Green and Rapid HS-GC-IMS Using Hydrogen Carrier Gas for Non-Targeted Volatilomic Fingerprinting and Authentication of Cocoa Liquors

Researchers explored a green and high-throughput chromatographic approach based on headspace gas chromatography–ion mobility spectrometry (HS-GC-IMS) for the non-targeted analysis of volatile organic compounds (VOCs) in cocoa liquors. Emphasizing the principles of green analytical chemistry, the method minimizes sample preparation, solvent use, and energy consumption while enabling rapid and robust volatilomic profiling.

A key innovation is the use of hydrogen as a carrier gas instead of the conventional nitrogen, significantly increasing linear velocity and reducing chromatographic run times without compromising separation efficiency. This results in substantially faster analyses compared to traditional HS-solid phase microextraction (SPME)-GC-MS methods, which typically involve lengthy extraction steps and extended run times.

LCGC International spoke to Philipp Weller and Lukas Bodenbender of Technical University of Applied Sciences Mannheim, Germany, two of the authors of a paper resulting from this study,¹ about their work.

How does headspace gas chromatography–ion mobility spectrometry (HS-GC-IMS) compare with headspace solid-phase microextraction gas chromatography-mass spectrometry (HS-SPME-GC-MS) for volatile fingerprinting of complex food matrices such as cocoa liquor?

The fundamental difference of HS-GC-IMS approaches over SPME-based (or any other online enrichment strategy, such as ITEX) GC-MS techniques is that we typically do not require any enrichment, because the GC-IMS systems are extremely sensitive.¹ You cannot generalize that point for all chemicals, but for polar to medium polar compounds, Hanna Schanzmann from the Sielemann group in Hamm, Germany demonstrated recently² that GC-IMS is often substantially more sensitive than a modern single quadrupole GC-MS system. Having said that, it becomes clear that fingerprinting-based analysis of complex matrices – such as cocoa liquors – profit from this higher sensitivity when it comes to trace compounds, that might be relevant features for authentication or quality analysis. This sensitivity is mainly a result of the soft ionization, paired with the fact that the drift tube ion mobility spectrometer (DTIMS) cell operates under near-ambient conditions. While electron ionization (EI) is a brilliant ion source, it has its limitations when it comes to labile molecules due to excessive fragmentation. A second aspect is that in GC-IMS, we have (nearly) complete freedom of choice of carrier gases, such as hydrogen, as demonstrated in our paper.

What are the advantages and potential limitations of using hydrogen instead of helium as a carrier gas in gas chromatographic separations?

The clear advantages are higher speed of analysis and lower cost for the carrier gas if generator hydrogen is used. The price of hydrogen is substantially lower than that of helium and the availability is not a relevant topic – for helium, it is, in times where natural gas sources are often in politically instable regions. Only yesterday, we have received a force majeure note from our gas supplier related to helium, so we will very soon see whether we must speed up our transition.

The hydrogen purity coming from such generator systems is remarkable, we have never managed to obtain such clean spectra with bottle hydrogen that was coming from our university installation, not even with the usual gas filter cartridges. The advantage of high sensitivity in GC-IMS sometimes also is a limitation here, as we see virtually everything between the gas bottle and the GC-IMS, mainly lubricants. Thinking about that is quite logic, as larger installations require more valves and regulators, that add potential contaminations sources. A hydrogen generator sits very close to the GC-IMS with short tubing, so the risk of contaminations are even less likely. These contaminations often show a high variance in such that they vary over time, which makes fingerprinting analyses more laborious.

A disadvantage is that hydrogen is not an inert gas, therefore it is possible that reactive compounds may produce artifacts. We have not spotted this effect in our cocoa samples to be a relevant problem for the data analysis, yet this is an aspect to keep in mind. Headspace methods typically work with relatively cold injection temperatures in comparison to liquid injections, which further reduces artifact formation.

However, this aspect can be a crucial one, depending on the platform you intend to use. On a GC-IMS system, we were able to demonstrate that hydrogen can be a game changer with substantially faster separations and higher sensitivity due to the narrow peak shapes. For non-target screening approaches in GC-IMS, where we typically have a focus on the characteristic fingerprint rather than specific substance identities it should further be of no relevance if other ion species are formed when hydrogen is used instead of helium or nitrogen – if all analyses, training and prediction are done on the same system configuration. Switching gases or operating different machines on different carrier gases might be problematic. This is one of the main reasons, why so far, GC-EI-MS users are often reluctant to use hydrogen as a carrier gas, as it substantially changes the fragmentation pattern, which ultimately reduces the hit rate in EI spectral databases. I have talked to several routine pesticide analysts who told me they switched back to helium. This is one of the reasons, why we are currently evaluating digital twins for such systems and to translate e.g. helium EI spectra to hydrogen spectra.

A second disadvantage is obviously the flammability and the potential to form explosive air/hydrogen mixtures. Clearly, this is a valid point to think about. My personal experience and opinion is however, that this seldomly a solid issue. While it is true that the IMS does not combust and therefore “disarms” the hydrogen carrier gas, at is done in a GC-FID, one has to keep in mind that the flow rates coming from the GC to the IMS range around 1-2 mL/min, while the IMS itself is using a nitrogen drift gas flow of around 150 mL/min. High split rates are typically not used in headspace GC injection, so I consider the risk of generating explosive gas mixtures to be of minor relevance in practice.

In case of generator hydrogen, a critical factor, safe storage of gas bottles is irrelevant, as these systems generate relatively small amounts of hydrogen and have safety shutdown once a leak is detected.

Yet, it requires precautious measures to ensure safe working environments, so it does make sense to have a hydrogen sensor in the GC and to ensure a proper ventilation. It might also be subject to specific regulations, whether hydrogen may be used as a carrier gas or not.

How does the Van Deemter equationexplain the improved separation speed achievable when hydrogen is used as a GC carrier gas?

Both the van Deemter equation and the more applicable Golay equation were developed in the 1950s, where nearly all routine GC was carried out using packed columns and isothermal conditions. As a result, these classical equations assume constant temperature, however, nearly all modern GC systems apply temperature ramping to minimize peak broadening. This means, that in a stricter sense, these equations should be taken with a grain of salt. One limitation for example is the different behavior of the gases when it comes to viscosity under different temperature conditions. However, I think they are still great to explain effects such as the higher performance of hydrogen in a simplified way to beginners and at least I do so in my courses.

The main effect of the higher performance of hydrogen is related to its higher diffusion coefficient. Now, to be fair, one must compare hydrogen with its direct competitor helium and not nitrogen. Nitrogen principally is a very performant carrier gas, when it is used in its optimum linear velocity of around 12 cm/s, it has the best performance of the three candidates, but at the price of substantially lower speed of analysis. When comparing hydrogen and helium, Graham´s law shows that hydrogen has a slight advantage of 1.41 fold higher diffusion speed.

In the simplified van Deemter/Golay explanation, this translates to the broader optimum and the lower slope towards higher linear velocities in the H/u plot. The main reason here is the resistance to mass transfer term C, which basically decreases with the diffusion speed of the analytes in the carrier gas. As these parameters are dependent on temperature and pressure, the van Deemter/Golay curves must be generated for discrete substances in discrete media under isobaric and isothermal conditions. If diffusion coefficients of discrete substances are compared under such conditions, hydrogen features the highest values.

In simple words, this means that analysts can use higher linear velocities while maintaining narrow peak shapes. This even becomes more relevant, when using smaller column diameters of 0.25 mm or below and it was quite fascinating to us, that we could easily go beyond the commonly recommended 50-60 cm/s to way above 100 cm/s and still obtaining reasonable peak shapes. This means that the column geometry does have a substantial impact and a combination of low inner diameter and low film thickness supports that effect. The price is obviously a lower linear capacity of such columns, which ultimately requires a more sensitive detection system, as less absolute amount of analyte reaches the detector – and voilà, here, our ³H-based DTIMS cells excel.

Why is non-target screening (NTS) increasingly important in chromatographic analysis for food authenticity and quality control?

There are several reasons why NTS approaches have gained relevance over the last years. One major aspect that needs to be clarified is that authenticity and quality is not the same thing. An arbitrary product might not be of the highest possible quality or may be inferior in terms of contaminants versus another but still be an authentic one – or vice versa. Generally, globalization of trade has made the world of food products more complicated in such, as markets have become more volatile on the one hand and consumer demands on price, quality and all-over-the-year availability have made fraudulent activity more attractive. At the same time, fraudsters have become more sophisticated in fine-tuning adulterated products, so that common test strategies based on target compounds might fail. Furthermore, there are more complex authentication challenges coming up and demand for a more holistic, fingerprinting-type of approach. One example could be a low-level blending of an expensive Apulian olive extra virgin oil with a cheaper olive oil from other countries or using cheaper Arabica coffee to blend with higher priced specialty Arabica, such as Catuai Vermelho. Both are examples, where simple target analysis will fail, as there are simply no targets to analyze. While genetic analysis will work on intact green coffees beans of different species, I seriously doubt that this would work for roast and/or ground coffees, particularly not for blended coffees. Here, NTS strategies, both LC- and GC-based, optimally both can be a valuable tool to detect anomalies. However, these approaches are like metabolomics approaches in such as they rely on a larger data set of samples and sophisticated machine or deep learning algorithms. Few samples won´t help much and this is also the Achilles heel of NTS strategies – getting hold of a reasonable number of authentic samples.

A further reason is related rather to the quality aspect of foods and extends towards food safety: we have built up substantial knowledge over the years concerning contaminants and residues, that has virtually proliferated target lists of compounds to be detected. Examples are PFAS and its breakdown products or masked mycotoxins, i.e. mycotoxins that undergo metabolic and process-related changes and are thus not on the usual suspect lists. It is quite tempting to use NTS-based strategies, e.g. based on HRMS systems or for specific use, also GC-IMS approaches to get screening information before diving deeper into compound search and quantitation. However, it has to be stressed that while this is a fast growing research field, there are still a lot of limitations, among of which are limited scope of ionization and classical RP chromatography, limited scope of substances amenable by GC enrichment procedures, limitations of comprehensive databases from GC and LC world, missing standardization of NTS screening approaches, so far still a relatively large share of “dark matter” (non-annotated features) and many more. Yet, I believe that we are on a good way, as there are numerous excellent research groups active in this field. The volatilomics field still needs more of a push into a broader use, but this is obviously my personal and biased point of view.

What role do chemometric tools such as principal component analysis (PCA) and partial least squares (PLS) regression play in interpreting complex chromatographic datasets from volatilomics studies?

A decisive one, without a doubt. Chemometrics or machine learning are crucial tools, or strategies rather to dig through that enormous pile of data that NTS volatilomic approaches generate. While the data space is more limited than in LC-based approaches, simply because there are fewer volatile substances than non-volatile ones, we are still talking about hundreds or thousands of signals that are being detected by our systems.

While complex data generally is a powerful source of knowledge, it also requires sophisticated data analysis strategies to distill meaningful features from useless variables. Such data sets are called high-dimensional, which refers to the issue that there are typically much more variables than objects (or samples), which makes life more complex. Working with such data in the sense of classification or regression typically requires dimensionality reduction, which e.g. can be done with Principal Component Analysis or Partial Least Squares algorithms. PCA is a brilliant tool if used correctly and to my experience, often underrated and used incorrectly or at least, with the wrong expectations. It does a great job in a first understanding of potential correlations or problems in your data, such as an improper preprocessing strategy or potential residual outliers. However, as an unsupervised algorithm, PCA is not good at separating highly variant multi-class problems, e.g. ten classes of olive oils with a relatively low number of samples. In chemometrics speech, PCA is maximizing total variance of your data, but it does not maximize inter-class variance as does PLS. PCA does not know anything about the identity of your data, i.e. it is unlabeled and even when users use color maps to graphically support the visual inspection, there is no connection to the algorithm.

PLS regression and its derivative PLS discriminant analysis are in contrast supervised algorithms, that in my experience pair up perfectly with PCA as a second step once you feel comfortable about the quality of your data and when you have enough data for a proper training and testing strategy. PLS-DA and its derivative OPLS-DA are great and interpretable tools not only for classification, but also for “biomarker” search in food matrices that might indicate an ambiguous authenticity. My experience however is that for complex, multi-class problems, often PLS-based approaches are less performant and robust than modern ensemble learning methods, such as bagging or boosting. If correctly handled, these have a reduced tendency of overfitting. One aspect however that is crucial for all strategies is a proper validation of your models and I cannot stress enough this point.

How can volatile organic compound (VOC) profiles obtained by GC-based methods be used to differentiate cocoa samples based on geographic origin or processing conditions?

I would suggest two approaches here. One could be an NTS strategy that is based on a pattern recognition analysis, for example by GC-IMS or GC-MS paired with a supervised algorithm, such as PLS-DA, Random-Forest or MCR-ALS. This would be a good option for the differentiation of geographic origins, as this is typically a rather “diffuse” question: the expectation of a sharp differentiation of geographies often is difficult, if the cultivation regions are close together or adjacent. Country boarders are man-made and often not characterized by distinct climatic conditions or soil properties that one would expect to reflect in the pattern of the metabolic profile of cocoa. This is getting clearer in cases where specifically different varieties are cultivated in the respective geographies to be differentiated, namely Criollo, Forastero and its hybrid Trinitario. This is somewhat comparable with the geographic differentiation of olive oils from different geographies: here, the main differentiation feature (most likely) is not the geography itself, but rather the fact that these countries mainly cultivate endemic varieties that partially show a distinct difference in their fingerprints. The advantage of this approach is that one could use mass spec data if available, but also less selective data, such as coming from GC-IMS will work well in this context. The latter would also allow to perform such NTS approaches on-site, while MS-based techniques typically would require a dedicated lab infrastructure.

The second option would be a semi-targeted NTS strategy, which could be applied to processing conditions. The rationale behind this is that depending on specific processing steps, artifacts could be formed, such as oxidation or hydrolysis products from the secondary plant metabolites or we might see specific fermentation products or spoilage markers. One example here could be terpene oxides that result from different fermentation, drying and roasting regimes, that leave a potentially characteristic fingerprint in the raw material. Here, an NTS approach, paired with an optimized search strategy for several substance groups (such as breakdown products from lipids or amino acids or terpene oxides), which then can be characterized in a semi-quantitative fashion, such as via fold-change plots. One highly promising strategy here is the use of molecular networks, which comes from classical metabolomics and allows to explain structural relationships of known and unknown compounds to be displayed in a network that can be characteristic for a specific variety or processing. This is however still limited in such as the volatilomics databases are still scarce and mostly limited to either commercial EI spectra databases, such as NIST or to inhouse proprietary databases. I hope that this will change in the future towards a direction as it is the way for LC-MS data. Similar is true for GC-IMS data, where the availability of open data bases is scarce.

What chromatographic parameters can be optimized to reduce GC analysis time while maintaining adequate resolution and sensitivity?

There are several parameters that can be fine-tuned, but the quickest results undoubtedly come from column geometry and chemistry, carrier gas type and temperature profiles. For the sake of simplicity, I would exclude specific hardware upgrade, such as direct heating columns. Also, I would limit this answer to helium and hydrogen as a carrier gas, as nitrogen is more susceptible towards faster linear velocities and is not an optimal choice when it comes to GC-MS. Again, GC-IMS have an advantage here, as these systems work fine with nitrogen and can handle high flow rates in wide-bore or multi-capillary columns.

A simple strategy to optimize speed of analysis while not changing too much in your system is changing the column geometry to smaller inner diameters, thinner films and shorter lengths, e.g. from the conventionally used 30m/0.25 mm ID/0.25µm film to 20m/0.18 mm ID/0.18 µm film, ramping up temperature a bit faster. There are great tools out there, that simplify method translation.

Assuming that you are coming from a helium-based method, my experience is that if you stay with helium, you should end up with a speed gain of around 30%. If you switch to hydrogen and optimize parameters again, my guess is an improvement of around 50-60%.

One aspect to keep in mind is that the price to pay for this step is a lower linear capacity of the column, which means you will need to decrease your load on the column, e.g. by higher split rates if possible to avoid overloading of columns. At the same time, peaks are virtually pushed to a narrower shape and consequently, higher intensities, which increases S/N ratios and to a certain extent, compensates the reduced absolute amount of analyte on the column.

I think, this impressively shows how little analysts would need to do to substantially speed up their analyses if they are willing to leave the comfort zone of using standard parameters.

Why does GC-IMS often eliminate the need for enrichment techniques such as solid‑phase microextraction that are commonly used in GC-MS workflows?

GC-IMS in the form we use it exclusively is based on a soft, ³H based ionization, that leads to the formation of proton-water cluster ions. This results in extremely high sensitivities for polar and medium polar compounds, as the ions remain intact. In contrast, most GC-MS workflows are based on EI ionization, which is a very hard, fragmentation-rich ionization process. While this leads to great spectra that are perfect for database search, such as NIST or WILEY, this often leads to low abundance of the molecular ion and to ambiguous fragmentations, e.g. for terpenes. Another reason for the higher sensitivity of GC-IMS is the substantially lower temperature in the ion source of around 60-120°C instead of 300-350°C in GC-MS, which also has an impact on sensitive molecules. Consequently, GC-MS workflows typically do not require enrichment procedures, such as SPME or ITEX. We commonly reach single digit ppbv values for compounds such as ketones and aldehydes.

How does drift tube ion mobility spectrometry (DTIMS) provide an additional separation dimension when coupled with gas chromatography?

DTIMS separation is based on the collisional cross section (CCS) of ions traveling through the drift tube which is filled with a drift or buffer gas. In our case, this is nitrogen, as we need a gas with a reasonable CCS itself. This property is orthogonal to the main property that drives GC separation, the vapor pressure. This is particularly interesting, as the ion source generates proton-water clusters, where the number of ion-water clusters attaching to a molecule depends on the structure and polarity. That means that for example certain aldehydes and ketones could have the same boiling point and as such, would not separate on the GC column, but might have a different CCS in the DTIMS separation – and ultimately show separated signals. As IMS happens in a millisecond scale and GC separations occur in the range of seconds or minutes, this is an optimal pairing.

This hyphenation is in particular interesting when it comes to data analysis, as orthogonal data features the so-called 2^nd order advantage, which allows us to use intelligent machine learning algorithms, such as MCR-ALS or PARAFAC to extract this information even if one of the dimensions do not show a separation and deconvolute the signals.³ Obviously, this also is true for MS data, where typically even more selectivity is generated by the fragment ions.

In the context of green analytical chemistry, how does the adoption of faster GC methods and minimal sample preparation improve the sustainability of chromatographic workflows in food analysis?

I think this question is more relevant than ever with an increasing number of conflicts that show us, how brittle our supply chains for resources, such as natural gases are. Helium will be a critical resource, where laboratories cannot be sure to reliably source supplies.

Faster GC methods translate to a higher efficiency of laboratories in terms of sample numbers, which is also true for minimal (i.e. shorter) sample preparation. Less sample preparation, less solvents used and the use of carrier gases from renewable sources, such as hydrogen mean less use of fossil resources and less emission of VOCs coming from sample analysis and less chemical waste to be disposed. There is several parameters that are evaluated in green chemistry concepts, such as in the AGREE scheme. A side effect of analyzing larger numbers of samples is that this requires more efficient workflows, where chemometrics can be a valuable tool for.

However, it is not only the chromatography domain that needs critical evaluation in view of sustainability, but also the detection systems used. In this context, DTIMS is a powerful strategy as often it can compete with GC-MS based approaches but can be operated more sustainable.⁴

References

1. Bodenbender, L.; Rohn, S.; Parastar, H. et al. Towards "Greener" Strategies in Quality Control: Rapid Volatilomics of Cocoa Based on HS-GC-IMS and Machine Learning. Anal Bioanal Chem. 2026. DOI: 10.1007/s00216-026-06415-3
2. Schanzmann, H.; Gaar, S.; Keip, S. et al. Comparison of the Quantification Performance of Thermal Desorption GC-IMS and GC-MS in VOC Analysis. Anal Bioanal Chem. 2025, 417, 4179–4198. DOI: 10.1007/s00216-025-05933-w
3. Parastar, H.; Yazdanpanah, H.; Weller, P. Non-Targeted Volatilomics for the Authentication of Saffron by Gas Chromatography-Ion Mobility Spectrometry and Multivariate Curve Resolution, Food Chem. 2025, 465 (Part 2), 142074. DOI: 10.1016/j.foodchem.2024.142074
4. Parastar, H.; Weller, L. Towards Greener Volatilomics: Is GC-IMS the New Swiss Army Knife of Gas Phase Analysis? TrAC Trends Anal. Chem. 2024, 170. DOI: 10.1016/j.trac.2023.117438