Assays for Measuring the Cell Permeability of Proteolysis-Targeting Chimeras (PROTACs): Performance, Correlations, Applicability and Recommendations Authors: Christina Brenner, Yunhai Cui, Denis Schmidt, Johannes Kirchmair, Lu Tan, Nina Braun Published: 2026 ABSTRACT Most Proteolysis-Targeting Chimeras (PROTACs) are beyond-Rule-of-Five compounds, and optimizing their ability to cross cell membranes is a central topic in PROTAC therapeutic research. In this study, we systematically investigate the performance and assay correlations of PROTAC cell permeability assays and the chemical space in which these assays provide reliable results. Our work builds on a comprehensive data set of 3271 PROTACs spanning diverse chemical and target spaces, providing unprecedented coverage of PROTAC cell permeability data. These compounds were measured in up to three different types of assays: the NanoBRET E3 ligase target engagement assay, the Caco-2 assay, and a P-glycoprotein (P-gp) efflux liability assay. We find a strong correlation between readouts from the NanoBRET E3 ligase target engagement assay (availability shift) and those from the Caco-2 assay (apparent intrinsic permeability, basolateral-to-apical permeability, and efflux ratio), while P-gp liability derived from the P-gp assay cannot easily be translated to the Caco-2 efflux ratio. In the process, we also investigate the impact of physicochemical properties included in the Rule-of-Five on the applicability of the assays. Based on these findings, we derive guidelines and data-driven recommendations for the use of cell permeability assays in PROTAC optimization. While this study focuses on PROTACs, some of the observations may be relevant to other modalities with beyond-Rule-of-Five physicochemical properties. Introduction Proteolysis-Targeting Chimeras (PROTACs) are bifunctional molecules that induce the formation of a ternary complex between the PROTAC, a protein of interest (POI) and an E3 ligase, leading to subsequent ubiquitination and targeted degradation of the POI through the ubiquitin–proteasome system (UPS, Figure A). − PROTACs and their characterization using cell permeability assays. (A) PROTAC mode of action: Interaction between the PROTAC, a protein of interest (POI, blue) and an E3 ligase (orange) enables the transfer of ubiquitin molecules (Ub) from a ubiquitin-conjugating enzyme (E2) onto the POI. Multiple ubiquitination steps lead to recognition and subsequent proteasomal degradation of the POI. (B) The Caco-2 assay serves as an in vitro absorption model for PROTACs and can be performed either with or without an additional incubation step. PROTACs are added to either side of a Caco-2 cellular monolayer, and their capability to permeate through this layer is quantified. (C) The NanoBRET assay is used to study E3-ligase engagement of PROTACs in intact and permeabilized cells. The ratio of the two resulting luminescence signals is called availability shift and serves as a measure of intracellular availability. PROTACs with a low availability shift show comparable potency in both conditions (blue), whereas those with a high availability shift are less potent in intact cells, indicating decreased permeability (orange). (D) The P-gp liability of PROTACs is assessed using a comparative CellTiter-Glo assay in cells expressing high and low levels of P-gp. The P-gp index is expressed as the ratio of IC50 measured in the high and low P-gp expression conditions. PROTACs without P-gp liability show comparable potency in both conditions (blue), while those with a high P-gp index are less potent in cells expressing high levels of P-gp, an indication of their P-gp liability (orange). PROTACs offer a range of unique properties that make them a promising new drug modality. While the target proteins are degraded by the UPS, the PROTACs themselves remain intact and can act in a substoichiometric fashion, offering favorable opportunities with respect to required dose, dosing frequency and toxicity compared to classical small-molecule inhibitors. , Moreover, PROTACs can interact with proteins at less conserved binding sites (e.g., allosteric pockets) rather than at largely conserved functional motifs, making them more selective for specific target proteins. Collectively, these properties open new avenues for proteins that were previously considered “undruggable”. , Current pharmaceutical development and clinical candidates are largely based on PROTACs binding to two distinct E3 ligases: the Cereblon (CRBN) and von Hippel-Lindau (VHL) ligases. ,− These enzymes are particularly promising for PROTAC development because they are both highly expressed in many tissues and have been extensively studied in drug discovery. Research efforts have been fueled by the early discovery of specific and potent ligands, such as immunomodulatory imide drugs (e.g., thalidomide and its derivatives ,, ) targeting CRBN and the peptidomimetic VH032 motif and its more drug-like follow-ups (e.g., VH298) for VHL. Both examples closely model the natural ligands of the two E3 ligases and have been used as starting scaffolds for PROTAC design. From a chemical perspective, the structural complexity of PROTACs is substantially higher than for classic small-molecule drugs. Most PROTACs violate one or more definitions of Lipinski’s Rule-of-Five, an indicator of oral bioavailability. ,, Herein lies one of the most challenging aspects of PROTAC research to date: designing PROTACs with favorable pharmacokinetic properties. One key pharmacokinetic property defining the in vivo efficacy of PROTACs is cell permeability. ,, Several in vitro systems for assessing cell permeability are established in pharmaceutical research. One of the most widely employed cell permeability tests is the Caco-2 assay, , which uses human colon cancer cells (Caco-2) to mimic the intestinal barrier (Figure B). The Caco-2 assay can account for both passive diffusion and for active export of molecules, e.g., by multidrug resistance proteins such as the P-glycoprotein (P-gp). , Although the handling requirements and extended cell culture times limit its suitability for large-scale screening and automation, the Caco-2 assay remains the gold standard for assessing the cell permeability of drug candidates, including PROTACs. − However, when working with beyond-Rule-of-Five compounds, some adjustments to the standard Caco-2 protocols are recommended, such as lower dosing and additional preincubation steps with analytes when using highly lipophilic compounds in order to compensate for nonspecific binding. ,, A relatively new system for assessing the intracellular availability of PROTACs is the NanoBRET target engagement assay (Figure C). , This assay system can be used to both quantify protein–protein interactions (PPI) by measuring spatial proximity of molecules via bioluminescence resonance energy transfer (BRET) and to study compound permeability. In the case of PROTAC permeability specifically, cells express an E3 ligase-nanoluciferase fusion construct (luminescence donor) that interacts with a known ligand modified with a suitable fluorescent dye (luminescence acceptor), creating a baseline signal. Displacing the luminescence acceptor with a (nonfluorescent) PROTAC results in a loss of BRET signal and indicates the degree of target engagement and thereby intracellular PROTAC availability. By measuring in both intact and permeabilized cells, the relative cellular permeability of PROTACs can be determined: The availability shift (IC50,intact/IC50,permeabilized) is inversely proportional to the intracellular availability of the PROTAC. A key aspect of intracellular availability is efflux, which is often mediated by the efflux transporter P-glycoprotein (P-gp). − Upregulation of P-gp is a common resistance mechanism and has previously been shown to limit PROTAC efficacy. , To investigate a PROTAC’s liability to P-gp-mediated efflux, phenotypic assays such as paired viability assays can be used to compare PROTAC efficacy in cells expressing high or low P-gp levels (Figure D). However, as we will show in this study, this indirect readout of P-gp impact on PROTAC efficacy does not directly translate into differences in Caco-2 efflux ratios, limiting its use for predicting this aspect of PROTAC permeability. This work sets out to systematically identify and investigate connections in three relevant assays for measuring the cell permeability of PROTACs: the Caco-2 assay, NanoBRET target engagement assay and P-gp liability assay. Identifying correlations among assay readouts can support the optimization of development pipelines, accelerating the identification of promising compounds. Simultaneously, the impact of molecular properties on cell permeability can be better understood and further applied in PROTAC design. Our study is based on a diverse set of 3271 PROTACs investigated at Boehringer Ingelheim, providing a broad range of assay setups, assay data abundance, and coverage of a wide chemical space. The goal of this work is to advance the understanding of PROTAC properties and their permeability in different cell assays. We also provide insights into assay applicability and workflow design, as well as recommendations to support efficient processes in PROTAC discovery pipelines. Experimental Section NanoBRET Target Engagement Assay The NanoBRET TE Intracellular E3 Ligase Assay kit (Promega Corporation) was used for measuring target engagement. The assay protocol follows the supplier’s guidelines and recommendations. HEK293 and MIA PaCa-2 cells (both ATCC) stably expressing either a VHL-NanoLuc or NanoLuc-CRBN fusion were cultured in Falcon 75 cm2 Rectangular Canted Neck Cell Culture Flasks (Falcon) and passaged twice a week for up to 25 passages using the following conditions: HEK293 cells were maintained in EMEM (Sigma-Aldrich) supplemented with GlutaMax (1x, Gibco), 1 mM sodium pyruvate (Gibco) and 10% fetal calf serum (HyClone), while MIA PaCa-2 cells were grown in DMEM (Sigma-Aldrich) with 10% fetal calf serum (HyClone). Cells were detached from flasks using Trypsin/EDTA (PAN-Biotech) or TrypLE Express (Gibco) and seeded into 384 well cell culture microplates (polystyrene, F-Bottom, white (Greiner)) at a density of 7000 cells per well in 60 μL Opti-MEM (Gibco) and incubated overnight for VHL assays, or at 10000 cells in 40 μL for direct use in the assay for CRBN assays. PROTACs (10 mM stock in DMSO) were added using an Access Laboratory Workstation with Echo 555 Liquid Handler (Beckman Coulter Life Sciences) to yield final concentrations of 0.1 nM to 10 μM in 0.5% DMSO. All dose response curves were measured in technical duplicates with negative controls (0.5% DMSO) added to each plate, and all plates were duplicated for parallel assessment of intact and permeabilized cells. For CRBN assays, CRBN Tracer (3 μL, final concentration 350 nM, Promega) was added directly after compound addition and cells were incubated for 2 h. For VHL assays, compounds were incubated with the cells for 2 h, after which an in-house VHL tracer (3 μL, final concentration 240 nM) was added and incubated for an additional 2 h. For the permeabilized assay mode, 5 μL digitonin (final concentration 44 μg/mL, Sigma-Aldrich) was added. The plates were shaken for 5–10 min at room temperature, followed by the addition of 25 μL NanoBRET Nano-Glo Substrate (1:166 dilution of N157B (Promega Corporation)) in Opti-MEM (Gibco) to all plates. Plates were shaken at room temperature for 2–3 min and emissions measured at 450 nm (donor) and 610 nm (acceptor) using an EnVision 2105 multimode plate reader (Revvity). Raw donor signals were monitored for stability using a cutoff of 3× standard deviation of the negative control to correct for assay interference by the test compounds. Data was processed using in-house statistical software to calculate the BRET ratio (Emission610 nm/Emission450 nm), normalized against the DMSO control and fitted to a four-parameter logistic regression model to determine the IC50 value. To quantify the intracellular availability of PROTACs, the “availability shift” was calculated as 1 Availability shift=IC50(intact cells)IC50(permeabilized cells) A higher availability shift indicates a bigger difference in potency between intact and permeabilized cells and therefore a lower permeability or intracellular availability for a PROTAC. PROTACs yielding IC50 values > 4 μM in the permeabilized condition were not included in further analysis steps due to insufficient potency for the determination of an availability shift. Caco-2 Permeability Assay Bidirectional permeability of PROTACs across the Caco-2 cell monolayer was measured as previously described. , Briefly, Caco-2 cells were seeded at a density of 16000 cells/cm2 on Transwell inserts (#3379, Corning, Wiesbaden, Germany), and cultured in DMEM (high glucose, with 10% fetal bovine serum) for 14–21 days. In the experimental setup with preincubation, cells were incubated for 24 h with culture media containing 1 μM test compound in DMSO (final concentration ≤ 0.5%); in the experimental setup without preincubation, this step was omitted. After removing culture media, fresh transport buffer with 1 μM test compound was added to the apical (apical to basal) or basal (basal to apical) compartment (donor). Transport buffer without test compound was added to the opposite (receiver) compartment. Samples were extracted at different time points for up to 2 h, and PROTAC in the samples was quantified using LC-MS/MS. Apparent permeability coefficients from the apical-to-basolateral side (Papp,AB), and from the basolateral-to-apical side (Papp,BA), were calculated as below, respectively: 2 Papp,AB=QAB(C0×s×t) 3 Papp,BA=QBA(C0×s×t) where Q is the quantity of PROTAC recovered in the receiver compartment after the incubation time t, C0 is the initial PROTAC concentration given to the donor compartment, and s is the surface area of the Transwell inserts. The apparent, intrinsic permeability (Papp,int) was calculated as the arithmetic mean of Papp,AB and Papp,BA. This value provides a good approximation of a compound’s passive permeability. 4 Papp,int=(Papp,AB+Papp,BA)2 5 RatioEfflux=Papp,BAPapp,AB P-gp Liability Assay For a subset of PROTACs, potential P-gp liability was determined using a comparative antiproliferative assay. Parental NCI-H358 cells (ATCC) as well as an engineered clone stably overexpressing P-gp (ABCB1) were cultured in Falcon 75 cm2 Rectangular Canted Neck Cell Culture Flasks (Falcon) in RPMI 1640 (ATCC) supplemented with 10% fetal calf serum (HyClone) for up to 25 passages. Cells were detached from flasks using Trypsin/EDTA (PAN-Biotech) and seeded into 384 well cell culture microplates (polystyrene, F-Bottom, clear bottom, black (Greiner)) at a density of 200 cells per well in 60 μL cell culture medium and incubated overnight. PROTACs (10 mM stock in DMSO) were added the next day using an Access Laboratory Workstation with Echo 555 Liquid Handler (Beckman Coulter Life Sciences) to yield final concentrations of 0.3 nM to 25 μM in 0.5% DMSO. All dose response curves were measured in technical duplicates with negative controls (0.5% DMSO) added to each plate. Cells were incubated for 120 h before addition of 10 μL CellTiter-Glo 2.0 Cell viability reagent (1x, Promega Corporation), followed by shaking at room temperature for 15–30 min and measurement of the luminescence signal using an EnVision 2105 multimode plate reader (PerkinElmer). Data was processed using in-house statistical software to normalize raw luminescence signals against the DMSO control and fit the normalized raw data to a four-parameter logistic regression model to determine the IC50 value. To quantify the P-gp liability of PROTACs, the ratio of IC50 in P-gp overexpressing vs parental cells was calculated as a P-gp index. A higher P-gp index indicates a bigger difference in potency between parental vs P-gp overexpressing cells and therefore a higher P-gp liability for a PROTAC. 6 P‐gp index=IC50(P‐gpoverexpressing cells)IC50(parental cells) P-gp Expression Analysis Using Automated Capillary Western Blot Cell lysates of 25000 HEK293, MIA PaCa-2 or NCI-H358 cells (parental or P-gp overexpressing) were prepared in triplicates in RIPA buffer (Sigma #R0278, 35 μL) containing Halt Protease Inhibitor Cocktail (Thermo Fisher #87786, (100x)) and analyzed for P-gp expression using a Protein Simple WES device (Biotechne, #SM-W004). P-gp expression was probed using a recombinant Anti-P Glycoprotein antibody (Abcam #ab170904, 1:25), while GAPDH expression was visualized with anti-GAPDH antibody (Abcam #ab9485, 1:50). Results and Discussions Coverage of Chemical Space and Comparison with Data in the Public Domain The Boehringer Ingelheim PROTAC data set explored in this study comprises 3271 compounds from a range of exploratory PROTAC projects, which were assessed in up to three different assay types. In total, 346 compounds were measured in all three assay types, 753 in two of the three and 2172 in only one assay type. The numbers of tested compounds in each individual assay are reported in Table S1. The following assay setups were used to derive the listed permeability parameters:The NanoBRET assay was conducted for VHL and CRBN PROTACs using the HEK293 cell line. VHL PROTACs were, in addition, tested in the MIA PaCa-2 cell line. Compound permeability is quantified using the availability shift, which describes the intracellular availability of a compound by comparing the IC50 values obtained in intact vs permeabilized cells.The CellTiter-Glo assay was conducted using the NCI-H358 cell line. The P-gp liability of a compound is quantified by the P-gp index, which indirectly compares the intracellular availability of a compound by assessing the antiproliferative potency values (IC50) obtained in P-gp-overexpressing vs parental cells.The Caco-2 assay was conducted with and without PROTAC preincubation, assessing compound permeability using Papp,AB, Papp,BA, Papp,int and the efflux ratio. The NanoBRET assay was conducted for VHL and CRBN PROTACs using the HEK293 cell line. VHL PROTACs were, in addition, tested in the MIA PaCa-2 cell line. Compound permeability is quantified using the availability shift, which describes the intracellular availability of a compound by comparing the IC50 values obtained in intact vs permeabilized cells. The NanoBRET assay was conducted for VHL and CRBN PROTACs using the HEK293 cell line. VHL PROTACs were, in addition, tested in the MIA PaCa-2 cell line. Compound permeability is quantified using the availability shift, which describes the intracellular availability of a compound by comparing the IC50 values obtained in intact vs permeabilized cells. The CellTiter-Glo assay was conducted using the NCI-H358 cell line. The P-gp liability of a compound is quantified by the P-gp index, which indirectly compares the intracellular availability of a compound by assessing the antiproliferative potency values (IC50) obtained in P-gp-overexpressing vs parental cells. The CellTiter-Glo assay was conducted using the NCI-H358 cell line. The P-gp liability of a compound is quantified by the P-gp index, which indirectly compares the intracellular availability of a compound by assessing the antiproliferative potency values (IC50) obtained in P-gp-overexpressing vs parental cells. The Caco-2 assay was conducted with and without PROTAC preincubation, assessing compound permeability using Papp,AB, Papp,BA, Papp,int and the efflux ratio. The Caco-2 assay was conducted with and without PROTAC preincubation, assessing compound permeability using Papp,AB, Papp,BA, Papp,int and the efflux ratio. Before investigating correlations between the above assay parameters, we assessed whether learnings from our study can be translated beyond the Boehringer Ingelheim PROTAC data set. This was done by conducting a comparative principal component analysis (PCA, Figure A) of the chemical space covered by the Boehringer Ingelheim PROTAC data set and PROTAC DB 3.0, the largest publicly available database on PROTACs. , For the PCA, we used RDKit , to calculate 12 physically meaningful descriptors (Figure S1A) for the 3271 PROTACs in the Boehringer Ingelheim data set and the 6111 in PROTAC DB 3.0. (A) PCA scatter plot for the PROTACs recorded in the Boehringer Ingelheim PROTAC data set (blue) and PROTAC DB 3.0 (orange). Selected PROTACs that have been tested in clinical studies are highlighted in white, with their structures shown below. (B) Distribution of the maximum pairwise Tanimoto coefficients (based on Morgan fingerprints with radius = 2 and nbits = 4048) computed between each compound from the Boehringer Ingelheim PROTACs data set and from PROTAC DB 3.0. (C) Structures of exemplary PROTACs involved in (pre)­clinical trials and highlighted in (A), with vendor names. As shown in Figure A, the chemical spaces represented by the data sets overlap in most areas. Nevertheless, some areas are not covered by the Boehringer Ingelheim data set. When analyzing the PCA separately for CRBN and VHL-targeting PROTACs, the general observations are similar. There is a large area in chemical space populated with PROTACs from both proprietary and public data, and there are also substantial areas populated with compounds from only one source (Figure S1C–D). Differences in the data distributions across the chemical space are observed primarily along the y-axis (principal component 2), which is dominated by the number of aromatic or unsaturated rings, the number of hydrogen bond donors, the number of rotatable bonds and the cLogP (Figure S1A). Many of the PROTACs in the public data set have more hydrogen bond donors, rotatable bonds, and sp3-hybridized carbon atoms than those in the Boehringer Ingelheim PROTAC data set. In contrast, the molecules included in the Boehringer Ingelheim PROTAC data set often contain more ring systems, both saturated and unsaturated (Figure S2). These differences likely stem from the fact that long aliphatic or PEG-based linkers are often used as starting points for proof-of-concept PROTACs, which might be more prevalent in the public data set. , The representative examples from structures in the public domain (Figure C) are compounds that have progressed to (pre)­clinical trials and were chosen to visualize the diversity in chemical space that can still be found in optimized candidates. The visualized compounds are primarily separated along the first principal component. This can be seen in a trend toward larger and therefore heavier molecules with a higher number of heteroatoms, especially when comparing structures 1 (ARV-110, Bavdegalutamide) and 4 (DT2216). To a lesser extent, a separation along the second principal component is also visible when comparing structures 2 (ARV-471, Vepdegestrant) and 3 (KTX-335) with respect to the fraction of sp3-hybridized C atoms, which is higher in compound 3, and their cLogP, which is higher for compound 2. For additional context, available assay data for two of the reference compounds shown in Figure are provided in Table S2. In addition to the PCA, we computed the maximum pairwise similarity between each PROTAC in the Boehringer Ingelheim PROTAC data set and each compound in PROTAC DB 3.0. As shown in Figure B, for approximately 6.67% (218) of the Boehringer Ingelheim PROTACs, the Tanimoto coefficient to the PROTACs reported in the public domain is lower than 0.3, indicating that the molecular structures of these compounds are clearly distinct. In contrast, 112 compounds (3.42%) have a Tanimoto coefficient ≥ 0.9, leaving the majority of the data with moderate similarities to publicly available compounds. Overall, we conclude that the covered chemical space overlaps sufficiently to extend the learnings from the study to other PROTACs. Compound Clustering Does Not Affect the Correlation of Assay Readouts As it must be expected that the Boehringer Ingelheim PROTAC data set contains many congeneric ligand series, we first sought to characterize its structural diversity. As shown in Figure S1B, the largest part of the Boehringer Ingelheim data set is represented by at least one other compound in the data set with a Tanimoto coefficient (based on Morgan fingerprints, radius = 2, nbits = 4048) above 0.9. This indicates a close structural relation. Due to this, we aimed to reduce the bias in chemical space representation. We employed Butina clustering based on Morgan fingerprints to reduce series of compounds to individual data points. By clustering the data set using different similarity thresholds, we found that a Tanimoto coefficient of 0.9 was most suitable for achieving this goal, while keeping compounds that are clearly distinct in their molecular structure as separate data points (examples in Figure S3A). Table S3 shows that the selected threshold has no substantial impact on the observations made in this work. As a result of the clustering procedure, the Boehringer Ingelheim PROTAC data set was reduced by 26.0% ± 11.3% across the investigated assays. To investigate correlations between the different cell permeability assay types, the Pearson correlation coefficient (PCC) was calculated for all 66 assay parameter pairs (Figure S3B). Nineteen assay parameter pairs had absolute PCC values of 0.5 or higher and were investigated in more detail to assess the predictive value of different permeability parameters. NanoBRET Availability Shift Shows Good Correlation to Permeability in the Caco-2 Assay In PROTAC design, both the NanoBRET and the Caco-2 assay can be used to assess different aspects of cellular permeability. The NanoBRET assay determines the intracellular availability of PROTACs, whereas the Caco-2 assay measures the extracellular concentration of PROTACs after crossing a cell monolayer. Since the experimental setup of the Caco-2 assay requires cultivation of cells in an intact monolayer and sophisticated LC-MS-based concentration measurements, high-throughput alternatives for prefiltering compounds are of great interest. Generally, the correlations in Figure indicate a noteworthy relationship between the NanoBRET availability shift and various parameters of the Caco-2 assay. For CRBN PROTACs specifically, we observed strong inverse correlations between the NanoBRET assay availability shift and Caco-2 apparent intrinsic permeability (Papp,int, PCC = −0.70, Figure A), Caco-2 BA-permeability (Papp,BA, PCC = −0.71, Figure B), and, to a lesser extent, Caco-2 efflux ratio (PCC = −0.54, Figure C). (A) to (C) Correlations of cell permeability readouts for CRBN PROTACs obtained with the NanoBRET assay and three readouts of the Caco-2 assay without preincubation. (D) Schematic diagram of the correlated parameters in the NanoBRET and Caco-2 systems and their rationale. All correlations depicted had an absolute PCC > 0.5. Black arrows indicate the direction of correlation; all parameters are positively correlated except for the NanoBRET availability shift, which shows inverse correlation with the other parameters. The permeability of a compound in the NanoBRET assay is largely determined by its intrinsic permeability and the activity of efflux transporters at the plasma membrane. Since P-gp expression levels are low in the HEK293 and MIA PaCa-2 cells used in the NanoBRET assay (Figure S4), the readout is mainly driven by intrinsic permeability. This is supported by our data, which shows a negative correlation between the availability shift (NanoBRET) and the intrinsic permeability (Papp,int) of a PROTAC (Caco-2). This inverse correlation supports the hypothesis that the availability shift readout is a strong indicator for intrinsic permeability (Figure A). In addition, we observed that the availability shift shows a strong correlation with Caco-2 BA-permeability (Figure B), which, in turn, is determined by intrinsic permeability and P-gp activity. Based on the theoretical considerations for permeability assays by Kalvass and Pollack, Caco-2 BA-permeability varies between one and two-fold of the Caco-2 apparent intrinsic permeability, depending on the efflux mediated by P-gp. Since this is in the range of the assay variability, a strong correlation between Papp,BA and Papp,int is expected in the Caco-2 assay, which is confirmed by a PCC of 0.98 between these two parameters and discussed in further detail later (Figure A). The data explains the strong correlation between the availability shift in NanoBRET and Caco-2 Papp,BA. Furthermore, we found a correlation between the availability shift as an indicator of intrinsic permeability and the Caco-2 efflux ratio, which relates P-gp activity to intrinsic permeability (Figure C). One plausible explanation is that both intrinsic permeability and P-gp activity are influenced, at least in part, by the lipophilicity of the compounds. Increasing lipophilicity may enhance intrinsic membrane permeability while also increasing the likelihood of interactions with P-gp. This is consistent with the observed correlations between assay readouts and the cLogP of the PROTACs (NanoBRET – cLogP: PCC = −0.42, Papp,int – cLogP: 0.26). However, the comparatively low correlation coefficients indicate that lipophilicity alone cannot explain the assay outcomes. Instead, these results highlight that permeability and efflux behavior are governed by multiple physicochemical and structural factors. Regarding the interassay correlation, we reason that the increase in intrinsic permeability is overcompensated by the increase in P-gp activity, which results in a direct correlation between Papp,int and the efflux ratio in Caco-2 (PCC = 0.50), and an inverse correlation between the availability shift in NanoBRET and Caco-2 efflux ratio (PCC = −0.54), although the comparatively low PCC suggests influence of other factors in this. Lastly, given the strong impact of P-gp activity on Caco-2 AB-permeability, no correlation is expected between this parameter and the availability shift readout. Our data confirms that the NanoBRET assay cannot account for the influence of P-gp in our assay setup (PCC = 0.07). For VHL PROTACs, similar correlations are observed between the NanoBRET assay and the Caco-2 assay with preincubation. The PCCs between the NanoBRET and Caco-2 (with preincubation) assay readouts are −0.81 for Papp,int, −0.79 for Papp,BA and −0.29 for efflux ratio (Figure S5A–C). The corresponding correlations between the NanoBRET and Caco-2 assay without preincubation are weaker for VHL PROTACs and not considered as predictive. As discussed above, the readouts from the NanoBRET assay are correlated with those from the Caco-2 assay with and without preincubation for VHL and CRBN PROTACs, respectively. This finding suggests that the choice of the Caco-2 assay setup is important for obtaining meaningful assay readouts for structurally different PROTACs. Indeed, earlier work investigating the applicability of the Caco-2 assay, with and without preincubation, showed that the measurement discrepancy between the two setups increases with the lipophilicity (cLogP) of the compounds. That study also concluded that the Caco-2 assay with preincubation is better suited for highly lipophilic compounds that do not conform to the Rule-of-Five chemical space, as it allows the system sufficient time to reach an equilibrium. To confirm this observation in our study, we analyzed the difference in cell permeability across the different setups as a function of compound lipophilicity. To this end, the apparent permeability values from the Caco-2 assay with preincubation were divided by the respective values in the standard Caco-2 assay without preincubation (Figure A). The resulting Papp,int ratio (Figure B) is correlated with the lipophilicity of compounds (PCC = 0.55), indicating a greater difference between the setups with increasing cLogP. Given the complex and multifactorial nature of intrinsic permeability, a PCC of 0.55 already indicates a meaningful contribution of lipophilicity, even though it is only one of several influencing parameters. VHL PROTACs clearly occupy the high-ratio and high-cLogP space, and these compounds are prone to nonspecific binding in experimental conditions without preincubation. This finding supports the use of the Caco-2 assay with preincubation for VHL PROTACs, yielding more meaningful observations. (A) Permeability values measured with the Caco-2 assay with and without preincubation (unity line in black), and (B) permeability ratios (with and without preincubation, y/x values from (A)) in dependence of the cLogP. Data points for CRBN PROTACs are shown in blue, data points for VHL PROTACs in orange. P-gp is a Contributing, but Not Determining, Factor of PROTAC Permeability The CellTiter-Glo assay measuring P-gp liability of PROTACs was originally included in this study to filter out unfavorable compounds before running the Caco-2 assay, with the rationale that PROTACs showing high P-gp liability should also exhibit unfavorable efflux ratios and are therefore unsuitable for further development. However, across the full data set, the P-gp liability assay does not show a meaningful correlation with any of the Caco-2 assay parameters, with the highest PCC between the two assay types observed for the efflux ratio in the Caco-2 setup without preincubation (PCC = 0.35; Figure S6A). We therefore investigated the distribution of the calculated P-gp indices across all compounds measured in both the P-gp and one of the Caco-2 assay setups. For the purpose of prioritizing PROTACs, P-gp indices of ≤10 were considered favorable, while P-gp indices >10 would exclude further investigation in the Caco-2 assay. As shown in Figure S6B, the majority of the data set does not meet the ≤10 criterion. To investigate whether the hypothesized correlation could be observed if these compounds were removed from the data set, compounds with P-gp indices below and above 10 were analyzed separately. For the correlation with the Caco-2 assay without preincubation, all compounds measured also in the preincubation setup were excluded to avoid bias from unsuitable compounds for this assay. As shown in Figure S6C, compounds with a P-gp index ≤ 10 show a reasonable correlation to the efflux ratio with a PCC of 0.58. At higher P-gp susceptibility, this correlation is clearly lost (Figure S6D, PCC = 0.26), and several compounds show favorable efflux ratios (≤10) despite a high P-gp index. For compounds measured in the Caco-2 assay with preincubation, no relevant correlations were observed with any parameters. The analysis highlights that we observe higher efflux ratios with increased P-gp susceptibility as expected, but only as long as the P-gp liability in the NCI-H358 cells is moderate (≤10). This indicates that although P-gp liability is an established factor influencing the intracellular availability of PROTACs, it is not straightforward to systematically link the two or to use the P-gp index based on the antiproliferative assay introduced in this study as a filtering criterion in the drug development pipeline. One limiting factor is the assay principle of the comparative CellTiter-Glo assay, to derive relative differences in PROTAC intracellular concentration by comparing the compounds’ ability to inhibit NCI-H358 proliferation in the presence of high or low levels of P-gp, which provides a rather indirect readout compared to the LC-MS based quantification of PROTAC passage through the Caco-2 cell monolayer. Additionally, as the measured range of P-gp indices is based on differences in P-gp expression in two different cell models, an observed high P-gp susceptibility in the engineered vs parental NCI-H358 cell lines, which show a significant difference in P-gp expression levels (see Figure S4), might not be fully reflected by the efflux ratio measured in Caco-2 cells. Nevertheless, identifying and optimizing compound properties with regard to P-gp liability remains of relevance in drug candidate optimization. To this end, cell-free assay systems measuring inhibition of P-gp via substrate competition in vesicle-based systems have been developed. Alternatively, by designing functional PROTAC assays based e.g., on target degradation that employ cell models with high endogenous P-gp expression in combination with a specific P-gp inhibitor such as zosuquidar, the impact of P-gp on cellular PROTAC efficacy can be studied more directly. While these functional assays focus on the PROTAC’s effect on a specific target protein and therefore have to be developed individually for each drug design campaign, they can inform on P-gp liability more specifically compared to the indirect assay setup based on proliferation introduced above and still support a high-throughput setup with fast cycle times. Lastly, the P-gp component of the Caco-2 efflux ratio can be studied individually, e.g., via coincubation with zosuquidar, or a similar transwell assay using MDCK1 cells modified to overexpress P-gp can be utilized. NanoBRET Readout is Largely Independent of the Cell Line Used In addition to the interassay correlations discussed so far, we also observed strong correlations for parameters measured within the same assay type. As described in the Experimental Section, the noncancerous cell line HEK293 was used as the sole assay system for CRBN PROTACs, while both HEK293 and MIA PaCa-2 cells (tumor cells) were used for VHL PROTAC profiling. This enables us to assess the impact of using alternative cell models for PROTAC permeability measurements. The readouts for both cell types are highly correlated (PCC = 0.81; Figure S5D), indicating that the cell models tested here have little impact on the NanoBRET assay readout or the performance of the test compounds. In fact, the data suggests that a widely used model, such as HEK293 cells, , can be employed as a surrogate for more project-specific cell lines. This enables the setup of efficient screening cascades and the cross-comparison of PROTACs targeting various POIs using a single assay platform. Comparison of Caco-2 Assay with and without Preincubation Several strong correlations among parameters measured in the Caco-2 assay are also worth noting. The strongest correlation observed is between Papp,BA and Papp,int: PCCs of 0.97 and 0.98 are seen in Caco-2 assay setups with and without preincubation, respectively (Figure A). As rationalized above, this is mainly due to the established small variation between Papp,int and Papp,BA, as well as a reduced sensitivity to P-gp influence on the BA side. Examples of strong correlations observed in readouts obtained within the Caco-2 assay without (left) and with 24 h preincubation of compounds (right). Correlation of (A) Papp,BA vs Papp,int, (B) Papp,BA vs efflux ratio (C) Papp,AB vs efflux ratio. Furthermore, the efflux ratio is correlated with both the Papp,BA (PCC = 0.62) and Papp,AB (PCC = −0.71) values in the Caco-2 assay without preincubation (Figure B and C, left panels). The correlation between these parameters is expected as the efflux ratio is derived directly from these parameters. In the Caco-2 assay with preincubation, the corresponding PCCs are 0.74 and −0.77, respectively (Figure B and C, right panels). The data clearly shows that the correlations between the assay parameters hold true for the modified assay setup with preincubation, which allows the accurate measurement of candidates with higher lipophilicity. To further characterize efflux behavior within the data set, we examined the distribution of efflux ratios across the PROTAC set. Using an efflux ratio (ER) < 2 as an indicator of negligible efflux, 35 (12%) and 149 (13%) PROTACs fall below this threshold for the Caco-2 assays with and without preincubation, respectively. In addition, we investigated whether ER could be related to the selected physicochemical descriptors used in the PCA analysis (Figure A, detailed in Figure S1A). However, no significant correlations were observed for any single descriptor, indicating that efflux behavior in PROTACsas already discussed for lipophilicity aboveis likely governed by a combination of structural and physicochemical factors rather than a single dominant parameter. Impact of Rule-of-Five Compliance on Assay Readout Correlations It is possible that some of the correlations with an absolute PCC > 0.5 do not hold for PROTACs with extreme physicochemical properties, including particularly large, lipophilic substances, as exemplified in the Caco-2 assay data set, where preincubation is required to reliably measure the large, lipophilic VHL PROTACs. To test this, we classified our data set by the number of Rule-of-Five violations per compound (Figure A) and analyzed the effect on the resulting correlations. Analysis of the population of the Rule-of-Five chemical space in the Boehringer Ingelheim PROTAC data set. (A) Number of violations of the Rule-of-Five. (B) Description of the physicochemical properties and their violations in the data set. (C) Pearson correlation coefficients of relevant assay correlations compared to correlations of a subset of data violating either 2 (white) or 3 (black) definitions of the Rule-of-Five. The correlations are based on the log10-transformed values, as for other correlations previously. Due to the nature of PROTACs, all compounds violate the molecular weight constraint, resulting in at least one violation for the whole data set. Most compounds, however, show additional violations, mostly due to the number of hydrogen bond acceptors and, to a lesser extent, lipophilicity (Figure B). An additional interesting observation is that most VHL-targeting compounds (87.6%) exhibit three violations of the Rule-of-Five, whereas CRBN compounds are more evenly distributed between two and three violations (Figure A). Due to the limited data available for one or four violations, only the correlations for compounds with two and three violations of the Rule-of-Five were analyzed in more detail. Generally, existing correlations are not strongly affected by considering the compound subset with two or three Ro5 violations (Figure S7; values in Table S4). This validates the general applicability of these correlations to diverse compound sets. However, some correlations not investigated in detail in previous sections show clearly different trends when looking only at the subsets mentioned. Specifically, correlations influenced by this categorization are between the NanoBRET CRBN assay and the Caco-2 assay with preincubation on the one hand, and between the two Caco-2 assay setups on the other hand (Figure C, values in Table S4). Here, stronger correlations (PCC > 0.5) are observed for compounds with two violations, whereas the subset with three violations shows weaker or similar correlations to the complete data set. When exploring the correlations between the NanoBRET CRBN and Caco-2 assay with preincubation, the cLogP is the most relevant parameter for differentiating the two subgroups, as virtually all CRBN-targeting PROTACs violate the recommendations for molecular weight and number of H-bond acceptors and simultaneously conform to the number of H-bond donors (Figure B). This leads to the conclusion that, for CRBN PROTACs with a cLogP > 5, approximating Caco-2 assay results using the availability shift (NanoBRET) should be done with caution. Such recommendations cannot be given for VHL-targeting PROTACs due to the limited availability of molecules violating only two Ro5. Interestingly, the correlation between the different Caco-2 assay setups also weakens when looking at the data set with three Ro5 violations. This goes hand-in-hand with the difference observed between the assay setups in Figure B, where it is clearly visible that the contrast between the assay setups correlates with the cLogP of compounds. The cLogP is likely also an important factor in this case, since it is the only parameter that does not show a pronounced bias toward either compliance or violation of the recommendations. However, the cellular uptake of PROTACs and beyond-Rule-of-Five compounds in general is a complex process, involving both passive and active uptake routes and cannot be accurately captured in one physicochemical parameter or assay alone. It is therefore important to develop efficient screening pipelines to identify the most promising compounds at early development stages through assays suited for high throughput, and to prioritize those for characterization in different assay setups that may be more time-consuming. Conclusions The molecular properties of most PROTACs go beyond what is widely accepted as favorable, drug-like properties. Consequently, one of the key challenges in PROTAC research is optimizing their ADME properties, including cellular permeability. Several technologies exist for measuring the cellular permeability of PROTACs. The NanoBRET assay measures the target engagement and indicates the intracellular availability of PROTACs. It is suitable for high-throughput screening, supports fast assay cycles and provides a straightforward readout. In contrast, the Caco-2 assay provides more information on PROTAC permeability in the intestinal barrier but requires more resources and time. In addition, the NanoBRET assay is better suited for high-throughput screening in 384- or 1536-well formats than in the 24- to 96-well systems used for the Caco-2 assay. Another assay relevant to the intracellular availability of PROTACs is the P-gp liability assay, which focuses on P-gp-dependent export as one aspect of permeability. Since multiple transport mechanisms govern a molecule’s intracellular availability, the P-gp assay may indicate whether molecules are heavily affected by P-gp export but cannot be used as a standalone assay to estimate overall permeability. So far, no study has been published that systematically investigates the agreement between the readouts obtained with different cell permeability assays for PROTACs. Therefore, we set out to determine such links, with the aim of understanding PROTAC behavior in these assays, investigating correlations, and providing insights for experimental design. We studied three assay technologies across varying experimental setups and 19 assay readouts, yielding 66 correlation pairs. Nineteen of these correlations exhibited an absolute PCC of >0.5 and were investigated in more detail. Among these correlations, we identified a particularly interesting link between the NanoBRET assay availability shift and the Caco-2 permeability parameters for both CRBN and VHL PROTACs. It is important to note that the correlations have implications for assay applicability. The Caco-2 assay without preincubation, for example, is suited to compounds with lower lipophilicity, which is more readily achieved with CRBN PROTACs than with VHL PROTACs. Our data shows that meaningful measurements for more lipophilic compounds are obtained with the Caco-2 assay, including preincubation. Moreover, lipophilicity is a relevant parameter for CRBN PROTACs, as we observe that higher cLogP weakens correlations for certain assay readouts. Especially when extrapolating these findings to data sets with particularly lipophilic compounds or with compounds far beyond the Rule-of-Five space, the correlation should be reassessed on an individual basis before implementation. Based on the insights gained during this work, we recommend the following assay workflows for determining the cell permeability of PROTACs, as also outlined in Figure : For both CRBN and VHL PROTACs, the compounds can be prescreened using the NanoBRET assay to identify promising compounds. CRBN PROTACs or low-lipophilicity PROTACs with favorable results should then be progressed for in-depth evaluation in the Caco-2 assay without preincubation. For VHL PROTACs or more lipophilic PROTACs with favorable results, we suggest progressing to the Caco-2 assay with preincubation. While lipophilicity alone does not fully determine permeability or efflux behavior, it represents a practical and readily interpretable proxy for guiding initial assay selection based on the trends observed in this data set. Specific cutoffs for the NanoBRET availability index will depend on the individual project and change throughout the drug design process, as other PROTAC parameters such as POI degradation efficiency need to be optimized in parallel. The P-gp liability assay can be run to provide additional insights into mechanisms influencing the compounds’ intracellular availability, but cannot be reliably used to predict the efflux ratio in the Caco-2 assay, especially for compounds with a P-gp index >10. Recommendation for the design of a screening pipeline for assessing PROTAC permeability. All compounds should be screened in the NanoBRET target engagement assay, and only those with promising results should be further investigated in more complex assay setups, such as the Caco-2 assay. For the Caco-2 assay, lipophilicity can be used as a practical guideline to differentiate compounds, with more lipophilic compounds typically benefiting from the inclusion of a preincubation step. We believe that the learnings and guidelines from this large-scale analysis will help develop more efficient and informative assay strategies for determining the cell permeability of PROTACs and other modalities with beyond-Rule-of-Five physicochemical properties. Supplementary Material