Analysis of the State of the Art: Near-Infrared Spectroscopy

In honor of Spectroscopy's celebration of 30 years covering the latest developments in materials analysis, we asked a panel of experts to assess the current state of the art of near-infrared (NIR) spectroscopy and try to predict how the technology will develop in the future.

Although near-infrared (NIR) spectroscopy is not a particularly sensitive technique, it can be implemented with little or no sample preparation and thus is well suited to applications such as process monitoring, materials science, and medical uses. We asked a panel of experts about recent advances in NIR, challenges faced by NIR users, application areas, and future developments in the field.

This article is part of a group of articles covering the state of the art of six spectroscopic techniques. The other techniques covered are infrared (IR) spectroscopy, Raman spectroscopy, inductively coupled plasma–mass spectrometry (ICP-MS), laser-induced breakdown spectroscopy (LIBS), and X-ray fluorescence (XRF) spectroscopy.

NIR Advances in the Past Decade

Our panel noted several recent key developments in NIR spectroscopy, mainly associated with new imaging systems and reduced instrument size.

Pierre Dardenne, who is a department head at Walloon Agricultural Research Centre (CRA-W), which is part of the regional government of Wallonia in Belgium, named two major advances for the technique: hyperspectral imaging systems and handheld, portable instruments. Gary McGeorge, a senior principal scientist at Bristol-Myers Squibb, agreed with Dardenne about the importance of hyperspectral imaging. "I think that the commercialization of and advances with imaging spectrographs and hyperspectral imaging cameras have made a dramatic change in the NIR field over the last decade," he said. "They provide the ability to understand the microscopic distribution of components within formulated pharmaceutical products and generate a link to understand the functional performance of pharmaceutical products," he added.

McGeorge noted that the applications for these instruments extend beyond pharmaceutical analysis to areas such as agriculture and food manufacturing, through integration into machine vision systems. "The speed of these systems enables real-time images in a matter of a few seconds for full spectral analysis if only a few wavelengths are required for analysis, which was simply not possible previously," he said.

Benoit Igne, a principal scientist at GlaxoSmithKline and president-elect of the Council for Near-Infrared Spectroscopy, feels the advent of smaller NIR systems was a key development. "I think the most important advance in NIR over the past 5–10 years is the arrival on the market of low-cost, miniature, fit-for-purpose sensors that can replace bulky systems in R&D, in the field, and manufacturing environments," he said. He noted that miniature Fourier transform NIR, microelectromechanical systems (MEMS) NIR, and linear variable filter (LVF) NIR systems are now available that can be used for many applications formerly reserved for research-grade instruments and that they have enabled more real-time, on-line monitoring and process control. "Their low cost, which almost allows them to be discarded rather than serviced, has the potential to completely change the traditional instrumentation lifecycle," he added.

Limitations and Challenges

We next asked our panel about the key challenges that spectroscopists developing or using NIR spectroscopy currently face, in terms of limitations of the technique, limits to our understanding of how it really works, and difficulties in using the technique.

"What makes NIR so interesting, but also so difficult to work with, is its sensitivity to the sample matrix," Igne said. He noted that unless a method is carefully designed and validated-a process that typically requires a significant amount of time, energy, and funds-it will be affected by changes in the sample matrix related to particle size, density, humidity, and temperature. "More work is needed to understand the absorption and scattering behaviors of diffuse reflectance and transmittance so that better algorithms can be developed, thus allowing the removal of matrix effects and improved model robustness," he said.

McGeorge agreed that achieving model robustness is a significant challenge in NIR. "For quantitative method development, one must ensure that the models are accurate and robust even in the presence of variable physical interferences that are unwanted," he said. "Designing this robustness a priori is hugely expensive during a pharmaceutical development program, and as spectroscopists we need to expand our understanding of how to mitigate these effects with minimal resource burn." Without the ability to adjust to the various effects, he added, it can be difficult to convince leadership that the benefits of a method outweigh the cost.

Pharmaceutical analysis, McGeorge added, presents an additional robustness issue: the innate variability in the excipients used in pharmaceutical tablets and other oral dosage forms. "Excipients are often naturally derived from plant material, so seasonal, geographical, and other sources of change may result in differences in the physicochemical composition of the material," he said. "These changes often have a direct impact in the NIR spectral features and need to be considered during method development and commercial sourcing discussions."

The lack of uniformity in instrument design is another important issue, according to Igne. "Because instruments are not similar, transferring a method from one unit to another is not always straightforward," he said. "This process often requires a significant investment and the use of potentially complex algorithms." The need for these algorithms is a major limitation to the traditional analytical framework of method development and transfer, he noted. "Many NIR practitioners build calibration models with the instruments that will be deployed in the field to avoid having to deal with transfer issues," he said.

Dardenne sees the implications of these problems in agricultural applications of NIR spectroscopy. "Very often we observe an obvious lack of actual validation," he said. "There are too many publications with only a classical random cross-validation, and real independent validations are rarely presented." He also mentioned that the limit of detection is a challenge with agricultural products, and that modeling analytes at levels below 0.1% is almost impossible. "The models seem to be working due to internal correlations, but the targeted constituent is lost in the noise," he said.

Igne added that there is a lack of acceptance of NIR methods in analytical laboratories. "People still tend to rely on traditional wet chemistry testing in applications for which NIR has proven successful," he said. Adoption of NIR increases, he said, when the information is needed in-line or on-line and in real time.

A Shrinking Talent Pool

One major challenge in NIR spectroscopy, which impedes its broader adoption, is that too few scientists have sufficient expertise with the technique. As a result, companies that want to use NIR can't find qualified people to help them do it. "It is very difficult finding the right people these days," said McGeorge. "A talent pool of experienced staff looking for employment or even replying to job applications simply doesn't exist."

McGeorge believes there are several reasons for this shortage. "In the pharmaceutical sector, the PAT initiative resulted in an explosion of interest in academia to explore how spectroscopy could be leveraged to understand pharmaceutical formulations and processes," he said. However, he noted, many schools simply used the tools without building an understanding of the spectrometers or the mathematics behind the modeling. "Additionally, I think many people have been swayed by advertisements from the NIR vendors, who try to present the units as simple to remove the apprehension that many feel towards the complex mathematics required to make robust models." Consequently, these schools are turning out great pharmacists or engineers who lack a strong spectroscopic background, he said, and students are entering the workplace without an adequate understanding of the spectroscopy.

Dardenne agreed that more NIR spectroscopy education is needed in universities and high schools, and mentioned an example of such an effort being undertaken by the International Council of Near Infrared Spectroscopy and its chair, Dr. Ana Garrido-Varo. They are developing a virtual platform for learning and teaching NIR spectroscopy (1), and, according to Dardenne, it will soon be operational and will provide credits for students.

How to Address the Challenges

"All manufacturers are working on methods to improve the transfer of calibration between units," said Igne. "I expect these methodologies to get better as manufacturing variability diminishes." He predicts, however, that calibration transfer will remain a problem well into the next decade.

As for matrix effects, Igne noted that academic groups are working on understanding scattering and absorption phenomena and are developing algorithms to better extract the relevant information. "I would expect these systems to remain in the academic sphere for quite a while," he said.

McGeorge sees some progress in addressing this problem. "As new challenges are being identified, the chemometricians are doing a wonderful job developing algorithms to meet these challenges," he said. "External parameter orthogonalization is one such example where known systematic interferants can be eliminated to ensure that the model is robust to variation in that feature." This approach can correct for systematic shifts in spectrometer performance or raw material properties that are changing from known sources, he added.

McGeorge mentioned another possible solution for improving method robustness. "An interesting new commercial technology relies on spatially offset analysis, where a portion of the sample is illuminated and another is interrogated," he said. "Through the maturation of such technology it may become turn-key to deploy NIR solutions that are devoid of any impact from physical perturbations of the spectral information, yielding more robust methods that are easier to develop."

In Igne's view, the ability to convince more laboratories to adopt NIR depends on researchers' achieving two things: coming up with ways to help users understand validation statistics and simplifying the calibration process. "Making the technology more accessible to non-chemometrics experts will be a significant step toward successful implementation," he said.

Application Areas

We've already touched on a number of applications for NIR so far in this article, and our panel members brought up several more when they were asked about important current applications and emerging new areas of application for the technique.

"NIR spectroscopy started with Karl Norris's work in the 1960s, and his work was on agriculture," said Dardenne. "I believe agriculture remains the main user of the technology, especially for the determination of the feeding value for animals," he added. Igne agreed that agriculture-including crops, animals, forestry, and soil-is certainly the most important area of application for NIR.

All three of our panel members agreed that the pharmaceutical industry is another important current application. "Within the pharmaceutical manufacturing area continuous manufacturing solutions have garnered huge interest by both the manufacturing companies and the regulators," said McGeorge. "This paradigm shift in manufacturing requires continuous process verification to ensure that the system is in steady state and is delivering product of acceptable quality." McGeorge further noted that without noninvasive sensors such as NIR spectrometers embedded in each unit operation to ensure the uniformity of drug formulations, this shift would not have been made so quickly.

Igne had a slightly less positive view of the use of NIR in the pharmaceutical industry, however. "Regulatory overheads have created comparability burdens between NIR and established analytical methods that limit the adoption of NIR," he said. "As the regulatory burden eases, it should be easier to embed NIR in the normal control of a manufacturing process, similar to a temperature or a pressure sensor."

McGeorge noted that the biggest challenge is trying to design spectroscopic solutions in a global regulatory landscape where each country or region has its own expectations. "At the moment these requirements are not clear and each applicant learns the hard way, by trial and error, with the individual health authorities," he said. "There is simply no clear regulatory roadmap for such applications." However, he sees progress being made. "This does appear to be changing with the EMA issuing a final guideline for the use of NIR spectroscopy in the pharmaceutical industry and with the FDA now issuing draft guidance that is currently under review," he said. In addition, the ASTM E55 committee is actively working to provide standards for various areas of pharmaceutical manufacturing that includes the use of in-line, on-line, and at-line spectroscopy. "Through these integrated efforts it is likely that the implementation landscape will become clearer and easier than it was for the early adopters."

Our panelists named biomedical analysis as a significant emerging area of application. "Most of the theoretical work on NIR (separation of absorption and scattering) has been done in biomedical analysis to improve our ability to detect and monitor tumors, control glucose, and so forth," said Igne. "NIR allows the laboratory to come to the bedside, and this is a significant improvement for patients." Dardenne noted that with NIR in biomedical analysis, validation must be rigorous because the consequences of a mistake would be too dire.

Advances in sampling and sample presentation have allowed NIR to be used more efficiently in these historical fields, said Igne, and they are driving new applications. He thinks the technology has reached a mature state with respect to a fundamental understanding of the technique but that challenges remain. "NIR is, however, still experiencing difficulties interfacing with processes to ensure signal quality and the collection of the most relevant information about the samples," he said.

A Role for NIR in Biopharma?

The recent large swing in the pharma business toward proteins may have created a new area of application for NIR.

"There is certainly a lot of interest from biotechnology groups to use NIR as a process analytical technology to monitor cell growth in upstream processes and to characterize proteins during purification," said Igne. "Numerous academic groups and industrials have invested time and funds deploying NIR in bioreactors with good success."

"Fermentation is a tricky application," cautioned Dardenne. "We have to remember that NIR sees NH and not necessarily the configuration of the NH bands in big molecules." He added that to be sure that a given protein or amino acid can be determined by NIR, the NIR values must be compared with reference values, not expressed as "as is" values but expressed relative to the total nitrogen content.

Both Igne and McGeorge noted a fundamental problem with the use of NIR in biopharma: the presence of water. "Water is a challenge for NIR since it has a very high absorptivity and broad lines that reduce the ability of NIR to provide meaningful information during a fermentation process," said McGeorge.

Hyphenation with NIR

In analytical chemistry, the term "hyphenated techniques" is usually used to describe separation techniques being combined with multiple detection approaches-for example, gas chromatography–mass spectrometry or liquid chromatography–IR. We asked our panel if a similar phenomenon exists for NIR.

"We have already used data fusion in our team: NIR plus mid-infrared, for instance, or NIR spectra of the forage plus NIR spectra of the feces to obtain a better estimate of the digestibility," said Dardenne. "We dream of a portable instrument combining NIR and Raman or even NIR–Raman–XRF."

Igne noted the emergence of NIR imaging as an example of this approach. "While remote sensing traditionally has been done from space and planes, the development of the drone industry will certainly reduce the cost of deployment of remote sensing methods that will allow NIR chemical imaging data to be obtained at low cost in the agriculture and natural resources management fields," he said. He added that NIR chemical imaging also can be used to monitor product quality in real time in the food industry (for example, with poultry or fresh fruits) and in the pharmaceutical industry in continuous unit operations.

Future Developments

Panel responses about future developments in NIR spectroscopy and instrumentation were predictably varied, but cost was a common theme.

"I anticipate that we will see increased adoption of low-cost miniaturized systems that are wirelessly connected to networks and deployed in a trivial fashion much like pH probes or pressure sensors in current manufacturing processes," said McGeorge. "Recent commercialization of ultracompact LVF-based spectrometers has shown proof of concept, and by increased adoption their limitations will be assessed so that they may evolve to meet even more demanding challenges." He noted that as their costs drop, such devices might be found in mainstream consumer stores and possibly could be designed to be coupled with or integrated into cell phones.

Dardenne mentioned the effects of reduced cost with respect to hyperspectral imaging systems. "In terms of instrumentation, the price of the hyperspectral camera will decrease, making the technology affordable for more applications." He predicted that hyperspectral imaging systems will perform like classical instruments but will be able to detect contaminants at a much lower level. He also predicted a development that would bring together the current technology of unmanned aerial vehicles with hyperspectral imaging. "We will see drones carrying light hyperspectral imaging systems, and these devices will enhance precision agriculture to improve the use of fertilizers and pesticides," he said.

According to Igne, improvements that have resulted in rugged, stable systems with impressive resolution and high signal-to-noise ratios have come with a cost that often prohibits the deployment of NIR spectroscopy in manufacturing sites. He expects the instrument industry to follow two tracks. "Historical manufacturers will keep improving their systems while newcomers will work hard at designing fit-for-purpose systems that answer the needs of particular applications without the bells and whistles that are needed in an R&D laboratory," he said. "This will drive the cost down and improve the penetration of this technology in plants where cost and maintenance have been the main burdens of implementation."

Igne also predicts that more instruments will be developed that incorporate myriad complementary technologies, which would allow data fusion and a more complete characterization of samples.

Acknowledgments

I would like to thank the panel members for their input and Spectroscopy Editorial Advisory Board member Gary McGeorge for his very helpful efforts in inviting panel members, formulating survey questions, and collecting the various responses.

Reference

(1) http://www3.uco.es/m1415/

Steve Brown is the technical editor of Spectroscopy magazine.