Experimental design for optimizing MALDI-TOF-MS analysis of palladium complexes

This paper presents optimization of matrix-assisted laser desorption/ionization (MALDI) time-of-flight (TOF) mass spectrometer (MS) instrumental parameters for the analysis of chloro(2,2',2"-terpyridine)palladium(II) chloride dihydrate complex applying design of experiments methodology (DoE). This complex is of interest for potential use in the cancer therapy. DoE methodology was proved to succeed in optimization of many complex analytical problems. However, it has been poorly used for MALDI-TOF-MS optimization up to now. The theoretical mathematical relationships which explain the influence of important experimental factors (laser energy, grid voltage and number of laser shots) on the selected responses (signal to noise – S/N ratio and the resolution – R of the leading peak) is established. The optimal instrumental settings providing maximal S/N and R are identified and experimentally verified.

The quality of the mass spectra obtained by matrixassisted laser desorption/ionization (MALDI) time--of-flight (TOF) is under strong influence of several key factors, such as the type and concentration of both matrix and the analyte, their mole ratio, technique used for the sample deposition (dried droplet, sandwich, double sandwich or thin layer) [1].However, different spectra quality can be obtained for the fixed combination of the matrix/analyte molar ratio and technique of sample deposition by adjustment of instrument parameters.Experimentally, the influence of instrumental parameters can be studied by varying (increasing/decreasing) one parameter while the remaining settings are on the fixed level.This approach has many disadvantages: the information about the spectra quality is obtained only for a few parameter settings, while there is no data about the system behavior for the settings between investigated points.Moreover, the potential interaction of the factors cannot be seen since they are not varied simultaneously.
Design of experiments (DoE) methodology provides systematic approach for the investigation of instrumental parameters influence on the spectra quality and identification of the settings that provide the most reliable data [2,3].To date, the DoE methodology has been applied to the optimization of chemical factors during the development of analytical methods, such as, chromatographic [4][5][6][7] electroanalytical [8][9][10][11], thermogravimetry [12], and liquid-liquid extraction [13,14].Additionally, the DoE is used for the optimization of instrumental parameters of equipment for analysis at graphite furnace atomic absorption spectrometer (GF AAS) [15,16], inductively coupled plasma optical emission spectrometer (ICP OES) [17,18], inductively coupled plasma mass spectrometer (ICP-MS) [19], and matrix-assisted laser desorption/ionization time-of--flight mass spectrometer (MALDI-TOF-MS) [1,[20][21][22][23][24][25].This methodology involves performing of small number of carefully planned experiments, which allows the establishment of mathematical relationship between investigated factors and important system responses.In that way, the influence of each parameter is quantified, and created relationship allows theoretical examination of experimental space searching for the optimal solutions, without the necessity for performing additional experiments.
In this work, DoE methodology is applied for the optimization of MALDI-TOF instrumental parameters for the analysis of one of the potential metallo-drugs, i.e., [Pd(terpy)Cl]Cl .2H 2 O, the structure of which is given in Figure 1.DoE methodology and related chemometrical strategies are poorly used for MALDI-TOF-MS optimization up to now [1,[20][21][22][23][24][25], and to the best of our knowledge this is the first time that this approach is applied for the MALDI MS investigation of transition metal complexes.
Matrix for MALDI-TOF-MS was (5,10,15,20-tetrakis-(pentafluorophenyl)porphyrin, MTTP.This matrix, methanol and chloroform were purchased from Sigma Aldrich (Taufkirchen, Germany) and were used without further purification.Working solution of Pd-complex was prepared dissolving in pure methanol in order to obtain the concentration of 1 mg mL -1 .

Equipment
All mass spectra were acquired in positive reflector mode on Voyager Biospectrometry DE Pro Workstation (PerSeptive Biosystems, Framingham, MA, USA) MALDI--TOF mass spectrometer.The mass spectrometer was equipped with a pulsed nitrogen laser operated at 337 nm with 3 ns duration pulses, and with 20.0 Hz laser repetition rate.The laser intensity (laser operates in the range from 750 to 3000 W/cm 2 ) and number of laser shots must be adjusted.Time of flight mass analyzers was operated on the delayed-extraction (DE) mode.Extraction delay time (time between laser flash and ion extraction) may be in the range 25-1000 ns.In this experiment in all cases the delay time of 100 ns was used.An accelerating voltage is applied to the sample plate in the range 15-25 kV.Besides, a grid 1-2 mm above the sample plate exists with an additional voltage to fine-tune ion acceleration; this voltage is called grid voltage.The value of grid voltage is expressed as a certain percentage of the accelerating voltage.

Acquisition of MALDI-TOF spectra
Spectra of the complex was obtained by applying 0.5 µL of working solution onto the MALDI sample plate, followed by the addition of 0.5 µL MTTP solution (prepared dissolving MTTP in chloroform in order to obtain concentration of 10 mg mL -1 ).Samples were left to evaporate and co-crystallize at room temperature.Spectra were calibrated by setting the peak of the 2,5--dihydroxybenzoic acid to its appropriate value (155.034Da).

Software
Box-Behnken design (BBD) was selected like an appropriate experimental design in DoE methodology for optimization of MALDI-TOF-MS instrumental parameters.The BBD was performed using Design-Expert® 7.0.0software (Stat-Ease Inc., Minneapolis, MN, USA).Three-dimensional (3D) response surfaces were obtained by using Statistica 7.For grid search point calculation, MATLAB was used.

RESULTS AND DISCUSSION
In the first phase of the research, preliminary experiments allowed the selection of MTTP matrix as a suitable for the analysis of chloro(2,2',2"-terpyridine)palladium(II) chloride dihydrate [27].The leading peak with the most intense signal is identified and its position was m/z 376.1, which corresponds to the complex with a loss of one Cl -ion (discussed later in text).Two responses are selected as relevant for the quality of the obtained spectra: signal to noise ratio (S/N signifies the level of a desired signal to the level of background noise) and resolution (R represents the ability of a mass spectrometer to distinguish between the ions of different m/z ratios).Since palladium metal complexes have higher number of isotopes, achieved resolution is of particular importance.For instance, a signal that arises from tested complex is composed of 9 individual peaks with various intensities [27], which correspond to theoretical presentation of the same spectra (data not shown).Thus, the optimization aim was the simultaneous maximization of both S/N and R. Due to the well-known variability of MALDI-TOF-MS spectra which interfere with adequate quantification, the coefficient of variation (CV) obtained for triplicate of identical combination complex-matrix-instrumental parameters was also monitored.
It is well known that variety of instrumental parameters can affect the spectra quality.For this reason, in the second phase, optimal values of input variables, namely, the laser intensity, the grid voltage and the number of laser shots were determined using DoE methodology in order to obtain the best values of S/N and R of the peak of interest within the range defined previously.The values of input variables are defined according to the earlier experience, that is, on the basis of the results obtained in preliminary experiments (Table 1).The DoE methods are often run at different values of input variables (or factors), which are called levels.Each level is placed at one of three equally spaced values usually coded as -1, 0, +1 (Table 1).Each run of a DoE method involves a combination of the levels of the selected factors.After the identification of input factors and responses of interest (S/N and R) the experimental plan is created following Box-Behnken design [28].This design is based on three-level incomplete factorial designs where two factors are arranged in a full two-level design, while the level of the third factor is set at zero (Figure 2).Points on the diagram represent the experimental runs that are done.It is nearly rotatable and especially useful when the points on one or more corners of the cube represent factor-level combinations that are for any reason unacceptable.Since Box-Behnken design investigates each factor on three levels, the definition of quadratic dependence of responses on investigated factors is enabled.For the investigation of three factors, twelve experiments plus central point replications (in this case three replications are selected) are required.The experimental plan defined by the applied design is presented in Table 2.
where y 1 is the response S/N and x i investigated factors (Table 1).The same equation for real factor values has the following form: The mathematical model in terms of coded factor values for the resolution is created in the following way: Where y 2 is the response R and x i investigated factors (Table 1).
The same equation for real factor values could be written as: The p-value is the basic for interpreting statistical results.The response models are created removing model terms whose coefficients had p value greater than 0.1, since they are considered unimportant.Statistical adequacy of the obtained relationship is confirmed by high values of coefficient of determination (>0.94), adjusted coefficient of determination (>0.89) and non-significant lack of fit test.The absolute value of the obtained coefficients in coded factor values explains the intensity of particular factor or interaction influence.Generally the factors also might have "interact", for example, a two-factor interaction occurs when the influence of one factor on the response is different at different levels of the second factor.Taking into account obtained results on basic of p-value, and also, keeping in the mind that certain factors have significant influence if p < 0.05, it can be concluded the following: the greatest influence on the signal to noise ratio exhibited laser intensity (both linear and quadratic term had p value < 0.0001), followed by the interaction between laser intensity and grid voltage (p = 0.02).Grid voltage by itself exhibited lower influence (p value 0.049 for linear and 0.09 for quadratic term), while the impact of number of laser shoots was characterized by p value around 0.06.In the case of response R, it is shown that number of laser shots does not have significant influence, while the linear terms of the remaining two factors appeared to be significant (p value < 0.001).Moreover, the interaction between laser shots and grid voltage had borderline significance (p = 0.1).Response surface plots enable simpler visualization of investigated factors influence and they are presented in Figure 3.
The response, S/N, exhibits quadratic dependence on laser intensity and number of laser shots.It can be seen on Figure 3 that the saddle point of both quadratic functions was within the investigated region, indicating that increasing these factors up to the certain level, the value of S/N increases, while the increase of these factors after the saddle point leads to decrease in S/N.On the other hand, the increase of number of laser shots leads to the moderate linear increase of S/N.It is important to notice the significance of laser shots and As far as resolution is concerned, the linear influence of laser intensity and grid voltage and their interaction can be seen in Figure 3, as well.
As far as the coefficient of variation (CV, a statistical measure of the dispersion of data points in a data series around the mean) is concerned, it is estimated for both S/N and R.However, statistical analysis has shown that the overall mean is better prediction of both CV than the mathematical models involving examined factors.Therefore, the system variability was not under control of investigated instrument parameters.
Response surface plots for S/N and R presented in Figure 3 can be used for identification of the optimal instrumental settings where maximal S/N and maximal resolution could be achieved.However, it can be seen that settings providing maximal S/N are different from those providing maximal R. Therefore, in order to resolve such multiobjective optimization problem the priority is given to S/N and the solution is searched as the point where maximal S/N value with satisfactory R value could be found.In this study, among several potentially adequate solutions, the following settings are chosen: laser intensity -1220 W/cm 2 , grid voltage -79% and number of laser shots -360.
The theoretical value of responses in this point is estimated by created mathematical models to be 5938 for S/N and 2368 for R. The experimental verification of this settings is performed in five replicates and one of the obtained spectra is presented in Figure 4.
Positive ion MALDI TOF mass spectrum of [Pd(terpy)Cl]Cl⋅2H 2 O complex is dominated by the signal at m/z 376.1, that corresponds to the cation [Pd(terpy)Cl] + .Much smaller signal that arises from [Pd(terpy)] + can be detected at m/z 340.7.Other signals, arising from matrix are indicated by an asterisk.Inset of the Figure 4 represents the expanded region of the signal at m/z 376.1, which is, in fact, composed of several isotopes (m/z indicated in the Figure 4).The average experimentally obtained value of signal to noise ratio was 5850, while the average value of resolution was 2456.The deviation of experimental S/N value from theoretical was 1.5%, while the deviation of R was calculated to be 3.7%.Moreover, the obtained signal to noise ratio in located optimal point is significantlly greater than all S/N values obtained for instrument settings combinations presented in Table 2. Therefore, the DoE methodology proved to be valuable assistance in optimization of MALDI-TOF-MS instrumental settings for the analysis of Pd metal complex.

Figure 1 .
Figure 1.Chemical structure of analyzed Pd metal complex.

Figure 2 .
Figure 2. Schematic presentation of Box-Behnken experimental matrix for three factors.Experiments are performed randomly and S/N and R values are obtained.Each factors combination is investigated in triplicate and the values of coefficient of variation (CV) are calculated.Multiple linear regression and least squares methods of approximation are used to create mathematical relationship between investigated the input factors and the responses.In the case of S/N response, the following equation in terms of coded factor values is obtained: equation for real factor values enables easy navigation of experimental space and calculation of the response by simple inserting of factors values.However, coefficients in this equation are influenced by different scales of factor values.Therefore, differences in magnitude of these coefficients cannot be used for factors comparison.On the other hand, equation in terms of coded factor levels enables comparison of factors influence on the response based on coefficient in front of each term.The larger the coefficient is, the greater influence will exhibit the factor.

Figure 3 .
Figure 3. Response surface plots presenting the dependence A) S/N = f (x 1 , x 2 ); B) S/N = f (x 1 , x 3 ); C) S/N = f (x 2 ,x 3 ); D) R = f (x 1 , x 2 ), where x i are the input variables or factors: x 1 is laser intensity, x 2 is grid voltage, and x 3 is number of laser shoots.

Figure 4 .
Figure 4. Positive ion MALDI-TOF mass spectrum of Pd metal complex obtained for optimal instrumental settings.Signals are indicated by their position, whereas matrix signals are indicated by an asterisk.Inset represents expanded m/z region from 368 to 380.To illustrate mass resolution achieved under experimental conditions.

Table 1 .
Investigated instrumental parameters (three input variables) and their levels

Table 2 .
Experimental plan according to Box-Behnken experimental design