Evaluation of PNU-159682 antibody drug conjugates (ADCs)

Keywords: antibody drug conjugate; PNU-159682; nemorubicin; doxorubicin

Abstract

PNU-159682 is a highly potent secondary metabolite of nemorubicin belonging to the anthracycline class of natural products. Due to its extremely high potency and only partially understood mechanism of action, it was deemed an interesting starting point for the development of a new suite of linker drugs for antibody drug conjugates (ADCs). Structure activity relationships were explored on the small molecule which led to six linker drugs being developed for conjugation to antibodies. Herein we describe the synthesis of novel PNU-159682 derivatives and the subsequent linker drugs as well as the corresponding biological evaluations of the small molecules and ADCs.

In a balance of efficacy and toxicity, highly potent (sub-nanomolar IC50) cytotoxic small molecules see some use in cancer therapy but too often lack the requisite therapeutic window. It has long been sought after to selectively deliver potent molecules directly to tumor tissue.1 One ongoing attempt to create a magic bullet lies in the antibody drug conjugate (ADC) platform, which seeks to combine the specificity of a monoclonal antibody with the potency of a cytotoxic drug (colloquially, “warhead”) to achieve an optimized therapeutic index.2,3 With the FDA approval of seven ADCs and more than a hundred ongoing clinical trials,4 companies globally are seeking to develop antibodies for new targets, linkers for controlled release, and new small molecule warheads to fine-tune the desired potency.5,6

Highly potent molecules such as calicheamicin or pyrrolobenzodiazapenes, previously shelved for toxicity issues, have seen resurgence in use under the ADC regime7,8 PNU-159682 (PNU, 1)9 is one such molecule which bears extreme potency and as such is being examined for use in ADCs.10-12 PNU is an oxidized secondary metabolite nemorubicin (MMDX, 2). It is substantially more potent than either MMDX (2) (800–2400-fold) or doxorubicin (3) (2100–6400-fold), a commonly used chemotherapeutic anthracycline (Figure 1).9,13 Due to their distinct mechanism of action from their naturally occurring parent, doxorubicin (3), and lower doses needed for efficacy, nemorubicin and PNU (1) may avoid the known dose limiting cardiotoxicities of doxorubicin.14 Although it is noteworthy that anthracyclines remain heavily prescribed as chemotherapies and some of the first ADCs ever developed were doxorubicin conjugates,15,16 to date none of the seven commercial ADCs utilize an anthracycline-based warhead: either PNU-159682 (1), its metabolic parent nemorubicin (2), or any other approved anthracycline.

Herein we seek to evaluate PNU-159682 for use in an ADC platform. Medicinal chemistry was utilized to assess structure activity relationships around both the C-14 alcohol side chain and, to a lesser extent, around the saturated tricyclic heterocycle. A variety of linker structures were evaluated to attempt to address some of the inherent issues with PNU-159682 linker drugs, including stability and solubility. The warheads were evaluated for in vitro cytotoxcity and antibody drug conjugates were synthesized, purified, and tested for in vivo efficacy.

Due to the potency of ADC warheads very small quantities are needed for biological evaluation. As such, starting with complex commercially available molecules such as PNU (1) or the synthetic derivative PNU-acid 4 is a practical approach for medicinal chemistry (Figure 1).17 To this end, several heterocyclic analogues of 1, compounds 6-11, were prepared through straightforward derivatization of PNU.

During the synthesis of PNU-159682 (1) derived linker drugs, the primary point of attachment was at the C-14 alcohol. Reacting PNU with bis(4-nitrophenyl) carbonate to functionalize the alcohol for linker attachment often led to a cyclic carbonate (5) being observed by LCMS as a side product, arising through intramolecular cyclization of the activated intermediate with the C-9 alcohol (Scheme 1). During the second step in the synthesis of one such linker, a clean sample of cyclic carbonate 5 was obtained through preparative HPLC purification. The purified sample was submitted for an in vitro cytotoxicity assay and to our delight it showed potency of similar levels to PNU-159682 (Table 1).

To this end, 1,2,4-triazole derivative 6 can be prepared from PNU-159582 by first using a known NaIO4 mediated oxidative cleavage of the α-hydroxyketone to give commercially available PNU-acid 4.17,28 Coupling of PNU-acid 4 with hydrazine hydrochloride followed by ring closure with ethyl acetimidate gave the desired 1,2,4-triazole 6. The 1,3-azole series (compounds 7-11) were all prepared through mesylation of the C-14 alcohol to form the reactive PNU-OMs 12, followed by cyclization with an appropriate 1,3-heteroatom containing building block (see Scheme 1).29

Each of the heterocycles 5-11 was analyzed by UPLC-MS to confirm that none of the parent PNU-159682 remained, whose extreme potency may have altered the results of in vitro potency assays. Gratifyingly, several of these PNU-heterocycles retained much of the remarkable potency of the parent compound 1. In the 2-methyl series: the 1,2,4- triazole (6) and 1,3-imidazole (7) analogues were less potent toward our diagnostic cell line overexpressing MDR pumps (MES SA/DX) with EC50 >10 nM (Table 1). For this reason, these targets were considered less interesting and were not pursued further as ADCs. On the other hand, the 2-methylthiazole derivative 8 showed noteworthy potency in MES SA, MES SA/DX, and 293T cell lines with potencies ~10 pM, approximately a single order of magnitude loss in potency from the parent compound, PNU-159682 (< 1 pM). This minor loss of potency was not considered to be an issue given the already remarkably potent nature PNU-159682 and having less potent analogues may enable learning about the toxicity profiles of these molecules in future studies.

Pushing the thiazole line of derivatives further, 2-aminothiazoles (9 and 10) were considered to be linkable target molecules for ADC warheads. PNU-2-Me2N-thiazole 11 was considered as a possible substrate for tertiary amine conjugation.30 However, this was not pursued, and it was only tested as a small molecule at this time. The presence of the 2- amino group establishes a handle on which to attach the linker to the antibody, and secondly, the synthesis of heterocycles 9-11 proved more facile than the 2-methylthiazole derivative 8 which was low yielding and required a Lewis acid for ring closure. This was presumably due to the greater nucleophilicity of the thiourea cyclization partners compared to the thioacetamide used in PNU-2-methylthiazole (8) (Scheme 1). The 2-aminothaizole series was prepared on ~10-50 mg scale in unoptimized yields ranging from ~40-70% (including the necessary preparative HPLC purification) under mild conditions: EtOH at 40 °C with no additional acid or base.

It is noteworthy to mention the instabilities observed and drawbacks when working with PNU-159682 as a small molecule, as well as some of the failed chemistries attempted. Focusing on the saturated, fused-tricyclic ring system of PNU- 159682 reveals the most obvious of the instabilities inherent to this molecule. This ring system is prone to various epimerizations and hydrolyses (Paths A and B, Scheme 2). In fact, this reactive core is likely related to the alkylation potential of this ring system and mechanism of action of the small molecule.31

These types of impurities and side products were routinely characterized by MS while doing chemistry on the small molecule and linker drugs. Epimerizations and hydrolyses most notably occurred under acidic conditions, particularly at elevated temperatures, which limits the chemistry that can be performed on 1. UPLC analyses and HPLC purifications were challenging until it was determined that a buffered 20 mM ammonium formate (pH 4.0) solution did not lead to decomposition of the PNU (1) warhead. Hydrolysis products with structures consistent with products 13 and 14 were isolated after incubating PNU (1) with acetic acid/water overnight. Interestingly, under acidic conditions there seems to be conversion between these hydrolysis products (13 and 14) which is observable by UPLC. Somewhat unexpectedly, when hydrolysis product 14 was isolated it showed five orders of magnitude lower potency than PNU (1) (see SI for details). Additionally, it was established that the commercial PNU-159682 contains ~5-10% of an unconfirmed diastereomer that presumably remains in most of the compounds generated (see SI for analytical details on 1). Given the early stages of this project, this isomer was not followed up on in detail. The most likely sites of epimerization are shown in path A.

Elimination products (Path C, Scheme 2) were observed by MS in PNU-159682 small molecules, linker drugs, and ADCs. This was observed to be a more significant problem in the ADCs than in the small molecules but were always in low abundances.Additionally, it is well established that the anthracycline core can undergo redox cycling in vivo (Path D, Scheme 2). This has not been demonstrated on PNU (1), however, when ADCs containing PNU-159682 (1) were characterized using mass spectrometry, the [M + 2] impurities were routinely observed. Attempts to repeat this chemistry on the small molecule or linker drug for more accurate MS analysis and structural determination proved unfruitful. This redox cycling may be promoted by various components of the drug conjugation (thiols), the plasma in the in vitro assay, or light. It was noted that a solution of PNU (1) decomposes nonspecifically when exposed to light for extended periods of time (>95% pure to ~70% over 4 days). As such, most of the chemistry of PNU-159682 was conducted shielded from light and samples of the small molecule and ADC were stored in brown amber vials.

For these reasons, the chemistry used to work on PNU-159682 small molecules, linker drugs, and ADCs was both challenging and quite limited. If these molecules were to be developed further, aspects related to inherent instability need to be understood in greater detail.Several chemistries were attempted to modify both the saturated heterocycle and anthracycline core of doxorubicin and PNU-159682 to in hopes of mitigating some of these stability issues. In one case, it has been demonstrated that doxorubicin (3) could be converted to the 5-imino-doxorubicin (15) through the use of ammonia at low temperature (Scheme 3).32,33 We hoped that this would change the redox properties of the anthracycline core, and may allow for another position to link to an antibody. This chemistry worked reasonably well on doxorubicin and was somewhat successful when repeated on PNU-159682 (1), affording 5-imino-PNU 16.

Another interesting attempt to modulate the stability of PNU-159682 toward epimerization, hydrolysis, and elimination involved the enzyme mediated conversion of doxorubicin into doxorubicin-7-thioglycoside 17.34 This known chemistry was demonstrated on doxorubicin (3) but unfortunately, conversion was low and optimization was not pursued. Several steps would be necessary to synthesize the heterocyclic sugar found in PNU-159682 but substituted with a monothioacetal (18) (Scheme 3).35-37

With a suite of small molecule warheads available, PNU (1), PNU-acid (4), and the novel heterocyclic compounds 6-11, work began on a set of differentiated linker drugs. The choice of linker and conjugation mechanism can dramatically affect the conjugation, payload stability, purification, release, pharmacokinetic properties, and ultimately the potency of the ADC.11,38,39 As such, it is remarkably challenging to apply rational designs to this multivariable problem. Nevertheless, different linkers can be used to test hypotheses and refine linker drug structure in much the same way rational design can apply to more traditional medicinal chemistry. To this end a number of linker drugs were prepared (Figure 2). One of the more complex syntheses, specifically of linker drug 19, is demonstrated in Scheme 4. The complete synthetic details on the more straightforward syntheses of all linker drugs shown in Figure 2 can be found in the supporting information.

The synthesis of linker drug 19 begins with the preparation of PNU-p-nitrocarbonate 20 by reaction of PNU (1) with bis(4-nitrophenyl) carbonate under basic conditions (Scheme 4). The reaction is concentrated and used rapidly, as compound 20 is unstable. The principle side reactions are the intramolecular cyclization to form cyclic carbonate 5 (shown in Scheme 1) and hydrolysis back to starting material. Reaction of compound 20 with N,N’-dimethylethylenediamine leads to the formation of secondary amine 21. Secondary amine 21 can be coupled with a synthetically advanced linker 22, the synthesis of which is disclosed elsewhere,40,41 utilizing Hünig’s base and 1-hydroxybenzotriazole (HOBt) to give the fully- protected linker drug 23. To avoid any diastereomer formation, a mild two-step deprotection procedure is used to remove all the protecting groups and hydrolyze the methyl ester. The Fmoc protecting group is removed first with excess diethylamine, followed by a low temperature lithium hydroxide hydrolysis of the methyl ester and acetates to afford the deprotected penultimate amine 24. Finally, to form linker drug 19 the reactive maleimide is installed last via peptide coupling using HOBt and Hünig’s base.

The structure of the advanced linker (22) attached to PNU-159628 in linker drug 19 was chosen based on known desirable properties of our linker drugs and ADCs. Working from left to right on the linker: the maleimido-glycine moiety is both utilized for conjugation, but also is easily hydrolyzed to an open form after conjugation. This forms a stable attachment to the antibody as the hydrolyzed form slows retro-Michael addition of cysteine (leading to uncontrolled loss of warhead).42,43 The valine-alanine (Val-Ala) linker is a relatively stable linker recognized by cathepsin B for controlled cleavage and release of the reactive drug. And finally, the p-aminobenzyl alcohol (PABA), utilized across the ADC space, has been adorned with a stable C-glycoside that makes the entire linker drug more soluble, facilitating conjugation and preventing aggregation of the ADC.40,41

In a similar fashion, five additional linker drugs were made to test other warhead potencies, chemical properties, and toxicity profiles (Figure 2). Linker drug 25 was prepared with a stable (hydrolysable) maleimide attachment, containing a cleavable linker, and using a m-substituted aniline as a spacer that contains a less basic functionality in order to maintain good permeability, as compared to the N,N’-dimethylethylenediamine spacer used in linker drug 19.

With highly potent molecules, off-target or early loss of the warhead is a concern. To this end, several “non- cleavable” linker drugs were designed for later release upon degradation of the ADC versus cleavage by cathepsin B. These linker drugs had varying linker lengths, as there is evidence having the warhead proximal to the antibody can shield the warhead from degradation or early release.11 Shorter and longer non-cleavable maleimide linker drugs were prepared (26 and 27, respectively). These two molecules also had different mechanisms of attachment between the linker and warhead moieties. In linker drug 26 an amide bond is used to directly tie the linker to PNU-acid (4) and in linker drug 27 a carbamate attachment is used between the linker and PNU-159682 (1).

To test the PNU-heterocycle series two linker drugs were prepared using PNU-2-aminothiazole 10 as the warhead. Linker drug 28 was designed to be a long linker with a solubilizing PEG8 chain in the linker and the shorter conjugate 29 is the canonical Val-Ala cathepsin B cleavable linker but without any additional spacers.

As these compounds moved through our selection funnel conjugate 27 was eliminated because poor conjugation led to a low drug-to-antibody ratio (DAR), presumably due to its relatively high lipophilicity. Additional optimization of the conjugation conditions may have led to favorable conjugations but were not pursued at this time. Regarding lipophilicities, solubility during conjugation, and ADC aggregation, conjugate 19 which was designed to be more soluble, and as such has a negative cLogP, high tPSA, and elutes much earlier on a reverse phase UPLC run. Comparing cLogP we see 27 > 26 and 29 > 28 (Table 2). These calculations are consistent with structural intuition about the lipophilicities of these molecules. At this time, compound 28 was not pursued further as it was isolated as a ~1:1 mixture of diastereomers.

Many of these linker drugs were tested for in vitro plasma stability as one of the screening funnels to move the most stable compounds forward for in vivo assessment. The goal of assessing the in vitro plasma stability for antibody drug conjugates (ADCs) is to understand the stability of linker drugs while still attached to the antibody. Also, this method allows us to identify clippings, degradation and biotransformations of the linker drugs in various plasma matrices. Briefly, ADCs 25, 26 and 29 were tested for in vitro plasma stability in both mouse and rat plasma using the method described in SI section. The assay read-out measured stability as a rate of recovery of these linker drugs in plasma at day 7 as compared to the time 0 hr. The assay cut-off is +/-25%. Specifically, linker drugs 25 and 26 narrowly failed to meet this criterion at day 7 with less than 75% recovery in both mouse and rat plasma. On the other hand, ADC 29 passed the in vitro plasma stability screening funnel.

Several ADCs were carried forward into in vitro cytotoxicity assays for potency evaluation. Antibodies conjugated with linker drugs 19, 25, 26, and 29 were evaluated with an anti-CD46 antibody (target), against an HuIgG1 as an isotype control ( Table 3). Two cell lines with high CD46 expression levels were evaluated for most ADCs, and in all cases the targeted antibody showed higher potency than the non-targeted isotype control. Linker drug 25 is noteworthy for the difference in potencies between target and isotype, particularly in the MES SA cell line. Although modest potency is shown by hCD46-29 against Naïve 293T cells, in another cell line, MES SA, there is a significant difference between target-29 and isotype-29. While some of the observed differences in ADC potencies may be attributed to the different linker release mechanisms, overall we observed a correlation between potency of the free payload and the corresponding ADC – highest potency payload PNU-159682 provided the highest potency ADC hCD46-19 (47 pM in HEK 293T cells), while its less potent thiazole analogue 10 resulted in lower potency ADC hCD46-29 (1.6 nM in HEK 293T cells).

Linker drug 19 was selected for further in vivo evaluation as a representative member of the cleavable linker drugs that would release PNU (1) as a payload. Patient-derived xenograft (PDX) models with high levels of CD46 expression were used for the evaluation. Given the potency of PNU-159682 and mechanism of action we decided to test the ADCs in cancer indications that are typically very difficult to treat and have a high degree of chemotherapy resistance and tumor recurrence. We have selected five non-small cell lung cancer (NSCLC) and six colorectal cancer (CR) PDX models for the evaluation. Figure 3 shows in vivo efficacy in two representative models, LU253 (NSCLC) and CR188 (CR). A single dose of 0.5 or 1.0 mg/kg hCD46-19 ADC resulted in durable responses. Tumor regrowth did not occur for more than 80 days in 10 of 11 PDX models tested. Observations of tumor re-growth less than 60 days post-dose occurred for CR120 model (dosed at 1 mg/kg) or LU253 (dosed at 0.5 mg/kg). Importantly, there was no response observed with treatment of 1.0 mg/kg non- targeted HuIgG1-19 ADC, for which the response curve overlapped with that of vehicle treatment.

PNU-159682 (1) and its novel analogues were investigated as possible payloads for ADC therapeutics. Six linker drugs (19, 25-29) encompassing four differentiated warheads were successfully generated. The high reactivity of PNU and its linker drugs led to significant challenges to chemistry being uncovered and disclosed. In the course of investigation, we have noted that PNU ADCs produce small quantities of unusual oxidative products in the in vitro plasma stability assay which may require further investigation in vivo. ADCs demonstrated target-specific cytotoxicity in vitro, and, most importantly, ADC 19 has been highly efficacious in the in vivo NSCLC and CR PDX models. Single dose treatment at 1.0 mg/kg resulted in complete tumor regression and durable responses in majority of the models. Further investigation is warranted based on this preliminary efficacy data.