Journal of Theoretical Biology 343 (2014) 2531

Contents lists available at ScienceDirect

Journal of Theoretical Biology

journal homepage: www.elsevier.com/locate/yjtbi

Assessing cytokines' talking patterns following experimental

myocardial damage by applying Shannon's information theory
Michele M. Ciulla a,n, Federico De Marco b, Elisa Montelatici c,d, Lorenza Lazzari c,d,
Gianluca L. Perrucci a, Fabio Magrini a,d
a
Department of Clinical Science and Community Health, Laboratory of Clinical Informatics and Cardiovascular Imaging,
University of Milan, 20122 Milan, Italy
b
Cardiology 1, Emodinamica, Ospedale Niguarda Ca' Granda, Milan, Italy
c
Cell Factory Franco Calori, Milan, Italy
d
Foundation I.R.C.C.S. Ca' Granda Ospedale Maggiore Policlinico, 20122 Milan, Italy

H I G H L I G H T S

We studied the information throughput of a multiplex serum cytokine arrays by using the formulas of Shannon modied accordingly;

Assuming that redundant informations do not increase the throughput, we derived Entropy and Redundancy before and after myocardial cryodamage
in rats;

High information throughput was associated to a low level of correlation between cytokines and vice versa for a low throughput;

After cryodamage, the information throughput deteriorates showing its maximum level in the control conditions, with great differences between
individual cytokines;

The contribution of cytokines to information throughput allows to formulate alternative approaches to describe the inammatory cascade after
cryodamage.

art ic l e i nf o

a b s t r a c t

Article history:
Received 29 May 2013
Received in revised form
28 October 2013
Accepted 30 October 2013
Available online 7 November 2013

Background: The simultaneous measurement of multiple cytokines in parallel by using multiplex

proteome arrays (MPA) is of great interest to understanding the inammatory response following
myocardial infarction; however, since cytokines are pleiotropic and redundant, increase of information
throughput (IT) attained by measuring multiple cytokines remain to be determined. We aimed this study
to assess the IT of an MPA system designed to assess 8 cytokines commercially available at the time of
the study serum levels, before (control state) and after experimental myocardial cryoinjury (activated
state) in rats.
Methods: By assuming that redundant information do not generally increase the IT, we derived Entropy
(H) and Redundancy (R) of information by using formulas of Shannon modied accordingly, where a high
IT (high H and low R) corresponds to a low level of correlation between cytokines and vice versa for a low
IT. The maximum theoretical level of IT and the contribution of each cytokine were also estimated.
Results: In control state, no signicant correlations were found between cytokines showing high IT; on
the contrary, in activated state, several signicant correlations were found supporting a complex crosstalk pattern between cytokines with low IT. Using as reference the maximum theoretical level of IT, in
activated state, H was reduced of 67.0% and R was increased of 77.4% supporting a reduction of IT.
Furthermore, the contribution of individual cytokines to H value of MPA was variable: in control state,
IL-2 gave the most contribution to H value, conversely during activated state IL-10 gave most
contribution. Finally during activated state, IL-1 was the only cytokine strongly correlated with values
of all other cytokines, suggesting a crucial role in the inammatory cascade.
Conclusions: Paradoxically, by analyzing an MPA system designed for redundant analytes such as
cytokines, translating the Shannon's information theory from the eld of communication to biology,
the IT system in our model deteriorates during the activation state by increasing its redundancy, showing
maximum value of entropy in the control conditions. Finally, the study of the mutual interdependence

Keywords:
Cytokines cross talk
Cryoinjury
Entropy
Redundancy

n
Corresponding author at: Department of Clinical Science and Community Health, Laboratory of Clinical Informatics and Cardiovascular Imaging, University of Milan,
20122 Milan, Italy. Tel.: 39 02 55033592; fax: 39 02 50320480.
E-mail addresses: michele.ciulla@unimi.it (M.M. Ciulla), federico.demarco@gmail.com (F. De Marco), ely.mega@gmail.com (E. Montelatici), lorenza.lazzari@policlinico.mi.it
(L. Lazzari), gianluca.perrucci@unimi.it (G.L. Perrucci), fabio.magrini@unimi.it (F. Magrini).

0022-5193/$ - see front matter & 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.jtbi.2013.10.019

26

M.M. Ciulla et al. / Journal of Theoretical Biology 343 (2014) 2531

between cytokines by the contribution to the IT may allow formulating alternative models to describe
the inammatory cascade after myocardial infarction.
& 2013 Elsevier Ltd. All rights reserved.

1. Introduction
Natural strategies aimed to cardiac repair after myocardial
infarction involve a variety of pleiotropic inammatory mediators
whose effect consists in collagen deposition in order to start the
healing process (Frangogiannis, 2012). Among these mediators,
cytokines have been proven to be activated in cascade and
responsible for both pro-inammatory and anti-inammatory
processes (Dinarello, 2010). Unfortunately the pattern of cytokine
signaling is the result of a rather complex cross-talk as a cytokine
can trigger the synthesis or the release of one or several cytokines
acting, both, in sequence or in parallel and having additive,
synergistic, or antagonistic effects (Prabhu, 2004). Therefore, in
principle, the simultaneous measurement of multiple cytokines
within the same sample, will be of great interest in revealing the
complicated cytokine cross-talk (Wang et al., 2002) and in understanding the biology of the inammatory response following
myocardial infarction. In addition, a systematic approach seems
to be unavoidable to understand the complex protein-based
signaling system as shown by previous experiences in models of
cytokine-induced apoptosis (Janes et al., 2005). Recent developments in protein microarray technology and the advent of multiplex proteome arrays (MPAs) have provided new opportunities to
obtain, with a single assay, more data, thus, increasing the nal
information throughput (Zhu and Snyder, 2003). However, by
assuming that MPAs are collection of analytes designed in light
of the current scientic knowledge, the results, in terms of
information throughput (IT), are highly affected by (1) availability
and choice of analytes and (2) the correct application to a
biological context. Furthermore, since cytokines exhibit pleiotropic
and redundant features, the increase in terms of IT attained by
measuring several cytokines in parallel remain to be determined.
Following our previous experience with MPAs for assessing
cytokine cross-talk (Ciulla et al., 2008) we approached the problem in terms of IT by applying for the rst time in the freezethaw myocardial infarction injury model the basic concepts and
the formulae of Entropy (H) and Redundancy (R) drawn by
Shannon (Shannon, 1997, 1998). Shannon was the founder of the
information theory and, by studying language, described H and R
of vowels and consonants, according to their relevance in understanding of the words and phrases. Obviously, vowels are few and
highly repetitive (low H, high R), on the contrary, consonants are
numerically more than vowels and low repetitive (high H, low R);
these features make the consonants more essential for the
comprehension of the words then vowels, thus, resulting in a
higher IT. We translated these concepts by conducting a parallelism between alphabet letters and inammatory cytokines, assuming that the IT of an MPA is affected by the correlation between
cytokines. In this model, low H and high R values, correspond to
high level of correlation (randomness) with a overall low IT. Since
the reciprocal activation of cytokines, according to the cascade
model, presupposes a highly level of mutual dependence following
activation, we used the correlationregression multiple analysis;
the conventional p-value, employed to correctly understand the
correlation between the variables, was used as a surrogate of H,
according to the formulae reported in Appendix. A commercially
available MPA system custom-designed to assess 8 cytokines in
parallel was used; cytokines quantitative data and their correlations were obtained from rats before (control state) and following
experimental myocardial damage (activated state), a well known

stimulus capable of activating the inammatory cascade (Ciulla

et al., 2004, 2008).

2. Methods
2.1. Animal model and experimental myocardial cryoinjury
A total number of 30 male adult rats (200250 g) were studied.
Animals were housed and handled in accordance with the Guide
for the Care and Use of Laboratory Animals (1996). To ensure the
permanent identication, at the arrival each rat was implanted
with a microchip device (MUSICC, AVID Microchip, Barcelona,
Spain). Experimental myocardial cryoinjury was produced by
freezethaw technique, previously described in detail (Ciulla
et al., 2004), that allows producing a predictable cardiac lesion.

2.2. Serum cytokine assay

Cytokine prole analysis was performed by ELISA assay
(SearchLight Proteome Array, USA) on 50 ml serum samples (dilution 1:5) obtained by direct puncture of the heart at 7 days from
the infarct before sacrice, following previous experience (Ciulla
et al., 2008). The array layout included the following 8 cytokines:
IL-1, IL-1, IL-2, IL-4, IL-6, IL-10, IFN- and TNF-. The lower
detection limit for each cytokine was set according to the manufacturer. The cytokines studied are a partial selection conditioned
by their commercial availability at the time of the study; furthermore, in the panel set-up, we did not include other cytokines that
we tested previously in the same model of myocardial injury
(Ciulla et al., 2008), whose values were below the lower detection
limit. According to the shortest half-life of cytokines studied (TNF E30 min), we minimized the delay between sampling and
storage of all samples before the analysis. All samples were stored
at  20 1C within 1520 min from collection in order to keep and
protect the stability of cytokines; during the storage, the  20 1C
freezing were under controlled and centralized alarm system.

2.3. Statistical analysis

Data were analyzed using computer statistical software (SPSSRel 6.1.1; SPSS Inc., Chicago, Ill). All the quantitative variables were
tested for Gaussian distribution with the KolmogorovSmirnov
test. The distribution was presented as mean 7standard deviation
(SD). Assuming that cytokines are part of coherent set of proteins,
activated in sequence according to the cascade model, we have
identied as probative the correlative-regressive model. This
assumption was conrmed preliminarily plotting the values of
cytokines in pairs after activation on a scatter diagram by
highlighting that the points of many of them were lying along a
straight line. Furthermore, to correctly understand the correlation
between the variables, we employed the p-value, that is used as a
parameter to read the most common statistical tests, since it was
fully compatible with the original formulas of Shannon. Conventionally a p-value o0.05 was considered signicant.

M.M. Ciulla et al. / Journal of Theoretical Biology 343 (2014) 2531

2.4. Entropy and Redundancy assessment

27

Table 2a
Correlation between cytokines in control state.

The cytokine MPA system was assumed as an information

source made up of 8 variables (cytokines). The states were control
and activated, following experimental myocardial infarction.
H and R of each state was calculated, for each cytokine and for
the MPA, according to the assumptions and formulae reported in
Appendix.

3. Results
The cardiac injury was completed and well tolerated in 17 rats.
At the end of the study, the mean area of infarction in the studied
animals measured in vivo by echocardiography was 21.57 3.1% of
the transverse left ventricular free wall. Deep examination of the
excised hearts conrmed the presence of a regular non-transmural
scar of the entire left ventricular free wall. Histology conrmed the
presence of a hemorrhagic infarct, without transition zones from
non-infarcted to infarcted areas.
In control state, very low levels of cytokines were found,
conversely in activated state high levels of them were found. The
average values obtained for each cytokine in control state (n 7)
and infarcted animals (n 17) are shown in Table 1.

IL-1
IL-1
IL-1
IL-2
IL-4
IL-6
IL-10
IFN-
TNF-

0.4181
0.4981
0.6808
0.9028
0.4779
0.5417
0.331

IL-1

IL-2

IL-4

IL-6

IL-10

IFN-

TNF-

0.4181

0.4981
0.9162

0.6808
0.2627
0.8211

0.9028
0.4847
0.5991
0.2220

0.4779
0.0590
0.976
0.2029
0.4248

0.5417
0.1236
0.9602
0.1391
0.3611
0.0638

0.331
0.0871
0.8291
0.3498
0.5718
0.1469
0.2107

0.9162
0.2627
0.4847
0.0590
0.1236
0.0871

0.8211
0.5991
0.976
0.9602
0.8291

0.2220
0.2029
0.1391
0.3498

0.4248
0.3611
0.5718

0.0638
0.1469

0.2107

Table 2b
Correlation between cytokines in activated state.

IL-1
IL-1
IL-2
IL-4
IL-6
IL-10
IFN-
TNF-

IL-1

IL-1

IL-2

0.0023
0.0012
0.0805
0.1668
0.5492
o0.0001
0.1257

0.0339
0.0012
o0.0001
0.0014
0.0053
o0.0001

0.0023

0.0012
0.0339
0.0397
0.3779
0.9980
o0.0001
0.2614

IL-4

IL-6

IL-10

IFN-

TNF-

0.0805
0.0012
0.0397

0.1668
o0.0001
0.3779
0.0096

0.5492
0.0014
0.9980
0.2053
o0.0001

o 0.0001
0.0053
o 0.0001
0.0655
0.3250
0.7618

0.1257
o0.0001
0.2614
0.0081
o0.0001
o0.0001
0.2363

0.0096
0.2053
0.0655
0.0081

o0.0001
0.3250
o0.0001

0.7618
o0.0001

0.2363

The statistically signicant p-values are shown in bold.

3.1. Correlations between cytokines

No signicant correlations were found between cytokines in
controls (Table 2a); on the contrary, following experimental
myocardial infarction, several signicant correlations were found
supporting a complex pattern of interactions between cytokines
(Table 2b). In particular, IL-1 values were strongly correlated with
all other 7 cytokines. Among these, statistical signicance was:
p o0.0001 for IL-6 and TNF-, p o0.005 for IL-1, IL-4, and IL-10
and po 0.05 for IL-2 and IFN-. Furthermore, IL-2, IL-4, IL-6 and
TNF- correlate with 4 cytokines, while IL-1, IL-10, IFN-
correlate with 3 cytokines.
3.2. Entropy and Redundancy in control and activated state
In control state, where no signicant correlations between
cytokines were found, the MPA system showed an overall high H
of 20.8189 if compared with the maximum theoretical H(Hmax)
calculated according to the formulae in Appendix that was
38.8162; accordingly a low R value of 46.4%, normalized for Hmax,
was found. In activated state, where signicant correlations
between cytokines were found, H values was 6.8449 supporting
a drastic reduction of H in MPA, representing an overall 67.0% H
loss. R was increased to the value of 82.3% during activated state,

Table 1
Cytokines' serum prole in control state and activated state.
Control state (n7)

IL-1
IL-1
IL-2
IL-4
IL-6
IL-10
IFN-
TNF-

Activated state (n 17)

pg/ml

7SD

pg/ml

7 SD

5.3
3.8
0.5
1.3
1.8
1.7
1.1
1.9

5.3
4.9
0.3
0.8
1.4
2.2
0.6
1.9

41.5
116.8
1031.0
18.4
614.8
330.1
615.2
378.3

86
140.8
1956.5
17.0
593.5
801.5
1370.1
388.5

The statistically signicant p-values are shown in bold.

p (ANOVA)

0.2865
0.0467
0.1826
0.0159
0.0131
0.2951
0.2544
0.0188

Table 3
Cytokines' entropy (H) and redundancy (R) in control state and in activated state.
Entropy
Control
state
System 20.8189
IL-1
3.0256
IL-1
1.8863
IL-2
4.0814
IL-4
3.1674
IL-6
2.8230
IL-10
1.8675
IFN-
1.9140
TNF-
2.0537

Redundancy %
Activated
state

Control
state

Activated
state

6.8649
0.7890
0.0416
1.3158
0.3739
0.7642
1.8833
1.1273
0.5698

 67.0
 73.9
 97.8
 67.8
 88.2
 72.9
0.8
 41.1
 72.3

46.4
37.6
61.1
15.9
34.7
41.8
61.5
60.6
57.7

82.3
83.7
99.1
72.9
92.3
84.3
61.2
76.8
88.3

77.4
122.6
62.2
358.5
166.0
101.7
 0.5
26.7
53.0

Redundancies of the whole system and of each single cytokine, expressed in

percentage, are normalized for the maximal theoretical Entropy of the whole
system [Hmax(8cyto) 38.8162] and for the maximal theoretical Entropy of each
one of its components [Hmax(cyto) 4.8520].

representing an overall 77.4% increase in R. When approaching

each cytokine, IL-1 showed the lowest H value in activated state
with 0.0416 and the highest R with 99.1%. The cytokine who
showed the higher H contribution to the overall IT of MPA during
control state was IL-2; on the contrary, IL-10 was the cytokine that
showed the higher H value during activated state with an inverse
trend demonstrated by the slight increase in H( 0.8%) and
decrease in R(  0.5%) from control to activated state. Table 3
shows H and R both in each cytokine and in the whole MPA
system. The contribution of each cytokine to the H of the MPA is
showed in Fig. 1.

4. Discussion
Multiple arrays for proteome assays have become a crucial tool
for large-scale and high-throughput studies. These devices allow
fast, easy and simultaneous detection of several molecular target
in a single experiment (Hu et al., 2011). Unfortunately, the IT that
these systems could offer is markedly affected by the arrays layout

28

M.M. Ciulla et al. / Journal of Theoretical Biology 343 (2014) 2531

Fig. 1. Single cytokine contribution in Entropy (H). Graph bars comparing the
individual contribution in Entropy of each cytokine during control (light gray) and
activated state (dark gray).

that reects the established level of knowledge in each specic

eld of application.
Furthermore, on the theoretical point of view, is possible to
postulate that the IT of a biological redundant signaling system,
made up of n mediators, is higher if the values of the individual
analytes are independent of each other. Moreover, whether if the
analytes exhibit a signicant level of correlation, it is possible to
estimate their concentration with the regression equation in a
reasonable range of condence. By following these general concepts,
in this study we proposed a new translational approach to quantify IT
level in redundant biological systems. The analogy with language is
not new, in this regard other authors have previously proposed to use
the metaphor of language as a possible interpretative key of the
cytokine network (Kawade, 1990; Sporn and Roberts, 1988). Our
approach was translated from the Shannon's theory of information
by modifying the formulae to assess H and R. Our approach suggest
that the IT, obtain by an MPA designed to assess the cytokines serum
levels, expresses its maximum level of H in control state and the
minimum level of H during a specic activation state consisting in an
experimental myocardial cryoinjury. Furthermore, by assessing the
contribution of individual cytokine, it is possible to formulate an
unconventional hypothetical model to alternatively describe the
sequence of activation of the inammatory cascade after experimental myocardial infarction. Even if the description of a comprehensive inammation cascade lie outside from the topics of this
article, however, the identication of some analytes more informative than others allows to hypothesize their possible role as markers
of the inammatory response. For example, IL-1b showing a low IT is
already known to be a key cytokine in the inammatory cascade, on
the others side, IL-10 following experimental myocardial infarction in
our study features a high IT supporting a possible pivotal role in the
inammatory cascade as demonstrated also by some other authors
(Fuse et al., 2005; Kaur et al., 2006). Thus, in choosing of analytes to
set a custom analysis, the criterion based on the content of information can be a viable alternative to the ready-to-use MPAs set-up by
the manufacturer.
To the discussion of the present work, we have built a search
query in PubMed by using as keywords the 8 cytokines used in this
study as interleukin X, associated with the term acute myocardial infarction in the tag-eld [TIAB] to explore the availability
of matching documents already present in the database limited to
title and abstract elds.
With descriptive purposes only, we introduced the Fig. 2, which
summarizes the number of matching records obtained by search
query (beginMarch 2013) on PubMed next to the activated state
entropy values obtained in our study (Fig. 2a and b, respectively), the
latter previously shown as Fig. 1 (red bars). Looking now at the trend,
it is possible to identify some similarities and differences; in particular
IL-6 and TNF- in the literature appear to be the most studied

cytokines in acute myocardial infarction, corresponding to a good

level of entropy in our model. Conversely other cytokines less studied
in the literature as the IL-10 have a high level of entropy in our model.
This simple observation opens up to new possible and original
investigations concerning cytokines with high value IT, limited evidence in the literature and involved in acute myocardial infarction.
The idea that biochemical signaling systems of multicellular
organisms dedicated to the processing of environmental stimuli are
like complex information networks is not new and, itself, tautological; in fact they are complex signaling systems that, therefore, has to
do with information. Interestingly, starting from the thesis of Sebeok
that living systems are made as systems of signs which represent a
language at various levels, a school of thought has developed around
the application of semiotics in biology; for deepening this contribution we suggest the reader a recently published book (Emmeche and
Kull, 2011). At this regard, Kawade was a pioneering author in
approaching communication processes at the microscopic level by
studying the cytokine network and in suggesting the analogy to
language (Kawade, 1990, 1992). The paradigm of information theory
applied to protein signaling networks includes already few published
papers (o5 papers from a recent review of Medline by using
keywords such as information theory AND cytokines). Among
these articles, we suggest the reader a review recently published on
the Shannon's communication model applied to biological systems
(Waltermann and Klipp, 2011); nevertheless, in cited papers, the
point of view of the experimenter as receiver/decoder has been
neglected. One of the reasons of novelty of our work is to have
addressed, for the rst time, the problem of protein-based information networks from the point of view of the researcher, attempting to
answer the question: Which one of the cytokines commercially
available should we measure in order to obtain non-redundant
information when environmental conditions change toward the
pathophysiology?. This approach also enables to place in the general
model of communication developed by Shannon (Fig. 3) the perspective of the clinician looking for a reliable clinical marker in order
to establish the prognosis and indicate the therapy of a patient with
acute myocardial infarction. The main problems focused and the
solutions proposed in the context of the papers inspired by the
Shannon's information theory are summarized in Table 4.
Taking into consideration the limitations of this study, it must be
remembered that the experimental myocardial cryoinjury is not
precisely comparable to that observed with the coronary arteries
ligation; however, as previously demonstrated, this model may reliably
induce the inammatory cascade activation (Ciulla et al., 2008).
Moreover, another limitation of the study consists in having
analyzed only the level of cytokines in the serum and not in the
myocardial tissue; these two compartments are actually distinct,
although they are communicating after a tissue injury; therefore
our calculations are possibly distinct from those that could have
been done sampling myocardial tissue. On the other hand, serum
levels of inammatory cytokines are surely more easily comparable with the data that can also be obtained in humans with a
peripheral blood sampling. Moreover, in multicellular organisms,
the endocrine perspective enables a systemic view of signaling
systems such as cytokines.
It should be remarked that, since the actual number of cytokines
that composes the whole system at the moment is not known, the
information source we used is, inevitably, a partial view of the system
which, in any case, represents the current state of knowledge. Moreover, by assuming the cytokine system as a sort of communication
channel to convey information within the immune system, the
partiality of information is a basic concept of Shannon's theory since
communication is established only within the recipient capabilities of
understanding. By generalizing, biomedical research focused on the
response to the damage seen in terms of communication inevitably
stumbles in the general information theory as the recipient is not

M.M. Ciulla et al. / Journal of Theoretical Biology 343 (2014) 2531

29

Fig. 2. Amount of information in literature and entropy level of each cytokine. Graph bars comparing the number of matching documents listed in PubMed for each cytokine
limited to acute myocardial infarction (A) with the levels of Entropy for each cytokine in activated state (B).

Fig. 3. Parallelism between domain of human verbal and intercellular biology communication. In the biological domain transmission of information takes place from cell to
cell through contact and mediation. In complex organisms evolved receptor structures are used to receive and transmit stimuli through protein informative channels.
The receipt of the message is accomplished with the structuring of a specic response. In this perspective, the researcher is located outside to observe a cellular communication
session with the possibility to sample the message at various levels, often without a precise idea of what, when and at what level to take a sample. In experimental
conditions where it is possible to produce a known stimulus in a complex organism, such as limited myocardial damage, the experimenter is in the possibility of comparing
two conditions, the basal and the activated. The question is whether to study a complex organism and its behavior is actually more informative the activated state if
compared to the baseline condition. To explore the themes of communication and nd interesting similarities, we recommend the technical report written by Sebeok in 1994
for a government agency of the United States, publicly accessible since 2006 (Sebeok, 1984).

Table 4
Shannon's information theory within systems biology.
Issue to solve

Approach adopted

Model used

Reference

Cascades of events and cross-talk

Cascades of events and cross-talk
Signal separation in overlapping pathways
Information quantication of protein domains
Information exchange in signaling pathways
Negative feedback and information transmission in signaling pathways
Credit assignment in complex networks
Noise and redundancy

Measure of specicity of a network

Intrinsic/extrinsic specicity
Analysis of steady-state function
Mutual information and redundancy
Measure of cooperation
Sensitivity analysis
Sensitivity analysis
Mutual information

(a) Yeast proteins

(b) Yeast pheromones
Bacteria Vibrio harveyi
Protein domains
Proteins
Yeast

Mouse broblasts

(Komarova et al., 2005)

(Schaber et al., 2006)
(Mehta et al., 2009)
(Lenaerts et al., 2008)
(Lenaerts et al., 2009)
(Yu et al., 2008)
(Ludtke et al., 2008)
(Cheong et al., 2011)

only the immune system but, also, the experimenter who reads the
information from the outside, according to his peculiar point of view
(Fig. 3). As a result of the plethora of information that can be obtained
from a systematic approach to signaling proteins, the reasoning in
terms of networks and the development of complex mathematical
models seems inevitable to biology (Janes and Lauffenburger, 2013).
Although other statistical models are available to analyze and
represent the mutual interdependence between a data cluster, the
comparison of different statistical tests was not among of objectives
of the study; we used the correlativeregressive model because of
high reciprocal dependence of variables, and, specically, the p-value,
because is (a) a concise parameter, (b) reproducible (c) widely used in
bio-medical research to determine the degree of association between
the variables in study and (d) compatible with the original formulas
of Shannon.

Finally it should be acknowledged that the Shannon's information theory and formulae used in this study for the rst time is
translated from his original eld to nd a possible application in
biology. It should also be remembered that often the application of
models borrowed from other disciplines allows a different perspective, able to open scenarios not previously observed or nd
applications with more functional approaches (e.g. application of
nanotechnology in biotechnology elds).

5. Conclusions
Assuming that MPA are informative devices, designed according to general knowledge about each analyte and its supposed
functions, the results are highly affected (a) by the initial choice of

30

M.M. Ciulla et al. / Journal of Theoretical Biology 343 (2014) 2531

Fig. 4. Information content and redundancy in a biological system, during control state and activated state. Graphical representation of a dialog between 8 cytokine
represented as comic characters, under control conditions and after activation, in this case an injury. In basal conditions each character speaks a different letter contributing
to a general increase in information content (high entropy and low redundancy). After stimulation, all the characters say substantially the same sentence, even if with a
different style, reducing the information content to the single word damage.

the analytes, often made by the manufacturer or the researcher,

and (b) by the correct application to a biological context. Furthermore, the increase in information obtained by simultaneous
measurement of several cytokines remains to be determined, since
cytokines exhibit pleiotropic and redundant qualities. The paradoxical deterioration of information content after activation of the
biological system has been depicted in Fig. 4.
The quantication of IT applied to redundant systems in biology
can provide interesting insights to understand the complex puzzle of
the network of cytokines, both local and systemic, and as dening
the role of each molecule in the cascade of inammatory activation.

Authors' contributions
MMC: conception of the translational model, design of the
study, echocardiograms of the animals, data analysis, drafting,
revising and nal approval of the paper; FDM: verication of the
mathematical model, design of the study, data analysis, revising
the paper; EM: design of the study, set-up of the analytes,
sampling and analysis of the sera, and analysis, revising and nal
approval of the paper; LL: design of the study, set-up of the
analytes, sampling and analysis of the sera, and analysis, revising
and nal approval of the paper; GLP: biological predictability of
the model, data analysis, drafting, revising and nal approval of
the paper; FM: drafting, revising and nal approval of the paper.

Acknowledgments
We thank Dr. Roberta Paliotti for the preliminary check of the
mathematical model.

Therefore, we dened the Entropy (H) of a system of n

cytokines as:
nn  1

Hn cyto

i1

ln1 pi

for i 1; 2; ; nn  1

H(n cyto) ranges from 0 in a hypothetical system where each

cytokine perfectly correlates with each other to a positive number
(maximum theoretical entropy).
It is proportional (continuous), i.e. changing a p-value of a little
amount changes the entropy of the system by a little amount.
Increasing the number of cytokines increases the maximum
theoretical entropy.
It is an additive value, i.e. total entropy of a system is the sum of
the single entropy of each one of its components.
Hn cyto Hcyto 1 Hcyto 2 Hcyto n

Entropy of a single cytokine in a system of n cytokines:

Hcyto

n  1

i1

ln1 pi

for i 1; 2; ; n  1

Maximum theoretical entropy of a single cytokine in a system of n

cytokines and maximum theoretical entropy of a whole system of
n cytokines, supposing no correlation between each two of the
cytokines (p 1 for every correlation in a linear regression model),
are:
H max cyto n  1 ln2



H max n cyto nn  1 ln2

We dened Redundancy (R) as the normalized measure of the

entropy of a system of n cytokines with regard to its maximal
theoretical entropy.
Hn cyto
H max n cyto

Appendix

Rn cyto 1 

Entropy of information describes how much randomness there

is in a signal or in a random event. In terms of information,
Entropy describes how much unique information are carried by
the signal. By assuming the signal as the complex signals given by
a multiplex proteome arrays, entropy is high when quantitative
measurements are unrelated to each other (random) and low
when measurements are highly correlated (randomness).

The higher the redundancy the more the cytokines are selfcorrelated, while the lower the redundancy the more the cytokines are independent one another.
Similarly, R of a single cytokine is
Rcyto 1 

Hcyto
H max cyto

M.M. Ciulla et al. / Journal of Theoretical Biology 343 (2014) 2531

And we can dene the contribution of each cytokine to total

entropy of the system:
Contribution

Hcyto
Hn cyto

References
Cheong, R., Rhee, A., Wang, C.J., Nemenman, I., Levchenko, A., 2011. Information
transduction capacity of noisy biochemical signaling networks. Science 334,
354358, http://dx.doi.org/10.1126/science.1204553.
Ciulla, M.M., Paliotti, R., Ferrero, S., Braidotti, P., Esposito, A., Gianelli, U., Busca, G.,
Ciof, U., Bulfamante, G., Magrini, F., 2004. Left ventricular remodeling after
experimental myocardial cryoinjury in rats. J. Surg. Res. 116, 9197. (United
States).
Ciulla, M.M., Montelatici, E., Ferrero, S., Braidotti, P., Paliotti, R., Annoni, G., De
Camilli, E., Busca, G., Chiappa, L., Rebulla, P., Magrini, F., Lazzari, L., 2008.
Potential advantages of cell administration on the inammatory response
compared to standard ACE inhibitor treatment in experimental myocardial
infarction. J. Transl. Med. 6, 30. (England).
Dinarello, C.A., 2010. Anti-inammatory agents. Present Future Cell 140, 935950,
http://dx.doi.org/10.1016/j.cell.2010.02.043.
Emmeche, C., Kull, K., 2011. Towards a Semiotic BiologyLife is the Action of Signs.
Imperial College Press 57, Shelton Street Covent Garden London. (WC2H 9HE).
Frangogiannis, N.G., 2012. Regulation of the inammatory response in cardiac
repair. Circ. Res. 110, 159173, http://dx.doi.org/10.1161/circresaha.111.243162.
Fuse, K., Kodama, M., Okura, Y., Ito, M., Kato, K., Hanawa, H., Aizawa, Y., 2005. Shortterm prognostic value of initial serum levels of interleukin-10 in patients with
acute myocarditis. Eur. J. Heart Fail 7, 109112. (Netherlands).
Guide for the Care and Use of Laboratory Animals. 1996. The National Academies
Press.
Hu, S., Xie, Z., Qian, J., Blackshaw, S., Zhu, H., 2011. Functional protein microarray
technology. Wiley Interdiscip. Rev. Syst. Biol. Med. 3, 255268, http://dx.doi.
org/10.1002/wsbm.118.
Janes, K.A., Lauffenburger, D.A., 2013. Models of signalling networkswhat cell
biologists can gain from them and give to them. J. Cell Sci. 126, 19131921, http:
//dx.doi.org/10.1242/jcs.112045.
Janes, K.A., Albeck, J.G., Gaudet, S., Sorger, P.K., Lauffenburger, D.A., Yaffe, M.B., 2005.
A systems model of signaling identies a molecular basis set for cytokineinduced apoptosis. Science 310, 16461653, http://dx.doi.org/10.1126/
science.1116598.
Kaur, K., Sharma, A.K., Dhingra, S., Singal, P.K., 2006. Interplay of TNF-alpha and
IL-10 in regulating oxidative stress in isolated adult cardiac myocytes. J. Mol.
Cell Cardiol 41, 10231030. (England).

31

Kawade, Y., 1990. Cytokine network in analogy to languagea general view of

interferon research from a distance. J. Interferon Res. 10, 101107.
Kawade, Y., 1992. A molecular semiotic view of biology. Interferon and homeokine'
as symbols. Riv. Biol. 85, 7178.
Komarova, N.L., Zou, X., Nie, Q., Bardwell, L., 2005. A theoretical framework for
specicity in cell signaling. Mol. Syst. Biol. 1, http://dx.doi.org/10.1038/
msb4100031. (2005.0023).
Lenaerts, T., Ferkinghoff-Borg, J., Stricher, F., Serrano, L., Schymkowitz, J.W.,
Rousseau, F., 2008. Quantifying information transfer by protein domains:
analysis of the Fyn SH2 domain structure. BMC Struct. Biol. 8, 43, http://dx.
doi.org/10.1186/1472-6807-8-43.
Lenaerts, T., Ferkinghoff-Borg, J., Schymkowitz, J., Rousseau, F., 2009. Information
theoretical quantication of cooperativity in signalling complexes. BMC. Syst.
Biol. 3, 9, http://dx.doi.org/10.1186/1752-0509-3-9.
Ludtke, N., Panzeri, S., Brown, M., Broomhead, D.S., Knowles, J., Montemurro, M.A.,
Kell, D.B., 2008. Information-theoretic sensitivity analysis: a general method for
credit assignment in complex networks. J. R. Soc. Interface 5, 223235, http:
//dx.doi.org/10.1098/rsif.2007.1079.
Mehta, P., Goyal, S., Long, T., Bassler, B.L., Wingreen, N.S., 2009. Information
processing and signal integration in bacterial quorum sensing. Mol. Syst. Biol.
5, 325, http://dx.doi.org/10.1038/msb.2009.79.
Prabhu, S.D., 2004. Cytokine-induced modulation of cardiac function. Circ. Res. 95,
11401153, http://dx.doi.org/10.1161/01.res.0000150734.79804.92.
Schaber, J., Kofahl, B., Kowald, A., Klipp, E., 2006. A modelling approach to quantify
dynamic crosstalk between the pheromone and the starvation pathway in
baker's yeast. FEBS J. 273, 35203533, http://dx.doi.org/10.1111/j.17424658.2006.05359.x.
Sebeok, T. A., 1984. Communication measures to Bridge Ten Millennia. (Contains
glossary). pp. Medium: ED; Size: Pages: 43.
Shannon, C.E., 1997. The mathematical theory of communication. 1963. MD
Comput. 14, 306317.
Shannon, C.E., 1998. Communication theory of secrecy systems. 1945. MD Comput.
15, 5764.
Sporn, M.B., Roberts, A.B., 1988. Peptide growth factors are multifunctional. Nature
332, 217219, http://dx.doi.org/10.1038/332217a0.
Waltermann, C., Klipp, E., 2011. Information theory based approaches to cellular
signaling. Biochim. Biophys. Acta 1810, 924932, http://dx.doi.org/10.1016/j.
bbagen.2011.07.009.
Wang, C.C., Huang, R.P., Sommer, M., Lisoukov, H., Huang, R., Lin, Y., Miller, T., Burke,
J., 2002. Array-based multiplexed screening and quantitation of human
cytokines and chemokines. J. Proteome Res. 1, 337343.
Yu, R.C., Pesce, C.G., Colman-Lerner, A., Lok, L., Pincus, D., Serra, E., Holl, M.,
Benjamin, K., Gordon, A., Brent, R., 2008. Negative feedback that improves
information transmission in yeast signalling. Nature 456, 755761, http://dx.
doi.org/10.1038/nature07513.
Zhu, H., Snyder, M., 2003. Protein chip technology. Curr. Opin. Chem. Biol. 7, 5563.
(England).