Subtle metabolic and liver gene transcriptional changes underlie diet-induced fatty liver susceptibility in insulin-resistant mice
- First Online:
- Cite this article as:
- Toye, A.A., Dumas, M.E., Blancher, C. et al. Diabetologia (2007) 50: 1867. doi:10.1007/s00125-007-0738-5
- 919 Downloads
Complex changes in gene expression are associated with insulin resistance and non-alcoholic fatty liver disease (NAFLD) promoted by feeding a high-fat diet (HFD). We used functional genomic technologies to document molecular mechanisms associated with diet-induced NAFLD.
Materials and Methods
Male 129S6 mice were fed a diet containing 40% fat (high-fat diet, HFD) for 15 weeks. Glucose tolerance, in vivo insulin secretion, plasma lipid profile and adiposity were determined. Plasma metabonomics and liver transcriptomics were used to identify changes in gene expression associated with HFD-induced NAFLD.
In HFD-fed mice, NAFLD and impaired glucose and lipid homeostasis were associated with increased hepatic transcription of genes involved in fatty acid uptake, intracellular transport, modification and elongation, whilst genes involved in beta-oxidation and lipoprotein secretion were, paradoxically, also upregulated. NAFLD developed despite strong and sustained downregulation of transcription of the gene encoding stearoyl-coenzyme A desaturase 1 (Scd1) and uncoordinated regulation of transcription of Scd1 and the gene encoding sterol regulatory element binding factor 1c (Srebf1c) transcription. Inflammatory mechanisms appeared to be stimulated by HFD.
Our results provide an accurate representation of subtle changes in metabolic and gene expression regulation underlying disease-promoting and compensatory mechanisms, collectively contributing to diet-induced insulin resistance and NAFLD. They suggest that proposed models of NAFLD pathogenesis can be enriched with novel diet-reactive genes and disease mechanisms.
KeywordsGene transcription Genome Insulin resistance High-fat diet Metabonomics NMR spectroscopy Transcriptomics
expressed sequence tag
ingenuity systems pathway knowledge base analysis
intraperitoneal glucose tolerance test
non-alcoholic fatty liver disease
polyunsaturated fatty acid
stearoyl-coenzyme A desaturase 1
sterol regulatory element binding factor 1c
The increasing incidence of type 2 diabetes mellitus and obesity is paralleled by a progression in a wide spectrum of clinical conditions often associated with insulin resistance. Non-alcoholic fatty liver disease (NAFLD) is the most common liver disease associated with insulin resistance [1, 2]. Genetic and environmental factors, including dietary excess and inadequate exercise, contribute to NAFLD either directly or indirectly through inter-organ hormonal or metabolic signalling events [3, 4].
A composite description of molecular mechanisms associated with liver triacylglycerol accumulation, e.g. increased fatty acid uptake and synthesis and reduced fatty acid beta-oxidation and export, is emerging from clinical and experimental data . Progress in gene expression profiling technologies and genome functional annotations can help improve knowledge of NAFLD-associated markers. Functional genomic tools, including transcriptomics and metabonomics, a technology designed to quantify metabolites in biological samples , generate high-density data sets underlying disease-related mechanisms at either end of the regulation of gene expression. Alterations in transcription regulation of hepatic carbohydrate and lipogenic pathways have a central role in mechanisms contributing to insulin resistance and NAFLD [7, 8].
Owing to the complex aetiology and pathogenesis of diabetes and obesity and ethical issues in relation to genomic investigations in patients and controls, rodent models are important tools for studying human disease conditions in controlled genetic and environmental situations . Feeding rodents a high-fat diet (HFD) provides an efficient system for consistently promoting obesity, insulin resistance and NAFLD in the vast majority of strains [10, 11], even after short periods of fat feeding . We and others have shown that 129S6 mice are susceptible to HFD-induced diabetes, obesity and fatty liver and that the molecular mechanisms involved are multifaceted [13, 14].
Here we report the detailed characterisation of physiological, metabolic and liver transcriptional responses to prolonged HFD feeding in 129S6 mice. Upregulated expression of genes involved in fatty acid uptake, intracellular transport, modification and elongation in response to HFD contributes to NAFLD despite enhanced expression of genes involved in beta-oxidation and lipoprotein secretion. These results indicate that mice susceptible to experimentally induced obesity and NAFLD adapt to HFD feeding with a complex pattern of transcriptional changes, which probably involve disease-causative mechanisms and secondary hepatic adaptations to obesity, glucose intolerance, hyperinsulinaemia and hyperlipidaemia.
Materials and Methods
Male mice from the 129S6/SvEv and BALB/c inbred strains were used. Mice were weaned at 3 weeks of age and housed in groups of ten. Mice were maintained on a 12-h light and dark cycle and fed ad libitum a standard carbohydrate (CD) chow (ERB, Witham, UK) from weaning. At 5 weeks of age, one group of mice was transferred to an HFD (Special Diets Services, Witham, UK), containing 32% lard oil and 8% corn oil. A separate age-matched group remained on the CD. All procedures were carried out in accordance with national license conditions.
Intraperitoneal glucose tolerance test
At 5 months 129S6 mice were fasted overnight and subjected to a simplified intraperitoneal glucose tolerance test (IPGTT) as previously described . Mice were anaesthetised by i.p. injection of sodium pentobarbital (Sagatal; Rhône Mérieux, Harlow, UK) and injected i.p. with a single dose of 2 g glucose/kg body weight. Blood samples were collected before glucose injection and 15, 30 and 75 min afterwards. Blood glucose concentrations were determined immediately using a glucose meter (Accucheck; Roche Diagnostics, Welwyn Garden City, UK). Blood samples were centrifuged and plasma was collected for insulin assays. Cumulative glycaemia (CumG) and insulinaemia (CumIRI; IRI is immunoreactive insulin) were calculated as the increment of the values of plasma glucose and insulin, respectively, during the IPGTT. Glucose-induced insulin secretion was calculated as the insulinogenic index (ΔIRI/ΔG), which is the ratio of incremental plasma insulin values above baseline, integrated over 75 min after glucose injection (ΔIRI), to the corresponding incremental plasma glucose values (ΔG) integrated over the same period.
Mice were allowed to recover for 4 days after the IPGTT. They were separately housed in metabolic cages for one night to determine food consumption. Body weight (BW) and length (BL) (distance from the tip of the nose to the base of the tail) were determined and BMI was calculated as BW/BL2. Following an overnight fast, mice were killed by cerebral dislocation and cardiac puncture. Blood was centrifuged and plasma was removed for lipid assay and metabonomic profiling. Heart, brown fat, and epididymal and retroperitoneal white fat pads were dissected and weighed. Adiposity index was calculated as the ratio between epididymal or retroperitoneal fat weight and body weight. Liver samples were snap-frozen in liquid nitrogen and stored at −80°C for RNA preparation. A similar procedure was applied to collect blood and organs from 129S6 mice maintained on an HFD for 3, 7 and 23 weeks, and from age-matched CD-fed controls.
Plasma IRI was determined using an ELISA kit (Mercodia, Uppsala, Sweden). Plasma lipids were measured using colorimetric detection assays (ABX Diagnostics, Shefford, UK) on a COBAS MIRA clinical chemistry analyser (ABX Diagnostics).
1H-NMR spectroscopy of mouse plasma
1H-NMR spectra of individual mice were acquired at 600.22 MHz on a Bruker Avance-600 spectrometer (Bruker, Coventry, UK) using a 100 μl plasma aliquot. Standard monodimensional spectra were acquired using a selective presaturation pulse sequence for water signal suppression as described elsewhere . For each acquisition, 128 transient-free induction decays (FID) were collected into 32,768 data points using a spectral width of 20 ppm. The FIDs were convoluted with an exponential weighting function corresponding to a line broadening of 0.3 Hz before Fourier transform. The acquired NMR spectra were corrected for phase and baseline distortions using in-house software and referenced to glucose α-anomeric signal (δ5.23) . The 1H-NMR spectra were imported as full-spectrum data . To remove the effects of variation in the suppression of the water resonance, the region (δ4.5–5.0) was discarded. The data were normalised using noise variance normalisation. The final data set was then mean-centred prior to analysis .
Total RNA was extracted from frozen liver using Trizol reagent (Invitrogen Life Technologies, Paisley, UK) and cleaned with RNeasy columns (Qiagen, Crawley, UK). RNA concentrations and integrity were assessed using an Agilent 2100 Bioanalyser (Agilent Technologies, Waldbronn, Germany).
Array hybridisation and scanning
Gene transcription profiling was carried out using Affymetrix expression arrays U430A and B (Affymetrix UK, High Wycombe, UK), which contain 22,690 (U430A) and 22,576 (U430B) probe sets, allowing the interrogation of the transcription level of 13 250 (U430A) and 7577 (U430B) independent gene and expressed sequence tag (EST) sequences. Hybridisation was carried out following cDNA synthesis and biotin-labelled cRNA synthesis was performed using 10 μg of total RNA from each liver sample. cRNA concentrations and integrity were assessed using an Agilent 2100 Bioanalyser. Biotinylated cRNAs were fragmented and individually hybridised in an Affymetrix oven to the arrays. Washing and staining procedures were performed in an Affymetrix Fluidics Station 450. Arrays were scanned at 560 nm using an Agilent array scanner (Affymetrix UK). All microarray experiments were compliant with MIAME (Minimum Information about a Microarray Experiment) and both protocols and data are publicly available through the ArrayExpress repository (http://www.ebi.ac.uk/arrayexpress/, last accessed in June 2007) under the accession E-MEXP-893.
Statistical analysis of microarray data
Microarray data were obtained from liver samples of 12 mice (six from each group). Analysis of the Affymetrix. CEL file data was performed using the Bioconductor packages in the R language and environment. The U430A and B chips were normalised separately. The analysis of significant differences between test groups for each probe set was also performed separately for each chip data set. Gene chip data were normalised by RMA quantile normalisation . We used the RMA expression index, ignoring mismatch (MM) values for background correction, even though the method may underestimate transcription ratio values. A linear model for microarray data (LIMMA) was used to calculate significant differences between groups. To correct for multiple testing, we used the false discovery rate , which provides a measure of the number false positives for a given t statistic level.
Pathways and biological network analyses
To assess the significance of difference in expression of gene categories between fat-fed and control mice, array-wide false discovery probability was derived from analysis of 1,000 false data sets, each created by permuting the class labels (GO or KEGG term) among the probe sets. Data are presented as significantly differentially expressed GO and KEGG gene clusters (chip-wide p < 0.05) or not significant. Ingenuity systems pathway knowledge base analysis (IPA) (Ingenuity Systems, Mountain View, CA, USA) was used to identify known biological networks and pathways associated with gene expression differences in response to HFD.
Quantitative real-time PCR
Assays were performed on a Rotor-Gene 3000 system (Corbett Research, Milton, UK) using the QuantiTect SYBR Green PCR kit (Qiagen). Analysis was performed using the standard curve method (Rotor-Gene Software 5.0.47; Corbett Research, Milton, UK). This software allowed further verification that a single PCR product signal was quantified. Gene expression was normalised against the expression of beta actin. Experiments were performed in duplicate with samples prepared from six animals per group. Statistical significance was determined by the two-tailed independent sample t test.
Statistical analysis of 1H NMR data
Orthogonal partial least squares discriminant analysis (O-PLS-DA) allows enhanced focus on diet-related variation whilst minimising other biological/analytical variation. The two experimental groups were modelled using the O-PLS algorithm, which derives from the PLS regression method, as described previously [14, 17]. The method was also applied to determine global metabolic changes derived from plasma 1H NMR data in response to short-term and prolonged fat feeding in 129S6 mice. Further details of standard O-PLS implementation in metabonomics have been given previously [17, 20].
Metabolic and hormonal adaptation to prolonged fat feeding
We verified that this group of fat-fed mice developed histological evidence of fatty liver (data not shown), as reported in 129S6 mice under identical experimental conditions .
NMR analysis of plasma metabolites
Biomarkers derived from plasma 1H NMR metabolic profiling
Amino acids and proteins
Fatty acids and lipoproteins
Cholesterol C18 HDL
Cholesterol C18 LDL
HDL + LDL CH2
Longitudinal effects of HFD on metabolic and physiological variables
Effects of HFD on genome-wide liver transcription profiling
Significant transcription changes induced in liver of 129S6 mice by prolonged high-fat diet feeding
ATP-binding cassette, sub-family A, member 1
ATP-binding cassette, sub-family B, member 11
Acetyl-Coenzyme A acetyltransferase 1
2.5 × 10−4
Acyl-Coenzyme A synthetase long-chain family member 5
Adipose differentiation related protein
6.2 × 10−4
Aldo-keto reductase family 1, member C19
Aldehyde dehydrogenase 1 family, member B1
Aldolase 2, B isoform
2.1 × 10−4
1.6 × 10−4
Argininosuccinate synthetase 1
1.9 × 10−4
Complement component 4 binding protein
Complement component 5a, receptor 1
Complement component 6
Cytochrome p450, family 2, subfamily a, polypeptide 5
Cytochrome p450, family 2, subfamily c, polypeptide 29
Cytochrome p450, family 2, subfamily c, polypeptide 37
Cytochrome p450, family 2, subfamily c, polypeptide 37
Cytochrome p450, family 2, subfamily c, polypeptide 37
Cytochrome p450, family 2, subfamily c, polypeptide 38
Cytochrome p450, family 2, subfamily c, polypeptide 39
Cytochrome p450, family 7, subfamily b, polypeptide 1
Elongation of very long chain fatty acids like 3
Fatty acid binding protein 5
Fatty acid desaturase 2
4.4 × 10−4
Glycerol phosphate dehydrogenase 2
Glutathione S-transferase, alpha 1
2.4 × 10−5
Glutathione S-transferase, alpha 2
8.6 × 10−9
Glutathione S-transferase, mu 1
Glutathione S-transferase, omega 1
Glutathione S-transferase, theta 2
Glutathione S-transferase, theta 3
Hydroxysteroid (17-beta) dehydrogenase 12
4.4 × 10−4
Inhibitor of kappaB kinase gamma
Leukemia inhibitory factor receptor
1.8 × 10−4
Mitogen-activated protein kinase 11
NADPH oxidase 4
Serum amyloid A4
Sterol C5 desaturase
Scavenger receptor class B, member 1
Stearoyl-coenzyme A desaturase 1
8.2 × 10−7
8.4 × 10−6
Serine (or cysteine) proteinase inhibitor, clade B, member 5
Sterol regulatory element binding factor 1
Transforming growth factor beta 1 induced transcript 4
4.4 × 10−4
7.4 × 10−6
1.5 × 10−4
Transforming growth factor, beta induced
Tumour protein p53 inducible nuclear protein 2
In fat-fed mice, significant changes in the abundance of transcripts for Aldo2 (−20%), Gpd2 (up to +90%) and Gyk (up to +40%) may drive metabolism to increased production of glycerol-3-phosphate, the key substrate for triacylglycerol synthesis, whilst parallel downregulation of Mgll expression may limit the degradation of intracellular triacylglycerol stores. The transcription of enzymes of the urea cycle (Arg1, Asl, Ass1) was reduced by HFD. No major changes in transcripts for enzymes involved in carbohydrate metabolism were noted. These data suggest that increased lipoprotein uptake, intracellular fatty acid transport and elongation in response to HFD may account for the NAFLD phenotype, despite apparent reduced de novo lipid biosynthesis and enhanced lipid and cholesterol efflux.
Adaptation to HFD feeding also involved changes in expression of genes associated with inflammation (increased C4bp, C6, Ikbkg, Lifr, Tgfbi; decreased C5ar1, Serpinb5, Mapk11), DNA damage (increased Trp53inp2), cell death (increased Casp8, decreased Mapk11), increased production of cellular reactive oxygen species (ROS) (Nox4) and decreased chaperones and free radical scavengers (Mt1) (Table 2).
In addition, expression of several genes encoding glutathione S-transferases (Gsta1, Gsta2, Gstm1, Gsto1, Gstt2, Gstt3), an aldo-keto reductase (Akr1c19) and an aldehyde dehydrogenase (Aldh1b1) was downregulated by HFD, suggesting reduced glutathione conjugation and resulting in reduced ROS deactivation processes and increased proinflammatory activities. These patterns of gene expression suggest further deterioration of liver histopathological features in fat-fed mice.
Gene transcription regulation assessed by quantitative real-time PCR
To test an effect of genetic background differences on gene transcription regulation, quantitative real-time PCR was repeated in fat-fed and control BALB/c mice, which are resistant to diet-induced obesity, diabetes and fatty liver . The two strains showed markedly contrasting gene transcription patterns induced by HFD, suggesting the involvement of strain-specific mechanisms of gene expression regulation underlying susceptibility or resistance to diet-induced obesity and NAFLD.
Longitudinal effects of HFD on Scd1, Srebf1c and Ppara gene transcription
We then investigated shorter- (3 and 7 weeks) and longer-term (23 weeks) effects of HFD in 129S6 mice on key genes regulating the transcription of fatty acid metabolism (Srebf1c, Ppara) and tested the paradoxical downregulation of Scd1 transcription in a model of obesity. These experiments were carried out with fat-fed mice that showed glucose intolerance (data not shown), extensive metabolic changes and increased adiposity (Fig. 3) when compared with CD fed controls and marked NAFLD after prolonged HFD feeding (23 weeks; data not shown). Gene expression data demonstrate the permanent effects of HFD on Scd1 downregulation and Srebf1c upregulation, even though the change in Srebf1c transcription at the last time point was not significant (Fig. 6). The abundance of Ppara transcripts was also reduced by HFD, but the effect was significant only in mice fat-fed for 15 weeks.
We have used global functional genomic approaches to document the impact of dietary changes on the regulation of gene expression in an inbred mouse strain susceptible to HFD-induced obesity, diabetes and NAFLD [13, 14]. Metabonomic and transcriptomic data provide an original representation of subtle adaptations of gene expression regulatory systems to sustained and standardised dietary changes and underline the involvement of a wide range of known and potentially novel molecular mechanisms.
Pathophysiological responses to prolonged HFD in 129S6 mice combine glucose intolerance, hyperinsulinaemia, enhanced insulin secretion, hyperlipidaemia, obesity and NAFLD. Most of these features, which develop soon after the start of an HFD and remain permanently altered, are consistently found in similar fat-feeding studies in this [13, 14] and other mouse strains [11, 13, 21]. They are the result of complex metabolic and hormonal alterations that plasma metabonomics and liver transcriptomics can characterise at the molecular level. Metabolic and gene transcription changes underlie primary causes of liver fat accumulation and impaired glucose regulation induced by HFD, secondary physiological adaptations to perturbed metabolism, hyperinsulinaemia and liver morphological alterations, and diet-specific transitions between fed and fasting states.
Metabonomic data reported here provide detailed analysis of quantitative changes in individual plasma metabolites that are contributed to by multiple organs, each having specific effects on the regulation of lipid, protein and carbohydrate fuel physiology, and showing specific responses to hormonal and neural stimuli. For example, elevated plasma lactate concentration in fat-fed mice may reflect the effect of hyperglycaemia on enhanced muscle lactate production. High concentrations of plasma lactate, pyruvate and alanine in fasted mice of the fat-fed group indicate increased availability of gluconeogenic substrates. Some diet-induced changes in metabolite concentrations can be linked to alterations in liver gene expression regulation. In fat-fed mice, elevated plasma alanine, which is used for the transport of nitrogen for excretion as urea in the liver, may be caused by the downregulation of genes encoding enzymes of the urea cycle. Increased plasma n-3 fatty acids and PUFA classes in HFD-fed mice may reflect the transcription downregulation of liver epoxygenases, resulting in PUFA-sparing function. The definition of coordinated changes in plasma metabolites and gene expression could help test causal relationships between insulin resistance and NAFLD.
Hepatic insulin- and glucose-responsive metabolic systems play a central role in the regulation of carbohydrate and lipid metabolism . Interpretation of gene expression profiling data relies on mechanistic models integrating molecular markers of disease onset and progression, which derive from clinical and experimental observations. A consensus model was proposed for NAFLD  in which hepatic lipid accumulation is promoted by increased fatty acid uptake and biosynthesis, and decreased degradation and secretion. The gene expression data reported here suggest that the tendency of 129S6 mice to develop NAFLD stems from the activation of genes for lipoprotein assembly and is compensated by an increase in the expression of genes involved in beta-oxidation. Apparent insulin resistance in 129S6 fat-fed mice would indicate accrued NEFA fluxes from peripheral tissues to the liver. This is supported by the activation of genes for cytoplasmic fatty acid binding and transport proteins and transporters for the movement of fatty acids into liver cells and organelles. In this model, however, de novo fatty acid synthesis is depressed, but there is activation of mechanisms for fatty acid modification and elongation of pre-existing fatty acids.
Our gene transcription data confirm a number of reports on the regulation of hepatic gene expression [7, 8], derived from different microarray platforms generally allowing the interrogation of largely overlapping sets of transcripts [10, 23]. In particular, stimulated Srebf1c transcription in response to short-term and prolonged fat feeding in 129S6 mice has been reported in models of spontaneous and HFD-induced insulin resistance and NAFLD [13, 24, 25]. SREBF1c is a transcription factor involved in carbohydrate and lipid metabolism . Its activation, which simulates lipogenic enzymes in the liver, is mediated by increased glucose and insulin concentrations [22, 26]. In fat-fed 129S6 mice, hyperglycaemia and hyperinsulinaemia could directly result in the upregulation of Srebf1c transcription, which can in turn enhance the transcription of Abca1, encoding an Apo-A1 binding protein controlling cholesterol efflux .
Insulin-mediated stimulation of Srebf1c expression also contributes to the upregulation of SCD1, the rate-limiting enzyme catalysing fatty acid synthesis . There is strong evidence that downregulation of SCD1 prevents or reduces the development of NAFLD, obesity and insulin resistance [29, 30, 31, 32]. Elevated hepatic transcription of Scd1 has been reported in fat-fed mice of the 129S6  and C57BL/6J [13, 23] strains. This is in sharp contrast to our results, which showed strong and permanent HFD-induced downregulation of Scd1 transcription, of similar magnitude to that previously reported in fat-fed rats . Differences in experimental protocols (e.g. HFD composition, duration of fat feeding) and genetic variation between strains can explain these discrepancies. We sequenced in the C57BL6J and 129S6 strains about 2 kb of the promoter and all exons and intron-flanking regions of the gene Scd1 (GenBank NM_009127) and found no variants that can explain altered Scd1 transcription regulation in 129S6 mice (data not shown). Although this effect may underlie NAFLD compensatory mechanisms, HFD-stimulated transcription of both Srebf1c and Abca1, a protein normally destabilised and downregulated by SCD1 , suggests that SCD1 inhibition may involve a factor uncoupling the coordinated regulation of Srebf1c and Scd1 transcription. This may be caused by coding variants in genes which are not differentially transcribed between fat-fed and control mice, including MLX interacting protein-like (Mlxipl, also known as ChREBP) and the nuclear receptor 1H3 (Nr1h3, also known as LXR-a).
The upregulation of Srebf1c transcription in 129S6 fat-fed mice is also relevant to the role of gut microbiota in NAFLD and insulin resistance . Srebf1c is overexpressed in experiments on the microbial colonisation of the gastrointestinal tract in germ-free mice . Gut microbiota could promote storage of triacylglycerol in hepatocytes and adipocytes by indirect modulation of cellular fatty acid uptake [14, 34]. There is growing evidence from studies in animals  and humans  that variation in gut microbial composition is an important factor in obesity and insulin resistance. The increased level of microbial-related metabolites (dimethylamine, trimethylamine, trimethylamine N-oxide) in the plasma of 129S6 fat-fed mice indicates important changes in gut microbiota that could contribute to NAFLD, obesity and insulin resistance. This model provides a useful system to test causal links between changes in gut microbiota and these pathologies.
We show here detailed changes in gene transcription and metabolic regulation that favour NAFLD and possibly subsequent phenotype progression involving inflammatory processes in 129S6 mice, despite activation of compensatory homeostatic mechanisms. Our results help establish consensus and strain-specific patterns of gene expression associated with NAFLD, obesity and diabetes, and identify gene targets for validation studies at the proteome level . With the emergence of transcriptomic data sets derived from NAFLD-susceptible strains [10, 13, 23], our results contribute to enrichment of the model proposed for disease pathogenesis  with novel insights into mechanisms regulating gene expression and novel diet-reactive molecular components, including predicted or poorly annotated genes. Further investigations in a genetic cross between 129S6 and a strain resistant to diet-induced NAFLD, obesity and diabetes are required to investigate causal relationships between altered transcriptomic and metabonomic patterns and disease traits .
The authors thank C. Hetherington (Biomedical Services, University of Oxford) for invaluable help with the implementation of the mouse colony. A. A. Toye, M. E. Dumas and C. Blancher contributed equally to this work. This work was funded by the Wellcome Functional Genomics Initiative BAIR (Biological Atlas of Insulin Resistance) (066786). D. Gauguier holds a Wellcome Senior Fellowship in Basic Biomedical Science (057733). S. P. Wilder is a recipient of a Wellcome Prize Studentship in Bioinformatics and Statistical Genetics.
Duality of interest
The authors declare that there is no duality of interest associated with this manuscript.