Characterization of Drug-like Chemical Space for Cytotoxic Marine Metabolites Using Multivariate Methods.

In the last few decades, marine metabolites have been exploited to find commercially viable products in several areas. In this article, molecular descriptors [log P, mass, total polar surface area (TPSA), H-bond donor, H-bond acceptor, and the number of rotatable bonds] for the marine-derived cytotoxic metabolites were calculated and compared with marketed anticancer drugs to understand their position in the drug-like space. Marine-based cytotoxic metabolites are divided into highly toxic (HT) and moderately toxic (MT) classes. The marketed anticancer drugs complied well with Lipinski's rule of five for all molecular descriptors. The majority of HT and MT metabolites complied solely with H-bond donors and a number of rotatable bonds with the Lipinski cutoff values. Hierarchical cluster analysis (HCA) and principal component analysis (PCA) were also performed using 73 molecular descriptors on an ensemble of highly cytotoxic or moderately cytotoxic marine metabolites and the marketed reference drugs. The HCA results showed that 12% of marine metabolites clustered with the marketed anticancer drugs and many of them had structural scaffold homology. The PCA results revealed the presence of a clear distinction between the cytotoxic marine metabolites and the marketed anticancer drugs. Results indicate that mass, TPSA, and log P are the vital parameters and the careful optimization of these parameters for marine cytotoxic metabolites may generate more meaningful anticancer candidates in the future.

Introduction

Cancer is one of the unbeatable causes of death worldwide. As per the GLOBOCAN statistics for 2012, there are 14.2 million new cancer cases, which are diagnosed and result in a mortality of 8.2 million globally.[1] In the last two decades, there has been an alarming increase in cancer cases in the lesser developed world. This is increasingly evident because of the 8–15% rise in mortality due to cancer observed in developed countries as compared to lesser developed ones. Prostate, colorectal, lung, and breast cancers have high incidence rates in developed countries, whereas lesser developed countries have witnessed an increase in liver, stomach, and cervical cancer types owed to the prevalence of less hygienic conditions.[2] Even though a battery of treatment options are available, except acute lymphoblastic leukemia, the human battle is ongoing against other cancer types. Chemotherapy is a primary treatment option that has undergone dramatic progress in the last few decades.[3] It has few important drawbacks such as toxicity due to lack of specificity to cancer cells[4] and multidrug resistance.[5] To address these pitfalls, the search for new drug candidates is essential. This forces seeking the novel sourcing of new drug candidates in both academia and the pharmaceutical industry.

Scientific exploration in the ocean opened up a new avenue for medicinal chemists and the pharmaceutical industry. MarinLit, a marine-based natural product database contains about 24,000 compounds and 26,000 research articles; it is a good piece of evidence for the growing interest in marine-sourced natural products. Excellent reviews on marine natural products give detailed accounts of their sources and the various classes of molecules derived from them.[6−9] Marine-based natural products have found wide application as pharmacologicals,[10] nutraceuticals,[10] cosmetics,[10] antifouling agents,[11] and surfactants.[12] For pharmaceutical application, many marine-based drugs are currently in the pipeline for various disorders such as tetrododoxin for pain,[13] DMXBA for neurological disorders,[14] and plitidepsin for cancer.[15] Cytarabine (Ara-C) and trabectedin are approved as anticancer drugs by the USFDA and EMA, respectively, having their origin in the ocean.[16] The marine-based drug pipeline includes a number of agents having exciting anticancer potential in various stages of clinical trials.[16]

Modern medicinal chemistry has acquired tools from chemoinformatics to identify and predict the drug-like and pharmacokinetic properties of new candidates.[17] Molecular descriptors explained by Lipinski—such as molecular weight (<500), log P (<5), H-bond donor (<5), and H-bond acceptor (<10)—are useful to understand the oral absorption of drug or drug-like compounds.[18,19] The extension of Lipinski’s rule to predict the drug-like properties of compounds include polar surface area and molar refractability.[20] Lipinski’s rule of five has been reduced to three [molecular weight (<300), log P (<3), H-bond donor (<3), and H-bond acceptor (<3)] to accommodate the increase in molecular weight and lipophilicity during the lead optimization stage in the drug discovery process.[21] This article attempts to identify the possible drug-like candidates from the different groups of marine-based cytotoxic metabolites by analyzing their molecular descriptors benchmarked with currently marketed anticancer drugs. Additionally, hierarchical cluster analysis (HCA) and principal component analysis (PCA) have been performed on the data to identify the marine metabolites in the drug-like space.

Results and Discussion

The log P distribution for highly and moderately cytotoxic marine metabolites and the marketed cancer drugs is shown in Figure .

Figure 1

Comparison of log P for cytotoxic marine metabolites and marketed anticancer drugs. Highly toxic (HT; blue); moderately toxic (MT; red); marketed drug (green).

About 90% of the marketed drugs obeyed Lipinski’s rule of five (log P < 5)[18,19] which includes negative values. For the HT and MT categories, 72 and 84% accounted for the range from −ve values to +5, respectively. Even while excluding negative values, about 64% of HT and MT were in the range of 0–5.

Marketed anticancer drugs showed a high (90% including negative values) compliance with Lipinski’s rule of five for log P, because they were modified to enhance the pharmacokinetic properties, particularly oral availability. The average log P of HT, MT, and marketed anticancer drugs are 3.65, 2.16, and 1.25, respectively. Three drugs (Ara-C, trebectedin, and eribilin indicated in Figure ) from marine origins among the marketed drugs category obeyed Lipinski’s rule of five (0–5). The aqueous marine environment may be a driving force for cyanobacteria or sponges to produce highly hydrophobic cytotoxic metabolites having a high log P value, which can be delivered to the target in their vicinity.

For the mass distribution shown in Figure , 62 and 26% of marketed anticancer drugs fall in the range of 0–500 and 500–1000, indicating a significant number of deviations from Lipinski’s rule of five (M < 500).

Figure 2

Comparison of mass for marine metabolites and marketed anticancer drugs. HT (blue); MT (red); marketed drug (green).

The majority of highly (77.94%) and moderately (73.39) toxic marine metabolites are in the range of 500–1000. Mass in the range of 1000–2000 is contributed between 8 and 20%; hence, synthetic and marine cytotoxic metabolites showed similar trends in this category. The average mass for marketed anticancer drugs, highly (77.94%) toxic marine metabolites, and moderately (73.39) toxic marine metabolites are 517, 827, and 897, respectively. Out of three marine-derived reference drugs, Ara-C obeyed Lipinski’s rule of five and other two (trabectedin & eribulin) showed a violation. There is a large difference in the average mass between the marketed anticancer drugs and other cytotoxic marine metabolites. Hence, mass and log P showed considerable deviation for bench marked anticancer drugs and marine-based cytotoxic metabolites.

Total polar surface area (TPSA) variation for marketed anticancer drugs and marine-based cytotoxic metabolites is shown in the Figure .

Figure 3

Comparison of TPSA for cytotoxic marine metabolites and marketed anticancer drugs. HT (blue); MT (red); marketed drug (green).

As per Lipinski’s rule of five, TPSA has the range of 0–140. In the case of marketed anticancer drugs, 64% complied with Lipinski’s rule of five, whereas marine-based cytotoxic metabolites showed a compliance of 21–30%. In the range 140–300, 56% of HT and MT categories of marine cytotoxic metabolites were observed. In the category of marine-derived reference drugs, Ara-C obeyed Lipinski’s rule of five, whereas trabectedin & eribulin violated by having a larger TPSA.

Lipinski’s rule of five limits the number of rotatable bonds to less than 10 (RB < 10) for a drug candidate. Figure depicts distribution of RB for marine metabolites and marketed anticancer drugs.

Figure 4

Comparison of number of rotatable bonds for cytotoxic marine metabolites and marketed anticancer drugs. HT (blue); MT (red); marketed drug (green).

The percentage distribution of marketed anticancer drugs, highly cytotoxic, and moderately cytotoxic categories for the range 0–10 are 88, 64, and 48, respectively. Interestingly, the marketed anticancer drugs including marine sources (Ara-C, trabectedin, eribulin) strictly obeyed Lipinski’s rule of five with the highest percentage distribution in the 0–10 range among the three different categories. A decrease in the number of rotatable bonds makes the molecule rigid, restricting the conformation freedom. Furthermore, it may freeze the molecule in bioactive conformation, which confers the drug status for the given candidate. This may account for a steady decrease in the percentage distribution from marketed anticancer drugs to highly cytotoxic and moderately cytotoxic categories for the range 0–10.

For H-bond donors, the percentage distribution for marketed anticancer drugs and cytotoxic marine metabolites is shown in the Figure .

Figure 5

Comparison of number of hydrogen bond donors for cytotoxic marine metabolites and marketed anticancer drugs. HT (blue); MT (red); marketed drug (green).

HT and marketed anticancer drugs account for about 80% for the 0–5 range. The maximum number of hydrogen bond donors as per Lipinski’s rule of five is five and its compliance is excellent for the marketed anticancer drugs including marine sources (Ara-C, trabectedin, eribulin) and the highly cytotoxic marine metabolites. Moderately, cytotoxic metabolites are lagging behind (66% in the range of 0–5) the other two categories in this respect.

Cluster Analysis

HCA was performed on all the ctyotoxic marine metabolites and marketed anticancer drugs, as shown in Figure .

Figure 6

Dendrogram of HCA showing cytotoxic marine metabolites and the marketed anticancer drug using Ward’s method (H denotes highly cytotoxic marine metabolite, M denotes moderately cytotoxic marine metabolite, R denotes marketed anticancer reference drug).

It is important to identity the cytotoxic marine metabolites clustering with marketed anticancer reference drugs, which is the primary goal of HCA in this article. Molecular scaffold plays an important role in the structure–activity relationship.[22] It is classified into structural and functional scaffolds. Hence, the comparison of structural and functional scaffolds of reference drugs with marine metabolites is used to understand whether the clustering similarity results from structural homology or nonstructural factors. The comparison of scaffolds of reference drugs and marine metabolites was based on the Bemis–Murcko framework, which presents the structural scaffold along with the functional scaffolds associated with them. All marine metabolites clustering with marketed reference drugs are mentioned in Table . The dendrogram showing HCA of all the cytotoxic marine metabolites and the marketed anticancer reference drugs is shown in Figure . It shows two major clusters 1 and 2. Cluster 2 consisted of two subclusters F and G. In the subcluster F, bleomycin (R8) showed a high degree of similarity with the family of largamides (D,E,F,G) (M41–44). Scaffolds showing the backbone and the rings are shown in the Supporting Information. The bleomycin scaffold showed that it is an acyclic peptide molecule with two aromatic five-membered rings at the N-terminal region and two nonaromatic six-membered rings. It has a long twelve carbon tail ending with two thiazole rings. Bleomycin exerts its cytotoxicity by intercalation of its peptide into the GC-rich region of the DNA and the two thiazole rings bind metal ions leading to the production of free radicals resulting in DNA cleavage.[23] Largamides have a cyclic peptide head, a smaller hydrophobic chain, and a six-membered aromatic ring at the end of the molecule. Even though both bleomycin and largamides share certain amounts of scaffold homology, largamides lack the functional scaffold (like thiazole rings) to enter the drug-like space.

Table 1

List of Cytotoxic Marine Metabolites Clustered with Reference Drugsa

S. No	marine metabolite—anticancer reference drug in the dendrogram	anticancer reference drug (R)	Cytotoxic marine metabolite
1	R8-M41  (sub-cluster G in Figure 6)	bleomycin (R8)	largamide D (M41)
2	R8-M42(sub-cluster G in Figure 6)	bleomycin (R8)	largamide E (M42)
3	R8-M43(sub-cluster G in Figure 6)	bleomycin (R8)	largamide F (M43)
4	R8-M44(sub-cluster G in Figure 6)	bleomycin (R8)	largamide G (M44)
5	R15-M16(sub-cluster G in Figure 6)	degarelix (R15)	halicylindramide D (M16)
6	R15-M36(sub-cluster G in Figure 6)	degarelix (R15)	koshikamide A2 (M36)
7	R17-H57(sub-cluster E in Figure 6)	erubilin (R17) marine derived	piperazimycin B (H57)
8	R17-H58(sub-cluster E in Figure 6)	erubilin (R17) marine derived	piperazimycin C (H58)
9	R3-H64 (sub-cluster D in Figure 6)	actinomycin D (R3)	thiocoraline (H64)
10	R25-M107 (sub-cluster D in Figure 6)	leuprolide (R25)	wewakazole (M107)
11	R24, R28-H15, H16 (sub-cluster A in Figure 6)	ixabepilone (R24), mitomycin (R28)	cryptophycin 1 (H15) cryptophycin-52 (H16)
12	R33-H18, H47 (sub-cluster F in Figure 7)	vinblastine (R33)	diazonamide (H18),microcolin B (H47)
13	R34-H18, H47 (sub-cluster F in Figure 7)	vinorelbine (R34)	diazonamide (H18), microcolin B (H47)
14	R32-H18, H47 (sub-cluster F in Figure 7)	trabectedtin (R32) marine derived	diazonamide (H18), microcolin B (H47)
15	R29-H18, H47 (sub-cluster F in Figure 7)	paclitaxel (R29)	diazonamide (H18), microcolin B (H47)
16	R4-H60 (sub-cluster D in Figure 7)	anastrozole (R4)	smenothiazole B (H60)
17	R4- H69 (sub-cluster D in Figure 7)	anastrozole (R4)	smenthiazole B (H69)
18	R15, R8-H55 (sub-cluster C in Figure 7)	degarelix (R15), bleomycin (R8)	palauamide (H55)
19	R24-M75 (sub-cluster A in Figure 8)	ixabepilone (R24)	pseudodysidenin (M75)
20	R30-M104, (sub-cluster A in Figure 8)	tamoxifen (R30)	virenamide B (M104)
21	R30-M105   (sub-cluster A in Figure 8)	tamoxifen (R30)	virenamide C (M105)
22	R30-M3,  (sub-cluster A in Figure 8)	tamoxifen (R30)	belamide A (M3)
23	R30-M103 (sub-cluster A in Figure 8)	tamoxifen (R30)	virenamide A (M103)

Cases 1–11, 12–18, and 19–23 are extracted from HT + MT + reference drugs, HT + reference drugs, and MT + reference drugs respectively.

Reference drug degarelix (R15) clustered with koshikamide A2 (M36) and halicylindramide D (M16). Degarelix is GnRH antagonist[24] and functional scaffold requirements are fused aromatic ring at the N-terminal and a linear peptide structure having aromatic rings on either side and a five membered ring aliphatic ring at the C-terminal region. Degarelix can be compared with leuprolide, which is a GnRH agonist showing structural similarity. In this cluster (R15-M36-M16), koshikamide A2 showed few structural similarities to degarelix; it has a six-membered aromatic ring at the C-terminal and lacks aromatic rings between the N- and C-terminal regions.

Eribulin (R17) clustered with piperazimycins (A,B,C-H57-59). Eribulin is a cylic macrolide and it destabilizes the microtubule assembly by acting as a hydrogen bond breaker.[25] However, the piperazimycins have a cyclic structure having both H-bond donors and acceptors. Because of this, the hydrogen bond breaking capacity of eribulin is higher than piperazimycins.

Actinomycin D (R3) clustered with thiocoraline (H64). As shown in the supplement 2, both peptides did not show any structural scaffolds. Actinomycin has a homodimer and the thiocoraline is a bicyclic peptide. Moreover, the most important functional scaffold in actinomycin D is the phenoxazone ring, which binds to the GC base pairs in the DNA,[26] preventing the RNA polymerase from binding to the DNA.

Leuprolide (R25) and wewakazole (M107) clustered and both the peptides showed structural variations. Leuprolide is a linear peptide and wewakazole is a cyclic peptide and the geometry of functional scaffolds is essential for bioactivity.

Ixabepilone[27] (R24) and mitomycin (R28) clustered with cryptophycin 1[28](H15) and Cryptophycin 52 (H16). Both the cryptophycins share cyclic peptide scaffold homology with ixabepilone and all three inhibit microtubule assembly.

In the dendrogram of HT and reference drugs, few clusters having both the marine metabolites and the reference drugs were observed in Figure . Paclitaxel (R29), trabectedin (R32), vinblastine (R33), and vinorelbine (R34) clustered with diazonamide (H18) and microcolin A (H47). Paclitaxel, trabectedin, vinblastine, vinorelbine, and diazonamide showed complex fused ring systems. Except trabectedin, the other four showed binding to microtubules.[27] In the second cluster, anastrozole (R4) clustered with smenothiazoles A and B. All three showed a heterocyclic five-membered ring connected to either a six-membered (anatrozole) or five-membered ring systems (smenothiazoles). Anastrozole binds to the aromatase in which the azole electron deficient nitrogen binds to the heam prosthetic group in the enzyme.[29] Similarly, smenothiazoles have thiazole rings, which may be responsible for bioactivity. Finally, degarelix (R15) and bleomycin (R8) clustered with palauamide (H55) in which degarelix and bleomycin showed structural homology having an open peptide structure with aromatic rings between the N- and C-terminal regions. Palauamide is a cyclic peptide with one benzene ring and does not share much with the other two reference drugs.

Figure 7

Dendrograms of HCA showing HT marine metabolites (H) and the marketed anticancer drug (R) using Ward’s method.

The dendrogram having moderately marine metabolites and reference drugs showed two interesting clusters in Figure . First, ixabepilone (R15) clustered with pseudodysidenin (M75) in which both showed the thiazole ring, as stated in earlier sections, thiazole is an important functional scaffold for many bioactive compounds. Ixabepilone inhibits the microtubules[27] and the exact mechanism of pseudodysidenin is not known; thiazole may have functional role in its bioactivity. The second cluster, tamoxifen (R30), clustered with virenamides (A–C) (M103, M104, and M105) and belamide (M3). Three ring connectivities are the structural scaffold in this cluster and they are observed in all four molecules except virenamide A. In the case of tamoxifen, three aromatic ring systems may be important for binding to the estrogen receptor[30] and the thiazole ring present in virenamides may be the important functional scaffold.

Figure 8

Dendrogram of HCA showing MT marine metabolites (M) and marketed anticancer drugs (R) using Ward’s method.

Principal Component Analysis

The suitability of PCA was assessed prior to the analysis. The overall Kaiser–Meyer–Olkin (KMO) measure and Bartlett’s test of sphericity for three data sets are given in Table .

Table 2

PCA Datasets and Their Quality Parameters

PCA dataset	dataset size	KMO values	Bartlett’s test of sphericity
HT + MT + reference drugs (dataset 1)	212	0.807	<0.0005
HT + reference drugs (dataset 2)	103	0.681	<0.0005
MT + reference drugs (dataset 3)	143	0.810	<0.0005

Dataset 1 & 3 showed the KMO value > 0.7 indicating that the classifications belong to middling to meritorious according to Kaiser.[31] Dataset 2 (HT + reference drugs) KMO falls between the mediocre to middling category and the KMO values of all the data sets indicate PCA is well applicable to all analyses using PCA. Bartlett’s test of sphericity for all three datasets was <0.0005, indicating that all three datasets were likely factorizable.

The PCA score plot for dataset 1 (HT, MT and reference drugs) is shown in Figure .

Figure 9

PCA score plot for HT (yellow), MT (green), and reference drugs (brown). Three marine derived anticancer reference drugs are R12 (Ara-C), R17 (eribulin), and R32 (trabectedin). H60 corresponds to smenthiazole A.

In the left upper and lower quadrant of the score plot, 25 reference drugs were grouped accounting for 74% and the rest co-exist with other marine metabolites. The four reference drugs spread over the right upper quadrant area which resulted from a higher mass (>1000); mass is one of the parameters resulting in the clear separation of the two groups. Except smenothiazole A (H60), no other marine metabolite was observed in the major group of reference drugs (brown). Out of three marine-derived reference anticancer drugs, R12 (marked R12 = Ara-C) falls into the group of 25 reference drugs (Brown), whereas the other two (marked R32 = trabectedin, marked R17 = eribulin) co-exist with other marine derived cytotoxic metabolites. Interestingly, eribulin was present in the group of marine metabolites and completely separated from other reference drugs. Three aggregated sub space groups (circled in red color) were observed and interestingly two reference drugs (R18-etoposide in the top and R17-eribulin in the bottom) were present in the aggregated sub space.

The score plot from the PCA analysis performed on HT and reference drugs is shown in Figure . The separation of two groups could be clearly observed and 85% of the reference drugs were seen on the left side of the red line. Two marine derived reference drugs (R12-Ara-C and R32-trabectedin) were grouped with a majority of the reference drugs. In the left side, 29 HT metabolites were present; still there was no complete mixing of reference drugs and HT metabolites. Interestingly, smenthiazole B (H69) and smenthiazole A (H60) were observed to closer to reference drug group compared to other marine metabolites.

Figure 10

PCA score plot for HT marine metabolites (yellow) and the marketed anticancer drugs (green). Three marine derived reference drugs are R12 (Ara-C), R17 (eribulin), and R32 (trabectedin).

The score plot from the PCA analysis on moderately cytotoxic marine metabolites and the reference drugs is shown in Figure . Most of the (76%) reference drugs were observed in the upper left quadrant of the plot. Similar to earlier PCA results, two marine-derived anticancer reference drugs (R12-Ara-C, R32-trabectedin) were observed with the other reference drugs. Majority of cytotoxic marine metabolites were clustered around the center of the plot. Eight cytotoxic marine metabolites circled in orange color [jasplakinolide J (M29), jasplakinolide M (M30), jasplakinolide Q (M31), marthiapeptide A (M50), microcystin-YR (M52), pipestelide A (M72), virenamide B (M104), and virenamide C (M105), reference drugs: daunorubicin (R14), doxorubicin (R16), etoposide (R18), irinotecan (R23), and methotrexate (R27)] co-existed with reference drugs.

Figure 11

PCA score plot for MT marine metabolites (yellow) and the marketed anticancer drugs (green). Three marine derived anticancer reference drugs are R12 (Ara-C), R17 (eribulin), and R32 (trabectedin).

Successful drug design must address structural, ADMET aspects for a drug candidate to become a marketable drug by any pharmaceutical company. Oral availability is governed by the log P of a drug and it has an impact on the effective serum concentration, eventually affecting the required efficacy at the target site. Majority of marketed anticancer drugs (88%) are highly hydrophilic having log P below 5 possessing excellent oral availability, whereas marine-based cytotoxic metabolites (HT and MT) are highly hydrophobic, resulting in poor oral availability. In the case of mass, about 62% of marketed anticancer drugs complied with Lipinski rule of five (0–500); this deviation is more pronounced with cytotoxic marine metabolites. TPSA compliance with Lipinski’s rule is higher (64%) for marketed anticancer drugs as compared to other two categories (21–30%). Drug candidates exhibit poor absorption when their TPSA is higher than 140 Å2, which is benchmarked for marketed drugs.[32−34] TPSA has a positive correlation with mass and the molecules with a mass higher than 500 are observed to have TPSA beyond the range of 0–140. Compliance with Lipinski’s rule of five for a total number of rotatable bonds and H-bond donors for cytotoxic marine metabolites is high and matched with reference drugs.

The comparison of cytotoxic marine metabolites clustered with reference drugs in the dendrogram obtained from HCA showed that many scaffolds of cytotoxic marine metabolites have a high degree of similarity with reference anticancer drugs. Bleomycin (R8)–largamides-E-F, degarelix (R15)–koshikamide A2 (M36), erubulin (R17)–piperazimycins, vinblastine (R33)–diazonamide (H18), anastrozole (R4)–smenothiazole B (H69), and tamoxifen (R30)–virenamides B-C (M104, M105) are good cases of structural scaffold homology. Apart from structural scaffolds, the placement and topology of functional scaffolds in the molecule is important for bioactivity and the selection of function scaffold depends on the protein target in the cell. The PCA results suggest a clear distinction between the reference anticancer drugs and the cytotoxic marine metabolites. The raw marine cytotoxic metabolites have to undergo modifications to reach the anticancer drug space. When PCA was run on HT and reference drugs, a sizable number of HT metabolites grouped with reference drugs. Smenothiazoles A and B (H60, H69) merged into the reference anticancer drugs space (Figure ). Hence, it offers optimistic picture on cytotoxic marine metabolites in the journey to become a successful drug candidate.

Molecular engineering on marine cytotoxic metabolites by applying de novo drug design principles may possibly remove the unnecessary hydrophobic group or fragment from marine-based cytotoxic metabolites. In this process, reduction with mass may optimize the factors such as mass, log P, and TPSA to improve the drug-like properties. In this process, the structural and functional scaffolds of the candidate should be retained for its biological activity.

Conclusions

Analysis of five molecular descriptors (log P, mass, TPSA, number of rotatable bonds, and H-bond donors) for reference anticancer drugs and cytotoxic marine metabolites showed that cytotoxic marine metabolites violated Lipinski’s rule of five with respect to mass and TPSA. The HCA results indicate that there are cytotoxic marine metabolites having structural scaffold homology with reference anticancer drugs. PCA results pointed out that cytotoxic marine metabolites have not reached the drug status in their current form. Necessary chemical modifications on cytotoxic marine metabolites controlling mass, log P, and TPSA without disturbing structural and functional scaffolds will result in more drug-like candidates for the chemotherapy route in the future.

Methods

Cytotoxic marine metabolites from cyanobacterial species were collected from the cyanobacterial database.[35] For sponges and other creatures (mollusk, ascidian, sea hare etc), a list of cytotoxic marine metabolites was obtained from excellent reviews.[36−39] Each marine source was divided into two categories—highly cytotoxic (Table ) (picomolar to 100 nM) and moderately cytotoxic (Table ) (101 nM to 15 μM). Highly and moderately cytotoxic groups had 69 and 109 metabolites, respectively. Thirty four currently marketed and widely used anticancer drugs, which include three agents from the marine origin, were benchmarked against the marine cytotoxic metabolites (Table ).

Table 3

List of Highly Cytotoxic Marine Metabolites

aplidine	diazonamide	jasplakinolide R1	neamphamide D
apratoxin A	didemnin	kulokekahilide-2	palauamide
apratoxin B	dolastatin 10	lagunamide A	piperazimycin A
apratoxin C	doliculide	lagunamide B	piperazimycin B
apratoxin D	dolstatin 15	largazole	piperazimycin C
apratoxin E	dolstatin 16	lissoclinamide 4	smenothiazole A
apratoxin F	geodiamolide A	lissoclinamide 5	smenothiazole B
aurilide B	geodiamolide B	lyngbyaballin A	symplocamide A
aurilide C	geodiamolide D	majusculamide C	symplocin A
bisebromoamide	geodiamolide I	malevamide D	tamandarin A
chondramide A	grassystatin A	mechercharmycin A	thiocoraline
chondramide B	grassystatin B	microcolin A	viequeamide A
chondramide C	grassystatin C	microcolin B1	yakuamide-A
chromopeptide	hemiasterlin	microcolin B3	yakuamide-B
coibamide A	hemiasterlin A	milnamide A	zygosporamide
cryptophycin 1	jasplakinolide	milnamide C
cryptophycin-52	jasplakinolide D	milnamide E
desmethoxymajusculamide C	jasplakinolide Q	molassamide

Table 4

List of Moderately Cytotoxic Marine Metabolites

antillatoxin	jasplakinolide Q	neamphamide B	theonellamide F
antillatoxin B	kahalalide F	neamphamide C	theopapuamide
belamide A	keenamide A	N-Methylsansalvamide	thiocoraline C
bistratamide J	kempopeptin A	nostocyclopeptide A1	ulongapeptin
bouillonamide	kempopeptin B	nostocyclopeptide A2	veraguamide A
callipeltin A	koshikamide A2	obyanamide	veraguamide-B
callipeltin B	koshikamide B	ohmyungsamycin A	veraguamide-C
cordyheptapeptide E	largamide A	ohmyungsamycin B	veraguamide D
cordyheptapeptide C	largamide B	onchidin	veraguamide E
cycloxazoline	largamide C	orbiculamide A	veraguamide G
geodiamolide E	largamide D	phakellistatin 12	veraguamide-K
geodiamolide F	largamide E	pipestelide A	veraguamide-L
grassypeptolide A	largamide F	pitipeptolide A	virenamide A
grassypeptolide B	largamide G	pitipeptolide B	virenamide B
gymnangiamide	laxaphycin B	pseudodysidenin	virenamide C
halicylindramide D	leucamide A	rolloamide A	vitileuvamide
haligramide A	lissoclinamide 7	roseotoxin B	wewakazole
haligramide B	lyngbyaballin B	sansalvamide A	wewakpeptin A
hantupeptin A	lyngbyastatin 1	scleritodermin A	wewakpeptin B
hantupeptin-B	marthiapeptide A	scytalidamide B
hantupeptin-C	microcyclamide	scytalidamide-A
hoiamide A	microcystin-YR	seragamide A
hoiamide B	microcystin-LR	seragamide E
homodolastatin 16	microcionamide A?	symplostatin 3
IB-01212	milnamide A	tasiamide A
jamaicamide A	milnamide D	tasiamide B
jamaicamide B	milnamide F	tasipeptin-A
jamaicamide C	milnamide G	tasipeptin-B
jasplakinolide J	milnamideC	theonellamide A
jasplakinolide M	mollamide B	theonellamide E

Table 5

List of Marketed Anticancer Drugs

abiraterone	daunorubicin	lomustine
actinomycin D	degarelix	methotrexate
anastrozole	doxorubicin	mitomycin
bendamustine	erubilin	paclitaxel
bexarotene	etoposide	tamoxifen
bicalutamide	exemestane	temozolomide
bleomycin	fludarabine	trabectedtin
busulfan	gemcitabine	5-flu-uracil
bhlorambucil	ifosfamide	vinblastine
cyclophosphamide	irinotecan	vinorelbine
ara-C	ixabepilone
dacarbazine	leuprolide

Marketed anticancer drugs were selected from all categories such as alkylating agents, antitumor-antibiotics, antimetabolites, microtubule inhibitors, DNA linking agents, and hormones. Furthermore, three recently FDA/EMA approved anticancer drugs from marine sources were included in the reference list of anticancer drugs. Five molecular descriptors (log P, mass, number of rotatable bonds, TPSA, and the number of hydrogen bond donors) were calculated using the Molinspiration software from Molinspiration Chemoinformatics for a comparison study on Lipinski’s rule of five between marine metabolites and the marketed reference drugs. The initial cutoff was set for molecular parameters adhering to Lipinski’s rule of five obtained from the relevant literature. The data was analyzed using an Excel spreadsheet.

For multivariate analysis, 73 molecular descriptors were derived from the CDK molecular descriptor calculator. All the molecular descriptors for all three group sets are provided in the Supporting Information.

HCA was executed using the Orange software (version 3.15.0) developed by Bioinformatics Lab at the University of Ljubljana, Slovenia, in collaboration with the open source community.[40] Furthermore, Ward’s method was used for HCA.

PCA was executed using SPSS25 statistics from IBM that were run on three data sets. The first data set had highly cytotoxic elements (HT-69), moderately cytotoxic elements (MT-109), and the 34 reference drugs. The second and third data sets had HT and MT candidates along with 34 reference drugs.

Scaffold Analysis

Scaffold analysis was performed using the Marvin sketch 18.20 module from ChemAxon software solutions. The Bemis–Murcko loose framework was selected for the extraction of scaffolds for a given molecule.