The ability to quantify metabolic fluxes in mammalian cells is crucial to understanding normal metabolic regulation and the pathophysiology of a broad spectrum of diseases. These include overtly metabolic conditions (such as diabetes) and conditions involving secondary metabolic derangements (such as cancer). Although previous studies of metabolic fluxes in mammalian cells have yielded many important results regarding specific pathways and branch points (see, e.g., refs. 38–40
), they have not been adequate to produce comprehensive metabolic flux maps. Microbial fluxomic methods based on steady-state isotope labeling patterns of amino acids do not translate readily to mammalian cells, because mammalian cells require diverse nutrient inputs (complicating interpretation of labeling pattern data) and cannot synthesize essential amino acids de novo
(limiting the information obtained from analyzing amino acids only).
We used the kinetics of assimilation of isotope-labeled nutrients into downstream metabolites to dissect metabolic fluxes in mammalian cells. Reliable flux determination is enabled by combining kinetic data with selective measurement of metabolite uptake and excretion rates and steady-state labeling patterns. As in microbial flux determination, computational data integration is achieved by a genetic algorithm that searches for flux combinations consistent with the experimental data41
. A distinguishing feature of this study is identification of a large set of flux combinations that came close to recapitulating the experimental results within their 95% confidence limits. In contrast to approaches that identify a single flux solution based on mean experimental results, this approach avoids overfitting and provides flux confidence limits. This enables significant flux changes (those cases in which the flux distributions in Supplementary Table 5
are nonoverlapping) to be reliably identified. A next step in flux deconvolution would include acquisition and incorporation of data relevant also to cofactor reactions—for example, oxygen consumption to gain insight into rates of NAD(H) oxidation and reduction.
Application of this flux measurement approach to mock- and HCMV-infected human fibroblasts revealed massive flux upregulation in the infected cells. Of 41 fluxes examined, 28 showed nonoverlapping flux distributions between the uninfected and infected cells, with flux greater in the infected cells in all cases (Supplementary Table 5
). Thus, HCMV results in nearly global metabolic upregulation. The mechanisms by which HCMV upregulates metabolic fluxes (including fatty acid biosynthesis) remain largely unknown. Virus-induced transcriptional changes may have a role for some pathways. For example, through microarray and quantitative PCR analysis, we have found that the phosphofructokinase-1 transcript is upregulated throughout infection, potentially contributing to increased glycolytic flux18
. In other cases, virally induced metabolic gene transcription does not seem to have a role. ACC transcription, for example, is not changed by HCMV infection14,18
. In such cases, post-translational modification of metabolic proteins could have a role in virally induced flux alterations.
Nucleotide biosynthesis, the target of current antimetabolites used in treatment of HCMV infection42
, is among the fluxes upregulated by HCMV. Thus, our approach effectively identified an upregulated pathway whose inhibition is known to be clinically relevant for HCMV treatment. Like nucleotide biosynthesis, fatty acid biosynthesis can be pharmacologically inhibited in mammals without severe side effects. Our observation that HCMV increased flux into fatty acid biosynthesis at least as much as nucleotide biosynthesis suggested that inhibitors of fatty acid biosynthesis, developed with the objective of treating hyperlipidemia and obesity, could be used to impair HCMV replication. This proved to be the case with TOFA, an inhibitor of the committed step of fatty acid biosynthesis, which reduced HCMV titers ~1,000-fold. Notably, TOFA also affected replication of influenza A, a virus with little in common with HCMV except for the presence of a lipid envelope.
The specific mechanism by which inhibition of fatty acid biosynthesis targets HCMV and influenza A remains to be determined. Possibilities include precluding changes in membrane composition required for viral budding, impairing synthesis of specialized envelope phospholipids43
, or impeding fatty acid modification of proteins44
. Notably, hemagglutinin release from influenza A–infected cells is sensitive to the nonspecific lipid synthesis inhibitor cerulenin45
. Similarly, hepatitis C virus replication is linked to host cholesterol synthesis through levels of geranylgeranyl pyrophosphate and activities of proteins found in cholesterol-rich membrane domains46
. Furthermore, a recent metabolomic analysis identified elevated phospholipase A2 activity in simian immunodeficiency virus–induced encephalitis47
, and a recent RNAi-based screen found that several genes involved in glycosphingolipid and inositol metabolism are important for HIV replication48
. Taken together, these results suggest that lipid metabolism is a useful therapeutic target to treat infection with various enveloped viruses. Moreover, inhibitors of lipid metabolism might have relatively broad-spectrum antiviral activity, enabling their use in patients with viral syndrome without the need to identify the specific underlying pathogen.
For any potential new therapeutic approach, it is important to weigh anticipated efficacy versus side effects. The clinical success of 3-hydroxy-3-methyl-glutaryl-CoA reductase inhibitors (statins) indicates that inhibitors of lipid metabolism can be safe and effective human therapeutics. For treatment of viral infection by inhibiting fatty acid biosynthesis, targeting of ACC may be more clinically practical than targeting FAS, as FAS inhibition can cause severe anorexia and weight loss49
. In mammals, ACC exists as two tissue-specific isoenzymes—ACC1 in adipose tissue and liver, and ACC2 in liver, heart and skeletal muscles. Although ACC1 is essential during embryogenesis32
, TOFA (which is not isozyme specific) is well tolerated in rats, including during pregnancy and postnatal development. Oral administration of TOFA (150 mg per kg per day) results in steady-state plasma concentrations (30 μg ml−1
) above those required here to block HCMV replication (10 μg ml−1
. This dose is associated with reductions in plasma cholesterol and fatty acids without obvious signs of toxicity or teratogenicity31
. Although extensive clinical testing would be required, this hints at the possibility of a favorable risk-benefit ratio for TOFA or another ACC inhibitor50
in treating HCMV infection or controlling an epidemic of influenza A resistant to current agents.
Links between cancer and viral infection have appeared repeatedly over the past decades. Viruses are important causes of cancer, and cancer and viruses both target specific genes, including tumor suppressors, to override normal control of the cell cycle and DNA replication. Our results extend these similarities to the arena of cellular metabolism. We find substantial virus-induced upregulation of nucleotide biosynthesis, glycolysis and lipid biogenesis. Increased nucleotide biosynthesis has long been known to be a hallmark of cancer, elevated glycolysis is an analog of the Warburg effect51
and, recently, increased flux from glucose into fatty acids has emerged as a feature of oncogenesis52
. It is likely that—beyond providing more specific means of targeting viral infection—understanding the mechanisms of virus-induced metabolic flux modulation will also inform cancer research.