Using Fluorescence-Based Sensing to Accelerate Process Development

ABSTRACT

Bioprocesses have long been ubiquitous in the production of modern pharmaceuticals and drugs. Contemporary bioprocesses are being increasingly used in the production of many other products, ranging from biodegradable plastics, packing materials, and other throwaways, to non-fossil fuels such as ethanol and biodiesel, and commonly needed human spare parts such as artificial skin and cartilage. Fluorescence-based sensing technologies, which can greatly decrease overall development time, labor, and costs, become an increasingly useful tool, particularly because their use permits a degree of miniaturization, scalability, and multiplexing that was previously unavailable.

Partly because of the growing number of applications, in the last few decades "faster, better, cheaper" has driven advances in bioprocess techniques in much the same way that it has driven advances in electronics, medical care, and other competitive industries. Even minor technological improvements in the early screening stages of drug discovery can improve speed to market and increase profits. Improvements in the pilot and production stages of development lead to reduced production costs that, for a long-lived drug (the ideal goal of any such development endeavor), can lead to increased profits over the longterm.

(New Brunswick Scientific)

The initial stages of a bioprocess development cycle, whether for a drug or other product, typically require an enormous number of experiments (thousands or even tens of thousands) to select an optimal cell line and media formulation. Even then, the cell line and formulation selected is merely the best of those tested. The more combinations tested, the better the chance that the combination selected will be close to optimal.

These experiments must be done quickly to minimize development time. This requirement mandates that the experiments be done, to the extent possible, in parallel. Until recently, this initial screening was done almost exclusively in well plates or flasks, vessels that were reasonably inexpensive and thus economically feasible for a high degree of parallelism. However, these systems run mostly blind, with little or no instrumentation to track and control relevant process parameters.

Typically, the only data available to the research scientist are the results of offline measurements, which are labor intensive to take and produce results of questionable accuracy because process parameters are subject to measurable change during the sampling process. Consequently, little data are obtained regarding optimal process parameters; process optimization is generally left to be done heuristically in bioreactors. The latter systems are adequately instrumented and capable of measuring and controlling salient process parameters. However, because of the discrete and expensive nature of experiments run in such systems, a fairly limited number of experiments may be performed in an attempt to determine what constitutes optimal culture conditions. Even small errors in what are determined to be (in contrast to what actually are) optimal conditions can result in enormous increases in production costs over time. In addition, if the range of allowable variances in such conditions is not adequately explored and documented, the possibility exists that valuable and viable batches of product may be discarded because of small excursions from so-called ideal culturing conditions, even though the excursions may actually be of little consequence.

Process parameters are needed even during the screening stages of development. This is illustrated in Figure 1, which shows an experiment performed with E. coli in a 50-mL flask. As shown, the dissolved oxygen (DO) level reaches its minimum (and hence the cell line its viable maximum) just three hours into the run. Commonly, such an experiment would be started in a flask, which would then be left to continue overnight without interruption until morning. Offline sampling at that point in time would be misleading because it would show only the end of run conditions and provide no information as to what actually transpired during the run.

Figure 1

Of course, small laboratory-scale bioreactors could be used for the initial screening of cell lines and media as well as for process optimization. However, purchasing large numbers of individual bioreactors for the screening stage would be prohibitively expensive and setting up, running, breaking down, and cleaning a large number of such bioreactors would require an enormous amount of labor. If bioreactors were to be used, the degree of parallelism (and hence the number of experiments performed in a given period of time) would be quite limited or else the time required for the initial stages of the development cycle would increase dramatically.

MODERN ADVANCEMENTS

In recent years, several systems have been introduced that attempt to address the conflicting needs of maximizing information while minimizing expense. Some of these systems, such as automated plate readers and similar devices, provide an adequate degree of parallelism while providing at least some information about what transpires during culturing. However, process control in such systems is generally absent or inadequate. Newer, novel systems using microreactors or chips provide increased capabilities but are quite expensive compared to traditional systems and raise the question of how best to scale up to large-capacity systems for production purposes.

Another approach to high throughput design is typified by systems, such as Infors Profors sparged column reactors, DasGIP Fedbatch-Pro stirred tank reactors, Infors Sixfors stirred tank reactors, and Sartorius Q stirred-tank reactors. These systems incorporate a number of stirred-tank bioreactors ranging from a few to a dozen or more, ganged together. The vessels are offered in a variety of small sizes (less than a few liters) that have volumes sufficiently large to allow repeated sampling without disruption of the process. These systems address the need for measurement and control of process parameters in a parallel processing environment, and simplify scale-up issues by means of the familiar nature of the vessels. Although they provide capabilities similar to those of an equivalent number of individual bioreactors, virtually all of the high throughput systems shown economize on bench space, provide a simple, common interface for laboratory utilities, and use a single integrated control system. Moreover, each of the systems uses common and universally accepted measurement technology so that the data generated by these systems can be readily used in the scale-up process.

The measurement technology used in these systems is both an advantage and a drawback. The advantage is that the technology is decades old and already familiar to the operator. The drawback is that the systems, while facilitating parallel processing, are quite labor intensive: setup, calibration, and teardown and cleaning times are virtually the same as they would be for multiple experiments in an equivalent number of individual reactors. The reason for this is that the systems use conventional probe-type sensors: each vessel has its own set which must be sterilized before use, calibrated in place, and cleaned again after use. In addition, these probes must be inserted into the vessel to contact the liquid to work. This limits the system's smallest practical vessel size, even if miniature probes are used. Because of this, the volumes of required media (and thus supply costs) are nearly the same as they would be if banks of individual bioreactors were used in place of these high throughput systems. Because there is no appreciable savings in supply costs or labor costs, the attractiveness of these high throughput systems is limited.

FLUORESCENCE-BASED SENSING

There is another approach to high throughput design using a technology that's been known for many years, but has not, until recently, seen widespread adoption. Fluorescence-based sensing technology combines novel chemistry, optics, and electronics in the design of a sensor that's virtually calibration-free, can be miniaturized to accommodate the smallest vessels or well plates, isolates the sensing electronics from the cell culture broth, and can be multiplexed easily, if needed, to measure values of process parameters in a large number of vessels without concern for contamination.

Figure 2 shows such a sensor schematically. The sensor consists of two major components: a sensing head that contains both optics and electronics; and a peel-and-stick sensing patch or foil that is affixed to the inside of the vessel being monitored.

Figure 2

One or more high intensity light emitting diodes (LEDs), after proper bandwidth filtering, are used commercially to illuminate the sensing patch, which is prepared with a dye that causes reflected light (measured at a different wavelength) to vary in a well-defined and consistent way with changes in the measured parameter.

As an example, a system for measuring DO can be designed based on the principal that oxygen quenches, in a dynamic and well-defined manner, the fluorescence of any fluorophore with a lifetime longer than 10 ns. Such a sensor will use a modulated, properly filtered light source (a high intensity blue LED, for example) to excite a patch treated with a specific compound, commonly a ruthenium-based or platinum-based fluorophore, although other compounds have been used.

The patch, when excited, will reflect light, which differs from the incident light in wavelength, phase, and intensity. The amount of phase shift and difference in intensity are dependent on the DO level in the immersion liquid. The reflected light, properly filtered, can be captured using a common photodetector. Either change in intensity or shift in phase can be used as the primary measured variable, although intensity is less desirable because of possible errors caused by background noise, ambient light conditions, and changes in incident intensity caused by aging of the excitation LED and associated electronics. Although more difficult to measure electronically, phase shift is thus the preferred parameter in this case.

Measurement instruments for pH, CO₂, and other parameters can be designed in a similar manner. Despite the differences in optics, chemistry, and methodologies, these parameters are measured with instruments using physically similar designs and apparatus, and are based on the same principal of exciting a patch treated with a specific fluorophore with one or more light sources, and measuring and analyzing the reflected light. A pH sensor, for example, might use a patch treated with a dye such as hydroxypyrene trisulfonic acid (HTPS), which exhibits two excitation wavelengths that correspond to the acid and its conjugate base. Using a pair of high intensity LEDs to excite the patch, one ultraviolet and one blue, the ratio of emission intensities can be used to determine pH. CO2 can be measured in a similar manner. Research is currently being directed toward the robust design of sensors for more exotic parameters such as alcohols, green fluorescent protein, and glucose.

Some care must be used in the design and fabrication of the sensing patch. Typically, such patches comprise a number of thin layers: a protective layer made of mylar or similar material to protect the adhesive until the patch is ready to be pressed in place; the adhesive layer; a support layer to provide some degree of rigidity to the patch; the sensing layer; and a cover layer that provides optical isolation for the sensing layer. The sensing layer contains the dye, which is usually immobilized in some fashion to minimize leaching of the dye into the medium. For example, for DO, the dye can be immobilized in an oxygen permeable polymer film or absorbed on silica gel. The cover layer is required to prevent ambient light from unduly affecting the sensor's readings.

With minor modifications, fluorescence-based sensors also can be used to measure process parameters in a gas stream as well as a liquid. By using three such sensors with a bioreactor system (to measure DO in the inlet and outlet gas streams and in the culture medium), it is easy to determine total oxygen uptake rate over a period of time. If the uptake rate of the cell line is known, this provides an easy way to determine viable cell density during a culture, and thus the optimal harvesting point.

Because there is no probe entering the culture vessel with a fluorescence-based sensor, these sensors can be used in vessels of only a few milliliters in size. This is because the illuminated area of the patch (the only component that must be placed in the vessel) need be no more than a few millimeters in diameter, or even smaller, if fiber optics are used as a light source and receiver in place of the excitation LED and photodetector. Of course, the use of fiber optics has its own unique set of problems, but it has enabled the technology to be used in applications that otherwise have very limited space for the accommodation of sensors for parametric measurement.

Fluorescence-based technology has a number of advantages over conventional probe-type sensors. The ability to inexpensively measure process parameters in small vessels allows key operational parameters to be accurately measured much earlier in the development cycle than would be possible otherwise, such as at the flask or well-plate stage rather than the bioreactor stage. This permits early determination of optimal media formulation and cell culturing conditions, and thus facilitates scale-up from glassware to laboratory-scale and production-scale bioreactors.

An important additional advantage of the technology (particularly in fully instrumented systems) is that sensing electronics are extremely stable, requiring only a yearly measurement of electronic offsets (a procedure that takes less than a minute), thus rendering obsolete the time-consuming calibration procedures required for the proper use of conventional probes. By virtually eliminating calibration, the amount of setup labor required to use these sensors is substantially less than with conventional probes. Moreover, because the electronics are never in contact with the process (only the sensing patches are in contact and, although they can be autoclaved, they are cheap enough to be discarded after each use), there is no need for cleaning and sterilization of sensors between experiments. Again, the savings in time (and thus the savings in labor costs) can be quite substantial.

A final advantage of the fluorescence-based sensors is that they can be readily multiplexed so that one sensor is used to measure process parameters in many vessels. For applications in which measurements need be taken only at infrequent intervals, a sensor can be moved from vessel to vessel to measure a given parameter. In such a case, it is only necessary to provide a sensing patch in each vessel. Because the patches cost only a few dollars each, the use of a single sensing head with multiple patches greatly decreases instrumentation costs.

Conversely, another approach to multiplexing uses robotics, X–Y tables, or rotary tables to carry vessels to fixed sensors. For example, the Fluorometrix CellStation HTBR-1 is a 12-reactor system that uses a rotary table to automatically position, at a programmed interval, each of the 12-reactors over a single fluorescence-based module that measures both pH and DO. The same scheme can be used with well plates and other similar devices. The number of vessels or wells that can be monitored with each fixed sensor depends only on the measurement time (which is generally a few seconds) and the maximum time allowed between readings.

PROCESS ANALYTICAL TECHNOLOGY

As demonstrated above, the use of fluorescence-based sensors during the screening phases of the bioprocess cycle can greatly reduce the elapsed time required to complete these phases and do so with great savings of labor, media, and even facilities costs. However, the greatest value of the technology is that it provides information about the process that might otherwise have to be determined during later phases of the development cycle. For this reason, its use in the screening application is in keeping with the FDA's Process Analytical Technologies initiative (PAT). The FDA encourages a rather broad application of PAT by defining it as:

"Tools and systems that utilize real-time measurements, or rapid measurements during processing, of evolving quality and performance attributes of in-process materials to provide information to ensure optimal processing to produce final product that consistently conforms to established quality and performance standards."

The ultimate goal of PAT is the realization of measurable process conditions that are linked directly to the quality of the product being developed. The ability to measure viable process conditions during the screening phases of the development cycle provides a wealth of information about these cell growth conditions, and therefore, allows an assessment of how closely process-screening devices can be used to replicate large-scale conditions. This data will provide a smoother transition to large-scale processes, requiring fewer experiments to be done in expensive and relatively labor-intensive bioreactors during the laboratory-scale phases. Also, it should be noted that the use of this sensing technology does not require the purchase of novel and costly systems for the screening phases. The sensors are compatible with the well plates, dishes, and flasks that have traditionally been used in small-scale screening operations, and can be readily used to retrofit existing equipment.

USE FOR LARGE-SCALE PROCESSES

Despite the focus on the screening phases of the development cycle, this sensing technology also has value during the later phases of the development and production cycle. Recent trends tend toward the use of disposables during these phases because of lower operation and validation costs.¹ Disposables primarily take the form of bag type bioreactors—plastic pillows that can be charged with a few to many hundreds of liters of cell culturing media. Because they are generally shipped sterile and disposed of after use, they require less labor during setup and teardown of the culturing system than conventional stirred-tank reactors. However, these bioreactors require the same careful monitoring as the conventional bioreactors. If conventional probe-type sensors are adapted for use in these disposable bioreactors, the system is then only "semi-disposable" because the sensors must be sterilized before and after use, and calibrated during use. To address this issue, some recent efforts have been made to develop miniature disposable probe-type sensors so that the entire system becomes truly disposable. These sensors, however, are still relatively costly for a disposable system, and fail to adequately address calibration issues and the consequent labor and time requirements.

A different approach to fully disposable systems that has been adopted by several manufacturers is the incorporation of fluorescence-based sensors in disposable-bag bioreactors.² The sensor technology is applicable to any size bioreactor because it uses no volume-dependent components. The sensing patches themselves may be placed in the bags during fabrication and then sterilized by irradiation or by autoclaving, and then sealed into a presterilized bag. Alternately, the sensing patches can be autoclaved on a carrier (a clear glass or autoclaveable plastic slide or disk) that can then be inserted into a pocket in the bag before final sealing. This latter technique has the added advantage of facilitating alignment of the sensing patch in the bag with respect to the sensing head or fiber. The patches, which cost only a few dollars, are far less costly than even a disposable miniature probe and can be disposed of along with the bag after use. The sensing head itself remains ready to use again. In addition to the cost savings resulting from the disposal of only a very inexpensive patch, elimination of the effective need for calibration saves substantial time and labor costs associated with the culturing process.

ADOPTION OF THE TECHNOLOGY

Given the many advantages to fluorescence-based sensing, it is only natural to ask why this technology, which is nearly three-decades old has been so slow to catch on and why, even now, its use is primarily associated with small-scale apparatus.³

There are several factors that have contributed to its slow adoption. First, the biotech community is notoriously slow to embrace new technologies, particularly during the pilot and manufacturing phases of a drug-development cycle. This is because the cost of drug development and approval often greatly outweighs the cost of drug production, and the elapsed time required for drug approval is often far greater than the time required for screening and laboratory-scale operations. The use of a decades-old technology provides the manufacturer with a certain level of assurance during the approval process, and might therefore be preferred over a newer technology even if the newer technology is less costly and more efficient.

Second, the fluorescence-based instruments are just now reaching a level of robustness and affordability that makes them suitable for use in bioprocessing. Although conventional probes are subject to inaccuracy because of drift over time, this artifact can be virtually eliminated through frequent and careful calibration of the probe, a time-consuming operation but one that is effective nonetheless. In contrast, early fluorescence-based probes had several sources of error: inherent electronic variance from sensor to sensor, variability of patch formulation from batch-to-batch, and the need to ensure that the vessel being used conformed to certain specifications. In recent years, these problems have been substantially addressed: by including a microprocessor on board each instrument, the electronics can be normalized from sensor to sensor, largely mitigating electronic variance. Patch-to-patch variation is addressed by associating a calibration code with each batch of patches, similar to the codes used with most diabetic test strips, and the measurement results have been shown to be largely independent of bioreactor-to-bioreactor variance. The referenced testing was done using the Fluorometrix CellStation HTBR-1 12 station system.⁴

Robustness was initially not the only drawback to the technology—equally important was instrument cost. The original prototype instruments used lasers as an excitation source and were consequently financially unsuitable for practical applications. The use of high-intensity LEDs for excitation, which have in the last years dropped in price to a couple of dollars, together with efficient optical and electronic designs, addresses that problem: fluorescence-based sensors are potentially cheaper than conventional probes. Much has been done on the design of a new generation of sensing patches that can be demonstrated to not leach into the culture media or otherwise affect the growing cells.

Even with all of these advances, the technology still has a potential disadvantage in that the sensing patches are subject to photobleaching over time. Because of improved formulations which greatly increase the lifetime of the patches, and the use of phase shift or ratiometric measurements, the sensor's measurements are immune to changes in the patch response caused by photobleaching, except after long-term use. In such a case, the signal to noise ratio drops with time until the measured signals become noisy and less accurate after an excessive number of measurements. Because the effective lifetime of the sensing patch is measured in tens of thousands of readings, this is a problem only with very long processes monitored at frequent intervals (seconds versus minutes or hours). Even in this case, an easy solution exists: a vessel can be equipped with two or more patches in place of a single patch and the sensing head can be moved to a fresh patch whenever the measurements start to become noisy. In this way, processes of arbitrary duration can be monitored with no loss of accuracy and no drift over time.

CONCLUSIONS

As the need for more cost-effective and rapid development of drugs and other bioprocess-related products increases, so does the tolerance for new technologies and approaches in what has been a rather conservative industry. Fluorescence-based sensing technologies, which can greatly decrease overall development time, labor, and costs, become an increasingly useful tool, particularly because their use permits a degree of miniaturization, scalability, and multiplexing previously unavailable. Although there are limitations, particularly in patch development, continued development and maturation in this area holds great promise for the development of a widespread paradigm-shifting technology in the near future.

Joe Qualitz is the president and CEO of Fluorometrix Corporation, Stow, MA, 978.461.2468, support@fluorometrix.com

REFERENCES

1. Rao G, Moreira A, Brorson K. Disposable bioprocessing: the future has arrived. Biotechnol Bioeng. 2009 Feb:102(2)348–56.

2. Rao G, et al. Noninvasive sensors as enablers of "smart" disposables. BioProc Int. 2009 Jan;26–9.

3. Kroneis HW, Marsoner HJ. A fluorescence-based sterilizable oxygen probe for use in bioreactor. Sens. Actuators. 1983:(4)587–92.

4. Xudong Ge X, Hanson M, Kostov Y, Moreira AR, Rao G. 2006. Validation of an optical sensor-based high-throughput bioreactor system for mammalian cell culture. J Biotechnol. 2006:122(3):293–306.