Characterization of novel natural compound derivatives with cancer-selective cytotoxicity

and but (aMan1) and Paeonol-1 more eciently induced cytotoxicity in HCT116, HT-29, and SW48 colorectal cancer cell lines than the parental compounds. Both, aMan1 and Pae1 arrested HCT116 cells in G1 and HT-29 and SW48 cells in G2/M phase of the cell cycle. aMan1 and Pae1 induced selective transcriptional responses in CRC cells involving genes related to metabolic stress and DNA damage response signaling pathways. Both aMan1 and Pae1 induced apoptosis in human cancer cells and organoids derived from tumor tissue without affecting the viability of human non-cancer cells and intestinal organoids derived from healthy tissue.


Background
Colorectal cancer (CRC) is the third most prevalent cancer in the world a malignancy that is frequently caused by life style, diet, and genetics [1,2]. CRC is also an age-related disease with increasing incidence rate with aging, especially after the age of 50. Currently, CRC has a very differential prognosis. At the early stage of diagnosis, the ve-year survival rate of the patient is 90%. However, if diagnosed at a metastases stage, the ve-year survival rate drops to 10% [3]. The current medical CRC treatments usually include a combination of multiple chemotherapeutic, targeted and/or immunotherapeutic drugs [4,5]. However, such therapies are often still ine cient to completely cure CRC and have numerous and severe side effects on the patients health. The side effects of drug-mediated therapies depend on the type, dose and duration of the treatment and are usually caused by the lack of a drug's cancer-speci city. Due to these severe side effects, chemotherapy is often designed taking in consideration the health status of the patient and the medical history. In recent years, a large effort has been devoted to provide psychological and medical support to patients dealing with severe side effects [6][7][8]. In summary, the high morbidity and mortality associated with CRC and the ine cacy of the current available drugs to selectively target cancer cells increases the demand to nd novel cost-effective anti-cancer agents.
Natural compounds have emerged as economic, practicable and effective therapeutic approaches for treatment of cancer. Natural compounds (phytochemicals) are substances with potentially bioactive properties produced by microbes or plants. Phytochemicals have been largely shown to suppress carcinogenesis in vitro studies and in pre-clinical models. Almost half of the approved chemotherapeutic drugs are derived from natural compounds or their derivatives [9][10][11][12]. One class of chemical natural phenolic compounds, the Xanthonoids, have shown potential anti-cancer, anti-in ammatory, and antioxidative properties [13,14] Among xanthonoids, the compound a-Mangostin derived from G. mangostana has been shown to have a broad functional activity. For example, a-Mangostin induces a variety of pharmacological functions: anti-oxidant, anti-carcinogenic, and anti-diabetic. Among them, the anti-cancer activity is the most promising [15]. a-Mangostin affects the growth of the tumor cells in vitro and in vivo including in high-grade malignancies. a-Mangostin inhibits the migration and invasion, and it reduces the actin cytoskeleton of human lung cancer cells, thereby it exhibits anti-metastatic activities [16]. a-Mangostin inhibits the activation of TAK1-NF-κB pathway, thus acting as an anti-in ammatory compound [17]. It also induces mitochondrial dysfunction [18]. Furthermore, it induces apoptosis and cell cycle arrest in human colon cancer cell lines DLD-1, HCT116 and HT-29 [19] and blocks tumor growth in mouse xenograft models [20].
Another class of natural phenolic compound Paeonol (2-hydroxy-4-methoxyacetophenone) is a bioactive component isolated from the root bark of P. suffruticosa Andr [21][22][23]. Paeonol has been widely used as an anti-in ammatory drug for repairing oxidative damage and enhancing immunity function. It also reduces the severity of liver brosis and prevents ox-LDL-induced endothelial cell apoptosis [24]. Paeonol inhibits the growth of colorectal cancer cell lines HCT116, SW620, and it down regulates the expression of COX-2 and PGE2 synthesis in colorectal cancer LoVo cells [25]. Paeonol signi cantly lowers the tumor growth and causees tumor regression in a gastric cancer cell line MFC tumor-bearing mice [22].
Despite the broad functional activity of a-Mangostin and Paeonol, both of these compounds have an intrinsically low solubility and low membrane permeability that are chemical properties that made them fail to reach clinical applications. We speculate that an enhanced solubility, improved cell membrane penetration and/or by adding new functional groups may potentiate their anti-cancer activity and selectivity as well as increase their therapeutic potential.
This study aimed to improve the cancer cytotoxicity and selectivity of a-Mangostin and Paeonol by addition of new functional groups. As the cytotoxicity of parental compounds have been tested on colon cancer cell lines and our lab is mainly focused on colon cancer, we investigated the effect of two very promising a-Mangostin and Paeonol derivatives against human colon cancer cell lines and human intestinal organoids. We found that the two derivatives showed a much lower toxic effect on not transformed human cells with respect to their parental compounds. Our ndings increase the knowledge about natural compounds derivatives as anticancer compounds and open new research options on the derivation of lead compounds such as novel CRC chemotherapeutic drugs that selectively target cancer, but not healthy cells.

Apoptosis analysis by ow cytometry
The cells BjhTERT, HCT116, HCT116 TP53_KO, HT-29 or SW48 cells (1X10 5 cells/ well) were seeded on a six-well plate. Then, the cells were incubated either with aMan (25 μM), aMan1 (25 μM), Pae (300 μM), or Pae1 (10 μM) for 48 h. After incubation, the adherent cells were trypsinized and harvested including the oating cells in culture medium. The cells were washed once with PBS and incubated in 100 μl of 1X Annexin V binding buffer for 20 minutes on ice. After washing, the cells were resuspended in PBS with 0.1% FBS containing DAPI and analyzed by ow cytometry with FACScanto™.
DNA Fragmentation assay SW48 cells (1X10 5 cells/ well) were seeded in a six-well plate, and the cells were treated with aMan1 (25 μM) or Pae1 (10 μM) for 72 hrs. After incubation, the adherent as well as oating cells were harvested and genomic DNA was isolated with the use of DNeasy blood and tissue kit (Qiagen). Genomic DNA (2 μg) was incubated with 2 μl of RNase cocktail. The genomic DNA of (1 μg) was resolved on 2% Agarose gel containing ethidium bromide. The image was acquired using Gel Doc (Bio-Rad).
Confocal Microscopy SW48 cells (1X10 5 cells/ well) were cultured in a six-well plate on glass cover slips in RPMI medium supplemented with FCS (10%). Then the cells were treated with aMan1 (25 μM) or Pae1 (10 μM) for 72 hrs. Followed by incubation, the cells were xed with paraformaldehyde (4%) and permeabilized with triton X100 (0.2 %) in PBS for 10 minutes. After washing with PBS, the cells were blocked with 5 % skim milk powder overnight and the nuclei were stained with DAPI for 1 h hour at room temperature. After three washes with PBS, the slides were mounted on glass slides using DPX mounting medium and cured for overnight. The images were captured using a Zeiss Apotome.

Cell cycle analysis
To investigate the effect of natural compounds on cell cycle, BjhTERT, HCT116, HT-29, or SW48 cells supplemented with 5% milk for 2 hours at RT, and then incubated with p53 WT speci c antibody (Sigma, A5316) and Beta actin (Santa Cruz, sc126) mouse antibodies in T-PBS + 5% BSA overnight at 4°C. After incubation with primary antibody, the membranes were washed six times with T-PBS for 30 min at RT, and then incubated with secondary anti moue HRPO antibody (1:3000) in T-PBS + 5% milk for 1 hour at RT. After incubation with secondary antibody, the membranes were washed six times with T-PBS for 30 min at RT, and then developed on Amersham Imager 600 (GE Healthcare).

Human intestinal crypt isolation and cultivation
Human intestinal crypts were isolated from either tumor tissue or adjacent healthy tissue from the same patient and cultured to intestinal organoids as previously shown [33].
Then the images of organoid were captured with the use of Zeiss AxioCam MRc 5 (Carl Zeiss). Live and dead organoids were enumerated by their morphological appearance as described [34].
Quanti cation of organoids cell death by ow cytometry Fastq les quality check was performed using FastQC v0.11.5. The fastq les were mapped to the hg19 genome using TopHat v2.1.0 [35] with the following parameters --bowtie1 --no-coverage-search -a 5. The number of reads covered by each gene is calculated by HTSeq-Count 0.11.2 [36] with -s no -a 0 -t exon -m intersection-nonempty parameters and hg19 gencode.v19 annotation. Before further analysis, all of the rRNA genes are removed from the count data. For calculating differentially expressed genes and normalized count, DESeq2 R package v1.20.0 [37] was used with the default parameters. For Pearson correlation analysis, principal component analysis (PCA), gene set enrichment analysis (GSEA) and plotting the expression, the normalized count (DESeq2) was used. For PCA and Pearson correlation, only the genes with more than 10 counts in at least 3 samples and with the minimum inter quadrille range (IQR) of 1.5 in log2 transformed normalized counts were used for calculation. For functional and pathway analysis, DESeq2 differentially expression analysis (adjusted p-value<0.01 & | log2 fold change | >=1) results were uploaded in IPA (Ingenuity Pathway Analysis v45868156 [38]. For gene set enrichment analysis, normalized counts (for each gene in all of the samples) were scaled using scale function in R (with center = TRUE, scale = TRUE parameters). The z-score was calculated by multiplying the scaled counts by +1 or -1 which shows the expected direction (+1 for up-regulated genes and -1 for down-regulated genes). For gure 5F and 6F, +1 is used for all of the genes in the 3 gene sets (same genes as gure 5E) except CDKN2B which -1 is used. For the canonical pathways in gure S4C and S5A, the gene list and the directions are extracted from DEGs (adjusted p-value<0.01 & | log2 fold change | >=1) in comparison between SW48 cells treated with aMan1 versus DMSO, and for gure S4D and S5B, DEGs from the comparison between SW48 cells treated with Pae1 versus DMSO is used. The average of z-scores were calculated for each group (one value for each gene per group) and used for plotting and statistical test.

Statistical Analysis
The statistical signi cance between two groups were analysed by two-tail unpaired T-test followed by Holm-Sidak correction for multiple comparison using Graphpad Prism 5 (GraphPad Inc., La Jolla, CA, USA). Differences with p < 0.05 were considered signi cant, and statistical signi cance is shown as *p < 0.05, **p < 0.01, ***p < 0.001.

Results
Cell viability screening of a-Mangostin, Paeonol, and their derivatives in CRC cell lines.
aMan1 with an aldehyde group was prepared to leave the xanthone core intact, to improve solubility and to reduce the cytotoxicity. Although Paeonol exhibits good anti-in ammatory/ oxidative activity and low cytotoxicity, it barely showed an anti-cancer effect. Moreover, its low solubility reduced its usefullness. To increase the solubility with facile synthesis, the boronic acid group was used. More importantly, when attached to a sugar (ex: a fructose), the boronic acid-fructose group can increase the selectivity of a compound concerning tumor cells. This idea is based on the design concept of boronophenylalaninefructose (BPA-fructose), an agent applied in a clinical trial of boron neutron capture therapy. Even without the attachment of a sugar moiety, the strong a nity of a boronic acid group to the cell membrane can enhance the bioactivity of a compound that has interactions on membrane proteins expressed more on tumor cells. In brief, we tried using a boronic acid group to improve solubility and, once needed, to attach a sugar moiety for "indirectly" increasing the selectivity over tumor cells. The overall cytotoxicity of Paeonol will be increased when transformed into a chalcone, but the selectivity over tumor cells became greater due to a boronic acid group allowing lower dosages.
Pae showed cytotoxic effects on cancer cell lines, but required a high concentration (300 μM) to reduce the viability of HCT116, HT-29, and SW48 cells by ~90% (Figure 2c). Pae1 required a consistently lower concentration (25 μM) to reduce the viability of HCT116, HT-29, and SW48 to 91%, 75%, and 86%, respectively (Figure 2d) and showed a higher cancer cell speci city compared to the parental compound. The other Pae derivatives showed no evident higher cancer-speci c cytotoxicity than the parental compound ( Figure S2a, b, and c). Therefore, Pae1 displays more effectiveness at a low concentration and more selectivity than the parental compound against human colon cancer cell lines. To further understand whether the cellular death was mediated by an apoptotic program, we performed FACS analysis with Annexin V staining (Figure 3c, d and S3b). In all of the three CRC cell lines the derivative compounds showed a signi cantly higher induction of apoptosis, while no apoptosis was observed in the BjhTERT cells (Figure 3c, d and S3b). Pae1 also allowed us to use a lower concentration compared to its parental compound (Figure d, bottom panel and S3b). Analysis of DNA fragmentation by agarose gel and DAPI uorescence imaging con rmed the induction of apoptosis following aMan1 and Pae1 treatment (Figure 3e, and f). Thus, aMan1 and Pae1 displayed a signi cantly higher apoptosisinducing activity in CRC cells, but not in non-transformed cells.
Characterization of the effect of a-Mangostin, Paeonol, and their two enhanced derivatives on the cell cycle in CRC cell lines.
As an apoptotic program may be induced by the arrest at speci c cell cycle phases, we analyzed the cell cycle status by ow cytometry (Figure 4). Both aMan and aMan1 arrested HCT116 cell lines in G1 phase of cell cycle (Figure 4a, quanti cation in b). Interestingly, these two compounds arrested HT-29 and SW48 cell lines in the in G2/M phase of cell cycle (Figure 4a, and b). However, we also observed, especially in HT-29 and SW48 cells, an increase of the overall DNA content indicating a potential appearance of polyploid cells perhaps due to events of endoreduplication suggesting that these two compounds may To better characterize the molecular phenotype, we performed transcriptomic analysis (RNA-seq) of the SW48 cells treated with aMan1 or Pae1 for 24 hours. Hierarchical clustering heat-map of Pearson correlation of whole-transcriptomes showed that the Pae1 or aMan1 treated SW48 cells clustered separately with respect to the untreated cells (2 replicates for each condition), in contrast to the treated and untreated BjhTERT cells which clustered together (Figure 5a and b). PCA analysis of the datasets further con rmed that BjhTERT were not transcriptionally affected by aMan1 treatment and slightly responded to the Pae1 treatment ( Figure S4a, and b). These two analyses highlight the fact that the Pae1 or aMan1 treatments induced a more massive transcriptional response in cancer cells than in nontransformed cells. The nding is con rmed by the number of signi cantly differentially expressed genes (DEGs) following treatments that was about ten times higher in SW48 cells than in BjhTERTcells ( Figure  5b).
We performed Gene Ontology (GO) analysis of the DEGs in cancer cells and we found that both aMan1 and Pae1 treatments induced transcriptional regulation of genes involved in cell cycle regulation and DNA damage response con rming the results of the experiments described in gure 3 and 4 (Figure 5c and d). In particular, proliferation-associated genes (e.g. CCND1, CCNA2, PCNA) were strongly downregulated, while cell cycle inhibitors like CDKN2B is upregulated in cancer cells after treatments (Figure 5e and f -left panel). Pae1, but not aMan1 slightly affected these genes also in the BjhTERT (Figure 5e and f -left panel). Remarkably, cancer cells treated with both the compounds showed a strong upregulation of genes involved in DNA damage response and in the EGR/NF-kB pathways (Figure 5e and f -middle and right panels) suggesting that NF-kB, EGR and GADD45 genes may be sequentially activated upon DNA damage following treatment with the compounds [29]. Interestingly this pathway is only activated in cancer, but not BjhTERT cells. GO analysis also showed enrichment of signaling pathways involving genes that might be regulated in metabolic stress response. In particular EI2F signaling, mTOR signaling and protein ubiquitination pathways may indicate an endoplasmic reticulum (ER) stress in cells treated with aMan1, while oxidative phosphorylation, sirtuin signaling and mitochondrial dysfunction pathways suggest the occurrence of a mitochondrial stress response in cells treated with Pae1 (Figure 5c and d). Geneset enrichment analysis of these pathways showed that ER stress response related genes were speci cally enriched only in SW48 cells after aMan1 treatment, while mitochondria stress response associated pathways were speci cally enriched in cancer cells treated with Pae1 ( Figure S4c, and d).
Cytotoxic activity of a-Mangostin-1 and Paeonol-1 and cellular stress response following treatment is largely maintained in TP53 de cient CRC cells.
Since we found enrichment of the P53 signaling pathway (Figure 5c and d) and it has been previously reported that aMan induces apoptosis in a P53 dependent manner [11], we questioned if the P53 pro ciency was a necessary condition to preserve the cytotoxicity and cancer selectivity of the two derivative compounds. Therefore, we tested the aMan1 and Pae1 on HCT116 cells having wild type (WT) or mutant (inactive) p53. Cells with mutant p53 showed a decreased apoptosis following aMan1 and Pae1 treatment (from ~41% to ~33% and ~44 to ~27%, respectively) suggesting that p53 is indeed important in inducing aMan1-mediated apoptosis, but also that additional p53-independent pathways are taking place to carry out this function (Figure 6a and b). RNAseq analysis of the HCT116 cells treated with aMan1 and Pae1 showed that, even though transcriptionally different, HCT116 WT or KO for TP53 responded similarly to the two derivative compounds (Figure 6c). TP53-WT HCT116 cells have a higher basal expression of the TP53 gene and its known targest P21 and P15 INK4b [30] as well as transcriptional upregulation of these genes in contrast to the TP53-KO HCT116 cells (Figure 6e). However, both the HCT116 cell lines have a downregulation of cell-cycle related genes, and upregulation of genes belonging to DNA damage and EGR/NF-kB signaling pathways as previously observed in the SW48 CRC cell line (Figure 6f). Remarkably, ER-stress response pathways were again observed only following aMan1 treatment (in both TP53-WT and TP53-KO HCT116 cells) ( Figure S5a). Mitochondrial stress response pathways were observed in both the cell lines and following the treatments with both aMan1 and Pae1 ( Figure S5b).
These results suggest that aMan1 or Pae1 may induce a cellular metabolic stress involving ER or mitochondria that can lead to DNA damage and activation of both TP53-dependent and -independent cell-cycle arrest and induction of the apoptotic programs. One candidate pathway involved in the DNA damage response is the NF-kB/EGR/GADD45 pathways that has already been demonstrated to work in a TP53-independent manner [31]. Very importantly, neither the metabolic stress nor the potential consequent DNA damage and DNA damage response is shown in non-cancer BjhTERT cells. These analyses revealed the importance of the cancer mutational pro le in modulating the anticancer activity of the aMan1 and Pae1 treatment, but also provide evidence that these compound derivatives may be also used in P53-de cient tumors.
Taken together these results show that aMan1 and Pae1 provoke a strong transcriptional response in CRC cells (not present in non-transformed cells), involving metabolic stress and DNA damage response signaling pathways, ultimately leading to cell cycle arrest and induction of apoptotic programs.
Characterization of the cytotoxic property of a-Mangostin, Paeonol, and their two derivatives in human primary intestinal organoids To investigate the effect of natural compounds on human intestinal organoids, we have isolated the crypts from colorectal tumor tissue as well as normal colonic epithelium from samples (surgery leftover) of the same patient. This procedure allows culturing the organoids side-by-side and testing and screening of drugs for their anticancer activity, and assessing the adverse toxic side effects on normal tissue. To assess whether natural compounds induce apoptosis selectively in tumor cells, human colon cancer organoids and healthy tissue organoids from the same patient were plated in equal number, and organoids were treated with various concentrations of aMan, aMan1, Pae or Pae1 for 48 h (Figure 7a). We used irinotecan (a camptothecin (CPT) analog) that is currently used as chemotherapeutic drug for CRCs and DMSO as a negative control. At the minimal concentration required to induce cellular death in all of the cancer organoids, irinotecan, aMan and Pae were also inducing cellular death in ~50% of the colon organoids derived from the cancer surrounding healthy tissue (Figure 7a, quanti cation in b). Importantly and re ecting the cell viability assays in the CRC cell lines, aMan1 and Pae1 did not show any cytotoxic effect in the healthy colon tissue cultures (Figure 7a, quanti cation in b). Furthermore, the cell death induced by the natural compound was analysed with the use of ow cytometry ( Figure S6). irinotecan, Man and Pae compounds induced cell death in cancer organoids and also healthy colon organoids. However, Man1 and Pae1 induced cell death only in cancer organoids without inducing any cytotoxicity in healthy organoids ( Figure S6).
In cancer organoids, aMan1 and Pae1 induced apoptosis (Figure 7c and d), but no apoptosis was observed in healthy organoids. Therefore, aMan1 and Pae1 showed a higher induction of cell death and apoptosis in cancer organoids but not in healthy organoids.
Taken together these results con rm that our observations from the 2D CRC cell lines are also applicable in ex-vivo human organoids, strongly suggesting that the two derivatives of the natural compounds α-Mangostin and Paeonol have a strong selectivity in targeting only cancer cells and promoting them as potential novel promising compounds for the treatment of CRC patients.

Discussion
Page 12/20 CRC is one the most frequent cancers in both men and women in the world. According to the American Cancer Society, this cancer is the cause of ~50 thousand deaths per year in the USA. Patients with metastatic CRC that are usually treated with multiple therapeutic lines involving potent chemotherapeutic drug cocktails have a higher mortality risk [32]. Most of these drugs have very severe adverse effects and, often, the therapy has to be designed in according to the medical history and the health condition of the patients. In many cases chemotherapy initially works; however, there is often problems of dosage, relapse, and adverse effects indicating an urgent need for more novel treatment options. Therefore, the development of novel chemotherapeutic drugs that selectively kills the tumor cells, leaving the healthy cells alive could provide a potential way for treating CRC. In the last years, many attempts (sometimes successful) have been done to develop more powerful cancer speci c drugs and to reduce the side effects of the treatment. For example, the development of molecular-targeted agents and immunotherapies that are targeting speci c molecules expressed only, or more abundantly, in the tumor cells. Drug development and testing for CRC is continuously evolving (e.g. in USA there currently are more than 50 different clinical trials in more than 1000 locations) and the promise of cancer-speci c therapeutic treatments still remains an unmet medical priority [32].
Natural compounds represent a wide variety of relatively cheap and suitable molecules that often have been used for various therapeutic aims since ancient times; for example, a-Mangostin and Paeonol derived from plants are used in the traditional Chinese medicine. However, these two compounds, although they showed mild anti-cancer properties, never reached a clinical phase because of some physical and chemical features (e.g. low solubility and low membrane permeability) that make them unsuitable as drugs.
In this study, we tested some chemical derivatives of these two compounds aimed to enhance their solubility, cell membrane penetration and/or cancer selectivity. We screened for those derivatives that performed better than the parental compounds, with particular focus on two important features (enhanced cytotoxicity and cancer selectivity) and two of them (named aMan1 and Pae1) were selected for further characterization. FACS analysis of the cell viability, cell death and apoptosis as well as transcriptome pro ling con rmed that these two derivatives have enhanced cytotoxicity and cancercell selectivity. RNA-seq analysis indicated that the two compounds might cause metabolic stress in the cells (probably by inducing ER or mitochondria stress) and to further activate the DNA damage response that leads to cell cycle arrest and apoptosis of cancer cells. We also showed that aMan1 and Pae1 have similar activity (slightly lower) in cancer cell lines de cient for TP53, one of the major players in mediating cell apoptosis probably through the P53-independent signalling of the NF-kB/EGR/GADD45 pathway [31]. Finally, we showed that aMan1 and Pae1 are able to kill organoids from colon cancer, and do not show any toxic effect on colon organoids from healthy tissue. Therefore, they performed much better than the parental compounds as well as some currently used chemotherapeutic drugs like irinotecan.

Conclusions
In this study, we have screened a series of natural compound derivatives to nd those that improved toxicity against cancer cells with respect to the parental compounds and to currently used chemotherapeutic drugs for the CRC treatment (like Irinotecan). Our data revealed two natural compound derivatives (named aMan1 and Pae1) high cancer-selectivity and cytotoxic activity. Cell viability assays, FACS analysis and transcriptome analysis con rmed that these two compounds induce cell death and apoptosis in CRC cell lines and in organoid derived from human colon tumors, but not in non-cancer human cells and in organoids derived from healthy colonic epithelium. Taken together, our data promote these two natural compound derivatives as potential promising anticancer compounds and provide breakthrough, novel ndings on functional groups that can be used in drug development. Additionally, these results point out that the selective targeting of cancer cells can be reached and it is a reasonable aim during drug design and lead compound optimization. We think that the identi cation of cancerspeci c molecules and pathways is one of the most desirable targets in cancer research and our data can open new horizons in this direction.