September 1, 2011 (Vol. 31, No. 15)
Angelo DePalma Ph.D. Writer GEN
Infrared and Raman Methods Gradually Breaking Through but Require Specific Skills
Spectrometric techniques like near-infrared (NIR) and Raman are among the most appealing alternatives to traditional sensors for bioprocess monitoring. These techniques are robust, scalable, and capture tremendous quantities of process data in real time. Their only drawbacks are that they take effort and knowledge not only of spectroscopy, but of chemometrics and multivariate analysis.
The use of NIR has evolved, initially from raw materials testing, to final product quality control, and most recently to process monitoring. Advances in computing and chemometrics have contributed to this growth, together with the realization that NIR is suitable to microbial, fungal, yeast, and mammalian cell cultures (although not equally).
NIR’s appeal stems from its advantages over traditional in- or at-line techniques: flexibility, simplicity of operation, rapidity of measurement, multiplicity of analysis, and the potential to predict chemical and physical parameters from a single spectrum.
Yet despite fulfilling nearly all objectives of FDA’s Process Analytic Technology (PAT) initiative, adoption of NIR has been less than inspiring.
Not Ready for Prime Time?
AstraZeneca for example has experimented with NIR at the bench. “But we haven’t yet had the opportunity to explore it at pilot scale,” admits Payal Roychoudhury, Ph.D., senior scientist at the company’s Mölndal, Sweden facility. “To be honest, it is not used as extensively in process monitoring as it should be.”
NIR has numerous advantages for PAT programs, particularly with respect to critical-to-quality factors. NIR provides rapid, real-time, noninvasive, cost-effective monitoring with no sampling or sample preparation. Furthermore, it may be multiplexed with respect to both parameters and bioreactors.
It is also scalable and transfers well. “This is really important,” Dr. Roychoudhury explains, “as in its life cycle a typical biopharm product will transfer across scales and relocate geographically. Analytical models that are robust in the face of such changes are potentially applicable throughout the product’s life cycle.”
NIR is not perfect. It is considered a secondary technique, meaning it does not directly measure analytes through intimate contact with a sensor. Users must expend significant effort into modeling, and the stability of fiber-optic probes it employs has been questioned.
For example, after repeated sterilization cycles, the NIR probe might undergo signal saturation or show drifting in spectral signals. “The fiber optics available today are generally silica or quartz and are quite fragile,” Dr. Roychoudhury notes. “Microfractures may, over time, lead to complete loss in transmission.”
NIR spectra are collected over the entire spectral range, from 800 to 2,200 nm, and provide multianalyte readings simultaneously in real time including cell counts, metabolite and nutrient uptake rates, and product levels. However, as Dr. Roychoudhury points out, the predictions from NIR are only as good as the reference assay method.
“Analytical modeling requires sophisticated mathematical treatments such as partial least squares in correlating the spectral data with the reference data for the respective analytes.”
NIR works especially well in cell culture (vs. fermentations) because the medium is clear and cell density, gassing, and agitation are low compared with microbial cultures. Barriers to adoption exist, however, including:
- investment of time, personnel, and equipment, particularly for one-off projects;
- validation and modeling;
- high level of user familiarity with the theory, mathematics, and instrumentation; and
- technology transfer issues, particularly transferring data from one system to another.
Functional groups do not absorb strongly in the NIR. This is a disadvantage for traditional benchtop analysis, but a powerful benefit in bioprocessing, says Robert Mattes, applications scientist at Foss NIRSystems.
NIR can simultaneously measure glucose glutamine, lactate, ammonia, pH, osmolality, and viable cell density. According to Mattes, NIR can distinguish between live and dead cells, which otherwise involves a time-consuming assay.
NIR is still considered an early-stage technology for bioprocessing. Biogen Idec and MedImmune have presented data on PAT applications. FDA appreciates the potential of NIR but has stopped short of endorsing it.
But NIR is not a simple plug-and-play technique. “It’s not like a pH meter where you stick in the probe and get a number. NIR requires someone with an understanding of spectroscopy,” Mattes says. “People get excited about it, but then other high-priority projects come up or those championing the technology leave the company.”
In addition, chemometrics, statistics, spectroscopy, and multivariate analysis take time and skill that most biotechnology companies lack.
Your Dissolved Oxygen Is Dangerously Low
Vendors have for years pushed wireless transmission of process-monitoring data as an alternative to cabled connections. Under this scenario, sensors are coupled with radio transmitters that broadcast data signals to receivers throughout the plant. Broadcast systems work well at very large industrial facilities like refineries but are not particularly suited to bioprocess plants, says Larry West, principal consultant with Aspenbrook Consulting.
Bioprocessors appear to have skipped the sensor-with-an-antenna step, however, as uptake was never as high as monitor/control firms anticipated. The next big thing is direct transmissions of process data to smartphones.
“Employees at biotech companies have been bugging management for smartphones for years. Management now realizes the business case for pushing data points out to key personnel, wherever they may be, through the existing communications infrastructure,” West explains.
The idea has “taken off like wildfire,” West says, at companies like Amgen, Pfizer, and Genentech.
The rationale is to leverage, in this era of downsizing and “de-skilling,” the contact that the ever-shrinking number of decision-makers have with critical processes.
While large manufacturing campaigns are attended around the clock by skilled operators, fewer and fewer people, says West, understand what to do when an alarm condition arises or the process goes even slightly out of spec. “Plenty of people know what to do, but many fewer understand why. These ‘knowledge centers’ need to be as accessible as possible, around the clock.”
For now these applications are limited to monitoring: Managers must still phone in with instructions. The next logical step, once regulatory and technical issues are ironed out, is to enable personnel to log in and effect necessary changes remotely.
Based on his ongoing work as a chemistry professor at the University of Iowa, Mark A. Arnold, Ph.D., vp at ASL Analytical is working on NIR monitors for Pichia pastoris fermentation. Dr. Arnold believes that Pichia fermentations are a more logical way to introduce NIR industry-wide.
From a monitoring perspective, Pichia fermentations are more straightforward than cell culture, and more similar to one another as well. “This allows us to build calibration models that are closer to platform monitoring than we could with CHO cells,” Dr. Arnold says.
The monitor measures glycerol, methanol, and biomass in real time. Pichia use glycerol as their primary carbon source, and methanol to induce protein production.
As with other spectroscopic methods, Dr. Arnold employs multivariate calibration methods to fish out analytes from a complex milieu. Eventually, he would like to incorporate feedback control for methanol levels.
Cell density, by contrast, is a general method that relates to light transmitted through the cell suspension and scattered by increasing cell mass.
NIR spectral data correspond to overtones and combinations of vibrations associated with O-H, N-H, and C-H bonds within the molecules of interest but not water. Species monitored during Pichia fermentations are present in significantly higher concentrations than protein products.
According to Dr. Arnold, many bioprocessing companies have attempted to develop NIR, Raman, mid-IR, and fluorescence techniques in-house for PAT-style analysis for online bioreactor monitoring. “They’ve all failed for the most part,” he says, which explains the slow uptake of spectroscopy in bioprocess monitoring.
Raman: The Anti-NIR?
Raman is another bioprocess monitoring modality that is long on promise, short on implementation—but not for lack of trying.
A group at Biogen Idec headed by John Paul Smelko has reported the use of in situ Raman for determining growth and metabolic profiles of CHO cell cultures in real time. Smelko’s objective was development of an inline PAT-enabling method.
Biogen Idec reported “high-quality quantitative results” for glucose, lactate, glutamine, glutamate, ammonium, osmolality, and viable cell density based on models developed at 3 L and 200 L scales.
Raman and NIR are complementary techniques. IR is based on absorption, Raman on scattering. The spectra in fact appear almost as mirror images. For both methods the location of the signal indicates what is being measured, and the height tells how much. And both techniques require chemometrics—a math-intensive technique for relating spectral bands from compounds of interest to the test system.
One downside of Raman, says Lee Smith, Ph.D., president of Process Instruments, is when highly colored or fluorescent materials are present in the sample. “They will swamp the weak Raman signal,” he says. A corollary is that Raman is not particularly sensitive below part-per-million concentrations.
Yet the technique has improved over the years. Instruments are smaller and much less expensive; advances in electronic and optical components have greatly improved Raman signals and expanded the technique’s suitability for monitoring bio- and other process industries.
Water interferes seriously in mid-IR, and to a much lesser degree in NIR, but in Raman the water interference peak is insignificant. “Without interference from water there is no need for spectral subtraction, and you get high chemical specificity for most of the components of interest in a bioprocess,” notes Maryann Cuellar, an applications scientist at Kaiser Optical Systems.
Functional groups absorb in similar regions in Raman and IR. Thus, users can apply traditional IR vibrational frequency charts to predict Raman signals.
Raman uses a laser light source, so the optical fiber delivery system can extend hundreds of meters from the bioreactor and probes can be quite robust. NIR has a much shorter useful range and requires “exotic materials,” according to Cuellar. Raman’s robustness has made it very popular in nonpharmaceutical process industries like materials, polymers, and chemicals, where the technique has been fully deployed as a process analytic for decades.
Small molecule drug makers are slowly warming up to Raman, particularly for solid dosage forms, but bioprocessors are apparently still sorting out details.
“We will need a paradigm shift for bioprocessors to adopt Raman fully,” says Cuellar. “There’s a serious lack of analytically trained scientists in bioprocessing, and most engineers and biologists are unfamiliar with spectroscopy. There is also a shortage of trained chemometricians capable of extracting information from Raman data. The technique is about more than just acquiring spectra.”
Monitoring for Process-Killing Pathogens
Monitoring contaminating viruses and bacteria during a bioprocess can be frustrating, as conventional tests are slow relative to the fermentation’s or cell culture’s life cycle. Undetected infections can affect production for months or longer, and negatively affect patients due to product supply constraints. Experts note that contamination is not a matter of if, but when.
For example, in 2009 Genzyme detected a virus, vesivirus 2117, in a bioreactor at its Allston Landing, MA, facility, after conventional testing failed to find the pathogen. Although benign to humans, vesivirus interferes with CHO cell growth and was responsible for productivity declines at Allston and at Genzyme’s Geel, Belgium, facility. These mishaps disrupted supplies of the company’s Cerezyme® and Fabrazyme® products.
Life Technologies sells tests, based on real-time PCR (RT-PCR), for detecting mouse minute virus, vesivirus, and mycoplasma, the three most troublesome pathogens in CHO cultures.
RT-PCR takes about three hours, and according to PAT jargon is an off-line (i.e., not real-time) technique. However, it is a vast improvement over conventional culturing methods that take up to a month. Three hour test times are more than adequate for determining whether to scale from smaller to larger bioreactors.
Despite the risks of contamination, the benefits of rapid testing, and success stories from some of the largest biotech players, not all companies are on board with RT-PCR as a risk-management tool. According to Mike Brewer, senior manager for bioproduction at Life Technologies, biomanufacturers may not be aware of the ease of implementation of kit-based RT-PCR risk-management programs.
“In the past, a possible drawback was the risk of a false positive result from an accidental cross-contamination of a test sample with the positive control, but we’ve eliminated that risk with our test design.”