Amyotrophic lateral sclerosis (ALS) is a fatal neurodegenerative disease with death from respiratory failure within 2 to 4 years of symptom onset.^1,2 While the primary pathology in ALS occurs in the central nervous system,³ changes in the metabolome,^4–6 energy homeostasis,⁷ and body mass index (BMI)^8,9 are an integral part of the disease process.

Altered metabolism in ALS is complex and multifaceted; ALS patients experience global hypermetabolism¹⁰ with elevated resting energy expenditure.¹¹ Metabolic changes occur across various tissues, including brain, spinal cord, and muscle, and result from alterations in both glucose and lipid metabolism as well as impaired mitochondrial function with inefficient tricarboxylic acid cycle and oxidative phosphorylation activity.⁹ This impaired metabolic function in ALS manifests by altered levels in energy substrates and intermediates; poor energy production in ALS could also impact downstream processes, inducing broad changes to the metabolome overall, including to bioactive metabolites and lipids.⁹ In cross-sectional analysis, specific metabolites significantly correlate with disease status, based on the revised ALS functional rating scale (ALSFRS-R),⁵ suggestive that metabolic status may reflect functional status.

Moreover, hypermetabolism¹² and steeper BMI changes¹³ associate with faster disease progression and poorer survival, indicating that metabolome changes may occur over the disease course longitudinally. However, despite well-established associations between the metabolome and ALS,^4,5,9,14 most previous studies were cross-sectional. Our goal in this study was to overcome limitations of cross-sectional analyses and assess longitudinal changes in the ALS metabolome throughout the disease course, including the terminal time point. We hypothesized that plasma metabolite changes would track with disease progression, serving as potential biomarkers and offering new insights into ALS pathophysiology.

We performed untargeted plasma metabolomics early in the disease course and longitudinally 6 and 12 months later. In tandem, postmortem spinal cord and brain motor cortex were also analyzed in a participant subset, the first analysis, to our knowledge, of various central nervous system tissues in human ALS. In plasma, we generated models that associated metabolite abundance to ALS functional status by ALSFRS-R across the disease course. An interaction model leveraged baseline metabolites to predict change in ALSFRS-R and progression model identified metabolites linked to a 20 and 50% drop in ALSFRS-R relative to baseline score. We also performed sex interaction analyses since sex impact metabolism¹⁵ and ALS,^16,17 including metabolites.⁵ In tissue, mirroring disease progression, we assessed differential metabolites in the onset versus opposing spinal cord segment to identify species potentially linked to temporal changes in disease.

By leveraging longitudinal metabolomic profiles from a prospective cohort of individuals with ALS, our analysis identified several plasma lipid metabolites and lipid sub-pathways that correlated with disease course, along with select amino acids and xenobiotics. In spinal cord, select sphingomyelin and long-chain saturated and monounsaturated fatty acid species were nominally more abundant in the onset tissue, whereas select phosphatidylcholines and phosphatidylethanolamines were less abundant. Mendelian randomization identified impaired carnitine and short chain acylcarnitine metabolism as possible genetic determinants in ALS, along with various antioxidant derivatives. Our findings reveal longitudinal changes in the ALS metabolome that associate with disease progression, yielding potential new ALS biomarkers and therapeutic targets.

ALS pathophysiology is characterized by changes to the metabolome^4–6 and energy homeostasis,^7,9 manifesting clinically by correlation of BMI^8,13,26 and hypermetabolism¹² to disease progression and survival. Low BMI at diagnosis or BMI loss in the years leading up to diagnosis may presage rapid ALS progression.^8,26 Similarly, hypermetabolism in ALS adversely affects survival,^12,27 with elevated resting energy expenditure.²⁸ Thus, overall, metabolic and metabolomic changes may underscore ALS progression. Indeed, we^5,8 and others¹⁴ demonstrated that specific plasma and cerebrospinal fluid (CSF) metabolites, lipids, and/or metabolic modules correlate with cross-sectional ALS functional status^5,14 and BMI trajectories.⁸ Specifically, we found that ALS plasma was characterized by altered lipid, amino acid, and xenobiotic sub-pathways, with several lipid species, including sphingomyelins and long-chain acylcarnitines, correlated to cross-sectional ALSFRS-R.

Herein, we hypothesized that ALS progression would be similarly marked by longitudinal changes in metabolome. We examined individual metabolites as well as sub-pathways since the sum of their collective member metabolites contributes to biological processes. We extend our prior findings in this new longitudinal dataset comprising granular metabolomic and functional data. Across the continuum of disease, individual lipid species and/or lipid sub-pathways continue to remain prominent in all Models 1, 2, and 3, including sphingomyelins, fatty acids, plasmalogens, phosphatidylcholines, and lysophospholipids. Interestingly, fatty acid, but not sphingomyelin, sub-pathways remained prominent in sub-pathway analysis, possibly suggesting broadly impaired fatty acid metabolism in ALS, linked to functional status, but only alterations in specific sphingomyelin species. The two models that integrated changes to ALSFRS-R, Models 2 and 3, selected several fatty acid sub-pathways, especially long-chain saturated acylcarnitine species, which may suggest that impaired baseline fatty acid metabolism may be prognostic of disease progression, aligned with hypermetabolism in ALS.^12,27

We also build on prior work by analysis of differential metabolite species between onset versus second spinal cord segment, as a surrogate of ALS progression and as a terminal time point. This metabolomic analysis is, to our knowledge, the largest to date, in terms of sample size and the types of central nervous system tissue assessed.²⁹ We found that most differential metabolites were lipids; among them, a few sphingomyelin and long-chain saturated and monounsaturated fatty acid species were nominally more abundant in the onset segment, whereas phosphatidylcholines and phosphatidylethanolamines were nominally less abundant.

Although impaired metabolism in ALS, especially of lipids,^4,5,30 is relatively well-established, few prior studies were longitudinal.^31–34 A relatively small analysis of ALS CSF identified altered glycolytic metabolite abundance at 2-year follow-up by proton nuclear magnetic resonance, a method of limited resolution, so the study lacked granular metabolomic information.³² A mass spectrometry analysis of serum found ALS cases (n=40) were uniquely perturbed in sphingolipids and glycerophospholipids at 2-year follow-up versus controls, as herein, but the study lacked correlation to functional status. Finally, a relatively small mass spectrometry lipidomics analysis of serum from ALS (n=28) versus control (n=22) participants identified 14 lipid classes that changed with time in ALS, especially diacylglycerols,³¹ which we noted as a differential class herein. Although the investigators did not examine correlation of lipid abundance to ALSFRS-R, they found that diacylglycerol (18:1/18:1) correlated to neurofilament light chain levels,³¹ a marker of neuronal damage and progression in ALS. Collectively, the reported literature, albeit using small sample sizes, supports our current findings.

Besides tracking with disease progression, and potentially serving as biomarkers, metabolomic changes can also shed insight into ALS pathophysiology and potential treatment avenues. We previously reported altered β-oxidation metabolites in cross-sectional ALS plasma analysis, evidenced by partially metabolized acylcarnitine intermediates.⁵ Here, we noted the presence of long-chain acylcarnitines linked to ALSFRS-R in plasma and spinal cord tissue, now also supporting aberrant β-oxidation in ALS progression.^35,36 Evidence from mutant SOD1 mouse models also uphold altered β-oxidation and fatty acid metabolism, including in spinal cord.^37,38 Enhancing β-oxidation capacity by administering L-carnitine to SOD1^G93A mice slows motor function deterioration and extends life.³⁹ However, although the precursor acetyl-L-carnitine has been considered as a potential ALS treatment,⁴⁰ an idea supported by the current findings, it is not approved clinically. Moreover, our prior⁴¹ and current MR analysis indicates possible genetically determined causality from specific fatty acid, acylcarnitine, and carnitine species, further bolstering dysfunctional β-oxidation in ALS. Enrichment analysis persisted in selecting carnitine and fatty acid metabolism, underscoring the centrality of this metabolic pathway in ALS, along with sub-pathways comprising antioxidant derivatives and oxidized byproducts, possibly linked to oxidative stress, another ALS hallmark.³

We also previously reported differential plasma sphingomyelins in ALS versus control participants, linked to cross-sectional ALSFRS-R.⁵ We expand these findings here by showing that sphingomyelins further correlate with disease progression longitudinally in plasma and in onset versus second spinal cord segment, adding to the growing literature on this lipid class in ALS.⁴² Cross-sectional studies have noted altered sphingolipids, including sphingomyelins, in ALS plasma⁴³ and spinal cord,^44–46 corroborated by longitudinal studies in early⁴⁵ and symptomatic stages.^33,37 Modifying activity of enzymes involved in specific sphingolipid species biosynthesis, e.g., glucocerebrosides, alters ALS progression rate in mouse models.^45,46 Moreover, the occurrence of monogenic childhood-onset ALS with mutant serine palmitoyltransferase subunit 1⁴⁷ further supports a causative or early role for sphingolipid metabolism in ALS. Indeed, our MR analysis identified a sphingolipid, sphingomyelin (d18:2/24:1, d18:1/24:2), with potentially genetically determined elevated levels in ALS, which was shared with metabolites selected by Models 1 and 3, as well as the CNS analysis. The specific role of this sphingomyelin (d18:2/24:1, d18:1/24:2) in ALS remains unclear, but warrants further investigation.

In addition to fatty acids and sphingomyelins, we found plasma plasmalogens, phosphatidylcholines, lysoplasmalogens, lysophospholipids, and diacylglycerols and/or sub-pathways correlated with ALSFRS-R over time. In spinal cord, phosphatidylcholine and phosphatidylethanolamine metabolites were more abundant in the second segment. ALS changes in plasmalogens, phosphatidylcholines, lysophosphatidylcholines, and/or phosphatidylethanolamines have been observed in cross-sectional⁴⁴ and longitudinal assessments.^31,33,37 Plasmalogens,⁴⁸ phosphatidylcholines, and phosphatidylethanolamines⁴⁹ are cell membrane components that regulate various biological functions, e.g., mitochondrial function, membrane fusion, proliferation, survival, and cholesterol efflux. Phosphatidylcholines and phosphatidylethanolamines are synthesized from diacylglycerols through the Kennedy pathway,⁵⁰ rendering the pool of these bioactive lipids interconnected.

Model 1 also selected bile acid lipid sub-pathways, that may be indicative of impaired lipid absorption and metabolism in the gut; indeed, ALS patients may have an altered microbiome⁵¹ and some microbiome-derived metabolites in the current study, e.g., indolelactate, correlated with ALSFRS-R. In our previous gut microbiome-plasma metabolome correlation, gut microbial structure at the phylum and genera level differed in ALS versus control participants.⁴¹ In the integrated analyses, several microbiota genera/species strongly correlated to multiple metabolites, primarily lipids, including bile acids, which associated to functional ALS status and differentiated cases from controls.

Besides lipids, amino acid metabolism and xenobiotics were recurrent in all three plasma models but to a lesser extent in spinal cord tissue. Altered amino acid metabolism in ALS has also been reported by cross-sectional^52–54 and longitudinal^52,55 human and animal studies, including of histidine, arginine, and proline,⁵² as well as creatine metabolism.⁴³ The latter, creatine, however, is likely a byproduct of muscle wasting. Selected xenobiotic metabolites in baseline and longitudinal analysis belonged mostly to xanthine and benzoate metabolism and food component/plant sub-pathways, as previously seen in cross-sectional studies.⁵ Sub-pathway analysis further selected various drug sub-pathways. Notably, “drug, gastrointestinal” and “drug, respiratory” abundance increased with disease progression, likely reflecting symptom management of ALS patients.

Our study has some limitations. First, a limitation intrinsic to most human studies, is the tissue issue, since our longitudinal analysis was restricted to accessible blood samples, which may not necessarily reflect the disease process in nervous tissue or may instead reflect downstream processes. However, there is no method to longitudinally capture brain and spinal cord tissue, which was only represented at the terminal time point by the spinal cord and brain cohort. Nevertheless, metabolomics analysis of plasma, though non-organ specific, can still reflect complex biochemical processes occurring throughout the body, including impaired systemic metabolism in ALS. Second, we had some cohort attrition in our longitudinal analysis, frequent in ALS studies,⁵⁶ which may have biased our findings. Third, our study involved ALS participants that had not fasted prior to providing blood collection since it was deemed unethical to request it. However, although diet influences circulating lipid levels, studies have shown that a plasma lipidomics signature of disease can occur independent of diet and/or fasting state.^57,58 Moreover, although we collected blood from ALS patients at the same time intervals, samples may have been collected at different time points in their disease course. Also, we assumed a linear rate of disease progression to impute missing ALSFRS-R scores, which is often, but not always, the case.⁵⁹ However, ALSFRS-R scores had to be imputed for only 3.9% of the data. A fourth limitation was the time window between a patient’s death and the time of autopsy, which may have allowed for postmortem changes in the metabolome. Additionally, the spinal cord and brain cohort was relatively small and did not include any tissues from healthy controls to facilitate direct comparisons to ALS, which instead relied on comparisons of the initial to the second affected spinal segment as a surrogate of disease progression. Finally, since this was an exploratory study that encompassed all detected metabolites, we lost power to detect significant correlations of specific metabolites and pathways to ALSFRS-R scores and relied on nominal p-values for Models 2 and 3, sex interaction analysis, and CNS metabolites. Future larger studies that analyze select target metabolites and pathways are needed to confirm our more exploratory findings.

Nevertheless, our study also had many strengths, including a longitudinal design using an agnostic metabolome-wide analysis platform that detects and identifies an extensive number of metabolites in plasma from a relatively large cohort of ALS participants. Additionally, the metabolomic platform included multiple standards to assure low day-to-day variability and ability to mitigate batch effects. In our data analysis pipeline, we performed a sex interaction analysis for the longitudinal plasma data, which is relevant to ALS, including differences to metabolites and lipids by male versus female patients;⁵ however, the spinal cord and brain cohort was too small to examine the impact of sex on central nervous tissue metabolomes. Finally, we also assessed both metabolites individually and cumulatively within sub-pathways. The former method identified robust single metabolites, whereas the second measured the sum of metabolites within sub-pathways, which considers minor but cumulative contributions from multiple metabolites, an appropriate approach for a heterogeneous disease like ALS.

In summary, we show for the first time, to our knowledge, that differences especially in plasma lipid species and sub-pathways, amino acids, xenobiotics, and various additional less frequently selected pathways, track with ALS progression, specifically with functional status by ALSFRS-R. Examination by lipid class pinpointed fatty acids and sphingomyelins, along with plasmalogens, phosphatidylcholines, and lysophospholipids. In spinal cord, sphingomyelin and long-chain saturated and monounsaturated fatty acid species were nominally more abundant in onset tissue, while phosphatidylcholines and phosphatidylethanolamines were less abundant. Overall, our findings suggest that metabolic and metabolomic changes may underscore disease severity and progression in ALS and offer new opportunities for both biomarker and therapeutic development.