Chemometrics and Intelligent Laboratory Systems
Real-time monitoring and chemical profiling of a cultivation process
Introduction
Handling the expression of enzymes in microbial cultivation on an industrial scale is a multi-faceted discipline. Several critical factors have to be controlled in order to ensure a profitable outcome. The bioprocess itself depends on available nutrients, cofactors and accumulated metabolites. Control of the bioprocess depends on representative sampling and relevant (preferably low cost, real time) assessment tools as well as insight into the critical factors affecting the cultivation. Enzyme production typically starts with a vial of dried or frozen microorganisms, which have been selected or genetically modified to produce large amounts of the enzyme(s) of interest. The process is initiated in, e.g., a Fernbach flask and after several cultivation steps resides in a large bioreactor, where conditions of nutrients, temperature, pH, and dissolved O2 are carefully controlled. When the main cultivation is complete, the mixture of cells, nutrients, and enzymes are transferred to a recovery step, where the enzyme is filtered from the broth. Real-time monitoring of a bioprocess is advantageous because it ensures product delivery by allowing keeping the process in control and it enables a much faster optimization effort and error handling. More importantly, real-time product assessment enables subsequent recovery steps to take the known variations in quality (enzyme activity) into account immediately. Several sensors for bioprocesses are available to the market but only a few of them are actually able to unfold the actual state of the cells, i.e., describe the intracellular condition in the cultivation at hand as well as describing the medium from which the cells are nourished. Changes in auto-fluorescence have previously been reported capable of describing relevant physical, chemical and biological parameters in cultivations [1], [2], [3], [4], [5], [6], [7], [8].
In the present study, fed-batch cultivations are monitored with a multi-channel fluorescence sensor. The measurements are performed at-line approximately 30 min after sampling the bioreactor. The primary scope of the monitoring is to enable an instant characterization of the batch with respect to enzyme activity replacing the currently used wet-chemical characterization, which takes from hours to days to perform, depending on the workload of the quality control laboratory. Secondly, the data are used for a more thorough description of inferential parameters in the cultivation process at hand. The full potential of the sensor is not yet utilized due to the at-line approach and the reason for that is the desire to optimize and validate the conditions of analysis before further advancing towards an in-line setup.
The paper is organized as follows. First, an experimental section is given that outlines the background for the present study; this includes an important reflection on sampling representativity issues. Second, a section with results and discussion of the results is provided, and finally, a conclusion.
Section snippets
Bioprocess
The fed-batch cultivation under investigation is typically performed in a 50 m3 bioreactor (larger bioreactors also exist). The microorganism is a genetically modified bacterium (Bacillus) expressing a protease. Bacteria are mainly nourished on complex medium consisting of potato protein (N-source) and sugar (C-source). The pH is adjusted with ammonia, which also acts as an additional N-source. Other vital components are various salts and trace metals.
In the present work 283 samples have been
Data analysis
As mentioned in Section 2.1, 25 batches (283 samples) are the basis for the present data analysis. Further, 3 batches (53 samples) function as an independent test set for the developed models.
One part of the data analysis compares different regression methods for calibrating the fluorescence data with the reference analysis for enzyme activity using either unfold Partial Least Squares Regression (unfold-PLSR), N-way PLSR, or PLSR on PARAFAC scores.
Another part of the data analysis visualizes
Enzyme activity
The scope of predicting the enzyme activity is to enable better process control and enable faster detection of the batch end point. Three different regression approaches between the reference analysis and the multi-channel fluorescence sensor are tested. Results are displayed in Table 1. The models are built using cross-validation and subsequently the predictive ability is tested on the test set.
It is noted that the test set is generally better predicted than the cross-validation results would
Conclusion
In conclusion, the presented approach, using an at-line multi-channel fluorescence sensor, enables immediate quality assessment (with a primary sampling representativity caveat) as well as a more profound analysis of what is going on in the bioprocess of making enzymes.
Three different methods for predicting the enzyme activity in 25 batches has been tested; bi-linear PLSR and tri-linear PLSR are found to be good and superior to bi-linear PLSR on PARAFAC scores. The uncertainty of the
References (31)
- et al.
Chemometric modelling with two-dimensional fluorescence data for Claviceps purpurea bioprocess characterization
J. Biotechnol.
(2003) - et al.
On-line cell mass monitoring of Saccharomyces cerevisiae cultivations by multi wavelength fluorescence
J. Biotechnol.
(2004) - et al.
Factors affecting culture fluorescence when monitoring bioreactors
J. Ferment. Bioeng.
(1992) - et al.
Representative sampling for reliable data analysis: theory of sampling
Chemom. Intell. Lab. Syst.
(2005) - et al.
Representative mass reduction in sampling—a critical survey of techniques and hardware
Chemom. Intell. Lab. Syst.
(2004) PARAFAC. Tutorial and applications
Chemom. Intell. Lab. Syst.
(1997)- et al.
Partial least-squares regression: a tutorial
Anal. Chim. Acta
(1986) - M.B. Haack, A. Eliasson, P. P. Mortensen, L. Olsson, Chemometric analysis of in-line multi-wavelength fluorescence...
- et al.
In situ fluorescence mass measurements of Saccharomyces cerevisiae using cellular tryptophan
Biotechnol. Prog.
(1993) - et al.
Monitoring concentration and activity by multiple excitation fluorometry
Biotechnol. Prog.
(1991)
Chemometric modelling based on 2D-fluorescence spectra without a calibration measurement
Bioinformatics
Estimation of fermentation biomass concentration by measuring culture fluorescence
Appl. Environ. Microbiol.
Handbook of Mixing Technology
Sampling for Analytical Purposes
Pierre Gy's Sampling Theory and Sampling Practice: Heterogeneity, Sampling Correctness, and Statistical Process Control
Cited by (36)
A review on multivariate curve resolution applied to spectroscopic and chromatographic data acquired during the real-time monitoring of evolving multi-component processes: From process analytical chemistry (PAC) to process analytical technology (PAT)
2022, TrAC - Trends in Analytical ChemistryCitation Excerpt :Therefore, the present review proposes a consensual and practical definition, which appropriately condenses the subtle differences between PAC and PAT in order to preserve the legacy of PAC [1,38], and, at the same time, to highlight the challenges that it has to face, aiming to meet “process” requirements, under the PAT framework, to establish a quality-by-design (QbD) approach [25]. Spectroscopic and chromatographic data acquired throughout the real-time monitoring of the evolution of multi-component processes [39–42] in combination with either MCR-ALS [13–15,20,33,34] or other resolution-based chemometric models, can be considered as: PAC protocols are mainly directed to monitor (bio)chemical reactions during a given evolving process [37], including the production stage, as it is confirmed by the well-documented evidence showing that one of the companies that mostly benefitted from these tools was The Dow Chemical Company [1].
Parallel factor analysis for monitoring data from a grape harvest in Qualified Designation of Origin Rioja including spatial and temporal variability
2015, Chemometrics and Intelligent Laboratory SystemsA fluorescence anisotropy method for measuring protein concentration in complex cell culture media
2014, Analytica Chimica ActaCitation Excerpt :Fluorescence spectroscopy is an attractive alternative for near real time on-line/at-line and non-destructive monitoring of biogenic process samples and raw materials quality control [12]. This is because some significant constituents of culture media are intrinsically fluorescent, like riboflavin, pyridoxine, and the amino acids phenylalanine (Phe), tryptophan (Trp), and tyrosine (Tyr), and furthermore they can be detected at low concentrations (mg–μg mL−1) [12–15]. However, the compositional complexity of cell culture media causes problems with fluorophore emission overlap.
Process control in biogas plants
2013, The Biogas Handbook: Science, Production and ApplicationsMultiway Methods
2013, Data Handling in Science and TechnologyCitation Excerpt :The models must be able to give a fast response to take a fast decision. Fermentation processes, for instance, are clear examples in which the combination of EEM with PARAFAC gives a perfect framework for a real-time and fast decision set-up [32,67–69]. Several attempts have been made for the automation of multiway models.