Metabolic design: how to engineer a living cell to desired metabolite concentrations and fluxes

Cell State Transition Models Stratify Breast Cancer Cell Phenotypes and Reveal New Therapeutic Targets

Understanding signaling patterns of transformation and controlling cell phenotypes is a challenge of current biology. Here we applied a cell State Transition Assessment and Regulation (cSTAR) approach to a perturbation dataset of single cell phosphoproteomic patterns of multiple breast cancer (BC) and normal breast tissue-derived cell lines. Following a separation of luminal, basal, and normal cell states, we identified signaling nodes within core control networks, delineated causal connections, and determined the primary drivers underlying oncogenic transformation and transitions across distinct BC subtypes. Whereas cell lines within the same BC subtype have different mutational and expression profiles, the architecture of the core network was similar for all luminal BC cells, and mTOR was a main oncogenic driver. In contrast, core networks of basal BC were heterogeneous and segregated into roughly four major subclasses with distinct oncogenic and BC subtype drivers. Likewise, normal breast tissue cells were separated into two different subclasses. Based on the data and quantified network topologies, we derived mechanistic cSTAR models that serve as digital cell twins and allow the deliberate control of cell movements within a Waddington landscape across different cell states. These cSTAR models suggested strategies of normalizing phosphorylation networks of BC cell lines using small molecule inhibitors.

On paradoxes between optimal growth, metabolic control analysis, and flux balance analysis

In Microbiology it is often assumed that growth rate is maximal. This may be taken to suggest that the dependence of the growth rate on every enzyme activity is at the top of an inverse-parabolic function, i.e. that all flux control coefficients should equal zero. This might seem to imply that the sum of these flux control coefficients equals zero. According to the summation law of Metabolic Control Analysis (MCA) the sum of flux control coefficients should equal 1 however. And in Flux Balance Analysis (FBA) catabolism is often limited by a hard bound, causing catabolism to fully control the fluxes, again in apparent contrast with a flux control coefficient of zero. Here we resolve these paradoxes (apparent contradictions) in an analysis that uses the 'Edinburgh pathway', the 'Amsterdam pathway', as well as a generic metabolic network providing the building blocks or Gibbs energy for microbial growth. We review and show that (i) optimization depends on so-called enzyme control coefficients rather than the 'catalytic control coefficients' of MCA's summation law, (ii) when optimization occurs at fixed total protein, the former differ from the latter to the extent that they may all become equal to zero in the optimum state, (iii) in more realistic scenarios of optimization where catalytically inert biomass is compensating or maintenance metabolism is taken into consideration, the optimum enzyme concentrations should not be expected to equal those that maximize the specific growth rate, (iv) optimization may be in terms of yield rather than specific growth rate, which resolves the paradox because the sum of catalytic control coefficients on yield equals 0, (v) FBA effectively maximizes growth yield, and for yield the summation law states 0 rather than 1, thereby removing the paradox, (vi) furthermore, FBA then comes more often to a 'hard optimum' defined by a maximum catabolic flux and a catabolic-enzyme control coefficient of 1. The trade-off between maintenance metabolism and growth is highlighted as worthy of further analysis.

DMPy: a Python package for automated mathematical model construction of large-scale metabolic systems

Background

Models of metabolism are often used in biotechnology and pharmaceutical research to identify drug targets or increase the direct production of valuable compounds. Due to the complexity of large metabolic systems, a number of conclusions have been drawn using mathematical methods with simplifying assumptions. For example, constraint-based models describe changes of internal concentrations that occur much quicker than alterations in cell physiology. Thus, metabolite concentrations and reaction fluxes are fixed to constant values. This greatly reduces the mathematical complexity, while providing a reasonably good description of the system in steady state. However, without a large number of constraints, many different flux sets can describe the optimal model and we obtain no information on how metabolite levels dynamically change. Thus, to accurately determine what is taking place within the cell, finer quality data and more detailed models need to be constructed.

Results

In this paper we present a computational framework, DMPy, that uses a network scheme as input to automatically search for kinetic rates and produce a mathematical model that describes temporal changes of metabolite fluxes. The parameter search utilises several online databases to find measured reaction parameters. From this, we take advantage of previous modelling efforts, such as Parameter Balancing, to produce an initial mathematical model of a metabolic pathway. We analyse the effect of parameter uncertainty on model dynamics and test how recent flux-based model reduction techniques alter system properties. To our knowledge this is the first time such analysis has been performed on large models of metabolism. Our results highlight that good estimates of at least 80% of the reaction rates are required to accurately model metabolic systems. Furthermore, reducing the size of the model by grouping reactions together based on fluxes alters the resulting system dynamics.

Conclusion

The presented pipeline automates the modelling process for large metabolic networks. From this, users can simulate their pathway of interest and obtain a better understanding of how altering conditions influences cellular dynamics. By testing the effects of different parameterisations we are also able to provide suggestions to help construct more accurate models of complete metabolic systems in the future.

Electronic supplementary material

The online version of this article (10.1186/s12918-018-0584-8) contains supplementary material, which is available to authorized users.

Mathematical analysis of the influence of brain metabolism on the BOLD signal in Alzheimer's disease

Blood oxygen level-dependent functional magnetic resonance imaging (BOLD-fMRI) is a standard clinical tool for the detection of brain activation. In Alzheimer's disease (AD), task-related and resting state fMRI have been used to detect brain dysfunction. It has been shown that the shape of the BOLD response is affected in early AD. To correctly interpret these changes, the mechanisms responsible for the observed behaviour need to be known. The parameters of the canonical hemodynamic response function (HRF) commonly used in the analysis of fMRI data have no direct biological interpretation and cannot be used to answer this question. We here present a model that allows relating AD-specific changes in the BOLD shape to changes in the underlying energy metabolism. According to our findings, the classic view that differences in the BOLD shape are only attributed to changes in strength and duration of the stimulus does not hold. Instead, peak height, peak timing and full width at half maximum are sensitive to changes in the reaction rate of several metabolic reactions. Our systems-theoretic approach allows the use of patient-specific clinical data to predict dementia-driven changes in the HRF, which can be used to improve the results of fMRI analyses in AD patients.

A method for analysis and design of metabolism using metabolomics data and kinetic models: Application on lipidomics using a novel kinetic model of sphingolipid metabolism

We present a model-based method, designated Inverse Metabolic Control Analysis (IMCA), which can be used in conjunction with classical Metabolic Control Analysis for the analysis and design of cellular metabolism. We demonstrate the capabilities of the method by first developing a comprehensively curated kinetic model of sphingolipid biosynthesis in the yeast Saccharomyces cerevisiae. Next we apply IMCA using the model and integrating lipidomics data. The combinatorial complexity of the synthesis of sphingolipid molecules, along with the operational complexity of the participating enzymes of the pathway, presents an excellent case study for testing the capabilities of the IMCA. The exceptional agreement of the predictions of the method with genome-wide data highlights the importance and value of a comprehensive and consistent engineering approach for the development of such methods and models. Based on the analysis, we identified the class of enzymes regulating the distribution of sphingolipids among species and hydroxylation states, with the D-phospholipase SPO14 being one of the most prominent. The method and the applications presented here can be used for a broader, model-based inverse metabolic engineering approach.

A modular modulation method for achieving increases in metabolite production

Increasing the production of overproducing strains represents a great challenge. Here, we develop a modular modulation method to determine the key steps for genetic manipulation to increase metabolite production. The method consists of three steps: (i) modularization of the metabolic network into two modules connected by linking metabolites, (ii) change in the activity of the modules using auxiliary rates producing or consuming the linking metabolites in appropriate proportions and (iii) determination of the key modules and steps to increase production. The mathematical formulation of the method in matrix form shows that it may be applied to metabolic networks of any structure and size, with reactions showing any kind of rate laws. The results are valid for any type of conservation relationships in the metabolite concentrations or interactions between modules. The activity of the module may, in principle, be changed by any large factor. The method may be applied recursively or combined with other methods devised to perform fine searches in smaller regions. In practice, it is implemented by integrating to the producer strain heterologous reactions or synthetic pathways producing or consuming the linking metabolites. The new procedure may contribute to develop metabolic engineering into a more systematic practice.

Systematic construction of kinetic models from genome-scale metabolic networks

The quantitative effects of environmental and genetic perturbations on metabolism can be studied in silico using kinetic models. We present a strategy for large-scale model construction based on a logical layering of data such as reaction fluxes, metabolite concentrations, and kinetic constants. The resulting models contain realistic standard rate laws and plausible parameters, adhere to the laws of thermodynamics, and reproduce a predefined steady state. These features have not been simultaneously achieved by previous workflows. We demonstrate the advantages and limitations of the workflow by translating the yeast consensus metabolic network into a kinetic model. Despite crudely selected data, the model shows realistic control behaviour, a stable dynamic, and realistic response to perturbations in extracellular glucose concentrations. The paper concludes by outlining how new data can continuously be fed into the workflow and how iterative model building can assist in directing experiments.

Enhancing the production of tropane alkaloids in transgenic Anisodus acutangulus hairy root cultures by over-expressing tropinone reductase I and hyoscyamine-6β-hydroxylase

Tropane alkaloids (TA) including hyoscyamine, anisodamine, scopolamine and anisodine, are used medicinally as anticholinergic agents with increasing market demand, so it is very important to improve TA production by metabolic engineering strategy. Here, we report the simultaneous introduction of genes encoding the branch-controlling enzyme tropinone reductase I (TRI, EU424321) and the downstream rate-limiting enzyme hyoscyamine-6β-hydroxylase (H6H, EF187826) involved in TA biosynthesis into Anisodus acutangulus hairy roots by Agrobacterium-mediated gene transfer technology. Transgenic hairy root lines expressing both TRI and H6H (TH lines) produced significantly higher (P < 0.05) levels of TA compared with the control and single gene transformed lines (T or H lines). The best double gene transformed line (TH53) produced 4.293 mg g(-1) TA, which was about 4.49-fold higher than that of the control lines (0.96 mg g(-1)). As far as it is known, this is the first report on simultaneous introduction of TRI and H6H genes into TA-producing plant by biotechnological approaches. Besides, the content of anisodine was also greatly improved in A. acutangulus by over-expression of AaTRI and AaH6H genes. The average content of anisodine in TH lines was 0.984 mg g(-1) dw, about 18.57-fold of BC lines (0.053 mg g(-1) dw). This is the first time that this phenomenon has been found in TA-producing plants.

Adjusting phenotypes by noise control

Genetically identical cells can show phenotypic variability. This is often caused by stochastic events that originate from randomness in biochemical processes involving in gene expression and other extrinsic cellular processes. From an engineering perspective, there have been efforts focused on theory and experiments to control noise levels by perturbing and replacing gene network components. However, systematic methods for noise control are lacking mainly due to the intractable mathematical structure of noise propagation through reaction networks. Here, we provide a numerical analysis method by quantifying the parametric sensitivity of noise characteristics at the level of the linear noise approximation. Our analysis is readily applicable to various types of noise control and to different types of system; for example, we can orthogonally control the mean and noise levels and can control system dynamics such as noisy oscillations. As an illustration we applied our method to HIV and yeast gene expression systems and metabolic networks. The oscillatory signal control was applied to p53 oscillations from DNA damage. Furthermore, we showed that the efficiency of orthogonal control can be enhanced by applying extrinsic noise and feedback. Our noise control analysis can be applied to any stochastic model belonging to continuous time Markovian systems such as biological and chemical reaction systems, and even computer and social networks. We anticipate the proposed analysis to be a useful tool for designing and controlling synthetic gene networks.

Stochastic gene expression at the single cell level can lead to significant phenotypic variation at the population level. To obtain a desired phenotype, the noise levels of intracellular protein concentrations may need to be tuned and controlled. Noise levels often decrease in relative amount as the mean values increase. This implies that the noise levels can be passively controlled by changing the mean values. In an engineering perspective, the noise levels can be further controlled while the mean values can be simultaneously adjusted to desired values. Here, systematic schemes for such simultaneous control are described by identifying where and by how much the system needs to be perturbed. The schemes can be applied to the design process of a potential therapeutic HIV-drug that targets a certain set of reactions that are identified by the proposed analysis, to prevent stochastic transition to the lytic state. In some cases, the simultaneous control cannot be performed efficiently, when the noise levels strongly change with the mean values. This problem is shown to be resolved by applying extra noise and feedback.

Robust signal processing in living cells

Cellular signaling networks have evolved an astonishing ability to function reliably and with high fidelity in uncertain environments. A crucial prerequisite for the high precision exhibited by many signaling circuits is their ability to keep the concentrations of active signaling compounds within tightly defined bounds, despite strong stochastic fluctuations in copy numbers and other detrimental influences. Based on a simple mathematical formalism, we identify topological organizing principles that facilitate such robust control of intracellular concentrations in the face of multifarious perturbations. Our framework allows us to judge whether a multiple-input-multiple-output reaction network is robust against large perturbations of network parameters and enables the predictive design of perfectly robust synthetic network architectures. Utilizing the Escherichia coli chemotaxis pathway as a hallmark example, we provide experimental evidence that our framework indeed allows us to unravel the topological organization of robust signaling. We demonstrate that the specific organization of the pathway allows the system to maintain global concentration robustness of the diffusible response regulator CheY with respect to several dominant perturbations. Our framework provides a counterpoint to the hypothesis that cellular function relies on an extensive machinery to fine-tune or control intracellular parameters. Rather, we suggest that for a large class of perturbations, there exists an appropriate topology that renders the network output invariant to the respective perturbations.

Ensemble modeling for aromatic production in Escherichia coli

Ensemble Modeling (EM) is a recently developed method for metabolic modeling, particularly for utilizing the effect of enzyme tuning data on the production of a specific compound to refine the model. This approach is used here to investigate the production of aromatic products in Escherichia coli. Instead of using dynamic metabolite data to fit a model, the EM approach uses phenotypic data (effects of enzyme overexpression or knockouts on the steady state production rate) to screen possible models. These data are routinely generated during strain design. An ensemble of models is constructed that all reach the same steady state and are based on the same mechanistic framework at the elementary reaction level. The behavior of the models spans the kinetics allowable by thermodynamics. Then by using existing data from the literature for the overexpression of genes coding for transketolase (Tkt), transaldolase (Tal), and phosphoenolpyruvate synthase (Pps) to screen the ensemble, we arrive at a set of models that properly describes the known enzyme overexpression phenotypes. This subset of models becomes more predictive as additional data are used to refine the models. The final ensemble of models demonstrates the characteristic of the cell that Tkt is the first rate controlling step, and correctly predicts that only after Tkt is overexpressed does an increase in Pps increase the production rate of aromatics. This work demonstrates that EM is able to capture the result of enzyme overexpression on aromatic producing bacteria by successfully utilizing routinely generated enzyme tuning data to guide model learning.

In silico strategy to rationally engineer metabolite production: A case study for threonine in Escherichia coli

Genetic engineering of metabolic pathways is a standard strategy to increase the production of metabolites of economic interest. However, such flux increases could very likely lead to undesirable changes in metabolite concentrations, producing deleterious perturbations on other cellular processes. These negative effects could be avoided by implementing a balanced increase of enzyme concentrations according to the Universal Method [Kacser and Acerenza (1993) Eur J Biochem 216:361-367]. Exact application of the method usually requires modification of many reactions, which is difficult to achieve in practice. Here, improvement of threonine production via pyruvate kinase deletion in Escherichia coli is used as a case study to demonstrate a partial application of the Universal Method, which includes performing sensitivity analysis. Our analysis predicts that manipulating a few reactions is sufficient to obtain an important increase in threonine production without major perturbations of metabolite concentrations.

Ensemble modeling of metabolic networks

Complete modeling of metabolic networks is desirable, but it is difficult to accomplish because of the lack of kinetics. As a step toward this goal, we have developed an approach to build an ensemble of dynamic models that reach the same steady state. The models in the ensemble are based on the same mechanistic framework at the elementary reaction level, including known regulations, and span the space of all kinetics allowable by thermodynamics. This ensemble allows for the examination of possible phenotypes of the network upon perturbations, such as changes in enzyme expression levels. The size of the ensemble is reduced by acquiring data for such perturbation phenotypes. If the mechanistic framework is approximately accurate, the ensemble converges to a smaller set of models and becomes more predictive. This approach bypasses the need for detailed characterization of kinetic parameters and arrives at a set of models that describes relevant phenotypes upon enzyme perturbations.

Mathematical model of phosphatidylinositol-4,5-bisphosphate hydrolysis mediated by epidermal growth factor receptor generating diacylglycerol

Phosphatidylinositol-4,5-bisphosphate (PIP2) is hydrolyzed in response to the tyrosine phosphorylation of the epidermal growth factor receptor (EGFR) and plays an important role in regulating cell proliferation and differentiation through the generation of second messengers diacylglycerol (DAG) and trisphosphate inositol (IP3) which lead to the activation of protein kinase C (PKC) and increased levels of intracellular calcium, respectively. In the paper, a mathematical model was established to simulate the accumulation of DAG due to PIP2 hydrolysis mediated by EGFR. Molecular mechanisms between DAG, PIP2, EGFR and phosphatidylinositol transfer protein (PITP) were explained successfully, and positive cooperativity which existed between phospholipase C-gamma1 (PLC-gamma1) and PIP2 was also explained. In the model the effects of parameters on simulation of PIP2 hydrolysis were analyzed and the efficacies of some molecular intervention strategies were predicted. To test the coherence between the model and the biological response to epidermal growth factor (EGF) in cells, the levels of DAG and the tyrosine phosphorylation-EGFRs in NIH3T3 mouse embryonic fibroblast (MEF) were determined by biochemical experiments which showed that the accumulation of DAG was a sigmoidal function of phosphorylation-EGFR concentration, and the consistency between the mathematical model and experimental results was confirmed. In brief, this mathematical model provided a new idea for the further study of the dynamic change of biological characteristics in inositol phospholipid hydrolysis, predicting the efficacy of molecular intervention and the relationship between the metabolisms of inositol phospholipid and other signal transduction pathways.

Elucidation and structural analysis of conserved pools for genome-scale metabolic reconstructions

In this article, we introduce metabolite concentration coupling analysis (MCCA) to study conservation relationships for metabolite concentrations in genome-scale metabolic networks. The analysis allows the global identification of subsets of metabolites whose concentrations are always coupled within common conserved pools. Also, the minimal conserved pool identification (MCPI) procedure is developed for elucidating conserved pools for targeted metabolites without computing the entire basis conservation relationships. The approaches are demonstrated on genome-scale metabolic reconstructions of Helicobacter pylori, Escherichia coli, and Saccharomyces cerevisiae. Despite significant differences in the size and complexity of the examined organism's models, we find that the concentrations of nearly all metabolites are coupled within a relatively small number of subsets. These correspond to the overall exchange of carbon molecules into and out of the networks, interconversion of energy and redox cofactors, and the transfer of nitrogen, sulfur, phosphate, coenzyme A, and acyl carrier protein moieties among metabolites. The presence of large conserved pools can be viewed as global biophysical barriers protecting cellular systems from stresses, maintaining coordinated interconversions between key metabolites, and providing an additional mode of global metabolic regulation. The developed approaches thus provide novel and versatile tools for elucidating coupling relationships between metabolite concentrations with implications in biotechnological and medical applications.

Effect of pmt gene overexpression on tropane alkaloid production in transformed root cultures of Datura metel and Hyoscyamus muticus

In order to increase the production of the pharmaceuticals hyoscyamine and scopolamine in hairy root cultures, a binary vector system was developed to introduce the T-DNA of the Ri plasmid together with the tobacco pmt gene under the control of CaMV 35S promoter, into the genome of Datura metel and Hyoscyamus muticus. This gene codes for putrescine:SAM N-methyltransferase (PMT; EC. 2.1.1.53), which catalyses the first committed step in the tropane alkaloid pathway. Hairy root cultures overexpressing the pmt gene aged faster and accumulated higher amounts of tropane alkaloids than control hairy roots. Both hyoscyamine and scopolamine production were improved in hairy root cultures of D. metel, whereas in H. muticus only hyoscyamine contents were increased by pmt gene overexpression. These roots have a high capacity to synthesize hyoscyamine, but their ability to convert it into scopolamine is very limited. The results indicate that the same biosynthetic pathway in two related plant species can be differently regulated, and overexpression of a given gene does not necessarily lead to a similar accumulation pattern of secondary metabolites.

Metabolic engineering of lactic acid bacteria, the combined approach: kinetic modelling, metabolic control and experimental analysis

Everyone who has ever tried to radically change metabolic fluxes knows that it is often harder to determine which enzymes have to be modified than it is to actually implement these changes. In the more traditional genetic engineering approaches 'bottle-necks' are pinpointed using qualitative, intuitive approaches, but the alleviation of suspected 'rate-limiting' steps has not often been successful. Here the authors demonstrate that a model of pyruvate distribution in Lactococcus lactis based on enzyme kinetics in combination with metabolic control analysis clearly indicates the key control points in the flux to acetoin and diacetyl, important flavour compounds. The model presented here (available at http://jjj.biochem.sun.ac.za/wcfs.html) showed that the enzymes with the greatest effect on this flux resided outside the acetolactate synthase branch itself. Experiments confirmed the predictions of the model, i.e. knocking out lactate dehydrogenase and overexpressing NADH oxidase increased the flux through the acetolactate synthase branch from 0 to 75% of measured product formation rates.

Comparative genome analysis and pathway reconstruction

Pathway reconstruction builds on genome and biochemical data with the aim of reconstructing higher level interactions between identified enzymes in a specific genome, in particular the different enzyme pathways (species or individual/patient). Metabolite flow in a pathway is analyzed by different tools, such as elementary mode analysis. This reveals key enzymes and pharmacological targets in the enzyme network. An overview of bioinformatic tools and algorithms for these tasks, application examples and recent results from these techniques are presented. Target selection, drug development and optimization can all be sped up using these approaches.

Product dependence and bifunctionality compromise the ultrasensitivity of signal transduction cascades

Covalent modification cycles are ubiquitous. Theoretical studies have suggested that they serve to increase sensitivity. However, this suggestion has not been corroborated experimentally in vivo. Here, we demonstrate that the assumptions of the theoretical studies, i.e., irreversibility and absence of product inhibition, were not trivial: when the conversion reactions are close to equilibrium or saturated by their product, "zero-order" ultrasensitivity disappears. For high sensitivities to arise, not only substrate saturation (zero-order) but also high equilibrium constants and low product saturation are required. Many covalent modification cycles are catalyzed by one bifunctional 'ambiguous' enzyme rather than by two independent proteins. This makes high substrate concentration and low product concentration for both reactions of the cycle inconsistent; such modification cycles cannot have high responses. Defining signal strength as ratios of modified (e.g., phosphorylated) over unmodified protein, signal-to-signal response sensitivity equals 1: signal strength should remain constant along a cascade of ambiguous modification cycles. We also show that the total concentration of a signalling effector protein cannot affect the signal emanating from a modification cycle catalyzed by an ambiguous enzyme if the ratio of the two forms of the effector protein is not altered. This finding may explain the experimental result that the pivotal signal transduction protein PII plus its paralogue GlnK do not control steady-state N-signal transduction in Escherichia coli. It also rationalizes the absence of strong phenotypes for many signal-transduction proteins. Emphasis on extent of modification of these proteins is perhaps more urgent than transcriptome analysis.

Control of the threonine-synthesis pathway in Escherichia coli: a theoretical and experimental approach

A computer simulation of the threonine-synthesis pathway in Escherichia coli Tir-8 has been developed based on our previous measurements of the kinetics of the pathway enzymes under near-physiological conditions. The model successfully simulates the main features of the time courses of threonine synthesis previously observed in a cell-free extract without alteration of the experimentally determined parameters, although improved quantitative fits can be obtained with small parameter adjustments. At the concentrations of enzymes, precursors and products present in cells, the model predicts a threonine-synthesis flux close to that required to support cell growth. Furthermore, the first two enzymes operate close to equilibrium, providing an example of a near-equilibrium feedback-inhibited enzyme. The predicted flux control coefficients of the pathway enzymes under physiological conditions show that the control of flux is shared between the first three enzymes: aspartate kinase, aspartate semialdehyde dehydrogenase and homoserine dehydrogenase, with no single activity dominating the control. The response of the model to the external metabolites shows that the sharing of control between the three enzymes holds across a wide range of conditions, but that the pathway flux is sensitive to the aspartate concentration. When the model was embedded in a larger model to simulate the variable demands for threonine at different growth rates, it showed the accumulation of free threonine that is typical of the Tir-8 strain at low growth rates. At low growth rates, the control of threonine flux remains largely with the pathway enzymes. As an example of the predictive power of the model, we studied the consequences of over-expressing different enzymes in the pathway.

Engineering a living cell to desired metabolite concentrations and fluxes: pathways with multifunctional enzymes

With molecular genetics enabling modulation of the concentrations of cellular enzymes, metabolic engineering becomes limited by the question of which modulations of the enzyme concentrations are required to bring about a desired pattern of cellular metabolism. In an earlier paper (Kholodenko et al. (1998). Biotechnol. Bioeng. 59, 239-247) we derived a method to determine the required modulations. This method, however, cannot be immediately applied to cellular pathways with enzymes catalyzing more than one step in metabolism (multifunctional enzymes). In the present paper we show to which extent the presence of multifunctional enzymes limits biotechological ambitions, which one might otherwise pursue in vain. In particular, it is impossible to change the concentration of a single intermediate and leave the rest of metabolism unperturbed if that intermediate interacts directly with a multifunctional enzyme. The analytical machinery of Metabolic Control Analysis is used to relate the desired and ensuing changes in the metabolic pattern. An explicit solution to this problem of engineering metabolism is then given in the form of a single matrix equation.