Mutasynthesis of C17- and C21-Substituted Ansamitocins for Use as ADC Payloads

Abstract

Strain engineering and process improvement were used to improve the titer of mutasynthetically generated ansamitocins generated by feeding 3-amino-5-hydroxybenzoic acid (AHBA) analogs to cultures of Actinosynnema pretiosum inactivated in AHBA biosynthesis. Ansamitocin analogs with fluorine and bromine substituents at C17 and C21 were then generated by feeding hydroxylated AHBA analogs. Fully processed C17 and C21 fluoro and bromo ansamitocins had cytotoxic activity similar to that of Ansamitocin P3. The C21 fluoro derivative was converted to a cytotoxic payload and an antibody drug conjugate (ADC).

Introduction

Antibody–drug conjugates (ADCs), the combination of a recombinant monoclonal antibody and covalently bound drug (usually a cytotoxic payload), have shown increasing success following the initial approval of Mylotarg (gemtuzumab ozogamicin) in 2000, with over 100 ADC candidates in different stages of clinical trials in 2022. Many factors are key in the clinical success of a new ADC; however, the properties of the monoclonal antibody (mAb), the linker, and the payload are most critical. Of these, the payload probably has the least breadth in development, with all approved ADCs based on one of two classesthe microtubule inhibitors or the DNA damaging agents. Two out of the 11 currently approved agents utilize maytansine derivatives (maytansinoids) as the payload: DM1 (1) in Kadcyla and DM4 (2) in Elahere.

Maytansinoids (Figure ) are natural product-based tubulin inhibitors with a 9-member ansa macrolide structure attached to a chlorinated benzene ring and were originally isolated from the shrub Maytenus ovatus. Even though DM1 and DM4 are highly potent and effective payloads, there is huge value in generating improved variants. In particular, increased potency, altered permeability, and solubility can all impact the therapeutic effect and ADC synthesis.

1.

General structure for maytansinoids, with examples.

While the biosynthesis of the plant natural product maytansine (3) has not yet been fully elucidated, the biosynthesis of ansamitocins, which are bacterially produced maytansinoids with an O-linked isobutyrate at the 3-position, was described in 2002 and requires the action of a modular polyketide synthase (PKS) initiated by the “starter unit” 3-amino-5-hydroxybenzoic acid (AHBA, a), Table ). The prototypic and most potent natural ansamitocin, ansamitocin P-3 (AP-3, 4), can be generated by scalable fermentation and then deacetylated via lithium aluminum reduction to form the 3-O-desacyl product maytansinol (5), which can then be subjected to further chemistry on the C3-hydroxyl position to make payloads such as DM1 and DM4.

1. Feeds Selected for Ansamitocin Analog Generation.

feed ID	R1	R2	R3	R4	R5
(a) (AHBA)	H	OH	NH2	H	H
(b)	H	Br	NH2	H	H
(c)	H	F	NH2	H	Et
(d)	Br	OH	NH2	H	H
(e)	H	OH	NH2·HBr	Br	H
(f)	F	OH	NH2	H	H
(g)	H	OH	NH2	F	H

Following generation of the macrocyclic core by the PKS, this is further decorated by a series of post-PKS processing enzymes, including a carbamoyltransferase (asm21), O-methyltransferase (asm7) at C20, chlorinase (asm12) at C19, N-methyltransferase (asm10), acylase at C3 (asm19), and epoxidase (asm11). Previous work has shown that the most highly processed analogs have the greatest activity, and analogs lacking an acyl group at the 3-O position are considerably less toxic. ,

Previous work by Kirschning and other authors has shown that mutasynthesis (mutational biosynthesis) can be a valuable tool for rapidly generating ansamitocin analogs. , In particular, by feeding analogs of AHBA to an ansamitocin producing strain with AHBA biosynthesis inactivated, such as in HGF073, ansamitocin analogs with altered substituents at the C19 and C20 positions have been isolated. This has included a wide range of substituents at these positions, along with varying levels of post PKS processing, such as altered acylation, carbamoylation, N-methylation, and epoxidation. There have been other targeting concepts used for maytansinoids. A folic acid-targeting conjugate and a magnetic nanoparticle have been reported. ,

However, to date, no generation of analogs with altered substituents at C17 or C21 have been described, and the potency of most mutasynthetic analogs isolated to date has been significantly lower than the potency of AP-3. In addition, the published titers of many of the products were low, potentially leading to issues with isolating enough material for downstream ADC linker chemistry.

Materials and Methods

Information on the Materials and Methods used in the preparation of this manuscript can be found in the Supporting Information associated with this publication.

Results and Discussion

To explore the formation of new analogs with improved activity that would also be amenable to linker chemistry to produce ADCs, we decided to expand the chemistry of mutasynthetic feeds incorporated into the ansamitocin template. The UV mutant strain HGF073 as used in the previous literature (ISOM-5711, DSM 21325), which harbors a large deletion of the genes involved in AHBA biosynthesis and is unable to make AP-3 without feeding AHBA, was utilized. This has previously been published as producing AP-3 at ∼65 mg/L when fed AHBA.

HGF073 was first subjected to selection of an improved strain to improve the consistency of ansamitocin analog production while retaining transformation efficiency. Isolated clones were screened for their amenability to incorporate an alternative starter unit (b, Table ) and carry out the post-PKS processing steps seen in AP-3 production. The isolate showing the most robust performance and highest titers of processed analogs was determined and designated as strain ISOM-5973. When AHBA was fed under standard fermentation conditions, this was seen to produce ∼100 mg/L AP-3.

In order to further increase the efficiency of the post-PKS processing steps, an additional copy of the asm19, asm11, and asm10 genes, responsible for 3-O-acylation, 4,5-epoxidation, and N-methylation, respectively, was introduced under control of the PermE* promoter via stable integration into the SV1 attP site of ISOM-5973 to create ISOM-6267. When ISOM-6267 was fed AHBA under standard fermentation conditions, this produced an average of ∼150 mg/L of AP-3.

Fermentation conditions were then optimized to produce more fully processed analogs. A range of conditions were assessed, and a key factor in production titer was found to be harvest date. For example, by extending the fermentation process to allow more time for the ansamitocin analogs to undergo the latter stages of post-PKS modification, the titer of 4 could be increased to ∼150 mg/L in ISOM-5973 and up to ∼225 mg/L with the modified ISOM-6267 strain. Additionally, the ISOM-6267 strain showed a significantly lower amount of partially processed less active analogs when fed AHBA, totaling ∼33 mg/L after 10 days, compared to ∼61 mg/L with the original ISOM-5973 strain. Therefore, not only did ISOM-6267 produce around half the amount of partially processed analogs compared to ISOM-5973, but the percentage of less active analogs was significantly reduced in ISOM-6267, with partially processed analogs making up 13% of total maytansinoid production in the improved strain, compared to 29% in the unmodified strain.

Following optimization of titer and AHBA incorporation, a range of AHBA analogs were added to culture broths and production of novel ansamitocin analogs assessed by HPLC, followed by isolation where titers were sufficient (see Tables and ). The compounds were then tested against a small panel of cell lines for cytotoxic activity (see Table ).

2. Ansamitocin Analogs and Their Antiproliferative Activity: Antiproliferative Activity Measured as the IC50 in nM .

chemical ID	feed used	R1	R2	R3	R4	R5	R6	R7	R8	C4–2(a)	HCT15(MDR+)(b)
4, AP-3	(a)	H	OMe	Cl	H	Me	Me	Me	epoxide	0.14	0.15
8	(b)	H	Br	H	H	Me	Me	Me	epoxide	0.28	0.75
6	(b)	H	Br	H	H	Me	Me	Me	alkene	0.49	2.4
7	(c)	H	F	H	H	H	Me	Me	alkene	2.1	4
9	(d)	Br	OMe	H	H	Me	Me	Me	epoxide	0.14	0.91
10	(e)	H	OMe	H	Br	Me	Me	Me	epoxide	0.13	0.26
11	(f)	F	OMe	Cl	H	Me	Me	Me	epoxide	0.13	0.21
12	(g)	H	OMe	Cl	F	Me	Me	Me	epoxide	0.14	0.29

^{a a}C4-2 cell line as a model for human prostate cancer ^bHCT15­(MDR+) cell line is a multi-drug-resistant model for colorectal cancer.

To show comparison to previously published work, 8, one of the C20–Br products (12c, disclosed in Knobloch et al. 2011), was generated. The AHBA analog b) (5b from Knobloch et al.) was fed to ISOM-5973, and incorporation was seen with the improved mean titer of 1.3 mg/L (stdev 0.1 mg/L) and up to 1.6 mg/L (compared to the previously published 0.3 mg/L). A further previously unreported analogue was also identified. This analog, 6, has most of the possible post-PKS processing changes present but is missing the 4,5-epoxidation step. While it defies the usually strictly defined order of later processing steps, with the epoxidation normally occurring as the penultimate step before the final N-methylation modification, similar products have been seen with C20–Cl and C20–I products (Taft et al. 2008 11d and 14c). This is likely a shunt product, where the N-methylation has occurred before the epoxidation across the 4,5-double bond. It may be that the change in chemical space, sterics, and/or electrics introduced by a halogen at C20 inhibits the activity of Asm11, which may not be able to process this molecule. In comparison, 7 has similar activity to 6, which indicates that this methylation is not essential for activity, and as expected, 6 was not as active as its fully processed counterpart 8 (see Table ).

While previous publications have shown incorporation of a range of AHBA analogs leading to new substituents at C19 and C20, positions usually occupied by the chlorine and methoxy group, respectively, in AP-3, no publications have previously shown the production or testing of ansamitocins with new substituents at C21 or C17, both usually unsubstituted. We therefore fed a range of AHBA analogs with either fluorine or bromine to incorporate at the C21 or C17 position.

At first, feed acceptance was poor, with either no incorporation or only trace analogs observed. However, the use of feeds containing a 3-hydroxy moiety allowed access to a range of analogs, including bromine or fluorine incorporated at C21 or C17 at reasonable titers. In comparison, the des-hydroxy equivalents were not readily accepted, producing either no ansamitocin analogs or only trace amounts of partially processed products.

When 3-amino-2-bromo-5-hydroxybenzoic acid (d) and 5-amino-2-bromo-3-hydroxybenzoic acid (e) were fed, no chlorine at C19 was evident, suggesting that the bulky bromine inhibited the action of the chlorinase due to either sterics or electronics. Even so, titers of the resultant products 9 and 10 were reasonable at mean titers of 3.8 mg/L (stdev 0.4 mg/L) and 10.4 mg/L (stdev 1.0 mg/L), respectively, and sufficient to allow isolation of enough material to confirm that both had similar activity to AP-3 in the C4–2 cell line but slightly worse activity in the HCT15 (MDR+) cell line.

When 5-amino-2-fluoro-3-hydroxybenzoic acid (f) and 3-amino-2-fluoro-5-hydroxybenzoic acid (g) were fed, a chlorine at C19 was evident in the major products, and compounds 11 and 12 were produced at good titers of 54.9 mg/L (whole broth average from 1.35 L fermentation) and a mean titer of 17.3 mg/L (stdev 1.5 mg/L), respectively. When isolated, they also showed comparable activity to AP-3 against C4–2 and 11 had similar activity against HCT15 (MDR+).

A range of other partially processed, alternatively acylated analogs and shunt products were seen in these fermentations, as is expected with this kind of mutasynthetic approach. Some of these products were present at high enough levels for isolation and analysis; however, it was the fully processed analog described more fully in this manuscript that proved most interesting.

Due to the reasonable titers of 11 and the cell line data showing similar activity to AP-3, chemistry was carried out to synthesize similar analogs reported previously ^,. Compounds 17 (payload) and 22 (linker-payload) were generated and tested against cell lines C4–2 and HCT15 as the naked payload and linker-payload conjugated to a targeting antibody for STEAP2, a known prostate tumor antigen.

Compound 11 was converted by first reducing off the isopropyl ester with LAH (Scheme ). The payload was synthesized in 3 steps from maytansinol analog 13. The known NCA 14 was coupled to maytansinol analog 13 using a zinc triflate-mediated esterification. A substituted nitro-benzoic acid 16 was coupled to compound 15 using HATU and the nitro group reduced using zinc powder under acidic conditions to yield the payload 17.

1. Chemical Synthesis of ADC Payload.

The linker-payload was synthesized in six steps (Scheme ). Starting from intermediate 18, phosgene conversion to the isocyanate was followed by coupling with Fmoc-Val-Ala-p-amino benzyl alcohol. Removal of the Fmoc group occurred during the formation of the carbamate to furnish dipeptide 19. Attachment of 20 to the Val amino group and deprotection of the carboxylic acid furnished 21. A HATU couple with 15 and piperidine removal of the Fmoc group yielded 22 that was ready for site-specific conjugation to the STEAP2 antibody.

2. Chemical Synthesis of Linker-Payload.

Two antibodies, an anti-STEAP2 antibody and nonbinding isotype control antibody derived from an immunological antigen having no relation to oncology both with an N297Q mutation on both heavy chains, were employed. Microbial transglutaminase (TG) conjugation was used to conjugate linker-payload 22 to the heavy-chain glutamines Q295 and Q297 site specifically.

Taking the payload and ADC, both derived from the semisynthesis above, we conducted cell viability assays (Table ). − Comparing compound 17 to the des-fluoro 17 in both prostate line C4–2 and the MDR line HCT15, we observed very similar potencies and degree of cell kill. The same was observed for the ADCs (fluoro- and des-fluoro). Either STEAP2 targeting ADC gave 2 nM activity even though the des-fluoro ADC had a slightly higher drug-to-antibody ratio (DAR). The isotype control ADCs were devoid of any activity, as expected.

3. Cytotoxic Analogs and ADCs with Their Antiproliferative Activity: Antiproliferative Activity Measured as the IC50 in nM .

entity	DAR	C4–2 IC50 (nM)	HCT15 (MDR) IC50 (nM)
compound 17 (des-fluoro)	n/a	0.348 (91.7)	1.84 (99)
compound 17	n/a	0.227 (91.6)	2.67 (98.2)
STEAP2–22 (des-fluoro)	3.8	2.00 (88.6)	n/a
isotype control-22 (des-fluoro)	3.7	>100	n/a
STEAP2–22	3.3	2.05 (86.4)	n/a
isotype control-22	3.2	>100	n/a

^aValues in parentheses are the percent of cell kill.

^{b a}C4-2 cell line as a model for human prostate cancer ^bhct15­(mdr+) cell line is a multi-drug-resistant model for colorectal cancer.

In summary, providing the mutasynthetic substrate with the hydroxyl already in situ appears to enhance incorporation rates and promote access to fully processed final products with halogen substituents at C17 and C21, which have cytotoxic activity similar to that of AP-3. Compound 11 was further converted to a potent cytotoxic payload and linker-payload. Once conjugated, the linker-payload was potent and selective in targeting the payload to antigen-positive cells. The isotype control ADC was devoid of any cytotoxic activity.

While this study demonstrates the feasibility of generating and evaluating novel C17- and C21-substituted ansamitocin analogs via mutasynthesis, several limitations should be acknowledged. The titers, degree of post-PKS processing, shunt products, and altered acylation varied significantly between different AHBA analog feeds, which impacted the consistency and yield of isolatable analogs. The natural variability of the strain and the relatively long cultivation times required to produce compounds for isolation also required careful handling and scheduling. In addition, the preparative isolation of closely eluting analogs posed a significant challenge, particularly when analogs differed by only minor structural features. This limited the ability to fully characterize some compounds using the full suite of analytical techniques (e.g., NMR, HRMS). The option to further scale up fermentations for all feed options would have enabled access to larger quantities of these analogs, facilitating more comprehensive structural elucidation and biological testing. The cost and availability of certain AHBA analogs were also restrictive to the scale of some experiments. Future studies could benefit from more cost-effective synthetic routes, including in-house synthesis of AHBA feeds, or optimization of alternative feed delivery mechanisms to improve uptake and incorporation efficiency. Despite these limitations, the study provides a strong foundation for further exploration of mutasynthetic strategies to diversify maytansinoid scaffolds.

The described methodology, in combination with strain and process improvements, enables the production of these novel ADC payload precursors. We also anticipate that this methodology will enable access to analogs with a broader range of substituents at C17 and C21, which may have improved properties and/or be more amenable to semisynthetic modification to generate further useful analogs.

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsomega.5c05529.

• Bioassays, synthetic methods, NMR characterization, and LCMS data (PDF)

Strain manipulations and/or bacterial fermentations were performed by B.G., J.K., C.M., J.P., and C.R.E. Chemistry analysis and/or isolation from bacterial fermentations was undertaken by N.C., J.F., A.H., H.H., J.L., and A.M. Chemical synthesis and conjugation work were completed by S.D., A.K., and B.W. Biological studies were performed by F.D., S.M., and F.Z. Intellectual guidance and experimental design for this work were supervised by A.S.-S., M.G., S.M., and T.N.

The authors declare the following competing financial interest(s): A.E.S.S., N.C., J.C.F., B.G., A.H., H.H., J.K., J.L., C.M., A.M., J.P., C.R.E., S.J.M. and M.A.G. are employees or ex-employees of and own shares or share options in Isomerase, F.D., A.K., S.M., F.Z., T.N. are employees of Regeneron and S.D. and B.W. are employees of Abzena.