Blog

ASMS is just around the corner, and we are all eager to meet both old and new colleagues, network, and seek inspiration from thousands of posters. For the galaxy of companies that orbit mass spectrometers (MS), it’s the prime stage to launch new products and concepts. As a user and consumer, I am also curious. Some of these will be pleasantly surprising and inspiring, while others will raise eyebrows. Regardless of the impression they make, this is the ideal time to voice it.

I previously wrote about the fact that high-resolution MS were designed for proteomics. TripleTOFs, Tribrids, timsTOF, Astral, and others were all conceived to improve the analysis of peptides. At the product launch, the benchmarks pivot around the number of peptide IDs per amount of sample or per time.

Small molecules, whether metabolites, lipids, PFAS, etc., are rarely considered because nobody has looked into them yet. Typically, this occurs only about 6 to 12 months later, when there is enough time for more exploratory demos with customers who have the funds to invest, including labs focused on small molecules.

This is understandably driven by sales and the need to meet specific commercial goals. The KPIs include revenue, double-digit growth, and simply the number of new instruments sold. Depending on sales performance, there will be more or fewer incentives to promote sales by the end of the reporting period.

Therefore, there is a window of opportunity to secure the best deal, but not necessarily the best instrument. This is because the driving force isn’t the application or the specific needs of the customer, but rather sales. The best deal can’t overshadow the fact that these remarkable instruments were developed with peptides in mind and offer characteristics or workflows that do not necessarily align with the needs of small-molecule analysis.

During testing, it’s quite common to discover flaws, bugs, or other design issues that compromise the ability to detect, quantify, or identify small molecules. Even in the recent years, we collected a long list simply by performing basic sanity checks, such as example loss of transmission in negative mode, the loss of MS2 fragments, incorrect DDA precursor selection, >300% scan overheads in DDA, faulty iterative fragmentation, incorrect centroiding, wrong deisotoping and feature detection…. The best we could do was report our observations to the vendor, but we rarely saw fixes arrive within a short timeframe. Enhancements were sporadically noticed only in later generations of instrumentation, which could not be applied to the instrument I purchased.

Therefore, if vendors offer you an option to “collaborate” with them to address unsatisfactory demo results and shift towards results that meet your expectations, don’t fall for it. It’s too late to obtain something in the near term. You are on your own.

How can we as a community improve the situation? First of all, by clearly formulating what’s needed, why it’s needed, and voicing it. The ASMS event is the perfect opportunity to share these insights with all solution providers! While we might not see immediate changes, let’s keep in mind the words of Sepp Herberger: after the game is before the game!. The feedback and conversations we engage in will play a pivotal role in shaping our future plans, which will start right after ASMS.

What are our needs?

Needs matter a lot, both to users and providers. A fundamental principle of a successful product is that it must solve a problem or address a clearly identifiable need (see JTBD, pain point, product-market fit). If we reverse the logic, the success of a product scales with the need it addresses. There are some notable exceptions, but there is no reason to believe that scientific equipment doesn’t adhere to the same principle.

Here, I aim to explain what I believe is genuinely needed in metabolomics, with a focus on discovery metabolomics and, consequently, high-resolution MS. Before looking into the future, however, we have to recap how our field has evolved since the beginning of the Millenium. The underlying problems have changed over the past decades, also to best complement other omics, and so have our needs in terms of technology.

These notes reflect discussions I frequently have with colleagues and vendors. From experience, I know it helps to reiterate how metabolomics and proteomics differ, and why we need something else. Here and there, I will elaborate despite the risk of saying trivial things. Repetita juvant.

Let’s roll back time…

Need #1 – Chemical formulas

In the early days of metabolomics, researchers discovered that low-resolution mass spectrometers (quads and ion traps) were inadequate for untargeted analysis. Too many distinct metabolites shared the same nominal mass, and without high resolving power, they appeared as a single peak or were indistinguishable.

The primary demand was therefore for high-resolution, accurate-mass instrumentation to confidently resolve peaks and assign molecular formulas. By the mid-2000s, high-resolution MS became widely adopted in metabolomics specifically because it dramatically improved the separation of quasi-isobaric compounds and the calculation of elemental compositions. Orbitraps, and with some delay, TOFs provided routine measurements with <5 ppm mass errors. With isotope ratios and heuristic rules (for example 7 golden rules), it vastly narrowed the list of possible formulas for a given m/z. This was a game-changer: a metabolomics experiment could detect hundreds of features and assign putative formulas based on accurate mass and isotope patterns, something previously feasible only in niche laboratories.

What I’d like to stress is that for many years, all high-resolution MS vendors have effectively met the necessary standards that are practically relevant in routine metabolomics. This is reflected in the fact that we have stopped worrying about resolution and mass accuracy. We are very pleased with the resolution and ppm we achieve with TOFs that are over 10 years old and have a resolution of 30-50k. I don’t recall pushing any of our Orbitrap instruments to 240-480k, even for tracing studies that rely on resolving fine isotopic patterns. The impact of higher resolutions is minimal. Needs have evolved.

Need #2 – Fast (enough) MS2

The next step involved moving beyond the formula and closer to structural elucidation, which motivated the shift toward fast MS2. Here, the ability to collect numerous MS2 spectra per unit of time has been more important than precise mass. This accounted for the initial success of Q-Exactives (= Q-Orbitrap) over early Q-TOFs, which provided the unique advantage of flexibly adjusting the resolution of MS1 and MS2 scans, and in turn, adapting the acquisition speed to the gradient and sample complexity. MS2 spectra collected at a resolution of 7500 in approximately 30 ms (and later 16 on the HF-X or the Exploris) are sufficient for library searches. For the record, the MS2 scan rate limit on Orbitraps was (and is?) 40 Hz.

In the meantime, TOF instruments have undergone significant improvements, and all top models offer nominal MS2 scan rates of 50-133 Hz in DDA and approximately 200 Hz in DIA, but at 5- 10 times higher resolution than what an Orbitrap can deliver with short transients. The best combination in terms of speed and resolution still belongs to the Astral (200 Hz in DIA and 80k resolution), which benefits from running MS1 and MS2 acquisitions in parallel on specialised detectors, while all Q-TOFs use a single detector and switch back and forth between MS1 and MS2 scans, resulting in substantial loss in effective MS2 events per time, which in DDA is further aggravated by the time needed to “decide” what to fragment next.

The key question is what we really need in terms of MS2 acquisition speed for metabolomics. Is it “the faster the better” as in proteomics? In my opinion, no, quite the opposite. Firstly, our samples contain by far fewer features than peptide digests. Secondly, and most importantly, our tests indicate that the fastest MS2 acquisition speed is only effective on the most abundant features. While it is possible to obtain good MS2 spectra for the 500-1000 compounds in a sample with a 5 ms time, mid- to low-abundant features require a longer acquisition time (or injection time on an Orbitrap) to collect sufficient precursors and the most informative fragments that allow discrimination of isomers. Therefore, we need to slow down by a factor of 5-10 to a range where time, and not speed, is the limiting factor.

The need to slow down MS2 acquisition is similar to what happens in single-cell proteomics to collect sufficient data, with one key difference: it doesn’t happen because we inject less material but because of the low signal produced by a large portion of analytes, resulting from broad concentration, poor ionization, or low abundance of the key fragments. While in proteomics it’s possible to “pick” the best flying peptides and most selective b/y fragments for identification and quantification, in metabolomics (and lipidomics) we are bound to measuring single or a few MS2 fragments for each compound.

Overall, in our lab we utilise the maximum acquisition speed (top 10, 5 ms per MS2 spectrum) only with very short gradients of 2-5 minutes. If the LC time permits (5-15 min), we allocate it to longer MS2 acquisition (e.g., top-10 up to 50 ms per MS2) instead of faster cycling (top-40 at 5 ms). This enhances MS2 quality and, consequently, identification.

Need #3 – Intrascan dynamic range

The primary aim of untargeted analysis is to identify and measure potentially numerous features. As I mentioned earlier, enhancing the resolution to levels at or exceeding 10_⁵ has minimal effect on coverage. Rather, the critical aspect is the intrascan dynamic range, which generally denotes the range between the maximum and minimum signal intensities that can be reliably detected.

To illustrate the dynamic range, vendors frequently present calibration curves with concentrations of a compound spanning several decades. Don’t fall for it: these are typically acquired under optimised conditions, perhaps using the quadrupole or other techniques to isolate a thin slice of precursor before entering the high-resolution detector. Such a test is only relevant if you wish to compare it against a QQQ and does not reflect the characteristics of the instrument when operating in untargeted mode.

In my view, a much better measure of performance would be how many compounds are detected and identified when analyzing a single representative sample. This is what vendors present for proteomics (# number of identified peptides, protein groups, etc.), but we never get to see a similar analysis for metabolomics.

In real life, instruments must simultaneously manage the complexity of natural samples, which span over 10 logs (link). Regardless of the length of chromatography, there will always be coelution of high and low-abundant compounds. For this reason, we need technologies that can detect a potentially wide range of intensities within the same scan.

While early TOFs had only 2-3 logs, modern detectors achieve 5-6 logs. Orbitraps fall somewhere in between, limited by the Fourier Transform (to approx. 3.7-4 decades). Additional limitations may arise in any instrument that uses a trap to accumulate ions before MS1 scanning (e.g., tims, C-trap). To prevent detrimental space-charge effects, they restrict the number of ions or charges processed in a single packet. Even when the number of charges is large (10⁵-10⁷), it can result in abundant ions outcompeting rare ones. The effects of limited dynamic range can be dramatic. I reported years ago on flow injection analysis (link), and we observed similar outcomes with fast LC: the lower the intrascan dynamic range, the fewer features are detected when scanning in full MS.

There is an additional, underappreciated aspect of ion traps that affects small molecules: they have a limit on the range of m/z values that they can efficiently trap, which affects both MS1 and MS2 acquisition. For instance, it might be technically impossible to observe light fragments if the precursor is >3x heavier (see low mass cutoff), or you might observe that a large m/z window is automatically split into smaller slices by the acquisition software to compensate for the lack of bandwith of traps, with obvious consequences for the scanning rate.

Vendors are well aware of the importance of intrascan dynamic range and adopt a variety of sophisticated methods to continually push instruments beyond the limits of the actual detector. Increasingly, we observe that instruments employ the same strategy seen in HDR photography: acquiring multiple shots at both low and high gain (i.e., exposure) and merging them into a single HDR image that exceeds the technical limit of the detector. This can leave scars, particularly at the baseline.

Moving into the future

These three needs have driven the field and sales over the past two decades. In my view, they largely explain why, currently, TOF-based instruments are better suited for untargeted metabolomics. Proteomics-optimised instruments, despite their amazing sophistication, frequently miss the mark for our field: there is simply no general need for >100 MS2 spectra per second, or resolution > 100’000, in particular if this comes at the cost of dynamic range.

What about the future? I’d like to address this question by starting with the needs, recognising that there are different ways to technical address them. Across all applications (from biomedical to natural products, exposomics, …) we are still hitting two “usual” barriers.

1. the detection of low-abundant compounds: we are still far from the 10+ decades, and many signaling molecules remain undetected
2. the more fine-grained annotation by MS2, i.e. resolution of isomers with identical elemental composition.

None of this is surprising; these are the same forces that have shaped the last twenty years. We’ve made significant progress, but there is still much more to achieve. I reflected on ways to overcome these barriers and identified two key areas that I view as game-changers in the coming years.

Need #4 – The Unrelenting Need for Full Scan Sensitivity

We continue to discover new molecules and biology (link), but we are aware that there is still a significant number of biologically relevant molecules that we do not detect at all or only by MS1. To detect and identify more of the chemical space, we need more sensitivity.

In an MS, sensitivity is the essential equivalent of horsepower for a motor. It determines our capacity to detect low-abundance compounds, capture rare MS2 fragments, enhance throughput, dilute samples to counteract matrix effects, and analyse tiny sample sizes (like single cells). We require significantly more sensitivity for both MS1 (detection) and MS2 (identification). There is essentially no upper limit: increased sensitivity allows us to delve deeper into the metabolome.

How can we increase sensitivity in full scan? It’s a complex task, much more challenging than enhancing the sensitivity of a QQQ. It requires an orchestrated design of multiple components to prevent collateral issues, such as space-charge effects, saturation, and loss of dynamic range. Despite being challenging, vendors understand how to address it. Yet, this doesn’t seem to be a priority. For instance, I am still perplexed that a 13-year-old instrument remains the best-in-class for high-throughput screening by flow injection (link). My take is that our community must advocate more effectively for the fundamental importance of sensitivity, prioritising it over speed, resolution, and other gimmicks. We should create a compelling business case that will encourage vendors to allocate resources in this fundamental direction.

Need #5 – Multimodal Information for Enhanced Structural Elucidation

We are encountering a fundamental barrier in metabolite annotation by MS2. This is not primarily due to insufficient MS2 libraries or inadequate AI/ML tools. The core issue is that Collision-Induced Dissociation (CID) spectra often lack sufficient information for complex small molecules. Dealing mainly with singly charged species, CID biases fragmentation towards polar functional groups. While useful for compound class identification, it often fails to resolve positional isomers, double bond locations, or aromatic ring substitution patterns.

To transcend this structural elucidation barrier, we must integrate additional orthogonal information. In my view, the most intriguing avenues are techniques such as IR spectroscopy, UV photodissociation (UVPD), and electron-induced dissociation (EID/Ecd)—all notable for their broad applicability. Having explored EID for years and invested in UVPD instrumentation, I am convinced of the profound benefits that alternative dissociation techniques and spectroscopic methods offer. It’s only a matter of time before computational methods advance to leverage this richer data for finer structural detail. I am encouraged to see several vendors incorporating lasers and electron sources into commercial platforms. This is just the beginning, but it’s unequivocally the path forward.

Summary and final thoughts

To detect and quantify more compounds, we need instruments with maximum intrascan dynamic range and full scan sensitivity. There is no limit. More sensitivity is always better: it provides more information, and can be converted into speed.

To identify more compounds, we need informative spectra and multimodality. For this, we have to integrate additional, orthogonal techniques (EID, UVPD, IR spectroscopy) and likely slow down to ensure that the rare diagnostic fragments are captured. No, we don’t need even more empty MS2 spectra.

Any innovation in these primary directions, while preserving established performance in resolution, scan speed, and dynamic range, will have a long-lasting impact on the entire metabolomics community—and would certainly make me one of the happiest participants at ASMS.

Everything else, at this point, is nice to have: resolution & accuracy, real-time search features, APIs, better DDA engines, acquisition APIs, higher-order calibration systems, integrated system suitability testing, …

What about ion mobility? I am in favor of it if it is used to address the core needs. For instance, by increasing sensitivity (#4, think of FAIMS) or by resolving isomers (#5). The latter, however, cannot come at the cost of sensitivity or dynamic range (#3). I have yet to see a reliable 4-dimensional feature detection algorithm, and I am still not convinced that CCSs are helpful in routine work. Happy to be proven wrong.