Dynamic modelling of an ACADS genotype in fatty acid oxidation – Application of cellular models for the analysis of common genetic variants

Background Genome-wide association studies of common diseases or metabolite quantitative traits often identify common variants of small effect size, which may contribute to phenotypes by modulation of gene expression. Thus, there is growing demand for cellular models enabling to assess the impact of gene regulatory variants with moderate effects on gene expression. Mitochondrial fatty acid oxidation is an important energy metabolism pathway. Common noncoding acyl-CoA dehydrogenase short chain (ACADS) gene variants are associated with plasma C4-acylcarnitine levels and allele-specific modulation of ACADS expression may contribute to the observed phenotype. Methods and findings We assessed ACADS expression and intracellular acylcarnitine levels in human lymphoblastoid cell lines (LCL) genotyped for a common ACADS variant associated with plasma C4-acylcarnitine and found a significant genotype-dependent decrease of ACADS mRNA and protein. Next, we modelled gradual decrease of ACADS expression using a tetracycline-regulated shRNA-knockdown of ACADS in Huh7 hepatocytes, a cell line with high fatty acid oxidation-(FAO)-capacity. Assessing acylcarnitine flux in both models, we found increased C4-acylcarnitine levels with decreased ACADS expression levels. Moreover, assessing time-dependent changes of acylcarnitine levels in shRNA-hepatocytes with altered ACADS expression levels revealed an unexpected effect on long- and medium-chain fatty acid intermediates. Conclusions Both, genotyped LCL and regulated shRNA-knockdown are valuable tools to model moderate, gradual gene-regulatory effects of common variants on cellular phenotypes. Decreasing ACADS expression levels modulate short and surprisingly also long/medium chain acylcarnitines, and may contribute to increased plasma acylcarnitine levels.


Introduction
Genome-wide association studies (GWAS) identified thousands of variants associated with diverse diseases [1]. Although inborn errors of metabolism provided numerous examples how genetics associates with metabolic traits [2], the mechanistic impact of common gene variants, often resulting from a mixture of related processes such as environmental exposures and identified loci [1,3,4], remains challenging. Expression and metabolic quantitative trait loci (eQTL, mQTL) can assist the identification of the underlying biological mechanisms that link a genotype to a phenotype, but this work requires proper cell models with the observed genetic background. However, the availability of human cell models for elucidating the functional role of common gene variants in human disease is limited [5].
Lymphoblastoid cell lines (LCL) are easily accessible in humans and frequently used and stored in biobanks as a choice of the patients' genetic material [6][7][8][9]. Here we tested if genotyped LCLs are a valuable tool to assess allele-dependent differences in metabolic pathways. We chose an association of C4-acylcarnitine plasma levels, the transport form of the mitochondrial fatty acid oxidation (FAO) product butyric acid, with the noncoding rs2014355T>C variant located in close proximity to the short chain acyl-CoA dehydrogenase (ACADS) gene locus [10][11][12][13][14] and in perfect linkage disequilibrium with the non-synonymous single nucleotide polymorphism (SNP) rs1799958G>A (c.625G>A, p.G209S). The gene encodes for an enzyme catalysing mainly the conversion of butyryl (C4)-CoA to acetyl (C2)-CoA in mitochondrial FAO [15]. To assess genotype-specific effects on FAO, we adopted an in situ assay of oleic acid oxidation [16] for use in donor LCLs. Moreover, we generated a tet-regulated ACADS knockdown in a human hepatocyte cell line, enabling assessment of dose-dependent effects of ACADS mRNA and protein expression levels in a cell type with known high FAO . The funder provided support in the form of salaries for author MS, but did not have any additional role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript. The specific roles of the author was related to Investigation, Methodology, and Writing -review & editing, as articulated in the ' author contributions section'.

Genotyping and sequencing of LCL cells
LCL immortalised with Epstein-Barr virus from three homozygous major and five homozygous minor allele carriers of the rs2014355T>C variant (age 55.5 ± 6.9 years, 6 female and 2 male donors) were obtained from the KORA cohort (Cooperative Health Research in the Augsburg Region), a population-based cohort of healthy adults from the Augsburg region in Southern Germany [18]. The Ethical Committee of the Bayerische Landesärztekammer has approved the study (#05004, 2 nd August 2005) and written informed consent was obtained. First, genotypes for the noncoding rs2014355 and the coding rs1799958 (NM_000017.2: c.625G>A; NP_000008.1:p.G209S) variants were verified by Sanger sequencing on an ABI3730 instrument (Applied Biosystems, Darmstadt, Germany) using standard protocols after PCR amplification of the SNP containing DNA fragment with genomic DNA as template (PCR primers: rs2014355-forward TGTCCTTAGGGTGACAGG; -reverse TCCTGCATCACTG CCGTT; rs1799958-forward TGGGCTGCTGTCATTTCT; -reverse AGTCCTCAAAGATGAGG TT). Additional sequencing of all ACADS exons 1-10 confirmed rs2014355 and rs1799958 genotypes and identified the synonymous variants rs3914 (c.321T>C; p.R107R) and rs3915 (c.990C>T; p.R330R). In all analysed LCLs these variants were found to be in perfect linkage disequilibrium. Moreover, sequencing confirmed the absence of any other mutation in the ACADS coding region in the here analysed LCL cells.

Plasmid construction
The optimal shRNA sequence for the ACADS knockdown system was identified using the pVal shRNA Validation Platform RNAiONE (Sirion Biotech, Martinsried, Germany, http:// www.sirion-biotech.com/pval_shrna_validation_platform.html,). In brief, a panel of 10 bioinformatically selected shRNA sequences with high predicted on-target activity was evaluated for activity in a cell model using the RNAiONE shRNA selection vector pVal. The best shRNA sequence showing 91% knockdown activity was used in subsequent studies. Both, shRNA revealing the best knockdown efficiency and a non-target shRNA (CAACAAGATGAAGAG CACCAA) were cloned into lentivirus inducible one-vector shmir platform (Sirion Biotech, Martinsried, Germany). Stable Huh7 cell pools (shACADS Huh7 and shNTC Huh7) were generated by packaging, transduction and stable integration of the tet-on expression vector and subsequent antibiotic selection by Sirion Biotech.

Doxycycline treatment
To achieve different levels of ACADS expression, stable Huh7 cells transduced with the tetinducible shRNA for down-regulation of ACADS (shACADS) or control shRNA (shNTC) were incubated with 10 ng/mL doxycycline (dox, Sigma-Aldrich, Steinheim, Germany) for 5 days and with 5 ng/mL dox for 3 days, and the medium was changed after 3 days. Knockdown efficiency was examined on RNA and protein level using qRT-PCR and western blot analysis, respectively.

In situ assay of FAO in LCL
The in situ assay of fatty acid beta-oxidation was carried out according to the method described by Ensenauer et al. [16]. Cell-type specific adaptations included counting of LCLs immediately before the assay using an automated cell counter (Countess, Invitrogen, OR, USA). After permeabilisation of 5 x 10 5 cells/cell line (triplicates) with digitonin (10 μg/mL) (Sigma-Aldrich, Steinheim, Germany) for five min, LCLs were treated with 500 μL incubation buffer containing 100 μmol/L oleic acid (Sigma-Aldrich, Steinheim, Germany) for 160 min in a water bath at 37˚C (GFL 1083, Hilab, Düsseldorf, Germany) while shaking gently. Acylcarnitines were extracted from both incubation medium and cell lysates by protein precipitation/ desalting and subjected to solid-phase extraction. Acylcarnitines analysed by ESI-MS/MS were normalised to the corresponding citrate synthase activity.
Citrate synthase (CS) activity measurement

Loading of Huh7 cells with palmitic acid-BSA
Assay conditions for preparation of palmitic acid-BSA conjugates and loading of Huh7 cells with palmitic acid-BSA in Huh7 cells, using Seahorse technology, were adopted from the Seahorse user protocols (http://seahorsebio.com/resources/tech-writing/protocol-fatty-acidoxidation.pdf, Seahorse Bioscience Inc., North Billerica, MA, USA). Palmitic acid (Sigma-Aldrich, Steinheim, Germany) was bound to fatty acid free BSA (Sigma-Aldrich, Steinheim, Germany) (molar ratio palmitic acid: BSA = 6:1) for solubilisation. Briefly, sodium palmitic acid was solubilised in 150 mmol/L sodium chloride at 70˚C in a water bath; BSA was dissolved in 150 mmol/L sodium chloride at 37˚C with continuous stirring. Solubilised palmitic acid was added to BSA at 37˚C with continuous stirring. The pH was adjusted to 7.4 with 1 mol/L NaOH. Cells treated with dox were seeded in 6-well plates at a density of 250,000 cells / well and cultured in a 5% CO 2 incubator at 37˚C. After two days, growth medium (DMEM medium containing 10% FBS, 1% penicillin-streptomycin and 0 ng/mL, 5 ng/mL or 10 ng/mL doxycycline) was changed to assay medium (111 mmol/L NaCl, 4.7 mmol/L KCl, 2 mmol/L MgSO 4 , 1.2 mmol/L Na 2 HPO4, 0.5 mmol/L carnitine (all components from Sigma-Aldrich, Steinheim, Germany) at a final volume of 1.8 mL. After 60 min 0.2 mL 2 mmol/L palmitic acid-BSA was added. At baseline and after 7, 14, 21 and 28 min 20 μL supernatant was given on a 6 mm filter paper punch. Cells were washed with PBS and harvested by scraping in 300 μL ice-cold 100% methanol. Both, supernatant and cells were shock frozen in liquid nitrogen.

Measurement of oxygen consumption
The Seahorse Extracellular Flux Analyzer XF96 (Seahorse Bioscience, North Billerica, MA, USA) measures oxygen consumption rate (OCR) in a 96 well format by sensing changes in oxygen content in a < 3 μL volume above the plated cells with a fluorescence biosensor. The measurements are non-invasive and made in short and repeated intervals. An assay medium composed of 111 mmol/L NaCl, 4.7 mmol/L KCl, 2 mmol/L MgSO 4 , 1.2 mmol/L Na 2 HPO 4 , 0.5 mmol/L carnitine (Sigma-Aldrich, Steinheim, Germany) and pH 7.4 was used in the XF analysis. The cells were seeded in a collagen pre-treated XF96 96 well cell culture microplate (Seahorse Bioscience, North Billerica, MA, USA) at 10,000 cells/well in 80 μL of growth medium (see cell culture methods) and incubated 2 days at 37˚C in a humidified atmosphere of 5% CO 2 . Prior to assay, growth medium was removed and replaced by 140 μL of assay medium. The cells were preincubated under these conditions for 24h at 37˚C in air. After four consecutive baseline OCR measurements palmitic acid was injected as a BSA-palmitic acid conjugate in a final concentration of 200 μM. To ensure that palmitic acid was metabolised in FAO and its products were used for respiration, the inhibitory effect of etomoxir on CPT-I was used which blocks the intake of palmitic acid into the mitochondrial matrix. Etomoxir (Sigma-Aldrich, Steinheim, Germany) was injected after a further 28 min at a final concentration of 50 μM.

Extraction of acylcarnitines from Huh7 cell culture supernatant
Filter paper punches soaked with 20 μL cell culture supernatant flow-through were vacuum dried in a speed vac (Savant SPD 111V SpeedVac Concentrator, Thermo Scientific, Dreieich, Germany) for approximately 45 min. 100 μL 5 mmol/L NH 4 Ac in LC-MS methanol containing internal standard (see above) was added to the dried filter paper punches which were shaken for 30 min at full speed and room temperature (Thermomixer comfort, Eppendorf, Hamburg, Germany). The supernatant was transferred into a Millipore filter plate and filtered by centrifugation at 1,500 g for 20 min. Flow-through was collected in glass vials and stored at -80˚C until measurement.

Acylcarnitine analysis in supernatant and Huh7 cell extracts
Chromatographic separations of metabolites in supernatants and cell extracts were conducted on a ZIC-HILIC column (150 x 4.6 mm, 5 μm) (Merck, Darmstadt, Germany). A QTRAP 5500 (AB Sciex, Framingham, MA, USA) triple-quadrupole tandem mass spectrometer with Turbo V spray electron spray interface was used in positive ion mode for detection. Analyst and MultiQuant software (AB Sciex, Framingham, MA, USA) were used for data acquisition and data processing. Limit of quantification was defined as signal/noise ratio > 9. Acylcarnitine measurements were normalised to the protein amount per well, data of all single measurements are given in S1A-S1D Table. Statistical analysis All data are expressed as mean ± SD if not stated otherwise. ACADS mRNA expression of LCLs and Huh7 cells were normalised to the reference gene index. Fold changes were calculated using the ΔΔCt method. Differences of gene expression in LCLs and Huh7 cells were assessed by two-tailed one sample t-test. C4-acylcarnitine levels in LCLs from the metaboliteflux assay were compared using Mann-Whitney-U test. Kruskal-Wallis one-way ANOVA with Dunn's post hoc test was used to compare C4-acylcarnitine accumulation in shACADS and shNTC Huh7 cells. Data were analysed by two-way ANOVA with Bonferroni post hoc test for multiple comparisons against control to study the effects of different knockdowns over time.
To quantify differences in the dynamics of time-dependent metabolite flux FAO was modelled in a linear cascade of subsequent, irreversible first-order reactions using ordinary differential equations with mass action kinetics [19][20][21] using acylcarnitine concentrations as a reflection for acyl-CoA concentrations. Model simulations were compared to acylcarnitine data on log 10 -scale to account for log-normally distributed metabolite concentrations and reaction kinetics [22], resulting in model parameters optimally describing the measured data [23]. A detailed description of equations and model selection is presented in S1 Text. All statistical and quantitative dynamical modelling analyses were performed using GraphPad Prism5 (GraphPad software, La Jolla, CA, USA), MATLAB (R2012a, The Mathworks Inc., Natick, MA) and the Data2Dynamics software package [24].

Reduced ACADS expression levels and altered FAO rate in LCLs from subjects homozygous for the rs2014355 and the coding rs1799958 minor allele
First, we tested the hypothesis that common ACADS variants modulate ACADS enzyme function in FAO [10][11][12][13][14]. We measured acylcarnitine flux in LCLs genotyped for two lead SNPs identified by mQTL; the noncoding variant rs2014355T>C and the non-synonymous coding variant rs1799958G>A (c.625G>A, p.G209S) which are in perfect linkage disequilibrium (LD = 1.0, 1000 Genomes Phase 3 [25]; confirmed by sequencing in our LCLs). Absence of further ACADS non-synonymous mutations was confirmed by sequencing. We included LCLs homozygous for the major or minor allele of rs2014355T>C and rs1799958G>A (c.625G>A, p.G209S), respectively. An in situ assay of FAO with oleic acid loading revealed a significant 1.8-fold increase of total C4-acylcarnitine levels in LCLs homozygous for the minor allele ( Fig 1A). Next, we assessed ACADS mRNA and ACADS protein expression levels that may explain the reduced C4-acylcarnitine conversion found in LCLs from minor allele carriers. Both, mRNA ( Fig 1B) and protein levels (Fig 1C and 1D) were significantly reduced in LCLs homozygous for the minor allele as compared to controls homozygous for the major allele.

In vitro model of reduced ACADS expression in Huh7 hepatocytes reflects the genotype-dependent metabolic C4-acylcarnitine phenotype
Next, we assessed how gradual changes of ACADS expression levels modulate FAO in a cell type with high FAO-capacity-here Huh7 hepatocytes-which may contribute to the observed association of plasma C4-acylcarnitine levels with the common ACADS genotypes FAO [10][11][12][13][14]. We established an inducible ACADS knockdown in Huh7 cells as in vitro model enabling the investigation of gradually reduced ACADS expression levels on intra-and extracellular acylcarnitine levels. We first confirmed the validity of the Huh7 cell system as a model for FAO by assessing the oxygen consumption rate in Huh7 cells after palmitic acid loading (S1A and S1B Fig). Huh7 cells were stably infected using a lentiviral expression system generating a doxycycline-inducible shRNA ACADS knockdown (shACADS). Evaluating the efficiency of the doxycycline-inducible ACADS knockdown by RT-qPCR we found a significant reduction of ACADS mRNA levels in the intermediate (shACADS med , doxycycline 5 ng/mL) and maximal (shACADS max , doxycycline 10 ng/mL) shACADS Huh7 cells (82% and 84%, respectively, p < 0.001, Fig 2A) as compared to the null knockdown (shACADS null , doxycycline 0 ng/mL). In western blot analyses, we found gradual 70% and 93% reduction of ACADS protein levels for shACADS med and shACADS max Huh7 cells, respectively (Fig 2B and 2C). Doxycyclinetreatment of Huh7 cells stably transduced with a non-targeting control shRNA (shNTC) revealed no effect on both, protein and mRNA levels (Fig 2A-2C). We found no differences when comparing mRNA expression levels in shACADS versus shNTC cells without doxycycline treatment (p = 0.96, one-sample t-test, n = 4), proving that the lentiviral shRNA construct is solely active upon doxycycline treatment without any leakiness. Moreover, we found that Analysis of baseline intracellular acylcarnitine levels in the shACADS Huh7 cell model revealed a knockdown-dependent accumulation of C4-acylcarnitine ( Fig 2D). Maximal knockdown in the shACADS cells caused a significant 2.6-fold increase of baseline C4-acylcarnitine levels (p < 0.05) as compared to shACADS null . The shACADS Huh7 hepatocyte knockdown model in this respect reflects the rs2014355/rs1799958 locus associated changes in C4-acylcarnitine levels also presented in the LCL model. No significant differences were observed in shNTC cells upon doxycycline-treatment, proving ACADS-specificity of the observed effect ( Fig 2D). We found no ACADS-dependent significant differences in baseline levels of any other measured acylcarnitine in either cell model (time-point t = 0, S2 and S3 Figs). Moreover, reflecting the results from numerous GWAS FAO [10][11][12][13][14] we found a significant 2.1-fold decrease of the intracellular C3/C4-acylcarnitine ratio (p < 0.05, S1F Fig) in shACADS max cells as compared to shACADS null cells.

Model-based quantification of time-dependent intracellular acylcarnitine levels in shACADS cells after palmitic acid loading infers novel FAO signatures
To assess the effect of reduced ACADS enzyme expression levels on FAO kinetics, we leveraged the shACADS Huh7 cell model and measured the changes of intra-cellular acylcarnitine levels over time after incubation of cells with palmitic acid as FAO pathway substrate. We focused analysis on even-numbered intermediates of the FAO cascade, as palmitic acid is an even-numbered, saturated fatty acid. Intracellular acylcarnitine concentrations increased after exposure of cells to palmitic acid, suggesting increased FAO (Fig 3A, S2 and S3 Figs, S1 Table). In shACADS max cells, C4-acylcarnitine levels are further increased at all time points as compared to shACADS null knockdown, significant effects are observed at baseline and seven min after palmitic acid induction (p < 0.01 and p < 0.05, respectively, Fig 3A). Notably, the measured levels of longer chain acylcarnitines were lower in shACADS max knockdown at most time points as compared to shACADS null , with significant effects found for some C6-, C8-, C10-and C12-acylcarnitines time points (Fig 3A, S2 Fig) Next, we aimed to approximate ACADS knockdown-dependent changes of FAO-kinetics from the here measured intracellular acylcarnitines C16 to C2 profiles (as FAO intermediate metabolite proxies [19][20][21]). Therefore, we first fitted a simplified mathematical FAO model (S1 Text) to the experimental data ( Fig 3B, S4 and S5 Figs). We used this mathematical FAO model to calculate reaction rates for different steps of FAO and found that shACADS max results in a significant decrease (p < 0.05, Fig 3C) of the reaction rate for C4 to C2 conversion, reflecting the predominant role of ACADS for this reaction [26] and the observed increase of C4-levels. Supporting the sensitivity and specificity of the regulated shACADS cell model, also intermediate shACADS med knockdown was sufficient to decrease specifically the C4 to C2 reaction rate (p < 0.05, S6 Fig). Interestingly, we also found a significant increase of the reaction rates for C16, C10, C8 and C6-acylcarnitine conversion (p < 0.05, Fig 3C)

Extracellular shACADS cell acylcarnitine profiles suggest direct contribution to plasma mQTL phenotype
Finally, we determined extracellular C4-acylcarnitine levels in the cell culture medium supernatant (Fig 4). Extracellular baseline C4-acylcarnitine levels were higher than intracellular levels (65.4±17.2 and 26.9±9.7 nmol/g protein, Figs 4 and 3A, respectively, S1 Table), which may be attributed to an efflux of intracellular C4 before ACADS knockdown and palmitic acid loading. However, matrix effects cannot be excluded, challenging direct comparability of absolute levels measured in cell extracts vs. medium.
At baseline no significant ACADS-dependent differences of extracellular C4-acylcarnitine levels were found in shACADS null , shACADS max and shNTC cells (time-point t = 0, Fig 4), whereas basal intracellular C4-acylcarnitine levels were increased 2.65-fold (p < 0.01, timepoint t = 0) in shACADS max as compared to shACADS null cells (Figs 2D and 3A). In both shA-CADS null and shNTC cells no effect of palmitic acid loading on extracellular C4-acylcarnitine levels was observed (Fig 4), despite the observed intracellular increase (Fig 3A). Strikingly, in shACADS max cells we found a significant increase of extracellular C4-acylcarnitine levels as compared to shACADS null cells at each of the four measured time points after palmitic acid loading (p < 0.05 or p < 0.01; Fig 4). This ACADS-dependent, significant increase of extracellular C4-acylcarnitine (Fig 4) reflects the significant increase of intracellular C4-acylcarnitine levels ( Fig 3A).

Fig 3. Time-course of intracellular acylcarnitine levels in ACADS knockdown cells-comparison of distinct timepoints and kinetic differences.
(A) Intracellular acylcarnitine levels, representing acyl-CoAs with corresponding chain length, were extracted and measured before palmitic acid loading and 7, 14, 21 and 28 min after loading in shACADS knockdown Huh7 cells, i.e. shACADS null and shACADS max cells treated with 0 and 10 ng/ml doxycycline (dox), respectively, for shRNA induction. Shown are levels of C4-acylcarnitine as primary outcome, C14-, C6-acylcarnitines representing long-and medium-chain acylcarnitines, and C16-acylcarnitine (note that data for acylcarnitines C2 to C16, including the here presented, are given in S2 Fig). Values of four independent experiments are shown as mean ± SD (original data of single measurements are given in S1A Table). Time point specific comparison between shACADS null and shACADS max , were performed using t-tests; � p < 0.05 and �� p < 0.01. Results for all acylcarnitines in S2 and S3 Figs for cells treated with shACADS and shNTC, respectively. (B) Results from fitting of linear FAO model to the measured data (note that data for acylcarnitines C2 to C16 are given in S2 Fig). (C) FAO model based quantification of differences in acylcarnitine flux. shACADS null and shACADS max are compared by the ratio α = k shACADSmax / k shACADSnull , reaction rates k are derived from the FAO model fits. α-values from best model fits are represented as dots with respective 95% confidence intervals. (B+C) # indicates significant difference (p < 0.05)

Discussion
Defining the mechanistic basis of how common gene variants affect QTLs or disease states can be difficult and demanding. Moreover, GWAS-inferred variants are frequently noncoding and suggested to affect the expression level of nearby or other genes. Population-based mQTL analysis identified altered plasma levels of C4-acylcarnitine levels associated with a common ACADS locus [10][11][12][13][14] in high linkage with previously described mutations [27,28]. ACADS codes for an enzyme involved in mitochondrial FAO, an important energy metabolism pathway, and these ACADS locus variants may affect the rate of FAO. Here, we selected the ACADS locus for assessing this association at the cellular and molecular level using two cell models, donor-specific genotyped LCLs and an inducible lentiviral hepatocyte knockout cell line. Combining these two cell models allowed the generation of a data set enabling an in vitro assessment on how a gradual, genotype-specific modulation of gene-expression contributes to the FAO pathway and phenotypes.
Decreased plasma C3/C4-acylcarnitine ratios in subjects carrying the common rs2014355 variant minor allele at the ACADS locus have been reported (minor allele frequency = 26%; 1,000 Genome Phase 1; EUR [25]) [10,11]. This suggested that there may be an effect on FAO capacity via the ACADS enzyme which catalyses the conversion of C4-CoA into C2-CoA [15]. We analysed intracellular acylcarnitine levels after oleic acid loading of LCLs selected for the perfectly linked noncoding rs2014355 and non-synonymous coding rs1799958 (c.625G>A, p. G209S) variants. The specific accumulation of C4-acylcarnitine-as proxy-metabolite of the ACADS substrate C4-CoA [19,20,29,30]-in LCL homozygous for the minor allele, supports that GWAS [10,11] and candidate gene study [31] association of the ACADS locus with plasma C4-acylcarnitine level may be caused by reduced intracellular conversion of C4-CoA species between reaction rate k shACADSmax and k shACADSnull of the respective knockdown experiments. For details of dynamical modelling, see materials and methods, S4-S6 Figs and S1 Text.  Palmitic acid was added to induce FAO in shACADS and shNTC Huh7 cells. C4-acylcarnitine, which accumulates within the cell due to the ACADS knockdown, also accumulates in the supernatant. Gradual expression of shRNA was induced using 0, 5 and 10 ng/mL doxycycline. C4 was measured in supernatants before palmitic acid loading and after 7, 14, 21 and 28 min. Values of four independent experiments are expressed as mean ± SD. (original data of single measurements are given in S1C and S1D Table). �� : p < 0.01, � : p < 0.05 comparison between shACADS dox 0 ng/mL and shACADS dox 10 ng/mL (two-sample t-test).
Our data highlight that, going beyond eQTL analysis-frequently and successfully performed in LCL [35][36][37][38][39] -, also genotype-dependent differences in metabolic phenotypes related to a differentially expressed gene, such as FAO and ACADS, can be experimentally studied in the LCL cell model. Yet, the reported genotype associations with plasma acylcarnitine levels [10,11] may not result from lymphocytes, from which LCL are derived and which oxidise fatty acids although the main energy substrates are glucose and glutamine [40]. Rather, also tissues where FAO serves as substantive energy source, such as muscle, adipose tissue or particularly liver, where genotype dependent ACADS expression was found [41], may be a source of plasma acylcarnitine levels.
Here, we used gradual knockdown of endogenous ACADS in the human Huh7 hepatocyte cell line as cell model to further assess our findings in LCL. We observed that gradual reduction of ACADS expression levels leads to an accumulation of intracellular C4-acylcarnitine species. Moreover, we found a concurrent increase of C4-acylcarnitine levels in the medium along with the intracellular accumulation of C4-acylcarnitine. This increase in extracellular space was highly specific for C4-acylcarnitine and not observed for any other acylcarnitine and relates to the observation in human plasma with elevated C4-acylcarnitine levels associated with the rs2014355 genotype in GWAS [10][11][12] and the LCL phenotype. Notably, in shACADS knockdown without palmitic acid loading we found increased intracellular, but not increased extracellular C4-acylcarnitine levels, suggesting that acylcarnitine release from cells may be regulated by fatty acid stimulation or fatty acid availability in the cells and supporting the importance to consider time-course experiments. It has been proposed that the organic cation transporter 2 (OCTN2) may mediate the efflux of acylcarnitines from cells, but a recent study failed to confirm this [42]. Thus, the identity of plasma membrane transporters involved in the export of these metabolites remains to be determined. For such studies, the here described Huh7 cell model may be useful. Moreover, such a lentiviral approach is suitable for the assessment of metabolic alterations caused by genotype-specific alterations in mRNA levels in an appropriate biological context. The combination of such cells models can complement recent advancements in genome editing [43,44] and provides a practical, efficient analysis of genotype-phenotype relationships and underlying mechanisms.
How reduced ACADS expression modulates the overall kinetics of FAO is largely unknown. Further leveraging the hepatocyte shACADS model, we assessed time-and ACADS-dependent changes in acylcarnitine levels after exposure of the cells to palmitic acid as a substrate. We applied a mathematical FAO model tailored to our experimental data, as differences of measured metabolite and model complexity limits transfer of models described literature [21,45,46], and trimmed the initial general description of the linear FAO pathway down to the reactions essential to describe the observed data. Such an approach enables data analysis on pathway dynamics over time taking into account the reaction kinetics between fatty acids; a type of information that cannot be obtained from statistics on single metabolite alone. Our model is of course a simplified representation of the FAO pathway revealing knockdown-specific differences in metabolite dynamics, but not delivering kinetic rates. However, using a similar strategy we previously modelled plasma acylcarnitine levels in a human study with a starvation-induced increase in FAO resulting in reasonable approximations [47]. Comparing the dynamic changes resulting from Huh7 liver cell ACADS knockdown we obtained a significant decrease in the C4-related reaction rates, reflecting both, the rs2014355 genotypespecific ACADS gene expression and C4 levels in LCL and GWAS data [10,11,31]. We note that in Huh7 experiments we assessed palmitic acid C16:0, but in LCL oleic acid C18:1 requiring an auxiliary upstream oxidation step, yet after the first cycles of FAO oleic acid yields the same intermediates as palmitic acid. Further experiments are necessary for direct comparison on how different fatty acids may affect the observed effects. We did not observe a decrease of C2 levels as a product of the ACADS reaction. Acetyl-CoA or its carnitine derivative can also originate from a number of other pathways and thus may not be a good indicator of an altered FAO flux. The unexpected increase in medium-and long-chain carnitine reaction rates and decreased specimen levels may reflect the C4 accumulation with effects on upstream products or reactions. Changes in the protein levels of medium-and long-chain acyl-CoA dehydrogenases ACADM, ACADL and ACADVL causing these effects were excluded by Western-blot analysis leaving changes in enzyme activity as the likely cause. Further experiments are needed to assess such effects in other cell types than liver Huh7, such as the here analysed LCL or in other cell types with importance for FAO such as adipose tissue.
In summary, we have shown genotype-dependent changes of C4-acylcarnitine levels in LCLs homozygous for the minor allele of a common ACADS locus. Impaired FAO may be explained by both, reduced mRNA and protein expression levels which may be affected by different mechanisms of noncoding variants tagged by rs2014355 and the rs1799958A (p.209S) coding variant, respectively. Using the Huh7 hepatocyte cell line we observed that an impaired palmitic acid-induced FAO flux-induced by a specific, gradual reduction of ACADS expression levels-results in an intracellular accumulation of C4-, medium-and long-chain acylcarnitine species, and an elevated efflux of C4-acylcarnitines. Regarding the limited availability of relevant cell systems, i.e. genotyped primary human cells [5] essential for functional analysis, the cell models presented here may prove helpful to assess the functional consequences of common genetic variants. . mRNA (C) and Protein (D-E) were harvested after 3 or 5 days of treatment with 0, 5, and 10 ng/mL doxycycline, respectively. One-sample t-test for (C) and (D) revealed no significant effect. (F) Decrease of intracellular C3/C4-acylcarnitine ratio in shACADS max Huh7 cells. Intracellular C3-and C4-acylcarnitine measurement in doxycycline-induced Huh7 shACADS null and shACADS max cells (treated with 0 and 10 ng/mL doxycycline, respectively). Values of four independent experiments are expressed as box plots (Boxes extend from first quartile to third quartile; median is indicated as a horizontal line; whiskers are drawn equal to 1.5 times the interquartile distance). � = p < 0.05; two-tailed unpaired t-test. (PDF)

S2 Fig. Time courses of intracellular acylcarnitines after palmitic acid loading in shACADS
Huh7 cells. Palmitic acid was added to induce fatty acid oxidation in shACADS knockdown cells. shACADS null , shACADS med and shACADS max cells treated with 0, 5 and 10 ng/mL doxycycline (dox), respectively, for shRNA induction. Intracellular acylcarnitines, assumed to represent acyl-CoAs with corresponding chain length, were extracted and measured before palmitic acid loading and after 7, 14, 21 and 28 min. Values of four independent experiments are expressed as mean ± SD (original data of single measurements are given in S1A Table). ND: concentration not measured. Time point specific comparison between shACADS null and shACADS max using t-test with �� p < 0.01, � p < 0.05. (PDF)

S3 Fig. Time courses of intracellular acylcarnitines after palmitic acid loading in Huh7
cells with non-target shRNA. Palmitic acid was added to induce fatty acid oxidation in cells transduced with a non-target shRNA. shNTC null , shNTC med and shNTC max cells treated with 0, 5 and 10 ng/mL doxycycline (dox), respectively, for shRNA induction. Intracellular acylcarnitines, assumed to represent acyl-CoAs with corresponding chain length, were extracted and measured before palmitic acid loading and after 7, 14, 21 and 28 min. Values of four independent experiments are expressed as mean ± SD (original data of single measurements are given in S1B Table). ND: concentration not measured. Time point specific comparison between shNTC null and shNTC max using t-test with � p < 0.05. (PDF) S4 Fig. Illustration of the mathematical fatty acid oxidation chain model. In each FAO reaction step of palmitic acid loaded Huh7 cells the carbon chain is shortened and C2 is produced. Fundamental chain and influx reactions for C16, C14-, C8-and C4-acylcarnitine are described by reaction rates (k 16 ,. . .,k 2 ,k input ,k 14in ,k 8in ,k 4in ). For a detailed description of the model and the data-driven selection of influx reactions see methods section and S1 Text. (PDF)

S5 Fig. Model-based analysis of intracellular acylcarnitine time course data in cells with maximal ACADS knockdown.
Results from fitting the linear fatty acid oxidation model to the null (shACADS null ) and maximal (shACADS max ) ACADS knockdown data. Intracellular acylcarnitine levels, representing acyl-CoAs with corresponding chain length, were extracted and measured before palmitic acid loading and 7, 14, 21 and 28 min after loading in shACADS knockdown Huh7 cells, i.e. shACADS null and shACADS max cells treated with 0 and 10 ng/ml doxycycline (dox), respectively, for shRNA induction. Values of four independent experiments are shown as mean ± SD (original data of single measurements are given in S1A Table). # indicates significant difference (p < 0.05) between reaction rate k shACADSmax and k shACADSnull of the respective knockdown experiments (see also Fig 3C). were extracted and measured before palmitic acid loading and 7, 14, 21 and 28 min after loading in shACADS knockdown Huh7 cells, i.e. shACADS null and shACADS med cells treated with 0 and 5 ng/ml doxycycline (dox), respectively, for shRNA induction. Values of four independent experiments are shown as mean ± SD (original data of single measurements are given in S1A Table). (B) FAO model-based quantification of differences in acylcarnitine flux dynamics. shACADS med and shACADS null are compared by the ratio α = k shACADSmed / k shACADSnull , reaction rates k are derived from the FAO model fits. α-values from best model fits are represented as dots with respective 95% confidence intervals. (A+B) # indicates significant difference (p < 0.05) between reaction rate k shACADSmed and k shACADSnull of the respective knockdown experiments. For details of dynamical FAO modelling, see materials and methods, S4-S6 Figs and S1 Text. Compared to the null knockdown, in the intermediate knockdown reaction rate k 4 is significantly decreased, whereas reaction rate k 8 is significantly increased. (PDF) S1 Text. Description of the fatty acid oxidation model. (PDF) S1 Table. S1A Table Intracellular acylcarnitine levels in Huh7 with shACADS knockdown -original data. Acylcarnitines were extracted as described in material and methods from the Huh7 cell line shACADS (shACADSnull, med and max indicates no treatment or treatment with 5 and 10 ng/mL doxycycline, respectively) before (time point 0 min) and after loading of the cells with palmitic acid (time points 7-28 min). In brief, after chromatographic separation of metabolites in the cell extracts triple-quadrupole tandem mass spectrometer was used for detection of metabolites, analyst and MultiQuant software for data acquisition and data processing. Limit of quantification was defined as signal/noise ratio > 9. The measured acylcarnitine levels are presented in nmol per g protein for each experiment. S1B Table. Intracellular acylcarnitine levels in Huh7 cells with shNTC control knockdown-original data. Acylcarnitines were extracted as described in material and methods from the Huh7 cell line shNTC (shNTCnull, med and max indicates no treatment or treatment with 5 and 10 ng/mL doxycycline, respectively) before (time point 0 min) and after loading of the cells with palmitic acid (time points 7-28 min). In brief, after chromatographic separation of metabolites in the cell extracts triple-quadrupole tandem mass spectrometer was used for detection of metabolites, analyst and MultiQuant software for data acquisition and data processing. Limit of quantification was defined as signal/noise ratio > 9. The measured acylcarnitine levels are presented in nmol per g protein for each experiment. S1C Table. Extracellular C4-acylcarnitine levels in Huh7 with shACADS knockdown-original data. C4-Acylcarnitines in the cell culture supernatant from the Huh7 cell line shACADS (shACADSnull, med and max indicates no treatment or treatment with 5 and 10 ng/mL doxycycline, respectively) before (time point 0 min) and after loading of the cells with palmitic acid (time points 7-28 min). In brief, after chromatographic separation of metabolites in the cell culture supernatant triple-quadrupole tandem mass spectrometer was used for detection of metabolites, analyst and MultiQuant software for data acquisition and data processing. Limit of quantification was defined as signal/ noise ratio > 9. The measured acylcarnitine levels are presented in nmol per g protein for each experiment. S1D Table. Extracellular C4-acylcarnitine levels in Huh7 cells with shNTC control knockdown-original data. C4-Acylcarnitines in the cell culture supernatant from the Huh7 cell line shNTC (shNTCnull, med and max indicates no treatment or treatment with 5 and 10 ng/mL doxycycline, respectively) before (time point 0 min) and after loading of the cells with palmitic acid (time points 7-28 min). In brief, after chromatographic separation of metabolites in the cell culture supernatant triple-quadrupole tandem mass spectrometer was used for detection of metabolites, analyst and MultiQuant software for data acquisition and data processing. Limit of quantification was defined as signal/noise ratio > 9. The measured acylcarnitine levels are presented in nmol per g protein for each experiment. (XLS)