Abstract :
[en] Retention indices (RIs) are widely used in GC and GC×GC as essential chromatographic descriptors to support compound identification, especially in non-targeted analysis. The ability to predict RIs for compounds not represented in existing databases can substantially strengthen identification workflows. However, developing robust prediction models first requires a clear understanding of the factors that influence RIs. In GC×GC, RIs values tend to be less consistent and show greater variability due to the additional operational parameters inherent to the technique.
To investigate these effects, a study was conducted using the Century Mix, a reference mixture containing around one hundred compounds representing diverse chemical families. A full factorial design of experiments was performed using both normal (RXI-5MS×RXI-17SilMs) and reversed (BPX50×RTX-5MS) column configurations to assess the influence of key instrumental parameters (temperature ramp, carrier-gas flow, modulation time, and oven offset) on RIs values. The temperature ramp emerged as the dominant factor driving RIs variation for several chemical classes, particularly phthalates, lactones, and PAHs, followed by carrier-gas flow. The compounds most affected were those containing π bonds, suggesting that π–π interactions are more sensitive to changes in operating conditions. In contrast, esters, ketones, and alcohol exhibited minor RIs shifts.
Complementary custom designs using three additional column configurations (Rxi-624Sil MS×Stabilwax, Stabilwax×Rxi-17Sil MS, and SLB-IL59×Rxi-17Sil MS) were conducted to evaluate the effect of stationary-phase chemistry and to determine whether the most sensitive compounds remained consistent across phases.
Overall, the results indicate that for certain compound classes, careful control of the temperature program is essential when comparing measured RIs with library values. Other classes, however, show greater stability across method variations. Notably, 2D-specific parameters such as modulation settings did not significantly affect RIs. The study highlights which operational parameters must be documented and integrated as input variables for reliable RI prediction models.
