3.1. Experimental Design

A scientific hypothesis needs to be clearly defined in order to properly design a metabolomics experiment (see Fig. 3). In other words, simply characterizing the metabolome of a cell, tissue or biofluid is likely to fail to provide any scientific insight or value without a logically designed comparative analysis. In this regards, a metabolomics experiment is fundamentally a comparison between two or more groups (e.g., control vs. drug treated), where a difference or similarity in a metabolome is expected to answer a scientific question. The next critical step is to identify an appropriate source for the metabolome or the type of biological samples to be compared (i.e., established or primary cell lines, tissue samples from an animal model, or human biofluids, etc.). The choice of system will then determine the number of biological replicates that can be practically obtained, which, in turn, will define the statistical significance of any observed results. Given a testable hypothesis and the type of biological samples, there are essentially an endless number of experimental variables that can be manipulated, where each permutation defines a new group. Practically speaking, it would be impossible to account for every possible permutation in a single experimental design. Instead, the identification of which groups to use is fundamentally driven by the scientific question the study is designed to answer (see Fig. 3) Thus, a general set of protocols is presented for a generic metabolomics experiment that can be applied to a drug discovery project. The protocol provides a means to determine a drug’s mechanism of action and to monitor a patient’s response to treatment.

(a) A summary of the design and planning steps in a typical metabolomics-based drug discovery project. (b) A workflow for a typical cell-based NMR metabolomics experiment focused on understanding the mechanism of action for a chemical lead.

The simplest experimental design for a metabolomics drug discovery project will consist of two groups: untreated samples and drug-treated samples. Nevertheless, it is typical to significantly expand the experimental design beyond just two groups (see Figs. 2 and ​and3).3). Common groups include:

Controls (e.g., wild type cells or healthy subjects).

Control with treatment (positive control e.g., wild type cells or healthy animals treated with a drug lead).

Experimental model (e.g., animal disease model or treatment naïve patients).

Experimental model with a drug treatment (e.g., animal disease model or patients treated with a drug lead).

Negative control (e.g., mutant cells with no treatment, patients with alternative disease).

Negative control with a drug treatment (e.g., mutant cell lines treated with a drug lead).

The metabolomics study may consist of any combination of biological samples originating from cell cultures, tissue samples and/or biofluids. Human or animal biofluids may consist of urine, serum, plasma, blood, fecal material, saliva, sweat, condensed breath, cerebrospinal fluid (CSF) or any other body fluid or excretion. Similarly, tissue samples usually consist of organs, such as heart, lungs, brain, liver or kidneys from sacrificed animals. Tissue samples from human subjects typically originate from biopsies (i.e., muscle, adipose, etc.), from surgeries (i.e., tumors) or tissue banks. Each biofluid, tissue type, or time-point would constitute a different group. Of course, there are numerous experimental protocols that could also be varied, such as different media, nutrients, temperature, diets, etc.

Besides sample type and experimental protocols, there are other variations that may be applied to form additional groups. For example, different groups may be formed by varying the drug dosage or by using a placebo. Of course, multiple drugs could be tested in a single study and group membership would be defined by what drug a cell culture or an individual receives. In addition, the number of drug treatments received by a cohort could vary. Time could be another variable in the experimental design (e.g., when samples are collected and when treatments are administered). Similarly, different groups may consist of different mutant or wild-type cell lines or strains. For example, a mutant cell line may correspond to the drug protein target being inactivated (i.e., negative control) through a variety of genetic mutations or gene-knockout methods. Alternatively, the mutant cell lines could be various human isolates with variable levels of resistance or susceptibility to the tested drug. The cell lines could just be different types of bacteria (e.g., E. coli., S. aureus, etc.) or different types of cancers (e.g., pancreatic, breast, etc.). Similarly, animal models or human cohorts may have different stages or severity of the disease, or even different diseases. Furthermore, multiple biofluid samples (i.e., both urine and serum) may be collected from the same animal or human participant at multiple time points during the study. Similarly, more than one tissue or organ sample may be harvested from each sacrificed animal.

Another important decision regarding the experimental design is the number of biological replicates required per group. In general, more biological replicates led to a better statistical significance in differentiating between the various groups. But of course, there are practical limits to the number of cell cultures, animals or human cohorts that can be prepared or recruited for any study. Facility and equipment capacities, the availability of personnel, and cost, are all major factors that greatly limit the number of samples or cohorts that are possible. There is also an obvious multiplication factor based on the number of desired groups and the number of replicates per group. Accordingly, there may be a tendency to maximize the number of groups at the expense of the number of replicates per group. This is inadvisable since it would likely lead to statistically insignificant conclusions. Instead, the number of groups should be reduced to achieve an acceptable minimal number of biological replicates per group. In fact, sample size calculation for metabolomics studies is complex and not well-defined, where recent efforts suggest sample sizes may need to be > 20 per group [66,67]. In the case of metabolomics studies involving human cohorts, the need for replicates is even greater and more challenging to estimate, but sample sizes > 60 to 100 cohorts per group are typical targets.

One final factor to consider for the design of a metabolomics study involving human cohorts is the inclusion and exclusion criteria. Simply, what individual characteristics, traits or behavior are needed for the goals of the study? Conversely, what are the potential confounding factors that may confuse or obscure the desired outcome of the study and should be avoided? The specific inclusion and exclusion criteria are likely unique to each study; and there are no general guidelines to apply. Criteria commonly considered include: age, body mass index, diet, comorbidities, ethnicity, gender, health test panel (e.g., blood pressure, lipid panel, metabolic panel, urinalysis, etc.), physical activity, race, and the use of alcohol, recreational drugs, or tobacco (see Table 1).

A list of typical exclusion criteria for a metabolomics clinical study