R-E-L-A-X. Successful metabolomics studies are not that hard
by Kirk Beebe
I like football (the American version that is), and I also like metabolomics. So, not surprisingly, as I was contemplating writing about how doing a successful metabolomics study is not really that difficult, these two topics collided. I thought back to the famous Green Bay Packers quarterback Aaron Rodgers and his comments to the media after a brief losing streak last season. His response to the panicked media was to spell a single word in an effort to assuage their concerns: “R-E-L-A-X”. People found this perplexing. Here was a guy at the helm of a position that many believe is the most complex in all sports, suggesting that it was simple; it was all under control. He backed it up as they won 11 out of their last 13 games.
As I was pondering a blog contribution about metabolomics studies and thinking that I really wanted to demystify a topic that, in my opinion is often over-mystified, I immediately recalled the above exchange.
I assure those of you contemplating or planning a metabolomics study, that the principles for success are actually very intuitive and straight-forward.
DNA, RNA, Protein – Metabolites are not aliens from Mars
Metabolomics is increasingly embraced by investigators who recognize that it’s a valuable tool for research, biomarker discovery and precision medicine. But, in our gene and protein world where the comfort level with genes and proteins (and the associated -omic tools) is generally very high, the metabolic world surveyed by metabolomics can seem complex and foreign. In contrast to this impression, a well-conceived metabolomics study can often be more tractable for finding and interpreting meaningful signals than more familiar types of -omic data.
Sure, there’s a learning curve, but it’s not as steep as some investigators perceive it. Don’t get hung-up on “how difficult” or “how important” a particular step is in the process and miss the forest for the trees. This is a natural and understandable response to something we don’t understand – grasp and try to control an aspect we understand.
Metabolon has successfully put metabolomics results into the hands of researchers for over a decade. While there is a massive amount of technology and engineering under the hood of our platforms that produce the data, a successful metabolomics study is fairly straight-forward if you follow these five steps.
1. Define a Clear Objective
I know what you’re thinking: “Well of course; who wouldn’t do that?” And, I know, it seems pretty basic. But, all -omic technologies have a seductive power, even to those of who are discerning and informed. The seductive power comes from an intuition that the girth of the technology, data and informatics applied to a set of samples will surely, by brute-force, render a clear result at the end. This notion is less prevalent today.
Almost all scientists using more established expression technologies understand that the exact opposite is true. When -omic experiments are not carefully and thoughtfully done, they can produce false positives that consume valuable resources and time. But, with metabolomics being less familiar and often regarded as a “newer” -omic compared to those for nucleic acids, the pitfall remains. Thus, investigators are sometimes tempted to bundle excessive numbers of groups, diverse samples, treatments, etc., into a single study. The result is often predictable – a lot of data with few concrete leads.
So, like ANY experiment, define your target (what are you precisely trying to answer?). Once defined, ensure that this target is suffocated with the proper controls.
2. Use Strong Study Design Elements
Controls for metabolomics studies are no different than any other experiment. Account for as many variables as possible to encircle the key question you are trying to address. And, (statisticians please skip ahead to #3, because I will horrify you with the next statement) design always trumps powering a study. Having both are optimal but, if for some reason you have to sacrifice one for the other, don’t compromise on design. I will offer one simple example of this after we discuss power.
3. Power Your Study for Success
Power ensures that when a meaningful difference is detected in the study, you can assign significance to it. Most investigators are well versed in the virtues of powering a study and the practice of power calculations. We’ve done more than 4,000 studies over the years, so we can provide solid guidelines for powering almost any experiment. However, given that a study is an experiment and effect sizes and variability within any experimental system can be different, we simply advocate for excellent design and a number of samples per group (power) that confidently works for nearly all systems we have studied to date. It’s not that complicated. Use more per group if you anticipate a subtle effect or want to have the highest probability of success; use fewer if you expect a large effect and or are willing to gamble a bit.
An exquisitely powered study with poor design may allow you to stand-back and admire the low p-values and get you a pat you on the back from your statistician colleague, but it may fail to advance the question you were aiming to answer.
For example, let’s say you want to discover the mechanism of a weight-loss drug and have designed the study to only collect samples during the acute phase of weight loss. Here, primary mechanistic changes most salient to the drug mechanism will be mired in covariates not salient to the drug mechanism (secondary effects related to large morphological changes related to the shedding of weight).
A design that maximizes the probability of success would be to have earlier time-points that could illuminate the primary effects related to the drug mechanism. Ironically, if this required that the study be less well-powered, biologically meaningful signals as well as the identification of outliers can frequently be discerned through the context of the time-points (albeit with larger p-values). The aim of a discovery metabolomics study should be to emerge with a strong hypothesis. Design is paramount for achieving this.
Finally, a brief disclaimer regarding the above scenario. Neither is ideal. One is just better (in my opinion) than the other. Optimizing design AND power maximizes the probability of success for your metabolomics study.
4. Choose the Right Technology for Your Goals
There are many potential ways to collect data on metabolites, but the most important advice I can offer is to refer to point #1 and tailor the choice to the research goal. In contrast to this point, it is not uncommon to encounter investigators insisting on an approach that, while perfectly indicated in one instance, is not the tool needed for addressing the research goal. Consider the following example:
An investigator is seeking to discover a pathway that could help them understand the function of a gene or a drug. Here, the most important tool would be something that can offer the widest screen of the metabolome, with the aim that a pathway or set of metabolites can offer a clue to the function of the gene or the drug. A broad discovery metabolomics method is therefore indicated. However, it is not uncommon to encounter those who lose focus on the need for a broad profiling approach and insist on the importance of methods that are indicated for very precise questions, such as flux. While it may be nice to have every feature imaginable in one technology or assay, you often have to choose between features.
The way we approach any discovery study is to use broad metabolomic profiling to detect unique signals.
To do this, we use a platform that automatically defines the identity of each metabolite in the sample (note: many “untargeted” workflows fail to do this and, therefore, have major pitfalls in revealing meaningful signals within a set of samples. More on this another day).
Once the unique signals are discovered, we use different methods to sharpen or validate the hypothesis. Sometimes this may involve elegant targeted or fluxomic methods for really understanding the regulation of the specific pathway illuminated in the discovery study. Many times it also involves a traditional cell and molecular biology method. So, in summary, the technology features should be ruthlessly assessed and selected based on your precise goal.
5. Have a Data Interpretation & Follow-up Plan in Place
A well-designed and executed experiment applied to the appropriate technology has just produced a stellar metabolomics data set. Unfortunately, after a metabolomics data set is produced, the hard work of interpreting it to generate a hypothesis is required. We typically do a lot of this for our clients and go into more detail around this point in a recently produced eBook, but I want to leave you with an extension to this point – one that is often overlooked.
As discussed above, sometimes -omic technologies provide such a powerful impression with the intended user that there is a sense that the end result of a study will be a Science paper with a bow around it. Unfortunately, even after a hypothesis is generated from the interpretation, significant work remains. It’s essential that an investigator have a resource plan in place for refining, validating or sharpening the hypothesis generated from a metabolomics study. Can it sometimes be straight-forward? Yes. But, some form of orthogonal experiment will be required at the end.
This final point can sometimes offer the impression that doing a metabolomics study is just going to create confusion and more work. Let me assure you that a well-conceived, designed and powered metabolomics study can create an immense short-cut to an important result. Think of it this way. This type of study is like a “cheat code” for a video game. You still have to play the game, but with the code in hand, the degree of difficulty and time required to complete a level is greatly diminished.
What we’ve described here and more is detailed in our new eBook. Download your free copy.