Total Synthesis of (-)-Luminacin D.

A second-generation synthesis of (-)-luminacin D based on an early stage introduction of the trisubstituted epoxide group is reported, allowing access to the natural product in an improved yield and a reduced number of steps (5.4%, 17 steps vs 2.6%, 19 steps). A full account of the optimization work is provided, with the reversal of stereoselection in the formation of the C4 alcohol in equally excellent diastereoselectivity as the key improvement.


Introduction
Angiogenesis is defined as the formation of new blood vessels from the pre-existing vascular network. 1 Through its involvement in numerous pathologies, including tumor growth and metastasis, angiogenesis and its associated regulation mechanisms have emerged as promising targets in drug discovery. In particular, remarkable efforts have been directed towards the identification of angiogenic modulators among the natural products. 2,3 The luminacin family of natural products, originally isolated from bacterial fermentation, contains numerous members that have been shown to exhibit potent antiangiogenic activity in several assays. Wakabayashi et al. notably demonstrated that luminacins operate by blocking the initial stages of the capillary tube formation in vitro, with luminacin D 1a (Chart 1) being the most active among the 12 members tested. 4 Later on, additional in vivo studies using luminacin C2 1b revealed that this molecule effectively inhibited the phosphorylation activity of Src tyrosine kinases, and was found to exert its unique mode of action by disrupting Src mediated protein-protein interactions. 5,6 Src tyrosine kinases play key roles in the regulation of numerous processes associated to angiogenesis, including growth, differentiation, migration and survival. 7 In addition, luminacin C2 was also found to inhibit breast cancer cell invasion and metastasis in vitro by disrupting the AMAP1-cortactin binding (proteinprotein interactions). 8 The recent isolation of two cancer cell migration inhibitors of similar structure (migracins A and B, 1c, 1d), highlighted once more the therapeutic potential of these molecules. 9

Chart 1. Structure of luminacins
Despite its promising anti-angiogenic activity as revealed by the original work of Wakabayashi, luminacin D has been less extensively studied in comparison with some other members of its family, and little information can be found regarding its mode of action and biological functions. To obtain further material to enable further biological investigations, chemical synthesis is the most efficient way given the modest yield from extraction (and the fact that a new extraction campaign would be required).
Apart from our recent contribution, 10 so far there have been four reported syntheses of luminacin derivatives, 11,12,13,14,15 each presenting shortcomings in term of length or selectivity. In particular, the efficiency of three from these four syntheses was dramatically compromised by the low or undesired stereoselectivity associated with the epoxidation step which, in addition, in each case took place at a late stage of the synthesis. In this context, we achieved a highly diastereoselective synthesis of (-)luminacin D in 19 steps. 10 As shown in Scheme 1, our synthetic approach relied on the stereoselective introduction of the epoxide moiety at an early stage of the synthesis starting from the enantiopure sulfoxide 5, and subsequently to utilize the chirality of the epoxide group in 4 for the diastereoselective completion of the aliphatic fragment. This was achieved via a chelation-controlled allylation procedure of the enantiopure α-epoxy aldehyde 4a, which proceeded in excellent yield and diastereoselectivity.
Unfortunately, the reaction led to the formation of the undesired diastereoisomer 6, and thus an inversion of the obtained alcohol stereocentre was required to complete the synthesis. As further shown in the retrosynthetic analysis, the formation of Luminacin D 1a was realized via arylation of the fully functionalized fragment 2, whose construction was envisaged via spontaneous hemiacetal formation and syn-aldol reaction from the key compound 3. A full account of the different approaches for the formation of the cyclic hemiacetal moiety, and further optimizations of several other steps are disclosed here. In particular, this includes our efforts towards the development of methodology that resulted in direct access to the key intermediate 3 from 4.

Synthesis of the epoxide precursor (ester-sulfoxide 9)
Starting from the α-sulfoxy-esters 5, we initially investigated the one-pot Knoevenagel procedure described by Tanikaga et al. 16 in order to access to the desired (E)-alkenes 8 Tol and (±)-8 Ph . This method proved unsuccessful when applied to our substrates (recovery of starting material). Hence, as described in our previous communication, 10 the formation of racemic and enantiopure α,β-unsaturated (E)alkenes (±)-8 Ph and 8 Tol was then accomplished in 2 steps from the corresponding β-sulfoxy-ester, as shown in Scheme 2. At first, following a known procedure, 17 an aldol-type condensation of 5 with propanal led to the β-hydroxy ester 7 as an impure mixture of diastereoisomers. It was found that treatment of this mixture with MsCl in pyridine afforded alkenes 8 in excellent yield and stereoselectivity. Further to Tanikaga's stereochemical assignment by chemical shift differences, the E configuration of 8 is now further confirmed by NOE analysis (see SI).

Scheme 2. Synthesis of (E)-alkenes 8
The subsequent epoxidation step had been achieved in a diastereoselective manner in our previous synthesis, using a procedure that was adapted from De La Pradilla's vinyl sulfoxide methodology (Table 1, entries 1 and 2). 18 The reaction proceeded in excellent yield and diastereoselectivity with the phenyl derivative 8 Ph (90%, dr 94:6), while the same reaction conditions applied with tolyl derivative 8 Tol led to lower yield and diastereoselectivity (77%, dr 88:12). In addition, the product 12 was obtained in 19% yield as mixture of diastereoisomers ( Table 1). The latter was thought to arise from the nucleophilic attack of n-BuLi onto the Michael intermediate 11, since an excess of n-BuLi was used compared to t-BuOOH (5 vs 4 equiv., respectively).
We then decided to investigate modified conditions for the epoxidation reaction. The first experiment was carried out with 8 Tol by using an excess of t-BuOOH compared to n-BuLi (Entry 3). Although the reaction proceeded without any formation of 12, the formation of undesired by-products could be observed by 1 H NMR, alongside with the expected trans-epoxides syn-9 Tol and anti-9 Tol . After column chromatography, the epoxides were isolated as a mixture of diastereoisomers in moderate yield (60%, dr syn-9 Tol /anti-9 Tol 95:5). A mixture of two unexpected products was also isolated in 16% yield, which allowed their assignment as the cis-epoxide isomers syn-10 Tol and anti-10 Tol . Following this, it was found that using a 1:1 ratio of t-BuOOH and n-BuLi, and reducing the reaction time allowed to minimize the formation of the cis-epoxides 10 Tol (Entry 4). The trans-epoxide 9 Tol was isolated in both excellent yield and diastereselectivity in these conditions (82%, dr syn-9 Tol /anti-9 Tol 91:9). Interestingly, the replacement of n-BuLi by NaH as base with the racemic derivative (±)-8 ph resulted in promoting the formation of cis-isomers 10 ph , with a good selectivity towards the syn-epoxide (±)-syn-10 ph (Entry 5).
The same outcome was observed when an excess of NaH compared to t-BuOOH was used with the tolyl derivative 8 Tol (Entry 6). The epoxidation reaction was carried out on 3 g scale (10 mmol) with the tolyl derivative 8 Tol using the optimised conditions, and enabled isolation of the expected transepoxides 9 Tol in a slightly improved yield and diastereoselectivity compared to our earlier procedure (Entry 7, 82%, dr syn/anti 92:8 vs 77% dr syn/anti 88:12). A minor quantity of the cis-epoxides 10 Tol was also obtained after separation (<2% yield). The assignment of configuration of all the epoxide stereomers was achieved by a combination of X-ray crystallographic analysis and a chemical correlation experiment. The configuration of the crystalline C3 ("pseudo"-) epimers syn-9 Tol and (±)-syn-10 Ph was established by X-ray analysis (see supporting information), as the syn-isomers for both 9 and 10 crystallized as pure diastereomers. The stereochemical relationship between the syn-and anti-epoxides was established by the oxidation (Scheme 3) of a mixture of isomers syn-10 Tol and anti-10 Tol (dr ~1:1), which led to a single sulfone 13 (as observed by 1 H NMR), which allowed unambiguous assignment of anti-10 Tol as the cis-antiepoxide (and by inference, also that of anti-10 Ph ).

Synthesis of the intermediate 3: the diastereoselective reduction approach
As already mentioned, we previously reported the development of chelation-controlled allylation methodology, which, when applied to aldehydes possessing a α-oxygenated center, proceeded with excellent diastereoselectivity. 10 The selectivity outcome was found consistent with the formation of a Furthermore, a number of methodologies for the metal-mediated diastereoselective reduction of β-keto esters, β-hydroxy ketones and α--epoxy ketones have been described, leading in general to excellent facial selectivity. 20,21,22,23,24,25,26

Scheme 4. Proposed reduction approach
In order to simplify the optimization studies, we first focused on the synthesis of the β-propyl keto ester 4c, whose formation was envisaged via acylation reaction of the sulfoxide 9 Tol (Table 2). This was achieved in moderate yield, via treatment of 9 Tol with t-BuLi and subsequent trapping of the resulting oxiranyl anion with methyl butyrate, under Barbier conditions. As these reactions were carried out on the 92:8 syn/anti mixture, an 84% product enantiopurity was obtained. Unfortunately, the selective crystallization procedure of 9 as explained above was only achieved after carrying out the experiments given in Table 2, but would give access to enantiopure material. As shown in table 2, several trials involving a Lewis acid to induce chelation-control during the reduction reaction were undertaken. 19 As a first experiment, treatment of 4c with NaBH 4 and MgBr 2 , in a mixture of THF/DCM gave no expected product. Instead, these conditions resulted in the formation of the bromohydrin 17 as major product (48% isolated yield), alongside with the reduced bromohydrin 18 as a mixture of diasteroisomers. The anti-product 18a was isolated in 9% yield. The epoxide opening issue was overcome by performing the reaction at 0 °C in MeOH, leading to the exclusive formation of products 16. To our surprise, the undesired anti-diastereoisomer 16a was obtained as major product (dr 16a/16b 71:29, Entry 2), which is not consistent with reaction via the transition state 14 (cf. Scheme 4). Replacing MgBr 2 by CaCl 2 as chelating metal 21,22 resulted in a similar outcome, with 15a obtained in good isolated yield and excellent diastereoselectivity (70%, dr 16a/16b 97:3, Entry 3). Following this, the use of Et 3 SiH or L-selectride as reducing agents with MgBr 2 was also attempted at -78 °C, though both conditions led to the exclusive formation of the bromohydrin 17 (Entry 4 and 5). Since the involvement of MgBr 2 /CaCl 2 led to undesired diastereoselectivity or unexpected reactivity, the reduction of the ketone 4c was attempted using L-selectride only (Entry 6). This time, the reaction proceeded in good yield and excellent diastereoselectivity towards the desired syn-product 16b (Entry 6, 90%, dr 16a/16b 1:9). Interestingly, employing the more hindered LS-selectride led to a drop of conversion and selectivity.  As shown in Figure 1, the selectivity observed when NaBH 4 /CaCl 2 and MgBr 2 were used could be explained by the 1,2-chelated transition state 19, assuming that the metal salt catalyzes the formation formation of alkoxyborohydrides NaBH 4 -n(OMe)n in MeOH. 26 The coordination between a Ca 2+ and the methoxy group of the borohydride species would therefore direct the hydride attack to the Re-face, leading to the anti-compound 16a. On the other hand, the models 20 and 21 are consistent with the selectivity observed when L or LS-selectride are employed, assuming that the Li cation is able to chelate between the carbonyl groups (model 20) or between the carbonyl group and the epoxide (model 21). Hydride attack from the least hindered Si-face in both cases would lead to the observed formation of the syn-compound 16b.

Figure 1. Possible rationalization of the selectivity outcome
The relative configuration of 16a and 16b was assigned by NMR comparison with the anti-alcohol, which was obtained after reduction of the double bond of previously synthesized 3 (Scheme 5). The regioselectivity of bromide mediated epoxide opening on 4c, and the relative configuration of the resulting 18a, were determined thanks to X-ray crystallographic analysis (See SI).

Scheme 5. Hydrogenation of 3 to allow assignment of the relative stereochemistry
Motivated by these results, the acylation/diastereoselective reduction procedure was then applied towards the luminacin D synthesis, using methyl but-3-eneoate 22 27 and L-selectride (Scheme 6). Since the intermediate 4b proved unstable to purification on silica gel (with double bond isomerization occurring during silica gel chromatography, not shown), the reduction reaction was attempted on the crude material, immediately after work-up. A first experiment was conducted on small scale with the racemic epoxide (±)-9 Ph and L-selectride as reducing agent. The syn-α-epoxy alcohol (±)-3 was obtained as major product in an encouraging yield (19 % over 2 steps), together with a minor quantity of the anti-diastereoisomer (±)-7 (1% over 2 steps, separation achieved by column chromatography).
Unfortunately, the reaction proved less efficient on 1 g scale, resulting in a drop of yield (14% for (±)-3 over 2 steps). Several parameters, including the volatility of intermediate 4b and the purification issues induced by the formation of numerous by-products over the 2 steps, made the process cumbersome.

Synthesis of the intermediate 3: the allylation approach
Given the moderate yield obtained with the previous approach, the original strategy involving an allylation reaction was reconsidered, with the aim of developing new conditions allowing access to the opposite selectivity outcome compared to the MgBr 2 -promoted allylation procedure. Given the unexpected stereochemical outcome of the reduction process using CaCl 2 as explained above, this additive was now used in a reinvestigation of the allylation of 4a. Hence, the aldehyde 4a (and (±)-4a) was re-synthesized through formylation of the epoxide precursors 9, applying similar conditions as used for the acylation procedure (Table 3). Pleasingly, the reaction proceeded in an improved yield compared to our previous procedure, 10 and is generally more efficient as it can be conducted at -78 °C (instead of -120 °C) without the need of CeCl 3 , which had to be dried under vacuum prior to the reaction and made the work up difficult.
We then examined the use of a modified procedure for the allylation reaction ( Table 3). The conditions of the reported procedure (Entry 1), but with CaCl 2 instead of MgBr 2 , were investigated first (Entry 2).
Despite the poor conversion obtained, we were pleased to notice that only the desired syndiastereoisomer 3 was formed during the reaction, as observed by 1 H NMR of the reaction mixture before chromatography. Increasing the temperature, concentration and reaction time resulted in a better conversion, with 3 obtained in a very good diastereoselectivity (Entry 3, dr 3/7 92:8). Based on these results, it was envisaged that CaCl 2 might not be involved in a chelated transition state, but would only act as a weak activator of the reaction. To confirm this hypothesis, investigations were directed towards the use of non-chelating conditions for the allylation reaction. A first experiment involving the reaction of 4a with allyltrimethylsilane and a sub-stoichiometric amount of TBAF led to the recovery of the starting material (Entry 4). 28 However, the allylation of 4a occurred using the more reactive pinacolyl allylboronate 23 in DCM, by raising the temperature from -78 °C to rt overnight (entry 5). 29 As predicted, the non-chelation control promoted the formation of the desired syn-diastereoisomer 3, in an excellent diastereoselectivity and isolated yield. This result mirrors the work of Mulzer and Prantz, who recently demonstrated that the selectivity of the allylation of 2,2-dialkyl-3-oxopropionates could be reverted by switching from chelation (TiCl 4 ) to non-chelation (BF 3 •OEt 2 ) mediated allylation. 30 It should be noted that both these Lewis acids are not compatible with the epoxide-containing substrate 4.
The optimised two-steps procedure was then carried out on 1.5 g (5 mmol) scale of sulfoxide 9 Tol (dr 92:8)(Entry 6). The slow addition of t-BuLi to the mixture via syringe pump over a period of 1 h was found to give the best results for the formylation reaction. After column chromatography, the aldehyde 4a was obtained in a mixture with minor impurities. Subsequent treatment with the pinacolyl allylboronate 23 using the optimised conditions enabled isolation of the syn-alcohol 3 as major product in 33 % yield over 2 steps, together with the minor anti-diastereoisomer 7, isolated in 1% yield.
Although an accurate dr determination was not possible by 1 H NMR due to the presence of impurities, the ratio of isolated yields of 7 and 3 is consistent with that observed on small scale. Similar results were obtained when the racemic phenyl epoxide (±)-9 Ph was used as starting material (Entry 7). In the context of the luminacin D synthesis, this new procedure represents a significant improvement compared to the previous route reported by our laboratory, which required two extra steps for the formation of 3, in a lower overall yield (24% over 4 steps). The excellent substrate control of this allylation reaction under non-chelating conditions can be rationalized ( Figure 2) by invoking the classic Cornforth-Evans (24) or polar Felkin-Anh (25) models, assuming that the C-O bond of the epoxide acts as the "polar substituent" in preference to the ester.

Completion of the aliphatic fragment: aldol reaction and attempted lactonization.
With access to the pure intermediate 3 (and (±)-3), the synthesis was pursued towards the formation of aldehyde 26 (and (±)-26), which was accomplished in two steps, following the reported procedure (Scheme 7). The β-chiral silyl ether center on 26 offered the possibility for remote stereocontrol, which had been exploited in the luminacin D synthesis by Shipman et al. 15 However, the use of a titanium enolate derived from an aromatic ketone (already containing the luminacin D aliphatic moiety) only led to modest stereocontrol (dr ~2:1, in favor of the desired isomer). Interestingly, while this type of remote stereocontrol has been mainly investigated for Mukaiyama aldol reactions, 31 we found no related investigations of the extent of remote stereocontrol for aldol reactions involving classic N-acyl oxazolidinone boron enolate reagents. Hence, at this juncture, it was decided to investigate this process using simplified model compounds in order to evaluate its potential usefulness in the luminacin D synthesis (Table 4). Aldehydes (±)-27 32 and (±)-28 32 were prepared according to standard procedures and subjected to aldol reactions with the boron enolate of 29. For the reaction between the ethyl oxazolidinone 28a and (±)-26, a low stereocontrol was obtained (Entry 1). As predicted from the Evans model, the major isomer contained the desired relative stereochemistry for our purposes (see SI for the determination of the product relative stereochemistry). Increasing the size of the protecting group (as in (±)-28) led to a slight increase of the desired selectivity (Entry 2). A further increase of the steric bulk by using 29b, the reagent required for the luminacin D synthesis, did give a reasonable 5:1 ratio (Entry 3). With this level of selectivity obtained, this diastereoselective aldol reaction was then performed on the racemic natural product intermediate (±)-32 with a TBDPS protecting group (Scheme 8).

Scheme 8.Translation of the diastereoselective aldol reaction to the natural product system
Given the modest diastereoselectivity favoured the desired stereomer, a matched double diastereodifferentiation process using a chiral oxazolidinone based auxiliary was then investigated.
This approach has also been used in the luminacin D synthesis by Maier et al. 14 Hence, the enantiopure oxazolidinone 35 33 was required (Scheme 9). For atom economy reasons, it was decided to use a TES protecting group as opposed to a TBDPS group. Initially, the racemic aldehyde (±)-26 was engaged in Evans-aldol reaction with the acyl chiral oxazolidinone, which led to the formation of two (among the four possible) aldol adducts ( 1 H NMR analysis) in a 1:1 dr. The two isomers could be separated by preparative HPLC after TES protection of the formed alcohol, allowing isolation of the expected aldol product 38 10 as well as the isomer 39, the latter resulting from the aldol reaction of the oxazolidinone 35 with the enantiomer of 26, since racemic starting material was employed. Given the low remote stereocontrol exerted by the alcohol chiral centre as shown above, it is thought that the auxiliary dominates the stereoselection, leading to the C2',C3'-syn-C3',C5'-syn diastereoisomer 37. With enantioenriched aldehyde 26 (er 92:8), exclusive formation of the aldol products 36 and 37 in a 91:9 dr was observed. From that mixture, alcohol protection and HPLC separation allowed isolation of 38 and 39 in 86 and 6% yields, respectively. As mentioned above, applying the selective crystallization procedure of 9 would avoid this separation issue, as in this case only aldol product 36 would be formed.

Cyclization of the aliphatic fragment: first approach
It was envisaged that the synthesis of the aliphatic fragment could be completed at this stage by acidcatalyzed t-Bu deprotection, which would initiate lactone formation that then could be reduced to the luminacin D lactol ring. The lactone formation was first investigated using the racemic aldol product 33 was used as model substrate (Scheme 10). To our surprise, heating with CSA in toluene led to a product with the t-Bu ester intact, but in which cyclization towards the epoxide group had occurred, leading to 41 in excellent yield (81%). When TFA in DCM was used, the desired lactone formation did occur, but only 11% of the 43 was isolated. Under these conditions, the same alternative cyclization leading to a tetrahydropyran group occurred, even if the resulting product 44 was isolated as the carboxylic acid. Presumably the slow t-butyl ester deprotection promoted tetrahydropyran over lactone formation, and the COOH deprotection leading to 44 could have occurred after the ring formation.
Assignment of the different cyclisation products was achieved by HMBC and NOE analyses (see supporting information).

Scheme 10. Deprotection and unexpected cyclisation of the aldol product 33
a Isolated in a mixture with 44 (see experimental section) Cyclization under basic conditions was also unsuccessful (Scheme 11). Treatment of the aldol product 33 with sodium hydride resulted in the formation of a product 45 in low yield, in which both elimination and oxazolidinone ring opening had occurred. Interestingly, when 33 was subjected to lithium ethylthiolate (see next section), the same elimination product was obtained in quantitative yield.
A mechanism of formation for this product 45 is proposed: deprotonation of the hydroxyl group initiates cyclization to the carbamate group, expelling the primary alkoxide 48, which could then be involved in carbon dioxide elimination to give 49, possibly via an intramolecular deprotonation pathway as shown. Finally, amide anion protonation, either by reaction with 33, or in the workup, leads to 45. The fact that no elimination/oxazolidinone opening product such as 45 was formed with lithium ethylthiolate when the alcohol group was protected (see next section) is consistent with the proposed mechanism.
Scheme 11. Base catalysed elimination of aldol product 33

Cyclization of the aliphatic fragment: second approach
Given the unsuccessful lactone formation, it was envisaged to postpone this step until after the introduction of the aryl fragment (Scheme 12). Hence, oxazolidinone removal was attempted via thioester formation. At high reagent concentration, the product 52, resulting from oxazolidinone opening with lithium ethyl thiolate was sometimes observed, alongside with the expected thioester 50.
Nevertheless, a fully chemoselective conversion of TES-protected aldol product 38 to the thioester 50 was achieved in excellent yield using dilute [EtSLi] conditions. The subsequent palladium-mediated reduction reaction produced the final aldehyde fragment 51. The yield of the reduction was significantly increased by adding the reagents at 0 °C rather than rt as reported in the previous procedure (96% vs 66-75%).

Completion of the synthesis
With the aliphatic fragment in hand, we pursued our efforts towards the synthesis of the bromoaryl derivatives 55 and 58, as potential substrates for the coupling reaction. As depicted in Scheme 13, these two compounds could be synthesized from the same intermediate 53, 10,34,35 and only differ from the choice of protecting groups. In the first case, O-lithiation of 53 and treatment with BOMCl enabled introduction of the benzyloxy moiety in moderate yield. 10 The obtained 54 was then brominated with NBS to yield the desired bromoaryl 55. For 58, an O-formylation reaction was followed by aldehyde reduction, silylation and finally bromination.

Scheme 13. Synthesis of aromatic fragments
The coupling reaction was then carried out in the presence of t-BuLi and an excess of the bromoaryl derivative (Scheme 14), leading in each case to the desired product 59 as a mixture of benzylic alcohol epimers in excellent yield. Pleasingly, the excess of aromatic compound could be easily recovered by column chromatography as an inseparable mixture of 57 and 58, and treatment with NBS allowed complete recycling of 58. The mixture of epimers 59a and 59b was then subjected to DIBAL-H reduction in order to convert the t-butyl ester to the corresponding aldehydes 60a-b (Scheme 14).
Surprisingly, the minor benzylic alcohol epimer was found to be unreactive towards reduction, and aldehydes 60a and 60b were obtained as a single diastereoisomer, together with the remaining isomerically pure starting material 59a-b (the alcohol configuration at C1' could not be determined).
Aldehyde 60b could separated from 59b by preparative HPLC, and was subsequently converted to the hemiacetal 61b after treatment with TBAF and spontaneous cyclisation. In the case of 60a, separation from its starting material was not possible, and the TBAF treatment was thus applied to the mixture.
This led to the formation of the desired hemiacetal derivative 61a, together with the residual starting material 62a, with separation now achieved by column chromatography.

Scheme 14. Formation of hemiacetals 61
Assuming that the lack of reactivity observed for the minor epimer 59a (and 59b) was due to conformational restrictions imposed by the alcohol configuration at C1', a sequential oxidation/reduction process towards the formation of 60a was attempted (Scheme 26). Thus, the benzylic alcohol was oxidised using Dess-Martin periodinane (DMP) in 73% yield, and the resulting ketone 63 was then treated with an excess of DIBAL-H. Although the benzylic ketone in C1' was effectively reduced, only trace amount of the aldehyde 60a could be observed by NMR. Instead, the compound 59a was obtained as a single epimer, whose configuration unfortunately corresponds to that of the previously observed unreactive isomer. Following this, no further investigation was attempted on this sequence, and the synthesis was pursued on the major epimer 61a.

Scheme 15. Attempted sequential oxidation/reduction process
Completion of the luminacin D synthesis was achieved in 2 further steps from the intermediate 61a (Scheme 16). At first, the treatment of 61a using DMP in the presence of NaHCO 3 enabled oxidation of the benzylic alcohols to give 64 in moderate yield. The oxidation step proved cumbersome, with the best yield (56%) obtained after termination of the reaction prior to completion (5 min), separation of the product from the starting material, and re-subjecting the remaining starting material to DMP. A longer reaction time (10 min or 1.5 h) led to a drop in yield (43% in each case). Finally, subsequent deprotection provided (-)-luminacin D 1a in 92% yield after column chromatography, and in 80% after HPLC purification.

Scheme 16: Completion of the synthesis from the first protecting group strategy
The final sequence was then investigated with the tri-benzylated 61b, as simultaneous deprotection of the benzyl ethers would enable to complete the synthesis with only bis-benzylic oxidation left to do (Scheme 17). However, the hydrogenolysis attempts were associated with numerous selectivity issues, and 65 was never obtained in a meaningful yield. It was found that the primary benzylic alcohol could easily be fully reduced to a methyl group, while the secondary benzyl alcohol was also found to be labile.

Scheme 17. Attempted hydrogenolysis of the tribenzylated 61b
In view of these unexpected results, deprotection conditions were investigated on a simple model substrate 66 (Scheme 18), resulting from the coupling reaction between 55 and propionaldehyde (not shown). It was envisioned that DDQ oxidation of the electron rich aromatic ring, similar to p-methoxy benzyl cleavage, would directly lead to the corresponding C1 aldehyde 68, alongside with BnOH. 36,37 However, despite considerable experimentation, this was not achieved. Surprisingly, this process did yield the ketone 67, which, though potentially useful for our purposes, was judged too low-yielding for application on the luminacin D system. Hence, the hydrogenolysis approach was reinvestigated, using the same model system.

Scheme 18. DDQ promoted debenzylation/oxidation of the secondary benzyl alcohol
Given its perceived instability, the secondary benzylic alcohol group was first oxidized to the ketone 69 (Scheme 19). Manganese dioxide was found ineffective at this transformation on small scale. The full debenzylation was now achieved under acidic conditions previously as used by Tatsuda 11 to give the triol 70 in excellent yield. In this reaction, control of the reaction time was required, as over-reduction to 71 occurred with longer reaction times, a side-reaction not reported by Tatsuda. 11

Scheme 19. Oxidation/reduction sequence
Finally, these successful reactions were applied to 61b (Scheme 20). Pleasingly, the initial oxidation to ketone 72 proceeded in quantitative yield, as did the subsequent debenzylation reaction to triol 73.
Luminacin D 1a was then obtained by a second Dess-Martin oxidation.
Scheme 20. Completion of the synthesis from the second protecting group strategy

Conclusions
A successful second generation synthesis of enantiopure (-)-luminacin D is reported in full. The synthetic strategy relies on a conventional key disconnection to give an aromatic and aliphatic fragment. The synthesis of the chiral aliphatic fragment relies on the diastereoselective introduction of the trisubstituted epoxide subunit, which is achieved by modified de la Pradilla sulfoxide methodology, with the sulfoxide then becoming a reactive handle for introduction of a formyl group. A key step is the subsequent diastereoselective allylation of this formyl group. Initial methodology relying on chelation control achieved this allylation in very high diastereoselectivity, but with the wrong relative stereochemistry. Subsequently, different allylation conditions under non-chelation control were found that achieved this process with the correct relative stereochemistry, in equally excellent de. As a complementary approach, we also showed that high levels of diastereoselectivity could be achieved through the reduction of β-keto ester containing an α-quartenary epoxide center, although this approach was hampered by the low-yielding acylation reaction of the sulfoxide derivative.
Completion of the aliphatic fragment was achieved by aldol reaction involving acyl-oxazolidinones. A first approach solely relying on remote stereocontrol induced by a β-OSiR 3 center was moderately successful (4:1 de), but the diastereoselection could be amplified by the use of a 'matched' chiral oxazolidinone. Installation of the cyclic hemiacetal group proved not possible at this stage, but was achieved after coupling with the aromatic fragment. Elaborate final deprotection investigations using two different protecting groups for the primary benzylic alcohol were required to arrive at a successful luminacin D synthesis. In spite of the extra oxidation step required to achieve the synthesis, the second aromatic protecting strategy described was found more satisfactory in term of yield than the first route described (40% over 6 steps vs 22% yield over 5 steps for the first route). The successful enantioselective formation of the trisubstituted epoxide and the diastereoselective installation of an adjacent chiral alcohol group will be of general applicability. To the best of our knowledge, remote stereocontrol by a β-OSiR 3 center of an achiral oxazolidinone based boron enol ether mediated aldol reaction had not been described before. Overall, this second generation synthesis enabled access to the natural product in an improved yield and a reduced number of steps compared to our previous approach (5.4%, 17 steps vs 2.6%, 19 steps).
Organic phases were combined, dried over Na 2 SO 4 and concentrated in vacuo to give the crude α-  , which was used in the next step without further purification.
The mixture was dissolved in DCM (8 mL) and cooled down at -78 °C, after which allylboronic acid pinacol ester (475 µL, 2.53 mmol, 0.5 equiv.) was added dropwise. The reaction was then allowed to warm up for 16 h (without removing the dry ice bath, T ~ 15 °C after 16 h). The mixture was then quenched at rt with H 2 O (8 mL), and stirring was continued for 5 min. The layers were separated, and the aqueous phase was extracted with Et 2 O (3×20 mL). Organic phases were combined, dried over Na 2 SO 4 and concentrated in vacuo to give the crude α-epoxy alcohol 3 and 7 as a mixture of diastereoisomers (dr n.d due to complexity of the crude mixture, see copy of 1 H NMR spectrum in SI).
The reaction mixture was stirred for 1.5 h at this temperature, before quenching with H 2 O (10 mL), followed by dropwise addition of HCl (0.5 M, 5 mL). The mixture was diluted with Et 2 O (10 mL) and the phases were separated. The aqueous phase was re-extracted with Et 2 O (2x25 mL) and the combined organic phases were washed with a saturated solution of NH 4 Cl (20 mL). The combined organic phases were dried over Na 2 SO 4 and concentrated in vacuo, to give the corresponding alcohol as a pale oil which was used without further purification.
The mixture was filtered through a pad of celite ® , washed with EtOAc (24 mL). The layers were separated, and the aqueous phase was extracted with EtOAc (5 mL). The combined organic layers were dried over anhydrous Na 2 SO 4 and concentrated under reduced pressure. Purification via column chromatography (pentane/Et 2 O 95:5 to 9:1) followed by preparative HPLC (hexane/Et 2 O 9:1) gave a mixture of aldehyde 60a and starting material 59a (86 mg), which was used in the next step without further purification.
The mixture (86 mg) was then dissolved in THF (3 mL), and TBAF (1M in THF, 520 µL, 0.52 mmol, 5.2 equiv.) was added dropwise at 0 °C. The resulting solution was stirred for 1 h at 0 °C, then the mixture was allowed to warm up to rt, and stirring was continued for 2.5 h at this temperature, before evaporating under reduced pressure. Purification via column chromatography (pentane/acetone 8:2 to 7:3) gave the hemiacetal 61a as a single epimer and as a colourless oil (35 mg, isolated with 5% of 62a, 54% over 2 steps), as well as an impure mixture of deprotected ester 62a, which was repurified by * To whom correspondence should be addressed. Fax: +44 23 8059 7574; e-mail: bruno.linclau@soton.ac.uk