The impactful and enduring role that natural products have played in improving the quality and duration of life for both humans and animals cannot be overstated. For example, rapamycin and its congeners have received Food and Drug Administration approval for various ailments, and this natural product continues to provide insights into basic cell biology1. Synthetic chemists are at the forefront of harvesting the full potential of natural products through synthetic efforts including classic total synthesis. Towards this goal, strategies to synthesize natural products have evolved significantly in recent years as more emphasis has been placed on biological function2. In Danishefsky’s ‘diverted total synthesis’ (DTS), a synthetic sequence is developed from simple building blocks employing classical retrosynthetic analysis and then various advanced intermediates previously employed in the synthesis effort are diverted towards simplified derivatives for biological analysis (Fig. 1a)3. In a strategy not necessarily directed towards total synthesis, Wender’s ‘function-oriented synthesis’ seeks to develop hypotheses regarding pharmacophores based on structural and computational analysis of distinct natural products and typically larger structure–activity relationship (SAR) data sets enabling the design and synthesis of simplified derivatives bearing a common pharmacophore4. Alternatively, Schreiber’s ‘diversity oriented synthesis’ (DOS) of natural product-like libraries5 and the derived ‘biology-oriented synthesis’ (BIOS) by Waldmann6 seeks to synthesize collections of compounds based on structural features found in natural products. Finally, Myers recently developed a convergent building block strategy for rapid access to a diverse array of structurally related scaffolds such as macrolides related to erythromycin for the discovery of novel antibiotics7; this was subsequently termed ‘analogue-oriented synthesis’ by Vanderwal8.

Fig. 1: Pharmacophore-directed retrosynthesis (PDR) applied to gracilin A and comparison to other synthetic strategies harvesting the rich information content of natural products.
figure 1

a, Strategies toward simplified, bioactive small molecules using natural products as starting points, including diversity-oriented synthesis (DOS) directed towards libraries of natural product-inspired compounds. b, PDR seeks to identify or hypothesize the pharmacophore of a natural product, dictating retrosynthesis by ensuring these features are present in multiple intermediates towards the natural product target.

The simultaneous alignment of total synthesis efforts with structure–activity relationship (SAR) studies has not been fully realized to the extent possible and in particular with novel natural products for which minimal SAR information exists. Several truncated natural products9, found to possess similar bioactivity to the parent natural product, are known (for example, eribulin mesylate from halichondrin10); however, these derivatives were typically identified following completion of a total synthesis11. In an example from our laboratory, a des-methyl, des-amino variant of the protein translation initiation inhibitor pateamine A was designed and synthesized following our total synthesis and found to have nearly equipotent activity to the natural product12. This led to a retrospective question of whether such a derivative may have been accessed en route to the natural product, leading us to pose the following question. Can the total synthesis of natural products, in particular with limited SAR or unknown or unconfirmed cellular targets, be more closely aligned to proposed biological activity during the retrosynthetic planning stages?

Here, we describe a type of retrosynthetic analysis that seeks to more closely align total synthesis efforts with concurrent biological studies. The strategy enables the identification of simplified versions of the natural product with similar potency or potentially new functions in the course of a total synthesis effort. We term this strategy ‘pharmacophore-directed retrosynthesis’ (PDR) to emphasize the importance of considering hypothesized pharmacophores at the retrosynthetic planning stage of a total synthesis effort. Although this approach increases the challenges of natural product total synthesis beyond important, contemporary goals including atom economy13, step and redox efficiency14, and protecting group avoidance15, it has the potential to greatly accelerate harvesting of the vast information content of natural products for basic cell biology and medicine. In PDR, we build on Wender’s notion of bringing function to the forefront of a synthetic endeavour (function-oriented synthesis), but employ the logic of retrosynthesis16 to target simplified intermediates that importantly possess the proposed minimal structural features required for bioactivity (the pharmacophore) en route to the natural product17. In applying PDR, a key first step is identification of a proposed pharmacophore. This may be based on (1) structural analysis with chemical intuition; (2) existing SAR from isolated natural product congeners; (3) the activity of structurally related compounds; (4) anticipated reactivity. A retrosynthesis is then devised that ensures that the proposed pharmacophore is present in multiple intermediates of increasing complexity, ultimately leading to the natural product. We selected a member of the spongiane diterpenoid family, gracilin A (1)18, to initiate assessment of the utility of PDR toward exploring its recently described immunosuppressive19 and neuroprotective activity20 given the limited SAR information and no prior synthetic work (Fig. 2a)21. In the retrosynthetic analysis, as complexity is increased towards the natural product, several intermediates possessing the proposed pharmacophore are specifically targeted, thus enabling SAR to be gathered as the total synthesis progresses (for example, 8 → 7 → 6 → 1 in Fig. 2b). It should be noted that several hypotheses regarding the pharmacophore of a particular natural product could be posited for PDR, leading to alternative retrosynthetic strategies.

Fig. 2: Select members of the gracilin A family and application of PDR to gracilin A.
figure 2

a, Naturally occurring spongiane diterpenoids bearing a common bis-acetoxy furanose moiety (red) selected as the hypothetical pharmacophore. b, PDR applied to gracilin A (1) leads to increasingly complex intermediates (that is, 8, 7, 6) throughout the course of total synthesis from key intermediate 9.

The gracilins, including the rare nor-diterpene gracilin A18, were originally isolated and characterized from the Mediterranean sponge Spongionella gracilis22. These diterpenes are structurally unique owing to the unusual diacetoxy furanose found in most members. The cytotoxic activity of gracilins B and G–I (isolated from Spongionella pulchella) against a diverse panel of 12 human cancer cell lines has been reported, but these compounds did not progress further in preclinical evaluation23. The absolute configuration of gracilin A was never established, while its relative configuration was based on X-ray analysis of the keto derivative of 9,11-dihydrogracilin A (3)24. Gracilin A was reported to be a potent inhibitor of phospholipase A2 (PLA2) with a 69% inactivation efficiency25. We previously reported that gracilin A was mildly cytotoxic against K562 and peripheral blood mononuclear cells (with half-maximum inhibitory concentration (IC50) values of 0.6 and 0.8 μM, respectively) and inhibited the epidermal growth factor receptor by 70% at 100 μM concentration22. We also recently posited that gracilin A mimics the effects of cyclosporin A (CsA) through interaction with cyclophilin A (CypA)19 and also improves the hallmarks of Alzheimer’s disease in vitro and in vivo20,26,27. These results prompted the current study to apply PDR to unravel and ideally differentiate the structural requirements for the immunosuppressive and neuroprotective effects observed with gracilin A.

The cyclophilins (Cyp) are highly conserved peptidyl–prolyl, cistrans isomerases (PPIase) involved in protein folding and trafficking28,29,30 and are found in multiple cellular compartments31. CypA, the cytoplasmic isoform, has several roles in cell metabolism and energy homeostasis32, with enhanced expression in inflammation and cancer33. The use of small molecules that selectively block the inflammation-related functions of CypA, is an important pharmacological strategy leading to effective immunosuppressive agents33. On the other hand, cyclophilin D (CypD), the mitochondrial isoform, translocates to form the mitochondrial permeability transition pore (mPTP) and its activity correlates with the mitochondrial dysfunction observed in Alzheimer’s disease that leads to neuronal death34. CypD inhibitors devoid of immunosuppressive activity through lack of binding to CypA but possessing the desired mitochondrial effect and appropriate blood–brain barrier permeability could provide an approach to address Alzheimer’s disease.

Herein, we demonstrate a proof-of-principle study of PDR through application to the spongiane diterpene, gracilin A, that has led to useful lead compounds for both immunosuppression and neuroprotection. This study revealed simplified gracilin A derivatives with high affinity for CypA and others that demonstrated significant selectivity for CypD over CypA, demonstrating their potential as neuroprotective agents devoid of immunosuppressive effects.

Results and discussion

Pharmacophore-directed retrosynthesis applied to gracilin A

The bis-acetoxy furanose moiety was selected as the pharmacophore of gracilin A based on several lines of evidence. Studies of the reactivity of macfarlandin E (which possesses a 1,4-dicarbonyl masked as a bis-acyloxy furanose) with lysine derivatives provided evidence for pyrrole formation through a Paal–Knorr process35, and mounting evidence suggests that this is possible with a number of other bis-acetoxy furanose-containing natural products36. Furthermore, a computational study demonstrated the potential of the bis-acyloxy furanoses of gracilin A and aplysulphurin-1 to bind divalent cations such as Ca2+, pointing to the potential importance of this moiety for bioactivity37. Finally, it is interesting to note that several bioactive members of the spongiane diterpenoid family possess a bis-acyloxy furanose including the structurally related gracilin L (2)22, 9,11-dihydrogracilin A (3), dendrillin (4)38 and tetrahydroaplysulphurin-1 (5)39. These considerations guided application of PDR to gracilin A and imposed a requirement that multiple intermediates along the synthetic route would bear or be converted to the proposed pharmacophore, namely the bis-acetoxy furanose (for example, derivatives 68, Fig. 2b). In this way, SAR studies could be conducted throughout the course of the total synthesis. We recognized that keto lactone 9, accessible in gram quantities through our recently described Diels–Alder lactonization organocascade40, would serve as a key intermediate to study the importance of the C8-exocyclic alkene (for example derivative 6), the C9-appended cyclohexyl moiety (for example, derivative 7). The importance of the bicyclic core of gracilin A would be ascertained by synthesis of the highly simplified monocyclic furanose 8 (Fig. 2b).

Sequential synthesis of increasingly complex gracilin A derivatives

We recognized that, in applying PDR, synthetic progress with concurrent biological assays of intermediates could best be achieved in stages based on the increasing complexity of intermediates in the synthetic sequence, as outlined in Figs. 3a–c and 4a. A final stage of diverted total synthesis could enable further refinement of the SAR profile through ‘gap filling’ with particular targeted derivatives to answer more specific questions building on information gathered during the initial stages. As applied to gracilin A, we first targeted the simple bis-acetoxy furanose 8 as the minimal pharmacophore that was readily obtained through a two-step oxidation/hydrogenation sequence to deliver the racemic syn-substituted acetoxy furanose (±)-8a along with the anti-diastereomer (±)-8b (d.r. 1.7:1). In a second stage, derivatives devoid of the cyclohexyl substituent and exocyclic alkene were targeted, namely bicyclic bis-acetoxy furanose 7a,b (Fig. 3b). The bicyclic lactone endo-15a (3:1 d.r., 94% e.e.) was obtained from a Diels–Alder-lactonization organocascade employing diene 12 and acryloyl chloride (11) with tetramisole as the Lewis base promoter. The required endo-diastereomer 15a could be isolated in 58% yield, and subsequent reduction with LiAlH4 followed by desilylation delivered ketodiol 16. Swern oxidation led to a dialdehyde that was directly subjected to acid-promoted acetylation to give the unstable keto bis-acetoxy furanoses 17 as a mixture of syn/anti diastereomers. A subsequent reduction with NaBH4 and dehydration with SOCl2 delivered the bicyclic bis-acetoxy furanoses 7a,b as a mixture of alkene regioisomers. The low yields obtained in this and subsequent four-step sequences leading to the bis-acetoxy furanoses were primarily a result of incomplete or non-simultaneous Swern oxidation of the diols leading to regioisomeric mono-acetoxy furanoses.

Fig. 3: Pharmacophore-directed retrosynthesis applied to gracilin A.
figure 3

a, Synthesis of a highly simplified analogue bearing the proposed pharmacophore. b, Assembly of the cis-fused, 6,5-bicyclic core 15a of gracilin A through an enantioselective, organocatalytic Diels–Alder-lactonization cascade and manipulation to a simplified bicyclic analogue 7. c, Annulation of the natural and a simplified cyclohexyl moiety. Inset, ORTEP representation of the single crystal X-ray structure of (–)-21b. TBS, tert-butyldimethyl silyl; r.r., regioisomeric ratio.

Fig. 4: Synthesis of gracilin A derivatives toward SAR profile gap filling.
figure 4

a, Alternative oxidation states of the original bis-acetoxy furanose. b, Select gracilin A derivatives in the enantiomeric series.

We next targeted derivatives devoid of the exocyclic alkene but bearing the cyclohexyl substituent (Fig. 3c). Desilylation of endo-Diels–Alder adduct 15a gave bicyclic lactone (–)-9, which was subjected to an allylzinc reagent derived from cyclohexenyl choride (±)-18 employing conditions we previously used in our synthesis of spongiolactone41. This delivered the tricyclic adducts 20 as a 1:1 mixture of diastereomers at the generated quaternary carbon but with high facial selectivity, leading to a single epimer at the tertiary alcohol centre. The contra-steric addition of the allylzinc reagent to the concave face of ketone (–)-9 was unexpected but likely due to stereoelectronic effects leading directly to a chair rather than a boat conformation and was verified by extensive NMR studies (see Supplementary Section C for details). In the case of addition of the gem-dimethyl cyclohexenyl zinc reagent derived from (±)-19, the stereochemistry was further verified by single-crystal X-ray analysis of adduct (–)-21b. However, this diastereoselectivity is inconsequential because the tertiary alcohol in (–)-21b is subsequently dehydrated. Hydrogenation reduced the cyclohexene of (–)-21b and a three-step process delivered the hydroxy bis-acetoxy furanose 22 as a 1:1 mixture of diastereomers. Dehydration with Burgess’ reagent42 gave alkenes 23 as a mixture of four diastereomers due to the alkene regioisomers produced. The four diastereomers were separable by preparative chiral HPLC, enabling biological analysis of each stereoisomer.

Gaps in the SAR profile were also back-filled following initial assays as described in the following (Fig. 4). This entailed application of aspects of diverted total synthesis to synthesize derivatives devoid of the cyclohexyl moiety (that is, 24), variations in the oxidation state and substituents of the tetrahydrofuran and cyclohexyl moiety (that is, 2627), and also exploration of the enantiomeric series (that is, 2829).

PDR enables sequential increases in complexity with correlation to immunosuppressive activity

As a preliminary biological screen, the initial gracilin A derivatives synthesized were analysed by surface plasmon resonance (SPR) with immobilized CypA43,44. Biological testing of gracilin A derivatives was performed in sets as the various stages of PDR were accomplished (Figs. 3a–c and 4a,b). Association curves (KD) were measured at various concentrations in comparison to gracilin A and CsA and the initially synthesized highly simplified derivatives, monocyclic (±)-8 and bicyclic 7 bis-acetoxy furanoses, exhibited no affinity for CypA up to 10 μM (Fig. 5). On binding to CypA, CsA and gracilin A modulate interleukin 2 (IL-2) release through the calcineurin pathway19, so the effects of these simplified derivatives on IL-2 production were also measured. Concanavalin A (ConA) was used to activate human T lymphocytes and induce IL-2 release45. T lymphocytes were pre-treated for 2 h with different non-toxic concentrations of compounds and then activated for 48 h in the presence of ConA (50 µg ml−1). After this time, levels of IL-2 released to the medium were measured by an enzyme-linked immunosorbent assay (ELISA) kit. As expected, the greatly simplified monocyclic (±)-8a,b and bicyclic furanoses 7a,b did not inhibit IL-2 production nor exhibit T-lymphocyte toxicity at concentrations up to 10 μM (MTT assay).

Fig. 5: Immunosuppressive activity of gracilin A derivatives.
figure 5

Kinetic equilibrium dissociation constants (KD, μM) for binding of gracilin A derivatives to CypA as measured by SPR (n = 4, data presented as mean ± s.e.m.), IL-2 release inhibition (IC50, n = 10, mean, CI) by ELISA, and cell viability of human T lymphocytes (half-maximal inhibitory concentration, EC50, n = 3, mean, CI) as determined by an MTT assay. CI (95% confidence interval); R2 values (0.91–0.98) were in the acceptable range (see Supplementary Section B for details). Inhibition of calcineurin (CN) phosphatase activity at 1 µM expressed in percentage of control cells (n = 4, mean ± s.e.m.). (>10.0 = not active up to 10 µM). ND, not determined.

Gracilin A derivative 23 bearing the pendant cyclohexyl moiety, began to display immunosuppressive activity (Fig. 5). The initial diastereomeric mixture of tertiary alcohols 22 derived from cyclohexenyl zinc addition was inactive. However, following dehydration to introduce the tri-substituted alkenes, immunosuppressive activity was observed with only a twofold decrease in KD relative to gracilin A (2.53 ± 0.40 μM) for alkene 23b bearing the 10S configuration (5.83 ± 3.33 μM) with moderate IL-2 inhibition (IC50 0.30 μM). The interplay between alkene regioisomers (Δ8,9 versus Δ9,11) and the C10 configuration was revealed through derivatives 23a and 23d, which exhibited nearly a 1,000-fold increase in KD toward CypA compared to gracilin A (5.34 ± 1.68 nM and 7.57 ± 1.61 nM, respectively), while isomers 23b and 23c were either significantly less active or completely inactive, respectively. Derivatives 29a and 23a displayed IC50 values for IL-2 inhibition of 0.12 and 0.15 µM, respectively whereas the C10 epimer of 23a, derivative 23b and the Δ9,11 regioisomer 23d, had higher values of 0.30 and 0.36 µM, respectively, while the epimeric, Δ9,11 regioisomer 23c exhibited a value of >10 µM, again pointing to the interplay between the alkene regiochemistry and the C10 stereochemistry for IL-2 inhibition. The discrepancy between binding (KD) to CypA as measured by SPR and the cellular inhibition of IL-2 expression may point to differential cell permeability or reflect differential binding to calcineurin, the presumed target of the CypA–gracilin A complex, impacted by the absence of the C8 exocyclic ethylidene. In addition, the conformation of the cyclophilins is highly dependent on the pH of the assay. This assay was performed at low pH (4.5) to demonstrate the activity of gracilin derivatives, which also lowers the KD for CsA compared to literature values (see Supplementary Section B and Discussion for details). To address this discrepancy, we also performed an enzymatic assay to measure calcineurin inhibition by gracilin A derivatives (at 1 μM) and, as expected, the most potent derivatives in the SPR assay showed comparable inhibition (22–31%) to CsA (23% at 1 μM). Despite differences in CypA binding affinity, these derivatives have comparable activity in both IL-2 and calcineurin assays, which supports the hypothesis that these derivatives lead to differential binding to calcineurin by the respective CypA–gracilin derivative complexes. These data, taken together, suggest that conformational preferences about the C9–C10 bond play a pivotal role in the immunosuppressive activity of the gracilin family.

To fill gaps in the SAR profile regarding immunosuppressive activity, we turned to the application of diverted total synthesis and targeted additional simplified derivatives (Fig. 5). The greatly simplified bicyclic mono-acetoxy ketone 24 and the bicyclic lactone 28, differing from the highly active bis-acetoxy furanose 23a only by the oxidation state of the tetrahydrofuran, were completely inactive. Epimerization of exo-15b enabled access to the enantiomeric series, leading to mono-acetoxy furanoses 29a,b epimeric at the quaternary C10 centre. Affinity to CypA (KD ≈ 3 μM) for both diastereomers 29a,b was comparable to gracilin A; however, differential inhibition of IL-2 production (IC50 = 0.12 versus >10 µM) was observed for both diastereomers while remaining relatively non-toxic to T cells (EC50 = 1–4 µM). The importance of the quaternary carbon stereochemistry and alkene regisomer on IL-2 inhibition, which directly impacts the conformation about the C9–C10 bond, is again highlighted by this enantiomeric series with the natural (S) configuration imparting the greatest activity.

Neuroprotective activity of gracilin A derivatives

Given the neuroprotective effects previously observed for gracilin A, the derivatives synthesized through application of PDR for immunosuppressive effects were also studied for potential neuroprotective activity and of particular interest was possible selectivity for CypD versus CypA PPiase activity (Fig. 6), a key property required for neuroprotective lead compounds devoid of immunosuppressive effects. All assays were performed at the indicated concentrations (0.001, 0.01, 0.1 or 1.0 µM) based on cytotoxicity determined with SH-SY5Y cells. In particular, compounds that displayed some toxicity against this cell line at 1 µM were tested at lower concentrations. Treatment of SH-SY5Y cells with a known potent oxidant such as H2O2 produces oxidative damage with a consequent increase in reactive oxygen species (ROS) release. ROS generation provokes mitochondrial dysfunction, increases cell death and affects the cellular antioxidant systems such as the glutathione (GSH) cycle46. Therefore, four parameters were measured to evaluate the neuroprotective effects of the simplified gracilin A derivatives: cell viability, mitochondrial membrane potential (ΔΨm), ROS release and GSH levels after cellular challenge with H2O2 (see Supplementary Section C for details). Several highly simplified gracilin A derivatives, namely monoacetoxy furanose 27a and the highly simplified diacetate 27b (both derived from 21a), the inseparable, tertiary alcohols 22 (1:1, d.r.) and the enantiomeric mono-acetoxy furanose 29a, but not its diastereomer 29b, and even the simplest bis-acetoxy furanose (±)-8a,b (1.7:1, syn/anti) displayed significant neuroprotective effects. In view of these results, the compounds with greater activity in oxidative stress assays (27a, 27b, 29a, 22 and 23c) were chosen to test their ability to block the opening of the mPTP. The simplest derivatives 8a,b and 7a,b displayed lower activities and only at one of the concentrations tested, so these compounds were not subjected to the following assays. In PPIase activity assays, selectivity for CypD over CypA was observed for derivatives 27a (~3-fold) and 29a (~18-fold) with the very simple diacetate 27b displaying the greatest differential CypD activity because it was inactive against CypA (up to 10 μM) but displayed activity against CypD (IC50 = 0.48 μM). In contrast, tertiary alcohols 22 had opposite selectivity for CypA over CypD (~4-fold). Our results suggest that the antioxidant effect of these gracilin A derivatives is mitochondrial-related, similar to gracilin A, mediated through interaction with CypD. These compounds protect cells from oxidative damage induced by H2O2 improving mitochondrial functioning measured by MTT and TMRM assays and increased GSH levels. Moreover, they block mitochondrial pore opening and induce CypD activity inhibition. Therefore, gracilin A derivatives hold potential as neuroprotective lead compounds that are devoid of immunosuppressive effects.

Fig. 6: Activity of gracilin A derivatives as neuroprotective agents.
figure 6

Cell viability against H2O2 toxicity, mitochondrial membrane potential restoration (TMRM), % inhibition of ROS release, % increase in cellular GSH levels, and mPTP blockage and IC50 values for inhibition of the enzymatic activity of CypD and CypA. The lowest effective concentration of each derivative is indicated in parentheses (μM). Values are calculated as the percentage of untreated control cells in comparison to H2O2-treated cells with the following values: H2O2 toxicity 55.6 ± 4.4, TMRM 75.9 ± 3, ROS 140.3 ± 4.4, GSH 82.9 ± 1.6, mPTP blockage 76.7 ± 2.8 (n = 4, mean ± s.e.m.). Cyp D and A activity are expressed as IC50 values (µM). CI (95% confidence interval); R2 values (0.91–0.99) were in the acceptable range (see Supplementary Section C for details); derivatives exhibiting neuroprotective effects while also showing differential activity of CypA versus CypD are highlighted (red boxes; n = 4, mean ± s.e.m.) (>10.0 = not active up to 10 μM). ND, not determined.

Summary of PDR and SARs gleaned from application of PDR to gracilin A

The total synthesis of natural products continues to be an important endeavour for the discovery of novel synthetic strategies and methods in addition to the exploration of biologically relevant chemical space. The described PDR approach brings biological function into the retrosynthetic planning stages to target multiple, simplified derivatives bearing a hypothesized pharmacophore en route to the natural product. PDR can be considered to be a subset of Wender’s function-oriented synthesis and will of course not be applicable to every natural product (that is, those where the majority of the structure is required for bioactivity). A caveat to this approach is that a balance must be found between synthetic convergency and targeting of bioactivity during the total synthesis. However, the application of PDR may only be limited by the creativity and intuition of synthetic chemists building on minimal SAR. For example, the SARs of isolated natural product congeners and other lines of evidence may direct one or more hypotheses regarding a proposed pharmacophore. Indeed, application of PDR can generate alternative strategies for a given natural product through alternative hypothesized pharmacophores in a similar way to consideration of various strategic bond disconnections in classic retrosynthesis. In the present study, the first SAR profiles of gracilin A derivatives were secured by providing proof-of-principle studies of PDR and support the notion of the bis-acetoxy furanose as a pharmacophore for immunosuppressive activity but not necessarily for neuroprotective activity (Fig. 7).

Fig. 7: SAR profile of gracilin A for both immunosuppressive and neuroprotective activity enabled through application of PDR.
figure 7

a, SAR profile in immunosuppressive assays. b, SAR profile in neuroprotective assays.

Evidence was gathered for our initial hypothesis invoking the bis-acetoxy furanose of gracilin A as the pharmacophore, as relates to immunosuppressive activity, given that derivatives with this moiety (for example, 23a and 23d) displayed the greatest activity (KD ≈ 5–7 nM), while those lacking this moiety were inactive (for example, (–)-28). Importantly, PDR disclosed that the C8-ethylidene is not required to elicit potent immunosuppressive activity and, with the availability of natural gracilin A as a comparator, it was unnecessary to complete a total synthesis.

While derivatives of CsA were previously investigated as potential neuroprotective agents through inhibition of CypD, they lacked effectiveness due to their high molecular weights and low blood–brain barrier permeability47,48,49. Several gracilin A derivatives (for example, 27a,b and 29a) accessed through this study, that interestingly includes the enantiomeric series, displayed significant neuroprotective activity, importantly while also displaying selectivity for CypD versus CypA inhibition. These gracilin derivatives serve as lead compounds for neurodegenerative diseases and other CypD-mediated diseases including atherosclerosis or autoimmune diseases50,51.

Although an ideal application of PDR will generally be challenging to implement, in particular when reactive functionality precludes a completely linear synthetic strategy as dictated by PDR, as in the present case of the gracilins, we expect that pharmacophore hypotheses brought into retrosynthetic planning will enable greater SAR information to be gathered en route to a natural product. It is also anticipated that application of PDR will provide an avenue for hypothesis-driven, natural product total synthesis efforts while simultaneously accelerating the exploration of natural product chemical space through total synthesis efforts premised on PDR.


Gram-scale Diels–Alder-lactonization organocascade providing γ-lactones (–)-15a and (+)-15b, the bicyclic core of gracilin A derivatives

To an oven-dried, 250 ml round-bottomed flask equipped with a magnetic stir bar was added silyloxydiene alcohol 1252 (2.14 g, 10.0 mmol, 1.0 equiv), (S)-(–)-TM·HCl (2.40 g, 10.0 mmol, 1.0 equiv.), 2,6-lutidine (3.5 ml, 30.0 mmol, 3.0 equiv.) and anhydrous CH2Cl2 (100 ml, to make a final concentration of silyloxydiene alcohol of 0.1 M) at ambient temperature (23 °C). With vigorous stirring, 11 (1.05 ml, 13.0 mmol, 1.3 equiv.) in CH2Cl2 (9.0 ml) was added over a period of 5 h by syringe pump addition. After stirring for an additional 13 h, the reaction mixture was filtered through a short pad of SiO2 and the filtrate was concentrated by rotary evaporation. Purification by automated flash chromatography (5 → 50% EtOAc/hexanes) afforded bicyclic γ-lactones (–)-15a (1.55 g, 58% yield, 94% e.e.) and (+)-15b (0.51 g, 19% yield, 94% e.e.). See Supplementary information section A for full characterization of 15a and 15b.

Surface activation, ligand immobilization and binding

A Biacore X SPR biosensor with Control Software and BIAevaluation software version 3.0 from Biacore (GE Healthcare) was used to check the binding between gracilin A derivatives and CypA. Sensor surface activation and ligand immobilization were performed by using Hank’s balance solution surfactant P20 (HBS-EP) as running buffer at a flow rate of 5 μl min−1 and 25 °C. CM5 sensor chips were used as the surface where Cyp A was immobilized as the ligand. The CM5 chip was a glass slide coated with a thin layer of gold with a matrix of carboxymethylated dextran covalently attached. The CM5 chip was activated using an amine coupling kit. Following the manufacturer’s instructions, a mixture (1:1 vol/vol) of 1-ethyl-3- (3-dimetylaminopropyl) carbodiimide hydrochloride (EDC) and N-hydroxysuccinimide (NHS) was applied for 2 min over the sensor chip. After activation, the ligand, 100 μg ml−1 of active human CypA protein dissolved in sodium acetate 10 mM at pH 4.5, was added to be immobilized over a CM5 sensor chip. Finally, ethanolamine-HCl was injected to deactivate the remaining active esters and to avoid non-specific binding. Analytes (CsA as positive control or synthetic compounds) were added to evaluate the binding with CypA. Once analytes were tested and interaction was observed, individual binding curves were analysed by determining the kinetic constants of the analytes–CypA binding, namely, the observed rate constant (Kobs), the association rate constant (Kass), the dissociation rate constant (Kdiss) and the kinetic equilibrium dissociation constant (KD). At equilibrium, by definition, Kdiss/Kass = KD. The pseudo-first-order association rate constants Kobs (s−1) were determined for each compound concentration by using the 1:1 Langmuir association model of BiaEvaluation software (BiaCore). A representation of Kobs against the corresponding concentration of each compound was then carried out. These plots follow a linear correlation coefficient. From the equation of these representations, Kass (M−1 s−1), the gradient of the plot, and Kdiss (s−1), the intercept of the plot, were obtained. From these two values, the kinetic equilibrium dissociation constant KD for each analyte–CypA binding was obtained.

The duration of the sample injection was 2 min at 10 μl min−1 flow rate. Next, dissociation of bound molecules in HBS-EP buffer flow was studied. The bound drugs were removed from the chip surface before the next injection by adding 1 M glycine-HCl at pH 2.5 for 1 min. The association phase was used to quantify the compound–CypA interactions. All experiments were performed four times.

Determination of intracellular ROS levels

Intracellular levels of ROS were determined with carboxy-H2DCFDA (5-(and-6)-carboxy-2′,7′-dichlorodihydrofluorescein diacetate). This dye diffuses through the cellular membrane and is converted by cellular esterases to carboxy-H2DCFH (non-fluorescent). When carboxy-H2DCFH is oxidized by ROS, it becomes fluorescent53. Cells were seeded at a density of 2.5 × 105 cells ml−1 in 96-well plates and allowed to attach for 24 h. Following treatment with compounds at various concentrations (1 µM → 1 nM) and 150 µM H2O2 for 6 h, SH-SY5Y cells were washed twice with serum-free culture medium. Carboxy-H2DCFDA 20 µM dissolved in serum-free culture medium was then added to each well and the cells were incubated for 1 h at 37 °C. After this incubation, the medium with the fluorescent dye was replaced with PBS and the plate was incubated for 30 min at 37 °C. Fluorescence was read at 527 nm, with an excitation wavelength of 495 nm. All experiments were performed four times.

Mitochondrial membrane permeability transition pore measurement

Blockage of the mitochondrial membrane permeability transition pore (mPTP) by compounds was determined with the MitoProbe Transition Assay Kit following the manufacturer’s instructions. Briefly, SH-SY5Y human neuroblastoma cells were resuspended in pre-warmed PBS/Ca2+ buffer at a final concentration of 1 × 106 cells ml−1. Cells were loaded with 0.01 µM calcein-AM and incubated at 37 °C for 15 min. Then, 0.4 mM CoCl2 and compounds at selected concentrations were added and incubated for 15 min at 37 °C. CsA at 0.2 µM was used as positive control. After this incubation, cells were centrifuged and resuspended in 100 µl of PBS. Just before analysis, 1 mM tert-butyl hydroperoxide was added to the samples to induce pore opening. Fluorescence intensity was measured at 488 nm excitation and 517 nm emission wavelengths by flow cytometry using ImageStreamMKII (Amnis Corporation, Merck-Millipore) and INSPIRE software. The fluorescence of 10,000 events was analysed with IDEAS Application version 6.0 (Amnis Corporation, Merck-Millipore). Experiments were carried out four times.

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.