Talanta 156-157 (2016) 87–94

Contents lists available at ScienceDirect

Talanta
journal homepage: www.elsevier.com/locate/talanta

Bio-assisted potentiometric multisensor system for purity evaluation

of recombinant protein A
Edita Voitechovič a,n, Anton Korepanov b, Dmitry Kirsanov a,c,nn, Igor Jahatspanian c,
Andrey Legin a,c
a
St. Petersburg State University, Institute of Chemistry, St. Petersburg, Russia
b
Protein Contour LLC, St. Petersburg, Russia
c
Laboratory of Artificial Sensory Systems, ITMO University, St. Petersburg, Russia

art ic l e i nf o a b s t r a c t

Article history: Recombinant proteins became essential components of drug manufacturing. Quality control of such
Received 18 February 2016 proteins is routine task, which usually requires a lot of time, expensive reagents, specialized equipment
Received in revised form and highly educated personnel. In this study we propose a new concept for protein purity evaluation that
18 April 2016
is based on application of bio-assisted potentiometric multisensor system. The model object for analysis
Accepted 2 May 2016
Available online 2 May 2016
was recombinant protein A from Staphylococcus aureus (SpA), which is commonly used for monoclonal
antibody purification. SpA solutions with different amount of host cell related impurities (Escherichia coli,
Keywords: bacterial lysate) were analyzed. Two different bio-transducers were employed: proteinase K from Tri-
Potentiometric sensor tirachium album and baker's yeast Saccharomyces cerevisiae. It was shown that both bio-transducers are
Baker's yeast
able to induce changes in pure and lysate-contaminated SpA samples. Different products of yeast di-
Proteinase K
gestion and proteolysis with proteinase of pure SpA and lysate were detected with size exclusion high-
Protein A
Protein purity performance liquid chromatography (SE-HPLC). The induced changes of chemical composition are de-
tectible with potentiometric multisensor system and can be related to SpA purity through projection on
latent structures (PLS) regression technique. The proposed method allows for estimation of the impurity
content with 12% accuracy using proteinase K and 16% accuracy using baker's yeast. The suggested ap-
proach could be useful for early contamination warning at initial protein purification steps. The analysis
requires no expensive materials and equipment, no bio-material immobilization, and its duration time is
comparable with other commonly used methods like chromatography or electrophoresis though the
main part of this time is related to the sample preparation.
& 2016 Elsevier B.V. All rights reserved.

1. Introduction major source of therapeutic recombinant proteins. Well-studied

Protein production for pharmaceutical use and biochemical
studies has become a highly developed technological discipline in
recent years [1]. Rapid evolution of modern medicine and bio-
technology significantly increases the demand for recombinant
proteins [2]. Since the first recombinant protein, insulin, was in-
troduced in early 1980s, more than 100 new recombinant proteins
were approved for therapeutic use [3]. About 30% of these phar-
maceuticals are produced in Escherichia coli, showing that it is the
n
Correspondence to: Saint Petersburg State University, Institute of Chemistry,
Mendeleev Center, Universitetskaya Nab. 7-9, Saint Petersburg 199034, Russia.
nn
Corresponding author at: St. Petersburg State University, Institute of Chem-
istry, St. Petersburg, Russia.
E-mail addresses: voitechovic.edita@gmail.com (E. Voitechovič),
d.kirsanov@gmail.com (D. Kirsanov).
genetics, fast growth dynamics and high target product yield make
E. coli a preferred choice for non-glycosylated protein expression
[4]. Besides pharmaceutical protein production, E. coli is widely
employed in production of proteins for a broad range of bio-
technological and biochemical needs. Different recombinant forms
of Staphylococcus aureus protein A (SpA) are examples of such
objects. SpA, a cell-wall protein, is extensively used for detection
and purification of antibodies including therapeutic monoclonal
antibodies due to its high affinity to immunoglobulins and rela-
tively high chemical stability, comparing to other im-
munoglobulin-binding molecules [5–7].
Quality control comes to the foreground when the protein of
interest is produced for pharmaceutical needs or as a tool for
pharmaceutical manufacturing. Competitive biopharmaceutical
market increases quality requirements for modern products,
creating a growing demand for quality monitoring instruments.

http://dx.doi.org/10.1016/j.talanta.2016.05.009
0039-9140/& 2016 Elsevier B.V. All rights reserved.
88 E. Voitechovič et al. / Talanta 156-157 (2016) 87–94

The proteins produced in any biological source including E. coli 2. Materials and methods
must be extremely well purified from any impurities that can
potentially cause adverse side effects [9,10]. The main group of 2.1. Materials
impurities in recombinant protein products is the host cell related
one including host cell proteins, endotoxins (e. g. pyrogenic lipo- Yeast extract and CaCl2 (dihydrate) were from “Fluka Analytical”
polysaccharides derived from Gram-negative bacteria like E. coli), (Germany), peptone (Bacto™ Tryptone) was from “BD Diagnostic
nucleic acids and cell debris [8,11]. Conventional list of methods, Systems” (United States). D( þ)glucose, Na2HPO4, KH2PO4, NaCl,
suitable for effective protein analysis and recommended by the KCl, glycerin, and NaOH were from “Vekton” (Russia), Tween-20
International Pharmacopoeia for purity control, includes high- was from “Helicon” (Russia) and HCl was from “Reaktiv” (Russia).
performance liquid chromatography (HPLC), mass spectrometry Tris(hydroxymethyl)aminomethane (Tris) was from “Merck” (Ger-
(MS), capillary electrophoresis (CE) and gel electrophoresis [12– many). All chemicals were of analytical grade.
14]. These methods provide reliable versatile data on the compo- Recombinant proteinase K from Tritirachium album was pur-
sition of protein solutions, thus being indispensable for pharma- chased from “Thermo Scientific” (Lithuania). The activity of en-
zyme solution in storage buffer (10 mM Tris–HCl (pH 7.5) with 50%
ceutical manufacturing. However, there are certain disadvantages
(v/v) of glycerol) was 60 U/ml and protein concentration was
of these methods such as expensive equipment, expendable sup-
2 mg/ml. Active dry baker's yeast Saccharomyces cerevisiae was
plies and a need for highly educated personnel involvement. HPLC,
from “SAF NEVA” (Russia) and was purchased from the local retail
MS, CE and gel electrophoresis usually require sample pre-treat-
store.
ment, utilization of toxic reagents and/or special dyes, labels,
Molecular weight markers for SE-HPLC: bovine thyroglobulin
sorbents [11,15,16]. Thus, there is always a need for novel analy-
(670 kDa), bovine γ-globulin (158 kDa), chicken ovalbumin
tical approaches that could allow solving the same tasks but at
(44 kDa), horse myoglobin (17 kDa) and vitamin B12 (1.35 kDa)
lower cost.
were from Bio-Rad Laboratories (USA). Other materials that were
It was shown earlier that multisensor systems based on elec-
used for bacteria growth, protein purification and HPLC analysis
trochemical methods can be used for analysis of complex and
were of analytical grade and were obtained from Sigma-Aldrich.
dispersed liquids, including fermentation broth [17,18], pharma-
ceutical solutions [19], beverages [20–22] and clinical samples
[23,24]. Simultaneous detection of endotoxins and other bacterial 2.2. Samples for analysis
lysis contaminating species in purified water using impedance
technology was reported in [25]. Multisensor systems based on SpA was expressed in E. coli BL21 (DE3) strain (Protein Contour
potentiometric sensors have got certain advantages compared to LLC, Russia) in the presence of 0.1 mM isopropyl-β-D-thiogalacto-
other methods such as rapid response, good reproducibility, the side for 3 h. The composition of growth media was: 1% peptone,
0.5% yeast extract and 1% NaCl and the culture was grown with
ease of construction and miniaturization [26,27]. However, po-
shaking according to the standard protocol. Cells were harvested
tentiometric sensors are normally not sensitive to biological
by centrifugation at 10,000  g for 30 min and re-suspended in
macromolecules and non-ionized compounds. A special transdu-
10 mM Tris–HCl buffer (pH 8.0), containing 1 mM EDTA, 0.1 mM
cer should be employed to analyze such substances. Low mole-
PMSF and 1 mg/ml lysozyme. Cells were sonicated and frozen at
cular weight compounds like sugars can be determined by po-
 70 °C. Defrosted lysate was divided into two parts. One part was
tentiometry using enzyme- or microorganism-catalyzed reactions
dialyzed against distilled water for preparation of the “E. coli ly-
that would increase sample acidity according to sugar concentra-
sate” sample. Another part was acidified by 85% phosphoric acid to
tion [28]. Such an analytical technique usually employs im-
pH ¼3 and centrifuged at 4000  g for 20 min Supernatant was
mobilized bio-recognition element and it can be applied for ana-
loaded onto C10/10 column (GE Healthcare Bio-Sciences AB,
lyte quantification in aqueous solutions [26,29,30]. In order to
Sweden), packed with 4 ml Sepharose SP (GE Healthcare Bio-Sci-
simplify potentiometric measurements and to avoid microorgan-
ences AB, Sweden), and eluted with 10 mM citrate buffer pH 5.5.
ism immobilization a new approach for sugar analysis employing a
Elution pool containing purified SpA was dialyzed against distilled
yeast-assisted pH glass electrode was recently suggested [31]. The
water for further use as “purified SpA” sample. Purification quality
overall idea of the approach is to pretreat a sample through yeast
was controlled using size-exclusion high-performance liquid
metabolism and to detect corresponding changes in ionic com-
chromatography (SE-HPLC) (Fig. 1). The “purified SpA” sample was
position by potentiometric sensors. This approach seems being
diluted by distilled water to match the concentration of SpA in the
applicable not only for sugar quantification but also for other non- “E. coli lysate” sample (2 mg/ml).
ionized substances.
In the present research we were aiming to study the applic-
ability of bio-assisted potentiometric multisensor system for im-
purity detection in complex protein solutions. The model objects
were purified SpA solution and SpA solutions, contaminated with
E. coli components from the source biomaterial. In order to modify
initial media sample pretreatment with two different bio-trans-
ducers was done: proteinase K from Tritirachium album and ba-
ker's yeast Saccharomyces cerevisiae. Proteinase K is an en-
dopeptidase, which hydrolyzes a broad range of proteins yielding
oligopeptides and amino acids and it is widely applied in different
molecular biology processes [32]. Baker's yeast can metabolize
macromolecules like proteins, oligopeptides, lipids, nucleic acids
into simple compounds and ions [33]. Thus, both proteinase K and
yeast can work directly as SpA sample modulators without their Fig. 1. SE-HPLC chromatogram of E. coli lysate containing SpA and purified SpA.
immobilization on sensor surface. Both samples were dialyzed against distilled water.
 E. Voitechovič et al. / Talanta 156-157 (2016) 87–94
2.3. Baker's yeast cultivation and preparation for analysis
Standard liquid medium for yeast cultivation was employed
(yeast-peptone-dextrose (YPD)), 1% yeast extract, 2% peptone and
2% glucose at pH 6.5–6.6. 1 g of dry yeast was added into 100 ml
sterile YPD medium and was cultivated under stirring at 150 rpm/
min at 30 °C for 1 h. In order to obtain induced yeast, 0.5 vol% of
dialyzed against water E. coli lysate was introduced into standard
YPD medium and the rest of the procedures were the same as in
the case of common yeast cultivation. After cultivation the yeast
cells were harvested by centrifugation at 4000  g for 5 min at 4 °C
and washed once by phosphate buffered saline solution (PBS,
10 mM Na2HPO4, 2 mM KH2PO4, 137 mM NaCl, 2.7 mM KCl, pH
7.3–7.6) and then centrifuged once over again under the same
conditions. After that 1.07 g of obtained yeast biomass were re-
suspended in 1 ml PBS buffer and were stored at 4 °C and em-
ployed for experiments within one week.
2.4. Sample preparation for analysis
2.4.1. Sample preparation for potentiometric measurements
Samples of dialyzed bacterial lysate and purified SpA were
mixed in different ratios to obtain the mixtures with varying im-
purity content (while SpA amount was the same). The ratios of
lysate (%) and SpA (%) were as follows: 100/0, 94/6, 92/8, 87.5/12.5,
82/18, 75/25, 62.5/37.5, 60/40, 50/50, 37.5/62.5, 25/75, 20/80, 18/
82, 8/92, 5/95 and 0/100. Two reaction mixtures were prepared:
one for digestion by yeast and another one for hydrolysis by pro-
teinase K. The reaction mixture for digestion by yeast contained
48% (v/v) of PBS buffer (pH 7.0), 50% (v/v) of prepared “purified
SpA” and “E. coli lysate” samples and 2% (v/v) of 1.07 g/ml yeast
suspension. This mixture was incubated without stirring at 30 °C
in water bath of Julabo FP-50 thermostat (Julabo GmbH, Seelbach,
Germany) for 1 h according to the recommendations of enzyme
producer. The reaction mixture for hydrolysis with proteinase K
contained 48% (v/v) of proteinase K working buffer (8 mM KH2PO4,
4 mM NaCl, 3 mM CaCl2, pH 7.5), 50% (v/v) of prepared purified
SpA and E. coli lysate sample and 2% of 2 mg/ml proteinase K so-
lution. The reaction mixture was incubated without stirring at
37 °C in water bath of Julabo FP-50 thermostat for 1 h.
2.4.2. Sample preparation for SE-HPLC
Samples for SE-HPLC were taken directly from the mixtures
that were prepared for potentiometric analysis. These samples
were centrifuged at 20,000  g for 5 min before SE-HPLC
measurements.
2.5. Potentiometric analysis
Potentiometric sensor array consisted of 8 chemical sensors for
determination of H þ , K þ , Na þ , NH4 þ , NO3  , HCO3  , Cl  and Pt
wire as redox-sensitive electrode. K þ , Na þ , NH4 þ , NO3  , HCO3 
sensors were based on PVC-plasticized membranes produced ac-
cording to Flukas company recipes [34] employing valinomycin,
sodium ionophore X, nonactine, tridodecylmethylammonium ni-
trate and 3-bromo-4-hexyl-5-nitrotrifluoroacetophenone as iono-
phores correspondingly. PVC membranes were employed together
with solid contact based on conductive graphite glue made of
graphite/PVC mixture dispersed in cyclohexanone. Cl  sensor was
polycrystalline membrane pressed of AgCl/AgS2 mixture. The
commercial pH glass electrode (ES-106017) from Izmeritelnaya
Tehnika (Moscow, Russia) was used for H þ determination. Elec-
trochemical measurements were performed against Ag/AgCl re-
ference electrode (Gamry Instruments, US) with 0.1 mV precision
using digital 32-channel high input impedance mV-meter (Sensor
Systems, LLC, St. Petersburg, Russia) connected to PC for data
89
acquisition and processing. Measurements were performed in
thermostated electrochemical cell using Julabo FP-50 thermostat
at 25 °C and analysis time was 3 min. To increase reproducibility of
results the sensor array was washed after each measurement ac-
cording to the following sequence: 1 min in distilled water, 2 min
in 0.05% (v/v) Tween 20 water solution and three - five times in
distilled water for 3 min. This procedure allowed for stable re-
sponse values of the sensors in distilled water for at least
2 months.
2.6. HPLC analysis
SE-HPLC analysis was performed using Agilent 1260 Infinity
HPLC system (Agilent Technologies, Germany) and Superose 12 10/
300 GL column (GE Healthcare Bio-Sciences AB, Sweden) with PBS
buffer (pH 7.5) as mobile phase, flow rate 0.75 ml/min and spec-
trophotometric detection at 280 nm.
2.7. Data processing
Exploratory analysis of the data from potentiometric sensor
array was performed using principal component analysis (PCA).
PCA is a widely applied tool in chemometrics [35]. This method
projects the initial multidimensional data onto a new coordinate
system formed with so called principal components – orthogonal
coordinate axis in the directions of maximal variance in the data.
One of the main outcomes of PCA is a score plot visualizing initial
data as points on a two dimensional plane according to their dif-
ference or similarity in terms of the initial variables (sensor re-
sponses in our case). A projection on latent structures (PLS)
technique was used for establishing quantitative relation between
response of sensor array and E. coli lysate content in the analyzed
samples. PLS is a multivariate regression technique which can be
employed for construction of predictive models for estimation of
dependent variable (E. coli lysate) from a set of independent
variables (sensor responses). Estimation of the predictive ability of
the PLS model is performed through validation procedure where
unknown samples are presented to the model and the resulted
model prediction is compared with real value of the parameter
obtained from a reference method. Since the available number of
samples was limited, full cross-validation was employed to assess
the model performance. As a quantitative measure for model as-
sessment root mean squared error of cross-validation (RMSECV) is
usually employed:
n
∑i = 1 (yi, pred − yi, ref )2
RMSECV =
n (1)
where n is a number of samples in the test set, yi,pred is the value
predicted by model, yi, ref is the reference value. The Unscramblers
9.7 (CAMO, Norway) software was used for PCA and PLS data
processing.
3. Results and discussion
3.1. Protein purity estimation with proteinase K-assisted sensor
array
Proteinase K from Tritirachium album was employed as pro-
teolytic enzyme for hydrolysis of SpA and bacterial lysate. Ex-
pected products of hydrolysis are different oligopeptides and
aminoacids, which can form zwitterions, cations or anions de-
pending on pH of a sample solution. At the first step the sensor
sensitivities to the intact SpA and lysate were explored. It was
found that even 0.2 mg/ml of pure SpA and lysate containing the
90 E. Voitechovič et al. / Talanta 156-157 (2016) 87–94
Fig. 2. PCA score plot for control experiments and for sample medium after digestion with proteinase K (A). PCA score plot for reaction mixtures with pure SpA and E. coli
lysate, containing SpA, after digestion with proteinase K (B). Symbol's shape stands for measurements at the same pH. White symbols correspond to the samples with pure
SpA, black symbols – samples with bacterial lysate, gray symbols with white border are controls. The arrows on the B plot correspond to the increase of reaction mixture
amount (concentration of hydrolysis products) in analyzed medium.
same SpA concentration did not affect the signal of the sensors.
The response of the sensors to the reaction mixtures after hydro-
lysis with proteinase K was observed. Aiming to obtaining opti-
mum conditions for the potentiometric measurements the amount
of reaction mixture in analyzed medium and the incubation time
with proteinase were varied. The best results (in terms of differ-
ence between pure and contaminated SpA samples assessed by
PCA score plots) were achieved for the samples containing at least
0.5% (v/v) of reaction mixture in the analyzed medium, which
corresponds to 5 mg/ml of initial SpA concentration. Proteolysis
duration was 60 min The shorter times did not yield significant
response of the sensor array in analyzed media with 0.5% reaction
mixture. Since pH value affects ionization of reaction products,
potentiometric measurements were performed at different pH
values. pH of the medium was adjusted by H2SO4 and NaOH. The
buffer systems were not employed since they can mask the
changes in sample acidity or basicity. Fig. 2A represents relation-
ship between control experiments when the incubation of bac-
terial lysate was performed without proteinase and reaction
mixtures when the incubation of lysate and SpA was made with
proteinase for 1 h. In this case pH sensor response was excluded
from the PCA analysis because the changes of its response would
have strong influence on the results that can suppress the impact
of the other sensors. Control and reaction mixtures are well dis-
criminated as one can see in Fig. 2A. First principal component
(PC1) describes more than 85% of the variance in sensor responses
and corresponds to the differences between control and reaction
medium. The second principal component (PC2) stands for 12% of
total variance and is likely related to pH since in the both cases
(control and reaction) the clusters of sensor responses obtained at
different pH are organized from the top to the bottom and fol-
lowing the increase of pH level.
Control experiments were excluded from PCA analysis in order
to find the most suitable pH for potentiometric measurements.
Obtained PCA score plot (Fig. 2B) represents distribution of sensor
response clusters at different pH values. PC1 describes more than
90% of total variance and can be related to pH, while PC2 explains
only 4% and can be associated with the amount of ionic species
produced by proteolysis products at extreme pH levels (1.7 and
9.7). Moreover, the distribution of points corresponding to differ-
ent amount of reaction mixtures (concentration of hydrolysis
products) in analyzed medium normally follows the PC2 direction.
An interesting observation is that the points corresponding to pH
levels 3.3, 6 and 7 are grouped closely together, probably due to
the fact that the ionization of the products is not that well pro-
nounced at these conditions. Another important issue is that the
measurements at pH 9.7 can provide for better discrimination
between proteolysis products of purified SpA and bacterial lysate.
Therefore, all further potentiometric measurements were per-
formed in strong basic medium with 0.5% (v/v) of proteolysis re-
action mixture. The studied pH values correspond to the following
solutions: 10 mM H2SO4, 0.1 mM H2SO4, 1 mM H2SO4, 0.035 mM
NaOH and 1 mM NaOH.
Once the optimal conditions for potentiometric measurements
were established proteolytic hydrolysis of mixture with different
ratio between pure SpA and lysate was performed. PLS regression
modelling of sensor responses was employed to establish pre-
dictive models for lysate amount in the reaction mixture (which
corresponds to the purity of SpA) (Fig. 3A). Full-cross validation
was employed to assess the model performance. Optimal PLS
model was based on five latent variables (according to the mini-
mum at #LV-RMSECV plot). In order to confirm the validity of
results we also employed permutation testing of the PLS model
[36]. The vector of dependent variable (lysate content) was ran-
domly rearranged 50 times and in each case PLS modelling with
full cross-validation was performed. The results are shown in
Fig. 3B. The model based on original non-permuted data has the
lowest RMSECV, which confirms statistical validity of the model-
ling result.
These results imply that quantification of lysate amount in SpA
solutions can be performed with the approach based on po-
tentiometric multisensor array and proteinase K and the corre-
sponding errors are around 12%.
3.2. Protein purity estimation with yeast-assisted sensor array
SpA can potentially interact with yeast and affect its metabolic
behavior. A standard growth medium was examined by potentio-
metric multisensor array in order to evaluate the effect of SpA on
yeast ability to metabolize D-glucose. No evolution of sensor po-
tentials was observed since SpA has no influence on yeast under
the studied experimental conditions. No effect can be detected
 E. Voitechovič et al. / Talanta 156-157 (2016) 87–94 91

Fig. 3. Regression model for prediction of lysate amount in the medium (A). Plot of RMSECV vs. random permutation number (B). (A) Black line stands for ideal
diagonal. Calibration: slope 0.970, offset 1.528, R2 0.970, RMSEC 5.61; validation: slope 0.966, offset 0.684, R2 0.880, RMSECV 12.22 (B) Bar graph represents RMSECV values
after 50 permutations of the original data. Black bar represents the original RMSECV with non-permuted data.

Fig. 4. PCA score plot for medium after digestion with yeast at acidic and basic conditions (A). PCA score plot for medium after digestion with commonly grown and induced
yeasts at pH 10 (B).

with multisensor array either. Bacterial lysate can be a good sub-
strate for the yeast due to its chemical composition. Depending on
the carbon sources and environment, yeasts, like other micro-
organisms, can induce special metabolic pathways for adaptation
and optimal utilization of nutrients. Consequently, it is possible to
enhance the sensitivity of wild type baker's yeast to the bacterial
lysate. Aiming to obtaining such yeast culture we performed yeast
cultivation following the common method in standard YPD med-
ium with 0.5% (v/v) of dialyzed against water E. coli lysate and in
typical YPD medium. The baker's yeast culture, which was ob-
tained with lysate cultivation, will be further referred to as “in-
duced yeast”. The reaction mixtures with pure SpA and lysate were
incubated with both yeast cultures during 1 h similarly to the case
of proteinase K. Potentiometric analysis of obtained reaction
mixtures was performed at pH 6 and 10. The concentration of
reaction mixture employed for electrochemical measurements
was 0.5% (v/v). PCA score plot derived from these data (Fig. 4A)
shows that 96% of total variance in the data (PC1) is related to pH
value while only 3% (PC2) is likely associated with the difference
between pure and lysate-contaminated SpA. The discrimination
was somewhat better when potentiometric measurements were
performed at pH 10 (wider span along the PC2 axis).
PCA score plot in Fig. 4B demonstrates the influence of induced
and commonly grown yeasts on reaction mixtures. The difference
between SpA and lysate contribute to about 69% of total variance
in the data and clear separation of corresponding samples along
PC1 can be seen. Separation along the PC2 is due to the difference
in nutrients digestion by common and induced yeast. The appli-
cation of induced yeast culture resulted in a more significant
changes in the analyzed medium composition and thus in better
92 E. Voitechovič et al. / Talanta 156-157 (2016) 87–94
Fig. 5. Regression model for prediction of lysate amount in the media (A). Plot of RMSECV vs. random permutation number for the validation model from plot A. (A) Black
line stands for ideal diagonal. Calibration: slope 0.959, offset 1.751, R2 0.959, RMSEC 6.69; validation: slope 0.874, offset 6.725, R2 0.804, RMSECV 16.22 (B) Bar graph
represents RMSECV values in 50 permutations of the original data. Black bar represents the RMSECV for original data.
separation of lystate and SpA samples. All further potentiometric
measurements were performed with induced yeast and at pH 10.
The obtained results confirm that the yeast can metabolize bac-
terial lysate, producing low molecular weight ionic species stipu-
lating the response of potentiometric sensors.
Reaction mixtures with different ratio of pure SpA and lysate
were also incubated with induced yeast. Data processing em-
ployed to establish the relationship between sensor signals and
lysate concertation in reaction mixtures was similar to the ex-
periments with proteinase K. PLS regression modelling with full
cross-validation yielded reasonable results (Fig. 5A) in prediction
of contamination degree of SpA with RMSECV values around 16%.
Optimal model was based on four latent variables. In order to
confirm statistical validity of this result, 50 permutations of ori-
ginal concentration vector were performed and the results are
shown in Fig. 5B. The original concentration vector allows for the
lowest RMSECV, which confirms statistical validity of the model-
ling result. It must be noted that the employment of proteinase K
for sample digestion allows for higher precision of contamination
assessment. Probably, this fact is associated to poorer standardi-
zation opportunities while employing the whole microorganism
culture rather than particular enzyme.
3.3. Results of SE-HPLC analysis
SE-HPLC was applied for determination of the main changes in
macromolecular sample composition after incubation with yeast/
proteinase K. Three control reaction mixtures were prepared be-
sides typical reaction mixtures. The first control mixture contained
bacterial lysate only (without proteinase or yeast). The second
control mixture contained SpA only (without proteinase or yeast).
The third control mixture contained proteinase (or yeast) but did
not contain SpA or lysate. Incubation of all reaction mixtures fol-
lowed the protocols described in the Section 2.4.2. The obtained
SE-HPLC chromatograms are presented in Fig. 6. The chromato-
gram of molecular weight markers is shown in Fig. 6A. Fig. 6B and
C demonstrate the differences between proteolytic hydrolysis of
pure SpA and bacterial lysate. SpA proteolysis is close to 100%
resulting in the disappearance of corresponding protein peak just
as expected. Fig. 6C indicates that proteinase K mainly affects the
low molecular part of lysate and leaves the most “heavy” peak
almost unchanged. The chromatogram in Fig. 6D displays a very
weak yeast effect on purified SpA, which leads to insignificant SpA
peak height reduction. The changes in the lysate chromatogram
after digestion by yeast pointed out that yeast mostly change the
left-hand peak corresponding to some heavy lysate components,
probably peptidoglycans of bacterial cell envelope.
SE-HPLC results imply that proteinase K decomposes SpA into
low molecular weight compounds, including ions providing ob-
served potentiometric response, while the yeast does not affect
the SpA molecule. At the same time yeast affects some lysate
components with high molecular weight, which are not affected in
the presence of proteinase K, and this interaction likely results in
formation of ions that can be detected by potentiometric multi-
sensor system.
4. Conclusion
We demonstrated the applicability of bio-assisted potentio-
metric multisensor system for quantification of bacterial lysate
impurities in protein samples. Such an approach including pre-
liminary sample modification by means of bioactive compounds/
microorganisms appears being promising for extension of po-
tentiometric multisensor system capabilities. These systems are
normally not applicable for quantitative analysis of complex bio-
logical molecules due to the lack of sensitivity towards these
compounds. However, an appropriate sample modification
through biological digestion of certain components in the media
provides for reproducible response towards various metabolic
products and thus, for indirect potentiometric analysis of target
analyte. Both enzyme and whole microorganism were employed
as media bio-modificators in this research and both have shown
promising performance in promoting potentiometric sensitivity.
An important advantage of the suggested approach is that it does
not require immobilization of a biological object onto sensor sur-
face, thus making the overall measurement procedure quite sim-
ple and inexpensive.
 E. Voitechovič et al. / Talanta 156-157 (2016) 87–94
Fig. 6. SE-HPLC chromatograms. (A) Molecular weight markers. SpA (B) and lysate
(C) before and after incubation with proteinase K. SpA (D) and lysate (E) before and
after incubation with induced yeast.
Acknowledgements
E. Voitechovic would like to acknowledge the financial support
from St. Petersburg State University Postdoc Grant
#12.50.1191.2014 and support from research resource center
“Molecular and cell technologies” of St. Petersburg State University.
This study was supported by Protein Contour LLC (Saint-Peters-
burg, Russian Federation). This work was partially financially
supported by Government of Russian Federation, Grant 074-U01.
Appendix A. Supplementary material
Supplementary data associated with this article can be found in
the online version at http://dx.doi.org/10.1016/j.talanta.2016.05.009.
93
References
[1] R. Assenberg, P.T. Wan, S. Geisse, L.M. Mayr, Advances in recombinant protein
expression for use in pharmaceutical research, Curr. Opin. Struct. Biol. 23
(2013) 393–402.
[2] K. Swiech, V. Picanco-Castro, D.T. Covas, Human cells: New platform for re-
combinant therapeutic protein production, Protein Expr. Purif. 84 (2012)
147–153.
[3] J. Zhu, Mammalian cell protein expression for biopharmaceutical production,
Biotechnol. Adv. 30 (2012) 1158–1170.
[4] C. Huang Jr., H. Lin, X. Yang, Industrial production of recombinant therapeutics
in Escherichia coli and its recent advancements, J. Ind. Microbiol. Biotechnol. 39
(2012) 383–399.
[5] S. Hober, K. Nord, M. Linhult, Protein A chromatography for antibody pur-
ification, J. Chromatogr. B 848 (2007) 40–47.
[6] M. Graille, E.A. Stura, A.L. Corper, B.J. Sutton, M.J. Taussig, J.-B. Charbonnier, G.
J. Silverman, Crystal structure of a staphylococcus aureus protein A domain
complexed with the Fab fragment of a human IgM antibody: structural basis
for recognition of B-cell receptors and superantigen activity, Proc. Natl. Acad.
Sci. USA 97 (2000) 5399–5404.
[7] D. Low, R. O’Leary, N.S. Pujar, Future of antibody purification, J. Chromatogr. B
848 (2007) 48–63.
[8] B. DiPaolo, A. Pennetti, L. Nugent, K. Venkat, Monitoring impurities in bio-
pharmaceuticals produced by recombinant technology, Pharm. Sci. Technol.
Today 2 (1999) 70–82.
[9] M.R. Schenauer, G.C. Flynn, A.M. Goetze, Identification and quantification of
host cell protein impurities in biotherapeutics using mass spectrometry, Anal.
Biochem. 428 (2012) 150–157.
[10] V.R. Anicetti, B.A. Keytand, W.S. Hancock, Purity analysis of protein pharma-
ceuticals produced by recombinant DNA technology, Trends Biotechnol. 7
(1989) 342–349.
[11] M. D’Hondt, N. Bracke, L. Taevernier, B. Gevaert, F. Verbeke, E. Wynendaele,
B. De Spiegeleer, Related impurities in peptide medicines, J. Pharm. Biomed.
Anal. 101 (2014) 2–30.
[12] The International Pharmacopoeia, in: World Health Organization Expert
Committee (Ed.), Fifth ed., 1.15 Electrophoresis, World Health Organization,
〈http://apps.who.int/phint/en/p/docf/〉, 2015.
[13] The International Pharmacopoeia, in: World Health Organization Expert
Committee (Ed.), Fifth ed., A.3 Evaluation of Chemical reference substances,
World Health Organization, 〈http://apps.who.int/phint/en/p/docf/〉, 2015.
[14] The International Pharmacopoeia, in: World Health Organization Expert
Committee (Ed.), Fifth ed., A.4 Chemical and physical methods used in eval-
uating chemical reference substances, World Health Organization 〈http://apps.
who.int/phint/en/p/docf/〉 2015.
[15] N.A. Lacher, R.K. Roberts, Y. He, H. Cargill, K.M. Kearns, H. Holovics, M.
N. Ruesch, Development, validation, and implementation of capillary gel
electrophoresis as a replacement for SDS-PAGE for purity analysis of IgG2
mAbs, J. Sep. Sci. 33 (2010) 218–227.
[16] A.R. Mire-Sluis, T. Gerrard, R. Gaines Das, A. Padilla, R. Thorpe, Biological as-
says: their role in the development and quality control of recombinant bio-
logical medicinal products, Biologicals 24 (1996) 351–362.
[17] P. Ciosek, W. Wróblewski, Potentiometric and hybrid electronic tongues for
bioprocess monitoring-an overview, Anal. Methods 7 (2015) 3958–3966.
[18] K. Esbensen, D. Kirsanov, A. Legin, A. Rudnitskaya, J. Mortensen, J. Pedersen,
L. Vognsen, S. Makarychev-Mikhailov, Y. Vlasov, Fermentation monitoring
using multisensor systems: feasibility study of the electronic tongue, Anal.
Bioanal. Chem. 378 (2004) 391–395.
[19] A. Rudnitskaya, D. Kirsanov, Y. Blinova, E. Legin, B. Seleznev, D. Clapham, R.
S. Ives, K.A. Saunders, A. Legin, Assessment of bitter taste of pharmaceuticals
with multisensor system employing 3 way PLS regression, Anal. Chim. Acta
770 (2013) 45–52.
[20] Z. Wei, J. Wang, L. Ye, Classification and prediction of rice wines with different
marked ages by using a voltammetric electronic tongue, Biosens. Bioelectron.
26 (2011) 4767–4773.
[21] X. Cetó, F. Céspedes, M. Del Valle, Bioelectronic tongue for the quantification of
total polyphenol content in wine, Talanta 99 (2012) 544–551.
[22] X. Cetó, J.M. Gutiérrez, A. Mimendia, F. Céspedes, M. Del Valle, Voltammetric
electronic tongue for the qualitative analysis of beers, Electroanalysis 25
(2013) 1635–1644.
[23] M. Gutiérrez, S. Alegret, M. del Valle, Potentiometric bioelectronic tongue for
the analysis of urea and alkaline ions in clinical samples, Biosens. Bioelectron.
22 (2007) 2171–2178.
[24] F. Phillips, K. Kaczor, N. Gandhi, B.D. Pendley, R.K. Danish, M.R. Neuman,
B. Tóth, V. Horváth, E. Lindner, Measurement of sodium ion concentration in
undiluted urine with cation-selective polymeric membrane electrodes after
the removal of interfering compounds, Talanta 74 (2007) 255–264.
[25] J.Y. Heras, D. Pallarola, F. Battaglini, Electronic tongue for simultaneous de-
tection of endotoxins and other contaminants of microbiological origin, Bio-
sens. Bioelectron. 25 (2010) 2470–2476.
[26] A. Bratov, N. Abramova, A. Ipatov, Recent trends in potentiometric sensor ar-
rays—a review, Anal. Chim. Acta 678 (2010) 149–159.
[27] T. Mottram, A. Rudnitskaya, A. Legin, J.L. Fitzpatrick, P.D. Eckersall, Evaluation
of a novel chemical sensor system to detect clinical mastitis in bovine milk,
Biosens. Bioelectron. 22 (2007) 2689–2693.
[28] A.L. Simonian, E.I. Rainina, Microbial biosensors based on potentiometric
94 E. Voitechovič et al. / Talanta 156-157 (2016) 87–94

detection, in: A. Mulchandani, K.R. Rogers (Eds.), Enzyme and Microbial Bio-
sensors. Techniques and Protocols, Humana Press Inc., New Jersey, 1998,
pp. 237–249.
[29] A. Dwevedi, A.M. Kayastha, Enzyme immobilization: a breakthrough in en-
zyme technology and boon to enzyme based industries, Proteom. Res. J. 3
(2012) 333–352.
[30] L. Su, W. Jia, C. Hou, Y. Lei, Microbial biosensors: a review, Biosens. Bioelectron.
26 (2011) 1788–1799.
[31] E. Voitechovic, D. Kirsanov, A. Legin, An approach to potentiometric sensing of
sugars: Baker’s yeast assisted pH electrode, Sens. Actuator B–Chem. 225 (2016)
209–212.
[32] W. Saenger, Proteinase K, in: N.D.R. Salvesen (Ed.), Handbook of Proteolytic
Enzymes, Academic Press, United States, 2013, pp. 3240–3242 (Chapter 714 ).
[33] R. Walmsley, P. Keenan, The eukaryote alternative: advantages of using yeasts
in place of bacteria in microbial biosensor development, Biotechnol. Biopro-
cess Eng. 5 (2000) 387–394.
[34] 〈http://www.sigmaaldrich.com/analytical-chromatography/analytical-re
agents/sensoric-applications.html〉 (accessed 14.04.16).
[35] M. Marcel, N. Yorck-Michael, Model-free analyses, in: M. Marcel, N. Yorck-
Michael (Eds.), Practical Data Analysis in Chemistry, Elsevier, United States,
2007, pp. 213–316.
[36] H. van der Voet, Comparing the predictive accuracy of models using a simple
randomization test, Chemom. Intell. Lab. Syst. 25 (1994) 313–323.