Development of a p-hydroxybenzyl alcohol (PHOBA)-linked glutamate prodrug for activation by Pseudomonas carboxypeptidase G2 (CPG2)

Abstract

Directed enzyme-prodrug therapies used for targeted drug delivery require prodrugs that are chemically stable and processed efficiently by the activating enzyme. We recently reported the development of AMS-6-Glu (2), a glutamate-masked version of the cytotoxic natural product 5´-O-sulfamoyladenosine (AMS, 1) that can be activated by Pseudomonas carboxypeptidase G2 (CPG2). Herein, we report the development of a second-generation prodrug, AMS-5´-PHOBA-Glu (5), that undergoes cleavage by CPG2 with >160-fold higher efficiency. Use of a p-hydroxybenzyl alcohol (PHOBA) self-immolative linker overcame unexpected chemical instability observed with a conventional p-aminobenzyl alchohol (PABA) linker.

Graphical Abstract

Chemotherapy with cytotoxic agents remains an important component of cancer therapy, but its effectiveness is generally constrained by off-tumor toxicity.¹ To address this problem, a number of strategies have been developed to target cytotoxic agents specifically to tumors,² most prominent among these being antibody–drug conjugates (ADCs).^3–5 However, because ADCs require that the drug be conjugated covalently to the antibody at sites that do not compromise antibody structure or antigen binding, only a very limited amount of the drug can be delivered to the tumor. As an alternative approach, a variety of directed enzyme–prodrug strategies have been developed in which an antibody or other targeting vehicle is used to deliver an enzyme, rather than the parent drug itself, to the tumor.^6,7 A non-toxic prodrug is then administered systemically, and is activated locally at the tumor by the enzyme, providing amplification of the drug through enzyme catalysis. Antibody-directed enzyme prodrug therapies (ADEPT) have advanced to clinical evaluation,⁸ and other implementations use bacteria, viruses, polymers, or cell-surface receptor ligands as delivery vehicles.^6,7 Recently, we have developed a novel platform for combined cellular and small-molecule therapy in which a chimeric antigen receptor (CAR)-T cell is engineered to express and secrete a prodrug-activating enzyme.⁹ This synthetic enzyme-armed killer (SEAKER) cell is administered first and undergoes antigen-activated logarithmic cell proliferation at the tumor site, with each cell expressing multiple copies of the enzyme, which then catalytically activate the systemically administered prodrug, providing three layers of drug amplification. Moreover, this ‘targetable cellular micropharmacy’ strategy combines the cytolytic activity of the T-cell with the orthogonal cytotoxic activity of the small-molecule drug, providing advantages over either individual modality.¹⁰

These approaches require prodrugs that are accepted as substrates by the activating enzyme and are otherwise chemically and metabolically stable. In our initial report on SEAKER cells, we engineered the cells to express and secrete a Pseudomonas sp. carboxypeptidase G2 (CPG2), which cleaves C-terminal glutamate moieties and has been used previously in ADEPT therapies (Figure 1a).⁸ We designed the prodrug AMS-6-Glu (2, Figure 1b), a masked version of the highly cytotoxic nucleoside natural product 5´-O-sulfamoyladenosine (AMS) (1),^11,12 and demonstrated specific activation by CPG2 both in vitro and in vivo.⁹ However, enzyme kinetic analysis indicated that AMS-6-Glu (2) was cleaved relatively inefficiently by CPG2 compared to other known substrates.^6,13 Thus, we sought to design a prodrug that could undergo more efficient activation by CPG2. Herein, we report the rational development of an improved prodrug, AMS-5´-PHOBA-Glu (5), which undergoes >160-fold more efficient CPG2 cleavage and exhibits enzyme-activated cytotoxicity in vitro. In the course of this work, we observed unexpected and pronounced chemical instability of a conventional p-aminobenzyl alcohol (PABA) self-immolative linker (cf. 4) and, based on mechanistic insights, used an NH-to-O modification in a p-hydroxybenzyl alcohol (PHOBA) linker to avoid this problem.

Figure 1.

CPG2-labile glutamate prodrugs. (a) Inactive prodrugs undergo cleavage of glutamate (blue) by CPG2 then spontaneous decomposition of linker (gray) to unmask active drug (red). (b) Structures of parent compound AMS (1), previously reported prodrug AMS-6-Glu (2), and proposed prodrug variants masked at the 6-amino position (3) or 5´-sulfamate nitrogen (4, 5).

Pseudomonas CPG2 is a bacterial hydrolase enzyme identified on the basis of its ability to cleave a C-terminal glutamate from folate derivatives.¹⁴ This enzyme has been used extensively in directed enzyme–prodrug therapy systems and cleaves glutamate substrates linked via N-amide, urea, and urethane bonds to aromatic moieties.^6,7 Importantly, the substrate specificity of CPG2 is distinct from that of human glutamate carboxypeptidase II (PSMA),¹⁵ enabling design of prodrugs that are specifically cleaved by CPG2 and not by endogenous PSMA.

In our earlier work, AMS (1) was masked with a glutamate moiety attached via a carbamate linker to the adenine 6-amino group in AMS-6-Glu (2) (Figure 1b).⁹ However, while AMS-6-Glu (2) was highly stable in vitro and in vivo, cleavage with recombinant CPG2 proceded with poor catalytic efficiency (k_cat/K_M = 0.00065 μM^–1s^–1), ≥1,600-fold lower than a known CPG2-labile substrate, methotrexate (k_cat/K_M = 1.1 μM^–1s^–1).⁹ In most CPG2-labile substrates reported previously, the aromatic moiety is para-substituted.^16–18 Thus, we postulated that steric encumbrance of the bicyclic adenine ring of AMS-6-Glu (2) might be responsible for the modest catalytic efficiency of CPG2 cleavage. We noted that self-immolative linkers containing a p-substituted aromatic moiety have been used successfully with CPG2.^16–18 In particular, p-aminobenzyl alcohol (PABA) linkers¹⁹ have been reported in prodrugs that are cleaved effectively by CPG2 (k_cat/K_M = 0.91–21.1 μM^–1s^–1),^16–18 leading to spontaneous elimination of the linker to form an aza-quinone methide byproduct.¹⁹

Accordingly, we sought to insert a PABA linker into AMS-6-Glu (2) in the second-generation prodrug AMS-6-PABA-Glu (3) (Figure 1b). In initial efforts, protected PABA-Glu alcohol 6 (Figure 2a) was activated with p-nitrophenylchloroformate (Et₃N, THF, 50–80%).¹⁸ However, attempted coupling of the resulting activated carbonate to the 6-amino group of protected adenosine derivative 7 provided only trace amounts of the desired product (NaH, DMF, 50 °C) or no reaction (pyridine, DMF, 50 °C). Use of other activating reagents gave no reaction (CDI²⁰ or CBMIT²¹) or dimerization of the alcohol (triphosgene²² or N,N´-disuccinimidyl carbonate²³). We then investigated an alternative route involving direct activation of 6-amino group of adenosine derivative 7 for subsequent coupling to alcohol 6, but observed no activation (CDI, NaH, DMF, rt) or dimerization of the adenosine fragment (phosgene, triphosgene, or p-nitrophenylchloroformate).

Figure 2.

Synthetic approaches to AMS prodrugs 3 and 4. (a) Structures of coupling partners used in attempted synthesis of AMS-6-PABA-Glu (3). (b) Attempted synthesis of PABA-linked prodrug 4. DMAP = 4-(dimethylamino)pyridine; DMF = N,N-dimethylformamide; TBS = t-butyldimethylsilyl; TFA = 2,2,2-trifluoroacetic acid.

Thus, we next investigated relocation of the masking group to the 5´-sulfamate nitrogen in prodrug AMS-5´-PABA-Glu (4) (Figure 1b). We anticipated that masking at this position would still abrogate the cytotoxicity of AMS, based on extensive previous studies of other 5´-N-acylated AMS derivatives developed as inhibitors of adenylate-forming enzymes.²⁴ Thus, protected AMS derivative 8 was coupled with p-nitrophenylcarbonate-activated PABA-Glu 9, providing the protected carbamate 10 (Figure 2b). However, although this protected intermediate could be observed by LC-MS analysis, it proved surprisingly unstable and decomposed spontaneously to form the parent compound AMS (1). Carrying the crude intermediate 10 directly on to deprotection afforded the desired product 4, which could be isolated by HPLC, but also rapidly decomposed to AMS (1).

Although PABA-linked glutamate prodrugs have been reported previously,^16–18 we postulated that the labile sulfamate leaving group in prodrug 4 might potentiate an elimination pathway via deprotonation of the aromatic urea nitrogen (Figure 3a). Consistent with this hypothesis, it has been reported for an analogous PABA-Glu prodrug of a nitrogen mustard that N-methylation of the aromatic urea nitrogen increases aqueous half-life by ≈3–4-fold, albeit with a concommitant ≈6-fold decrease in the efficiency of activation by CPG2.¹⁶ Thus, we envisioned that this elimination pathway might also be avoided by replacing the aromatic urea nitrogen with an oxygen, which could not undergo deprotonation and might retain efficient enzymatic activation. Along these lines, analogous PHOBA-Glu nitrogen mustard prodrugs have been investigated previously, although differential stability of the linker was not investigated explicitly.^16–18

Figure 3.

Decomposition of PABA-linked prodrugs. (a) Proposed mechanism of AMS-5´-PABA-Glu (4) decomposition. (b) Structures of model substrates PNP-PABA-Glu (11) and PNP-PHOBA-Glu (12). (c) Chemical stability of 11 in the presence or absence of base in CH₃CN, as assessed by HPLC analysis. (d) Chemical stability of 12, assessed as above.

To test this hypothesis, we evaluated the chemical stability of PABA- and PHOBA-linked p-nitrophenyl carbonates 11 and 12 (Figure 3b), synthesized as previously reported.¹⁸ Decomposition was monitored by HPLC in the presence or absence of 1 equiv Et₃N. The PABA-linked probe 11 underwent rapid decomposition in the presence of base (t_1/2 ≈ 2.4 h) (Figure 3c). In contrast, the PHOBA-linked probe 12 was completely stable for at least 12 h (Figure 3d). These results suggested that, somewhat counterintuitively, urethane-based PHOBA-linked prodrugs would be chemically more stable than urea-based PABA-linked congeners.

Thus, we next synthesized the corresponding AMS-5´-PHOBA-Glu prodrug 5 by triphosgene activation of the protected PHOBA-Glu fragment 13, followed by coupling with protected AMS derivative 14 (Figure 4). Deprotection of the coupling product 15 in two steps yielded the desired PHOBA-linked prodrug 5. As hoped, both the protected intermediate 15 and the prodrug 5 were chemically stable during isolation and purification, in stark contrast to the corresponding PABA-linked intermediate 10 and prodrug 4.

Figure 4.

Synthesis of PHOBA-linked prodrug 5. DIPEA = N,N-diisopropylethylamine.

With a chemically stable AMS-5´-PHOBA-Glu prodrug 5 in hand, we evaluated its cleavage by purified, recombinant CPG2. In an Amplex Red assay for glutamate release, the PHOBA-linked prodrug 5 was cleaved successfully with qualitatively higher efficency than the original prodrug AMS-6-Glu (2) (Figure 5a). Formation of AMS was also observed by MS analysis, confirming spontaneous self-immolation of the linker. Enzyme kinetic parameters were then determined using a SPE-TOF-MS (solid-phase extraction/time-of-flight mass spectrometry; Agilent RapidFire 360 with 6520 TOF-MS) assay for substrate consumption, which indicated cleavage of AMS-5´-PHOBA-Glu (5) at >1,600-fold higher efficiency compared to AMS-6-Glu (2), and within 10-fold of a canonical substrate methotrexate (16) (Figure 5b). Consistent with our hypothesis about poor substrate recognition of AMS-6-Glu (2) by CPG2, the K_m for AMS-5´-PHOBA-Glu (5) was >170-fold lower by comparison.

Figure 5.

Cleavage and activation of AMS prodrugs by CPG2. (a) Cleavage of prodrugs AMS-6-Glu (2) and AMS-5´-PHOBA-Glu (5) as determined by glutamate release in an Amplex Red assay. (b) Enzyme kinetics parameters for cleavage of prodrugs 2 and 5, and a known substrate methotrexate (16), as determined using a a SPE-TOF-MS assay. (c) Cytotoxicity of prodrugs 2 and 5 in the presence of recombinant, purified CPG2, and in comparison to prodrug 5 alone or parent drug AMS (1) alone, as assessed by CellTiter-Glo assay with SET-2 cells (24 h).

Next, we tested whether recombinant CPG2 could activate the cytotoxicity of the prodrugs against a SET-2 human megakaryoblastic leukemia cell line, which we have used previously to evaluate AMS-based prodrugs.⁹ AMS-5´-PHOBA-Glu (5) exhibited dose-dependent cytotoxicity, with increased potency (IC₅₀ = 41 nM) compared to AMS-6-Glu (2) (IC₅₀ = 230 nM) under these experimental conditions, consistent with more efficient cleavage of the former by CPG2 (Figure 5c). Further, CPG2-activated AMS-5´-PHOBA-Glu (5) exhibited cytotoxicity comparable to that of the parent compound AMS (1) (IC₅₀ = 83 nM). In the absence of CPG2, the prodrug AMS-5´-PHOBA-Glu (5) exhibited 14-fold weaker cytotoxicity, consistent with selective activation by the enzyme. We attribute the enzyme-independent cytotoxicity of the parent prodrug to formation of small amounts of AMS observed over extended incubation times (see Supporting Informationfor complete details). Control experiments with deallylated 13 (PHOBA-Glu; Pd(PPh₃)₂Cl₂, Et₃SiH, 18% from 13) indicated that this occurred via direct hydrolysis of the distal urethane linkage. Thus, further optimization of the linker for very potent drugs such as AMS will be of interest in the future.

In conclusion, we have developed a novel prodrug AMS-5´-PHOBA-Glu (5) that is cleaved efficiently by CPG2, activating its cytotoxic activity. Initial efforts to advance a more conventional PABA-linked prodrug 4 were thwarted by unexpected chemical instability, which was attributed to a spontaneous elimination pathway. This problem was overcome by replacement of the PABA urea nitrogen with a PHOBA urethane oxygen, which somewhat counterintutively provided increased chemical stability. Although a modest level of enzyme-independent cytotoxicity was observed, the PHOBA linker design dramatically improved enzyme cleavage by CPG2 and vastly increased chemical stability compared to the corresponding PABA linker. We envision that this strategy for increasing the chemical stability of PABA linkers by conversion to PHOBA linkers (NH-to-O) may be applicable to other prodrug systems in the future.

Data Availability Statement

The data underlying this study are available in the published article and its Supporting Information.