Leading the Way in Life Science Technologies

GEN Exclusives

More »

Columns

More »
November 15, 2014 (Vol. 34, No. 20)

Spectroscopy’s Real-Time Role in Process Analytics

Companies Slowly Beginning to Adopt the Technology for Close-Contact PAT Applications

  • Promulgated in 2004, and since supplemented by more recent guidances on quality by design (QbD) and risk-based manufacturing, the FDA’s process analytic technology (PAT) initiative has been slow on the uptake. However, biomanufacturers have made significant progress during the last few years. PAT’s difficulties arise from one of analytical chemistry’s immutable facts: the closer the analysis modality to the process, and the farther upstream, the more difficult its implementation.

    Of the several analysis techniques available for close-contact PAT, spectroscopy offers perhaps the greatest opportunity for real-time analysis of multiple parameters with minimal or no surface preparation.

    Hui Li, Ph.D., of Bruker Optics differentiates between primary and secondary PAT in small-molecule manufacture. Primary PAT monitors raw materials and chemical reactions; secondary PAT focuses on blend uniformity, moisture content, hardness, tablet integrity, etc. One could imagine an analogous model for biomanufacturing where upstream events (viability, nutrient depletion, presence/absence of gases) constitute primary monitoring, whereas homogeneity, physical characteristics, and yield-quantified downstream operations constitute secondary monitoring.

    Dr. Li prefers near-infrared (NIR) spectroscopy over mid-infrared and Raman.NIR’s advantages include speed, simplicity, lack of sample prep, accuracy, nondestructiveness, suitability for multicomponent analysis, and remote sensing capabilities. On the negative side, NIR requires some knowledge of chemometrics, provides less information than mid-infrared or Raman, is influenced by a sample’s physical characteristics, and requires long-term support for calibration.

  • Top-Down Deployment

    Adoption of spectrophotometric PAT is occurring slowly, from the top down. “Several large firms are already doing it,” says Emil Ciurczak, president, Doramaxx Consulting. These include Pfizer, which was a pioneer on the small molecule side, and Amgen.

    Ciurczak believes that bioprocessors are missing an opportunity by over-thinking the application of spectroscopy to biological processes. “You can insert an NIR or a Raman probe into a fermentation, and even if you don’t know what’s happening when the signal stops changing, you know the ‘reaction’ or some aspect of it is over,” asserts Ciurczak. “You can do kinetics because you’re looking at changes.”

    For example, bioprocessors currently rely on standared operating procedures to sample a production cell culture at various times, in anticipation of adding feed or adjusting pH. Correlating these events with changes in Raman or NIR signals, while not trivial, is not terribly challenging, according to Ciurczak.

    “If you have a spectrometric probe in there the whole while, and if you have taken enough samples over enough processes, two things will occur,” he continues. “You will eventually have an inline method to indicate when to take future readings, and without a lot of extra work, you will have a spectrophotometric monitoring method, even if you prefer not to use it for control purposes.”

  • Not So Fast

    Click Image To Enlarge +
    The NIRS XDS MultiChannel Process System from Metrohm may be fitted with different probes. Pictured here is the Interactance Immersion probe, which is suitable for transmission-style measurements on clear to partly clear liquids. The path length of the tip is set by use of a spacer.

    Not everyone agrees that modeling complex processes is a walk in the park. Cell culture and fermentations are strongly correlated processes, meaning that all interesting parameters follow certain trends based on dependencies of analytes. Nutrients convert to metabolites and inversely correlate with cell count, which correlates with product titer, etc.

    “It is therefore impossible to separate the impact of different parameters on a spectroscopic signal,” explains Marek Hoehse, Ph.D., who works on photonics and chemometrics at Sartorius Stedim Biotech. “Multivariate models frequently incorporate correlations and causalities. However, models based on correlations are valid only as long as the batches used to build the model behave the same way as production batches.”

    One way to avoid correlations in models is through spiking batches with quantities of the target analyte, which decouples it from the concentration of other analytes.

    Rather than employing optical fiber lightguides, the Sartorius Stedim Biotech approach employs free-beam NIR mounted directly to the process, which Dr. Hoehse claims is “more robust compared to fiber-optic NIR.” The larger spot size also compensates for fluctuation in the measurement signal.

    Because NIR is a secondary analysis technology, developing a model or method for continuous monitoring involves correlating NIR measurements with primary values on the same exact sample through reference-worthy methodology. “The accuracy of the primary method and the primary data will greatly influence the accuracy of the resulting NIR model,” says Hari Narayanan, Ph.D., senior product manager for spectroscopy at Metrohm USA.

    Some groups have noted difficulties in measuring multiple process parameters simultaneously with NIR. Dr. Narayanan attributes this to low instrument sensitivity or the types of probes used to collect NIR data. For bioprocess monitoring, Dr. Narayanan recommends microbundle fibers (40 illumination/40 collection).

    “Microbundle fibers allow the noise-free and highly reproducible collection of quality spectrum from the reaction medium, which facilitates building a calibration model,” explains the scientist. “When developing a robust model, it is important to include all common sources of variation that may affect the absorbance of the analytes being modeled.”

  • Why So Slow?

    If spectroscopy is so accessible and informative, why has it taken so long to catch on, and why doesn’t everyone use it? The time lag between regulatory guidance, concept, and implementation has been long for most PAT and QbD ideas for reasons mentioned earlier: Everyone wants to see how top companies do it. Dr. Narayanan believes that the obstacle to implementing NIR has been the initial time it takes to develop a model.

    But there’s a cultural issue here as well. A small molecule manufacturing facility has a “ton of chemists,” Ciurczak observes. “Lots of people are used to instrumentation that performs chemical analysis.” Not so at biotech companies, where biologists predominate. “It’s much easier to convince people who are already using infrared and Raman to apply it to their manufacturing processes.”

    In fact, spectroscopic PAT was not whole-heartedly embraced at small-molecule manufacturers, either. Management, says Ciurczek, initially feared that adding a layer of quality testing might uncover defects that had previously gone unnoticed, and for which they might have to start answering to regulators.

    Ciurczek cites the heavy burden of cGMPs that are strongly inculcated into pharmaceutical manufacturing engineers. “These people are bright, but they’re afraid of quick-and-dirty [approaches], of trying something that hasn’t been approved,” Ciurczak says. He cites as an example the long lag between the availability of reliable HPLC and inclusion of that method in pharmaceutical monographs.

    For Dr. Hoehse, the reason for slow adoption has been a lack of system robustness. For years in-process spectroscopy was limited to reconfigured laboratory instruments. “It took a while,” he recalls, “for NIR instruments to be designed specifically for process applications.”

    Dr. Hoehse echoes Ciurczek’s point on expertise: “It takes a while to perform successful trials and to convince people to trust NIR predictions, and to apply them with the correct multivariate data evaluation tools.” “For some users,” Dr. Hoehse adds, “this is still a kind of witchcraft.” 

  • Check Out UV

    Ultraviolet is an uncommon but noteworthy modality for process analytical technology. Crystal IS has specifically been promoting ultraviolet-C light-emitting diodes (UVC-LEDs) for development and bioprocess monitoring. Examples include the detection of protein collection start and end points and the real-time, at-line monitoring of cleaning operations.

    “Companies frequently use light-induced fluorescence sensors for this process, but UV LEDs provide tighter control and detection with small footprint, low power consumption, and the detection sensitivity of LEDs,” notes Hari Venugpalan, director of global product management at Crystal IS.

    UVC LEDs excel at measuring one specific parameter, for example, absorbance wavelengths for proteins or residual materials, according to Venugpalan. “Working with a single parameter is cost-effective because unwanted wavelengths do not need to be filtered out,” he maintains.

Related content