Characterization of Lipophilicity and Blood Partitioning of Pyrrolizidine Alkaloids and Their N-Oxides In Vitro and In Silico for Toxicokinetic Modeling

Abstract

Lipophilicity and blood partitioning are important determinants for predicting toxicokinetics using physiologically-based toxicokinetic modeling. In this study, the logarithm of the n-octanol : water partition coefficient and the blood-to-plasma concentration ratio were, for the first time, experimentally determined for the pyrrolizidine alkaloids intermedine, lasiocarpine, monocrotaline, retrorsine, and their N-oxides. Validated in vitro assays for determination of the n-octanol : water partition coefficient (miniaturized shake-flask method) and the blood-to-plasma conentration ratio (LC-MS/MS-based depletion assay) were compared to an ensemble of in silico models. The experimentally determined octanol : water partition coefficient indicates a higher affinity of pyrrolizidine alkaloids and their N-oxides to the aqueous compared to the organic phase. Depending on the method, in silico determined n-octanol : water partition coefficients overpredicted the experimental values by ≥ 1 log unit for three out of four pyrrolizidine alkaloids (SPARC), four out of six (CLOGP), five out of eight (KowWIN), and three out of eight (S+logP) pyrrolizidine alkaloids and their N-oxides. The blood-to-plasma concentration ratio obtained in vitro suggested a low binding affinity of pyrrolizidine alkaloids and their N-oxides towards red blood cells. For all eight pyrrolizidine alkaloids and their N-oxides, in silico predicted blood-to-plasma ratios deviated from experimental values by less than 50%. In conclusion, for physiologically-based toxicokinetic modeling of pyrrolizidine alkaloids and their N-oxides, the experimental octanol : water partition coefficient should be preferred, while the blood-to-plasma concentration ratio predicted by the acid/base classification model is a suitable surrogate for experimental data.

Keywords

logP - octanol : water partition coefficient - blood-to-plasma ratio - drug distribution - pharmacokinetics - PBTK modeling

Introduction

Pyrrolizidine alkaloids (PAs) are a large group of secondary plant metabolites of which the 1,2-unsaturated PAs are of particular concern due to their hepatotoxicity, genotoxicity, and potential carcinogenicity [1]. PAs and their corresponding N-oxides (PANOs) usually coexist in plants, while the latter are reduced to toxic PAs in both the intestine and liver [2]. The relative toxicity of structurally different PA congeners is substantially influenced by their individual kinetics, i.e., by their absorption, distribution, metabolism, and excretion (ADME) characteristics [3]. Physiologically-based toxicokinetic (PBTK) modeling is a powerful tool to explore the ADME characteristics of PAs within an organism. PBTK models have been successfully used for PAs to quantitatively predict in vivo liver toxicity and in vivo genotoxicity from in vitro toxicity data [4], [5]. In conjunction with other physico- and biochemical properties, lipophilicity and blood partitioning are important determinants in PBTK models for predicting whole body toxicokinetics.

Since the in vivo analysis of tissue distribution requires animal testing that, apart from ethical concerns, is costly and time intensive, mechanistic models for an a priori prediction of tissue distribution have been established for PBTK modeling. In these mechanistic approaches, e.g., by Rodgers and Rowland [6], [7] or by Schmitt [8], [9], the composition of the different tissues and the interaction of compounds with the tissue constituents is taken into consideration. For prediction of compound partitioning into the tissue lipids, a measure of lipophilicity is needed as a parameter in the models.

The n-octanol : water partition coefficient K _o : w (P) is a measure of lipophilicity in absence of relevant speciation and refers to the neutral form of a substance. K _o : w is defined as the concentration ratio in an organic phase, typically n-octanol, and an aqueous phase:

K o : w   =   P =   C o r g a n i c C a q u e o u s {K}_{o:w}\ =\ P=\ {{C}_{organic}\over{{C}_{aqueous}}}

K _o : w is usually expressed as base 10 logarithm logP.

For ionizable substances, where neutral and charged species coexist to a relevant extent in solution, the corresponding distribution ratio D specifically accounts for neutral and ionized species [10]:

D =   C n e u t r a l ,     o r g a n i c   +   C i o n i z e d ,   o r g a n i c C n e u t r a l , a q u e o u s   +   C i o n i z e d ,   a q u e o u s D=\ {{C}_{neutral,\ \ organic}\ {+}\ {C}_{ionized,\ organic}\over{{C}_{neutral,aqueous}\ {+}\ {C}_{ionized,\ aqueous}}}

Since the degree of ionization depends on pH, so does D. D is usually given as base 10 logarithm logD. The distribution coefficient logD can be calculated for a specific pH from logP and the acid dissociation constant pKa using the Henderson-Hasselbalch relation:

l o g D ( p H ) = l o g P   -   l o g ( 1   +   10 ( p H - p K a ) Δ i )   logD(pH)=logP\ \hbox{‐}\ log(1\ {+}\ {10}^{(pH\hbox{‐}pKa)}\Delta i)

with Δi = 1 for acids and Δi = − 1 for bases [11].

PAs are weak bases, since the nitrogen atom of the tertiary amine becomes protonated under neutral and acidic conditions. PANOs also show weak basic behavior, attributed to protonation of the oxygen atom of the N-oxide under acidic conditions. The weak basic characteristics of PAs and PANOs have been confirmed by in silico predicted pKa values (see Table 1S, Supporting Information).

Experimentally, the logP value can be determined using direct or indirect methods [12]. Direct experimental methods, such as the OECD shake flask method [13] or miniaturized adaptations [14], obtain logP directly from the concentration ratio of the compound partitioned between the organic and aqueous phase. In indirect experimental methods, like reversed-phase HPLC (RP-HPLC) [15], logP is estimated through the compound retention in a hydrophobic stationary phase. During experimental logP determination for ionizable substances, the impact of pH on the ionization state needs to be considered so that partitioning exclusively refers to their neutral form. In addition, a vast number of in silico methodologies have been developed to predict lipophilicity. While substructure-based models derive logP from cutting molecules into fragments or down to the single-atom level, property-based models use empirical approaches, the 3D structure of a molecule, or topological descriptors [16]. The lipophilicity of a compound is related to many biological properties, such as solubility and membrane permeability, thereby strongly affecting in vivo compound ADME characteristics and its pharmacodynamic and toxicological profile [17].

Certain compounds are highly bound to or distributed into blood cells. Like plasma protein binding, this can significantly reduce a compoundʼs free plasma concentration, i.e., the fraction that is available for its pharmacological or toxicological action and its elimination. Blood partitioning, therefore, can strongly determine the toxicokinetic and toxicodynamic profile of a compound. The blood-to-plasma concentration ratio (R_b = C _blood/C _plasma) is a measure of compound distribution within whole blood at equilibrium [18].

The importance of a precisely determined R_b value can be illustrated on, for example, the drugs thioridazine and maprotiline. For thioridazine, which shows an R_b value of 0.55 [19] and therefore almost no distribution into blood cells, exposure calculated from plasma equals twice the exposure calculated from whole blood. For maprotiline, which is distributed into blood cells, showing a R_b value of 2.1 [19], exposure determined from plasma is half the exposure determined from whole blood. In the latter case, clearance calculated from plasma data would highly overestimate whole blood clearance if R_b was not considered and could even exceed hepatic blood flow.

Experimentally, the R_b value can be obtained in vitro by a separate measurement of compound concentrations in equilibrating plasma and whole blood (or blood cells) or, more laborious, in vivo within a pharmacokinetic study [20]. In addition, in silico methodologies for R_b prediction have been developed. Some methodologies are based on mechanistic prediction of distribution into red blood cells [6], [8], while others comprise regression models that relate a variety of molecular descriptors to the R_b value using statistical algorithms such as partial least squares regression and artificial neural networks [18], [21], [22]. In the absence of experimental data, it is often assumed R_b  = 1 for neutrals and bases, and R_b  = 1-hematocrit = 0.55 for acids and zwitterions (acid/base classification) [22].

Due to the need to characterize lipophilicity and blood partitioning of PAs for PBTK modeling, the aim of this study was to determine logP and R_b values of representative 1,2-unsaturated PA congeners. Therefore, PAs of different ester types were analyzed, including the monoester intermedine, the cyclic diesters monocrotaline and retrorsine, and the open-chained diester lasiocarpine. Additionally, the corresponding PANOs were included in the study. The logP and human R_b values were determined in vitro using established experimental methods and compared to published literature values. The performance of established in silico models for the prediction of logP and R_b values was assessed for PAs and PANOs. LogP values were predicted using two substructure- and two property-based models and R_b values were predicted by an acid/base classification model and a mechanistic model. Finally, in silico predictions were evaluated against experimental values.

Results

LogP values were determined by bidirectional equilibration from n-octanol to water and vice versa (see section titiled Determination of n-octanol : water partition coefficient). Method evaluation with reference substances showed a convincing linear correlation (R² = 0.988) between determined logP values and literature logP values (Fig. 1Sa, Supporting Information). Except for warfarin, determined logP values did not differ by more than 0.2 log units from those reported in literature demonstrating a high accuracy (Table 2S, Supporting Information). The experimental method was precise, as indicated by relative standard deviations below 10% for sulfanilamide, tolbutamide, and warfarin (caffeine: 20%, theophylline: 47%).

For all PAs and PANOs, measured logP values were negative and ranged between − 1.93 (intermedine) and − 0.302 (lasiocarpine N-oxide) ([Table 1] and [Fig. 1], circles). There were no significant differences in logP values depending on the direction of equilibration, i.e., starting from the organic phase or starting from the aqueous phase. LogP values of PANOs were between 0.02 log units (retrorsine N-oxide) and 0.2 log units (lasiocarpine N-oxide), higher compared to logP values of the respective PAs.

Table 1 Experimentally determined and predicted logP values of PAs and PANOs.

Pyrrolizidine alkaloids and their N-oxides	Logarithm of the n-octanol:water partition coefficient (logP)
	Experimentala		In silico prediction
			Substructure-based		Property-based
			KowWIN [28]	CLOGP [29]	SPARC [30]	S + logP [31]
aMean ± standard deviation of equilibration from the organic to the aqueous phase (n = 3) and from the aqueous to the organic phase (n = 3), blogP predictions not available
Intermedine	− 1.93	± 0.168	0.91	− 0.96	− 0.85	− 0.045
Intermedine N-oxide	− 1.36	± 0.0346	0.15	–b	–b	− 0.875
Lasiocarpine	− 0.487	± 0.0220	2.43	0.65	1.21	1.59
Lasiocarpine N-oxide	− 0.302	± 0.0269	1.66	0.77	–b	0.319
Monocrotaline	− 1.92	± 0.0727	− 1.18	− 0.93	− 1.01	− 0.686
Monocrotaline N-oxide	− 1.61	± 0.0352	− 1.95	–b	–b	− 1.56
Retrorsine	− 1.26	± 0.0610	− 0.25	− 0.62	− 0.24	− 0.298
Retrorsine N-oxide	− 1.24	± 0.0421	− 1.01	− 0.50	–b	− 1.26

Fig. 1 Experimentally determined and predicted logP values of PAs and PANOs. Experimental data are shown as the mean ± standard deviation (SD) of equilibration from the organic to the aqueous phase (n = 3, yellow circles) and from the aqueous to the organic phase (n = 3, blue circles). LogP predictions were not available for intermedine N-oxide, monocrotaline N-oxide (CLOGP), and for PANOs (SPARC).

The logP predicted by KowWIN deviated from experimental values by < 1 log unit for monocrotaline, monocrotaline N-oxide, and retrorsine N-oxide and by ≥ 1 log unit for the other PAs and PANOs ([Table 1] and [Fig. 1], asterisks). LogP calculated by the CLOGP method overpredicted experimental values by < 1 log unit for retrorsine and retrorsine N-oxide and by ≥ 1 log unit for the other PAs and PANOs. Regarding SPARC, logP values were overpredicted by < 1 log unit for monocrotaline and by ≥ 1 log unit for the other PAs. Note that logP predictions were not available for PANOs in SPARC. LogP predicted by the S + logP method (ADMET predictor) deviated from experimental values by < 1 log unit for all PANOs and retrorsine and by ≥ 1 log unit for intermedine, lasiocarpine, and monocrotaline.

The R_b value was determined using the LC-MS/MS-based depletion assay that measures the compound concentration (i) in plasma that has been equilibrating with red blood cells and (ii) in a plasma reference (see section titled Determination of blood-to-plasma concentration ratio). The study was conducted in accordance with the Declaration of Helsinki and approved by the Ethics Committee of the Charité University Medicine Berlin (EA4/183/19, November 12, 2019). Method evaluation with reference substances showed a good linear correlation (R² = 0.908) between determined R_b and literature R_b (Fig. 1Sb, Supporting Information). Except for imipramine and warfarin, the determined R_b did not deviate by more than 10% from previously published R_b values, which shows the high accuracy of the method (Table 2S, Supporting Information). Relative standard deviations below 10% indicate that the experimental method was precise. An effect of the compound concentration on the R_b value was not observed, within the range of 4 to 1000 ng/mL exemplarily determined for imipramine, caffeine, and warfarin (Fig. 2S, Supporting Information).

Measured R_b values of PAs and PANOs ranged between 0.689 (intermedine N-oxide) and 1.12 (intermedine) ([Table 2] and [Fig. 2], circles). The R_b value of the PAs intermedine, monocrotaline, and retrorsine was identical (R_b  = 1.1). Except for lasiocarpine, determined R_b values of PAs were 30 to 40% higher compared to that of their corresponding N-oxide. Within the range of 4 to 1000 ng/mL, no concentration dependency was observed, exemplarily determined for retrorsine and its N-oxide ([Fig. 3 a, b]).

Table 2 Experimentally determined and predicted R_b values of PAs and PANOs.

Pyrrolizidine alkloids and their N-oxides	Blood-to-plasma concentration ratio (Rb )
	Experimentala		In silico prediction
			Acid/base classification modelb (Eq. 6)	Mechanistic model (Eq. 7)
aMean ± standard deviation (n = 6), bat a physiological pH of 7.4, PAs are basic, while PANOs are neutral
Intermedine	1.12	± 0.0853	1.00	0.807
Intermedine N-oxide	0.689	± 0.0499	1.00	0.867
Lasiocarpine	0.736	± 0.0438	1.00	0.601
Lasiocarpine N-oxide	0.897	± 0.0718	1.00	0.605
Monocrotaline	1.07	± 0.0865	1.00	0.672
Monocrotaline N-oxide	0.709	± 0.0430	1.00	0.674
Retrorsine	1.08	± 0.0859	1.00	0.717
Retrorsine N-oxide	0.736	± 0.228	1.00	0.830

Fig. 2 Experimentally determined and predicted R_b values of PAs and PANOs. Experimental data are shown as the mean ± standard deviation (SD) (n = 6).

Fig. 3 Effect of the concentration (4, 20, 100, 500, and 1000 ng/mL) on the R_b value on the example of the PA retrorsine (a) and its N-oxide (b). Experimental data are shown as the mean ± standard deviation (SD), with n = 6 (n = 3: retrorsine 500 ng/mL).

In silico predictions of R_b according to the acid/base classification model deviated from experimentally determined R_b by less than 10% for monocrotaline and retrorsine and by more than 10% but less than 50% for the other PAs and PANOs, showing a tendency for over prediction ([Table 2] and [Fig. 2], asterisks). Regarding the mechanistic model, in silico predictions of R_b deviated from experimentally determined R_b by less than 10% for monocrotaline N-oxide and by more than 10% but less than 50% for the other PAs and PANOs, with a tendency for under prediction.

Discussion

We herein report, for the first time, lipophilicity and blood partitioning of representative PAs and their corresponding N-oxides, both being important predictors of compound ADME characteristics and key parameters in PBTK models. Measures of lipophilicity and blood partitioning, logP and R_b values, were determined using published experimental assays as well as established in silico methods.

For all PAs and PANOs, logP values were negative, showing that the compounds have a higher affinity to the aqueous phase. The observed lipophilicity of analyzed PAs can be ranked as intermedine < monocrotaline < retrorsine < lasiocarpine. Regarding the PA ester type, these results suggest the following lipophilicity order, which is, however, limited by the small number of PAs assessed: monoester < cyclic diester < open-chained diester. In reversed-phase liquid chromatography, logP is correlated with the retention time, i.e., hydrophilic compounds elute first and lipophilic last [15]. Our observed lipophilicity ranking is supported by the elution order of 22 structurally different PAs in the reversed-phase chromatogram [23]. Monoesters and cyclic diesters tend to have shorter retention times, which implies that they have a lower lipophilicity compared to open-chained diesters. The experimental results indicate, except for lasiocarpine N-oxide, that PANOs are slightly more lipophilic compared to their respective PAs. This is in opposition to the N-oxidation of pyridines, aromatic weak bases with a chemical structure related to PAs, which was reported to increase hydrophilicity [24].

LogP values calculated using in silico methodologies showed a trend for overprediction for PAs and PANOs. Predictions by substructure-based methods deviated upward from the experimental values by ≥ 1 log unit for five out of eight (63%) PAs and PANOs (KowWIN) and four out of six (67%) PAs and PANOs (CLOGP). Predictions by property-based methods deviated upwards by ≥ 1 log unit for three out of four (75%) PAs (SPARC) and three out of eight (38%) PAs and PANOs (S + logP). The best performance was achieved with S + logP, which showed particularly good results for all PANOs, but not PAs.

For comparison, KowWIN, SPARC, and S + logP (CLOGP not available) showed better accuracy of logP for reference substances than for PAs and PANOs, i.e., deviations were < 1 log unit for five out of five (100%, KowWIN) and four out of five (80%, SPARC, S + logP) reference substances (also see Fig. 3S, Supporting Information). The accuracy of logP prediction methods is strongly dependent on the compounds included in a training dataset and is limited by the complexity of molecules, in particular by their size [16].

Exposure-related kinetic parameters, such as clearance, especially in relation to blood flow and volume of distribution, can be misleading if interpreted based on plasma instead of whole blood data, which can be prevented if the R_b value of a compound is known. R_b values below 1 (C _plasma > C _blood) indicate stronger binding to plasma proteins than to red blood cells, while R_b values above 1 (C _blood > C _plasma) show that binding to blood cells is larger than plasma protein binding [18]. For PAs and PANOs, determined R_b values were in the range between 0.689 and 1.12, with R_b values of N-oxides tending to be smaller compared to PAs. These results suggest that PAs and PANOs have a low binding affinity towards red blood cells and distribution within red blood cells is not expected. This is in line with the absence of a concentration dependency of the R_b value, as was exemplarily shown for retrorsine and its N-oxide. By contrast, for compounds that are highly bound or distributed into red blood cells, concentration-dependent saturation of binding or active uptake would have been expected [25], [26].

In the absence of experimental data, the R_b value is often assumed to be 1 for neutrals and bases or 0.55 (1-hematocrit) for acids and zwitterions [22]. Since PAs are weak bases and PANOs are neutrals under physiological conditions, their R_b value is 1 according to the acid/base classification model. While predictions by this acid/base classification model were slightly superior to predictions by the mechanistic tissue distribution model, both methods showed predictions close to experimental values, with deviations below 1.5-fold for all PAs and PANOs. In particular, R_b values deviated from the experimental values by less than 10% for two out of eight (acid/base classification model) and one out of eight (mechanistic model) PAs and PANOs, and by more than 10% but less than 50% for six out of eight (acid/base classification model) and seven out of eight (mechanistic model) PAs and PANOs. It was reported that the performance of more complex artificial neural network (ANN) models was superior to that of mechanistic models for tissue distribution [18], [22]. However, the application of ANN models is less feasible, since in addition to logP and fraction unbound in plasma (f _u,p), more compound-specific prior information (> 10 molecular descriptors) is required.

For PAs and PANOs, we recommend predicting tissue distribution with experimentally determined logP values, since the outlined deviations of in silico predictions from validated in vitro measurements were ≥ 1 log unit. A direct experimental method for logP determination, such as the miniaturized shake flask method used in this study, should be preferred over indirect methods to avoid erroneous results [27]. While in silico prediction of logP values was insufficient, the in silico predicted R_b value using the acid/base classification model (deviations below 1.5-fold) can be regarded as a suitable surrogate for PAs and PANOs when experimental data on blood partitioning are not available.

Material and Methods

Materials

Intermedine (purity > 95%) and lasiocarpine (purity > 97%) were purchased from Phytoplan Diehm & Neuberger GmbH. Retrorsine (purity ≥ 90%) was obtained from Sigma-Aldrich Chemie GmbH. Monocrotaline (purity ≥ 90%), intermedine N-oxide (≥ 90%), monocrotaline N-oxide (≥ 90%), and retrorsine N-oxide (purity ≥ 90%) were received from Phytolab GmbH & Co. KG. Lasiocarpine N-oxide (purity 96.7%) was purchased from Cfm Oskar Tropitzsch GmbH. Caffeine (purity ≥ 99%), imipramine hydrochlorid (purity ≥ 99%), sulfanilamide (purity ≥ 99%), theophylline (purity ≥ 99%), tolbutamide (purity 99.8%), warfarin (purity ≥ 98%), and quinidine (purity ≥ 80%) were obtained from Sigma-Aldrich.

1-Octanol Chromasolv (purity ≥ 99%) was purchased from Fisher Scientific GmbH. Water LiChrosolv, acetonitrile LiChrosolv (purity ≥ 99.9%), and methanol LiChrosolv (purity ≥ 99.9%) were obtained from Merck KGaA. Ammonium formate Chromanorm (purity ≥ 99%) was provided by VWR International GmbH.

Collection of whole blood and plasma samples

Blood samples were collected from two healthy donors (female, 44 years; male, 25 years). Collection of blood samples was conducted in accordance with the Declaration of Helsinki and was approved by the Ethics Committee of the Charité University Medicine Berlin (EA4/183/19, November 12, 2019). Whole blood was collected from the cubital vein into blood collection tubes with heparin (S-Monovette Lithium Heparin, Sarstedt AG & Co. KG). To obtain plasma, collection tubes were centrifuged for 10 min at 11 180 g and 8 °C and the plasma was separated from blood cells. Whole blood and plasma samples were stored at 4 °C until analysis.

Determination of the n-octanol : water partition coefficient P

The n-octanol : water partition coefficient P was determined according to the miniaturized shake flask method [14]. Presaturated solutions of n-octanol and water were prepared. Therefore, equal volumes of n-octanol and water were mixed thoroughly in a separating funnel. After equilibration for 24 h, the organic and aqueous phases were separated for further use. The compounds were dissolved either in presaturated water or in presaturated n-octanol to measure bidirectional partitioning from the aqueous to the organic phase and vice versa. Next, 350 µL of compound solution were added to a reaction tube and the volume was made up to 500 µL with the same phase to achieve a concentration of 1000 ng/mL. Then, 500 µL of the other phase were added to obtain a total volume of 1 mL. Reaction tubes were shaken overnight at 500 rpm and 22 °C. After centrifugation for 20 min at 16 099 g and 20 °C, separated phases were removed using a glass syringe to avoid phase contamination. Of each sample, 200 µL were transferred to glass vials and analyzed by LC-MS/MS. Experiments were performed in triplicate for equilibration from the organic to the aqueous phase (n = 3) and from the aqueous to the organic phase (n = 3).

The logP values were calculated as logarithm of the mass-spectrometric peak area ratio (Eq. 4). Analyte concentrations were assumed proportional to peak areas since volumes of aqueous and organic phase were identical and samples were not diluted.

l o g P = l o g P e a k   a r e a   ( A n a l y t e   i n   n - o c t a n o l ) P e a k   a r e a   ( A n a l y t e   i n   w a t e r ) logP=log{Peak\ area\ (Analyte\ in\ n\hbox{‐}octanol)\over{Peak\ area\ (Analyte\ in\ water)}}

The experimental method was validated with the five structurally different reference substances caffeine, sulfanilamide, theophylline, tolbutamide, and warfarin. Linear correlation between determined and literature logP values was analyzed visually and numerically using the coefficient of determination R². PAs and PANOs as well as reference compounds were neutral (≥ 96%) under experimental conditions (see Table 1S, Supporting Information).

The performance of in silico methodologies for prediction of the logP value was assessed against experimentally determined logP values for PAs and PANOs. To this end, two substructure-based prediction methods, KowWIN [28] and CLOGP (values published in [29]), were compared to two property-based methods, SPARC [30] and S + logP (ADMET Predictor [31]).

Determination of blood-to-plasma concentration ratio

Conventionally, the partitioning of a compound between blood and plasma is determined by separate analysis of compound concentrations in equilibrating plasma and whole blood or red blood cells. Therefore, separate standards in their respective matrix, i.e., in plasma and whole blood or red blood cell lysates, are used. Here, we used an LC-MS/MS-based depletion assay. The blood-to-plasma concentration ratio R_b was determined by measurement of compound concentrations in the plasma sample against a defined plasma reference [19].

A whole blood sample and a plasma reference sample of the same volume were spiked with the compound to achieve a final concentration of 1000 ng/mL. Additional concentrations were assessed at 4, 20, 100, and 500 ng/mL to analyze concentration dependency of R_b . The whole blood was incubated on a shaker for 60 min at 37 °C to allow equilibration of the compound between the plasma and red blood cells. Then, the whole blood was centrifuged for 10 min at 11 180 g and 8 °C to yield the plasma. The plasma reference was treated the same way as the whole blood regarding incubation and centrifugation. Then, 50 µL of plasma were collected from the separated whole blood and from the plasma reference, respectively. The collected plasma samples were quenched with 450 µL of ice-cold acetonitrile, mixed with a vortex mixer and centrifuged at 21 913 g for 10 min to remove plasma protein. Supernatants were collected and analyzed using LC-MS/MS. Experiments for PAs and PANOs were performed with n = 6 for PAs and PANOs and n = 4 for reference substances.

The R_b value was calculated using the analyte peak area responses of the plasma samples (Eq. 5). Since volumes and matrix of the reference plasma sample and the equilibrating plasma sample were identical, analyte concentrations were proportional to the peak areas.

R b = P e a k   a r e a   A n a l y t e   i n   p l a s m a   r e f e r e n c e P e a k   a r e a   A n a l y t e   i n   e q u i l i b r a t i n g   p l a s m a {R}_{b}={Peak\ area\ \left(Analyte\ in\ plasma\ reference\right)\over{Peak\ area\ \left(Analyte\ in\ equilibrating\ plasma\right)}}

The experimental method was validated with the six structurally diverse reference substances, caffeine, imipramine, theophylline, tolbutamide, warfarin, and quinidine. The linear correlation between determined and literature R_b values was assessed visually and numerically (R²).

The performance of in silico models for the R_b value was evaluated against the experimentally determined R_b values for PAs and PANOs. To this end, an acid/base classification model and a mechanistic model for R_b value prediction were used. In the acid/base classification model, the R_b value was assumed 1 to be for neutrals and bases or 0.55 (1-hematocrit) for acids and zwitterions [22]:

R b = 1             f o r   n e u t r a l s   a n d   b a s e s 0.55           f o r   a c i d s   a n d   z w i t t e r i o n s {R}_{b}=\left\{\matrix {{c}1\ \ \ \ \ \ for\ neutrals\ and\ bases\cr 0.55\ \ \ \ \ for\ acids\ and\ zwitterions}\right.

PAs and PANOs are weak bases (see Table 1S, Supporting Information, for pKa predictions). Under physiological conditions, i.e., at pH 7.4, PAs are present as bases, while PANOs are uncharged and therefore present as neutrals. Hence, according to the acid/base classification model, PAs and PANOs are assigned an R_b value of 1.

In the mechanistic model, R_b values are calculated from red blood cells to the unbound plasma partition coefficient K _rbc,pla by the Open Systems Pharmacology (OSP) Suite (https://www.open-systems-pharmacology.org/, accessed October 18, 2023) based on a mechanistic tissue distribution model. In this model, compound distribution into red blood cells is approximated by the distribution into the cellular constituents water, lipids, and proteins, instead:

R b = ( K r b c , p l a ×   f u , p - 1 )   × h c t - 1 {R}_{b}={(K}_{rbc,pla}\bullet \ {f}_{u,p}\hbox{‐}1)\ \bullet hct\hbox{‐}1

K r b c , p l a = ( f r b c , w a t e r +   f r b c , l i p i d s × 10 l o g P +   f r b c , p r o t × K p r o t {K}_{rbc,pla}=({f}_{rbc,water}{+}\ {f}_{rbc,lipids}\bullet {10}^{logP}{+}\ {f}_{rbc,prot}\bullet {K}_{prot}

with K _rbc,pla = red blood cells to the unbound plasma partition coefficient; f _u,p = fraction unbound in plasma; hct = hematocrit; f _rbc,water = volume fraction of red blood cells that is water; f _rbc,lipids = volume fraction of red blood cells that is lipids; logP = logarithm to base 10 of the n-octanol : water partition coefficient; f _rbc,prot = volume fraction of red blood cells that is proteins; K _prot = water-protein partition coefficient.

LogP values were experimentally determined (see [Table 1]). The f _u,p values were taken from our in-house dataset (intermedine: 0.890, intermedine N-oxide: 1.08, retrorsine: 0.600, retrorsine N-oxide: 0.960; rapid equilibrium dialysis) or were predicted (lasiocarpine: 0.227, lasiocarpine N-oxide: 0.237, monocrotaline: 0.456, monocrotaline N-oxide: 0.461 [32]).

Sample analysis by liquid chromatography and high-resolution mass spectrometry

The chromatographic measurements were performed using an UltiMate 3000 UHPLC system (Thermo Fisher Scientific) with a C18 Hypersil Gold column (150 mm × 2.1 mm; 1.9 µm particle size) and guard column (Thermo Fisher Scientific). The column temperature was set at 40 °C, while the flow rate and injection volume were adjusted to 0.3 mL/min and 2 µL, respectively. UHPLC was carried out by a binary mobile phase consisting of water (A) and methanol (B) as mobile phases, whereby 0.1% formic acid and 5 mM ammonium formate in each mobile phase was integrated. The gradient system was configured as follows: 0 – 7.0 min A: 95%, 7.0 – 7.5 min A: 50%, 7.5 – 7.6 min A: 20%, 7.6 – 10.1 min A: 0%, 10.1 – 15 min A: 95%.

High-resolution mass spectrometry was conducted using an Orbitrap mass spectrometer (Thermo Fisher Scientific) in the positive ionization mode (precursor ions [M + H]⁺) to determine ion scans of reference substances caffeine, imipramine, sulfanilamide, theophylline, tolbutamide, quinidine, and warfarin. Ion scans (ddMS2) with collision energies in the range of 10 – 35 eV and a resolution of 17 500 were recorded, whereby three fragments

(product ions) with the highest relative abundances of each substance were identified (Table 3S, Supporting Information). Representative ddMS2 data of the reference substances achieved with 25 eV are shown in Figs. 4S-10S, Supporting Information. For further analyses, samples containing reference substances were measured by data-independent acquisition (vDIA) consisting of full scans for quantification and MS2 data by fragmentation of four mass range windows (m/z: 50 – 150, 130 – 300, 280 – 400, and 380 – 600) with a fixed collision energy of 25 eV for verification. Data were acquired and processed with TraceFinder 4.1 and Xcalibur 4.0 (Thermo Fisher Scientific).

Low-resolution mass spectrometry was carried out using an Agilent 6495 Triple Quadrupole system combined with an Agilent 1290 Infinity II LC System (Agilent Technologies) to measure PAs and PANOs. Compounds were ionized by electrospray ionization (ESI) in the positive mode and collision energies were individually adjusted. For identification, three product ions of each analyte were selected in each measurement (Table 3S, Supporting Information). Data were acquired and processed with the Mass Hunter Workstation (Agilent Technologies).

Guidance [33] criteria for recovery (70 – 120%) and intraday precision (≤ 20%) were tested and fulfilled for reference substances in matrices water, n-octanol, and human plasma (also see Table 4S, Supporting Information).

Data availability

All data are available in the GitHub repository https://github.com/al901 010/Supplement_Lipophilicity_BloodPartitoning_PAs.

Contributorsʼ Statement

Conceptualization: A. Lehmann and C. Hethey; Methodology: A. Lehmann, M. Haas, A. These and C. Hethey; Validation: A. Lehmann, M. Haas, A. These and C. Hethey; Formal analysis: A. Lehmann and M. Haas; Investigation: M. Haas and J. Taenzer; Data curation: A. Lehmann; Writing – original draft: A. Lehmann; Writing – review & editing: M. Haas, J. Taenzer, G. Hamscher, C. Kloft, A. These and C. Hethey; Visualization: A. Lehmann; Supervision: G. Hamscher, C. Kloft, A. These and C. Hethey; Project administration: A. These

Conflict of Interest

Charlotte Kloft reports funding grants from an industry consortium (AbbVie Deutschland GmbH & Co. KG, Astra Zeneca, Boehringer Ingelheim Pharma GmbH & Co. KG, Grünenthal GmbH, F. Hoffmann-La Roche Ltd., Merck KGaA, Novo Nordisk, and Sanofi) for the PharMetrX PhD program and from the Innovative Medicines Initiative-Joint Undertaking ('DDMoRe').
The other authors declare that they have no conflict of interest.

Acknowledgements

The authors thank Anja Gessele for her skillful technical support. Open Access funding was enabled and organized by Projekt DEAL.

• References

• 1 Dusemund B, Nowak N, Sommerfeld C, Lindtner O, Schäfer B, Lampen A. Risk assessment of pyrrolizidine alkaloids in food of plant and animal origin. Food Chem Toxicol 2018; 115: 63-72
• 2 Mengbi Y, Ma J, Ruan J, Ye Y, Fu P, Lin G. Intestinal and hepatic biotransformation of pyrrolizidine alkaloid N-oxides to toxic pyrrolizidine alkaloids. Arch Toxicol 2019; 93: 2197-2209
• 3 Widjaja F, Alhejji Y, Rietjens IMCM. The role of kinetics as key determinant in toxicity of pyrrolizidine alkaloids and their N-oxides. Planta Med 2022; 88: 130-143
• 4 Chen L, Peijnenburg A, de Haan L, Rietjens IMCM. Prediction of in vivo genotoxicity of lasiocarpine and riddelliine in rat liver using a combined in vitro-physiologically based kinetic modelling-facilitated reverse dosimetry approach. Arch Toxicol 2019; 93: 2385-2395
• 5 Lehmann A, Geburek I, Hessel-Pras S, Hengstler JG, Albrecht W, Mielke H, Müller-Graf C, Yang X, Kloft C, Hethey C. PBTK modeling of the pyrrolizidine alkaloid retrorsine to predict liver toxicity in mouse and rat. Arch Toxicol 2023; 97: 1319-1333
• 6 Rodgers T, Leahy D, Rowland M. Physiologically based pharmacokinetic modeling 1: Predicting the tissue distribution of moderate-to-strong bases. J Pharm Sci 2005; 94: 1259-1276
• 7 Rodgers T, Rowland M. Physiologically based pharmacokinetic modelling 2: Predicting the tissue distribution of acids, very weak bases, neutrals and zwitterions. J Pharm Sci 2006; 95: 1238-1257
• 8 Schmitt W. General approach for the calculation of tissue to plasma partition coefficients. Toxicol In Vitro 2008; 22: 457-467
• 9 Schmitt W. Corrigendum to: “General approach for the calculation of tissue to plasma partition coefficients” [Toxicol In Vitro 2008; 22:457–467]. Toxicol In Vitro 2008; 22: 1666-1666
• 10 Kah M, Brown CD. LogD: Lipophilicity for ionisable compounds. Chemosphere 2008; 72: 1401-1408
• 11 Aliagas I, Gobbi A, Lee ML, Sellers BD. Comparison of logP and logD correction models trained with public and proprietary data sets. J Comput Aided Mol Des 2022; 36: 253-262
• 12 Soares JX, Santos Á, Fernandes C, Pinto MMM. Liquid chromatography on the different methods for the determination of lipophilicity: An Essential analytical tool in medicinal chemistry. Chemosensors 2022; 10: 340
• 13 OECD. Test No. 107: Partition Coefficient (n-octanol/water): Shake Flask Method. Paris, France: OECD Publishing; 1995
• 14 Bharate SS, Kumar V, Vishwakarma RA. Determining Partition Coefficient (Log P), Distribution Coefficient (Log D) and Ionization Constant (pKa) in Early Drug Discovery. Comb Chem High Throughput Screen 2016; 19: 461-469
• 15 OECD. Test No. 117: Partition Coefficient (n-octanol/water), HPLC Method. Paris, France: OECD Publishing; 2022
• 16 Mannhold R, Poda GI, Ostermann C, Tetko IV. Calculation of molecular lipophilicity: State-of-the-art and comparison of log P methods on more than 96000 compounds. J Pharm Sci 2009; 98: 861-893
• 17 Testa B, Crivori P, Reist M, Carrupt PA. The influence of lipophilicity on the pharmacokinetic behavior of drugs: Concepts and examples. Perspect Drug Discov Des 2000; 19: 179-211
• 18 Paixão P, Gouveia LF, Morais JAG. Prediction of drug distribution within blood. Eur J Pharm Sci 2009; 36: 544-554
• 19 Yu S, Li S, Yang H, Lee F, Wu JT, Qian MG. A novel liquid chromatography/tandem mass spectrometry based depletion method for measuring red blood cell partitioning of pharmaceutical compounds in drug discovery. Rapid Commun Mass Spectrom 2005; 19: 250-254
• 20 Liu XR, Wu KC, Huang Y, Sun JB, Ke XY, Wang JC, Lu WL, Zhang X, Zhang Q. In vitro and in vivo studies on plasma-to-blood ratio of paclitaxel in human, rabbit and rat blood fractions. Biol Pharm Bull 2008; 31: 1215-1220
• 21 Uchimura T, Kato M, Saito T, Kinoshita H. Prediction of human blood-to-plasma drug concentration ratio. Biopharm Drug Dispos 2010; 31: 286-297
• 22 Mamada H, Iwamoto K, Nomura Y, Uesawa Y. Predicting blood-to-plasma concentration ratios of drugs from chemical structures and volumes of distribution in humans. Mol Divers 2021; 25: 1261-1270
• 23 Geburek I, Preiss-Weigert A, Lahrssen-Wiederholt M, Schrenk D, These A. In vitro metabolism of pyrrolizidine alkaloids – Metabolic degradation and GSH conjugate formation of different structure types. Food Chem Toxicol 2020; 135: 110868
• 24 Caron G, Carrupt PA, Testa B, Ermondi G, Gasco A. Insight into the lipophilicity of the aromatic N-oxide moiety. Pharm Res 1996; 13: 1186-1190
• 25 Snoeck E, Jacqmin P, Van Peer A, Danhof M, Ver Donck K, Van Belle H, Woestenborghs R, Crabbé R, Van Gool R, Dupont A, Heykants J. The implications of non-linear red blood cell partitioning for the pharmacokinetics and pharmacodynamics of the nucleoside transport inhibitor draflazine. Br J Clin Pharmacol 1996; 42: 605-613
• 26 Dash RP, Veeravalli V, Thomas JA, Rosenfeld C, Mehta N, Srinivas N. Whole blood or plasma: What is the ideal matrix for pharmacokinetic-driven drug candidate selection?. Future Med Chem 2021; 13: 157-171
• 27 Colson RO. Unexpected results of some simple exercises in equilibrium melting based on experimentally determined partition coefficients. Int Geol Rev 1998; 40: 936-943
• 28 US EPA. Estimation Programs Interface Suite^™ for Microsoft^® Windows, 2019. Accessed 2024-07-15 at https://www.epa.gov/tsca-screening-tools/epi-suitetm-estimation-program-interface
• 29 Merz KH, Schrenk D. Interim relative potency factors for the toxicological risk assessment of pyrrolizidine alkaloids in food and herbal medicines. Toxicol Lett 2016; 263: 44-57
• 30 ARChem LLC. SPARC Automated Reasoning in Chemistry. Accessed 2025-07-15 at http://www.archemcalc.com/sparc.html
• 31 SimulationsPlus. ADMET Predictor 12. ADMET Property Estimation and Model Building, 2024. Accessed 2024-10-29 at https://www.simulations-plus.com/software/admetpredictor/
• 32 Watanabe R, Esaki T, Kawashima H, Natsume-Kitatani Y, Nagao C, Ohashi R, Mizuguchi K. Predicting fraction unbound in human plasma from chemical structure: Improved accuracy in the low value ranges. Mol Pharm 2018; 15: 5302-5311
• 33 EU Reference Laboratories. Guidance document on analytical quality control and method validation procedures for pesticide residues and analysis in food and feed. SANTE/11813/2017. 2017. Accessed October 26, 2023 at: https://www.eurl-pesticides.eu/userfiles/file/EurlALL/SANTE_11813_2017-fin.pdf

Publication History

Received: 29 July 2024

Accepted after revision: 05 January 2025

Accepted Manuscript online:
23 January 2025

Article published online:
21 February 2025