Biological Control Systems:

Systems Biology of Diseases and the Design of Effective Treatments

ABSTRACT

The mammalian organism maintains stable, efficient and “near-optimal” performance

and homeostasis in the face of external and internal perturbations via distinct biological
systems ranging from the large-scale physiological (nervous, endocrine, immune,
circulatory, respiratory, etc.), to the cellular (growth and proliferation regulation, DNA
damage repair, etc.), and the sub-cellular (gene expression, protein synthesis, metabolite
regulation, etc). “Biological Control Systems,” a sub-topic of Control Theory, arises from
a control engineering perspective of the function, organization, and coordination of
these multi-scale biological systems and the control mechanisms that enable them to
carry out their functions effectively.

In this presentation, we provide an overview of how physiological life is made possible

by control, and demonstrate the usefulness of a control engineering perspective of
pathologies for diagnosis, design, and implementation of effective treatments. The
concepts and principles will be illustrated using specific examples with significant
research and clinical implications.

OVERVIEW

This is a presentation in three parts:

• In Part 1 (Introduction) is where I make the central point that “control” is

everywhere in biological systems; in fact, physiological life as we know it is not
possible without control.
• In Part 2 (Pathologies & Control Engineering Perspectives) is where I follow up
with another important point: that if control is involved in a biological system
and one is unaware of this, a naïve analysis of input/output data will quite likely
lead to erroneous conclusions, especially when one wants to diagnose anomalous
behavior. Using a control engineering perspective protects us from such errors,
and helps clarify potential confusion
• In Part 3 (Engineering Control Systems and Disease Treatment) is where I make
the final point: that when a biological control system fails, engineering control
systems can provide viable alternatives.

1. INTRODUCTION

In 1775, Charles Blagden (1748-1820), and some research associates, performed a

seminal experiment: some human subjects, a dog and a piece of beefsteak were confined
for 45 minutes in a chamber heated up to 127°F, with the following results: over this
period, the human subjects sweated, the dog panted, but all survived; the beefsteak, on
the other hand, cooked. The central conclusion was that living organisms are able to
adapt to the external conditions and thereby maintain what became known as
“homeostasis”; dead organism could not.

It is now well established that, overall, the mammalian system consists of a hierarchical
organization of multiple levels of physiological systems at different length scales: from
the chemical level (with atoms C, H, O, N, P, and DNA) to the cellular level consisting of
an organization of atoms and molecules; to the tissue level, consisting of a systematic
collection of cells; to the organ level, which is a collection of tissues, to the system level
which is a collection of organs and tissues organized into a coherent unit system with a
specific task and objective (for example, the digestive system); to the final overall
organismal level. With an overall objective of maintaining physiological function
(equivalent to keeping physiological variables within specification limits), the
mammalian system must rely on specialized “control systems” in order to meet this
overarching objective because of persistent perturbations that threaten to displace the
physiological systems away from homeostatic conditions. Table 1 below shows a sample
of typical “specification limits” for constituents of the extracellular fluid in humans.

TABLE 1: Important Constituents and “Specification Limits” of the Human Extracellular

Fluid: The specs for blood glucose are highlighted

Important Constituents and “Specs” of

Extracellular Fluid

Normal Value Normal Range Short-Term Unit

Nonlethal Limit
Oxygen 40 35-45 10-1000 mm Hg
Carbon dioxide 40 35-45 5-80 mm Hg
Sodium ion 142 138-146 115-175 mmol/L
Potassium ion 4.2 3.8-5.0 1.5-9.0 mmol/L
Calcium ion 1.2 1.0-1.4 0.5-2.0 mmol/L
Chloride ion 108 103-112 70-130 mmol/L
Bicarbonate ion 28 24-32 8-45 mmol/L
Glucose 85 75-99 20-150 mg/dl
Body Temperature 98.4 (37.0) 98-98.8 (37.0) 65-110 (18.3-43.3) °F (°C)
Acid-base 7.4 7.3-7.5 6.9-8.0 pH

A short term but significant departure of physiological variables from normal limits is
referred to as an “illness” where recovery occurs when the perturbed variables are
returned to within normal limits. “Chronic illnesses” are characterized by long term (or
systemic) departures from normal limits, but with the affected variables still within non-
lethal limits. Death occurs when critical physiological variables fall outside non-lethal
limits and are not restored within a reasonable period of time.

Physiological Control Systems

The mammalian system maintains homeostasis and desired physiological function in
the face of constant perturbations with the aid of a collection of physiological systems,
each with its dedicated function and objectives, and organized as follows:

The MASTER REGULATORS

1. The Nervous System (for overall control and regulation, among other functions such
as memory, behavior, etc.)
 2. The Endocrine System (for chemical regulation, among other functions such as
growth and development, etc.)
The DEFENDERS, PROTECTORS & SUPPORTERS
3. The Immune/Lymphatic System (for defense and internal protection, among other
functions including fluid balance and fat absorption, etc.)
4. The Integumentary System (for protection and heat transfer, among other functions
such as Vitamin D production etc.)
5. The Musculoskeletal System (for movement and mechanical support, among other
functions such as minerals warehousing and blood cell production, etc.)
The UTILITIES
6. The Cardiovascular System (for transport and supplies distribution)
7. The Respiratory System (for oxygen supply and CO2 elimination, among other
functions such as vice production and olfaction, etc.)
8. The Digestive System (for nutrient processing and supply)
9. The Urinary System (for waste extraction and elimination, among other functions
such as Vitamin D synthesis, etc.)
The SPECIES PROPAGATOR
10. The Reproductive System (for species propagation from one generation to the next)

Biological Control Systems Configurations

Each of the ten physiological systems noted above consists of exquisitely calibrated
control systems that enable it carry out its functions efficiently. These control systems
consist of the same functional components as the engineering control system: sensors,
controllers, actuators, and the controlled process itself. And, as with engineering control
systems, depending on how these components are organized relative to one another,
one finds a variety of control configurations: Feedback, Feedforward, Cascade, and
unconventional multi-loop configurations that do net exist in tradition engineering
control systems. The most common configuration is feedback control, an example of
which is the blood pressure control system whose block diagram is shown in Figure 1.

Baroreceptor Reflex: Simplified Block

Diagram
Input from other Disturbance
brain centers
e.g., Exercise

Parasympathetic,
Activity Heart Rate
SA/AV
Nodes Cardiac
Cardiovascular Output
Cardiac
Center Muscle Controlled
Stroke Arterial
Process
Volume BP
Sympathetic, TPR
Activity . Arterioles

Baroreceptor
neurons

©Babatunde A. Ogunnaike Lecture Notes on Biological Control Systems 11

Fig 1. Block Diagrammatic Representation of the Baroreceptor Reflex: The Blood Pressure
Control Systems
We showed in the presentation that feedback control allows for robustness and
performance without the need for detailed process knowledge. The cascade structure is
the stereotypical structure of choice for endocrine control systems, an example of which
is shown in Fig 2. Many more structures, far more complex than commonly seen in
engineered systems, exist in biological control systems (for example, Fig 3 shows a
simplified representation of the immune system block diagram).

Example Endocrine Control Cascade:

Cortisol and Cell Metabolism

Disturbance
Endocrine Controller
CNS (Hypothalamus) (Anterior Pituitary) Adrenal Cortex Cortisol Output
Primary Secondary Controlled Controlled
Controller CRH Controller Process 2 Process 1
ACTH
Target
Cell

Cell Metabolic rate

Sensor

CRH—Corticotropin releasing hormone

ACTH—Adrenocorticotropic hormone

©Babatunde A. Ogunnaike Lecture Notes on Biological Control Systems 15

Fig 2. Block Diagrammatic Representation of a typical endocrine control cascade

2. PATHOLOGIES & CONTROL ENGINEERING PERSPECTIVES

Many diseases arise from the failure or malfunction of one or more control system
components. For example, diabetes arises as a result of the failure in the control system
responsible for blood glucose regulation. However, the type of diabetes depends on
which component of this control system failed. Type I diabetes arises from an actuator
failure because, in this case, the pancreas is incapable of producing insulin. On the other
hand, Type II diabetes arises from impaired insulin sensitivity by the “process” to be
controlled. Diagnosis and the prescription of appropriate treatments often involve
clinical observations which are tantamount to the collection and analysis of “input-
output” data, in the sense that the physician “perturbs” the system in question and
observes the patient’s response through various assays of samples taken from the
patient, for example, blood sample analysis following fasting to determine “fasting
blood glucose” levels in a potentially diabetic patient.
 Overall Specific Immune Response
Block Diagram
*Each involves multiple processes
lumped into indicted single box
Pathogen
Infection stimulus
B-cell Antibodies
Effector Population
System*
IL-4, 5, 6
Helper IFN-Gamma Macrophage IFNs concn Controlled Antigen
T-cells Other System* Process Population
Cytokines
Cytotoxic T-cells
T-cell Population
IL-2 Effector
System* Suppressor T-cells
Population

T-cell Antigen
Specific
MHC-I Antigen
Receptors processing and
B-cell Antigen MHC-II presentation
Specific
Receptors

©Babatunde A. Ogunnaike Lecture Notes on Biological Control Systems 16

Fig 3. Simplified Block Diagrammatic Representation of the Immune Response

The two thematic concepts central to this part of the presentation may be stated as
follows:

1. If the involvement of feedback control in a biological system is unrecognized,

naïve input-output data analysis will likely lead to erroneous conclusions;
2. A control engineering perspective of diseases decreases the possibility of
“faulty” diagnosis caused by confounding and conflating “component
malfunctions”

With the first concept, the central issue is that a control system response to an external
disturbance is drastically different under open loop conditions as opposed to closed loop
feedback control. To illustrate, consider a seminar room with a single door that opens
directly to the outside. Further consider that it is winter, with a temperature 0 degree
Celsius outside, but the temperature inside the room is maintained at a comfortable 25
degrees Celsius by a “perfect” thermostat. Under these circumstances, frequent and
periodic opening of the door, which exposes the room to the outside temperature, will
be countered by the action of the perfect controller with the net effect that the room
temperature will remain unchanged. Without recognizing the presence of the thermostat
in the room, observers analyzing the response of the room temperature to the
perturbations caused by exposure to the outside elements will come to the erroneous
conclusion that the room temperature is fundamentally unaffected by exposure to the
outside temperature.
The second concept is predicated on the fact that a control system consists of an
interconnection of four functional modules/components: the sensor, the controller, the
actuator, and the controlled process itself. Thus, a control system can fail or
malfunction when one or more of these components fails or malfunctions, in which case,
a rational approach to rectifying the problem must begin with the determination of
which specific component malfunction is responsible for the observed system failure.
But, a sensor failure, for example, is not the same as an actuator failure, in precisely the
same sense in which Type I diabetes (an actuator failure) is fundamentally different
from Type II (a process malfunction). However, because the components are connected,
an indiscriminate analysis of the overall system response to any arbitrary perturbation,
conflates the responses of the components making it difficult, if not impossible, to
identifying the malfunctioning component.

TGF-b regulation of cell population

The first concept is illustrated by the role of TGf-β in late stage cancer where, because
the amount of TGF-β, a known tumor suppressor, was clinically observed to be unusually
high in the worst cases, it was proposed that TGF-β somehow switches function from a
tumor suppressor to a tumor promoter. The premise was that since an observed increase
in tumor size corresponded with increasing amounts of TGf-β, then this growth
inhibitor must have transformed into a growth promoter in late stage cancers. This
paradox was resolved in Chung et al., (2012)1 by taking a control engineering approach
and showing that the observation and misleading conclusion is due to the presence a
control system that employs TGF-β to regulate cell proliferation, with the following
conclusions arising from an analysis of the mathematical model of the control system:

• Under Normal Conditions the controller regulates growth, inhibits proliferation

effectively using the tumor suppressor ligand, TGF-β; however,
• Under Cancerous Conditions (characterized by TGF-β resistance, in a manner
reminiscent of Type II diabetes), the role of TGF-β is unchanged; the control
system still intact; but it must now secrete more of TGF-β in a futile attempt to
achieve the level of tumor suppression attainable with normal, responsive cells.

Thus, the increased level of TGF-β is consistent with its unchanged role as a tumor
suppressor; the observed correlation was confused with causality because the presence
of a control system went unnoticed.

Calcium Regulation
The second concept is illustrated with differential diagnosis of three different
manifestations of Hypercalcemia (a condition where the calcium level is higher than
normal) using a quantitative model of the Ca2+ regulation system in the human body2
based on the control engineering block diagram in Fig. 4.

The three different manifestations of hypercalcemia, are:

• Primary Hyperparathyroidism (PHPT), responsible for ~70% of all cases
• Familial Benign Hypercalcemia (FBH) and
• Humoral Hypercalcemia of Malignancy (HHM)

1
S.W. Chung, C. R. Cooper, M. C. Farach-Carson, and B. A. Ogunnaike, “A control engineering approach to
understanding the TGF-β paradox in cancer”, J. R. Soc. Interface (2012), 9, 1380-1397
2
C. Christie, L. E. K. Achenie and B. A. Ogunnaike, 2014, “A control engineering model of calcium regulation”, J
Clin Endocrinol Metab, 99(8): 2844–2853
 ACTUATOR DISTURBANCE
Intravenous
Infusion
KIDNEYS INTESTINE Ca(i)
Calcitriol CTL Calcium
Production Absorption PO4(i)
PROCESS
CONTROLLER KIDNEYS Ca(u) Σ Σ Ca(p)
(PLASMA)
(PT GLAND) PTH Calcium
Ca2+ Pool
PTH Secretion Reabsorption PO4(u) Σ
PO43- Pool

BONE
Formation
Ca(b)
&
Resorption PO4(b)

PO4(p)

SENSOR
Ca(sens) (PT Gland)
Calcium Sensing

Fig 4: The Calcium Regulation Control System.

The primary challenge is that they are all classified as “Hypercalcemia” and show similar
clinical presentations (High Ca and High Parathyroid Hormone-like effects), making
differential diagnosis difficult. However, PHPT is caused by abnormal enlargement of
the parathyroid glands, in which case this is equivalent to a “controller defect”. On the
other hand, FBH is a hereditary genetic disorder involving a missense mutation of the
Ca-sensing receptors—clearly a “sensor defect”. Finally, HHM arises from complications
of some cancers (Breast; Lung; Hematologic), where tumor cells produce PTH-related
protein (PTHrP), which acts like PTH. When PTH receptors detect/respond to PTHrP, the
result is “high PTH”-like response, characteristic of other hypercalcemic responses.
While it is not as obvious, we showed in the presentation that this is an “actuator
disturbance”.

Having thus identified the component malfunction responsible for each manifestation
of hypercalcemia, we used the (validated) mathematical model to investigate input
stimuli design to generate theoretical differentiated responses. We showed that even
with standard induced hypercalcemia pulses, the post infusion dynamic response
profiles of the parathyroid hormone and calcitriol are sufficiently distinguishable for
each of the three different cases of hypercalcemia.

3. ENGINEERING CONTROL SYSTEMS & DISEASE TREATMENT

When natural biological control systems fail, engineered control systems can provide
effective substitutes. With proper recognition of the intrinsic complexity of biological
system dynamics, control engineering principles can be (and have been) applied
successfully to compensate. The most well known example is the “Artificial Pancreas”,
an engineered device designed for Type I Diabetes patients. Based on engineering
control principles, a glucose sensor is combined with an embedded controller and an
insulin pump, in a single device that delivers precise amounts of insulin for patients
suffering from Type I diabetes.

Here we present a case study of a control system designed to achieve platelet count
control for an Immune Thrombocytopenic Purpura (ITP) patient.3

3
C.-H. Tsai, J. B. Bussel, A. A. I mahiyerobo, S. I. Sandler, and B. A. Ogunnaike, Platelet count control in immune
thrombocyropenic purpura patient: Optimum romiplostim dose profile, J. Process Control, 45, (2016) 76-83.
Immune Thrombocytopenic Purpura (ITP)
The clinical definition of ITP is a platelet count lower than 150×109 cells/L, putting the
patient at a high risk of excessive bleeding. (See table below.)

ITP is characterized by increased platelet destruction in patients, and/or reduced

platelet production; its root cause is unknown. It is common to treat ITP with periodic
injections of romiplostim, a thrombopoietin (TPO)-mimetic, which functions like
endogenous TPO. Natural TPO, which is synthesized in the liver, is central to the
production of platelets according to the following mechanism:
• TPO binds to its receptor, c-Mpl, on the plasma membrane of the platelet
precursor (e.g., megakaryocyte)
• This leads to megakaryocyte differentiation and eventually to platelet production
• However, platelets also have TPO receptors, c-Mpl; so that binding TPO lowers its
concentration in circulation, providing a natural feedback regulation of the
amount of TPO.

Romiplostim thus increases platelet count by stimulating platelet production in the

same manner as endogenous TPO. However, with romiplostim therapy, the platelet
count response is often oscillatory for many patients, with very dangerous lows and
highs, making it difficult to maintain steady platelet count. In addition, romiplostim is
expensive and its administration is accompanied by long delays before its effect (which
depends nonlinearly on platelet count) can be observed.

Platelet Count Control

The specific objective of this study may be stated as follows: For an actual clinical
patient, determine “optimum” dose profile required to maintain the ITP patient’s
platelet count at 70×109 cells/L, subject to the following constraints:

• Treatment is limited to pulses (injections) of fixed magnitudes at discrete points

in time;
• Current practice is limited to weekly or biweekly treatments

The approach is to develop and validate a custom pharmacokinetic-pharmacodynamic

(PK-PD) model for the patient in question, analyze the model for insight, and use it to
develop/evaluate control schemes. See Ref 3 for details, which show that:

1. This particular patient has platelets with average lifespans less than half those
of healthy subjects; and
2. A model predicted dose response curve shown in Fig 5 below:
 500
5 µg/kg
3 µg/kg
400
1 µg/kg

PLT Count (*109/L)

0.5 µg/kg
300 0.2 µg/kg

200

100

0
0 10 20 30 40
Days
Fig 5: Predicted dose response curve from an initial low platelet count, for the patient in question: Legend
indicates various doses of Romiplostim.

The following control schemes were studied in simulation using the PK/PD model to
represent the patient dynamics:

• Fixed dose, Open-Loop (current practice)

• Optimally-Tuned PI Control
• “Variable Dose” Optimal Open-Loop Control

with the following results. Fig 6 shows the results of fixed doses (2 µg/kg; 1 µg/kg, and
0.5 µg/kg) applied weekly or bi-weekly. The platelet count is oscillatory and the
objective of a steady count of 70×109 cells/L (the red line) has not been attained.
Fig 7 shows the results for an optimally tuned PI controller strategy. In this case,
the controller parameters KP and KI are determined, along with u(k=0), using fminsearch
in Matlab to minimize:

nk

å(y
k =0
k - y*) 2

where y* is the desired set-point. We observe that the bi-weekly measurements are not
representative of true dynamics as a result of aliasing due to sampling.
Rather than use a PI controller, if the implemented dose, u(k), is determined
instead by optimal control, where the decision variable determined by minimizing the
same sum-of-squared deviations shown above is u(k), directly, and not the controller
parameters, the results are shown in Fig 8. Not surprisingly, the responses are similar to
those in Fig 7.

As a result of these simulations, we are able to make the following observations:

• With carefully designed controllers, it is possible to stabilize platelet count, but
only with weekly injections;
• Bi-weekly injections still produced oscillations, regardless of the controller type.

The latter observation prompted a closer look at the intrinsic characteristics of the
patient’s response to romiplostim as indicated in Fig 9.
 Fig 6: Simulation results of platelet count response to a fixed-dose open loop control strategy for weekly and bi-
weekly treatemtns of 2 µg/kg; 1 µg/kg, and 0.5 µg/kg

2 Dose (µg/kg) 2
Dose (µg/kg)

Romiplostim
Romiplostim

1 1

0 0
0 50 100 150 200 0 50 100 150 200
200 80
Simulation
PLT Count (*109/L)
PLT Count (*109/L)

Target 60
150
measurements taken by PI controller
100 40
Simulation
50 20
Target
measurements taken by PI controller
0 0
0 50 100 150 200 0 50 100 150 200
Days Days

Weekly Bi-Weekly
Fig 7: Simulation results of platelet count response under an optimally-tuned PI control strategy

Weekly Bi-Weekly
Fig 8: Simulation results of platelet count response under variable-dose optimal control.
 Fig 9: Basic dose-response curves for Romiplostim and bi-weekly platelet count response under variable-dose
optimal control of Fig 8).

From this figure, we may now observe why it will be virtually impossible to eliminate
oscillations in bi-weekly injections. The pharmacology of romiplostim is such that from
the application of the injection to the peak platelet count is approximately 2 weeks, for
this specific patient. And since the observed platelet response is nothing but a
“nonlinear convolution” of this theoretical dose response “basis function”, because the
2-week implementation period coincides almost precisely with the natural period of the
dose response, the steady state response will be expected to be a repetition of the dose
response curve on the left over and over again. Hence, the oscillations are not due to an
overactive controller; rather, they are due to the natural period of the response to
romiplotism coinciding with the treatment period. Any treatment period that is less
than two seeks should produce better results, as Figs 7 and 8 have shown with weekly
treatments. We are thus able to make the following recommendations:

• Current practice did not work well for the candidate in question because of the
natural pharmacodynamics of romiplostim in this patient;
• Constant dose injections (regardless of actual platelet count), weekly or bi-
weekly, can never produce stable platelet count
• However, stable platelet control possible with variable dose, measurement-based,
weekly discrete treatment, even when determined via PI control.

4. CONCLUSIONS
Because control is a central and inevitable feature of biological systems, it is not
surprising that “Control Engineering” has a lot to offer any systematic study of
biological control systems, especially with regard to diseases and their treatment. In
such an endeavor, a control engineering perspective offers the following advantages:

• It provides a means for efficient modeling, particularly making the resulting

models analytically tractable while retaining high fidelity;
• It facilitates diagnosis and prevents the potential problem of confounding the
root causes of diseases, especially those that are caused by the malfunction of a
control system component.
• Finally, it provides a rational basis for engineered treatment.