You are on page 1of 6

Journal of Cereal Science 58 (2013) 31e36

Contents lists available at SciVerse ScienceDirect

Journal of Cereal Science


journal homepage: www.elsevier.com/locate/jcs

Cereal protein analysis via Dumas method: Standardization of a


micro-method using the EuroVector Elemental Analyser
S. Serrano a, *, F. Rincón a, J. García-Olmo b
a
Department of Food Science and Technology, Charles Darwin Building Annex, Campus of Rabanales, University of Córdoba, E-14014 Córdoba, Spain
b
Central Service for Research Support (SCAI), Campus of Rabanales, University of Córdoba, E-14014 Córdoba, Spain

a r t i c l e i n f o a b s t r a c t

Article history: The Dumas method (DM) of protein analysis in foods was standardized using two cereal references in
Received 13 November 2012 order to identify the main sources of error during analysis; specifically, the study sought to establish
Received in revised form which of the apparent critical factors (ACFs) were really significant influencing factors (IF) for the
27 March 2013
application of the DM to protein analysis in cereals.
Accepted 8 April 2013
Using a non-geometric Plackett-Burman experimental design augmented to resolution IV via fold-over
by increasing to twenty-four runs, the following eleven ACFs were investigated: sample weight (WS),
Keywords:
capsule sealing process (CW), carrier gas flow (PC), purge gas flow (FHe), oxygen volume (O2), oxygen
Dumas method
Protein analysis
pressure (DO2), sample delay time (SD), sample run time (SR), front furnace temperature (FFT), rear
Kjeldahl method furnace temperature (RFT), and GC oven temperatures (GCT).
PlacketteBurman experimental design Two IFs were identified for each of the factors RFT and GCT, so an optimization process was carried out,
based on a central composite design.
Optimized conditions, ensuring enhanced accuracy and repeatability when determining protein
content in cereals using the DM, are reported.
Ó 2013 Elsevier Ltd. All rights reserved.

1. Introduction has it slowly begun to replace the traditional KM as the method of


choice for protein analysis.
The Dumas method (DM) for the quantitative determination of Certain difficulties have been reported with the quantitative
nitrogen is based on the Dumas method originally employed to conversion of organic nitrogen to ammonia (Ma and Rittner, 1979).
determine the molecular weight of an unknown gas; it is used as an Some forms of nitrogen are difficult, if not impossible, to convert to
alternative to the classical Kjeldahl method (KM) for analyzing the ammonia (Sebranek and Cassens, 1973) and the KM achieves low
protein content of foods (Jiménez and Lahda, 1993). recoveries of nitrogen from refractory compounds with NeO or Ne
The two methods are based upon very different principles. The N bonds such as nitrite, nitrate, oximes azo-, nitro- and nitroso-
DM consists in combusting a sample of known mass at high tem- compounds (Kubota et al., 2011); as a result, values obtained using
perature in the presence of oxygen; the gases produced are then the DM tend to be slightly higher than those provided by the KM. In
reduced by copper and dried while the CO2 is trapped, and finally the case of cereal products, that difference is reported to range from
nitrogen is quantified using a universal detector. In the KM, by 2% to 4% (Bellomonte et al., 1987) with a mean value of 1.7%
contrast, the sample is wet digested to convert organic nitrogen (Thompson et al., 2002). Most authors agree that DM gives a more
into ammonia nitrogen by boiling sulfuric acid; the ammonia reliable measure of organic nitrogen than KM and offers greater
liberated is distilled, collected and titrated and a value for crude precision in the determination of protein content in cereals and
protein is calculated. Although the DM was introduced by Jean- oilseeds (Beljkas et al., 2010). In addition, DM has been shown to
Baptiste Dumas in 1831, and thus predates the KM by over 50 have certain major advantages: it is faster, more economical and
years (Kjeldahl, 1883), only over the last ten years e thanks to less polluting than KM (Chiacchierini et al., 2003) and probably
improvements in dry combustion nitrogen analyzer technology e safer, according to our self experience.
The next step must therefore be to devise a suitable protocol for
each type of sample, with a goal of optimizing the analytical process
* Corresponding author. Tel.: þ34 9 57 212654; fax: þ34 9 57 212000. using the DM. Some authors have attempted to optimize the DM
E-mail address: bt2sejis@uco.es (S. Serrano). using a 2k experimental design (Saint-Denis and Goupy, 2004).

0733-5210/$ e see front matter Ó 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.jcs.2013.04.006
32 S. Serrano et al. / Journal of Cereal Science 58 (2013) 31e36

However, this experimental way has proved that it is not the best 2.2. Reference materials
way of optimizing a process because two consecutive steps are
required. First, we need to determine which of the apparent critical The material used to calibrate analytical instruments for DM
factors (ACFs) are actually of major importance to the analytical protein analysis is generally selected on the basis of sample nitro-
procedure (step one, or screening phase). To address step two (or gen content, although there is usually a marked difference in matrix
optimization phase), experimental designs with three- to five-levels characteristics between calibration material and actual samples.
per factor are necessary (Myers and Montgomery, 1995), because For example, atropine (4.84% N) or acetanilide (10.36%) have been
when a variable is measured at three or more values, a quadratic used in infant-foods analysis (Bellomonte et al., 1987) and NaN3
response surface can be estimated by least-squares regression and solutions for milk (Jakob et al., 1995). Here, the reference standard
the predicted optimal value (i.e. the stationary point of the system) used for instrument calibration was 2, 5-bis (5-tert-butyl-benzox-
can be found from the estimated surface (SAS, 1999). azol-2-yl) thiophene (BBOT).
Previous obtained results unpublished on lycopene extraction Two cereal reference materials, BCR-381 rye (Secale cereale L.)
from tomato (Solanum lycopersicum L.) products highlighted the flour, 1.249  0.1 %N) and BCR-382 wheat (Triticum aestivum) flour,
differences between maximization (using a 2-levels- per-factor 2.12  0.1% N) were used for screening the factors affecting the DM
design, such as that employed by Saint-Denis and Goupy, 2004) analytical process.
and optimization (using a 5-levels-per-factor design, as reported
here). Marked quantitative differences were reported between the 2.3. Experimental design
two processes and between the end results, confirming that
maximization and optimization are clearly very different concepts. 2.3.1. Screening phase
Since the design to be used depends on the research stage An n  n matrix H ¼ hij is a Hadamard matrix of order n if the
involved e in the early stages, the goal is to detect or confirm the entries of H are either þ1 or 1, and such that HHt ¼ nI, where Ht is
importance of the variables or factors included in the experimental the transpose of H and I is the order n identity matrix. In other
design e and since screening designs are the most commonly-used words, a (þ1, 1) matrix is a Hadamard matrix if the inner product
factorial designs for determining the robustness of an analytical of two distinct rows is 0 and the inner product of a row with itself is
process (Youden and Steiner, 1975), the aim of this study was to n. Taking Hadamard matrices as their starting-point, Plackett and
analyze the effect of certain ACFs on the DM analytical process, Burman (1946) developed a screening-experiment design (PB)
using a PlacketteBurman experimental design to identify IFs and enabling estimation of main effects, assuming e on the basis of the
later to optimize the IF combination. effect sparsity principle e that all interactions are negligible when
compared with the few important main effects. Later, Box and
Hunter (1961) demonstrated that resolution IV designs can be ob-
2. Material and Methods
tained by folding over the PB experimental design.
Although other authors (e.g. Saint-Denis and Goupy, 2004)
2.1. Analytical instrument
considered only three factors for DM optimization, a review of the
instrument manual, a survey of the literature going back 10 years
The Dumas method is based on combustion of the whole sample
and a brainstorming session led to the identification of eleven
in an oxygen-enriched atmosphere at a high temperature in order
apparent critical factors (ACFs); a non-geometric PB design was
to ensure complete combustion. A EuroVector Elemental Analyser
therefore selected, because the number of experiments was a mul-
EA3000 equipped with Callidus software (EuroVector SpA, Milan,
tiple of four but not a power of two, so twelve runs were required
Italy) was used. The analyzer was generally operated in accordance
(k ¼ n  1). The ACFs examined and their experimental ranges are
with manufacturer’s instructions, but also bearing in mind the
shown in Table 1. Finally, to obtain resolution IV, the number of ex-
experimental design used for the screening phase (Table 1).
periments included in the experimental design was increased to
The present study included two samples obtained from the
twenty four runs via fold over. The experimental design is shown in
Community Bureau of Reference (BCR). The role of certified refer-
Tables 2 and 3 for CRM-381 and CRM-382, respectively.
ence materials (CRMs) in analytical chemistry is now well estab-
lished, particularly for calibration and for verifying the accuracy of
2.3.2. Optimization phase
analytical methods (Belliardo and Wagstaffe, 1988). The general
The simplicity of the screening phase lies in the assumption of a
point of using CRMs is to improve accuracy in an optimization
linear relationship between the settings of the factors and the re-
process such as that reported in this paper.
sponses. However, the optimization phase requires a more complex
Table 1 experimental design because only when each variable is measured
Eleven apparent critical factors (ACF) investigated, and experimental ranges tested at 3 or more levels, can a quadratic response surface be estimated
for the PlacketteBurman experimental design at resolution IV level. by least-squares regression and the optimal values (or stationary
ACF Symbol Levels point of the system) be found from the estimated surface (SAS,
1999). A two-level experimental design such as that used by
Code Actual 1 þ1
Saint-Denis and Goupy (2004) can thus never be used to optimize a
Sample weight, mg X1 WS 2 4 process. This is a fundamental consideration, since two points can
Capsule sealing processa X2 CW A B
Carrier gas flow, ml min1 X3 PC 80 120
only serve to estimate a linear model, but never a quadratic model.
Purge gas flow, ml min1 X4 FHe 60 100 Hence, at least a three-level design, e.g. a full factorial or Boxe
Oxygen volume, ml X5 O2 7 14 Behnken experimental design, must be used for optimization
Oxygen pressure, kPa X6 DO2 20 40 purposes.
Sample delay time, seg X7 SD 10 20
After the screening process, as indicated later under Results and
Sample run time, seg X8 SR 150 250
Front furnace temperature,  C X9 FFT 800 1000 Discussion, the number of key factors to be optimized turned out to
Rear furnace temperature,  C X10 RFT 500 750 be just 2. It was therefore possible to use a central composite design
GC oven temperature,  C X11 GCT 75 95 (CCD) (Box and Wilson, 1951) which is a sum of a two-level factorial
a
A ¼ capsule sealed with pliers to obtain the shape of a small bag. B ¼ capsule design and a star design, to obtain 5 levels per factor, as shown in
sealed with the help of tweezers to obtain a spherical shape. Table 4.
S. Serrano et al. / Journal of Cereal Science 58 (2013) 31e36 33

Table 2
PlacketteBurman experimental design and results obtained for rye flour CRM-381 (R1 and R2) and wheat flour CRM-382 (R3 and R4). The responses considered were R1 and
R3 ¼ mean difference between detected and certified values of four replicates; R2 and R4 ¼ SD of replicate determinations.

Trial Run X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 R1 R2 R3 R4

1 14 1 1 1 1 1 1 1 1 1 1 1 1.831 2.094 0.711 0.844


2 10 1 1 1 1 1 1 1 1 1 1 1 0.013 0.046 0.020 0.075
3 17 1 1 1 1 1 1 1 1 1 1 1 0.120 0.102 0.129 0.069
4 15 1 1 1 1 1 1 1 1 1 1 1 0.438 0.463 0.479 0.551
5 20 1 1 1 1 1 1 1 1 1 1 1 6.532 5.686 22.135 13.946
6 22 1 1 1 1 1 1 1 1 1 1 1 0.153 0.034 0.325 0.022
7 19 1 1 1 1 1 1 1 1 1 1 1 0.181 0.144 0.267 0.024
8 13 1 1 1 1 1 1 1 1 1 1 1 0.120 0.030 0.200 0.032
9 9 1 1 1 1 1 1 1 1 1 1 1 0.005 0.015 0.001 0.013
10 2 1 1 1 1 1 1 1 1 1 1 1 0.101 0.014 0.175 0.032
11 18 1 1 1 1 1 1 1 1 1 1 1 0.459 0.354 0.423 0.227
12 6 1 1 1 1 1 1 1 1 1 1 1 0.467 0.089 0.349 0.141
13 4 1 1 1 1 1 1 1 1 1 1 1 0.202 0.184 1.078 0.428
14 1 1 1 1 1 1 1 1 1 1 1 1 5.911 2.269 7.886 2.764
15 5 1 1 1 1 1 1 1 1 1 1 1 0.501 0.030 0.550 0.062
16 7 1 1 1 1 1 1 1 1 1 1 1 0.022 0.025 0.022 0.007
17 3 1 1 1 1 1 1 1 1 1 1 1 0.029 0.022 0.007 0.018
18 12 1 1 1 1 1 1 1 1 1 1 1 0.009 0.010 0.017 0.037
19 23 1 1 1 1 1 1 1 1 1 1 1 0.748 0.039 1.469 0.253
20 21 1 1 1 1 1 1 1 1 1 1 1 0.033 0.029 0.028 0.015
21 24 1 1 1 1 1 1 1 1 1 1 1 0.688 0.040 1.106 0.098
22 11 1 1 1 1 1 1 1 1 1 1 1 0.141 0.018 0.285 0.140
23 16 1 1 1 1 1 1 1 1 1 1 1 0.690 0.214 1.989 0.445
24 8 1 1 1 1 1 1 1 1 1 1 1 0.312 0.246 0.833 0.541

Accuracy is the degree of concordance between the detected Effects identified as IF (p  0.25) for each of the responses
value and the certified reference value, so the R1 response measures considered are shown in Table 4. Certain IFs were identified
the difference between detected (mean of four replicates) and amongst the apparent critical factors (ACF) included in the exper-
certified values. Precision, by contrast, is the variability of the re- imental screening design (Table 1). On the basis of the effects
sults obtained when analyzing the same sample using the same detected, a number of factors (X1 to X4 and also X6) appeared not to
analytical method for validation purposes. The R2 response mea- be critical for the implementation of the DM, i.e. differences
sures the SD of four replicate measurements. between 1 and þ1 levels were not significant at the p  0.25
Because accuracy is the concordance between the detected value confidence level. Thus, over the experimental range considered
obtained and the certified reference value, the R1 (CRM-381) and R3 (Table 1), sample weight, capsule sealing process, carrier gas flow,
(CRM-382) responses will measure the difference between both purge gas flow and oxygen pressure, appeared not to be critical in
detected (mean of four replicates) and certified values (Table 2). the case of cereal samples. A smaller difference between detected
Precision is the variability of results obtained by analyzing the same and certified nitrogen content (R1 and R3 responses, Table 4) and a
sample using the same method of analysis to validation, so R2 (CRM- lower SD (R2 and R4 responses, Table 4), therefore needed to be
381) and R4 (CRM-382) responses were used to measure the SD of considered during the stage in optimization, as did the low levels
four replicates determinations. In summary, factors affecting both recorded for X1, X2, X4 and the high levels recorded for X3 and X6.
accuracy and precision of DM were detected and later optimized. Similar results for the influence of sample weight were reported by
For both screening and optimization phases, the Design-Expert Saint-Denis and Goupy (2004), who concluded that DM, though
(Stat-Ease, Inc. Minneapolis) and Statistical (StatSoft, Inc, Tulsa) faster than KM, was less precise for milk samples. By contrast,
software packages were used to generate designs, fit the response Kubota et al. (2011) reported that the DM provided excellent
surface model to the experimental data, and draw response sur-
faces figures at optimization phase. In addition, runs were per- Table 4
formed in random order (trial order) because randomization helps Effects in terms of coded factors on each response considered for strong effects. A
conventional criterion was considered to define a strong effect such as p  0.25 and
protect the experimenter, again reaching erroneous conclusions
so to identify the factor as an influencing factor (IF).
due to extraneous sources of variability (Joglekar and May, 1987).
Responsesa

Rye flour Wheat flour


3. Results and Discussion
Factors R1 R2 R3 R4
The results obtained for each of the trials included in the X1
experimental design for both rye and wheat CRM samples are X2
X3
shown in Table 2.
X4
Table 3 X5 0.84 2.67 1.36
Influencing factors to be optimized in a central composite experimental design, with X6
experimental ranges expressed in coded and actual units. X7 1.13 0.81 2.35 1.37
X8 0.89 0.52 2.57 1.36
Symbol Levels X9 0.52 1.38
X10 1.10 0.75 2.38
Code Actual 1.41 1 0 1 1.41
X11 1.11 0.87 2.36 1.50
Rear furnace temperature,  C X10 RFT 609 650 750 850 891 a
R1 and R3 ¼ difference between detected and certified values; R2 and R4 ¼ SD of
GC oven temperature,  C X11 GCT 53.8 60 75 90 96.2
replicate determinations.
34 S. Serrano et al. / Journal of Cereal Science 58 (2013) 31e36

precision for fish samples, while Daun and DeClercq (1994) found every sample type requires its own specific conditions in order to
that DM yielded more accurate measurements than KM in oilseed ensure the best possible analytical performance.
samples and Nozawa et al. (2007) reported that DM provided A similar analysis can be carried out for X7 (sample delay time,
excellent precision and was at least as accurate as KM for soy Table 1) leading to the same conclusion. A longer waiting-time
(Glycine max L.) sauce samples. It must therefore be concluded that before the sample is placed in the oven-reactor leads to a larger
certain factors in DM implementation are critical for ensuring amount of air in the sample container, which would readily account
adequate precision. The poorer precision reported by Saint-Denis for the effects observed (Table 4).
and Goupy (2004) may be ascribed to use of an experimental X8 (sample run time, Table 1) exerted a negative effect on ac-
range (700e1300 mg) much larger than that used here (Table 1) curacy in both samples, the effects were especially marked in the
and also greater than the sample size recommended by the wheat sample (Table 4, R1 and R3). The mean detected value at
manufacturer (150e200 mg) for a similar instrument of the same level 1 was 2.16% N, compared with a level þ1 value of 1.27% N in
brand (Beljkas et al., 2010). the rye sample (certified value 1.249% N), and with 4.49% N and
With regard to the IFs detected (Table 4), some preliminary 2.09% N, respectively, in the wheat sample (certified value 2.12% N).
comments are warranted. First, similar IFs were detected for rye and In other words, as sample run time increased, so did measurement
wheat samples: X5 and X7 to X11. These factors (oxygen volume, accuracy. At the same time, the effect on both R2 and R4 was also
sample delay time, sample run time, front and rear furnace tem- negative, i.e. SD was smaller and therefore measurement precision
perature and GC oven temperature) therefore needed, a priori, to be was greater.
optimized in a subsequent step. Second, although nitrogen content Briefly, X8 displayed a negative linear effect on all the responses
in wheat samples was 1.7 times that of rye samples, the effects of studied (Table 4), indicating that, in the optimization phase, the
factors identified as IF in both sample types displayed a similar trend value to be used should lie at the higher end of the experimental
(Table 4), reflecting the similar composition of the two cereal sam- range tested (Table 1, X8). The effects recorded for this factor can
ples. However, this difference in nitrogen concentration prompted readily be accounted for, given that a decrease in chromatogram
a difference in uncertainty, a parameter e associated with the result duration leads to the appearance of chromatographic peaks that fail
of a measurement e that characterizes the dispersion of the values to reach the baseline, thus prompting errors in peak integration due
that could reasonably be attributed to the measurand, according to to poorer definition of baseline peaks. Saint-Denis and Goupy
Horwitz’s criteria (Horwitz, 2003), uncertainty for rye and wheat (2004) obtained outliers when integration time was short, while
samples was 4.8  104 and 7.5  104, respectively. Williams et al. (1998) note that instrument malfunction, i.e. the use
X5 (oxygen volume, Table 1) exerted a positive effect on R1 and R3 of non-optimized instrument conditions, should be included
in both samples, the effect being particularly marked for wheat among the most important sources of error in Dumas testing,
(Table 4, R3). The higher the value, the further it lay from the real which is why the equipment operating manual should specify that
value. Specifically, values rose from 0.4 to 1.24 units for rye optimum conditions need to be established for each sample type.
(effect ¼ 1.24  0.4 ¼ 0.84, Table 4, R1) and from 0.35 to 3.02 units for Regarding X9 (front furnace temperature, Table 1), either of the
wheat (effect ¼ 3.02  0.35 ¼ 2.67, Table 4, R3). In other words, as tested values for this parameter (800  C and 1000  C, Table 1) were
oxygen volume increases, measurement becomes less accurate, sufficient to ensure complete sample combustion. Hence, there was
leading to a higher-than-actual result for N content. Moreover, this no difference between detected % N and certified % N for these
loss in accuracy is accompanied by a sharp increase in R2 and R4 (2.6 samples (Table 4). For example, a furnace temperature of 870  C, i.e.
for rye and 8 for wheat), thus prompting a major loss in precision. lower than the maximum included in the experimental range used
Briefly, oxygen volume (Table 1, X5) displayed a positive linear here (Table 1), has been used successfully for other types of food
effect both for R1 (rye sample) and for R3 and R4 (wheat sample), including soy sauce (Nozawa et al., 2007). However, this factor did
indicating that as the value of this parameter increases, both the display a marked effect on the precision of the method (Table 4, R2
accuracy (Table 4, R1 and R3) and the precision (Table 4, R2 and R4) and R4). Precision was considerably improved at 1000  C, and this
of the method decline. Consequently, the value to be used for this temperature should therefore be used in the optimization phase.
factor should lie at the lower end of the experimental range tested What was a factor in the screening phase thus becomes a constant in
(Table 1). Clearly, during the optimization phase, the value of X5 the optimization phase. At 800 C, the combustion of N-steam
should be kept at the 1 level (Table 1, 7 ml), so that factor X5 generating compounds was ensured, higher temperatures enabled
becomes a constant in the optimization phase. The presence of an greater and better combustion of other substances whose presence in
excessive volume of O2 during sample combustion leads to (i) the combustion chamber and subsequent presence in the reactor
greater oxidation of the reactor reducing agent (Cu) and a short- leads to considerable background noise and hence to loss of preci-
ening of its useful life, (ii) the appearance of partially-overlapping sion. For larger sample sizes (700e1300 mg), and thus for larger
peaks, preventing their adequate integration, with the result that amounts of interfering gaseous substances reported in the literature
the measured amount of N is higher than the real amount, and (iii) (Saint-Denis and Goupy, 2004), it has even been suggested that
an increase in background noise and thus a loss of precision. temperatures of up to 1150  C be used. When using a nitrogen
Saint-Denis and Goupy (2004) failed to consider O2 volume as a analyzer based on the DM, the intensive steam production during
possible critical factor e since it was not included as a factor in the injection may result in poor nitrogen recoveries for volatile nitrogen
experiment design used e so that use of an excessive oxygen vol- compounds such as ethylene diamine (Jakob et al., 1995). By contrast,
ume (not shown by the authors) might have contributed to the poor heavy samples and low furnace temperatures are associated with a
precision they reported in comparison to the KM. Despite the very high number of low-end outliers (Saint-Denis and Goupy, 2004).
critical role of oxygen volume in the acceptability of the analytical Similar temperatures to those recommended in this paper (1000  C)
result obtained, a role highlighted in the present study, many au- have been used by other authors, including 950  C (AOAC, 1990;
thors fail to optimize this parameter in accordance with the sample Krotz et al., 2008; Beljkas et al., 2010) and, in some cases, 870 C
analyzed (Daun and DeClercq, 1994; Nozawa et al., 2007; Beljkas (Horwitz, 2003) or even 850  C (Daun and DeClercq, 1994).
et al., 2010; Kubota et al., 2011) opting simply to follow manufac- Slightly different results were obtained for rear furnace tem-
turers’ instructions for any sample type. The results obtained here perature (X10) and GC oven temperature (X11) (Table 1). The in-
underline the fact that the generic instructions provided by the crease in values for X10 (Table 4) was accompanied by a decrease of
manufacturer are not the most suitable for all sample types; rather, the difference with respect to certified values from 1.37 to 0.27 for
S. Serrano et al. / Journal of Cereal Science 58 (2013) 31e36 35

rye and from 2.88 to 0.5 for wheat. In other words, greater accuracy 1,41
is obtained at the þ1 level. Since the oven-reactor temperature 0,1 0,00 A
should not be kept constantly at 1000  C, two stand-by tempera- 1,00
tures were tested (500  C and 750  C; Table 1). It was found that 0,02
when a stand-by temperature of 500  C was used, the values ob-
0,04
tained for the first sample were lower than real values, probably
due to insufficient combustion of that sample. As a result, the mean 0,2 0,06
value for the four replicates decreased, prompting a loss of accu-

11
racy. A few minutes should therefore be allowed to elapse before 0,00

X
the sample is inserted, even though the instrument may signal that 0,08 0,1
the initial set temperature has been reached. A stand-by temper-
ature of 600  C has been used by other authors (Nozawa et al.,
2007) with satisfactory results in other types of sample (soy sauce). 0,0
Similarly, better results were obtained for R2 (Table 4) at -1,00 -0,1
level þ1; indeed, precision was improved 6.8-fold for rye and 7.7-
fold for wheat with respect to level 1. The effects of factor X11 were -0,2
-1,41
the reverse of those recorded for X10 (Table 4). The smallest dif- -1,41 -1,00 0,00 1,00 1,41
ference with respect to certified values was obtained at level 1 X10
(Table 4, R1 and R3), and a lower value for SD (Table 4, R2 and R4)
was also obtained at this level. Gases leaving the oven-reactor pass 1,41
into a chromatography column, whose temperature has a crucial
impact on the shape and position of the peaks obtained. Inadequate 1,00 0,00
B
integration and possible overlapping of these peaks might account
for the effects obtained. 0,2
0,02
The difference between factors X10 and X11 and the other factors 0,1
examined earlier is that their effect on detected nitrogen levels in 0,0
the sample appears ostensibly to be a quadratic rather than a linear
11

0,00
effect, as shown in Fig. 1. The presence of expected levels (certified 0,04
X

values) amongst the levels obtained points to the existence of -0,1


quadratic effects for both factors, which cannot be reliably detected
using a PlacketteBurman experimental design. Both these factors
should therefore be carried forward into the optimization phase, -0,2
0,06
using a central composite design, as outlined under Material and -1,00
Methods and detailed in Table 3. The results of this experimental
design enabled the optimization of conditions for parameters X10 -1,41
and X11 in the case of rye (Fig. 2A) at 870  C and 92  C for X10 and X11, -1,41 -1,00 0,00 1,00 1,41
respectively, and in the case of wheat (Fig. 2B) at as 720  C and X10
95  C, for X10 and X11, respectively. Beljkas et al. (2010) used a rear
furnace temperature of 850  C for N determination in wheat sam- Fig. 2. Optimization graphics. Quadratic effect detected for rye (A) and wheat (B), and
identification of system stationary point.
ples. Although they did not implement a fully-fledged optimization
process, simply following manufacturer’s instructions, this tem-
perature recommendation would agree with the results obtained In order to validate the conditions suggested here for the use of
here for rye, but not with those obtained for wheat. This highlights the DM for protein measurement in cereals, five measurements
the need to optimize the analytical process for each different were performed for wheat (2.1067, 2.1253, 2.1254, 2.1365 and
sample type. Various combinations of these factors have been used 2.1162, mean 2.122 e real values 2.12%- and SD 0.011) and rye
by other authors, including 840  C and 50  C for X11 and X12, (1.2534, 1.2728, 1.2655, 1.2685 and 1.2533, mean 1.262 ereal value
respectively (Krotz et al., 2008), and 850  C (Beljkas et al., 2010) or 1.249%- and SD 0.009).
600 C for X11 (Nozawa et al., 2007). For the measurement of crude protein in grains and oilseeds using
the combustion method, standard deviations for repeatability and
reproducibility are reported to range from 0.10 to 0.37 and from 0.25
to 0.54, respectively, while relative standard deviations for repeat-
Rye Wheat ability and reproducibility range from 0.77 to 2.57% and from 1.24 to
= certified value: 1.249 % N = certified value: 2.12 % N 3.15%, respectively (Bicsak, 1993). In a more recent review, Moore
et al. (2010) reported that intra- and interlaboratory relative stan-
-1 +1 -1 +1
dard deviations for different matrices range from 0.30% to 8.3% and

X10 0.62e8.2%, respectively. In general, a significantly lower analytical


precision has been reported (Sweeny and Rexroad, 1987) for the
2.40 N level detected 1.03 4.75 N level detected 1.84
combustion method (0.013e0.052%) as compared with KM (0.006e
0.035%). Similar results are noted by Schmitter and Rihs (1989). The
-1 +1 -1 +1
results obtained in the present study suggest that the poorer preci-

X11 sion of DM compared to KM reported in the literature may be


attributable to the use of non-optimized instrumental conditions.
1.10 N level detected 2.33 1.94 N level detected 4.64
Beljkas et al. (2010) concluded that since the DM procedure is
Fig. 1. N levels detected as a function of X10 and X11 factor levels, and relationship with very simple and almost completely automated, there are relatively
certified value for rye and wheat samples (BCR-381 and BCR-382, respectively). few sources of uncertainty compared to the KM. The present results
36 S. Serrano et al. / Journal of Cereal Science 58 (2013) 31e36

show that edespite fewer sources of uncertainty than in the KM e Daun, J.K., DeClercq, D.R., 1994. Comparison of combustion and Kjeldhal Methods
for determination of nitrogen in Oilseeds. Journal of the American Oil Chemists’
the failure to use instrumental conditions perfectly optimized for
Society 71 (10), 1069e1072.
each sample type may constitute a major source of error when Horwitz, W., 2003. The certainty of uncertainty. Journal of AOAC International 86
implementing the DM. (1), 109e111.
This study confirmed that measurement of nitrogen content Jakob, E., Sievert, C., Sommer, S., Puhan, Z., 1995. Automated determination of total
nitrogen in milk by Dumas method. Zeitschrift für Lebensmittel-Untersuchung
using the DM requires the prior optimization of analytical condi- und eForschung 200 (4), 239e243.
tions for each sample type, rather than simple reliance on the Jiménez, R.R., Ladha, J.K., 1993. Automated elemental analysis: a rapid and reliable but
general recommendations provided by the equipment manufac- expensive measurement of total carbon and nitrogen in plant and soil samples.
Communications in Soil Science and Plant Analysis 24 (15e16), 1897e1924.
turer. Use of non-optimized conditions may be a significant cause of Joglekar, A.M., May, A.T., 1987. Product excellence through design of experiments.
error in analytical measurements. Cereal Foods World 32, 857e868.
In this study, application of a non-geometric PlacketteBurman Kjeldahl, J., 1883. A new method for the determination of nitrogen in organic
matter. Zeitschrift fur Analytische Chemie 22, 366e382.
experimental design enabled the identification of two critical fac- Krotz, L., Ciceri, E., Giazzi, G., August/September 2008. Protein determination in
tors showing quadratic effects: RFT (rear furnace temperature) and cereals and seeds. New combustion methods and technologies eliminate sam-
GCT (GC oven temperatures). Accordingly, an optimization process ple preparation. Food Quality Magazine, 9e12.
Kubota, T., Oshida, T., Yanai, K., Inoue, Y., 2011. Improvement of the conditions for
was performed, based on a central composite design in order to the determination of total nitrogen in fish meal in Kjeldahl method and its
obtain perfectly-optimized conditions for the implementation of comparison with Dumas Method. Bunseki Kagaku 60 (1), 67e74.
the DM. Ma, T., Rittner, R.C., 1979. Modern Organic Elemental Analysis. Marcel Dekker, New
York.
Moore, J.C., DeVries, J.W., Lipp, M., Griffiths, J.C., Abernethy, D.R., 2010. Total
Appendix A. Supplementary data protein methods and their potential utility to reduce the risk of food protein
adulteration. Comprehensive Reviews in Food Sciences and Food Safety 9,
330e357.
Supplementary data related to this article can be found at http:// Myers, R.H., Montgomery, D.C., 1995. Introduction to response surface methodol-
dx.doi.org/10.1016/j.jcs.2013.04.006. ogy. In: Response Surface Methodology. Process and Product Optimization
Using Designed Experiments. John Wiley & Sons Inc, New York, pp. 1e16.
Nozawa, S., Kasama, H., Suzuki, T., Yasui, A., 2007. Application of improved Dumas
References method to the determination of total nitrogen in soy sauce. Bunseki Kagaku 56
(3), 179e183.
AOAC, 1990. Official Methods of Analysis, 4th Supplement (1993). 992.23 Method, Plackett, R.L., Burman, J.P., 1946. The design of optimum multifactor experiments.
15th ed. AOAC International, Arlington, Virginia. Biometrika 33, 305e325.
Beljkas, B., Matic, J., Milovanovic, I., Jovanov, P., Misan, A., Saric, L., 2010. Rapid Saint-Denis, T., Goupy, J., 2004. Optimization of a nitrogen analyzer based on the
method for determination of protein content in cereals and oilseeds: validation, Dumas method. Analytica Chimica Acta 515, 191e198.
measurement uncertainty and comparison with the Kjeldhal method. Accred- SAS, 1999. The RSREG procedure. In: SAS/SAT User Guide, fourth ed., vol 2. SAS
itation and Quality Assurance 15, 555e561. Institute Inc, Cary, NC 27513 (USA), pp. 1457e1478.
Belliardo, J.J., Wagstaffe, P.J., 1988. BCR reference materials food and agricultural anal- Schmitter, B.M., Rihs, T., 1989. Evaluation of a macro-combustion method for total
ysis: and overview. Fresenius Zeitschrift für Analytische Chemie 332, 533e538. nitrogen determination in feedstuffs. Journal of Agricultural and Food Chem-
Bellomonte, G., Costantini, A., Giammarioli, S., 1987. Comparison of modified istry 37, 992e994.
automatic Dumas method and the traditional Kjeldhal method for nitrogen Sebranek, J.G., Cassens, R.G., 1973. Partial recovery of nitrite nitrogen by the Kjeldahl
determination in infant food. Journal of the Association of Official Analytical procedure in meat products. Journal of Food Sciences 38, 1085e1086.
Chemists 70 (2), 227e229. Sweeny, R.A., Rexroad, P.R., 1987. Comparison of LECO FP-228 nitrogen determinator
Bicsak, R.C., 1993. Comparison of Kjeldahl method for determination of crude with AOAC copper catalyst Kjeldahl method for crude protein. Journal of the
protein in cereal-grains and oilseeds with generic combustion method. Journal Association of Official Analytical Chemists 70, 1028e1030.
of AOAC International 76 (4), 780e786. Thompson, M., Owen, L., Wilkinson, K., Wood, R., Damant, A., 2002. A comparison of
Box, G.E.P., Hunter, J.S., 1961. The 2k-p fractional factorial designs: part I. Techno- the Kjeldahl and Dumas methods for the determination of protein in foods,
metrics 3, 311e351. using data from a proficiency testing scheme. Analyst 127, 1666e1668.
Box, G.E.P., Wilson, K.B., 1951. On the experimental attainment of optimum condi- Williams, P., Sobering, D., Antoniszyn, J., 1998. Protein testing methods at the Ca-
tions. Journal of the Royal Statistical Society Series B 13, 1e15. nadian grain Commission. In: Wheat Protein Symposium, Saskatoon, Canada,
Chiacchierini, E., Dáscenzo, F., Restuccia, D., Vinci, G., 2003. Milk soluble whey March 9e10.
proteins: fast and precise determination with Dumas method. Analytical Letters Youden, W.J., Steiner, E.H., 1975. Statistical Manual of the Association of Official
36 (11), 2473e2484. Analytical Chemist. The Association of Official Analytical Chemist, Washington, DC.

You might also like