Analytical methodologies applied to CDM

respectively. Data were acquired using MassHunter B.06.00 software. The mass spectrometer was operated in positive dynamic multiple reaction monitoring (dMRM) mode and set to unit resolution. MS/MS product ion spectra for each analyte were recorded using direct injection of single compound references into the ESI source. Reference compounds were dissolved in a mixture of water/methanol 50:50 (v/v). The cell accelerator voltage (CAV) was set to 7 V, while fragmentation voltage (UF) and collision energy (CE) were optimized for each transition. Retention times, transition pairs and MS/MS acquisition parameters for each analyte are shown in Table 3.

Table 13 Chromatographic gradient for dMRM method. Mobile phase A was 5 mM ammonium formate in water and mobile phase B was a 80:20 methanol/water (v/v) mixture with 0.025% v/v formic acid. Maximum column back pressure was set to 250 bar.

Method validation In order to investigate the developed LC-QqQ-MS method in more detail a validation after scientific quality standards has been conducted. If possible, all the described method validation parameters were performed according to the FDA guidelines for industry.⁵⁵⁸

Selectivity Method interferences were evaluated through signal analysis with SNR≥3 as cut-off criterion for artificial matrix analysis. Based on the high number of compounds included in the method (≥ 50% coverage of CDM compounds) preparation of blank matrix for individual compounds was not feasible. Matrix effect assessment was thus performed by comparing the reference stock solution spiked into the artificial medium matrix and into water (1:2 dilution). This was done for solutions at 5 different concentration levels (1:10, 1:50, 1:100, 1:500 and 1:100). The norm of the correlation matrix and the percent difference in absolute values was used to estimate the impact of the matrix effect. With differences below 30% and correlation norms ≤0.98, an impact of the matrix was rejected. It has to be noted that ion suppression caused by coeluting analytes was not investigated.

Calibration curves Calibration curves for each component were obtained by analyzing freshly thawed aliquots of the reference stock solution at 15 dilution levels (1:1.5:1.33:1.25:1.4:

1.28:1.38:1.48:1.35:1.5:1.33:2:2.5:2:2). Serial dilution of the stock was performed in water. Based on the measurement time and the variety of analyte concentrations in cell culture medium, multiple dilution levels ensuring linear calibration curves were not feasible for all compounds. To

time [min]

mobile phase A [vol.% ]

mobile phase B [vol.%]

flow rate [mL/min]

0 95.0 5.0 0.18

4 77.5 22.5 0.18

5 74.5 25.5 0.18

6 71.5 28.5 0.18

7 68.5 31.5 0.18

15 49.3 50.7 0.18

30 0.0 100.0 0.18

32 0.0 100.0 0.18

33 95.0 5.0 0.18

36 95.0 5.0 0.80

40 95.0 5.0 0.80

40.5 95.0 5.0 0.18

solve this problem, either quadratic or power regression models were used. For regression model decision, comparison of R² values of forced linear fit and heteroscedasticity test was performed.

The chi-square distribution test calculated after Breush, Pagan and Koenker in Matlab was used to determine if heteroscedasticity affected the calibration data.¹ If the calculated pValue was

≤0.05, the null hypothesis of homoscedasticity was rejected and heteroscedasticity assumed. For low R square (forced linear fit) and distinctively higher R square for power or quadratic fit linear regression was rejected.

Accuracy and Precision Method accuracy was assessed through comparison of measured analyte concentrations in QC samples with known concentration levels of model medium 2. The measurement was considered accurate when deviations of ≤ 15% between known and measured analyte concentrations could be demonstrated. For inter-batch comparability, accuracy of QC samples was determined for each batch and plotted on a Xbar-R chart. Intra- and inter-day precision was evaluated as coefficient of variation (CV). For this purpose, compound concentrations in QC samples were determined over 6 days with three replicates per day. Sample comparability was ensured through single sample preparation (aliquots frozen at -80°C) and aliquot thawing just prior to measurements. The method was considered precise when CV was ≤5%.

Sensitivity and carryover The lower limit of quantification (LLOQ) was determined according to the mathematical equation 𝐿𝐿𝑂𝑄 = 10 ×^𝑠

𝑏, where b depicts the calibration graph slope and s the standard deviation of intercept. For limit of detection (LOD) determination, 𝐿𝑂𝐷 = 3.3 ×^𝑠

𝑏

was assumed.⁵⁵⁶ Logarithmic transformation of the calibration curve was performed for compounds with power regression type. For quadratic regression, SNR ≥3.3 and SNR ≥10 were set as criteria for LOD and LLOQ respectively. Carryover was assessed by comparison of analyte peak areas of blank injections measured after and before QC samples and reported as the difference relative to peak area in replicate QC samples.

Reproducibility and stability Reproducibility was determined by QC sample incorporation within each measurement batch. Coefficients of variation below or equal to 20% were considered acceptable. Furthermore, compound stabilities within the measurement time were examined for both QC and calibration curve samples.

6.3.3 FTIR spectroscopy

Fourier transform infrared microspectroscopy (FTIR) analysis was performed in two collaborating laboratories. MVA incorporated used an Olympus BX-51 compound microscope coupled to a SensIR IlluminatIR FTIR spectrophotometer.

Intertek AG in Switzerland used an Attenuated Total Reflectance (ATR) – FTIR approach. The IR spectra of the samples were recorded on a Varian 3100 FT-IR Spectrometer.

6.3.4 Raman Microscopy

All the precipitate analyzed were filtered on mixed cellulose ester filter membranes. Since the layer of precipitate was usually too thin to be scratched off the membranes the dried solid precipitate has been analyzed directly on the filter membrane. Having the material of interest in a pure layer would have been advantageous for the Raman analysis. In order to be able to distinguish signal of the filter membrane from the precipitate, a pure membrane has been measured on an aluminum foil wrapped slide (Figure 62). All slides used have been wrapped in aluminum foil in order to prevent the glass to mask the Raman signal of the sample. In a first step the sample was focused with either a 50x or 100x objective. All results shown were measured with

the 532 nm wavelength laser. For CDM precipitate samples the laser power was generally set between 0.2 and 2 mW to prevent burning. Standard settings for slit width and pinhole were 50 µm. Investigative analysis as baseline subtraction and data base searches have been conducted in the Thermo software used to control the instrument (Omnic^TM Spectra^TM). After raw data acquisition and matching to Thermo data base the spectra have been exported in

*.csv format. Plotting and analysis was done later with GraphPad Prism 7.03 software.

Figure 62: Blank filter membrane measured on aluminum foil wrapped slide. The reference spectrum was used to judge if sample spectra were impacted by mixed cellulose ester background.

6.3.5 Scanning electron microscopy – energy dispersive x-ray (SEM-EDX)

Scanning electron microscopy coupled with energy dispersive x-ray detector (SEM-EDX) was performed in a collaborating laboratory within BI. The samples were prepared by excising little pieces of the filter membrane with the dried filter cake. Subsequently, the little pieces were mounted on aluminum sample holders with sticky coal pads. The samples were first investigated with the high-resolution secondary electrons detector (SE). The microscopic investigations with the SE-detector gave good knowledge of the sample topology and the fine-structural composition.

In a second step, the sample was scanned with a back-scattered-electrons detector (BSE). The BSE-detector type allows conclusions on elemental composition because elements with higher order number reflect electrons more efficiently due to their bigger nuclei than elements with lower order number. Therefore, elements with lower order number appear darker in the image.

BSE-detector is marked in the images of the results and discussion part as angular selective backscattered (AsB). Both SE-detector and BSE-detector images are shown as results.

BSE-detector images are coupled with K-series images highlighting local hotspots of specific elements.

The EDX analysis has been performed on local, representative fields marked in the AsB images with white rectangles. The exciting electron beam causes the elements with elemental order number ≥5 to emit x-rays with element specific energy [keV]. The EDX measurements with high local resolution have been conducted with a process time of 5 and an acceleration voltage of 10-25 kV. For the interpretation of EDX spectra the elements of the sample holder have to be considered as background.

6.3.6 Inductively coupled plasma mass spectrometry (ICP-MS) and inductively coupled plasma optical emission spectroscopy (ICP-OES)

All the ICP-MS and ICP-OES analysis have been conducted externally at the contract research organization Solvias AG in Kaiseraugust. The precipitate on filter membrane was dissolved with acid (HCl, H3PO4 or HNO3). The dissolution has been conducted in quartz glass vessels with PTFE caps that were carefully rinsed with acid and water. Afterwards, the sample material was dissolved in digestion apparatus with high pressure autoclave and microwave heating. The measurement has been conducted with settings proprietary to Solvias AG. The data has been provided in elemental concentrations in mg/kg or w/w.

6.3.7 Conditions for radioactivity measurements – liquid scintillation counting (LSC)

All samples were determined for 14C-radioactivity using a LS 6500 multi-purpose scintillation counter (Beckman Coulter). The sample aliquots were transferred into a mini-Vial and filled up with 4.5 ml LSC-cocktail (Ultima Gold).

All parameters for the measurement were documented on the protocol and were as follows:

Count conditions:

Nuclide C-14

Count time 2 min Count unit dpm Quench set:

Quench limits (low) 29.062 Quench limits (high) 198.56 Background subtract Off Low CPM threshold Off Count region 0.0 – 156.0 kEV Count corrections:

Color quench correction On

The samples were measured in the C-14 protocol because of the lack of S-35 LSC-Standards. The energy and the counting yield of these isotopes are nearly the same (C-14: 96%, S-35: 97%).