Comparison of vibrational spectroscopy methods for real-time monitoring of anaerobic fermentation

Abstract

A central idea in the transition to a circular economy is to recover resources from wastes. An emerging biorefinery for carbon recovery from waste streams focuses on the production of carboxylates or volatile fatty acids (VFA) by anaerobic fermentation. VFA have value on their own but also as precursors to other chemicals, biofuels and biopolymers constituting the so-called carboxylate platform. A challenge in the development of the carboxylate platform is the poor selectivity in a given VFA (e.g. acetic, propionic, butyric acid…) and the lack of thorough understanding of how changes in operational conditions (pH, retention time, substrate nature and concentration) have an impact on the VFA produced. Currently, the analysis of products, often present in aqueous solutions at moderate/low concentrations, tends to be done by gas chromatography (GC) or high-performance liquid chromatography (HPLC) which prevents real-time or near real-time monitoring.

Spectroscopic methods can enable faster monitoring rates if configured at-, in- or online. The interest on vibrational spectroscopy (Raman, infrared) has steadily grown in the last 15 years thanks to the decreasing cost of the monitoring equipment. The potential of vibrational spectroscopy in anaerobic fermentation monitoring lies in the possibility of quantifying the main metabolites and substrates (e.g. glucose, ethanol, starch, amino acids). However, there are important challenges too: a major obstacle for infrared spectroscopy in bioprocess monitoring lies in its high sensitivity to water with high bands that overlap other analytes. In contrast, Raman spectroscopy is particularly interesting for bioprocesses, precisely, because water is mostly Raman inactive. However, a large number of molecules present in mixed-cultures lead to fluorescence in Raman. In this work, we compare the capability of Fourier-Transform Infrared (FTIR) and Raman spectroscopies to effectively monitor the main metabolites of anaerobic fermentation.

Aqueous mixtures containing acetic, propionic and butyric acid (n = 30; range 0,1 g/L to 20 g/L) were prepared with concentrations designed by latin hypercube sampling as described by Cabaneros-Lopez et al. (2020) to ensure high coverage of the sampling space while keeping at minimum in-sample correlation. The mixtures were measured by Attenuated-Total Reflectance (ATR) FTIR in the medium infrared range and by Raman spectroscopy between 95 and 3500 cm-1 Raman shift wavenumber using a 785 nm excitation laser. Spectra were preprocessed (baseline correction, Savitzky-Golay noise filtering) and used as input for several chemometric models. Spectra were split in calibration (20 samples) and validation (10 samples) sets. Partial least square (PLS) models for each of the analytes were finally selected based on the goodness of fit for the calibration and validation models (Figure 1) for the three analytes. Acetic and propionic acid were satisfactorily estimated while butyric acid estimation is being improved.

Despite the influence of water in the obtained spectra, FTIR results were superior to Raman spectroscopy results for the three analytes at all range. When tested in the fermentation media using glucose as a substrate, Raman spectroscopy led to severe fluorescence phenomena, hence preventing an accurate estimation of the VFA in the solution. (data not shown). Infrared spectroscopy was hence selected for VFA monitoring.

Upcoming work will deal with the combined use of mathematical models, in particular observers (e.g. nonlinear versions of Kalman filters) to integrate the information obtained by spectroscopy with models of anaerobic fermentation (Regueira et al. 2021) in order to integrate all the available information