Genome-wide association studies (GWAS) investigate genomic variants (single-nucleotide polymorphisms (SNPs)) across different individuals to identify associations with a specific trait or phenotype. Over the past 15 years, GWAS have uncovered many variants associated with complex traits, but identifying the causal gene(s) is still a major challenge. More recently, integrating the GWAS dataset with other omics data has emerged as a promising approach to identify these causal genes (Mountjoy, E. et al. 2021).

Identifying causal genes for genetic diseases is not only of academic interest but pivotal for the development of therapeutic approaches. Traditional drug development is a knowledge-based approach that starts with a hypothesis based on limited data. The hypothesis is then put to the test in cell lines to qualify a potential target, before moving to preclinical research in animal models that only moderately represent human disease. More than 90% of all drug candidates fail in the preclinical stage (Sun et al. 2022), mostly because of safety concerns or lack of efficacy. In contrast, precision medicine uses a human-centric, hypothesis-free approach to drug discovery grounded in real-world data, using multiomics datasets from large populations evaluated with deep learning for target relevance and causality. Switching to this approach increased the success rate of Astra Zeneca’s drug development (from drug candidate to approval) from 4% to 19% (Morgan et al. 2018).

Why metabolomics?

Combining GWAS with metabolomics (mGWAS) shows particular promise. Metabolomics is used to analyze pathophysiological processes and map metabolites to biochemical pathways. When combined with GWAS datasets, metabolite profiling helps establish functional links to genes associated with clinical disease phenotypes. This way, mGWAS identifies genetically influenced metabotypes that correspond to phenotype-converting genetic variations. mGWAS qualifies genetic associations by effect size — indicating the phenotype-converting potential — and thus reveals potential drug targets for drug development.

Including metabolomics has an advantage over other omics in GWAS due to a marked p-value gain when using metabolite ratios. Metabolite ratios represent the flux through a biochemical pathway when a pair of metabolites is connected. For example, the ratio of an enzymatic reaction product to the source metabolite characterizes enzyme activity much better than either metabolite concentration alone. This comes with statistical benefits:

• Metabolite ratios increase the statistical power by reducing the overall biological variability.
• They reduce the impact of systematic experimental errors.

The p-gain statistic is a measure for whether a ratio between two metabolite concentrations carries more information than the two corresponding metabolite concentrations alone (Petersen et al. 2012). When this is the case, and metabolite concentrations are affected by the gene in question, including metabolite ratios in the GWAS analysis will lead to markedly lower p-values, highlighting the relevant associations.

Including metabolite sums instead of single metabolites can also make sense when studying lipids, as lipid-converting enzymes usually process several lipids with similar configuration such as similar fatty acid side chain lengths. The biocrates MetaboINDICATOR™ is a useful tool exploring metabolite sums and ratios relevant to health and disease.

Discovery of potential drug targets with mGWAS

Montasser and colleagues published a study on cardiovascular diseases that illustrates the potential of mGWAS (Montasser et al. 2022). They performed a GWAS on a discovery cohort of 650 individuals from the Old Order Amish founder population. The chance of identifying previously unknown disease associations is elevated in founder populations because certain variants are enriched to a higher frequency due to genetic drift. They discovered about eight million genetic variants.

Trying to identify the relevant ones is like looking for a needle in a haystack. To narrow down the results, they also performed lipidomics and integrated their concentration results on 355 lipid species from 14 different classes with their GWAS outcomes. This mGWAS analysis resulted in 12 significant associations. Seven of these were already known, and the five remaining significant associations represented novel associations between SNPs and cardio protection, cholecystitis, atherosclerosis, blood pressure, and inflammation, respectively. Integrating metabolomics in their GWAS thus resulted in five novel potential drug targets instead of hundreds of associations with uncertain relevance.

Naturally, the chances for identification of relevant variants increase with the number of metabolites covered. For example, metabolomics results obtained with the MxP Quant 500 XL kit, which covers up to 1,019 metabolites from 39 biochemical classes, bear a higher likelihood of success than a smaller kit or assay.

From weak associations to causality

The full power of mGWAS is realized when combined with Mendelian randomization. Mendelian randomization uses the measured variation in candidate genes to assess the causal effect a gene variant has on a disease by affecting metabolite concentrations or ratios, diminishing the need for an additional randomized controlled validation study. Mendelian randomization assumes that genetic variants are:

• Associated with the disease
• Not related to confounders
• Not associated with the disease through an alternative pathway in which the metabolite is not involved.

The causal effect is calculated as the variable ß̂ for each association:

Significant results in Mendelian randomization mean that a metabolite or a metabolite ratio is related to a specific disease via a specific genetic variant (Davies et al. 2018). Thus, that variant may become a target for that disease.

In other words, Mendelian randomization refines mGWAS data by testing for causality. The direct link between a disease or other phenotype and the corresponding genetic variant is usually quite weak, due to environmental influences. However, metabolites act as an intermediate phenotype involved in the development of the disease. The link between genetic variant and metabolite concentration is usually stronger than the association between genetic variant and disease because the relationship is more direct and independent of environmental exposures. The strength of an association is further increased when using metabolite ratios.

The same is true for the association of a certain metabolite concentration or metabolite ratio with a certain phenotype. Mendelian randomization uses the strong association between genetic variant and metabolite and the strong association between metabolome and clinical phenotype instead of the weak association between genetic variant and clinical phenotype to calculate causality. Because the environmental factors play a smaller role for these associations, they can be ignored for the causality calculation (Genetics meets metabolomics 2012). This also minimizes the risk of reverse causation, where the disease itself might appear to cause the differences in metabolite concentrations (Bowden et al. 2019; Le Chang et al. 2023).

A few examples for mGWAS studies combined with Mendelian randomization are provided in the following table:

A new general strategy for analyzing genomic variance

Given the robust causal relationships found between certain clinical phenotypes, metabolites, and genetic variants, combining GWAS and metabolomics into mGWAS, followed by evaluation with Mendelian randomization, should become the new standard for identifying genomic variance and potential therapeutic targets. The key steps should include the following:

1. In a population study with available genomic, metabolomic and longitudinal clinical information, select a phenotype of interest (e.g. myocardial infarction, type 2 diabetes, cancer).
2. Conduct mGWAS by jointly evaluating GWAS and metabolomic data (especially metabolite ratios) to uncover highly relevant associations and/or new functional relationships to underlying genetic variance.
3. Use Mendelian randomization to establish causality between genetic variant and phenotype. The shortlist of associations with proven causality can then be mined for feasible therapeutic drug targets.
4. Repeat the Mendelian randomization with subgroups defined by specific longitudinal clinical information, such as drug exposure, to identify causal relationships between treatment and outcome (phenotype), and enable prediction of drug intervention effects.

Where to start

Despite the undisputable advantages of combining GWAS with metabolomics and Mendelian randomization, it is still not a common approach. Among the more than 9000 GWAS that have been conducted, there are less than 70 mGWAS, and only about a dozen have incorporated Mendelian randomization, as shown in the table above. These numbers suggest that many existing GWAS could be combined with metabolomics to qualify results and derive causal links between gene variants and a clinical phenotype.

Many researchers performing GWAS may be unfamiliar with how to integrate metabolomics and conduct Mendelian randomization. To bring out this untapped potential, here are some helpful links:

• Prof. Karsten Suhre explains the principle behind mGWAS and Mendelian randomization very well in the book “Genetics meets metabolomics” (Genetics meets metabolomics 2012). For a preview, you can listen to Prof. Suhre talking about mGWAS and metabolite ratios in our podcast “The Metabolomist”.
• The mGWAS-Explorer lists details of 65 manually curated mGWAS studies, along with an mGWAS R package for download.
• biocrates’ multiomics data analysis service enables integration of metabolomics with genomics and other omic datasets. 

With these resources, you should be able to use metabolomics to significantly improve fine-mapping of genomic data to phenotypes and identify causally supported phenotype-converting therapeutic targets.

Interested in conducting a broad metabolomics analysis to be integrated with your GWAS? Find out more about the biocrates MxP® Quant 500 XL kit.

References

Mountjoy, E. et al.: An open approach to systematically prioritize causal variants and genes at all published human GWAS trait-associated loci. (2021). Nat Genet | DOI: 10.1038/s41588-021-00945-5.

Sun, D. et al.: Why 90% of clinical drug development fails and how to improve it? 2022. Acta Pharm Sin B | DOI: 10.1016/j.apsb.2022.02.002.

Morgan, P. et al.: Impact of a five-dimensional framework on R&D productivity at AstraZeneca. 2018. Nat Rev Drug Discov | DOI: 10.1038/nrd.2017.244.

Petersen, A-K. et al.: On the hypothesis-free testing of metabolite ratios in genome-wide and metabolome-wide association studies. 2012. BMC Bioinformatics | DOI: 10.1186/1471-2105-13-120.

Montasser, ME. et al.: An Amish founder population reveals rare-population genetic determinants of the human lipidome. 2022. Commun Biol | DOI: 10.1038/s42003-022-03291-2.

Davies, NM. et al.: Reading Mendelian randomisation studies: a guide, glossary, and checklist for clinicians. 2018. BMJ | DOI: 10.1136/bmj.k601.

Genetics meets metabolomics: From experiment to systems biology. 2012. New York, Heidelberg: Springer | DOI: 10.1007/978-1-4614-1689-0.

Bowden, J. et al.: Meta-analysis and Mendelian randomization: A review. 2019. Res Synth Methods | DOI: 10.1002/jrsm.1346.

Chang, L. et al.: mGWAS-Explorer 2.0: Causal Analysis and Interpretation of Metabolite–Phenotype Associations. 2023. Metabolites | DOI: 10.3390/metabo13070826.

Yin, X. et al.: Genome-wide association studies of metabolites in Finnish men identify disease-relevant loci. 2022. Nat Commun | DOI: 10.1038/s41467-022-29143-5.

Karjalainen, MK. et al.: Genome-wide characterization of circulating metabolic biomarkers reveals substantial pleiotropy and novel disease pathways. 2022. medRxiv | DOI: 10.1101/2022.10.20.22281089.

Surendran, P. et al.: Rare and common genetic determinants of metabolic individuality and their effects on human health. 2022. Nat Med | DOI: 10.1038/s41591-022-02046-0.

König, E. et al.: Whole Exome Sequencing Enhanced Imputation Identifies 85 Metabolite Associations in the Alpine CHRIS Cohort. 2022. Metabolites | DOI: 10.3390/metabo12070604.

Feofanova, EV. et al.: A Genome-wide Association Study Discovers 46 Loci of the Human Metabolome in the Hispanic Community Health Study/Study of Latinos. 2020. Am J Hum Genet | DOI: 10.1016/j.ajhg.2020.09.003.

Rueedi, R. et al.: Genome-wide association study of metabolic traits reveals novel gene-metabolite-disease links. 2014. PLoS Genet | DOI: 10.1371/journal.pgen.1004132.

Lotta, LA. et al.: Genetic Predisposition to an Impaired Metabolism of the Branched-Chain Amino Acids and Risk of Type 2 Diabetes: A Mendelian Randomisation Analysis. 2016. PLoS Med | DOI: 10.1371/journal.pmed.1002179.

Qin, Y. et al.: Genome-wide association and Mendelian randomization analysis prioritizes bioactive metabolites with putative causal effects on common diseases. 2020. medRxiv | DOI: 10.1101/2020.08.01.20166413.

Wang, Z. et al.: Genome-wide association study of metabolites in patients with coronary artery disease identified novel metabolite quantitative trait loci. 2021. Clin Transl Med | DOI: 10.1002/ctm2.290.

Panyard, DJ. et al.: Cerebrospinal fluid metabolomics identifies 19 brain-related phenotype associations. 2021. Commun Biol | DOI: 10.1038/s42003-020-01583-z.

The right tools for the job

Automated data interpretation

If you are used to working with automated workflows for sample or data processing, you may be living in hope that there is also one for data interpretation. Well… there isn’t. This makes data interpretation both beautiful and a little painful.

I often say that data interpretation is an art, which may not sit well with those aiming to describe the hard facts of biology. But this is how biology knowledge grows. We perform and analyze measurements, then interpret what the results (most likely) mean, sometimes adding a pinch of theory that remains to be proven or refuted by future works.

Happily, there are bioinformatic tools that help us get closer to our final story. The basic starting point is often the use of classical statistical methods to identify statistically significant differences between groups. More elaborate tools exist to harness the power of machine learning (data-driven tools), of pre-existing biological knowledge (biology-driven tools) or of structural similarities between metabolites (chemistry-driven tools).

Some of these tools are online software with easy-to-use graphical user interfaces, others are scripts provided by publications and research institutes. Many are free, and often come with tutorials and related publications to help us understand how best to use them.

All of them require data preparation (or pre-processing) to make sure your dataset will be properly handled. For example, you’ll need to prepare your data to remove low quality information, remove data not relevant to your analysis, add compatible identifiers, and more.

Stay up-to-date

In my experience, the first limiting factor to using any tool is to even know of its existence. In the same way that you want to be aware of the latest technological advances for the type of instrument you use, you need to know the latest approaches to prepare and handle your data type.

Some of the best strategies to stay on top of this ever-growing list of resources are to:

• discuss with colleagues
• attend conferences and webinars
• screen the literature for new methods
• read reviews focused on bioinformatic tools.

To help with this, there are people in the community generous enough to compile regularly updated lists of software and code, often shared as peer-reviewed publications. Once you find them in PubMed, don’t hesitate to set up an automated alert so you know when their next publication or review comes out. This is a great way to get regular updates on the latest developments in the field.

The worst case scenario is to discover the “perfect” tool right at the end of a project and wonder if you should repeat your work. To protect yourself from this nightmare, make sure to scour the literature for new tools that may be relevant to your research.

Choosing the right tool

The right tool for the job satisfies multiple needs and possibilities, not all of which are scientifically “honorable.” Ideally, we would choose tools that are the best fit for our data. In real life, we must consider cost, analysis time, learning curves, influence from colleagues and other factors.

We’re also limited by our ability to run certain tools. What should you do if you have no programming skills, but your perfect tool runs exclusively on R? It may be time to ask a colleague or even the developers of the code in question if they want to collaborate and check how well the tool works for your data.

In reality, compromise is likely.

Another factor that may influence your choice is comparability. If you want to be able to compare your results with your last three studies, you might opt for the exact same analytical workflow that you used before. If a new tool or algorithm has become available since running those past studies, it may be an idea to re-run all four with that new tool, to see if this brings you more information. It may be time-consuming, but it could provide the material for another publication on a new meta-analysis of past data, in addition to your current project.

Know your tools

A critical aspect of running any data analysis is to know what your bioinformatics tools can and cannot do and what type of data they take and don’t take. In short: know your tools’ limitations.

No tool can do everything. This is why combining several tools is often a great way to approach an interpretation project. We often run more tools than those described in the final paper. Don’t be afraid to perform many analyses if there is a sound reason to do so.

Whichever tool you use, imagine a warning label that reads, “no tool will write your paper for you.”
No tool will tell you exactly what’s happening in your biological system and how this fits into the larger context of biomedicine today. That’s a job for the human operating the tools.

Many tools can help identify the mechanisms that will form the core of your interpretation, but at the end of the day, you’ll still have to make the last connections yourself by studying your data in its biological context. This is where your knowledge of metabolism comes in.

Knowledge of metabolism

How much is enough to start?

If you have master’s level knowledge in a biology-related field, you have the understanding required to begin interpreting metabolomics. You don’t need to know everything, but you need to know enough about biology to make sense of the literature while you research your topic further.

A first year PhD student may have less chance of coming up with a thick plot of what’s happening in a metabolomics experiment than someone with 10 or 20 years’ experience, but that doesn’t mean that beginners shouldn’t play.

In fact, this type of work is exactly how experience is gained. Inexperience can be compensated for through hard work and automated tools that help focus the work on crucial metabolites or pathways.

While proficiency in molecular biology and metabolism is certainly a plus, it can be refined along the way. You can begin a metabolomics interpretation project without an extensive knowledge of metabolism and expect to know quite a bit more by the end of it.

Best source of knowledge

I would argue that working on a data interpretation project is possibly the best way to acquire knowledge about metabolism and biology in general. Here is an example of how that may play out in practice.

Let’s say you measured a broad panel of metabolites in blood samples from patients with Alzheimer’s disease and found that the levels of several bile acids are different compared to healthy controls. If your knowledge on this class of metabolites is a bit fuzzy, all you need to do is go to the literature to fill in the gaps with the latest knowledge on these molecules.

You may begin at the more generic level, using textbooks and web searches to get an overview of what’s known about these metabolites. You may discover that bile acids are synthesized primarily in the liver, but require the gut microbiome for certain steps.

You may learn about primary and secondary bile acids, about so-called “toxic” bile acids, and about their roles in digestion and as signaling molecules. Other metabolites may be less well documented, indicating that more research is needed to understand their roles in biology.

Next, you may look into what’s known about these metabolites in relation to your disease of interest, Alzheimer’s disease. You will find several publications describing changes in bile acids in Alzheimer’s. You’ll compare levels in different sample types (blood-derived, cerebrospinal fluid, and maybe even tissues) and see what others have concluded on the topic. You will also ask the literature for answers to questions such as “do bile acids cross the blood-brain barrier, and how?” or “can bile acids be synthesized in the brain too?”

After that, you may want to look into how these metabolites have been studied in other contexts. Much of the knowledge we have on metabolites is repeated across different fields of biomedicine, and what is common knowledge to a cardiology specialist, may be groundbreaking to a neurologist. So don’t hesitate to jump around at the beginning of your search to get inspired about the possibilities of your dataset.

Someone with more experience may already know (or think they know) the answers to these questions, but there’s nothing to stop you learning about this with a thorough literature search.

Finally, you will connect these pieces of information together to build the story of what may be happening in your experiment at the level of bile acids. You’ll look at how this relates to other metabolites in your dataset or with the disease in general, including known symptoms and risk factors.

By combining a tailored literature search with your own findings, you’ll create a storyline that becomes your version of what’s happening in the experiment. This story becomes the narrative you use to explain your work to yourself and to others. It will also form one of the many building blocks for your own knowledge of metabolism. That is, if you stick to your plan…

Perseverance

The final asset needed for data interpretation is perseverance. Perseverance may seem like a soft skill of little relevance, but it’s incredibly important if you ever want to complete your project.

It’s easy to progress through the first exciting steps of collecting information. But connecting the dots and coming up with an enticing story to tell about your experiment can be a bit more challenging.

You need perseverance to stick to an interpretation when the results are not as interesting as you expected, when there are too many things to discuss, or when you feel overwhelmed by the number of possible directions to pursue.

You certainly need perseverance when your supervisor makes a suggestion that is more than a suggestion that requires you to start over. The list of things that can get in your way is literally endless. When the road is already difficult to navigate, it can be hard to stick to the plan.

The challenge is compounded by the fact that there’s no protocol for data interpretation. You’re doing well to stay motivated when there’s no step-by-step process to tell you how to go from start to finish, or even when the interpretation is finished.

The STORY principle

In an attempt to answer this need, I’ve developed the STORY principle. This is a 5-step framework designed to help you construct your own plan, avoid the pitfalls commonly associated with data interpretation projects, and execute your interpretation as smoothly as humanly possible.

The STORY principle is based on my own experience and that of others who have experience planning and executing metabolomics data interpretation projects. I describe the five steps in detail in a book that will soon be published, and in a webinar that you can sign up for today.

I have fallen into all sorts of traps over the years, from interpretation quicksand, where you can’t seem to see the end of a project, to chasing butterflies, where other priorities distract you from your goal. These pitfalls are typical of omic data analysis and very much apply to metabolomics as well.

The STORY principle is my best advice on how to steer clear of these traps so you can be as efficient as possible in your data interpretation and enjoy it at the same time.

But here is perhaps the most important point: data interpretation can (and should) be fun!

It’s a great way to learn about metabolism and biology, and every project is a new opportunity to refresh your knowledge of the latest bioinformatic tools. Lastly, sharing the insightful interpretations that you made of your data is an entryway into the metabolomics community where you’ll get to connect with like-minded scientists.

Are you ready to get started?
If you want to learn more about strategies to plan and execute your data interpretation projects, register for my webinar on the STORY principle.

Let’s cut to the chase: there is no one-size-fits-all answer to this question. If you were looking for a quick answer, you might want to stop reading now. But if you’re curious about why “it depends” and how you can figure out a good starting point for your study, then read on.

While there’s no single correct approach to figuring out your sample matrix, there certainly are things that could be done wrong. Here, we dive into some of the advantages and disadvantages of the sample matrices you might consider.

The most important question

Before we discuss individual samples, let’s get an even more important question out of the way. What do you want to achieve with your metabolomics study? If you properly define the outcome you desire, the rest often falls neatly into place.

For example, if you want to learn about a biological process that cannot be investigated in human studies, perhaps because it would require multiple tissue samples from sites difficult to access, you have no choice but to move toward animal models or other model systems for the project.

If the goal is to find a stratification biomarker and human biofluids are available, then those biofluids are your logical matrix. If you have other data that seems to make a specific metabolic pathway relevant, such as from transcriptomics experiments, you’d clearly want a metabolomics approach that covers the respective pathways.

The scientific question defines the feasibility of metabolomics from a specific matrix.

An overview of benefits and challenges:

Cells and tissues (incl. novel model systems)

The benefit of using cell and tissue sample matrices is obvious. You can directly investigate the cell or organ type that you are interested in. Any signals should be expected to represent the target organ or cell type. Organoids and organ-on-a-chip technologies offer amazing new opportunities for cell- and tissue-based research as well.

Several factors must be considered to obtain meaningful results from cell- and tissue-based studies:

• Generally, cells and tissues can provide insights into local effects. However, organisms (and biological systems in general) are complex, so one must carefully consider the representativeness of any findings.
  
  
• With tissue samples, variability between locations within organs must be considered. You can counter this problem by using homogenates of samples larger than what the metabolomics method requires, or by measuring tissue samples from multiple sites.
  
  Most metabolomics studies use fresh tissue rather than fixed tissues, mainly because the fixation step has vast consequences on metabolite levels.
  
  
• In organs that are highly perfused, eliminating blood contamination can also be a step to increase validity.
  
  
• Small changes in the conditions for culturing and/or collection can have a major impact on the findings. If cells change their phenotype and behavior in a culture, they may no longer represent what you are intending to investigate.
  
  Co-culturing of cell types can be an interesting option to better simulate physiological conditions. The type of medium can also influence the results.
  
  
• Finally, consider the extraction of metabolites from cells and tissues. Metabolites vary considerably in their chemical and physical properties.
  
  Consequently, one needs to consider whether a single-step extraction works well enough across metabolites of interest, whether a multiple-step extraction protocol could yield better results across metabolite classes, or whether an extraction optimized for polar or apolar is the preferrable option.
  

Biofluids

Blood components

Blood components are probably the most frequently used matrix in metabolomics studies. As a routinely used clinical sample, it’s the go-to matrix for biomarker studies. Blood represents signals from the whole organism, making it a great matrix for research on complex diseases.

While no one doubts the relevance of blood-based samples as a matrix for metabolomics, there’s intense debate around the exact type of sample to use.

Some proponents in the field prefer a serum matrix because it removes a larger proportion of cellular components, leading to slightly higher metabolite concentrations and thus higher sensitivity in biomarker discovery. For some areas of (targeted) metabolite analyses, such as steroid hormone analysis, serum is the conventional matrix of choice.

Others argue that serum is more prone to pre-analytical issues such as oxidation, making serum-based metabolomics vulnerable to effects of impaired sample quality. In addition, some blood cells remain metabolically active during coagulation and may release metabolites into the fluid component of the sample. For example, platelets can metabolize arachidonic acid and release high amounts of eicosanoids during coagulation. This may limit the relevance of such groups of metabolites if analyzed in serum rather than plasma.

For this reason, many say that plasma should be be preferred over serum. However, it’s not clear which anti-coagulant is most suitable. Ethylenediamine tetraacetic acid (EDTA) plasma is most-commonly used, while citrate plasma is often discouraged for use in LC-MS based metabolomics studies.

Whole blood is rarely used as a matrix for metabolomics because of the potential interference from cellular components. That said, dried blood spot analysis is a well-established metabolomics approach in routine newborn screening (NBS).

Recently, we’ve also seen considerable interest in novel means of biofluid sampling, such as dried plasma spot sampling. This could keep samples (more) stable at room temperature and alleviate logistical barriers to using metabolomics (traditional methods are hampered if blood samples cannot be frozen and transported safely).

Sorted cells or cellular components, plus extracellular vesicles collected from blood samples, can be a very interesting matrix to use to answer specific questions. For example, tumor-secreted extracellular vesicles could provide a non-invasive sample to inform researchers about the metabolome of a tumor. Similarly, metabolomics performed from sorted immune cells can reveal the intricacies of immune regulation.

Urine

Urine is generally considered an interesting matrix for metabolomics due to the easy and non-invasive means of sampling. It’s also highly stable compared to blood-based samples, even at higher storage temperatures.

The gold standard is a 24-hour urine sample, but spot urine is also common. Researchers should watch out for the high salt content in urine when performing analysis. In addition, normalizing for creatinine concentrations is a common strategy to improve quality and reduce variability.

It’s also worth noting that in diagnostic settings, you often look for what is not supposed to be there, e.g. glucose or proteins. Many detectable metabolites are deliberately excreted by the organism to maintain homeostasis within the organism. This is a conceptual issue that is also relevant to the use of feces, as will be discussed later.

Other biofluids

• Saliva and tear fluid are also interesting matrices due to the non-invasive means of sampling, though compared to other biofluids mentioned here, metabolomics in these matrices is in its infancy.
  
• Cerebrospinal fluid (CSF) is an interesting surrogate for the central nervous system, but is difficult to sample and it can be very difficult to obtain healthy control samples for ethical reasons.
  
• Likewise, bronchoalveolar lavage fluid (BALF) can be a great sample for research on respiratory tract disorders, but sampling challenges make sampling at scale impractical.
  
• Sweat could also be a promising matrix due to the easy and non-invasive sampling, but again remains in its infancy.

Other non-invasive sample types

Non-invasive sampling is extremely attractive because it’s accessible, acceptable to study subjects, and easily repeated. Feces and hair are the main matrices to mention here, but we see metabolomics applications described for several other sample types, such as skin lavage, earwax, nasal mucus and nasal lavage, among others.

The use of feces as a matrix for metabolomics studies has seen a steep rise recently. This is partly due to the rise of microbiome studies and acknowledgement by microbiome researchers that metabolomics can enhance functional understanding of host-microbiome interaction. However, ensuring quality in metabolomics studies from fecal samples poses several challenges.

Firstly, the structure and consistency of fecal samples can vary considerably, which affects the abundance of biomolecules. Secondly, both recent and habitual diet can be important confounders. In addition, the method of extraction has a major effect on what is measured in the fecal metabolome.

With separation of fecal water or soft extraction methods, you may be able to mostly base your analysis on the cell-free parts of the fecal sample. With harsher extraction methods you may lyse or rupture microbial cells to varying degrees. This can lead to higher metabolite coverage and be of scientific interest as well.

After all, bacteria interact with the organism extensively and cells may produce and excrete substances that are swiftly taken up by the organism and might thus not be represent in fecal water at significant amounts.

Finally, and most importantly, the fecal metabolome may represent what the organism actively sheds rather than the organism’s metabolic composition. The interaction between the microbiome and the metabolome is more significant in the duodenum than in the lower intestines.

Consequently, the fecal metabolome is a poor surrogate for the metabolome in the upper intestines. The use of cecal duodenal contents may be a better matrix to assess host-microbial interaction, at least in basic research.

In in-vivo human settings, sampling and analyzing samples from the upper intestines has not yet moved beyond proof-of-concept studies and no generally accepted method has evolved yet.

See our article on fecal metabolomics to learn more: Best practices for feces metabolomics – biocrates life sciences gmbh

Hair is another accessible matrix that’s particularly interesting for disease monitoring because it can capture changes that have occurred over weeks or even months prior to sampling.

Although hair analysis has become an established method for measuring xenobiotics and analyzing selected metabolites such as cortisol, its clinical utility remains in question.

The scarcity of metabolomics studies from hair compared to other indications may make it even harder to confirm the relevance of biomarker signatures. Again, variability is a major concern. This can stem from hair characteristics such as pigmentation, but also from external factors such as hair treatment with shampoo and other chemicals.

There are also no accepted standard procedures specific to the analysis of the hair metabolome, which makes comparison between studies even more difficult.

Concluding thoughts

What’s clear is that metabolomics can be and has been done from a vast variety of sample matrices, each with its pros and cons. In that respect, the question of which sample matrix to use remains as relevant as it is difficult to answer.

Thankfully, if you are considering more than one matrix, you don’t always have to choose: why not combine multiple matrices to optimize results and maximize insights?

Here are a few examples of how combining matrices could be of value:

• Combining plasma and urine analyses for research on kidney diseases can add insights compared to using only a single matrix, as relative changes between plasma and urine could be biologically meaningful.

• For biomarker research based on blood components, the question of where the signal originates can be answered by adding tissue of the target organ or multiple organs which are known to contribute to a disease. In other words, combining circulatory biofluids and tissues can provide information on BOTH the local effects and the systemic contribution.

• Combining metabolomics from fecal samples and plasma can provide clues as to the probable effects of microbiome changes directly in the gastrointestinal tract, and which actions it can exert in systemic circulation.

There’s also plenty to say about combining metabolomics with imaging technologies or other -omics technologies – but we’ll save that for another article.

Before you embark on your metabolomics project, you have some serious thinking to do. Choose wisely, and you’ll be rewarded with exciting findings.

What do you think? What is your favorite matrix and why? Have we missed a matrix that you find important? Reach out to us if you have questions about the methodological considerations for metabolomics studies in the matrix of your interest

Nobody wants to invest thousands of dollars and countless hours of effort in a metabolomics study that turns out to be a flop. So, what steps can researchers take to boost the chances of obtaining robust and reliable results?

The first crucial ingredient is to use standardized and quality-controlled quantitative metabolomics technology.

The second is to understand and control potential confounding factors during the pre-analytical phase. This helps to ensure good sample quality, which is a prerequisite for reliable metabolomics results.

The molecular composition of a biological sample is constantly changing. Some molecules – especially metabolites – are unstable and may be subject to oxidation, aggregation or degradation.

Even ex vivo, these dynamic changes need to be kept to a minimum to best reflect the subject’s physiological status. If essential pre-analytical steps and standardized experimental protocols aren’t meticulously controlled and followed, the results could be inconsistent or corrupted by unwanted artifacts in the data (Ulaszewska et al. 2019).

Is sample type important?

Serum, plasma and urine are the most common sample types in metabolomic studies. Every sample type has its own advantages and disadvantages, so it’s important to choose the most appropriate sample matrix for the target metabolites and study aims (Wang et al. 2018).

For example, there’s not much difference between the lipidomic profiles of serum and plasma, but the profiles of polar metabolites are significantly dependent on sample type (Jørgenrud et al. 2015).

The type of anticoagulant should also be carefully selected, as explained later in this article. If the sample matrix is not the right fit for the research question, and if protocols aren’t adhered to, the study won’t generate high-quality data (Jørgenrud et al. 2015).

What are the main considerations before sampling?

Multiple factors affect the results of a metabolomics study, even before sample collection has begun. For example, alcohol consumption, drug intake, physical exercise, fasting status and diseases can all influence an individual’s metabolic profile.

Studies have shown that alterations in the amino acid profile, especially that of branched-chain amino acids, are associated with obesity (Morris et al. 2012). Furthermore, both pathological and molecular subtypes of diseases can differ in their metabolic profiles (Fan et al. 2016; Wojakowska et al. 2015).

Therefore, the study protocol should specify which information has to be recorded and how. The collected metadata needs to be available for downstream data analysis. Defining these factors in advance is crucial, especially for large-scale, longitudinal studies.

Some important confounding factors that may affect metabolomics results are (Davies et al. 2014; Hardikar et al. 2020):

• Fasting status
• Age
• Gender
• Body mass index (BMI)
• Circadian and physiological rhythm
• Time of sample collection
• Diet, incl. alcohol consumption
• Physical exercise
• Sleep deprivation
• Drugs
• Diseases

What are the main considerations for blood sample collection?

Samples are often taken by different people in different centers. The use of a standardized sample collection protocol (and strict adherence to these protocols) may help to prevent deviations in results (Kirwan et al. 2018; Paglia et al. 2018).

Plasma samples are collected using a set of anticoagulants, including ethylenediaminetetraacetic acid (EDTA), heparin, sodium fluoride/potassium oxalate (NaF/KOx), and sodium citrate. The type of anticoagulant has a strong effect on the concentration of several metabolites.

For example, sodium citrate impairs the reliable determination of citric acid and compounds with similar elution times (Jørgenrud et al. 2015). Anticoagulant residues in the final extract can also affect the final mass spectrometry analysis by forming sodium and potassium formate clusters and causing ion suppression and enhancement (Barri et al. 2013; Jørgenrud et al. 2015).

Choosing the right anticoagulant for metabolomics is still a topic under discussion, and there is no universal answer. Nevertheless, the Human Serum Metabolomics Association (HUSERMET) recommends the use of heparin for metabolomics research (Dunn et al. 2011).

What are the main considerations during blood sample processing?

Before centrifugation and short-term storage

For all omics studies, the general rule is to cool the sample whenever possible. Cooling slows down metabolic activity, thus preserving the sample while it awaits the next preparation step (e.g. centrifugation).

Never freeze a blood sample before centrifugation if you aim to prepare a plasma or serum sample. Freezing breaks cell membranes, resulting in the release of intracellular contents in the sample.

Before centrifugation, the temperature at short-term storage during sample processing can affect the metabolic profile (Hebels et al. 2013; Jørgenrud et al. 2015). Due to continuous blood cell metabolism and release of intracellular compounds, exposure of whole blood to room temperature (RT) for more than two hours is a major risk in the pre-analytical phase (Bi et al. 2020).

In one study, incubation of both whole blood and plasma samples at room temperature for six hours resulted in significantly altered levels of metabolites compared to 0°C and 4°C, respectively (Cao et al. 2019).

Analyses with EDTA plasma samples showed large differences in the stability of specific metabolites, especially lipids, when stored at RT for four days. Degradation of lipids starts already after one day (Jørgenrud et al. 2015).

A similar situation was observed in serum samples. Several studies demonstrated that metabolites are not stable at RT or even at -20°C after being stored for one month (Haid et al. 2018; Hernandes et al. 2017; La Frano et al. 2018).

Following a prolonged storage at RT, about 20% of the metabolites were significantly increased, while 4% were decreased (Kamlage et al. 2018). In fact, degradation and metabolite profile changes began after just 12 hours storage at RT. For example, glutamate levels increase when stored at RT (Kamlage et al. 2014).

Even on ice, certain changes were visible (Anton et al. 2015). This shows that the time window between coagulation and storage is critical for metabolite stability. Luckily there are a few metabolites that may be used as indicators to evaluate pre-analytical sample quality (Schwarz et al. 2019).

Serum is one of the most common sample types used for metabolomic studies. In the first step to obtain serum, whole blood is incubated at RT for 30-60 minutes for coagulation. If clotting time is shorter, coagulation may be incomplete; if it is longer, hemolysis can occur. Both may lead to altered metabolic profiles (Olshansky et al. 2022).

In large-scale multicenter studies, samples may need to be shipped. This introduces another pre-analytical pitfall and may cause delay in sample processing and variability between samples based on different shipping conditions (Breier et al. 2014).

Therefore, fast processing of the sample is recommended. Researchers must pay attention to temperature and clotting time to prevent variations in this step.

While artefacts from pre-analytics may arise, one way to limit their impact on a study is to harmonize the pre-analytical steps and randomize samples during preparation to avoid introducing pre-analytical bias such as batch effects.

Hemolysis

When working with blood samples, hemolysis is another frequent pre-analytical confounding factor (Blanckaert et al. 2008). Hemolysis is the release of hemoglobin and other intracellular components, including metabolites and enzymes from erythrocytes, following disruption of the cell membrane.

Although hemolysis typically results in a red color, it can be hard to spot and requires careful visual inspection during sample preparation.

Considering that intracellular metabolite concentrations can be more than 10 times higher than extracellular, hemolysis can lead to a significant increase in many metabolite concentrations in plasma or serum.

Yin et al. found that 69 metabolites changed significantly in hemolytic samples (Yin et al. 2013). Thus, hemolysis should be avoided, and hemolytic samples should be flagged for statistical data analysis and their results interpreted with caution.

After centrifugation

When preparing plasma, the plasma supernatant should be carefully removed after centrifugation without touching the buffy layer, to avoid contamination with blood cells that can otherwise affect the plasma metabolome (Kamlage et al. 2018).

The ratio of blood sample volume to anticoagulant can also have an effect on the metabolite concentrations in plasma samples. Drawing variable sample volumes may change this proportion and may cause unwanted variations in results (Olshansky et al. 2022).

What to consider when collecting urine samples

Urine collection is relatively easy compared to plasma and serum. But there are still several factors that may cause variability of metabolite profiles (Stevens et al. 2019).

The concentration of metabolites may change up to 15-fold in the same sample volume due to physiological factors, water intake, and external environment (Warrack et al. 2009). Usually, urine samples are collected by the patient. This may introduce variations regarding collection time, sterilization, and sample volume.

As a solution, collection time and fasting status should be specified, an antibacterial additive or filtration should be used to prevent contamination, and normalization strategies should be applied to overcome the volume issue; e.g. normalizing to creatinine concentration (Emwas et al. 2016).

The European Consensus Expert Group Report has published several recommendations for urine biobanks (Yuille et al. 2010).

Long-term sample storage and metabolite stability

After the sample has been processed, it should be transferred to -80°C or lower for long-term storage as quickly as possible to suppress enzymatic activities affecting metabolite levels. However, even at -80°C, certain plasma metabolites are unstable after five years (Haid et al. 2018).

This is mainly due to mechanical degradation. Since most long-term storage of serum, plasma, and urine samples is at -80°C, the decrease in sample quality should be taken into account for long-term studies (Abuja et al. 2015; Jurowski et al. 2017; Rotter et al. 2017; Wagner-Golbs et al. 2019).

Another issue related to sample storage is the frequency of freezing and thawing. Reports on the stability of metabolites during freeze-thaw cycles vary (Abuja et al. 2015; Gratton et al. 2016; Helmschrodt et al. 2014; Moriya et al. 2016).

To minimize the effects on the metabolome, freeze-thaw cycles should be limited, and all samples in each study should be treated consistently, with quality controls applied (Cao et al. 2019).

One way to limit the number of free-thaw cycles is to prepare multiple aliquots in independent storage tubes. This allows to thaw only a small portion of the original sample for a given measurement. The downside is that sample volume also influences the speed of sample degradation during long-term storage.

As always with pre-analytical considerations, the specifics of the study (number of expected measurements, volume required, etc…) are the main driver to choose the optimal long-term storage conditions to ensure sample quality.

Pre-analytics: all’s well that starts well

In summary, pre-analytical confounding factors can have a significant effect on some metabolites and should be kept in mind when comparing metabolomics results. Robust, standardized and detailed sample collection and processing protocols are essential to prevent unwanted variations and to ensure data quality.

All confounding factors and deviations from the study protocol should be recorded and accessible for statistical data analysis. A homogenously treated sample set will be a promising starting point for reliable and reproducible results from a metabolomics study.

If you’d like expert advice on which sample matrix to use for your metabolomic study and how to handle it, please contact us.

References

Abuja P. et al.: Alterations in Human Liver Metabolome during Prolonged Cryostorage. (2015) J Proteome Res | https://doi.org/10.1021/acs.jproteome.5b00025

Anton G. et al.: Pre-analytical sample quality: metabolite ratios as an intrinsic marker for prolonged room temperature exposure of serum samples. (2015) PLOS ONE | https://doi.org/10.1371/journal.pone.0121495

Barri T. et al.: UPLC-ESI-QTOF/MS and multivariate data analysis for blood plasma and serum metabolomics: effect of experimental artefacts and anticoagulant. (2013) Analytica Chimica Acta | https://doi.org/10.1016/j.aca.2013.01.015

Bi H. et al.: The key points in the pre-analytical procedures of blood and urine samples in metabolomics studies (2020) Metabolomics | https://doi.org/10.1007/s11306-020-01666-2

Blanckaert N. et al.: Haemolysis: an overview of the leading cause of unsuitable specimens in clinical laboratories (2008) Clin Chem Lab Med | https://doi.org/10.1515/CCLM.2008.170

Breier M. et al.: Targeted metabolomics identifies reliable and stable metabolites in human serum and plasma samples (2014) PLOS ONE | https://doi.org/10.1371/journal.pone.0089728

Cao Z. et al.: An Integrated Analysis of Metabolites, Peptides, and Inflammation Biomarkers for Assessment of Preanalytical Variability of Human Plasma (2019) J Proteome Res | https://doi.org/10.1021/acs.jproteome.8b00903

Davies S. et al.: Effect of sleep deprivation on the human metabolome (2014) Proc Natl Acad Sci U S A | https://doi.org/10.1073/pnas.1402663111

Dunn W. et al.: Procedures for large-scale metabolic profiling of serum and plasma using gas chromatography and liquid chromatography coupled to mass spectrometry (2011) Nat Protoc | https://doi.org/10.1038/nprot.2011.335

Emwas A. et al.: Recommendations and Standardization of Biomarker Quantification Using NMR-Based Metabolomics with Particular Focus on Urinary Analysis (2016) J Proteome Res | https://doi.org/10.1021/acs.jproteome.5b00885

Fan Y. et al.: Human plasma metabolomics for identifying differential metabolites and predicting molecular subtypes of breast cancer (2016) Oncotarget | https://doi.org/10.18632/oncotarget.7155

Gratton J. et al.: Optimized Sample Handling Strategy for Metabolic Profiling of Human Feces (2016) Anal Chem | https://doi.org/10.1021/acs.analchem.5b04159

Haid M. et al.: Long-Term Stability of Human Plasma Metabolites during Storage at -80 °C (2018) J Proteome Res | https://doi.org/10.1021/acs.jproteome.7b00518

Hardikar S. et al.: Impact of Pre-blood Collection Factors on Plasma Metabolomic Profiles (2020) Metabolites | https://doi.org/10.3390/metabo10050213

Hebels D. et al.: Performance in omics analyses of blood samples in long-term storage: opportunities for the exploitation of existing biobanks in environmental health research (2013) Environmental Health Perspectives | https://doi.org/10.1289/ehp.1205657

Helmschrodt C. et al.: Preanalytical standardization for reactive oxygen species derived oxysterol analysis in human plasma by liquid chromatography-tandem mass spectrometry (2014) Biochem Biophys Res Commun | https://doi.org/10.1016/j.bbrc.2013.12.087

Hernandes V. et al.: A review of blood sample handling and pre-processing for metabolomics studies (2017) Electrophoresis | https://doi.org/10.1002/elps.201700086

Jørgenrud B. et al.: The influence of sample collection methodology and sample preprocessing on the blood metabolic profile (2015) Bioanalysis | https://doi.org/10.4155/bio.15.16

Jurowski K. et al.: Comprehensive review of trends and analytical strategies applied for biological samples preparation and storage in modern medical lipidomics: State of the art (2015) TrAC Trends in Analytical Chemistry | https://doi.org/10.1016/j.trac.2016.10.014

Kamlage B. et al.: Quality markers addressing preanalytical variations of blood and plasma processing identified by broad and targeted metabolite profiling (2014) Clin Chem | https://doi.org/10.1373/clinchem.2013.211979

Kamlage B. et al.: Impact of Prolonged Blood Incubation and Extended Serum Storage at Room Temperature on the Human Serum Metabolome (2018) Metabolites | https://doi.org/10.3390/metabo8010006

Kirwan J. et al.: Preanalytical Processing and Biobanking Procedures of Biological Samples for Metabolomics Research: A White Paper, Community Perspective (for “Precision Medicine and Pharmacometabolomics Task Group (2018) The Metabolomics Society Initiative). Clin Chem | https://doi.org/10.1373/clinchem.2018.287045

La Frano M. et al.: Impact of post-collection freezing delay on the reliability of serum metabolomics in samples reflecting the California mid-term pregnancy biobank (2018) Metabolomics | https://doi.org/10.1007/s11306-018-1450-9

Moriya T. et al.: Intensive determination of storage condition effects on human plasma metabolomics (2016) Metabolomics | https://doi.org/10.1007/s11306-016-1126-2

Morris C. et al.: The relationship between BMI and metabolomic profiles: a focus on amino acids (2012) Proceedings of the Nutrition Society | https://doi.org/10.1017/S0029665112000699

Olshansky G. et al.: Challenges and opportunities for prevention and removal of unwanted variation in lipidomic studies. (2022) Progress in Lipid Research | https://doi.org/10.1016/j.plipres.2022.101177

Paglia G. et al.: Influence of collection tubes during quantitative targeted metabolomics studies in human blood samples (2018) Clin Chim Acta | https://doi.org/10.1016/j.cca.2018.08.014

Rotter M. et al.: Stability of targeted metabolite profiles of urine samples under different storage conditions (2017) : Metabolomics | https://doi.org/10.1007/s11306-016-1137-z

Schwarz N. et al.: Quality Assessment of the Preanalytical Workflow in Liquid Biobanking: Taurine as a Serum-Specific Quality Indicator for Preanalytical Process Variations (2019) Biopreservation and Biobanking | https://doi.org/10.1089/bio.2019.0004

Stevens V. et al.: Pre-Analytical Factors that Affect Metabolite Stability in Human Urine, Plasma, and Serum : A Review (2019) | https://doi.org/10.3390/metabo9080156

Ulaszewska M. et al.: Nutrimetabolomics: An Integrative Action for Metabolomic Analyses in Human Nutritional Studies (2019) | https://doi.org/10.1002/mnfr.201800384

Wagner-Golbs A. et al.: Effects of Long-Term Storage at -80 °C on the Human Plasma Metabolome (2019) MDPI | https://doi.org/10.3390/metabo9050099

Wang Z. et al.: Comparison of Fecal Collection Methods for Microbiome and Metabolomics Studies (2018) Frontiers in Cellular and Infection | https://doi.org/10.3389/fcimb.2018.00301

Warrack B. et al.: Normalization strategies for metabonomic analysis of urine samples (2009) Science Direct | https://doi.org/10.1016/j.jchromb.2009.01.007

Wojakowska A. et al.: Detection of metabolites discriminating subtypes of thyroid cancer: Molecular profiling of FFPE samples using the GC/MS approach (2015) Science Direct | https://doi.org/10.1016/j.mce.2015.09.021

Yin P. et al.: Preanalytical aspects and sample quality assessment in metabolomics studies of human blood (2013) Clinical Chemistry | https://doi.org/10.1373/clinchem.2012.199257

Yuille M. et al.: Laboratory Management of Samples in Biobanks: European Consensus Expert Group Report (2010) Biopreservation and Biobanking| https://doi.org/10.1089/bio.2010.8102

Soft tissue sarcomas (STS) are a heterogenous group of cancers that affect tissues such as adipose tissue, muscle, cartilage, vessels, and nerves. In this publication, the authors combined metabolomic data with more classical biomarkers of cancer progression and survival to train models to predict the outcome of antitumor therapy with trabectedin in patients with metastatic forms of STS.

Profiling of the patients’ amino acids, amino acid related metabolites and bile acids levels in serum enabled the identification of a metabolic pattern in weak responders to the antitumor cancer treatment. Three patients with a particularly poor outcome clearly clustered separately from the other patients in a principal component analysis (PCA). The separation of these two clusters was driven by lower levels of amino acids in the patients with poor outcomes, while bile acids were not involved. Using univariate Cox proportional hazards regression analysis, the authors identified two amino acids, citrulline and histidine, that were associated with the overall survival of the patients.

Next, a risk prediction model was developed that combined metabolomic data with endpoints such as ECOG performance status (a therapy tolerance indicator), hemoglobin levels and tumor grading. The model was able to identify two groups of patients with either a high risk or a low to medium risk for overall survival after treatment.

This model focused on a small number of patients (n = 24), however, and it would require to be validated using a larger independent cohort of patients. Nevertheless, this promising study is a good example of the approach that can be taken to integrate metabolomic data with clinical variables to improve current diagnostic and prognostic biomarkers for cancer and other diseases.

Intrigued about the uses of metabolomics in research? Check out our applications page.

Miolo G, Di Gregorio E, Saorin A, Lombardi D, Scalone S, Buonadonna A, Steffan A, Corona G: Integration of serum metabolomics into clinical assessment to improve outcome prediction of metastatic soft tissue sarcoma patients treated with trabectedin (2020) Cancers (Basel) | doi: 10.3390/cancers12071983