What can mathematical modelling say about CHO metabolism and protein glycosylation? Academic research paper on "Industrial Biotechnology"

001

CSB1-0167; No ofPages 10 ARTICLE IN PRESS

Computational and Structural Biotechnology Journal xxx (2017) xxx-xxx

ELSEVIER

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

What can mathematical modelling say about CHO metabolism and protein glycosylation?

Sarah Galleguillosa b, David Ruckerbauer a-b, Matthias P. Gerstla-b, Nicole Borth3 Michael Hanschoa-b, Jürgen Zanghellinia

aDepartment of Biotechnology, University of Natural Resources and Life Sciences, Vienna, Austria b Austrian Centre of Industrial Biotechnology, Vienna, Austria

080 081 082

100 101 102

110 111 112

120 121 122

ARTICLE INFO

Article history:

Received 12 November 2016

Received in revised form 9 January 2017

Accepted 12 January 2017

Available online xxxx

MSC2010: 00-01 99-00

Keywords: CHO cells

Metabolic modelling

Glycosylation

Kinetic model 13C-labeling

ABSTRACT

Chinese hamster ovary cells have been in the spotlight for process optimization in recent years, due to being the major, long established cell factory for the production of recombinant proteins. A deep, quantitative understanding of CHO metabolism and mechanisms involved in protein glycosylation has proven to be attainable through the development of high throughput technologies. Here we review the most notable accomplishments in the field of modelling CHO metabolism and protein glycosylation.

© 2017 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license

(http://creativecommons.org/licenses/by/4.0/).

1. Introduction

Mammalian cells, more specifically immortalized Chinese hamster ovary (CHO) cells, are the dominant biological platform for the production of many therapeutic recombinant proteins [1]. CHO cells are not only able to correctly fold these proteins, but they are also capable of performing human-compatible post-translational modifications (e.g. glycosylation) [2,3]. This is important for the correct functioning of the proteins and to prevent immunogenic responses in humans. In addition, CHO cells show high and stable expression of heterologous proteins and they easily adapt to growth in suspension. Both features are essential for industrial-scale production [4]. Furthermore, CHO cells are considered to be "safe", since most human pathogenic viruses do not replicate in CHO [5]. All

Abbreviations: CHO, Chinese hamster ovary; FBA, Flux Balance Analysis; GSMR, genome-scale metabolic reconstruction; MFA, metabolic flux analysis; PPP, pentose phosphate pathway; TCA, tricarboxylic acid.

* Corresponding author at: Department of Biotechnology, University of Natural Resources and Life Sciences, Vienna, Austria.

E-mail address: juergen.zanghellini@boku.ac.at (J. Zanghellini).

of these characteristics have contributed to a steep increase in the number of approvals for products expressed in this system compared to those produced in non-mammalian cells [6].

Due to their major role in the biopharmaceutical industry, several efforts have been focused on optimizing the culture process [7,8]. In the past two decades, these efforts were mainly based on experimental observations of the metabolic profiles during cell culture [9,10]. However, the advent of -omics technologies and associated modelling approaches facilitated a better and more detailed understanding of cell behaviour and intercellular processes. In particular, the development of constraint-based modelling techniques contributed tremendously to our understanding of metabolic processes, pathways and networks, so that these techniques have become one of the most (if not the most) successful modelling approaches in systems biology. Key to this success is the analysis of genome-scale metabolic reconstructions (GSMR). Combined with constraint-based modelling approaches, these models provide a mechanistic basis to investigate and elucidate genotype-phenotype relationships [11,12].

Here we will review recent progress in the computational modelling of CHO cells. Specifically, we will focus on and analyze

http://dx.doi.org/10.1016/jxsbj.2017.01.005

2001-0370/ © 2017 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).

ARTICLE IN PRESS

S. Galleguillos et al. / Computational and Structural Biotechnology Journal xxx (2017) xxx-xxx

two main issues associated with recombinant protein production:

(i) metabolic burdens affecting growth and thus protein yield and

(ii) understanding of the correct glycosylation process of the protein of interest, which is one of the major criteria for product quality.

2. CHO metabolism

The cultivation of CHO cells in bio-reactors is characterized by fast consumption of the main carbon and energy sources, glucose and glutamine, with the concomitant production ofammonia and lactate. The production of lactate not only indicates inefficient metabolisa-tion of the carbon sources [two molecules of ATP compared to 36 if glucose was completely oxidized in the tricarboxylic acid (TCA) cycle], but also has a negative effect on pH and osmolarity [13], which reduces the specific growth rate [14,15] and protein yield [16]. High ammonia concentration in the medium has similar adverse effects on cell growth, productivity and glycosylation [17-20]. Several strategies have been devised to overcome the accumulation of these by-products: rational supplementation of glucose and glutamine in fed-batch cultures [21,22], use of alternative carbon sources [7] or cell engineering [23,24], among others. These approaches were, however, based on trial and error and lack deterministic, quantitative justification.

2.1. Modelling CHO metabolism

To gain mechanistic understanding of these processes, appropriate metabolic models are required that allow one to estimate cellular flux distributions. This can be done in two ways: (i) in a time-dependent or dynamic manner (kinetic analysis) or (ii) in a constraint-based, steady-state analysis. The former approach aims to assess the evolution of the concentrations of metabolites over time and requires a large number of kinetic parameters. Due to the lack of accurate, quantitative data, this approach is currently not feasible on a genome-scale level, but restricted to small-scale models that consider several tens of selected reactions and interactions. The latter approach, on the other hand, avoids the need for detailed kinetic information by focusing on the steady-state behaviour inside the cell. Disregarding dynamic processes makes this approach, called metabolic flux analysis (MFA), scalable and suitable for genome-wide analysis. For better understanding the modelling approaches are briefly reviewed in Box 1.

In the following section we review current advances in metabolic modelling of CHO cells (listed chronologically in Fig. 1), focusing on those that investigate the accumulation of the two main metabolic by-products that are detrimental to cell growth, i.e. lactate and ammonia.

2.1.1. The metabolic fate of lactate

Altamirano et al. [31] investigated the metabolic fate of lac-tate on a metabolic network of CHO core metabolism. They argued that, when re-metabolized, lactate is not used as an energy source, as their experimentally measured low oxygen uptake rate was inconsistent with a full oxidation of lactate via the TCA cycle. Consequently, they proposed alternative pathways for the non-oxidative decarboxylation of pyruvate, which are known to exist in cancer cells [32], to be present in CHO cells too. Nevertheless, the accumulation of the end product of these pathways, i.e. acetoin, was not experimentally proven. In a more recent work, Martinez et al. [33] were able to refute this hypothesis. In their study, they analyzed the metabolic switch from lactate production to lactate uptake by means of FBA in a reduced mouse-derived metabolic model. Contrary to Altamirano et al., Martinez et al. showed that their oxygen uptake rate measurements were consistent with lactate oxidation in the

Common modelling approaches.

MFA (Metabolic Flux Analysis): pathway analysis method based on the stoichiometry of metabolic reactions and mass balances under pseudo-steady-state assumption [25]. It can be implemented in several ways. Among them:

• FBA (Flux Balance Analysis): an implementation of MFA based on the optimization of a cellular function (such as growth) under specific constraints (experimental metabolic uptake and secretion rates, thermodynamic data, etc.) [26,27].

• 13C MFA: isotope-labelled substrates are added to the culture media and, once the isotopic steady-state is reached, the distribution of the isotopes is measured via nuclear magnetic resonance or gas chromatography — mass spectrometry [28].

Markov chain Monte Carlo sampling: the glycosylation process is described as a series of states with transition probabilities from one state to the other. In the references reviewed herein, it is used to overcome the lack of kinetic parameters (metabolic and glycosylation enzymes) [29]. Artificial Neural Network models: aim to predict the behaviour of complex, non-linear systems by detecting and "learning" patterns and relationships within a training set which can be applied then to the input data [30].

TCA cycle. This suggests that the metabolic network of Altamirano et al. might have been too simplistic to capture the metabolic changes between the phases. Compared to Martinez, Altamirano's model lacked fatty acid, steroid and glycogen metabolism. In addition, the prediction of the ATP yield per mol carbon identified lactate consumption to be energetically more efficient than glucose consumption. Furthermore, they showed that the estimation of ranges for the metabolic fluxes (due to the insufficient amount of experimentally measured data in an underdetermined network) provides a valuable, semi-quantitative description of the changes between the two metabolic states. This concept was also supported by Zamorano et al. [34], who performed MFA in an under-determined network containing 100 reactions of the core metabolism and obtained narrow intervals for the fluxes with a relatively low amount of extracellular measurements.

FBA can be combined with isotopomer analysis to improve the accuracy of the predicted fluxes. Sengupta et al. [35] studied the main metabolic fluxes in a simplified network during the stationary phase of cell culture by 13C MFA. This phase is typically characterized by reduced production of lactate and high protein yields. Likewise, Templeton et al. [36] performed 13C MFA to understand the metabolic changes between growth and stationary phases in a producer CHO cell line. They found that, during the antibody production peak (stationary phase), fluxes through the TCA cycle were maximal while lactate was not produced. Moreover, this increased activity of the TCA cycle correlated with increased fluxes through the oxidative pentose phosphate pathway (PPP) when compared to the exponential phase, where high glycolytic fluxes predominate. They provide several explanations for the activation of the oxidative PPP: to regenerate NADPH/NADP+, to compensate reduction during exponential growth, to suppress oxidative stress or to cover NADPH requirements during protein folding and secretion. Irrespective of the ultimate reason, these findings point towards metabolic engineering to increase oxidative TCA cycle (CO2-producing reactions) and PPP fluxes which would help achieve higher protein yields.

ARTICLE IN PRESS

S. Galleguillos et al. / Computational and Structural Biotechnology Journal xxx (2017) xxx-xxx

Nyberg et ¿1 1999« MFA

33 reactions continuous culture Y-CHO

Altamirano et al 20011 MFA 43reactlons continuous culture CHO Tf TOR

Provost et al. EFMs + dynamic model 16 (macroscopic) reactions batch culture CHO-320

Goudaretal. 2006'

ORT-MFA 65 reactions perfusion culture CHO-(?l

Xing et al "I0'

Markov-Cham MC kinetics 6 ODES, 18 parameters fed-batch culture CHO (proprietary)

Altamirano et al Provost et al.

MFA 3 MFAs (each culture phase)

24 reactions 24 reactions (growth phase) batch culture batch culture CHO TF 70R CH0-320

Zamorano et al MFA

100 reactions batch culture CHO-320

Xing et al. Naderi et al.

MFA MFA » dynamic growth

23 reactions 34 reactions

continuous culture batch culture CHO (proprietary) CHO dhfr '

Ahnetal. J011' non-stationary 13C MFA 73 reactions fed-batch culture CHO-K1

Selvarasu et al 20121

MFA 1.540 reactions fed-batch culture CHO-DG44

Nolan et al. dynamic MFA 34 reactions fed-batch culture CHO Kl

Wilkens et al.

MFA 31 reactions batch culture CHO TF 7OR

Sengupta et at. "C MFA 58 reactions batch culture GS-CHO

Xu et al. CHO-K1 genome sequence

Sheikholeslami et al. "C MFA 68 reactions semi-continuous culture CHO-Cum2

Martine; et al

MFA 272 reactions batch culture CHO-XL99

Templeton et al "C MFA 71 reactions fed-batch culture CHO dhfr -

Ahn et al. "C MFA 79 reactions fed-batch culture CHO-K1

Ghorbaniaghdam et al kinetic model 30 reactions batch culture CHO-17)

Zamorano et al. EFMs + dynamic model 19 (macroscopic) reactions batch culture CHO-320

Sheikholeslami et al. "C MFA 46 reactions semi-continuous culture CHO-Cum2

Chen et al. 10,4

kinetic modeling 23 reactions (glucose metabolism) batch and fed-batch culture GS-CHO

Wahrtieit et al dynamic MFA 85 reactions batch and fed-batch culture CHO-K1

Nlcolae et al. non-stationary 1JC MFA 60 reactions batch culture CHOK1

Wahitieit et al dynamic MFA 70 reactions batch culture CHOK1

Hefzi et al. CHO genome-scale metabolic model

2ÔÎ61

Fig. 1. Metabolic modelling efforts in CHO listed in chronological order. Abbreviations: QRT, quasi-real-time; dhfr, dihydrofolate reductase.

2.1.2. Lactate as a beneficial medium component?

More recently, Chen et al. [37] even suggested that adding small amounts of lactate at the beginning of the culture process increases the metabolic efficiency. They used a kinetic model of the central carbon metabolism (i.e. glycolysis, PPP and TCA cycle) coupled with a model of the population dynamics and computed the time-dependent yield of lactate with respect to glucose. They found this yield decreased with increasing (yet not toxic) initial extracellular concentrations of lactate, meaning more efficient use of glucose. These findings were supported by Li et al. [38], who found that lac-tate can be fed as a major carbon source when glucose concentrations are kept low in culture.

Lactate uptake in the presence of galactose was also studied by flux balance analysis (FBA) in tissue plasminogen activator producing CHO cells in batch cultures [39]. Main changes were observed to

occur in the pyruvate metabolism; the slow utilization of galactose as compared to glucose does not provide enough pyruvate to fulfill the energy requirements. This causes lactate dehydrogenase to reverse its mode of operation, transforming lactate into pyruvate, which then enters the TCA cycle. Consequently, intracellular pyruvate and lactate concentrations are reduced, which activates the monocarboxylate transporter towards lactate uptake.

The importance of taking compartments into consideration when modelling metabolism has been demonstrated by analyzing enzyme localized activity together with non-stationary 13C techniques. These allow a more accurate assessment of metabolic fluxes [40], mostly for those pathways that cannot be resolved using steady state approaches, such as cyclic or parallel pathways (e.g. glycolysis and PPP). In this study, Nicolae et al. also discussed the sources of lactate production in both cytosol and mitochondria. Taking into account

ARTICLE IN PRESS

S. Galleguillos et al. / Computational and Structural Biotechnology Journal xxx (2017) xxx-xxx

not only the time-evolution of the metabolites, but also their spatial localization, proved that there is an additional control factor of precursor availability for both glycolysis and TCA cycle [41].

Likewise, Ahn et al. [42,43] performed high precision 13C MFA on a network containing 79 reactions and resolved metabolic fluxes accurately. During the exponential phase, characterized predominantly by high fluxes through glycolysis, 70% of the glucose was converted to lactate. They also observed a decrease in glycolytic fluxes and an increase in the oxidative PPP in the stationary phase, as reported previously [36].

2.1.3. What makes a "good"growth medium?

As already mentioned, the addition of alternative energy feedstocks can reduce the accumulation of undesired by-products. The effects of these alternative carbon sources on metabolism and protein production were studied with MFA on a reduced metabolic network by Altamirano et al. [44]. They showed that replacing glutamine by glutamate indeed resulted in reduced accumulation of ammonia, although at the price of a lower glucose uptake rate. This lowered metabolism has a negative impact on the specific protein production rate, as carbon is predominantly captured to sustain growth, leaving little for protein production.

In a follow-up work, Altamirano et al. [31] considered co-feeding strategies with galactose added to the medium, as galactose-glutamate media are known to significantly reduce by-product formation, but unfortunately, also cell growth. However, they showed that after glucose depletion, cells were able to maintain growth on galactose by simultaneously utilizing previously produced lactate. Interestingly, CHO cells do not metabolize lactate when it is offered as the sole carbon source.

MFA has also been applied for media optimization. Xing et al. performed MFA in continuous culture to assess the metabolic demands (in terms of amino acids) of antibody producing CHO cells [45], which resulted in a modified medium where final concentrations of ammonia and lactate were reduced and higher viable cell densities and higher productivities were achieved.

The steady state assumption might be problematic when modelling the inherently time-dependent fed-batch processes [46]. Hence, several efforts have been made to perform kinetic metabolic analysis while keeping a reduced, tractable set of reactions to avoid dealing with too many kinetic parameters. One of the first attempts in this direction was made by Nolan et al. [47], who included kinetic expressions in a reduced, lumped model containing 34 reactions. They studied the metabolic lactate switch by linking glucose concentration in the medium to cytosolic levels of NADH and lactate metabolic rate (lower levels of cytosolic NADH leading to net lactate consumption). This study also analyzed the intracellular concentrations of 24 metabolites in different cell lines and found that 20 of them either remained constant during the process or that their concentration changes were negligible compared to the fluxes, supporting the validity of the pseudo-steady state assumption [25] also for fed-batch processes.

Goudar et al. [48] made remarkable progress towards quasi realtime estimation of the metabolic rates in perfusion culture of CHO cells for optimal process control based on metabolite balancing. They observed that reducing the initial concentrations of glucose and glutamine resulted in an increased flux towards the TCA cycle and decreased production of waste metabolites, mainly lactate.

Xing et al. [49] applied a Markov chain Monte Carlo method to develop a kinetic model of fed-batch cultures and predicted optimal initial concentrations of glucose and glutamine that minimized the production of ammonia and lactate.

The effects of decreasing concentrations of glutamine in the media, namely the increased uptake of other carbon sources and the reduction of secreted ammonia and other products, was studied by dynamic MFA on fed-batch CHO cultures with different glutamine

concentrations [50]. They show how controlled feeding prevents 463

glutamine metabolism to be coupled to waste producing pathways 464

and, moreover, stabilizes the flux through the TCA cycle. 465

Similarly, Sheikholeslami et al. [51] used 13C MFA to compare 466

two semicontinuous cultures grown on chemically defined media 467

with 1 mM and 5 mM glutamine, respectively, and found that low 468

glutamine uptake (in the 1 mM culture) was more metabolically effi- 469

cient in terms of the proportion of pyruvate that enters the TCA 470

cycle (and therefore is not converted to lactate). Furthermore, the 471

CHO cell line used in this study was found to be particularly effi- 472

cient, mostly under hypothermic conditions, as confirmed on their 473

previous work [52]. In this case, the use of 13C MFA was simplified 474

by analyzing only extracellular 13C -labelled metabolites and then 475

performing MFA to predict the intracellular fluxes. 476

Another interesting feeding strategy was suggested by Naderi et 477

al. [53]. In their work, they used MFA to reduce the metabolic net- 478

work to a set of significant reactions and coupled them to a dynamic 479

cell growth model to asses the differences between growing and 480

apoptotic cells. They highlighted the differences on the metabolic 481

rates for the different cell subpopulations (growing, resting and 482

apoptotic cells) and suggested a feeding strategy based on the 483

"aging" of the cell culture: when glutamine is in excess in late phases 484

of the process (where the non-growing cells become predominant), 485

there is a switch from glycolytic reactions towards deamination of 486

glutamine (and concomitant ammonia accumulation), which could 487

be prevented by gradually lowering the concentrations of glutamine 488

in the feed as the culture ages. 489

Some other compounds, such as sodium butyrate, have shown 490

to improve productivity in CHO cells [54]; Ghorbaniaghdam et al. 491

[55] used a kinetic model to assess the effects of this compound on 492

metabolism in a non-compartmentalized model assuming Michaelis- 493

Menten kinetics. They found cells to become more energetically 494

efficient (in terms of the lactate to glucose ratio) when sodium 495

butyrate was added at the mid-exponential phase. Moreover, they 496

made noteworthy improvements in describing energy metabolism 497

(in terms of ATP) and redox potential (in terms of NADH, NAD+, 498

NADPH and NADP+). Adding sodium butyrate to the media gener- 499

ates an increased flux through the TCA cycle and a high cell redox 500

potential, while not significantly changing the ATP production rates. 501

MFA has also been combined with statistical analysis methods 502

(such as principal component analysis) to determine key metabolites 503

linked to the accumulation of ammonia and lactate. In their study, 504

Selvarasu et al. [56] analyzed profiles of extracellular and intracellu- 505

lar species and integrated this information in a mouse-derived GSMR 506

with the goal of finding pathways related to growth limitation. In 507

addition to glucose and glutamine, they identified asparagine to be 508

correlated with the accumulation of ammonia in the medium, most 509

probably via its conversion to aspartate, then glutamate and finally 510

a-ketoglutarate. 511

2.1.4. The future starts now: iCHO1766, a comprehensive, 513

genome-scale metabolic reconstruction of CHO 514

As outlined above, the results derived from a model-based analy- 515

sis have significantly improved our understanding of the underlying 516

metabolic processes. This is all the more remarkable as, so far, a truly 517

CHO-specific GSMR was missing. All the applications summarized 518

above used either small-scale metabolic models or adapted recon- 519

structions developed for related organisms like mouse or humans. 520

However, after the complete genomic sequence of CHO-K1 was 521

published in 2011 [57], several research groups around the world 522

joined forces in creating the first community-curated GSMR of CHO, 523

which just now became available [58]. This model consists of 4455 524

metabolites participating in 6663 reactions and contains 1766 anno- 525

tated genes. In a first demonstration of possible applications of this 526

CHO GSMR, typical process engineering strategies were analyzed for 527

their effects on the predicted maximum product yield. In all tested 528

ARTICLE IN PRESS

S. Galleguillos et al. / Computational and Structural Biotechnology Journal xxx (2017) xxx-xxx

cases, the model suggested that these processes are not even close to tapping the full potential of CHO cells.

Furthermore, the transcriptome [59] and proteome [60] of CHO cells can be now used to obtain strain-specific models that provide a more precise characterization of metabolic capabilities [61]. Metabolomics data can further refine these models to make better predictions under the given culture conditions. Thus, given the advances in high-throughput technology, we expect that the model based-analysis of systems-level data like the transcriptome and proteome will help to further unravel the complexity of CHO metabolism.

Regardless of these promising results, model performance has to be further evaluated. Ever since the first modelling approaches appeared, the accuracy of experimental measurements has been shown to be an important factor to obtain meaningful results [62]. Moreover, it has been shown that biomass composition varies among different cell lines [56]. It is also known that the biomass composition has a great effect on model predictions [63]. Therefore, factors influencing the robustness of CHO metabolic models is a question that still remains to be addressed.

3. Glycosylation

Modelling metabolism aims at reducing the metabolic burden on the cells induced by the recombinant production of the protein of interest. It aims to increase the protein yield. However, the biopharmaceutical industry is not only faced with the problem of producing therapeutic proteins efficiently, but also to produce them at high quality. A major quality attribute of many biopharmaceuticals is correct glycosylation, as the correct function of most therapeutic proteins depends on it[64]. Glycosylation consists of the addition of an oligosaccharide chain to an amino acid residue, predominantly asparagine (N-linked) or serine/threonine (O-linked glycosylation) and takes place in the Endoplasmic reticu-lum and Golgi apparatus along the protein secretory pathway. These sugar modifications play a fundamental role in protein conformation, stability, solubility, receptor recognition and antigenicity as well as cytotoxicity [65-68]. Thus glycosylation essentially modifies the pharmacological properties of a protein.

Glycosylation patterns are naturally and in general heterogeneous. There are two main sources of variability in glycosylation: macroheterogeneity, which refers to the fact that a particular site in the protein might or might not be glycosylated; and micro-heterogeneity, when different glycan structures can be found on the same site. However, this natural variability presents a particular challenge for the production of biosimilars, were the glycosyla-tion patterns of the primary drugs have to be reproduced within tight tolerance regions defined by regulatory authorities.

3.1. Modelling glycosylation in CHO

Many factors are known to influence glycosylation in cell culture: concentration of metabolites in the medium (both substrate and waste products), pH, temperature and cell viability [69,70]. The mechanisms by which these factors affect micro- and macrohetero-geneity remain, however, unclear. Thus a systematic analysis is called for. Computational modelling provides a powerful framework for such an analysis. In fact, there have been remarkable advances in the development of mathematical models of glycosyla-tion, supported by the detailed knowledge of the glycosylation pathways [71]. Generally, these models aim to reduce the combinatorial explosion in the number of possible glycan distributions. To this end, models make some general assumptions, while keeping compartmentalization (each compartment is modelled differently since they contain different sets of enzymes) and finally linking glycosylation to metabolism. The complexity of the process, together

with the many intervening factors, makes modelling glycosylation 595

quite a challenging task. 596

One of the first attempts to deterministically describe protein 597

glycosylation focused on macroheterogeneity. In 1996, Shelikoff et al. 598

[72] proposed a mathematical model to predict how site-occupancy 599

is affected by different factors such as the expression levels of 600

glycotransferases, the protein production rate, the concentrations 601

of nucleotide sugars and the mRNA elongation rate. They used a 602

plug-flow reactor-based model and included protein folding as a 603

competing event that occurs concurrently with glycosylation. 604

Shortly after, Monica et al. [73] modeled sialylation of N-linked 605

oligosaccharides in a single, isotropic compartment (trans-Golgi). 606

The predictions were in agreement with experimental data of CD4 607

glycoprotein produced in CHO cells. 608

Umana and Bailey (1997) [74] presented the first attempt to 609

model glycoform microheterogeneity based on expression and spa- 610

tial localization of the enzymes involved in N-linked glycosylation. 611

Parameters such as the half-life of the protein in the Golgi, the 612

protein productivity and the volume of the Golgi compartments were 613

also included in this model. Furthermore, they modified the model 614

to take the competition for the glycosylation machinery between 615

endogenous and recombinant proteins into account. Kontoravdi et al. 616

used this model of glycosylation and included it in a simple dynamic 617

mathematical model of cell growth, death and metabolism. With this 618

reduced model they predicted the evolution of oligosaccharide molar 619

fractions over time. However, these results could not be validated 620

due to the lack of experimental data [75]. 621

Several years later, in 2005, Krambeck and Betenbaugh [76] 622

extended Umana's model (which contained 33 glycan structures and 623

33 reactions), by adding around 7500 oligosaccharide structures and 624

more than 22,000 reactions. Among these, reactions for fucosylation 625

and sialylation were included in the model, which are of special rel- 626

evance for recombinant proteins [77,78]. In contrast to the model of 627

Umana and Bailey, this model adjusts enzyme concentrations to fit 628

an experimentally observed glycopattern, thereby calibrating it to a 629

specific protein. They argue that the reason for having a case-specific, 630

adjusted model is the inherent variability of glycosylation: the glycan 631

structures do not only depend on the specific protein, but also on the 632

glycosylation site. Their results were validated with N-glycan struc- 633

tures observed in recombinant human thrombopoietin expressed in 634

CHO cells [79]. This model was then used as a prototype for further 635

development by other research groups. 636

In 2009, Krambeck et al. applied the previously developed model 637

to predict enzyme expression that resulted in an observed mass spec- 638

trometry spectrum. Reciprocally, the model was used to automati- 639

cally annotate spectra to the corresponding glycan structures [80]. 640

Both models (Umana and Bailey, Krambeck and Betenbaugh) 641

were combined in two different studies to predict the sensitivity of 642

N-Glycan branching with respect to the hexosamine flux [81] and 643

key enzymes involved in glycan branching [82]. 644

Senger and Karim [83] used a plug-flow reactor model to describe 645

the differences in glycosylation of recombinant tissue plasminogen 646

activator in CHO under shear stress conditions. They found decreased 647

site occupancy to be related to low residence times of the protein 648

in the Endoplasmic reticulum due to high protein production rates, 649

caused by increasing levels of shear stress. 650

In a follow-up study, Senger and Karim used artificial neural 651

network models to predict glycosylation from primary sequence 652

information around the glycosylation site (glycosylation window). 653

The model was used to classify macroheterogeneity as either robust 654

(invariant with culture conditions) or variable, according to this 655

sequence information [84]. They improved this approach further by 656

using information about the secondary structure and solvent acces- 657

sibility, resulting in the prediction of two main types of glycan 658

branching: high mannose type and complex-type [85]. Artificial 659

Neural Networks had already been applied to predict glycosylation 660

ARTICLE IN PRESS

S. Galleguillos et al. / Computational and Structural Biotechnology Journal xxx (2017) xxx-xxx

sites [86,87]. The complexity of the impact of protein conformation in the surroundings of the glycosylation site on glycotransferase activity hinders the creation of a mathematical model that could describe the process deterministically. Therefore, they presented the Neural-Network approach as a valuable workaround to construct prediction tools. The main advantage of this approach with respect to the previous models is that it does not require a large number of parameters, but only the protein sequence (from which they predict the secondary structure). In addition, it highlights the influence of protein secondary and tertiary structure on the accessibility of the enzymes. In another instance, Gerken et al. [88] considered the inhibitory effect of the presence of glycan structures on neighboring sites of glycosylation.

Built on the premise that glycan biosynthesis is controlled by the expression of glycotransferases, Kawano et al. [89] predicted a set of glycan structures from DNA microarray data. This set was further expanded by Suga et al. [90] with the prediction of new structures (Kawano's set of predicted glycans was limited to those included in the database of known structures). This approach was refined several years later with high-throughput RNA microarray data [91].

Hossler et al. [92] compared the prediction performance of two main models for protein maturation in the Golgi: four continuous mixing-tanks (4CSTR) for vesicular transport and four plug-flow reactors (4PFR) in series for the maturation model. They claimed that the latter describes the process more accurately and they emphasised the importance of the residence time in the Golgi and enzyme localization as key parameters to be considered when modelling glycosylation.

The plug-flow reactor model was then used to describe monoclonal antibody (mAb) glycosylation [93]. The major improvement over the previous model was to include the transport of nucleotide sugar donors. This was the first step towards coupling cellular metabolism (and therefore measurable variables like glucose uptake) to glycosylation. Kaveh et al. [94] pursued this goal and performed a dynamic analysis of extracellular metabolite concentrations via MFA and linked those of glutamine and glucose to nucleotide sugar biosynthesis and glycolysis using the previous models (del Val 2011 [93] and Hossler 2007 [92]). The model successfully predicted dynamic trends of the glycopatterns of mAb produced in CHO batch culture. In another study [95], they combined dynamic MFA with the GLYCOVIS software developed by Hossler et al. [96] to predict, based on experimentally observed glycopatterns, how different concentrations of glutamine, glucose, ammonia and different pH values affect the glycosylation process. Yet more progress was made by Jedrzejewski et al. [97], who used a dynamic model for cell death and growth together with the dynamic model from del Val [93] to predict glycosylation patterns. In this case, experimental data from mAb producing mouse hybridoma cells was used for the calculations. A similar study was applied to mAb producing CHO fed-batch cultures [98]. As a result, recent models have succeeded in linking cell growth, metabolism, protein production rate and glycosylation [99].

The majority of these models describe N-glycosylation. Liu et al. [100] presented a reaction network for the formation of the O-glycosylation of the sialyl Lewis-X epitope. In their work, they introduce the concept of "subset-modelling", where the whole set of reactions in the network is divided into "sub-networks" and then a search is performed for the one that fits the experimental data best. Furthermore, they use genetic algorithm-based optimization, hierarchical clustering and principal component analysis to fit subsets of reaction networks to the observed glycan structure distribution, thereby reducing the parameterisation of the model. Recently, the same group developed a software for the automated creation, analysis and visualization of glycosylation reaction networks, called GNAT (Glycosylation Network Analysis Toolbox) [101,102]. GNAT was further expanded to include a higher number of enzymes [103].

Kim et al. [104] also exploited the modularity of the glycosylation 727

pathways to propose new engineering strategies based on targeting 728

modules instead of specific enzymes. 729 In a simpler approach, FBA was applied to assess the effect of 730

low temperature conditions on metabolism and nucleotide sugar 731

availability for glycosylation in mAb producing CHO cells [105]. A 732

similar MFA-based method was applied to analyze the effects of dif- 733

ferent concentrations of glutamine in the media on nucleotide sugar 734

intracellular concentrations and N-glycan content of recombinant 735

human chorionic gonadotrophin in CHO cells [106]. 736

In the past year, a simple stoichiometric model was also used to 737

compute the nucleotide sugar demands for glycosylation of recom- 738

binant proteins in CHO for rational feeding strategies [107]. 739

In order to avoid the requirement of a high number of kinetic 740

parameters, Spahn et al. [108] used a Markov chain model to describe 741

glycosylation as a stochastic process in which each glycan state 742

has a transition probability to reach the next glycan state. These 743

probabilities are linked to the steady state solution given by FBA 744

for a reduced network of the reactions contributing to the observed 745

glycoprofile. By using this protein-specific model, they successfully 746

predicted the effect of an enzyme knock-down on an antibody 747

producer CHO cell line [109]. 748

3.2. Parameters and general assumptions 750

The parameters involved in glycosylation include reaction kinetic 752

parameters, compartment residence times, enzyme distributions 753

between compartments, compartment volumes, total glycan con- 754

centration and donor cosubstrate concentrations. These parameters 755

are either obtained via optimization or taken from literature [110]. 756

Imaging techniques for green fluorescent protein-labelled proteins 757

can be used to measure residence time and protein flux through 758

the secretory machinery [111]. Kinetic parameters are commonly 759

derived from independent enzymology experiments [112], which 760

are arduous and should be carried out for each enzyme. However, 761

there have been remarkable advances on high-throughput technolo- 762 gies that allow more accurate assessment of kinetic parameters of 763

glycosyltransferases [113]. 764

Due to the sequential nature of glycosylation, models have to 765

incorporate time-dependent equations. The majority of the kinetic 766

models reviewed herein assume Michaelis-Menten Kinetics. Over 767

time, more terms were included in these models' equations, with 768

increasing complexity, e.g. competitive inhibition terms in their 769

enzyme-kinetic expressions. 770

The main limitation of glycosylation models is the high grade 771

of parameterisation required to describe the process. Moreover, 772

most of the parameters are derived from in vitro experiments, even 773

though they might be different in an intracellular environment. As 774

previously mentioned, various factors influence glycosylation at dif- 775

ferent points of the process [70] and the effects are cell line [114], 776

glycoprotein [115] and even glycosylation site specific [74], which 777

reduces the general applicability of the models. Thus, despite the 778

tremendous advances achieved over the last years in this field, the 779

ultimate goal of predicting the effect of cell line specific behaviour 780

of different protein sequences or structures, or of process related 781

changes, on glycosylation still requires further work and ptimisation 782 to be fully achieved.

4. Conclusions and future perspectives

Metabolic modelling of mammalian cells has been hampered by 787

the inherent complexity of the cell structure (compartmentalization) 788

and the large variability of media compositions and process pertur- 789

bations under which the culture processes are carried out. Never- 790

theless, with the rampant progress in scope and reliability of -omics 791

technologies, it is for the first time that we can access cell metabolism 792

ARTICLE IN PRESS

S. Galleguillos et al. / Computational and Structural Biotechnology Journal xxx (2017) xxx-xxx

810 811 812

820 821 822

in a systems-level manner. The main applications are the rational improvement of both the culture process (media optimization) and the cells themselves (via targeted genetic engineering).

The natural evolution towards more complex metabolic models including compartmentalization and dynamic analysis has put emphasis on the necessity to have accurate measurements (intra- and extracellular) as well as accurate values for the biomass and media composition [56,116]. As for glycosylation, it has been recently shown that only a limited amount of CHO proteins account for the majority of glycosylation, which could ease the approaches

Umana and Batley

- reactors in series model

- competitive Inhibition

Shelikoff et al.

■ models site occupancy

- plu g-flow reactor model . |Jlow reactor m0(Jel

- concurrent protein folding

dealing with the dynamic evolution of glycosylation by focusing solely on these highly contributing proteins [107].

To date, the vast majority of modelling approaches in CHO have been applied in a reduced set of reactions. These usually include glycolysis, the TCA cycle, the PPP and the amino acid metabolism. However, in the past year, a full genome-scale metabolic model of CHO has become available, unleashing the capabilities of genome-scale metabolic modelling.

We have also addressed the second main challenge concerning the production of recombinant proteins in CHO. Glycosylation is

861 862

UDP GDP UDP CMP-

Villlger et al.

Fig. 2. Models for protein glycosylation in CHO listed in chronological order. Abbreviations: MS, Mass-spectrometry.

880 881 882

ARTICLE IN PRESS

S. Galleguillos et al. / Computational and Structural Biotechnology Journal xxx (2017) xxx-xxx

a highly complex, variable process, in which many factors are involved. Glycan patterns vary from batch to batch and from strain to strain, making it difficult to model the process deterministi-cally. Even though the mechanisms by which the culture conditions and enzyme expression affect glycosylation are still unknown, the modelling efforts discussed here (see Fig. 2) have taken a significant step forward in media optimization by linking glycosylation to metabolism. A future step in this direction would be including gly-can compounds in the biomass stoichiometric equation, since it has been shown that the metabolic demands towards glycosylation of both recombinant and host proteins are significant [107].

Therefore, given the combinatorial nature of the process, there are still major achievements to be reached in controlling the glyco-form, since it plays a key role in product quality.

References

Kyriakopoulos S, Kontoravdi C. Analysis of the landscape of biologically-derived pharmaceuticals in Europe: dominant production systems, molecule types on the rise and approval trends. EurJ Pharm Sci 2013;48(3):428-41. Lim Y, Wong NS, Lee YY, Ku Sc, Wong DC, Yap MG. Engineering mammalian cells in bioprocessing-current achievements and future perspectives. Biotechnol Appl Biochem 2010;55(4):175-89.

Jayapal KP, Wlaschin KF, Hu W, Yap MG. et al. Recombinant protein therapeutics from CHO cells-20 years and counting. Chem Eng Prog 2007;103(10):40. ZangM, Trautmann H, GandorC, Messi F, Asselbergs F, Leist C, Fiechter A Reiser J. Production of recombinant proteins in Chinese hamster ovary cells using a protein-free cell culture medium. Nat Biotechnol 1995;13(4):389-92. Wurm FM. 1.4 Aspects of gene transfer and gene amplification in recombinant mamman cells. Mamm Cell Biotechnol Protein Production 1997; Walsh G. Biopharmaceutical benchmarks 2014. Nature biotechnol 2014;32(10):992-1000.

Altamirano C, Paredes C, Cairo J, Godia F. Improvement of CHO cell culture medium formulation: simultaneous substitution of glucose and glutamine. Biotechnol Prog 2000;16(1):69-75.

Rajendra Y, Kiseljak D, Baldi L, Hacker DL, Wurm FM. Reduced glutamine concentration improves protein production in growth-arrested CHO-DG44 and HEK-293e cells. Biotechnol Lett 2012;34(4):619-26.

Lao MS, Toth D. Effects of ammonium and lactate on growth and metabolism of a recombinant Chinese hamster ovary cell culture. Biotechnol Prog 1997;13(5):688-91.

Dietmair S, Hodson MP, Quek L-E, Timmins NE, Chrysanthopoulos P, Jacob SS, Gray P, Nielsen LK. Metabolite profiling of CHO cells with different growth characteristics. Biotech Bioeng 2012;109(6):1404-14.

Winden WA, DamJC, Ras C, Kleijn RJ, VinkejL, GulikWM, HeijnenJJ. Metabolic-flux analysis of Saccharomyces cerevisiae CEN. PK113-7d based on mass isotopomer measurements of 13C-labeled primary metabolites. FEMS Yeast Res 2005;5(6-7):559-68.

Schuetz R, Zamboni N, Zampieri M, Heinemann M, Sauer U. Multidimensional

optimality of microbial metabolism. Sci 2012;336(6081):601-4.

Li F, Vijayasankaran N, Shen AY, Kiss R, Amanullah A. Cell culture processes for

monoclonal antibody production. MAbs 2010;2(5):466-79.

Hu W, Dodge T, Frame K, Himes V. Effect of glucose on the cultivation of

mammalian cells. Dev Biol Stand 1986;66:279-90.

Kurano N, Leist C, Messi F, Kurano S, Fiechter A. Growth behavior of Chinese hamster ovary cells in a compact loop bioreactor. 2. Effects of medium components and waste products. J biotechnol 1990;15(1-2):113-28. Glacken M, Fleischaker R, Sinskey A. Reduction of waste product excretion via nutrient control: possible strategies for maximizing product and cell yields on serum in cultures of mamMalian cells. Biotechnol Bioeng 1986;28(9):1376-89.

Schneider M, Marison IW, von Stockar U. The importance of ammonia in mammalian cell culture. J Biotechnol 1996;46(3):161-85. Hansen HA, Emborg C. Influence of ammonium on growth, metabolism, and productivity of a continuous suspension Chinese hamster ovary cell culture. Biotechnol Prog 1994;10(1):121-4.

Andersen DC, Goochee CF. The effect ofammonia on the O-linked glycosylation of granulocyte colony-stimulating factor produced by chinese hamster ovary cells. Biotechnol Bioeng 1995;47(1):96-105.

Yang M, Butler M. Effects of ammonia on CHO cell growth, erythropoietin production, and glycosylation. Biotechnol Bioeng 2000;68(4):370-80. Ljunggren J, Häggström L. Catabolic control of hybridoma cells by glucose and glutamine limited fed batch cultures. Biotechnol Bioeng 1994;44(7):808-18.

Xie L, Wang DI. Fed-batch cultivation of animal cells using different medium design concepts and feeding strategies. Biotechnol Bioeng 1994;43(11):1175-89.

Zhou M, Crawford Y, Ng D, Tung J, Pynn AF, Meier A, Yuk IH, Vijayasankaran N, Leach K, Joly J. et al. Decreasing lactate level and increasing antibody

[28 [29

[32 [33

[35 [36

[47 [48

[52 [53

production in Chinese hamster ovary cells (CHO) by reducing the expression of lactate dehydrogenase and pyruvate dehydrogenase kinases. J. biotechnol. 2011;153(1):27-34.

Kim SH, Lee GM. Functional expression of human pyruvate carboxylase for reduced lactic acid formation of Chinese hamster ovary cells (DG44). Appl Microbiol Biotechnol 2007;76(3):659-65.

Stephanopoulos G, Aristidou A, Nielsen J. Metabolic engineering: principles and methodologies. Elsevier Science; 1998.

Varma A, Palsson BO. Metabolic flux balancing: Basic concepts, scientific and practical use. Bio/technology 1994;12.

Orth JD, Thiele I, Palsson B 0. What is flux balance analysis? Nature biotechnol. 2010;28(3):245-8.

Wiechert W. 13 C metabolic flux analysis. Metab Eng2001;3(3):195-206. Hastings WK. Monte Carlo sampling methods using Markov chains and their applications. Biometrika 1970;57(1):97-109.

Shanmuganathan S, Samarasinghe S. Artificial neural network modelling, studies in computational intelligence. Springer International Publishing;

Altamirano C, Illanes A, Becerra S, Cairo JJ, Godia F. Considerations on the lactate consumption by CHO cells in the presence of galactose. J Biotechnol 2006;125(4):547-56.

Baggetto L. Deviant energetic metabolism of glycolytic cancer cells. Biochimie 1992;74(11):959-74.

Martinez VS, Dietmair S, Quek LE, Hodson MP, Gray P, Nielsen LK. Flux balance analysis of CHO cells before and after a metabolic switch from lactate production to consumption. Biotechnol Bioeng 2013;110(2):660-6. Zamorano F, Wouwer Av, Bastin G. A detailed metabolic flux analysis of an underdetermined network of CHO cells. J biotechnol 2010;150(4):497-508.

Sengupta N, Rose ST, Morgan JA. Metabolic flux analysis of CHO cell metabolism in the late non-growth phase. Biotechnol Bioeng 2011;108(1):82-92. Templeton N, Dean J, Reddy P, Young JD. Peak antibody production is associated with increased oxidative metabolism in an industrially relevant fed-batch CHO cell culture. Biotechnol Bioeng 2013;110(7).

Chen N, Bennett MH, Kontoravdi C. Analysis of Chinese hamster ovary cell metabolism through a combined computational and experimental approach. Cytotechnology 2014;66(6):945-66.

Li J, Wong CL, Vijayasankaran N, Hudson T, Amanullah A. Feeding lactate for CHO cell culture processes: impact on culture metabolism and performance. Biotechnol Bioeng 2012;109(5):1173-86.

Wilkens CA, Altamirano C, Gerdtzen ZP. Comparative metabolic analysis of lactate for CHO cells in glucose and galactose. Biotechnol Bioproc Eng 2011;16(4):714-24.

Nicolae A, Wahrheit J, Bahnemann J, Zeng AP, Heinzle E. Non-stationary 13 C metabolic flux analysis of Chinese hamster ovary cells in batch culture using extracellular labeling highlights metabolic reversibility and compartmentation. BMCSyst Biol 2014;8(1):1.

WahrheitJ, NiklasJ, Heinzle E. Metabolic control at the cytosol-mitochondria interface in different growth phases of CHO cells. Metab Eng 2014;23:9-21.

Ahn WS, Antoniewicz MR. Metabolic flux analysis of CHO cells at growth and non-growth phases using isotopic tracers and mass spectrometry. Metab Eng 2011;13(5):598-609.

Ahn WS, Antoniewicz MR. Parallel labeling experiments with [1, 2-13 C] glucose and [U-13 c] glutamine provide new insights into CHO cell metabolism. Metab Eng 2013;15:34-47.

Altamirano C, Illanes A, Casablancas A, Gamez X, Cairo J, Godia C. Analysis of CHO cells metabolic redistribution in a glutamate-based defined medium in continuous culture. Biotechnol Prog 2001;17(6):1032-41. Xing Z, Kenty B, Koyrakh I, Borys M, Pan SH, Li ZJ. Optimizing amino acid composition of CHO cell culture media for a fusion protein production. Process Biochem 2011;46(7):1423-9.

Deshpande R, Yang TH, Heinzle E. Towards a metabolic and isotopic steady state in CHO batch cultures for reliable isotope-based metabolic profiling. Biotechnolj 2009;4(2):247-63.

Nolan RP, Lee K. Dynamic model of CHO cell metabolism. Metab Eng 2011;13(1):108-24.

Goudar C, Biener R, Zhang C, Michaels J, Piret J, Konstantinov K. Towards industrial application of quasi real-time metabolic flux analysis for mammalian cell culture. Cell Culture EngineeringSpringer.2006. p. 99-118. Xing Z, Bishop N, Leister K, Li ZJ. Modeling kinetics of a large-scale fed-batch CHO cell culture by Markov chain Monte Carlo method. Biotechnol Prog 2010;26(1):208-19.

Wahrheit J, Nicolae A, Heinzle E. Dynamics of growth and metabolism controlled by glutamine availability in Chinese hamster ovary cells. Appl Microbiol Biotechnol 2014;98(4):1771-83.

Sheikholeslami Z, Jolicoeur M, Henry O. Elucidating the effects of postinduction glutamine feeding on the growth and productivity of CHO cells. Biotechnol prog 2014;30(3):535-46.

Sheikholeslami Z, Jolicoeur M, Henry O. Probing the metabolism of an inducible mammalian expression system using extracellular isotopomer analysis. J biotechnol 2013;164(4):469-78.

Naderi S, Meshram M, Wei C, McConkey B, Ingalls B, Budman H, Scharer J. Development of a mathematical model for evaluating the dynamics of normal and apoptotic Chinese hamster ovary cells. Biotechnol Prog 2011;27(5):1197-205.

ARTICLE IN PRESS

S. Galleguillos et al. / Computational and Structural Biotechnology Journal xxx (2017) xxx-xxx

McMurray-Beaulieu V, Hisiger S, Durand C, Perrier M, Jolicoeur M. Na-butyrate [84

sustains energetic states of metabolism and t-PA productivity of CHO cells. J

Biosci Bioeng 2009;108(2):160-7. [85

Ghorbaniaghdam A, Henry O, Jolicoeur M. A kinetic-metabolic model based on

cell energetic state: study of CHO cell behavior under Na-butyrate stimulation. [86

Bioprocess Biosyst Eng 2013;36(4):469-87.

Selvarasu S, Ho YS, Chong WP, Wong NS, Yusufi FN, Lee YY, Yap MG, Lee DY. Combined in silico modeling and metabolomics analysis to characterize fed- [87

batch CHO cell culture. Biotech Bioeng 2012;109(6):1415-29. Xu X, Nagarajan H, Lewis NE, Pan S, Cai Z, Liu X, Chen W, Xie M, Wang W, Hammond S. et al. The genomic sequence of the Chinese hamster ovary (CHO)- [88

K1 cell line. Nature biotechnol 2011;29(8):735-41.

Hefzi H, Ang KS, Hanscho M, Bordbar A, Ruckerbauer D, Lakshmanan M, Orellana CA, Baycin-Hizal D, Huang Y, Ley D. et al. A consensus genome- [89

scale reconstruction of Chinese hamster ovary cell metabolism. Cell Syst 2016;3(5):434-43.

Becker J, Hackl M, RuppO,JakobiT, Schneider J, Szczepanowski R, BekelT, Borth [90

N, Goesmann A, Grillari J. et al. Unraveling the Chinese hamster ovary cell line transcriptome by next-generation sequencing. J biotechnol 2011;156(3):227-35. [91

Baycin-Hizal D, Tabb DL, Chaerkady R, Chen L, Lewis NE, Nagarajan H, Sarkaria V, Kumar A, Wolozny D, Colao J. et al. Proteomic analysis of Chinese hamster ovary cells. J Proteome Res 2012;11(11):5265-76. [92

Lewis NE, Nagarajan H, Palsson BO. Constraining the metabolic genotype-phenotype relationship using a phylogeny of in silico methods. Nat Rev [93 Microbiol 2012;10(4):291-305.

Nyberg GB, Balcarcel RR, Follstad BD, Stephanopoulos G, Wang DI. Metabolism of peptide amino acids by Chinese hamster ovary cells grown in a complex [94 medium. Biotechnol Bioeng 1999;62(3):324-35.

Dikicioglu D, Krdar B, Oliver SG. Biomass composition: the "elephant in the room" of metabolic modelling. Metabolomics 2015;11(6):1690-701. http://dx. doi.org/10.1007/s11306-015-0819-2. [95

Durocher Y, Butler M. Expression systems for therapeutic glycoprotein production. Curr Opin Biotechnol 2009;20(6):700-7. http://dx.doi.org/10.1016/j. copbio.2009.10.008.

Lu D, Yang C, Liu Z. How hydrophobicity and the glycosylation site of glycans [96

affect protein folding and stability: a molecular dynamics simulation. J Phys

Chem B 2012;116(1):390-400. http://dx.doi.org/10.1021/jp203926r.

Jefferis R. Glycosylation as a strategy to improve antibody-based thera- [97

peutics. Nat Rev Drug Discov 2009;8(3):226-34. http://dx.doi.org/10.1038/

nrd2804.

Cumming DA. Glycosylation of recombinant protein therapeutics: control and functional implications. Glycobiology 1991;1(2):115-30. [98

Varki A. Biological roles of oligosaccharides: all of the theories are correct. Glycobiology 1993;3(2):97-130.

Andersen DC, Goochee CF. The effect of cell-culture conditions on the [99

oligosaccharide structures of secreted glycoproteins. Curr Opin Biotechnol 1994;5(5):546-9.

Hossler P, Khattak SF, Li ZJ. Optimal and consistent protein glycosylation in [100

mammalian cell culture. Glycobiology 2009;19(9):936-49.

Kornfeld R, Kornfeld S. Assembly of asparagine-linked oligosaccharides. Annu

Rev Biochem 1985;54(1):631-64. [101

Shelikoff M, Sinskey A, Stephanopoulos G. A modeling framework for the study

of protein glycosylation. Biotechnol bioeng 1996;50(1):73-90.

Monica TJ, Andersen DC, Goochee CF. A mathematical model of sialyla- [102 tion of N-linked oligosaccharides in the trans-golgi network. Glycobiology 1997;7(4):515-21. [103

Umana P, Bailey JE. A mathematical model of N-linked glycoform biosynthesis. Biotechnol bioeng 1997;55(6):890-908.

Kontoravdi C, Asprey SP, Pistikopoulos EN, Mantalaris A. Development of [104 a dynamic model of monoclonal antibody production and glycosylation for product quality monitoring. Comput Chem Eng 2007;31(5):392-400. [105

Krambeck FJ, Betenbaugh MJ. A mathematical model of N-linked glycosylation. Biotech Bioeng 2005;92(6):711-28.

Jenkins N, Curling EM. Glycosylation of recombinant proteins: problems and [106

prospects. Enzym MicrobTechnol 1994;16(5):354-64.

Weikert S, Papac D, Briggs J, Cowfer D, Tom S, Gawlitzek M, Lofgren J, Mehta

S, Chisholm V, Modi N. et al. Engineering Chinese hamster ovary cells to

maximize sialic acid content of recombinant glycoproteins. Nature biotechnol [107

1999;17(11):1116-21.

Inoue N, Watanabe T, Kutsukake T, Saitoh H, Tsumura H, Arai H, Takeuchi M. Asn-linked sugar chain structures of recombinant human thrombopoietin [108 produced in Chinese hamster ovary cells. Glycoconj J 1999;16(11):707-18. Krambeck FJ, Bennun SV, Narang S, Choi S, Yarema KJ, Betenbaugh MJ. A mathematical model to derive N-glycan structures and cellular enzyme activities [109 from mass spectrometric data. Glycobiology 2009;19(11):1163-75. Lau KS, Partridge EA, Grigorian A, Silvescu CI, Reinhold VN, Demetriou M, Dennis JW. Complex N-glycan number and degree of branching cooperate to regulate cell proliferation and differentiation. Cell 2007;129(1):123-34. McDonald AG, Hayes JM, Bezak T, Gluchowska SA, Cosgrave EF, Struwe WB, [110 Stroop CJ, Kok H, van de Laar T, Rudd PM. et al. Galactosyltransferase 4 is a major control point for glycan branching in N-linked glycosylation. J Cell Sci [111 2014;127(23):5014-26.

Senger RS, Karim MN. Effect of shear stress on intrinsic CHO culture state

and glycosylation of recombinant tissue-type plasminogen activator protein. [112

Biotechnol Prog 2003;19(4):1199-209.

Senger RS, Karim MN. Variable site-occupancy classification of N-linked glycosylation using artificial neural networks. Biotechnol Prog2005;21(6):1653-62. Senger RS, Karim MN. Prediction of N-linked glycan branching patterns using artificial neural networks. Math Biosci 2008;211(1):89-104. Hansen JE, Lund O, Tolstrup N, Gooley AA, Williams KL, Brunak S. NetOglyc: prediction of mucin type O-glycosylation sites based on sequence context and surface accessibility. Glycoconj J 1998;15(2):115-30.

Julenius K, M0lgaard A, Gupta R, Brunak S. Prediction, conservation analysis, and structural characterization of mamman mucin-type O-glycosylation sites. Glycobiology 2005;15(2):153-64.

Gerken TA Kinetic modeling confirms the biosynthesis of mucin core 1 (ß-Gal (1-3) a-GalNac-O-Ser/Thr) O-glycan structures are modulated by neighboring glycosylation effects. Biochem 2004;43(14):4137-42.

Kawano S, Hashimoto K, Miyama T, Goto S, Kanehisa M. Prediction of glycan structures from gene expression data based on glycosyltransferase reactions. Bioinformatics 2005;21(21):3976-82.

Suga A, Yamanishi Y, Hashimoto K, Goto S, Kanehisa M. An improved scoring scheme for predicting glycan structures from gene expression data. Genome Inform 2007;18:237-46.

Bennun SV, Yarema KJ, Betenbaugh MJ, Krambeck FJ. Integration of the transcriptome and glycome for identification of glycan cell signatures. PLoS Comput Biol 2013;9(1):e1002813.

Hossler P, Mulukutla BC, Hu W-S. Systems analysis of N-glycan processing in mamMalian cells. PloS one 2007;2(8):e713.

Jimenez del Val I, Nagy JM, Kontoravdi C. A dynamic mathematical model for monoclonal antibody N-linked glycosylation and nucleotide sugar donor transport within a maturing Golgi apparatus. Biotechnol Prog 2011;27(6):1730-43. Kaveh O, Hengameh A, Johannes G, Murray M-Y, Raymond LL, Jeno S, Hector BM. Novel dynamic model to predict the glycosylation pattern of monoclonal antibodies from extracellular cell culture conditions. IFAC Proc Vol 2013;46(31):30-5.

Aghamohseni H, Ohadi K, Spearman M, Krahn N, Moo-Young M, Scharer JM, Butler M, Budman HM. Effects of nutrient levels and average culture pH on the glycosylation pattern ofcamelid-humanized monoclonal antibody. J biotechnol 2014;186:98-109.

Hossler P, Goh L-T, Lee MM, Hu W-S. GlycoVis: visualizing glycan distribution in the protein N-glycosylation pathway in mammalian cells. Biotechnol Bioeng 2006;95(5):946-60.

Jedrzejewski PM, del Val IJ, Constantinou A, Dell A, Haslam SM, Polizzi KM, Kontoravdi C. Towards controlling the glycoform: a model framework linking extracellular metabolites to antibody glycosylation. Int. J. Mol. Sci. 2014;15(3):4492-522.

Villiger TK, Scibona E, Stettler M, Broly H, Morbidelli M, Soos M. Controlling the time evolution of mAb N-linked glycosylation - part II: Model-based predictions. Biotechnol Prog2016;32(5):1135-48.

Jimenez del Val I, Fan Y, Weilguny D. Dynamics of immature mAb glyco-form secretion during CHO cell culture: an integrated modelling framework. BiotechnolJ 2016;11(5):610-23.

Liu G, Marathe DD, Matta KL, Neelamegham S. Systems-level modeling of cellular glycosylation reaction networks: O-linked glycan formation on natural selectin ligands. Bioinformatics 2008;24(23):2740-7.

Liu G, Puri A, Neelamegham S. Glycosylation network analysis toolbox: a MATLAB-based environment for systems glycobiology. Bioinformatics 2013;29(3):404-6.

Liu G, Neelamegham S. A computational framework for the automated construction of glycosylation reaction networks. PloS one 2014;9(6):e100939. Hou W, Qiu Y, Hashimoto N, Ching WK, Aoki-Kinoshita KF. A systematic framework to derive N-glycan biosynthesis process and the automated construction of glycosylation networks. bMc bioinformatics 2016;17(7):240. Kim PJ, Lee DY, Jeong H. Centralized modularity of N-linked glycosylation pathways in mammalian cells. PloS one 2009;4(10):e7317. Sou SN, Sellick C, Lee K, Mason A, Kyriakopoulos S, Polizzi KM, Kontoravdi C. How does mild hypothermia affect monoclonal antibody glycosylation? Biotechnol Bioeng 2015;112(6):1165-76.

Burleigh SC, van de Laar T, Stroop CJ, van Grunsven WM, O'Donoghue N, Rudd PM, Davey GP. Synergizing metabolic flux analysis and nucleotide sugar metabolism to understand the control of glycosylation of recombinant protein in CHO cells. BMC biotechnol 2011;11(1):1.

Del Val IJ, Polizzi KM, Kontoravdi C. A theoretical estimate for nucleotide sugar demand towards Chinese Hamster Ovary cellular glycosylation. Sci Rep 2016;6(28547):610-23.

Spahn PN, Hansen AH, Hansen HG, Arnsdorf J, Kildegaard HF, Lewis NE. A Markov chain model for N-linked protein glycosylation-towards a low-parameter tool for model-driven glycoengineering. Metab Eng 2016;33:52-66. Imai-Nishiya H, Mori K, Inoue M, Wakitani M, Iida S, Shitara K, Satoh M. Double knockdown of a1, 6-fucosyltransferase (FUT8) and GDP-mannose 4, 6-dehydratase (GMD) in antibody-producing cells: a new strategy for generating fully non-fucosylated therapeutic antibodies with enhanced ADCC. BMC biotechnol 2007;7(1):1.

Taniguchi N, Honke K, Fukuda M. Handbook of glycosyltransferases and related genes. Springer Science & Business Media.; 2011.

Hirschberg K, Lippincott-Schwartz J. Secretory pathway kinetics and in vivo analysis of protein traffic from the Golgi complex to the cell surface. FASEB j off publ Fed Am Soc Exp Biol 1999;13:S251.

Marathe DD, Chandrasekaran E, Lau JT, Matta KL, Neelamegham S. Systemslevel studies of glycosyltransferase gene expression and enzyme activity that

ARTICLE IN PRESS

10 S. Galleguillos et al. / Computational and Structural Biotechnology Journal xxx (2017) xxx-xxx

1189 are associated with the selectin binding function of human leukocytes. FASEB J [115] Thaysen-Andersen M, Packer NH. Site-specific glycoproteomics confirms that 1255

n90 2 0 08;22(12):4154-67. protein structure dictates formation of N-glycan type, core fucosylation and 1256

[113] Yang M, Brazier M, Edwards R, Davis BG. High-throughput mass-spectrometry branching. Glycobiology 2012;22(11):1440-52.

1191 monitoring for multisubstrate enzymes: determining the kinetic parameters [116] Quek L-E, Dietmair S, Kromer JO, Nielsen LK. Metabolic flux analysis in mam- 1257

1192 and catalytic activities of glycosyltransferases. ChemBioChem 2005;6(2):346- malian cell culture. Metab Eng 2010;12(2):161-71. 1258

1193 5 7. 1259

[114] Zhang P, Chan KF, Haryadi R, Bardor M, Song Z. CHO glycosylation mutants

1194 as potential host cells to produce therapeutic proteins with enhanced efficacy. 1260

1195 Future Trends in BiotechnologySpringer.2012. p. 63-87. 1261

1196 1262

1197 1263

1198 1264

1199 1265

1200 1266

1201 1267

1202 1268

1203 1269

1204 1270

1205 1271

1206 1272

1207 1273

1208 1274

1209 1275

1210 1276

1211 1277

1212 1278

1213 1279

1214 1280

1215 1281

1216 1282

1217 1283

1218 1284

1219 1285

1220 1286

1221 1287

1222 1288

1223 1289

1224 1290

1225 1291

1226 1292

1227 1293

1228 1294

1229 1295

1230 1296

1231 1297

1232 1298

1233 1299

1234 1300

1235 1301

1236 1302

1237 1303

1238 1304

1239 1305

1240 1306

1241 1307

1242 1308

1243 1309

1244 1310

1245 1311

1246 1312

1247 1313

1248 1314

1249 1315

1250 1316

1251 1317

1252 1318

1253 1319

1254 1320