There has been a precipitous decrease in new molecular entity (NME) approved drugs by the US Food and Drug Administration (FDA) over the last 20 years. For example, the count of 45 agents approved in 1990 decreased to 21 in 2010.^{1, 7} In our view, there appears to be a positive correlation with the diminished focus on natural products as sources of new therapeutic leads and the drop in the number of NME approved drugs. Whether these trends are causal or coincidental is open to debate, but few would disagree on the significant role of natural products in providing sources and inspiration for new therapeutic leads.⁸ One bright spot amidst this controversy is that interest in natural products-based discovery programs in the developing world has increased dramatically since the adoption of the Convention on Biological Diversity (CBD) in 1993.⁶ As another development, several companies engaged in natural products research including Sequoia Sciences,⁹ AnalytiCon Discovery,¹⁰ and Wyeth¹¹ plus a small number of academic groups have published first generation results showing that high throughput (HT) HPLC purification methods can be interfaced with modern HTS bioassays.^12–15 Surprisingly, only a few studies of this type from academic groups have taken the next step involving the use of such HT approaches culminating in the disclosure of compounds with completely new biological activities or molecular structures.^13–17 As one exception, the UC Santa Cruz consortium has recently revealed that combining HT HPLC methods with a HT Yeast Halo assay successfully pinpointed the unreported anti-fungal bioactivity of crambescidin 800.¹⁴ We have now further optimized this strategy to identify lead compounds through an approach that incorporates systematic LC-MS-UV-ELSD analysis. The impetus for these changes stemmed from partnerships developed in a multi-disciplinary campaign as part of a natural products-based International Cooperative Biodiversity Group (ICBG)⁶ initiative. We have formed a powerful alliance, which involves contributions from Indonesian professionals working alongside investigators from four University of California campuses.

A key tool introduced to guide our ICBG programs consists of a refined HT screening paradigm. The goal is to accelerate identifying compounds with unreported bioactivity and or new structures and it involves the four step process outlined in Scheme 1. 1) Raw materials, which can vary from marine sponges, tropical plants, or culture broths from microorganisms, are pre-fractionated using traditional methods¹⁸ or with our previously described HT approach of accelerated solvent extraction (ASE)¹⁹ which reduces the extraction cycle times from hours/days to minutes. This creates semi-crude extract assortments (SCEAs), and only the MeOH plant extracts are further pre-treated in the first step using solid phase extraction (SPE) cartridges, to remove polyphenols, which can act as false bioassay positives.¹² 2) The SCEAs are evaluated in a panel of HT bioassays²⁰ involving cytoskeleton activity,²¹ immune modulation,¹⁸ parasites,²² tumor cell lines^{18, 23} and opioid receptors.²⁴ 3) Prioritized active extracts are then selected for LC-MS-UV-ELSD library creation into 96 well plates with subsequent follow up HT bioassay evaluation. 4) When fractions (containing potential lead compounds) that either exceed the potency of the SCEAs or possess new m/z values are identified, further processing is immediately initiated. This involves using the same column and LC-MS-UV-ELSD conditions of step 3 to scale up into 20 mL vials (or 50 mL test tubes), followed by evaporation, NMR, dereplication and or structure elucidation. This protocol addresses several key HT parameters^{11, 25} including: a) modest expense accompanied by an easy setup and implementation, b) minimal volumes of solvent are required to process ASE extracts,¹⁹ 96 well plate, or 20 mL vial scale up fractions, c) nuisance substances (e.g. salts or polyphenols) that can interfere with bioassays are effectively removed, d) the rapid processing provides high quality purification followed by concentration of lead compounds into a library for direct HTS bioassay, and e) frontline acquisition LC-MS m/z ion, UV and ELSD data analysis serves to jumpstart dereplication and structure elucidation efforts.

We implemented the process outlined in Scheme 1 to test the hypothesis that using a focused HT strategy would provide rapid identification of lead compounds with important activity properties and the discovery of new compounds from even well studied species. The proof of concept trial involved investigation of individual specimens of the Indo Pacific sponge Cacospongia mycofijiensis. Extracts of this sponge have been the source of seven major structural classes having various forms of bioactivity, some with unique mechanisms of action. The structures include: the latrunculins (anti-tumor, microfilament disruption),^{26, 27} fijianolides (syn laulimalide, anti-tumor, microtubule stabilizing),^{28, 29} dendrolasen (cytotoxic, target unknown),³⁰ mycothiazole (solid tumor selective, mitochondria complex I),^{31, 32} dactylolide (cytotoxic, microtubule stabilizing),³³ CTP-431 (cytotoxic, target unknown)³⁴ and the aignopsanes (moderate anti-parasitic, mechanism unknown).³⁵ The results of our HT survey of C. mycofijiensis plus additional insights discovered during our brief evaluation using LC-MS-UV-ELSD of other assemblages are discussed below.

The next task was to correlate the activity of the parent extract XFD (Table 1) with that of specific compounds by evaluating the LC-MS and NMR data of selective library fractions. A small subset of fractions were active in the MF and MT or both assays. The entries 2, 4, 7 and 12 in Table 1 displayed MF disrupting effects at 20 µg/mL. The LC-MS m/z data in Table 1 and ¹H NMR (see Figures S6–S9, S15) of these fractions, plus entry 3, all corresponded to the latrunculin A (10) chemotypes (1–4).²⁷ Not surprisingly, entries 8, 10–11 and 14 exhibited microtubule stabilizing effects at 20 µg/mL with m/z values and ¹H NMR data in agreement with the fijianolides (5, 8–9, see Figures S10, S13–S14).²⁸ The ¹H NMR (see Figure S33) and microtubule stabilizing effects of entry 14 (Fraction H57, 513 m/z) was similar but did not match any previously reported fijianolides^{28, 36} and it was set aside for further characterization.

All 17 fractions listed in Table 1 were evaluated against T. brucei, the parasite responsible for Human African trypanosomiasis (HAT).²² Only six of the samples (entries 2, 7, 8, 10, 11, 12) demonstrated good inhibitory activity (~1 µg/mL) and these contained the latrunculins (1, 4, 10)²⁷ and fijianolides (5, 8–9).²⁸ Less potent were five fractions (entries 3, 4, 6, 9, 13) which also showed modest inhibition (< 10 µg/mL), and the active constituents contained the latrunculins (2, 3, 14)²⁷, an aignopsane (6)³⁵ and mycothiazole (11)³¹. The values of the first set are on par with natural products investigated by others³⁷ and are less potent than the clinically used but broadly cytotoxic agent melarsoprol (IC₅₀ = 0.0026 µM).³⁸ The LC-MS m/z ions of the various fractions were used to pinpoint the presence of these compounds and most were confirmed (i.e., see Figures S11, S16). The exceptions to this analysis included the following: the LC-MS m/z ion (404 amu) and ¹H NMR data (see Figure S27) of entry 6 resembled latrunculin A (10), but were not an exact match. The LC-MS m/z ion (513 amu) and ¹H NMR data (see Figure S33) of entry 14 also did not match that of previously described compounds. Thus, this material was added to the list of compounds needing additional analysis. Some further noteworthy points pertaining to the more bioactive compounds in this group are as follows. The compound latrunculol A (1) is being pursued as a preclinical anti-tumor candidate³⁹ due to its selective cytotoxicity for colon cancer (C38) vs normal bone marrow (CFU-GM) cell lines.²⁷ Other latrunculin analogs have also recently emerged as potential anti-tumor, anti-microbial, anti-fungal, and anti-protozoal therapeutic leads.^{27, 40–43}

The final analysis involved assaying the library fractions against solid tumor cell lines including melanoma (MDA-MB-435), colon (HT-29), non small cell lung (H522-T1) and lymphoma (U937). Considerable cytotoxicity (< 2.5 µg/mL) was observed for the parent extract, entry 1 as well as for library fractions, entries 2, 7, 10–13. These active constituents corresponded to compounds 1, 4, 8–11 and similar results have been reported previously.^{27, 28, 31, 44} Although 1, 4, 8–10 were roughly equally active in the assays, mycothiazole (11) showed distinct solid tumor selectivity for U937 cells (0.01 µg/mL) versus MDA-MB-435 (1.0 µg/mL), HT-29 (0.7 µg/mL) and H522-T1 (1.0 µg/mL) cell lines. Compound 11 has previously shown selectivity in the NCI panel of 60 cell lines,³¹ but data for U937 has not been reported. These results highlight the utility of using this HT approach to identify lead compounds with unreported selectivity for additional cancer cell lines.

The compositions of four other fractions (entries 5, 15–17) were also investigated based on their unique LC-MS and ¹H NMR data. Entry 5 (Fraction H30, 269 m/z) gave a ¹H NMR spectrum (see Figure S18) similar to the aignopsanes³⁵ but lacked a UV chromophore, and was visible only by ELSD. It was designated for further structure elucidation work. Entry 15 (Fraction H59, 265 m/z) had LC-MS and ¹H NMR data (see Figure S12) identical to methyl aignopsanoate 7.³⁵ Entry 16 (Fraction H61, 249 m/z) was identified as discussed below as the ketal derivative of 6 based on ¹H NMR (see Figure S39). Entry 17 (Fraction H65, 463 m/z) was also only visible by ELSD. Its LC-MS m/z and NMR spectroscopic data (see Figure S17) matched that of sacrotride A (12),⁴⁵ a common bioactive compound from sponges,⁴⁶ but unreported from this species.

At this point it was clear that four new compounds were present in the fractions (entries 5–6, 14, 16), which also exhibited LC-MS m/z, NMR and or bioactivity data not previously reported from C. mycofijiensis. The first to be analyzed was entry 5 (Fraction H30, 269 m/z) which had a metabolite whose molecular formula of C₁₅H₂₄O₄ from HRESITOFMS was based on the [M+Na]⁺ ion m/z 291.15343. This varied from that of aignopsanoic acid A (6)³⁵ by addition of H₂O. These two compounds exhibited parallel ¹H and ¹³C NMR data (see Table S1) with the single distinction that the vinyl resonances were replaced by the two methine signals at δ_H 3.20 and δ_H 3.86, plus that of an OH (δ_H 8.6). The C-11 (δ_C 72.4) sp³ carbon was deduced to be a secondary alcohol from DEPT data and key 2D NMR data in Figure 3 showing diagnostic^2–3 J_H,C HMBC correlations observed from H-11 to the carbonyl carbons C-9 (δ_C 215.9) and C-12 (δ_C 178.8). Additional evidence for the cis decalin chair-chair orientation was derived from the 1–2 trans diaxial couplings (J = 14.4 Hz) observed from H-6_ax to H-7_ax. NOESY data in Figure 3 showed that all three methyl groups plus H-8 were on the same side of a plane. This was consistent with reports for 6, where the absolute configuration at positions C-4, C-5, C-10 was confirmed using CD measurements.³⁵ Based on a biosynthetic analogy to 6, we have named this new compound aignopsanoic acid B (13) and conditionally assign its four stereocenters as 4S, 5R, 8S, 10S with position 11 remaning undefined. It is reasonable to conclude that 13 could serve as an important biogenetic precursor to 6 and vice versa.

Analysis of the compound listed as entry 6 (Fraction H33, 404 m/z) began by noting it displayed diagnostic δ_H, δ_C values (see Table S2) similar to latrunculin A (10).²⁷ However key resonances of the hemiacetal and thiazolidinone moieties were altered. A molecular formula of C₂₂H₃₁NO₅SNa was set from HRESITOFMS data based on the [M + Na]⁺ ion m/z 444.1782, which indicated eight degrees of unsaturation. These matching empirical data suggested a constitutional isomer of 10. The ¹H and ¹³C NMR shifts indicated a polyketide pattern from C-1 to C-10 as in 10 but with a terminal carboxylic acid group and modified thiazolidinone side chain suggesting an acyclic derivative. This was confirmed by significant ¹H-¹H COSY data of H-4 to H-11 and^2–3 J -HMBC correlations of H-2, H₃-21 and H₃-22 shown in Figure 3. Evidence of the free acid was supported by the existence of a broad singlet (δ_H 10.2) and acid carbonyl at C-1 (δ_C 169.6). Key differences observed for the remaining atoms C-11 to C-20 involved: a) the replacement of a methine by a methylene at C-15 (δ_C 19.9), suggesting the opened macrolide ring and b) the substitution of a hemiacetal carbon with an α,β-unsatured carbonyl at C-17 (δ_C 188.7) adjacent to sp² carbons C-18 (δ_C 134.2) and C-19 (δ_C 114.8). These new functionalities, along with the presence of a carbonyl C-20 (δ_C 171.7) accounted for all eight degrees of unsaturation. Placement of the sp² carbons was confined to the 4-thiazolin-2-one ring as evident from^2–3 J -HMBC correlations of H-19 (δ_H 7.10) to C-17, C-18 and C-20. Additional COSY data of H-15 (δ_H 1.74, 1.81) to H-16 (δ_H 2.73. 2.74) along with^2–3 J -HMBC correlations from H-11 to C-13 (δ_C 71.3); H-13 (δ_H 3.61) to C-15, and H-15, H-16 to C-17 secured the working structure presented in Figure 3. Overall this structure bears resemblance to the acyclic latrunculin B derivative latrunculin T⁴³ therefore we have named this latrunculin A (10) derivative apo latrunculin T (14). Latrunculin T has shown superior antifungal activity against Saccharomyces cerevisiae compared to either of the macrolides latrunculin A or B.⁴⁰ Interestingly 14 displayed activity in Table 1 against the parasite T. brucei but did not disrupt MF assembly at 20 µg/mL suggesting a mode of action independent of the actin pathway similar to reports by others.^{27, 42}

The geometries at Δ^2,3 and Δ^8,9 of 14 were designated Z based on the ¹³C chemical shift value for CH₃-21 (δ_C 24.5) and observed J values for H-8 and H-9 (10.8 Hz) while position Δ^6,7 was assigned as E from J- based analysis of H-6 and H-7 (15.0 Hz). These values were in accordance of those reported for 10.²⁷ The orientation of the remaining asymmetric centers was provisionally assigned as 10S, 13S from NOESY data in Figure 3 showing correlations of H-10 (δ_H 2.61) to H-13 suggesting both protons were in the β position. Also from a biosynthetic perspective, all previously reported natural latrunculins retain these configurations (or the equivalent as a function of differing numbering schemes).^{27, 43} Compound 14 could serve as a putative biogenetic precursor to 10 in a similar manner previously described for the production of latrunculin B from latrunculin T.⁴³

The analysis of entry 14 (Fraction H57, 513 m/z) NMR data (see Table S3) showed it was a fijianolide A (9)²⁸ derivative with an OCH₃ (δ_H 3.15, δ_C 49.2) present. This was supported by the molecular formula of C₃₁H₄₄NO₈ derived from HRESITOFMS data based on the [M+Na]⁺ ion m/z ion of 567.29184 indicating an extra carbon, oxygen and two hydrogens. Major NMR shift differences observed for H-17 (δ_H 5.65), H-19 (δ_H 4.61) and C-20 (δ_C 109.0) compared with those of 9,²⁸ suggested the OCH₃ was near the furan ring. This was supported by 2D NMR in Figure 3 which showed a key² J HMBC correlation from the OCH₃ to C-20 followed by^2–3 J -HMBC correlations from H-18 (δ_H 2.60, 2.03), H-19 to C-20 and H-21 (δ_H 5.84) to C-20, C-22 (δ_C 137.5) along with ¹H-¹H COSY data of H-18 to H-19 and H-22 (δ_H 6.35) to H-23 (δ_H 3.90). The ¹³C δ values of the remaining atoms were ≤ 1.0 ppm of those of 9, except for C-2, C-3 and C-13, which were verified with^2–3 J- HMBC correlations. These data secured the overall working structure as 20-methoxy-fijianolide A(15). The configurations at positions Δ^2,3, Δ^6,7 and Δ^25,26 were assigned as Z based on the identical (≤ 1.0 ppm) ¹³C values reported for 9.²⁸ The configurations of six of the nine stereogenic centers were designated as 5R, 9R, 11S, 15S, 16S, 23S based also on matching ¹³C δ values. The remaining three positions were provisionally assigned 17R, 19S, 20R from NOESY data in Figure 3 that showed correlations from the OCH₃ to H-16 (δ_H 4.01) and H-17 (δ_H 5.65) indicating both sets were orientated on the same side of the plane. Modest MT stabilizing effects were seen for 15 in Table 1 confirming this assay can rapidly identify new compounds with MT cytoskeletal activity.

The final new compound, entry 16 (Fraction H61, 249 m/z) possessed a molecular formula of C₁₇H₂₈O₃ determined from the HRESITOFMS data based on the [M+Na]⁺ m/z ion of 303.1918. This represented the loss of one degree of unsaturation versus that of 6. The ¹H and ¹³C NMR data (see Table S4) indicated a substitution of the exocyclic double and acid moiety for a di-methoxy ethyl side chain with OCH₃ groups (δ_H 3.33, δ_C 53.5, 53.7), methylenes C-11 (δ_H 2.54, 2.46; δ_C 34.1), and a methine C-12 (δ_H 4.51, δ_C 103.4). A rearranged α, β-unsatured ketone C-9 (δ_C 204.1) with an endocyclic double C-8 (δ_C 133.1), C-7 (δ_H 6.53, δ_C 142.7) was also apparent. These conclusions accounted for the loss of one degree of unsaturation and were supported by 2D NMR data in Figure 3 showing² J HMBC correlations from the OCH₃ groups to C-12, ¹H - ¹H COSY data from H-12 to H-11 and^2–3 J -HMBC correlations from H-11 to C-7 - C-9. Additional COSY data from H-7 to H-6 (δ_H 2.39, 2.29) followed by^2–3 J -HMBC correlations from H-7 to C-9; H-6 to C-5 (δ_C 41.4), C-10 (δ_C 50.7) and H₃-15 (δ_H 0.98) to C-9, C-10 linked the B ring together in only one way as aignopsane ketal (16). The remaining atoms (C-1 to C- 5 and C-13 to C-15) displayed ¹³C NMR with values ≤ 1 ppm of those of 6, confirming the existence of the A ring and the overall working structure of 16. The geometry of Δ^7,8 was assigned as Z based on the observed vicinal coupling of 6.0 Hz between H-6 and H-7 that was in agreement with blancasterol⁴⁷ which shares a similar α,β-unsaturated ketone motif. Determination of the relative configuration of the asymmetric centers was set from NOESY data in Figure 3 which paralleled 13, indicating the 4S, 5R, 10S orientation. It’s possible to conclude that the unique structure of 16 arose from either 6 or 13 after a 48h exposure to MeOH-H₂O during processing and or transport, which may have lead to the formation of it’s dimethyl acetal functionality.

Although the above exemplar served to rapidly identify distinct bioactive natural products and several new compounds, one particular issue involving its application deserves discussion. Further efforts screening extracts of additional sponges and microorganisms led to several false positives with selected examples involving the compounds displayed in Figure 4. These compounds were identified as active library fractions based on inaccurate fraction weight concentrations used when assaying library wells. Typically natural product HPLC well plate library fractions are assayed based on averaging the amount injected on the column divided by the number of library fractions wells to arrive at an assumed amount per well.^13–15 This approach proved practical when assaying complex extracts such as in Figure 1. However when assaying extracts consisting of only 1–2 major metabolites contained in only a few library fraction wells, with the remaining library fraction wells being devoid of compounds, these libraries plates must be viewed with caution. Our results indicated these library fractions could appear an order of magnitude more potent versus data obtained from re-assay using accurate weights.

A further illustration of misleading analyses of libraries containing just a few major compounds culminating in the conclusion of false positives involves the situation depicted in Figure 5. The hexanes extract (sample coded FH) of a sponge (Spongia species, coll. no. 92503) appeared to have several metabolites with varying concentrations using ELSD and UV (230nm) detection. Although LC-MS analysis with ELSD is often regarded as a reliable indicator of actual sample concentrations, detector response factors can be affected by the nature of the solvent and analytes.¹⁰ As an example the ¹H NMR spectrum (see Figure S45) of this extract indicated the presence of just one major metabolite that was later confirmed as the active component in the LC-MS-UV-ELSD library fraction 92503 FH H27 in Figure 5a when tested at 10 µg/mL with an assumed weight of 0.1 mg. This library fraction displayed cytotoxicity on par with the standard doxorubicin⁴⁸ at 10 µM against macrophage (RAW 264.7) cells. After scale up isolation / de-replication the structure proved to be the common sponge diterpene, spongia- 13(16),-14-dien-19-oic acid (17).⁴⁹ Unfortunately when re-assayed as a pure compound with a measured weight (in mgs) it required a concentration ≥ 90 µM to achieve the same cytotoxicity seen previously alongside doxorubicin at 10 µM.

We now believe one way to avoid the bioassay variability described above is to use removable 96 well library plates, which can provide actual measured weights per well for the assay activity calculation. An example of this approach is outlined in Figure 5b. Reevaluation of the extract coll. no. 92503 FH indicated the library fraction 92503 FH H27 accounted for approximately ≥ 4.0 mg of the 10.0 mg extract and when the fraction was assayed using this amount (at 10 µg/mL) it was significantly less cytotoxic then doxorubicin at 10 µM. In terms of the bioassay data seen here, this sample would not constitute a priority lead to undergo automated scale up HPLC. Similar results were encountered with bioactive microbial extracts provided by the Phaff yeast collection⁵⁰ UC Davis as well as from Indonesian fungal samples. Selected examples include coll. nos. UCDFST 05565 L, and LIPI 010A5 L (see Figures S46, S48). Using the standard library approach these samples displayed cytotoxicity against prostate (PC3) cancer cell lines equal to or greater then doxorubicin at 10 µM. Upon reexamination using removable well libraries, we concluded that the initially active library fractions from samples 05565 L H10 and 010A5 L H29 which contained penicillic acid (18)⁵¹ and hexylcinnamaldehyde (19)⁵² (Figure 4) were false positives as they displayed markedly diminished cytotoxicity.

1_si_001