Aminoaciduric canine hypoaminoacidemic hepatopathy syndrome (ACHES) is defined by severe hypoaminoacidemia,1–3 hepatocutaneous-associated hepatopathy,4–8 and aminoaciduria.1,3 This is the most common cause of superficial necrolytic dermatitis (SND), a distinctive and painful skin condition, in dogs.4 This rare condition was first described in the English literature in 1986 as a diabetic dermatopathy.9 However, in some dogs, the distinct degenerative hepatopathy and other syndromic features precede the development of cutaneous lesions, favoring the use of ACHES as a more appropriate acronym. In dogs with SND, ACHES is synonymous with the traditional terminology of hepatocutaneous syndrome. In humans, SND is a rare condition associated with several metabolic disorders. It was first described in humans in 1942 and is most commonly associated with glucagonoma in that species.10–12 Therefore, while the typical etiopathogenesis of SND differs between species, low plasma amino acid (AA) concentrations are common to known forms of SND and are thought to impair healing at sites of cutaneous mechanical microtrauma.10
The most prominent metabolic alteration in dogs with ACHES is the reduction in most plasma AAs; however, the broad metabolic consequences of this finding remain unclear. Additionally, dogs with ACHES have a high incidence of diabetes mellitus,1 which further broadens the metabolic pathways that may be affected in ACHES. Although hypoaminoacidemia is believed to be the primary pathophysiological mechanism for SND, abnormal fatty acid metabolism and epidermal inflammation have also been postulated as contributing factors with systemic metabolic implications. Despite these speculations, a paucity of data exists investigating metabolic alterations in ACHES. Nevertheless, they have informed treatment modalities that primarily involve increasing protein and amino acid provisions through diet or parenteral AA administration.13 Therefore, a better understanding of metabolic perturbation in ACHES may also yield information that will lead to enhanced or new treatment strategies.
Amino acids are crucial intermediates in various cellular pathways, including the TCA cycle, the urea cycle, ketone body formation, and fatty acid and protein synthesis. The features of ACHES described above support a hypothesis that widespread metabolic perturbations affecting multiple pathways occur in affected dogs.1–3 Metabolomics is a powerful tool that can measure hundreds of metabolites in a biological sample, identifying biomarkers and affected metabolic pathways in disease. Increasingly, metabolomic profiling provides new insights into veterinary conditions, and its use in canine and feline medicine is expanding.14–20
Given the need for a better understanding of the metabolic perturbations in ACHES, we undertook an untargeted metabolomic analysis approach. The purpose of this study was threefold: (1) to identify novel metabolic abnormalities in dogs with ACHES, (2) to ascertain the metabolic pathways affected in ACHES, and (3) to identify candidate plasma and urine biomarkers.
Methods
Animals
We recruited dogs with ACHES for this prospective case-control study (Figure 1) with informed client consent between February 18, 2015, and April 18, 2018. The Cornell University Institutional Animal Care and Use Committee approved this protocol (2017-0094). Animal numbers were targeted based on previous studies case-control studies of untargeted metabolomics in small animals.15,17,20
Overview of the methodologic workflow in a case-control study conducted between February 18, 2015, and April 18, 2018, investigating the plasma and urine metabolomes of dogs with aminoaciduric canine hypoaminoacidemic hepatopathy syndrome (ACHES) using an untargeted metabolomic approach via liquid chromatography tandem mass spectrometry (LC-MS/MS). The Loftus Lab organized case recruitment and sampling. Stored samples were submitted to the Metabolomics Facility for LC-MS/MS, compound identification, and data filtering with Compound Discoverer 3.1 (CD 3.1) software. The filtered data set was analyzed by the Loftus lab with MetaboAnalyst software as described in the methods.
Citation: American Journal of Veterinary Research 84, 12; 10.2460/ajvr.23.08.0186
ACHES group—The previously described diagnostic criteria were applied for dogs with ACHES,1,3,13,21 and all but 1 dog in this study was included in previous reports. Briefly, cutaneous lesions were definitively diagnosed as SND by histopathology or clinically diagnosed by characteristic appearance and distribution. Hepatocutaneous-associated hepatopathy was definitively diagnosed by histopathology or clinically diagnosed by distinct ultrasound (nodular, “Swiss cheese” appearance) and compatible hepatic enzymopathy (predominantly alkaline phosphatase increases). A pattern of hypoaminoacidemia consistent with ACHES (generalized reductions and prominently amino acids involved in the urea cycle, cystathionine, 1-methylhistidine, proline, and hydroxyproline) was necessary for diagnostic inclusion. Dogs with concurrent regulated diabetes mellitus were permitted due to their association with ACHES. Exclusion criteria included: (1) glucagonoma, (2) failure to develop skin lesions in cases without a histologic diagnosis, or (3) skin or liver histopathology results were inconsistent with SND or HCH. Dogs with concurrent ACHES and diabetes mellitus (managed on insulin) were permitted. One author (J.P.L.) evaluated all medical records and assured a definitive diagnosis of ACHES.
CON group—We recruited healthy control dogs from the primary author's institution based on (1) an unremarkable physical exam by a veterinarian, (2) a history lacking indication of disease, and (3) an unremarkable CBC and serum biochemistry profile. These dogs had similar breed and age demographics to the ACHES group.
Sample collection—Blood samples were acquired by venipuncture of peripheral veins or through sampling catheters. The phlebotomist chose the site at their discretion. In most cases, blood was collected contemporaneously with diagnostic blood samples as close to the time of diagnosis as possible. Whole blood was collected into 4 mL sodium heparinized tubes (BD Vacutainer®). Tubes were centrifuged at 676 X g for 10 minutes at room temperature, yielding platelet-enriched plasma. Plasma was collected with transfer pipettes and stored in 2 mL plastic tubes. Urine was collected by cystocentesis, urinary catheterization, or by free-catch at the discretion of the treating veterinarian. No urine samples were grossly hematuric, and no dog had a concurrent urinary tract infection (based on inactive sediment, negative culture, or both). Plasma and urine samples were collected within 12 hours of each other, and stored at −80 °C until conducting the metabolomic analyses in August 2018.
Sample preparation
Samples were prepared for metabolomic analysis, similar to previous studies.22,23 Cold methanol (600 µL) was added to 200 µL of plasma sample, transferred into a clean microcentrifuge tube, and vortexed for 30 seconds. The samples were placed at 4 °C for 1 hour for protein precipitation, then removed from 4 °C and let sit at room temperature for 20 minutes. After that, the samples were centrifuged for 10 minutes at 16,200 X g at 4 °C. The supernatant was evaporated to dryness under a vacuum. Samples were re-suspended in 150 µL 60% acetonitrile before liquid chromatography (LC) tandem mass spectrometry (MS; LC-MS/MS) analysis. Sulfadimethoxine, 13 Pyruvic Acid, and 13C valine were added to all samples in the same concentration (100 ppb) for LC-MS/MS quality assurance.
Urine creatinine was measured enzymatically by the New York State Veterinary Diagnostic Laboratory with a commercial chemistry analyzer (Roche Cobas c501). Samples were diluted with purified (Milli-Q®, MilliporeSigma) water to a final volume to achieve the same creatinine concentration as the sample with the lowest urine creatinine concentration. Sulfadimethoxine, 13 Pyruvic Acid, and 13C valine were added to all samples in the same concentration (100 ppb) for LC-MS/MS quality assurance.
LC-MS/MS analysis and metabolomics profile
Untargeted metabolomic analysis was conducted similar to previous studies in Cornell Proteomics and Metabolomics Facility.22,23 Chromatographic separation (Vanquish UHPLC system with a 5 µm, 2.1 X 150 mm SeQuant ZIC pHILIC column) was performed at 35 °C coupled to a high-resolution mass spectrometer (Q Exactive™ Hybrid Quadrupole-Orbitrap, Thermo Fisher Scientific). The mobile phase consisted of (A) 10 mM ammonium acetate in water, pH = 9.8, 0.1% formic acid, and (B) acetonitrile. The gradient was as follows: 0–15 min, 90% to 30% solvent B; 15–18 minutes, isocratic 30% solvent B; 18–19 minutes, 30–90% solvent B; 19–27 minutes, 90% solvent B; followed by 3 minutes of re-equilibration of the column before the next run. The flow rate was 250 μL/minute, and the injection volumes were set to 2 μL. To avoid possible bias, the sequence of injections was randomized (Microsoft Excel, RAND function).
All samples were analyzed by positive (ESI+) and negative electrospray ionization (ESI−) modes in full scan MS mode. Nitrogen, as sheath, auxiliary, and sweep gas, was set at 50, 8, and 1 U, respectively. Data were acquired under resolving power of 120,000 (at m/z 200); automatic gain control target, 3e6 ions; maximum injection time, 100 ms; scan range, 67–1,000 m/z; spray voltage, 3.50 kV; and capillary temperature at 275 °C. ESI ± data-dependent MS/MS spectra were generated for quality control samples containing an equal amount of each sample. For identification, MS/MS data were acquired with a full scan followed by the top 15 MS/MS scans with resolving power of 15,000 (at m/z 200); automatic gain control target, 1e5 ions; maximum injection time, 50 ms; isolation window, 0.4 m/z; and stepped NCE 20, 30, and 40. The acquired data set, composed of full MS and data-dependent MS/MS raw files, was processed using commercial software (Compound Discoverer 3.1). An untargeted metabolomics workflow with putative compound identification through ChemSpider, Kyoto Encyclopedia of Genes and Genomes (KEGG), Small Molecule Pathway Database (SMPDB), and mzCloud databases were used for processing the raw data and for compound annotation. The software parameters for alignment were 5 ppm mass tolerance for the adaptive curve model and 0.5-min maximum shift for alignment. The software parameters for detecting unknown compounds were 5 ppm mass tolerance for detection, 30% intensity tolerance, 3 for the sensitivity and noise threshold, and 2 X 106 minimum peak height.
Metabolomics data filtering
Data filtering was conducted similarly to a previously reported approach.22 A commercial software package (Compound Discoverer 3.1) was used for identification, and any features not annotated were removed as previously described.22 The resulting mass/charge (m/z) results of the samples were analyzed with different databases: Chemspider, bioCyc, HMDB, Food metabolome, Phenol Explorer, mzCloud, and MassList. The initially identified molecules in plasma samples were filtered out through background subtraction and exclusion of false positive or repetitive features without MS2 spectra and removal of compounds not found in QC samples. In plasma, ESI− mode detected 6,905 features, and 466 compounds were annotated while ESI+ mode yielded 5,446 features, and 61 compounds were annotated. After removing duplicate and un-named compounds, 228 annotated compounds in plasma remained and were used for data analysis. In urine, ESI− mode detected 5,378 features, and 889 compounds were annotated while ESI+ mode detected 4,563 features, and 650 compounds were annotated. After removing duplicate and un-named metabolites, 323 compounds in urine remained and were used for data analysis. Nonzero peak intensity values were measured for all annotated metabolites in all samples. The Human Metabolome Database24 was searched for reference information on selected metabolites.
Statistical analysis
The proportion of males and females between ACHES and CON groups was tested by the Fisher's exact test with commercial online software. Dog ages and weights were compared between groups by the Mann-Whitney test with commercial software (Prism 9.0 or later, GraphPad). A P value of < .05 established significance.
Statistical analyses and data visualization of the LC-MS/MS data were performed on the peak intensity data derived from data filtering with online software (MetaboAnalyst 5.0) that was used for generating all figures.25 Plasma and urine data sets were analyzed separately. Data were filtered by interquartile range, log-transformed (base-10), normalized by the median, and Pareto data scaling was applied. Duplicate metabolites were removed from the positive mode data set. Raw P values were reported unless otherwise specified. Analysis parameters were set at the software default unless otherwise specified.
Fold-change and T-tests (1-factor statistical analysis) were conducted to identify metabolites that significantly increased >2-fold and were depicted by volcano plots. The P-value threshold was set at a .05 false discovery rate.
The sparse partial least squares-discriminant analysis (sPLS-DA) method (1-factor statistical analysis) was chosen for data reduction analysis.26 Settings were as follows: number of components = 5, variables per component = keep the same number (10). Validation method = 5-fold CV (2-component sPLS-DA error rates were < 15% for plasma and < 4% for urine). Random forest analysis: number of trees = 500, number of predictors = 7, randomness = on.
Metadata analysis included sex, age, and weight and were used as covariates for hierarchical cluster analysis depicted by heatmap, linear model correlation, and random forest analysis with the following settings.25 Heatmap (hierarchical clustering analysis): Data source = normalized data, standardization = autoscale features, distance measure = Euclidean, clustering algorithm = Ward, sample arrangement = blanked. We selected “display top 75 based on average” to avoid potential bias from displaying top metabolites by smallest P values. Linear model settings were as follows: Based on limma. Primary metadata = group, reference group = control; contrast = all contrasts (ANOVA style); covariates (control for) = age, sex, and weight; blocking factor = unspecified; P value cutoff = .05. Random forest analysis settings were as follows: primary metadata = group; metadata for predictors = age, sex, and weight; randomness = on.
We conducted enrichment analysis using the over-representation analysis (globaltest method)27 using the SMPDB. The significant metabolites identified by the linear model adjusting for covariates described above were used. All compounds in the library were selected for the reference metabolome.
We conducted both multivariate and classical univariate biomarker analysis to identify the best independently performing biomarkers. The multivariate analysis was conducted using random forest analysis (for classification and ranking method). Settings for univariate methods were as follows: zero or missing values were replaced by 1/5 of the min positive value for each variable. Optimal cut-off = closest to left corner. Show 95% confidence band selected.
Results
Patient Demographics
The ACHES group (n = 19) included the following breeds: mixed breed (n = 4), Shetland Sheepdog (n = 4), Maltese (n = 2), Schipperke (n = 2), Labrador Retriever (n = 2), West Highland White Terrier (n = 2), Cocker Spaniel (n = 1), German shepherd dog (n = 1), and Shih Tzu (n = 1). The median age was 10.5 years (range = 5–14; age not reported in one dog), and the median weight was 12.7 kg (range = 5–40.5). Dogs’ sexes were as follows: castrated male (n = 13) and spayed female (n = 6). Two ACHES dogs were diabetic at the time of sample collection (ACHES7 and ACHES13). Apparently healthy dogs comprised the CON group (n = 9). Breeds included: Cocker Spaniel (n = 2), Maltese (n = 2), Shetland Sheepdog (n = 2), Siberian Husky (n = 1), mixed breed (n = 1), and Schipperke (n = 1). The median age was 8 years (range = 4–10.5), and their median weight was 10.2 kg (range = 3.5–25.7). The CON group comprised castrated males (n = 7) and spayed females (n = 2). The sex proportions and weights were not statistically different between the ACHES and CON groups; however, age was significantly different between groups (P = .02).
Plasma metabolomic profiles
Fold-Change and T-test analysis showed 67 compounds higher and 46 lower, significantly greater than 2-fold in ACHES dogs (Figure 2). Data reduction of significant compounds by the sPLS-DA method revealed robustly distinct clustering of ACHES dogs from CON dogs. Additionally, hierarchical cluster analysis of plasma metabolites demonstrated segregation by the group but no consistent trend in clustering by age, sex, or weight. Because we detected a significant difference in ages between groups and age, sex, and breed or size (weight) can affect the canine metabolome,28 we adjusted for these covariates in additional analyses. After adjusting for covariates, the linear model analysis identified 111 (vs 107 before adjustment) significant plasma metabolites (Supplementary Table 1). The supervised machine learning technique of random forest analysis (RFA) was used to identify plasma metabolites that best-distinguished ACHES from CON dogs. After adjusting for covariates, RFA resulted in an out-of-bag error rate of 0.0714 and separate clustering was present by principal component analysis (Supplementary File 1). Metabolites that were the best RFA model predictors included N-(1-ethoxy-1-oxo-4-phenyl-2-butanyl)alanylproline, pyruvic acid isomer, cis-4-hydroxy-D-proline.
Analyses and data visualization of plasma metabolomes in aminoaciduric canine hypoaminoacidemic hepatopathy syndrome (ACHES) in a case-control study. (A) Volcano plot of plasma metabolites significantly (P < .05, with false discovery threshold of .05) higher (red) or lower (blue) greater than 2-fold without adjusting for covariates in ACHES dogs (n = 19) compared to healthy control dogs (n = 9) are depicted by single dots. (B) Sparse Partial Least Squares-Discriminant Analysis (sPLS-DA) demonstrating clustering by group without adjusting for covariates. Dots indicate individual dogs, and shading indicates the 95% confidence region. Covariate adjustment was applied for subsequent analyses (C–E). (C) Heatmap displaying the top 75 metabolites (based on average) were segregated into clades (by Euclidian distance and Ward methods). Dog groups, age, sex, and weight designations are depicted by colors in the legend. Dogs designated ACHES7 and ACHES13 were diabetic and indicated by asterisks. (D) Results of linear model (limma) analysis accounting for covariates (age, sex, and weight). (E) Random forest analysis after controlling for covariates (age, sex, and weight). Incomplete names in order are N-(1-Ethoxy-1-oxo-4-phenyl-2-butanyl)alanylproline, pyruvic acid isomer, cis-4-hydroxy-D-proline, glycerol 3-phosphate, N-(5-Amino-5-carboxypentyl)glutamic acid (saccharopine), (3α,5β,6α)-3,24-dihydroxy-24-oxocholan-6-yl β-D-glucopyranosiduronic acid, (-)-menthyl O-β-D-glucoside, 5’-O-β-D-glucosylpyridoxine, but-3-en-2-one, N-glycolylneuraminic acid, and tetradecan-1-yl acetate.
Citation: American Journal of Veterinary Research 84, 12; 10.2460/ajvr.23.08.0186
Urine metabolomic profiles
Fold-Change and T-test analysis showed 137 higher and 76 lower compounds, significantly greater than 2-fold in ACHES dogs (Figure 3). Data reduction of significant compounds by the sPLS-DA method revealed robustly distinct clustering of ACHES dogs from CON dogs. Additionally, hierarchical cluster analysis of urine metabolites demonstrated segregation by the group but no consistent trend in clustering by age, sex, or weight. After adjusting for age, sex, and weight, the linear model analysis identified 204 (vs 197 before adjustment) significant urine metabolites (Supplementary Table 2). We conducted RFA of urine metabolites, adjusting for age, sex, and weight, which resulted in an out-of-bag error rate of 0.0357 and separate clustering was present by principal component analysis (Supplementary File 1). Metabolites that were the best RFA model predictors included alanine-2 (ie, an alanine isomer), 3-sialyllactose, and bis-D-fructose 2’,1:2,1’-dianhydride.
Analyses and data visualization of creatinine-normalized urine metabolomes in aminoaciduric canine hypoaminoacidemic hepatopathy syndrome (ACHES) in a case-control study. (A) Volcano plot of urine metabolites significantly (P < .05, with false discovery threshold of .05) higher (red) or lower (blue) greater than 2-fold without adjusting for covariates in ACHES dogs (n = 19) compared to healthy control dogs (n = 9) are depicted by single dots. (B) Sparse Partial Least Squares-Discriminant Analysis (sPLS-DA) demonstrating clustering by group without adjusting for covariates. Dots indicate individual dogs, and shading indicates the 95% confidence region. Covariate adjustment was applied for subsequent analyses (C–E). (C) Results of linear model (limma) analysis accounting for covariates (age, sex, and weight). (D) Heatmap displaying the top 75 metabolites (based on average) were segregated into clades (by Euclidian distance and Ward methods). Dog groups, age, sex, and weight designations are depicted by colors in the legend. Dogs designated ACHES7 and ACHES13 were diabetic and indicated by asterisks. (E) Random forest analysis after controlling for covariates (age, sex, and weight). Incomplete names in order are 3-sialylactose, bis-D-fructose 2’,1:2,1’-dianhydride, (2S)-1-hydroxy-3-(phosphonooxy)-2-propanyl D-alanyl-D-alaninate, γ-aminobutyric acid (GABA), β-D-glucopyranuronic acid, N-α-acetyl-L-asparagine, indole-3-lactic acid, 5-Phospho-β-D-ribosylamine, 6-methylquinoline, 4-Indolecarbaldehyde-2, 1-(β-D-ribofuranosyl)-1,2-dihydropyrimidine, 1-methylguanosine, and 1-alanyl-l-glutamine. Alanine-2 is an alanine isomer.
Citation: American Journal of Veterinary Research 84, 12; 10.2460/ajvr.23.08.0186
Notable metabolites
Several significant compounds that the authors noted as potentially having salient relationships with ACHES were identified as significantly different in the plasma and urine of ACHES dogs. These include higher plasma metharbital and 2-Amino-3-(5-methyl-3-oxo-2,3-dihydro-1,2-oxazol-4-yl) propanoic acid (AMPA), higher urinary choline (without significantly higher plasma choline), and higher exogenous catecholamines.
Enrichment analysis
Several metabolic pathways were over-represented in their number of metabolites differentially quantified in ACHES biological samples. Enrichment analyses independently identified overrepresented SMPDB pathways in plasma and urine (Figure 4). Many of these pathways were related to AA metabolism, cellular energetics (eg, TCA cycle), and lipid biosynthesis (eg, phospholipid and fatty acid biosynthetic pathways).
Enrichment (over-representation) analysis visualization in aminoaciduric canine hypoaminoacidemic hepatopathy syndrome (ACHES) in a case-control study. The top 25 Small Molecule Pathway Database Pathways in plasma (A) and urine (B) with over-represented significant metabolites are displayed and ranked by P value.
Citation: American Journal of Veterinary Research 84, 12; 10.2460/ajvr.23.08.0186
Candidate biomarkers
Multivariate and univariate methods were used to identify novel plasma and urine candidate biomarkers. The best-performing biomarkers were selected through RFA, and then individually assessed by classical univariate receiver-operator characteristics (ROC). The best-performing model for plasma included 50 features, and the model correctly predicted 17/19 ACHES dogs, with 2 dogs predicted in the control group (Figure 5). Similarly, the best RFA model for urine metabolites included 50 features with similar group prediction performance (Figure 6). Although both models performed well, the model generated by urine metabolites performed slightly better than plasma. For this reason, we chose to display univariate ROC curves for the top 4 plasma and the top 6 urine features. The 4 top-performing plasma candidate biomarkers were pyruvic acid isomer, glycerol-3-phosphate, N-(1-ethoxy-1-oxo-4-phenyl-2-butanyl)alanylproline, and succinic acid. The 6 best-performing urine candidate biomarkers were alanine-2 (ie, an isomer of alanine), 3-sialyllactose, bis-D-fructose 2’,1:2,1’-dianhydride, (2S)-1-hydroxy-3-(phosphonooxy)-2-propanyl D-alanyl-D-alaninate, choline, and glycine anhydride.
Integrated results of multivariate and univariate biomarker analysis of plasma metabolites in aminoaciduric canine hypoaminoacidemic hepatopathy syndrome (ACHES). Multivariate analysis using random forest analysis (RFA) was conducted to identify the metabolites that were the best-performing biomarkers to distinguish ACHES (n = 19) and healthy control (CON; n = 9) dogs. (A) Receiver-operator characteristic (ROC) plot comparing RFA model performance with the indicated number of included features. (B) The predictive accuracy of RFA models based on included number of features demonstrating the RFA model with 50 features was the best performer with the corresponding ROC plot in the inset. (C) Class prediction of the 50-feature RFA model. (D) The 15 best-performing features from the RFA model (50 features) are ranked by selected frequency. (E) Univariate ROC curves and corresponding box and whisker plots to the right for the 4 top features selected by RFA. Values in the upper left corner of each plot depict the (semi-quantitative) cut-off value and, in parenthesis, specificity and sensitivity, respectively. AUC = area under the curve, parenthesis depict the AUC confidence interval. Shading depicts the ROC curve 95% confidence band. All P values were < .05.
Citation: American Journal of Veterinary Research 84, 12; 10.2460/ajvr.23.08.0186
Integrated results of multivariate and univariate biomarker analysis of urine metabolites in aminoaciduric canine hypoaminoacidemic hepatopathy syndrome (ACHES). Multivariate analysis using random forest analysis (RFA) was conducted to identify the metabolites that were the best-performing biomarkers to distinguish ACHES (n = 19) and healthy control (CON; n = 9) dogs. (A) Receiver-operator characteristic (ROC) plot comparing RFA model performance with the indicated number of included features. (B) The predictive accuracy of RFA models based on included number of features demonstrating the RFA model with 50 features was the best performer with the corresponding ROC plot in the inset. (C) Class prediction of the 50-feature RFA model. (D) The 15 best-performing features from the selected RFA model (50 features) are ranked by selected frequency. (E) Univariate ROC curves and corresponding box and whisker plots to the right for the 6 top features selected by random forest analysis. Values in the upper left corner of each plot depict the (semi-quantitative) cut-off value and, in parenthesis, specificity and sensitivity, respectively. AUC = area under the curve, parenthesis depict the AUC confidence interval. Shading depicts the ROC curve 95% confidence band. All P values were < .05.
Citation: American Journal of Veterinary Research 84, 12; 10.2460/ajvr.23.08.0186
Discussion
We conducted untargeted metabolomic profiling of dogs with ACHES to elucidate the comprehensive metabolic signature of this condition. Metabolomic profiling provides valuable insight into the metabolic pathways involved in disease and enables novel candidate biomarker identification. Metabolomic profiling of dogs with ACHES disclosed a distinct signature implicating the involvement of diverse biological pathways with widespread metabolic perturbations. Among these, several metabolites warrant additional comment due to hypothetical links to ACHES pathophysiology.
Several significant metabolites warrant additional comment due to potential links to ACHES pathophysiology. Metharbital (Figure 2) is a barbiturate, the same drug class as phenobarbital. Metharbital is no longer manufactured, and none of the dogs were treated with barbiturates. Therefore, we speculate this compound may be microbiome-derived as it has been previously identified as a metabolite from the microbe Xanthomonas perforans.29 Alternatively, this analyte signal may represent a chemically similar substance erroneously annotated as metharbital. In either case, this finding is interesting because of the historical association between phenobarbital and ACHES with SND development.30 Metharbital and phenobarbital metabolism and clearance by p450 cytochromes depend highly on functional hepatic mass and enzyme induction. Thus, finding a metharbital or metharbital-like signal may coincide with impaired hepatic detoxification of an environmentally derived compound, reflecting the functional impact of hepatocutaneous hepatopathy.
We also identified higher levels of 2-Amino-3-(5-methyl-3-oxo-2,3-dihydro-1,2-oxazol-4-yl) propanoic acid (AMPA, also α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid) in dogs with ACHES (Figure 2). We did not find evidence in the literature that AMPA is endogenously produced; therefore, we speculate it is also exposure-derived (environment or conceivably host microbiome). Interestingly, AMPA is the prototypical agonist of the AMPA receptor, which is part of the glutamate receptor family.31 It not only functions in neural transmission but is also involved in glucagon secretion in a rodent model.32 Hyperglucagonemia can provoke SND and reduce plasma amino acids. Therefore, altered glucagon homeostasis has been speculated to play a role in ACHES.1 While recent work has demonstrated lower glucagon levels in ACHES dogs,21 this does not necessarily entirely rule out a role for glucagon in ACHES pathophysiology. Lower glucagon concentrations could still be inappropriate in the face of hypoaminoacidemia and will require more work to determine if this is the case. Thus, AMPA may be pathomechanistically important if it contributes to inappropriate glucagon secretion.
Amino acids are vital intermediates in many metabolic pathways. We anticipated hypoaminoacidemia associated with ACHES would affect various AA synthetic and metabolic pathways, which was supported by our data. The liver is a major site of AA metabolism, which is influenced by glucagon signaling, creating a liver-alpha cell axis.33,34 We previously identified urea cycle AA intermediates reduced in ACHES dogs.3 Enrichment analysis corroborated that the urea cycle is affected. As previously speculated, this may be an important driver of ACHES hepatopathy, as urea cycle defects in human patients are associated with vacuolar hepatopathy.3,35 However, cytosolic vacuolar change in ACHES hepatopathy is degenerative and is phenotypically designated as a mixed type of cytosolic vacuolation. This involves a diaphanous nonmembrane-bound glycogen-type vacuolation and a membrane-bound macro- and microvesicular lipid vacuolation. Additional to the vacuolar and degenerative hepatocyte changes is the development of uniquely proliferative nodules of hepatocytes.1,6,8
Metabolomic profiling in the present study corroborates our previous documentation of reduced plasma AA concentrations essential for glutathione synthesis (lower L-cysteine, Figure 2, Supplementary Table S1) and urea cycle function (pathway analysis, Figure 4).1,3 We also identified dysregulations in several pathways that might contribute to the development of ACHES hepatopathy, a consistent phenotypic feature of ACHES (with or without cutaneous lesions). These include perturbations in fatty acid biosynthesis, glycerolipid pathways, and disruptions in several bioenergetic pathways.
We identified several significantly affected metabolic pathways (Figure 4). There were 3 main metabolic categories of over-represented pathways, AA metabolism and synthesis, cellular energy pathways, and lipid metabolism and synthesis. The breadth of the affected pathways corroborates our hypothesis that widespread metabolic perturbations exist in ACHES dogs. We previously identified abnormal concentrations of amino acids associated with glutathione metabolism and the urea cycle,3 corroborated by our pathway analysis. Two other cellular energy-related pathways significantly affected include the cardiolipin pathway and the glycerol phosphate shuttle. Cardiolipin is an essential component of the inner mitochondrial membrane, where it plays a crucial role in energetics, particularly in oxidative phosphorylation. Cardiolipin alterations and mitochondrial oxidative stress have been described in nonalcoholic fatty liver disease.36 Thus, perturbations in the cardiolipin pathway may be a novel finding if this is a contributing mechanism for hepatic alternations described in ACHES. Veterinarians have incorporated parenteral lipid solution administration in conjunction with AA solutions for dogs with ACHES and SND due to its use in people with SND to correct documented fatty acid deficiencies.37 While a previous study did not find a survival benefit to administering parenteral lipids to ACHES patients,13 the implication of lipid perturbations in this study may further support the practice until more data is available. Metabolomic profiles in ACHES dogs also implicated altered nucleotide synthesis.
Plasma pyruvic acid and a pyruvic acid isomer were higher in the ACHES group. Glucagon infusions attenuate plasma AA concentrations12 while concurrently increasing pyruvate concentrations.38 Pyruvate, the conjugate base, has been evaluated in diabetic patients and is variably higher in affected people, with 1 study finding normal concentrations in ambulatory patients.39 To the authors’ knowledge, studies measuring pyruvate in dogs with naturally occurring diabetes mellitus have not been conducted. However, pyruvate has been measured in pancreatectomized dogs, where increases in pyruvate were predominantly in response to exercise.40 The highest relative concentrations of pyruvic acid were not measured in the diabetic dogs in this study (Figure 3). However, measuring this biomarker in diabetic dogs without ACHES would be necessary to determine if it can distinguish ACHES from diabetes mellitus. Additionally, a significant increase in plasma succinate levels in ACHES implicates impaired mitochondrial utilization for energy production.41 Still, like the circumstance with pyruvate, the complexity of this metabolic pathway precludes simple deductions from metabolomic profiling. Nevertheless, increased pyruvate would further support our previous finding that many dogs with ACHES had microcytic anemia associated with more severe cutaneous disease.1 We speculated this likely reflects protein-calorie malnutrition, such as in kwashiorkor, where pyruvate is also increased.42
Significantly higher urine alanine isomer (alanine-2) and alanine levels in this study contradict findings in our prior investigation of urine AA profiles in ACHES-affected dogs.1,3 This discordance likely reflects methodologic differences in AA profiling, which uses ion exchange liquid chromatography (Biochrom 30 Amino Acid Analyzer). The previous reports described quantitative methodology only measuring urine AA moieties without identifying multiple isomers. D-alanine, which is derived from bacteria, may have contributed to measured urinary alanine isomer levels.43 If true, this may also suggest fecal microbiome differences in ACHES dogs.
Higher urinary choline was observed in the absence of higher plasma choline. Although this may reflect physiologic differences in plasma vs urine analytes (ie, plasma provides a sample reflecting a single point in time while urine reflects excretion over time), this finding may suggest altered renal choline handling. Renal choline transport in dogs is bidirectional.44 Therefore, it is possible that dietary choline is higher in ACHES cases and greater urinary excretion occurs to maintain total body choline homeostasis. Alternatively, higher urinary choline could result from decreased expression or function of the choline transporter, SLC44A.45 Several affected pathways, such as cysteine and methionine metabolism, are also relevant to choline metabolism. In a mouse model, feeding a choline and methionine-deficient diet resulted in microvesicular hepatic steatosis and hepatocellular proliferation that often gave rise to hepatocellular carcinoma.46 Although low plasma choline levels were not detected in this cohort of ACHES dogs, it is conceivable that reductions in methionine and other metabolites related to choline metabolism could be specific contributors to ACHES hepatopathy. Furthermore, 2 authors (J.P.L. and S.A.C.) have encountered a suspicious proportion of ACHES cases that developed hepatocellular carcinoma.13 It is intriguing to speculate that these metabolic perturbations may drive the development of ACHES hepatopathy and, in some cases, predispose afflicted dogs to hepatocellular carcinoma development.
The sample size in the present study was modest; however, this is a rare condition with a metabolically robust phenotype. A larger, longitudinal study is needed to corroborate our findings and to further delineate metabolic signatures between subsets of ACHES-affected dogs, with subcategorization based on the presence or absence of cutaneous lesions or diabetes mellitus or before and after treatment. Additionally, healthy dogs as the control group precludes the discovery of metabolomic changes and biomarkers unique to ACHES as this would require comparisons to dogs with conditions similar to ACHES. Thus, a targeted approach evaluating candidate biomarker performance in distinguishing ACHES dogs from dogs with similar conditions is necessary to determine their diagnostic utility. In this study, we permitted phenotype heterogeneity among cases. For example, diabetes mellitus and skin lesion status were not used to subcategorize cases because of the small number of studied dogs. Additionally, due to the small number of dogs of each breed, we used size as a surrogate for covariate analysis. Although in 1 study, size was not a significant factor affecting the canine plasma metabolome, size, and breed did correlate,28 prompting us to use weight as a covariate. Sufficient dietary and environmental data were unavailable for all dogs, making conclusions about these factors impossible. However, given the paucity of information regarding exogenous sources of metabolites discussed above, it is unlikely that these data would have yielded more refined interpretations than we stated. Identified candidate biomarkers require additional work to evaluate their performance in the clinical setting and may be limited by a lack of routine testing availability. Normalization of urine samples for untargeted metabolomic approaches is somewhat controversial. Although creatinine normalization is cited as common,47 urine osmolarity or analytical approaches are gaining recognition as potentially better normalization alternatives. Finally, it is worth noting that comparisons to other canine metabolomics data must be made cautiously as untargeted metabolomics approaches are semi-quantitative, and sample type (eg, serum vs platelet-rich vs platelet-poor plasma) can influence metabolic signatures; thus, additional target-based verification analysis for the identified candidate biomarkers using a large scale of samples is needed.
The results of our study achieved our goals of improving our understanding of ACHES pathophysiology and identifying candidate plasma and urine biomarkers. Our data corroborate the anticipated widespread metabolic effect of hypoaminoacidemia in dogs with ACHES and implicate pathways that may be involved in ACHES pathophysiology that were previously unknown, for example, lipid metabolism and biosynthesis. Several metabolites, such as metharbital, are likely exogenously derived and plausibly related to the microbiome and suggest a role for the “exposome” in ACHES. We identified several candidate biomarkers that may yield improved diagnostic avenues in the future. Further studies are necessary to corroborate our findings, validate the clinical relevance of several metabolites of interest, and delineate the sequence of metabolic aberrations in this condition.
Supplementary Materials
Supplementary materials are posted online at the journal website: avmajournals.avma.org.
Acknowledgments
We thank the Cornell Institute of Biotechnology's Proteomic and Metabolomic Facility (Research Resource Identifier: RRID: SCR_021743). We thank the following individuals for providing case samples: Drs. Jennifer Adler, Kathy Arrington, Martha Cline, Mary Ann Crawford, Hathaway Fiocchi, Tiffany Green, Timothy Hui, Julie Landrum, Robert Mason, Heather Peikes, Jennifer Prieto, and Jonathan Schnier. We thank Seth A. Peng for his technical assistance. We also thank Dr. Elisa Benedetti for providing a critical appraisal of the manuscript.
Disclosures
The authors have nothing to disclose. No AI-assisted technologies were used in the generation of this manuscript.
Funding
A Hill's Pet Nutrition Resident Clinical Study Grant supported this study.
References
- 1.↑
Loftus JP, Center SA, Astor M, Miller AJ, Peters-Kennedy J. Clinical features and amino acid profiles of dogs with hepatocutaneous syndrome or hepatocutaneous-associated hepatopathy. J Vet Intern Med. 2022;36(1):97–105. doi:10.1111/jvim.16259
- 2.
Outerbridge CA, Marks SL, Rogers QR. Plasma amino acid concentrations in 36 dogs with histologically confirmed superficial necrolytic dermatitis. Vet Dermatol. 2002;13(4):177–186. doi:10.1046/j.1365-3164.2002.00295.x
- 3.↑
Loftus JP, Center SA, Lucy JM, et al. Characterization of aminoaciduria and hypoaminoacidemia in dogs with hepatocutaneous syndrome. Am J Vet Res. 2017;78(6):735–744. doi:10.2460/ajvr.78.6.735
- 4.↑
Byrne KP. Metabolic epidermal necrosis-hepatocutaneous syndrome. Vet Clin North Am Small Anim Pract. 1999;29(6):1337–1355. doi:10.1016/s0195-5616(99)50131-9
- 5.
Brenseke BM, Belz KM, Saunders GK. Pathology in practice. Superficial necrolytic dermatitis and nodular hepatopathy (lesions consistent with hepatocutaneous syndrome). J Am Vet Med Assoc. 2011;238(4):445–447. doi:10.2460/javma.238.4.445
- 6.↑
Hall-Fonte DL, Center SA, McDonough SP, et al. Hepatocutaneous syndrome in Shih Tzus: 31 cases (1996–2014). J Am Vet Med Assoc. 2016;248(7):802–813. doi:10.2460/javma.248.7.802
- 7.
DeMarle KB, Webster CRL, Penninck D, Ferrer L. Approach to the diagnosis of hepatocutaneous syndrome in dogs: a retrospective study and literature review. J Am Anim Hosp Assoc. 2021;57(1):15–25. doi:10.5326/JAAHA-MS-7072
- 8.↑
Gross TL, Song MD, Havel PJ, Ihrke PJ. Superficial necrolytic dermatitis (necrolytic migratory erythema) in dogs. Vet Pathol. 1993;30(1):75–81. doi:10.1177/030098589303000110
- 9.↑
Walton DK, Center SA, Scott DW, Collins K. Ulcerative dermatosis associated with diabetes-mellitus in the dog-a report of 4 cases. J Am Anim Hosp Assoc. 1986;22(1):79–88.
- 10.↑
Tierney EP, Badger J. Etiology and pathogenesis of necrolytic migratory erythema: review of the literature. MedGenMed. 2004;6(3):4.
- 11.
Norton JA, Kahn CR, Schiebinger R, Gorschboth C, Brennan MF. Amino acid deficiency and the skin rash associated with glucagonoma. Ann Intern Med. 1979;91(2):213–215. doi:10.7326/0003-4819-91-2-213
- 12.↑
Boden G, Rezvani I, Owen OE. Effects of glucagon on plasma amino acids. J Clin Invest. 1984;73(3):785–793. doi:10.1172/JCI111272
- 13.↑
Loftus JP, Miller AJ, Center SA, Peters-Kennedy J, Astor M. Treatment and outcomes of dogs with hepatocutaneous syndrome or hepatocutaneous-associated hepatopathy. J Vet Intern Med. 2022;36(1):106–115. doi:10.1111/jvim.16323
- 14.↑
Gookin JL, Mathews KG, Cullen J, Seiler G. Qualitative metabolomics profiling of serum and bile from dogs with gallbladder mucocele formation. PLoS ONE. 2018;13(1):e0191076. doi:10.1371/journal.pone.0191076
- 15.↑
Barko PC, Williams DA. Untargeted analysis of the serum metabolome in cats with exocrine pancreatic insufficiency. PLoS ONE. 2021;16(9):e0257856. doi:10.1371/journal.pone.0257856
- 16.
González-Arostegui LG, Rubio CP, Rubić I, et al. Changes in the salivary metabolome in canine hypothyroidism: a pilot study. Res Vet Sci. 2022;151:189–195. doi:10.1016/j.rvsc.2022.08.012
- 17.↑
Marsilio S, Chow B, Hill SL, et al. Untargeted metabolomic analysis in cats with naturally occurring inflammatory bowel disease and alimentary small cell lymphoma. Sci Rep. 2021;11(1):9198. doi:10.1038/s41598-021-88707-5
- 18.
Allaway D, Alexander JE, Carvell-Miller LJ, et al. Suitability of dried blood spots for accelerating veterinary biobank collections and identifying metabolomics biomarkers with minimal resources. Front Vet Sci. 2022;9:887163. doi:10.3389/fvets.2022.887163
- 19.
Adin DB, Haimovitz D, Freeman LM, Rush JE. Untargeted global metabolomic profiling of healthy dogs grouped on the basis of grain inclusivity of their diet and of dogs with subclinical cardiac abnormalities that underwent a diet change. Am J Vet Res. 2022;83(9):ajvr.22.03.0054. doi:10.2460/ajvr.22.03.0054
- 20.↑
Montague B, Summers A, Bhawal R, et al. Identifying potential biomarkers and therapeutic targets for dogs with sepsis using metabolomics and lipidomics analyses. PLoS ONE. 2022;17(7):e0271137. doi:10.1371/journal.pone.0271137
- 21.↑
Holm SM, Peng SA, Holter MM, Cummings BP, Loftus JP. Plasma concentrations of glucagon and glucagon-like peptide-1 are reduced in dogs with aminoaciduric canine hypoaminoacidemic hepatopathy syndrome versus healthy dogs: a preliminary study. Am J Vet Res. 2023;84(4):ajvr.22.10.0174. doi:10.2460/ajvr.22.10.0174
- 22.↑
Noordwijk KJ, Qin R, Diaz-Rubio ME, et al. Metabolism and global protein glycosylation are differentially expressed in healthy and osteoarthritic equine carpal synovial fluid. Equine Vet J. 2022;54(2):323–333. doi:10.1111/evj.13440
- 23.↑
Ledet MM, Harman RM, Fan JC, et al. Secreted sphingomyelins modulate low mammary cancer incidence observed in certain mammals. Sci Rep. 2020;10:20580. doi:10.1038/s41598-020-77639-1
- 24.↑
Wishart DS, Tzur D, Knox C, et al. Human Metabolome Database. Accessed March 28, 2023. https://hmdb.ca/
- 25.↑
Pang Z, Zhou G, Ewald J, et al. Using MetaboAnalyst 5.0 for LC–HRMS spectra processing, multi-omics integration and covariate adjustment of global metabolomics data. Nat Protoc. 2022;17(8):1735–1761. doi:10.1038/s41596-022-00710-w
- 26.↑
Lê Cao KA, Boitard S, Besse P. Sparse PLS discriminant analysis: biologically relevant feature selection and graphical displays for multiclass problems. BMC Bioinformatics. 2011;12:253. doi:10.1186/1471-2105-12-253
- 27.↑
Goeman JJ, van de Geer SA, de Kort F, van Houwelingen HC. A global test for groups of genes: testing association with a clinical outcome. Bioinformatics. 2004;20(1):93–99. doi:10.1093/bioinformatics/btg382
- 28.↑
Puurunen J, Ottka C, Salonen M, Niskanen JE, Lohi H. Age, breed, sex and diet influence serum metabolite profiles of 2000 pet dogs. R Soc Open Sci. 2022;9(2):211642. doi:10.1098/rsos.211642
- 29.↑
Jibrin MO, Liu Q, Guingab-Cagmat J, et al. Metabolomics insights into chemical convergence in Xanthomonas perforans and metabolic changes following treatment with the small molecule carvacrol. Metabolites. 2021;11(12):879. doi:10.3390/metabo11120879
- 30.↑
March PA, Hillier A, Weisbrode SE, et al. Superficial necrolytic dermatitis in 11 dogs with a history of phenobarbital administration (1995–2002). J Vet Intern Med. 2004;18(1):65–74. doi:10.1111/j.1939-1676.2004.tb00137.x
- 31.↑
Gouaux E. Structure and function of AMPA receptors. J Physiol. 2004;554(Pt 2):249–253. doi:10.1113/jphysiol.2003.054320
- 32.↑
Bertrand G, Gross R, Puech R, Loubatières-Mariani MM, Bockaert J. Glutamate stimulates glucagon secretion via an excitatory amino acid receptor of the AMPA subtype in rat pancreas. Eur J Pharmacol. 1993;237(1):45–50. doi:10.1016/0014-2999(93)90091-U
- 33.↑
Galsgaard KD, Winther-Sørensen M, Ørskov C, et al. Disruption of glucagon receptor signaling causes hyperaminoacidemia exposing a possible liver-alpha-cell axis. Am J Physiol Endocrinol Metab. 2018;314(1):E93–E103. doi:10.1152/ajpendo.00198.2017
- 34.↑
Richter MM, Galsgaard KD, Elmelund E, et al. The liver–α-cell axis in health and in disease. Diabetes. 2022;71(9):1852–1861. doi:10.2337/dbi22-0004
- 35.↑
Yaplito-Lee J, Chow CW, Boneh A. Histopathological findings in livers of patients with urea cycle disorders. Mol Genet Metab. 2013;108(3):161–165. doi:10.1016/j.ymgme.2013.01.006
- 36.↑
Paradies G, Paradies V, Ruggiero FM, Petrosillo G. Oxidative stress, cardiolipin and mitochondrial dysfunction in nonalcoholic fatty liver disease. World J Gastroenterol. 2014;20(39):14205–14218. doi:10.3748/wjg.v20.i39.14205
- 37.↑
Bach JF, Glasser SA. A case of necrolytic migratory erythema managed for 24 months with intravenous amino acid and lipid infusions. Can Vet J Rev Veterinaire Can. 2013;54(9):873–875.
- 38.↑
Yamamoto T, Moriwaki Y, Takahashi S, et al. Effect of glucagon on the plasma concentration of uridine. Metabolism. 1998;47(6):695–698. doi:10.1016/S0026-0495(98)90033-6
- 39.↑
Smith MJH, Taylor KW. Blood pyruvate and a-ketoglutarate in normal and diabetic subjects. Br Med J. 1956;2(5000):1035–1038. doi:10.1136/bmj.2.5000.1035
- 40.↑
Himwich WA, Himwich HE. Pyruvic acid in exercising depancreatized dogs and diabetic patients. J Biol Chem. 1946;165(2):513–519. doi:10.1016/S0021-9258(17)41164-1
- 41.↑
Chouchani ET, Pell VR, Gaude E, et al. Ischaemic accumulation of succinate controls reperfusion injury through mitochondrial ROS. Nature. 2014;515(7527):431–435. doi:10.1038/nature13909
- 42.↑
Whitehead RG, Harland PSEG. Blood glucose, lactate and pyruvate in kwashiorkor. Br J Nutr. 1966;20(4):825–831. doi:10.1079/BJN19660085
- 43.↑
Konno R, Oowada T, Ozaki A, et al. Origin of D-alanine present in urine of mutant mice lacking D-amino-acid oxidase activity. Am J Physiol. 1993;265(4 Pt 1):G699–703. doi:10.1152/ajpgi.1993.265.4.G699
- 44.↑
Vander AJ. Renal excretion of choline in the dog. Am J Physiol-Leg Content. 1962;202(2):319–324. doi:10.1152/ajplegacy.1962.202.2.319
- 45.↑
Hedtke V, Bakovic M. Choline transport for phospholipid synthesis: an emerging role of choline transporter-like protein 1. Exp Biol Med Maywood NJ. 2019;244(8):655–662. doi:10.1177/1535370219830997
- 46.↑
Ikawa-Yoshida A, Matsuo S, Kato A, et al. Hepatocellular carcinoma in a mouse model fed a choline-deficient, L-amino acid-defined, high-fat diet. Int J Exp Pathol. 2017;98(4):221. doi:10.1111/iep.12240
- 47.↑
Nam SL, de la Mata AP, Dias RP, Harynuk JJ. Towards standardization of data normalization strategies to improve urinary metabolomics studies by GC×GC-TOFMS. Metabolites. 2020;10(9):376. doi:10.3390/metabo10090376
