APPLICATION Note

No. RB 3528434
Determination of moisture in a
protein sample using a portable
NIR instrument
Suzanne K. Schreyer, Ph.D. and Michelle A. Pressler, Ph.D., Thermo Scientific, Tewksbury, MA

Abstract
Near infrared (NIR) spectroscopy has been used effectively
and reliably to measure moisture in a wide variety of
samples. Due to the large overtone band for water in
NIR, levels of accuracy tend to be high and limits of
detection are lower than for most materials measured by
NIR. In this feasibility study the performance of a portable
NIR instrument was evaluated to monitor moisture
levels within a protein sample, as proof of concept to
monitor lyophilized samples in a pharmaceutical process.
Specifically, a multivariate Partial Least Squares (PLS)
model was developed and evaluated for the ability to
predict the levels of moisture in samples of casein protein.
The results generated with the portable NIR instrument
are also evaluated and compared with a laboratory based
NIR instrument in order to demonstrate the comparable
performance and improved utility of the portable instrument.

Introduction Thermo Scientific’s microPHAZIR RX can provide raw material

Lyophilization is a well-established procedure for identification in seconds
the preservation and storage of proteins and other
number of issues inherent in this type of testing. This
biopharmaceutical materials in both the food and the
method assumes that the selected samples are actually
pharmaceutical industries. Monitoring protein moisture
representative of the larger population and renders
content is an important process as moisture affects both
sampling data vulnerable to potential inconsistency. The
the physical and the chemical state of the materials. In
method is destructive so samples cannot be used again
particular, lyophilization of proteins in pharmaceutical
causing unnecessary loss of product. The process is
processes requires monitoring of these samples as any
expensive and time-consuming due to both the materials
excess moisture, or moisture lower than optimum, will
involved and the specialized training required to perform
impact the usability and the shelf life of the samples.
the relevant protocols.
Current methods of monitoring the samples rely on
destructive techniques whereby a subset of samples are
pulled, opened and analyzed for moisture. There are a
In contrast the use of NIR technology is fast (seconds), Results and Discussion
easy to use, and allows for accurate determination of the
moisture levels in a closed sample vial. Scanning of the
Data preparation Sample LOD, %

sample can be performed through the glass vial without

and generation 1 8.93
Spectral data from the 2 8.98
compromising the material; the vial can then be returned
microPHAZIR RX was 3 9.85
to the sample set as viable product improving operational
combined with the 4 10.02
efficiency. It is possible to perform 100% inspection of
reference moisture values 5 10.66
samples to comply with regulatory requirements and ensure
from the LOD balance for 6 11.18
that anomalous samples will be identified. The deployment
all samples. The moisture 7 12.16
of a portable NIR analyzer allows the instrument to be
values ranged from 8 12.11
taken directly to the sample source—eliminating the time
approximately 9 to 9 12.43
and protocols required to move samples to the lab.
13 wt/wt% as shown 10 12.60
Experimental in Table 1. 11 8.93
Replicate casein samples were made using casein from The collected spectra of
12 8.98
bovine milk (Sigma Aldrich). Twenty standard samples the moisture in casein
13 9.85
were placed in glass vials in a saturated NaCl solution samples are shown in
14 10.02
within a humidity chamber set to produce 75% humidity. 15 10.66
Figure 1. Color coding is
Samples were sealed in the humidity chamber for varying 16 11.18
by sample number with
periods of time to produce a range of moisture content. 17 12.73
each sample containing
At various time intervals, duplicate samples were taken, 18 13.07
different moisture levels.
mixed thoroughly, and capped to equilibrate moisture 19 12.43
within the sample. dry 8.92
Table 1.
Once mixed and equilibrated, samples were scanned
on the portable Thermo Scientific™ microPHAZIR™ RX
NIR instrument and the lab-based Thermo Scientific
Antaris™ NIR instrument. The microPHAZIR RX has
effective wavelength range from 1595-2395 nm, at a
spectral resolution of (the default) 8 nm. Five sample
scans were co-averaged and triplicate scans were taken
of the sample. Spectral data was collected in diffuse
reflectance mode and evaluated in the proprietary
software Method Generator (MG).

Spectra collected on the Antaris FT-NIR had 32 scans

co-averaged and triplicate scans were generated
with each sample. The wavelength scanned was from
4000cm -10,000cm -1 (1000nm -2500nm) at a resolution Figure 1. The collected spectra of the casein samples at various
moisture levels
of 32 cm -1. TQ Analyst was used to evaluate and
process the spectra.
Spectra preprocessing was done to optimize the
Once the spectra were collected, the samples were robustness of subsequent models. In this case
measured using a Mettler Toledo LOD balance to preprocessing consisted of Standard Normal Variates
determine the moisture content. These reference values (SNV) to compensate for packing or density differences,
were then combined with spectral data and PLS models and Savitsky Golay (1st derivative-5 point smooth-2nd
were created to predict moisture in the casein protein. order polynomial). Preprocessing was kept to a minimum
Results from the Antaris were used for comparison due to the strong water absorbance peaks commonly seen
to those generated by the microPHAZIR RX; model in NIR. The wavelength was selected for the strong water
development will focus solely on the microPHAZIR RX overtone region (1800-2100 nm) as shown in Figure 2.
in the following sections.
 The resulting PLS model was built to utilize these 3
factors. The subsequent correlation plot of reference
moisture values and the predicted moisture values is
shown in Figure 4. The plot shows the calibration (blue)
and the validation data results (red). As clearly seen, both
calibration and validation results are linear, indicating
good agreement between predicted and actual results.
The results from the Antaris data set are also shown for
comparison. Again, a high degree of agreement is seen
with the predicted and reference values. Both plots also
show a similar response to expected model predictive
behavior as they correspond to a very similar linear
pattern.
Figure 2. Wavelength range was selected to observe the strong
water overtone

Model Development
Initially the data set was split into a calibration and a
validation data set. A PLS model was used to generate
the quantitative model for moisture prediction using the
calibration data only. Based on the conditions described
above, the PLS model was developed in MG software.
The model was then used to predict results on the
validation data. Cross validation results on the calibration
data were recorded.

Initially the data was evaluated to determine the required

number of factors to use for the PLS model. A plot of the
number of factors and their contribution to the model
(Screen plot, Figure 3) indicates that 3 loadings would
provide useful information to the model and limit the
noise contribution.

Figure 4. Correlation plots between calibration and validation

values for microPHAZIR RX and Antaris, respectively

Figure 3. A plot of the number of factors and their contribution to

the model
 Analysis of the residuals from the microPHAZIR RX
3 Factor Antaris microPhazir RX
PLS model also indicate the model only has random
variations in the residuals, as seen in Figure 5 for both RMSEC 0.14 0.11

the calibration and the validation results. Residuals RMSEV 0.12 0.15
are a way to evaluate if the model has any systematic RMSECV 0.20 0.16
variations that may interfere with the PLS model. Only
R2 0.99 1.00
random variations are observed providing evidence of
model robustness. In every instance it is further noted StDev, Residual 0.11 0.11
that calibration and validation data display similar trends;
Table 2. RMSE comparison of Antaris and microPHAZIR RX
while validation results have greater residuals than methodologies
calibration data, this is to be expected due to slightly
higher root mean square errors (RMSE).
Conclusion
Predictive models for moisture levels in casein protein
were built on a benchtop laboratory and handheld
NIR spectrometer. Both systems allowed for the
development of an effective, robust PLS model that
provided accurate results with consistency. In terms of
practical implementation, the slight decrease in accuracy
of the handheld system (RMSEV = 0.12 [Antaris] vs.
0.15 [microPHAZIR]) is overcome by the advantages of
performing the analysis at the point of sampling, allowing
a reduction in testing time and expense and lost product
and enabling 100% material inspection.
Figure 5. Residuals analysis of microPHAZIR RX results,
demonstrating random variation about zero Based on this feasibility study, it is expected that
moisture analysis comparable to those currently
The similarity between the Antaris and the microPHAZIR obtained via benchtop systems should be readily
RX are further corroborated in the RMSE results. RMSE, producible with a handheld platform allowing for rapid
which incorporates standard deviation of prediction and robust monitoring of a substantial range of protein
errors, is a consistent indicator of model performance samples. Deployment of the microPHAZIR RX can
and robustness, and as Table 2 shows, both instruments provide a valuable tool for the monitoring of moisture in
are expected to have excellent accuracy predicting raw materials, intermediate production stages, or final
moisture in the casein protein samples. Accuracy with products at a degree of convenience that is unmatched
respect to the RMSE for calibration and cross validation by conventional laboratory based systems.
are similar for the Antaris and the microPHAZIR RX. Note
the low and consistent standard deviations indicating
both precision in the obtained results and consistency
between devices observed.

No. America Central and So. America Europe, Middle East, Africa Asia Pacific
Boston, USA Sao Paulo, Brazil Munich, Germany Mumbai, India
Shanghai, China
Tokyo, Japan

Find out more at thermofisher.com

For Research Use Only. Not for use in diagnostic procedures. © 2019 Thermo Fisher Scientific Inc.
All other trademarks are the property of Thermo Fisher Scientific and its subsidiaries. RB 3528434 0719 v1