September 1, 2015 (Vol. 35, No. 15)
Angelo DePalma Ph.D. Writer GEN
Your Surveillance of Cell Culture Conditions Can Be As Unblinking As the Eye of Providence
To cope with increasingly complex cell culture media and protocols, developers are looking for ways to optimize feed additions to fed-batch cultures. “Cell culture-based biotechnology is lagging in terms of available control and measurement techniques,” says Bill Campbell, CEO of Stratophase. Developers are also trying to deal with the shift from several-liter benchtop bioreactors to mini- and microbioreactors.
This shift also raises control and measurement issues. In particular, it calls for improvements in sensors at the mini- and microscales. These sensors need to mimic more accurately the control capabilities of pilot- and manufacturing-scale bioreactors.
Traditional glass or stainless-steel sensor probes have served bioprocesses from several liters and up, but they are impractical for, say, TAP Biosystems’ 15 and 250 mL parallel bioreactors. Even most single-use sensors, of the type integrated into plastic bioreactors, are too bulky and costly for operations at very small scale.
Campbell believes his company has solved the problem through commercialization of technology developed at the University of Southampton in 2003. This technology, incorporated in the company’s Ranger line of controllers and sensors, makes it possible to track compositional changes via refractive index (RI) measurements. RI is an under-utilized detection mode in bioprocessing.
The sensing component is an optical waveguide—a 10 micron diameter optical fiber. Campbell calls it “the optical equivalent of an integrated circuit.”
Laser light is sent through the control unit and down the fiber into the process. An optical component (a Bragg grating) at the end reflects a very narrow band of light back down the fiber so that an instrument can read it.
Unlike conventional sensors for, say, dissolved oxygen, this device does not quantify discrete process parameters, but rather the overall state of the cell culture. Specifically, it monitors RI values indicative of metabolic activity. When those values fall, the controller triggers a feed addition.
RI sensors suitable for repeated use in large bioreactors are typically built into stainless steel probes that are about the size of a pencil. But because the active sensing area has the dimensions of an optical fiber—about 1 mm across if you include the protective jacket—these sensors can be made extremely small.
“We have significantly miniaturized the device itself without miniaturizing the optical sensing component or losing performance,” Campbell asserts. “The only size limitation is how small you can make the silicon to be able to handle the sensor, mount it, and present it to the media in the bioreactor.”
The sensors provide real-time measures of process kinetics. Unlike spectroscopic methods, they require no “learning” because RI is a relative measurement.
Another benefit is tight control over feeding. According to Campbell, cultures that are fed lean are more productive: “This technique lets the cells tell you when they want to be fed, rather than leaving you to impose your own feed schedule on the culture.”
Process Analytic Spectroscopy
Perhaps the most significant trend in bioprocess sensing is in spectroscopy, specifically, Raman and near-infrared (NIR) spectroscopy. Both promise to advance process analytical technology (PAT), a quality-by-design (QbD) approach that uses online monitoring and automated controls to enable real-time process adjustments.
The big Raman player is Kaiser Optical Systems, which manufactures the RamanRxn3™ line of Raman process analyzers. A leader in NIR is Sartorius Stedim North America. The company has developed the BioPAT® Spectro in situ NIR spectrometer for pilot and manufacturing use.
Sartorius Stedim’s Dan Kopec, a field marketing manager for process analytics, is confident of NIR’s utility. Ask him what is measurable with NIR, and Kopec responds, “A better question is, what can’t you measure with it?”
BioPAT Spectro takes an NIR snapshot of the cell culture, from 950–1,750 nm, every 15 seconds. It quantifies nutrients, metabolites, product, and cell viability by analyzing specific characteristic wavelengths of chemical components, and through light scattering for viability.
NIR spectra may also be used qualitatively by applying multivariate analysis to a spectrum’s fingerprint region. The software, developed in partnership with data analysis specialists Umetrics, is first calibrated or trained offline on simulated processes or historical data. During an actual run, operators generate a multivariate data analysis batch trajectory, which is a simplified representation of all parameters under investigation including inputs from classic sensors and spectrometers.
“We would not be able to detect glucose, lactate, glutamine, glutamate, ammonia, etc. with univariate analysis,” Kopec explains. “You absolutely need multivariate software to manage that data.
NIR has its limitations, particularly with respect to concentration. For example if glucose concentration falls below 0.5 g/L, accuracy and precision fall. And glutamate concentration is obtained from spectral correlations rather than direct measurement.
Although NIR poses interpretive complexities and concentration limitations, Kopec remains bullish on the technology. How, then, does he respond to the question, why isn’t everyone already using it? Kopec mentions the usual barriers to new technology adoption in bioprocessing: “From my experience in the food and chemical industries, bioprocessing is far behind in terms of analytics. More than 10 years have passed since PAT was promulgated, and biotech is just now beginning to appreciate advanced analytical sensors.”
Sensors for pH, dissolved oxygen, carbon dioxide, redox, temperature, and pressure are “pretty much in the bag” for stainless-steel bioreactors, according to Ken Clapp, senior global product manager, Xcellerex, a GE Healthcare’s Life Sciences brand that focuses on bioreactors and fermentors. “Manufacturers enjoy plenty of choices with respect to technologies and suppliers,” says Clapp, “for both cell culture and fermentation.”
Sterilization of sensor and probes remains challenging, however. Sterilization or autoclaving are harsh treatments, particularly for cutting-edge sensing technologies. Another limitation, for stainless and plastic alike, has been the need, or desire, to maintain the sensing device as close as possible to the process, preferably inside the tank. Bioprocessors avoid external recirculation loops whenever possible.
“Some of the more interesting spectroscopy modes—infrared, Fourier transform infrared, and Raman—were migrated from other industries,” notes Clapp, “which makes proximity to the process even more difficult.”
Acquisition and deployment costs, and requisite calibration, add to the difficulties.
Emerging interests, spurred in part by PAT and QbD initiatives, include monitoring intermediate metabolites, carbon source consumption, and protein concentration. Sensors exist for those measurements, but adapting them for bioprocessing cost-effectively has been tough.
“There has certainly been progress,” Clapp observes. He also expresses optimism that as modes such as NIR and Raman become more widely accepted, vendors will have the resources to work through the key difficulties:
- Process proximity
The original Fourier transform infrared devices were spectrometers, not true probes. Their origin was the chemical industry. The optics required to make infrared work did not lend themselves to small, clean, confined biomanufacturing. “We spent a lot of time transferring those technologies to fiber optics to help make them more manageable,” Clapp remarks. “That helped adoption as well.”
With the form factor problem cleared, the next hurdle was cost. Early on, Clapp estimated the cost per measurement at $100,000, which is beyond the reach of most companies. He defines “measurement” as a continuous reading for one species throughout the run. Switching to fiber optics allowed the use of an optical multiplexer that collects data from up to four sensors, thus reducing cost per measurement by two-thirds and enabling the acquisition of four signals simultaneously.
“Yes, that’s still expensive,” Clapp admits, “but if a certain parameter is critical to process quality, then it’s probably worthwhile.”
Naysayers or Realists?
Not everyone is enamored with process analytic spectroscopy. “Unless you have a really well-defined spectrum with sharp peaks, NIR works only for qualitative determination,” states Mark Selker, Ph.D., CTO at Finesse Solutions, which specializes in point sensors. “You have noise and ambiguity in any signal and amplification chain, so it’s hard to tell where the peaks are. That noise and ambiguity translate to lack of accuracy and lack of precision.”
So much for NIR. Dr. Selker, who earned his doctorate studying Raman scattering, is slightly friendlier to that technique: “It’s nice, but it also has limitations.”
“While Raman has the capability of delivering in a bioprocess setting, the way it’s currently used generally will not provide the desired results,” Dr. Selker explains. “There are great Raman spectrometers and holographic gratings out there, but they are used with a training set and partial least squares to do the spectral fitting/concentration prediction, and that simply won’t cut it for bioprocessing. Too much noise. Unlike small molecules, large molecule combinations in cell culture are too complex.”
Govind Rao, Ph.D., director of the Center for Advanced Sensor Technology and professor of chemical, biochemical, and environmental engineering at the University of Maryland, Baltimore County, also points to difficulties in applying spectroscopy to bioprocesses: “Interpreting the signals and correlating them to process events is nontrivial. Even with a very well defined process, the best you can hope for is a fingerprint of sorts.”
“Training will be involved,” he continues. “You will do repeated runs under exactly the same conditions and see what the measured profile looks like. Then you will correlate that to process or product characteristics.”
Unlike direct measurement of pH or temperature, data acquired for one process may not apply to another. And spectroscopy is expensive to conduct in a bioprocess environment.
However, Dr. Rao believes that advances in computing power and analytical hardware may eventually make techniques such as NIR and Raman more widely used than they are currently despite the complexity of fermentation and cell culture processes.
Quantitative PCR as a Process Sensor?
At the University of Maryland, Baltimore County, Govind Rao, Ph.D., has been investigating quantitative polymerase chain reaction (qPCR) techniques to monitor cells’ metabolic status. Dr. Rao’s group has identified a panel of genes associated with metabolism, DNA damage, apoptosis, and other indicators of culture health, whose expression is independent of scale. The assay involves running qPCR for all genes of interest and noting whether these genes are up- or downregulated compared with a reference value.
Because the expression of these genes is scale-independent, they serve as indicators that can show whether process conditions were preserved during scale-up or scale-down. “The point of process development,” insists Dr. Rao, “is to have cells behaving the exact same way, so your product quality is consistent.”
His technique might also be applied during clone selection or cell line development.
“The downside is, you need to sample the bioreactor,” admits Dr. Rao. “But I think it’s well worth it because of the enormous insight you obtain regarding cellular metabolism. RT/PCR is an almost real-time measurement.”