3.1 Determination of lipophilicity

Experimental lipophilicity of the phthalimide derivatives 3–23 was determined by RP-TLC and UPLC/MS. The retention factors RM (Tables 1AS, 1BS, 2AS, and 2BS in Supplementary Material) of the investigated compounds 3–23 obtained from RP-TLC decreased linearly with increasing methanol concentration in the mobile phase [from 40% to 80% (V/V) in 5% increments]. The experimental relative lipophilicity (RM0) and logkUPLC/MS values are presented in Table 1.

Table 1 The relative lipophilicity (RM0) values for tested compounds 3–23

Based on the RM0, logk values and literature data for reference compounds (Table 3S in Supplementary Material), we obtained equations (Fig. 2) and calculated logPRP-TLC and logPUPLC/MS, respectively. Based on these data, the logP values of investigated compounds obtained using the RP-TLC method were in the range from 2.54 (compound 10) to 6.35 (compound 7), while obtained by UPLC/MS method ranged from 1.11 (compound 14) to 5.62 (compound 5) (Table 2).

Fig. 2

The relationship between logP from literature and experimental RM0 RP-TLC (a chart on the left; R = 0.9163, p < 0.00001, s = 0.4495, F = 57.61) and RM0 UPLC/MS (b chart on the right; R = 0.9260, p < 0.000001, s = 0.5034, F = 138.35)

Table 2 The experimental and computational partition coefficient for the investigated compounds 3–23

In the case of logP experimentally determined by the UPLC/MS method, the influence of substituents and the length of the carbon linker on lipophilicity is not so evident (Table 2). The lowest logP values have compounds with the following imidazole < pyridine < benzimidazole substituent, consistent with in silico results. For the derivatives with benzimidazole, phenyl, and pyridine fragments, the increase of predicted logP and elongation of the carbon linker was also confirmed. However, for derivatives with decahydroisoquinoline, 1,2,3,4-tetrahydroisoquinoline, and 6,7-dimethoxy-1,2,3,4-tetrahydroisoquinoline substituents (compounds 3–8), the effect of these fragments on lipophilicity varied depending on the length of the carbon linker. In these cases, compounds with a three-carbon linker had lower lipophilicity than compounds with a two-carbon linker. Similarly, in experimentally determined logP values by the RP-TLC method, the increase in lipophilicity along with the elongation of the carbon linker is unambiguous in the case of decahydroisoquinoline and phenyl substituents (compounds 5 vs. 8 and 12 vs. 16). However, for derivatives with 1,2,3,4-tetrahydroisoquinoline, 6,7-dimethoxy-1,2,3,4-tetrahydroisoquinoline, benzimidazole (except for compounds 17 and 22) and pyridine fragments, the experimentally determined logPRP-TLC value increases with the elongation of the carbon linker. Moreover, compounds with a 4-methoxy group in phthalimide ring (11, 15, 17–19) showed slightly lower lipophilicity values, determined by RP-TLC and UPLC/MS methods, than their analogs without 4-methoxy substituent (20–23).

The in silico lipophilicity parameter was determined using the following programs: ChemDraw, Marvin, and MarvinSketch, as well as the Swiss ADME website module (Table 2) [18]. These programs use various calculation algorithms to increase the accuracy of logP prediction. Atomistic, topological, and hybrid methods (XLOGP3, MLOGP, Silicos-IT) are used, among others, to calculate in silico logP, and the mean value of the predicted methods used is called logP consensus.

Analysis of the clogP values for the investigated compounds (Table 2), it can be observed that regardless of the length of the carbon linker, the in silico predicted lipophilicity increases in the following order of moieties: imidazole < pyridine < benzimidazole < 1,2,3,4-tetrahydroisoquinoline < 6,7-dimethoxy-1,2,3,4-tetrahydroisoquinoline < phenyl/decahydroisoquinoline. Although logP values calculated using in silico methods generally correlate pretty well with those determined experimentally, one should draw detailed conclusions from them carefully. Possible incorrect predictions may result, among others, from the fact that the programs do not consider differences in the spatial conformation of compounds, which may significantly impact the observed lipophilicity within a specific group of compounds. The statistical analysis revealed that the clogP values obtained with the ChemDraw program showed the best correlation with logP determined by the UPLC/MS method (R = 0.7626, p = 0.000058, Table 3).

Table 3 Correlation between calculated logP and logP values determined by the UPLC/MS method

It is important to note that logP RP-TLC values generally showed a slightly weaker correlation with logP calculated in silico compared with those obtained with UPLC/MS. The reason for the lower correlation between experimental and computational data in RP-TLC may be due to both the limitations of this technique (such as limited separation efficiency and limited sensitivity) and its inability to separate highly polar or ionized compounds. In the case of RP-TLC, the Marvin program provided the best results using the “consensus” option, rejecting one outlier point: compound 14 (R = 0.6490, p = 0.001960, Table 4).

Table 4 Correlation between calculated logP and logP values determined by the RP-TLC method

In the next step, multivariate statistics were performed using PCA to assess if there is any relationship between the experimental logP and computational logP variables. PCA is a chemometric technique that reduces the dimensionality of the original data set by finding the linear combinations of the variables called “principal components.” In the obtained new set, the variables were uncorrelated and ordered so that the first few components retain most of the variation in all original variables. As a result, the plot was obtained as a graphical representation of the PCA’s eigenvalues and proportions of explained variance. Based on the plot (Fig. 3a), it was decided to investigate the first two components (PC1 74.25% and PC2 14.95%), which explained approximately 89.2% of the data set’s variability.

Fig. 3

A screen plot of principal components in principal component analysis (a); PC1–PC2 score plots of logPRP-TLC and logPUPLC/MS and calculated partition coefficients (b); logPRP-TLC; logPUPLC/MS; clogPChemDraw; logPMarvin (Consensus); logPMarvinSketch; logPSwissAdme(XLOGP3) and logP

The obtained results of PCA (Fig. 3b) revealed that calculated logP variables, provided by the Marvin program and the Swiss ADME website module, were highly correlated. Moreover, these data correlated reasonably well with logP obtained by MarvinSketch and ChemDraw programs. The predicted in silico logP did not correlate well with the experimental logP values. However, the analysis shows (Fig. 3b) that using the ChemDraw program to predict logP values in the tested group of compounds would be slightly better.

3.2 Correlation of lipophilicity with PDE10A inhibition activity

In the case of the compounds with benzimidazole moiety, which displayed significant PDE10A inhibitory activity (15, 17–19), it could be observed that the PDE10A inhibition activity [13] increased with increasing lipophilicity determined by the UPLC/MS method, i.e., with an elongation of the linker from one to four methylene groups. But then, despite the increase in lipophilicity associated with the further extension of the linker to five methylene groups, a decrease in inhibitory activity against PDE10A was observed. Moreover, despite slightly lower lipophilicity, compounds 11, 15, and 17–19 with 4-methoxy substituent in phthalimide ring showed higher activity than their analogs without 4-methoxy substituent (20–23).

3.3 Analysis of the influence of drug-likeness parameters on PDE10A activity

In the next step, the drug-likeness properties in line with the Lipinski “rule of five” (RO5) for the final compounds were assessed using the SwissAdme website (SI, Table 4S). The criteria of Lipinski rules are molecular weight (MW) ≤ 500 Da, lipophilicity values (logP) ≤ 5, number of hydrogen bond donors (NHD) ≤ 5, and number of hydrogen bond acceptors (NHA) ≤ 10 [19]. The logP values of the most tested compounds (either experimentally or computationally) meet the criteria for orally available drugs (logP < 5). Considering additionally the molecular weight of the compounds (267.28–410.47, Table 4S, Supplementary Material) as well as the number of hydrogen bond donors (0–1) and hydrogen bond acceptors (3–6), all the studied compounds meet the criteria of Lipinski’s rule of five [19]. This suggests their good bioavailability after oral administration, which is also supported by the fact that they also fulfill Veber rules; thus, they have fewer than ten rotating bonds and a topological polar surface area (TPSA) value below 140 Å2 [20, 21]. According to the BOILED-Egg diagram generated in SwissADME, most of the tested compounds (except compounds 11 and 14) can cross the blood‒brain barrier due to their favorable TPSA (< 79 Å2) and logP (0.4–6.0) values [14]. This is also indicated by most compounds’ relatively low molecular weight (optimum ≤ 360) and the absence of only one hydrogen bond donor [22]. The studied compounds also fulfill the drug-likeness criteria proposed by Ghose, which, in addition to the logP mentioned above and molecular weight, also include molar refraction (optimum 40–130, studied compounds 74.65–119.57) and the total number of heavy atoms (optimum 20–70, studied compounds 20–30) [23].

Analysis of the influence of the physicochemical properties of the studied compounds on the activity toward particular biological targets showed that the compounds with increased potency on PDE10A had significantly higher TPSA values (Fig. 4). For the other properties, no meaningful correlations with activity were found.

Fig. 4

Comparison of TPSA values for active (≥ 40% inhibition in 10−5,5 M) and inactive compounds on PDE10A in Student’s t test. TPSA mean value for inactive compounds 58.98, standard deviation (SD) 10.49; TPSA means value for active compounds 73.44, SD 4.13; p = 0.00787

Several lipophilicity indices have been introduced in recent years, which link a molecule’s lipophilicity or molecular weight with its activity and facilitate the selection of compounds for further research [15]. In this case, we calculated for the active compounds with benzimidazole moiety (15, 17–19, 22) two parameters: ligand lipophilicity efficiency (LLE) and binding efficiency index (BEI). The LLE parameter combines both potency and lipophilicity, whereas the BEI parameter relates to activity and molecular weight.

The LLE parameters of the tested compounds ranged from 0.24 to 2.26 for logPRP-TLC and from 2.24 to 3.80 for logPUPLC/MS, respectively (Table 5). The LLE values for the most active compound 18 were 2.26 and 3.27, respectively, deviating from the optimum of 5–7. The results were similar for the BEI parameter, where the optimum is between 20 and 30, and the tested compounds ranged from 14 to 18, with both compounds 17 and 18 showing the highest value of the BEI parameter (> 17).

Table 5 LLE and BEI parameters were calculated for the most active PDE10A inhibitors and reference drug—papaverine3.4 Permeability assessment

Blood‒brain barrier permeability studies are particularly crucial for compounds that exhibit activity in the central nervous system. In the early drug discovery process, optimization compounds are screened for their oral absorption to modify the physicochemical properties of hit compounds if they display poor permeability through membranes. Parallel artificial membrane permeability assays (PAMPA) have become helpful in predicting in vivo drug permeability [24]. In our case, it was used to predict the blood‒brain barrier permeability of selected, the most potent compound 18. Additionally, we tested nine commercially available drugs as phenobarbital, propranolol, theobromine, barbital, theophylline, mefenamic acid, naphazoline, loratadine, and sulfasalazine and classified their permeability through the blood‒brain barrier as good to average (CNS+) or poor (CNS–), respectively (Table 6).

Table 6 Permeability (logPe) in the PAMPA assay and prediction of the CNS penetration

The PAMPA assay results are consistent with the in silico predictions for the blood‒brain barrier permeability value of compound 18 described above. The results indicate that compound 18 can cross the blood‒brain barrier and interact with PDE10A in the central nervous system.

3.5 Pharmacokinetic in silico prediction

In preliminary studies, the balance between activity and pharmacokinetic parameters is crucial for effective drug candidates [2]. Thus, computational approaches have been developed to optimize pharmacokinetic properties that allow the evaluation of these parameters for newly synthesized compounds. One of these predictive models was used to assess the in silico pharmacokinetic parameters of the investigated compounds 3–23 (for detailed information, please see Tables 5AS and 5BS in Supplementary Material) with a phthalimide core.

The absorption of orally administrated drugs was assessed in the Caco-2 cell prediction model, in which all the compounds showed average predicted values (> 0.90). The obtained prediction results revealed that the investigated compounds 3–23 would be absorbed from the intestinal truck (> 80%) but display relatively poor water solubility. The investigated compounds showed average predicted values in the blood‒brain barrier permeability test ranging from −0.905 to 0.003, except for compound 14, which is poorly distributed to the brain, probably due to good water solubility and low lipophilicity. According to prediction results, most compounds are P-glycoprotein substrates, whereas only a few compounds act as P-glycoprotein inhibitors. CYP3A4 likely metabolizes most investigated compounds but not CYP2D6. Moreover, tested compounds with monoamine cyclic moiety will probably be cytochrome CYP2D6 (3–8, 17–23) inhibitors, whereas compounds with phenyl or pyridine and benzimidazole ring will be cytochrome CYP1A2 (9, 11–13, 15–23) or cytochrome CYP2C19 (9, 10, 12, 16–19, 21–23) inhibitors. In addition, compounds containing a benzimidazole moiety linked to a phthalimide fragment by a butyl or pentyl linker will probably inhibit CYP3A4 and CYP2C9 (18, 19, 23).

3.6 Hemolytic activity

An additional assay was taken to measure the potential toxicity toward mammalian cells, such as erythrocytes, for the most active compound 18. Compounds with good permeability can damage the membrane, causing a structural destabilization of the membrane of red blood cells. Our test monitored the ability to lysis human erythrocytes by estimating the level of released hemoglobin. The results confirmed that 18 is safe for this type of cell, as no hemolysis was detected when used in the concentration range 10−6–10−4 M. Our results follow findings obtained by Bogdanov et al. [25]. They found that isatin derivatives also do not have high hemolytic activity. Moreover, it also corresponds well with the results reported by Nayab et al., who proved that some N-substituted phthalimides showed less than 20% hemolysis [26]. Thus, the lack of toxicity 18 toward this type of human cell suggests that this compound is very promising for further studies.