
(G(s

k
)) f
3
(G(s

k
))f
4
(I(s

k
)) + f
5
(I(s

k

2
)), k = 1, 2, 3, ....
42
Then, notice that f
4
and f
5
0,
0 = G
in
f
2
(G(s

k
)) f
3
(G(s

k
))f
4
(I(s

k
)) + f
5
(I(s

k

2
))
G
in
f
2
(G(s

k
)) f
3
(G(s

k
))f
4
(I) + f
5
(I), k = 1, 2, 3, ....
Thus,
0 G
in
f
2
(G) f
3
(G)f
4
(I) +f
5
(I).
Apparently,

G = G implies

I = I due to (2.4.5). If

I = I, then (2.4.6) and
(2.4.7) together lead to f
2
(

G) f
2
(G) f
4
(

I)(f
3
(G) f
2
(

G)) 0. That is,

G = G.
This complete the proof of following
Theorem 2.4.1 Let (G(t), I(t)) be a solution of the Model (2.3.1). Then

G = G and

I = I imply each other.

Following proposition proves the model (2.3.1) is persistent.
Proposition 2.4.3 Model (2.3.1) is persistent, that is, all solutions of Model (2.3.1)
are bounded by a pair of positive constants from above and below, respectively.
Proof. Notice that f
2
(0) + f
3
(0) = 0 and f
4
(x) < M
4
for all x 0. Then (2.4.7)
implies that
G
in
f
2
(G) + f
3
(G)M
4
, for all t > 0. (2.4.8)
Thus
G
> 0, t
G
> 0, such that G(t) >
G
for t > t
G
> 0. Therefore (2.4.5)
implies that I(t) is bounded below.
43
On the other hand side, (2.4.3) and (2.4.5) imply that I(t) and G(t) are bounded
above.
5. Global Stability of Steady State
In this section, we will give one result of globally asymptotically stable equilib-
rium of this model using Lemma 2.4.1.
Theorem 2.5.1 Let
F(x, y) = f
3
(x)f
4
(d
1
i
f
1
(y)) + f
5
(d
1
i
f
1
(x)), x, y 0. (2.5.1)
If
F(x, y) F(y, x), x y 0, (2.5.2)
then the steady state (G

, I

) of (2.3.1) is globally asymptotically stable.

Proof Let (G(t), I(t)) be a solution of (2.3.1). Due to Lemma 2.4.1, we have
G
in
f
2
(G) f
3
(G)f
4
(I) + f
5
(I) G
in
f
2
(G) f
3
(G)f
4
(I) + f
5
(I)
that is,
0 [f
2
(G) + f
3
(G)f
4
(I) f
5
(I)] [f
2
(G) + f
3
(G)f
4
(I) f
5
(I)]
= [f
2
(G) + f
3
(G)f
4
(I) +f
5
(I)] [f
2
(G) + f
3
(G)f
4
(I) + f
5
(I)]
[f
2
(G) f
2
(G)] + [(f
3
(G)f
4
(d
1
i
f
1
(G)) + f
5
(d
1
i
f
1
(G)))
(f
3
(G)f
4
(d
1
i
f
1
(G)) + f
5
(d
1
i
f
1
(G)))]
= [f
2
(G) f
2
(G)] + [F(G, G) F(G, G)]
f
2
(G) f
2
(G)
44
due to (2.5.2). Thus G = G.
Remark Notice that f
5
(d
1
i
f
1
(x)) f
5
(d
1
i
f
1
(y)) for x y 0 means higher hep-
atic production of glucose helps to make oscillations happen (the case that (G

, I

) is
unstable).
Remark Notice that f
3
(G) can be linear and f
4
is bounded. If the glucose concentra-
tion G is big enough and there is no hepatic production (f
5
0), then the steady state
(G

, I

) will be globally stable and thus there is no oscillation.

6. Linearization and Local Analysis
We need following theorem for two special cases, where one of the two time
delays equals to zero. When both delays equal to zero, the linearized system of the
model (2.3.1) becomes a trivial 2-dimensional ODE. Now we state theorem here without
proof. For a proof, see Kuang ([54], 1993)(Theorem 3.1, page 77).
Theorem B In the following second order real scalar linear neutral delay equation
x

(t) + x

(t ) + ax

(t) + bx

(t ) +cx(t) + dx(t ) = 0, (2.6.1)

where 0. Assume || < 1, c + d = 0 and a
2
+ b
2
+ (d c)
2
= 0. Consider the
characteristic equation of (2.6.2)

2
+
2
e

+ a + be

+ c + de

= 0. (2.6.2)
The number of dierent imaginary roots with positive (negative) imaginary parts of
(2.6.2) can be zero, one, or two only.
(I) If there are no such roots, then the stability of the zero solution does not
change for any > 0.
45
(II) If there are any imaginary roots with positive imaginary part, an unstable
zero solution never becomes stable for any 0. If the zero solution is asymptotically
stable for = 0, then it is asymptotically stable for <
0
, and it becomes unstable
for >
0
where
0
> 0 is a constant. It undergoes a supercritical Hopf bifurcation at
=
0
.
(III) If there are two imaginary roots with positive imaginary part, i
+
and i

,
such that
+
>

> 0, then the stability of the zero solution can change (when changes
from stable to unstable, the zero solution undergoes a supercritical Hopf bifurcation) a
nite number of times at most as is increased, and eventually it becomes unstable.
The number of such roots are determined by the following conditions.
If c
2
d
2
, then there is only one such root.
If c
2
> d
2
, then there are two such roots provided that
(A) b
2
+ 2c a
2
2d > 0, and
(B) (b
2
+ 2c a
2
2d)
2
> 4(1
2
)(c
2
d
2
).
Otherwise, there is no such solution.
Now we try to linearize the model (2.3.1). Let G(t) = G
1
(t) + G

and I(t) =
I
1
(t) + I

. Then system (2.3.1) becomes

G

1
(t) = G
in
f
2
(G
1
(t) + G

) f
3
(G
1
(t) + G

)f
4
(I
1
(t) + I

) +f
5
(I
1
(t
2
) + I

)
= [f

2
(G

) + f

3
(G

)f
4
(I

)]G
1
(t) f
3
(G

)f

4
(I

)I
1
(t) + f

5
(I

)I
1
(t
2
)
I

1
(t) = f
1
(G
1
(t
1
) + G

) d
i
(I
1
(t) + I

)
= f

1
(G

)G
1
(t
1
) d
i
I
1
(t).
We still use G(t) and I(t) to denote G
1
(t) and I
1
(t), respectively. Thus the linearized
46
system of (2.3.1) can be written as
_

_
dG(t)
dt
= AG(t) BI(t) CI(t
2
)
dI(t)
dt
= DG(t
1
) d
i
I
1
(t)
(2.6.3)
where
A := f

2
(G

) + f

3
(G

)f
4
(I

) > 0,
B := f
3
(G

)f

4
(I

) > 0,
C := f

5
(I

) > 0,
D := f

1
(G

) > 0.
Let
_
_
_
_
_
G(t)
I(t)
_
_
_
_
_
= e
t
_
_
_
_
_
G
0
I
0
_
_
_
_
_
, G
0
, I
0
> 0, C, t > 0
be a solution of (2.6.3). Then
e
t
_
_
_
_
_
G
0
I
0
_
_
_
_
_
=
_
_
_
_
_
AG
0
e
t
BI
0
e
t
CI
0
e
(t
2
)
DG
0
e
(t
1
)
d
i
I
0
e
t
_
_
_
_
_
= e
t
_
_
_
_
_
A B Ce

2
De

1
d
i
_
_
_
_
_
_
_
_
_
_
G
0
I
0
_
_
_
_
_
.
So the characteristic equation of (2.6.3) is given as
det
_
_
_
_
_
_
_
_
_
_
0
0
_
_
_
_
_

_
_
_
_
_
A B Ce

2
De

1
d
i
_
_
_
_
_
_
_
_
_
_
= det
_
_
_
_
_
+ A B + Ce

2
De

1
+ d
i
_
_
_
_
_
47
= ( +A)( + d
i
) + De

1
(B + Ce

2
)
=
2
+ (A + d
i
) +d
i
A + DBe

1
+ DCe
(
1
+
2
)
= 0.
We denote the characteristic equation as
() =
2
+ (A + d
i
) + d
i
A + DBe

1
+ DCe
(
1
+
2
)
= 0. (2.6.4)
Note (0) = d
i
A + DB + DC > 0. So = 0 is not a solution of the characteristic
equation (2.6.4). So if there is any stability switch of the trivial solution of the linearized
system (2.6.3), there must exist a pair of pure imaginary roots of the characteristic
equation (2.6.4).
If
1
= 0 and
2
= 0, the original model (2.3.1) is an ODE model. The charac-
teristic equation of its linearized equation is given by
() =
2
+ (A + d
i
) + d
i
A + DB + DC = 0.
Then due to A + d
i
> 0 and d
i
A +DB + DC > 0, the steady state (G

, I

) is stable.
If
2
= 0 but
1
> 0, the characteristic equation of the linearized system takes
the following form.
() =
2
+ (A +d
i
) + d
i
A + (DB + DC)e

1
= 0. (2.6.5)
Then due to Theorem B ([54]), d
i

D(B+C)
A
means there exists only one positive root
of (2.6.5). That is, there exists an
10
> 0 such that the trivial solution of the linearized
system (2.6.3) is stable when
1
(0,
10
) and unstable when
1

10
.
Similarly if
1
= 0 and
2
> 0, then d
2
i
2DB A
2
implies the trivial solution
of the linearized system (2.6.3) is stable. If d
2
i
< 2DB A
2
and 2DB + D
2
C
2
>
A
2
+d
2
i
+ (d
i
A+DB)
2
, then the trivial solution of the linearized system (2.6.3) has at
most nite number of stability switches and eventually is unstable.
48
Now assume both
1
> 0 and
2
> 0. Let = i, > 0, be such an eigenvalue
in (2.6.4), then we have
(i) =
2
+ (A + d
i
)i + d
i
A + DB(cos
1
i sin
1
) + DC(cos (
1
+
2
) i sin (
1
+
2
))
= [
2
+d
i
A + DBcos
1
+ DC cos (
1
+
2
)]
+i[(A + d
i
) DBsin
1
DC sin (
1
+
2
)]
= 0.
That is,
_

2
+ d
i
A +DBcos
1
+ DC cos (
1
+
2
) = 0,
(A + d
i
) DBsin
1
DC sin (
1
+
2
) = 0.
(2.6.6)
This leads to
(
2
+ d
i
A)
2
+ ((A + d
i
))
2
= (DBcos
1
+ DC cos (
1
+
2
))
2
+ (DBsin
1
+ DC sin (
1
+
2
))
2
,
that is,

4
2d
i
A
2
+ d
2
i
A
2
+ (A
2
+ 2d
i
A + d
2
i
)
2
= D
2
(B
2
+ C
2
+ 2BC cos
2
) D
2
(B
2
+ C
2
+ 2BC) = D
2
(B + C)
2
.
So

4
+ (A
2
+ d
2
i
)
2
+ [d
2
i
A
2
D
2
(B + C)
2
] 0.
This is impossible in the case that d
i
A D(B+C). We summarize the results obtained
in this section as follows.
Proposition 2.6.1 In the linearized system (2.6.3),
49
(a) When
1
= 0 and
2
= 0, the steady state of (2.6.3) is a stable spiral point.
(b) When
1
> 0 and
2
= 0, if d
i
D(B+C)/A, then there exists a stability switch,
i.e., there exists an
10
> 0 such that the trivial solution of the linearized system
(2.6.3) is stable when
1
(0,
10
) and unstable when
1

10
.
(c) When
1
= 0 and
2
> 0,
(c.1) if d
2
i
2DB A
2
, then the trivial solution of the linearized system (2.6.3)
is stable.
(c.2) If d
2
i
< 2DB A
2
and 2DB + D
2
C
2
> A
2
+ d
2
i
+ (d
i
A + DB)
2
, then the
trivial solution of the linearized system (2.6.3) has at most nite number of
stability switches and eventually is unstable.
(d) When
1
> 0 and
2
> 0, if d
i
D(B + C)/A, then the steady state of the
linearized system (2.6.3) is stable.
Thus Proposition 2.6.1 leads to the following trivial result for the model (2.3.1).
Theorem 2.6.1 In Model (2.3.1),
(a) when
1
= 0 and
2
= 0, the steady state (G

, I

) is a stable spiral point.

(b) When
1
> 0 and
2
= 0, if d
i
f
1
(G

)(Bf

5
(I

))/(f

2
(G

) +f

3
(G

)f
4
(I

)), then
there exists a stability switch, i.e., there exists an
10
> 0 such that the trivial
solution of the steady state (G

, I

) is stable when
1
(0,
10
) and unstable when

1

10
.
(c) When
1
= 0 and
2
> 0,
50
(c.1) if d
2
i
2f

1
(G

)B(f

2
(G

) +f

3
(G

)f
4
(I

))
2
, then the steady state (G

, I

)
is stable.
(c.2) if
d
2
i
< 2f
1
(G

)f
3
(G

)f

4
(I

) (f

2
(G

) + f

3
(G

)f
4
(I

))
2
,
and
2f
1
(G

)f
3
(G

)f

4
(I

) + (f
1
(G

)f

5
(I

))
2
> (f

2
(G

) +f

3
(G

)f
4
(I

))
2
+ d
2
i
+
+(d
i
(f

2
(G

) +f

3
(G

)f
4
(I

)) + f

1
(G

)(f
3
(G

)f

4
(I

)))
2
then there are at most a nite number of stability switch and eventually steady
state (G

, I

) is unstable.
(d) When
1
> 0 and
2
> 0, if the insulin degradation rate
d
i

f

1
(G

)(f
3
(G

)f

4
(I

) f

5
(I

))
f

2
(G

) + f

3
(G

)f
4
(I

)
:= d
0
, (2.6.7)
the steady state (G

, I

) is stable.
Remark If the parameters and functions f
i
, i = 1, 2, 3, 4, 5, take the values in (2.7.1)
to (2.7.5) and Table (2.2.1) and (2.2.2), then the threshold value d
0
= 0.6669 when
G
in
= 0.54. So when d
i
= 1/26 = 0.03849 < d
0
. So, (2.6.7) does not hold. In fact, the
insulin and glucose oscillation is sustained provided that
2
= 36 and
1
is suciently
large (greater than 5.2).
To further analyze the stability of the steady state of the model (2.3.1) and the
cases of the oscillations to be sustained, we will apply Rouch` e

s Theorem to analyze
51
when
1
> 0 and
2
> 0 that the steady state (G

, I

) is unstable. Recall following

Rouch` e

s Theorem ([19], p.125-126).

Rouch` es Theorem Given two functions f(z) and g(z) analytic in a simple connected
region A C with boundary , a simple loop homotopic to a point in A. If |f(z)| >
|g(z)| on , then f(z) and f(z) +g(z) have the same number of roots in A.
We start from a more generic equation and leave the system (2.6.3) as a special
case.
Let
S
1
= {
2m
2n 1
: m, n Z
+
, m, n 1}
and
S
2
= {
2m1
2n
: m, n Z
+
, m, n 1}.
Clearly Q
+
= S
1
S
2
and S
1
S
2
= . Further we have
Lemma 2.6.1 S
1
and S
2
are dense in Q
+
thus in R
+
.
Proof. r Q
+
\ S
1
, p, q Z
+
such that r =
2p1
2q
. Thus
r
k
=
2p 1
2
2k
2q
1
2k
=
(4kp 2k 2)/2k
(4kq 1)/2k
=
2(2kp 2k 1)
2(2kq) 1
S
1
k = 1, 2, 3, ...
and lim
k
r
k
= (2p 1)/2q = r. That is, S
1
= Q
+
. Similarly, S
2
= Q
+
.
Proposition 2.6.2 For characteristic equation

k
+
k1

j=1
a
j

j
+ b + ce

1
+de

2
= 0, k 2,
1
,
2
> 0, (2.6.8)
where b, c, d > 0, a
j
R, j = 1, 2, 3, ..., k,, if b < d c or b < c d, then
10
> 0 and

20
> 0 such that the characteristic equation (2.6.8) has at least one root with positive
real part for
1
>
10
and
2
>
20
and
1
/
2
S
1
or
1
/
2
S
2
.
52
We need following lemmas to prove Proposition 2.6.2.
Lemma 2.6.2 For the equation

k
z
k
+
k1

j=1
a
j

j
z
j
+ b + ce
p
1
z
+ de
p
2
z
= 0, k 2, p
1
, p
2
> 0, z C (2.6.9)
where b, c, d > 0, a
j
R, j = 1, 2, 3, ..., k, assume
(i) b < d c, and p
1
/p
2
S
1
, or
(ii) b < c d, and p
1
/p
2
S
2
.
Then,
0
> 0 such that for all , 0 < <
0
, the equation (2.6.9) has at least one root
with positive real part.
Proof. Let
f(z) = b + ce
p
1
z
+ de
p
2
z
.
We show that f(z) has a zero with positive real part. Since p
1
and p
2
are S
1
related
in case (i) or S
2
related in case (ii), there exist integer m, n 1 such that
p
1
p
2
=
2m
2n1
for case (i), or
p
1
p
2
=
2m1
2n
for case (ii). Let z = x + qi, where q = 2m/p
1
= (2n
1)/p
2
for case (i) or q = (2m1)/p
1
= 2n/p
2
for case (ii). Then
f(z) = b + ce
p
1
x
e
p
1
qi
+de
p
2
x
e
p
2
qi
= b + ce
p
1
x
cos p
1
q + de
p
2
x
cos p
2
q i(ce
p
1
x
sin p
1
q +de
p
2
x
sin p
2
q)
= b + ce
p
1
x
cos 2m + de
p
2
x
cos (2n 1)
(= b + ce
p
1
x
cos (2m1) + de
p
2
x
cos 2n for case (ii))
= b + ce
p
1
x
de
p
2
x
(= b ce
p
1
x
+ de
p
2
x
for case (ii))
:= H(x).
53
Notice that H(0) = b+cd < 0 (H(0) = bc+d < 0 for case (ii)) and lim
x
H(x) =
b > 0, therefore H(x) has at least one zero x
0
(0, ). So f(z) has at least one zero
z
0
= x
0
+ qi with x
0
> 0.
We perturb f(z) by g

(z) given by
g

(z) =
k
z
k
+
k1

j=1
a
j

j
z
j
, > 0, (2.6.10)
with small > 0 and show that f(z) + g

(z) has the same number of zeros as f(z) if

is small. To this end, we rst construct a simple loop homotopic to a point and then
show |f(z)| > |g

(z)| on .
Let z = x, x (, ), then
|f(z)| = b + ce
p
1
x
+ de
p
2
x
> b.
Let z = x + 2qi, x (, ), then
|f(z)| = |b + ce
p
1
x
e
2qp
1
i
+ de
p
2
x
e
2qp
2
i
|
= |b + ce
p
1
x
cos 2qp
1
+de
p
2
x
cos 2qp
2

i(ce
p
1
x
sin 2qp
1
+ de
p
2
x
sin 2qp
2
)|
= |b + ce
p
1
x
cos 4m + de
p
2
x
cos 2(2n 1)|
(= |b + ce
p
1
x
cos 2(2m1) + de
p
2
x
cos 4n| for case (ii))
= b + ce
p
1
x
+de
p
2
x
> b.
Let z = Kx
0
+ yi, y [0, 2q], where K > 1 such that b ce
p
1
Kx
0
de
p
2
Kx
0
> b/2.
Then
|f(z)| = |b + ce
p
1
Kx
0
e
p
1
yi
+ de
p
2
Kx
0
e
p
2
yi
|
b ce
p
1
Kx
0
de
p
2
Kx
0
> b/2.
54
Let z = yi, y [0, 2q], then
|f(z)| = |b + ce
p
1
yi
+ de
p
2
yi
|

_
d c b, for case (i),
c d b, for case (ii)
:=
0
> 0.
Let

0
:= min{
0
, b/2}. Denote
:= {z = x + yi C : z = x or z = x 2qi, x [0, Kx
0
]
or z = yi or z = Kx
0
+ yi y [0, 2q].

:= {z = x + yi C : 0 < x < Kx
0
, 0 < y < 2q}.
Clearly, is a simple loop homotopic to the original, z
0
= x
0
+ qi

and |f(z)| >

0
on . Choose r
0
> 0 such that A := {z C : |z| < r
0
}. Denote A := {z C :
|z| = r
0
}. Thus z A, z = r
0
e
i
, [0, 2], we have
|g

(z)| = |
k
z
k
+
k1

j=1
a
j

j
z
j
|
k
r
k
0
+
k1

j=1
|a
j
|
j
r
j
0
. (2.6.11)
Obviously
0
> 0 such that , 0 < <
0
,
|g

(z)| <

0
, z A.
z A, z = re
i
, then r < r
0
, and
|g

(z)| = |
k
z
k
+
k1

j=1
a
j

j
z
j
|
k
r
k
+
k1

j=1
|a
j
|
j
r
j
<
k
r
k
0
+
k1

j=1
|a
j
|
j
r
j
0
.
Thus
|g

(z)| <

0
for all z .
55
Therefore |f(z)| > |g

(z)| on . By Rouch` e

s Theorem ([19], p125-126), f(z) and

f(z) +g

(z) have the same number of zeros in

. That is, f(z) +g

(z) = 0 has at least

one root z

.
Proof of Proposition 2.6.2. Assume b < d c, and
1
/
2
S
1
(or b < c d,
and
1
/
2
S
2
). In Lemma 2.6.2, choose p
10
and p
20
such that p
10
/p
20
S
1
(or
p
10
/p
20
S
2
). Suppose
0
is given by (2.6.11) in the proof of Lemma 2.6.2. Let
10
=
p
10
/
0
and
20
= p
20
/
0
. Then
1
>
10
,
2
>
20
and
1
/
2
S
1
(or
1
/
2
S
2
), ,
0 < <
0
such that

1
= p
1
/ >
10
and
2
= p
2
/ >
20
.
Let = z. Then (2.6.8) becomes (2.6.9) in Lemma 2.6.2 and thus the conclusion
follows.
Remark In Lemma 2.6.2, given p
1
and p
2
that are S
1
or S
2
related, if we carefully
choose
0
in the proof of Lemma 2.6.2, an estimate of unstable region of
1
and
2
can
be given. For the special case k = 2, r
0
and
0
can be chosen as
r
0
=
_
K
2
x
2
0
+ q
2

2
and
0
= (
_
a
2
1
+ 4

0
a
1
)/2r
0
.
Let k = 2 and we apply the Proposition 2.6.2 to the linearized system (2.6.3) and we
have
Proposition 2.6.3 If d
i
A < D|C B|, then there exist
10
> 0 and
20
> 0 such that
the characteristic equation of the system (2.6.3) has at least one root with positive real
part if
(i) d
i
A < D(C B),
1
>
10
,
1
+
2
>
20
and
1
/(
1
+
2
) S
1
, or
(ii) d
i
A < D(B C),
1
>
10
,
1
+
2
>
20
and
1
/(
1
+
2
) S
2
56
Proof. This is straight forward if in Proposition 2.6.2 choose k = 2, a
1
= A+d
i
, b =
d
i
A, c = DB, d = DC,
1
=
1
,
2
=
1
+
2
.
Therefore, we have
Theorem 2.6.2 In model (2.3.1), if
d
i
<
f

1
(G

)|f
3
(G

)f

4
(I

) + f

5
(I

)|
f

2
(G

) +f

3
(G

)f
4
(I

)
(2.6.12)
then there exist
10
> 0 and
20
> 0 such that if
1
>
10
and
2
>
20
, the steady state
(G

, I

) is unstable if
(i) d
i
<
f

1
(G

)(f
3
(G

)f

4
(I

)+f

5
(I

))
f

2
(G

)+f

3
(G

)f
4
(I

)
,
1
>
10
,
1
+
2
>
20
and
1
/(
1
+
2
) S
1
, or
(ii) d
i
<
f

1
(G

)(f
3
(G

)f

4
(I

)+f

5
(I

))
f

2
(G

)+f

3
(G

)f
4
(I

)
,
1
>
10
,
1
+
2
>
20
and
1
/(
1
+
2
) S
2
Remark Theorem 2.6.2 indicates the insulin concentration oscillation in Model (2.3.1)
is sustained when the (2.6.12) holds. In next section, intensive simulations are performed
with the functions given in (2.2.2) to (2.2.6). When the d
i
= 0.03849 and G
in
= 0.54,
(2.6.12) does hold. If
1
= 6 and
2
[25, 45], the insulin concentration oscillation
is sustained (see Figure 2.7.7). When the d
i
= 0.03849 and G
in
= 2.54, (2.6.12) also
holds. But, in this case, the delay parameters are required to be very large, for example,

1
= 70 and
2
= 90, and the steady state is unstable.
7. Numerical Analysis of Stability Switches and Bifurcations
In this section, we further explore numerical analysis on a group of particular
functions and parameters (Table 2.7.1) developed from experiments ([79] and [84]) and
also used in [31] and [4]. The functions f
i
, i = 1, 2, 3, 4, 5, take the following forms
57
Table 2.7.1. Parameters of the functions in Two Time Delay Model (2.3.1).
Parameters Units Values
V
g
l 10
R
m
Umin
1
210
a
1
mg l
1
300
C
1
mg l
1
2000
U
b
mg min
1
72
C
2
mg l
1
144
C
3
mg l
1
1000
Parameters Units Values
U
0
mg min
1
40
U
m
mg min
1
940
1.77
C
4
Ul
1
80
R
g
mg min
1
180
lU
1
0.29
a
1
Ul
1
26
f
1
(G) =
0.1R
m
1 + exp((C
1
G/V
g
)/a
1
)
, (2.7.1)
f
2
(G) = 0.1U
b
(1 exp(G/(C
2
V
g
))), (2.7.2)
f
3
(G) =
G
10C
3
V
g
, (2.7.3)
f
4
(I
i
) = 0.1U
0
+
0.1(U
m
U
0
)
1 + exp(ln(I
i
/C
4
(1/V
i
+ 1/Et
i
)))
, (2.7.4)
f
5
(x) =
0.01R
g
1 + exp((x/V
p
C
5
))
, (2.7.5)
and the parameters are dened in table (2.7.1) based on [79] and [84]. The glucose
G and insulin I in Sturis-Tolic models (2.2.1)([79], [84]) and later the models studied
by K. Engelborghs, V. Lemaire, J. Belair and D. Roose ([31]) and D. L. Bennett and
S. A. Gourley ([4]) are in the unit mg and U, respectively. We divide G and I by 100
and 10, respectively, in the Model (2.3.1) (therefore the functions f
i
, i = 1, 2, 3, 4, 5, in
(2.7.1), (2.7.2), (2.7.3), (2.7.4) and (2.7.5)) so that the glucose G is in the unit of mg/dl
58
and insulin I is in the unit of U/ml. But, some of the simulations are done simply
using the original f
i
, i = 1, 2, 3, 4, 5.
We analyze the dynamics of the glucose and insulin numerically in following
cases.
I Take the insulin response time delay
1
as the bifurcation parameter and let other
parameters are xed;
II Take the glucose infusion rate G
in
as the bifurcation parameter and let other
parameters are xed;
III Take the insulin degradation d
i
as the bifurcation parameter and let other param-
eters are xed;
IV Take the hepatic glucose time delay
2
as the bifurcation parameter and let other
parameters are xed;
V Take both insulin response time delay
1
and glucose infusion rate G
in
as bifur-
cation parameters;
VI Take both insulin response time delay
1
and the insulin degradation d
i
as bifur-
cation parameters;
VII Take both the insulin degradation d
i
and the glucose infusion rate G
in
as bifur-
cation parameters.
Let d
i
= 0.03849 be xed, then d
0
= 0.6669 in Theorem 2.6.1. So Theorem 2.6.1
does not apply and lead to a stable steady state (G

, I

).
59
0 2 4 6 8 10 12 14 16 18 20
0
20
40
60
80
100
120
140
tau
1
(min) (Gin = 0.54, tau
2
= 36)
G
l
u
c
o
s
e
(
r
e
d
, m
g
/
d
l
)
, I
n
s
u
l
i
n
(
b
l
u
e
, m
U
/
m
l
)
0
5
10
15
20
20
40
60
80
100
120
140
0
5
10
15
20
25
30
tau
1
(min)
Periodic Solutions in (tau
1
, G, I) when (Gin = 0.54 (mg/dl/min), tau
2
= 36 (min))
G (mg/dl)
I

(
m
U
/
m
l)
Figure 2.7.1. Bifurcation diagram of
1
where
1
[0, 20], d
i
= 0.03849,
2
= 36 and G
in
= 0.54. Left: the upper bifurcation diagram is for
glucose G and the lower one is for insulin I; right: Periodic Solutions.
7.1. Insulin Response Time Delay
1
. In this subsection we take the insulin
response time delay
1
as a bifurcation parameter and study the behavior of the system.
We let parameter
1
change from 0 to 20 and G
in
= 0.54, d
i
= 0.03849 and

2
= 36 are xed. Figures 2.7.1 and 2.7.2 show the stability regions, limit cycles and
period of periodic solutions in this case. To summarize the observations of these three
gures, we have
Numerical Observation 2.7.1 In model (2.3.1), assume d
i
= 0.03849, G
in
= 0.54,

2
= 36 and
1
changes in [0, 20]. Then there exists a
10
, 5.1 <
10
< 5.2, such that
the model (2.3.1) has a stability switch at
10
, that is
(1) The unique steady state (G

, I

) is stable when
1
[0,
10
) and unstable when

1
[
10
, 20].
(2) When
1
[
10
, 20], there exists a periodic solution. While
1
changes from

10
to 20, the amplitudes and periods of periodic solutions increase signicantly.
60
0 5 10 15 20 25
0
20
40
60
80
100
120
140
160
180
200
Period of Periodic Solutions of G and I (overlap)
tau
1
(min) (Gin = 0.54, tau
2
= 36)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
0
50
100
150
200
G
l
u
c
o
s
e
c
o
n
c
e
n
t
r
a
t
i
o
n
(
m
g
/
d
l
)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
4
6
8
10
12
I
n
s
u
l
i
n
c
o
n
c
e
n
t
r
a
t
i
o
n
(
m
U
/
l
)
di (mU/l/min), Gin=54 (mg/dl/min), tau
1
=7.5 (min), tau
2
=36 (min)
Figure 2.7.2. Periods of periodic solutions when
1
[0, 20] and bifurcation diagram of d
i
Left: where d
i
= 0.03849,
2
= 36 and G
in
= 0.54 are xed This curve indicates the period of the
periodic solutions increase signicantly when
1
increases from 0 to 20 (min). Right: bifurcation
diagram of d
i
[0.001, 0.7] (U/(lmin)).
7.2. Glucose Infusion Rate G
in
. We take the glucose infusion rate G
in
as a
bifurcation parameter. We consider the case
1
= 6.0 and the case
1
= 7.5, respectively.
All other parameters are xed.
Let parameter G
in
vary from 0 to 2.16, d
i
= 0.03849,
2
= 36,
1
= 6.0 or

1
= 7.5 are xed. Figures 2.7.3, 2.7.4 and 2.7.5 show the stability regions, limit cycles
and period of periodic solutions in this case. To summarize the observations of these
three gures, we have, for the case
1
= 6.0,
Numerical Observation 2.7.2 In model (2.3.1), assume
1
= 6,
2
= 36 and G
in
changes in [0, 2.16]. Then there exist G
01
in
and G
02
in
, 0.05 < G
01
in
< 0.075, 0.7 < G
02
in
<
0.8, such that the model (2.3.1) has stability switches at G
01
in
and G
02
in
, that is
(1) The unique steady state (G

, I

) is stable when G
in
[0, G
01
in
), unstable when
G
in
[G
01
in
, G
02
in
] and stable when G
in
(G
02
in
, 2.16].
(2) When G
in
[G
01
in
, G
02
in
], there exists a periodic solution. There exists an
a
0
[G
01
in
, G
02
in
], while G
in
changes from G
01
in
to G
02
in
, the amplitudes increase when
61
0 0.5 1 1.5 2 2.5
0
10
20
30
40
50
60
70
80
90
100
Gin (mg/dl/min) ( tau
1
= 6 (min), tau
2
= 36 (min))
G
l
u
c
o
s
e
(
r
e
d
, m
g
/
d
l
)
, I
n
s
u
l
i
n
(
b
l
u
e
, m
U
/
m
l
)
0 0.5 1 1.5 2 2.5
0
10
20
30
40
50
60
70
80
90
100
Gin (mg/dl/min) ( tau
1
= 7.5 (min), tau
2
= 36 (min))
G
l
u
c
o
s
e
(
r
e
d
, m
g
/
d
l
)
, I
n
s
u
l
i
n
(
b
l
u
e
, m
U
/
m
l
)
Figure 2.7.3. Bifurcation diagram of G
in
[0, 2.16]
These two gures indicate when
1
= 6.0, there are two bifurcation points when G
in
changes from 0
to 2.16 (mg/(dlmin), and there is only one bifurcation point when
1
= 7.5. Left:
2
= 36,
1
= 6.0.
Right:
2
= 36,
1
= 7.5 .
G
in
< a
0
and decreases when G
in
> a
0
. The periods of periodic solutions change
insignicantly in the range (102, 108).
For the case
1
= 7.5, we have
Numerical Observation 2.7.3 In model (2.3.1), assume
1
= 7.5,
2
= 36 and G
in
changes in [0, 2.16]. Then there exist G
03
in
, 0.8 < G
03
in
< 1.0 such that the model (2.3.1)
has a stability switch at G
03
in
, that is
(1) The unique steady state (G

, I

) is unstable when G
in
[0, G
03
in
], stable when
G
in
(G
03
in
, 2.16].
(2) When G
in
[0, G
03
in
], there exists a periodic solution. There exists an a

0

[0, G
03
in
], while G
in
changes from 0 to G
03
in
, the amplitudes increase when G
in
< a

0
and
decreases when G
in
> a

0
. The periods of periodic solutions vary insignicantly in the
range (115, 122).
62
0
0.5
1
1.5
2
2.5
60
70
80
90
100
6
8
10
12
14
16
18
20
Gin (mg/dl/min)
Periodic Solutions in (tau
1
, G, I) when (tau
1
= 6 (min), tau
2
= 36 (min))
G (mg/dl)
I
(
m
U
/
m
l
)
0
0.5
1
1.5
2
2.5
40
60
80
100
5
10
15
20
Gin (mg/dl/min)
Periodic Solutions in (tau
1
, G, I) when (tau
1
= 7.5 (min), tau
2
= 36 (min))
G (mg/dl)
I

(
m
U
/
m
l)
Figure 2.7.4. Limit cycles in (G
in
, G, I)-space when G
in
[0, 2.16]
Left:
2
= 36,
1
= 6.0. Right:
2
= 36,
1
= 7.5.
0 0.5 1 1.5 2 2.5
0
20
40
60
80
100
120
Period of Periodic Solutions
Gin (mg/dl/min) ( tau
1
= 6 (min), tau
2
= 36 (min))
G
(
r
e
d
,

m
g
/
d
l
)
,

I
(
b
l
u
e
,

m
U
/
m
l
)
0 0.5 1 1.5 2 2.5
0
20
40
60
80
100
120
140
Period of Periodic Solutions
Gin (mg/dl/min) ( tau
1
= 7.5 (min), tau
2
= 36 (min))
G
(
r
e
d
,

m
g
/
d
l
)
,

I
(
b
l
u
e
,

m
U
/
m
l
)
Figure 2.7.5. Periods of periodic solutions when G
in
[0, 2.16]
These two gures show clearly that the periods of periodic solutions change insignicantly when G
in
varies in [0, 2.16] (mg/(dlmin) and
2
= 36,
1
= 6.0 (left gure) and
1
= 7.5 (right gure).
63
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
0
20
40
60
80
100
120
140
P
e
r
i
o
d

(
m
i
n
)

o
f

p
e
r
i
o
d
i
c

s
o
l
u
t
i
o
n
s
di (mU/l/min), Gin=54 (mg/dl/min), tau
1
=7.5 (min), tau
2
=36 (min)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
0
5
10
15
20
25
30
T
i
m
e

(
m
i
n
)

s
h
i
f
t

b
e
t
w
e
e
n

p
e
a
k
s

o
f

G

a
n
d

I

o
f

p
e
r
i
o
d
i
c

s
o
l
u
t
i
o
n
s
di (mU/l/min), Gin=54 (mg/dl/min), tau
1
=7.5 (min), tau
2
=36 (min)
Figure 2.7.6. Periods and peak time dierences when d
i
changes in [0.001, 0.7]
Left: periods of the periodic solutions decreases along with d
i
increases. The decrease is dramatic when
d
i
is small and insignicant when d
i
is larger. Right: the time shift that insulin peaks after glucose
peaks. The changes along with d
i
is in the pace with the periods.
7.3. Insulin Degradation Rate d
i
. We take the insulin degradation rate d
i
as a bifurcation parameter and let d
i
vary from 0.001 to 0.07 (U/(lmin)) while other
parameters are xed.
The right hand side gure in Figures 2.7.2 shows the stability region when d
i
changes from 0.001 to 0.7 (U/(lmin)) and G
in
= 0.54,
2
= 36 and
1
= 7.5 are
xed. The dotted line is the steady state when it is unstable and the solid lines are
the amplitude of the periodic solution bifurcated when the steady state switches its
stability. The left gure in Figure 2.7.6 shows the periods of the periodic solutions
when d
i
changes in [0.001, 0.7]. It is noticeable that the period of the oscillations
decreases dramatically when d
i
increases and tend to be smaller. The right gure shows
in each cycle, how long it takes the insulin concentration to peak after the glucose
concentration level peaks.
7.4. Hepatic Glucose Production
2
. Let the hepatic glucose production
2
be the bifurcation parameter and others are xed.
64
25 30 35 40 45
6500
7000
7500
8000
8500
G
l
u
c
o
s
e
(
r
e
d
, m
g
)
25 30 35 40 45
70
75
80
85
90
95
100
tau
2
(min) when tau
1
= 6 (min), Gin = 0.54 (mg/dl/min)
I
n
s
u
l
i
n
(
b
l
u
e
, m
U
)
25
30
35
40
45
70
75
80
85
90
95
100
6500
7000
7500
8000
8500
tau
2
(min)
Periodic Solutions in (tau
2
, I, G) when (tau
1
= 6 (min), Gin = 0.54 (mg/dl/min))
I (mg/dl)
G

(
m
U
/
m
l
)
Figure 2.7.7. Hepatic production delay has no impact to sustained oscillations
Periodic solutions (right) and Amplitude of periodic solutions (left) when
2
[25, 45] and
1
=
6.0, G
in
= 0.54, d
i
= 0.03846 xed. Theses two gures indicates the hepatic time delay
2
does not
have impact to the oscillations when
2
[25, 45]
Figure 2.7.7 shows the stability regions, limit cycles and period of periodic so-
lutions when
2
(25, 45). To summarize the observations of these three gures, we
have
Numerical Observation 2.7.4 In this case we observe that the hepatic delay
2
is
not sensitive in its physiologically meaningful range.
7.5. Parameter
1
vs. G
in
. In this subsection, we take both the insulin re-
sponse delay
1
and the glucose infusion rate G
in
as bifurcation parameters. We try to
identify the stability regions in the (
1
, G
in
) space.
Let d
i
= 0.03849 and
2
= 36 are xed. We compute the bifurcation points for
the parameters
1
and G
in
varying in
1
, G
in
[4.9, 20] [0, 2.16]. Figure 2.7.8 shows
the stability regions and stability region in the (
1
, G
in
)-plane. Summarize all above,
we have
Numerical Observation 2.7.5 In model (2.3.1), assume
2
= 36, d
i
= 0.03849, then
there exists a curve composed of bifurcation points and divides the rectangular [4.9, 20]
65
0 2 4 6 8 10 12 14 16 18 20
0
20
40
60
80
100
120
140
160
180
200
220
Stable Region
Unstable Region
tau_1
G
_
i
n
5 5.5 6
0
10
20
30
40
50
60
70
80
90
Unstable Region
Stable Region
Figure 2.7.8. Stability Region in (
1
, G
in
)-plane
Left: Stability Region in (
1
, G
in
)-plane (d
i
= 0.03846,
2
= 36). This gures demonstrates that the
oscillations will not be sustained if the glucose infusion rate G
in
is very large or the insulin response
time delay is very small. Right: Zoomed in left gure.
[0, 2.16] in (
1
, G
in
)-plane into two regions. The steady state of the model (2.3.1) is stable
in one of these two regions and unstable in the other region (refer to Figure 2.7.8). In
more details,
(1) There exists G
0
in
> 1.40 such that if G
in
< G
0
in
, there is a unique stability
switch when
1
varies from 0 to 20. That is, there exists a
0
, 0 <
0
< 20 depending
on G
in
such that the steady state (G

, I

) is stable for
1
[0,
0
) and unstable for

1
[
0
, 20].
(2) There exists a
01
, 5.1 <
01
< 5.2, such that the steady state is always stable
if 0 <
1
<
01
.
(3) There exists a
02
,
01
<
02
< 6.2, such that when
1
(
01
,
02
), stability
switches twice while G
in
increase from 0 to 2.16. That is, there exist G
01
in
and G
02
in
depending on
1
, 0 < G
01
in
< G
02
in
< 90, such that when G
in
G
01
in
or G
in
G
02
in
, the
steady state (G

, I

) of Model (2.3.1) is stable; when G

in
(G
01
in
, G
02
in
), the steady state
(G

, I

) of Model (2.3.1) is unstable.

66
4
5
6
7
8
9 0
0.5
1
1.5
2
2.5
0
10
20
30
40
50
60
70
80
90
100
Gin (mg/dl/min)
1
(min)
G (mg/dl)
I (U/l)
4
5
6
7
8
9
0
0.5
1
1.5
2
2.5
0
20
40
60
80
100
120
140

1
(min)
P
e
r
i
o
d

o
f

p
e
r
i
o
d
i
c

s
o
l
u
t
i
o
n
s

(
m
i
n
)
Gin (mg/dl/min )
Period (min)
Unstable region
Stable region
Figure 2.7.9. Bifurcation diagrams and stability regions in (
1
, G
in
)-space
Left: The stability regions in 3D-mesh show for G (top) and I (bottom) in (
1
, G
in
) [4.9, 9] [0, 2.16]
while
2
= 36. Each plane
1
= c [4.9, 9] and each plane G
in
= c [0, 2.16] intersect the 3D-mesh
at the bifurcation diagrams in G
in
[0, 2.16] and
1
[4.9, 9], respectively. Right: Periods of Periodic
Solutions. This gure shows clearly that the periods jump out at bifurcation points.
(4) When
1
>
02
, there exists a

G
in
(
1
) > 80 such that (G

, I

) is stable if
G
in
<

G
in
(
1
) and unstable if G
in


G
in
(
1
).
The right gure in Figure 2.7.9 shows the periods of the periodic solutions for
each point (
1
, G
in
) [4.9, 9] [0, 2.16]. This gure shows clearly that the periods jump
out at bifurcation points. The left gure shows the stability region in 3D mesh for G
(top) and I (bottom) in (
1
, G
in
) [4.9, 9] [0, 2.16]. Each plane
1
= c [4.9, 9]
and each plane G
in
= c [0, 2.16] intersect the 3D mesh at the stability region in
G
in
[0, 2.16] and
1
[4.9, 9], respectively.
7.6. Parameter
1
vs. d
i
. In this subsection, we take both the insulin response
delay
1
and the insulin degradation rate d
i
as bifurcation parameters. We try to identify
the stability regions in the (
1
, d
i
) space.
To study the relationship between the parameter
1
vs. d
i
, we compute the
bifurcation points for the parameters
1
and d
i
varying in [4.9, 20] [0.0025, 0.7]. Figure
67
5 6 7 8 9 10 11
0
0.1
0.2
0.3
0.4
0.5
0.6
tau
1
(min) (Gin=54, tau
2
=36)
d
i
(
m
U
/
l
/
m
i
n
)
: f
r
o
m
0
.0
0
2
5
t
o
0
.7
Unstable Region
Stable Region
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
0.1
0.2
0.3
0.4
0.5
0.6
d
i

(
m
u
U
/
l
/
m
i
n
)
Gin (mg/dl/min)
Unstable Region
Stable Region
Figure 2.7.10. Stability Regions in (
1
, d
i
)-plane and (G
in
, d
i
)-plane
Left: stability region in (
1
, d
i
)-space, G
in
= 0.54,
2
= 36. Right: stability region in (G
in
, d
i
)-plane,

1
= 6.0,
2
= 36
2.7.10 shows the stability region in this case. To summarize the observations of these
three gures, we have
Numerical Observation 2.7.6 In model (2.3.1), assume
2
= 36, G
in
= 0.54, then
there exists a curve composed of bifurcation points and divides the rectangular [4.9, 20]
[0.0025, 0.7] in (
1
, d
i
)-plane into two regions. The steady state of the model (2.3.1) is
stable in one of these two regions and unstable in the other region (refer to Figure
2.7.10).
7.7. Parameter G
in
vs. d
i
. In this subsection, we take both the insulin degra-
dation rate d
i
and the glucose infusion rate G
in
as bifurcation parameters. We try to
identify the stability regions in the (d
i
, G
in
) space.
The stability region in (G
in
, d
i
)-plane is shown in the right hand side gure
of Figure 2.7.10. We computed the bifurcation points in the rectangular (G
in
, d
i
)
[0, 85] [0.075, 0.7] while
1
= 6,
2
= 36 are xed.
Numerical Observation 2.7.7 In model (2.3.1), assume
1
= 6,
2
= 36, then a
68
curve composed with bifurcation points and divides the rectangular [0.0075, 0.7][0, 0.85]
in (
1
, d
i
)-plane into two regions. The steady state of the model (2.3.1) is stable in one
of these two regions and unstable in the other region (refer to Figure 2.7.10).
7.8. Insulin Concentration Peaks after Glucose Concentration Peaks.
We observed that in one cycle of the insulin secretion ultradian oscillation, the insulin
concentration peaks after glucose concentration peaks in a 15-35 min range depending
on the parameters
1
,
2
, G
in
and d
i
. Figure 2.7.11 shows four cases indicating the
dierences of the peaks. From the physiological point of view, the insulin secretion
is triggered by the increased glucose concentration level and, in turn, the increased
insulin concentration level helps the cells to consume the plasma glucose and thus
bring the glucose concentration level down. The left hand side gure in Figure 2.7.11
demonstrates that the oset of the peaks between the glucose concentration level and
the insulin concentration level increase when the insulin response time delay
1
increases
from its bifurcation point
10
( 5.15) to 20 minutes.
8. Discussion
Lets recall the physiological background rst. The experiments have revealed,
both in-vivo and in-vitro, that the insulin secretion ultradian oscillations sustain in the
daily life of healthy subjects in the interval 80-150 minutes ([73], [79], [60], [74], [75],
[67] and their cited references). Due to the complex chemical reactions in the glucose-
insulin endocrine metabolism, the secretion of insulin stimulated by glucose contains two
signicant time delays: one is so called hepatic glucose production time delay caused
by the time lag between the appearance of insulin in the plasma and its inhibitory
eect on livers converting glucagon into glucose ([15], [69]); the other one is due to
69
7800 7810 7820 7830 7840 7850 7860 7870 7880 7890 7900
71
72
73
74
75
76
77
78
79
G
l
u
c
o
s
e

(
m
g
/
d
l
)
Insulin Concentration Peaks 22.0761 min after Glucose Concentration
7800 7810 7820 7830 7840 7850 7860 7870 7880 7890 7900
7.8
8
8.2
8.4
8.6
8.8
I
n
s
u
l
i
n

(
m
U
/
l
)
t (min) (tau
1
= 5.5, tau
2
= 36, Gin = 0.54, di = 0.03849)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
0
5
10
15
20
25
30
T
i
m
e

(
m
i
n
)

s
h
i
f
t

b
e
t
w
e
e
n

p
e
a
k
s

o
f

G

a
n
d

I

o
f

p
e
r
i
o
d
i
c

s
o
l
u
t
i
o
n
s
di (mU/l/min), Gin=54 (mg/dl/min), tau
1
=7.5 (min), tau
2
=36 (min)
Figure 2.7.11. Glucose concentrations peak before insulin does
Left: in one cycle of the insulin secretion ultradian oscillation, insulin concentration level peaks after
glucose concentration. Right: the dierence of time (min.) between the insulin concentration and the
glucose concentration level peak changes from 20 to 35 min: d
i
(0.001, 0.7) while other parameters
are xed
chemical reactions and physiological actions of insulin on the uptakes of glucose ([88],
[66]). This oscillation has been modelled by J. Sturis, K. S. Polonsky, E. Mosekilde
and E. Van Cauter ([79], 1991) and I. M. Tolic, E. Mosekilde and J. Sturis ([84], 2000)
(model (2.2.1)), K. Engelborghs, V. Lemaire, J. Belair and D. Roose ([31], 2001) (model
(2.2.8)) and D. L. Bennett and S. A. Gourley ([4]) (model (2.2.9)).
In this chapter, we introduced a more realistic model (Model (2.3.1)) with two
time delays. One delay,
2
, is the same as introduced in [4] and the other one is for
the delay,
1
, caused by the chemical reactions in Langerhans islets. So the eort of
breaking insulin into two dierent compartment is unnecessary and thus the number of
equations is reduced by one. Our analytical and numerical analysis yield the following
observations and exhibit intrinsic insulin secretion ultradian oscillations.
Lets recall the results obtained in [79] [84] and [4] and compare with our results.
We list the major results in [79] and [84] as [STx], the major results in [4] as [BGx], our
results as [Ax], where x is a number.
70
ST1 The ultradian insulin secretion oscillation is critically dependent on hepatic glu-
cose production, that is, if there is no hepatic glucose production, then there is
no insulin secretion oscillation. ([79] and [84])
A1 From Theorem 2.5.1, we can notice that f
5
(d
1
i
f
1
(x)) f
5
(d
1
i
f
1
(y)) for x y 0
means higher hepatic production of glucose helps to make oscillations happen (the
case that (G

, I

) is unstable). Comparing with [ST1], notice that f

3
(G) can be
linear and f
4
is bounded, which is the case used by [79] and [84]. If G is big enough
and there is no hepatic production (f
5
0), then the steady state (G

, I

) will be
globally stable and thus the insulin oscillation will not occur.
A2 If the hepatic glucose production
2
= 0, then no insulin oscillation is observed
due to Theorem 2.6.1. This also conrms above [ST1]. This seems a surprising
observation. But, from physiological point of view, the hepatic glucose produc-
tion is much smaller relative to the glucose infusion thus its impact should be
relatively small, but is not ignorable. Neither is the time delay, for the normal sit-
uation (refer to IVGTT models in Chapter 3 for the case that the hepatic glucose
production triggered by insulin can be completely ignored.)
ST2 When the hepatic glucose production time delay
2
(25, 50), the period of the
periodic solutions of both insulin and glucose is in interval (95, 140) (min.), that
is, (95, 140). ([79] and [84])
A3 Figure 2.7.7 indicates that the hepatic glucose production time delay
2
[25, 45]
has no impact to the insulin oscillation if the glucose infusion rate, the insulin
degradation rate and the insulin secretion time delay
1
are in their right ranges,
respectively (Numerical Observation 2.7.4).
71
A4 Figure 2.7.11 indicates that during one cycle of the insulin and glucose concen-
tration oscillations (95-135 min depending on
1
, G
in
and d
i
), the insulin concen-
tration level peaks after the glucose concentration level does. This reects the
physiological fact that glucose stimulates insulin secretion. On the other hand,
the glucose concentration level bottoms before the insulin concentration level bot-
toms. This reects that higher insulin concentration helps the glucose uptake by
cells. Our [A2] and [A3] conrms [ST2].
ST3 To obtain the ultradian oscillation (periodic solutions), it is necessary to break
the insulin into two separate compartments, the plasma and interstitial tissues.
([79] and [84])
A5 Theorem 2.6.1 reveals that if there is no insulin response to glucose stimulation
time delay (
1
= 0), insulin secretion oscillation will not be sustained. Comparing
with [ST3], the eort of compartment split on insulin is overcome by introducing
the insulin response time delay.
BG2 If the hepatic glucose production time delay and the insulin degradation rates
between the plasma and interstitial compartments t
i
and t
d
are suciently small,
then there is no sustained oscillations. For larger delay or large glucose infusion
rate, oscillatory solutions become possible [4].
A6 Theorem 2.6.2 and Theorem 2.6.1 help to understand the role of insulin degra-
dation rate. For a subject, if ones insulin degradation rate is suciently small
or suciently large (d
i

f

1
(G

)(f
3
(G

)f

4
(I

)f

5
(I

))
f

2
(G

)+f

3
(G

)f
4
(I

)
), the insulin secretion ultradian
oscillation does not occur. The oscillation can be sustained if the degradation rate
is moderate (d
i
<
f

1
(G

)|f
3
(G

)f

4
(I

)+f

5
(I

)|
f

2
(G

)+f

3
(G

)f
4
(I

)
). The numerical observation indicates
72
that if the insulin degradation is too small, the glucose maintains in a low basel
level (see Figure 2.7.10, Figure 2.7.10). This is considered to be hypoglycemic.
On the other hand, if the insulin degradation is too large, the glucose basel level
is high which may lead to hyperglycemia (see Figure 2.7.10, Figure 2.7.10). This
provides more insightful information than the general statements in [BG2].
A7 When the glucose infusion rate and insulin degradation rate are moderate (not
too high and not too low), the insulin oscillation is sustained if the time delay

1
> 5.15 and
2
(25, 45) (Figure 2.7.1). In this case, along with the increase of
time delay
1
from 5.15 to 10 minutes, the amplitude of oscillation increases and
the period of the oscillation increases signicantly from about 98 to 140 minutes
(Figure 2.7.2). All these are consistent to the experiments ([73], [79], [60], [74],
[75], [67] and their cited references). Comparing with [BG2], the bifurcation
diagram in Figure 2.7.1 quanties the behaviors of the dynamics in the particular
case.
Besides, we also have following observations from our model (2.3.1).
A8 Theorem 2.4.1 guarantees that there is no such case for a healthy subject, he/she
neither experiences a stable glucose concentration at basel level and at the same
time have his/her insulin concentration oscillates nor does he/she have his/her
insulin concentration stabilizes at its basel level and the glucose concentration
oscillates. In other words, the insulin concentration and glucose concentration
levels vary simultaneously with an observed variation (see Figure 2.7.11).
A9 From Proposition 2.4.1, we know that the glucose concentration and insulin con-
centration basel levels are unique.
73
A10 Considering the small glucose infusion rate v.s. the insulin response time delay,
Figure 2.7.8 reveals that when the insulin response to glucose stimulation time
delay
1
is in the range of 5.15 minutes to 6.15 minutes, small glucose infusion
rate G
in
< 0.135 (mg/dl/min) would not make insulin oscillations to be sustained.
(See also left gures in Figure 2.7.3 and Figure 2.7.4.) This indicates if the glucose
infusion is small but the insulin response time is quick, then the small amount
of glucose added into the bloodstream would be consumed quickly and no more
insulin would need to be secreted.
On the other hand, Figure 2.7.8 also reveals that when the insulin response to
glucose stimulation time delay
1
is approximately greater than 6.15 minutes, the
insulin oscillations sustain at small, even zero, glucose infusion rate (refer to the
right gures in Figure 2.7.3 and Figure 2.7.4.) This is possibly due to the slow
response of insulin secretion to the glucose stimulation and thus the glucose uptake
by cells are not that fast.
So only moderate glucose infusion rate 0.135 < G
in
< 0.675 (mg/dl/min) can
make insulin secretion ultradian oscillations to be sustained when the insulin
response to glucose stimulation is fast.
A11 When considering the large glucose infusion rate v.s. the insulin response time
delay. Again, Figure 2.7.8 shows when the glucose infusion rate is high, G
in
> 1.40
(mg/dl/min), no matter how fast or slow the insulin responds to the glucose
stimulation, the insulin oscillation will not be sustained. This indicates that
the insulin can not be produced and released from -cells to uptake large amount
glucose in the plasma and thus sustains oscillation. This explains why the IVGTT
models in the Chapter 3 focus only on the study of asymptotically stable steady
74
state (the basel levels of glucose and insulin concentrations).
A12 From Figure 2.7.5 and Figure 2.7.2, when d
i
= 0.03849,
2
= 36, the larger the
glucose stimulation delay
1
(5.15, 15), the period is signicantly larger; the
larger the glucose infusion rate G
in
, the smaller the period, but not signicantly.
So, our analytical results and numerical observations have conrmed the exist-
ing research results or observations and also provided more insightful and quantitative
information.
CHAPTER 3
Modeling Intra-Venus Glucose Tolerance Test
1. Introduction
Due to the increased occurrence of pathological conditions such as diabetes (8% of
American people in 2002 ([94])) and obesity (one third, or currently 18 million American
adults ([92]), the quantication of insulin sensitivity from some relatively non-invasive
tests has gained increased interest and importance in physiological research. This lead
to some new studies on the existing models ([10], [8], [60]) and the introduction of some
alternative ones ([23], [63], [60]).
To detect the onset of diabetes or to diagnose the potential of having diabetes,
several tests or protocols are currently in use. These include Oral Glucose Tolerance Test
(OGTT), Fasting Glucose Tolerance Test (FGTT), Intra-venous Glucose Tolerance Test
(IVGTT) and the frequently sampled Intra-venous Glucose Tolerance Test (fsIVGTT)
([93]). They all test the insulin sensitivity or response to high the plasma glucose
concentration with a big bolus of glucose infusion. The most eective and accurate test
is the Intra-venous Glucose Tolerance Test (IVGTT) and the frequently sampled Intra-
venous Glucose Tolerance Test (fsIVGTT). During IVGTT, with designated frequently
sampled the plasma glucose concentration levels after a big bolus of glucose intra-venous
infusion, for example, 0.33g/kg body weight [23], the rich data can reveal how ecient
76
the insulin sensitivity of the subjects glucose-insulin metabolic regulatory system is.
Many mathematical models study the Intravenous Glucose Tolerance Test (IVGTT)
([10], [8], [23], [57], [60], [63] and [61]). The most widely used is the so called Mini-
mal Model ([10], [8]), which is challenged by [23]. [23] proposed an alternative model
called Dynamic Model. Besides the exogenous glucose intake, the liver produces a
small amount of glucose and converts glycogen into glucose when the plasma glucose
concentration level is low, with a delay of about 30 to 50 minutes ([79], [84]). As pointed
out in the numerical observation 2.7.5 in Chapter 2, due to the insulin response delay
(about 5 to 15 minutes), although the huge glucose intravenous infusion, the small
amount of hepatic glucose production is still sustained until a large amount of insulin
is released from the cells. But, in a study of dynamics in a short period (about 30
min), the insulin response time delay of the liver glucose production is usually ignored
(the term f
5
(I(t
2
)) in model (2.2.1), (2.2.9), (2.2.7), (2.2.8) and (2.3.1). This is in
agreement with the absence of this term in these models ([10], [8], [23], [57] and [63]).
As shown in Chapter 2, the absence of this term or the time delay often admits global
asymptotically stable steady state (Section 8 in Chapter 2, [79] and [84]).
This chapter is organized as follows. In Section 2, we summarize the current
status with focus on the minimal model ([10], [8]) and dynamic model ([23], [63]). Then
we present our genetic model and two other less general ones (still more general than the
dynamic model [23]) in Section 2, and present our generic models in Section 3. The basic
properties of the models we present in Section 3 will be discussed in Section 4. Section 5
contains some general global stability results. Section 6 presents a formal derivation of
linearization and characteristic equations. Section 7 provides some delay independent of
local stability results and Section 8 gives delay dependent stability conditions. Guided
77
by the results of Section 8, we performed extensive computer simulation (using clinic
data) and succeeded in nding periodic solutions in our discrete delay model, which is
shown in Section 9. Discussions on the implications of our results are summarized in
Section 10.
2. Current Research Status
The dynamic relationship between glucose and its controlling hormone insulin
has been mathematically modelled and studied by many researchers since the sixties
([11], [45], [55], [70], [79], [84], [85], [31], and the references cited in [29]). Most of
these models consist of several ordinary dierential equations, the number of equations
is often proportional to that of factors considered. Some of these equations are simply
linear and were judged unacceptable for various reasons ([10], [8]), such as parameters
are not identiable or have poor ts to experimental data. Nevertheless, the most
noticeable model, the so called Minimal Model which contains minimal number of
parameters ([10], [8]), is widely used in physiological research work to estimate glucose
eectiveness (SG) and insulin sensitivity (SI) from intravenous glucose tolerance test
(IVGTT) data by sampling over certain periods. Also a few are on the control through
meals and exercise ([25]). According to ([6]), 2002), there are now approximately 50
major studies published per year and more than 500 can be found in the literature,
according to the same author, which involve the minimal model.
Currently, the most widely used model in physiological research on the metabolism
of glucose is the so-called minimal model, which describes intra-venous glucose tol-
erance test (IVGTT) experimental data well using the smallest set of identiable and
meaningful parameters ( [10], [65]). After incorporating the insulin dynamics, it takes
78
the form of [23])
_

_
dG(t)
dt
= G

= [b
1
+ X(t)]G(t) +b
1
G
b
,
dX(t)
dt
= X

= b
2
X(t) +b
3
[I(t) I
b
],
dI(t)
dt
= I

= b
4
[G(t) b
5
]
+
t b
6
[I(t) I
b
].
(3.2.1)
The initial conditions are:
G(0) = b
0
, X(0) = 0, I(0) = b
7
+ I
b
.
Here G(t) [mg/dl],I(t) [UI/ml] is the plasma glucose, insulin concentration at time t
[min], respectively. X(t) [min
1
] is an auxiliary function representing insulin-excitable
tissue glucose uptake activity, roughly proportional to insulin concentration in a dis-
tant compartment. G
b
[mg/dl], I
b
[UI/ml] is the subjects baseline glycemia, insuline-
mia, respectively. b
0
[mg/dl] is theoretical glycemia at time 0 after the instantaneous
glucose bolus intake. b
1
[min
1
] is the insulin-independent constant of tissue glucose
uptake rate. b
2
[min
1
] is the rate constant describing the spontaneous decrease of
tissue glucose uptake ability. b
3
[min
2
(UI/ml)
1
]is the insulin-dependent increase in
tissue glucose uptake ability, per unit of insulin concentration excess over the baseline.
b
4
[(UI/ml)(mg/dl)
1
min
1
] is the rate of pancreatic release of insulin after the intake
of the glucose bolus, per minute per unit of glucose concentration above the target
glycemia b
5
[mg/dl]. b
6
[UI/ml] is the rst order decay rate for insulin in the plasma.
b
7
[UI/ml] is the plasma insulin concentration at time 0, above basal insulinemia,
immediately after the glucose bolus intake.
While the above minimal model has a minimal number of constants (b
0
b
7
),
and has been very useful in physiological research works, this minimal model has been
79
challenged by De Gaetano and Arino [23] from both physiological and modeling aspects
recently. [23] argues that it has the following three drawbacks associated with it. For
this model, the parameter tting is to be divided into two separate parts: rst, using
the recorded insulin concentration as given input data in order to derive the parameters
in the rst two equations in the model, then using the recorded glucose concentration as
given input to derive the parameters in the third equation. However, the system is an
integrated physiological dynamic system and one should treat it as a whole and be able
to conduct a single-step parameter tting process. Secondly, some of the mathematical
results produced by this model are not realistic. Specically, it can be shown that the
minimal model dose not admit an equilibrium and the solutions may not be bounded.
Finally, the non-observable auxiliary variable X(t) is articially introduced to delay the
action of insulin on glucose. An alternative and natural way is to explicitly introduce the
time delay in the model. To address these issues, De Gaetano and Arino [23] introduced
the following aggregated delay dierential model which they named as dynamic model
by use of certain simple and specic functions and introduces a time delay in a particular
way. It takes the form of
_

_
dG(t)
dt
= G

= b
1
G(t) b
4
I(t)G(t) + b
7
,
dI(t)
dt
= I

= b
2
I(t) +
b
6
b
5
_
t
tb
5
G(s)ds.
(3.2.2)
The initial condition now takes the form of
G(0) = G
b
+ b
0
, I(0) = I
b
+ b
3
b
0
, and for t [b
5
, 0), G(t) = G
b
.
As in the minimal model, G(t) [mg/dl],I(t) [UI/ml] is the plasma glucose, insulin
concentration at time t [min], respectively. The G
b
, I
b
, b
0
, b
1
, b
2
and b
3
are the same
80
as, or similar to that in the minimal model with same units. b
4
[min
1
pM
1
] is the
constant measuring the insulin-dependent glucose disappearance rate per unit [pM]
of the plasma insulin concentration. b
5
[min] is the number of minutes of the past
period whose the plasma glucose concentrations inuence the current pancreatic insulin
secretion. b
6
[min
1
pM/(mg/dl)] is the constant describing the second-phase pancreatic
insulin secretion rate per unit of average the plasma glucose concentration throughout
the previous b
5
minutes. b
7
[(mg/dl)min
1
] is the constant increase in the plasma glucose
concentration due to constant baseline liver glucose release.
The outcome is that the model always admits a globally asymptotically stable
steady state.
One of the objectives of this chapter is to nd out if and how this outcome
depends on the specic choice of functions and the way delay is incorporated. To
this end, we generalize the dynamical model to allow more general functions and an
alternative way of incorporating time delay. Our ndings show that in theory, such
models can possess unstable positive steady states and produce oscillatory solutions.
However, for all the clinic data reported in [23], such unstable steady states do not
exist. Hence, our work indicates that the dynamic model does provide qualitatively
robust dynamics for the purpose of clinic application. We also perform simulations
based on data from a clinic study reported in [23] and point out some plausible but
important implications.
3. More Generic IVGTT Model
While the dynamic model solves the problems of minimal model, it implicitly or
explicitly made a few assumptions that may not be necessary or realistic. Specically
81
some of the interaction terms are too special and thus too restrictive. For example, the
term b
4
I(t)G(t) assumes mass action law applies here. A more popular, general and
realistic alternative is to replace this term by b
4
I(t)G(t)/(G(t) +1). Since in a unit of
time, a unit of insulin can only process a limited amount of glucose. Also the way the
delay is introduced is somewhat restrictive, the justication of which consists of only
one subjective assumption the delay term refers to the pancreatic secretion of insulin:
eective pancreatic secretion at time t is considered to be proportional to the average
value of glucose concentration in the b
5
minutes preceding time t. This naturally invites
other plausible ways of incorporating the time delay. The most noteworthy outcome of
the dynamical model is that it always admits a globally asymptotically stable steady
state.
We propose the following general and more realistic model for the interaction
of glucose and insulin. This model includes the dynamic model ( 3.2.2) as a special
case:
_

_
dG(t)
dt
= G

(t) = f(G(t)) g(G(t), I(t)) +b

7
,
dI(t)
dt
= I

(t) = p(I(t)) + q(L(G

t
)).
(3.3.1)
The initial condition is
G(0) = G
b
+ b
0
, I(0) = I
b
+b
3
b
0
, and G(t) G
b
, for t [b
5
, 0),
where G
t
() = G(t + ), t > 0, [b
5
, 0]. G
b
[mg/dl], I
b
[UI/ml] is the subjects
baseline glycemia, insulinemia, respectively. The parameters b
0
, b
3
, b
5
and b
7
are the
same as in model (3.2.2). Functions f, g, p, q satisfy the following general conditions.
(i) f(0) = 0, f() = , 0 < f

(x) < for x > 0

82
(ii) g(0, 0) = 0, 0 < g
x
(x, y) < , 0 < g
y
(x, y) < for x > 0
g(x, 0) = 0, g(0, y) = 0, g(, y) < , and g(x, ) = when x = 0
(iii) p(0) = 0, p() = , 0 < p

(x) < for x > 0

(iv) q(x) = 0, if and only if x = 0; L : C[b
5
, 0] C[b
5
, 0] is a linear operator
dened as L() =
_
0
b
5
(s)d((s)), where (s) is nondecreasing with
_
0
b
5
d((s)) = 1.
We will consider two cases for L(G
t
): the discrete and distributed cases. For discrete
delay, L(G
t
) = G(tb
5
) and for distributed delay, L(G
t
) =
1
b
5
_
0
b
5
G(t+)d. We assume
q(L(G
t
+
t
)) > q(L(G
t
)) for
t
C[b
5
, 0] with
t
() > 0, [b
5
, 0]. Furthermore,
we assume that |q(
1
) q(
2
)| < k
1

2
for
1
,
2
C[b
5
, 0], where is C
norm. So q(L()) : C R is Lipschitz.
We always assume that the model (3.3.1) has a unique equilibrium point (G

, I

)
in R
2
+
= {(x, y) : x > 0, y > 0}.
We also present two other less general models (still more general than the dy-
namic model in [23]), for the convenience of analysis and applications. We rst propose
the following specic model of glucose-insulin interaction.
_

_
dG(t)
dt
= G

(t) = b
1
G(t)
b
4
I(t)G(t)
G(t) + 1
+ b
7
dI(t)
dt
= I

(t) = b
2
I(t) + b
6
G(t b
5
)
(3.3.2)
with the same initial conditions of model (3.3.1). The parameters have the same mean-
ing as those in dynamic model (3.2.2).
Comparing with the dynamic model (3.2.2), model (3.3.2) has two notable and
important dierences: First, no mass action law is assumed for glucose concentration
change due to the insulin-dependent net glucose tissue uptake. We assume instead
that insulin-dependent net glucose tissue uptake takes the more general and realistic
83
Michaelis-Menten form G(t)/(G(t) + 1) which has a maximum capacity b
4
/. The
parameter in the response function G(t)/(G(t) + 1) is non-negative. 1/ is the
half-saturation constant. The reason for this is simply due to the limit of time and the
capacity of insulins ability of digesting glucose. Second, we assume that the eective
pancreatic secretion (after the liver rst-pass eect) at time t is aected by the value of
glucose concentration in the b
5
minutes preceding time t instead of the average amount
in that period.
When
L(G
t
) =
1
b
5
_
0
b
5
G(t + )d,
and the Michaelis-Menten kinetics is assumed, the model (3.3.1) becomes
_

_
G

(t) = b
1
G(t)
b
4
I(t)G(t)
G(t) + 1
+ b
7
I

(t) = b
2
I(t) +
b
6
b
5
_
0
b
5
G(t + )d
(3.3.3)
Clearly, both models ( 3.3.2) and (3.3.3) have a unique equilibrium point (G

, I

)
in R
2
+
= {(x, y) : x 0, y 0}, where
G

= 2b
7
/
_
(b
1
b
7
) +

(b
1
b
7
)
2
+ 4b
7
_
b
1
+
b
4
b
6
b
2
__
and I

=
b
6
b
2
G

.
Obviously, model (3.3.2)((3.3.3)) is a special case of model (3.3.1), where f(x) =
b
1
x, g(x, y) = b
4
xy/(x + 1), p(x) = b
2
x and q(L(x
t
)) = b
6
x(t b
5
) (q(L(x
t
)) =
(b
6
/b
5
)
_
b
6
tb
5
x(ts)ds). For the same choice of f, g, p and q(L(x
t
)) = (b
6
/b
5
)
_
t
tb
5
x(s)ds
and = 0, model (3.3.1) reduces to the dynamic model (3.2.2).
4. Preliminary Analysis
The following basic proposition is important for our study. Its proof is straight-
forward.
84
Proposition 3.4.1 All solutions of model (3.3.1) exist for all t > 0, and are positive
and bounded.
Proof Since the |f

(x)|, |g
x
(x, y)|, |g
y
(x, y)| and |p

(x)| are bounded for x, y > 0, they

are Lipschitz and completely continuous for x > 0, y > 0. Notice that we assume q(x) is
Lipschitz in for x 0 and L : C[b
5
, 0] C[b
5
, 0] is linear, then q(L()) is Lipschitz
in C[b
5
, 0]. By Theorem 2.1, 2.2 and 2.4 on page 19 and 20 in [54], the solution of
equation (3.3.1) with given initial condition exists and unique for all t 0.
Let (G(t), I(t)) be a solution of (3.3.1). If G(t
0
) = 0 for some t
0
> 0 and
G(t) > 0 for 0 < t < t
0
, then G

(t
0
) 0. However, at t
0
, due to the assumptions that
f(0) = g(0, y) = 0, we have G

(t
0
) = f(G(t
0
)) g(G(t
0
), I(t
0
)) + b
7
= b
7
> 0. This
contradiction shows that G(t) > 0 for all t in the interval of existence. If I(

t
0
) = 0
for some

t
0
> 0, then I

t
0
)) 0 and 0 I

t
0
) = p(I(

t
0
)) + q(L(G
t
0
)) = q(L(G
t
0
)).
Since G
t
0
() > 0 for [b
5
, 0], q(L(G
t
0
)) > 0 by (iv) and thus I(t) > 0 for all t in
the interval of existence.
As for the boundedness of G(t), by the rst equation of (3.3.1),
G

(t) = f(G(t)) g(G(t), I(t)) + b

7
f(G(t)) + b
7
.
Thus G(t) is bounded by M
G
= max{G
b
+ b
0
, f
1
(b
7
)}. And hence I(t) is bounded by
M
I

= max{I
b
+ b
3
b
0
, p
1
(q(M
G
))} due to
I

(t) = p(I(t)) + q(G

t
) p(I(t)) + q(M
G
).
This completes the proof.
Let (G(t), I(t)) be a solution of (3.3.1). Throughout this paper, we dene
G = limsup
t
G(t), G = liminf
t
G(t) I = limsup
t
I(t), I = liminf
t
I(t).
85
Due to the Proposition 3.4.1, we see that these limits are nite.
As in Chapter 2, we apply following elementary lemma to obtain a few prelimi-
nary results. See [48] for a proof.
Lemma A Let f : R R be a dierentiable function. If l = liminf
t
f(t) <
limsup
t
f(t) = L, then there are sequences {t
k
} , {s
k
} such that for all
k, f

(t
k
) = f

(s
k
) = 0, lim
k
f(t
k
) = L and lim
k
f(s
k
) = l.
The following lemma is useful for establishing the fact that model (3.3.1) is
always persistent, which implies that both components of solutions of the model are
eventually bounded by positive constants from both above and below. Such bounds are
independent of initial data.
Lemma 3.4.1 Consider model (3.3.1). If I < I, then
p
1
(q(G)) I < I p
1
(q(G)).
If G < G, then
f(G) g(G, I) + b
7
0, and f(G) g(G, I) + b
7
0. (3.4.1)
Proof Since I < I, by Lemma A, there exists {t
k
} , {s
k
} , such that I

(t
k
) =
I

(s
k
) = 0, lim
k
I(t
k
) = I and lim
k
I(s
k
) = I. Notice that p, q are continuous,
q

() > 0 and (G(t), I(t)) is a solution of (3.3.1). Hence, we have

0 = I

(t
k
) = p(I(t
k
)) + q(L(G
t
k
)) for all k.
For any > 0, there exists k
0
> 0, such that
G + > G
t
k
(), [, 0] for all k > k
0
.
86
Hence condition (iv) implies that q(L(G
t
k
)) q(G + ) for k > k
0
. Therefore,
0 = p(I(t
k
)) + q(L(G
t
k
)) p(I(t
k
)) + q(G +).
By letting k and 0, we have
p(I) q(G). (3.4.2)
Similarly, we have
p(I) q(G). (3.4.3)
(3.4.2) and (3.4.3) lead to
q(G) p(I) < p(I) q(G)
and then
p
1
(q(G)) I < I p
1
(q(G)).
If G < G, by Lemma A there exists {t

k
} , {s

k
} , such that G

(t

k
) =
G

(s

k
) = 0, lim
k
G(t

k
) = G and lim
k
G(s

k
) = G. Thus we have
0 = G

(t

k
) = f(G(t

k
)) g(G(t

k
), I(t

k
)) + b
7
and
0 = G

(s

k
) = f(G(s

k
)) g(G(s

k
), I(s

k
)) + b
7
for all k.
Since f, g are continuous and g
2
(x, y) > 0 for all x > 0, without loss of generality,
assuming lim
k
I(t

k
) and lim
k
I(s

k
) exist, we have
0 = lim
k
(f(G(t

k
)) g(G(t

k
), I(t

k
))) + b
7
= f(G) g(G, lim
k
I(t

k
)) +b
7
f(G) g(G, I) + b
7
87
and
0 = lim
t
(f(G(s

k
)) g(G(s

k
), I(s

k
))) + b
7
= f(G) g(G, lim
k
I(s

k
)) + b
7
f(G) g(G, I) + b
7
.
This completes the proof.
Proposition 3.4.2 The model (3.3.1) is persistent. That is, solutions are eventually
bounded by positive constants from both above and below.
Proof For a solution (G(t), I(t)) of (3.3.1), by Proposition 3.4.1,
G

(t) = f(G(t)) g(G(t), I(t)) + b

7
f(G(t)) + b
7
.
Using Lemma A, we can obtain that G f
1
(b
7
), where G

= limsup
t
G(t).
Notice that (3.4.1) implies
f(G) + g(G, p
1
(q(f
1
(b
7
)))) f(G) + g(G, p
1
(q(G))) b
7
,
which shows that G > 0 due to f(0) = 0 and g(0, y) = 0 for y 0. This together with
Lemma 3.4.1 shows that the model (3.3.1) is persistent.
5. Global Stability of Steady State
In this section, we provide several global stability results for the steady state
(G

, I

). The same method was used by [23] to establish the global stability of the pos-
itive steady state. As we shall see, for the general model (3.3.1), global asymptotically
stability of (G

, I

) is conditional.
Using mainly uctuation type argument and Lemma 3.4.1, we can obtain
88
Theorem 3.5.1 For model (3.3.1), if
g(x, p
1
(q(y))) g(y, p
1
(q(x))) 0, (3.5.1)
for all x y > 0, then the unique equilibrium point (G

, I

) of (3.3.1) is globally
asymptotically stable.
Proof If I < I, then from Lemma 3.4.1,
p
1
(q(G)) I < I p
1
(q(G)).
Thus G < G and
f(G) g(G, p
1
(q(G))) + b
7
f(G) g(G, I) + b
7
0,
f(G) g(G, p
1
(q(G))) + b
7
f(G) g(G, I) + b
7
0.
Therefore
(f(G) f(G)) + (g(G, p
1
(q(G))) g(G, p
1
(q(G)))) 0.
Due to (3.5.1), we have
g(G, p
1
(q(G))) g(G, p
1
(q(G))) 0.
Hence
f(G) f(G) 0,
which indicates G = G and thus I = I. Since (G

, I

) is the only equilibrium point of

(3.3.1), we have
lim
t
G(t) = G

and lim
t
I(t) = I

.
The proof is completed.
Theorem 3.5.2 For model (3.3.1), if
89
f

(x) +g
x
(x, p
1
(q(y))) g
y
(x, p
1
(q(y)))
q

(y)
p

(p
1
(q(y)))
> 0 (3.5.2)
for all x, y > 0, then the unique equilibrium point (G

, I

) of (3.3.1) is globally asymp-

totically stable.
Proof If I < I, then from Lemma 3.4.1,
p
1
(q(G)) I < I p
1
(q(G))
we see that G < G, and
(f(G) f(G)) + (g(G, p
1
(q(G))) g(G, p
1
(q(G)))) 0. (3.5.3)
Let
F(x, y) = f(x) + g(x, p
1
(q(y))), (x, y) R
2
+
= {(x, y) : x > 0, y > 0}.
Then (3.5.3) is equivalent to
F(G, G) F(G, G) 0. (3.5.4)
By the mean value theorem, there exists a (0, 1) such that
F(G, G) F(G, G) = (GG)(F
x
(, ) F
y
(, )) (3.5.5)
where

= G + (GG) and

= G(GG). Notice that
F
x
(x, y) = f

(x) +g
x
(x, p
1
(q(y)))
and
F
y
(x, y) = g
y
(x, p
1
(q(y)))
q

(y)
p

(p
1
(q(y)))
.
90
From (3.5.2)
F
x
(, ) F
y
(, ) = f

() + g
x
(, p
1
(q()))
g
y
(, p
1
(q()))
q

()
p

(p
1
(q()))
> 0.
(3.5.6)
On the other hand, (3.5.4) and (3.5.5) lead to
(GG)(F
x
(, ) F
y
(, )) 0.
Clearly, (3.5.6) leads to G G 0 and thus G = G and I = I. Since (G

, I

) is the
only equilibrium point of (3.3.1), we have
(G(t), I(t)) (G

, I

) as t .
This completes the proof.
In model (3.3.1), if g(x, y) takes the special form
g(x, y) = g
1
(x, y)/g
2
(x)
where
(v) g
1
(0, 0) = 0, g
1
(x, 0) = g
1
(0, y) = 0 for all (x, y) > 0,
(vi) g
2
(x) c > 0 for some constant c. g
2
() = .
(vii) (g
1
)
x
(x, y) > 0, (g
1
)
y
(x, y) > 0, for all x, y > 0.
(viii) g
1
(x, ) = g
1
(, y) = for x, y > 0, xy = 0.
(ix) g

2
(x) > 0 for all x > 0,
then we have
Theorem 3.5.3 For model (3.3.1), g(x, y) = g
1
(x, y)/g
2
(x), where g
1
, g
2
satisfy (v)
(ix). Then (G

, I

) is globally asymptotically stable if

91
(a) (f(x) b
7
)g
2
(x) is increasing for all x > 0
(b) g
1
(x, p
1
(q(y))) g
1
(y, p
1
(q(x))) 0 for all x y > 0.
Proof We shall show that G = G and I = I. By Lemma 3.4.1, if I < I, then G < G
and
p
1
(q(G)) I < I p
1
(q(G)),
f(G) g
1
(G, p
1
(q(G)))/g
2
(G) + b
7
0
and
f(G) g
1
(G, p
1
(q(G)))/g
2
(G) + b
7
0.
Thus
(f(G) b
7
)g
2
(G) g
1
(G, p
1
(q(G))) 0 (3.5.7)
and
(f(G) b
7
)g
2
(G) g
1
(G, p
1
(q(G))) 0. (3.5.8)
Hence, (3.5.7) and (3.5.8) imply
((f(G) b
7
)g
2
(G) (f(G) b
7
)g
2
(G)) + (g
1
(G, p
1
(q(G))) g
1
(G, p
1
(q(G)))) 0.
(3.5.9)
This together with the assumptions (a) and (b), we obtain
(f(G) b
7
)g
2
(G) (f(G) b
7
)g
2
(G) = 0.
This implies that G = G, and therefore I = I. This completes the proof.
Corollary 3.5.1 For model (3.3.1), assume g(x, y) = g
1
(x, y)/g
2
(x), where g
1
, g
2
sat-
isfy (v)(ix). If (b) in Theorem 3.5.3 is replaced by
(b

)

x
g
1
(x, p
1
(q(y)))

y
g
1
(x, p
1
(q(y)))
q

(y)
p

(p
1
(q(y)))
0 for all x, y > 0,
then (G

, I

) is globally asymptotically stable.

92
Proof We shall show that (b) in Theorem 3.5.3 is true when (b

) holds. Let
u(x, y) = g
1
(x, p
1
(q(y))), x, y > 0.
By the mean value theorem, for x y > 0, there exists (0, 1) such that
u(x, y) u(y, x) = u
x
(, )(x y) +u
y
(, )(x y)
= (x y)(u
x
(, ) u
y
(, )) 0,
where = y + (x y), = x (x y). This completes the proof.
The following are direct results of the applications of the above theorems to the
specic models (3.3.2) and (3.3.3). For convenience, we dene two new parameters.
a
1
:= b1
_
b
7
and := b
4
b
6
_
b
1
b
2
Corollary 3.5.2 For model (3.3.2) ((3.3.3)), if , then the only equilibrium point
(G

, I

) of (3.3.2) ((3.3.3)) is globally asymptotically stable.

Proof We shall apply Theorem 3.5.2 to model (3.3.2) ((3.3.3)). For model (3.3.2), we
have
f

(x) + g
x
(x, p
1
(q(y))) g
y
(x, p
1
(q(y)))
q

(y)
p

(p
1
(q(y)))
= b
1
+
b
4
(b
6
/b
2
)y
(x + 1)
2

b
4
x
x + 1
b
6
b
2
b
1
_
1
x
x + 1
_
= b
1
(( )x + 1)/(x + 1) > 0
for all x > 0, if .
Corollary 3.5.3 For model (3.3.2) ((3.3.3)), if a
1
= b
1
/b
7
, then the only equilib-
rium point (G

, I

) of (3.3.2)((3.3.3)) is globally asymptotically stable.

93
Proof We shall apply Theorem 3.5.3 to model (3.3.2). For (a) to hold, we need
d
dx
((f(x) b
7
)g
2
(x)) =
d
dx
((b
1
x b
7
)(x + 1))
= 2b
1
x + (b
1
b
7
) > 0
for all x > 0. This is true if b
1
b
7
0, i.e., b
1
/b
7
.
For (b) to hold, we need only to observe that
g
1
(x, p
1
(q(y))) g
1
(y, p
1
(q(x))) 0 for all x, y > 0.
Combining the above two corollaries, we immediately arrive at the following
conclusion
Corollary 3.5.4 For model (3.3.2) ((3.3.3)), if a
1
, i.e.,
b
1
/b
7
b
4
b
6
_
b
1
b
2
,
then (G

, I

) is globally asymptotically stable for all 0 and b

5
> 0.
6. Local Stability of Steady State and Stability Switch
Although we have obtained several global stability results, we have yet to study
the local stability systematically. One of our motivations to study the general model
(3.3.1) is to see if an oscillatory solution can exist for certain parameter values. Recall
that both the minimal and dynamic models permit only globally asymptotically stable
positive steady states. We would like to know if and how this may change for more
realistic models. To this end, we need to obtain the characteristic equations associated
to our models.
Consider rst model (3.3.1). Let
G
1
(t) = G(t) G

, I
1
(t) = I(t) I

. (3.6.1)
94
then model (3.3.1) is translated to
_

_
G

1
(t) = f(G
1
(t) + G

) g(G
1
(t) + G

, I
1
(t) + I

) + b
7
I

1
(t) = p(I
1
(t) +I

) + q(L((G
1
)
t
+ G

))
(3.6.2)
and has a unique equilibrium point at (0, 0).
Thus the linearized system of (3.6.2) is given by
_

_
G

1
(t) = (f

(G

) + g
x
(G

, I

))G
1
(t) g
y
(G

, I

)I
1
(t),
I

1
(t) = p

(I

)I
1
(t) + q

(G

)L((G
1
)
t
).
For convenience, we still use G(t) and I(t) to represent G
1
(t) and I
1
(t), respec-
tively, and dene
A = f

(G

) + g
x
(G

, I

), B = g
y
(G

, I

), C = q

(G

), D = p

(I

).
Then A, B, C and D are positive. The linearized system of (3.6.2) can be rewritten as
_

_
G

(t) = AG(t) BI(t)

I

(t) = DI(t) + CL(G

t
).
(3.6.3)
It is easy to see that
G

(t) = (A + D)G

(t) ADG(t) BCL(G

t
). (3.6.4)
Denote that
a = A + D, c = AD, d = BC and = b
5
.
Then (3.6.4) can be rewritten as
G

(t) + aG

(t) + cG(t) + dL(G

t
) = 0. (3.6.5)
If L(G
t
) takes the discrete delay form, i.e., L(G
t
) = G(t ), t > 0, then the charac-
teristic equation of (3.6.5) is given by
D()

=
2
+ a + c + de

= 0. (3.6.6)
95
If L(G
t
) takes the form of distributed delay, i.e., L(G
t
) = (1/)
_
t
t
G()d, t > 0, then
the characteristic equation of (3.6.5) is given by

D()

=
2
+ a + c +
d

_
0

d = 0. (3.6.7)
7. Delay Independent Stability Results for Discrete Delay Model
In this section we consider only the case of discrete delay in model (3.3.1) and
therefore the results are applicable to model (3.3.2). We will need theorem B stated in
Section 6 of Chapter 2 (Theorem 3.1 on page 77 in [54], 1993). First, we have
Lemma 3.7.1 In (3.6.6), the stability of (0, 0) is determined as follows.
Case (1). If AD > BC, then the stability of (0, 0) does not change as 0 is
increasing;
Case (2). If AD BC, then the stability of (0, 0) can at most change once from
stable to unstable, i.e., if (0, 0) is stable, when = 0, then (0, 0) becomes unstable when

0
> 0 for some
0
> 0; if (0, 0) is unstable when = 0, (0, 0) remains unstable for
all > 0.
Proof Compare (3.6.6) with (2.6.1). We have
= 0, a = A + D, b = 0, c = AD, d = BC.
Thus,
b
2
+ 2c a
2
2d = 2c a
2
= 2AD (A + D)
2
= (A
2
+D
2
) 0.
Hence, (A) is violated, and hence the case (III) in (Theorem B in Section 6 of Chapter
2). That is, (0, 0) cannot have multiple stability switches.
96
Since A, B, C, D > 0, we see that c
2
> d
2
(c
2
d
2
) is equivalent to c > d (c d).
Thus we proved Case (1) and Case (2).
From Lemma 3.7.1, we have
Theorem 3.7.1 For (3.3.1), we have the following results on the local stability of (0, 0).
Case (i). If
(f

(G

) + g
x
(G

, I

))p

(I

) g
y
(G

, I

)q

(G

),
then (G

, I

) has at most one stability switch as 0 increases.

Case (ii). If
(f

(G

) + g
x
(G

, I

))p

(I

) > g
y
(G

, I

)q

(G

),
then stability of (G

, I

) does not change for any 0.

Proof Notice that (3.6.3) is the linearized system of (3.3.1) at (G

, I

), and A =
f

(G

) + g
x
(G

, I

), D = p

(I

), B = g
y
(G

, I

), C = q

(G

).
Corollary 3.7.1 For model (3.3.2), the local stability of (G

, I

) can be determined as
follows.
Case (i). If >
1
2
(11 + 5

5)a
1
, then there is at most one stability switch of
(G

, I

). Specically, if (G

, I

) is stable when b
5
= 0, then there exists
0
> 0 such that
(G

, I

) is stable for b
5
[0,
0
) and unstable for b
5

0
; if (G

, I

) is unstable when
b
5
= 0, then (G

, I

) is unstable for all b

5
> 0.
Case (ii). If
1
2
(11 + 5

5)a
1
, then the stability of (G

, I

) does not change

for all b
5
0.
97
Proof By Theorem 3.7.1, we need only to check the sign of
(f

(G

) + g
x
(G

, I

))p

(I

) g
y
(G

, I

)q

(G

). (3.7.1)
Notice that
G

= 2b
7
_
((b
1
b
7
) +
_
(b
1
b
7
)
2
+ 4b
7
(b
1
+ b
4
b
6
/b
2
)), and I

=
b
6
b
2
G

, (3.7.2)
and
f

(x) = b
1
, p

(x) = b
2
, q

(x) = b
6
,
g
x
(x, y) =
b
4
y
(x + 1)
2
, g
y
(x, y) =
b
4
x
x + 1
.
Let

= b
4
b
6
/b
2
(= b
1
).
Then (3.7.1) becomes
b
2
_
b
1
+
G

(G

+ 1)
2
_

b
4
b
6
G

+ 1
=
b
2
(G

+ 1)
2
(b
1
(G

+ 1)
2
G
2
)
=
b
2
(G

+ 1)
2
(
_
b
1
(G

+ 1) +
_

)(
_
b
1
(G

+ 1)
_

).
Let
w()

=
_
b
1
(G

+ 1)
_

=
_
b
1
(1 + (

=
_
b
1
w
1
().
Then sign (5.3) = sign w
1
(). We have
w
1
() = 1 + (

= 1 +
2b
7

)
b
1
b
7
+
_
(b
1
b
7
)
2
+ 4b
1
b
7
( + )
=
_
(b
1
b
7
)
2
+ 4b
1
b
7
( + ) + (b
1
+b
7
2b
7

)
b
1
b
7
+
_
(b
1
b
7
)
2
+ 4b
1
b
7
( + )
.
98
Let
v() =
_
(b
1
2b
7
)
2
+ 4b
1
b
7
( +) + (b
1
+ b
7
2b
7

)
then sign (5.3) = sign v().
v() =
_
(b
1
+ b
7
)
2
+ 4b
1
b
7
+ (b
1
+ b
7
2b
7

)
=
4b
1
b
7
+ 4b
1
b
7

+ 4b
2
7

3/2
4b
2
7

_
(b
1
+ 2b
7
)
2
+ 4b
1
b
7
(b
1
+ b
7
2b
7

)
=
4b
2
7

(
3/2

+a
1

+ a
1

)
_
(b
1
+ b
7
)
2
+ 4b
1
b
7
(b
1
+ b
7
2b
7

)
.
Let =

and
J() =
3

2
+a
1
+ a
1

, 0.
Then
sign (5.3) = sign v() = sign J().
Notice that
J

() = 3
2
2

+ a
1

J
= 4 12a
1
= 4( 3a
1
)
J

() = 0 gives two extreme points if and only if

J
> 0. So if
J
0, J() > 0. If

J
> 0, J() has the possibility to assume negative value.
Assume now that
J
> 0, i.e., 3a
1
> 0. We shall nd the minimum value of
J(), 0. Solving J

() = 0, we obtain

1,2
=
1
6
(2

3a
1
) =
1
3
(

3a
1
).
Clearly
0
=
1
3
(

3a
1
) is the minimum point of J(), 0.
J(
0
) = a
1

+
0
(
2
0

0
+ a
1
)
99
= a
1

+
1
27
(

3a
1
)( + 2
_
( 3a
1
) + 3a
1
3 3
_
( 3a
1
) + 9a
1
)
= a
1

+
1
27
(

3a
1
)(
_
( 3a
1
) + 6a
1
)
= a
1

+
1
27
(2

+ 9a
1

2( 3a
1
)

3a
1
)
=
1
27
(36a
1

2( 3a
1
)

3a
1
)
=
2
27
((18a
1
)

( 3a
1
)

3a
1
)
0 if 18a
1
.
Assume 3a
1
< < 18a
1
, then
J(
0
) =
2
27
(18a
1
)
2
( 3a
1
)
2
( 3a
1
)
(18a
1
)

+ ( 3a
1
)

3a
1
.
Let h(a
1
, )

= (18a
1
)
2
( 3a
1
)( 3a
1
)
2
, then sign J(
0
) = sign h(a
1
, ). We
have
h(a
1
, ) = (15a
1
( 3a
1
))
2
( 3a
1
)( 3a
1
)
2
= 27a
1
(
2
11a
1
a
2
2
)
= 27a
1
_
+
11 + 5

5
2
a
1
__

11 + 5

5
2
a
1
_
< 0 i >
11 + 5

5
2
a
1
.
This completes the proof.
8. Delay Dependent Stability Conditions
The stability results in the previous section do not depend on the values of the
delay(b
5
). However, they do suggest that for some parameter values(delay is included),
the positive steady state may become unstable. It is thus interesting to locate such
100
parameter values. Such a task turns out to be very complex. Instead, we shall single
out parameter region where no stability switch can take place. Specically, we would
like to obtain some upper bound on the delay length(while holding other parameters
steady) for the positive steady state to remain stable. Specically, we shall seek the
relationship of such upper bound with the value of . We shall consider both model
(3.3.2) and (3.3.3). Recall that the following notations in Section 6,
A = f

(G

) + g
x
(G

, I

), B = g
y
(G

, I

), C = q

(G

), D = p

(I

).
and
a = A + D, c = AD, d = BC, = b
5
.
Dene
H() =
f

(G

) + g
x
(G

, I

) + p

(I

)
g
y
(G

, I

)q

(G

)
.
8.1. The case of discrete delay. We shall consider here the characteristic
equation (3.6.6) below. Recall that
D() =
2
+ a + c + de

= 0.
If there exists a
0
> 0 such that the trivial solution of (3.6.3) is unstable for all
0
,
then there must be an > 0 such that D(i) = 0. Thus

2
+ ai + c +d(cos i sin ) = 0,
and hence
_

2
= c + d cos ,
a = d sin .
(3.8.1)
Since a > 0 and > 0, (3.8.1) implies that a/d. This leads to the following
101
Theorem 3.8.1 In linear equation (3.6.5), if L(G
t
) = G(t ), t > 0, > 0, and
< a/d, then the trivial solution of (3.6.5) is globally asymptotically stable.
Applying theorem 3.8.1 to model (3.3.2), we have
Corollary 3.8.1 In model (3.3.2), if L(G
t
) = G(t ), t > 0, > 0, and < H(),
then the equilibrium point (G

, I

) is locally asymptotically stable.

8.2. The case of distributed delay. We now consider the characteristic equa-
tion (3.6.7) of linear equation (3.6.5) for the case of L(G
t
) =
1

_
t
t
G()d, t > 0, > 0,

D() =
2
+ a + c +
d

_
0

d = 0. (3.8.2)
If the trivial solution of (3.6.5) is unstable for some > 0, then there exist u 0 and
v > 0 such that = u +iv is a solution of (3.8.2), i.e.,
(u + iv)
2
+a(u + iv) + c +
d

_
0

e
u
(cos v + i sin v)d = 0.
Thus
u
2
v
2
+ au + c +
d

_
0

e
u
cos vd = 0, (3.8.3)
and
2uv + av +
d

_
0

e
u
sin vd = 0. (3.8.4)
Since v > 0, we see that (3.8.4) implies
2u + a =
d

_
0

e
u
sin v
v
d.
Therefore
2u +a
d

_
0

e
u

sin v
v
d

_
0

||d =
1
2
d.
Hence we have u
1
2
_
1
2
d a
_
. This leads to the following
102
Theorem 3.8.2 In linear equation (3.6.5), if L(G
t
) = (1/)
_
0

G(t + )d, t > 0,

> 0 and < 2a/d, then the trivial solution of (3.6.5) is globally asymptotically stable.
From Theorem 3.8.2, we have
Corollary 3.8.2 In model (3.3.2), if L(G
t
) = (1/)
_
0

G(t + )d, t > 0, > 0, and

< 2H(), then the equilibrium point (G

, I

) is locally asymptotically stable.

8.3. Expression of H(). In the following, we shall give the explicit expression
of H() for the Model (3.3.2) and (3.3.3). This is useful in applications and in planning
computer simulations. Recall that Model (3.3.1) reduces to the Model (3.3.2) ((3.3.3))
if
f(x) = b
1
x, g(x, y) =
b
4
xy
x + 1
, p(x) = b
2
x, and q(L(u
t
)) = b
6
L(u
t
), u
t
C.
In this case, we have
f

(x) = b
1
, g
x
(x, y) =
b
4
y
(x + 1)
2
, g
y
(x, y) =
b
4
x
x + 1
, p

(x) = b
2
, x 0
and q

(L(u
t
)) = b
6
for u
t
C.
Recall that Model (3.3.2) ((3.3.3)) assumes the existence of an unique steady
state (G

, I

) in R
2
+
, where G

and I

are given by (5.4). Let S() = +

1
G

, then
S() = +
1
2b
7
_
(b
1
b
7
) +
_
(b
1
b
7
)
2
+ 4b
7
(b
1
+b
4
b
6
/b
2
)
__
=
1
2b
7
_
(b
1
+ b
7
) +
_
(b
1
+ b
7
)
2
+ 4b
4
b
6
b
7
/b
2
_
.
Therefore
H() =
f

(G

) + g
x
(G

, I

) + p

(I

)
g
y
(G

, I

)q

(G

)
=
b
1
+ b
2
+ (b
4
I

/(G

+ 1)
2
)
(b
6
b
4
G

/G

+ 1)
103
=
b
1
+ b
2
b
4
b
6
G

+ 1
G

+
1
b
2
(G

+ 1)
=
b
1
+ b
2
b
4
b
6
S() +
(G

)
1
b
2
S()
.
Let = (b
1
+ b
7
)
2
+ 4b
4
b
6
b
7
/b
2
. Notice that
(G

)
1
S()
=
(b
1
b
7
) +
_
(b
1
b
7
)
2
+ 4b
7
(b
1
+ (b
4
b
6
/b
2
))
(b
1
+ b
7
) +

=
[(b
1
b
7
) +

][(b
1
+ b
7
)

]
4b
4
b
6
b
7
/b
2
=
b
2
4b
4
b
6
b
7
[ + (b
1
b
7
)

(b
1
+ b
7
)

(b
2
1
b
2
7

2
)]
=
b
2
2b
4
b
6
b
7
[b
2
7

2
b
7

+b
1
b
7
+ 2b
4
b
6
b
7
/b
2
].
Thus,
H() =
b
1
+ b
2
2b
4
b
6
b
7
[(b
1
+ b
7
) +

]
+
1
2b
4
b
6
b
7
[b
2
7

2
b
7

+ b
1
b
7
+ 2b
4
b
6
b
7
/b
2
]
=
1
2b
4
b
6
b
7
[b
2
7

2
+ (b
1
+ b
2
b
7
)

+(2b
1
+b
2
)b
7
+ 2b
4
b
6
b
7
/b
2
].
9. Numerical Simulations
In a clinic study reported in [23], ten healthy volunteers participated(5 males
and 5 females). All of them had negative family and personal histories for diabetes
mellitus and other endocrine diseases, were on no medications and had maintained
a constant weight for the six months preceding that study. For detailed experiment
description, see [23]. They were able to show that the dynamic model does produce
solutions that t well with the data collected from that experiment. The parameter
values for these individuals are listed in their Table 1 ([23]). Using these parameters
and Corollary 3.5.4, we found that for all these persons except the sixth person (female)
104
Table 3.9.1. Parameters for subjects 6 and 7 in IVGTT Models (b
5
= 23min.)
P G
b
I
b
b
0
b
1
b
2
b
3
b
4
b
6
b
7
U
mg
dl
pM
mg
dl
1
min
1
min
dlpM
mg
1
minpM
dlpM
mgmin
mg
dlmin
6 88 68.6 209 2E-4 0.0422 1.64 1.09E-4 0.033 0.68
7 87 37.9 311 1E-4 0.2196 0.64 3.73E-4 0.096 1.24
and the seventh person (male), and all values of and delay lengths, solutions always
tend to the positive steady state for both models (3.3.2) and (3.3.3). Since one of our
main objectives is to see if unstable positive steady state is possible for our models and
some of these ten persons, and our theoretical results failed to exclude such possibility,
we thus focused our simulations on the subjects 6 and 7 (sixth and seventh persons).
The relevant parameter values (obtained through the above mentioned clinic study) for
these two persons are summarized in table 3.9.1 (De Gaetano and Arino [23]), where P
stands for parameters, U stands for units and 6, 7 stand for subject 6, 7, respectively..
Even for subjects 6 and 7, according to our extensive simulation work, the steady
state is globally asymptotically stable in all practically meaningful values of time delay
= b
5
and . Nevertheless, we did nd that the positive steady state can be unstable(if
a
1
< and (a
1
, ) provided the delay is large enough). Figures in gure 3.9.1
illustrate such instability with nontrivial periodic solutions for the discrete delay model
(3.3.2), using the data (except b
5
and ) given in the table 1 for subjects 6 and 7. For
the discrete delay model (3.3.2), we need the time delay to be as close to 550 minutes
in order to observe sustainable oscillatory solutions for both subjects 6 (550) and 7
(600). The values of are 0.01 and 0.05 for subject 6 and subject 7, respectively. Since
for both subjects, the actual delay length is only 23 minutes, we believe that it is very
unlikely to observe any sustainable oscillatory solutions in real life experiments for these
105
1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2
x 10
4
0
50
100
150
200
250
300
time t
G
, I
0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1.3 1.4 1.5
x 10
4
0
100
200
300
400
500
600
700
800
900
time t
G
, I
Figure 3.9.1. Periodic solutions for the discrete delay model (3.3.2) for subject 6 and 7
Left: The large amplitude curve (solid line) is G, the other one (dash-dot line) is I. Here = 0.01,
and b
5
= 550. Right: The large amplitude curve (solid) is G, the other one (dash-dot line) is I. Here
= 0.05, and b
5
= 600.
subjects. In other words, the global asymptotical results reported in [23] is conrmed
at least practically by the more general model (3.3.1).
10. Discussion
From a purely theoretical point of view, we can still make a few insightful state-
ments. For both very large or small values of , the positive steady state is believed to
be globally asymptotically stable for any conceivable person. For relatively small and
large delay, oscillatory solutions become possible. This indicates that it is important to
have in the model, since the value of is most likely small, but not too small. Indeed,
from the simulation results (Figure 3.9.1), we see is in the range of 0.01 to 0.05, which
translates into a range of 20 to 100 for half saturation value of G. This range is very
close or comparable to the experiment values of G depicted in gures 2 and 3 in [23],
where G starts at values around 250 and drop down quickly (in about 40 minutes) to
values close to 100. The added work for measuring should not be overwhelming, since
106
more than enough measurements are recorded. The additional work is mathematical.
We would like to point out here that including in the model may very well change
the values of other parameters, in particular that of b
4
.
From Corollary 3.7.1, we know that the positive steady state may be unstable if
>
1
2
(11 +5

5)a
1
, which is equivalent to b
4
b
6
b
7
>
1
2
(11 +5

5)b
2
1
b
2
. In view of the fact
that the values of b
1
(from 0.0001 to 0.0565) and b
4
(from 3.51E-08 to 3.73E-04) change
signicantly from one person to another, we see that the likely candidates for having
sustainable oscillatory glucose and insulin levels are subjects that have high values of
b
4
(insulin-dependent glucose disappearance rate) and low values of b
1
(the spontaneous
or insulin-independent glucose rst order disappearance rate). That is, those subjects
who can not process their glucose quick enough when insulin level is low but respond
very well with added insulin.
Roughly speaking, when the delay is short (less than 60 minutes), we probably
will not see any sustainable oscillatory solutions. However, for large enough delays, this
may be possible. In such cases, the larger the half-saturation constant (i.e., 1/), the
more likely that the steady state is to be unstable. Even when the delay is small and the
steady state is stable, the solutions may converge to the steady state in an oscillatory
way. This can give rise to subsequent peaks of glucose and insulin, as observed in the
experiment of De Gaetano and Arino [23].
If = 0, then we know the positive steady state is always locally stable. If
> 0, the positive steady state remains stable for small and medium delays. Only
for large values of delay and appropriate values of , the positive steady state may
lose its stability. Therefore, delay or the saturating glucose uptake functional response
(G/(G + 1)) alone will not destabilize the steady state. In other words, for model
107
(3.3.1), oscillatory solutions resulted from delayed suitable nonlinear glucose uptake
mechanism. Refer [79], [84] and Chapter 2 and Chapter 4. This is in striking contrast
to the well known predator-prey dynamics (see [54]), where either alone will be enough
to produce periodic solutions.
A potential application of our study here is to nd better ways of delivering
insulin and timing of the intake of glucose. Previous studies are largely done on linear
compartment models ([80], [81]) or on the minimal type models [36]. Our theoretical and
simulation work shows that in clinic applications, we can safely assume that after about
40 minutes, solutions are close and stay close to steady state levels. Another possible
usage of our work is to design eective ways to estimate the involved parameters for
clinic applications. For minimal model, such eort is documented in [24]. The recent
development of PET technology provides a possibility of eectively monitor subjects
glucose and insulin levels through noninvasive method [18]. This should be very helpful
in estimating parameters for individuals in order to design proper controls for their
glucose and insulin levels.
CHAPTER 4
The Eects of Active -Cells: A Preliminary Study
1. Introduction
It is known that the cells are the only hormone that secretes insulin ([89]).
Dysfunctional cells can lead to diabetes ([9]) due to the plasma glucose can not
be utilized suciently and cause hyperglycemia. Hyperglycemia is known to induce
insulin resistance ([71]) and a decit in the mass of cells, reduced insulin secretion
[53]. This supports the hypothesis that a primary insulin secretory defect that causes
hyperglycemia could lead to insulin resistance and diabetes via increased glucose levels
([9]). It is believed that insulin resistance can cause -cell defects, and hence diabetes
either by overworking the cells or by toxic eects of hyperglycemia on the cells ([9]).
The cells are contained in the Langerhans islets through out the pancreas.
(Refer to Figure 3.1 for Langerhans islet.) The insulin is contained in the granules in
the cells. There are about one million of the Langerhans islets and each islet contains
approximately three hundreds of cells. Each cell contains around one thousand of
granules. The distribution is very complex ([85]).
The formation and growth (via neogenesis or replication) and death of the cells
are still not completely clear ([9], [35], [28], [85] and [12]). It is believed that the cells of
each individual, shortly after his/her birth, do not replicate. When a cell is damaged,
109
it will not be able to function as it did originally. If ones cells are damaged in large
amounts, he/she would have to suer diabetes due to cell dysfunction. However, this
hypothesis is challenged by S Bonner-Weir [12] in 2000. The new perspective ([12],
2000) assumes that the -cells can be replicated and neogenesis, which is supported by
in vivo and in vitra experiments ([13], [14], [34], [82] and [49]).
The purpose of this chapter is to study the eects of the -cell mass to the
glucose-insulin endocrine regulatory system. Due to it is still not clear if the -cells can
replicated or neogenerated (for example, from stem cells) ([28], [12]), we consider only
the active -cell mass in the dynamics. Next section will introduce the current research
status and we present a model taking active -cell mass as a factor in the glucose-insulin
regulatory system in Section 3. In Section 4 we will exhibit our numerical simulation.
Then we will discuss our simulation results in Section 5.
2. Current Research Status
The rst study on -cell mass kinetics was given in 1995 by D. T. Finegood,
L. Scaglia and S. Bonner-Weir ([35], 1995). Based on experimental data, a simple
single variable (-cell mass) system was developed as follows to study the cell mass
growth and renewal (neogenesis or replication) vs. death.

(t) = REP(t) +NEO(t) DEATH(t)

where
REP(t) - replication rate = 16.10e
0.065t
+ 2.31
NEO(t) - islet neogenesis rate
110
DEATH(t) - death rate.
The net growth or death rate is given by
NEO(t) DEATH(t) =

(t) 16.10e
0.065t
+ 2.31
The increased -cell death rate is consistent with experiment reports of increased islet
cell apoptosis ([9] and its references). However, this model consists of only cell mass as
its variable. The -cell mass is related to the glucose concentration level ([82], [49] and
[85]) and the glucose concentration level is regulated by the insulin secreted from the
celss. Obviously it is necessary to consider glucose, insulin and -cell mass together
as a whole system when study the pathways to the Type 2 diabetes which is caused by
-cell dysfunction and insulin resistance ([9], [85] and their references).
In 2000, B. Topp, K. Promislow, G. De Vries, R. M. Miura and D. T. Finegood
([85]) introduced a novel model, according to Mari ([60]), taking cell mass as a
dynamically growing variable together with the glucose and insulin concentrations. The
model takes following form.
_

_
G

(t) = R
0
E
G
0
G(t) S
I
I(t)G(t),
I

(t) =
(t)G
2
(t)
+G
2
(t)
kI(t),

(t) = (d
0
+ r
1
G(t) r
2
G
2
(t))(t)
(4.2.1)
where
S
I
(mlU
1
d
1
) is the total insulin sensitivity.
E
G0
(d
1
) is the total glucose eectiveness at zero insulin.
111
Table 4.2.1. Parameters of the Model 4.2.1
Parameters Values Units
S
I
0.72 mlU
1
d
1
E
G0
1.44 d
1
R
0
864 mgdl
1
43.2 Uml
1
d
1
2000 mg
2
dl
2
Parameters Values Units
k 432 d
1
d
0
0.06 d
1
r
2
0.24x10
5
mg
2
dl
2
d
1
r
1
0.84x10
3
mg
2
dl
2
d
1
R
0
(mgdl
1
) is the net rate of production at zero glucose.
The term
(t)G
2
(t)
+G
2
(t)
indicates the insulin secretion. All cells are assumed to
secrete insulin at the same maximal rate (Uml
1
d
1
). The term kI stands for
the insulin clearance by the liver, kidneys and insulin receptors at the rate k(d
1
)
The term (d
0
+ r
1
G(t) r
2
G
2
(t)) is the cell growth/death rate which is de-
pendent on glucose concentration G(t). d
0
(d
1
) is the death rate at zero glucose
concentration.
All the parameter values of the model (4.2.1) are listed in Table 4.2.1.
B. Topp, K. Promislow, G. De Vries, R. M. Miura and D. T. Finegood [85] studied
the above model assuming -cell reaction to the insulin-glucose slowly, namely, -cell
dynamics moves slowly and try to understand the complex dynamics regarding -cell
mass, insulin and glucose. This model predicts, by the numerical study results, that
rapidly falling insulin sensitivity leads to the onset of hyperglycaemia and then -cell
dysfunction ([85], [9]). Another result induced from this model is that the increases in
insulin sensitivity mediated by agents such as rosiglitazone prevents the plasma glucose
levels to go higher and thus preserves cell functions from overworking ([85], [9]).
112
3. Active -Cell Model
While the model (4.2.1) exhibits some of the physiological meanings, the model
leaves few places to improve. The authors assume the -cell mass growth-death rate
take a quadratic function due to that the growth rate was shown to be non-linear
([82], [49], [85]. The authors stated that extremely high hyperglycemia may reduce
the replication of the cells ([85]). But the lower glucose concentration level does not
trigger the cells to release insulin. This appears not considered by the GI mode
(4.2.1. We present a new model to consider the eect of -cell mass dynamics in the
glucose-insulin regulatory system. Instead of considering the whole -cell mass, we only
let the healthy and active cells be involved in the our model. This is due to the fact
that if a cell has been damaged, it does not secret insulin even when the glucose
concentration level is high.
_

_
G

= G
in
f
2
(G(t)) f
3
(G(t))f
4
(I(t)) +f
5
(I(t
2
)),
I

= f
1
(G(t
1
))(t) p(I(t)),

= g(G(t
1
)) k(t),
(4.3.1)
where the initial condition I(0) = I
0
> 0, G(0) = G
0
> 0, G(t) G
0
for all t [
1
, 0]
and I(t) I
0
for t [
2
, 0] with
1
,
2
> 0. In model (4.3.1), G (mg/dl/min) and
I (U/ml/min) stands for the glucose and insulin concentration, respectively. The
(mg) stands for the mass of active cells which will release insulin into the bloodstream
when stimulated by the glucose concentration level. Additionally, in the third equation,
the term g(G(t
1
)) mimics that the transient and mild glucose can increase active
113
G
l
u
c
o
s
e

u
t
i
l
i
z
a
t
i
o
n
G
l
u
c
o
s
e

p
r
o
d
u
c
t
i
o
n
Liver converts
glucagon and
glycogen to
glucose
I
n
s
u
l
i
n

p
r
o
d
u
c
t
i
o
n
I
n
s
u
l
i
n

c
l
e
a
r
a
n
c
e
Insulin independent:
brain cells, and
others
Insulin dependent:
fat cells, and
others
Insulin degradation:
receptor, enzyme, and
others
Delay
Delay
Glucose Infusion:
meal ingenstion,
oral intake,
enteral nutrition,
constant infusion
Glucagon
secrete
Glucose Controls
insulin secretion
Glucose Controls
glucagon secretion
Insulin helps cells consume glucose
Insulin secretion
Insulin Controls
Hepatic
glucose production
Glucose
Insulin
Pancreas
Liver
-cells
Active
-cell mass
Figure 4.3.1. Glucose-Insulin with Active -cell Interaction Diagram
The divide lines (dash-dot-dot) indicate insulin controlled hepatic glucose production with time delay;
the dash-dot lines indicate the insulin secretion from the active -cells stimulated by elevated glucose
concentration level with time delay; the dashed lines indicate low glucose concentration level triggers -
cells in pancreas to release glucagon; and the dot line indicates the insulin accelerates glucose utilization
in cells.
114
cell mass, but hyperglycemia (216 mg [85]) will cause cells to overwork and damage
the cells. The term k(t) indicates the active cell mass decreases at the rate k. The
term f
1
(G(t
1
))(t) in the second equation stands for insulin secretion from the
pancreas and is proportional to the active -cell mass. > 0 (mg
1
) is a parameter
indicates how ecient the active -cells release the insulin in to blood stream. Insulin is
stored in -cell granules. Glucose is the primary stimuli of insulin secretion from the
cells. The delay is due to the complex electric processes inside of islet (refer to Chapter
2). The function f
1
(G(t)), in sigmoidal shape, assumes the same condition as that in
the model (2.3.1) in Chapter 2. The term p(I(t)) indicates the insulin degradation and
p(0) = 0, p(x) > 0 and p

(x) > 0 for x > 0. All terms in the rst equation assume the
same physiological meanings and conditions as that in the model (2.3.1) in Section 3 of
Chapter 2. We do not repeat here.
The function g(G) takes the following form and the shape is shown in Figure
4.3.2.
g(G) =
G
5
(1 G/e)
bG + 1
(4.3.2)
where = 1E 6, b = 10E7 and e = 2.16E4.
4. Numerical Simulations
In this section, we show some simulation results of the GI model with a group
of particular functions (2.2.2) (2.2.6) and parameters given in Table 2.2.1, 2.2.2 and
2.7.1.
Through out this section, when we let parameters be xed, they always as-
sume following values.
1
= 5(min),
2
= 36(min), G
in
= 0.54(mg/(dlmin)), d
i
=
115
6000
4000
2000
-2000
0
G
20000 15000 5000 0 10000
8000
Figure 4.3.2. Function g(G) in GI-Model
where
1
= 1E 6, b = 10E7, and e = 2.16E4.
0.03846(U/(mlmin), k = 0.1(mg/min), and = 0.005(mg
1
).
4.1. Insulin Response Delay and Hepatic Glucose Production Are Crit-
ical for Sustain Insulin Secretion Oscillations. Insulin response time delay and
hepatic glucose production are critical for insulin secretion oscillation to sustain. Our
numerical simulation (Figure 4.4.1) indicate that the insulin secretion oscillations do
not take place if (a) the insulin response time delay is assumed to be zero (
1
= 0) or
(b) there is no hepatic production (f
5
0).
4.2. Insulin Response Time Delay
1
as a Bifurcation Parameter. In
last paragraph our simulation shows the insulin response time delay
1
is critical to the
insulin secretion oscillations to sustain. In this paragraph, we take
1
as a bifurcation
parameter and let
1
change from 0 to 20 with step size 0.01 (min.) and try to nd out
the bifurcation point at which a periodic solution can be bifurcated out. We let
2
= 36,
G
in
= 0.54, d
i
= 0.03846, k = 0.1, and = 0.005 be x and let
1
= 5 change from 0.01
116
0
20
40
60
80
100
0
100
200
300
400
0
2
4
6
8
10
12
14
16
18
G
beta
I
0
20
40
60
80
100
0
100
200
300
400
0
5
10
15
20
G
beta
I
Figure 4.4.1. Orbits of (G, I, ) of GI model
Left: the case
1
= 0; right: the case f
5
0. These indicate that both hepatic production and insulin
response looks very necessary to the sustained oscillation.
to 20 (min). We observe from our numerical simulation (shown in Finger 4.4.2) that
when the delay
1
changes from 0 to 20, when
1
is suciently small (
1
< 1.5), the
steady state of the model (4.3.1) is stable. When
1
increases large enough (
1
> 2), the
steady state becomes unstable and the model (4.3.1) has a stable periodic solution the
insulin oscillation takes place. That means, there is one bifurcation point
10
(1.5, 2.0)
for insulin response time delay parameter
1
such that the steady state is always unstable
for
1
>
10
and thus the insulin oscillation sustains. The left of Figure 4.4.2 shows the
bifurcation diagram when
1
changes from 0 to 20 and the right gure is the periods
of the periodic solutions when
1
changes. Approximately speaking, the longer delay of
the insulin response to the glucose, the larger the period of the secretion oscillation.
4.3. Glucose Infusion Rate G
in
as a Bifurcation Parameter. Let the
glucose infusion rate G
in
vary from 0 to 3.0 (mg/dlmin) and all other parameters
be xed. The numerical simulation indicates that there exists a bifurcation point
G
0
in
(1.65, 1.75) such that the steady state is unstable when G
in
[0, G
0
in
] and stable
117
0 2 4 6 8 10 12 14 16 18 20
0
50
100
150
0 2 4 6 8 10 12 14 16 18 20
0
50
100
0 2 4 6 8 10 12 14 16 18 20
0
500
1000
tau
1
0 2 4 6 8 10 12 14 16 18 20
0
50
100
150
200
250
P
e
r
i
o
d

o
f

p
e
r
i
o
d
i
c

s
o
l
u
t
i
o
n
s
tau
1
Figure 4.4.2. Bifurcation diagram of
1
[0, 20]
Left: bifurcation diagrams (top - glucose; middle insulin and bottom active -cell mass) of
1

[0, 20] (min); right: periods of periodic solutions
when G
in
> G
0
in
. This shows that if the glucose infusion rate is not too high, then
the insulin oscillations sustain. When the infusion rate is large enough, the oscillations
disappear. This conrms again that the insulin secretion ultradian oscillations do not
take place in the intravenous glucose tolerance test (IVGTT) as a big bolus (0.33 mg
per kg of body weight [22] of glucose infusion. See the left gure in Figure 4.4.3.
Another agreement to the physiologic glucose-insulin regulatory system is exhib-
ited in the right hand side gure in Figure 4.4.3. The top line is the time dierences
(11 to 13 minutes) between the peaks of the glucose concentration and the peaks of the
active -cell mass. The middle line is the time dierences (5 to 7 minutes) between the
active -cell mass and the glucose concentration. The bottom line is the time dierences
(18 to 20 minutes) between the active -cell mass and the insulin concentration. This
can be restated as follows. When the glucose infusion rate G
in
varies in the range that
the insulin secretion oscillation sustain, in one cycle, the active cells mass reaches its
peak about 11 to 13 minutes after the glucose concentration reaches its peak. Then it
takes about 5 to 7 minutes that the insulin concentration reaches its peak. Therefore
118
0 50 100 150 200 250 300
0
50
100
150
0 50 100 150 200 250 300
0
20
40
60
B
i
f
u
r
c
a
t
i
o
n

d
i
a
g
r
a
m
0 50 100 150 200 250 300
0
20
40
60
80
Gin 0 50 100 150 200 250 300
0
2
4
6
8
10
12
14
16
18
20
Figure 4.4.3. Bifurcation diagram of G
in
[0, 3.0]
Left: bifurcation diagram when G
in
changes from 0 to 3.0(mg/dlmin). If G
in
< 1.65, the oscillations
are sustained. Right: in one cycle, the active cells mass reaches its peak about 11 to 13 minutes
after the glucose concentration reaches its peak (the middle line). Then it takes about 5 to 7 minutes
that the insulin concentration reaches its peak (the bottom line). Therefore it takes around 18 to 20
minutes from increased glucose concentration to the insulin is released (the top line).
it takes around 18 to 20 minutes from increased glucose concentration to the insulin is
released.
Several periodic solutions are shown in the left gure in Figure 4.4.4 when G
in
changes from 0 to 3.00 (mg/dlmin) while other parameters are xed. The right hand
side gure shows the periods of periodic solutions decrease from around 138 minutes to
110 minutes when the G
in
increases from 0 to G
0
in
.
4.4. Peaks of Oscillations in One Cycle. Our simulation results exhibit that
in one oscillation cycle, the glucose concentration peaks before the active -cell mass
peaks; the active -cell mass peaks before the insulin concentration peaks. On the other
hand, the glucose concentration bottoms before the active -cell mass does; the active
-cell mass bottoms before the insulin concentration does. Figure 4.4.5 exhibits the
shifts among these three variables.
For example, the gure in the right hand side of Figure 4.4.3 demonstrates,
119
30
40
50
60
70
80
90
100
110
120
0
5
10
15
20
25
30
35
40
45
50
0
500
1000
G
I
b
e
t
a
0 50 100 150 200 250 300
0
20
40
60
80
100
120
140
Gin
P
e
r
i
o
d

o
f

p
e
r
i
o
d
i
c

s
o
l
u
t
i
o
n
s
Figure 4.4.4. Periodic solutions and periods when G
in
[0, 3.0]
Left: periodic solutions; right: periods of periodic solutions. The periods decreases from 138 to 110
(min) along with the increase of G
in
from 0 to G
0
in
(1.65, 1.75) (mg/(dlmin)).
when the glucose infusion rate G
in
changes from 0 to 3.00 (mg/(dlmin)), the glucose
concentration peaks around 11-13 minutes before the active -cell mass peaks and the
active -cell mass peaks around 5-7 minutes before the insulin concentration peaks.
This is in agreement with the physiological stimulation chain, that is, increased
glucose concentration level triggers active cells to release insulin, and then the insulin
helps the cells to utilize the plasma glucose. These shifts among the peaks also reect
the time delay of the insulin response to increased glucose concentration.
4.5. -cell Deactivation Rate k [0.01, 2] as a Bifurcation Parameter.
The parameter k is the rate at which the -cell deactivated. After the cells release
the insulin contained in the granules, the cells have to reproduce insulin before next
secretion. We call this as -cell deactivation. From the Figure 4.4.6, we can observe
that when k [0.01, 2] and
1
= 5,
2
= 36, G
in
= 0.54, d
i
= 0.03846 and = 0.005,
the insulin oscillation always sustain but the amplitudes and the periods of the periodic
solutions vary along with the changes of the -cell deactivation rate parameter k from
120
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
0
50
100
150
m
g
/
d
l
G
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
0
500
1000
1500
m
g
beta
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
0
50
100
150
m
U
/
m
l
time (min)
I
2050 2100 2150 2200
0
50
100
150
200
250
300
350
400
450
500
time (min)
Figure 4.4.5. Peaks of Oscillations in One Cycle
Left: a periodic solution of Model (4.3.1) when
1
= 5(min),
2
= 36(min), G
in
= 0.54(mg/(dl min)),
d
i
= 0.03846(U/(ml min)), k = 0.1(mg/min) and = 0.005. G - top line; I - bottom line; and -
middle line. Right: In one cycle, the glucose peaks before active -cell mass peaks (middle line) and
the active -cell mass peaks before Insulin concentration peaks (bottom line). (G and I are recalled
for comparison.)
0.01 to 2. It appears that the best range for k is around 0.05 so that the glucose
concentration level of subjects is within the normal range.
4.6. Parameter as a Bifurcation Parameter. We take the parameter
(mg
1
) as the stimulation eciency coecient as its value controls how much insulin
is released from cell mass. According to our numerical simulation result (shown
in the left gure in Figure 4.4.7), The larger the parameter , the higher the insulin
concentration, the less need of the -cell mass and the lower the glucose concentration.
The other gure in Figure 4.4.7 indicates when the parameter changes from 0.001
to 0.1, the periods of periodic solutions increase from 119 minutes to 128 minutes
approximately. In particular, the increase of the periods near = 0.0001 is dramatic.
The Figure 4.4.8 simply shows, from two dierent directions, several limit cycles in this
parameter range.
121
0 0.5 1 1.5 2 2.5
0
50
100
150
0 0.5 1 1.5 2 2.5
0
10
20
30
0 0.5 1 1.5 2 2.5
0
200
400
600
800
0 0.5 1 1.5 2 2.5
80
100
120
140
0 0.5 1 1.5 2 2.5
80
100
120
140
P
e
r
i
o
d

o
f

p
e
r
i
o
d
i
c

s
o
l
u
t
i
o
n
s
0 0.5 1 1.5 2 2.5
80
100
120
140
Figure 4.4.6. Bifurcation diagram of k [0.01, 2]
Left: bifurcation diagram; right: periods of periodic solutions
0 0.002 0.004 0.006 0.008 0.01 0.012
0
50
100
150
200
0 0.002 0.004 0.006 0.008 0.01 0.012
0
10
20
30
40
B
i
f
u
r
c
a
t
i
o
n

d
i
a
g
r
a
m
0 0.002 0.004 0.006 0.008 0.01 0.012
0
50
100
150
200
sigma
0 0.002 0.004 0.006 0.008 0.01 0.012
119
120
121
122
123
124
125
126
127
128
sigma
P
e
r
i
o
d

o
f

p
e
r
i
o
d
i
c

s
o
l
u
t
i
o
n
s
Figure 4.4.7. Bifurcation diagram of [0.0001, 0.1]
Left: bifurcation diagram; right: periods of periodic solutions
122
20
40
60
80
100
120
140
0
10
20
30
40
0
200
400
600
800
1000
1200
I
G
b
e
t
a
20
40
60
80
100
120
140
0
5
10
15
20
25
30
35
0
200
400
600
800
1000
1200
I G
b
e
t
a
Figure 4.4.8. Limit Cycles when [0.0001, 0.1]
0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1 0.11
0
50
100
150
0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1 0.11
0
10
20
30
0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1 0.11
0
200
400
600
di
30 40 50 60 70 80 90 100 110
0
5
10
15
20
25
0
100
200
300
400
500
600
G
I
b
e
t
a
Figure 4.4.9. There is no bifurcation when d
i
[0.005, 0.01]
Left: the amplitude of the periodic solutions; right: the limit cycles
4.7. The Changes of Insulin Degradation Rate d
i
[0.025, 0.1] Do Not
Aect the Oscillations. Similar to the parameter [0.0001, 0.1], the changes of
insulin degradation rate d
i
[0.025, 0.1] has no impact on stability of the stead state.
The insulin oscillations sustain. (Refer to Figure 4.4.9.)
123
5. Discussion
In this chapter, we tried to model the glucose-insulin endocrine regulatory system
involving the active -cell mass. we performed numerical simulations for the model
(4.3.1) and have following observations.
We conrmed that both the insulin response delay and the hepatic glucose pro-
duction are critical to the oscillatory insulin secretion from the active cells.
We observed that the insulin response time delay needs to be big enough (e.g.,

1
2 for the insulin secretion oscillations to be sustained. Besides, the longer
delay of the insulin response to the glucose, the larger the period of the oscillation.
The glucose infusion rate has to be moderate to observe the insulin secretion
oscillations. When the Glucose infusion rate is extremely high, for example G
in
=
2.16 mg/(dlmin), the insulin secretion becomes damped. This conrms again
that in IVGTT the hepatic glucose production is insignicant comparing to the
large glucose infusion.
We observed that the glucose concentration peaks around 11-13 minutes before
the active -cell mass peaks and the active -cell mass peaks around 5-7 minutes
before the insulin concentration peaks when G
in
changes. This is in agreement
with the physiological stimulation chains. Last, according to our numerical sim-
ulations, in the physiologically meaningful ranges, the insulin degradation rate
d
i
[0.025, 0.1], the stimulation eciency coecient [0.0001, 0.1] and the -
cell deactivation rate k [0.01, 2] do not aect the insulin secretion oscillations.
According to simulations we performed, no insulin secretion pacemaker is ob-
served for the ultradian oscillations. More investigation is need to nd out if the -
124
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
40
60
80
100
m
g
/
d
l
G
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
0
100
200
300
m
g
beta
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
5
10
15
m
U
/
m
l
time (min)
I
3600 3800 4000 4200 4400 4600 4800 5000
74.9611
74.9611
74.9611
74.9611
m
g
/
d
l
G
3600 3800 4000 4200 4400 4600 4800 5000
206.165
206.17
206.175
206.18
m
g
beta
3600 3800 4000 4200 4400 4600 4800 5000
8.2859
8.2859
8.2859
8.2859
m
U
/
m
l
time (min)
I
Figure 4.5.1. Possible -cell pulsatile oscillation?
Left: solutions of Model (2.3.1 when
1
= 5,
2
= 36, G
in
= 0.54, d
i
= 0.03846, k = 0.1, and
= 0.005); right: Zoomed in gure which shows the active -cell mass oscillates in a period around
50 minutes while the glucose concentration G and the insulin concentration I do not oscillate.
cells-selves can be the pacemaker only or some other factors have to be considered.
Nevertheless, an interesting numerical observation of the Model (4.3.1) (displayed in
Figure 4.5.1 is that the active -cell mass oscillates in an almost invisible range while
both glucose and insulin concentrations do not oscillate. The period is around 50 min-
utes.
References
[1] M. Ader and R. N. Bergman, Importance of transcapillary insulin trans-
port to dynamics of insulin action after intravenous glucose, Am. J.
Physiol. 266 (1994), E17E25.
[2] Ashcroft, M. Frances and P. Rorsman, Electrophysiology of the pancre-
atic -cell, Biophysics Molecular Biology, 54 (1989), 87-143.
[3] R. M. Barbosa, A. M. Silva, A. R. Tom e, J. A. Stamford, R. M. San-
tos and Rosario L. M. , Control of pulsatile 5-HT/insulin secretion from
single mouse pancreatic islets by intracellular calcium dynamics, J. Phys-
iol., 510 (1998), 135143.
[4] D. L. Bennett and S. A. Gourley, Asymptotic properties of a delay dier-
ential equation model for the interaction of glucose with the plasma and
interstitial insulin, Applied Mathematics and Computation, 151 (2004),
189207.
[5] R. G. Bennett, F. G. Hamel and W. C. Duckworth, An insulin-
degrading enzyme inhibitor decreases amylin degradation, increases
amylin-induced cytotoxicity, and increases amyloid formation in insuli-
noma cell cultures, Diabetes, 52 (2003), 23152320.
[6] R. N. Bergman, Pathogenesis and prediction of diabetes mellitus: lessons
from integrative physiology, Irving L. Schwartz Lecture, Mount Sinai J.
of Medicine, 60 (2002), 280290.
[7] R. N. Bergman, D. C. Bradley, and M. Ader, On insulin action in vivo:
the single gateway hypothesis, Adv. Exp. Med. Biol., 334 (1993), 181198.
[8] R. N. Bergman and C. Cobelli, Minimal modeling/partition analysis and
the estimation of insulin sensitivity, Federation Proceedings, 39 (1980),
110115.
126
[9] R. N. Bergman, D. T. Finegood, S. E. Kahn, The evolution of beta-
cell dysfunction and insulin resistance in type 2 diabetes, Eur. J. Clin.
Invest., 32 (2002), (Suppl. 3), 3545.
[10] R. N. Bergman, Y. Z. Ider, C. R. Bowden and C. Cobelli, Quantitative
estimation of insulin sensitivity, Am. J. Physiol., 236 (1979), E667E677.
[11] V. W. Bolie, Coecients of normal blood glucose regulation, J. Appl.
Physiol., 16 (1960), 783788.
[12] S Bonner-Weir, Perspective: postnatal pancreatic -cell growth, En-
docrinology, 141 (2000), 19261929.
[13] S Bonner-Weir, D. Deery, J. L. Leahy and G. C. Weir, Compensatory
growth of pancreatic -cells in adult rats after short-term glucose infu-
sion, Diabetes, 38 (1989), 4953.
[14] S. Bonner-Weir, L. A. Baxter, G. T. Schuppin and F. E. Smith, A second
pathway for gegeneration of adult exocrine and endocrine the pancreas:
a possible recapitulation of embryonic development, Diabetes, 42 (1993),
17151720.
[15] D. C. Bradley, R. A. Poulin, R. N. Bergman, Dynamics of hepatic and
peripheral insulin eects suggest common rate-limiting step in vivo, Di-
abetes 42(1993), 296-306.
[16] A. Caumo, R. N. Bergman and C. Cobelli, Insulin sensitivity from meal
tolerance tests in normal subjects: a minimal model index, J. Clin. En-
docrinol. Metab., 85:11 (2000), 4396-4402.
[17] E. Cerasi, Insulin deciency and insulin resistance in the pathogenesis
of NIDDM: is a dovorce Possible?, Diabetologia 38, 992997.
[18] K. Chen, D. Bandy, E. Reiman, S.-C. Huang, M. Lawson, D. Feng,
L.-S. Yun, and A. Palant, Noninvasive quantication of the cerebral
metabolic rate for glucose using positron emission tomography, F18-
Fluoro-2-Deoxyglucose, the Patlak method, and an image-derived input
function, J. Cerebral Blood Flow and Metabolism, 18 (1998), 716723.
[19] J. B. Conway, Functions of One Complex Variable, Second Edition,
Springer-Verlag, 1973, Corrected fourth printing, 1986.
127
[20] D. L. Cook, Isolated islets of Langerhans have slow oscillations of elec-
trical activity, Metabolism, 32:681685, 1983.
[21] W. F. Crowley and J. G. Hoer, (eds) The Episodic of Hormones, New
York: Wiley and Sons, ASIN: 0471849928, 1987.
[22] A. De Gaetano and O. Arino, A statistical approach to the determina-
tion of stability for dynamical systems modeling physiological processes,
Mathematical and Computer Modeling, 31 (2000), 4151.
[23] A. De Gaetano and O. Arino, Mathematical modeling of the intravenous
glucose tolerance test, J. Math. Biol., 40 (2000), 136168.
[24] A. De Gaetano, G. Mingrone and M. Castagneto, NONMEM improves
group parameter estimation for the minimal model of glucose kinetics,
Am. J. Physiol., 271 (1996), E932E937.
[25] M. Derouich, A. Boutayeb, The eect of physical exercise on the dy-
namics of glucose and insulin, J. Biomechanics, 35 (2002), 911917.
[26] S. Dryselius99, E. Grapengiesser and B. Hellman, E. Gylfe, Voltage-
dependent entry and generation of slow Ca
2+
oscillations in glucose-
stimulated pancreatic -cells, Am J. Physiol., 276:E512E518, 1999.
[27] W. C. Duckworth, R. G. Bennett and F. G. Hamel, Insulin degradation:
progress and potential, Endocr. Rev., 19 (1998), 698-624.
[28] I. B. Efanova, S. V. Zaitsev, B. Zhivotovsky, M. Kohler, S. Efendic,
S. Orrenius and P. O. Berggren, Glucose and tolbutamide induce apop-
tosis in pancreatic -cells. J. Biol. Chem., 273(1998), 3350133507.
[29] L. Edelstein-Keshet, Mathematical Models in Biology, McGraw-Hill,
New York 1988.
[30] K. Engelborphs, DDE-BifTool: a Matlab package for bifurcation anal-
ysis of delay dierential equations, Technical Report TW-305, De-
partment of Computer Science, K. U. Leuven, Leuen, Belgium, 2000.
http://www.cs.kuleuven.ac.be/ keon/delay/ddebiftool.shtml.
[31] K. Engelborghs, V. Lemaire, J. Belair and D. Roose, Numerical bifur-
cation analysis of delay dierential equations arising from physiological
modeling, J. Math. Biol., 42(2001), 361-385.
128
[32] The report of the Expert Committee on the diagnosis and classication
of diabetes mellitus, Diabetes Care, 20 (1997), 11831197.
[33] W. Farris, S. Mansourian, Y. Chang, L. Lindsley, E. A. Eckman,
M. P. Frosch, C. B. Eckman, R. E. Tanzi, D. J. Selkoe and S. Gunette,
Insulin-degrading enzyme regulates the levels of insulin, amyloid -
protein, and the amyloid precursor protein intracellular domain in vivo,
PNAS (Proceedings of the Natural Academy of Sciences), 100 (2003), 7,
4162-4167.
[34] A. Fernandes, L. C. King, Y. Guz, R. Stein, V. E. Wright and G. Teilel-
man, Dierentiation of new insulin-producing cells is induced by injury
in adult pancreatic islets, Endocrinology, 138 (1997), 17501762.
[35] D. T. Finegood, L. Scaglia, and S. Bonner-Weir, Dynamics of -cell mass
in the growing rat the pancreas. Estimation with a simiple mathematical
model, Diabetes, 44 (1995), 249256.
[36] M. E. Fisher, A semiclosed-loop algorithm for the control of blood glu-
cose levels in diabeties, IEEE Tran. Bio. Eng., 38 (1991), 5761.
[37] L. Getty, M. Hamilton-Wessler, M. Ader, M. K. Dea, and
R. N. Bergman, Biphasic insulin secretion during intravenous glucose
tolerance test promotes optimal interstitial insulin prole, Diabetes, 47
(1998), 19411947.
[38] P. Gilon and J. C. Henquin, Inuence of membrane potential changes
on cytoplasmic Ca
2+
concentration in an electrically excitable cell,
the insulin-secreting pancreatic B-cell, J. Biol. Chem., 267:2071320720,
1992.
[39] P. Gilon, M. A. Ravier, J. C. Jonas, and J. C. Henquin, Control mech-
anisms of the oscillations of insulin secretion in vitro and in vivo, Dia-
betes, 51 (2002) (Suppl. 1), S144S151 .
[40] A. Goldbeter, (eds) Cell to Cell Signalling: from Experiments to Theo-
retical Models, Landon: Academic Press, 1990.
[41] T. A. Gresl, R. J. Colman, T. C. Havighurst, L. O. Byerley, D. B. Alli-
son, D. A. Schoeller and J. W. Kemnitz, Insulin sensitivity and glucose
eectiveness from three minimal models: eects of energy restriction and
body fat in adult male rhesus monkeys, Am J. Physiol. Regul. Integr.
Comp. Physiol., 285 (2003), R1340R1354.
129
[42] H. F. Gu, S. Efendic, S. Nordman, C.-G. stenson, K. Brismar,
A. J. Brookes, and J. A. Prince, Brief genetics report: quantitative
trait loci near the insulin-degrading enzyme (IDE) gene contribute to
variation in the plasma insulin levels, Diabetes, 53 (2004), 21372142.
[43] E. Gylfe, E. Grapengiesser and B. Hellman, Propagation of cytoplasmic
Ca
2+
oscillations in clusters of pancreatic -cells exposed to glucose, Cell
Calcium, 12 (1991), 229240.
[44] E. Gylfe, M. Ahmed, P. Bergsten, H. Dansk, O. Dyachok, M. Eberhard-
son, E. Grapengiesser, B. Hellman, J. M. Lin, T. Sundsten, A. Tengholm,
E. Vieira and J. Westerlund, Signaling underlying pulsatile insulin se-
cretion, Ups. J. Med. Sci., 105 (2000), 3551.
[45] P. Hagander, K. G. Tranberg, J. Thorell and J. Distefano, Models for
the insulin response to intravenous glucose, Math. Biosci., 42 (1978),
1529.
[46] J. C. Henquin, H. P. Meissner and W. Schmeer, Cyclic variations of
glucoseinduced electrical activity in pancreatic cells, Pugers Arch
393 (1982), 322327.
[47] J. C. Henquin, Triggering and amplifying pathways of regulation of in-
sulin secretion by glucose, Diabetes, 49:17511760, 2000.
[48] W. M. Hirsch, H. Hanish and J. -P. Gabriel, Dierential equation model
of some parasitic infections: methods for the study of asymptotic be-
havior, Comm. Pure. Appl. Math., 38(1985), 733-753.
[49] S. R. Hugl, M. F. Write and C. J. Rhodes, Insulin-like growth fac-
tor I (IGF-I)-stimulated pancreatic -cell growth is glucose-dependent,
J. Biol. Chem., 273 (1998), 1777117779.
[50] F. C. Jonkers, J. C. Jonas, P. Gilon and J. C. Henquin, Inuence of
cell number on the characteristics and synchrony of Ca
2+
oscillations in
clusters of mouse pancreatic islet cells, J. Physiol., 520:839849, 1999.
[51] J. Keener and J. Sneyd, Mathematical Physiology, Springer-Verlag, New
York, 1998.
[52] B. Keymeulen, Z. Ling, F. K. Gorus and et al., Implantation of stan-
dardized beta-cell grafts in a liver segment of IDDM patients: graft
130
and recipient characteristics in two cases of insulin-independence under
maintenance immunosuppression for prior kidney graft, Diabetologia, 41
(1998), 452-459.
[53] G. Kloppel, M. Lohr, K. Habich, M. Oberholzer and P. U. Heitz, Islet
pathology and the pathogenesis of type 1 and type 2 diabetes mellitus
revisited, Surv. Synth. Path. Res. 4, 110125.
[54] Y. Kuang, Delay Dierential Equations with Applications in Popula-
tion Dynamics, Vol. 191 in the series of Mathematics in Science and
Engineering, Academic Press, Boston 1993.
[55] H. D. Landahl and G. M. Grodsky, Comparison of models of insulin
release, Bull. Math. Biol., 44 (1982), 399-409.
[56] G. Leng, Pulsatility in Neuroendocrine Systems, Boca Raton: FL, CRC
Press.
[57] J. Li, Y. Kuang and B. Li, Analysis of IVGTT glucose-insulin interaction
models with time delay, Discrete Contin. Dyn. Syst. Ser. B., 1(2001), No.
1, 103124.
[58] Y. X. Li and A. Goldbeter, Frequency specicity in intercellular com-
munication, Biophys. J. 55(198), 125145.
[59] A. Makroglou, J. Li and Y. Kuang, Mathematical models and soft-
ware tools for the glucose-insulin regulatory system and diabetes: an
overview, accepted, 2004.
[60] A. Mari, Mathematical modeling in glucose metabolism and insulin se-
cretion, Curr. Opin. Clin. Nutr. Metab. Care, 5(2002), 495-501.
[61] A. Mari, L. Stojanovska, J. Proietto and A.W. Thorburn, A circula-
tory model for calculating non-steady-state glucose uxes: validation
and comparison with compartmental models, Computer Methods and
Programs in Biomedicine, 71(2003), 269-281.
[62] H. P. Meissner, Electrophysiological evidence for coupling between
cells of pancreatic islets, Nature, 262(1976), 502504.
131
[63] A. Mukhopadhyay, A. DeGaetano and O. Arino, Modeling the intra-
venous glucose tolerance test: a global study for a single-distributed-
delay model, Discrete Contin. Dyn. Syst. Ser. B., 4(2004), No. 2, 407
417.
[64] G. Nucci, C. Cobelli, Models of subcutaneous insulin kinetics. A criti-
cal review, Computer Methods and Programs in Biomedicine, 62(2000),
249257.
[65] G. Pacini and R. N. Bergman, MINMOD, a computer program to calcu-
late insulin sensitivity and pancreatic responsivity from the frequently
sampled intravenous glucose tolerance test, Comput. Methods Programs
Biomed. 23 (1986), 113122.
[66] R. A. Poulin, G. M. Steil, D. M. Moore, M. Ader and R. N. Bergman,
Dynamics of glucose production and uptake are more closely related to
insulin in hindlimb lymph than in thoracic duct lymph, Diabetes, 43
(1994), 180-190.
[67] N. Prksen, M. Hollingdal, C. Juhl, P. Butler, J. D. Veldhuis and
O. Schmitz, Pulsatile insulin secretion: detection, regulation, and role
in diabetes, Diabetes, 51 (2002), S245S254.
[68] V. Poitout, An integrated view of -cell dysfunction in type-II diabetes,
Annu. Rev. Med. 1996. 47:6983.
[69] R. Prager, P. Wallace and J. M. Olefsky, In vivo kinetics of insulin action
on peripheral glucose disposal and hepatic glucose output in normal and
obese subjects, J. Clin. Invest., 78 (1987), 472-481.
[70] G. S. Rao, J. S. Bajaj, and J. S. Rao, A mathematical model for insulin
kinetics II. Extension of the model to include response to oral glucose
administration and application to insulin-dependent diabetes mellitus
(IDDM), J. Theor. Biol., 142 (1990), 473483.
[71] L. Rossetti, Glucose toxicity: the implications of hyperglycemia in the
pathophysiology of diabetes mellitus, Clin. Invest. Med., 18(1995), 255-
260.
[72] R. M. Santos, L. M. Rosario, A. Nadal, J. Garcia-Sancho, B. Soria and
M. Valdeolmillos, Widespread synchronous [Ca
2+
] in oscillations due to
bursting electrical activity in single pancreatic islets, Pugers Arch., 418
(1991), 417-422.
132
[73] E. T. Shapiro, H. Tillil, K. S. Polonsky, V. S. Fang, A. H. Rubenstein,
and E. Van Cauter, Oscillations in insulin secretion during constant
glucose infusion in mormal man: relationship to changes in the plasma
glucose, J. Clin. Endocrinol. Metab., 67 (1988), 669674.
[74] C. Simon and G. Brandenberger, Ultradian oscillations of insulin secre-
tion in humans, Diabetes, 51 (2002), S258S261.
[75] C. Simon, G. Brandenberger and M. Follenius, Ultradian oscillations of
the plasma glucose, insulin, and C-peptide in man during continuous
enteral nutrition, J. Clin. Endocrinol. Metab., 64 (1987), 307314.
[76] G. M. Steil, A. Volund, S.E. Kahn and R.N. Bergman, Reduced sample
number for calculation of insulin sensitivity and glucose eectiveness
from the minimal model, Diabetes, 42 (1993), 250-256.
[77] J. I. Stagner, E. Samols and G. C. Weir, Sustained oscillations of insulin,
glucagon, and somatostatin from glucose concentration, J. Clin. Invest.,
65 (1980), 939942.
[78] R. Steele, Inuence of glucose loading and of injected insulin on hepatic
glucose output, Ann. NY Acad. Sci., 82 (1959), 420430.
[79] J. Sturis, K. S. Polonsky, E. Mosekilde, E. Van Cauter, Computer-model
for mechanisms underlying ultradian oscillations of insulin and glucose,
Am. J. of Physiol., 260 (1991), E801E809.
[80] G. W. Swan, An optimal control model of diabetes mellitus, Bull. Math.
Biol., 44 (1982), 793808.
[81] G. W. Swan, Applications of Optimal Control Theory in Biomedicine,
Marcel Dekker, New York 1984, chap. 3.
[82] I. Swenned, The role of glucose in the in vitro regulation of cell cycle
kinetics and proliferation of fetal pancreatic -cells, Diabetes, 31 (1982),
754760.
[83] G. Toolo, R. N. Bergman, D. T. Finegood, C. R. Bowden, C. Cobelli,
Quantitative estimation of beta cell sensitivity to glucose in the intact
organism: a minimal model of insulin kinetics in the dog, Diabetes, 29
(1980), No. 12, 979990.
133
[84] I. M. Tolic, E. Mosekilde and J. Sturis, Modeling the insulin-
glucose feedback system: the signicance of pulsatile insulin secretion,
J. Theor. Biol., 207 (2000), 361375.
[85] B. Topp, K. Promislow, G. De Vries, R. M. Miura and D. T. Finegood, A
Model of -cell mass, insulin, and glucose kinetics: pathways to diabetes,
J. Theor. Biol. 206 (2000), 605619.
[86] S. Wanant and M. J. Quon, Insulin Receptor Binding Kinetics: Modeling
and Simulation Studies, J. Theor. Bio., 205 (2000), 355-364.
[87] J. Warberg, Human Physiology, Fourth Edition, Polytechnic Forlag,
1998, ISBN 87-502-0810-1.
[88] Y. J. Yang, I. D. Hope, M. Ader, and R. N. Bergman, Insulin trans-
port across capillaries is rate limiting for insulin action in dogs, J. Clin.
Invest., 84 (1989), 16201628.
[89] http://arbl.cvmbs.colostate.edu/hbooks/pathphys/endocrine
/pancreas/index.html
[90] http://en.wikipedia.org/wiki/Insulin receptor
[91] http://web.indstate.edu/thcme/mwking/diabetes.html
[92] http://www.barixclinics.com/surgery/obesity facts.jsp?src=
overture&srcdesc=OvObesityAG29
[93] http://www.diabetes.org
[94] http://www.diabetes.org/diabetes-statistics/national-diabetes-fact-
sheet.jsp
[95] http://www.diabetesselfmanagement.com/article.cfm?sid=2&tid=
42&aid=290&sk=5WZA
[96] http://www.endocrineweb.com/insulin.html
[97] http://www.labtestsonline.org/understanding/analytes/glucose
/test.html
[98] E. Y. Skolnik, Insulin receptor signaling pathways,
http://www.med.nyu.edu/research/skolne01.html