Divergent resistance pathways amongst SARS-CoV-2 PLpro inhibitors highlight the need for scaffold diversity

Abstract

Drug-escape, where a target evolves to escape inhibition from a drug, has the potential to lead to cross-resistance where drugs that are structurally related or share similar binding mechanisms all become less effective. PLpro inhibitors are currently under development and many emerging PLpro inhibitors are derived from GRL0617, a repurposed SARS-CoV PLpro inhibitor with moderate activity against SARS-CoV-2. Two leading derivatives, PF-07957472 and Jun12682, demonstrate low nanomolar activity and display activity in mice. WEHI-P8 is structurally distinct but binds to a similar pocket adjacent to the active site as GRL0617-like compounds. Using deep mutational scanning, we assessed the potential for PLpro to develop resistance to PF-07957472, Jun12682, and WEHI-P8. PF-07957472 and Jun12682 exhibited largely overlapping escape mutations due to their shared scaffold and binding modes, whereas WEHI-P8 resistance mutations were distinct. These findings underscore the importance of developing structurally diverse inhibitors to minimize resistance risks and ensure that viral mutations against one compound do not compromise the efficacy of others.

Fig 1. PLpro inhibitors, cellular assay and DMS workflow.

(A) Structures of PLpro bound to the inhibitors (in yellow). From top to bottom: PF-07957472 (PDB: 9CSY, PLpro in green) [19], Jun12682 (PDB: 8UOB, PLpro in blue) [6], and WEHI-P4 (9CYD, PLpro in red) [26]. The central panel zooms in on these compounds and their surroundings, highlighting key interaction residues. Chemical structure formulae are shown in the right panel. (B) Schematic of the FRET biosensor design [18], which links mClover3 to mRuby3 via a PLpro cleavage motif. Upon excitation of mClover3, energy is transferred to mRuby3, producing a FRET signal; however, cleavage of the linker by PLpro disrupts this transfer, resulting in diminished signal. Created in BioRender. Wu, X. (2025) https://BioRender.com/aopcy0t (C) Dose-response curves showing the cellular efficacy of PF-07957472 (green, top), Jun12682 (blue, middle), and WEHI-P8 (red, bottom). The x-axis represents the log-transformed inhibitor concentration (unit: M), while the y-axis represents the normalized FRET signals. The signals from no PLpro treatment and no inhibitor treatment are normalized to 100 and 0 respectively. The plots are representative experiment of two biological replicates, while reported EC50 values represent the averages obtained from two biological replicates. Error bar: mean ± SD. (D) The DMS workflow. PLpro variants controlled by tet-on and containing a barcode after the stop codon are transduced into 293T cells and sorted based on FRET signal. These populations are harvested for genetic material, which is sequenced by Illumina and scored using DiMSum [29].

Fig 2. Data processing pipeline and identification of drug escape hotspots.

(A) Transformation of DMS data to FRET- ratio. The raw DiMSum fitness scores are converted to FRET ratios using a custom equation (see methods). An inhibition score is then calculated from the FRET- ratios obtained from activity [18], leaky, and drug treatment DMS data. This inhibition score is inversed to calculate an escape score, which is subsequently normalized into a Z-score, based on the distribution of synonymous wildtype variants. Created in BioRender. Wu, X. (2025) https://BioRender.com/zp95rys (B) Comparison of data before (top) and after (bottom) processing. The top panel shows the DiMSum data, while the bottom panel displays data after processing through our pipeline. Variants are plotted by their residue number (x-axis) and mutation (y-axis). Variants with either a DiMSum fitness score over 1 or a Z-score above 2 are highlighted in red. Refer to S4–S6 and S8–S10 Figs for detailed heatmaps. (C) Identification of drug escape hotspots. For each drug treatment, the Z-scores (y axis) are plotted for all variants arranged by residue position. Variants with a Z-score greater than 2 are highlighted in red, indicating drug escape variants. Their corresponding residue positions are labeled on the graph. These scores are also plotted in S8–S10 Figs in sequence-function heatmaps.

Fig 3. Inhibitor potency correlates with Z-score.

(A) The table displays the IC50 values for 17 variants plus wildtype in response to three compounds PF-07957472 (the 2nd column), Jun12682 (the 3rd column), and WEHI-P8 (the 4th column). The IC50 values represent the mean of two independent biological replicates; corresponding dose-response curves are shown in S13–S15 Figs. (B) The heatmap illustrates the log-transformed IC50 fold-change relative to wildtype of the data in a), with a color scale ranging from 0 (white)-2 (red). Rows correspond to different protein variants as labeled, while columns represent the treatment (left-right: PF-07957472, Jun12682, and WEHI-P8). (C) The plots show the relationship between Z-scores with log-transformed IC50 values, for each drug treatment (left-right: PF-07957472, Jun12682, and WEHI-P8). The linear regression is performed in R using “lm” function, with the Pearson coefficients displayed in the top left corner of each plot. The shadow area indicates the 95% confidence interval of the regression.

Fig 4. Biochemical properties of selected variants and inhibitor escape profiles.

(A) Comparison of substrate cleavage speeds of variants with wildtype against different PLpro substrates (as labelled), with data from “no PLpro” control set to 0 and data from wildtype to 100. Two independent biological replicates were measured, with each dot representing the average of three technical replicates. (B) Summary of data in A) where speed relative to wildtype is shown on a blue-white-red gradient scale from 0 to 200%. Wildtype-like scores ranging from 75-125% are shown in white. Variants cleaving faster than wildtype are in red, while those cleaving slower are in blue. (C) Variants present in GISAID colored according to the number of observations on a log scale from 2.0 (i.e., 100 observations, white) to 3.5 (i.e., ~ 3200 observations, red). (D) Activity of circulating PLpro variants as a function of number of sequences in GISAID dataset. (E) Representative thermal shift assay of M208 variants. The data are normalized so that the baseline is 0 and the peak is 100. Three technical replicates are depicted as dots, with error bars indicating mean ± SD. The reported VC50 is the average value from two biological replicates. (F-G) Area scaled Venn diagrams of drug escape variants from DMS data. F) shows the overlap patterns of confident escape variants identified via DMS data. G) shows variants accessible via a single base-pair mutation. PF-07957472 (green); Jun12682 (blue); WEHI-P8 (red). (H) Activity fitness scores from previous work [18] of confident escape variants are plotted by treatment (PF-07957472, green; Jun12682, blue; WEHI-P8, red). The activity scores are normalized so that wildtype has an activity score of 1 and nonsense variants have a score of 0. Violin plots show the distribution of fitness scores of variants within each treatment group, with the short bar indicating the mean value for that group. Pairwise p-values were calculated using Dunn’s test, and p-values below 0.05 are marked with an asterisk to indicate significance. (I) Escape variants observed in GISAID dataset with number of observations plotted on the x-axis and Z-score on the y-axis. Vertical lines delineate the same bins shown in (H).