Chemometrics Applications in Biotech Processes: A Review

Anurag S. Rathore, Nitish Bhushan, and Sandip Hadpe

Dept. of Chemical Engineering, Indian Institute of Technology Delhi, Hauz Khas, New Delhi 110016, India

DOI 10.1002/btpr.561
Published online February 28, 2011 in Wiley Online Library (wileyonlinelibrary.com).

Biotech unit operations are often characterized by a large number of inputs (operational
parameters) and outputs (performance parameters) along with complex correlations amongst
them. A typical biotech process starts with the vial of the cell bank, ends with the final prod-
uct, and has anywhere from 15 to 30 such unit operations in series. The aforementioned pa-
rameters can impact process performance and product quality and also interact amongst
each other. Chemometrics presents one effective approach to gather process understanding
from such complex data sets. The increasing use of chemometrics is fuelled by the gradual
acceptance of quality by design and process analytical technology amongst the regulators
and the biotech industry, which require enhanced process and product understanding. In
this article, we review the topic of chemometrics applications in biotech processes with a
special focus on recent major developments. Case studies have been used to highlight some
of the significant applications. V
C 2011 American Institute of Chemical Engineers Biotechnol.

Prog., 27: 307–315, 2011

Keywords: chemometrics, MVDA, multivariate data analysis, bioprocessing

Introduction process analyzers. PCA has been used to transform 12 cross-

Chemometrics as a separate field has grown considerably
in the past 4 decades. One of the popular definitions is that
from Wold—‘‘How to get chemically relevant information
out of measured chemical data, how to represent and display
this information, and how to get such information into
data.1’’ Chemometrics is being increasingly used in both ba-
sic research and applied scientific fields and has enabled the
diagnostic evaluation of parameter interactions that were pre-
viously undefined. For complex datasets, univariate or bivari-
ate analysis is often inefficient and is likely to result in
misinterpretation of data.2–4 Use of projection methods can
effectively deal with challenges such as multidimensionality
of the data set, multicollinearity, missing data, and variabili-
ty from experimental error and noise.5
The chemical industry was early in recognizing and adopt-
ing chemometrics as a quick and economical method of
extracting real-time information from data and, thus, leading
to improved process monitoring and control. Visible spec-
troscopy, near-infrared (NIR) spectroscopy, mid-infrared
(MIR) spectroscopy, nuclear magnetic resonance (NMR)
spectroscopy, and Fourier transform infrared (FTIR) spec-
troscopy are some of the commonly used process analyzers
that have been used in the chemical industry. Principal com-
ponent analysis (PCA), partial least squares (PLS) regression,
principal component regression (PCR), canonical variable
analysis (CVA), and modified soft independent modeling of
the class analogy (SIMCA) are some of the statistical tools
that have been used to facilitate analysis and modeling of
the abundant data that are provided by the aforementioned
Correspondence concerning this article should be addressed to A. S.
Rathore at asrathore@biotechcmz.com.
correlated variables of an activated sludge wastewater treat-
ment into three uncorrelated principal components, which
were able to account for 78% of the system variability. This
enabled easier analysis, monitoring, and diagnosis of the sys-
tem.6 Combination of visible (Vis) and NIR spectroscopy
and chemometrics has been used for discrimination between
samples of Australian commercial white wines of different
varietal origins—Chardonnay and Riesling.7 Models devel-
oped using PCA, PCR, and discriminant PLS regression
gave an excellent discrimination between samples of the two
varietal origins under consideration with an accuracy up to
98%. These models could be used by the wine industry for
identification of white wine varieties or their blends. FTIR-
PCA has been used to study oxidation of lubricating base
oils.8 The principal components generated were able to
explain 99.99% of the variance, with the second PC showing
that iron favored formation of alcohols and esters and thus
influenced the oxidation process. Applications of chemomet-
rics in pattern recognition and multivariate calibration have
been shown to result in greater profits for industry because
of better process control, faster verification of raw material
identification and quality, and faster analysis of wastewater.9
Chemometrics has also been applied for authentication of
meat products, on-line estimation of the carcinogenic poten-
tial of lubricant base oil, and rapid analysis of a gaseous
effluent from a heterogeneously catalyzed reaction.10–12 As
demonstrated by the aforementioned applications, the combi-
nation of chemometrics and a suitable process analyzer has
been proven beneficial for the chemical industry.
The use of chemometrics in the pharmaceutical industry
has been relatively more recent with a lot of work being
done in the last 2 decades. NIR spectroscopy has emerged as
the analyzer of choice for a wide range of applications. NIR,

C 2011 American Institute of Chemical Engineers

V 307
308
in combination with PCA and SIMCA, has been used for
routine testing of packed pharmaceutical substances directly
in warehouses using a fiber optic probe.13 The proposed
approach allowed for measurements through closed polyeth-
ylene bags, and a trichotomy classification procedure was
proposed, which allowed an operator to identify raw materi-
als of satisfactory quality. NIR, in combination with PCA
and SIMCA, has also been used for the identification of
counterfeit drugs.14 The tool was capable of identifying
subtle alterations in drug composition with 100% accuracy
and could be used as a tool for rapid, nondestructive testing.
Other NIR-based applications include analysis of high-shear
granulation, characterization and determination of azithromy-
cin polymorphs, and determination of the crystalline form
present in amorphous miokamycin.15–17 UV–vis spectros-
copy, attenuated total reflectance-Fourier transform infrared
(ATR-FTIR) spectroscopy, infrared imaging, and Raman
spectroscopy are some of the other commonly used process
analyzers in the pharmaceutical industry. NIR transmittance
and Raman spectroscopy with PLS have been used for the
prediction of active substance content in tablets.18 Successful
calibrations were developed with prediction errors less than
3.7% and comparable to the error of the chromatographic
reference method. PLS of absorbance spectra has been used
for simultaneous determination of benzyl alcohol and diclo-
fenac in pharmaceutical formulations.19 The proposed
method was found to be simple, precise, inexpensive, requir-
ing no complex pretreatment, and thus an effective means
for quality control of pharmaceutical products.
The last decade has seen a flurry of activities in the area
of development of chemometrics applications for the bio-
pharmaceutical processes.20,21 They include analysis of NIR
spectral information for an antibiotic production process,
multivariate statistical process monitoring for processing of
pharmaceutical granules, the assessment of seed inoculum
quality from a manufacturing process, and development of
an integrated on-line multivariate statistical process monitor-
ing, product attributes prediction, and fault diagnosis frame-
work for a fed-batch penicillin fermentation.22–25 A flexible
process monitoring method has been applied for analysis of
pilot plant cell culture data for fault detection and diagno-
sis.26 A PCA model was constructed from 19 batches, and
the model was shown to successfully detect abnormal pro-
cess conditions and diagnose root causes. Feasibility of using
chemometrics for supporting key activities required for suc-
cessful manufacturing of biopharmaceutical products, includ-
ing scale-up, process comparability, process characterization,
and fault diagnosis, has been examined.27 Representative
data from small-scale (2 L) and large-scale (2,000 L) cell
culture batches were analyzed in this study. Scores plots,
loadings plots, and variable importance for the projection
(VIP) plots were utilized for assessing scale-up and compara-
bility of the cell culture process. Batch control charts were
also shown to be useful for fault diagnosis during routine
manufacturing. A subsequent publication examined the use-
fulness of chemometrics for root cause analysis to identify
scale-up differences and parameter interactions that
adversely impacted cell culture process performance.28 We
will discuss this application in more detail in the next sec-
tion. In an even more recent publication, NIR-PCA was
effectively used for screening of lots of basal medium pow-
ders based on their impact on process performance and prod-
uct attributes for a cell culture process.29 These lots had
identical composition as per the supplier and were manufac-
Biotechnol. Prog., 2011, Vol. 27, No. 2
tured at different scales using an identical process. A com-
bined NIR-PCA approach made it possible to fingerprint the
raw materials and to distinguish between the good and poor
performing media lots.
In the following sections, we review chemometrics appli-
cations in upstream and downstream processing of biotech
products. It will be evident that because of the complexity
present in most biotech unit operations, implementation of
quality by design (QbD) and process analytical technology
(PAT) initiatives will require an increased use of chemomet-
ric tools and approaches.30–35 We wish to provide to the
readers a summary of major research that has been done on
this topic recently.
Chemometrics Applications in Upstream Processing
Chemometric tools have been used in the cell culture
operations in the last decade as summarized in Table 1. PCA
has been used for detection and diagnosis of abnormal pro-
cess conditions in an industrial fed-batch cell culture process.
The model was successfully able to detect abnormal process
conditions, which resulted from three known fault types,
namely irregular thermal heating, elevated dissolved oxygen
values, and large variation in agitation.26 PLS calibration
models of NIR spectra have been utilized for the measure-
ment of glucose, lactate, glutamine, and ammonia in
undiluted serum-based cell culture media.37 Robust, analyte-
specific models were generated, and the low values of stand-
ard errors of prediction for each analyte demonstrate that the
models can be used to (off-line) determine the important nu-
trient and byproduct content in a serum-based cell culture
medium. A novel PLS approach called evolving PLS has
been compared with the traditional PLS using data from an
industrial fed-batch mammalian cell culture process for pre-
diction of intermediate and final quality variable values.40
Use of in situ 2D fluorometry in combination with chemo-
metrics has been evaluated for monitoring the concentration
of viable cells and the concentration of recombinant proteins
in mammalian cell culture.42 PCA was used to filter the large
volumes of redundant spectral data, while PLS correlated the
reduced data with the target state variables. Both viable cells
density and glycoprotein concentration were accurately esti-
mated, which strongly suggests that the combination of 2D
fluorometry with suitable chemometric techniques is a con-
sistent technique for monitoring of a cell culture medium.
Table 2 presents a summary of chemometrics applications
targeting the fermentation unit operation. FTIR combined
with multiple linear regression has been demonstrated to be
useful for measurement of fermentation components includ-
ing proteins, polysaccharides, lipids, and microbes that are
otherwise difficult to measure using sensors.51 The resulting
calibration model could be used for measuring component
concentrations. A combination of attenuated total reflec-
tance-mid-infrared spectroscopy (ATR-MIR) and PLS has
been used to monitor the concentrations of key analytes in a
Streptomyces clavuligerus bioprocess for the synthesis of
clavulanic acid.45 Quantitative, at-line models having low
prediction error were developed and validated, demonstrating
the utility of ATR-MIR along with chemometrics for real-
time monitoring of complex fermentation process. Utility of
including seed-quality information into data-based models
for estimation of final productivity in an industrial antibiotic
fermentation process has been examined.24 Multiway PCA
and PLS regressions were applied to assess the seed quality
Biotechnol. Prog., 2011, Vol. 27, No. 2 309

Table 1. Chemometrics Applications in Mammalian Cell Culture Unit Operation of Biotech Processes
Application
Development of rapid peptone screening
method to optimize the efficiency and
consistency of protein production
Fault detection and diagnosis for an
industrial fed-batch process
Chemometric approach for root
cause analysis
Measurement of multiple analytes in undiluted
serum-based cell culture media
Measurement of glucose and glutamine in
insect cell culture media
Routine screening of cell culture media used
in industrial biotechnology
Use of chemometrics in data analysis
Prediction of product titer in an industrial
fed-batch cell culture process
Noninvasive glucose monitoring
Monitoring concentration of viable cells and
recombinant proteins
Process Analyzer Chemometric Tool Observations
NMR spectroscopy PCA and PLS NMR-PLS was used to analyze
spectral data for prediction of cell
culture titer for a given peptone lot.36
PCA Model successfully detected abnormal
process conditions and diagnosed root
cause for process underperformance.26
PLS Model successful in identifying the root
cause and designing experimental
conditions to demonstrate and correct it.28
NIR spectroscopy PLS Robust, analyte-specific models were
generated with low prediction errors.37
NIR spectroscopy PLS Accurate measurements were
possible with low error values.38
Fluorescence Multiway robust PCA, Robust methods could identify compositional
spectroscopy NPLS-DA, and NPLS changes and predict product yield.39
PLS Chemometrics used for scale-up,
comparability, process characterization,
and fault diagnosis.27
PLS, ‘‘evolving’’ PLS Method was able to accurately predict
both intermediate and final quality
measurements
as well as detect process faults.40
NIR spectroscopy PLS A successful multivariate calibration
model was built for glucose
detection with very low error values.41
In situ 2D fluorometry PCA, PLS Accurate measurements over a wide range of
reactor operating conditions.42

Table 2. Chemometrics Applications in Microbial Fermentation Unit Operation of Biotech Processes

Application Process Analyzer Chemometric Tool Observations
Yeast fermentations Mid-infrared (MIR) PLS Prediction of concentrations of
spectroscopy glucose and ethanol.43
Process and quality control of FT-MIR and PCR and PLS High prediction errors with Raman
fermentative production of ethanol FT-Raman spectroscopy. MIR could be used
spectroscopy for rapid detection of analytes.44
Complex antibiotic fermentation Attenuated total PLS At-line models with low prediction error
reflectance-MIR values for monitoring of ammonium,
spectroscopy glucose, methyl oleate, and biomass.45
Influence of seed quality on productivity Multiway PCA Model could distinguish between high
of an industrial antibiotic fermentation and PLS and low productivity using seed fermentation
data from both pilot and production scales.24
Automated production support in an PCA, multiway Algorithm provided concise indicators of
industrial fermentation system PCA, ‘‘moving process faults and assisted operators by
window’’ PCA taking corrective measures.46
Monitoring and control of fed-batch PLS Accurate inference of difficult-to-measure
fermentation processes quality variables; detection and isolation of
fault conditions.47
On-line batch fermentation NIR PCA, PLS, MSPC Easy and efficient identification of abnormal
process monitoring charts fermentation runs at an early stage of the
fermentation.48
Monitoring of industrial MPCA, MPLS Detection of dissimilar batches and irregular
fermentation processes quality variables, prediction of final
product concentration.49
’’ NMR PCA First two principal components accounted for
over 99% of the sample variance. Loading
plot predicts the analyte variation.50
Solid-state fermentations FTIR Multiple linear Standard errors of prediction were less than 5%
regression for measurement of proteins, polysaccharides,
lipids, and microbes.51
Improvement of the fermentation process Gas chromatography Artificial neural Prediction of accurate optimal conditions
of docosahexaenoic acid production NETWORK (ANN) without any experimentation.52
Rapid analysis of the expression of Mass spectrometry, ANN, PLS regression Quantitation of expression of a heterologous
heterologous proteins in Escherichia coli FTIR spectroscopy protein directly in fermentation broths.53
310 Biotechnol. Prog., 2011, Vol. 27, No. 2

Figure 1. Illustration of using chemometrics for root cause analysis for a mammalian cell culture step.
A: Variable importance for the projection plot shows which variables have the most influence on the process. B: The score scatter plot shows a cor-
relation between the raw material type and product attributes. Adapted from Refs. 28 and 29.

and estimate the final productivity, respectively. Using these
chemometric techniques, it was possible to extract seed fer-
mentation features related to the final productivity at both
pilot and production scales.
Case study illustrating use of chemometrics for root cause
analysis
Figure 1 illustrates an application involving use of chemo-
metrics for root cause analysis to identify scale-up differen-
ces and parameter interactions that adversely impacted cell
culture process performance.29 Data from a total of 171
batches run at small scale (2 L), pilot scale (2,000 L), and
commercial scale (15,000 L) were used in this analysis.
Input parameters examined included continuous on-line
measurements of operating parameters (e.g., pH, dissolved
oxygen, and temperature), daily measurements of dissolved
CO2, metabolic indicators, and cell growth parameters. The
output parameters included product attributes such as product
titer, viable cell density, cell viability, and osmolality. Time
course performance variables (daily, initial, peak, and end
point) were also evaluated. A total of 119 output variables
from raw materials, product attributes, time course, and seed
inocula trains were evaluated using chemometrics. The VIP
plot in Figure 1A summarizes the observations made from
the score and loading plots by showing the relative impor-
tance of each included variable in the analysis. It is observed
that large-scale raw material type and culture metabolism
evolution (e.g., elevated lactate and osmolality) exert the
greatest influence on product attributes. These two parame-
ters were further examined for their impact and interactions.
Next, a score scatter plot (shown in Figure 1B) was gener-
ated based on data analysis. It provides a visual summary of
the process behavior over time, with the score vectors for
the first two principal components, t[1] and t[2], plotted
against each other. It is seen in Figure 1B that a correlation
exists between raw material type and product attributes. On
the basis of the chemometric analysis, the authors were able
to conclude that the root cause of process underperformance
was the combination of large-scale media and high pCO2
conditions in the large-scale bioreactor. This hypothesis was
later confirmed by experiments performed at pilot scale
using the different types of raw materials and CO2
concentrations.29
As is evident from Tables 1 and 2, significant work has
been done in the mammalian cell culture and microbial
Biotechnol. Prog., 2011, Vol. 27, No. 2
Table 3. Summary of Downstream Applications
Application
Use of on-line HPLC for making
real-time decisions for pooling
of process scale chromatography
columns based on product quality
attributes to get consistent product
quality
’’
’’
Competitive binding for partial
filling affinity capillary
electrophoresis (ACE)
’’
Separation of lysozyme from
chicken egg white using
ultrafiltration
Thin layer chromatography
for analysis of amino acids
Peptide profiling of cheese extracts
Real-time control of antibody loading
during protein A affinity
chromatography
Bioprocess monitoring and control
of flocculation from yeast cell
homogenate
Properties of benzoic acid derivative
with polypeptide
(benzoyl poly-L-lysine) materials
Determination of structural parameters
from chromatography retention data
fermentation unit operations involving the use of chemomet-
rics. As upstream processing is a critical part of the biotech
process, with the product of interest being expressed as well
as most of the product-related and host-related impurities
formed, this will be an area that will continue to see interest
in use of chemometric tools and approaches for process
modeling and optimization.
311
Process Analyzer Chemometric Tool Observations
HPLC Linear regression Scale ability demonstrated of using
on-line anion exchange HPLC for
operations at pilot scale and
eventually at manufacturing scale
for pooling of a hydrophobic
interaction chromatography (HIC)
column.54
UPLC ’’ UPLC offered significantly faster
separations (7 min) while retaining
the separation efficiency.55
HPLC ’’ Feasibility of using a commercially
available on-line high-performance
liquid chromatography (HPLC)
system for real-time pooling of process
chromatography column.56
Capillary Response surface Successful estimation of affinity binding
electrophoresis methodology (RSM) constant and prediction of the
significance of injection time, voltage,
and ligand concentration on
protein–neutral ligand binding.57
’’ RSM-Box Behnken design Identification of the greatest influential
factor in reaching a targeted response
out of capillary length, voltage, and
injection time on protein–ligand
binding.58
UV–vis Parameter scanning Assessment of the suitability of the two
membranes (Biomax 30-kDa
polyethersulfone and Ultracel Amicon
YM 30-kDa regenerated cellulose)
and determination of significant
process parameters.59
NIR spectroscopy PCA Fast and easy method for classification
of L-form amino acids either in
dissolved or solid condition in
blood samples.60
HPLC Fuzzy approach (FA), Fuzzy approach used for data reduction
PCA, MDS, with results matching that from
nonhierarchical visual matching.61
cluster analysis
On-line HPLC, Linear regression Accurate control of antibody loading
Off-line ELISA on a protein A column in
a CHO-derived recombinant
antibody purification process and
assessment of chromatography
column performance.62
NIR spectroscopy PLS Cost-effective and useful approach for
monitoring of cell debris, protein,
and ribonucleic acid (RNA) with
a [90% correlation between the
predictions and the actual data.63
HPLC, NMR, circular PCA Model based on the data matrix of 18
dichroism (CD) conjugates described by 19 variables
and the six principal components
explained 85.7% of the total
data variance.64
Chromatography Quantitative structure Structural parameters of a series of
retention relationships benzodiazepines (BDZS) were
(QSSRs) identified determining quantitative
structure–enantiospecific retention
relationships (QSERRs), and the
topography of BDZ-binding sites
on HSA was determined.65
Chemometrics Applications in Downstream Processing
In comparison to upstream applications, chemometrics appli-
cations are somewhat sparse in the area of downstream proc-
essing. This could be due to the fact that a lot of the analyzers
used in downstream processing, such as high-performance liq-
uid chromatography (HPLC) or capillary electrophoresis, have
312
Figure 2. The fluorescence spectra of the permeate and reten-
tate stream obtained using a Varian fiber optic
probe (FOP) assembly connected to a Varian Cary
Eclipse spectrofluorometer equipped with a xenon
flash lamp as the light source.
Reprinted with permission from Elshereef et al., Biotechnol
Prog., 2010, 26, 168-178, V
C AIChE.
relatively longer times of analysis (compared to spectroscopy-
based methods such as UV–vis) and also yield discrete data
points. Table 3 presents a summary of these applications. Most
of them are focused on chromatography and membrane separa-
tions. The chemometric tools used in these applications com-
monly include PCA, PLS, quantitative structure retention
relationships, response surface methodology, and multidimen-
sional scanning (MDS). PCA has been used to analyze data
from inactivated polio vaccine (IPV) purification process for
process optimization, enhancement of efficiency, and robust-
ness.66 This application will be later discussed in this section in
the form of a case study.
Chemometric analysis has also been used for optimizing
and predicting protein–ligand binding conditions in affinity
capillary electrophoresis.57,58 Response surface methodology
(Box Behnken design) has been used to model the effect of
injection time, voltage, and ligand concentration on protein–
neutral ligand binding. The model was subsequently vali-
dated by a series of experimental runs. Parameter scanning
technique, in combination with UV detection, was used for
screening ultrafiltration membranes for the fractionation of
Biotechnol. Prog., 2011, Vol. 27, No. 2
proteins from real biological solutions.59 The authors suc-
cessfully carried out separation of lysozyme directly from
natural chicken egg white solutions with 30-kDa membranes.
Fuzzy approach and PCA have been used to make inferential
analysis easier by greatly reducing the number of dependent
variables to enable chemometric analysis of peptide profiles
from cheese extracts.61 Combination of PLS and NIR has
been used for monitoring of a process that involves selec-
tively removing contaminants such as cell debris, host cell
proteins, and nucleic acids by flocculation of yeast cell ho-
mogenate.63 The approach was very cost effective and
yielded a [90% correlation between the predictions and the
actual data. PCA has been used to analyze data from
high-performance size exclusion chromatography to study
chromatographic behavior of conjugates of benzoic acid de-
rivative with polypeptide (benzoyl poly-L-lysine) materials.64
Analysis was performed on the data matrix of 18 conjugates
described by 19 variables and the six principal components.
The resulting model could explain 85.7% of the total var-
iance in the data and was able to describe the structural
behavior of each conjugate compound.
Case study illustrating use of chemometrics
as a PAT enabler
Use of on-line HPLC has been demonstrated as a feasible
approach for analysis that can facilitate real-time decisions
for pooling of a hydroxyapatite process chromatography col-
umn based on product quality attributes.55 Linear regression
fit was used to make this predictive model. The bottle neck
of this approach was the analysis time (13 min) of the frac-
tions that impact the yield. The analysis time was lowered to
7 min in a subsequent publication by using ultra-perform-
ance liquid chromatography (UPLC) for pooling a diethyl
amino ethyl process chromatography column.56 In a more
recent publication, a combination of the use of monolithic
columns and UPLC was able to reduce the analysis time to
1.3 min.54 This study involved pooling of a hydrophobic
interaction process chromatography column. The control
scheme was demonstrated to work at pilot scale (a 2,294-
fold scale-up from a 3.4-mL column in the lab to a 7.8-L
column in the pilot plant) and eventually to manufacturing
scale (a 45,930-fold scale-up from a 3.4-mL column in the
lab to a 158-L column in the manufacturing plant). Although
chemometrics is not used here in the same manner as in the
upstream applications, process modeling is an integral part
of these applications involving pooling of process chroma-
tography columns as despite the faster analysis, the analysis
lags elution by at least one fraction and so a model is needed
to be able to predict the purity of the (n þ 1)th fraction
based on the purity data available till nth fraction.54–56
Case study illustrating use of chemometrics for
process modeling
A study involving monitoring of fractionation of a whey
protein isolate (WPI) during dead-end membrane filtration
using fluorescence and chemometric tools was published
recently.67 The model system of WPI consisted of a-lactal-
bumin (a-LA), b-lactoglobulin (b-LG), and a small propor-
tion of bovine serum albumin (BSA). The objective of the
study was to quantify the concentrations of the three species
in the permeate and the retentate streams during ultrafiltra-
tion. Fluorescence spectroscopy was used because of its
Biotechnol. Prog., 2011, Vol. 27, No. 2
greater sensitivity in comparison to UV–vis spectroscopy
and its ability to detect concentrations as low as 10
10–
10
12 M. The technique also provided multidimensional ex-
citation emission spectra, which required chemometric tools
for analysis. The composition of a-LA, b-LG, and BSA in
feed, permeate, and retentate streams was analyzed using an
HPLC system equipped with the size exclusion column. Fig-
ure 2 illustrates a fluorescence spectra of the permeate and
retentate stream acquired using a HPLC measurements of a-
LA, b-LG, and BSA, which were treated as output matrix Y,
whereas the fluorescence measurements were treated as input
data matrix X for the purpose of calibrating PLS model.
Sixty-four synthetic ternary mixtures containing the individ-
ual whey proteins in different proportions were randomly
designed and used to develop the calibration models.
The number of factors in the PLS model was selected by a
cross-validation method that consisted of leaving out one
sample at a time. Nonlinear tools such as artificial neural
network were used in combination to PLS for prediction of
samples that have higher protein content (i.e., relating fluo-
rescence to protein concentration). The estimated correlation
coefficients for a-LA, b-LG, and BSA were 0.99, 0.98, and
0.87, respectively. Good agreement was obtained between
predicted and measured values, indicating the applicability
of the proposed method for simultaneous determination of
the three species under discussion.
Case study illustrating use of chemometrics for
analysis of manufacturing data
A recently published study involved chemometric analysis
of data obtained from historical manufacturing batches of the
Salk-IPV production.66 The objective of this study was to
extract useful information from the manufacturing data for
potential use in process optimization and enhancement of ef-
ficiency and robustness. The dataset comprised of both on-
line (measured continuously and recorded digitally) and off-
line (measured occasionally and recorded by paper trace)
data of over 50 production runs of three polio viruses, Maho-
ney (Type 1), MEF-1 (Type 2), and Saukett (Type 3), on the
Vero cell line at two different bioreactor scales—2  350 L
(38 batches) and 2  750 L (12 batches)—and the ensuing
unit operations that involved virus purification and inactiva-
tion. Individual unit operations were analyzed first before
developing a model for the entire process. Random distribu-
tion of points for the cell culture step in the two bioreactors
in the score plot (350-L scale) indicated that both the bio-
reactors operated in a similar way. PCA analysis of the viral
inactivation showed that the polio virus cultivation was inde-
pendent of the virus strain type. PCA analysis of downstream
data revealed two groups representing the two production
scales. To study the operational behavior, the data were cor-
rected for volume, and the analyzed data were found to be
randomly distributed, implying similar operational behavior
at both production scales, indicating consistent manufactur-
ing. A five factor PLS model was developed to predict the
specific activity (defined as the ratio of D-antigen and pro-
tein nitrogen content) at 700-L production scale, resulting in
R2 of 93% and model error of about 0.6. Thus, using this
model, specific activity could be predicted reasonably well
from the process data. This example highlighted the useful-
ness of chemometrics in extracting useful information from
complex datasets and leading to increased robustness, better
troubleshooting, and possibilities for future improvements.
313
Although the chemometrics applications in downstream
processing are sparse compared to those in upstream process-
ing, the growing push for process understanding from QbD
and PAT implementation in the biotech industry will ensure
that more and more is accomplished in using chemometric
approaches toward understanding functioning of complex
downstream unit operations such as chromatography.
Conclusions
This article reviews chemometric applications in biotech
processes. We hope to have demonstrated that chemometric
tools can be very useful and powerful in extracting useful
process information through analysis of the readily available
data in order to maximize process understanding. In view of
the complexity of biotech processes and products, we expect
chemometrics to continue to act as a significant enabler of
the QbD and PAT initiatives.
Literature Cited
1. Wold S. Chemometrics; what do we mean with it, and what do
we want from it? Chemom Intell Lab Syst. 1995;30:109–115.
2. Kourti T. Process analytical technology and multivariate statisti-
cal control, Part 1. Process Anal Technol. 2004;1:13–19.
3. Kourti T. Process analytical technology and multivariate statisti-
cal control, Part 2. Process Anal Technol. 2005;2:24–28.
4. Kourti T. Process analytical technology and multivariate statisti-
cal control, Part 3. Process Anal Technol. 2006;3:18–24.
5. Martin EB, Morris AJ. Enhanced bio-manufacturing through
advanced multivariate statistical technologies. J Biotechnol.
2002;99:223–235.
6. Tomita RK, Park SW, Sotomayor OAZ. Analysis of activated
sludge process using multivariate statistical tools—a PCA
approach. Chem Eng J. 2002;90:283–290.
7. Cozzolino D, Smyth HE, Gishen M. Feasibility study on the use
of visible and near-infrared spectroscopy together with chemo-
metrics to discriminate between commercial white wines of dif-
ferent varietal origins. J Agric Food Chem. 2003;51:7703–7708.
8. Gracia N, Thomas S, Bazin P, Duponchel L, Thibault-Starzyk F,
Lerasle O. Combination of mid-infrared spectroscopy and chemo-
metric factorization tools to study the oxidation of lubricating base
oils. Catal Today. 2010;155:255–260.
9. Seasholtz MB. Making money with chemometrics. Chemom
Intell Lab Syst. 1999;45:55–63.
10. Al-Jowder O, Defernez M, Kemsley EK, Wilson RH. Mid-infra-
red spectroscopy and chemometrics for the authentication of
meat products. J Agric Food Chem. 1999;47:3210–3218.
11. Lima FSG, Araújo MAS, Borges LEP. Determination of the
carcinogenic potential of lubricant base oil using near infrared
spectroscopy and chemometrics. Trib Int. 2003;36:691–696.
12. Wilkin OM, Maitlis PM, Haynes A, Turner ML. Mid-IR spec-
troscopy for rapid on-line analysis in heterogeneous catalyst
testing. Catal Today. 2003;81:309–317.
13. Rodionova OY, Sokovikov YV, Pomerantsev AL. Quality con-
trol of packed raw materials in pharmaceutical industry. Anal
Chim Acta. 2009;642:222–227.
14. Scafi SHF, Pasquini C. Identification of counterfeit drugs using
near-infrared spectroscopy. Analyst. 2001;126:2218–2224.
15. Rantanen J, Wikström H, Turner R, Taylor LS. Use of in-line
near-infrared spectroscopy in combination with chemometrics
for improved understanding of pharmaceutical processes. Anal
Chem. 2005;77:556–563.
16. Blanco M, Valdés D, Bayod MS, Férnandez-Marı́ F, Llorente I.
Characterization and analysis of polymorphs by near-infrared
spectrometry. Anal Chim Acta. 2004;502:221–227.
17. Blanco M, Coello J, Itturiaga H, Maspoch S, Pérez-Maseda C.
Determination of polymorphic purity by near infrared spectrom-
etry. Anal Chim Acta. 2000;407:247–254.
18. Dyrby M. Chemometric quantitation of the active substance
(containing C¼N) in a pharmaceutical tablet using near-infrared
314
(NIR) transmittance and NIR FT-Raman spectra. Appl Spec-
trosc. 2002;56:579–585.
19. Ghasemi J, Niazi A, Ghobadi S. Simultaneous spectrophotomet-
ric determination of benzyl alcohol and diclofenac in pharma-
ceutical formulations by chemometrics method. J Chin Chem
Soc. 2005;52:1049–1054.
20. Gabrielsson J, Lindberg NO, Lundstedt T. Multivariate methods
in pharmaceutical applications. J Chemom. 2002;16:141–160.
21. Johnson R, Yu O, Kirdar AO, Annamalai A, Ahuja S, Ram K,
Rathore AS. Applications of multivariate data analysis in biotech
processing. Biopharm Int. 2007;20:130–134, 136–138, 140–144.
22. Vaidyanathan S, Arnold SA, Matheson L, Mohan P, McNeil B,
Harvey LM. Assessment of near-infrared spectral information
for rapid monitoring of bioprocess quality. Biotechnol Bioeng.
2001;74:376–388.
23. Ündey C, Cinar A. Statistical monitoring of multistage, multi-
phase batch processes. IEEE Control Syst Mag. 2002;22:40–52.
24. Cunha CCF, Glassey J, Montague GA, Albert S, Mohan P. An
assessment of seed quality and its influence on productivity esti-
mation in an industrial antibiotic fermentation. Biotechnol Bio-
eng. 2002;78:658–669.
25. Ündey C, Ertunç S, Çinar A. Online batch/fed-batch process per-
formance monitoring, quality prediction, and variable-contribution
analysis for diagnosis. Ind Eng Chem Res. 2003;42:4645–4658.
26. Gunther JC, Conner JS, Seborg DE. Fault detection and diagno-
sis in an industrial fed-batch cell culture process. Biotechnol
Prog. 2007;23:851–857.
27. Kirdar AO, Conner JS, Baclaski J, Rathore AS. Application of
multivariate analysis toward biotech processes: case study of a
cell-culture unit operation. Biotechnol Prog. 2007;23:61–67.
28. Kirdar AO, Green KD, Rathore AS. Application of multivariate
data analysis for identification and successful resolution of a root
cause for a bioprocessing application. Biotechnol Prog.
2008;24:720–726.
29. Kirdar AO, Chen G, Weidner J, Rathore AS. Combining near-
infrared (NIR) spectroscopy and multivariate data analysis
(MVDA) for screening of raw materials used in the cell culture
medium for the production of a recombinant therapeutic protein.
Biotechnol Prog. 2010;26:527–531.
30. Kozlowski S, Swann P. Considerations for biotechnology prod-
uct quality by design. In: Rathore AS, Mhatre R, editors. Qual-
ity by Design for Biopharmaceuticals: Perspectives and Case
Studies. NJ: Wiley; 2009:9–30.
31. Read EK, Park JT, Shah RB, Riley BS, Brorson KA, Rathore
AS. Process analytical technology (PAT) for biopharmaceutical
products: concepts and applications—Part 1. Biotechnol Bioeng.
2010;105:276–284.
32. Read EK, Park JT, Shah RB, Riley BS, Brorson KA, Rathore
AS. Process analytical technology (PAT) for biopharmaceutical
products: concepts and applications—Part 2. Biotechnol Bioeng.
2010;105:285–295.
33. Rathore AS, Bhambure R, Ghare V. Process analytical technol-
ogy (PAT) for biopharmaceutical products. Anal Bioanal Chem.
2010;398:137–154.
34. Rathore AS, Winkle H. Quality by design for pharmaceuticals:
regulatory perspective and approach. Nat Biotechnol.
2009;27:26–34.
35. Rathore AS. A roadmap for implementation of quality by design
(QbD) for biotechnology products. Trends Biotechnol.
2009;27:546–553.
36. Luo Y, Chen G. Combined approach of NMR and chemomet-
rics for screening peptones used in the cell culture medium for
the production of a recombinant therapeutic protein. Biotechnol
Bioeng. 2007;97:1654–1659.
37. Rhiel M, Cohen MB, Murhammer DW, Arnold MA. Non de-
structive near-infrared spectroscopic measurement of multiple
analytes in undiluted samples of serum-based cell culture media.
Biotechnol Bioeng. 2002;77:73–82.
38. Riley MR, Rhiel M, Zhou X, Arnold MA, Murhammer DW. Si-
multaneous measurement of glucose and glutamine in insect
cell culture media by near infrared spectroscopy. Biotechnol
Bioeng. 1997;55:11–15.
39. Ryan PW, Li B, Shanahan M, Leister KJ, Ryder AG. Prediction
of cell culture media performance using fluorescence spectros-
copy. Anal Chem. 2010;82:1311–1317.
Biotechnol. Prog., 2011, Vol. 27, No. 2
40. Gunther JC, Conner JS, Seborg DE. Process monitoring and
quality variable prediction utilizing PLS in industrial fed-batch
cell culture. J Process Control. 2009;19:914–921.
41. Jung B, Lee S, Yang IH, Good T, Cote GL. Automated on-line
noninvasive optical glucose monitoring in a cell culture system.
Appl Spectrosc. 2002;56:51–57.
42. Teixeira AP, Portugal CAM, Carinhas N, Dias JML, Crespo JP,
Alves PM, Carrondo MJT, Oliveira R. In situ 2D fluorometry
and chemometric monitoring of mammalian cell cultures. Bio-
technol Bioeng. 2009;102:1098–1106.
43. Mazarevica G, Diewok J, Baena JR, Rosenberg E, Lendl B. On-
line fermentation monitoring by mid-infrared spectroscopy. Appl
Spectrosc. 2004;58:804–810.
44. Sivakesava S, Irudayaraj J, Demirci A. Monitoring a bioprocess
for ethanol production using FT-MIR and FT-Raman spectros-
copy. J Ind Microbiol Biotechnol. 2001;26:185–190.
45. Roychoudhury P, Harvey LM, McNeil B. At-line monitoring of
ammonium, glucose, methyl oleate and biomass in a complex
antibiotic fermentation process using attenuated total reflec-
tance-mid-infrared (ATR-MIR) spectroscopy. Anal Chim Acta.
2006;561:218–224.
46. Lennox B, Kipling K, Glassey J, Montague G, Willis M, Hiden
H. Automated production support for the bioprocess industry.
Biotechnol Prog. 2002;18:269–275.
47. Zhang H, Lennox B. Integrated condition monitoring and con-
trol of fed-batch fermentation processes. J Process Control.
2004;14:41–50.
48. Jørgensen P, Pederson JG, Jensen EP, Esbensen KH. On-line
batch fermentation process monitoring (NIR)—introducing ‘bio-
logical process time’. J Chemom. 2004;18:81–91.
49. Ferreira AP, Lopes JA, Menezes JC. Study of the application
of multiway multivariate techniques to model data from an
industrial fermentation process. Anal Chim Acta. 2007;595:
120–127.
50. Clark S, Barnett NW, Adams M, Cook IB, Dyson GA, Johnston
G. Monitoring a commercial fermentation with proton nuclear
magnetic resonance spectroscopy with the aid of chemometric-
s,Anal Chim Acta. 2006;563:338–345.
51. Gordon SH, Green RV, Wheeler BC, James C. Multivariate
FTIR analysis of substrates for protein, polysaccharide, lipid
and microbe content: potential for solid state fermentations.
Biotechnol Adv. 1993;11:665–675.
52. Rosa SM, Soria MA, Vélez CG, Galvagno MA. Improvement
of a two-stage fermentation process for docosahexaenoic acid
production by Aurantiochytrium limacinum SR21 applying sta-
tistical experimental designs and data analysis. Bioresour Tech-
nol. 2010;101:2367–2374.
53. McGovern AC, Ernill R, Kara BV, Kell DB, Goodacre R. Rapid
analysis of the expression of heterologous proteins in Esche-
richia coli using pyrolysis mass spectrometry and Fourier trans-
form infrared spectroscopy with chemometrics: application to
a2-interferon production. J Biotechnol. 2009;72:157–167.
54. Rathore AS, Parr L, Dermawan S, Lawson K, Lu Y. Large scale
demonstration of a process analytical technology application in
bioprocessing: use of on-line high performance liquid chromatog-
raphy for making real time pooling decisions for process chroma-
tography. Biotechnol Prog. 2010;26:448–457.
55. Rathore AS, Wood R, Sharma A, Dermawan S. Case study
and application of process analytical technology (PAT)
towards bioprocessing. II. Use of ultra-performance liquid
chromatography (UPLC) for making real-time pooling deci-
sions for process chromatography. Biotechnol Bioeng. 2008;
101:1366–1374.
56. Rathore AS, Yu M, Yeboah S, Sharma A. Case study and
application of process analytical technology (PAT) towards
bioprocessing: use of online high performance liquid chroma-
tography (HPLC) for making real time pooling decisions for
process chromatography. Biotechnol Bioeng. 2008;100:306–
316.
57. Montes RE, Hanrahan G, Gomez FA. Use of chemometric
methodology in optimizing conditions for competitive binding
partial filling ACE. Electrophoresis. 2008;29:3325–3332.
58. Hanrahan G, Montes RE, Pao A, Johnson A, Gomez FA. Imple-
mentation of chemometric methodology in ACE: predictive
Biotechnol. Prog., 2011, Vol. 27, No. 2
investigation of protein–ligand binding. Electrophoresis.
2007;28:2853–2860.
59. Wan Y, Lu J, Cui Z. Separation of lysozyme from chicken egg
white using ultrafiltration. Sep Purif Technol. 2006;48:133–142.
60. Heigl N, Huck CW, Rainer M, Najam-ul-Haq M, Bonn GK.
Near infrared spectroscopy, cluster and multivariate analysis
hyphenated to thin layer chromatography for the analysis of
amino acids. Amino Acids. 2006;31:45–53.
61. Piraino P, Parente E, McSweeney PLH. Processing of chromato-
graphic data for chemometric analysis of peptide profiles from
cheese extracts: a novel approach. J Agric Food Chem.
2004;52:6904–6911.
62. Fahrner RL, Blank GS. Real-time control of antibody loading
during protein-A affinity chromatography using an on-line
assay. J Chromatogr A. 1999;849:191–196.
63. Yeung KSY, Hoare M, Thornhill NF, Williams T, Vaghjiani
JD. Near infrared spectroscopy for bioprocess monitoring and
control. Biotechnol Bioeng. 1999;63:684–693.
315
64. Bolzacchini E, Consonni V, Lucini R, Orlandi M, Rindone B.
HP-SEC behaviour of substituted benzoylpoly-L-lysines by prin-
cipal component analysis and molecular dynamics simulation. J
Chromatogr A. 1998;813:255–265.
65. Kaliszan R. Chemometric analysis of biochromatographic data:
implications for molecular pharmacology. Chemom Intell Lab
Syst. 1994;24:89–97.
66. Thomassen YE, van Sprang ENM, van der Pol LA, Bakker
WAM. Multivariate data analysis on historical IPV production
data for better process understanding and future improvements.
Biotechnol Bioeng. 2010;107:96–104.
67. Elshereef R, Budman H, Moresoli C, Legge RL. Monitoring the
fractionation of a whey protein isolate during dead-end mem-
brane filtration using fluorescence and chemometric methods.
Biotechnol Prog. 2010;26:168–178.
Manuscript received Sept. 29, 2010, and revision received Dec. 1,
2010.