POLITECNICO DI MILANO

FACOLTÀ DI INGEGNERIA DEI SISTEMI

Corso di Laurea Specialistica in Ingegneria Matematica

A MODEL FOR CELL GROWTH

IN BATCH BIOREACTORS

Relatore
Dott. Nicola Parolini

Tesi di laurea di
Susanna Carcano
matricola 721016
Anno Accademico 2009/2010
Contents

Introduction 1

Introduzione 5

1 State of the art on cell growth models 9

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.2 Simple microbial kinetics: primary models . . . . . . . . . . . 12
1.2.1 The Malthusian model . . . . . . . . . . . . . . . . . . 12
1.2.2 The logistic model . . . . . . . . . . . . . . . . . . . . 13
1.2.3 The Gompertz model . . . . . . . . . . . . . . . . . . . 14
1.2.4 The modied logistic and the modied Gompertz
models . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.3 Secondary models . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.3.1 The Monod model . . . . . . . . . . . . . . . . . . . . 16
1.3.2 Modied Monod models . . . . . . . . . . . . . . . . . 19
1.3.3 Secondary models based on the Arrhenius law . . . . . 22
1.3.4 The gamma concept . . . . . . . . . . . . . . . . . . . 24
1.4 Structured kinetic models . . . . . . . . . . . . . . . . . . . . 25
1.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

2 Characterization of CHO cells physiology 28

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.2 Cell growth curve . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.3 Cellular metabolism . . . . . . . . . . . . . . . . . . . . . . . 33
2.3.1 Glycolysis . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.3.2 Anaerobic respiration . . . . . . . . . . . . . . . . . . 36
2.3.3 Aerobic respiration . . . . . . . . . . . . . . . . . . . . 37

I
CONTENTS II

2.3.4 Glutamine oxidation . . . . . . . . . . . . . . . . . . . 40

2.4 Environmental factors . . . . . . . . . . . . . . . . . . . . . . 41
2.4.1 Temperature . . . . . . . . . . . . . . . . . . . . . . . 41
2.4.2 Oxygen concentration . . . . . . . . . . . . . . . . . . 42
2.4.3 pH . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

3 Batch bioreactor modeling 45

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.2 CHO cell culture . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.3 Bioreactor classication . . . . . . . . . . . . . . . . . . . . . 47
3.4 Balance equations for the batch fermenter . . . . . . . . . . . 49
3.5 Mass transfer . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.5.1 Oxygen and carbon dioxide balances for gas-liquid trans-
fer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.5.2 Models for oxygen transfer in large scale bioreactors . 53

4 A new cell growth model 54

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
4.2 0D model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.2.1 Cell growth rate . . . . . . . . . . . . . . . . . . . . . 57
4.2.2 Substrate uptake rate . . . . . . . . . . . . . . . . . . 59
4.2.3 Product formation rate . . . . . . . . . . . . . . . . . 61
4.2.4 Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.2.5 Numerical approximation . . . . . . . . . . . . . . . . 64
4.3 1D model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.3.1 Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . 67
4.3.2 Numerical approximation . . . . . . . . . . . . . . . . 68
4.4 3D model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.4.1 Numerical approximation . . . . . . . . . . . . . . . . 76
4.5 Environmental factors . . . . . . . . . . . . . . . . . . . . . . 77
4.5.1 Temperature dependance . . . . . . . . . . . . . . . . 77
4.5.2 pH estimation . . . . . . . . . . . . . . . . . . . . . . . 79

5 Numerical simulations 85
5.1 Preliminary remarks . . . . . . . . . . . . . . . . . . . . . . . 85
5.1.1 Cell culture . . . . . . . . . . . . . . . . . . . . . . . . 85
CONTENTS III

5.1.2 Units of measurement . . . . . . . . . . . . . . . . . . 86

5.1.3 Values of parameters . . . . . . . . . . . . . . . . . . . 86
5.2 0D model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
5.2.1 pH estimation . . . . . . . . . . . . . . . . . . . . . . . 88
5.2.2 Temperature dependance . . . . . . . . . . . . . . . . 90
5.2.3 KL a dependance . . . . . . . . . . . . . . . . . . . . . 91
5.3 1D model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
5.4 3D model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101

Closing remarks and future work 111

Ringraziamenti 115
List of Figures

1.1 Balance regions for dierent types of kinetic models (from [5]) 11
1.2 Solution of the Malthusian model with X0 = 0.3 · 105 [cell/ml]
and µ = 0.03 [1/h] . . . . . . . . . . . . . . . . . . . . . . . . 13
1.3 Solution of the logistic model with X0 = 0.3 · 105 [cell/ml],
µ = 0.08 [1/h] and K = 2 · 106 [cell/ml] . . . . . . . . . . . . 14
1.4 Solution of the Gompertz model with a = 2 · 106 [cell/ml],
b = 10, c = 0.05 [1/h] . . . . . . . . . . . . . . . . . . . . . . . 15
1.5 Solution of the Monod model with X0 = 0.3 · 105 [cell/ml],
S0 = 0.2 [g/l], µmax = 0.1 [1/h], KS = 0.1 [g/l], YXS = 0.1 . . 18
1.6 Solution of the multiple-substrate Monod model with X0 =
0.3 · 105 [cell/ml] . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.7 Solution of the double-Monod model with X0 = 0.3·105 [cell/ml] 20
1.8 Solution of the diauxic Monod model with X0 = 0.3 · 105
[cell/ml] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.9 Solution of the Monod model with the maintenance substrate
uptake with X0 = 0.3 · 105 [cell/ml] and m = 0.01 [kg sub-
strate/kg cells·h] . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.10 Eect of temperature on reaction rate predicted using the Ar-
rhenius law (solid line) and the estimated eect of temperature
on the cell growth rate (dashed line) . . . . . . . . . . . . . . 23

2.1 Cell growth curve . . . . . . . . . . . . . . . . . . . . . . . . . 30

2.2 Glycolysis pathway . . . . . . . . . . . . . . . . . . . . . . . . 35
2.3 Lactic fermentation . . . . . . . . . . . . . . . . . . . . . . . . 36
2.4 Anaerobic respiration . . . . . . . . . . . . . . . . . . . . . . . 37
2.5 Pyruvate oxidative decarboxylation . . . . . . . . . . . . . . . 38

IV
LIST OF FIGURES V

2.6 TCA cycle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

2.7 Cell concentration at dierent temperatures (from [1]) . . . . 42
2.8 Cell concentration at dierent oxygen concentrations: 20 %
(red) and anoxia (green) (from [1]) . . . . . . . . . . . . . . . 43
2.9 pH at dierent oxygen concentrations: 20 % (red) and anoxia
(green) [1] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

3.1 Shaken bioreactor under batch conditions . . . . . . . . . . . 46

4.1 1D domain . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64

5.1 Cell growth: comparison between model results and experi-

mental results (points) [1] . . . . . . . . . . . . . . . . . . . . 89
5.2 Cell metabolism: comparison between model results and expe-
rimental results (points) [1] . . . . . . . . . . . . . . . . . . . 89
5.3 Cell metabolism: oxygen (left) and carbon dioxide (right) con-
centration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
5.4 pH estimation from the model results . . . . . . . . . . . . . . 90
5.5 Maximum specic growth rate as a function of temperature . 92
5.6 Cell growth as a function of temperature: model results (left)
and experimental results (right) . . . . . . . . . . . . . . . . . 93
5.7 Cell growth model: results for dierent temperatures . . . . . 93
5.8 Cell growth as a function of the mass transfer coecient KL a:
model results . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
5.9 Model results as a function of the mass transfer coecient KL a 95
5.10 pH as a function of the mass transfer coecient KL a: model
results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
5.11 CultiFlask50 Bioreactor . . . . . . . . . . . . . . . . . . . . . 97
5.12 Cell growth: comparison between the 1D model and the 0D
model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
5.13 Comparison between the 1D model and the 0D model . . . . 99
5.14 1D cell growth model: spatial dependance . . . . . . . . . . . 100
5.15 1D model: spatial dependance . . . . . . . . . . . . . . . . . . 100
5.16 Velocity eld (5.2) . . . . . . . . . . . . . . . . . . . . . . . . 101
5.17 Cell growth: comparison between the 3D model and the 0D
model with the steady uid . . . . . . . . . . . . . . . . . . . 104
LIST OF FIGURES VI

5.18 Comparison between the 3D model and the 0D model with

the steady uid . . . . . . . . . . . . . . . . . . . . . . . . . . 104
5.19 Cell growth: comparison between the 3D model and the 0D
model (velocity eld (5.2) with A = 0.003 m/s) . . . . . . . . 105
5.20 Comparison between the 3D model and the 0D model (velocity
eld (5.2) with A = 0.003 m/s) . . . . . . . . . . . . . . . . . 105
5.21 Cell growth: comparison between the 3D model and the 0D
model (velocity eld (5.2) with A = 0.03 m/s) . . . . . . . . . 106
5.22 Comparison between the 3D model and the 0D model (velocity
eld (5.2) with A = 0.03 m/s) . . . . . . . . . . . . . . . . . . 106
5.23 Cell growth: comparison between the 3D model and the 0D
model (velocity eld (5.2) with A = 0.03 m/s) and KLCO2 a =
0.8 · KLO2 a . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
5.24 Comparison between the 3D model and the 0D model (velocity
eld (5.2) with A = 0.03 m/s) and KLCO2 a = 0.8 · KLO2 a . . . 107
5.25 3D model results: comparison between the steady uid (left)
and the transport eld (5.2) (right) at t = 120 h . . . . . . . . 108
5.26 3D model results: oxygen concentration at t = 0 h, t = 40
h, t = 80 h, t = 120 h with the transport eld (5.2) with
A = 0.03 m/s . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
5.27 3D model results: carbon dioxide concentration at t = 0 h,
t = 40 h, t = 80 h, t = 120 h with the transport eld (5.2)
with A = 0.03 m/s . . . . . . . . . . . . . . . . . . . . . . . . 110
Introduction

The use of animal cells for the production of recombinant proteins used
in human therapy has gained great relevance at both development and ap-
plication level in industrial production processes.
Today, in particular, Chinese hamster ovary (CHO) cells are the most
commonly used mammalian hosts in biological and medical research and
commercially in large scale recombinant protein production. CHO cells are
a cell line derived from the ovary of the Chinese hamster. These cells are
widely used for the production of therapeutic proteins such as prothrombin,
a glycoprotein precursor of thrombin, that is necessary for the coagulation
of blood, and thrombopoietin, a glycoprotein hormone that regulates the
production of platelets by the bone marrow. The CHO cell line is widely used
as it is known for being a highly stable host with respect to the expression
of heterologous genes and for its immense adaptive ability.
Despite the signicant advances and improvement during the last past
decade in cell line development, as well as in medium optimization, process
monitoring and control, the current paradigms of cell line development re-
main, to large extent, empirical [12], and the development of cultures for
this type of cells still presents a series of problems. These diculties prima-
rily stem from our inadequate understanding of the biology and physiology
of mammalian cells. Secondly, there is a considerable variability and very
little understanding of the sources of variation in mammalian cell culture
processes. Moreover, actually there are no reliable methods for predicting
or modeling the growth characteristics and production capabilities in large-
scale bioreactors. Indeed, the cells selected and characterized in bench-top
reactors may not behave similarly in large-scale bioreactors, despite employ-
ing identical process parameters.

1
 These problems can be faced in distinct ways. The rst approach consists
in the genetic manipulation of the cell. It is possible, for example, to enhance
cell viability and to modify and improve its metabolic pathways in order to
reduce the production of toxic by-product and to increase the production
of recombinant proteins. The second approach consists in the improvement
of the culture strategies, for example by modifying the bioreactor operative
conditions or the culture medium composition. Fed-batch strategies for the
addition of metabolic substrates and the elimination of the inhibitory com-
pounds have also been considered; however, both these strategies require a
high degree of instrumentation to be carried out in an optimal way. While
the rst approach, thanks to the sequencing of mammalian genomes, is based
on the exact knowledge of the gene-coding and regulatory element sequences
of the organism, the second approach is actually completely empirical, that
is based on experimental observations.

In predicting the cell growth behavior, the use of mathematical models is

gaining more and more attention because it guarantees an increasing insight
into the process. For process design and optimization, simulation studies
can be very useful to reduce the number of expensive and time-consuming
experiments. Accumulating sucient data on the dynamics of cells growth,
in fact, requires an extensive amount of work and it is costly. In addiction,
even if data can describe cell growth, they provide little insight into the re-
lationship between physiological processes and growth or survival. One way
this link can be made is through the use of mathematical models.

An investigation of bioreactor performance and cell growth behavior

might conventionally be carried out in an almost entirely empirical manner.
In this approach, the bioreactor behavior should be studied under practi-
cally all combinations of possible operating conditions and the results then
expressed as series of correlations, from which the resulting performance
might hopefully be estimated for any given set of new operating conditions.
This empirical procedure can be carried out in a very routine way and is
reasonably simple to set up. While this might seem to be rather conve-
nient, the procedure has actually many disadvantages, since very little real

2
understanding of the process would be obtained. Moreover very many ex-
periments would be required in order to obtain correlations that would cover
every process.
Compared to this, the modeling approach attempts to describe both
actual and potential bioreactor performance, by means of well-established
theory, which describes, in mathematical terms, a working model for the
process. In carrying out a modeling exercise, it is necessary to consider the
nature of all the important parameters of the process, their eect on the
process and how each parameter can be dened in quantitative terms. Thus
the very act of modeling is one that forces a better understanding of the
process, since all the relevant theory must be critically assessed. In addition,
the task of formulating theory into terms of mathematical equations is also
a very positive factor that forces a clear formulation of basic concepts.
Once formulated, the model can be solved and the behavior predicted by
the model compared with experimental data. Any dierences in performance
may be used to further redene or rene the model until good agreement is
obtained. Once the model is established, it can then be used, with reasonable
condence, to predict performance under dierent process conditions, and
it can also be used for such purposes as process design, optimization and
control. An input of experimental data is, of course, required in order to
assess or validate the model, but the quantity of experimental data required,
when compared to that of the empirical approach, is considerably reduced.
Apart from this, the major advantage gained is the increased understanding
of the process that one obtains simply by carrying out the modeling exercise.

Starting from these introductive considerations, our purpose is to develop

a mathematical model to describe CHO cell growth dynamics as a function
of the culture medium composition and the bioreactor operative conditions.
In this study, we follow a deterministic approach: we investigate the physical
and chemical properties of the process of CHO cell growth in order to under-
stand which are the governing factors. In particular, we analyze deeply cell
physiology, paying particular attention to the the mechanism of metabolism
that allows the cell to grow, maintain and reproduce. Then we try to un-
derstand in which way the environmental properties, like temperature and
concentrations of nutrients, and the bioreactor operative conditions can in-

3
uence cell growth dynamics. Consequently, the model that we are going
to introduce, is based on the investigation of some of the most important
mechanisms that govern the cell growth process.

This work has been carried out in collaboration with the Chair of Mode-
ling and Scientic Computing (CMCS), the Cellular Biotechnology Labora-
tory (LBTC) and the Hydraulic Machines Laboratory (LMH) of the EPFL
in the framework of the project Fluid dynamics and mixing behavior in or-
bitally shaken bioreactors for mammalian cell cultivation (Sinergia Research
Proposal 2009-2011), approved by the Swiss National Science Foundation
(FNS). In particular, we will refer to some of the experiments carried out at
the LBTC in order to obtain important experimental information.

The work is structured in ve chapters. In Chapter 1 we present and

discuss the state of the art on cell growth models in order to detect the
advantages and the limits related to the application of the most popular
models in the literature before presenting our own cell growth model.
In Chapter 2 we analyze the most important characteristics of CHO cells,
in particular focusing on cell metabolism and on the main environmental
factors that aect cell growth process.
In Chapter 3 we describe the main properties of CHO cell culture and
we introduce in particular a model for the batch fermenter.
In Chapter 4 we present a new model for cell growth which is based on
the analysis and on the considerations made in the previous chapters. In this
model the cell dynamics is function of time, substrates concentrations and
temperature. Later on, the model is extended in order to take into account
the spatial dependance.
In Chapter 5 we discuss the most signicant numerical results obtained
with the models introduced in the previous chapter and we make some com-
parison with the experimental results.

4
Introduzione

L'impiego di cellule animali per la produzione di proteine a scopo tera-

peutico ha acquistato una grande importanza sia a livello di ricerca, sia a
livello industriale.
Oggi, in particolare, le cellule CHO (Chinese Hamster Ovary cells) sono
le cellule maggiormente utilizzate nella ricerca medica e nella produzione di
proteine su larga scala. Questo tipo di cellule è utilizzato ad esempio per
la produzione di protrombina, una glicoproteina precursore della trombina,
necessaria per la coagulazione del sangue, e della trombopoietina, una glico-
proteina che regola la produzione di piastrine nel midollo osseo.
Nonostante i signicativi progressi e miglioramenti compiuti nel corso
degli ultimi dieci anni nello sviluppo di colture cellulari, così come nell'otti-
mizzazione del mezzo di coltura, nel monitoraggio e nel controllo del processo,
i correnti paradigmi relativi allo sviluppo di colture cellulari sono essenzial-
mente di carattere empirico. La progettazione di colture presenta ancora
una serie di problematiche. Queste dicoltà derivano principalmente da
una inadeguata ed incompleta conoscenza delle caratteristiche siologiche
delle cellule dei mammiferi. In secondo luogo, c'è una notevole variabilità
ed comprensione una molto limitata dei fattori che inuenzano il processo di
coltura cellulare. Inne, ad oggi, non esistono metodi accurati per predire e
modellare la dinamica del processo.
Questa tipologia di problemi può essere arontata in dierenti modi.
Il primo approccio consiste nella manipolazione genetica delle cellule. E'
possibile infatti incrementarne la probabilità di sopravvivenza e migliorarne
il metabolismo in modo tale da ridurre la produzione di sostanze tossiche
e aumentare la produzione di proteine. Il secondo approccio consiste nel
migliorare le strategie di coltura, per esempio modicando le condizioni ope-

5
rative del bioreattore o la composizione del mezzo di coltura. Sono state
anche considerate strategie basate sull'aggiunta di sostanze nutritive e sulla
rimozione di sostanze inibitrici. Entrambe queste strategie richiedono però
una complessa e costosa strumentazione per essere applicate in maniera ef-
cace ed ottimale.
Mentre il primo approccio, grazie al sequenziamento del genoma dei mam-
miferi, è basato sull'esatta conoscenza del codice genetico ed in particolare
delle sequenze che inuiscono sulle funzioni regolatrici sull'organismo, il se-
condo approccio è attualmente basato su conoscenze di tipo empirico, cioè
su osservazioni sperimentali.

Nel predire la dinamica del processo di crescita cellulare, l'utilizzo di

modelli matematici sta guadagnando sempre maggiore attenzione in quanto
garantisce una migliore e più profonda conoscenza del processo. Le simu-
lazioni numeriche, applicate alla progettazione e all'ottimizzazione del pro-
cesso, possono essere estremamente utili per ridurre il numero dei costosi
e lunghi esperimenti attualmente necessari per sviluppare una tecnologia
eciente. Inoltre, nonostante i dati sperimentali raccolti siano utili per de-
scrivere il processo, essi non forniscono maggiori informazioni sulle relazioni
che intercorrono tra siologia cellulare, condizioni operative e produzione di
cellule e proteine. Uno dei modi possibili per approfondire queste relazioni
è basato sull'uso di modelli matematici.

Lo studio del rendimento di un bioreattore e del processo di crescita cellu-

lare può essere portato avanti adottando un approccio quasi interamente em-
pirico. Il comportamento del bioreattore dovrebbe essere analizzato in tutte
le possibili condizioni operative ed i risultati quindi espressi come serie di
correlazioni, dalle quali stimare il rendimento per ogni dato insieme di nuove
condizioni operative. Questa procedura empirica risulta essere ragionevol-
mente semplice, ma non favorisce una reale comprensione del processo. In-
oltre per poter ottenere informazioni relative ad ogni possibile condizione
operativa è necessario un gran numero di esperimenti.
Al contrario un approccio di tipo modellistico ha lo scopo di descrivere il
rendimento attuale e potenziale del bioreattore sulla base di teorie accettate
che descrivono, in termini matematici, il processo. Per formulare il mod-

6
ello, è necessario considerare la natura di tutti i più importanti parametri
ed il loro eetto sul processo, cercando di darne una denizione in termini
quantitativi. L'esercizio stesso di modellazione, di conseguenza, stimola una
migliore comprensione del processo.
Una volta formulato, il modello può essere risolto ed il comportamento
previsto dal modello confrontato con i risultati sperimentali. Ogni dierenza
riscontrata può essere sfruttata per ridenire o ranare il modello nché
non si ottiene un buon accordo con i risultati attesi. Una volta riconosciuta
l'adabilità del modello, questo può essere applicato per predire il rendi-
mento in dierenti condizioni operative, per la progettazione, l'ottimizzazione
e il controllo del processo.
Anche adottando un approccio modellistico sono necessari dati speri-
mentali per validare il modello, ma la quantità di informazioni richieste è
inferiore se confrontato con quella necessaria adottando un approccio intera-
mente empirico. Il maggiore vantaggio risulta comunque essere quello di una
più profonda comprensione dei meccanismi alla base del processo.

Partendo da queste considerazioni introduttive, l'obiettivo di questa tesi è

stato lo sviluppo un modello matematico per descrivere il processo di crescita
delle cellule CHO in funzione della composizione del mezzo di coltura e delle
condizioni operative del bioreattore. Nello sviluppo del modello si è de-
ciso di adottare un approccio di tipo deterministico: si sono analizzate le
proprietà siche e chimiche del processo di crescita cellulare con l'obiettivo
di comprendere quali fossero i fattori determinanti. Sono stati analizzati i
principali aspetti della siologia cellulare, in particolare i processi metabol-
ici che permettono alla cellula di crescere, mantenersi e riprodursi. In se-
guito si è cercato di comprendere le modalità in cui le condizioni ambientali,
come temperatura e concentrazione di nutrienti, e le condizioni operative del
bioreattore possono inuenzare la dinamica del processo. In conclusione, il
modello così sviluppato è dunque basato sull'analisi di alcuni dei principali
meccanismi che governano il processo di crescita cellulare.

Questo lavoro è stato portato avanti in collaborazione con il Laborato-

rio CMCS (Chair of Modeling and Scientic Computing) ed il Laboratorio
LBTC (Laboratoire de Biotechnologie Cellulaire) dell'Ecole Polytechnique

7
LIST OF FIGURES 8

Fédérale de Lausanne (EPFL) nell'ambito del progetto "Fluid dynamics and

mixing behavior in orbitally shaken bioreactors for mammalian cell cultiva-
tion" nanziato dalla Swiss National Science Foundation (FNS). In partico-
lare la validazione del modello è stata svolta in collaborazione con il Labora-
torio LBTC, che ha messo a disposizione i risultati delle prove sperimentali
eettuate per ricavare alcune importanti informazioni relative al processo
analizzato.

Il lavoro è strutturato in cinque capitoli. Nel Capitolo 1 è presentato e

discusso lo stato dell'arte sui modelli di crescita cellulare con l'obiettivo di
individuare i principali vantaggi e limiti legati all'applicazione dei modelli
più diusi in letteratura.
Nel Capitolo 2, vengono analizzate le principali caratteristiche delle cel-
lule CHO; in particolare vengono studiati in dettaglio il metabolismo cel-
lulare e i più importanti fattori ambientali che inuenzano il processo di
crescita cellulare.
Nel Capitolo 3, vengono descritte le proprietà che contraddistinguono le
colture di cellule CHO e viene introdotto un opportuno modello matematico
per il bioreattore.
Nel Capitolo 4, viene presentato un nuovo modello di crescita cellulare
basato sulle analisi e sulle considerazioni discusse nei capitoli precedenti.
Il modello introdotto descrive la dinamica cellulare in funzione del tempo,
della concentrazione di nutrienti e della temperatura. In seguito, il modello
è ampliato al ne di considerare la dipendenza del fenomeno dalla profondità
e dal moto del mezzo di coltura.
Nel Capitolo 5, inne, vengono discussi i più signicativi risultati nu-
merici ottenuti applicando i modelli introdotti nel capitolo precedente.
Chapter 1

State of the art on cell growth

models

1.1 Introduction
The growth of bacteria and cells has been the subject of much study over
the years and in the scientic literature we can quite easily nd dierent
examples of cell growth models. In this chapter we want to introduce and
compare some of the most popular models, in order to detect the advantages
and the limits related to the application of each one before developing our
own cell growth model.

Various types and classications of models are possible. For example, we

can divide models into empirical and mechanistic models. Empirical models
are essentially pragmatic, and simply describe a set of data in a convenient
mathematical relationship with no consideration of underlying phenomena.
On the contrary, mechanistic models are build up from theoretical bases
and, if they are correctly formulated, can allow the response to be inter-
preted in terms of known physical, chemical and biological phenomena. In
one sense, if this approach is adopted, it follows that the parameters in such
models might be readily interpretable properties of the system under study,
and that the mathematical form of the model would enable interpretation of
the interactions between those factors. Interpretability of model parameters
is a feature highly valued by many authors in the predictive microbiology

9
CHAPTER 1. STATE OF THE ART ON CELL GROWTH MODELS 10

literature. Although the development of predictive microbiology has seen

the embedding of more and more mechanistic elements in the models, or
at least the development of models whose structure and parametrization re-
ects known or hypothesized phenomena, in practice many models currently
available, that we are going to introduce now, are not purely empirical nor
purely mechanistic.

Cell growth models can be divided into structured, unstructured and se-
gregated models [5]. A structured model attempts to explicitly describe intra-
cellular processes in both a structural and physiological sense, and thus of-
fers the most realistic representation of a cell. Structured models are usually
single-cell models that look in some detail at individual processes and reac-
tions. However, whereas cell behavior to varying growth conditions is not
thoroughly understood, a structured model is not applicable to most of cell
lines, for example to CHO cells. Therefore, the majority of the kinetic models
in literature are categorized as unstructured: they rely on the global rela-
tionships between cell growth and environmental properties of the culture,
and neglect intracellular processes. A third type of models are the segregated
ones, which consider dierent stages of cell cycle and therefore a distribution
of cell stages in a culture, without structuring the cell composition. In the
most realistic, but most complex situation, models can be both structured
and segregated.
The dierences among these models can be better described by their dif-
ferent balance regions, as shown in Figure 1.1. In the non structured-non
segregated model, the balance region is the total biomass volume; the struc-
tured model, instead, considers as balance regions dierent compartments
inside the cell; nally, in the segregated model, the balance regions are dif-
ferent biomass parts.

Consistently with the widely accepted terminology introduced in [2], we

term those models that describe the response of microorganisms to a single
set of conditions over time as primary models. Models that describe the
eect of environmental conditions, e.g., physical and chemical features, on
the values of the parameters of a primary model are referred to as secondary
models.
CHAPTER 1. STATE OF THE ART ON CELL GROWTH MODELS 11

(a) Non structured - non segregated

(b) Structured - non segregated model
model

(c) Non structured - segregated model (d) Structured - segregated model

Figure 1.1: Balance regions for dierent types of kinetic models (from [5])
CHAPTER 1. STATE OF THE ART ON CELL GROWTH MODELS 12

In this chapter we adopt this last classication to introduce the most

popular cell growth models in the literature. First, we are going to present
some example of primary model, then we will deeply describe the Monod
model, probably the most applied secondary model, and dierent derived
secondary models. Finally, we will briey introduce some aspects of the
structured approach.

1.2 Simple microbial kinetics: primary models

The models that we present in this section are known as primary models.
A primary model for cell growth aims to describe the kinetics of the process
with as few parameters as possible, while still being able to accurately de-
scribe the distinct stages of growth.

1.2.1 The Malthusian model

The simplest cell growth model is the Malthusian model [17], that is
often referred to as exponential law. Under ideal conditions for growth,
when a batch fermentation is carried out, it can be observed experimentally
that the quantity of biomass, and therefore also its concentration, increases
exponentially over time. This behavior can be explained by the fact that all
cells have the same probability to multiply. Thus the overall rate of biomass
formation is proportional to the biomass itself. This leads to an autocatalytic
reaction, which is described by a rst order rate expression as

rX = µX,

where rX is the rate of cell growth (kg cell/m3 ·h), X is the cell concentration
(kg cell/m3 ) and µ is a kinetic growth constant (1/h), called specic growth
rate. For a batch system, this is equivalent to
dX
= µX(t).
dt
The analytical solution of this simple, rst order dierential equation is
shown in Figure 1.2 and it is of the form

X(t) = X0 exp(µt),
CHAPTER 1. STATE OF THE ART ON CELL GROWTH MODELS 13

where X0 is the initial cell concentration at time t = 0. The cell concentration

that is obtained from the Malthusian model has the rather unrealistic feature
of getting larger without bound as t increases: the model ignores that cells
require resources to grow and that these resources are limited. Basically,
the Malthusian model does not fulll the rst kinetic principle proposed by
Penfold and Norris [20], namely that the cell growth rate µ is best described
by a saturation type of curve.

1.8
Cell concentration [106 cell/ml]

1.6

1.4

1.2

0.8

0.6

0.4

0.2

0
0 20 40 60 80 100 120
Time [h]

Figure 1.2: Solution of the Malthusian model with X0 = 0.3 · 105 [cell/ml]
and µ = 0.03 [1/h]

1.2.2 The logistic model

A second model that we can introduce is the logistic model [28], that was
rst published by Pierre-Francois Verhulst in 1838 after he read Malthus'
work. It takes the following form:
µ ¶
dX X
= µX 1 − ,
dt K

where now, besides µ, there is a second parameter K (kg cell/m3 ), often

called carrying capacity, which is the maximum concentration of cells that
the environment can support. The logistic equation has the major disadvan-
tage that the carrying capacity K cannot be measured other than by growing
the organism until it stops growing. Moreover, there is no theoretical under-
pinning for it.
CHAPTER 1. STATE OF THE ART ON CELL GROWTH MODELS 14

The logistic equation can be integrated exactly and has solution (Figure
1.3):
K
X(t) = ,
1 + CK exp(−µt)
where C = 1/X0 − 1/K is determined by the initial condition X0 .

2
Cell concentration [106 cell/ml]

1.5

0.5

0
0 20 40 60 80 100 120
Time [h]

Figure 1.3: Solution of the logistic model with X0 = 0.3 · 105 [cell/ml],
µ = 0.08 [1/h] and K = 2 · 106 [cell/ml]

1.2.3 The Gompertz model

The Gompertz function [8] is a sigmoid function, as the logistic curve.
As the previous model, the curve shows a slowest growth at the start and at
the end of the time period. In contrast to the logistic function in which both
the asymptotes are approached by the curve symmetrically, in the Gompertz
model the future value asymptote of the function is approached much more
gradually by the curve than the lower valued asymptote.
The Gompertz curve has the following expression (Figure 1.4):

X(t) = a exp[−b exp(−ct)].

It is possible to verify immediately that a is the upper asymptotic value,

b is the x displacement and c sets the growth rate. Moreover, the initial cell
concentration is X0 = a exp(−b).
CHAPTER 1. STATE OF THE ART ON CELL GROWTH MODELS 15

Cell concentration [106 cell/ml] 1.5

0.5

0 20 40 60 80 100 120
Time [h]

Figure 1.4: Solution of the Gompertz model with a = 2·106 [cell/ml], b = 10,
c = 0.05 [1/h]

1.2.4 The modied logistic and the modied Gompertz

models
Sigmoidal functions have been the most popular ones used to t microbial
growth data since these functions consist of three phases, similar to the
microbial growth curve. The most commonly used in the literature are the
modied logistic model [18]:

C
log X(t) = A +
1 + exp(−B(t − M ))

and the modied Gompertz model:

¡ ¢
log X(t) = A + C exp − exp[−B(t − M )] ,

where X(t) is the number of cells at time t, and A, B , C and M are model
parameters. The original logistic and Gompertz functions are considered to
be mechanistic; however, the modied functions are empirical: it is dicult,
in fact, to give an interpretation of the model parameters, that have to be
statistically estimated from experimental results.

The Gompertz equation and its modied version have been used exten-
sively by researchers to t a wide variety of growth curves from dierent
microorganisms with good results [33]. There are, however, some limitations
CHAPTER 1. STATE OF THE ART ON CELL GROWTH MODELS 16

associated with the use of these functions: the Gompertz rate µmax is always
the maximum rate, regardless of the actual culture medium composition, and
it occurs at an arbitrary point of inection; thus the generation time is not
estimated properly. In addition, since the slope of the function cannot be
zero, the lower asymptote must be lower than the inoculum level, giving a
negative lag phase duration for some data sets.

1.3 Secondary models

Secondary models describe cell growth dynamics as a function of envi-
ronmental conditions. The development of secondary models is based on the
knowledge of the eects of the environmental conditions on the cell behav-
ior. In particular we present those models that introduce the relationship
between cell growth, substrate concentration and temperature.

1.3.1 The Monod model

During the last half century, the concepts in microbial growth kinetics
have been dominated by the relatively simple semi-empirical model proposed
by Monod [19]. The Monod model diers from the primary models in the way
that it introduces the concept of a growth controlling substrate. This sub-
strate is called limiting substrate to indicate that the microbial growth rate
is dictated by the actual concentration of a particular metabolite. Whereas
there is a causal relationship between the exhaustion of the limiting substrate
and the end of growth, the Monod model may be considered deterministic.
In the 1930s and 1940s, Jacques Monod performed experiments on bacte-
ria feeding on a single limiting nutrient in order to see if the logistic equation
accurately described bacterial growth. He found it did not and therefore he
developed a new model to describe his results. If S(t) denotes the con-
centration (kg/m3 ) of the nutrient in the media at time t, his experiments
suggested that the specic growth rate
S
µ = µmax (1.1)
KS + S
seemed to best t the data. Consequently, the dynamics of the cell concen-
CHAPTER 1. STATE OF THE ART ON CELL GROWTH MODELS 17

tration X(t) is described by

dX S
= µmax X(t).
dt KS + S
The expression (1.1) is often called the Monod function. As a function of
S it is monotonically increasing with limit µmax as S → ∞. The parame-
ter µmax is therefore called maximum specic growth rate and KS is called
half-saturation constant, or anity constant, because when S(t) = KS the
µmax
specic growth rate becomes 2 , that is half the maximum specic growth
rate. The key feature of the Monod function is that the specic growth rate
increases with nutrient concentration S as expected, but it levels out at low
nutrient concentrations. For high substrate concentration, the relation (1.1)
approaches zero order and the rate of reaction is thus independent of sub-
strate concentration and is constant at the maximum value. In this case the
growth is said to be in conditions of non-limiting nutrients.

The link between growth and substrate utilization was made by Monod,
who linearly related the specic rate of biomass growth and the specic rate
of substrate consumption through the yield coecient YXS , a measure for the
conversion eciency of a growth substrate into cell material. The relation
between cell growth and substrate consumption is given by
dS 1 dX
=− ,
dt YXS dt
where
dX
YXS =.
dS
The complete Monod model is then composed by two coupled dierential
equations with two model parameters:

 dX S

 dt = µmax K + S X(t)

S
(1.2)

 dS = − 1 µmax
 S
X(t),
dt YXS KS + S

that can be solved if the initial values X0 and S0 are given. In Figure
1.5 the solution for the cell and the substrate concentrations is plotted: we
can observe that cell growth gradually stops as the substrate is consumed.
CHAPTER 1. STATE OF THE ART ON CELL GROWTH MODELS 18

Although very simple, the Monod model often describes experimental data
for growth rates reasonably well.

12
0.2
10
Cell concentration [106 cell/ml]

Substrate concentration [g/l]

0.15
8

6 0.1

4
0.05

0
0
0 20 40 60 80 100 120 0 20 40 60 80 100 120
Time [h] Time [h]

Figure 1.5: Solution of the Monod model with X0 = 0.3 · 105 [cell/ml],
S0 = 0.2 [g/l], µmax = 0.1 [1/h], KS = 0.1 [g/l], YXS = 0.1

Biological meaning of µmax and KS

A comment on the biological meaning of the parameters KS and µmax ,
which are used to characterize microbial growth for given growth conditions,
is necessary.
The maximum specic growth rate µmax is a characteristic of all orga-
nisms and it is related to their ability to reproduce themselves in nutritional
conditions of plenty. It is simply dened as the increase of biomass per
unit time under optimal feeding conditions, in which there are not limiting
nutrients.
While the interpretation of µmax as the maximum specic growth rate is
straightforward, the biological meaning of KS is less obvious. The empirical
constant KS is the substrate concentration at which organisms are substrate
limited to a growth rate of half the prevailing maximum value. This constant
is closely related to the mechanism of transport of the substrate over the
cell membrane, so it depends on cell membrane properties and intracellular
conditions, on the type of transporter proteins in the cell and on the substrate
properties. For this reason, the value of KS can be interpreted as a reection
of the anity of the cell towards the substrate S .
The values of these model parameters depend primarily on properties of
the organism itself, which are determined by its genome. Moreover, because
of the adaptive ability of organisms, both µmax and KS can vary with the
CHAPTER 1. STATE OF THE ART ON CELL GROWTH MODELS 19

environmental conditions, for example with the nature of the growth medium
and the duration of exposure to it, with temperature and pressure. For these
reasons the values of µmax and KS have to be estimated for each specic
couple cell-substrate and under constant conditions of temperature, pressure
and medium composition.

1.3.2 Modied Monod models

Starting from the Monod model, dierent growth rate expressions have
been proposed in the literature in order to model more complex situations.
In this section we analyze the most common modied Monod models.

Multiple-substrate Monod kinetics

If we want to describe the inuence of many substrates, for example
two substrates S1 and S2 , we can consider the following form of the Monod
equation: µ ¶µ ¶
S1 S2
µ = µmax .
KS1 + S1 KS2 + S2
In this model either substrate may be limiting under conditions when
the other substrate is in excess. In Figure 1.6 we can observe that cell
growth stops when the second substrate is totally exhausted, even if the rst
substrate has been only partially consumed. An example of such kinetics is
the simultaneous requirement of glucose and oxygen by aerobically growing
organism.

Figure 1.6: Solution of the multiple-substrate Monod model with X0 =

0.3 · 105 [cell/ml]
CHAPTER 1. STATE OF THE ART ON CELL GROWTH MODELS 20

Double-Monod kinetics
The Monod equation can also be written for two substrates that can be
used by the cell with parallel reactions:
µ ¶µ ¶
1 S1 S2
µ = µmax k1 + k2 .
k1 + k2 KS1 + S1 KS2 + S2

We can observe that each substrate allows a dierent maximal growth rate
(Figure 1.7). An example of this kinetics is the parallel use of alternative
carbon substrates, such as glucose and glutamine.

Figure 1.7: Solution of the double-Monod model with X0 = 0.3·105 [cell/ml]

Diauxic Monod growth

In a medium containing two carbon sources, cells can display a growth
curve that is called diauxic. Diauxic growth can be observed in many or-
ganisms. Monod rst observed this phenomenon when he grew Escherichia
Coli in a medium containing glucose and lactose. Under these conditions,
glucose is rst utilized as energy source and after the exhaustion of glucose,
lactose is utilized. This led him to conclude that some bacteria preferentially
utilize certain carbon substrates.
Diauxic Monod growth can be modeled, for two substrates S1 and S2 ,
by the relation
S1 S2
µ = µ(1)
max + µ(2)
max S12
.
KS1 + S1 K2 + S2 + KI

In this way the consumption of substrate S2 will be inhibited until S1 is

exhausted, for suitably low values of inhibition constant KI (Figure 1.8).
CHAPTER 1. STATE OF THE ART ON CELL GROWTH MODELS 21

Figure 1.8: Solution of the diauxic Monod model with X0 = 0.3·105 [cell/ml]

Maintenance substrate uptake

The classical Monod equation does not consider the fact that cells may
need substrate, or may synthesize product, even when they do not grow.
In fact cells require energy just to maintain cellular viability and replace
degraded proteins, even when they are not dividing. For this reason, the
original Monod equation was modied by introducing the terms of mainte-
nance, expressed as maintenance rate m, originally proposed in [10], and by
assuming that the per unit mass maintenance energy is constant.
The maintenance factor m (kg substrate/kg cells·h) is dened as the mass
of substrate that one unit of biomass requires for non-growth functions in one
unit time. The total substrate utilization for cell maintenance is, of course,
taken to be proportional to the total quantity of cells, and therefore for a
batch reactor it is proportional to the cell concentration X . If we introduce
the maintenance substrate uptake in the classic Monod model (1.2) we obtain
 dX S

 dt = µmax K + S X(t) − m YXS X(t)

S
(1.3)

 dS = − 1 µmax S X(t).

dt YXS KS + S
The solution of the model (1.3) is plotted in Figure 1.9. We can see that,
when the substrate is exhausted, cell growth stops and, since cells don't have
any energy source to maintain, they start dying.

A literature survey has shown that the rate of maintenance energy m is

similar for many microorganisms [25] and does not depend signicantly on
the nature of the carbon source used in catabolism to generate the mainte-
nance energy. This is understandable because maintenance relates to biomass
CHAPTER 1. STATE OF THE ART ON CELL GROWTH MODELS 22

9 0.25

8
0.2
Cell concentration [106 cell/ml]

Substrate concentration [g/l]

6
0.15
5

4 0.1

0.05
2

1
0
0

0 20 40 60 80 100 120 0 20 40 60 80 100 120

Time [h] Time [h]

Figure 1.9: Solution of the Monod model with the maintenance substrate
uptake with X0 = 0.3 · 105 [cell/ml] and m = 0.01 [kg substrate/kg cells·h]

which has already been synthesized and for which viability must be main-
tained; it does not relate to new biomass that is being formed. It is obvious
that the energy needed for maintenance is generated in a catabolic reaction,
where catabolic substrates and products are involved. Hence maintenance
is not only characterized by m, but the generation of this energy leads to
associated chemical maintenance rates of all the catabolic substrates and
products which are consumed and produced in the catabolic reaction.
Moreover it has been observed from experimental results that the main-
tenance energy is negligible during the growth phase, that is when cells are
dividing, and it becomes relevant only when growth is inhibited [30].

1.3.3 Secondary models based on the Arrhenius law

In order to introduce the temperature dependance on the specic growth
rate, the empirical Arrhenius-van't Ho relationship is often considered
µ ¶
∆Ea
µ = A exp ,
RT

where the parameters may be interpreted as follows:

• A is the constant related to the number of collisions between reactants

per unit time;

• Ea is the activation energy;

• R = 8.314 J/K · mol is the gas constant;

CHAPTER 1. STATE OF THE ART ON CELL GROWTH MODELS 23

• T is the temperature in Kelvin degrees.

This law is well established in chemistry to describe the eects of tempe-

rature on the rate of chemical reactions. If the logarithm of rate is plotted
against T , the resulting plot is a straight line over temperature ranges rele-
vant to microbial growth, as shown in Figure 1.10.
It has been argued that because all life processes are the result of chemical
reactions, the growth rate of organisms should also be described by Arrhenius
kinetics. Within a narrow range of temperature this is true. In fact if we
plot also the estimated eect of temperature on microbial growth rate for
a cell, we can see that in a limited range of temperature, termed normal
physiological range (NPR), the microbial growth rate follows the Arrhenius
model's prediction. At temperatures above or below the NPR, microbial
growth rate deviates markedly from that predicted by the Arrhenius model
(Figure 1.10).

Figure 1.10: Eect of temperature on reaction rate predicted using the Ar-
rhenius law (solid line) and the estimated eect of temperature on the cell
growth rate (dashed line)
CHAPTER 1. STATE OF THE ART ON CELL GROWTH MODELS 24

A range of secondary models, based on adherence to the reaction kinetics

described by the Arrhenius model, but including terms to account for the
observed deviations, have been proposed. These models can be divided into
those based on mechanistic modications of the Arrhenius model and those
based on empirical modications.
An example of the rst class of models is represented by the model pro-
posed by Hinshelwood [11]. This model is based on the hypothesis that at
high temperatures denaturation of enzymes involved in cell metabolism takes
place and cell growth is inhibited. The model has the following form:
µ ¶ µ ¶
Ea Ea,high
µ = A exp − − B exp − ,
RT RT

where A and B are pre-exponential factors (1/h), Ea is the activation en-

ergy of cell growth (J/mol) and Ea,high is the activation energy of the high-
temperature denaturation of the rate-limiting enzyme.
In practice, few of these modied models have been routinely applied
in predictive microbiology, possibly because the models are highly nonlinear
and parameter estimates are dicult to determine. Furthermore the con-
cept that a single enzyme is rate-limiting under all environmental conditions
seems questionable. Finally, several studies demonstrated that empirical
models provide an equally good t as those mechanistic models, and are
usually easier to work with. In the literature there are dierent examples
of empirical model for the dependance of the cell growth rate on tempera-
ture. Many of them expresses the growth rate as a polynomial function of
temperature and without any physical or chemical interpretation [27].

1.3.4 The gamma concept

The idea of dimensionless growth factors, now known as gamma concept,
was introduced in predictive microbiology by Zwietering [34]. The gamma
concept relies on the observation that

• many factors that aect microbial growth rate act independently;

• the eect of each measurable factor on growth rate can be represented

by a discrete term that is multiplied by terms for the eect of all other
growth rate aecting factors;
CHAPTER 1. STATE OF THE ART ON CELL GROWTH MODELS 25

• the eect on growth rate of any factor can be expressed as a fraction

of the maximum growth rate.

Under completely optimal conditions each microorganism has a repro-

ducible maximum growth rate. As any environmental factor becomes sub-
optimal the growth rate declines. The relative inhibitory eect of a specic
environmental variable is described by a growth factor gamma, a dimension-
less measure that has a value between 0 and 1.
The relative inhibitory eect can be determined from the distance be-
tween the optimal level of the factor and the minimum (or maximum) level
that completely inhibits growth. In the gamma model approach, the refe-
rence growth rate is µmax , so that the reference levels of every environmental
factor are those that are the optimum for growth rate.
If this approach is adopted, the problem shifts from the modeling of the
cell growth rate to the modeling of the eects of each inhibitory factor on
cell kinetics.

1.4 Structured kinetic models

In many cases the characterization of biological activity by considering
only the total biomass is insucient for a true model representation, in parti-
cular under varying environmental conditions. It has been proved experimen-
tally that variations of temperature during cell growth cause the composition
and behavior of the cell to change, for example in terms of maximum specic
growth rate and substrate uptake, that unstructured models can not simu-
late. It has also been veried that a variation in the biomass activity per
unit biomass concentration may be caused by morphological changes, vari-
ation of enzyme content of the cells or accumulation of intracellular storage
materials.
Such variations in biomass activity and composition require a complex
description of the cellular metabolism and a structured approach to the mo-
deling of cell kinetics. Models that are based on a compartmental description
of the cell mass are called structured models. These models look in some
details at individual cell processes and reactions, in order to obtain a more
realistic description of the behavior of the cell. In particular they focus on
the cell physiology, that includes the cell mechanism and the cell interaction
CHAPTER 1. STATE OF THE ART ON CELL GROWTH MODELS 26

with its environment, and are built up starting from appropriate intracellular
mass balances.
However, in general it is very dicult experimentally to obtain su-
cient mechanistic knowledge about the cell physiology and metabolism for
the development of a realistic structured model. Parameter estimation may
be almost impossible because it is dicult to take direct measurement of
quantities and properties related to one single cell, whose dimension is often
lower than 1µm. Moreover, the mechanism of cell growth is complex and not
yet completely understood. For this reason, unstructured models are more
desirable than structured ones, that are seldom used for design or control.

1.5 Conclusions
The idea that the relationship between µ and S is best described by a
saturation type of curve has been widely accepted [20]: at high substrate
concentrations the organisms should grow at a maximum rate µmax inde-
pendent of the substrate concentration. Although Monod model fullls the
requirement, the fact that even Monod own data did not indisputably sup-
port his proposed mathematical model gave rise to many more studies. A
variety of other mathematical expressions have been put forward to describe
this hyperbolic curve. However, the development of structured mechanistic
models for quantifying microbial growth kinetics is still limited because the
mechanism of cell growth is complex and not yet completely understood.
Therefore, most of the proposed growth models are unstructured and empi-
rical. In principle, three methods are used to design such rened equations
for the growth kinetics of cells:

• incorporating additional constants into the original Monod model that

provided corrections of, for example, substrate or product inhibition;

• proposing dierent kinetics concepts, resulting in both empirical and

mechanistic models;

• describing the inuence of physicochemical factors on the Monod growth

parameters.
CHAPTER 1. STATE OF THE ART ON CELL GROWTH MODELS 27

These approaches give rise to new empirical models, such that presented
in the previous sections, that however partially move away from the ideal of
creating a completely mechanistic model, in which each parameter has its
physical or chemical meaning. In the literature there are only a few examples
of cell growth models applied to batch culture system that includes multiple
nutrients, metabolites and toxic by-products with adequate accuracy [14, 30]
and, up to our knowledge, there are no examples of models that take into
account the spatial dependance of the cell growth process.
Chapter 2

Characterization of CHO cells

physiology

2.1 Introduction
Chinese hamster was rst used as laboratory specimen in 1919 in place
of mice for typing Pneumococci. Subsequent eorts in the mid-20th century
led to the development of spontaneous hereditary diseases due to inbreeding,
spurring research interest in hamster genetics. It was noted during that time
that the low chromosome number of Chinese hamsters (2n = 22) made them
particularly useful models in tissue culture studies. In 1957, for the rst time,
an ovary cell from a female Chinese hamster was isolated and established in
culture plates. It soon became obvious that these cells were quite resilient
and lent themselves to in vitro cultivation with relatively fast generation
times.
Until the later part of the 20th century, isolation and characterization
of mammalian cell mutants for cytogenetic studies was challenging exercise,
fraught with failures because, unlike microbes, mammalian cells are generally
diploid, so have two homologous copies of each chromosome. However, CHO
cells have, thereafter, been used in numerous biomedical studies ranging
from analysis of intermediary metabolism and cell cycle to toxicology stu-
dies, so much so, that they have been termed as the mammalian equivalent
of the model bacterium, Escherichia Coli. Among the historically important
medical and cell biology studies conducted in CHO, it was the early work

28
CHAPTER 2. CHARACTERIZATION OF CHO CELLS PHYSIOLOGY 29

involving mutagenesis of these cells that facilitated their migration from la-
boratory benches to industrial reactors. These mutants exhibited particular
nutritional requirements for maintaining growth and viability over long cul-
ture periods.
While the primary reason behind the isolation of these mutants was fun-
damental research, it was fortuitous that the nutritional requirements of
these cells could be put to use for selection of cells expressing exogenous
proteins. This ability of transfect, select, amplify and express biologically
active proteins soon became an immense boon for pharmaceutical companies
involved in the business of large-scale protein therapeutic synthesis. The im-
mense adaptive ability of CHO cells and their ease of maintenance have been
exploited in many elds of basic biomedical research. The most important
factors that enable the adoption of CHO cells as the industry workhorse are
their adaptability and their ease of genetic manipulation. CHO cells are
quite adaptable and can grow to very high density in suspension cultures
that are scaled up to 10.000 liters bioreactors.
One of the major challenges in using CHO and other mammalian cell
lines as recombinant protein production hosts is that the volumetric yields
of protein produced from processes using these cells are relatively low. The
productivity of mammalian cell culture processes is typically about 10-100
fold lower than what can be achieved using microbial host systems [12]. This
requires the construction and the maintenance of very large and costly pro-
duction facilities. This is one of the reasons why it is of great interest the
expansion of culture volumes and the optimization of culture strategies for
higher yields.

Despite the signicant advances and improvement during the last past
decade in cell line development, these are still a lot of diculties related to
these technologies, primarily stemmed from an inadequate understanding of
the biology and physiology of mammalian cells. First, their metabolic cha-
racteristics, especially the energy metabolism and regulation modes, have
not been well recognized. It has been noted, for example, that the cells have
consumption rates of the principal sources of carbon and energy far above
the rates strictly required to give support to the cellular metabolism. As
a result, situations of exhaustion of metabolites and accumulation of toxic
CHAPTER 2. CHARACTERIZATION OF CHO CELLS PHYSIOLOGY 30

by-products are frequently reached, with the consequent initiation of the pro-
grammed cell death apoptosis. Finally, there is only an empirical knowledge
about the environmental factors that determine variations in the cell growth
process.

The purpose of this chapter is to analyze the most important charac-

teristic of CHO cells growth, focusing in particular on cell physiology and
metabolism. Due to the lack of knowledge about the mechanism related
to cell behavior, we will take some important information from experimen-
tal results. This analysis is useful for the identication of which aspects and
factors related to cell growth should be taken into account in our new model.

2.2 Cell growth curve

The curve which represents the cell concentration in the culture as a
function of time is called growth curve. A typical cell growth curve is shown
in Figure 2.1.

Figure 2.1: Cell growth curve

The cell growth curve can be divided into a number of distinct phases,
CHAPTER 2. CHARACTERIZATION OF CHO CELLS PHYSIOLOGY 31

referred to as the lag, exponential, stationary and death phase respectively.

Each of these phases is characterized by a dierent growth rate and can be
inuenced by dierent factors:

• lag phase: the beginning of the growth is generally delayed because of

the changing of environment. Cells have to rearrange their metabolism
and to adapt it to the new environment, for example to a dierent
temperature or nutrient concentration. Lag phase is relatively poorly
understood. However, it is known that factors aecting the occur-
rence and extent of the commonly observed initial lag phase are the
initial cell concentration, the past environment, the new environment,
the magnitude of the environmental change, the rate of the environ-
mental change, the growth status of the inoculated cell culture, and
the variability between individual cell lag phases. In conclusion, these
environmental changes may involve nutritional and chemical, as well
as physical, changes. The lag phase can last from minutes to several
hours. The length of the lag phase can be controlled to some extent
because it is dependent on the type of medium as well as on the initial
inoculum size. For instance, if the inoculum is taken from a statio-
nary phase culture and is placed into fresh medium, there will be a lag
phase as the stationary phase cells adjust to the new conditions and
shift physiologically from stationary phase cells to exponential phase
cells. Similarly, if the inoculum is transferred from a medium with
glutamine as substrate to a new medium with only glucose as the sole
carbon source, a lag phase will be observed while the cells reorganize
and shift physiologically to synthesize the appropriate enzyme for glu-
cose metabolism;

• exponential phase: in this phase the cells have adjusted to their new en-
vironment and multiply rapidly. Growth rate is virtually independent
of nutrient concentration, as nutrient are in excess. The exponential
phase is characterized by the most rapid growth possible under the
actual conditions of the system. During exponential growth the rate
of increase of cells in the culture is proportional to the number of cells
present at any particular time. The exponential phase ends due to
either depletion of one or more essential nutrients, the accumulation
CHAPTER 2. CHARACTERIZATION OF CHO CELLS PHYSIOLOGY 32

of toxic by-products of growth or the reaching of a saturation limit of

cell concentration;

• stationary phase: it can be dened as a state in which, even if cells

still grow and divide, there is no net growth. In this phase growth is
balanced by an equal number of cells dying. There are several reasons
why a culture may reach the stationary phase. One common reason
is that the carbon and energy source becomes completely used up.
When carbon source is exhausted, it does not necessarily mean that
all growth stops. This is because dying cells can lyse and provide
a source of nutrients. A second reason why stationary phase may be
observed is that waste products build up to a point where they begin to
inhibit cell growth or are toxic to cells. This generally occurs in cultures
with high cell density. As a result of nutrient stress, stationary phase
cells are generally smaller and rounder than cells in the exponential
phase. Even if the cell concentration is constant, the environment keeps
changing because of the accumulation of toxic by-product produced by
cell functions dierent from growth;

• decline phase: the growth rate becomes smaller than the death rate
due to a lack of nutrients and toxic products increasing. It results in a
net loss of viable cells. The death phase is often exponential, although
the rate of cell death is usually slower than the rate of growth during
the exponential phase.

The growth phases that are of most interest to microbiologist, and in

particular to biologist involved in the development of cell cultures for pro-
duction of recombinant proteins, are the lag phase, the exponential phase
and the time to onset of the stationary phase. In particular there is not
a great interest in modeling the death phase because, if it is known that
after a certain time the stationary phase, hence the maximal cell concentra-
tion, is reached, then the growth process will be interrupt at that time. For
these reasons in the development of our model we will focus mainly on the
modeling of the lag, the exponential and the stationary phase.
CHAPTER 2. CHARACTERIZATION OF CHO CELLS PHYSIOLOGY 33

2.3 Cellular metabolism

Cellular metabolism can be dened as the sum of all the processes in-
volved in energy conversion in the cell. They regulate cellular conditions such
that a state of a metabolic homeostasis, that is a stable supply of energy and
metabolites, is maintained.
Metabolic processes are organized into complex sequences of controlled
chemical reactions referred to as metabolic pathways, that form the basis of
all cellular activities. Many dierent pathways are responsible for nutrient
processing, energy acquisition and energy conversion in the cell. Metabolic
pathways can be broadly categorized as anabolic and catabolic. Anabolic
pathways are energy-requiring (endoergonic) pathways that result in synthe-
sis of complex biomolecules such as proteins, nucleic acid and membranes
from smaller ones. In contrast, catabolic pathways are energy-releasing
(exergonic) pathways that break down molecules into smaller components.
The ensemble of catabolic processes in the cell, in which molecules are
oxidized to carbon dioxide and water, is termed cellular respiration. Cellular
respiration, also known as oxidative metabolism, is one of the key ways a
cell gains useful energy. It includes anaerobic pathways, that do not require
molecular oxygen, and aerobic pathways, that directly or indirectly require
molecular oxygen.
Energy released from cellular respiration has to be converted in some form
that is useful to drive energy-requiring processes in the cell. In most cases,
the energy that is used by most processes is stored in adenosine triphosphate
(ATP) molecule, that consists of three phosphate attached to the ribose of
adenosine. The energy is obtained from ATP through the hydrolysis process,
in which the ATP molecule is split by reacting with a molecule of water. Hy-
drolysis of ATP is an exergonic reaction with a standard free energy change
(δG0 ) of -7.3 kcal/mol. This is the energy which is available to do work.
Cellular reactions do not occur under standard conditions; thus, the actual
amount of energy released is likely greater than the calculated δG0 . In any
case, it is clear that hydrolysis of ATP releases a substantial packet that can
be used to drive endoergonic reactions. To maintain usable energy, the cell
must continually resynthesize ATP. Synthesis of ATP is an energy requiring
process, that is fed by the catabolic reactions that comprise cellular respira-
CHAPTER 2. CHARACTERIZATION OF CHO CELLS PHYSIOLOGY 34

tion. Hence, ATP acts as an intermediary that transfers energy from cellular
respiration to the energy-requiring processes in the cell.

In the model that we are going to present in the next chapters, we want
to take into account both the anaerobic and the aerobic respiration. Conse-
quently in this section we analyze these chemical processes in order to derive
information that may be useful for the development of our model. In parti-
cular we consider the main catabolic reactions that involve as energy source
glucose and glutamine, that are known to be the most important carbon and
energy source for animal cells [23].

2.3.1 Glycolysis
The rst step in cellular respiration is glycolysis, that includes the initial
reactions required for carbohydrate metabolism [29]. Glycolysis is a series
of cytosolic reactions that converts one molecule of glucose to two 3-carbon
pyruvate molecules (Figure 2.2):

Glucose + 2ADP + 2P i + 2N AD+ →

2P yruvate + 2AT P + 2N ADH + 2H + + 2H2 O.

Glycolysis is a 10-step process, in which usable energy is stored in two

ATP molecules and in two pairs of high energy electrons passed to nicoti-
namide adenine dinucleotide (N AD+ ). Two ATP represent only a small per-
centage of the total energy that can be obtained from the glucose molecule
(686 kcal/mol), and the nal product, pyruvate, still contains the bulk of
the total energy initially present in the glucose molecule. Nonetheless, gly-
colysis is a pivotal pathway in cellular respiration because it is the initial
step for oxidation of glucose, and it generates energy under anaerobic con-
ditions. Anaerobic metabolism, in fact, can be critically important when
oxygen availability is limited.
If glycolysis were to continue indenitely, all of the N AD+ would be used
up, and glycolysis would stop. To allow glycolysis to continue, the cell must
be able to oxidize N ADH back to N AD+ .
CHAPTER 2. CHARACTERIZATION OF CHO CELLS PHYSIOLOGY 35

Figure 2.2: Glycolysis pathway

CHAPTER 2. CHARACTERIZATION OF CHO CELLS PHYSIOLOGY 36

Molecules produced during glycolysis can enter dierent pathways. In

fact, pyruvate, the end product of glycolysis, can enter the anaerobic process
of fermentation or it can be converted to acetyl coenzyme A (acetyl-CoA) and
enter aerobic respiration. The employment of pyruvate will be determined
by the availability of oxygen.

2.3.2 Anaerobic respiration

In the absence of oxygen, N AD+ is regenerated when N ADH passes
electrons to pyruvate, forming lactate, but without producing ATP [29].
This process is called lactate fermentation (Figure 2.3):

P yruvate + N ADH + H + → Lactate + N AD+ .

This anaerobic fermentation allows cell to use glucose as its only energy

Figure 2.3: Lactic fermentation

resource, because it reconverts N ADH back to N AD+ that is needed by

glycolysis. If we assume that all the products of glycolysis, namely pyru-
vate, N ADH and H + , are instantaneously converted to lactate and N AD+
through lactic fermentation, the overall reaction of anaerobic respiration can
be simplied as follows (Figure 2.4):

Glucose + 2ADP + 2P i → 2Lactate + 2H2 O + 2AT P.

Plant and yeast cells also use fermentation under anaerobic conditions but
in these cells the electrons are not passed to pyruvate. Pyruvate is rst
decarboxylated to acetaldehyde, which is then reduced by electrons from
CHAPTER 2. CHARACTERIZATION OF CHO CELLS PHYSIOLOGY 37

Figure 2.4: Anaerobic respiration

N ADH to form ethanol. Because the process yields ethanol, it is called

alcoholic fermentation. This kind of fermentation, carried out by yeast and
plant cells, is used, for example, to make bread, wine and beer.

2.3.3 Aerobic respiration

Anaerobic fermentation, that converts glucose to lactate, releases only a
small portion of the total energy available in the glucose molecule. However,
if oxygen is available, glucose can be completely oxidized and a much larger
amount of energy can be extracted. In presence of oxygen, pyruvate produced
during glycolysis is transported into the matrix of the mitochondrion and it is
involved in four dierent processes: pyruvate decarboxylation, tricarboxylic
acid (TCA) cycle, electron transport chain and oxidative phosphorylation
[29].
The rst step of this process is pyruvate oxidative decarboxylation (Fi-
gure 2.5):

P yruvate + CoA + N AD+ → acetyl − CoA + CO2 + N ADH.

In this reaction, pyruvate is converted to acetyl-CoA and carbon dioxide,

thanks to the Coenzyme A (CoA). Oxidation of pyruvate to acetyl-CoA is
a highly exergonic process. Then acetyl-CoA is committed to entering the
CHAPTER 2. CHARACTERIZATION OF CHO CELLS PHYSIOLOGY 38

Figure 2.5: Pyruvate oxidative decarboxylation

TCA cycle, where it is fully oxidized to carbon dioxide:

Acetyl − CoA + 2H2 O + 3N AD+ + F AD + ADP + P i →

2CO2 + 3N ADH + 3H + + F ADH2 + CoA + AT P.

The TCA cycle is an eight-step cycle (Figure 2.6). The only usable en-
ergy produced in the TCA cycle is that stored in one molecule of ATP.
This molecule of ATP is in addition to the two ATP produced by glycolysis.
Thus, the complete oxidation of one molecule of glucose generates 4AT P ,
10N ADH and 2F ADH2 . The energy in these stored electrons can be con-
verted in ATP and N AD+ thanks to the electron transport chain and the
oxidative phosphorylation, that require molecular oxygen.
The theoretical maximum ATP yield for complete oxidation of a molecule
of glucose is 38 ATP. If one assumes that hydrolysis of ATP in the cell yields
approximately 10 kcal/mol, then oxidation of glucose provides 380 kcal/mol
of usable energy. Whereas the δG0 for oxidation of glucose is 686 kcal/mol,
the eciency of energy conservation in aerobic respiration is about 55.4 %.
In the hypothesis that we made in the previous section, the overall reac-
tion of oxidation of glucose can be simplied as follow:

Glucose + 6O2 → 6CO2 + 6H2 O + 38AT P.

CHAPTER 2. CHARACTERIZATION OF CHO CELLS PHYSIOLOGY 39

Figure 2.6: TCA cycle

CHAPTER 2. CHARACTERIZATION OF CHO CELLS PHYSIOLOGY 40

2.3.4 Glutamine oxidation

Glutamine is one of the twenty amino acids and has a variety of bio-
chemical functions, in particular it is an important cellular energy source,
next to glucose. Therefore glutamine is commonly added to the media in
cell culture.
Glutamine is involved in three dierent metabolic reactions [9]. The rst
one is a single-step reaction: in presence of oxygen, glutamine oxidizes and
the products of this reaction are carbon dioxide, ammonia and lactate:

Glutamine + 2O2 → 2CO2 + 2N H3 + Lactate.

Glutamine is also involved in two reactions that have respectively alanine and
aspartate as intermediate products. The rst two-step reaction has alanine
as intermediate product:

Glutamine + 3O2 → 2CO2 + N H3 + Alanine

Alanine + H2 O → CO2 + N H3 + Lactate,

while the second two-step reaction involves aspartate as intermediate pro-

duct:
Glutamine + 3O2 → CO2 + N H3 + Aspartate

Aspartate + H2 O → CO2 + N H3 + Lactate.

In order to simplify the model, we suppose that all the aspartate and the ala-
nine produced in the rst step of the reactions are consumed instantaneously
in the second step, so that the concentration of both alanine and aspartate
in the culture medium is constant at every time. This hypothesis is sup-
ported by the fact that alanine and aspartate need only water, which is not
a limiting-substrate, to react and produce carbon dioxide, ammonia and lac-
tate. However, this hypothesis is quite strong and depends on the chemical
properties of these substances, therefore it should be tested with appropriate
experiments. Under this hypothesis, the two-step reactions become

Glutamine + 3O2 → 3CO2 + 2N H3 + Lactate

and
Glutamine + 3O2 → 2CO2 + 2N H3 + Lactate.
CHAPTER 2. CHARACTERIZATION OF CHO CELLS PHYSIOLOGY 41

In conclusion, if we want to consider in our model the metabolic pathways

introduced in this section, we need to take into account the concentrations
of three substrates, that are glucose, glutamine and oxygen, and the con-
centrations of three toxic by-products, that are carbon dioxide, lactate and
ammonia. In order to build up the model that describes the dynamics of
these metabolites, we will consider the stoichiometric coecients of the re-
actions that we have just introduced. However, it is necessary to underline
that the reactions proposed in this chapter are ideal and simplied reactions
and not the actual reactions that take place in the cells. For example, there
are evidences that glutamine is not totally converted into lactate by the cell,
but it is also partially deaminated and used to produce other amino acids.
In fact, unless the amino acids like glutamine are taken up by the cells in
"right" stoichiometric ratios, conversions among them will be necessary to
meet the cellular needs.

2.4 Environmental factors

Before moving to the description of the cell growth model, it is neces-
sary to detect which environmental factors inuence the cell kinetics. In
this section, in particular, we want to analyze the eect on cell growth of
temperature, oxygen concentration and pH of the culture medium.

2.4.1 Temperature
Animal cells are most commonly cultured at 37◦ C. However, since the
culture temperature would aect such cellular events as growth, viability,
protein synthesis and metabolism, it is an important factor which has been
widely studied to realize an ecient process for protein production by ani-
mal cell culture [7]. It has been shown in the previous section that cellular
metabolism consists of a network of reactions. Whereas it is known that
these individual biochemical reactions are temperature dependent, the fun-
damental question of whether cell growth kinetics is temperature dependent
must be asked. To determine this, many comparisons of growth kinetics at
dierent temperatures were carried out in the literature.
With appropriate experiments it has been veried that the specic growth
CHAPTER 2. CHARACTERIZATION OF CHO CELLS PHYSIOLOGY 42

rate is temperature-dependent (Figure 2.7) and, in particular, the maximum

specic growth rate was found at 37◦ C [13, 32]. Temperatures above 37◦ C
decrease cell growth capacity by lowering the specic growth rate and the
maximum cell concentration, while at temperatures above 39◦ C cell growth
stops and cell viability decreases dramatically. Moreover temperatures below
37◦ C decrease growth capacity non linearly by lowering the specic growth
rate, while at temperatures below 35◦ C cell division is inhibited but cells
maintain a high viability over time, medium consumption is suppressed and
the production of toxic by-products is lowered [7].

Figure 2.7: Cell concentration at dierent temperatures (from [1])

2.4.2 Oxygen concentration

In the previous section, it has been shown that cells can obtain energy
both from aerobic and from anaerobic respiration. However, we have also
estimated that glucose oxidation produces an amount of ATP that is ap-
proximately 20 times larger than the amount produced by anaerobic glucose
consumption. This is the reason of the decrease in viability and the sup-
pression of growth in anoxic conditions [1], namely in absence of oxygen in
the atmosphere above the culture medium (Figure 2.8). Moreover, it has
been proved that the continuous exposure of CHO cells to an atmosphere of
98% of oxygen (hyperoxia) induces a disturbance of energy metabolism that
results in growth inhibition after an initial doubling [24].
CHAPTER 2. CHARACTERIZATION OF CHO CELLS PHYSIOLOGY 43

Finally, it has been proven that a variation of the oxygen concentration

between 5% and 35% does not aect signicantly cell growth and viability [1].
However, usually cell culture is carried on with 20% of oxygen concentration
in the gas phase.

Figure 2.8: Cell concentration at dierent oxygen concentrations: 20 % (red)

and anoxia (green) (from [1])

2.4.3 pH
The pH is commonly considered a good indicator for cell metabolism,
whereas it depends directly on the concentration of toxic by-products of cell
growth that dissociate in water. In CHO culture medium, the pH is mainly
determined by the concentrations of carbon dioxide and lactate, which cause
the reduction of the pH.
In particular, experiments have demonstrated a marked decrease in sur-
vival of CHO cells when they are incubated under oxygen-limited conditions
at low pH. Lethal eects are observed when the pH becomes lower than 6.4.
In CHO cells, glucose consumption and lactate production are suppressed as
the pH of the medium is lowered from 7.0 to 6.0, conrming that glycoly-
sis is inhibited at low pH. Thus, one possible explanation for the cell death
observed under conditions of hypoxia and acidity is that cells incubated in
hypoxia, a condition in which aerobic respiration is inhibited, and low pH,
CHAPTER 2. CHARACTERIZATION OF CHO CELLS PHYSIOLOGY 44

a condition in which glycolysis is inhibited, might die of energy deprivation

[22]. These conditions often occur in bioreactors when the oxygenation of
the culture medium is not ecient (Figure 2.9).
In order to limit the reduction of the pH in the culture medium, bicar-
bonate is often introduced in the bioreactor [26].

Figure 2.9: pH at dierent oxygen concentrations: 20 % (red) and anoxia

(green) [1]
Chapter 3

Batch bioreactor modeling

3.1 Introduction
In the previous chapter we analyzed some aspects of CHO cell growth
kinetics primarily related to the type of cell and its physiology, focusing in
particular on the cell metabolism and on the interactions between cells and
environment. However, it is necessary to highlight that the environmental
properties like oxygen concentration and pH are mainly determined by the
type of culture system. The consumption and production rates, for exam-
ple, depend upon the level of concentrations, and it has been proven that
concentration levels in the culture system depend on its type and operation
mode. Dierent operation modes can lead to dierent rates of cell growth,
to dierent rates of product formation and hence to substantially dierent
productivity. In this chapter we are going to analyze the CHO cell cul-
ture system, focusing in particular on the relationship between the mode of
operation and the properties of the culture environment.

3.2 CHO cell culture

Cell culture is the complex process by which cells are grown under con-
trolled conditions. Cells are maintained at an appropriate temperature and
gas mixture in a bioreactor, that is a vessel in which a chemical reaction is
carried out involving organisms and biochemical substances.
The mode of bioreactor operation depends on the type of cell line. CHO
cells, in particular, can be grown in suspension, without been attached to

45
CHAPTER 3. BATCH BIOREACTOR MODELING 46

a surface, in vessels called shaken bioreactors under batch conditions (Fi-

gure 3.1). This implies that the bioreactor is rst charged with medium,
inoculated with cells, and the cells are allowed to grow for a sucient time,
such that the culture achieves the required cell density or the optimum pro-
duct concentrations. The bioreactor contents are nally discharged and the
bioreactor is prepared for a fresh charge of medium.

Figure 3.1: Shaken bioreactor under batch conditions

During the period of cell growth, no additional material is either added

to or removed from the bioreactor, apart from minor adjustments needed for
control of pH or foam, the removal of samples and, of course, a continuous
supply of air needed for aerobic metabolism. Concentrations of biomass, nu-
trients and products thus change continuously with respect to time, as the
various constituents are either produced or consumed during the time course
of the reaction.

Although it has been over two decades since the employment of CHO
cell culture in recombinant therapeutic production, and a decade since cell
culture bioprocessing was proclaimed to be a "mature technology", the new
surge in the quantities of products required and the phenomenally high in-
vestment cost for a manufacturing plant have spurred a new drive to enhance
cell culture bioprocess technology and modeling.
Both physical and biological information are required in the design and
CHAPTER 3. BATCH BIOREACTOR MODELING 47

interpretation of biological reactor performance. Physical factors that aect

the general hydrodynamic environment of the bioreactor include such param-
eters as liquid ow pattern and circulation time, air distribution eciency
and gas holdup volume, oxygen mass transfer rates, intensity of mixing and
the eects of shear. These factors are aected by the bioreactor and agitator
geometries and by physical property eects, such as liquid viscosity. They
have a large eect on both liquid and gas phase hydrodynamics. It has been
proved that there are considerable interactions between the bioreactor hy-
drodynamic conditions and the cell biokinetics, morphology and physiology.
For example, in large scale bioreactors, some cells may suer local starvation
of essential nutrients owing to a combination of long liquid circulation time
and an inadequate rate of nutrient supply, caused by inadequate mixing or
inecient mass transfer. Agitation and shear eects can aect cell morphol-
ogy and hence liquid viscosity, which will also vary with cell density. This
means that the process of cell growth aects the bioreactor hydrodynamics in
a very complex and interactive manner. Changes in the cell physiology, such
that the cell processes are switched from production of biomass to that of a
secondary metabolite or product, can also be aected by selective limitation
on the quantity and rate of supply of some essential nutrient in the medium.
This can in turn be inuenced by the bioreactor hydrodynamics and also by
the mode of the operation of the bioreactor. These are the reasons why it is
not possible to develop a cell growth model without considering the coupling
between cell kinetics and bioreactor hydrodynamics.

3.3 Bioreactor classication

The various types of bioreactor can be classied according to the mode
of operation as batch, semi-continuous or continuous reactors.
As mentioned in the previous section, in a batch reactor all the materials
needed by the cell culture are introduced at the beginning of the process and
there is no introduction or removal of substances or fresh medium.
In semicontinuous or fed batch operation, additional substrate is fed into
the bioreactor, thus prolonging operation by providing an additional con-
tinuous supply of nutrients to the cells. No material is removed from the
reactor, apart from normal sampling, and therefore the total quantity of ma-
CHAPTER 3. BATCH BIOREACTOR MODELING 48

terial within the reactor will increase as a function of time. However, if the
feed is highly concentrated, then the reactor volume will not change much
and can be regarded as essentially constant.
In continuous operation, fresh medium is added continuously to the biore-
actor, while at the same time depleted medium is continuously removed. The
rates of addition and removal are such that the volume of the reactor con-
tents is maintained constant. The depleted material contains any products
that have been excreted by the cells and, in the case of suspended-cell cul-
ture, also contains euent cells from the bioreactor.

Otherwise, on the basis of the mixing technology, we can distinguish three

main bioreactor types: stirred tank, airlift and orbitally shaken bioreactor.
The stirred tank reactor is composed of a vessel and a mixer, such as a
stirred or a turbine, usually mounted with a shaft at the top of the vessel
to achieve mixing inside the tank. This is the most popular in industry and
represents a well established technology.
In the airlift reactor the culture medium is kept mixed and gassed by the
introduction of air or another gas mixture at the base of the vessel equipped
either with a draught tube or another device by which the reactor volume is
separated into a gassed and an ungassed region with dierent density, thus
generating a vertically circulating ow. In general, these reactors are less
expansive than the stirred tank bioreactors, however the mixing properties
are likely to be more critical.
The orbitally shaken bioreactor is a vessel, usually cylindrical, put in
motion by a shaker. The mixing is archived by the uid motion resulting
from friction force which is exerted by the liquid-contacting vessel wall, and
most of all, by the high inertial forces due by the shaking movement of the
tank. The shaken bioreactors have been widely used in the biotechnological
eld since they are easily to handle and inexpensive. In spite of their large
practical importance, very little is known about the characteristic properties
of shaken cultures. These are the reasons why in the last years many studies
about cell growth have been carried out using this alternative technology
[31] and models for shaken bioreactors have been recently proposed [4].
In this work, in particular, we focus on cell growth in shaken bioreactors
under batch conditions.
CHAPTER 3. BATCH BIOREACTOR MODELING 49

3.4 Balance equations for the batch fermenter

As advanced in the previous section, CHO cells are cultivated mainly
under batch conditions. Inherent in this system is that the cell environment
and hence the cell composition and physiological state change during the
culture period.
Starting from an inoculum X0 at t = 0 and an initial quantity of limiting
substrate S0 at t = 0, the biomass will grow, perhaps after a short lag phase,
and consume substrate. As the substrate becomes exhausted, the growth
rate will slow and become zero when substrate is completely depleted. If
we suppose that the volume V of the bioreactor is constant, we obtain the
following general balances:

• total volume balance

dV
= 0,
dt
• substrate balance
dS
V = rS V,
dt
• biomass balance
dX
V = rX V.
dt
Suitable rate expressions for rS and rX and the specication of the initial
conditions would complete the batch fermenter model, which describes the
growth process.
These balances can be applied to all the components of the culture
medium that are neither added nor removed from the bioreactor, for example
glucose, glutamine and cells, but they can not be applied to substances, like
oxygen and carbon dioxide, that are exchanged between the culture medium
and the atmosphere. Therefore in the next section we are going to analyze
the mass transfer process in order to complete the batch fermenter model.

3.5 Mass transfer

The bioreactor is a multiphase system in which the transport of mate-
rial between the gaseous phase and the liquid phase plays an essential role.
Usually one of the reactants is transferred from one phase into a second
CHAPTER 3. BATCH BIOREACTOR MODELING 50

phase, in which the reaction takes place. The typical example is the aera-
tion of the bioreactor culture medium and the supply of oxygen to the cells.

Concentration gradients are the driving forces for mass transfer. Actual
concentration gradients in the very near vicinity of the gas-liquid interface,
under mass transfer conditions, are very complex. They result from an inter-
action between the mass transfer process and the local uid hydrodynamics,
which changes gradually from stagnant ow, close to the interface, to perhaps
fully-developed turbulence within each of the bulk phases.
According to the Two-Film Theory [15], the actual concentration proles
can be approximated by linear gradients. A thin lm of uid is assumed
to exist at either side of the interface. Away from these lms, each uid
is assumed to be in fully developed turbulent ow. There is therefore no
resistance to mass transfer throughout each relevant phase. At the phase
interface itself, it is assumed that there is no resistance to mass transfer, and
the interfacial concentrations, CGi and CLi , are therefore in local equilibrium
with each other. All the resistance to mass transfer must occur within the
lms. In each lm, the ow of uid is assumed to be stagnant, and mass
transfer is assumed to occur only by molecular diusion and therefore to
be described by Fick's law, which says that the ux jA (mol/s·m2 ) for the
molecular diusion of the component A is given by
dC
jA = −D ,
dZ
where D is the molecular diusion coecient (m2 /s) and dC/dZ is the steady
state concentration gradient (mol/m3 · m). Thus applying the same concept
to mass transfer across the two lms we obtain:
CG − CGi CL − CLi
jA = DG = DL ,
ZG ZL
where DG and DL are the eective diusivity of each lm and ZG and ZL
are the respective thickness of the two lms. The above equations can be
expressed in terms of mass transfer coecient kG and kL (m2 /s) for the gas
and liquid lms:

jA = kG (CG − CGi ) = kL (CLi − CL ).

The total rate of mass transfer Q (mol/s) is given by

Q = jA A,
CHAPTER 3. BATCH BIOREACTOR MODELING 51

where A is the total interfacial area available for mass transfer. Thus we
obtain
Q = kG A (CG − CGi ) = kL A (CLi − CL ).

Since the mass transfer coecient ki and the interfacial area A depend
on the same hydrodynamic conditions and system physical properties, they
are frequently combined and referred to as a "kL A value" or "kG A value",
or more properly as mass transfer capacity coecients.

In the above theory, the interfacial concentrations CGi and CLi can not
be measured, and are therefore of relatively little use, even if the values
of the lm coecient were known. For this reason, by analogy to the lm
equations, overall mass transfer rate equations are dened, based on overall
coecients of mass transfer, KG and KL , and overall concentration driving
force terms:
∗
Q = KG A(CG − CG ) = KL A(CL∗ − CL ).
∗ and C ∗ are the respective equilibrium concentrations, corre-
Here, CG L
sponding to the bulk phase concentrations CL and CG respectively.
Equilibrium relationships for gas-liquid systems, at low concentrations of
component A, usually obey Henry's law, which is a linear relation between
∗ :
gas partial pressure, pA , and equilibrium liquid phase concentration, CLA

∗
pA = HA CLA ,

where HA (bar·m3 /kg) is the Henry's law constant for component A in the
medium. Henry's law is generally accurate for gases with low solubility, such
as of oxygen in water or in culture media.

For gases of low solubility, such as oxygen and carbon dioxide in water,
the concentration gradient through the gas lm is very small, compared to
that within the liquid lm. This results from the relatively low resistance
to mass transfer in the gas lm, as compared to the much greater resistance
to mass transfer in the liquid lm. The main resistance to mass transfer is
predominantly within the liquid lm. This causes a large jump in concen-
tration CLi − CL , since the resistance is almost entirely on the liquid side of
the interface.
CHAPTER 3. BATCH BIOREACTOR MODELING 52

At the interface, the liquid concentration CLi is in equilibrium with that

of the gas, CGi , and since CGi is very close in magnitude to the bulk gas
concentration, CLi must then be very near to equilibrium with the bulk gas
phase concentration CG . This is known as liquid lm control and corre-
sponds to the situation where the overall resistance to mass transfer resides
almost entirely within the liquid phase. The overall mass transfer capacity
coecient is KL A. Hence the overall mass transfer rate equation used for
slightly soluble gases in terms of the specic interfacial area a (interfacial
area per unit liquid volume) is

Q = KL a (CL∗ − CL ) VL ,

where CL∗ is in equilibrium with CG , as given by Henry's law:

CG = HCL∗ .

Mass transfer coecients in bioreactor are therefore generally referred to

in terms of KL values or KL a values for the case of mass transfer capacity
coecients.
Generally the KL a value depends on the mode of bioreactor operation,
for example, the dimensions of the vassel and the shaken velocity, and it can
be estimated from experimental results.

3.5.1 Oxygen and carbon dioxide balances for gas-liquid trans-

fer
In order to characterize aeration eciency, to predict dissolved oxygen
concentration, or to follow the biological activity, it is necessary to develop
models which include expressions for the rate of oxygen transfer and the rate
of oxygen uptake by the cells. Well-mixed phase region, in which the oxygen
concentration can be assumed uniform, can be described by simple balancing
methods. Situations in which spatial variations occur require more complex
models. The following generalized oxygen balance equation is derived for
well-mixed phases, using the well mixed tank concept.
If VL is the total volume of the liquid phase we obtain
dO2
VL = KLO2 a(O2∗ − O2 )VL .
dt
CHAPTER 3. BATCH BIOREACTOR MODELING 53

Here O2∗ is the equilibrium solubility of oxygen corresponding to the oxy-

gen partial pressure in the gas phase PO2 and is calculated by Henry's law,
according to the relationship:
PO2 nH2 O
O2∗ = ,
HO2

where nH2 O is the molar concentration of water and HO2 is the Henry con-
stant for oxygen, which depends on temperature. In this balance we are
neglecting the consumption of oxygen by the cells.
In the same way we can derive the balance equation for the carbon dioxide
concentration in the batch bioreactor:
dCO2
VL = KLCO2 a(CO2∗ − CO2 )VL .
dt

3.5.2 Models for oxygen transfer in large scale bioreactors

Large-scale industrial bioreactors can generally be expected to exhibit
deviations from the idealized ow conditions of perfect mixing. Thus the
assumption of completely mixed liquid phase may not be valid. Little expe-
rimental information is available on concentration inhomogeneities or con-
centration gradients within large bioreactors. Residence time distribution
information, from which a physical and mathematical model could be esta-
blished, is also generally not available.
Convection currents within the liquid phase of a bioreactor are usually
caused by the mechanical energy inputs of agitation and aeration. It is often
reasonable to assume that slowly changing quantities, such as biomass con-
centration, substrate concentration, pH and temperature are uniform within
the whole mass of bioreactor liquid. Oxygen must be considered, however,
as a rapidly changing substrate, owing to its low solubility in culture media.
It is therefore necessary to consider that dierences in oxygen transfer and
uptake will create oxygen concentration gradients throughout the reactor.
The possibility that oxygen transfer parameters can vary with the dimen-
sions and the conditions in which the bioreactor works, like rotational speed,
introduces a much greater degree of complexity to the problem of modeling
the bioreactor.
Chapter 4

A new cell growth model

4.1 Introduction
The purpose of this chapter is to present a new model for cell growth.
In particular the rst step is to derive an expression for the cell growth rate
as a function of the environmental properties that we have analyzed in the
previous sections. Then, we derive the rates for the substrates consumption
and the products formation as a function of the cells dynamics. In this rst
model, that we are going to introduce in the next section, we neglect the
spatial dependance of the concentrations of cells and metabolites and we
consider only the time dependance. Thus, we refer to it as 0D model.
Later on, we extend the 0D model in order to consider the eect of
the distance between cells and the culture medium surface. In particular,
we expect a dependance of the aerobic growth rate from the depth of the
medium, whereas oxygen diuses in the medium through the uid surface.
This model will be referred to as 1D model.
Finally, we present a 3D model in which we introduce the transport ef-
fect on cell growth. We expect that the dynamics of the uid would help the
mixing of the metabolites in the medium and the exchange of oxygen and
carbon dioxide between the gas phase and the liquid phase, thus enhancing
cell growth. In this work, we consider a transport eld u in the cell growth
model, without solving the uid problem. The coupling of the growth model
with the real free-surface hydrodynamics in the bioreactor is envisaged, how-
ever its implementation goes beyond the objective of this work.

54
CHAPTER 4. A NEW CELL GROWTH MODEL 55

4.2 0D model
It has been shown that the kinetic behavior of microorganism can be
described by the values of biomass specic rates, namely the specic rate of
substrate consumption, the specic cell growth rate and the specic rate of
product formation. The values of these rates depend on properties of the or-
ganism itself, which are determined by its genome, as well as on environmen-
tal conditions, such as the concentrations of compounds in the extracellular
environment.
It is obvious that a decrease in the extracellular substrate concentration
will lead, at suciently low concentration, to a decrease in the biomass spe-
cic substrate uptake rate. Since the substrate uptake rate and the specic
cell growth rate are intimately linked, it also follows that the cell growth
rate decreases at lower substrate concentration. Finally the rate of product
formation is also expected to change. Thus, the aim is to nd the algebraic
functions, called the kinetic functions, which describe how the rates depend
on the environmental conditions.
The kinetic functions are algebraic equations describing how changes in
extracellular environment modify the value of the rate ri of the component
i of the culture medium that we are considering.
These functions are nonlinear and should in principle contain the eects
of all dierent concentrations present in the growth medium on ri . Be-
cause the growth medium contains so many (> 30) dierent compounds,
the complete kinetic function for each ri would be very complex. Such
complexity would prohibit practical use, both with respect to the deter-
mination of the function itself as well as the determination of the values of
the kinetic parameters and with respect to the application in mathematical
models. Fortunately, a general property of biological systems allows con-
siderable simplication: during an experiment several of these factors are
usually kept constant. For example, the extracellular concentrations of some
of the substrates, such as water, are so high that the rates can be considered
independent of the extracellular concentration values.
However, it is always possible to identify one, or more, substrates that
can be considered limiting-nutrients. For example, as mentioned in the pre-
vious chapters, an exceeding concentration of oxygen can be toxic for CHO
CHAPTER 4. A NEW CELL GROWTH MODEL 56

cells. Moreover it has been proved that an excessive metabolic ux and
excretion of toxic by-products is diminished greatly when the external con-
centrations of glucose and glutamine are low. Therefore, it is critical that
the glucose and glutamine concentrations in a bioreactor are controlled at a
low level so as to decrease the lactate and ammonia accumulation [16] and
that the oxygen concentration in the gas phase above the culture medium is
maintained at constant, quite low, level.

In the case of CHO cells, three limiting-nutrient components of the cul-

ture medium can be identied:

• glucose Glc;

• glutamine Gln;

• oxygen O2 .

Each of them is involved in more than one metabolic pathway: under

the hypothesis introduced in the previous chapters and assuming that all
the other substances, enzymes and metabolites involved are not limiting-
components, the reactions can be simplied as follows:

• aerobic consumption of glucose

Glc + 6O2 → 6CO2 + 6H2 O, (4.1)

• anaerobic consumption of glucose

Glc → 2Lac + 2H2 O, (4.2)

• aerobic consumption of glutamine

3Gln + 8O2 + 2H2 O → 7CO2 + 6Amm + 3Lac. (4.3)

Whereas water (H2 O) can be neglected since it is not a limiting-component

of the culture medium, the products that we have to introduce in the model
are

• lactate Lac;
CHAPTER 4. A NEW CELL GROWTH MODEL 57

• ammonia Amm;

• carbon dioxide CO2 .

In conclusion, our model has to describe the dynamics of seven com-

ponent, i.e. cells, three substrates and three products, which are coupled
through the reactions (4.1), (4.2) and (4.3).

4.2.1 Cell growth rate

As introduced in the previous chapter, we consider a constant volume V
of culture medium in which we have an initial cell concentration X0 . Whereas
in a batch culture there is no introduction or removal of cells during the
process, the balance equation for the cells is:

 dX V = rX V
dt (4.4)

X(0) = X0 ,

where rX is the cell growth rate. The equation (4.4) can be written intro-
rX
ducing the specic growth rate µX = :
X

 dX = µX X
dt (4.5)

X(0) = X0 .

The expression for the specic growth rate that we are now introducing is
based on the following considerations:

• in presence of oxygen, the cell takes energy from the aerobic reactions
of glucose and glutamine, which release an higher amount of energy
respect to the anaerobic respiration;

• the anaerobic respiration takes always place, also in presence of oxygen,

but the process is inhibited until oxygen becomes limited;

• the rate of cell growth changes when the cell shifts from aerobic and
anaerobic respiration; because of the dierent amount of energy re-
leased in these two processes, we expect that cell growth would be
faster during aerobic respiration than during anaerobic respiration;

• the inhibition of cell growth is due to substrate exhaustion.

CHAPTER 4. A NEW CELL GROWTH MODEL 58

If we assume that aerobic and anaerobic respirations determine dierent

growth rates, we can decompose the specic growth rate into two terms:
¡ ¢
µX = µmax γaer µaer + γanaer µanaer ,

with γaer + γanaer = 1. The coecient µmax (1/h) is the maximum specic
growth rate, and it is function of all the environmental conditions except the
metabolites concentrations.
The term µaer represents the cell growth rate due to aerobic respiration,
while µanaer is the rate due to anaerobic respiration. These rates have the
properties that, under non-limiting nutrient conditions, they approach 1.
The constants γaer and γanaer help expressing the growth rate as a weighted
mean between the growth rate that we would have if we had only aerobic
respiration and the growth rate under anaerobic conditions. For example,
if γaer = 0.8 and γanaer = 0.2, under non limiting conditions, the specic
growth rate approaches the maximum specic growth rate and cell growth is
attributable to aerobic respiration for the 80% and to anaerobic respiration
for the 20%.

If we assume that the anaerobic consumption of glucose is inhibited at

high concentrations of oxygen, we can introduce the following expression for
the specic anaerobic growth rate:
Glc
µanaer = O22
,
Kanaer + Glc + KI

where Glc is the glucose concentration, O2 the oxygen concentration, KI is

the inhibition parameter and Kanaer is the anity constant for anaerobic
consumption of glucose. We can observe that this term becomes zero when
glucose is exhausted and approaches 1 when glucose is in excess and the
oxygen concentration is low.

In order to take into account in the expression of the aerobic growth rate
µaer of both the glucose and the glutamine consumption, we introduce the
following decomposition:

µaer = γGlc µGlc + γGln µGln ,

CHAPTER 4. A NEW CELL GROWTH MODEL 59

where the coecients γGlc and γGln , with γGlc + γGln = 1, have the same
meaning of γaer and γanaer .
The expression for µGlc is based on the consideration that the cell, in
order to obtain energy from glucose oxidation, needs both glucose and oxygen
at the same time. Thus µGlc depends on the product of glucose and oxygen
concentrations as follows:
µ ¶µ ¶
Glc O2
µGlc = .
KGlc + Glc KO2 + O2

We can observe that if either oxygen or glucose runs out, the growth rate
due to aerobic consumption of glucose becomes zero, while if both glucose
and oxygen are in excess, µGlc approaches 1.
In the same way, we dene the growth rate due to glutamine oxidation:
µ ¶µ ¶
Gln O2
µGln = .
KGln + Gln KO2 + O2

In conclusion, the model for µX that we have introduced considers the

inhibitory eect on cell growth due to substrate consumption and exhaustion.
In particular, if all the substrates are not limiting-nutrients, the specic cell
growth rate will approach µmax , while as the substrates are consumed, the
growth rate becomes zero.

4.2.2 Substrate uptake rate

The dynamics of the substrates that we are considering, namely glucose,
glutamine and oxygen, is coupled with the cell dynamics through the adi-
mensional yield coecients YS , that represents the amount of cells produced
for unit of substrate S consumed.
The generic balance equation for the substrate S has the following ex-
pression: 
 dS = r = − 1 µ X
S
dt YS X|S (4.6)

S(0) = S0 ,
where µX|S is the specic cell growth rate due to the substrate S consump-
tion. Starting from the expression for µX , we can derive the consumption
rates for glucose, glutamine and oxygen.
CHAPTER 4. A NEW CELL GROWTH MODEL 60

Glucose consumption rate

Whereas glucose consumption is related both to aerobic oxidation and
anaerobic metabolism, its balance equation becomes:
 µ ¶
 dGlc = −µ 1
γaer γGlc µGlc +
1
γanaer µanaer X
max
dt YGlc Yanaer (4.7)

Glc(0) = Glc0 ,

where YGlc is the amount of cells produced thanks to glucose aerobic con-
sumption and Yanaer is the amount of cells produced thanks to glucose anae-
robic consumption. We can observe that the glucose consumption would be
zero either when the cell concentration is zero or when glucose concentration
is zero.

Glutamine consumption rate

Glutamine is involved in aerobic respiration and its balance equation
takes the following form:

 dGln = −µ 1
γaer γGln µGln X
max
dt YGln (4.8)

Gln(0) = Gln0 ,

where YGln is the amount of cells produced for unit of glutamine consumed.
Glutamine consumption is zero either when cell concentration is zero, or
when glutamine or oxygen are zero.

Oxygen consumption rate

Oxygen dynamics is related only to aerobic respiration and, if we neglect
the oxygen transfer between gas phase and liquid phase, it can be described
by the following balance equation:
 µ ¶
 dO2 = −µmax γaer
 1
γGlc µGlc +
1
γGln µGln X
dt YO2 |Glc YO2 |Gln (4.9)

 O (0) = O .
2 20

The yield coecients YO2 |Glc and YO2 |Gln represent the amount of cells
produced for unit of oxygen consumed respectively through glucose and glu-
tamine oxidation. The value of these yield coecients are dependent on the
yield coecients for glucose and glutamine. In fact, from the stoichiometric
CHAPTER 4. A NEW CELL GROWTH MODEL 61

coecients of the reactions introduced in the previous section, we know that

we need 6 moles of oxygen to oxidize 1 mole of glucose and we need 2 moles
of oxygen to oxidize 1 mole of glutamine.
If we consider the consumption of oxygen through the glucose aerobic
reaction, whereas for one mole of glucose the cell consumes 6 moles of oxygen,
we obtain:
¯
X produced ¯¯ X produced Glc consumed 1 · MGlc
YO2 |Glc = = · = YGlc ,
O2 consumed ¯Glc Glc consumed O2 consumed 6 · MO2

where MGlc and MO2 are the molar masses of glucose and oxygen respec-
tively. In the same way we can express the yield coecient YO2 |Gln as a
function of the glutamine yield coecient YGln :
¯
X produced ¯¯ 1 · MGln
YO2 |Gln = ¯ = YGln ,
O2 consumed Gln 2 · MO2

where MGln is the molar mass of glutamine.

As introduced in the previous chapter, the batch culture system is char-
acterized by the exchange of oxygen between the culture medium and the
atmosphere above the liquid. If we introduce the mass transfer term in the
balance equation for the oxygen (4.9), we obtain:
 µ ¶
 dO2 1 1

 = −µmax γaer γGlc µGlc + γGln µGln

 dt YO2 |Glc YO2 |Gln
¡ ¢ (4.10)
 +KLO2 a O2∗ − O2 X




O2 (0) = O20 .

4.2.3 Product formation rate

As done in the previous section for the substrate consumption, we can
derive the rate of product formation from the cell growth rate µX . The
generic balance equation for the product P has the following form:

 dP = r = 1 µ X
P
dt YP X|P (4.11)

P (0) = P0 ,

where µX|P is the specic cell growth rate related to the product P formation.
In this section we derive the balance equations for lactate, ammonia and
carbon dioxide from the expression for µX .
CHAPTER 4. A NEW CELL GROWTH MODEL 62

Lactate production rate

Lactate is produced by the process of anaerobic respiration and by the
process of glutamine oxidation. We obtain the following expression:
 µ ¶
 dLac = µmax
 1
γanaer µanaer +
1
γaer γGln µGln X
dt YLac|Glc YLac|Gln

 Lac(0) = Lac .
0
(4.12)
The yield coecients of lactate related to the anaerobic reaction YLac|Glc
and to the aerobic reaction YLac|Gln depend on the glucose and glutamine
yield coecients:
¯
X produced ¯¯ 1 · MGlc
YLac|Glc = ¯ = YGlc
Lac produced Glc 2 · MLac

and ¯
X produced ¯¯ 3 · MGln
YLac|Gln = ¯ = YGln .
Lac produced Gln 3 · MLac
where MLac is the molar mass of lactate.

Ammonia production rate

Ammonia is produced only in the process of glutamine oxidation. We
obtain the following balance equation:
 dAmm 1
 = µmax γaer γGln X
dt YAmm|Gln (4.13)

Amm(0) = Amm0 ,

where the yield coecient YAmm|Gln is related to the glutamine yield coe-
cient YGln :
¯
X produced ¯¯ 3 · MGln
YAmm|Gln = = YGln .
Amm produced ¯Gln 6 · MAmm

where MAmm is the molar mass of ammonia.

Carbon dioxide production rate

Carbon dioxide is produced both in the process of glucose oxidation and
in the process of glutamine oxidation. Moreover, we have to consider the
CHAPTER 4. A NEW CELL GROWTH MODEL 63

exchange between the culture medium and the gas phase. We obtain:
 µ ¶
 dCO2 1 1

 = µmax γaer γGlc µGlc + γGln µGln X+

 dt YCO2 |Glc YCO2 |Gln
¡ ¢
 +KLCO2 a CO2∗ − CO2




CO2 (0) = CO20 ,
(4.14)
where the yield coecient YCO2 |Glc of carbon dioxide through glucose oxida-
tion depends on glucose yield coecient:
¯
X produced ¯¯ 1 · MGlc
YCO2 |Glc = = YGlc ,
CO2 produced ¯Glc 6 · MCO2

and the yield coecient YCO2 |Gln of carbon dioxide through glutamine oxi-
dation depends on glutamine yield coecient:
¯
X produced ¯¯ 3 · MGln
YCO2 |Gln = ¯ = YGln .
CO2 produced Gln 7 · MCO2

where MCO2 is the molar mass of carbon dioxide.

4.2.4 Remarks
The cell growth model introduced in this section consists of a system of
seven ordinary dierential equations (4.5), (4.7), (4.8), (4.10), (4.12), (4.13),
(4.14), that are coupled through the reaction terms ri .
The reaction rates are algebraic functions of the substrate concentrations
Glc, Gln and O2 . The relationships between the concentrations and the rates
are given by dierent types of model parameters:

• the fundamental yield coecients YGlc and YGln ;

• the derived yield coecients Yi|j , that depend on the fundamental yield
coecients through the stoichiometric coecients;

• the weight coecients γaer , γanaer , γGln and γGlc ;

• the half-saturation constants KGlc , KGlc , KO2 and Kanaer ;

• the maximum specic growth rate µmax ;

• the mass transfer coecients KLO2 a and KLCO2 a.

CHAPTER 4. A NEW CELL GROWTH MODEL 64

Except for the derived yield coecients, that are estimated from the
stoichiometric coecient of the metabolic reactions, these parameters should
be estimated through appropriate experiments.

4.2.5 Numerical approximation

The 0D model introduced in the previous section is solved with Matlab
and in particular the solver ode45, which is based on an explicit Runge-Kutta
formula. The results obtained from the 0D model are showed and discussed
in the next chapter.

4.3 1D model
In his section we want to introduce in the cell growth model the spatial
dependance, in particular the eect on the growth kinetics of the distance
from the interface between the culture medium and the atmosphere.
The new model that we are now introducing will be referred to as 1D
model, whereas we take into account of only one spatial coordinate (Figure
4.1). While in the 0D model the eect of the hydrodynamics was included
indirectly through the mass transfer coecient KL a, in the 1D model we are
not able to consider this eect on the growth kinetics.

Figure 4.1: 1D domain

CHAPTER 4. A NEW CELL GROWTH MODEL 65

Our model is based on the following further hypotheses:

• the dependance of cell growth on the distance from the cylinder axis
is negligible compared with the dependance on the distance from the
uid surface;

• the uid is steady;

• at the beginning, the cell concentration and the concentrations of glu-

cose, glutamine, oxygen and carbon dioxide are uniform in the culture
medium, while the concentrations of lactate and ammonia are zero;

• all the substances diuse in the culture medium: we dene Di the

diusivity coecient of the substance i;

• cells diusivity is negligible respect to the diusivity of the other com-

ponents; whereas the uid is steady, this hypothesis implies that each
cell is almost xed in its initial position;

• oxygen and carbon dioxide diuse from the gas phase to the liquid
phase, and conversely; there is no exchange of cells, substrates and
products between culture medium and atmosphere, except for oxygen
and carbon dioxide.

The second hypothesis that we have introduced is the most restricting: in

fact from experimental results we know that cell growth is strongly dependent
on the uid dynamics, which contributes to make the distribution of cells
and substrates uniform in the medium.
In order to build up the model, we dene the domain Ω = [0, L] and we
suppose that the point z = 0 represents the interface between gas phase and
culture medium, as shown in Figure 4.1. We obtain the following system of
partial dierential equations:

• Cell concentration X


 dX d2 X

 − D X − µX X = 0, in Ω, ∀ t > 0
 dt
 dz 2


 X(z, 0) = X0 (z), in Ω
d (4.15)

 X(0, t) = 0, ∀t>0



 dz

 d X(L, t) = 0,

∀t>0
dz
CHAPTER 4. A NEW CELL GROWTH MODEL 66

• Glucose concentration Glc



 dGlc d2 Glc

 − D Glc + µGlc Glc = 0, in Ω, ∀ t > 0

 dt dz 2


 Glc(z, 0) = Glc0 (z), in Ω
d (4.16)

 Glc(0, t) = 0, ∀t>0



 dz

 d Glc(L, t) = 0,

∀t>0
dz
• Glutamine concentration Gln


 dGln d2 Gln

 − D Gln + µGln Gln = 0, in Ω, ∀ t > 0

 dt dz 2


 Gln(z, 0) = Gln0 (z), in Ω
d (4.17)

 Gln(0, t) = 0, ∀t>0



 dz

 d Gln(L, t) = 0,

∀t>0
dz
• Oxygen concentration O2

 dO2 d2 O2

 − D O + µO2 O2 = 0, in Ω, ∀ t > 0

 dt 2
dz 2


 O (z, 0) = O (z), in Ω
2 20
(4.18)

 O2 (0, t) = O2∗ , ∀t>0





 d O (L, t) = 0,
2 ∀t>0
dz
• Lactate concentration Lac


 dLac d2 Lac

 − DLac = rLac , in Ω, ∀ t > 0

 dt dz 2


 Lac(z, 0) = Lac0 (z), in Ω
d (4.19)

 Lac(0, t) = 0, ∀t>0



 dz

 d Lac(L, t) = 0,

∀t>0
dz
• Ammonia concentration Amm


 dAmm d2 Amm

 − D Amm = rAmm , in Ω, ∀ t > 0

 dt dz 2


 Amm(z, 0) = Amm0 (z), in Ω
d (4.20)

 Amm(0, t) = 0, ∀t>0



 dz

 d Amm(L, t) = 0,

∀t>0
dz
CHAPTER 4. A NEW CELL GROWTH MODEL 67

• Carbon dioxide concentration CO2


 dCO2 d2 CO2

 − D CO = rCO2 , in Ω, ∀ t > 0

 dt 2
dz 2


 CO (z, 0) = CO (z), in Ω
2 20
(4.21)

 CO2 (0, t) = CO2∗ , ∀t>0





 d CO (L, t) = 0,
2 ∀ t > 0.
dz

The 1D model is composed of seven parabolic diusion-reaction equa-

tions: whereas the rate coecients µi depend on the unknown quantities,
the equations are non linear and coupled through the reaction terms.

4.3.1 Remarks
Unlike the 0D model, in the 1D model we introduce the mass diusivity
Di of each substance in water. It is possible to nd the value of the diusivity
coecients in the literature or to estimate them from the Einstein-Stokes
equation:
kB T
D= ,
6πηr
where kB is Boltzmann constant, T is the absolute temperature, η is water
viscosity and r is the radius of the molecule.

The coecient µi is the growth (or consumption) rate of the i-th sub-
stance. The specic cell growth rate µX has been already introduced in the
previous section and in the 1D model it maintains the same expression of
the 0D model. We can observe that, whereas the concentrations depend on
the spatial coordinate z , now also the cell growth rate depends on z .
The specic consumption rate µS for the substrate S is dened as
rS
µS = −
S
where rS is the consumption rate for S introduced in the 0D model, while
rP is the same production rate introduced in the 0D model. We can observe
that the production rate is independent from the concentration of P itself,
therefore in the balance equations for the product P we don't have the
reaction term but a forcing term coupled with the other equations.
CHAPTER 4. A NEW CELL GROWTH MODEL 68

In order to make the model well-posed, appropriate boundary condi-

tions have been imposed to describe the mass exchanges between the culture
medium and the external environment. In particular we use homogeneous
Neumann conditions to impose the null ux at z = L for each substance and
at z = 0 for each substance except oxygen and carbon dioxide. Moreover,
in agreement with the Two Film theory introduced in the previous chapter,
we impose Dirichlet boundary conditions equal to the saturation values at
z = 0 for oxygen and carbon dioxide.

Unlike the 0D model, the 1D model does not contain the mass transfer
parameter KL a, that measure the capacity of the uid to exchange mass
with the gas phase. The KL a is an experimental parameter that depends on
the gas diusivity coecient, on the bioreactor geometry and on the biore-
actor operational mode, in particular on the uid dynamics. It is a "global"
parameter, because, whereas in the 0D model the spatial dependance is not
considered, it is as if the liquid phase would exchange mass with the gas phase
in each point of its volume. Because of the hypothesis of steady uid, in the
1D model the exchange of oxygen and carbon dioxide between gas phase and
liquid phase is determined only by the Dirichlet boundary conditions and the
gas diusivity in water, but not by the uid motion. Consequently, with the
1D model we do not expect the same results of the 0D model, since it is
well known that the uid dynamics highly aect the mixing and the mass
exchanges in the bioreactor and hance the substances concentrations.

4.3.2 Numerical approximation

In order to solve the 1D model introduced in the previous section, the
governing equations must be discretized in both space and time. The tempo-
ral discretization is based on a rst-order semi-implicit scheme. We consider
a uniform decomposition of the time interval [0, T ] into Nt subintervals. We
denote the time step with ∆t = T /Nt and we use the index n to denote
variables at time tn = n∆t, with n = 0, . . . , Nt .
The generic expression for the time evolution of cells and substrates is
given by:
dS(t) d2 S(t)
− DS + µS (t)S(t) = 0.
dt dz 2
CHAPTER 4. A NEW CELL GROWTH MODEL 69

where µS (t) represents the time dependent reaction coecient.

If the time derivative is discretized using the backward nite dierences,
the rst-order time discretization is given by:
S n+1 − S n d2 S n+1
− DS + µS (t∗ )S(t∗∗ ) = 0.
∆t dz 2
If we consider t∗ = tn and t∗∗ = tn+1 , the time discretization can be
referred to as semi-implicit, and we obtain:

1 n+1 d2 1 n
S − DS 2 S n+1 + µS (tn )S n+1 = S .
∆t dz ∆t
In the same way, if we consider the generic expression for the time evo-
lution of products:
dP (t) d2 P (t)
− DP = rP (t),
dt dz 2
and we apply the previous approach, we obtain:
1 n+1 d2 P n+1 1 n
P − DP 2
= P + rPn .
∆t dz ∆t
We can observe that the semi-implicit time discretization permits to de-
couple the balance equations:
 n+1

 X d2 n+1 1 n

 − D X 2
X + µnX X n+1 = X

 ∆t dz ∆t

 Glcn+1 d2 1



 − DGlc 2 Glcn+1 + µnGlc Glcn+1 = Glcn

 ∆t dz ∆t



 Glnn+1 d2 1

 − D Gln Glnn+1 + µnGln Glnn+1 = Glnn

 ∆t dz 2 ∆t
 n+1
O2 d2 n+1 1 n
− D O O2 + µnO2 O2n+1 = O


 ∆t 2
dz 2 ∆t 2

 Lacn+1 d2 1

 − D Lacn+1 = Lacn + rLac
n

 Lac 2

 ∆t dz ∆t

 Ammn+1 d2 1



 − DAmm 2 Ammn+1 = Ammn + rAmmn

 ∆t dz ∆t

 n+1 2
 CO2 − DCO d COn+1 = 1 COn + rn

2 2 CO2
∆t 2
dz 2 ∆t

The spatial discretization is based on a nite-element approach. We

consider the general equation for cells and substrates concentrations on the
domain Ω = [0, L]:
dS d2 S
− DS 2 + µS (z, t)S = 0, (4.22)
dt dz
CHAPTER 4. A NEW CELL GROWTH MODEL 70

with Dirichlet boundary conditions on ΓD and homogeneous Neumann boun-

dary conditions on ΓN . We suppose that we are looking for a solution S in
an appropriate space V and we derive the week formulation of (4.22) by
multiplying for a test function v such that v ∈ V and v|ΓD = 0:

Find S ∈ V such that

Z L Z L Z L
dS dS dv
v dz + DS dz + µS (z, t) S v dz = 0
0 dt 0 dz dz 0

for all v ∈ V , with v|ΓD = 0.

We consider a uniform discretization Th of the domain Ω = [0, L] into

Nh subintervals Ki , with i = 1, . . . , Nh , and we dene h = L/Nh their
dimension. Moreover, we use the index i to denote variables in the node
zi = ih, with i = 0, . . . , Nh . We introduce the space Xh1 = {vh ∈ C 0 (Ω) :
vh |Ki ∈ P1 , ∀Ki ∈ Th }, where P1 is the space of polynomial function with
grade 1, and the space Vh = {vh ∈ Xh1 } ⊂ V . Finally, we consider a
lagrangian base {ϕi , i = 0, . . . , Nh } for the space Vh .
The nite element formulation for the problem is:

Find Sh ∈ Vh such that:

Z L Z L Z L
dSh dSh dvh
vh dz + DS dz µS (z, t) Sh vh dz = 0 (4.23)
0 dt 0 dz dz 0

for all vh ∈ Vh , with vh |ΓD = 0.

If we express Sh as a linear combination of the base functions ϕj :

Nh
X
Sh = Sj (t) ϕj (z)
j=0

and we impose that the equation (4.23) is valid for all vh ∈ Vh , we obtain
N Z Nh Z N Z
d Xh L X L
dϕj dϕi Xh L
Sj (t) ϕj ϕi dz+DS Sj (t) dz+ Sj (t) µS ϕj ϕi dz = 0
dt 0 0 dz dz 0
j=0 j=0 j=0

and then
dS
M + DS AS + RS S = 0,
dt
CHAPTER 4. A NEW CELL GROWTH MODEL 71

where we have introduced the stiness matrix

Z L
dϕj dϕi
[A]ij = dz,
0 dz dz
the mass matrix Z L
[M]ij = ϕj ϕi dz
0
and the reaction matrix
Z L
[RS ]ij = µS (z, t) ϕj ϕi dz.
0

Since we are using linear nite elements, the stiness matrix and, in
general, the reaction matrix will be tridiagonal while the mass matrix will
be diagonal. In order to build up the reaction matrix, in particular, we need
to introduce an integration rule in order to estimate its terms, because the
reaction coecient is not constant on the domain. We apply the trapezoidal
rule, that is:
Z b
b−a £ ¤
f (z) dz = f (a) + f (b) .
a 2
We can observe that, thanks to the properties of the linear nite elements,
we obtain that (
0 if j 6= i,
[RS ]ij =
h µS (zi , t) if j = i.
In conclusion, the reaction matrix that we obtain is diagonal as if we
applied the mass lumping technique to the reaction-diusion problem [21].
Consequently we don't expect instabilities related to dominating reaction.

Finally, if we apply the semi-implicit time discretization already intro-

duced, we obtain:
1 1
MSn+1 + DS ASn+1 + RnS Sn+1 = MSn
∆t ∆t
In the same way, we can derive the nite element formulation for the
general product equation:

dP d2 P
− DP = rP (z, t). (4.24)
dt dz 2
We obtain
dP
M + DP AP = rP (t)
dt
CHAPTER 4. A NEW CELL GROWTH MODEL 72

and
1 1
MPn+1 + DP APn+1 = MPn + rnP ,
∆t ∆t
where Z L
rP,i = rP (z) ϕi dz = h rP (zi ).
0
In conclusion, we obtain the following set of linear systems:

 1 1


 MXn+1 + DX AXn+1 + RnX Xn+1 = MXn

 ∆t ∆t

 1 1

 MGlcn+1 + DGlc AGlcn+1 + RnGlc Glcn+1 = MGlcn

 ∆t ∆t


 1
 1


 MGlnn+1 + DGln AGlnn+1 + RnGln Glnn+1 = MGlnn

 ∆t ∆t
1 1
MOn+1
2 + DO2 AOn+1
2 + RnO2 On+1
2 = MOn2 (4.25)

 ∆t ∆t

 1 1

 MLacn+1 + DLac ALacn+1 = MLacn + rnLac



 ∆t ∆t

 1 1

 MAmmn+1 + DAmm AAmmn+1 = MAmmn + rnAmm

 ∆t ∆t



 1 MCOn+1 + DCO2 ACOn+1 = 1 MCOn2 + rnCO

2 2
∆t ∆t 2

These linear systems are implemented and solved using Matlab. Thanks
to the semi-implicit time discretization, the systems are decoupled and con-
sequently they can be solved sequentially.
The algorithm can be synthesized as follows:

• definition of the model parameters;

• definition of the initial values;
• construction of the mass matrix M and the stiffness matrix A;
• temporal loop: for n = 1 ... Nt do
- loop on the systems: for i = 1 ... 7 do
- construction of the reaction matrices R and the
right-hand-side using the solution at the time step
n-1;
- imposition of the boundary conditions;
- solution of the system i at the time step n;
- end for;
• end for.
CHAPTER 4. A NEW CELL GROWTH MODEL 73

Boundary conditions
We have observed in the previous section that, since the boundary con-
ditions of all the balance equations are either homogeneous Neumann condi-
tions or Dirichlet conditions, we obtain the same weak formulation for each
variable, except for the reaction term. In particular the boundary terms
that derive from the integration are all equal to zero. Consequently, the last
step to be done before solving the system (4.25) is to impose the Dirichlet
boundary conditions for the oxygen and the carbon dioxide concentrations
on the interface between liquid and gas at z = 0. We decide to impose the
boundary value directly by replacing the equation related to the node z0
with
O2,0 = O2∗

for the oxygen and

CO2,0 = CO2∗

for the carbon dioxide. This approach requires only the modication of one
row of the system and it is quite simple.

4.4 3D model
The development of the 3D model for the cell growth process requires
the coupling of the growth model for the cells with the real free-surface
hydrodynamics inside the bioreactor. While the extension of the cell growth
model to a 3D domain is quite straightforward, the last problem requires
the solution of the Navier-Stokes equations for an incompressible Newtonian
uid on a moving domain Ω ⊂ R3 :

 ∂u

 ρ − µ∆u + ρu · ∇u + ∇p = 0, in Ω(t), ∀ t > 0

 ∂t



 ∇ · u = 0, in Ω(t), ∀ t > 0


u(x, 0) = u0 (x), in Ω(t) (4.26)





 u(x, t) = g(x, t), su ΓD (t), ∀ t > 0



 µ ∂ u − pn = h(x, t)

su ΓN (t), ∀ t > 0
∂n
where u is the velocity eld, p is the pressure, ρ is the uid density and µ is
the uid viscosity.
CHAPTER 4. A NEW CELL GROWTH MODEL 74

The implementation of the real bioreactor hydrodynamics is quite com-

plex and goes beyond our objective of testing the global properties of the
cell growth model. Therefore in this work we do not solve the uid dynamics
and we suppose that the transport eld u is known.

We assume that the uid transports the cells, the substrates and the
products and that the transport is passive, that is the uid dynamics is not
aected by the concentrations of the substances. This assumption is justied
by the the fact that the cell density is similar to the water density [26]. Under
this hypothesis, if we dene Γup the interface between the liquid phase and
the atmosphere, we obtain the following set of partial dierential equations:

• Cell concentration X

 ∂X

 − DX ∆X + u · ∇X − µX X = 0, in Ω, ∀ t > 0

 ∂t
X(x, 0) = X0 (x), in Ω (4.27)



 ∂ X(x, t) = 0,

su ∂Ω
∂n

• Glucose concentration Glc


 ∂Glc

 − DGlc ∆Glc + u · ∇Glc + µGlc Glc = 0, in Ω, ∀ t > 0

 ∂t
Glc(x, 0) = Glc0 (x), in Ω



 ∂ Glc(x, t) = 0,

su ∂Ω
∂n
(4.28)

• Glutamine concentration Gln


 ∂Gln

 − DGln ∆Gln + u · ∇Gln + µGln Gln = 0, in Ω, ∀ t > 0
 ∂t

Gln(x, 0) = Gln0 (x), in Ω



 ∂ Gln(x, t) = 0,

su ∂Ω
∂n
(4.29)
CHAPTER 4. A NEW CELL GROWTH MODEL 75

• Oxygen concentration O2

 ∂O2

 − DO2 ∆O2 + u · ∇O2 + µO2 O2 = 0, in Ω, ∀ t > 0

 ∂t


 O2 (x, 0) = O∗ (x), in Ω
2
(4.30)

 ∂

 O (x, t) = 0, su ∂Ω \ ΓU p



∂ n 2

O2 (x, t) = O2∗ , su ΓU p

• Lactate concentration Lac


 ∂Lac

 − DLac ∆Lac + u · ∇Lac = rLac , in Ω

 ∂t
Lac(x, 0) = Lac0 (x), in Ω (4.31)



 ∂ Lac(x, t) = 0,

su ∂Ω
∂n

• Ammonia concentration Amm


 ∂Amm

 − DAmm ∆Amm + u · ∇Amm = rAmm , in Ω

 ∂t
Amm(x, 0) = Amm0 (x), in Ω (4.32)



 ∂ Amm(x, t) = 0,

su ∂Ω
∂n

• Carbon dioxide concentration CO2


 ∂CO2

 − DCO2 ∆CO2 + u · ∇CO2 = rCO2 , in Ω

 ∂t


 CO2 (x, 0) = CO∗ (x), in Ω
2
(4.33)

 ∂

 CO2 (x, t) = 0, su ∂Ω \ ΓU p

 ∂n


CO2 (x, t) = CO2∗ , su ΓU p

The diusivity coecients Di and the rates µi and ri of the 3D model

are exactly the same as the 1D model. The only dierences between the
two models are represented by the transport term u · ∇(·) and the domain
dimension.

We can observe that the main dierence between the 0D model and the
3D model is that the compounds in the culture medium are transported by
the uid, thus helping the mixing of the substrates and the mass exchange
between the liquid phase and the gas phase. However we want to highlight
CHAPTER 4. A NEW CELL GROWTH MODEL 76

that we are not considering the free-surface hydrodynamics, which is known

to further enhance the mass exchange. As previously mentioned, the full
coupling with the complete hydrodynamics of the ow inside the bioreactor
is currently under development.
Moreover in our model we do not consider the eect of shear stresses on
the cell physiology. It has been observed, in fact, that low shear stresses can
increase cell activity, whereas high shear stresses can reduce cell viability.
In conclusion, the 3D model that we have presented allows us to under-
stand how a better mixing of the compounds in the culture medium inu-
ences the dynamics of cell growth, but not the eect of the real bioreactor
hydrodynamics.

4.4.1 Numerical approximation

In order to solve the 3D model introduced in the previous section, the
equations are discretized both in time and in space with the same approach
adopted for the 1D model discretization, that is with semi-implicit nite
dierences for time discretization and linear nite element for spatial dis-
cretization.
We obtain the following set of linear systems:

 1 1


 MXn+1 + DX AXn+1 + Bn+1 Xn+1 + RnX Xn+1 = MXn

 ∆t ∆t

 1 1

 MGlcn+1 + DGlc AGlcn+1 + Bn+1 Glcn+1 + RnGlc Glcn+1 = MGlcn

 ∆t ∆t



 1 1


 MGlnn+1 + DGln AGlnn+1 + Bn+1 Glnn+1 + RnGln Glnn+1 = MGlnn

 ∆t ∆t
1 1
MOn+1
2 + DO2 AOn+1
2 + Bn+1 On+1
2 + RnO2 On+1
2 = MOn2

 ∆t ∆t

 1 1

 MLacn+1 + DLac ALacn+1 + Bn+1 Lacn+1 = MLacn + rnLac



 ∆t ∆t

 1 1

 MAmmn+1 + DAmm AAmmn+1 + Bn+1 Ammn+1 = MAmmn + rnAmm

 ∆t ∆t



 1 1
 MCOn+1
2 + DCO2 ACOn+1
2 + Bn+1 COn+1
2 = MCOn2 + rnCO2 .
∆t ∆t
(4.34)
We have introduced the transport matrix B that is dened as:
Z
[B]ij = (un+1 · ∇ϕj ) ϕi dΩ.
n+1
Ω

where un+1 is the assigned velocity eld at time tn+1 .

CHAPTER 4. A NEW CELL GROWTH MODEL 77

As observed previously, the semi-implicit time discretization permits to

decouple the linear systems, that can be solved sequentially.

The discretized equations are solved using LifeV [6], a nite element
library providing implementation of mathematical and numerical methods.
In particular we use the advection-diusion-reaction solver, that we have op-
portunely extended in order to solve our problem. The decoupled equations
that we obtained for each time step with the semi-implicit scheme are solved
on the domain Ω, which is discretized using tetrahedral elements. Each lin-
ear equation is discretized at time tn using linear nite elements (P1), that
are implemented in the library. Moreover, the LifeV library implements the
Interior Penalty method in order to stabilize the advection-dominated prob-
lem [3]. The algebraic problem (4.34) obtained from the discretization of the
advection-diusion-reaction equations is solved using the GMRES method.
The linear systems are preconditioned using Ifpack, an object-oriented al-
gebraic preconditioner package, and in particular Amesos, that implements
the complete LU factorization.

4.5 Environmental factors

In this section we introduce in the previous models the eect of tempe-
rature on cell growth dynamics through two distinct terms. We rst present
a simple model for the maximum growth rate as a function of temperature.
Moreover, we introduce the relationship between temperature and the satu-
ration concentrations of oxygen and carbon dioxide. These expressions are
exactly the same for the three models introduced in the previous sections.
Finally, we derive an estimate of the pH in the culture medium as a
function of the concentrations of the main compounds in the medium.

4.5.1 Temperature dependance

Maximum growth rate µmax
It has been highlighted in the previous chapters that cell metabolism and
growth are highly inuenced by temperature. If we consider cell growth as
a chemical reaction, we can express the maximum specic cell growth rate
CHAPTER 4. A NEW CELL GROWTH MODEL 78

µmax as a function of the temperature T using the Arrhenius equation, as

proposed by [11]:
µ ¶ µ ¶
Ea Ea,high
µmax = A exp − − B exp − .
RT RT

The Arrhenius law has four parameters that have to be estimated; in

particular, if only one reaction is considered in the model, the coecients
Ea and Ea,high can be interpreted respectively as the activation energy and
the denaturation energy for the reaction. Consequently, whereas we con-
sider three dierent reactions, if we apply the Arrhenius law to estimate the
maximum specic growth rate in our model, we loose the mechanistic in-
terpretation of its parameters. Moreover, if we try to apply this law in our
model for CHO cells, it seems not to provide good results.
Thus, we introduce the following expression for µmax as a function of the
temperature T : µ ¶
(T − Topt )2
µmax = µopt exp − . (4.35)
r2
In the gaussian model (4.35) there are three parameters:

• Topt is the optimal temperature for cell growth;

• r represents the size of the interval of temperature in which cell growth

is not signicantly inhibited by temperature;

• µopt represents the maximum specic growth rate at optimal tempera-

ture conditions.

The advantage of using the gaussian model instead of the Arrhenius law is
related rst to a more clear signicance of the model parameters, even when
the growth model include more than one substrate. Secondly, as suggested in
[34], if we neglect the coecient µopt , the temperature eect on cell growth
is included in the model as an independent multiplicative factor that is equal
to 1 under optimal temperature conditions, while it decreases to 0 when the
environment moves away from optimal temperature conditions. Finally, the
number of parameters to e estimated is reduced from four to three.
CHAPTER 4. A NEW CELL GROWTH MODEL 79

Gas solubility
Temperature inuences the solubility of oxygen and carbon dioxide in
water and consequently their saturation concentration. We have seen in the
previous sections that the oxygen saturation concentration in water is given
by the Henry's law:
PO2 nH2 O
O2∗ = ,
HO2
where the Henry constant HO2 depends on temperature through the Van't
Ho equation: · µ ¶¸
1 1
H(T ) = H(Ts ) exp − C − .
T Ts
The constant Ts refers to the standard temperature, which is 289 K ,
T is the temperature of the uid and C is a constant that depends on the
considered gas. In particular for oxygen we have CO2 = 1700 K and for
carbon dioxide CCO2 = 2400 K . Moreover at standard temperature it results
HO2 (Ts ) = 4.259 · 104 atm and HCO2 (Ts ) = 0.163 · 104 atm.

4.5.2 pH estimation
If we want to recover an estimate of the pH in the culture medium, we
need to take into account the eects of four of the main component of the
medium:

• Ammonia N H3 : it is a base and in water it releases one hydroxyl anion

OH − :
N H3 → N H4+ + OH −

The basic dissociation constant of ammonia is KbAmm = 1.9 · 10−5 .

• Lactic acid C3 H6 O3 : it is an acid and in water it releases one hydrogen

ion H + :
Lac → Lac− + H + .

The acid dissociation constant is KaLac = 1.3 · 10−4 .

• Carbon dioxide CO2 : it reacts with water forming carbonic acid H2 CO3 .
The carbonic acid is a diprotic acid and its dissociation constants are
CHAPTER 4. A NEW CELL GROWTH MODEL 80

KaCA = 4.3 · 10−7 and Ka2 = 5.6 · 10−11 . In rst approximation, we

can neglect the second dissociation:

CO2 + H2 O → H2 CO3 → HCO3− + H + ,

• Bicarbonate HCO3− : the bicarbonate ion behaves like a base when it

is in solution with a stronger acid and it behaves like an acid when
it is in presence of a stronger base. The basic dissociation constant
of bicarbonate is KbB = 2.2 · 10−8 and the acid dissociation constant
is Ka = 5.6 · 10−11 . In the culture medium, in rst approximation,
bicarbonate behaves like a base:

HCO3− + H2 O → H2 CO3 + OH − .

If we want to calculate the pH of a solution of one acid and one base

exactly, we have to resolve a system of coupled equations that express:

• acid number of moles balance;

• base number of moles balance;

• charge balance;

• acid dissociation equilibria;

• base dissociation equilibria;

• water dissociation constant.

For example, if we consider only the stronger acid and the stronger base
that are dissolved in the culture medium, which in our case are lactic acid
and ammonia, we need to solve the following set of coupled equations:

• acid and base dissociation equilibria:



 [Amm+ ][OH − ]

 Kb Amm =
[Amm]

 [Lac ][H + ]
−

 KaLac =
[Lac]
CHAPTER 4. A NEW CELL GROWTH MODEL 81

• charge balance:

[Lac− ] + [OH − ] = [Amm+ ] + [H + ]

• acid and base moles balances:


 Ca = [Lac] + [Lac− ]
 C = [Amm] + [Amm+ ]
b

where Ca and Cb are the analytical concentration of the acid and the
base respectively;

• water dissociation constant:

Kw = [H + ][OH − ]

where Kw is the ion product of water.

The unknown quantities are the equilibrium concentrations [OH − ], [H + ],

[Lac− ], [Amm+], [Lac], [Amm]. If we manipulate the system, we obtain a
fourth degree equation for [H + ].
From this example we can understand that, even if we consider only one
acid and one base, the system that we have to solve is complex and it is not
possible to obtain the exact value of the pH. It becomes necessary to make
some approximations.
CHAPTER 4. A NEW CELL GROWTH MODEL 82

Let us now consider the set of equations that we obtain if we consider

the presence in the culture medium of both the two acids and the two basis:
 + −
 KbAmm = [Amm ][OH ]Amm



 [Amm]





 [H2 CO3 ][OH − ]B

 Kb B =

 [HCO3− ]





 [Lac− ][H + ]Lac

 Ka Lac =

 [Lac]



 [HCO − +
3 ][H ]CA

 Ka =

 CA
[H2 CO3 ]




 Cb +
Amm = [Amm] + [Amm ]
(4.36)

 −
 CbB = [H2 CO3 ] + [HCO3 ]






 CaLac = [Lac] + [Lac− ]







 CaCA = [H2 CO3 ] + [HCO3− ]





 Kw = ([H + ]Lac + [H + ]CA )([OH − ]Amm + [OH − ]B )







 (OH − )Lac + [OH − ]B + [Lac− ] + [HCO3− ] =





 = [Amm+ ] + [H2 CO3 ] + [H + ]Lac + [H + ]CA

Now we suppose that the fraction of each acid and base that dissociates
is negligible respect to the analytical concentration, e.g. [Amm] ' CbAmm .
This assumption is correct because the acids and the basis that we are con-
sidering are weak (Ka, Kb < 10−3 ).
The rst fourth equations of the system (4.36) become:
 + −
 KbAmm = [Amm ][OH ]Amm



 CbAmm





 [H2 CO3 ][OH − ]B

 Kb B =
 CbB
(4.37)

 [Lac ][H + ]Lac
−

 Ka Lac =

 CaLac



 [HCO3− ][H + ]CA

 Ka =

 CA
CaCA
CHAPTER 4. A NEW CELL GROWTH MODEL 83

Using the charge balance equations, from these relations we obtain the
following expression for the unknown concentrations [H + ] and [OH − ]:

CbAmm KbAmm CbB KbB

+ + [H + ]Lac + [H + ]CA =
[OH − ]Amm [OH − ]B
(4.38)
CaLac KaLac CaCA KaCA
= [OH − ]Amm + [OH − ]B + +
[H + ]Lac [H + ]CA

Now we can dene the total H + concentration:

[H + ] = [H + ]Lac + [H + ]CA

and the total OH − concentration:

Kw
[OH − ] = [OH − ]Amm + [OH − ]B = .
[H + ]

We obtain
CbAmm KbAmm [OH − ]B + CbB KbB [OH − ]Amm
+ [H + ] =
[OH − ]B [OH − ]Amm
(4.39)
Kw CaLac KaLac [H + ]CA + CaCA KaCA [H + ]Lac
+
[H + ] [H + ]Lac [H + ]CA

Now we suppose that the eects of lactic acid and carbonic acid on pH are
comparable, so that [H + ]Lac ' [H + ]CA ' [H + ]/2, and, similarly, that the ef-
fects of bicarbonate and ammonia are comparable, [OH − ]B ' [OH − ]Amm '
[OH − ]/2. This hypothesis is quite strong and should be tested with appro-
priate experiments. We obtain the following second degree equation for the
H + concentration:
[H + ]
2(CbAmm KbAmm + CbB KbB ) + [H + ] =
Kw
(4.40)
Kw 1
+ 2(CaLac KaLac + CaCA KaCA ) + .
[H + ] [H ]

From equation (4.40) we obtain an explicit estimate of the H + concen-

tration in the medium:
· ¸
+ 2Kw (CaLac KaLac + CaCA KaCA ) + Kw2 1/2
[H ] = , (4.41)
Kw + 2(CbAmm KbAmm + CbB KbB )

and nally the pH:

pH = − log10 [H + ]. (4.42)
CHAPTER 4. A NEW CELL GROWTH MODEL 84

As mentioned in the previous chapter, the pH inuences cell growth

and, when it becomes lower than a critical value that depends on the cell
physiology, growth is inhibited and cells start dying. In this work, however,
since we are not modeling the death phase of the growth curve, we will not
consider the pH as a factor that inuences cell growth directly, but we will
use the pH value in order to evaluate cells activity and to test our model.
Chapter 5

Numerical simulations

5.1 Preliminary remarks

In order to analyze and test the cell growth models introduced in the
previous chapter, it is necessary to dene some important properties of the
culture system. In particular we refer to the experiments carried out at
the Laboratoire de Biotechnologie Cellulaire (LBTC) of the EPFL [1, 26]
to obtain information about standard operative conditions that are usually
applied in CHO cell culture.

5.1.1 Cell culture

We consider suspension adapted CHO DG44 cells that are grown in
serum-free ProCHO5 medium; the culture have the following properties:

• the cells are incubated in an orbital shaker and their initial concentra-
tion is 3 · 105 cell/ml;

• the atmosphere above the culture medium is maintained with a con-

stant composition: the partial pressure of CO2 is 5% and the partial
pressure of oxygen is 20%;

• cell density is measured microscopically with a hemocytometer by the

trypan blue exclusion method. Glucose, glutamine, lactate and ammo-
nia concentrations are determined with a BioProle 200 Analyzer;

85
CHAPTER 5. NUMERICAL SIMULATIONS 86

• cells are cultivated in CultiFlask50 Bioreactor (Sartorius AG, Switzer-

land) with 10 ml of culture medium. The vessel is agitated at 180 rpm
with 5 cm shaking diameter;

• the shaker is placed in a warm cabinet in order to control temperature.

5.1.2 Units of measurement

During the experiments, the concentrations of cells, substrates and pro-
ducts are measured with dierent techniques and the values of concentrations
in the literature are given with dierent units of measurement. In order to
make our model more clear, all the quantity are expressed using the Inter-
national System of Units (SI). Therefore, to compare the results, we need to
introduce the correct conversion rates.

Usually the cell concentration is measured in cells per milliliter. In order

to convert the cell concentration in grams per liter, we have estimated the
average cell density knowing that the biomass density is about 1.2 kg/m3
and the cell diameter is about 15 µm [26]. Starting from these experimental
estimates, we obtain that the cell volume is 1.77 · 10−15 m3 and the mass of
one cell is about 2 · 10−15 kg. Finally, we obtain the conversion factor for the
cell concentration:
cell kg g
1 = 2 · 10−9 = 2 · 10−9 .
ml m 3 l
While glucose and lactate concentrations are measured in grams per mil-
liliter, glutamine and ammonia concentrations are measured in millimoles
per liter. To convert these values in grams per liter, it is simply necessary to
know the molar masses of these chemical species, respectively MGln = 146
g/mol and MAmm = 17 g/mol.

5.1.3 Values of parameters

In the development of the cell growth models, we have introduced some
parameters. All these parameters can be estimate with appropriate experi-
ments and several of them have already been estimated in the literature, for
example the mass transfer coecient KL a. However, whereas these parame-
ters are highly inuenced by the experimental set up, the range in which they
CHAPTER 5. NUMERICAL SIMULATIONS 87

can vary is often quite large. Consequently, for our simulations, we have se-
lected a particular operative set up and, when possible, we have chosen the
corresponding parameters values from the literature or, when these data
were not available, we estimated an average value from dierent experimen-
tal results. Finally, when appropriate experimental results were available,
we estimate the parameters values from them.

5.2 0D model
In this section we analyze the results obtained from the model under the
following conditions:

• temperature T = 37◦ C ;

• mass transfer coecient KL a = 20: this value has been estimated for
the CultiFlask50 Bioreactor with 10 ml of culture medium at 180 rpm
with 5 cm shaking diameter;

• initial cell concentration X0 = 3 · 105 cell/ml;

• initial glucose concentration Glc0 = 8.0 g/l;

• initial glutamine concentration Gln0 = 0.74 g/l;

• duration of the culture Ttot = 120 h;

• oxygen partial pressure PO2 = 20 % and carbon dioxide partial pressure

PCO2 = 5 %.

In Figure 5.1 (left) we can observe that the stationary phase is reached
after more than 4 days of culture and the maximal cell concentration is about
8.5 · 106 cell/ml. These values are comparable with the experimental results
plotted in Figure 5.1 (right), where we can observe that the maximum cell
concentration is about 9.0 · 106 cell/ml after 5 days of culture [1, 26]. The
lag phase of the real growth curve is longer than the one obtained from the
0D model, thus suggesting that an accurate model for the lag phase should
be introduced.
In Figure 5.2 we plot the results about the metabolites and the toxic
by-products concentrations. We can observe that, as expected from expe-
rimental results, glutamine is completely consumed in the rst 4 days of
CHAPTER 5. NUMERICAL SIMULATIONS 88

culture and the model reproduces quite well the real glutamine dynamics.
Moreover, as expected from experiments, glutamine is consumed faster than
glucose. However, while from experimental results we expect that glucose is
not completely consumed during the culture, from the model we obtain that
this substrate is exhausted at the end of the process.
We can observe that the maximum ammonia and lactate concentrations
are reached after 4 days of culture both in the model and in the experi-
mental results. The maximum lactate concentration in the culture medium
estimated by the model is 2.0 g/l, as expected from experimental results.
The maximum ammonia concentration that we obtain from the model is
0.07 g/l, which is equal to 4.2 mmol/l, while the maximum concentration
that is measured in the real culture is about 4.5 mmol/l [1, 26]. While the
model results about the ammonia concentration reproduce quite well the real
ammonia dynamics, the lactate formation obtained from the model is slower
than the real one.
In Figure 5.3 we plot the oxygen and the carbon dioxide concentrations.
We can observe that the oxygen is consumed during the cell growth phases
and when cell growth stops its concentration approaches the saturation value.
In the same way, the carbon dioxide is produced and accumulated in the rst
three days of culture and then it is released gradually to the atmosphere.

5.2.1 pH estimation
If we apply the equations (4.41) and (4.42) to estimate the pH in the
culture medium at T = 37◦ C , we obtain the result plotted in Figure 5.4.
If we compare the model results with the experimental results analyzed in
Section 2.4.3, we can observe that the model calculates an underestimate of
the real pH in the culture medium and the dynamics of the pH that the model
simulates is partially correct. In fact, in Figure 5.4 we can observe that the
pH diminishes quickly during the rst day of culture and later on it remains
almost constant. However, the model does not reproduce the increment of
the pH in the last hours of the process, as we can observe from experimental
results (Figure 2.9) [1, 26]. The dierences between the experimental results
and the model estimation could be probably related to the approximations
that we have introduced in order to obtain a simple estimation of the pH
from the compounds concentrations.
CHAPTER 5. NUMERICAL SIMULATIONS 89

10

9
Cell concentration [106 cell/ml]

0
0 20 40 60 80 100 120
Time [h]

Figure 5.1: Cell growth: comparison between model results and experi-
mental results (points) [1]
9 0.8

8
0.7
Glutamine concentration [g/l]

7
Glucose concentration [g/l]

0.6
6
0.5
5
0.4
4
0.3
3
0.2
2

0.1
1

0 0
0 20 40 60 80 100 120 0 20 40 60 80 100 120
Time [h] Time [h]

2.5 0.08

0.07

2
Ammonia concentration [g/l]
Lactate concentration [g/l]

0.06

0.05
1.5

0.04

1
0.03

0.02
0.5

0.01

0 0
0 20 40 60 80 100 120 0 20 40 60 80 100 120
Time [h] Time [h]

Figure 5.2: Cell metabolism: comparison between model results and experi-
mental results (points) [1]
CHAPTER 5. NUMERICAL SIMULATIONS 90

0.01 0.065

Carbon dioxide concentration [g/l]

Oxygen concentration [g/l]

0.008

0.06
0.006

0.004
0.055

0.002

0 0.05
0 20 40 60 80 100 120 0 20 40 60 80 100 120
Time [h] Time [h]

Figure 5.3: Cell metabolism: oxygen (left) and carbon dioxide (right) con-
centration

7.2

6.8
pH

6.6

6.4

6.2

6
0 20 40 60 80 100 120
Time [h]

Figure 5.4: pH estimation from the model results

5.2.2 Temperature dependance

In this section we analyze the behavior of our model as a function of tem-
perature. In order to estimate the maximum cell growth rate as a function of
temperature, we need to measure cell concentration under non-limiting nu-
trients conditions. Under this hypothesis, in fact, the multiplicative factor
related to the substrate concentrations is equal to 1 and the balance equation
for the cells is uncoupled from the balance equations for the substrates. We
obtain:
dX
= µmax X, (5.1)
dt
where µ ¶
(T − Topt )2
µmax = µopt exp − .
r2
CHAPTER 5. NUMERICAL SIMULATIONS 91

The equation (5.1) can be solved analytically and we obtain:

µ · ¸¶
(T − Topt )2
X(t) = X0 exp(µmax t) = X0 exp µopt exp − t ,
r2

where X0 is the initial cell concentration. If we know the cell concentration

as a function of time t and temperature T under non-limiting nutrients con-
ditions, we can estimate the model parameters µopt , Topt and r.

Whereas we don't have this kind of experimental results, we have to

proceed in a dierent way: we estimate the model parameters µopt , Topt and
r from the value of µmax estimated in [1] with the Gompertz model.
The maximum specic growth rate as a function of temperature that
we obtain with this strategy is plotted in Figure 5.5, while the values of
the parameters are showed in Table 5.1. For easy of reading, the values of
temperature are expressed in Celsius degrees.
If we apply the gaussian model for temperature to the 0D model for cell
growth, we obtain the results plotted in Figure 5.6 and in Figure 5.7. It is
possible to observe that temperature inuences both the maximum specic
cell growth rate, the lag phase duration and the maximum cell concentration,
as expected from experimental results (see section 2.4.1).
We can observe that when T = 37◦ C we obtain the maximum cell con-
centration. When T = 35◦ C the lag phase is longer and the maximum cell
concentration at the end of the process is a bit lower. The nal cell con-
centration decreases further when temperature increases up to T = 39◦ C .
When temperature becomes lower than 35◦ C and higher than 39◦ C , cell
growth is highly inhibited, and for T = 31◦ C and T = 41◦ C it is nearly zero,
as expected from experimental results (Figure 5.6).
Finally, in Figure 5.7 we can observe that the substrates and the toxic
by-products concentrations at dierent temperatures essentially reect cell
dynamics.

5.2.3 KL a dependance

In this section we want to test the model at dierent values of the mass
transfer parameters KL a, with T = 37◦ C . The results are plotted in Figure
5.8 and in Figure 5.9.
CHAPTER 5. NUMERICAL SIMULATIONS 92

Parameter Value
Topt [◦ C] 36.4
µopt [1/h] 1.882
r [∆◦ C] 3.12

Table 5.1: Parameters of the Gaussian model

1.5
µmax [1/h]

0.5

31 33 35 37 39 41
Temperature [°C]

Figure 5.5: Maximum specic growth rate as a function of temperature

CHAPTER 5. NUMERICAL SIMULATIONS 93

10
T = 31°C
9
T = 33°C
8 T = 35°C
Cell concentration [106 cell/ml]

T = 37°C
7 T = 39°C
T = 41°C
6

0
0 20 40 60 80 100 120
Time [h]

Figure 5.6: Cell growth as a function of temperature: model results (left)

and experimental results (right)
9
0.8
8
0.7
Glutamine concentration [g/ml]

7
Glucose concentration [g/ml]

0.6
6
0.5
5
0.4
4
0.3
3
T = 31°C T = 31°C
2 T = 33°C 0.2
T = 33°C
T = 35°C T = 35°C
1 T = 37°C 0.1
T = 37°C
T = 39°C T = 39°C
0 T = 41°C 0
T = 41°C
−1 −0.1
0 20 40 60 80 100 120 0 20 40 60 80 100 120
Time [h] Time [h]

2.5 0.08
T = 31°C T = 31°C
T = 33°C 0.07 T = 33°C
2 T = 35°C T = 35°C
Ammonia concentration [g/ml]

T = 37°C 0.06 T = 37°C

Lactate concentration [g/ml]

T = 39°C T = 39°C
T = 41°C 0.05 T = 41°C
1.5

0.04

1 0.03

0.02
0.5
0.01

0 0

−0.01
0 20 40 60 80 100 120 0 20 40 60 80 100 120
Time [h] Time [h]

Figure 5.7: Cell growth model: results for dierent temperatures

CHAPTER 5. NUMERICAL SIMULATIONS 94

We can observe that, as expected from experimental results [1], when

KL a is equal to 0, that is when there is not exchange of oxygen and carbon
dioxide between the culture medium and the atmosphere, the oxygen in the
culture medium is consumed in a few hours. Consequently, there is no pro-
duction of ammonia and consumption of glutamine, that is metabolized by
cells only under aerobic conditions. At the same time, carbon dioxide pro-
duced in the rst hours of the culture is not released to the atmosphere and it
causes a remarkable reduction of the pH in the culture medium (Figure 5.10).
Finally, whereas cells recover energy primarily from aerobic respiration, the
maximum cell concentration that is reached when KL a = 0 is signicantly
lower respect to the concentration that is reached when the KL a becomes
higher. From these observation we can infer why one of the aim of biore-
actor design is the increment of the coecient KL a of the culture system.
An increment of the KL a, in fact, causes the increment of the maximum cell
concentration reached during the culture period and helps to maintain the
pH above toxic values.
CHAPTER 5. NUMERICAL SIMULATIONS 95

8
KLa = 0
7 K a=5
L

Cell concentration [106 cell/ml]

KLa = 10
6

0
0 20 40 60 80 100 120
Time [h]

Figure 5.8: Cell growth as a function of the mass transfer coecient

KL a: model results
9 0.9

8 0.8
Glutamine concentration [g/ml]
Glucose concentration [g/ml]

7 0.7

6 0.6

5 0.5

4 0.4

3 0.3
KLa = 0 KLa = 0
2 0.2
KLa = 5 KLa = 5
1 K a = 10 0.1 K a = 10
L L

0 0
0 20 40 60 80 100 120 0 20 40 60 80 100 120
Time [h] Time [h]

1.8 0.08
KLa = 0 KLa = 0
1.6 0.07
KLa = 5 K a=5
L
KLa = 10
Ammonia concentration [g/ml]

1.4 KLa = 10 0.06

Lactate concentration [g/ml]

1.2 0.05

1
0.04
0.8
0.03
0.6
0.02
0.4
0.01
0.2
0
0
−0.01
0 20 40 60 80 100 120 0 20 40 60 80 100 120
Time [h] Time [h]

−3
x 10
0.07
KLa = 0 KLa = 0
9 KLa = 5 KLa = 5
0.068
Carbon dioxide concentration [g/ml]

8 KLa = 10 KLa = 10
Oxygen concentration [g/ml]

0.066
7

6 0.064

5 0.062
4
0.06
3
0.058
2

1 0.056

0
0.054
−1
0 20 40 60 80 100 120 0 20 40 60 80 100 120
Time [h] Time [h]

Figure 5.9: Model results as a function of the mass transfer coecient KL a

CHAPTER 5. NUMERICAL SIMULATIONS 96

7.5
KLa = 0
KLa = 5
KLa = 10
7

6.5
pH

5.5

5
0 20 40 60 80 100 120
Time [h]

Figure 5.10: pH as a function of the mass transfer coecient KL a: model

results

5.3 1D model
In this section we test the 1D model and in particular we compare its
results with the 0D model results. Whereas in the 1D model we are not
considering the uid motion and, consequently, the mass transfer is deter-
mined only by the boundary conditions and the diusivity of each substance
in water, we need to set the correct value of the mass transfer parameter
KL a in the 0D model in order to make a signicant comparison.
From experimental results [26] we know that when the uid in the biore-
actor is steady, the mass transfer coecient is usually lower than 2 and de-
pends only on the bioreactor geometry and the gas diusivity. In Figure 5.12
and in Figure 5.13 we plot the results obtained with KLO2 a = KLCO2 = 0.7
and T = 37◦ C . Whereas the CultiFlask50 Bioreactor (Figure 5.11) that we
are considering can be approximated with a cylinder with diameter equal
to 30 mm and height equal to 115 mm, we obtain the desired culture vol-
ume of 10 ml if we suppose that the uid in the bioreactor is about 3 cm deep.

One of the main diculties in the comparison of the 1D model and the
0D model is related to the meaning of the values of concentration obtained
with the 0D model and the measurement techniques adopted during the
experiments.
CHAPTER 5. NUMERICAL SIMULATIONS 97

Figure 5.11: CultiFlask50 Bioreactor

As explained in [26], the measurement of the concentration of the com-

pounds in the culture medium is done in dierent ways. For example, the
cell concentration is usually estimated by sampling the culture medium only
on the surface of the liquid, while the oxygen concentration is estimated by
sampling on the bottom of the bioreactor. Moreover, up to our knowledge,
in the literature there are neither results about the spatial distribution of
the concentrations in the bioreactor, nor results about the average concen-
trations in the volume. Therefore, it is not easy to make a comparison both
between the 1D model and the 0D model and between the 1D model and
the experimental results.
Consequently, we decide to compare the concentrations obtained from
the 0D model as they were an averaged value in the bioreactor with the av-
eraged concentrations obtained with the 1D model. Moreover, we consider
also the minimum and the maximum values as reference concentrations.

In Figure 5.12 we can observe that the averaged cell concentration that
we obtain by solving the 1D model is almost equal to the cell concentration
that is estimated with the 0D model when we assume that the uid is sta-
tionary. However we can see that the maximum cell concentration in the
domain is much larger than the minimum cell concentration: in particular
the cell concentration is quite uniform in the domain except in proximity of
the surface (Figure 5.14), where the maximum cell concentration is reached.
This behavior is related to the fast consumption of the oxygen by the cells
and its low solubility and diusivity in the culture medium. Whereas the
CHAPTER 5. NUMERICAL SIMULATIONS 98

uid is steady, the liquid that is far away from the surface is not oxygenated
(Figure 5.15) and cell growth is thus inhibited. In conclusion cell growth
is signicant only in proximity of the interface between the culture medium
and the surface (Figure 5.14). As a consequence, glucose and glutamine
are mainly consumed near the surface, as well as lactate and ammonia are
mainly produced in that zone of the domain (Figure 5.15).

In Figure 5.13 we can observe that, while there is correspondence between

the averaged oxygen concentration in the 1D domain and the oxygen con-
centration obtained from the 0D model, the concentrations of carbon dioxide
obtained with the two models are quite dierent and in particular the carbon
dioxide estimated with the 0D model is lower than the one obtained with the
1D model. This behavior can be probably explained with the choice of the
same KL a value for both oxygen and carbon dioxide. As observed in [26],
in fact, the KL a value for the carbon dioxide is usually lower than the value
for the oxygen, as seemingly predicted from the 1D model. We will focus on
this problem in the next section.
CHAPTER 5. NUMERICAL SIMULATIONS 99

10
Min 1D
9 Max 1D
Mean 1D
8 0D

Cell concentration [106 cell/ml]

7

0
0 20 40 60 80 100 120
Time [h]

Figure 5.12: Cell growth: comparison between the 1D model and the 0D
model
10 1
Min 1D Min 1D
9 Max 1D 0.9 Max 1D
Mean 1D Mean 1D
8 0D 0.8 0D
Glutamine concentration [g/l]
Glucose concentration [g/l]

7 0.7

6 0.6

5 0.5

4 0.4

3 0.3

2 0.2

1 0.1

0 0
0 20 40 60 80 100 120 0 20 40 60 80 100 120
Time [h] Time [h]

2
Min 1D Min 1D
1.8 0.025
Max 1D Max 1D
1.6 Mean 1D Mean 1D
Ammonia concentration [g/l]

0D 0D
Lactate concentration [g/l]

1.4 0.02

1.2
0.015
1

0.8
0.01
0.6

0.4 0.005

0.2

0 0

0 20 40 60 80 100 120 0 20 40 60 80 100 120

Time [h] Time [h]

−3
x 10
Min 1D Min 1D
0.08
9 Max 1D Max 1D
Mean 1D Mean 1D
8 0D 0D
0.075
Oxygen concentration [g/l]

7
CO2 concentration [g/l]

6 0.07
5

4 0.065

3
0.06
2

1
0.055
0

−1 0.05
0 20 40 60 80 100 120 0 20 40 60 80 100 120
Time [h] time [h]

Figure 5.13: Comparison between the 1D model and the 0D model

CHAPTER 5. NUMERICAL SIMULATIONS 100

30
t = 24 h
t = 48 h
25 t = 120 h
Cell concentration [106 cell/ml]

20

15

10

0 0.5 1 1.5 2 2.5 3

z [cm]
centering

Figure 5.14: 1D cell growth model: spatial dependance

11 1
t = 24 h t = 24 h
10 t = 48 h 0.9 t = 48 h
t = 120 h t = 120 h
9 0.8
Glutamine concentration [g/l]
Glucose concentration [g/l]

8
0.7
7
0.6
6
0.5
5
0.4
4
0.3
3

2 0.2

1 0.1

0 0
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3
z [cm] z [cm]

t = 24 h 0.025 t = 24 h
1.8 t = 48 h t = 48 h
t = 120 h t = 120 h
1.6
0.02
Ammonia concentration [g/l]
Lactate concentration [g/l]

1.4

1.2
0.015
1

0.8
0.01
0.6

0.4
0.005
0.2

0 0
−0.2
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3
z [cm] z [cm]

−3
x 10
8
t = 24 h t = 24 h
7 t = 48 h 0.08 t = 48 h
t = 120 h t = 120 h
6 0.075
Oxygen concentration [g/l]

CO2 concentration [g/l]

5
0.07
4

3 0.065

2
0.06
1
0.055
0

−1 0.05
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3
z [cm] z [cm]

Figure 5.15: 1D model: spatial dependance

CHAPTER 5. NUMERICAL SIMULATIONS 101

5.4 3D model
In this section we test the 3D model and we compare its results with the
ones obtained with the 0D model. However we have to take into account
that we are not considering the real geometry and the real hydrodynamics
of the shaken bioreactor. Consequently, with these simulations we do not
have the aim to reproduce the real process, but we want only to test the
qualitative eects of transport on the global growth process.
We select the domain Ω = [−1, 1]3 [dm] and the following simple velocity
eld (Figure 5.16):
 
0
 µ ¶ µ ¶ 
 πy πz 
u= −A cos sin . (5.2)

 µ 2¶ µ 2¶ 

πy πz
A sin cos
2 2

with A ∈ R [m/s].

Figure 5.16: Velocity eld (5.2)

As in the previous simulations, we don't know exactly the value of the

KL a coecient related to the geometry and the hydrodynamics that we are
considering, however it is possible to make some signicant comparisons. As
in the previous section, the results obtained with the 0D model are compared
with the averaged values of the concentrations in the domain Ω, as well as
CHAPTER 5. NUMERICAL SIMULATIONS 102

with the minimum and the maximum values.

In Figure 5.17 we compare the results obtained from the 3D model and
from the 0D model in the case of steady uid. We can observe that the
averaged cell concentration obtained with the 3D model with the considered
operative setting corresponds to the one obtained from the 0D model with
KL a ' 5. In Figure 5.18 we can observe that the correspondence between
the 3D model and the 0D model is quite good except for the carbon diox-
ide concentration: the values obtained with the 3D model are greater than
the values obtained with the 0D model with KL a = 5 for both the oxy-
gen and the carbon dioxide, as previously observed also in the 1D model
simulations. This result conrms the experimental observation that usually
the mass transfer coecient for the carbon dioxide is lower than the one for
the oxygen [26]. We can observe that, while the substances concentrations,
except for the oxygen and the carbon dioxide concentrations, are almost
uniform in the culture medium, since the minimum and the maximum val-
ues are almost equal, the cell concentration varies quite a lot in the volume
(Figure 5.25), as already observed for the 1D model. The maximum cell
concentration that is reached after 120 hours is nearly the double than the
minimum concentration. This behavior is related primarily to the dierent
oxygen concentration in the culture volume: in the lower part of the domain
the oxygen is soon consumed by the cells and consequently cell growth is
inhibited (Figure 5.25). These results conrm what previously discussed in
Section 3.5.2.
In Figure 5.19 we plot the results obtained with the 3D model when the
transport eld (5.2) with A = 0.003 m/s is applied. We can observe that
the cell concentration obtained in the considered operative setting can be
obtained from the 0D model with KL a ' 10. This is conrmed also by the
results showed in Figure 5.20. We can observe that, thanks to the uid mo-
tion, the concentration of cells in the volume is almost uniform.

In Figure 5.21 we plot the results obtained with the 3D model when the
transport eld (5.2) with A = 0.03 m/s is applied. We can observe that the
concentrations obtained with this transport eld can be obtained from the
0D model with KL a ' 23. We can observe that the maximum cell concen-
CHAPTER 5. NUMERICAL SIMULATIONS 103

tration that is reached at the end of the growth phase is greater that the
ones obtained in the previous cases. We can observe that also in this case
the 0D model seems to underestimate the carbon dioxide concentration if we
impose that the oxygen and the carbon dioxide have the same KL a value.
From experimental results it is known that the mass transfer rate of CO2
is about 10%−20% lower than the oxygen transfer rate. If we impose that
KLCO2 a = 0.8 · KLO2 a, we obtain the results plotted in Figure 5.23 and in
Figure 5.24. We can observe that the carbon dioxide concentration obtained
with the 3D model and the 0D model are now comparable, thus conrming
what expected from experimental results.

In Figure 5.25 we compare the spatial distribution of cells, oxygen and

carbon dioxide in the culture medium at the end of the process in the case
of steady uid (left) and with the eect of the transport eld (right). In
particular we plot the results in one section yz . We can observe that the
oxygen and the carbon dioxide concentrations are highly aected by the uid
motion, that helps to oxygenate the lower part of the bioreactor and to make
the cell concentration uniform in the volume.
In Figure 5.26 and 5.27 we plot the concentrations of oxygen and car-
bon dioxide respectively at four dierent time when the transport eld with
A = 0.03 m/s is applied. We can observe that, while the cell concentration
remains uniform in the volume (Figure 5.23), the oxygen, starting from a
uniform distribution, is consumed and almost exhausted only in the central
part of the volume when cell growth is faster. Later on, when cell growth
reaches the stationary phase, the oxygen concentration starts increasing. An
analogous behavior is observed for the carbon dioxide concentration.
CHAPTER 5. NUMERICAL SIMULATIONS 104

9
Min 3D
8

Cell concentration [106 cell/ml]

Max 3D
7 Mean 3D
0D K a = 5
L
6

0
0 20 40 60 80 100 120
Time [h]

Figure 5.17: Cell growth: comparison between the 3D model and the 0D
model with the steady uid
9 0.9
Min 3D
8 0.8 Max 3D
Glutamine concentration [g/l]

Mean 3D
Glucose concentration [g/l]

7 0.7 0D KLa = 5

6 0.6

5 0.5

4 0.4

3 Min 3D 0.3

2 Max 3D 0.2
Mean 3D
1 0D K a = 5 0.1
L

0 0
0 20 40 60 80 100 120 0 20 40 60 80 100 120
Time [h] Time [h]

0.07
0.4 Min 3D Min 3D
Max 3D 0.06 Max 3D
Ammonia concentration [g/l]
Lactate concentration [g/l]

0.35 Mean 3D Mean 3D

0D KLa = 5 0.05 0D K a = 5
0.3 L

0.25 0.04
0.2
0.03
0.15
0.02
0.1
0.05 0.01
0
0
−0.05
0 20 40 60 80 100 120 0 20 40 60 80 100 120
Time [h] Time [h]
−3
x 10
8
Min 3D Max 3D
7 0.075
Mean 3D Mean 3D
Oxygen concentration [g/l]

6 0D K a = 5 0D K a = 5
CO2 concentration [g/l]

L L
0.07
5

4
0.065
3

2
0.06
1

0 0.055

−1
0 20 40 60 80 100 120 0 20 40 60 80 100 120
Time [h] Time [h]

Figure 5.18: Comparison between the 3D model and the 0D model with the
steady uid
CHAPTER 5. NUMERICAL SIMULATIONS 105

9
Min 3D
8

Cell concentration [106 cell/ml]

Max 3D
7 Mean 3D
0D K a = 10
L
6

0
0 20 40 60 80 100 120
Time [h]

Figure 5.19: Cell growth: comparison between the 3D model and the 0D
model (velocity eld (5.2) with A = 0.003 m/s)
9 0.9
Min 3D Min 3D
8 Max 3D 0.8 Max 3D
Glutamine concentration [g/l]

Mean 3D Mean 3D
Glucose concentration [g/l]

7 0D KLa = 10 0.7 0D KLa = 10

6 0.6

5 0.5

4 0.4

3 0.3

2 0.2

1 0.1

0 0
0 20 40 60 80 100 120 0 20 40 60 80 100 120
Time [h] Time [h]

0.5
Min 3D 0.08 Min 3D
Max 3D Max 3D
Ammonia concentration [g/l]

0.4
Lactate concentration [g/l]

Mean 3D 0.07 Mean 3D

0D KLa = 10 0.06 0D KLa = 10
0.3
0.05
0.04
0.2
0.03
0.02
0.1
0.01

0 0
−0.01
0 20 40 60 80 100 120 0 20 40 60 80 100 120
Time [h] Time [h]
−3
x 10
7 0.075
Min 3D
Mean 3D Max 3D
6 Mean 3D
0D KLa = 10
Oxygen concentration [g/l]

0.07 0D KLa = 10
CO2 concentration [g/l]

4
0.065
3

2 0.06

1
0.055
0

0 20 40 60 80 100 120 0 20 40 60 80 100 120

Time [h] Time [h]

Figure 5.20: Comparison between the 3D model and the 0D model (velocity
eld (5.2) with A = 0.003 m/s)
CHAPTER 5. NUMERICAL SIMULATIONS 106

9
Min 3D
8

Cell concentration [106 cell/ml]

Max 3D
7 Mean 3D
0D KLa = 23
6

0
0 20 40 60 80 100 120
Time [h]

Figure 5.21: Cell growth: comparison between the 3D model and the 0D
model (velocity eld (5.2) with A = 0.03 m/s)
9 0.9
Min 3D Min 3D
8 Max 3D 0.8 Max 3D
Glutamine concentration [g/l]

Mean 3D Mean 3D
Glucose concentration [g/l]

7 0D KLa = 23 0.7 0D KLa = 23

6 0.6

5 0.5

4 0.4

3 0.3

2 0.2

1 0.1

0 0
0 20 40 60 80 100 120 0 20 40 60 80 100 120
Time [h] Time [h]

0.45
Min 3D Min 3D
0.4 0.08
Max 3D Max 3D
Ammonia concentration [g/l]

Mean 3D
Lactate concentration [g/l]

0.35 Mean 3D 0.07

0D KLa = 23 0D K a = 23
L
0.3 0.06

0.25 0.05

0.2 0.04

0.15 0.03
0.02
0.1
0.01
0.05
0
0
−0.01
0 20 40 60 80 100 120 0 20 40 60 80 100 120
Time [h] Time [h]
−3
x 10
10
Min 3D Max 3D
0.075
Mean 3D Mean 3D
0D K a = 23 0D KLa = 23
Oxygen concentration [g/l]

8 L
CO2 concentration [g/l]

0.07

6
0.065
4

0.06
2

0.055
0
0 20 40 60 80 100 120 0 20 40 60 80 100 120
Time [h] Time [h]

Figure 5.22: Comparison between the 3D model and the 0D model (velocity
eld (5.2) with A = 0.03 m/s)
CHAPTER 5. NUMERICAL SIMULATIONS 107

10
Min 3D
Max 3D

Cell concentration [106 cell/ml]

8 Mean 3D
0D

0
0 20 40 60 80 100 120
Time [h]

Figure 5.23: Cell growth: comparison between the 3D model and the 0D
model (velocity eld (5.2) with A = 0.03 m/s) and KLCO2 a = 0.8 · KLO2 a

9 0.8
Min 3D Min 3D
8 Max 3D 0.7 Max 3D
Glutamine concentration [g/l]

Mean 3D Mean 3D
Glucose concentration [g/l]

7 0D 0D
0.6
6
0.5
5
0.4
4
0.3
3
0.2
2

1 0.1

0 0
0 20 40 60 80 100 120 0 20 40 60 80 100 120
Time [h] Time [h]

0.45 0.09
Min 3D Min 3D
0.4 Max 3D 0.08
Max 3D
Ammonia concentration [g/l]
Lactate concentration [g/l]

0.35 Mean 3D 0.07 Mean 3D

0D 0D
0.3 0.06

0.25 0.05

0.2 0.04

0.15 0.03

0.1 0.02

0.05 0.01

0 0
0 20 40 60 80 100 120 0 20 40 60 80 100 120
Time [h] Time [h]
−3
x 10
0.075
Min 3D Max 3D
8 Mean 3D Mean 3D
0D KLa O2 = 23 0D KLa(CO2) = 0.8 KLa(O2)
Oxygen concentration [g/l]

7 0.07
CO2 concentration [g/l]

5 0.065

3 0.06
2

1 0.055
0
0 20 40 60 80 100 120 0 20 40 60 80 100 120
Time [h] time [h]

Figure 5.24: Comparison between the 3D model and the 0D model (velocity
eld (5.2) with A = 0.03 m/s) and KLCO2 a = 0.8 · KLO2 a
CHAPTER 5. NUMERICAL SIMULATIONS 108

Figure 5.25: 3D model results: comparison between the steady uid (left)
and the transport eld (5.2) (right) at t = 120 h
CHAPTER 5. NUMERICAL SIMULATIONS 109

Figure 5.26: 3D model results: oxygen concentration at t = 0 h, t = 40 h,

t = 80 h, t = 120 h with the transport eld (5.2) with A = 0.03 m/s
CHAPTER 5. NUMERICAL SIMULATIONS 110

Figure 5.27: 3D model results: carbon dioxide concentration at t = 0 h,

t = 40 h, t = 80 h, t = 120 h with the transport eld (5.2) with A = 0.03
m/s
Closing remarks and future
work

This work has been focused on the development of a new model for CHO
cell growth under batch conditions, primarily based on the study of the cell
metabolism.
In the rst part, we have analyzed in a critical manner the state of the
art on cell growth models, with the objective of detecting the advantages
and the limits of the dierent approaches adopted in the literature.

We studied the main aspects of cell physiology, focusing in particular on

cell metabolism and analyzing the eects of dierent environmental factors,
like temperature, oxygen concentration and pH, on the cell behavior.
Then we analyzed the CHO cell culture system, focusing in particular on
the relationship between the mode of the bioreactor operation and the pro-
perties of the culture environment. Whereas in the framework of the project
carried out at LBTC and CMCS CHO cell growth is studied in shaken biore-
actors under batch conditions, we introduced an appropriate model for the
batch fermenter and in particular for the oxygen and carbon dioxide transfer
between the culture medium and the atmosphere.

We proposed a 0D model for cell growth, in which the rate of cell pro-
duction is based on the analysis of the cell metabolism. The knowledge of
the metabolic reactions helps to couple the balance equation that describes
the cell dynamics with the ones that describe the substrates consumption
and the toxic by-products formation. In particular we considered glucose,
glutamine and oxygen as main energy sources and lactate, ammonia and

111
carbon dioxide as main metabolic products.
We introduced the temperature as a independent factor that inuences
cell growth. We proposed a gaussian model such that, under optimal tem-
perature conditions, we obtain the maximum growth rate that is possible
under the actual substrates concentrations, whereas cell growth is inhibited
when temperature exceeds the normal physiological range.
The 0D model is then extended in order to introduce the spatial depen-
dance in the cell growth model. The 1D model that we proposed allows us to
estimate the eect on cell growth dynamics of the distance from the culture
medium surface.
Finally, we introduced a 3D model in which cell growth is aected by
the uid motion. However, here we don't consider the real hydrodynamics
of the shaken bioreactor, but we only analyze the qualitative eects of the
uid transport on cell growth. The integration of the new cell growth model
with the real free-surface hydrodynamics is currently under development at
CMCS.

The models introduced were tested and the results obtained were quali-
tatively compared with the experimental results. It is necessary to highlight,
however, that further experimental measurement will be required for a full
validation of the model. The parameters values have been selected, when it
was possible, in the ranges of values proposed in the literature. Despite that,
the results obtained with the proposed models are encouraging. The die-
rent scale models are currently being tested by LBTC in order to perform a
calibration based on additional experiments.
Firstly, the 0D model reproduces the three main phases of the growth
curve, namely the lag phase, the exponential phase and the stationary phase.
Moreover, the maximum cell concentration that is reached is a quite good
estimation of the real concentration obtained from experiments under the
same operative conditions. The estimated concentrations of ammonia and
lactate are comparable with the ones measured during experiments. More-
over the 0D model seems to reproduce quite well the real cell dynamics under
dierent conditions of temperature and mass transfer rate.

Later on, the 1D model and the 3D model were tested. Whereas, up

112
to our knowledge, in the literature there are not results about the spatial
dependance of cell concentration, the results obtained with these models
were compared only with the 0D model's results.
Analyzing the results of the 1D model, we observed that, because of
the low oxygen solubility and the absence of mixing related to the uid
motion, the oxygen is rapidly consumed and the medium that is far away
from the liquid surface is badly oxygenated. Consequently cell growth is
inhibited except in the proximity of the surface, where the maximum cell
concentration is reached.
The same results were obtained with the 3D model in the case of steady
uid. On the contrary, if a transport eld is imposed, the compounds mixing
is enhanced and consequently the concentrations become almost constant in
the bioreactor volume. The uid motion helps to oxygenate the lower part of
the culture medium, thus enhancing cell growth and permitting to reach an
averaged cell concentration that is greater than in the case of steady uid.
Moreover, the averaged concentrations obtained with the 3D model are com-
parable with the ones obtained with the 0D model under the same operative
conditions and with the corresponding mass transfer coecient.

The results obtained with the 3D model show that, as expected, if the
uid is moving, the concentrations of cells and metabolic compounds tend
to become homogeneous in the bioreactor volume, and that the averaged
values in the volume are close to those obtained with the 0D model. We
could use the 3D model to simulate and optimize the operating setting, e.g.
velocity and geometry, in order to maximize the mixing and to make the
concentrations as much homogeneous as possible in the culture medium. If
the operative setting of the bioreactor is optimized, than we can proceed by
estimating the corresponding mass transfer coecient and by simulating the
cell growth process using the less expensive 0D model. However we need to
highlight that the current 3D cell growth model is not considering the eect
of shear stresses on cells viability. In shaken bioreactors, in fact, well-mixed
conditions are generally obtained with turbulent regimes of motion, which
are characterized by high shear-stress. As a consequence better results would
probably be obtained by solving the full 3D cell growth model with the real
free-surface hydrodynamics introducing the eect of shear-stress on cells vi-

113
ability.

In conclusion, we want to highlight that an accurate validation of the

models is necessary. In order to do that, appropriate experimental results
are needed in order to estimate the eect of each considered factor on cell
growth. Wherear the factors that inuence cell growth in our model are
independent the one from the others, each of them should be studied when
all the other factors are not limiting. In particular it is necessary to estimate
the eect of aerobic glucose oxidation, anaerobic glucose consumption and
glutamine aerobic oxidation on the cell growth rate. Moreover, as mentioned
in the previous chapter, the eect of dierent temperatures should be studied
under non limiting-nutrient conditions. Finally, further improvements can
also be envisaged thanks to the coupling of the cell growth model with the
real bioreactor hydrodynamics, that is now being implemented at CMCS.

114
Ringraziamenti

Sono giunta nalmente a scrivere la pagina più importante, una pagina

dedicata a tutte le persone che in questi anni mi hanno aiutato a raggiungere
questo importante traguardo.

Il primo e sincero ringraziamento va al Dott. Nicola Parolini, per avermi

seguito con pazienza e disponibilità in questi mesi.
Desidero ringraziare il Prof. Alo Quarteroni ed il Prof. Marco Discacciati,
che mi hanno ospitato presso il CMCS dell'EPFL di Losanna e che mi hanno
dato in questo modo la possibilità di lavorare in un ambiente molto sti-
molante. In particolare ringrazio Samuel, che ha condiviso con me, oltre
all'ucio, i "segreti" di LifeV. Un Grazie aettuoso a Matteo e Laura, per
avermi accolto e per avermi incoraggiato durante questa trasferta.
Ringrazio il Laboratorio di Biotecnologie Cellulari (LBTC) dell'EPFL ed
in particolare la Dottoressa Stéphanie Tissot, che con pazienza ha risposto
alle mie numerose e-mail ed ha chiarito ogni mio dubbio sui processi di
crescita cellulare.

Ringrazio tutti gli Ingegneri Matematici che hanno condiviso con me

questo percorso, in particolare Grazie a Laura e Claudia, con le quali ho
arontato e superato gli ostacoli di questi anni.
Grazie a Giovanni, che non ho ancora conosciuto di persona, ma che ha
fatto resuscitare il mio portatile quando avevo quasi perso le speranze.
Grazie a Matteo, che mi ha aiutato a installare LifeV prima della mia
partenza, e che anche a distanza è sempre stato disponibile per aiutarmi.

115
 Il Grazie più grande è per i miei genitori, che da sempre hanno creduto
in me, mi hanno sostenuta nei momenti di dicoltà e consolata nei momenti
di sconforto, come solo una mamma e un papà sanno fare. Grazie al mio
fratellino (ma se dico fratellone forse si capisce di più...) per aver sopportato
i miei schizzi di pazzia.

Un Grazie speciale alle mie amiche di sempre, Alessandra, Elisa e Paola,

per i nostri "Martedì Sera", per le "Corse al Regalo" e per gli "Appunta-
menti Disneyani", quelli passati e ovviamente per quelli futuri.

Per concludere, Grazie Leonardo. Per ogni istante condiviso e per il tuo
Amore. Il Politecnico ha voluto che anche il giorno della Laurea rientrasse
tra i giorni condivisi! Come hai detto tu, un altro Segno...

116
Bibliography

[1] Broccard G., "Development of a predictive model for cell growth in

shaken bioreactors as a function of the eects of temperature, oxygen
transfer rate and seeding concentration", Master Thesis, LBTC/EPFL,
2009

[2] Buchanan R.L., Whiting R.C., "A classication of models for predictive
microbiology", Food Microbiol., 10: 175-177, 1993

[3] Burman E., Hansbo P., "Edge stabilization for Galerkin approximations
of convection-diusion-reaction problems", Comput. Methods Appl.
Mech. Engrg., 193: 1437-1453, 2004

[4] De Sanctis D., Perrone M., "Numerical analysis of uid dynamics and
oxygen diusion in shaking bioreactors", Master Thesis, Università degli
Studi di Roma "Tor Vergata", 2007

[5] Dunn I.J., Heinzle E., Ingham J., Prenosil J.E., "Biological reaction
engineering", Wiley-VCH Verlag, 2003

[6] Fourestey G., Deparis S., Formaggia L., Gerbeau J.F., Prud'homme C.,
"LifeV developer manual", http://www.lifev.org, 2010

[7] Furukawa K., Ohsuye K., "Eect of culture temperature on a recombi-

nant CHO cell line producing a C-terminal α-amidating enzyme", Cy-
totechnology, 26: 153-164, 1998

[8] Gompertz B., "On the nature of the function expressive of the law of
human mortality, and on a new mode of determining the value of life
contingencies", Philos. Trans. R. Soc. Lond., 115: 513-585, 1825

117
 [9] Hauser H., Wagner R., "Mammalian cell biotechnology in protein pro-
duction", W. de Gruyter, 1997

[10] Herbert D., "Continuous culture of microorganisms; some theoretical as-

pects", in Malek I., "Continuous cultivation of microorganisms; a sym-
posium", pp. 45-52, Czech Academy of Science, 1958

[11] Hinshelwood C.N., "Inuence of temperature on the growth of bacteria",

in "The chemical kinetics of the bacterial cell", Clarendon press, 254-
257, 1946

[12] Jayapal K.P., Wlaschin K.F., Hu W.S., Yap M.G.S., "Recombinant pro-
tein therapeutics from CHO cells - 20 years and counting", Chem. Eng.
Prog., 103: 40-47, 2007

[13] Kovarova K., Zehnder A.J.B., Egli T., "Temperature-dependent growth

kinetics of Escherichia Coli ML 30 in glucose-limited continuous cul-
ture", J. Bacteriol., 178: 4530-4539

[14] Kovarova K., Egli T., "Growth kinetics of suspended microbial cells:
from single-substrate-controlled growth to mixed-substrate kinetics", Mi-
crobiol. Mol. Biol. Rev., 62(3): 646-666, 1998

[15] Lewis W.K., Withman W.G., "Principles of gas absorption", Ind. Eng.
Chem., 16(12): 1215-1220, 1924

[16] Lu S., Sun X., Zhang Y., "Insight into metabolism of CHO cells at low
glucose concentration on the basis of the determination of intracellular
metabolites", Process Biochem., 40: 1917-1921, 2005

[17] Malthus T., "An essay on the principle of population", (1798 1st edition)
with a "A summary view", and introduction by Professor Antony Flew,
Penguin Classics, 1830

[18] McKellar R.C., Lu X., "Modeling microbial responses in food", CRC

Series in Contemporary Food Science, 2004

[19] Monod J., "Recherches sur la croissance des cultures bacteriennes", Ed.
Hermann and Cie, 1942

118
[20] Penfold W., Norris D., "The relation of concentraion of food supply to
generation time of bacteria", J. Hyg, 12: 253-256, 1912

[21] Quarteroni A., "Modellistica numerica per problemi dierenziali",

Springer-Verlag Italia, 2006

[22] Rotin D., Robinson B., Tannock I.F., "Inuence of hypoxia and an acid
environment on the metabolism and viability of cultured cells: potential
implications for cell death in tumors", Cancer Rer., 46(6): 2821-2826,
1986

[23] Seth G., Hossler P., Yee J.C., Hu W.S., "Engineering cells for cell cul-
ture bioprocessing - Physiological fundamentals", Adv. Biochem. En-
gin./Biotechnol., 101: 119-164, 2006

[24] Schoonen W.G.E.J., Wanamarta A.H., Van der Klei-Van Moorsel J.M.,
Jakobs C., Joanje H., "Respiratory failure and stimulation of glycolysis
in Chinese hamster ovary cells exposed to normobaric hyperoxia", J.
Biol. Chem., 265: 11118-11124, 1990

[25] Smolke C.D., "The metabolic pathway engineering handbook", CRC

Press, 2010

[26] Tissot S., "Experimental results on CHO cell growth", Internal report,
LBTC/EPFL, 2010

[27] Van Impe J.F., Nicolaï B.M., Martens T., De Baerdemaeker J., Vande-
walle J., "Dynamic mathematical model to predict microbial growth and
inactivation during food processing", Appl. Environ. Microbiol., 58(9):
2901-2909, 1992

[28] Verhulst P.F., "Notice sur la loi que la population poursuit dans son
accroissement", Correspondance mathématique et physique 10: 113-
121, 1838

[29] Waite G.N., Waite L.R., "Applied cell and molecular biology for engi-
neers", McGraw-Hill, 2007

[30] Xing Z., Bishop N., Leister K., Li Z.J., "Modeling kinetics of a large-
scale fed-batch CHO cell culture by Markov Chain Monte Carlo method",
Biotechnol. Prog., 26: 208-219, 2010

119
[31] Zhang X., Bürki C.A., Stettler M., De Sanctis D., Perrone M.,
Discacciati M., Parolini N., DeJesus M., Hacker D.L., Quarteroni A.,
Wurm F.M., "Ecient oxygen transfer by surface aeration in shaken
cylindrical containers for mammalian cell cultivation at volumetric
scales up to 1000 L", Biochemical Engineering Journal, 45: 41-47, 2009

[32] Zwietering M.H., De Koos J.T., Hasenack .E., De Wit J.C., Van't Riet
K., "Modeling of bacterial growth as a function of temperature", Appl.
Environ. Microbiol., 57(4): 1094-1101, 1991

[33] Zwietering M.H., Jongenburger I., Rombouts F.M., Van't Riet K., Mo-
deling of the bacterial growth curve, Appl. Environ. Microbiol., 56(6):
1875-1881, 1990

[34] Zwietering M.H., Wijtzes T., De Wit J.C., Van't Riet K., "A decision
support system for prediction of the microbial spoilage in foods", J. Food
Prot., 55: 973-979, 1992

120