When I first suggested using Near-Infrared (NIR) for biological processes back in the late 1980s, I was told that they were very complex and there were too many components to “train” a NIR equation. My first response was that when the spectra stopped changing, the reaction was finished. One observation that I have made, with the aid of 20/20 hindsight, was the make-up of the staffs of a small molecule (SM) facility and a macro-molecule (biotech) facility. In short, the technical staff of a SM producer is quite varied, consisting of clinicians, organic chemists, analytical chemists, formulators, operators, and a host of other tech types. Basically, they employ chemists, pharmacists, and biochemists. When a process stream is being monitored, there are all disciplines available to help the production staff, each skilled in a different technology.

In a biotech facility, the staff is largely homogeneous and steeped in biochemistry. That is, the basic R&D, scale-up, production, and analyses are almost all performed by scientists and technicians trained in the same biotech disciplines. This lack of diversity of disciplines almost precludes familiarity with all the newest technologies available for SM process control. That alone means a slower “warming” to “out-of-the-box” thinking for real time measurements by, for example, NIR, Raman, etc., which means slower adaptation to PAT or QbD. On top of that, the majority of the chemistry is the process. That is, in a SM process, the synthesis of the active is merely a first step in a long process. In Bio, this “synthesis” is the lion’s share of the process, with packaging being a freezing, freeze drying, etc. step.

While I do not have formal training in the field of synthesizing/generating macromolecules, I have some familiarity with fermentations. I do remember, for example, that the source of the “catalyst” for fermentation can be as weird as the sludge from a brewery. In many cases, “natural” starter can not only vary from batch-to-batch, but even a 5-gallon pail varies within itself. In batch-fed fermentations, the classic method of bringing the fermentation to its maximum yield, was to take a sample from the reactor, analyze it, and either add more “food” or end the reaction and harvest the materials.

The sampling was usually based on an SOP, wherein the sampling times were based on averages of past batches, not what was actually occurring in the fermentation. There are several problems with this approach:

1.  The operator could be exposed to the bacteria (often e-Coli) or could expose the anaerobic reaction to oxygen or even contaminants.
2.  The intermediate test could come too early or too late to be optimal, reducing the yield or causing side reactions.
3.  The traditional off-line tests add time to the process, tying up equipment for extra hours. This means, to meet quotas, more equipment and operators are needed, causing the COGS (cost-of-goods-sold) to increase. That means charging more or having a lower profit margin.

So, without major modification to the equipment or major capital outlays, what may be done to monitor the process in real time? I’m glad you asked. Every fermentation vessel has opening for adding materials, sampling the mixture, and, quite often, viewing ports. If we use Near Infrared or Raman spectroscopy, we can use a port to insert a fiber optic probe or view the reaction through the window. By a fortunate historical accident, NIRS was made “mainstream” by the U.S. Dept of Agriculture.

By a rule in the CFR (Code of Federal Regulations), only stainless steel and sapphire are allowed to come in contact with foodstuffs. As a consequence, since USDA was the driving force in remote NIR measurements, stainless and sapphire became the de facto composition for fiber optic probes. That composition makes them quite useful for bioreactor monitoring: they are easily cleaned and unreactive with biological materials.

What can we measure?

Let’s look at NIRS first. Some early work (1900s) for a batch-fed Escherichia-Coli fermentation was able to calibrate and predict levels of acetate, ammonia, biomass, and glycerol.¹ Clearly, neither the equipment nor algorithms were at 21st century levels (Figure 1), but it showed that it was possible to monitor a bioprocess in real time.


In the early 2000s, workers at NIRSystems examined CHO (Chinese Hamster Ovary) cell culture process for ammonia, glucose, titer, methionine, lactate, glutamate, and glutamine. Samples were taken at 25-minute intervals, while the NIR spectrometer (probe inserted in reactor) took continuous readings at 15-second intervals. The physical samples were assayed by compendial methods while the NIR instrument used a calibration based on control samples, previously generated. Figure 2 shows the comparison of NIR vs. chemical tests for glucose. Figure 3 shows the results of “wet” chemistry vs. NIR for the titer. Obviously, in 2023, these 1900s and 2009 methods have been greatly improved.



So, let’s also look at some of the potential of Raman. In a nutshell, water is a pain for NIR work, being the largest absorber, by far. But, since Raman works best on molecules without a permanent dipole, it hardly “sees” water, making it an outstanding choice for aqueous work. Unfortunately, the LASERs used for Raman generate a fluorescence signal that is often far larger than the Raman spectrum.

Recently, several techniques have been developed to obviate the effects of the fluorescence. Some work had been done using longer wavelength LASERs, but, while the fluorescence diminishes with longer wavelengths, so does the Raman signal. If, for example, the point where the radiation impinges upon the sample is a few millimeters from the fiber optic that channels the radiated light to the detector, the Raman signal predominates. This is especially strong in condensed media, since the fluorescence has a strong tendency to reabsorb in the sample, while the Raman signal does not. This is called SRS (spatially resolved spectroscopy).

Another company uses an algorithm that estimate the fluorescence signal and subtracts it from the combined spectra, generating a usable Raman spectrum. This may then be run through typical Chemometric equations to predict values in a rapid time frame. The third approach is called “time-gating,” where a short pulse (femtoseconds?) of LASER light is impinged upon the sample. The mechanism for fluorescence involves electrons absorbing photons, being excited to a higher energy level, then returning to ground state and emitting photons of a different wavelength.

As rapid as that mechanism is (~10-10 seconds or shorter), Raman is not a “proper” spectroscopy, but involves elastic and non-elastic collisions, where the photons can either gain or lose energy. These scattered photons emerge ever so slightly before the photons from the fluorescence. In a typical spectrometer, reading for seconds at a time, the two signals overlap and the Raman is buried.

In the time-gated instrument, the LASER pulse strikes the sample and almost immediately the window opacifies. The time between the LASER pulse and when the window closes is sufficient to allow the Raman signal to enter, but too short for the fluorescence signal to strike the detector. Figure 4 shows the gated instrument following the changes in cell density over time. Figure 5 shows how the medium and the cells themselves may be measured, independently.



These tools may be used to assist the operators as to when to take samples, add feedstock, or harvest the batch. They may also be included in a full-fledged PAT program. In either case, the time savings and increased yields will make them well worth investigating. 

References

1. “NIR Spectroscopic Determination of Acetate, Ammonium, Biomass, and Glycerol in an Industrial Escherichia coli Fermentation” J.W. Hall et al., Applied Spectroscopy, 50 (1), 102 (1996).