Identification of Antiprotozoal Compounds from Buxus sempervirens L. by PLS-Prediction

Various nor-triterpene alkaloids of Buxus (B.) sempervirens L. have shown remarkable in vitro activity against the causative agents of tropical malaria and East African sleeping sickness. To identify further antiprotozoal compounds of this plant, 20 different fractions of B. sempervirens L., exhibiting a wide range of in vitro bioactivity, were analyzed by UHPLC/+ESI-QqTOF-MS/MS. The analytical profiles were investigated by partial least squares regression (PLS) for correlations between the intensity of LC/MS signals, bioactivity and cytotoxicity. The resulting models highlighted several compounds as mainly responsible for the antiprotozoal activity and thus, worthwhile for subsequent isolation. These compounds were dereplicated based on their mass spectra in comparison with isolated compounds recently reported by us and with literature data. Moreover, an estimation of the cytotoxicity of the highlighted compounds was derived from an additional PLS model in order to identify plant constituents with strong selectivity. In conclusion, high levels of antitrypanosomal and antiplasmodial activity were predicted for eight and four compounds, respectively. These include three hitherto unknown constituents of B. sempervirens L., presumably new natural products.


Introduction
Unicellular eukaryotic ("protozoan") parasites cause enormous morbidity and mortality [1]. One of the most widespread and serious protozoan diseases is malaria.
Malaria is caused by protozoans of the genus Plasmodium (P.), which occur in erythrocytes and in liver parenchyma cells depending on their developmental stage. The disease is transmitted by insect vectors, female Anopheles mosquitos. Several humanpathogenic Plasmodium species (P. falciparum, P. vivax, P. ovale, P. malariae, P. knowlesi and P. cynomolgi) cause different clinical manifestations of the disease. The World Health Organization (WHO) estimated the annual malaria incidence at 229 million with a mortality of 409,000 cases in 2019 [2]. The human-pathogenic species P. falciparum (Pf ) is responsible for the majority of deaths. The increasing development of resistance, especially of Pf, complicates the therapy and makes the development of new antimalarial drugs inevitable [3].
Protozoans of the genus Trypanosoma cause very severe infections considered neglected tropical diseases, namely Chagas disease in Latin America (caused by T. cruzi) and human African trypanosomiasis (HAT, "sleeping sickness", caused by T. brucei) in sub-Saharan Africa. Although these infections are widespread and lead to severe and long-lasting illnesses, trypanosomiases still do not receive sufficient attention in drug research and development. This is mainly because they affect the poorest part of the population in already low-income countries. Trypanosoma brucei rhodesiense (Tbr) is the causative agent of 2. Results and Discussion 2.1. Antiprotozoal Activity of Crude Extract and Fractions from B. sempervirens L.
The alkaloids (ALOF) of B. sempervirens L. leaves were enriched from crude dichloromethane extract (GBUS) by acid-base extraction and subsequently separated by CPC into 20 different fractions [12]. The samples were tested in vitro against Pf and Tbr as well as against mammalian control cells (L6 rat skeletal myoblasts) to evaluate the cytotoxicity (Cytotox.) of the fractions ( Table 1).
The dichloromethane extract (GBUS) and the two subfractions LNB (lipophilic and neutral constituents) and ALOF (alkaloids) showed clear differences in their antiprotozoal activity in the in vitro assays ( Table 1). The alkaloid fraction very clearly displayed the highest activity against both pathogens, Pf and Tbr, as well as considerable antiplasmodial selectivity with a selectivity index (SI) of 34. The antiplasmodial activity of the alkaloid fraction was most pronounced with an IC 50 value of 0.99 µg/mL. The growth inhibition against Tbr with an IC 50 value of 5 µg/mL was weaker but still interesting. The strong increase of activity and selectivity of the alkaloid rich fraction (ALOF) in comparison with the crude extract and, particularly, with the lipophilic and neutral fraction (LNB) implied that the alkaloids are mainly responsible for the antiprotozoal activity of the dichloromethane extract of B. sempervirens L. Therefore, the alkaloid fraction was fractionated by CPC for further study.
The 20 CPC fractions displayed a wide range of activities (Table 1) in the antiprotozoal in vitro tests while the cytotoxicity against L6 rat-skeletal myoblasts, also tested as a marker for selectivity, was generally rather low and did not vary to a significant degree, with IC 50 values ranging from 12 µg/mL in case of fraction 4 to 55 µg/mL in case of fraction 20.
Regarding activity against Tbr, fraction 2 was the most active with an IC 50 value of 0.63 µg/mL. Fraction 3 and fraction 12 also exhibited IC 50 values below 1 µg/mL. Fraction 18 was the least active against Tbr with an IC 50 value of 39.3 µg/mL.
Several of the fractions (fractions 2, 3, 4, 5, 7-13) displayed a high level of antiplasmodial activity against Pf with IC 50 values below 1 µg/mL. Fraction 2 was the most potent fraction also in this case, with an IC 50 value of 0.35 µg/mL. In addition, this fraction had a high SI of 43 against Pf. For the other mentioned active fractions, the concentration range of antiplasmodial IC 50 values was also distinctly below that of cytotoxicity (all SI > 10). This suggests a selective inhibitory effect of the Buxus-alkaloids, as opposed to an unspecific cytotoxic effect. 0.01 ± 0.004 GBUS: crude dichloromethane extract of B. sempervirens L. leaves; LNB: lipophilic and neutral fraction; ALOF: alkaloid fraction; SI: selectivity index. All IC 50 values are expressed in µg/mL. While * n = 3 was reported as the mean value from three independent measurements with the standard deviation, the other values were determined with n = 2 as the mean value from two independent measurements with the fluctuation range and n = 1 for the cytotoxicity data.

PLS Modeling
Correlations between analytical fingerprint data of complex mixtures, such as extracts and fractions and their bioactivity data can be investigated by PLS regression and the resulting models used to identify constituents mainly responsible for the biological effect under study [8][9][10]. In order to apply such an approach to the present set of fractions, they were all analyzed by LC/MS ( Figure S1, Supplementary Materials) and the fingerprint data were processed using the software DataAnalysis 4.1 and ProfileAnalysis 2.1 (Bruker Daltonik GmbH, Bremen, Germany). The resulting data matrix ("bucket table") of variables representing MS signal intensity within defined retention time and m/z value intervals consisted of 1420 [m/z: t R ] variables × 20 analyses. These data constituted the X-matrix (independent variables) while the respective negative logarithmic IC 50 values (pIC 50 ) of the fractions were treated as the Y-matrix (dependent variables) to generate PLS regression models in SIMCA 16.0.1 (Sartorius Stedim Data Analytics AB, Umeå, Sweden). Three different PLS models were calculated, namely, for the activity against Pf and Tbr as well as for cytotoxic activity against L6 rat skeletal myoblasts used as mammalian control cells to assess selectivity. High values for R 2 (coefficient of determination of values predicted during model calibration) and, particularly, for Q 2 (coefficient of determination of values predicted in leave-one-out cross validation) obtained with few PLS components in each case indicated that stable models with good internal predictive power had been generated ( Table 2). The PLS model for antiplasmodial activity consisted of two PLS components and yielded R 2 and Q 2 values of 0.9 and 0.82, respectively ( Table 2). Scores and loadings plots are presented in Figure 1. In the scores plot of the PLS, the different fractions from B. sempervirens L. are grouped along their PLS components, which form the axes of the coordinate system. The influence of individual X variables ("buckets") on the samples' positions in the scores plot is reflected in the loadings plot. For instance, if a fraction is located in the upper right quadrant of the scores plot (high values on both, the first and the second PLS component), the buckets responsible for the position of this fraction are found in this area of the loadings plot. At the same time, these variables are closest to the position of the Y variable, which means that they are most strongly correlated with activity.
constituted the X-matrix (independent variables) while the respective negative logarithmic IC50 values (pIC50) of the fractions were treated as the Y-matrix (dependent variables) to generate PLS regression models in SIMCA 16.0.1 (Sartorius Stedim Data Analytics AB, Umeå, Sweden). Three different PLS models were calculated, namely, for the activity against Pf and Tbr as well as for cytotoxic activity against L6 rat skeletal myoblasts used as mammalian control cells to assess selectivity. High values for R 2 (coefficient of determination of values predicted during model calibration) and, particularly, for Q 2 (coefficient of determination of values predicted in leave-one-out cross validation) obtained with few PLS components in each case indicated that stable models with good internal predictive power had been generated (Table 2). The PLS model for antiplasmodial activity consisted of two PLS components and yielded R 2 and Q 2 values of 0.9 and 0.82, respectively ( Table 2). Scores and loadings plots are presented in Figure 1. In the scores plot of the PLS, the different fractions from B. sempervirens L. are grouped along their PLS components, which form the axes of the coordinate system. The influence of individual X variables ("buckets") on the samples' positions in the scores plot is reflected in the loadings plot. For instance, if a fraction is located in the upper right quadrant of the scores plot (high values on both, the first and the second PLS component), the buckets responsible for the position of this fraction are found in this area of the loadings plot. At the same time, these variables are closest to the position of the Y variable, which means that they are most strongly correlated with activity.   Figure 2).

Identification of the Antiplasmodial Compounds Highlighted by the PLS Model
In the loadings plot, four bucket variables are emphasized as mainly important for the antiplasmodial activity of the alkaloid fractions ( Figure 2). In order to assign and identify these compounds, the chromatographic and mass spectral data of the LC/MS measurements were evaluated ( Table 3). The +ESI-QqTOF MS/MS spectra corresponding to the highlighted variables were examined in particular for the occurrence of prominent core fragments, which initially provide information about the substitution present and the occurrence of double bonds in the structures of the compounds. In previous studies, characteristic core fragments of Buxaceae alkaloids have already been identified for the amino-/amidosteroids in Pachysandra [8] and Sarcococca [13]. For the nor-triterpene alka-loids of the nor-cycloartane type, occurring in the genus Buxus, a few prominent fragments were also discovered, which can be used for structural assignment [14]. These characteristic diagnostic fragments were supplemented in the present work by further ones as shown in Figure 3.
the antiplasmodial activity of the alkaloid fractions ( Figure 2). In order to assign and identify these compounds, the chromatographic and mass spectral data of the LC/MS measurements were evaluated ( Table 3). The +ESI-QqTOF MS/MS spectra corresponding to the highlighted variables were examined in particular for the occurrence of prominent core fragments, which initially provide information about the substitution present and the occurrence of double bonds in the structures of the compounds. In previous studies, characteristic core fragments of Buxaceae alkaloids have already been identified for the amino-/amidosteroids in Pachysandra [8] and Sarcococca [13]. For the nor-triterpene alkaloids of the nor-cycloartane type, occurring in the genus Buxus, a few prominent fragments were also discovered, which can be used for structural assignment [14]. These characteristic diagnostic fragments were supplemented in the present work by further ones as shown in Figure 3.  Figure 1) with the buckets that, according to the PLS model, have a high impact on the antiplasmodial activity of B. sempervirens L. * Structural assignment based only on mass spectral data; stereochemistry tentatively assumed in analogy with known compounds [7]. Table 3. LC/MS-characteristics of the compounds with predicted antiplasmodial activity (compare Figure 2).

Bucket
Adduct   4), respectively, both isolated in our previous study and proven to be highly active against Pf in the in vitro assay (IC50: 1.05 µM (1); IC50: 1.76 µM (4)) [7]. The finding of these two compounds with already proven antiplasmodial activity can be taken as evidence for the validity of the calculated PLS model.  Figure 1) with the buckets that, according to the PLS model, have a high impact on the antiplasmodial activity of B. sempervirens L. * Structural assignment based only on mass spectral data; stereochemistry tentatively assumed in analogy with known compounds [7]. Table 3. LC/MS-characteristics of the compounds with predicted antiplasmodial activity (compare Figure 2).

Bucket
Adduct  The occurrence of a core fragment at m/z 323 allowed assignment of compound 2 (4.33 min: 501.412 m/z) as a C-3/C-20 diamine with an additional single substitution or unsaturation in the core skeleton ( Figure 3). The fragment m/z 401 indicated the neutral loss of a C5-acid moiety (100 Da; a tiglic acid ester is assumed in analogy to compound 1 from the [M + H] + ion of 2. The presence of two monomethylated amino groups at C-3 and C-20 was evident from the dual neutral losses of 31 Da (m/z 385  354  323) ( Figures S4  and S5, Supplementary Materials). The detailed fragmentation pathway is reported in  4), respectively, both isolated in our previous study and proven to be highly active against Pf in the in vitro assay (IC 50 : 1.05 µM (1); IC 50 : 1.76 µM (4)) [7]. The finding of these two compounds with already proven antiplasmodial activity can be taken as evidence for the validity of the calculated PLS model.
The occurrence of a core fragment at m/z 323 allowed assignment of compound 2 (4.33 min: 501.412 m/z) as a C-3/C-20 diamine with an additional single substitution or unsaturation in the core skeleton ( Figure 3). The fragment m/z 401 indicated the neutral loss of a C 5 -acid moiety (100 Da; a tiglic acid ester is assumed in analogy to compound 1 from the [M + H] + ion of 2. The presence of two monomethylated amino groups at C-3 and C-20 was evident from the dual neutral losses of 31 Da (m/z 385 → 354 → 323) ( Figures S4 and S5, Supplementary Materials). The detailed fragmentation pathway is reported in Figure S6, Supplementary Materials. The structure, thus determined from the mass spectral data (Figure 2), could not be found in the literature, so that we assume, to the best of our knowledge, that compound 2 is a new natural product. It should be noted that the stereochemistry of 2, which is not accessible on grounds of the mass spectra alone, is assumed in analogy with related compounds of known configuration (e.g., reported in [7]). The generic name of the new compound, O-tigloylcyclomicrophylline-D, was chosen based on the existing classification for Buxus-alkaloids [15,16].
Compound 3 (14.73 min: 310.312 m/z) displayed a completely different chromatographic and mass spectral behavior from the Buxus-alkaloids (+ESI-QqTOF MS/MS spectra see Figures S7 and S8, Supplementary Materials). Furthermore, the molecular formula C 20 H 39 NO with only two double bond equivalents is not compatible with a triterpenoid alkaloid and does not match any other known constituent from B. sempervirens L. According to the data, it could be an oleamide structure, but this was not investigated further in this study. The identification and clarification of this compound's contribution to the antiplasmodial activity must await its isolation in subsequent studies.

PLS-Prediction of Antitrypanosomal Compounds from B. sempervirens L.
The PLS model for antitrypanosomal activity is composed of four PLS components and shows good statistical performance with R 2 and Q 2 values of 0.98 and 0.89, respectively ( Table 2). The resulting scores and loadings plots are depicted in Figure 4.   Figure 5).

Identification of the Antitrypanosomal Compounds Highlighted by the PLS Model
In the loadings plot of the PLS model, eight buckets were located in the upper right quadrant and thus predicted to be mainly responsible for the antitrypanosomal activity  Figure 5). a box in the upper right quadrant are predicted to possess high antitrypanosomal activity by the PLS model (Buckets marked and labeled in green are detailed in Figure 5).

Identification of the Antitrypanosomal Compounds Highlighted by the PLS Model
In the loadings plot of the PLS model, eight buckets were located in the upper right quadrant and thus predicted to be mainly responsible for the antitrypanosomal activity ( Figure 5). The data of the LC/MS analysis of the underlying compounds are listed in Table 4.

Identification of the Antitrypanosomal Compounds Highlighted by the PLS Model
In the loadings plot of the PLS model, eight buckets were located in the upper right quadrant and thus predicted to be mainly responsible for the antitrypanosomal activity ( Figure 5). The data of the LC/MS analysis of the underlying compounds are listed in Table 4. Table 4. LC/MS-characteristics of the compounds with predicted antitrypanosomal activity (compare Figure 5). Cyclomicrophylline-A (4) (3.09 min: 445.388 m/z) was also found as an active compound in the model against Pf (see Section 2.2.2 above, Figure 2, Table 3) and had already demonstrated activity in the in vitro assay against Tbr with an IC 50 value of 2.3 µM [7]. In addition, compound 11 (3.55 min: 415.375 m/z) could be assigned ( Figures S28-S30, Supplementary Materials) to cyclovirobuxeine-B, which had also been isolated and displayed promising antitrypanosomal activity (IC 50 : 1.5 µM) in our previous study [7]. These findings, as in the case of the antiplasmodial model above, corroborate the validity of the calculated model.

Bucket
Compound 8 (2.30 min: 443.371 m/z) exhibited analytical data analogous to compound 11 (Figures S20-S22, Supplementary Materials). The structural formula C 28 H 46 N 2 O 2 derived from the [M + H] + ion (m/z 443) differed in the occurrence of additional oxygen and carbon atoms in comparison to 11. According to these mass spectral data compound 8 could be assigned to the known Buxus-alkaloid N-formylcyclovirobuxeine-B. This alkaloid has already been isolated from Buxus malayana Ridl. [17]. Antitrypanosomal activity of this compound has not been reported so far.
Compound 6 (4.67 min: 483.402 m/z) showed a characteristic fragmentation pattern (Figures S14-S17, Supplementary Materials) almost identical to compound 1. In contrast to O-tigloylcyclovirobuxeine-B (1), the presence of a primary amino group could be concluded in the structure of 6 from the neutral loss of NH 3 observed in the fragment at m/z 466 [483-NH 3 ] + . The detailed fragmentation pathway is reported in Figure S17, Supplementary Materials. Based on the fragmentation, it could not be determined whether the primary amino group is present at position 3 or 20. To the best of our knowledge, neither structure has been described in the literature so that 6 certainly represents a new natural product. In accordance with the systematic naming of such alkaloids [15,16], this compound must be either O-tigloylcyclovirobuxeine-E or F (E = R 1 : N(CH 3 ) 2 ; R 2 : NH 2 ; F = R 1 : NH 2 ; Although the cytotoxic activity of the alkaloid fractions with good activity against the protozoan parasites was usually rather low and their selectivity indices high, especially in the case of antiplasmodial activity, it was nevertheless interesting to create a PLS model for the cytotoxicity data of the alkaloid fractions. In this case, a model consisting of three PLS components resulted, which yielded good statistical quality with an R 2 of 0.96 and a Q 2 of 0.83 ( Table 2). The scores and loadings plots ( Figure 6) showed the distribution of fractions and buckets according to their effect on cytotoxicity, respectively.
Of the 11 compounds with predicted antiprotozoal activity, the cytotoxicity model included compounds 7 (2.51 min: 302.251 m/z), 9 (4.71 min: 354.321 m/z), 10 (4.74 min: 385.364 m/z) and 11 (3.55 min: 415.375 m/z), which were identified as the four compounds with the greatest impact on cytotoxicity (Figure 7). This might suggest that the antitrypanosomal activity of these four compounds could be due to an unspecific cytotoxic effect rather than a selective impact on the parasites. However, the cytotoxicity of compound 11 (3.55 min: 415.375 m/z) against the same L6 cell line was already tested in our previous study [7] and it was much lower than the antitrypanosomal activity (IC 50 value against Tbr of 1.5 µM vs. 35.5 µM for cytotoxicity against L6 cells; SI: 24). With some caution, this result may probably be extrapolated to the other three compounds dominating the PLS model for cytotoxicity. Thus, it can be concluded that even if these four ingredients have a significant influence on the cytotoxicity of the boxwood leaves, they may still possess a selective antiprotozoal effect since the overall cytotoxicity is rather low. the protozoan parasites was usually rather low and their selectivity indices high, especially in the case of antiplasmodial activity, it was nevertheless interesting to create a PLS model for the cytotoxicity data of the alkaloid fractions. In this case, a model consisting of three PLS components resulted, which yielded good statistical quality with an R 2 of 0.96 and a Q 2 of 0.83 ( Table 2). The scores and loadings plots ( Figure 6) showed the distribution of fractions and buckets according to their effect on cytotoxicity, respectively.  This might suggest that the antitrypanosomal activity of these four compounds could be due to an unspecific cytotoxic effect rather than a selective impact on the parasites. However, the cytotoxicity of compound 11 (3.55 min: 415.375 m/z) against the same L6 cell line was already tested in our previous study [7] and it was much lower than the antitrypanosomal activity (IC50 value against Tbr of 1.5 µM vs. 35.5 µM for cytotoxicity against L6 cells; SI: 24). With some caution, this result may probably be extrapolated to the other three compounds dominating the PLS model for cytotoxicity. Thus, it can be concluded that even if these four ingredients have a significant influence on the cytotoxicity of the boxwood leaves, they may still possess a selective antiprotozoal effect since the overall cytotoxicity is rather low.

Plant Material
The collection, identification and other details of the plant material used in this study were reported previously [12].

Plant Material
The collection, identification and other details of the plant material used in this study were reported previously [12].

Extraction and Fractionation of B. sempervirens L. Leaves
The extraction of plant material, the acid-base extraction to obtain the alkaloid fraction (ALOF) and the separation of ALOF into 20 different fractions by centrifugal partition chromatography (CPC) was described previously [12].

UHPLC/+ESI-QqTOF-Mass Spectrometry
LC/MS measurements for multivariate data analysis: The previously described method and parameters were used to obtain these data [12]. The 20 fractions were analyzed in a continuous sequence to obtain optimal comparability of retention and intensity data as needed for multivariate data analysis. The sample concentration of all fractions was 2 mg/mL. LC/MS measurements to identify constituents of particular interest: To investigate in more detail the characteristic MS fragmentation of several constituents with predicted antiprotozoal activity (i.e., after analysis of the PLS models), a separate experiment was performed in MRM mode, for which the respective [M + H] + ion (deviation m/z ± 10) was selected beforehand. The collision energy was 40 eV. Based on these spectra, fragmentation pathways were reconstructed ( Figures S6, S11, S17, S22, S27 and S30, Supplementary Materials). The mass deviation of the fragments taken into consideration was <40 ppm in all cases.
For further processing, the obtained molecular features were imported into the software ProfileAnalysis 2.1 (Bruker Daltonik GmbH, Bremen, Germany). A descriptor table ("bucket table") was created in which each bucket value corresponds to mass signal intensity at a given m/z value and a given retention time [min]. The following parameters were applied: data selection and processing: find molecular features, retention time range: 1-15 min, mass range: 100-1500 m/z, advanced bucketing: width of retention time window 0.1 min; width of m/z window 30 ppm, split buckets with multiple compounds: yes, bucket filter: value count of bucket ≥ 10%, allow empty group attributes: yes, bucket value transformation: none and display bucket values in table: yes. The resulting bucket table consisted of 1420 buckets and served as a data matrix for the calculation of the following statistical models.

PLS Modeling
The bucket table (1420 bucket variables × 20 analyses) was imported to the software SIMCA 16.0.1 (Sartorius Stedim Data Analytics AB, Umeå, Sweden) and constituted the X-matrix (independent variables) of the PLS model. The bioactivity data of all fractions were transformed to logarithmic scale (pIC 50 = −log IC 50 [µM]) and used as the Y-matrix (dependent variables). Three different PLS models were generated with the activity data against Pf and Tbr, respectively, as well as the cytotoxicity values against mammalian cells (L6 cell line). All variables were scaled to unit variance (scaling factor: standard deviation) in order to assign equal weights to every variable in the model. The leave-one-out crossvalidation was used to fully cross-validate the resulting models.
Buckets with a value ≥ 0.09 for w*c(1) (combined loading weight vector of X and Y for the first PLS component) and a w*c(2) ≤ 0.41 (combined loading weight vector of X and Y for the second PLS component) (pIC 50 Pf 0.145381 w*c(1) and 0.237369 w*c(2)) were analyzed in the loadings plot of the PLS model with prediction of antiplasmodial compounds (Figures 1 and 2). In the antitrypanosomal PLS model buckets with a w*c(1) ≥ 0.08 and a w*c(2) ≤ 0.13 (pIC 50 Tbr 0.121195 w*c(1) and 0.119258 w*c(2)) were evaluated (Figures 4 and 5).

In Vitro Bioassays
In vitro assays for the bioactivity of the 20 CPC fractions against Tbr (bloodstream trypomastigotes, STIB 900 strain) and Pf (intraerythrocytic forms, NF54 strain), and cytotoxicity tests against mammalian cells (L6-cell line from rat-skeletal myoblasts), were performed according to established standard protocols [18].

Conclusions
In conclusion, PLS models with good statistical performance were obtained with the LC/MS-data of 20 alkaloid fractions from the leaf extract of B. sempervirens L. for their antiplasmodial and antitrypanosomal activity as well as their cytotoxicity against mammalian cells. Using these models, antiprotozoal activity was predicted for eleven constituents. Of the four compounds responsible for the antiplasmodial activity, compound 1 and 4 have already been tested for their growth inhibition against Pf in the in vitro assay and exhibited promising IC 50 values (IC 50 : 1.05 µM (1); IC 50 : 1.76 µM (4)). Compound 2 represents a previously undescribed natural product. This new Buxus-alkaloid is of interest for subsequent isolation and testing, especially because it is not predicted by the respective PLS model to significantly contribute to cytotoxicity.
Among the eight compounds highlighted by the PLS model for antitrypanosomal activity, compound 4 and 11 have already been tested against Tbr and displayed reasonable activity (IC 50 : 2.3 µM (4); IC 50 : 1.5 µM (11)). Moreover, two new natural products, compound 6 and 10, were predicted to be effective against Tbr. Compound 8 could be dereplicated as N-formylcyclovirobuxeine-B which, to the best of our knowledge, is described for the first time as a constituent of B. sempervirens L. Since they did not significantly contribute to the PLS model for cytotoxicity, compounds 6 and 8 are the most interesting compounds for further studies in terms of antitrypanosomal Buxus-alkaloids.
The present study thus represents a new, impressive example showing that the chosen PLS approach is an auspicious means of identifying active natural products in a timeand resource-saving manner that can rationally guide the subsequent isolation of such compounds. Since the concentrations of compounds 2, 6, 8 and 10 in their respective fractions were quite low, their isolation will have to start from a larger amount of plant material which can now be performed in a target-oriented manner.