Untargeted Metabolomics Yields Insight Into ALS Disease Mechanisms
Total Page:16
File Type:pdf, Size:1020Kb
Neuromuscular J Neurol Neurosurg Psychiatry: first published as 10.1136/jnnp-2020-323611 on 14 September 2020. Downloaded from ORIGINAL RESEARCH Untargeted metabolomics yields insight into ALS disease mechanisms Stephen A Goutman ,1 Jonathan Boss,2 Kai Guo,3 Fadhl M Alakwaa,1 Adam Patterson,1 Sehee Kim,2 Masha Georges Savelieff ,1 Junguk Hur,3 Eva L Feldman 1 1Department of Neurology, ABSTRACT cytoskeletal organisation and axonal transport.2 3 University of Michigan, Ann Objective To identify dysregulated metabolic pathways Metabolic abnormalities are also implicated, such Arbor, Michigan, USA 6–8 2 as amino acid, pyruvate and lipid metabolism. Department of Biostatistics, in amyotrophic lateral sclerosis (ALS) versus control University of Michigan, Ann participants through untargeted metabolomics. Increasing the pathological complexity is the gene- Arbor, Michigan, USA Methods Untargeted metabolomics was performed time- environment hypothesis of ALS, which states 3Department of Biomedical on plasma from ALS participants (n=125) around 6.8 that environmental exposures superimposed on a Sciences, University of North months after diagnosis and healthy controls (n=71). genetic risk profile trigger metabolic abnormalities Dakota, Grand Forks, North 9 Dakota, USA Individual differential metabolites in ALS cases versus that initiate neurodegeneration. controls were assessed by Wilcoxon rank- sum tests, Metabolites ultimately reflect the coordi- Correspondence to adjusted logistic regression and partial least squares- nated influence of genetics, epigenetics and tran- Dr Eva L Feldman, Department discriminant analysis (PLS- DA), while group lasso scriptomics, as well as serving as evidence of of Neurology, University of explored sub- pathway- level differences. Adjustment environmental exposure through xenobiotics. Michigan, Ann Arbor, MI 48109, parameters included sex, age and body mass index Metabolites are also a reflection of dysregulated USA; efeldman@ umich. edu (BMI). Metabolomics pathway enrichment analysis cellular processes and pathological state. For Received 23 April 2020 was performed on metabolites selected by the above instance, a recognition that oxidative stress is an Revised 3 August 2020 methods. Finally, machine learning classification ALS hallmark, through the detection of oxidised Accepted 6 August 2020 algorithms applied to group lasso-selected metabolites metabolites in biosamples, led to clinical trials were evaluated for classifying case status. of the antioxidant edaravone,10 which is now US copyright. Results There were no group differences in sex, age Food and Drug Administration (FDA)- approved for and BMI. Significant metabolites selected were 303 by treating ALS. Thus, despite the genetic and clinical Wilcoxon, 300 by logistic regression, 295 by PLS- DA heterogeneity of ALS, metabolites may be a unifying and 259 by group lasso, corresponding to 11, 13, 12 feature through shared cellular processes and envi- and 22 enriched sub- pathways, respectively. ’Benzoate ronmental contact. Metabolomics has emerged as a metabolism’, ’ceramides’, ’creatine metabolism’, ’fatty new frontier for understanding pathological mech- acid metabolism (acyl carnitine, polyunsaturated)’ and anisms, biomarkers and evaluating environmental ’hexosylceramides’ sub- pathways were enriched by all impact on disease.11 Metabolomics is the untar- methods, and ’sphingomyelins’ by all but Wilcoxon, geted, system- wide and simultaneous analysis of indicating these pathways significantly associate with all metabolites present in a sample. Its untargeted ALS. Finally, machine learning prediction of ALS cases nature allows agnostic evaluation of metabolites http://jnnp.bmj.com/ using group lasso- selected metabolites achieved the to identify diagnostic or prognostic biomarker best performance by regularised logistic regression with panels, understand complex pathophysiology, iden- elastic net regularisation, with an area under the curve of tify drug target candidates, or reveal potentially 0.98 and specificity of 83%. novel, hypothesis- generating avenues.12 In ALS, Conclusion In our analysis, ALS led to significant a small number of studies have identified altered metabolic pathway alterations, which had correlations metabolites in biofluids, but from metabolite data- 13–15 to known ALS pathomechanisms in the basic and clinical sets of around 400 metabolites or fewer. Our on September 24, 2021 by guest. Protected literature, and may represent important targets for future aim was to identify differential metabolites and ALS therapeutics. enriched pathways in ALS versus control partici- pants using a commercial untargeted metabolomics platform, which characterises an extensive number © Author(s) (or their employer(s)) 2020. Re- use of compounds. Metabolomics analysis could lead to permitted under CC BY- NC. No INTRODUCTION novel hypotheses, therapeutic targets and potential commercial re- use. See rights Amyotrophic lateral sclerosis (ALS) is a progressive, biomarkers for ALS. and permissions. Published 1 by BMJ. fatal neurodegenerative disease of motor neurons, characterised by complex genetics2 and disease METHODS To cite: Goutman SA, Boss J, mechanisms,3 as well as environmental influ- Participants and biosamples Guo K, et al. J Neurol ences.4 5 Although a handful of genes are strongly ALS and neurologically healthy control partici- Neurosurg Psychiatry Epub 2 ahead of print: [please linked to ALS, genetic causes are not known in the pants were enrolled at the University of Michigan 5 include Day Month Year]. majority of sporadic cases. However, ALS genes (UM), as previously reported. Briefly, all patients doi:10.1136/jnnp-2020- affect several shared cellular processes, including seen at the UM ALS clinic over 18 years and able 323611 proteostasis, autophagy, mitochondrial function, to communicate in English were asked to provide Goutman SA, et al. J Neurol Neurosurg Psychiatry 2020;0:1–10. doi:10.1136/jnnp-2020-323611 1 Neuromuscular J Neurol Neurosurg Psychiatry: first published as 10.1136/jnnp-2020-323611 on 14 September 2020. Downloaded from plasma shortly after diagnosis. Sex-matched and age-matched using the R package mixOmics.18 Score plots illustrate differ- control subjects also provided plasma. Participant demographics ences between case versus control groups. The variable impor- were collected, for example, sex, age, height, weight, ALS disease tance in projection (VIP) score of each metabolite, a weighted characteristics. Blood was drawn from participants that had not sum of the squared correlations between PLS- DA components been asked to fast, as it was deemed irresponsible to request this and metabolites,19 contributed significantly to the separation of from a large number of ALS participants, who have blood drawn case versus controls for VIP >1.20 in conjunction with their standard clinical care. Blood samples were collected following good clinical practice into lavender EDTA tubes and stored temporarily for a maximum of 2 hours at Group lasso 4°C. Tubes were then centrifuged at 2000g for 10 min at 4°C and Group lasso regressed all metabolites against case/control status the plasma supernatant was aliquoted into cryovials and directly simultaneously, adjusting for sex, age and BMI (see online transferred to the −80°C freezer for storage. supplemental table S4). Group lasso selects entire sub-pathways to simultaneously account for within- sub- pathway correlation Metabolomic profiling structure. The gglasso R package was used to implement group Plasma samples were shipped on dry ice to Metabolon (Durham, lasso with natural log- transformed and standardised metabo- North Carolina), and stored at −80°C at their facility until they lite data. Fivefold cross- validation was used to select the tuning were analysed for untargeted metabolomics profiling using ultra- parameter corresponding to a sparse model within one SE of high performance liquid chromatography- tandem mass spectros- the minimum cross- validation error. Once the tuning parameter, copy (UPLC- MS/MS) following their published protocol.16 17 corresponding to the group lasso penalty was finalised, group In brief, metabolites were extracted from plasma using meth- lasso was refit to the full dataset to obtain the final model. anol and recovery and internal standards were added to assess extraction efficiency and instrument performance, respectively. Metabolites were then analysed by reverse- phase UPLC- MS/MS Metabolism pathway analysis in both positive and negative ion mode and hydrophilic inter- Pathway enrichment analysis was performed using our in- house R action chromatography UPLC- MS/MS. A total of 1051 known package richR (https:// github. com/ hurlab/ richR/). Sub- pathways metabolites (see online supplemental table S1) were identified (115 in total), annotated by Metabolon and 1051 identified by retention time/index, mass- to- charge ratio and chromato- metabolites were used as background pathway and metabo- graphic data against authenticated standards, followed by data lite sets, respectively. The significant metabolites identified by curation to ensure correct chemical identification. Day- to- day unadjusted Wilcoxon, adjusted logistic regression, group lasso and PLS- DA were evaluated for over- representation in each sub- variability was accounted for by normalising metabolite levels. copyright. This was accomplished daily by equating the metabolite median pathway by modified Fisher’s exact test. A hypergeometric test across samples for that day