Controlled masking and targeted release of redox-cycling ortho-quinones via a C–C bond-cleaving 1,6-elimination

Natural products that contain ortho-quinones show great potential as anticancer agents but have been largely discarded from clinical development because their redox-cycling behaviour results in general systemic toxicity. Here we report conjugation of ortho-quinones to a carrier, which simultaneously masks their underlying redox activity. C-benzylation at a quinone carbonyl forms a redox-inactive benzyl ketol. Upon a specific enzymatic trigger, an acid-promoted, self-immolative C–C bond-cleaving 1,6-elimination mechanism releases the redox-active hydroquinone inside cells. By using a 5-lipoxygenase modulator, β-lapachone, we created cathepsin-B-cleavable quinone prodrugs. We applied the strategy for intracellular release of β-lapachone upon antibody-mediated delivery. Conjugation of protected β-lapachone to Gem-IgG1 antibodies, which contain the variable region of gemtuzumab, results in homogeneous, systemically non-toxic and conditionally stable CD33+-specific antibody–drug conjugates with in vivo efficacy against a xenograft murine model of acute myeloid leukaemia. This protection strategy could allow the use of previously overlooked natural products as anticancer agents, thus extending the range of drugs available for next-generation targeted therapeutics.

T he application of new chemical entities with innovative mechanisms of action is necessary for the development of next-generation therapeutics. Natural products are often a source of anticancer agents or act as an inspiration for rational molecular design 1,2 . However, many natural products identified with medicinal value in vitro remain underexploited due to dose-limiting toxicity in vivo 3 . Toxicity can stem from low cancer-cell specificity, off-target reactivity or, sometimes, built-in metabolic reactivity of functional groups. Such liabilities can be addressed with prodrug strategies that can mask pharmacophores to prevent secondary pharmacology or accelerated metabolism and can widen the therapeutic window 4 . Drug targeting strategies can also be useful. For example, antibody-mediated delivery for cancer-cell discrimination has facilitated the use of auristatins, maytansinoids and calicheamicins 5 . However, some medicinal compounds contain a reactive moiety that cannot be effectively protected by existing chemistry or prodrug strategies. Furthermore, only highly functionalized molecules that contain amine or hydroxyl groups tend to be suitable for conjugation to carriers. Toxic molecules that do not contain such amenable functional groups are often discarded if a carrier cannot be attached without laborious synthetic derivatization. Therefore, improvements in protection of functionalities and attachment of drug carriers to alternative functional groups will extend treatment options with already-discovered compounds 6,7 .
One underused family of cytotoxic natural products is the ortho-quinones. Several low-molecular-weight compounds that contain an ortho-quinone group, such as β-lapachone 8 , tanshinones (I, IIA, IIB and crypto) 9 , mansonones A-G 10 , dunnione 11 , miltirone 12 , salvicine 13 and caryopteron A 14 , display wide antiproliferative effects in vitro. For example, β-lapachone (1, Fig. 1a), a natural product from Brazilian lapacho tree bark 15 exhibits good efficacy against leukaemias 16 and NQO1+ cancers, such as breast 17 , non-small-cell lung 18 and pancreatic 19 . However, untargeted ortho-quinones have dose-limiting toxicity and metabolic liabilities because of their NQO1-dependent redox-cycling behaviour that results in the formation of reactive oxygen species (ROS; Fig. 1a) 20 . ROS disrupt the function of proteins 21 and can lead to irremediable DNA oxidation, PARP1 hyperactivation and cell death 22,23 . Although this mechanism is of use against cancer cells, systemic propagation of ROS is undesirable and may lead to anaemia and methaemoglobinemia, as observed in clinical trials of 1 (refs. 24,25 ). Additional ortho-quinone toxicity can result from their ability to react as electrophiles with critical cellular proteins, peptides, nucleic acids or glutathione, which interferes with redox homeostasis 26 .

Controlled masking and targeted release of redox-cycling ortho-quinones via a C-C bond-cleaving 1,6-elimination
We hypothesized that if delivered to target cells without systemic side effects, ortho-quinones would have substantial potential as a monotherapy or in combination with mainstay anticancer agents 27,28 . Their metabolic interference may be a potential avenue to tackle malignancies 29,30 , and a strategy to overcome cancer drug resistance 31 . Some tumour selectivity of 1 has been reported, as overexpression of oxidoreductase enzymes (for example, NQO1) can increase redox-cycling rates selectively in the tumour 32 . However, no therapeutic ortho-quinone has yet reached the clinic due to generalized ROS-mediated systemic toxicity. In contrast, para-quinones, such as doxorubicin (Adriamycin), geldanamycin, mitoxantrone and mitomycin C, have lower redox-cycling rates and are successfully used in the clinic 22 . If the side effects of ortho-quinones could be prevented through selective protection and targeting strategies, this may lead to new cancer treatment options.
Prodrug approaches to mask redox activity and limit side effects of ortho-quinones, such as 1, have been described. Boothman and co-workers developed a number of approaches, including formation of esterase-cleavable hydroquinone alkyl esters 33 , and pH-sensitive aryl imine, acyl hydrazone, ketal 34 , aminoalkyl alcohol and amino aromatic phenol prodrugs 35 . However, the disadvantages of these strategies have prevented their widespread adoption. For example, esters and hydrazones are often too labile under physiological conditions for effective targeting, whereas ketals are not sufficiently labile in tumours 34 . Modern prodrug strategies and linkers, including those that connect targeting antibodies and their payloads, use enzymatically activatable trigger groups and release drug functionality by means of self-immolative spacers. For example, para-aminobenzyl carbamate linkers release amines and para-aminobenzyl ether linkers release alcohols upon specific protease-triggered hydrolysis 36 . We hypothesize that a similar strategy is possible for ortho-quinones.
Here we present a modular strategy that uses self-immolative benzyl linkers for protection and controlled release of ortho-quinones. The self-immolative 1,6-elimination of a para-aminobenzyl linker attached at the quinone carbon as a benzyl ketol enables release of the hydroquinone. The released hydroquinone then oxidizes spontaneously to give the desired quinone by using the redox-cycling ability of the payload (Fig. 1b). The strategy is compatible with peptide linkers, such as those activated by the cysteine protease cathepsin B 37 . In this work, we explore the mechanism and properties of this C-C bond-cleaving elimination and prospectively apply it to protect and target 1 as an antibody-drug conjugate (ADC) for treatment of acute myeloid leukaemia (AML).

Synthesis of self-immolative quinone models
Initially, the synthesis of model derivatives of quinones that fragment upon removal of a protecting group was investigated using 9,10-phenanthrenequinone (PhQ), 4, as a model compound (Fig. 2a). Phenolic alcohol-containing drugs are typically protected as para-aminobenzyl ethers 38 . However, when we investigated reductive alkylations of 4 with Boc-protected para-aminobenzyl bromide linker 5 (Fig. 2a), with a view to forming O-benzyl derivatives, we instead obtained the C-benzyl derivative Boc-para-aminobenzyl phenanthrene-ketol (Boc-PAB-PhQ), 6, in good yield (60-70%). Indeed, O-benzyl derivatives of hydroquinones have been reported to be unstable and to rearrange to their corresponding C-benzyl isomers 39,40 . Intrigued by 6, we tested its stability following N-Boc deprotection to give para-aminobenzyl phenanthrene-ketol (PAB-PhQ), 7, which to our surprise appeared to be unstable, and as it was consumed, formation of quinone 4 was observed ( Fig. 2b and Supplementary Figs. [1][2][3][4][5][6]. Complete consumption of 7 and formation of 4 was confirmed by 1 H NMR spectroscopy. Appearance of 4 proceeded with a half-life (t 1/2 ) of 2.94 ± 0.08 h at 25 °C in a moderately acidic methanolic solution with complete release within 20 h. Data fitting suggested that release followed first-order kinetics and the rate of consumption of 7 (k obs = 6.18 ± 0.12 × 10 -5 s -1 ) was consistent with the rate of formation of product 4 (k obs = 6.55 ± 0.22 × 10 -5 s -1 ; Fig. 2c and Supplementary Figs. 4-6). Through 1 H NMR spectroscopy, we identified the presence of an aminobenzyl side product from the spontaneous reaction of methanolic solvent with an aza-quinone methide generated by self-immolative elimination ( Supplementary Figs. 2 and 18). This, and the necessity for aniline deprotection for quinone release, suggests a self-immolative elimination mechanism. Under basic conditions, appearance of 4 was slower (k obs = 1.16 ± 0.18 × 10 -6 s -1 ) and not complete after 40 h (Supplementary Figs. 4-6), which suggested a pH dependence for the elimination reaction.
We anticipated that self-immolative release might generate an unstable hydroquinone intermediate (8) that could auto-oxidize to the corresponding quinone. 9,10-Phenanthrenediol hydroquinone 8 was not observed when elimination was performed under standard, oxygenated conditions as a result of its instability to auto-oxidation ( Supplementary Fig. 2

pH-dependent elimination profiles of benzyl ketol derivatives
The mechanism and kinetics of the elimination of model compound PAB-PhQ 7 were investigated. Initially, deprotected (R = H) derivative 7 and a theoretical protected (R = Ac) derivative 20 were compared (Fig. 3a). The energy profile along the breaking C-C bond (green) was calculated quantum mechanically at the PCM(H 2 O)/M06-2X/6-31+G(d,p) level of theory with the purpose of locating the transition state (TS) structures for the elimination. However, no maximum was detected, indicating that neutral species 7 remained unreactive and did not eliminate, contrasting with the observed release of 4. Comparisons were made with different heteroatom-based leaving groups for the elimination of similar neutral para-amino benzyl species ( Supplementary  Fig. 106). This showed that reagents with chloride or activated alcohol (mesyl or triflate) leaving groups had low enough activation barriers (7.8-16.9 kcal mol -1 ) to eliminate. A carbamate leaving group was calculated to have a higher activation barrier (28.1 kcal mol -1 ). However, carbon-based leaving groups needed three electron-withdrawing groups (CN or NO 2 ) or protonated carbonyl groups to have acceptable activation barriers (15.3-28.1 kcal mol -1 ) for elimination.
These results suggested that charged species might be necessary to promote the C-C bond elimination reaction. In aqueous solution, several charged species are in equilibrium albeit in various proportions. We considered nine possible charged states for 7 (Fig. 3b). Three of the nine species (7 NH-, 7 OH+ and 7 Zw1 ) were reactive enough to allow an effective flow of electronic density from the para-amino group to the ketone to promote elimination of the hydroquinone, by either making the former more nucleophilic (NH -) or the latter more electrophilic (C = OH + ). The activation barriers calculated for these species (ΔG ‡ NH-= 11.8 kcal mol -1 ; ΔG ‡ OH+ = 6.5 kcal mol -1 ; ΔG ‡ Zw1 ≤ 0.1 kcal mol -1 ) suggest fast to extremely fast intrinsic elimination rates ( Fig. 3c; k -= 3.13 × 10 4 s -1 ; k + = 1.70 × 10 8 s -1 ; k zw = 5.49 × 10 12 s -1 ). However, the global rate depends on the concentration of these reactive species in solution, which are exceedingly low due to the high pK a values for aniline deprotonation (≥16) and low pK a for carbonyl protonation (≤−2), values well beyond the practical range in aqueous solution. Hence, the theoretical global kinetic constant (k theo ) can be modelled as a function of the intrinsic kinetic constants for each reactive species (k r ), the equilibrium constants for all species in solution both reactive and non-reactive (K i ), and pH; this expression for the theoretical rate constant can be related to the experimentally observed rate constant (k obs ) ( Fig. 3b (equation) and Supplementary Information).
It became apparent that despite many attempted approximations to cancel out some of the terms in the equation, k theo calculated at different pH values was exceedingly dependent on the equilibrium constants involving the extremely low populated, but very reactive species (K NH-, K OH+ and K Zw1 ), which resulted in erratic pH dependency profiles depending on the numbers used. With no experimental values available for some equilibrium constants, and because tabulated 41,42 and calculated values by using different programs such as Marvin 19.19.0 (2019), ChemAxon (https://chemaxon.org) or Epik (Schrodinger Suite) 43,44 were not reliable ( Supplementary  Fig. 109), we decided to examine the elimination reaction kinetics for 7 at different pH values to derive more realistic equilibrium constants by using experimental data.
For N-protected derivative 20, Ac-PAB-PhQ, we also considered nine different species. We found three of them to be reactive, but with higher elimination activation barriers (ΔG ‡ NAc-= 24.8 kcal mol −1 , ΔG ‡ OH+ = 10.7 kcal mol −1 , ΔG ‡ Zw1 = 4.5 kcal mol −1 ; Fig. 3c, Supplementary Fig. 111 and Supplementary Table 5) relative to those calculated for the unprotected (R = H) derivative 7. The activation barrier for the negatively charged acetanilide (20 NAc-) was substantially increased with respect to the deacetylated analogue due to the large decrease in nucleophilicity of the nitrogen lone pair delocalized along the acetyl group. The dramatic obstruction of this reaction channel causes the global elimination rate to slow down by 2-5 orders of magnitude at pH 3-10 ( Supplementary Fig. 110). Therefore, as for the widely used para-aminobenzylalcohol linkers 45 , elimination is self-immolative and faster with free aniline.
For comparison, we also modelled the elimination of the aminobenzyl ketol species derived from the medicinal natural product of most notable interest, β-lapachone, PAB-BL 10 and dunnione, PAB-DN 18 (Supplementary Fig. 111 and Supplementary Table 5). Similarly to PAB-PhQ 7, the neutral species of PAB-BL 10 and PAB-DN 18 were unreactive to elimination. We considered the possible reactive species in a similar manner. An additional species-arising from protonation of the cyclic vinyl ether group-was considered but was calculated to be unreactive. In comparison to unprotected phenanthrenequinone derivative 7, slightly higher activation barriers were obtained for the three charged reactive species of PAB-BL 10 (ΔG ‡ NH-= 13.5 kcal mol −1 , ΔG ‡ OH+ = 13.5 kcal mol −1 , ΔG ‡ Zw1 = 1.4 kcal mol −1 ) and PAB-DN 18 (ΔG ‡ NH-= 13.0 kcal mol −1 , ΔG ‡ OH+ = 14.1 kcal mol −1 , ΔG ‡ Zw1 = 1.6 kcal mol −1 ) and were noticeably higher for cationic 10 OH+ and 18 OH+ . This increase is due to the reduced aromaticity around the carbonyl groups in 10 and 18, which in turn affects their protonation ability (K OH+ ). Given the calculated differences in the intrinsic kinetic accessibility to the different reaction channels, and the expected thermodynamic differences in the relative populations of each reactive charged species, it was likely that both PAB-BL 10 and PAB-DN 18 had a slightly different pH-rate dependence relative to 7.
To experimentally validate predictions and determine the pH-rate dependence of the elimination and the critical reaction parameters in aqueous solution, we synthesized models protected with a penicillin-G-amidase-cleavable phenylacetamide group 46 . Addition of penicillin G amidase to the protected models triggered amide hydrolysis and formation of the desired para-aminobenzyl ketol in neutral aqueous solution (Fig. 4a).
With this strategy, we showed that at physiological pH and temperature (pH 7.4, 37 °C), formation of PhQ 4 occurred at rate k obs = 1.15 ± 0.66 × 10 -5 s -1 following addition of the enzyme to 21. Intermediate 7, PAB-PhQ, could be identified ( Fig. 4b and Supplementary Figs. [24][25][26] and elimination of 7 occurred at rate k obs = 9.72 ± 2.88 × 10 -6 s -1 . Without added enzyme, 21 was stable and formation of 4 negligible. As mentioned above, hydroquinone 8 was not observed by NMR spectroscopic analysis of non-degassed solutions, suggesting that in oxygenated aqueous solution the intermediate has a very short t 1/2 relative to that of species 7. As enzyme-mediated amide hydrolysis was fast (<5 min), immediate kinetic analysis of the subsequent elimination rate was possible without use of acidic or basic deprotection reagents, thus allowing control of pH. Focusing on the phenanthrenequinone, β-lapachone  Equation  and 20 along the breaking C-C bond (green). b, General chemical scheme of nine different species denoted as Q (where subscript indicates neutral, cationic, anionic and zwitterionic) that are in equilibrium for the protected and unprotected para-aminobenzyl ketol derivatives in an aqueous solution. Species inside a dashed box (Q NH-, Q OH+ and Q Zw1 ) are those for which an elimination TS structure was found. The equation describes the effect of pH on the observed kinetic rate (k obs ) and its derivation is described in the Supplementary Information. c, TS structures for the elimination of reactive species calculated at the same level of theory. Green dashed lines show interatomic distances (in angstroms) for the breaking C-C bond. Colour blue denotes nitrogen, orange denotes oxygen, grey denotes carbon and white denotes hydrogen. Intrinsic first-order kinetic constants (k -, k + and k zw ) were calculated from the theoretical activation energies (ΔG ‡ in kcal mol -1 ) at 310 K using the Eyring equation. and dunnione derivatives, elimination from 7, 10 and 18 generated enzymatically from 21, 22 and the (R,R/S,S) pair of enantiomers of 23, respectively, was followed across a range of pH values at 37 °C ( Fig. 4c and Supplementary Figs. 27 and 38). Reactivity of the quinone products under basic conditions precluded analysis of product formation at pH >10. As predicted computationally, the reaction rate was dependent on pH. Compound 7 tends to show acid-promoted elimination, which peaks at pH 5-6 (k obs pH 6 = 1.7 ± 0.3 × 10 -5 s -1 ).   data presented as calculated k obs value from representative experiment ± 95% confidence interval. Lines represent fitting of the experimental data to theoretical models for k obs . ii: From this fitting, values for intrinsic kinetic (k i ) and equilibrium (K i ) constants were derived. a Intrinsic rate constants (k i ) calculated from the theoretical activation barriers were fixed during the fitting.
The k obs derived at different pHs were fitted into the equation shown in Fig. 3b; the calculated k i were used as fixed parameters and the individual K i values were left as adjustable parameters. A good fit of k obs to the theoretical kinetic model was obtained (7, χ 2 = 0.270, R 2 = 0.998; 10, χ 2 = 4.577, R 2 = 0.998; 18, χ 2 = 8.726, R 2 = 0.994), despite the different pH-dependent profile obtained for quinone derivatives 7, 10 and 18, which reflects the high quality of our kinetic model. The empirically derived equilibrium constants were all within expected ranges, except for the unusually facile deprotonation of the tertiary alcohol (pK O-≈ 8-12) (Fig. 4c(ii)).
With the calculated rate values, we became interested in the potential applications of the acid-dependent release. Such a profile is unusual because drug release from related para-hydroxybenzyl ether linkers is promoted in basic conditions 36 .

C-benzylation prevents quinone redox activity
The toxicity of the benzyl ketol pharmacophore generated upon quinone derivatization was investigated to determine the usefulness of the protection strategy for prodrug generation. C-benzylation disrupts the quinone scaffold, so it was expected that the redox-cycling ability of an ortho-quinone and associated undesirable systemic toxicity would be lost upon derivatization. We focused our attention on β-lapachone, the most well-studied medicinal ortho-quinone. Redox activity of a stable, non-releasable model of benzyl β-lapa-ketol, 24, which lacks a para-amino-group necessary for self-immolative release, was compared to parent drug 1, and control non-redox cycling protected 1,4-dioxine derivative 25. Compound 24 was found to be non-redox active in an in vitro redox cycling assay ( Fig. 5a and Supplementary Fig. 39) and, unlike 1, did not generate detectable ROS in AML cell line HL-60, a cell line in which H 2 O 2 generation by 1 has been previously reported (Fig. 5b) 47 . Compound 24 displayed decreased toxicity by >20-fold to a range of cancer cell lines, including leukaemia, breast, colon and cervical cancer (Extended Data Fig. 1b and Supplementary  Fig. 42). The IC 50 of 24 against HL-60 was unchanged by addition of antioxidant N-acetyl cysteine (NAC; Fig. 5c). This is in stark contrast to 1, for which the IC 50 value more than doubled upon NAC addition as NAC alleviated ROS-mediated toxicity. These results together confirm that 24 does not redox cycle.
In previous work, we showed that 1 is, cumulatively, an reversible allosteric inhibitor of the 5-lipoxygenase enzyme (5-LO also known as 5-LOX or ALOX5), an enzyme that catalyses the oxidation of arachidonic acid, and that its anticancer activity was dependent on 5-LO expression 48 . Activity of compounds 24 and 25 against 5-LO was tested. β-Lapachone 1 inhibited 5-LO activity in the presence of the reducing agent dithiothreitol (DTT) that promotes redox cycling and formation of hydroquinone 2. Unlike 1, the inhibition profiles of 24 and 25 were not impacted by addition of DTT (Fig. 5d). Therefore, benzyl ketol β-lapachone derivatives are not strong inhibitors of 5-LO, unlike 1.
Methaemoglobin formation, a major side effect of ortho-quinones in vivo, did not occur for the protected compounds 24 and 25 in an in vitro ovine blood model. In contrast, β-lapachone 1 caused dose-dependent methaemoglobin formation detectable at dose concentrations as low as 10 µM after 1 h of incubation (Fig. 5e, Extended Data Fig. 1a and Supplementary Fig. 40). Haemolysis of haemoglobin, related to anaemic side effects, was also lower for the protected models ( Supplementary Fig. 41). These promising results indicated that the benzyl protection unit, while intact, has the capacity to mitigate the blood-borne redox-mediated side effects of quinones.

Cathepsin B can trigger the release of β-lapachone from dipeptide prodrugs
Confident that the protection strategy produces redox-inactive derivatives, it was next investigated whether the self-immolative linker strategy was compatible with enzymatic cleavage of cathepsin-B-labile dipeptides. Dipeptide units 49 Cbz-Val-Cit-, Cbz-Val-Ala-and Cbz-Phe-Cit-were added by sequential amide coupling from deprotected aniline intermediate 10 (26-28, Fig. 5f). After incubation of 26-28 with cathepsin B (Fig. 5g and Supplementary Figs. 53-55), compound 1 was observed. Aniline 10 was identified as a common intermediate and release of 1 from 10 occurred as previously observed. The Cbz group was not necessary for cathepsin B action ( Supplementary Fig. 56). This demonstrated that the dipeptide-para-aminobenzyl ketols are compatible with cathepsin B, in an identical manner to dipeptide-para-aminobenzyl alcohol and carbamate linkers. Like benzyl model 24, dipeptide prodrugs 26-28 generated less methaemoglobin in blood in an in vitro model relative to 1 (Fig. 5h). Additionally, they generated less methaemoglobin than an acyl-hydroquinone β-lapachone prodrug 29, previously described by Ma et al. 33 due to the increased stability of the dipeptide over labile ester bonds.
Cbz dipeptide prodrugs 26-28 were fully stable in human serum (Supplementary Figs. 58 and 59) and did not exhibit toxicity to AML cell line MOLM-13 at 5 µM concentration, unlike 1 and derivative 29 (Extended Data Fig. 1c and Supplementary Fig. 57) that achieved complete cell death at this concentration. Compound 26 exhibited 10-fold higher IC 50 relative to 1 in AML cell lines HL-60 and MOLM-13 ( Supplementary Fig. 57). After lysosomal deacidification (NH 4 Cl) 50 or inhibition (E64d) 51 , 26 was less toxic, suggesting that acidic compartments help promote toxicity of the prodrug (Extended Data Fig. 1d). It was interesting that with NH 4 Cl at 30 mM, an increase in toxicity of 1 was seen, a result previously described for menadione 52 (Supplementary Fig. 60). Prodrugs 26-28 may have suitability for treatment of NQO1+ solid tumours with concurrent overexpression of proteases, for example, cathepsins 53,54 and acidic pH 55 in the extracellular malignant environment.

β-Lapachone is a promising treatment for AML
In our work, we were particularly interested in testing quinone 1 for the treatment of AML, a cancer for which 1 displays potent toxicity, as described in ref. 48 and confirmed by us on a panel of AML cell lines ( Supplementary Fig. 43). AML is the most common form of acute leukaemia among adults 56 . It is characterized by immature myeloid cell proliferation and bone marrow failure and is a cancer for which new treatments are urgently needed. AML has a poor 5 yr survival rate of ∼20% and a large proportion of patients relapse 57 . Interference in redox homeostasis is an appealing treatment angle for AML 58 . Recent observations note that despite differences between the mechanisms of action of clinically used AML therapeutics, most share oxidative stress as a mediator of the cytotoxic effect 59 . For example, anthracyclines and arsenic trioxide induce rapid ROS accumulation 60,61 .
Quinone 1 also has relevant protein targets in AML. As mentioned above, in previous work 48 , we show that β-lapachone strongly inhibits enzyme 5-LO, an enzyme that is a candidate target for therapeutic targeting of the stem-cell-like blasts in AML 62,63 . In our experiments, 5-LO gene knock-out (KO), using CRISPR-Cas9 gene editing, impaired proliferation of leukaemia cell line HEL . Experiments also suggested that KO or shRNA-mediated knockdown (K D ) may make AML cells more resistant to 1 . Both redox sensitivity and 5-LO target relevance made β-lapachone a valuable experimental drug for AML treatment. For AML, the data strongly supports the development of an intracellular targeting strategy to enable application of the quinone prodrugs.

ADCs release β-lapachone in CD33 + AML cells in vivo
Antibody conjugation can enable targeting and assisted delivery of payloads inside AML target cells. We have shown that an ortho-quinone protection strategy can eliminate redox toxicity from 1, and indeed, protection should therefore prevent any potential redox damage to the antibody carrier.
We reasoned that cell killing should depend on effective cellular trafficking of a conjugate to a low-pH cellular compartment inside target cells because pH 4-5 is required for the efficient release of 1 in vitro. Furthermore, in the system described, the slow release of 1 from the quinone protection unit PAB-BL 10 at physiological pH should limit systemic drug release, that is, any linker deconjugated from the antibody while in circulation should not release drug, even if a protecting dipeptide unit is broken. This can limit the toxicity profile of the ortho-quinone to that produced by non-specific cell uptake.
To test this strategy, we designed IgG1 antibodies for payload conjugation with a binding region based on the well-characterized CD33 targeting antibody gemtuzumab 64  site-specific cysteine conjugation (Fig. 6a). By using variants with engineered cysteine mutations, we generated homogeneous conjugates with a drug-antibody ratio (DAR) of 2. Three single-cysteine mutation sites were compared, one in the light chain and two in the heavy chain (Fig. 6b). Complete conversion was achieved for all conjugates. The positioning of the engineered cysteines on the antibody influenced the ease of conjugation. A Gem-HC-239iC mutant with a cysteine introduced in the hinge region of the heavy chain at position 239 offered favourable reactivity over Gem-LC-V205C and Gem-HC-S442C mutants with cysteine residues in more exposed positions on the light chain (Val to Cys mutant at position 205) and heavy chain (Ser to Cys mutant at position 442), respectively. With Gem-HC-239iC, complete conversion was achieved using 20 equivalents of linker per cysteine for 6 h to generate conjugate Gem-HC-239iC-BL. Gem-HC-V205C and Gem-HC-S442C mutants required more equivalents of linker (40 and 30 equivalents per cysteine, respectively) to achieve complete conversion. These conjugates may also be termed antibody-prodrug conjugates (APDCs) 68

DAR 2 ADC inhibits AML tumour growth in mice
We evaluated the antileukaemic effect of DAR 2 ADC Gem-HC-239iC-BL in a subcutaneous mouse model of human AML. NOD-SCID immunocompromised mice were inoculated with HEL cells to induce tumour growth and were treated intravenously with either ADC Gem-HC-239iC-BL, the native antibody Gem-IgG1 or PBS. In vitro, against HEL, Gem-HC-239iC-BL had similar toxicity to 1 while non-targeted NIP228-HC-239iC-BL and Gem-IgG1 showed no apparent toxicity (Fig. 6d). In vitro toxicity results were promising although variable. Results varied depending on the cell-counting method used and potentially because of conjugate aggregation arising from the hydrophobic 69 linker and drug.
In the in vivo model, a strong antitumour activity was observed with the administration of only two doses of ADC at 7.5 mg kg −1 . The ADC-treated mice had decreased tumour volumes relative to the controls (Fig. 6e-g). Moreover, the antileukaemic activity of the ADC significantly improved the survival of the ADC-treated animals (mean of 52 days versus 24 days for control; Fig. 6h). Importantly, two out of the five ADC-treated animals were alive 76 d post-treatment, and no signs of toxicity or metastasis were seen in the liver, spleen, kidney, brain, bone marrow, heart and lung of control mice and mice treated with the ADC as demonstrated by histopathological analysis (Supplementary Figs. 100 and  101 and Supplementary Table 1). To verify drug release, we tested samples of fixed tumour tissue of mice treated with ADC Gem-HC-239iC-BL by LC tandem MS, and in ADC-treated mice we were able to detect and quantify 1, which was present at 0.327 ng g −1 (Fig. 6i, Supplementary Figs. Tables 2 and 3). These data provide evidence for the successful in vivo release of 1 through the acid-dependent C-C bond-cleaving self-immolative 1,6-elimination mechanism reported in this work. Together, these results demonstrate that the ADC antitumour activity in vivo has great potential to be further developed as an antileukaemic treatment.

102-105 and Supplementary
Alternative payload classes are an important strategy to identify new treatments and combat drug resistance in the fight against cancer. Here we leverage prodrug chemistry and antibody-mediated delivery to develop a method for intracellular targeting and release of ortho-quinones from a protected redox-inactive form. This work begins to solve a fundamental problem of application of ortho-quinone therapeutics: the reactivity of ortho-quinones with cellular metabolism. Through protection and targeting we aim to increase the concentration of redox-active quinone in the target tissue and decrease the concentration in healthy tissues, thereby lowering associated dose-limiting toxicities.
Unlike previous approaches, our ortho-quinone protection strategy is truly modular and could be adapted to any peptide pro-moiety. The self-immolative 1,6-elimination mechanism described is unprecedented because it cleaves a C-C bond between the benzyl linker and quinone carbon. The elimination is triggered within an acidic pH range, depending on the protected quinone. We demonstrate the generality of the methodology, which we applied to protect model 9,10-phenanthrenquinone and four additional ortho-quinones. β-Lapachone, (±)-dunnione and cryptotanshinone are all ortho-quinones of medicinal interest. 3-Hydroxy-β-lapachone adds additional possibilities for functionalization compatible with the protection strategy.
As an example of the application of the ortho-quinone protection strategy, we synthesized cathepsin-B-cleavable prodrugs of β-lapachone, followed by ADCs using the protection chemistry. For β-lapachone, 1, minimal release of the payload from the quinone-protection unit occurs at physiological pH even after peptide deprotection, which for an ADC means the chemistry offers a built-in protection against premature in-circulation drug release upon linker deconjugation. The rapid cell death triggered by the ADC relative to the untargeted prodrug suggests that released protected quinone species 10 reaches a low pH compartment (∼4) necessary for its fast elimination. In many cells the FcRn receptor will recycle the antibody to the cell surface directly from the endosome (pH ∼5.5-6) and prevent access to more acidic compartments 70 , but our results suggest that our ADC does not undergo notable endosomal recycling 71,72 . The activity of β-lapachone is believed to be independent of endosomal/lysosomal escape, (that is, it should give a toxic effect as soon as it is released from 10).
This work also highlights the relevance of β-lapachone for the treatment of AML and verifies the relevance of 5-LO as a respective target. ADC therapy is well established in the AML disease area, with reapproval of CD33-targeting Mylotarg (gemtuzumab ozogamycin) for relapsed or refractory AML 73 and with many more ADCs in development 74 . Our protected β-lapachone payload is efficacious in a xenograft AML model following ADC targeting and can effectively expand future treatment options upon further development and optimization. It should also be noted that β-lapachone is a lower-potency payload than those commonly used in ADCs; for example, most marketed ADCs have subnanomolar-potency payloads 5 . However, the concept that ADCs must contain a subnanomolar toxic payload has been recently challenged and efficacious cancer-cell killing has been obtained with lower-potency payloads 75 . Side effects, and hence the high attrition rate of ADCs in clinical studies, often result from payload-mediated toxicity; thus a lower-toxicity drug that is masked while in circulation with built-in deconjugation toxicity protection may be advantageous 76 . In this respect, β-lapachone or other ortho-quinone derivatives offer opportunities for the development of ADCs with a distinct mode of action. The protection strategy described here is expected to be applicable to other ortho-quinones and may encourage the development of improved synthetic routes to, or new derivatives of, these high-value natural products. Derivatives of β-lapachone with higher toxicity that retain the quinone pharmacophore have been reported 77 . New prodrugs, drug-targeting strategies and combinations of the two will enable the application of compounds previously deemed inappropriate for medicinal use. Here, we have developed enabling chemistry that facilitates the integration of bioactive ortho-quinones into these strategies.

Online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/ s41557-022-00964-7. To form the active eliminating species, Boc-PAB-PhQ 6 was dissolved in dichloromethane/trifluoroacetic acid 4:1 at 0 °C and stirred for 30 min, upon which the reaction appeared complete by TLC. Product 7 was dried in vacuo before being dissolved in MeOD. The basicity of the solution was altered by addition of triethylamine (drops). pH was estimated using pH paper. pH values are only indicative because analysis is performed in MeOD. Elimination rates were determined from the aromatic region. Kinetics were measured by monitoring the integral of one peak, normalized to the number of protons present (that is, the integral of 1H), divided by the total integral of the aromatic region (δ = 6.2-8.6 ppm, that is, the integral of 12 H). Analysis was performed using Mestrenova v.14.2.0 processing software and rate constants were calculated using GraphPad Prism v.8.0 software.
Penicillin-G-amidase-mediated kinetic analysis of elimination at physiological pH. Stocks of 2-phenylacetyl-para-aminobenzyl ketol were made up in DMF at a concentration of 10 mM. For a reaction, 17.5 µl of ketol stock was added to 17.5 µl of a 10 mM stock of internal standard in DMF and 310 µl of PBS at pH 7.4. To initiate a reaction, 5 µl of a suspension of penicillin G amidase (PenG, Sigma-Aldrich enzyme 76427, 5-10 U mg −1 ) was added. In control samples 5 µl of PBS was added in place of the enzyme. Test and control samples were then incubated at 37 °C with shaking. At the recorded time points, 20 µl aliquots of the reactions were analysed by HPLC (254 nm, ThermoFisher U3000). HPLC column: Phenomenex Kinetex C18, 5 µm, 50 In vitro redox-cycling ability analysis. The in vitro redox-cycling capability of compounds was assessed by phenol red/horseradish peroxidase (HRP) assay by adaption of a previously described protocol 78 . Briefly, stocks were prepared as follows: (1) DTT to a concentration of 2.4 µM in PBS pH 7.4 buffer; (2) compound stocks in PBS pH 7.4 buffer to 3× the desired concentration by 1,000× dilution of DMSO stocks; (3) a stock of phenol red/HRP detection reagent containing 300 µg ml −1 phenol red and 180 µg ml −1 HRP enzyme in PBS pH 7.4 buffer; (4) a stock of 1 M NaOH in water; (5) a stock of 100 µM H 2 O 2 in PBS prepared by addition of a 30% solution of H 2 O 2 in water into 1,000 µl of PBS. Into each well of a 96-well clear flat-bottomed plate was then added 40 µl of compound stock in PBS followed by 40 µl of DTT stock, followed immediately by 40 µl of phenol red/ HRP reagent. A positive control with 100 µM H 2 O 2 added in place of compound was included. The plate was incubated in the dark at room temperature for 15 min. Following this, 20 µl of 1 M NaOH was added to each well to terminate the reaction. Absorbance was immediately measured at 610 nm (MiniMax i3x Imager, Molecular Devices). Absorbance of treated wells was normalized to control wells treated with PBS only (zero redox cycling).

Cellular ROS detection.
To determine ROS generation by compounds in leukaemia cells, HL-60 cells were washed with PBS, then incubated in serum-free media (10 ml of 1.6 × 10 6 ml −1 cell concentration) with 2,7-dichlorofluorescin diacetate at 20 µM for 30 min at 37 °C. Following this, the cells were washed with PBS and resuspended in Fluorbrite-DMEM media (ThermoFisher) (10 ml). Then, 450 µl of the cell mixture was seeded into wells of a clear flat-bottomed 12-well plate, to which 50 µl of a 10× stock of the compounds in media had been already added, to achieve a final compound concentration of 50 µM. The 10× media stocks were made by 100-fold dilution of DMSO compound stocks, to ensure a final in-plate DMSO concentration of 0.1%. DMSO 0.1% only was used as a negative control. H 2 O 2 was used as a positive control, for which 10 µl of a 30% aqueous solution was dissolved into 1 ml DMSO, then diluted 100-fold into media for plating, to give a final H 2 O 2 concentration of 1 µM. After plating the cells were immediately incubated at 37 °C in the dark for 15 min before the fluorescence of the wells (excitation, 485 nm; emission, 535 nm) was recorded (MiniMax i3x Imager, Molecular Devices).
Methaemoglobin analysis. Defibrinated Oxoid sheep's blood (ThermoScientific) was diluted to a 5% v/v suspension in PBS pH 7.4. In a 96-well microtitre plate, 150 µl of blood suspension was added to 50 µl of serially diluted compound in PBS containing a 4× stock of each compound of interest. Serial dilution of compounds from DMSO stocks into PBS was performed to achieve an in-plate DMSO concentration of 0.1%. PBS buffer alone was used as a negative control. The plate was incubated for the specified time at 37 °C. The plate was then centrifuged at 3,500 r.p.m. for 5 min. The supernatant was removed and cells lysed by resuspension in 100 µl of a solution of Triton-X-100 (1%) in PBS. Following this, 80 µl of lysed cell contents were transferred to a second microtitre plate for ultraviolet measurement. Ultraviolet absorbance was measured immediately at 630 nm (MiniMax i3x Imager, Molecular Devices). Methaemoglobin increase was determined by ( where A 630 is the absorbance of the test well at 630 nm, and A 0 630 is the absorbance of the negative control at 630 nm.

Haemolysis analysis.
To analyse haemolysis, defibrinated Oxoid sheep's blood (ThermoScientific) was diluted to a 5% v/v suspension in PBS pH 7.4. In a 96-well microtitre plate, 190 µl of blood suspension was added to 10 µl of serially diluted compound in PBS containing a 20× stock of each compound of interest. Serial dilution of compounds from DMSO stocks into PBS were performed to achieve an in-plate DMSO concentration of 0.2%. PBS buffer with 0.2% DMSO was used as a negative control. Triton-X (1% v/v) was used as a positive haemolysis control. The plate was incubated for 6 h at 37 °C. Following incubation, the plate was centrifuged at 3,500 r.p.m. for 5 min at room temperature. Cell pellets were lysed by resuspension in 100 µl of a solution of Triton X-100 (1%) in PBS. Then, 80 µl of lysed cell contents were transferred to a second microtitre plate for ultraviolet measurement. Ultraviolet absorbance was measured immediately at 540 nm. The percentage of haemolysis was determined by ( where A 540 is the absorbance of the test well at 540 nm, A 0 540 is the absorbance of the negative control at 540 nm and A total 540 is the absorbance of the positive haemolysis control (100% haemolysis at 540 nm).
Expression, purification and cell-free activity assay for recombinant human 5-LO. Escherichia coli (BL21) was transformed with pT3-5-LO plasmid, and recombinant 5-LO protein was expressed at 30 °C as described 79 . Cells were lysed in 50 mM triethanolamine/HCl pH 8.0, 5 mM EDTA, 1 mM phenylmethanesulfonyl fluoride, soybean trypsin inhibitor (60 μg ml −1 ) and lysozyme (1 mg ml −1 ), homogenized by sonication (3 × 15 s) and centrifuged at 40,000g for 20 min at 4 °C. The 40,000g supernatant (S40) was applied to an ATP-agarose column to partially purify 5-LO as described. Aliquots of semipurified 5-LO (0.5 μg) were diluted with 1 ml ice-cold PBS containing 1 mM EDTA. Samples were preincubated with the test compound or vehicle (0.1% DMSO) with or without 1 mM DTT, as indicated. After 10 min at 4 °C, samples were prewarmed for 30 s at 37 °C, and 2 mM CaCl 2 plus 20 μM arachidonic acid was added to start the formation of 5-LO products. After 10 min, the reaction was stopped by addition of 1 volume of ice-cold methanol, and the formed 5-LO products were analysed by reverse-phase HPLC as described 80 . 5-LO products include the all-trans isomers of LTB4 (trLTB4 isomers) and 5(S)-hydroperoxy-6-trans-8,11,14-cis-eicosatetraenoic acid and its corresponding alcohol 5(S)-hydroxy-6-trans-8,11,14-cis-eicosatetraenoic acid. Antibody expression and purification. The variable heavy-chain and light-chain sequences for gemtuzumab were identified as previously described [81][82][83] . Strings of the variable domains were produced from Geneart, Invitrogen and cloned into mammalian expression vectors encoding either standard human IgG1 constant regions or site-specific cysteine variants such as Gem-LC-V205C, Gem-HC-239iC and Gem-HC-S442C using NEBuilder assembly from NewEnglandBiolabs. Antibodies were transiently expressed in Chinese hamster ovary cells under serum-free conditions as previously described 84 . Cleared culture supernatant was loaded directly onto MapSelect SuRe column equilibrated with PBS (pH 7.2). Antibodies were step eluted with 0.1 M glycine pH 2.7. Pooled fractions were buffer exchanged into PBS.
Antibody-linker conjugation reactions. Antibodies in PBS pH 7.2 + 1 mM EDTA buffer were refolded prior to reaction. The refolding procedure was performed as follows. First, antibodies were reduced with tris(2-carboxyethyl)phosphine hydrochloride (20 equiv., to a final concentration of 10% DMF) for 30 min at 37 °C. Antibodies were desalted with Zeba spin desalting columns (0.5 ml, 7 kDa molecular weight cut-off, size exclusion columns, ThermoFisher) into PBS pH 7.2 + 1 mM EDTA. (l)-Dehydroascorbic acid (20 equiv., to a final concentration of 10% DMF) was then added and the antibodies were incubated for 3.5 h at 25 °C. Antibodies were desalted with Zeba spin desalting columns into NaPi 20 mM pH 8. General cell culture conditions. Cells were incubated in a humidified 10% CO 2 /90% air atmosphere at 37 °C. HL-60 cells were cultured in RPMI medium (Gibco) with 10% heat-inactivated fetal bovine serum. HEL cells were grown in RPMI media with 20% heat-inactivated fetal bovine serum. Leukaemia cells were maintained at a density of 1 × 10 6 cells ml −1 . Cells were split every second day to keep them in the exponential growth phase. SKBR-3 and HCT-116 cells were grown in McCoy's 5A Modified Medium (Gibco) with 10% heat-inactivated fetal bovine serum. MCF-7 cells were grown in Dulbecco's Modified Eagle Medium (Gibco) with 10% heat-inactivated fetal bovine serum. Adherent cell lines were passaged by addition of trypsin-EDTA (0.25%) (Gibco). HL-60 was kindly donated from the group of Prof. Bruno Silva-Santos, iMM Lisbon. HEL was kindly donated from the group of Prof. Bruno Silva-Santos from iMM, Lisbon (in vivo studies) or Dr Isaia Barbieri, Department of Pathology, University of Cambridge (in vitro ADC studies). HeLa, SKBr3 and HCT-116 were purchased from ATCC. MCF-7 was kindly donated from the group of Dr Sérgio Almeida, iMM Lisbon.
Viability assays with small-molecule lapachone models. Assays assessing cytotoxicity of β-lapachone and derivatives were performed by CellTiter-Blue assay (Promega) according to the manufacturer's instructions. Briefly, adherent cancer-cell lines were plated at 20,000 cells per well in 96-well plates or suspension cell lines were plated at a concentration of 2 × 10 5 cells ml −1 in 24-well plates. Compound stocks in media were made by dilution of DMSO stocks containing the compound of interest to obtain a final in-plate DMSO concentration of only 0.1%. Incubation was performed for 48 h. Following incubation, media was replaced with media containing CellTiter-Blue (Promega) in 1:10 dilution and the plates were incubated for 1.5-4 h. The fluorescence of the plates was recorded (excitation, 555 nm; emission, 585 nm; MiniMax i3x Reader, Molecular Devices). Cell viability was calculated by division of the fluorescence intensity of treated wells by that of the calculated average fluorescence intensity of replicate negative control wells containing cells with 0.1% DMSO only. IC 50 values were calculated using GraphPad Prism 8 software.
N-Acetyl cysteine assay. For assays involving cytotoxicity of β-lapachone derivatives with and without antioxidant N-acetyl cysteine, HL-60 cells were plated into 24-well plates at a concentration of 2.5 × 10 5 cells ml −1 in 450 µl. To each well was then added 25 µl of drug diluted in PBS to obtain the correct in-plate drug concentration, and 25 µl of PBS or a stock of N-acetyl cysteine at 12 mM in PBS.
DMSO was maintained at an in-plate concentration of 0.1%. After 96 h incubation, toxicity was assessed by counting of live cells using the trypan blue exclusion method with a Countess II Automated Cell Counter (ThermoFisher Scientific) according to the manufacturer's instructions. Cell viability was determined by dividing the average live cell number of treated wells by the average live cell number for control wells treated with 0.1% DMSO. Viability for wells treated with N-acetyl cysteine was assessed by comparison with control wells treated with N-acetyl cysteine and 0.1% DMSO. IC 50 values were calculated using GraphPad Prism 8 software.
Viability assays with small-molecule prodrugs and ADCs using the trypan blue exclusion method. For assays assessing cytotoxicity of β-lapachone versus small-molecule prodrugs, antibodies and ADCs, suspension cells were plated into 24-well plates at a density of 2.5 × 10 5 cells ml −1 in 240 µl. Stocks of ADC or antibody in PBS were normalized to 12.5 µM. Serial dilutions of compound were made up in PBS. Then, 60 µl of each stock of either PBS or compound in PBS was added to the cells to achieve the desired final concentration. After the specified incubation time, toxicity was assessed by counting of live cells using the trypan blue exclusion method with a Countess II Automated Cell Counter (ThermoFisher Scientific) according to the manufacturer's instructions. Cell viability was determined by dividing the average live cell number of treated wells by the average live cell number for control wells treated with PBS only. IC 50 values were calculated using GraphPad Prism software. Note: for the assay performed over 96 h, 5% human serum was added to media instead of 10% fetal bovine serum, to limit non-specific antibody internalisation. This had little effect on ADC toxicity.
Lysosomal inhibition assay with small molecule prodrugs. Assays to assess the effect of lysosomal activity on prodrugs were performed as follows: HL-60 cells were plated into 96-well plates at a concentration of 2.5 × 10 5 cells ml −1 in 180 µl media. Cells were treated with 10 µl of a stock of NH 4 Cl in PBS (to obtain an in-plate concentration of 3 or 30 mM), 10 µl of a dilution of a DMSO stock of E64d in PBS (to obtain an in-plate concentration of 30 μM), PBS only or a dilution of DMSO in PBS. Cells were incubated for 30 min. Following this, 10 µl of a stock of test compound made by dilution of a DMSO stock into media was added. DMSO was present in-plate at a maximum of 0.2%. Cells were incubated overnight for 16 h. Following incubation, media was replaced with media containing CellTiter-Blue (Promega) in 1:4 dilution and the plates were incubated for 6 h. The fluorescence of the plates was recorded (excitation, 555 nm; emission, 585 nm; MiniMax i3x Reader, Molecular Devices). Cell viability was calculated by division of the fluorescence intensity of treated wells by that of the average fluorescence intensity of the relevant negative control wells.
Protein mass spectrometry. LC-MS to analyse protein samples was performed on a SQ Detector 2 connected to an Acquity UPLC system using an Acquity UPLC BEH300 C4 column (1.7 μm, 2.1 mm × 50 mm). Flow rate was set at 0.2 ml min −1 with eluents of solvent A (water with 0.1% formic acid) and solvent B (71% acetonitrile, 29% water with 0.075% formic acid). The gradient was from solvent A/B (72:28) to 100% B over 25 min, followed by solvent B for 2 min and then up to solvent A/B (72:28) over 18 min. A capillary voltage of 2.0 kV and a cone voltage of 40 V were used for the electrospray source. The desolvation gas was nitrogen at a total flow of 850 l h −1 . Mass spectra were reconstructed using the MaxEnt algorithm preinstalled on MassLynx software (Waters) from the ion series. To obtain the ion series described, the major peak(s) of the chromatogram were selected for integration and further analysis.
Native protein MS. Native MS evaluations were performed on a Synapt High Definition Mass Spectrometer (Waters). All protein samples were buffer exchanged into 200 mM aqueous ammonium acetate buffer solution in water using Zeba spin desalting columns (0.5 ml, 7 kDa molecular weight cut-off, size-exclusion columns, ThermoFisher). The samples were further diluted to 2-10 µM before native MS evaluation. Sample aliquots of 2.5 µl were transferred to a borosilicate emitter (Thermo Scientific) before being introduced into the mass spectrometer. The instrument was set up for the detection of high mass complexes. The conditions used for these experiments were: capillary voltages, 1.6-2.5 kV; sample cone voltages, 100-180 V; source temperature, 80 °C, extraction cone voltage, 3-8 V; nanoflow gas pressure, 0.02-0.5 bar; backing pressure, 3.5-3.9 mbar; trap pressure, 3.2 × 10 2 mbar, ion mobility spectrometry (N 2 ) pressure, 6.1 × 10 -1 mbar; and time-of-flight pressure, 7.1 × 10 -7 mbar. Spectra were calibrated externally using caesium iodide. Data acquisition and processing were performed using MassLynx 4.1 (Waters).
In vivo AML xenograft study. The AML cell line used in this study, HEL cells, were a kind donation of Prof. Bruno Silva-Santos, iMM, Lisbon. The cells were maintained in a humidified incubator at 37 °C under 5% CO 2 and grown using 1× RPMI 1640 medium without l-glutamine (Invitrogen, Life Technologies) supplemented with 10% heat-inactivated fetal bovine serum (FBS) (Gibco, Thermo Scientific), 1× MEM NEAA (Gibco, Thermo Scientific), 1× sodium pyruvate (Gibco, Thermo Scientific), 1× GlutaMAX (Gibco, Thermo Scientific), 200 U ml −1 penicillin and 200 µg ml −1 streptomycin (Gibco, Thermo Scientific) and 10 mM HEPES (Gibco, Thermo Scientific). All animal experiments were conducted at the Instituto de Medicina Molecular João Lobo Antunes (Lisbon). Animal work was performed in strict accordance with Portuguese law (Portaria 1005/92) and European Guideline 86/609/EEC and follows the Federation of European Laboratory Animal Science Associations guidelines and recommendations concerning laboratory animal welfare. Furthermore, all animal experiments were approved by the Portuguese DGAV and the IMM Animal Ethics Committee (authorization AWB_2017_11_GB_AMLeukemia). A localized model of AML was established in 8-week-old female NOD-SCID mice (purchased from Charles River) by inoculating 5 × 10 6 HEL cells subcutaneously in the flank. Tumour growth was monitored over time, by performing bilateral vernier caliper measurements every 3-4 d and mean tumour volumes were calculated using the formula (length × width 2 )/2. Treatments were initiated when tumours reached approximately 100 mm 3 (approximately 10 d after tumour induction), with the mice having been randomly assigned to receive DAR-2 ADC or native antibody and PBS as controls. Treatments were administered intravenously in two injections with a 3 d interval. Animals were observed every 3-4 d; tumours were measured as described before and mouse weight was evaluated throughout the study. Once tumours reached 1,000 mm 3 animals were killed, and maximal tumour burden was not exceeded. No signs of animal suffering or discomfort were observed, including no weight loss. The light/dark cycle was 14 h light/10 h dark (lights on at 07:00; lights off at 21:00). The temperature was 20-24 °C and the relative humidity was 55 ± 10%. The type of food was autoclaved diet pellets RM3A (P), from SDS Special Diets Services (product code 801030). Food was placed in a grid inside the cage and provided ad libitum to animals. The type of water was sterile water treated by reverse osmosis. Water was provided ad libitum to animals through bottles with a capillary hole. The data collected were analysed using GraphPad Prism8. Once the mice had been killed (by isoflurane overdose) comprehensive necropsy was performed. Macroscopic findings were recorded, and heart, lung, left and right kidney, liver, spleen, central nervous system and primary tumours were collected for histopathology. Samples were immersion fixed in 10% neutral buffered formalin, routinely processed for paraffin embedding, sectioned at 4 µm, and stained with haematoxylin and eosin. Lesions were classified according to previously published criteria (International Harmonization of Nomenclature and Diagnostic Criteria for Lesions in Rats and Mice) and scored according to a six-tier severity scale: 0, absent; 1, minimal; 2, mild; 3, moderate; 4, marked; 5, severe. Distant metastasis were scored according to a five-tier severity scale: 0, absent; 1, minimal; 2, mild; 3, moderate; 4, marked. Representative haematoxylin and eosin pictures were obtained using NDP.view2 software (Hamamatsu) in slides digitally scanned in the Hamamatsu NanoZoomerSQ (20× for heart, lung, liver, spleen, kidney and primary tumour; and 10× for the central nervous system).
Quantitative LC-MS/MS of β-lapachone in formol-fixed xenograft mouse tumour and formol supernatant. Concentrations of β-lapachone in formalin-fixed tumour and the respective formol supernatant were quantified using reversed-phase ultra-high-performance liquid chromatography tandem mass spectrometry as previously described 85 . Care was taken during sample handling to minimize exposure to light, and amber glass/plastic and foil wrapping were used where possible. Tumour tissue was homogenized to a concentration of 200 mg ml −1 with PBS. An aliquot of homogenate or formol supernatant sample was transferred to a fresh tube with cryptotanshinone (internal standard) in ethyl acetate for liquid-liquid extraction. Samples were mixed, centrifuged and the ethyl acetate layer transferred to a 96-well plate, evaporated and reconstituted in 0.1% formic acid in acetonitrile:water 70:30. PBS was used as the surrogate matrix for calibration standards and quality-control samples. The calibration ranges were 0.25-250 ng g −1 (homogenate) and 0.05-50 ng ml −1 (formol). The analytical batches included quality-control samples at low, medium and high concentrations. The analysis was performed on a Shimadzu Nexera X2 UHPLC coupled to a Sciex TripleTOF 6600 mass spectrometer. The extracted samples were injected onto a Phenomenex Kinetex EVO C18, 1.7 µm, 50 × 2.1 mm analytical column, at 35 °C. Gradient elution was performed using 0.1% formic acid in water and an increasing percentage of 0.1% formic acid in acetonitrile at 0.5 ml min −1 , with a total run time of 4 min. The mass spectrometer was operated with positive electrospray ionization and enhanced high-sensitivity product ion scans of 100-500 Da, from precursor ions m/z 243.0 (β-lapachone) and m/z 297.0 (cryptotanshinone). Extracted ion chromatograms were produced from product ions m/z 187.0358 (β-lapachone) and m/z 251.1411 (cryptotanshinone), with a range of ±0.025 Da. The peak area ratios from the integrated analyte and internal standard chromatograms were used to back-calculate the concentrations of β-lapachone against a linear 1/x 2 weighted calibration curve.
Quantum mechanical calculations. Full geometry optimizations were carried out with Gaussian 16 (ref. 86 ), using the M06-2X hybrid functional 87 and 6-31+G(d,p) basis set in combination with ultrafine integration grids. Bulk solvent effects in water were considered implicitly through the IEF-PCM polarizable continuum model 88 . The possibility of different conformations was taken into account for all structures. All stationary points were characterized by a frequency analysis performed at the same level used in the geometry optimizations from which thermal corrections were obtained at 313.15 or 298.15 K. Frequency analyses were carried out at the same level used in the geometry optimizations, and the nature of the stationary points was determined in each case according to the appropriate number of negative eigenvalues of the Hessian matrix. The quasiharmonic approximation reported by Truhlar and co-workers was used to replace the harmonic oscillator approximation for the calculation of the vibrational contribution to enthalpy and entropy 89 . Scaled frequencies were not considered. Mass-weighted intrinsic reaction coordinate calculations were carried out using the Gonzalez and Schlegel scheme 90,91 to ensure that the TSs indeed connected the appropriate reactants and products. Single-point energies were alternatively calculated on the optimized geometries using combinations of different density functionals (M06-2X and ωB97x-D 92 ), basis sets (6-31+G(d,p) and 6-311++G(2d,p)) and implicit solvation models (IEF-PCM and SMD 93 )) (Supplementary Table 4). Gibbs free energies (ΔG) were used for the discussion on the relative stabilities of the considered structures. The lowest-energy conformer for each computed stationary point was considered in the calculation of the elimination activation barriers; all the computed structures can be obtained from authors upon request. Computed molecular structures were depicted using open-source PyMol 2.3 (https://pymol.org). Electronic energies, entropies, enthalpies, Gibbs free energies and lowest frequencies of the calculated structures are summarized in Supplementary Table 5. Cartesian coordinates of the lowest-energy structures calculated with PCM(H2O)/M06-2X/6-31+G(d,p) are shown in Supplementary Table 6. Theoretical kinetic and equilibrium constants derived from fitting of experimental and computed data were calculated using Microcal Origin Pro 2020b.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
Data supporting the findings of this study are available within the paper and its Supplementary Information. The Supplementary Information reports experiments described within the manuscript in greater detail, and describes synthetic procedures and characterization data. Requests for materials should be addressed to G.L.J.B. and G.J.O. Source data of HPLC traces used to calculate the kinetics of PAB-PhQ, PAB-BL and PAB-DN elimination in aqueous solution are provided with the manuscript. All computed geometries, energies and fitting data can be accessed through the Zenodo repository (https://doi.org/10.5281/zenodo.6325898). No restrictions on data availability apply. Source data are provided with this paper. Fig. 1 | Additional in vitro assays of the benzyl ketol protection unit. a, (i) Concentration-dependent in vitro methaemoglobin generation by 1, (ii) appearance of methaemoglobin peak at 630 nm for samples at 100 µM. Line represents mean values, with dotted borders representing ± SEM (n = 3). This experiment was repeated once with similar results. b, Toxicity of 1, 24 and 25 to cancer cell lines: HL-60 (acute myeloid leukaemia), HeLa (endocervical adenocarcinoma) and HCT-116 (colon carcinoma). Data are presented as mean values fit to 4-PL model with error bars representing ± SEM (n = 3). These experiments were replicated once with similar results. c, Viability of leukaemia cell line MOLM-13 at 24 h timepoints following treatment with peptide prodrugs at 5 µM compared to compounds 1 and 29. Values are mean normalised cell viability from three biological replicates ± SEM (n = 9). d, Lysosome deacidification and inhibition assay in HL-60 over 24 h. Significant difference marked: paired two-tailed t test for 26, untreated vs. NH 4 Cl 30 mM treated viability: p = 0.0199. Bar height displays mean values from one representative experiment with error bar displaying ± SEM (n = 3).