Dupuytren's: a systems biology disease

Dupuytren's disease (DD) is an ill-defined fibroproliferative disorder of the palm of the hands leading to digital contracture. DD commonly occurs in individuals of northern European extraction. Cellular components and processes associated with DD pathogenesis include altered gene and protein expression of cytokines, growth factors, adhesion molecules, and extracellular matrix components. Histology has shown increased but varying levels of particular types of collagen, myofibroblasts and myoglobin proteins in DD tissue. Free radicals and localised ischaemia have been suggested to trigger the proliferation of DD tissue. Although the existing available biological information on DD may contain potentially valuable (though largely uninterpreted) information, the precise aetiology of DD remains unknown. Systems biology combines mechanistic modelling with quantitative experimentation in studies of networks and better understanding of the interaction of multiple components in disease processes. Adopting systems biology may be the ideal approach for future research in order to improve understanding of complex diseases of multifactorial origin. In this review, we propose that DD is a disease of several networks rather than of a single gene, and show that this accounts for the experimental observations obtained to date from a variety of sources. We outline how DD may be investigated more effectively by employing a systems biology approach that considers the disease network as a whole rather than focusing on any specific single molecule.

Systems biology combines mechanistic modelling with quantitative experimentation in studies of networks [27][28][29][30][31][32][33][34] and aims at understanding how the interaction of multiple components within a cell, tissue, organ or indeed individual leads to much of biological function and obfuscates correlations with single genes. Systemslevel approaches have begun to help comprehension of network control, (dys-)regulation, and function [35][36][37][38]. Th is has improved the understanding of certain disorders [39], and has provided new rationales for drug discovery [40][41][42]. Th e complex biology of DD may constitute an invitation to a systems level approach. In this review, we outline such an approach.

Dupuytren's disease and its many faces
Histopathology Clinical examples of fi brosis include renal interstitial fi brosis [43], scleroderma [44], sarcoidosis [45], idiopathic pulmonary fi brosis [46], retroperitoneal fi brosis [47] and DD [48]. DD tissue shows increased deposition of collagen III relative to collagen I and increased levels of collagen hydroxylation and glycosylation [49]. DD is thought to arise either from a defect in the wound repair process or from an abnormal response to wounding. Th e presence of immune cells and related phenomena in DD tissue suggests DD may be immune-related [50][51][52][53].

Abstract
Dupuytren's disease (DD) is an ill-defi ned fi broproliferative disorder of the palm of the hands leading to digital contracture. DD commonly occurs in individuals of northern European extraction. Cellular components and processes associated with DD pathogenesis include altered gene and protein expression of cytokines, growth factors, adhesion molecules, and extracellular matrix components. Histology has shown increased but varying levels of particular types of collagen, myofi broblasts and myoglobin proteins in DD tissue. Free radicals and localised ischaemia have been suggested to trigger the proliferation of DD tissue. Although the existing available biological information on DD may contain potentially valuable (though largely uninterpreted) information, the precise aetiology of DD remains unknown. Systems biology combines mechanistic modelling with quantitative experimentation in studies of networks and better understanding of the interaction of multiple components in disease processes. Adopting systems biology may be the ideal approach for future research in order to improve understanding of complex diseases of multifactorial origin. In this review, we propose that DD is a disease of several networks rather than of a single gene, and show that this accounts for the experimental observations obtained to date from a variety of sources. We outline how DD may be investigated more eff ectively by employing a systems biology approach that considers the disease network as a whole rather than focusing on any specifi c single molecule.
Cellularity (quantifi ed as the cellular density) of the DD nodules (see below) is indicative of the activity of the disease [4]. DD has been classifi ed into three stages coexisting in the same specimen, that is, proliferative, involutional and residual, further subdivided into the essentially fi brous nodules, reactive tissue, and residual tissue. It contains two structurally distinct fi brotic elements: the nodule is a highly vascularised tissue containing many fi broblasts, with a high percentage being recognised as myofi broblasts due to their expression of the αsmooth muscle actin; and the cord is relatively avascular, acellular, and collagen-rich with few myofi broblasts. Th e nodule may develop into the cord as the disease progresses over time or the two structures represent indepen dent stages of the disease. Macroscopically, neither the deep retinacular tissue that includes the transverse palmar ligament or fascia, also known as 'Skoog's fi bres' , nor the fi brous fl exor tendon sheaths appear to be involved in DD. Other areas are aff ected macroscopically but at irregular depth and distribution, with the more superfi cial layers and ulnar side of the palm being aff ected most.
Th e specialised mesenchymal cells expressing smooth muscle α-actin may explain the contractility observed in DD [11,[54][55][56]; they resemble the myofi broblasts of granulation tissue thought to be responsible for contraction during wound healing. Th e Dupuytren myofi broblast synthesizes fi bronectin, an extracellular glycoprotein that connects myofi broblasts and connects them to the extracellular stromal matrix through an integrin.
According to genome-wide gene expression profi les, fi broblasts come in various subtypes [57], perhaps due to 'topographic diff erentiation'; that is, distinct phenotypes persisted in vitro even when fi broblasts were isolated from the infl uence of other cell types [58]. Chang et al. [57] did not evaluate diversity or cell heterogeneity in DD. All of those evaluated had the morphology of elongated, spindle-shaped cells. Fibroblast cultures were uniformly positive for a mesenchymal immunofl uorescence marker, but negative for markers of epithelial, smooth muscle, endothelial, perineural, and histiocytic cells. Diff erent passages of the same fi broblast culture clustered with each other, indicating that their in vitro phenotypes were stable. Several components implicated as modulators of transdiff erentiation of DD fi broblasts into myofi broblasts have been reported [59][60][61][62][63][64][65][66][67][68][69]. Among the cytokines, transforming growth factor-β is thought to be a signifi cant inducer of myo fi broblast transdiff er en tiation because of its ability to up-regulate α-smooth muscle actin and collagen in fi broblasts, both in vivo and in vitro [65].

Genetics
A study performed in a fi ve generation Swedish family suggested that DD was inherited in an autosomal dominant pattern [70]. Linkage analysis implicated a single region of approximately 6 cM between markers D16S419 In stage B the disease spreads up the fascia and into the fi ngers, leading to the development of a cord. In stage C the disease spreads up the fi ngers, eventually creating a tight cord such that the fi ngers are forced to progressively bend, and are unable to straighten, eff ecting an irreversible contracture. Reproduced with consent from Bayat et al. [6].

Stage C Stage B Stage A
and D16S3032 at a logarithm of the odds (LOD) score >1.5. Genotyping of four siblings aff ected by the disease but from another branch of the family together with the use of additional microsatellite markers produced a maximal LOD score of 3.2 (for D16S415), with four other markers producing LOD scores >1.5. When a disease is dominant, it is likely to be caused by a single allele of a single gene, and by the molecule it encodes. From this perspective, the above fi ndings would suggest that DD is a single gene disease. To date, however, linkage to a single gene has not been reported at a LOD that is much more signifi cant than the marginal value of 3 in this Swedish study and the penetrance in this study was incomplete. In addition, the disease develops at an advanced age, there are many more sporadic cases of DD, and there are few such families for which the genetic analysis has been performed. Indeed, other studies have shown association of the disease with other loci, including a positive association with HLA-DRB1*15 on chromosome 6 in Caucasians [71]. A study of 20 British DD patients with a maternally transmitted inheritance pattern demonstrated a mutation within the mitochondrial genome (mito chondrial 16S ribosomal RNA region) in 90% of patients [72]. Th e defective mitochondria generated abnormally high levels of free radicals and induced defects in apoptotic mechanisms.

Reactive oxygen species
A relation between localised ischaemia, reactive oxygen species (ROS) and DD was projected from a study in which palmar fascia from 10 DD individuals were subjected to 0 to 60 minutes of tourniquet ischaemia DD [73]. Th e concentration of hypoxanthine was six-fold higher in Dupuytren's palmar fascia compared to the palmar fascia from ten suitable control patients (having carpal tunnel decompression), implying that measuring metabolites directly in tissue could help understand DD. Xanthine oxidase catalyzes the removal of hypoxanthine, generating super oxide free radicals and hydrogen peroxide as by-products that would damage the peri vascular connective tissue, which fi broblasts would attempt to repair. Upon addition of free radicals to fi broblast cultures from DD palmar fascia, lower concentrations of ROS stimulated fi broblast proliferation, explaining the observed increase in collagen type III [73]. Th e hypoxanthine was more abundant in nodular areas than in the tight fi brous cords. We propose that microvessel narrowing, leading to localised hypoxic conditions, may be one cause of DD, secondary to age, smoking and other environ mental factors as discussed by Shih and Bayat [3].

Transcriptomics
Alterations of gene expression in Dupuytren's nodules [25], Peyronie's plaques [74], and cultured fi broblasts have been reported. Because of the complexity and hetero geneity of the disease, we carried out a new transcriptome analysis, optimizing for unbiased experimental design, sample size and suffi ciently large data sets by considering the nodule and the cord as two separate entities, and by adding a pathway oriented approach. We compared diseased Dupuytren tissue biopsies (both nodules and cords) with corresponding healthy tissue (the transverse palmar fascia adjacent to the diseased site) from the same patients as well as with profi les from the palmar fascia of individuals not aff ected by DD [18]. Th e genes we confi rmed and established as altered in expression in DD [13,60] are involved in the immune response, angiogenesis, apoptosis, cell adhesion and cellmatrix adhesion, cell cycle and proliferation, cell diff er entiation, transcription, development, signalling and signal transduction, protein synthesis and folding, oxygen transport, and carbohydrate metabolism ( Figure 2). A study comparing fi broblasts isolated from DD patients with those from controls found tens of genes to be altered at the mRNA expression level, although these diff ered between the two microarray platforms used. Th e downregulation of three of the genes was confi rmed by quantitative PCR; these encode a proteoglycan, a fi bulin and type XV collagen alpha 1 chain [75]. Also using PCR, Ulrich and colleagues [76] found that DD tissue amplifi ed mRNA encoding one metalloprotease and two tissue inhi bi tors of metalloproteases. Th e genes localised between markers D16S419 and D16S3032 encode hemoglobin α1 and 2, cadherin 11 type 2, OB-cadherin (osteoblast), matrix metallopeptidase 2, periplakin, tryptase α/β1, and tryptase β2. Th e transcriptomics results were therefore not particularly supportive of the early hypothesis that altered expression of a single autosomal gene on chromo some 16 is solely responsible for DD. Th e observation that genes with fairly obvious functional connections to DD, such as those encoding metalloproteases, proteo glycans and collagen components, have altered expres sion brings home the message that even if the disease were to have a singlegene origin, its aetiology is likely to involve multiple regulatory pathways and genes down stream. Up to now the hunt for the single DD gene has not only failed but also weakened its own motivation; many genes, both upstream and downstream, may be involved in causing DD.

Proteomics
Th e DD proteomics venture began in 2006 with the study of protein expression profi les in an attempt to identify potential disease biomarkers [77,78]. In one study, twodimensional gel electrophoresis was performed to extract proteins from diseased tissue (nodule and cord), the Skoog's fi bres, and normal control tissues. MALDI-TOF-MS (matrix-assisted laser desorption ionization time-offl ight mass spectrometry) generated a peptide mass fi nger print that was used to search protein databases; however, the authors did not report names of identifi ed proteomic changes in their abstract. SELDI-TOF-MS (surface enhanced laser desorption/ionization time-offl ight mass spectrometry) using Ciphergen's SELDI-TOF-MS Protein Biological System II (PBSII) ProteinChip reader [78] revealed several diff erentially expressed low molecular weight (<20 kDa) tissue proteins and identifi ed three disease-associated protein features (4,600.8 Da, 10,254.5 Da, and 11,405.1 Da) that were elevated (5-, 12-, and 4-fold, respectively). Th ree potential low molecular weight protein markers (p4.6DC, p1ODC, p11.7DC) for DD were identifi ed. An integrative proteomic-interactomic approach [79] coupling two-dimensional gel electrophoresis with mass spectrometry compared the proteomic profi le of DD tissue with that of unaff ected patient-matched palmar fasciae tissue and found several proteins correlated with DD. Th e fi ndings were used to create a protein-protein interaction network (inter actome) map on the basis of the proposed interactions in the Human Interactome Map (HiMAP) [80] and the Search Tool for the Retrieval of Interacting Proteins (STRING) [81]; however, several proteins were added to fi ll gaps in order to yield a complete network, which then involved extra-and intracellular signalling, oxidative stress, cytoskeletal changes, and alterations in cellular metabolism. In particular, ERBB-2 and insulin-like growth factor 1 receptor (IGF-1R) and the Akt signalling pathway emerged as novel components of pro-survival signalling in Dupuytren's fi broblasts. One should exercise care, however, not to over-interpret these results, as they are partly based on inference from other protein interaction data obtained in diff erent contexts. In addition, increased activity of pathways need not involve increased protein levels [82] and increased pathway expression may be homeostatic rather than aetiological.
The dilemma: more or less data -less or more understanding As more and more aspects associating with DD are revealed, we see less and less forest (understanding of the

Biological processes Biological processes
Cellular component Molecular Function disease) for the trees (its many molecules) [83]. Even if the disease were set in motion by a single genetic factor, its aetiology would involve many diverse processes such that DD will be co-determined by the many factors that regulate those processes. If indeed the networks governing diff erentiation of normal fi brocytes of the palm of the hand are perturbed irreversibly so that they diff erentiate into muscle-like tissue without the proper controllers of contraction and relaxation, then diff erent sets of genetic perturbations could lead to DD. In this context DD may be much like cancer [39]. Th e dilemma is that although we now have an unprecedented set of methodologies for the identifi cation and analysis of all the molecules in living cells, that methodology alone is not enough. We need something substantially more to understand how all those molecules interact to create functional networks. Seeing more mole cules may not help our understanding; seeing the connections between them and more mechanism might.

Systems biology disease versus molecular disease
Th e disease in Figure 3a may seem to depend on a single molecule (gene) only, or at least that is how it is often conceived. But of course, a disease cannot depend on a gene (if defi ned as the corresponding DNA sequence) alone: it will depend on its gene product (F in Figure 3a), and in fact on the molecular function of the latter. For instance, a muscular dystrophy could result from a mutation in the gene encoding myosin, the molecular function of which is muscle contraction. If that muscular dystrophy were only found when the myosin gene has been mutated and if the severity of the disease was not infl uenced by other factors, then that muscular dystrophy would be a single-gene disease. In actuality there are many diff erent genetic lesions that lead to similar muscular dystrophies, including lesions in mitochondrially encoded genes [84]. A better candidate for a mono-gene disease may be phenylketonuria, an inherited (autosomal recessive) metabolic disease that is largely due to mutations in the phenylalanine hydroxylase (PAH) gene [85]. However, its therapy (dietary restriction) shows that the disease can be infl uenced by external factors, mutations in genes involved in the synthesis of a cofactor of the phenylalanine hydroxylation reaction also lead to the disease, and there are multiple alleles of the PAH gene that confer diff erent severities. Hence, even this disease exhibits characteristics of systems biology diseases.
Most diseases have multiple genes associated with them. Such diseases might be considered to be a group of single-gene diseases; that is, many diff erent diseases each being caused by a diff erent single gene lesion, but all with similar phenotypes [86]. Th is would explain the asso ciation of multiple genes with the disease. In the case of a group of single gene diseases, no other faulty molecules should be important for that individual disease and no other gene changes (for example, polymorphisms) or conditions (for example, diet) should infl uence the disease severity. Notably, a single patient's transcriptome should then show only changes in the single molecular culprit and not in other factors controlling the network leading to the disease; and in the transcriptomes of diff erent patients suff ering from the same disease group, that single molecular culprit should be diff erent. For DD this is not what is observed (see above). As illustrated in Figure 3b, in a systems biology disease the function that is compromised depends cooperatively on a number of pathways, the functioning of each of which again depends on many cooperating molecular factors. In systems biology diseases one would typically fi nd multiple changes in the transcriptome or proteome of each patient, diff ering between individual patients but such that all have a very similar disabling eff ect on network function. Identifying a disease as a systems biology or network disease does not dispel molecules from its pathology: molecules are always involved. Th e issue is whether the change in networking of the molecules is crucial for the disease, that is, whether the disease is more a consequence of faulty networking than of an individual malfunctioning molecule.
What diff erence should this all make for research, diagnosis and therapy? Th e answer is straightforward: when dealing with a network disease, one should deal with the network; when dealing with a molecular disease, one should concentrate on the molecule. For systems biology diseases, transcriptome patterns should be mapped onto the known cellular pathways, network fl uxes, and the disease. Th e concept 'candidate pathway' or even 'candidate network' should be substituted for 'candidate gene' . In addition, one should investigate at the proteome, metabolome, and functional levels [87], and not each independently but all together, and then one may need to examine multiple network functions (Figure 4). Malignant cancer, for instance, may involve proliferation, lack of apoptosis, metastasis and multiple drug resistances.

Is Dupuytren's disease a single-molecule or a systems biological disease?
DD has been identifi ed as a disease inherited in an autosomal dominant pattern [70] (and see above). It was linked to a single 6cM region on chromosome 16. Th is would suggest that all DD patients should have a mutation in this part of their genome, and that transcriptomes of DD patients should be altered in terms of the level of the transcripts encoded by this part of the genome or in terms of the coding sequence of one of those transcripts. However, the dominance was incom plete, weak, and has only been observed in a single Swedish family. Th is suggests that the genes on chromosome 16 are only dominant when other genes in the genome are of certain allelic forms. Moreover, in many other cases the expression levels of many other mRNAs were changed, although it remains to be analyzed whether in those studies there was always a change in mRNAs from the 6 cM region on chromosome 16. In our own studies, DD nodule transcriptomes of individual patients have all exhibited multiple changes in mRNA levels, and although these changes overlapped, they were not identical between individuals. Th e proteome did not point to a single causative protein either. Th e functional studies pointed to myofi broblast enrichment, although not clearly as the sole cause, and neither was a causal relationship between a gene on chromosome 16 in the 6 cM region and diff erentiation of myofi broblasts established. Th is all shows that DD is not a single-gene disease and suggests that it is not just a group of pure single-gene diseases either. It is much more likely to be a systems biology disease.

Treatment
Surgical intervention is still the current mainstay of treatment for DD, usually involving fasciotomy, fasciectomy or dermofasciectomy [88]. A variety of non-operative techniques have been practiced but have failed to give long-lasting benefi ts. More recently, clinical In the molecular, or single-gene disease, a mutation in or around a piece of DNA causes a change in function of the gene product F. F is solely responsible (or the rate limiting step) for the physiological function that is impaired in the disease, or for the pathology itself. (b) In the systems biological or network disease, the biological function that is impaired in the disease, or the new pathological function, depends on many factors (called Z here) at the same time. Factors Z themselves depend on many other factors, on genes and environmental (for example, nutritional, hormonal, age) factors, and ultimately even on the development of the pathology itself. In terms of transcriptomics, changes in any factors shown could correlate somewhat with the disease, in either type of disease. In the molecular disease (a), however, the correlation between the disease and changes in the single causative disease gene should be 100%. When, as in systems biology, the cause-eff ect relationships are investigated, the correlations should be time-and perturbation-dependent and consistent with the network drawn (for example, a deletion of Y might not aff ect the disease totally, but should destroy the causal correlation between gene 2 and disease). The systems biology paradigm is not soft, however, as in that case the correlation between disease and network state should be 100%.

(a) (b)
investi gation (phase II and III), including the use of clostridial collagenase injections, have shown encouraging results in some DD patients [89,90], but long-term follow-up results are required before this can be advocated as standard procedure in place of surgery.

Conclusion
Since association of DD with a single gene is inconclusive, the present mainstream research paradigm may be unlikely to lead to a full understanding of it. Th e experimental data appear to be more consis tent with DD being a systems biology disease. Th erefore, a diff erent approach to the disease should be considered; analysis, diagnosis and therapy should target pathways rather than genes or their products. Th e concept of a 'candidate gene' should be replaced with that of 'candi date pathway(s)' . Studies should be aimed at elucidating cause-eff ect chains, rather than disease corre lations. From the experimental data, alterations in path ways should be inferred. Using transgenic and antisense approaches in cell lines, these pathway alterations should then be induced and the predicted development into a DD cellular phenotype tested. Th e pathways are expected to be integrals of gene expression, signalling and metabolic networks, as should be the approach and data analysis. A hypothesis-driven systems biology would be based on a priori observations in human, in vitro or in vivo (linkage and expression studies, for example), or on knowledge of related diseases (such as plantar fi bromatosis, peyronies, musculo-aponeurotic fi bromatosis and even keloid disease). Inter-relationships would be sought between hypothesized underlying mechanisms governing these fi brotic disorders and physiological changes predicted based on molecular and environmental changes impacting on those mechanisms. Th is could then be extended to understand inter-versus intra-individual variability. Altering the networks using multiple mole cular inter ven tions in a tissue culture model system for DD would enable the hypotheses to be tested.
Such an approach should also help put into perspective existing inconclusive discoveries and maximize the utilization of data obtained from molecular approaches computation-assisted analysis and experimental design can lead to understanding of the disease and rational and optimally eff ective therapies. This fi gure is for illustrative purposes only and the precise network structure has yet to be fully determined.  In the top-down branch of the systems biology approach, data maps generated by large scale experiments fi rst need to be annotated and subjected to statistical analysis in order to extract biologically relevant information. That information should then be used to generate hypotheses concerning patterns of molecular behaviour or dynamic parameters of the networks. Phenomenological or partly mechanistic mathematical modelling can already help here to weed the impossible from the possible and to enable one to put multiple complex interactions into single testable hypotheses. Then, predictions can be made and tested. This may spiral through iterations of top-down systems biology into an ever improving set of hypotheses that may become more and more mechanistic. A bottom-up systems biology branch of the research may begin with proposed mechanisms (such as stimulation of fi broblast growth because of enhanced reactive oxygen species production) and develop mathematical models of these in order to assist with experimental design. By spirally testing and adjusting the hypothesis this will ultimately lead to a hypothesis that is better and better tested and involves more and more of the network. At each step, data will be consolidated, reducing the amount of unnecessary information while increasing their accuracy, quality and usefulness to improve and generate stronger models of the DD cell. A metabolic or signalling network can then be represented in silico and its properties studied using computer-simulated perturbations. For instance, the fl ux balance model could be applied to predict the behaviour of metabolic networks upon perturbation of the optimised metabolites within a metabolic pathway.  [95] and Systems Biology Graphical Notation (SBGN) [96], and modelling tools such as COmplex PAthway SImulator (COPASI) [97], Cytoscape [98] and Pathway databases (for example, Ingenuity pathway analysis software) facilitate data representation and inter-operability from leading multidisciplinary research groups [99]. Additionally, the Java Web Simulation (JWS) facility, which quality controls kinetic models and puts them into a live repository, enables through-web experimentation in silico for scientists naive with respect to modelling, although it is experimentdependent [100]. BioModels is a parallel model repository not built for in silico experimentation but focusing on annotation [101]. If we are right, and the systems biology research paradigm is adopted, the rewards should be substantial: no longer will the data collected in this fi eld disappear into the diaspora of the experimental literature; they will be analyzed in terms of network models and, when informative, connected with proper data annotation [102]. Once the network hypotheses are proven, they will under pin the development of new rational biomarker strategies, and become the starting point for potential therapeutic interventions and prophylaxis.
More concretely, even though myofi broblasts obtained from diff erent stages of DD may exhibit features that could trigger contraction and uncontrollable growth, neither the diversity of these cells nor the extent or nature of their local specifi city in situ with regard to diff er entiation have been examined systematically. Remodel ling of vascular connective tissue should be of funda mental importance as DD progresses over time and the ability of that tissue to be remodelled could be an important factor in the development of the disease. Matrix remodelling and matrix turnover are controlled by a complex network of cell-cell and cell-matrix inter actions [103]. Mathematical modelling of these networks combined with targeted experimentation should help deduce the net outcome of the balance between proliferative and degradative processes.
Together, genomics, transcriptomics, proteomics, fl uxomics, lipidomics, interactomics, glycomics and secretomics studies of biofl uids within the DD system have the potential to improve our understanding of the disease immeasurably [104][105][106]; however, the 'together' concept requires an integration that is only plausible through systems approaches.
Investigating this complex deforming fi bromatosis as part of a systems biology approach ( Figure 5) will benefi t not only understanding of the specifi c diseased phenotypes, but may also address the eff ects on the extracellular matrix and excreted by-products, and could off er further suggestions for early diagnosis.
Recognition of DD as a systems biology disease should also aff ect therapy. Treatment need not target the reversal of a molecular event but rather return a network aff ected by a combination of molecular culprits (for example, resulting from single-nucleotide polymorphisms) to its normal state. Th is return may be attempted by interventions in the network that may otherwise be unrelated to the molecular culprits. In this sense the recognition that DD is a network disease should increase the number of treatment options. More specifi cally, one should look at a number of intracellular networks (metabolic, signal transducing, or otherwise) aff ecting the disease and then consider treatment with combinations of existing drugs (and behavioural advice) that target all those networks. Th e relative dosages should be optimized, allowing for the optimum to diff er between patients.

Competing interests
The authors declare that they have no competing interests.