From classic medicinal chemistry to state‐of‐the‐art interdisciplinary medicine: Recent advances in proteolysis‐targeting chimeras technology

Proteolysis‐targeting chimeras (PROTACs) is a targeted protein degradation (TPD) technique effected by hijacking the ubiquitin‐proteasome system (UPS) of the cells. A PROTAC molecule specifically binds to its protein of interest (POI) and recruits an E3 ligase to assemble a ternary complex. The POI was subsequently ubiquitinated, followed by being degraded by proteasomes. After 20 years of development, PROTAC technology has made a significant progress, and quite some candidates have entered clinical trials. Along with the excitement of realizing that PROTAC technology can develop therapeutics toward those traditionally believed “undruggable” targets; there are still unmet demands in PROTAC design, screening, and intracellular availability. PROTAC technology is rapidly advancing from employing traditional medicinal chemistry methodologies to integrating state‐of‐the‐art chemical biology technologies, becoming a typical example of interdisciplinary medicine. This review summarizes the progress made in PROTAC technology in recent years, including expanding new targets, developing dual‐target PROTACs, rational design and screening strategies, regulatory activation, targeted delivery, and developing biological macromolecule‐contained PROTACs.

ubiquitin-carrying E2 and the protein substrate. E3 ligase acts like a protein scaffold, making the E2 enzyme close to the substrate and promoting the transfer of ubiquitin to the substrate. In mammals, two members of the E1 family can transmit ubiquitin to all about 40 different E2 enzymes, which can deliver ubiquitin to substrates by more than 600 known E3 ligases. 1 The ubiquitination generally occurs on the lysine (Lys) residue of proteins, with an isopeptide bond formed between the carboxyl-terminal of ubiquitin and the primary amine of Lys residue. Proteins can be linked with only one ubiquitin monomer or multiple ubiquitin molecules to form polyubiquitin chains. Ubiquitin is a highly conserved 76-amino acid polypeptide containing seven Lys residues, which can all serve as docking points when the polyubiquitin chain forms, yielding different connection types (K6, K11, K27, K29, K33, K48, and K63linked chains). 2 These different types of polyubiquitin chains also determine the fate of ubiquitinated proteins, 3 among which K48-linked chains are believed to mediate the protein degradation by proteasomes. 4 Ubiquitination is a proximity-driven reactionubiquitin modification can usually be carried out as long as the target protein is close enough to E3 ligases. 5 This proximity-driven mechanism was employed in developing the targeted protein degradation (TPD) technique, namely Proteolysis-targeting chimeras (PROTACs, Figure 1B). The PROTAC is a heterogeneous bifunctional molecule, which usually consists of three parts: a warhead, a linker, and an E3 ligand. Warheads are responsible for targeting and binding to the protein of interest (POI), while E3 ligands recruit E3 ligases. A PROTAC molecule can thus induce ubiquitination via the proximity effect, and the POI is, therefore, down-regulated along with the resultant proteasome degradation.
In 2001, Sakamoto et al. reported the first PROTAC molecule, which consists of a covalent inhibitor ligand of methionine aminopeptidase-2 (MetAP2) and an IκBα phosphopeptide fragment capable of recruiting the F-box protein β-TRCP, successfully inducing the degradation of MetAP2 in Xenopus egg extracts. 6 In 2008, Schneekloth et al. first reported a small-molecule PROTAC, which targets the androgen receptor (AR) and recruits the E3 ligase MDM2. 7 Subsequently, the E3 ligase used by PROTACs was extended to inhibitors of apoptosis protein (IAP). 8 In 2019, ARV-110 for AR and ARV-471 for estrogen receptor (ER) became the first PROTAC molecule to be advanced into clinical trials ( Figure 1C). 9 After 20 years of development, PROTACs have become one of the most attractive areas in both academic research and the pharmaceutical industry. Nowadays, PROTACs in clinical trials have a wide range of targets, including kinases (AKT, AAK, and ALK), receptors (AR, ER, AXL, CCR9, and EGFR), epigenetic regulatory proteins (BET), and so on. 10 The E3 ligases commonly utilized by PRO-TACs are VHL, CRBN, MDM2, and IAP. 11 According to the PROTAC-DB database, as of 6/25/2022, 3270 small molecule-based PROTACs have been reported, consisting of 365 warheads, 82 E3 ligands, and 1501 linkers. 10 The development of PROTACs is used to engage classic medicinal chemistry methodologies-molecule design, organic synthesis, functional screening, and modification of the molecule until the pharmacological goal is fulfilled. The rapidly developing chemical biology technologies have vastly enriched the toolbox for building PROTACs. Newly released reports have constantly focused on developing new PROTAC concepts and principles integrating interdisciplinary technologies. This article mainly reviews the progress in the past 3 years, including the expansion of target types, the development of dual-target PROTACs, rational design and screening strategies, regulatory activation and targeted delivery, and the development of macromolecule-based PROTACs.

| DEALING WITH THE "UNDRUGGABLE": EXPANSION OF TARGET TYPES
One of the most attractive features of PROTACs is that they can down-regulated those traditionally believed "undruggable" targets. Different from traditional small-molecule drugs devoted to inhibiting the function of the target, PROTACs only require affinity binding to the POI, with the challenges more on ubiquitin tagging the POI by recruiting the E3 ligase and forming the ternary complex. In this scenario, molecules that are affinity binders to their targets but with poor inhibitory effects can be turned into building blocks to develop PROTACs, thus downregulating those targets that are used to be unable to inhibit. Besides, down-regulating target proteins instead of simply inhibiting them offers much more possibilities, largely expanding the target options in therapeutic design (Table 1).

| NF-κB1 p105
N-nuclear factor-κB (NF-κB) is a family of transcription factors (TFs) consisting of five different proteins (RelA, RelB, c-REL, p50, and p52) that play an essential role in processes such as cell proliferation and differentiation, cell survival, and immune and inflammatory responses. 12,13 The activation of NF-κB is initiated by extracellular signals and is highly dependent on the UPS. For example, p50 derives from the limited processing of its longer precursor p105 (NF-κB1) by the UPS that involves ubiquitin ligase KPC1. 14,15 Not commonly seen in ZHAO ET AL. the UPS, p105 can be either partially or completely degraded, which is regulated by different signaling mechanisms with different ubiquitin ligases. 16 It has been suggested that a long Gly-Ala repeat in the middle region of the p105 sequence generates the signal of interrupting the proteasome degradation and producing the p50 active subunit. 16 That said, the exact mechanism regulating the degree of degradation is not well-understood. NF-κB is often considered to be oncogenic; yet, the overexpression of KPC1 and the consequent accumulation of p50 have shown tumor-suppressive effects. 14, 15 Goldhirsh et al. found that the WILVRLW sequence in KPC1 is the key to its interaction with p105 and the ubiquitination process. 17 They developed a PROTAC based on the WILVRLW peptide as the warhead, connecting the VHL E3 ligase with a polyethylene glycol (PEG) linker in an attempt to induce limited processing from p105 to p50. It was confirmed that the ubiquitination of p105 outside the cell and the increase of p50 intracellularly could be achieved. However, no significant tumor-suppressive activity was observed, possibly due to the limited stability of the peptide-based PROTAC.

| FOXM1
The Forkhead box M1 (FOXM1) protein is a member of the Forkhead/Winged-Helix (FOX) family. It is found overexpressed in a variety of human cancer cells, such as liver, lung, and breast cancers. 18 FOXM1 contains a conserved DNA-binding domain (DBD) that binds the common promoter sequences of target genes. Activated via phosphorylation by the cell cycle complexes CDK4/6, FOXM1 can act as a TF to regulate the expression of genes promoting cell proliferation. 19 Besides, it can also interact with other proliferation-related proteins, including β-catenin and Smad3 to activate signaling pathways such as Wnt/β-catenin to promote cell growth and metastasis. 20 Therefore, the blockage of FOMX1 interacting with the promoter of the target genes is not sufficient to inhibit its function. Instead, a PROTAC to degrade FOXM1 provides an alternative therapeutic strategy.
Wang et al. screened a FOXM1-targeting peptide via phage display technology and designed a PROTAC composed of this peptide, E3 ligand pomalidomide, and a cell membrane penetrating peptide TAT. 21 The resultant PROTAC can effectively enter cells, induce degradation of FOXM1 protein and strongly inhibit the survival, migration, and invasion of a variety of cell lines. However, the effective concentrations are as high as 20-100 μM, which may be related to the short half-life of the peptide drug. Another strategy reported by Guoshun Luo et al. utilized computer simulation to screen FOXM1-DBD binding ligands, designed an inhibitor blocking FOXM1-DBD interacting with its DNA target, and converted it into a CRBN-based PROTAC. It effectively induced FOXM1 degradation in MDA MB-231 cells with a half-maximal degradation concentration (DC 50 ) value of 1.96 μM, caused G2/M cell cycle arrest, apoptosis, and growth inhibition of tumor cells, and inhibited migration and invasion of Triple-negative breast cancer (TNBC) cells. 22

| AR-V7
Prostate cancer is one of the most lethal cancers in men, and AR is critical in its development. Patients with T A B L E 1 Newly developed Proteolysis-targeting chimeras (PROTACs) toward new targets.
In Figure 2 Protein target Ref.

| BRAF V600E
The RAF family regulates cell proliferation, growth, differentiation, and survival and plays a role in the RAS-ERK pathway. 28 Mutations in the RAF family are related to cancer development, with BRAF V600E being one of the most well-known oncogenic mutant proteins. Small molecule inhibitors targeting BRAF V600E are clinically effective, despite the drug resistance caused by longterm use, which is attributed to the dimerization activation process of RAF kinases. Inhibitors can block the active site of RAF kinase to inhibit its catalytic function, but they are not ideal for preventing RAF dimerization. In contrast, some inhibitors even have the potential to promote the dimerization process. 29 Here the advantage of the degradation-inducing effect of the PROTACs is highlighted, as PROTACs function by inducing complete degradation of the target protein, thus preventing RAF dimerization. Ganna Posternak et al. developed a PROTAC, P4B (DC 50 = 12 nM; D max = 82%), by screening the combinations of two BRAF binders and three E3 ubiquitin ligase ligands (pomalidomide, thalidomide, VH032), with flexible inkers of different lengths. 30 Interestingly, P4B inhibits BRAF WT and BRAF V600E similarly in vitro but differs dramatically in its effect on inducing degradation. The authors speculated that the mutation in BRAF V600E made its active state different from the wild type and prone to form the ternary complex necessary for PROTAC-induced degradation. It also highlights the selectivity advantage of PROTACs over small molecule inhibitors as a strict ternary complex is required to be formed when taking effect.

| Wee1
Cells rely on the cell cycle checkpoints to repair DNA and maintain genomic integrity. In cancer, mutations in p53 often result in dysregulation of the G1/S cell cycle checkpoint, making cancer cells dependent on the G2/M checkpoint for DNA repair to prevent excessive DNA damage. 31 Wee1, a member of the serine/threonine kinase family, inhibits CDK1 kinase activity by phosphorylating the Tyr15 site of CDK1 to regulate the G2/M checkpoint and inhibit cells from entering mitosis. 32 AZD1775 is a Wee1 inhibitor currently in clinical trials, but studies have shown that AZD1775 has dose-dependent adverse effects. In addition, as an ATP-competitive Wee1 inhibitor, AZD1775 exerts a similar inhibitory effect on PLK1, which shares a highly conserved ATP binding site with Wee1, but has an essential role in negatively regulating Wee1 activity. 33 Li et al. developed ZNL-02-096, based on AZD1775 and pomalidomide, which exhibits a capability of inducing Wee1 degradation with a sub-micromolar dose, which could cause G2/M accumulation at doses by 10 fold lower than AZD1775 and acts synergistically with Olaparib in ovarian cancer cells. 34 Although ZNL-02-096 maintains the same biochemical selectivity as AZD1775, it cannot degrade PLK1 in cell-based assays. Marine C. Aublette et al. also achieved the use of AZD1775 as a warhead to construct selective Wee1 degraders by using the other two E3 ligases, VHL and CRBN. 35

| NSD3
The nuclear receptor binding SET domain protein 3 (NSD3) belongs to the NSD protein family, sharing high sequence homology with other proteins in this family. The NSD3 gene encodes two splice variants, NSD3-Short (NSD3S) and NSD3-Long (NSD3L). NSD3L is a histone lysine methyltransferase that catalyzes the mono-and dimethylation of histone H3 lysine 36 (H3K36) to regulate the chromatin structure and gene expression. NSD3S does not possess a C-terminal methyltransferase structural domain but retains an N-terminal structural domain that includes an acidic trans-activating structural domain, an H3K36me3/2-binding structural domain (PWWP1), and a complex scaffolding region. 36 NSD3 regulates cell cycle protein G1 (CCNG1) and NEK7 expression. It can also form a complex with BRD4 and interact with MYC to activate its oncogenic effects. 37 A selective antagonist of the NSD3-PWWP structural domain, BI-9321, has been reported, but this antagonist does not inhibit other functions of NSD3. 38 Xu et al. developed MS9715 based on BI-9321 and a VHL ligand. 39 It induces the degradation of NSD3 with a concentration-and time-dependent form in human hematological tumor models such as acute myeloid leukemia (AML), acute lymphoblastic leukemia (ALL), and multiple myeloma (MM). Using a similar idea, Yaoliang Sun et al. developed dNSD3, which induces the degradation of NSD3 in NCI-H1703 and A549 cells with DC 50 values of 1.43 and 0.94 μM, respectively. It also shows better selectivity for NSD3 than NSD1 and NSD2. 40

| SHP2
SRC homology 2 (SH2) domain-containing phosphatase 2 (SHP2) functions in a variety of cellular signaling pathways, such as RAS-ERK. 41 Over-activation of SHP2 caused by mutations has been found to be associated with various cancers, including solid tumors 42 and hematological malignancies. 43 It has also been suggested that SHP2 is essential in T-cell programmed cell death and immune escape. 42 SHP2 is thus believed to be a promising anti-tumor target. The mainstream inhibitors of SHP2 share some structural similarities, generally having aniline and saturated azido ring structures.
Mingliang Wang et al. modified an SHP2 inhibitor and designed SHP2-D26 with DC 50 values of 6.0 and 2.6 nM in KYSE520 and MV4-11 cells, respectively and >95% degradation of SHP2 at 30 nM in both cells. 44 Xiangbo Yang et al. also reported ZB-S-29 with a DC 50 value of 6.02 nM in MV4; 11 cells. 45 Instead of ligating the linker to the aniline group, Mengzhu Zheng et al. developed SP4 taking the primary amine on the saturated nitrogen heterocycle at the other end of the SHP2 inhibitor as the anchor site. 46 It effectively and specifically degrades SHP2 in Hela cell lines at low concentrations, and the IC 50 is 4.3 nM.

| Class I HDACs
Acetylation of lysine side chains is an important and dynamic post-translational modification (PTM) of histones. The regulation of histone deacetylation is associated with cancer, neurological diseases, and immune disorders. 47 The PTM is introduced by histone acetyltransferases (HATs) and can be removed by histone deacetylases (HDACs). HDACs can be classified into different classes: Class I (HDAC 1, 2, 3, and 8), Class IIa (HDAC 4, 5, 7, and 9), Class IIb (HDAC 6 and 10), and Class IV (HDAC 11). HDAC 1, 2, and 3 only function in the nucleus and are the HDACs with the most significant gene expression. 48 Smalley et al. designed a PROTAC using the benzamide HDAC inhibitor CI-994, which is selective for HDAC 1, 2, and 3 and verified that dHDAC-1 effectively induced degradation of HDAC 1, 2, and 3 in a concentration-dependent manner, resulting in increased overall acetylation levels as well as loss of cell viability in HCT116 cells. 49 Martin Roatsch et al. reported dHDAC-2 based on a macrocyclic tetrapeptide inhibitor selective for Class I HDACs. 50 It leads to the selective degradation of HDAC 1, 2, and 3 in a time-and concentrationdependent manner without significant cytotoxic effects. It is expected to be used to study Class I HDACs by shortterm deletion of the biological role of HDAC enzymes.
Class I HDAC 1, 2, and 3 can alter the acetylation status of TFs such as NF-κB in the inflammatory response, which in turn affects the expression of inflammatory genes. It has been shown that siRNA-mediated knockdown of HDAC3 upregulated IL-10 expression in lipopolysaccharide (LPS)/interferonγ (IFNγ)-treated RAW 264.7 macrophages. 51 However, RGFP966, an inhibitor of HDAC3-selective deacetylase activity, did not cause significant changes in gene expression in RAW 264.7 macrophages. 52 This discrepancy implies that HDAC3 plays a non-catalytic function in inflammatory signaling and, therefore, inhibition of enzymatic activity by HDAC3 selective inhibitors is insufficient to produce antiinflammatory effects. This has become another typical problem that could be addressed by selective degradation of HDAC3 by PROTACs.
Xiao et al. obtained XZ9002 through screening, which can degrade HDAC3 in different HDAC isozymes specifically and has the potential to overcome the toxicitylimiting dose associated with conventional HDACi. 53 Cao et al. developed the o-amino aniline-based HD-TAC7 for the degradation of HDAC3 with a DC 50 value of 0.32 μM in RAW 264.7 macrophages. 54 But compared to the HDAC inhibitors Eninostat and CI-994, HD-TAC7 has no significant effect on the transcription of genes for IL-10, iNOS, IL-6, and TNFa in LPS/IFNγ -stimulated RAW 264.7 macrophages. Yufeng Xiao et al. speculated that this phenomenon is because of the side effect of pomalidomide downregulating the levels of NF-κB p65 subunit.
HDAC8 catalyzes the deacetylation of non-histone proteins such as cohesins 55 and interacts with TFs STAT3 and CREB to regulate some gene expression. 56 It has been associated with the development of cancers such as T-cell lymphoma/leukemia. 57 Also, some studies suggest it to be a potential therapeutic target of pediatric neuroblastoma. 58 Chotitumnavee et al. reported that dHDAC8-1, based on a selective inhibitor of HDAC8, could efficiently degrade HDAC8 without affecting the HDAC1, HDAC2, and HDAC6 levels with a more ZHAO ET AL. substantial toxic effect on Jurkat cells than its inhibitor parent. 59 Darwish et al. developed dHDAC8-2 with excellent HDAC8 selective degradation and promotion of neuroblastoma phenotype differentiation. 60

| PD-L1
PD-L1 is expressed on the surface of tumor cells and binds to PD-1 on the surface of activated immune cells, leading to immunosuppression and promoting immune escape. Therapeutics that block PD-L1/PD-1 interactions include monoclonal antibodies (nivolumab, cymplimab, pembrolizumab, atezolizumab, avelumab and dulvalumab), macrocyclic peptides, polypeptides, peptide mimetics and non-peptide structured small molecules (CA-170, CA-327, and BMS-986189). 61 Despite being one of the popular targets in cancer therapy, one factor limiting the development of PROTACs targeting PD-L1 is that there are few ligands targeting the intracellular structural domain of PD-L1.
Wang et al. suggest that PD-L1 on the cell membrane is continuously "recycled" by endocytosis. Therefore PROTACs can be constructed with ligands targeting the extracellular region of PD-L1 to degrade newly generated PD-L1 and "recycled" PD-L1 in the cytoplasm. 62 They tested their conjecture by developing dPD-L1 based on the diaryl ether molecule BMS-37 (the IC 50 is 6 nM for PD-L1/PD-1) binding to the extracellular structural domain of PD-L1, with a CRBN ligand attached to the ethylenediamine side chain. dPD-L1 can dose-and timedependently induce the degradation of PD-L1 in different cancer cells.
Cotton et al. took a completely different approach. 63 They reported a type of antibody-based PROTACs-AbTACs. This fully recombinant bispecific antibody binds to the target protein PD-L1 and the E3 ligase RNF43 on the cell membrane, causing endocytosis and degradation of PD-L1 by RNF43 with a D max of 63%.

| Other targets
In addition to the above targets, Xiao et al. reported the first RPOTAC MD13 targeting macrophage movement inhibitory factor (MIF), which induced almost complete degradation of MIF in A549 cells at submicromolar concentrations with a DC 50 of approximately 100 nM; it showed antiproliferative effects in a 3D tumor sphere model. 64 Liu et al. developed deEF2K using A484954, an inhibitor of eukaryotic elongation factor 2 kinase (eEF2K), which degraded eEF2K and induced apoptosis of human breast cancer MDA-MB-231 cells in a dose-dependent manner. 65 Mingxing Hu et al. reported that dIDO1, which targets the immune checkpoint indoleamine 2,3-dioxygenase 1 (IDO1), significantly induced IDO1 degradation in HeLa cells with a Dmax of 93%. 66 Lu et al. first reported that PROTACs, dTTK-1 and dTTK-2, which target threonine tyrosine kinase (TTK), induced TTK degradation in human colon cancer cell line COLO-205. The DC 50 is 1.7 and 3.1 nM, respectively. 67 71 The reported PROTAC has a high affinity for CREPT (Kd = 0.34+/−0.11 μM). It can penetrate the cell via the membrane transport peptide RRRRK, successfully inducing the degradation of CREPT and inhibiting tumor growth. Their work validates the possibility of using protein dimerization structure-based interaction motifs to design PROTACs.
Aside from the tumor-related targets described above, inflammation-related targets have also been reported to be degraded by PROTACs. Degorce et al. reported a degrader targeting interleukin-1 receptor-associated kinase 3 (IRAK3). 72 Their identification of an IRAK4 inhibitor led them to find a by-product as a selective IRAK3 ligand, and they converted it to a PROTAC dIRAK3 with a D max of 98% and DC 50 of 2 nM for degrading IRAK3 in THP1 cells. Also, targets are associated with pigmentation and dermatopathy. Fu et al. developed a PROTAC targeting human tyrosinase, which is expected to be further developed as a drug for the treatment of pigmented skin diseases. 73

STONE: DUAL-TARGET PROTACS
Given the design principle of PROTAC, it is also possible to target more than one target with one molecule. The strategies include incorporating dual-target warheads that bind to POIs sharing structurally similar domains or, more impressively, combining warheads toward two different targets in one PROTAC molecule. Medicinal chemistry methodologies of molecule design, chemical synthesis, and modification are critical and highly desired in this process.

| BCL-xL and BCL-2
Both BCL-xL and BCL-2 proteins, having approximately 45% sequence homology, belong to the anti-apoptotic BCL2 family, and both protect cancer cells from apoptosis. 74 As the most common member of the BCL2 family, BCL-xL is overexpressed in various solid tumors and circulatory malignancies. It is thus often chosen as a target in cancer therapy. However, either BCL-xLselective or dual BCL-xL/BCL-2 inhibitors can produce thrombocytopenic adverse effects and limit the dose used. 75 Researchers have developed PROTACs that target BCL-xL, 76 but degrading BCL-xL alone is only effective for BCL-xL-dependent cancer types, not for cancers that rely on both BCL-xL and BCL-2 or rely on BCL-2 only. Therefore, Pratik Pal et al. developed a dual target PROTAC based on the dual BCL-xL/BCL-2 inhibitor ABT-263 77 ( Figure 3A). By optimizing the type and length of the linker, they constructed PZ703b capable of targeting both BCL-xL/BCL-2. The PROTAC effectively degrades BCL-xL while not BCL-2 but enhancing its inhibition. Previously, they developed the PROTAC DT2216, 78 inducing selective degradation of BCL-xL, also based on ABT-263. DT2216 has a linker site on the piperazine ring that replaces the ABT-263 morpholine ring. Given the possible influence of the linker site on the degradation effect of PROTACs, they changed the linker site of PZ703b to one of the two methyl groups on the ABT-263 cyclohexene ring. It was found that the Rconfiguration is the most effective configuration. Interestingly, the R-configuration of PZ703b could effectively induce the formation of the VCB-PROTAC-BCL-2 ternary complexes but not the degradation, whereas DT2216 does not induce the formation of the BCL-2 containing ternary complexes intracellularly.
Both dual-target PROTACs could induce the formation of the VCB-PROTAC-BCL-XL ternary complexes, showing that the ternary complex formation was not a sufficient condition. In contrast, the VCB-PROTAC-BCL-2 ternary complexes do not effectively promote the degradation of BCL-2 but enhance its inhibition: when the VHL ligand VH032 competed with the R-configuration PZ703b in RS4; 11 cells, the PROTAC activity was lost by nearly 20-fold, whereas when VH032 competed with DT2216, DT2216 activity was lost by only 1.5-fold. At the same time, pretreatment of RS4; 11 cells with VH032 disrupted the inhibition of the interaction between BCL-2 and the pro-apoptotic protein Bim by R-configuration PZ703b, but not by ABT-263. R-configuration PZ703b simultaneously induces BCL-xL degradation and enhances the inhibitory effect of BCL-2, and is more helpful in killing cancer cells. Later in the same year, the research group again reported a dual-target PROTAC 753b ( Figure 3B), also based on ABT-263, with the main difference from PZ703b being an additional piperazine loop in the linker, again with the R-configuration most potent. 79 However, it can induce both BCL-xL and BCL-2 degradation, with a DC 50 of 6 nM for BCL-xL in HEK293T cells (30 nM for DT2216) and 48 nM for BCL-2 degradation. But the group did not compare the difference in tumor inhibition between 753b and PZ703b.

| dCDK12/13
Using similar design principles, Yang et al. reported a dual-target PROTAC dCDK12/13 based on an optimized CDK12/13 inhibitor-induced protein degradation with DC 50 values of 2.2 nM for CDK12 and 2.1 nM for CDK13 in MDA-MB-231 cells ( Figure 3C). The proteomic identification revealed excellent selectivity of this PROTAC. 80 This PROTAC effectively inhibits the growth of HRdeficient MFM223 and MDAMB-436 TNBC cells with IC 50 values of 47 and 197.9 nM, respectively, with positive implications for the treatment of TNBC.

| dFLT3/CDK9
Rezníckov et al. constructed dFLT3/CDK9 capable of inducing both FLT3-ITD and CDK9 degradation based on the FLT3/CDK9 kinase inhibitor BPA311, but it does not have an anti-proliferative activity as high as the parental inhibitor BPA311 against leukemia cell lines, possibly because the linker types and lengths were not optimized when designing the PROTAC 81 ( Figure 3D). ZHAO ET AL.

| EGFR and PARP
Different from the design concept limited by a dual-target conjugate and structural similarity of the target proteins, Zheng et al. constructed a "heterogeneous trifunctional" PROTAC molecule targeting two completely different types of proteins. 82 The key to the design of such PRO-TACs is to find a suitable 'star linker,' and they chose natural amino acids such as tyrosine and serine as the two contain hydroxyl, amino, and carboxyl functional groups that can be used as attachment sites for two target protein ligands and an E3 ligase ligand. A total of eight dual-target PROTACs of different linkers were constructed based on epidermal growth factor receptor (EGFR) inhibitor gefitinib and the poly(ADP-ribose) polymerase (PARP) inhibitor olaparib linking to serine, some of which can induce the degradation of both EGFR and PARP. The IC 50 value of the most effective molecule, DP-V-4, is 19.92 � 1.08 μM in H1299 cells, which is not significantly better than the parental inhibitors (6.56 � 1.07 nM and 35.93 � 1.05 nM for gefitinib and olaparib, respectively) ( Figure 3E).
Meanwhile, Mengzhu Zheng et al. did not design experiments to compare the dual-target PROTAC versus two single-target PROTAC combinations in terms of activity. This type of dual-target PROTAC molecule has the F I G U R E 3 Structures of dual-target Proteolysis-targeting chimeras (PROTACs). potential to achieve better efficacy by inducing the degradation of key target proteins of two different signaling pathways, as well as to improve target selectivity and reduce side effects through a more complex multiplex interaction pattern. But, it will increase the difficulty of rational design and optimization of PROTAC molecules, as well as potentially produce poorer PK and reduced permeability. As a result, the bioavailability of PROTACs may be affected. In any case, Mengzhu Zheng et al. could inspire ideas for the design of PROTAC molecules.

| Two different domains of SIM1
It is noteworthy that Satomi Imaide et al. used a similar strategy to construct a trivalent PROTAC SIM1, except that they targeted two repetitive structural domains of the same protein to improve the affinity with POI. 83 They linked a bivalent BET inhibitor to the E3 ligand to construct a trivalent PROTAC with DC 50 values of only 0.7-3.3 nM in HEK293 cells, showing a higher and more sustained degradation efficiency than the bivalent PRO-TAC (DC 50 values are 25-923 nM in HEK293 cells), as well as a more robust anticancer activity.

RATIONAL DESIGN AND SCREENING STRATEGIES
In the recent 2 decades, tremendous efforts have been devoted to improving PROTAC technologies, with one focus on how to speed up the lead discovery process. In addition to the advanced synthetic strategies to assemble optimized E3 ligands, linkers, and warheads, more interdisciplinary technologies, such as artificial intelligenceassisted molecule design and innovative chemical biology screening assays have been applied in the PRO-TAC development process, and great success has been obtained.

| Improved E3 ligands and linkers
In the process of developing PROTACs, inhibitors or agonists with high affinity are generally chosen as POI ligands. E3 ligands are commonly used, according to the different E3 ligases, such as VH032 for VHL and thalidomide and pomalidomide for CRBN. However, thalidomide and pomalidomide have defects in that their phthalimide rings are easily hydrolyzed to form α-(carboxybenzamide) glutarimide in body fluids with pH 6-8. 84, 85 Min et al. reported using phenylglutarimide (PG) as a CRBN ligand to replace thalidomide. 86 Compared with thalidomide, PG maintains its affinity with CRBN and is more stable because the phenyl of PG can reduce the hydrolysis of the glutarimide ring. They synthesized PROTAC SJ995973 targeting BET based on PG. It induced the degradation of BRD4 with DC 50 as low as 0.87 nM, and D max was as high as 99%. The IC 50 value for MV4-11 cells was only 3 p.m.
In addition to POI ligands and E3 ligands, it is imperative to optimize the junction site, type, and length of the linker because it directly affects whether PROTACs can form effective ternary complexes with E3 ligases and POI to ubiquitinate successfully. The importance of the linker is also reflected in its ability to achieve some other functions. It determines the selectivity of the PROTAC by affecting the formation of ternary complexes. The type of the linker sometimes also determines the cell membrane permeability of the PROTAC (Table 2).
Andrea Testa et al. enhanced the discrimination of the PROTAC targeting BRD4 between the first and the second bromodomains of BET proteins through macrocyclizing linker design. 87 Based on the crystal structure of the BRD4 BD2 -MZ1-VHL ternary complex with the aid of computational calculations and molecular dynamics (MD) simulations, they designed the connection strategy of a macrocyclic linker, that is, adding a second linker between the two ligands of MZ1 to form a circle. Meanwhile, the research group proposed that amide-to-ester substitution can improve the membrane permeability of PROTACs. 88 PROTACs are usually linked by an amido bond because it is more stable, although it leads to poor pharmacokinetic (PK) properties. In this respect, estercontaining compounds may have an advantage over amides because of their relatively high lipophilicity. Although esters are often unstable in plasma, increasing steric hindrance in the chemical space around them can reduce the occurrence of hydrolysis. They designed some ester-containing VHL-based BET degradants, which have a stronger ability to induce degradation than their corresponding amide-containing compounds and show some stability in cells, although esters sacrifice part of their binding capacity.
With regard to the membrane permeability of PRO-TACs, it is generally considered that relative molecular mass is the key factor affecting the permeability. However, Atilaw et al. described the relationship between molecular conformation and permeability of PROTACs. 89 Their research object, VHL-based PROTAC 1, has good cell permeability despite its high molecular mass, polarity, and rotatable bonds. They studied its conformation by nuclear magnetic resonance (NMR) spectroscopy and found that PROTAC 1 had different conformations in different solutions. In DMSO and DMSO-water (simulating extracellular and intracellular environment, respectively), they have elongated and polar conformations, while in chloroform (simulating the internal environment of the cell membrane), under the action of intramolecular non-classical hydrogen bonds, π-π interaction and shielding of amide groups from the solvent, the conformation folds, resulting in smaller polar surface area, which promotes cell permeability by minimizing the size and polarity. They believe that the ability to change the conformation of PROTAC 1 is the main reason for the entry into target cells. This opinion has some implications for the design of PROTAC linkers. It can be noted whether the linker contributes to molecular folding to produce a smaller polar surface area and thus improve the membrane permeability.

| Advanced synthetic strategies
One of the speed-limiting steps in the process of PROTAC development is that it is often necessary to synthesize dozens or even hundreds of quasi-PROTAC molecules containing different kinds and lengths of linkers one after another, which is quite an energy-and time-consuming work. Developing rapid synthesis methods is a way to deal with this problem (Table 3). For instance, it is meaningful to aim at the commonly used E3 ligands, such as VH032, thalidomide, and so on. Yan et al. reported a rapid and large-scale synthesis of VH032 amine. 90 This method does not need column chromatography. The total yield is 65%, and 42.5 g VH032 amine hydrochloride can be synthesized in one week with a purity of 97%.
There is also a need for a method that can quickly connect each component into a complete PROTAC. Based on the amine library of IMiDs (Immunomodulatory drugs composed of thalidomide and its derivatives pomalidomide and lenalidomide) with different linkers, Haixia Liu et al. constructed the corresponding azide library efficiently by rapid one-step conversion reaction, which can be used to quickly connect POI ligands through 'click chemistry' reaction. 91 Xu et al. developed a method of solid-phase organic synthesis. 92 They designed azide resins with different linkers to which the E3 ligand pomalidomide was pre-attached, and they also developed similar amino and carboxyl resins. These resins can react with alkynyl, carboxyl, and amino groups of POI ligands to form complete PROTACs. Le Guo et al. reported on the Rapid-Tac platform. 93 By introducing hydrazide groups into POI ligands and reacting quickly with preestablished VHL and CRBN ligand libraries containing acetaldehyde functional groups and various linkers, they could prepare dozens of compounds in one or two days. After that, these PROTACs were directly used for screening at the cellular level, and the hydrolytic hydrazide structure of the most active compounds was replaced T A B L E 2 Improved E3 ligand and linkers. with amides to obtain stable PROTACs. They applied this method to the development of PROTACs targeting fibroblast growth factor receptor 1 (FGFR1) and finally got LG1188. The IC 50 values for DMS114 and NCI-H1581 cells were 139 and 160 nM, respectively.
Some researchers try to avoid the heavy work of linker optimization directly. For example, George M. Burslem et al. put forward the idea of Scaffold hopping. 94 They replaced the POI ligand part of the developed GMB-475 targeting BCR-Abl with a new recruitment element while keeping the E3 ligand and linker unchanged and constructed a new GMB-805 targeting BCR-Abl, which increased its activity by 10 fold. However, their research is only an isolated case, which is highly accidental. This method is limited to the targets with PROTACs reported already, and there is nothing it can do for new targets.

| Computer simulation and virtual filtering
Beneficial from the development of information technology, computer simulation and virtual filtering has become reliable methods in PROTAC design. Because the successful ubiquitination and degradation of POI are dependent on the formation of POI-PROTAC-E3 ternary complexes, the computer simulation mainly focuses on the modeling of such ternary complexes to evaluate whether the design of linkers is reasonable.
Drummond et al. proposed a modeling method 95 ( Figure 4A). First, two binary protein-ligand complexes are needed: the E3 ligase and POI bound to their binders, completely matching those binding constituent parts of PROTACs. Then protein-protein docking is used to form a set of interactions between the two complexes. A robust conformational ensemble is then generated for the input PROTAC in the protein-protein interaction conformations. Each constituent part of the PROTAC is constrained to maintain its binding conformation during the conformational search procedure.
Zaidman et al. reported ProsettaC Protocol 96 ( Figure 4D), which also needs to input the SMILES string of PROTACs and the complex structures of the POI and E3 ligase binding to their own ligands, and two anchor atoms in the two ligands are defined as part of the input. First, the distance between the anchor points is roughly sampled, and the limited distance is generated according to the PROTACs' random conformations that can successfully match the ligands. Second, PatchDock is used to carry out global protein-protein docking according to the  distance restriction in the previous step. Then, fine local docking without constraining the positioning distance is conducted using RosettaDock. After that, 100 complete PROTAC conformations suitable for the position of ligands in these complexes are constructed, and the best conformations of linkers are selected by RosettaPacker to form ternary complex models. Finally, the complex models are clustered and sorted according to the number in each cluster.
Weng et al. developed PROTAC-model 97 ( Figure 4E). First, the POI and E3 ligases containing their respective binders are docked by FRODOCK. After filtering and screening, the conformations with more than one residue in the binding sites of the two proteins are retained; the complete PROTAC is modeled as docking conformations by RDKit, and the conformational energies are calculated by ObEnergy, eliminating the model with an unfavorable atomic collision; the binding modes of PROTACs with protein-protein complexes are evaluated by VINA, and the complexes with energy scores less than 0 kcal/mol are retained. Then, the remaining models are reordered using VoroMQA and clustered by the fraction of common contact (FCC) method. Finally, the best model in each cluster is selected and refined with RosettaDock, which is optional.
The idea of the above method is first to construct the interaction models of two proteins and then construct the conformations of linkers in these models; of course, there is another different way of thinking, that is, first construct the binding modes of proteins with a certain degree of complementarity, and meanwhile screen the dominant conformations of PROTAC linkers and then combine the two.
Bai et al. proposed such a method 98 ( Figure 4C). First, two proteins are manually placed near ligands to face each other, and Rosetta is used in generating the POI and E3 ligase docking models containing ligand information. Then the top 10% of models are screened according to the interaction energy. At the same time, the OMEGA program is used to generate conformational assemblies of the linker (with chemical "stubs" of ligands at both ends, that is, parts of the POI ligand and E3 ligand). Then, according to the root mean square deviation (RMSD) between the ligand and the chemical "stub" in the ligand, the conformations of the linker compatible with a reasonable protein-protein interaction model are selected to form ternary complexes. Finally, each ternary complex is refined. After modeling the ternary complex, it is necessary to judge the difference between the ternary complex and the natural configuration, which is F I G U R E 4 Computer simulation and virtual filtering. (A) Comparison between the foramal approach method 4 and newly-developed method 4B which improved the accuracy by clustering the results. 95 (B) MD-based scoring approach to score PROTAC-mediated ternary complex poses. 99 (C) Models of the PROTAC-mediated ternary complex consist of docking and component fragments building. Reproduced with permission. 98 Copyright 2021, American Chemical Society. (D) Schematic protocol of the PRosettaC. This method first looked into the distance of the sample, following global docking and local docking, and finally clustered the top results. Reproduced with permission. 96 Copyright 2020, American Chemical Society. (E) Workflow of the PROTAC-model protocol that consists of the docking, filtering, rescoring, and refinement steps. 97 generally verified by experimental methods, but the combination of virtual verification can also be considered to improve the efficiency.
Liao et al. proposed a method based on MD to evaluate the heat resistance of ternary complexes 99 ( Figure 4B). They believe that the complexes approximate to natural configurations are thermally stable and have a longer occupancy time both at room temperature and higher temperatures. In addition to using the calculation method to judge the rationality of the designed linkers, there are also studies on the direct design and virtual screening of full PROTACs by using artificial intelligence.
AS et al. developed a PRTOAC design method based on artificial intelligence and combined it with the MD method to evaluate the stability of the ternary complexes, 100 while the virtual screening method designed by Zheng et al. took the PK of PROTACs into consideration. 101 However, whether it is the computer simulation of complex ternary configurations or the use of artificial intelligence virtual screening, the computing-based structure always has a certain deviation from reality, and this deviation cannot be predicted and quantified, so it is necessary to weigh the dependence on virtual computing in the process of PROTAC development.

| Screening assays
Another speed-limiting step in the PROTAC development process is the verification experiment. Western Blot is usually used for verification at the cellular level, but the flux of this method is not high, and the sensitivity is low. Reporter gene analysis is common as well, but it often requires tags which could lead to a loss of accuracy. Some studies have proposed experimental designs for rapid or high-throughput screening, and most of these methods are to detect complex ternary formation.
Lin et al. reported that the formation of ternary complexes could be characterized by time-resolved fluorescence energy transfer (TR-RET) to determine the potency of PROTACs 102 ( Figure 5A). They constructed GST-BRD2 (BD1) and His-CRBN fusion proteins, stained them with Tb-anti-GST and AF488-anti-His, respectively, and evaluated the ability of some PROTACs to induce ternary complexes by collecting FRET signals. This method is similar to reporter gene analysis, which requires the fusion of proteins with tags. capture dissociation (ECD), collision-induced dissociation (CID), and ion mobility spectroscopy (IMS) based on NMS to obtain additional information about the interaction to measure the effectiveness of PROTACs. 105 However, the formation of the ternary complex is not a sufficient condition for ubiquitination, 77-79 so Gross et al. proposed to directly detect the ubiquitination of natural POI mediated by PROTACs based on the ability of tandem ubiquitin-binding entities (TUBEs) 106 to bind to the ubiquitin chain specifically 107 ( Figure 5C). They fixed TUBEs to an orifice plate to capture ubiquitinated POI in cell lysates induced by PROTACs and then detected it by using a method similar to Elisa. The concentration of PROTACs in the group with the highest ubiquitination level is named Ub max . They used PRO-TACs that target proteins like BRD3, Aurora A Kinase, and KRAS for verification and observed an excellent correlation between Ub max s and the DC 50 values obtained by traditional Western Blot.
Maria-Anna Trapotsi et al. reported that Cell Painting is an unbiased high-content imaging method that can be used for high-throughput analysis of cell morphological changes caused by PROTACs 108 ( Figure 5D). They used a Cell Painting assay to quantitatively study the phenotypic characteristics of cells induced by a series of different PROTAC and non-PROTAC compounds and screened out active compounds. When they looked at specific examples of published PROTACs, they observed that PROTACs targeting BRD4 showed some activity; the cell phenotype of the PROTAC group targeting CDK9 was consistent with the function of CDK9 in the cell cycle. At the same time, they trained machine learning models to predict the mitochondrial toxicity of drugs and demonstrated that Cell Painting profiles could provide safety assessment information for compounds. Their approach is expected to be used for high-throughput PROTAC screening and to influence drug design based on the safety information provided.

REGULATORY ACTIVATION AND TARGETED DELIVERY
Modern pharmaceutics has advanced to precision medicine that requires precise administration and controlled release of the therapeutics to the diseased tissue or even achieving intracellular delivery. This is particularly important for PROTACs as they can only function by recruiting intracellular UPS systems and regulating intracellular targets. Recently, precision medicine principles of stimuli-responsive activation and targeted delivery have been successfully applied to PROTAC technologies, giving good examples of how interdisciplinary knowledge is nowadays helping the modern development of medicines.

| Photoactivation
In most cases of therapeutic development, tissue-specific targeted delivery or precise activation needs to be considered to avoid side effects on normal tissues. This is particularly critical for therapeutics with catalytic activity like PROTAC. The most common way to regulate PRO-TAC activation is via photoactivation, usually by adding a photolysis group to E3 ligands to prevent the recruitment of E3 ligases without designated light conditions. Cyrille S. Kounde 110 The addition of a nitroveratryloxycarbonyl group, which can be photolyzed when irradiated by ultraviolet A (UV-A) in vitro, to the glutarimide NH of pomalidomide can block its interaction with E3 ligase CRBN. They added the cage group to pomalidomide and two other PROTACs, dBET1 (BRD4 degrader) and dALK (ALK degrader), and observed the light-induced degradation of IKZF1, BRD4, and ALK proteins. The former report degraded BRD4 for hours, while the latter one took 15 min to degrade EML-ALK fusion protein and the dosage of drugs were both 1 μmol.
The application of adding photolysis groups to a specific E3 ligand is limited to several types of E3 ligases and the corresponding ligands. For other common E3 ligases such as MDM2, IAP, RNF114, 111 DCAF16, 112 KEAP1, 113 and so on, it is necessary to design new caging methods. Adding groups with light-responsive configuration changes is another common way to achieve lightcontrolled activation.
Yu-Hui Jin et al. added an azobenzene group to the linker part of PROTACs, optimized the length of the linker, and constructed Azo-PROTAC to induce the degradation of Abl and BCR-Abl. 114 When the PROTAC molecule is exposed to UV-C light, the trans-structure will be transformed into the cis-form; when exposed to white light, the cis-configuration will be transformed into the trans-configuration, which is a configuration with activity to induce degradation. They observed that the activity of the PROTAC with azobenzene switch could be regulated by UV irradiation in myeloid leukemia K562 cells.
Qisi Zhang et al. also reported an azo photo switch, arylazopyrazole, which can reach a photostationary state (PSS) condition with 75% trans-isomers after 457 nm UV irradiation and can be converted to 99% cis-isomers under 365 nm UV irradiation, and the trans-configuration is also more active. 115 (Figure 6A) By constructing the PROTAC molecule containing the photo switch, they achieved UV-regulated degradation of BRD2/4 and kinases such as FAK, AURORA-A, TBK1, and so on.
Placing a photo switch in the linker will vastly expand the suitable E3 ligases and ligands, but it will increase the difficulty of optimization of PROTAC linkers. And it is not applicable for some reported PROTAC molecules with considerable activity because it needs to be redesigned. None of the light-activated PROTAC molecules mentioned above has been used in in vivo tests. In fact, light-dependent regulatory activation may have some defects in clinical use, such as UV radiation may cause DNA damage, and the tissue transmittance of light irradiation is low, which limits the use of these methods.

| Other stimuli-responsive regulation methods
In addition to light activation, there are also some stimuli-responsive regulation methods designed according to the stimuli of focus in diseases, mainly tumors. Cheng et al. reported on hypoxia-activated PROTACs 116 ( Figure 6B). Different from normal tissue cells, tumor tissue blood vessels cannot meet the oxygen needs of tumor cells, which renders the tumor cells in a state of hypoxia. 117 Hypoxia-activated departure groups (HALGs) can be removed from the maternal structure under hypoxia conditions and are used to construct prodrugs for tumor hypoxia activation. 118 They linked two kinds of HALGs, (1-methyl-2-nitro-1H-imidazol-5-yl) methyl and 4-nitrobenzylgroups, respectively, to the 4-NH of the 4aniline quinazoline core of gefitinib, a listed EGFR inhibitor, which led to a sharp decrease of the binding affinity between the ligand and EGFR, thus constructing hypoxia-activated Ha-dEGFR-1 and Ha-dEGFR-2. The activities of inducing EGFRDel19 degradation of the two PROTACs were lower than those without HALGs, while the degradation degree of EGFRDel19 in HCC4006 cells induced by hypoxia was significantly higher than that under normoxia.
Liang et al. reported the regulation of enzymecatalyzed activation 119 (Figure 6D). They attached the group containing trimethylhydroquinone structure to the hydroxyl group of the VHL ligand, and the endogenous overexpression of NAD (P) H-quinone dehydrogenase 1 (NQO1) in cancer cells 120 could reduce trimethylhydroquinone and cause prodrugs to remove chemical modification, thus activating PROTACs.
Maslah et al. reported the regulation of the combination of enzyme catalysis and reactive oxygen species (ROS) activation. They linked the aryl boronic acid group, a ROS-induced leaving group, 121 to the same position of the VHL ligand. In tumor cells, the products of catalytic reduction of β-Lapachone by NQO1 can enhance oxidative stress and produce ROS. ROS can remove aryl boronic acid groups and release active PROTAC components. They used this method to improve the cell selectivity of PROTAC-induced degradation of BRD4.
Liu et al. also reported PROTACs activated by ROS 122 ( Figure 6C). They constructed ROS-PROTAC by introducing arylboronic acid into the parental PROTACs and observed that it could be activated by endogenous ROS and effectively degrade the target protein BRD3. Concentration-dependent asssay shows that 50nM Pre-PROTAC was able to degrade its traget BRD3 after 24 h. However, similar to photoactivation, these regulatory modes often require additional switching groups to the already large PROTAC molecules, which may lead to a further decrease in PK and cell membrane permeability.

| Targeting moiety conjugates for targeted delivery
The idea of targeted delivery is nowadays common in therapeutics development strategies. The accumulation at the focus can produce higher local administration concentration and lower systemic concentration, which can enhance efficacy and meanwhile reduce side effects. The commonly used targeted delivery strategies for PROTACs include drug conjugates and nanomaterial delivery.
Maneiro et al. used antibody-PROTAC conjugates to target HER2-positive cells to induce BRD4 degradation 123 ( Figure 7A). They linked PROTACs with the azide group at the hydroxyl site of the VHL ligand to dibromomaleimide-strained alkyne-functionalized trastuzumab. They obtained an antibody-PROTAC complex containing four PROTAC molecules. It targets HER2positive breast cancer cell lines through antibodies and releases active PROTAC molecules after being internalized and hydrolyzed by lysosomes. They observed that the complex selectively degraded BRD4 in HER2-positive cancer cells. This method can achieve targeted protein degradation with tumor tissue specificity. Still, it depends on antibodies against tumor antigens, and there are inherent disadvantages of monoclonal antibody drugs ZHAO ET AL.
-17 of 32 F I G U R E 6 Photo-activated PROTACs. (A) AP-PROTAC is more active at trans-configuration and can be converted to 99% cis-isomers under 365 nm UV irradiation. Therefore it can be easily regulated; 115 (B) Only by discarding the HALG with the help of hypoxia conditions can the Ha-PROTAC be activated, or HALG is going to hinder EGFR from being targeted; 116 (C) ROS is able to remove arylboronic acid which blocks E3 ligase from the PROTAC and activate it, therefore induces BRD4 degradation; 122 (D) The regulation of PROTAC can also be regulated by enzyme-catalyzed activation. With the help of overexpressed NQO1 in cancer cells, Caged-PROTAC will be released and activated. 119 and antibody-coupled drugs, such as poor stability, short plasma half-life, potential immunogenicity, difficulty in cell uptake, high synthesis cost, poor uniformity of drug binding sites and quantities, and so on.
Liu et al. reported folate-modified PROTACs 124 ( Figure 7B). The folate group is connected to the hydroxyl of the VHL ligand. Folate can recognize folate receptor α (FOLR1) in cancer cells. 125 FOLR1 is highly expressed in many cancer types, such as ovarian cancer, lung cancer, and breast cancer, while the expression is low in normal tissues. 126 The strategy of targeting FOLR1 is commonly seen in the field of tumor imaging and tumor-targeted drug delivery. 125 Folate-modified PROTACs tend to be transported to cancer cells with high expression of FOLR1, where it is hydrolyzed by hydrolase and removing folate modification and releasing active PRO-TACs to induce the degradation of endogenous target proteins. They constructed BRD PROTAC (folate-ARV-771), MEK PROTAC (folate-MS432), and ALK PROTAC (folate-MS99) and observed that they all degraded the target protein in a FOLR1-dependent manner. However, folate modification will increase the molecular weight of PROTACs and may affect the PK properties and bioavailability. Although folate molecules are hydrophilic with electric charge, it is still possible to enter the cell through osmosis, resulting in a certain non-specific effect.
He et al. reported the targeting strategy of nucleic acid aptamer-PROTAC coupling 127 ( Figure 7C). They coupled the nucleolin targeting aptamer AS1411 to the hydroxyl group of the VHL ligand of the PROTAC targeting BET F I G U R E 7 Targeting moiety conjugates for targeted delivery. (A) Antibody-Proteolysis-targeting chimeras (PROTAC) conjugates are designed to target HER2-positive cells to induce BRD4 degradation; 123 (B) E3 ligand can be linked to folate, which is able to recognize FOLR1 highly expressed in cancer cells and low in normal tissues; 124 (C) Aptamer AS1411 binds to nucleolin, thus inducing internalization of its coupled PROTAC and is going to be reduced by GSH in tumor cells, releasing active PROTAC. 127 ZHAO ET AL.
-19 of 32 via a dithiocarbonate linker. AS1411 targets and binds to the nucleolin, 128 which is highly expressed on the tumor cell membrane, 129 to make the complex enter the cell through internalization. Then, the rich endogenous glutathione (GSH) in the tumor cell 130 reduces the disulfide bond, and the free sulfhydryl group attacks the carbonate bond, thus releasing active PROTACs. They observed that the complex had better tumor targeting in the MCF-7 transplant tumor model than BET PROTACs without conjugating AS1411, which could enhance antitumor activity while reducing side effects. However, similar to antibody conjugates, this method also depends on tumor-specific aptamers and needs to face problems such as poor serum stability and tumor permeability of nucleic acids. Its efficacy will also be affected by the heterogeneous expression of receptors in different types of tumor cells.

| Targeted delivery and enhanced therapy with nanomaterials
PROTAC drugs usually have poor pharmacokinetic properties, while nano-based drug delivery systems (NDDS) can improve the behavior of drugs in vivo, enhance drug stability, and achieve targeted and space-time-controlled drug delivery through reasonable design.
The other NP reported in the same work is extracellular tumor acidity-activatable pre-targeted NPs, which are called PED NPs, which is formed by selfassembly of diblock copolymer mPEG113-P(EPA m -r-DBCO n ) (EPA refers to ethylenepropylamine). NPs with PEG corona can prolong the blood circulation of PRO-TACs. The removal of PEG corona after the GG site is cut by MMP-2 in the tumor can make NPs retain and accumulate in the tumor and enhance the internalization of NPs by tumor cells. When PED NPs are injected intravenously, the protonation of EPA tertiary amine causes PED NPs dissociation in the acidic microenvironment (pH < 6.8) in tumor tissue, and the DBCO group is exposed to the extracellular matrix (ECM). The click reaction with the azide group of N 3 @PGDA7 NPs could further increase the intratumoral retention of drug-loaded NPs.
When the uncrowned N 3 @PGDA7 NPs enter the tumor cell, the intracellular acidic environment (pH < 6.2) causes the protonation of the DPA tertiary amine and leads to the dissociation of the drug-loaded NPs. Meanwhile, the intracellular GSH 130 reduces the disulfide bond to make the dithiocarbonate side chain self-cleavage, thus releasing the active ARV771 to induce BRD4 degradation, which promotes the occurrence of the tumor cell apoptosis dependent on caspase-3. On the other hand, the PPa in PGDA released by NPs dissociation can be used in photodynamic therapy (PDT). ROS can be induced by irradiating PPa with a 671nm laser to promote apoptosis further. They used this platform to achieve the accumulation of drug-loaded NPs in tumors and tumor tissue infiltration. A combination of PROTACs and PDT therapy inhibited more than 95% of MDA-MB-231 TNBC tumor growth. This delivery technique utilizes the acidic conditions of the tumor microenvironment and rich MMP-2, as well as the acidic conditions and reduction environment in tumor cells.
Compared with the methods of incorporating tumortargeting moieties, these nanomaterial-based delivery techniques have broader applications. It is also applicable to be developed as a drug delivery platform combining PROTACs and other therapies. A key stage of this platform is the removal of PEG corona by MMP-2 cleavage. Moreover, MMP-2 is not only specific to tumor tissue but also highly expressed in other diseases such as vascular injury, 133 inflammation-related diseases, 134 and oxidative stress, 135 Figure 8B). They linked poly (cyclo-pentadithiophene-alt-benzothiadiazole) (PCB) to PROTACs via PEG and peptides that can be cut by the cancer biomarker Cathepsin B (CatB) 138 and the polymer self-assembled to form NPs. Semiconducting polymer PCB has near-infrared absorption and can be used as a photothermal element for photothermal therapy (PTT). They selected two types of POI as PROTAC targets, indoleamine 2,3-dioxygenase (IDO) 136 and cyclooxygenase 1/2 (COX-1/2). 137 When taking effect, PCB can produce singlet oxygen ( 1 O 2 ) to kill tumor cells, induce immunogenic cell death (ICD), release tumor-associated antigens (TAAs), promote DC maturation and T cell activation, and enhance anti-tumor cell immune response. Meanwhile, CatB cleavage releases active PROTACs, induces IDO degradation, reduces the metabolic transition from tryptophan (Trp) to kynurenine (Kyn), reverses immunosuppression and promotes the activation of effector T cells, 136 Or induces the continuous degradation of COX-1/2, and reduce its metabolite prostaglandin E 2 (PGE 2 ), to reduce the activation of PGE 2 on immunosuppressive cells such as regulatory T cells (Tregs), M2-type macrophages (M2-MACs), and myeloid suppressor cells (MDSCs). 137 Moreover, Zhang et al. developed a multi-functional nano-PROTAC, which uses pH/GSH responsive disulfide bond-linked poly(lactic-co-glycolic acid)(DS-PLGA) to encapsulate the PROTAC (dBET6) targeting BRD4 and camouflage it with engineered CRV highly expressed Lewis lung carcinoma (LLC) cell membranes to construct an intelligent NDDS CREATE with a biomimetic strategy. 139 (Figure 9A) CREATE can target both lung cancer F I G U R E 8 Targeted delivery with nanomaterials with cleavable linkers. (A) Pre-targeted NP will dissociate in the tumor microenvironment, and DBCO group exposed can react with POLY-PROTAC NP to increase intratumoral retention. POLY-PROTAC NP will gradually dissociate with the help of MMP-2, H+, and GSH in the tumor microenvironment and sequentially release the active PROTAC. 131 (B) PROTAC molecule is linked to PEG by the cancer biomarker CatB, and the other side of PEG is PCB which can inhibit tumor cells as well. 136 ZHAO ET AL.

F I G U R E 9
Targeted delivery with nanomaterials loaded with ligand in aid of recognizing the target. (A) BRD4-targeting PROTAC is encapsulated with DS-PLGA, and the complex is camouflaged with CRV highly expressed LLC cell membranes. This system CREATE is able to target both lung cancer cells and tumor-associated macrophages. 139 (B) PATU-8988 cell membrane is extracted and encapsulated PIPD by ultrasound. Therefore this system is endowed with good serum stability and biocompatibility. 140 (C) BRD4-targeting PROTAC is encapsulated with Dox in micelles formed by SP-PEG-PDLLA and mPEG-PDLLA. SP can specifically bind to NK-1R in gliomas. 141 (D) BRD4-targeting PROTAC is encapsulated in nanoliposomes modified by galactosylceramide, so the nanoliposomes are able to target hepatocytes specifically. 145 cells and tumor-associated macrophages (TAMs) and release dBET6 in the cells to induce apoptosis. Direct simultaneous elimination of tumor cells and TAMs can help to reshape the tumor microenvironment and achieve tumor inhibition. They observed that the NPs could inhibit tumor growth in subcutaneous and orthotopic tumor-bearing mice.
Similarly, Fan et al. extracted the cell membrane of PATU-8988 cells and encapsulated PIPD (A PROTAC that induces PDE δ degradation and has an excellent antitumor effect on tumors with KRAS mutation.) by ultrasound 140 ( Figure 9B). The camouflage of the tumor cell membrane gives the system good serum stability and biocompatibility, as well as the ability to target homologous PC cells.
Using the same principle, He et al. constructed NPs modified with cyclic (Arg-Gly-Asp-d-Phe-Lys) (cRGDfk) peptides, 143 which encapsulated both ARV-825 and doxorubicin (DOX), targeting gliomas based on the high specificity and stable binding ability of cRGDfk to α v β 3 integrin. 144 Aishwarya Saraswat et al. developed ARV-825 nanoliposomes modified by galactosylceramide 145 (Figure 9D). Galactose can specifically recognize the asialoglycoprotein receptors (ASGPRs), which are mainly expressed on the surface of hepatocytes, 146 so the nanoliposomes can target hepatocytes for the treatment of liver cancer. They observed that the nanosystem showed the ability to induce apoptosis in 2D cells and 3D multicellular liver cancer models.

| MEDICINAL CHEMISTRY TO CHEMICAL BIOLOGY: BIOLOGICAL MACROMOLECULES-BASED PROTACS
PROTACs work through the "event-driven" mechanism. Compared with the "occupancy-driven" mechanism of small molecules, PROTACs do not require a high affinity for POI or bind POI in an active and tractable pocket, so it has a broader range of targets. However, because the development of POI mainly depends on small molecular ligands, for proteins that are difficult to develop small molecular ligands due to the lack of binding sites or larger interaction surfaces, such as most TFs and skeleton proteins, it is still challenging to develop PROTACs. In this situation, according to the biological characteristics of POI, PROTACs combined with biological macromolecules (mainly nucleic acid) as targeting ligands arise at the historic moment.
Ghidini et al. reported an RNA-PROTAC, which targets RNA binding protein (RBD). 147 They used the sequence AGGAGAU, which can be recognized by the zinc finger domain of the RBP Lin28A 148 (Figure 10A), as a warhead to construct the RNA-PROTAC utilizing VHL. Because single-stranded RNA (ssRNA) is easily degraded, they replaced the phosphodiester skeleton with thiosulfate and alkylated ribose 2 0 hydroxy to resist nuclease cleavage. The PROTAC can induce the degradation of Lin28 and inhibit the growth of tumor cells. This method is expected to be extended to more RBP targets. Liu et al. and Shao et al. reported TF-PROTACs 149 ( Figure 10A) and PRO-TACs, 150 respectively. The principles of the two are similar. The double-stranded DNA (dsDNA) sequences specifically recognized by TFs are used as warheads to construct PROTACs. Using this method, Jing Liu et al. developed VHL-based PROTACs targeting NF-κB and E2F, which can effectively degrade intracellular p65 and E2F1 proteins. 149 Jingwei Shao et al. developed a VHL-based PROTAC targeting lymphoid enhancer-binding factor 1 (LEF1) and a CRBN-based PROTAC targeting ETS related-gene (ERG), with DC 50 values of 25 and 182.4 nM, respectively. 150 The results show that this idea can be used as a general strategy for the targeted degradation of "undruggable" TFs. Similarly, Patil et al. constructed a G4-PROTAC using G-quadruplex (G4) to target and induce the degradation of the G4-binding protein RHAU 151 ( Figure 10A). Because the function of G4-binding proteins is not very clear at present, G4-PROTACs are expected to be a tool to explore the function by making use of the properties of immediately inducing degradation of PROTACs.
The aptamer is a class of oligonucleotides, either ssDNA or ssRNA, obtained by high-throughput screening, which can bind to the target protein. Aptamers usually bind to their targets with excellent specificity and affinity, making them a type of molecule with great potential as a PROTAC warhead. Zhang et al. linked the aptamer AS1411 128 with VHL ligands to construct PRO-TAC ZL216, which effectively degraded nucleolin in vivo and in vitro and can inhibit the proliferation of breast cancer cells 152 ( Figure 10B). In view of the high expression of nucleolin in tumor cells 129   (a) Aptamer AS1411 was utilized as a warhead and linked to VHL ligand to compose and constituted PROTAC ZL216, which can specifically target nucleolin in tumor cells and degraded nucleolin in vitro and can inhibit the proliferation of breast cancer cells. 152 (b) AS1411 can also be added to a small molecule PROTAC to target the tumor cell. 127 Nowak et al. developed a similar system. 155 The tag is a mutant BRD4 BD1 L94V . By modifying the ligand I-BET762, the PROTAC molecule XY-06-007 they constructed using CRBN ligands can degrade the mutant. The function of these protein tags is similar to that of scaffolds, connecting target proteins to PROTAC molecules.
Xu et al. developed PROTACs by using RNA as scaffolds. 156 ( Figure 11A) The RNA scaffold is mainly composed of two RNA aptamers modules: one module is for POI targeting, and the other one is for recruiting a fluorogenic group DFHBI-thalidomide chimeric, namely Dth. 157 By recruiting Dth and the sequential CRBN, the RNA scaffold shortens the distance between POI and promotes ubiquitination. Using this strategy, they successfully induced the degradation of p50, p65, and E2F1 proteins in the cells.
They further expanded the RNA scaffold to include three aptamers, two of which target two different proteins. It can help degrade both p50 and E2F1 proteins at F I G U R E 1 1 Biological macromolecules function as a scaffold and connect target proteins and PROTAC molecules. (A) Aptamer was considered as a scaffold and recruited its ligand, which was linked to E3 ligand, and POI was recruited by another aptamer at proximity and degraded. 156 (B) The single site of bromodomain of BRD4 was mutated, and the warhead of its PROTAC was modified as well to target the mutated protein specifically. Therefore, the mutated protein can be utilized as Bromotag and fused to the undruggable POI. 154 C. CRISPR-Cas9 system was utilized as a "scaffold" to develop a TRAFTAC system. HT7 was fused with dCas9 to recruit HaloPROTAC, and TRAFTAC can target dCas9 with TF recruited by its dsDNA simultaneously. TF was finally ubiquitinated and degraded. 158 the same time and has a better inhibitory effect on MCF-7 cells than that of a single target. This method transfers the pressure of optimizing the small molecule PROTAC linker part to that between the two aptamers of the RNA scaffold, which is more straightforward in design and usually only needs to set different lengths of A bases, and the RNA scaffold dramatically reduces the complexity of chemical synthesis in the process of developing small molecule PROTACs.
Similarly, in double targeting or double binding degradation, it has similar advantages over the heterogeneous trifunctional PROTACs of Mengzhu Zheng et al. 82 and trivalent PROTACs of Satomi Imaide et al. 83 However, in order to achieve a better degradation effect, their strategy requires the concentration of Dth to reach dozens μM, which may be due to the addition of RNA scaffolds that increase the number of interaction pairs and the more complicated structure of the complex required for ubiquitination, which remains a problem to be solved.
Kusal T.G. Samarasinghe et al. used the CRISPR-Cas9 system as a "scaffold" to develop a TRAFTAC system for TFs degradation. 158 (Figure 11C) They covalently connect the TF-bound dsDNA to one end of the CRISPR-RNA that can bind to the ectopic expressed dCas9-Halotag7 fusion protein (dCas9HT7). Halo-PROTAC molecules can recognize Halo-tag and recruit VHL E3 ligases to ubiquitinate the TF binding to the dsDNA region. Through this strategy, they achieved the targeted degradation of p50 and Brachyury.
In the context of these three approaches, the Broma-Tag method can be applied to any protein in theory but must be labeled at the gene level. Employing RNA as a scaffold does not require modification of the gene, but it depends on the development of RNA aptamers. The warhead of TRAFTAC comes from an endogenous binding sequence of the target protein, but it is only suitable for TFs. And compared with TF-PROTACs and O'PROTACs, the TRAFTAC system is too complex.

| DISCUSSION AND PERSPECTIVE
The emergence and fast development of PROTAC technology largely complement the unmet needs in the development of small molecular inhibitors. First, PRO-TACs designed with small molecular inhibitors as the targeting moiety usually have better selectivity compared to the small molecular inhibitors themselves. 30,34,35 It is common that small molecular inhibitors have less satisfactory selectivity when confronting proteins in the same family or kinases with similar structures. When a PRO-TAC is designed on top of a small molecular inhibitor, it adds another condition to the affinity between the ligand and the target, that is, the formation of ternary complexes, which makes it possible to produce better selectivity than small molecular inhibitors.
Next, it is challenging for small molecular inhibitors to inhibit the non-enzymatic action of the target protein. 21,22,39,40,53,54 Small molecular inhibitors mainly bind to the enzyme active pocket to inhibit catalytic activity, but other functions of proteins, such as scaffold function, cannot be inhibited or cannot be inhibited at the same time as enzyme activity. PROTACs, instead, completely degrade the target protein, which makes the protein lose all its functions.
Moreover, the dosage is limited by adverse reactions. 34,35,53,54,59,60,77,79 Because the small molecule inhibitor works through the "occupancy driven" principle, it requires not only the affinity but also the drug concentration, but the toxicity and side effects will also increase with the dose, which in turn limits the dosage so as not to reach the concentration needed for a better curative effect. PROTACs work through the "eventdriven" principle, have catalytic properties, and can usually produce better therapeutic results at low doses.
More importantly, PROTACs could overcome the drug resistance of small molecular inhibitors. 26,27,30 The mechanisms of drug resistance include drug resistance mutation and protein overexpression. 159 Because the affinity of PROTACs is not as high as that required by small molecular inhibitors, and it can be reused, it can better deal with the drug resistance caused by these mechanisms. Another kind of drug resistance is achieved through the bypass signal pathway. 160 Because the degradation of target proteins by PROTACs is immediate and lasting, it can kill cells before they adapt to the pressure, which is an advantage compared with small molecules.
Of course, this advantage is not absolute. Some studies have shown drug resistance in PROTACs, and an important factor is a mutation in E3 ligase complexes, such as CRBN itself and Cullin 2 scaffolds in VCB. 161 Therefore, it is of great significance to develop more E3 ligands and explore the possibility of PROTAC-induced POI degradation by other E3 ligases in the future. At the same time, it is also beneficial to develop tissuespecific PROTACs according to the different tissue expression of E3 ligases and to develop cell cycle-specific PROTACs or concomitant drugs according to the dynamic expression process of E3 ligases in the cell cycle. 162 In addition, in order to better deal with the drug resistance caused by the bypass signal pathway mechanism, it is also worth exploring suitable bypass targets, designing a reasonable combination strategy, or developing dualtarget PROTAC degradants.
In spite of all advantages listed above, PROTACs cannot replace small molecular inhibitors. After all, the development of PROTACs is closely related to that of small molecular inhibitors. Although PROTACs can use peptides or other targeting moieties as warheads, PRO-TACs based on small molecules are the main force, given the humoral instability and poor membrane permeability of peptides. Therefore, the development of PROTACs often depends on small molecular inhibitors. The most common design idea is through the crystal structure of target proteins together with small molecular inhibitors or computer simulation models to select appropriate solvent exposure sites to connect different kinds and lengths of linkers that have a commonly used E3 ligand at the other end.
And also, PROTACs have poor PK properties and bioavailability compared with small molecular inhibitors. In order to solve this problem, it can start with PROTACs themselves, such as developing E3 ligands with lower molecular mass and more stability in the humoral environment and exploring factors other than "rule-of-five" 11 to guide structural design or developing a reasonable delivery system. Another idea is to improve the efficiency of PROTAC-induced POI degradation to make up for the deficiency in PK and permeability, which can be achieved by developing ligands with higher affinity, constructing multivalent PROTACs 83 and covalent reversible PRO-TACs, 163 and reasonably designing PROTAC molecules.
Reasonable design is a major difficulty in the field of PROTACs because there are no guiding principles at present. Artificial intelligence-aided methods can accelerate the design process of PROTACs to a certain extent, but some current methods [95][96][97][98][99][100][101] are not very mature. For one thing, computer-based methods in the field of PRO-TACs cannot provide structure-activity relationship (SAR) information and guide molecular design like the development of small molecular drugs targeting proteins, which may be due to the excessive complexity of ternary complexes induced by PROTACs and the lack of crystal structure and efficacy data.
For another, the current computer simulation methods (also some other biochemical methods [102][103][104][105] for rapid screening focus on the formation of ternary complexes induced by PROTACs, which is unfortunately affected by some environmental factors, such as DT2216, can induce the formation of VCB-PROTAC-BCL-2 complex in extracellular, but not in intracellular. 78 These methods do not take these factors into account. Moreover, the ternary complex is not a sufficient condition for the ubiquitination of POI. For example, PZ703b can induce the ternary complex of VCB-PROTAC-BCL-2 but cannot degrade it. 77 The deep-seated influencing factor may be the accessibility of both the physical space and chemical reactivity of lysine primary amines to the E2 enzyme catalytic site. Some studies have proposed that due to the limitation of the rotation range of Cullin 2 and Rbx1 in the complex of VCB-PROTAC-POI, the POI region that can be touched by the E2 catalytic site is a band shape, and lysins outside this region cannot be ubiquitinated. At the same time, if the side chain amino group of lysins in this region has electrostatic interaction, hydrogen bonding interaction etc., with other residues so that when it is inert in chemical reactivity, it cannot be ubiquitinated either. 79 In addition to the two aspects discussed above, nucleic acid-based PROTACs, such as RNA-PROTACs, TF-PROTACs, O'PROTACs, Aptamer-PROTACs and so on, also need to optimize linkers, but the current modeling methods are not suitable for nucleic acid-PROTACs. Nucleic acid-PROTACs have good water solubility and can be delivered with the help of state-of-theart nucleic acid drug delivery systems. Compared with PROTACs based on polypeptides or small molecules, the larger relative molecular mass will not be a limiting factor for the development of this kind of PROTAC. Among them, aptamer-PROTACs are to be more widely used, which are not limited to the degradation of RNA-binding proteins, DNA-binding proteins, or G4-binding proteins. In theory, selecting from a library consisting of modified bases 164 and artificial bases, 165 in addition to the four natural nucleobases, combining the high-throughput screening methods, aptamer targeting any protein of interest can be obtained, and the selectivity and affinity of aptamer can be improved by adjusting the screening pressure. 166 Therefore, aptamer-PROTACs have a good development prospect.