ADME-optimized algorithm prediction for activity of the CYP2C19 and CYP2D6 variants

To assess the potential functional impact of the rare genetic variants in CYP2C19 and CYP2D6, we used the ADME-optimised algorithm developed by Zhou et al., which is based on the LRT, Mutation Assessor, PROVEAN, VEST3 and CADD models [14]. As shown in Table 2 and Supplementary Table 2, all variants resulted in ADME scores above 0.5, indicating deleterious effects and impaired enzyme function. The ADME scores were converted into percentages representing the activity of the protein variant compared to the reference enzymes CYP2C19.1 and CYP2D6.1 (Table 2), based on a previously determined correlation between ADME scores and the experimentally determined functional in vitro activity of protein variants [14]. The CYP2C19 A297V variant had the highest ADME score of 1, which corresponds to an activity of 0% compared to CYP2C19.1. The R186H and M339T variants had the lowest ADME scores of 0.6, which corresponds to about 20–30% of the activity of CYP2C19.1. Other protein variants had ADME values of 0.8, which corresponds to 0–10% of the activity of CYP2C19.1.

Table 2 Comparison of the results predicted by the ADME optimized algorithm for the CYP2C19 and CYP2D6 variants, and the results obtained in the in vitro experiment

Predictions for all CYP2D6 enzyme variants also showed ADME values ≥ 0.6, indicating impaired CYP2D6 function. Two of the variants, R133C and A305T, were predicted to lack catalytic activity, with ADME scores of 1. The D301N and L314M variants had scores of 0.6 and 0.8, respectively, corresponding to 0–30% of normal activity.

Relative catalytic activity determined for CYP2C19 and CYP2D6 variants.

To assess the catalytic activity of the seven CYP2C19 variants in comparison to CYP2C19.1, we measured the production of 5-hydroxy-omeprazole from omeprazole. As shown in Fig. 1A, six of the variants exhibited significantly lower catalytic activity compared to the CYP2C19.1 enzyme. In particular, the H251Q and A297V variants showed the lowest catalytic activity, while the M339T variant appeared to have normal catalytic activity. The remaining CYP2C19 variants showed low-to-moderate metabolizing capabilities for omeprazole compared to CYP2C19.1.

Fig. 1

Relative catalytic activity by different CYP2C19 (A, omeprazole hydroxylation) and CYP2D6 (B, bufuralol hydroxylation) variants compared to CYP2C19.1 and CYP2D6.1, respectively. Data are presented as average ± SD and is from incubations with cell lysates from three independent expressions in HEK293 cells

The catalytic activities of the four CYP2D6 variants compared to CYP2D6.1 were determined using a bufuralol hydroxylation assay (see Fig. 1B). Two of the variants, R133C and A305T, showed very low activity, while D301N showed almost no catalytic activity. The CYP2D6 variant L314M showed slightly lower activity than CYP2D6.1.

Comparison of in vitro metabolism with prediction by algorithms

The omeprazole hydroxylation capacities of the seven CYP2C19 variants were compared to the predictions made by the ADME algorithm (Table 2). Notably, the H251Q and A297V variants showed minimal discrepancies between the predicted and observed catalytic activities, suggesting that the algorithm effectively captured the functional impacts of these two variants. This observation is consistent with the broader trend, where the algorithm demonstrated a 93% accuracy rate in predicting proteins with 0–10% functionality compared to the reference enzyme variant [14].

For other variants with 20–40% catalytic activity relative to CYP2C19.1, such as F114C, C164G, and K275M, the algorithm's predictive accuracy was notably less reliable. Additionally, the M339T variant, which exhibited near-normal function, was incorrectly predicted by the algorithm to have a deleterious impact. These discrepancies highlight the algorithm's limitations in accurately predicting proteins with intermediate activity levels (20–60% and 70–80% of the reference CYP2C19.1 activity), as reflected by a decline in the ADME score for these variants [14].

For the CYP2D6 variants, the predictions for R133C and A305T aligned closely with the catalytic activities observed in the bufuralol hydroxylation assay (Table 2). The R133C variant was predicted to lack enzymatic activity, and the in vitro assay confirmed this with only 5% of the activity of CYP2D6.1. Similarly, the A305T variant was predicted to be inactive and exhibited 10% of the catalytic activity of the reference enzyme. However, the algorithm was less accurate for two other variants. The D301N variant was predicted to retain 20–30% of CYP2D6.1 activity, but the assay revealed an almost complete loss of function during bufuralol incubation. Conversely, the L314M variant was predicted to have 0–10% activity but demonstrated nearly 60% of the reference CYP2D6.1 activity in the experimental results.

Structural effects of CYP2C19 and CYP2D6 genetic variants

To investigate the diminished metabolic capabilities of the CYP2C19 and CYP2D6 variants, in silico modelling was conducted. The stereoscopic structures at each mutation site were analysed to assess the interactions between the mutated residues and their surrounding atoms (Fig. 2 and Table 3). Using the CYP2C19.1 template, structural analysis revealed that F114 and A297 are integral to the active site of omeprazole, corroborating previous findings [24].

Fig. 2

In silico modelling of substrates binding to CYP2C19.1 and CYP2D6.1 compared to variant enzymes. A Molecular docking shows that omeprazole (in green sticks) interacts with Phe114 and Ala297 (in orange, interactions presented as dashed pseudo-bonds), and forms unfavourable clashes upon Ala297Val mutation (presented as purple arrows and purple pseudo-bonds), whereas the docking site is distant from five other variant positions. B Only Asp301 is predicted to be in close contact with bufuralol (in green sticks). Clashes were observed in all CYP2D6 variant structures, except for Arg133Cys, suggesting that this substitution does not cause significant structural changes. Heme is shown in red sticks. Pseudo-bonds with distance labelled indicate hydrogen bonds formed between the hydrogen bond donor and acceptor

Table 3 The differences in the number of interactions formed between the references CYP2C19.1 and CYP2D6.1 and the different variants residue and the neighboring residues/molecules

The A297V substitution resulted in a significant increase in interactions with neighbouring residues, the heme iron-containing prosthetic group, and omeprazole. Among these interactions, 12 unfavourable clashes were identified, which likely caused residue 297 to be repelled from neighbouring residues and the substrate omeprazole (Fig. 2A, Supplementary Fig. 1). These structural conflicts likely disrupted the original conformation of the CYP2C19 active site, impairing omeprazole binding (Supplementary Fig. 2).

Similarly, the H251Q substitution introduced multiple collisions with a neighbouring helix, leading to a marked reduction in catalytic activity. The high number of collisions may have caused substantial conformational changes in the helices, producing a structure unfavourable for omeprazole binding.

In contrast to A297V, residue F114 interacts directly with omeprazole and neighboring residues (Fig. 2A) and thus stabilizes the binding of omeprazole to the active site. By replacing this phenylalanine to cysteine, this residue no longer interacts with omeprazole, and the original number of interactions between F114 and the neighboring residues is significantly reduced from 74 to 8 (Table 3). This decrease in interactions could destabilize the surrounding residue.

Which could change the orientation of the side chains of the neighboring residues and lower the number of interactions with omeprazole. The overall weaker interaction between C114 and omeprazole thus would be disadvantageous to the binding of omeprazole to the active site of this CYP2C19 variant (Supplementary Fig. 2).

For C164G and M339T, only minimal changes in the interactions were observed (Fig. 2A). However, the significantly lower catalytic activity of C164G compared to M339T suggests that C164 plays a crucial role in favoring omeprazole binding. Similarities of atomic interactions after amino acid replacement in residues 186 and 275 were noted (Fig. 2A). Nevertheless, R186 appears to play a more critical role in favoring omeprazole binding, as shown by the in vitro data, than the exchange for histidine.

For the four variants of CYP2D6, the exchange of amino acids only slightly altered the number of interactions with neighboring residues (Fig. 2B, Table 3). Clashes were predicted for D301N, A305T and L314M, indicating their influence on protein structure (Supplementary Fig. 1). Of note, both D301 and A305 are located in the active center of CYP2D6 [20]. It is therefore to be expected that amino acid exchange of both residues, respectively, impair the stability of substrate binding and corresponds well to the results from the in vitro experiments.

Effects on access and egress channels

Access and exit channels are pathways through which substrates and catalytic products of CYP2C19 and CYP2D6 must pass [22]. These channels provide important information about how substrates bind to the active center and how water and products exit. Since the mutation sites are located far away from the membrane, the influx and efflux channels were modelled and analyzed. Different CYP450s have different channels, and the entry and exit channels of CYP2C19 and CYP2D6 were modelled using MOLEOnline (Fig. 3). For CYP2C19, F114, R186, H251 and A297 are predicted to be near the entry and exit channels (Fig. 3A, B). The amino acid exchanges likely limit the entry of substrate and the exit of water molecules or metabolites. The entry and exit channels of CYP2D6 are also likely to be affected by the D301N and A305T exchanges, Fig. 3C, D).

Fig. 3

Access/egress channels identified in CYP2C19 and CYP2D6. A, B The predicted access and egress channel in CYP2C19 (in cyan) indicates that substrate binding is likely to be affected by mutations at Phe114, Arg186, His251, and Ala297 (in yellow sticks), due to their close proximity, but not by mutations at Cys164, Lys275, and Met399. C, D Similarly, access and egress of CYP2D6 substrate is likely to be affected by mutations at Asp301 and Ala305, but not by mutations at Arg133 and Leu314. Heme is shown in red sticks