Highlights from the HPLC2026 Conference through the Lens of LC Troubleshooting

In June, I was fortunate to attend and contribute to the 55th International Symposium on High Performance Liquid Phase Separations and Related Techniques, better known in the liquid chromatography (LC) community as “HPLC 2026,” held in Indianapolis (https://hplc2026-symposium.org). I describe this meeting to my students as the Super Bowl of liquid chromatography, and as usual, it did not disappoint. It was great to reconnect with many friends and acquaintances, and, of course, to talk about some of the most recent and emerging science in our field. I highly encourage anyone in the field who has never been to this meeting to consider attending. It is held annually, with locations alternating between North America, Europe, and Asia. There is a variety of content for everyone, ranging from short courses at the introductory level in the two days before the main conference begins, to keynote lectures from world-leading experts in specific research areas under the umbrella of liquid chromatography. Next summer, the meeting will be held in Innsbruck, Austria, and there will also be a 2027 meeting in Fukuoka, Japan.

In this installment, I will present several vignettes describing presentations I saw at the HPLC 2026 meeting that relate to liquid chromatography troubleshooting. As I say frequently, the most successful LC troubleshooting exercises are those that follow effective prevention, so that the problem never develops into something that requires troubleshooting. So, these vignettes are related to troubleshooting but are sometimes more about preventing problems from developing rather than solving them once they’ve become too large to ignore.

In my preparation for this installment of “LC Troubleshooting,” I reread some older installments written by John Dolan. One that stood out to me was the March 1992 column entitled “LC Problems – Past, Present, and Future.”¹ It was fascinating to read John’s perspective at that time, some of the important problems with LC instrumentation, and his predictions about solutions the future might hold. The following quote from 1992 could just as easily been overheard at the recent HPLC 2026 meeting: “During [the previous decade], LC has progressed from a rapidly growing technology to one that shows signs of maturity…the chromatographic knowledge base and technical skill level of the typical worker [practicing LC] has dropped…When such changes take place, the frequency of hardware and separations problems tends to increase. The reverse has been true for the most part, however, primarily because manufacturers have done a fine job improving the reliability of LC products.” I would say this is largely true in 2026, which is why I think it is so important to continue to raise the level of the chromatographic knowledge base through this column and other educational resources.

In his discussion of major hardware-related problems from that time, John highlighted five major topics: 1) pump seals; 2) air bubble-related problems; 3) detector lamps; 4) column failure; and 5) fittings. He indicated that manufacturers had made significant progress in addressing these problems. It is interesting to reflect on further progress in these areas over the last 35 years. From my point of view, pump seals and detector lamps have become so good that we don’t view them as major problems anymore. I have not replaced a pump seal in my laboratory in the last 10 years. In the area of fittings, there continues to be a remarkable level of innovation, ranging from compression-style fittings that are reusable and finger-tight up to 1200 bar, to face-sealing fittings for low-dispersion, finger-tight connections. In my view, air bubbles remain problematic with pumps, and columns fail more often than we would like. As you will see in the discussion below, perhaps these are areas where increased “smartness” of our instruments will help improve operational quality of life.

In his thoughts about the future—in 1992—John highlighted several aspects related to instrument “smartness” and more intelligent use of instrument time and resources: 1) built-in diagnostics; 2) remote servicing; and 3) push-button experts. John concluded by saying, “The above developments are not just fantasy. All the technology exists today, and, in one form or another, these features are commercially available today.” Here again, I think these statements are as relevant in 2026 as they were in 1992. Perhaps a scientist from 1992 might be a little disappointed to see that there has not been enough progress in these features to make diagnosing a problem with an air bubble more straightforward. Nevertheless, what I saw at the HPLC 2026 meeting is that manufacturers are making important investments in instrument “smartness” that users will be benefiting from soon, and the future looks exciting in this regard. Perhaps it is a coincidence, but I have the feeling we are seeing a step change in the development of these features that is probably related to the wave of change surrounding generative artificial intelligence we are seeing in society in general. I’m not brave enough to gaze into the crystal ball as John did in 1992, but I am sure that LC is going to be quite different in 2036 compared to what we see today.

Smart LC Instruments are Getting Smarter

Development of Machine Learning Models for Automated Detection of Obstructions in LC Flow Paths

One of the topics I have written about most in this column is troubleshooting pressure issues in LC systems.² In the case of pressures that are higher than expected, sometimes the pressure increase from normal is sudden, and sometimes the increase is gradual. Experienced users know that it is important to monitor the pressure over time (that is, across many analyses), even if it is at the back of our minds. Changes in pressure when running the same method repeatedly can be an early warning sign that something is not right. But, as far as I know, no commercial chromatography data systems (CDSs) provide this kind of early-warning feedback based on acquired pressure data.

A poster presented at the HPLC 2026 meeting by Waters Corporation’s Erika Stark gave some insight into work she is doing to fill this gap. The poster, entitled “Method Development of Artificial LC Clogs for Machine Learning Applications,” described the development of conditions that could reliably produce partial obstruction of an LC flow path for the purpose of generating datasets that can be used to train machine learning models that, in turn, could be used as the basis of an early-warning system for pressure-related problems. The poster described the choice of a sample composition that, when injected repeatedly, led to slow increases in pressure over time—exactly the type of phenomenon we observe in real experiments. Although it might be overly ambitious for individual users to acquire these data and build a machine learning model to support early detection of pressure problems, the work presented in the poster suggests that we may soon see such early-warning systems in commercial CDS offerings. This is an exciting development, in line with John Dolan’s predictions in 1992 about smarter LC systems.

Development of Machine Learning Models for More Reliable Integration of “Messy” Chromatograms

Experienced LC users have encountered situations where the peaks in a chromatogram are not well resolved or have undesirable features such as tails or shoulders, despite extensive attempts to improve the chromatography to yield well-defined and resolved peaks. In my experience, this is not uncommon when working with biomolecules, which can come to us as highly heterogeneous mixtures even when we are supposed to be working with a single protein, for example. This type of situation can be problematic when trying to quantify the various species present in the sample, because clearly defining the features of the chromatogram (for example, baselines and start and stop points for peak integration) is difficult to do in a highly reproducible way, even for a single analyst. When multiple analysts are involved in the workflow, different views of how baselines should be drawn or where peak integrations should be started and stopped can introduce even more variability into quantitative results.

In an oral presentation in Indianapolis, Scott Trinkle of Waters Corporation gave a talk entitled “An Ensemble Machine Learning Approach for Custom Chromatographic Peak Integration.” As the title suggests, this work aims to address this problem of poor reproducibility when dealing with “messy” chromatograms that don’t have well-formed and cleanly resolved peaks. Trinkle presented results from a workflow that involves training a machine learning algorithm using human-curated chromatograms. This process effectively teaches the algorithm the patterns that exist in the chromatograms manually integrated by humans, whether those users consciously thought about the underlying pattern or not. The improvements shown in the reproducibility of the integration process for challenging chromatograms acquired for bovine serum albumin and beta-casein were encouraging. Trinkle showed a two- to three-fold reduction in the root-mean-square variation in the start and stop integration points for the machine learning algorithm compared to automated integration using an existing commercial CDS. It seems that soon we can also look forward to increased “smartness” in commercial CDSs that improve reproducibility when dealing with messy chromatograms.

Deeper Understanding of Instrument Characteristics Can Mitigate Problems with Method Transfer

In a poster presentation entitled “Addressing the Challenges of Thermal Mismatch on Sub-ambient Temperature Separations of siRNA Oligonucleotides,” Alnylam Pharmaceuticals’ Gregory Jones discussed some of the challenges associated with transferring oligonucleotide analysis methods between LC instruments from different manufacturers. The focus of Jones’ presentation was on methods involving sub-ambient temperatures. In the discussion, he reminded us that instruments from mainline LC manufacturers do not measure and report the temperature of the LC column itself, nor the mobile phase as it enters the column (I am aware of a commercial column thermostatting apparatus available as a standalone device that does measure the mobile phase temperature at the column inlet and outlet). Rather, the temperature of the mobile phase is modulated as it passes through a heat exchanger, fixed to a block of material that is heated or cooled depending on the column setpoint temperature specified in the CDS. The extent to which the actual column temperature deviates from the setpoint temperature depends upon a number of factors, including the flow rate, mobile phase composition, temperature relative to the setpoint, volume of the heat exchanger, and the materials used in the construction of the heat exchanger and the block it is fixed to. Moreover, because some of these factors are manufacturer- and design-dependent, the extent to which the column temperature deviates from the setpoint will also depend on the make and model of the instrument. These differences can lead to different chromatographic results when the same method is executed on different models of instruments from the same manufacturer, or on instruments from different manufacturers, particularly when the separation outcomes are especially sensitive to temperature.

If it is anticipated that a new method may at some point be transferred between different instruments, then it is critically important during method validation to evaluate the method’s robustness with respect to variations in temperature. Understanding how column temperature is or is not controlled, and what the temperature setpoint in the CDS actually means, puts us in a position to rationalize differences in chromatographic outcomes we see with different instruments, and potentially even implement correction factors for use in method transfer if the behaviors of the different instruments and sensitivity of chromatographic outcomes to temperature are understood in sufficient detail.

Deeper Understanding of Analyte Interactions with Mobile and Stationary Phases Can Support Development of Better Methods

If the spirit of the preceding section can be represented with the axiom “know thy instrument” (which Xiaoli Wang of Agilent Technologies attributed to the late Professor Peter Carr during our memorial session honoring him in Indianapolis), this section could be represented by “know the chemistry of thy method.” As Martin Gilar and I have been discussing in a series of recent “LC Troubleshooting” articles, the chemistries of different types of separations of oligonucleotides are not well understood (for example, see Figure 1 in Part II of the series³). In an oral presentation entitled “The pH Modifier Moonlighting on the Stationary Phase: HFIP’s Unrecognized Role in Oligonucleotide IP-RP Selectivity,” Eli Lilly’s Joshua Jones provided insights on the role of hexafluoroisopropanol (HFIP) in these separations, which is commonly used as a mobile phase additive in ion-pairing reversed-phase separations. As the intriguing presentation title alludes to, the presence of HFIP adsorbed on the reversed-phase (RP) stationary phase may be playing a role in determining retention and selectivity for oligonucleotides (ONs) that is more pronounced than was previously appreciated. Given the vigorous research activity around ON separations in general, and the desire for a more detailed understanding of the chemistries of how these separations work, I think we can all look forward to the full publication describing this work of Jones and his colleagues. More detailed understanding of how our analytes interact with the mobile and stationary phases invariably improves both the efficiency of the method development process and the robustness of the resulting methods.

Summary

In this installment of “LC Troubleshooting,” I have highlighted several presentations from the recent HPLC 2026 conference in Indianapolis related to various aspects of problem prevention and troubleshooting in LC. Several presentations at the meeting indicated that we can anticipate higher levels of instrument “smartness” in the near future. Along with the proliferation of machine learning and artificial intelligence that we are seeing in society in general, we are also seeing the development of new capabilities based on machine learning in LC instruments and their associated chromatography data systems. Specifically, I highlighted two presentations that discussed the use of machine learning for early warning about potential pressure problems and for improved reproducibility when integrating peaks in messy chromatograms. As I reflected on two other presentations, I viewed them as helping us to better “know our instruments” (specifically, the effect of differences in column temperature control on method transfer) and “know the chemistry of our methods” (specifically, the role of hexafluoroisopropanol in ion-pairing reversed-phase separations of oligonucleotides). Building up this knowledge positions us to construct better and more robust methods in the first place and to troubleshoot problems more efficiently when we do encounter them.

References

[1] Dolan, J. LC Problems – Past, Present, and Future. LCGC 1992, 10, 202–210.

[2] Stoll, D. R.; Grinias, J. P. Treat It Like a Circuit, Part II: Applications and Troubleshooting. LCGC Int. 2024, 1 (5), 6–12. DOI: 10.56530/lcgc.int.wy6073a3

[3] Gilar, M.;, Stoll D. R., Challenges and Solutions in Oligonucleotide Analysis, Part II: A Detailed Look at Ion-Pairing Reversed-Phase Separations. LCGC Int. 2026, 3 (2), 8–15. DOI: 10.56530/lcgc.int.mh1387e9