*3.1. TOM Models with Sampling Port*

The TOM models reproducibly showed an unstructured and hyperproliferative epithelial layer with pleomorphic tumor cells, also separating from the epithelial layer into the lamina propria. Neither the sampling port nor the cultivation within the autosampler of the UHPLC-MS/MS device influenced the tumor growth. The effects of docetaxel on the tumor size in TOM models by supplementing the differentiation medium with either two drug doses or the vehicle control were determined. Whereas the vehicle control did not change the tumor morphology (Figure 1b), docetaxel caused a dose-dependent reduction of tumor size with abundant epithelial cell death (Figure 1c). The average tumor size declined from 347 ± 72 µm (untreated) to 100 ± 45 µm (max docetaxel concentration, *n* = 4 each).

#### *3.2. Docetaxel Epimerization and Degradation Products*

During electrospray ionization, docetaxel mainly forms a sodium adduct ([M+ Na]<sup>+</sup> theor=830.3358), which is used as precursor ion for all MS/MS experiments. As shown in Figure 2 (top), the product ion spectrum of docetaxel shows three major fragments at *m*/*z* 549.2095 (taxane nucleus (10-deacetylbaccatin III, 10DABIII), [C29H34O<sup>9</sup> + Na]+, exact mass *m*/*z* 549.2095, mass error ∆*m*/*z* = 0 ppm), *m*/*z* 304.1159 (phenylpropionic acid side chain, [C14H19NO+Na]+, exact mass *m*/*z* 304.1155, ∆*m*/*z* = −1.17 ppm), and *m*/*z* 248.0537 (side chain with loss of the tert-butyl moiety, [C10H11NO<sup>5</sup> + Na]+, exact mass *m*/*z* 248.0529, ∆*m*/*z* = −3.05 ppm). The two main fragments *m*/*z* 549. 1 and *m*/*z* 304.1 were later chosen for MRM transitions in real-time quantitation (method A).

**Figure 2.** Docetaxel structure fragmentation. (**a**) Chemical structure of docetaxel and main fragmentation products. (**b**) Product ion spectra of docetaxel (top) and its potential 7-epimer (bottom), precursor [M + Na]<sup>+</sup> theor = 830.3358 indicated with black rhombus.

Analyses of docetaxel reference substance as well as cell culture media without cells, and TOM models, revealed a second peak with almost identical MRM transitions (method A, chromatogram showing transitions in Figure S1) and product ions (method B, Figure 2 (bottom)). It already appeared only minutes after preparing the samples for analysis with serum-free medium as sample diluent. This degradation product is postulated to be the 7-epimer of docetaxel (epi-docetaxel), which is known to occur in basic and acidic conditions [18,19]. Since the epimerization could not be avoided in calibration or quality control samples as well, the combined peak areas of docetaxel and the 7-epimer were considered for all further quantitation experiments of docetaxel.

Based on accurate mass data, we postulated further degradation products beside the main degradant epi-docetaxel. Oxidized species of docetaxel and several hydrolysis products (ester and carbamate hydrolysis), as well as oxidation of the products of ester hydrolysis and their respective epimers (Table 2, Figures S3 and S4) are assigned. Two oxidized species of docetaxel show abundant sodium adducts in MS1 of *m*/*z* 828.3200 (oxo-docetaxel, RT: 9.94 min, [C43H51NO<sup>14</sup> + Na+] <sup>+</sup>, exact mass *m*/*z* 828.3202, ∆*m*/*z* = 0.19 ppm) and *m*/*z* 828.3192 (epi-oxo-docetaxel, RT: 11.07 min, [C43H51NO<sup>14</sup> + Na+] <sup>+</sup>, exact mass *m*/*z* 828.3192, ∆*m*/*z* = 1.16 ppm). Their MS/MS spectra show abundant fragments at *m*/*z*oxo-docetaxel 772.2534 and *m*/*z*epi-oxo-docetaxel 772.2584 ([C39H43NO<sup>14</sup> + Na]+, exact mass *m*/*z* 772.2576, ∆*m*/*z*oxo-docetaxel =−1.07 ppm and ∆*m*/*z*epi-oxo-docetaxel =5.41 ppm), which may originate from the loss of the tert-butyl residue. They both show a fragment corresponding to an oxidation at the taxane nucleus at *m*/*z*oxo-docetaxel 547.1927 and *m*/*z*epi-oxo-docetaxel 547.1955 ([C29H32O<sup>9</sup> + Na]+, exact mass *m*/*z* 547.1939, ∆*m*/*z*oxo-docetaxel = 2.11 ppm and ∆*m*/*z*epi-oxo-docetaxel = −3.01 ppm). Analogously to docetaxel, the fragment *m*/*z* 304.1173 originated from the intact phenylpropionic acid side chain ([C14H19NO<sup>5</sup> + Na]+, exact mass *m*/*z* 304.1155, ∆*m*/*z*oxo-docetaxel = −5.77 ppm).

Further degradation products of docetaxel resulted from the ester hydrolysis of the taxane nucleus and the phenylpropionic acid side chain and are postulated here as 10DABIII (*m*/*z* 545.2378, RT: 4.55 min, [C29H36O<sup>10</sup> + H]+, exact mass *m*/*z* 545.2381, ∆*m*/*z* = 0.60 ppm) and epi-10DABIII (*m*/*z* 567.2196, RT: 5.31 min, [C29H36O<sup>10</sup> + Na]+, exact mass *m*/*z* 567.2201, ∆*m*/*z* = 0.83 ppm). A loss of benzoic acid, acetic acid and two losses of water from 10DABIII resulted in the fragment *m*/*z* 327.1587 ([C20H22O<sup>4</sup> + H]+, exact mass *m*/*z* 327.1591, ∆*m*/*z* = 1.18 ppm). Epi-10DABIII showed a fragment at *m*/*z* 445.1791 ([C22H30O<sup>8</sup> + Na]+, exact mass *m*/*z* 445.1833, ∆*m*/*z* = 9.41 ppm) which may correspond to the loss of the benzoic acid moiety and *m*/*z* 385.1615 ([C20H26O<sup>6</sup> + Na]+, exact mass *m*/*z* 385.1622, ∆*m*/*z* = 1.71 ppm), which indicates a subsequent loss of acetic acid.

These two hydrolyzed esters most likely exist in an oxidized form as well, which are proposed as oxo-10DABIII (*m*/*z* 565.2041, RT: 5.40 min, [C29H34O<sup>10</sup> + Na]+, exact mass *m*/*z* 565.2044, ∆*m*/*z* = 0.56 ppm) and epi-oxo-10DABIII (*m*/*z* 565.2040, RT: 5.84 min, [C29H34O<sup>10</sup> + Na]+, exact mass *m*/*z* 565.2044, ∆*m*/*z* = 0.74 ppm) based on their accurate mass data. They both show a distinct fragment at *m*/*z*oxo-10DABIII 443.1661 and *m*/*z*epi-oxo-10DABIII 443.1680 ([C22H28O<sup>8</sup> + Na]+, exact mass *m*/*z* 443.1676, ∆*m*/*z*oxo-10DABIII = 3.47 ppm and ∆*m*/*z*epi-oxo-10DABIII = −0.81 ppm), most likely originating from the loss of the benzoic acid moiety.

Furthermore, the hydrolysis of the carbamate function of docetaxel revealed two more products: 'Carbamate' showed an *m*/*z* 708.3010 in MS1 ([C38H45NO<sup>12</sup> + H]+, exact mass *m*/*z* 708.3015, ∆*m*/*z* = 0.64 ppm), and an abundant fragment of *m*/*z* 182.0818 in MS/MS which may originate from the cleavage of the remaining phenylpropionic acid side chain and the taxane nucleus ([C9H11NO<sup>3</sup> + H]+, exact mass *m*/*z* 182.0812, ∆*m*/*z* = −3.46 ppm). 'Epi-carbamate' showed a similar product ion spectrum with the same base peak of *m*/*z* 182.0820 ([C9H11NO<sup>3</sup> + H]+, exact mass *m*/*z* 182.0812, ∆*m*/*z* = −4.56 ppm) and *m*/*z* 708.3004 ([C38H45NO<sup>12</sup> + H]+, exact mass *m*/*z* 708.3015, ∆*m*/*z* = 1.47 ppm) in MS1.

An exemplary chromatogram of the degradation products following two applications of 70 µg/mL docetaxel for 48 h each is shown in Figure S2. We found only trace amounts of docetaxel degradation products in the TOM model media following the two applications of 7 µg/mL docetaxel for 48 h each. Therefore, we did not consider the degradation products in the real-time pharmacokinetic analyses.

#### *3.3. Validation*

As method A is used for quantitation in the online hyphenation of the tumor model with UHPLC based analysis, it was validated according to the guideline of the EMA [15].

Selectivity and carry-over: The method fulfilled the criteria for selectivity (<20% response in blank artificial matrix compared to response obtained at LLOQ) with a maximum of 7.45% detector response. Carry-over was a more critical parameter since the concentration range was very broad. Even after the optimization of the injector wash procedures, the detector response of analyte-free matrix samples exceeded the allowed 20% LLOQ detector response with a maximum of 29.42% after injection of HQC samples. Therefore, additional blank sample injections were included after samples of high concentrations resulted in successful prevention of carry-over.

Lower limit of quantitation and calibration: The concentration of 0.001 µg/mL showed acceptable accuracy (92.42–114.17%) and precision (6.58%CV) and was therefore chosen as the lowest point of the calibration. Based on the EMA guideline, the calculated LLOQ was 0.16 ng/mL and LOD 0.05 ng/mL, respectively.

For the calibration function, a quadratic fit after log-log transformation of the data provided the best results in terms of a combination of low residuals and best overall accuracy. All CAL samples met the requirements by EMA.

Accuracy and precision: The method (A) fulfilled the requirements given by EMA. Calculated concentrations of QC and CAL samples were within ±15% of the nominal values (Table 3), only 8.33% (within-day) and 15% (between-day) with only individual values outside.


**Table 3.** Accuracy and precision. c: docetaxel concentration, CV: coefficient of variation, RE: Relative error as measure of accuracy, LLOQ: lower limit of quantitation, LQC: lower quality control, MQC: middle quality control, HQC: higher quality control.

#### *3.4. Docetaxel Pharmacokinetics in TOM Models*

The area under the concentration curves (AUC), the maximum concentration (Cmax), and the time to maximum concentration (tmax) as main pharmacokinetic parameters for the concentration versus time profiles of docetaxel within the sampling port are summarized in Table 4.

The course of the concentration versus time curves was comparable between the applied drug doses (Figure 3). Following the administration of docetaxel by supplementing the differentiation medium of TOM models in the reservoir at 0 h, the drug concentration increased until a plateau phase. The time to maximum concentration tmax, 39 ± 7.9 h was almost independent of the administered docetaxel dose, while the Cmax depended on the administered docetaxel dose. Following the exchange of the differentiation medium, again supplemented with the same docetaxel doses, we detected 2.4 to 9.1-fold increased maximum concentrations and 2.4- to 8.8-fold increased AUCs in the sampling port compared to the respective values following the first docetaxel administration. Furthermore, we detected about 4- to 7-fold higher docetaxel concentrations in the sampling port compared to the applied docetaxel concentration (Figure 3b,c). This effect was not observed when applying 7 or 7000 ng/mL docetaxel (Figure 3a,d). Again, the tmax values were close to the end of the treatment cycle with values ranging between 82 and 89 h.


**Table 4.** Main pharmacokinetic parameters following 1st docetaxel application (0–48 h) and 2nd docetaxel application (48–96 h). c: docetaxel concentration, AUC: area under the curve.


**Figure 3.** Concentration-time curves of docetaxel. Docetaxel concentrations in the sampling port of TOM models following the application of (**a**) 7, (**b**) 70, (**c**) 700 or (**d**) 7000 ng/mL docetaxel. Docetaxel was supplemented to the differentiation at 0 and 48 h (arrows), *n* = 4 for each concentration.

Moreover, the concentration versus time curves showed a different shape in two experiments (blue and black curve vs. red and green curve in Figure 3a–c). The slope of the blue and black curves markedly differed from the slope of the red and green curves after the second docetaxel administration. The relatively constant docetaxel concentrations within the sampling port could result from evaporation of medium from the reservoir (Figure 1a(3)), causing in loss of contact of the model with the reservoir. Evaporation also affected the accessibility of the sample fluid for the autosampler needle, since we did not measure docetaxel from certain time points on (e.g., black curve in Figure 3c).

#### **4. Discussion**

An automated UHPLC-MS/MS method with online sampling in TOM models is presented here. This proof-of-concept study demonstrated the feasibility of real-time monitoring of drug levels in TOM models without any sample preparation. To achieve this, both the analytical method for docetaxel quantitation in human blood samples [20] and the culture of TOM models [14] needed to be adapted only slightly. Our approach was validated according to EMA guidelines [15] and is easily transferrable to other in vitro disease models.

In vitro studies frequently use drug doses far higher than the maximum tolerated dose in patients [21]. This overdosing causes effects in vitro that are not reproducible in vivo, contributing to the high attrition rate of investigational new drugs in clinical trials. Even if the patients tolerate such high doses, they will be prone to off-target effects. Aside from the bench-to-bedside extrapolation of drug doses, clinical data can be useful to conduct more relevant studies for investigation of personalized adaptations. Considering the maximum plasma concentration at the highest single dose recommended in the drug product of marketed drugs provides an upper limit for in vitro studies [22]. This concept is particularly useful to test potential new indications for approved drugs. We used docetaxel as a model drug to develop our analytical approach since both the efficacy and the pharmacokinetics of docetaxel are well-known [22]. After calculating a steady-state concentration of 74 ng/mL docetaxel in patients following an intravenous application of 75 mg/m<sup>2</sup> (for details, see [14]), we selected 7; 70; 700; and 7000 ng/mL as test concentrations in TOM models. The AUC within the TOM models ranged between 66.32 and 85,658.15 h × ng/mL following the first, and between 151 and 211,171 h×ng/mL following the second docetaxel application. Together with Cmax values below 2492 and 5920 ng/mL, these in vitro results were in range of the clinical application of 100 mg/m<sup>2</sup> , which results in an AUC of 4600 h × ng/mL and Cmax of 3700 ng/mL [23].

Focusing on the nominal concentration of 70 ng/mL, we detected less docetaxel in the sampling port than has been found in human blood samples. This discrepancy supports the hypothesis of the poor uptake of anticancer drugs into solid tumors [6]. Likewise, paclitaxel penetrates only to the periphery of spheroids [5]. Nevertheless, docetaxel uptake into the TOM models increased following the second drug application. Since apoptosis results in enhanced drug uptake into inner cell layers of solid tumors [8], tumor cells dying after the first application should favor docetaxel uptake into TOM models.

Moreover, our method provides an in-depth insight into the formation of docetaxel degradation products. Since docetaxel epimerization is associated with a loss of potency and tumor resistance development in vivo [24], the considerable epimer formation will affect the efficacy of docetaxel. In contrast, the trace amounts of oxidation products and carbamates should not limit docetaxel effects in TOM models, although they are 10- to 40-fold less active [25]. The degradation products were identified by QTOF-MS and related to degradation products known from the literature [26]. Nevertheless, our approach allows for only limited insights into clinically relevant clearance due to the absence of hepatic metabolism and biliary excretion. If tumor cells metabolize the applied drugs, the quantitation of local metabolites will be feasible as well, but in the case at hand, we observed docetaxel epimerization and formation of degradation products as artifacts also in cell-free medium.

Differences between docetaxel concentrations in human patients and TOM models also arise from differences in protein binding. Whereas plasma protein binding of docetaxel is 97% in the patients [22], protein binding in medium containing fetal bovine serum is saturable. Paclitaxel, close in chemical structure to docetaxel, shows a protein binding between 79% at 500 ng/mL and 20% at 15,000 ng/mL [27]. Thus, we expect higher amounts of free drug available compared to the patients, especially following the application of 7000 ng/mL docetaxel. Nevertheless, we were not able to discriminate free against total docetaxel concentration, since the membrane of the sampling port has a

pore size of 400 nm. Most protein sizes range between 1 and 100 nm, making protein diffusion into the sampling port likely. This might also explain higher Cmax values in the sampling port than the administered concentration in the reservoir, since we quantified all docetaxel within the sampling port. The first docetaxel administration saturated the protein binding and intracellular fluids, the second application directly increased the concentration in the interstitial fluid of TOM models and the sampling port. However, we assume complete equilibration between the interstitial fluid of the TOM model and the sampling fluid within two hours, equal to the time interval we selected between two measurements. Thus, signals of the concentration over time curve earlier between zero and two hours might not recapitulate the concentration within the interstitial fluid of the TOM model, but provide an insight into the lag-time between docetaxel application and first appearance within the sampling port. As to be expected, the lag-time decreases with increasing docetaxel concentrations: 11.1 ± 3 h (7 ng/mL docetaxel application) compared to 1.82 ± 0.6 h (7000 ng/mL docetaxel application).

Since classical microdialysis already allowed insights into tissue-specific drug [28] and cytokine levels [29], the automated determination of pharmacokinetic profiles will enable patient-specific analyses in higher throughput. PK-PD modelling already improved dose selection and characterization of drug effects on tumor growth, overall survival and safety [30], but requires relevant data for the patient and his/her tumor. Nonclinical testing together with pharmacometrics may provide a more detailed insight by testing drugs in patient-specific models and extrapolating drug concentrations in tumors to adapt dose regimen for patients.

UHPLC-MS/MS again proved as the method of choice as it was already useful for a wide range of applications in pharmacology, toxicology, and forensics [31–33]. Despite first dilute and inject attempts to reduce the time-consuming sample preparation [34–39], UHPLC-MS/MS analyses still often utilizes extensive sample preparation to separate the molecule of interest from interfering proteins and potential enzymatic degradation processes [40]. Our method (A) used for quantitation of docetaxel was successfully validated in terms of selectivity, carry-over, lower limit of quantitation (LLOQ), calibration function, accuracy, and precision according to EMA guidelines for bioanalytical method validation. A very broad concentration range of 1–10,000 ng/mL was covered compared to already published methods [20,41], allowing the analysis of docetaxel administered ranging from 7 to 7000 ng/mL. The method proved to be accurate and precise, showed acceptable carry-over after including blank injections between high and low concentration samples, as well as fitness-for purpose in LLOQ. Furthermore, the method was fast, being able to separate docetaxel and 7-epi-docetaxel in less than 3.7 min (total run-time including cleaning of injector 9.75 min).

Future studies will compare differences between the patients' drug responses and drug delivery systems to optimize the dose regimen and application form. For increased efficacy, model size and sampling volume may be further optimized in the direction of high-throughput, and therefore, enhance personalized medicine.

#### **5. Conclusions**

We developed and evaluated a real-time approach to automatically measure docetaxel concentrations in TOM models. Partial epimerization and neglectable amounts of degradation products were detected instantaneously upon application of docetaxel to the medium. The courses of concentration versus time curves for 96 h were comparable among four different docetaxel concentrations. The first drug application resulted in an increase of docetaxel concentration, followed by a plateau phase, and exceeded after the second drug application. This proof-of-concept study paves the way for real-time pharmacokinetic and further online investigations in 3D tumor models and beyond, and thus, helps to improve preclinical drug development and personalized medicine.

**Supplementary Materials:** The following are available online at http://www.mdpi.com/1999-4923/12/5/413/s1, Figure S1: Multiple reaction monitoring (MRM) chromatogram, Figure S2: Overlay of extracted ion chromatograms of docetaxel and degradation products, Figure S3: Product ion spectra of degradation products, Figure S4: Suggested chemical structure of docetaxel and degradation products, structural differences in comparison to docetaxel are displayed in red color.

**Author Contributions:** Conceptualization, J.F.J., L.G., B.W., C.Z., and M.K.P.; methodology, J.F.J., L.G., B.W., C.Z., and M.K.P.; validation, J.F.J., L.G., and C.Z.; formal analysis, J.F.J., L.G. and C.Z.; investigation, J.F.J., L.G., J.G., L.M.C..; resources, C.Z., U.K., M.K.P.; data curation, J.F.J., L.G., J.G.-M., L.M.C.; writing—original draft preparation, J.F.J., L.G., C.Z.; writing—review and editing, J.F.J., L.G., U.K., C.Z., M.K.P.; visualization, J.F.J., L.G., J.G., C.Z.; supervision, C.Z., M.K.P.; project administration, U.K., C.Z., M.K.P.; funding acquisition, C.Z., U.K., M.K.P. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by the Freie Universität Berlin (Focus Area Disease in Human Aging, "DynAge").

**Acknowledgments:** The authors thank Kathleen Sauvetre, Université Angers, France, for technical help with the sampling port and Luis Ilia, Freie Universität Berlin, Germany, for help with the pharmacokinetic analyses. The publication of this article was funded by Freie Universität Berlin.

**Conflicts of Interest:** The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.
