Stereochemical Studies of the Karlotoxin Class Using NMR and DP4 Chemical Shift Analysis and Insights into their Mechanism of Action (MoA)

Abstract

Following the publication of karlotoxin 2 (KmTx2, 1), the harmful algal bloom dinoflagellate Karlodinium sp. was collected from oceans and bays throughout the world and scrutinized to identify additional biologically active complex polyketides. The structure of 1 was validated and revised at C49 using computational NMR tools including J-based configurational analysis (JBCA) and chemical shift (CS) calculations. The characterization of 2 new compounds [KmTx8 and 9 (2–3)] was achieved through overlaid 2D HSQC NMR techniques, while the relative configurations were determined by comparison to 1 and computational chemical shift calculations. The detailed evaluation of 2 using the NCI-60 cell lines, NMR binding studies and an assessment of the literature supports an MoA of targeting cancer-cell membranes especially of cytostatic tumors. This is uniquely different from the MoA of current agents employed in the control of cancers such as leukemia and lung cancer for which 2 shows sensitivity.

Graphical abstract

Marine Cytotoxins as a Potential New Class of Cancer Therapeutics: Harmful algal bloom toxins derived from the dinoflagellate Karlodinium sp. produce and interesting suite of complex polyketides that have been shown to have hemolytic, cytotoxic and ichthyotoxic activity. Better understanding the mechanism of action for this class of compounds could lead to the development of a new cancer-cell membrane targeting agent.

The karlotoxins are a suite of compounds that have hemolytic, cytotoxic and ichthyotoxic activity.^[1] At least eight distinct strains of Karlodinium sp. have been isolated from around the world, and each produce their own suite of compounds.^[2] There has been a tremendous amount of research on the configurational assignment and MoA of the KmTx’s, driven by the structural similarity of 1^[3] to amphidinol 3 (Am3, 4) isolated from Amphidinium sp (Figure 1).^[4] While they share remarkable structural similarity and are probably derived from the same biosynthetic machinery, the pyran sections (C28–C49 in 1 and C30–C51 in 4) of the molecules are published with the enantiomeric configuration.^[4] The absolute configuration of 4 was determined based on JBCA, NOE interactions, and a modified Mosher’s method,^[4] while 1 was assigned using JBCA, NOE interactions, degradation with Grubb’s II catalysis, and periodate-mediated diol degradation.^[3a] Am3 (4) has already undergone two relative configuration revisions, C2 in 2008 and C51 in 2013.^[5] With the advancements of computational methodologies, the computationally-assisted JBCA and CS calculations of 1 were undertaken to further assess the assigned configuration of the KmTx class.

Figure 1.

Complex polyketide natural products derived from Karlodinium and Amphidinium sp. KmTx2 (1),^[3a] KmTx8 (2), and KmTx9 (3) and Am3 (4)^[4] with revised stereochemistry at C49 for 1. The red box indicates the deviation in the stereochemistry of the pyran systems between KmTx2 (1) and Am3 (4).

The revision for the relative configuration of the C50–C51 bond in 4^[5a] warranted verifying the assignment for the corresponding bond on 1 (C48–C49). The revision of 4 at C51 was confirmed using the gauge-including atomic orbital (GIAO) NMR CS calculation. This served both as validation for the method on a molecule of this size and further confirmation of the 4 revision (Figure 2). The ²J_{(C51, H50)} value of −2.5 Hz is considered an “intermediate” value, which may have led to the misassignment of C51.^[6]

Figure 2.

Confirmation of Am3 (4) structural revision using DP4 calculations. Diastereomers A–D for the relative configuration revision at C51 showing DP4 probability supporting the structural revisions made by Ebine et al.^[5a] Original JBCA for 4. ²J_{(C51, H50)} value of −2.5 Hz is considered an “intermediate” value resulting in the misassignment of C51. The “large” value for ³J_{(H44, H43)} leads to an unambiguous assignment.

The same method was explored to identify if a similar revision was necessary for 1. The large values (7–10 Hz for this diol system) for ³J_H,H are indicative of a structure with no significant conformational changes occurring, thus giving a reliable JBCA.^[6] Four possible diastereomers of C40–C51 were designed. The bonds between C42–C43, C47–C48 and the pyran were constrained based on NOE interactions and JBCA as previously described. Even though ²J_(C47,H48) fell in the “intermediate” range,^[3a] the C47–C48 bond was also constrained given the large ³J_(H47,H48) value^[3a] indicating no major conformational conversion occurred in this bond.^[6] The DP4 probability analysis was conducted with experimental and predicted CS values of the four diastereomers and the results indicated that 41S,49R was the most relevant with a probability of 86.5% (Figure 3). This implies that the original relative assignment of C48–C49 based on JBCA is incorrect and the revision for 4 holds true for 1 also. Like in 4, the ³J_(C50,H48) value of 3.5 Hz in 1 is considered an “intermediate” value, which may have resulted in the misassignment of C49. These calculations suggest that JBCA assignments should be employed with care especially when dealing with “intermediate” and “small” coupling values and, when appropriate, confirmed using the CS calculation method unless a vicinal coupling constant of two protons in an associated bond is “large”.^[6]

Figure 3.

Confirmation of karlotoxin structural revision using DP4 calculations. Four diastereomers investigating the relative configuration revision at C49 and C41 based on the amphidinol 3 revision. Original JBCA for 1. ³J_{(C50, H48)} value of 3.5 Hz is considered an “intermediate” value resulting in the misassignment of C49. The “large” value for ³J_{(H42, H41)} leads to an unambiguous assignment of configuration.

The next approach to verifying the absolute configuration of 1 was to use computation methods to validate the JBCA. The JBCA of C28–C29 resulted in an intermediate vicinal homo-coupling constant value.^[3a] As such, computation analysis of the three staggered conformers at that position gave us calculated values to compare with the measured values. The C28 position was determined based on degradation analysis to be the R configuration,^[3a] so for the analysis this position remained fixed. For comparison, the three conformers for the opposite configuration starting at C29 were also analyzed (Figure 4). The relative configuration of C29–C32 on both models was fixed based on the JBCA and the structures were minimized. The NOE data from 1 were also analyzed and integrated to establish distance restraints for the tetrahydropyran rings. After restraints were applied, a conformation search was conducted using Schrödinger MacroModel. Using an 80% or higher Boltzmann distribution cutoff, the conformers were optimized and the coupling constant values were calculated using Gaussian 09. The total absolute deviation (TAD) compared the absolute value of the difference between the calculated and the experimental values. The CS values for 1 were also analyzed using both mean absolute error calculations and DP4 probability statistics.^[7] The computation work supports the published relative configuration of C28–C29 of 1 in the anti conformation, which has the lowest TAD value at 2.89.

Figure 4.

Computational analysis of the coupling constants for C27–C30 assignment of KmTx2 (1) and inversion of stereochemistry starting at C29 in three conformation (gauche⁺, gauche⁻, anti) compared with the experimental values by total absolute deviation calculations. Dashed yellow line in the 3D representation indicates NOE correlations employed to restrain the tetrahydropyran ring.

Given that the GIAO NMR CS calculations have been validated to correctly assign the relative configuration of certain stereogenic centers in 1, the extension of this methodology was attempted to verify ambiguities of the JBCA in the original assignment. It is important to note that earlier attempts at analyzing these data using only fragments of the structure provided unreliable results depending on the length of the fragment being analyzed. When analyzing the larger, flexible fragments, the entire structure of 1 was analyzed using the DP4 applet.^[7a] The results showed that the relative configuration assignments of the published structure of 1 for C36–C37 and C28–C29, highly flexible bonds deduced from the medium values of ³J_(H36,H37) and ³J_(H28,H29), were correct (Figure 5).

Figure 5.

Four possible diastereomers for GIAO NMR shift analysis of KmTx2 (1). (A) The published structure of 1 with C49 modification, (B) inversion of stereochemistry at C29–C49, (C) inversion of stereochemistry C37–C49. (D) inversion of stereochemistry at C29–C36.

Capitalizing on the extensive NMR data collected for 1, a 2D Heteronuclear Single Quantum Coherence (HSQC) NMR overlay experiment was utilized to determine the planar structure of KmTx8 (2) isolated from an Australian K. veneficum strain.^[8] The NMR data alluded to the loss of an olefinic moiety between C4 and C5 with the addition of the oxidation of the hydroxy group at C6 to a carbonyl functionality. The appearance of a carbon resonating at δ211.94 indicated the presence of a ketone moiety, and the alteration of the CS values for C5 and C7 to δ51.94 and δ52.04, respectively, by HMBC assigned its position. The loss of the hydroxy group on C2 was also observed. The remaining differences in oxidation from C1–C18 were confirmed via HMBC and COSY correlations that walk the correlations from C1 through C5 and then again from C7 through C16. KmTx9 (3) was isolated from a closely related species, K. conicum, which came from the most southern and coldest location of all of the species isolated to date.^[2] The HSQC data for 2 and 3 overlaid well with the exception of increased signal intensity for 2 of the C53–C56 region. This coupled with the loss of 28 amu between the masses of 2 and 3 was a strong indication of a shortening of the aliphatic chain by 2 methylenes. The relative configuration of 2 and 3 was assigned by comparison to 1. For the portions of 2 and 3 that were too different for direct comparison by NMR, CS calculations were completed to confirm the relative configuration of the new stereogenic centers and link them to the known absolute configuration of C21 established in 1 by degradation chemistry.

Next, the GIAO NMR CS calculations were used to establish the relative configuration of stereogenic centers in the polyol section of 2 which is identical to 3. For validation of the method on a polyol chain, four different diastereomers of 1 were employed with varying configurations at C6 and C10 and the GIAO NMR CS analysis was performed. Based on these results, the DP4 method provided data that were consistent with the relative configuration of 1 that was established by Mosher’s ester analysis. The 6R,10R diastereomer showed the lowest TAD and a “100%” probability of being the correct assignment. Using the same technique, the relative configuration of the polyol chain of 2 was analyzed. Every carbon that did not have a direct overlap with the HSQC of 1 was analyzed. This not only allowed for the determination of the relative configuration of the polyol chain, but also relayed it back to the absolute configuration determined for 1 at C21. This method gave a conclusive answer for all but one of the undetermined stereocenters (4S,12R with C8 remaining ambiguous). The proton TAD and DP4 calculations supported one diastereomer and the carbon TAD and DP4 supported another diastereomer with only one stereocenter being opposite. Interestingly, when both carbon and proton data were taken into account for the DP4 analysis a different diastereomer with no discernible pattern was supported. It is clear that this occurs because that diastereomer strikes a balance between the lowest TAD for proton and carbon, which is statistically not unexpected.

Studies have elucidated the ecological role of 1 as being that of a prey capture mechanism. A pore is formed in the cell membrane with certain sterol interactions thus disturbing the osmotic balance of the cell through membrane depolarization, and ultimately leading to cell death through osmotic lysis.^[9] This MoA is dependent in part on the sterol composition of the cell.^{[2, 9b]} Cells containing 4α-methyl sterols are resistant to 1, whereas cells containing 4-desmethyl sterols (like cholesterol) are sensitive to 1.^[10] Studies also show that this lysis event is preceded by the permeabilization of the plasma membrane to various cations including Ca²⁺, Na⁺ and K⁺. Presumably, these same pores also allow for the disruption of the osmotic balance in the cell leading to cell death.^[11] Studies for 4 have shown a similar affinity for phospholipid membranes that contain sterols and transmembrane proteins. Sterols and protein-containing membranes play an important role in the mechanism of action which has been shown to be pore or lesion formation.^[12] Whether this pore follows the hypothesized barrel-stave model as predicted for amphotericin B or a toroidal pore model as predicted for 4 is still under investigation.^[13] The configuration of the sterol 3-OH group has also been shown to be important for activity in this class of molecules. For 4, pore formation was only seen when 3β-OH sterols were present in the membrane and not in the presence of 3α-OH sterols such as epicholesterol.^[14]

HSQC overlay studies were conducted to provide further insight into the sterol binding interactions of the karlotoxins. Using a series of increasing ratios of cholesterol added to 2, HSQC NMR experiments were conducted. Figure 7b summarizes the atoms that were perturbed by the addition of cholesterol. There were definitive shifts for certain carbons mainly around the tetrahydropyran ring systems, while other carbons on the exterior of the structure like C1 were untouched by the addition of cholesterol. Based on the cholesterol interactions, the molecule appears to prefer a hairpin formation which has not been seen when the molecule is freely rotating without other molecular interactions. ROESY experiments were also performed with these samples; however, the solubility limitations restricted the solvent choices for 2 with cholesterol and no meaningful NOE interactions were observed.

Figure 7.

Hypothesis for MoA of KmTx. A) Proposed pore formation of multiple KmTx molecules in a top down view of the cell membrane. B) Diagram of atoms affected by cholesterol interaction based on HSQC perturbation experiments. C) Diagram of KmTx interactions forming pores inside the cell membrane.

KmTx8 (2) exhibited hemolysis using rainbow trout erythrocytes with a HD₅₀= 0.049 ± 0.004 µM making it the most hemolytic compound at over 26 times the potency of KmTx2 (1) (HD₅₀ = 1.3 ± 0.3 µM). While KmTx9 (3) was the least active of any compounds assayed to date (HD₅₀ = 3.0 ± 0.3 µM). This indicated that the length of the lipophilic chain is important to the overall MoA of the molecule perhaps through perturbations in lipid bilayer interactions. When 1 was injected into 10 female Balb/c mice IP, no lethality was observed even at a high dose of 500 µg/kg. The only physical effects noted were reversible anorexia and a single case of necrotizing pancreatitis.^[15]

In the NCI-60 cell panel, KmTx8 (2) was most sensitive for leukemia cell lines SR (GI₅₀=0.100 µM) and CCRF-CEM (GI₅₀=0.686 µM), non-small cell lung cancer lines HOP-62 (GI₅₀=0.986 µM), NCI-H23 (GI₅₀=0.903 µM) and HOP-92 (GI₅₀=0.501 µM), ovarian cancer cell line IGROV1 (GI₅₀=0.631 µM), renal cancer cell line SN12C (GI₅₀=1.000 µM) and breast cancer cell line BT-549 (GI₅₀=0.501 µM). Figure 6 illustrates how 2 compares to other FDA-approved cancer drugs [eribulin (NSC 707389), taxol (NSC 125973), methotrexate (NSC 740), and cisplatin (NSC 119875]) as well as amphotericin B (AmB) (NSC 527017) which has a similar structure and mechanism known to involve sterol binding.^[16] KmTx8 (2) was also screened in HeLa cells (GI₅₀=369 nM, TGI=609 nM, LC₅₀= 1064 nM). In addition, a cell cycle analysis was performed and showed that there were no changes in cell cycle distribution after treatment with 2. This indicates that the MoA is independent of cell division, thus opening the door to treating both aggressively replicating and cytostatic cancer types and possibly reducing off-target toxicity issues commonly associated with various parts of the cell cycle.

Figure 6.

Comparison of select NCI 60 cell panel GI₅₀ for KmTx8 (2) vs. controls.

Comparison of 2 against all publically available NCI-60 data using the CellMiner Database Version 1.4,^[17] the Compare^[18] algorithm and Pearson correlation coefficients were performed. The Pearson correlation for 2 with AmB is 0.254. Any value under 0.7 is considered statistically insignificant. Therefore, 2 and AmB do not share a similar MoA based on their NCI-60 data. However, it is possible that the activity of 2 is due to interactions with multiple targets in addition to sterol interactions. This would explain the low correlation to AmB. Further studies are needed to address the similarity of the MoA. The closest correlation to 2 was a synthetic product not in clinical trials where the MoA has not been defined (citratohydragtogallium (III), 0.766). Using this same resource, the mining of gene correlations seen to be either up- or down-regulated for 2 was also explored. Using the Search Tool for the Retrieval of Interacting Genes/Proteins (STRING) 9.0 database, further downstream and upstream targets in the MoA pathway were identified for the top 10 gene correlations.^[19] All but four correlations share a common pathway. This pathway appears to be heavily influenced by extracellular receptors (IL2RA and IL2B) imbedded in the cell membrane.

Given that the already established MoA is the permeabilization of the plasma membrane, our hypothesis is that the sterol binding interactions of 2 are affecting the stability of the cell membrane thus having an impact on membrane integrity and possibly cellular signaling that depends on extracellular receptors to function properly. The pore formed by 2 appears similar to an aquaporin or water channel based on activity. This is further supported by the 0.644 correlation with Aquaporin 9. While the value is only marginally significant, it implies that a pore forming mechanism affecting osmotic balance might be responsible for the activity. ABCA1, a protein responsible for transfer of cellular cholesterol across the plasma membrane, has been shown to play a role in the buildup of cholesterol in cancer cells over normal cells leading to avoidance of the normal cell-death signaling cycle.^[20] These data further support the mechanistic rationale for cholesterol targeting as a potential cancer treatment strategy.

Given the aqueous nature of the cellular environment, it is hypothesized that KmTx forms a pore with multiple molecules in a micelle-like interaction. This formation is also seen in previous studies on 4.^[13a] While it is understood that the limitations of solubility of KmTx’s and sterols limits the ability of the NMR binding experiment to closely mimic the aqueous external membrane environment, the results from this study corroborate what has previously been seen for 4. These interactions would aid the pore formation of KmTx in the sterol-containing cellular membrane. The proposed interaction of KmTx on a cell membrane is shown in Figure 7. The data supports that the sterol interactions are a vital part of the MoA, while it is hypothesized that the length of the lipophilic chain interacts with another portion of the membrane, possibly the phospholipids, to strengthen the interaction. The 3-OH of the sterol group is likely heavily involved in hydrogen bonding at the apex of the proposed hairpin structure, while van der Waals forces strengthen the binding of the polyene section of KmTx to the lipophilic tail of the sterol. Having a sufficient chain length is critical to activity as seen in the difference of hemolysis of 2 and 3. It is likely that the shorter chain length of 3 does not allow for multiple molecules to stabilize each other as well as the longer chain length does. The lack of stabilization results in fewer pores being able to form leading to the decrease in activity. The degree of hydroxylation of the molecule is hypothesized to be equally important to provide stabilizing hydrogen bonds with the transmembrane proteins and other KmTx molecules.

This suite of molecules exhibits an interesting cytotoxic MoA through cholesterol and other membrane binding interactions. Based on computational data and comparison with the structural revisions made to Am3 (4), a minor stereochemical revision was made to the KmTx scaffold at C49.^[5a] Furthermore, this new data using the DP4 calculations supports the published relative configuration of KmTx2 (1) at C28 through C37. By mining the NCI-60 cytotoxic activity of KmTx8 (2), it is clear that this class of molecules provides cell sensitivity for certain tumor cell groups. The specific interactions and structural details which yield tumor cell sensitivity for NSCLC and leukemia remains to be determined; however, the data presented here reveals that preparing tumor cell selective, pore forming agents is a possibility that must be explored in the development of new chemotherapeutic agents for the control of cancer.