**Development of Dietary Supplement Label Database in Italy: Focus of FoodEx2 Coding**

**Alessandra Durazzo 1,\* , Emanuela Camilli <sup>1</sup> , Laura D'Addezio <sup>1</sup> , Ra**ff**aela Piccinelli 1, Angelika Mantur-Vierendeel 2, Luisa Marletta <sup>1</sup> , Paul Finglas 3, Aida Turrini <sup>1</sup> and Stefania Sette <sup>1</sup>**


Received: 20 November 2019; Accepted: 23 December 2019; Published: 27 December 2019

**Abstract:** The sector of food supplements is certainly varied and growing: an ever wider offer of new products is launched on the market every year. This is reflected in new reorganization of drug companies and new marketing strategies, in the adoption of new production technologies with resulting changes in dietary supplements regulation. In this context, information on composition reported in labels of selected dietary supplements was collected and updated for the development of a Dietary Supplement Label Database according to products' availability on the Italian market and also including items consumed in the last Italian Dietary Survey. For each item, a code was assigned following the food classification and description system FoodEx2, revision 2. A total of 558 products have been entered into the database at present, trying to give a uniform image and representation of the major classes of food supplements, and 82 descriptors have been compiled. Various suggestions on how the number of FoodEx2 system descriptors could be expanded were noted during the compilation of the database and the coding procedure, which are presented in this article. Limits encountered in compiling the database are represented by the changes in the formulation of products on the market and therefore by the need for a constant database update. The database here presented can be a useful tool in clinical trials, dietary plans, and pharmacological programs.

**Keywords:** Dietary Supplement Label Database; dietary supplements; food description; food classification; FoodEx2

#### **1. Introduction**

The sector of food supplements is certainly varied and growing: a wider and wider selection of new products is launched on the market every year. This is reflected in new reorganization of drug companies, in new marketing strategies, and in the adoption of new production technologies with resulting changes in the dietary supplements regulation. The growth of this sector is encouraged by growing interest of consumers in improving their health and physical and mental wellbeing, often to compensate for an incorrect lifestyle [1].

Dietary supplements are considered in epidemiological studies and in the analysis of food consumption patterns [2]. There are several implications in dietary adequacy assessment especially with regard to the issue of upper limits in daily intake of certain nutrients. Moreover, several factors may influence the use of dietary supplements, such as gender, age, socio-economic status, educational level, dietary habits, etc. A first attempt to harmonize information on food supplements between European countries was performed by EFSA [3] with the purpose of producing a food

composition database including both foods and food supplements to estimate nutrient intakes in European Countries. In this regard, it is worth mentioning some ongoing initiatives such as Global Dietary Database (GDD) (https://globaldietarydatabase.org/) and FAO/WHO Global Individual Food consumption data Tool (FAO/WHO GIFT) (http://www.fao.org/gift-individual-food-consumption/en/) aimed at the harmonization of dietary datasets worldwide for global diet monitoring using a common food classification and description system [4,5]. Considering the importance of dietary supplements in the evaluation of dietary intake, it is worth mentioning in particular the Dietary Supplement Label Database (DSLD) (https://dsld.nlm.nih.gov/dsld/) by the National Institutes of Health [6,7]; at present, it contains label information (brand name, ingredients, amount per serving, and manufacturer contact information) of more than 71,000 dietary supplements present and consumed in the U.S. marketplace [8,9]. The DSLD can be used to track changes in product composition and capture new products entering the market. Browsing options were developed and organized to search by product, ingredient, or contact of manufacturer, representing a unique resource that policymakers, researchers, clinicians, and consumers may find valuable for multiple applications [8,9].

In this context, information on composition reported on the product labels of selected dietary supplements has been collected and updated for the development of a Dietary Supplement Label Database for Italy, according to products' availability on the Italian market and also including items from both the third Italian National Food Consumption Survey, INRAN-SCAI 2005-06 database [2] and the ongoing Italian national dietary survey IV SCAI. The design and construction of a food database requires above all identifying foods through an adequate food nomenclature and a precise description. The FoodEx2 system has been used for the classification and description of dietary supplements in the aforementioned database. FoodEx2 is a standardized food classification and description system developed by EFSA to better describe characteristics of foods and dietary supplements in exposure assessment studies; this system, nowadays at revised version 2, consists of flexible combinations of classifications and descriptions based on a hierarchical system for different food safety-related domains (i.e., food consumption, chemical contaminants, pesticide residues, zoonoses and food composition) [10–14]. This system is characterized by a compromise between comprehensiveness (sufficiently detailed description) and feasibility in different areas of food data collection. In fact, it consists of a fixed and sufficiently large set of food categories or groups (food classification—organization of terms identifying/assigning different food items into groups) defined at high level of detail that constitute the "core list" and represent the minimum recommended level for coding during data collection [15]. More detailed terms can be found on the "extended list"; terms present in the core and extended lists may be aggregated in a hierarchical parent–child relationship in several ways according to different food safety domains. Descriptors, defined "facets", are aimed at registering all relevant food items characteristics and can be used to add details to create new categories responding to particular study requirements.

This work has been undertaken to study the application of FoodEx2 system starting from FoodEx2 categories (or terms) belonging to the FoodEx2 group "Products for non-standard diets, food imitates and food supplements" (A03RQ) for classifying the items that make up the Italian Dietary Supplement Label Database here presented.

#### **2. Materials and Methods**

The starting set of supplements has been drawn from the nationwide dietary surveys including the third Italian National Food Consumption Survey, INRAN-SCAI 2005-06 database [2] and items from preliminary results of the ongoing Italian national dietary survey IV SCAI. National food consumption surveys were designed with the aim of representativeness of the total population at national level and in the four main geographical areas, taking energy intake as the referring parameter.

Subsequently, products' labels had been searched on the internet using the following keywords in Italian: dietary supplements, botanical, herbal formulations, vitamin-based supplements, mineral-based supplements, protein-based supplement, carnitine-based supplements, prebiotic

formulations, probiotic formulations, algae-based formulations, enzyme-based formulations, yeast-based formulations, common supplements.

Afterwards, label surveys visiting retail points to directly observe products on shelves were carried out.

The official register of supplements authorized by the Italian Ministry of Health (http://www. salute.gov.it/imgs/C\_17\_pagineAree\_3668\_listaFile\_itemName\_1\_file.pdf) was consulted.

The coding procedure was carried out by a qualified compiler who constantly follows the FoodEx2 system updates, taking part in training courses organized by system developers [14]. Another qualified compiler double-checked the codes.

Procedurally, information on composition of dietary supplements was taken from labels, and a code was assigned to each item following the food classification and description system FoodEx2, revision 2; the exposure hierarchy was used for coding [10–14]. The FoodEx2 categories (terms) belonging to the FoodEx2 group "Products for non-standard diets, food imitates and food supplements" (A03RQ) were considered for classification of the items.

FoodEx2 system consists of 21 clearly defined food groups. Detailed food groups represent the basis of the systems; a food only fits in one group and a parent–child structure is present within the food groups. Facets descriptors, of which there are 28 in total, can be viewed as characteristics of foods from different points of view; the facets give additional information for a peculiar aspect of food, i.e., part nature, ingredient, packaging material, production method, qualitative information, process, target consumer. Peculiarity of FoodEx2 is that each food group lists term with included implicit facet descriptors, to which further descriptors of different characteristics can be added; during compilation procedure, in FoodEx2 for each food item, the terms may be aggregated in different ways according to the needs [16]. "Implicit facets" means facets proper of the base term chosen for classification, and therefore, implicitly assigned to it, whereas "added facets" means the facet descriptors that are added by the coder to the chosen base term while coding a food item. The procedure consists of organizing them to reduce the coding time and prevent general imprecision.

For each food item, the terms may be aggregated in different ways according to the needs, without following a general scheme; a typical case is given by a base term, followed (optionally) by a hashtag "#" and a sequence of facet descriptors separated by dollar character "\$".

During this practical experience of compiling the Dietary Supplement Label Database, feedbacks and suggestions for possible enhancement of FoodEx2 were formulated and forwarded to system developers. These suggestions can be grouped as "Additional items", "Clarifications", and "Typing suggestions".

#### **3. Results and Discussion**

A total of 558 products have been entered into the database at present, as an attempt to provide an adequate representation of the major categories of food supplements, and 82 descriptors have been compiled. Particular attention has been given to supplements/formulations based on medical herbs and plant extracts, one of the classes currently emerging [17,18].

#### *3.1. Database Description*

Items in the Dietary Supplements Label Database are organized in groups defined by base terms and additional facets.

Table S1 (Supplementary Material) reports ingredients and nutritional composition of the 558 products and Table S2 (Supplementary Material) reports the FoodEx2 codes of the 558 products.

#### 3.1.1. Base Terms

The base terms reported for describing the 558 products are distributed in the subgroups as follows: 73 Mixed supplements/formulations [A03TC], 28 Vitamin only supplements [A03SL], 27 Mineral only supplements [A03SM], 49 Combination of Vitamin and mineral only supplements [A03SN], 6 Bee-produced formulations [A03SQ], 7 Fiber supplements [A03SR], 283 Herbal formulations and plant extracts [A03SS], 14 Algae-based formulations (e.g., spirulina, chlorella) [A03ST], 8 Probiotic or prebiotic formulations [A0F3Y], 15 Formulations containing special fatty acids (e.g., Omega-3, essential fatty acids) [A03SX], 10 Protein and amino acids supplements [A03SY], 2 Coenzyme Q10 formulations [A03SZ], 1 Enzyme-based formulations [A03TA], 4 Yeast-based formulations [A03TB], 10 Other common supplements [A03SV], 3 Protein and protein components for sports people [A03SA], 6 Micronutrients supplement for sports people [A03SB], 7 Carnitine or creatine-based supplement for sports people [A03SC], 2 Nutritionally complete formulae [A03SE], 3 Imitation yoghurt, non-soy [A03TZ].

#### 3.1.2. Facets

Additional facets used for describing the dietary supplements are: FACET F03 "PHYSICAL STATE", FACET F04 "INGREDIENT", FACET F23 "TARGET CONSUMER", FACET F33 "LEGISLATIVE CLASSES". FACET F03 defines the physical state of a product such as: Tablets [A06JH], Powder [A06JD], Liquid [A06JL].

FACET F04 defines the characterizing ingredients. Common terms used are aggregation term Chemical elements [A0EVF], including core terms Calcium [A0EXH], Magnesium [A0EXF], Iron [A0EXD], Potassium [A0EXJ], Zinc [A0EXE], Fluorine [A0F3A]; aggregation term Vitamins [A0EVG], including core terms Vitamin C (Ascorbic acid) [A0EXN], Vitamin D (Cholecalciferol) [A0EXM], Vitamin E (Tocopherols, tocotrienols) [A0EXL, Vitamin A (retinol, carotenoids) [A0EXZ], Vitamin B9 (Folic acid, folinic acid) [A0EXQ], etc.; aggregation term Special fatty acids [A0EVS], including core terms Omega-3 fatty acids [A0EVV] and Omega-6 fatty acids [A0EVT]; aggregation term Phytochemicals [A0EVM], including core terms Phytosterols [A0EVQ], Polyphenols [A0EVP], Carotenoids [A0EVN]; the core term Dietary fiber [(A0EVR]; the extended terms Carnitine [A0F4N] and Creatine-creatinine [A0F4P]; the aggregation term Bee-produced fortifying agents [A0EVH], including the core term Royal jelly [A0CVG]; the aggregation term Live microorganisms for food production [A048X] including core terms Yeast cultures [A048Z]; the core term Caffeine [A0EVK]; the core term Algae-based fortifying agents (e.g., spirulina, chlorella) [A0EVL].

It is worth mentioning that the most used terms to indicate the ingredients present in the food supplements from the category Herbal formulations and plant extracts [A03SS] are Powdered extract of plant origin [A0ETZ], Liquid extract of plant origin [A0EVA], Extracts of plant origin [A0ETY], Dried herbs [A016T], Dried vegetables [A00ZQ], Dried fruit [A01MA], Dehydrated/powdered fruit juice [A03CG], Dehydrated/powdered vegetable juice [A03DA].

Examples of descriptors for FACET 23, used to indicate the target consumers for whom the product is intended are Children's food [A07TL], including the Children's food 4–8 years [A07TM] and Children's food 9–15 years [A07TN]; Infant or toddler's food [A07TF].

Within FACET F33, defining the legislative class, descriptors from the classification defined in the food additives legislation (Regulation (EC) No. 1333/2008) are used for dietary supplements as follows: FA-17.1 Food supplements supplied in a solid form including capsules and tablets and similar forms excluding chewable forms [A0C16], FA-17.2 Food supplements supplied in a liquid form [A0C15], FA-17.3 Food supplements supplied in a syrup-type or chewable form [A0C14].

#### 3.1.3. Groups' Description

Examples of group Vitamin only supplements [A03SL] are dietary supplements containing vitamin D such as the product coded by [A03SL#F03.A06JH\$F04.A0EXM\$F33.A0C16] (Re-coded: Vitamin only supplements, STATE = Tablets, INGRED = Vitamin D (cholecalciferol), LEGIS = FA-17.1 Food supplements supplied in a solid form including capsules and tablets and similar forms, excluding chewable forms) or the one by [A03SL#F03.A06JL\$F04.A0EXM\$F33.A0C15] (Re-coded: Vitamin only supplements, STATE = Liquid, INGRED = Vitamin D (cholecalciferol), LEGIS = FA-17.2 Food supplements supplied in a liquid form); dietary supplements containing vitamin C such

as the product coded by [A03SL#F04.A0EXN\$F33.A0C14] (Re-coded: Vitamin only supplements, INGRED = Vitamin C (ascorbic acid), LEGIS = FA-17.3 Food supplements supplied in a syrup-type or chewable form); dietary supplements containing vitamin B9 such as the product coded by [A03SL#F03.A06JH\$F04.A0EXQ\$F33.A0C16], (Re-coded: Vitamin only supplements, STATE = Tablets, INGRED = Vitamin B9 (folic acid, folinic acid), LEGIS = FA-17.1 Food supplements supplied in a solid form including capsules and tablets and similar forms, excluding chewable forms); etc.

For group Mineral only supplements [A03SM], including all supplements based only on minerals, examples are dietary supplements containing, i.e., iron, potassium, magnesium, zinc, or combination such as potassium and magnesium widespread used. An example of FoodEx2 code of a product containing potassium and magnesium is as follows: FoodEx2 code [A03SM#F03.A06JH\$F04.A0EXJ\$F04.A0EXF\$F33.A0C16], Re-coded: Mineral only supplements, STATE = Tablets, INGRED = Potassium, INGRED = Magnesium, LEGIS = FA-17.1 Food supplements supplied in a solid form including capsules and tablets and similar forms, excluding chewable forms.

Examples for group Combination of vitamin and mineral only supplements [A03SN] that comprises all supplements based only on formulations including both minerals and vitamins are dietary supplements containing vitamin D and Calcium, i.e., FoodEx2 Code: [A03SN#F04.A0EXM\$F04.A0EXH], Re-coded: Combination of vitamin and mineral only supplements, INGRED = Vitamin D (cholecalciferol), INGRED = Calcium; dietary supplements containing vitamin C and iron, i.e., FoodEx2 Code: [A03SN#F03.A06JH\$F04.A0EXN\$F04.A0EXD\$F33.A0C16], Re-coded: Combination of vitamin and mineral only supplements, STATE = Tablets, INGRED = Vitamin C (ascorbic acid), INGRED = Iron, LEGIS = FA-17.1 Food supplements supplied in a solid form including capsules and tablets and similar forms, excluding chewable forms.

The Herbal formulations and plant extracts [A03SS] include any type of supplement based on herbal formulations and/or plant extracts. Typical ingredients are ginkgo biloba, dog rose, star anise, tamarind, aloe, rhubarb, acacia, dandelion, astragalus, psyllium, holy basil, sage and others; these occur as dried products or liquid or powdered extracts. In addition to the classic medical herbs just mentioned, there are also foods with functional components such as artichoke, garlic, pineapple, black currant, whose use has become frequent. An example of coding of artichoke- based product is FoodEx2 Code: [A03SS#F03.A06JL\$F04.A0EVA\$F33.A0C15], (Re-coded: Herbal formulations and plant extracts, STATE = Liquid, INGRED = Liquid extract of plant origin, LEGIS = FA-17.2 Food supplements supplied in a liquid form), and in the Remark is noted: "The ingredient indicated as Liquid extract of plant origin is hydroalcoholic extract of artichoke leaves". For a garlic-based product, FoodEx2 Code is [A03SS#F04.A0ETZ\$F33.A0C16] (Re-coded: Herbal formulations and plant extracts, INGRED = Powdered extract of plant origin, LEGIS = FA-17.1 Food supplements supplied in a solid form including capsules and tablets and similar forms, excluding chewable forms) and in the Remark is noted: "The ingredient indicated as Powdered extract of plant origin is garlic bulb dry extract".

The group Other Common Supplements [A03SV], referring to any type of other common supplements, includes formulations containing, as the main ingredient, compounds such as alpha lipoic acid, beta glucans, lactoferrin, melatonin, etc. An example of product containing lipoic acid is given by the code [A03SV#F03.A06JH\$F04.A0F4M\$F33.A0C16], (Re-coded: Other common supplements, STATE = Tablets, INGRED = Co-factors to metabolism, LEGIS = FA-17.1 Food supplements supplied in a solid form including capsules and tablets and similar forms, excluding chewable forms) and in the Remark is noted "The ingredient indicated as Co-factors of metabolism is α-lipoic acid".

Another example is given by a product containing melatonin coded as [A03SV#F03.A06JH\$F04.A0EVM\$F33.A0C16] (Re-coded: Other common supplements, STATE = Tablets, INGRED = Phytochemicals, LEGIS = FA-17.1 Food supplements supplied in a solid form including capsules and tablets and similar forms, excluding chewable forms) and in the Remark is noted "The ingredient indicated as phytochemicals is melatonin".

The group Mixed supplements/formulations [A03TC] includes any type of supplements combining different principles without a strong prevalence of one. Moreover, various products belonging to this group present bioactive molecules among the ingredients, such as rutin, quercetin, coenzyme Q10. An example is given by product containing powdered extract of plant origin and fish oil, quercetin, vitamin C, vitamin B5, and Methylsulfonylmethane; it is coded as [A03TC#F03.A06JH\$F04.A0ETZ\$F04.A0EVP\$F04.A0EXN\$F04.A0EXT\$F04.A038M\$F04.A0EVV\$F33. A0C16] (Re-coded by Mixed supplements/formulations, STATE = Tablets, INGRED = Powdered extract of plant origin, INGRED = Polyphenols, INGRED = Vitamin C (ascorbic acid), INGRED = Vitamin B5 (pantothenic acid), INGRED = Fish oil, INGRED = Omega-3 fatty acids, LEGIS = FA-17.1 Food supplements supplied in a solid form including capsules and tablets and similar forms, excluding chewable forms) and in the Remark is noted "The ingredients indicated as licorice root dry extract, plantain leaves dry extract, chamomile flowers dry extracts, nettle aerial parts. The ingredient indicated as Polyphenols is quercetin. Methylsulfonylmethane is also contained".

#### *3.2. Feedback and Suggestions for FoodEx2 Revision 2 Implementation: Focus on Dietary Supplements*

Here we report the feedback and suggestions for implementation of FoodEx2 formulated during the development and updating of Dietary Supplement Label Database in Italy. Concerning additional items, supplementary aggregation terms for proteins and amino acids should be added, including core terms for main amino acids used in dietary supplements as well as within "Special fatty acids", extended terms for "Omega-3 fatty acids" and "Omega-6 fatty acids". Considering the widespread growth in the consumption of herbal remedies, additional items, such as powdered dried fruit, powdered dried vegetables and powdered dried herbs would be very useful. At the same time, attention should be given to additional terms linked to description of bioactive compounds; in this order, within the aggregation term "Phytochemicals", several core terms, i.e., alkaloids, nitrogen-containing compounds, organosulfur compounds, should be added, including their corresponding extended terms. Moreover, extended terms should be associated to the core terms just present in FoodEx2 System, "Carotenoids" and "Polyphenols".

In line with technological progress, facet descriptors, i.e., capsules, softgels, opercula, chewable tablets, and gastro-resistant tablets should be considered.

Details on additional items proposed were reported in Table 1.

"Clarification" about "scope notes"(textual information helping describing the selected term) of "protein and amino acids supplements" and "protein and protein components for sports people" should be underlined; differences in "protein and amino acids supplements [A03SY]" and "protein and protein components for sports people [A03SA]" should be clarified as well as if dietary supplements containing fiber with a marked prebiotic activity should be included in "Fiber supplements [A03SR]" or "Probiotic or prebiotic formulations [A0F3Y]". Moreover, several typing suggestions were indicated, i.e., "Chemical elements" should be replaced by "Minerals" and "Fiber" by "Fibre".


**Table 1.** Proposed descriptors for implementation of FoodEx2 distinguished by type of term in FoodEx2 hierarchical structure \*.

\* Proposed FoodEx2 descriptors are in upper case and bold.

#### **4. Conclusions**

A total of 558 products have been entered into the database at present, with the aim of providing an adequate representation of the major classes of food supplements, and 82 descriptors have been compiled.

This paper represents one of first works describing the procedure of coding dietary supplements through the FoodEx2 classification system and could be a useful tool/guide for other compilers and users.

The Dietary Supplement Label Database here presented is intended to be a first example of building a database of information on marketed dietary supplements and provides several suggestions for improving the adopted classification coding system. This database is intended as a basis for a dynamic database that can be expanded as new products are offered on the market. The main feature of a database dedicated to food supplements is its intrinsic dynamism linked to the frequent changes in the formulation of food supplements, with the consequent need to monitor the market and update the database regularly, both by inserting new formulations and expanding the number of descriptors. A precise and available description of the dietary supplements through coding is essential to recognize the type, the main ingredients, and the target consumers by users from different countries.

This database will help consumers to make healthy choices and will represent a valid tool for dietary intake calculations. This database can be useful in different contexts, such as, for example, in clinical trials, dietary plans and pharmacological programs, but also to expand the food composition databases for the purpose of daily nutrient intake estimations.

Considering the integrity of the labels of dietary supplements and whether they reflect the actual amount of each ingredient contained in the product or not, as properly pointed by Betz et al. [19], a new challenge is given by the development of analytically validated laboratory-derived dietary supplement databases. A valid, rapid, and environmental friendly tool in this direction could be represented by the use of infrared spectroscopy joined with chemometrics in the perspective of integrated research approach; as for instance, the development of a "fingerprint spectra database" of dietary supplements could be useful for further researches and applications in the assessment of quality and safety, i.e., monitoring production and/or shelf life of a product, identifying contaminants, and confirming an incoming product.

**Supplementary Materials:** The following are available online at http://www.mdpi.com/2072-6643/12/1/89/s1, Table S1: Ingredients and nutritional composition of the 558 products; Table S2: FoodEx2 codes of the 558 products.

**Author Contributions:** Conceptualization, A.D., A.T. and S.S.; Investigation, A.D., L.M., P.F. and A.T.; Data curation, A.D., L.D., E.C. and S.S.; Validation, A.D., L.D. and S.S.; Writing, Review & Editing, A.D., E.C., L.D., R.P., A.M.-V., L.M., P.F., A.T. and S.S. All authors have read and agreed to the published version of the manuscript.

**Funding:** The present work has been performed within the sub-contract agreement 'ITALIAN AND GREEK FOOD, RECIPES AND DIETARY SUPPLEMENTS COMPOSITION DATABASES AND DAILY REFERENCE VALUES FOR ITALY AND GREECE' (PD\_Manager) awarded by EuroFIR and within the Service Contracts OC/EFSA/DATA/2014/02-LOT1-CT03 ("The children's survey") and OC/EFSA/DATA/2014/02-LOT2-CT05 ("The adults' survey"); "Support to National Dietary Surveys in Compliance with the EU-Menu methodology (fourth support)" call.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Article* **Bioactivity Evaluation of a Novel Formulated Curcumin**

**Se-Chun Liao 1, Wei-Hsiang Hsu <sup>2</sup> , Zi-Yi Huang 2, Kun-Lin Chuang 2, Kuan-Ting Lin 3, Chia-Ling Tseng 4, Tung-Hu Tsai <sup>5</sup> , Anh-Hoang Dao 6, Chun-Li Su 4,7,\* and Chi-Ying F. Huang 1,2,8,\***

	- Fax: +886-223639635 (C.-L.S.); +886-228224045 (C.-Y.F.H.)

Received: 9 October 2019; Accepted: 19 November 2019; Published: 6 December 2019

**Abstract:** Curcumin has been used as a traditional medicine and/or functional food in several cultures because of its health benefits including anticancer properties. However, poor oral bioavailability of curcumin has limited its oral usage as a food supplement and medical food. Here we formulated curcumin pellets using a solid dispersion technique. The pellets had the advantages of reduced particle size, improved water solubility, and particle porosity. This pellet form led to an improvement in curcumin's oral bioavailability. Additionally, we used the C-Map and Library of Integrated Network-Based Cellular Signatures (LINCS) Unified Environment (CLUE) gene expression database to determine the potential biological functions of formulated curcumin. The results indicated that, similar to conventional curcumin, the formulated curcumin acted as an NF-κB pathway inhibitor. Moreover, ConsensusPathDB database analysis was used to predict possible targets and it revealed that both forms of curcumin exhibit similar biological functions, including apoptosis. Biochemical characterization revealed that both the forms indeed induced apoptosis of hepatocellular carcinoma (HCC) cell lines. We concluded that the formulated curcumin increases the oral bioavailability in animals, and, as expected, retains characteristics similar to conventional curcumin at the cellular level. Our screening platform using big data not only confirms that both the forms of curcumin have similar mechanisms but also predicts the novel mechanism of the formulated curcumin.

**Keywords:** curcumin; formulated curcumin; pharmacokinetics; aurora kinase A; hepatocellular carcinoma

#### **1. Introduction**

The use of nutraceutical or functional and medical foods as alternative medicine, in addition to supplementary foods, has been on the rise in recent years [1,2]. The delivery of active ingredients is important in order to obtain beneficial effects for the human body.

Curcumin has been used for many years as a naturally occurring alternative medicine and functional food for the treatment of many diseases. Curcumin is a polyphenol extracted from the rhizome of *Curcuma longa* L., which has phenolic groups and conjugated double bonds [3]. It has strong anti-oxidant, anti-inflammatory, anti-septic, anti-proliferative, and wound-healing properties [4–8]. In addition, curcumin can reverse multidrug resistance of cancer cells, suggesting that it can also serve as a supplement to traditional chemotherapy [9,10]. Several lines of evidence show that curcumin exerts potent anticancer effects against a broad range of human cancer cells, including prostate, colon, breast, ovarian, lung, and liver cancers, and can induce cancer cell apoptosis, for example, in liver cancer cell lines, including but not limited to HepG2, SK-Hep-1, Hep3B, SUN449, and Huh7 cells, with low cytotoxic effects on normal cells [11–16].

Despite curcumin's beneficial effects, its low oral bioavailability (due to its low absorption in the gut because of low solubility in water, fast metabolism by the liver, and rapid systemic elimination) has limited its applications [17–19]. Various methods have been developed to improve the oral bioavailability of curcumin, such as the use of a natural enhancer, a curcumin-phospholipid complex, cyclodextrin and microemulsions, and the development of curcumin analogs [20–22]. In this study, we selected the pellet form, a multiple-unit dosage form, as a vehicle for compound delivery. Pellets disperse freely in the gastrointestinal tract, so they invariably maximize drug absorption [23]. In addition, to improve the solubility of curcumin, we used a solid dispersion technique to formulate pellets. In solid dispersions, the particle size of poorly soluble drugs is reduced and their wettability and dispersibility enhanced, thereby improving their dissolution and absorption rate [24].

The Connectivity Map (C-Map) is a systematic database that establishes the relationship between diseases, genes, and compounds [25]. Recently, the database has been expanded and renamed C-Map and Library of Integrated Network-Based Cellular Signatures (LINCS) Unified Environment (CLUE) (https://clue.io/) [26]. Briefly, C-Map and CLUE use gene expression profiles to describe the biological states of cultured human cancer or normal cells to determine their chemical or genetic constructs (short hairpin RNA [shRNA] constructs). The new, low-cost, high-throughput generic solution for gene expression profiles is termed "L1000." CLUE not only expands the 1309 compounds listed in C-Map to 19,811 small molecules but also includes 5075 shRNA and overexpression genes. In addition, users find CLUE relatively convenient to quickly search for a drug class with a similar mechanism of action (MOA) as a target drug or the same gene family of genetic perturbagens, they codify the class-level annotation required considerable effort, perturbagen classes (PCLs). For example, users can compare their target, such as a disease gene signature or a novel compound, with C-Map and CLUE through pattern-matching algorithms and predict dissimilarities (search for a drug to reverse a disease) or similarities (search for a similar MOA via known compounds). They could upload gene expression profiles to C-Map and CLUE to calculate the connectivity score of each profile. A positive connectivity score would indicate a degree of similar mechanism, while a negative connectivity score would denote the reverse.

HCC is the fifth-most common malignancy worldwide. More than 75% of HCC cases occur in the Asia-Pacific region. The high mortality rate because of HCC is due to the difficulty in diagnosis and poor prognosis. Chemotherapy is a traditional choice for inoperable HCC, but drug resistance limits the therapeutic effect [27,28]. Sorafenib is a multi-kinase inhibitor that targets Raf kinases as well as vascular endothelial growth factor receptor (VEGFR)-2/VEGFR-3, platelet-derived growth factor receptor beta (PDGFR-β), Flt-3, and c-Kit. Because of its potential in providing a survival advantage of two to three months, as per results of two-phase III clinical studies, sorafenib is a Food and Drug Administration (FDA)-approved, first-line targeted therapy agent for treating advanced HCC patients [29,30]. However, the low tumor response rate and side effects of sorafenib indicate the need for investigating other new potential drugs or supplementary foods for HCC [31,32]. In this study, we investigated the anticancer activity of conventional and formulated curcumin and their combination with sorafenib in order to determine whether this combination can induce HCC cell apoptosis and autophagy and inhibit HCC cell proliferation. Formulated curcumin can be used as a functional food and alternative medicine in cancer therapy as it not only causes mitotic defects and cell cycle arrest in cancer cells but also alters chemosensitivity toward anticancer drugs by inducing Aurora-A suppression.

#### **2. Materials and Methods**

#### *2.1. Materials and Methods Used in Manufacturing Formulated Curcumin*

#### 2.1.1. Preparation of Curcumin Solid Dispersion Loaded Pellet

A curcumin standard with a purity > 95.6% was purchased from Sigma-Aldrich.

We used a solid dispersion technique to enhance the solubility and dissolution rate of curcumin [24]. Briefly, the process of making formulated curcumin included dispersing curcumin powder into a solid dispersion solution and spraying it onto sugar spheres. Tumeric extract powder contained 95% curcumin in 80 g (as the active drug). The excipients used in the preparation process of solid dispersion curcumin were Polyvinylpyrrolidone #k30 800 g (PVP K30, as the non-volatile polymer solvent for curcumin) and alcohol 3200 g (as the volatile solvent for curcumin). The turmeric extract powder was mixed with the excipients. The drug-polymer interaction evenly dispersed curcumin in the solvent. Next, the solvent containing solid dispersion curcumin was loaded onto sugar spheres by spray-drying to make solid dispersion pellets; the solvent evaporated during fluid-bed granulation (Figure 1).

**Figure 1.** Preparation of solid dispersion curcumin. Turmeric extract powder, containing 80 g of 95% curcumin (as the active drug), was mixed with excipients (800 g PVP as the nonvolatile polymer solvent in the presence of alcohol as the volatile solvent for curcumin). The drug-polymer interaction evenly dispersed curcumin in the solvent. Then, the solvent containing solid dispersion curcumin was loaded onto sugar spheres by fluid-bed granulation to make solid dispersion curcumin pellets. PVP, polyvinylpyrrolidone.

#### 2.1.2. Measurements of Particle Size and Zeta Potential

Particle size (Z-average, nm), polydispersity index (PDI), and ζ-potential (ZP, mV) of curcumin particles after re-dispersion in water were determined at 25 ◦C by dynamic light scattering (DLS) using a Zetasizer Nano ZS90 (Malvern Instruments, Malvern, UK). Measurements were performed with a detector at a fixed angle of 90◦, in triplicate, and results were shown as mean ± SD. About 60 mg of pellets were dispersed into 2 mL water and centrifuged at 2000 × g for 1 minute to remove starch particles originated from sugar spheres. Supernatant was aspirated to measure size and ZP.

#### 2.1.3. Dissolution Test

A dissolution test was conducted by the US Pharmacopeia 41 basket method (apparatus 1) using a dissolution tester (708-DS Dissolution apparatus, Agilent, USA). The samples of 89 mg conventional curcumin (equivalent to 72 mg curcumin) and 2180 mg formulated curcumin (equivalent to 72 mg curcumin) were placed into 900 mL of dissolution medium containing 1% sodium dodecyl sulfate (SDS) at 37 ◦C ± 0.5 ◦C, under a stirring speed of 100 ± 2 rpm. A 5 mL sample was withdrawn at each time interval (5, 10, 15, 20, 30, 45, and 60 minutes) and was mixed and filtered through a 0.45-μm pore membrane. Then, 2 mL of filtrate was diluted with mobile phase so that the total volume became 10 mL and subjected to HPLC analysis (Agilent 1260, USA). The HPLC program consisted of a mobile phase of tetrahydrofuran: 0.1% citric acid solution (4:6). The column used was 4.6 mm × 20 cm with 5-μm packing L1. The flow rate was 1 mL/minute, and the injection volume was 20 μL using a 420 nm wavelength detector. The percentages of curcumin dissolved from the conventional curcumin and pellets (formulated curcumin) into the medium were calculated and compared.

#### *2.2. Pharmacokinetic Study*

#### 2.2.1. Animal Model

All animal treatment procedures followed the Guide for the Care and Use of Laboratory Animals (National Institutes of Health (NIH) publication, 85–23, revised 1996) as well as the Animal research: Reporting *in vivo* experiments (ARRIVE) guidelines, and were approved by the Animal Research Committee at National Yang-Ming University, Taipei, Taiwan, under Institutional Animal Care and Use Committees (IACUC) approval no: 990103. All surgeries and experimental procedure were carried out under anesthesia with all efforts to minimize animal suffering.

Twelve male Sprague Dawley (SD) rats (270 ± 15 g body weight) were obtained from Bio-Lasco, Taipei, Taiwan. Water was provided *ad libitum*, regardless of administration route. All animals were acclimatized and quarantined in quarantine room of the Rosetta animal facility for about 1 week, and then transferred to feeding room. The humidity and temperature were well controlled as 30%–70% and 19–25 ◦C. The light and dark cycle was set as 12 h: 12 h. Food and drinking water were allowed *ad libitum* during housing.

Rats were randomly divided into two groups treated with curcumin and formulated curcumin. Curcumin (conventional; 500 mg/kg, *n* = 6) and 500 mg/kg formulated curcumin (equal to curcumin 60 mg/kg, *n* = 6) were administrated by gavage to the freely moving rats, respectively. A 300 μL blood sample was collected from the tail vein into a tube rinsed with heparin at 0, 0.25, 0.5, 1, 1.5, 2, 4, 6 h after oral administration.

#### 2.2.2. Sample Pretreatment

Plasma was obtained by centrifuging the blood sample at 4000 rpm for 10 minutes at 4 ◦C. The 10 μL plasma was mixed with 50 μL of the internal standard solution containing 0.1 ng/μL of agomelatine. The samples were vortexed and centrifuged at 13,000 rpm for 5 minutes. The 50 μL of supernatant was transferred to the 1.5 mL tube contained 50 μL solution of 25% acetonitrile and 0.1% acetic acid. After that, 50 μL of the solution was injected onto LC-MS system.

#### 2.2.3. LC/MS/MS Conditions and Data Analysis

Curcumin concentrations in the samples were determined by positive ion electrospray tandem mass spectrometry using multiple reaction monitoring (MRM). Separation of curcumin was conducted on a Cosmosil column (5C18-MS-IIPacked column, 120 Å, 5 μm, 4.6 mm I.D. x 150 mm; NACALAI TESQUE, Inc., Japan) with a mobile phase of acetonitrile-water-formic acid. MS/MS conditions consisted of a declustering potential of 50 V, desolvation temperature of 550 ◦C, spray needle of 5500 V, and collision energy of 30 V.

Pharmacokinetic analysis was calculated using a non-compartmental model with the Phoenix WinNonlin®(Version 8.0) software. The area under the drug concentration-time curve (AUC) was used to measure the total amount of curcumin reaching the systemic circulation. The relative oral bioavailability (BA) of curcumin was calculated according to the following equation: BA (%) = 100 × [(AUCformulated curcumin/doseformulated curcumin]/[(AUCcurcumin/dosecurcumin)]. The pharmacokinetic results were represented as the mean ± SD. Statistical analysis was performed by *t* test (SPSS version 10.0) to compare the differences between groups. The level of significance was set at *p* < 0.05.

#### *2.3. Cell Lines and Cell Culture*

The Huh7 and PLC5 cell lines were obtained from National Taiwan University Hospital, Taiwan. The Mahlavu cell lines were provided by Dr. Muh-Hwa Yang (Institute of Clinical Medicine, National Yang-Ming University, Taiwan). Hep3B cells were obtained from American Type Culture Collection (ATCC), Rockville, MD, USA.

HCC cell lines were cultured in Dulbecco's modified Eagle's medium (DMEM, GIBCO) supplemented with 10% (*v*/*v*) fetal bovine serum (FBS, GIBCO), non-essential amino acids (NEAA, GIBCO), L-glutamine (GlutaMAX™-I Supplement, GIBCO) and 10% penicillin-streptomycin (GIBCO). These cells were maintained in a humidified incubator with 5% CO2 at 37 ◦C and were regularly subcultured every 2–3 days.

#### *2.4. Drug Preparation and Cell Exposure*

The conventional curcumin was prepared as a 30 mM stock solution in dimethyl sulfoxide (DMSO; Sigma) and stored at -20 ◦C. Final curcumin concentrations of 1–90 μM were obtained by dilution in culture medium so that the final concentration of DMSO was less than 1%. Controls contained 0.1% DMSO in all experiments.

The formulated curcumin was prepared as a 30 mM stock solution (based on the weight of curcumin) in double distilled water and stored at -20 ◦C. The formulated curcumin was diluted with cell culture medium to obtain the concentration indicated.

#### *2.5. Proliferation and Viability Assays*

Cells were seeded into a 96-well plate (1500–2000 cells/well) overnight and then treated with curcumin and the formulated curcumin respectively for 0, 24, 48, 72, 96 and 120 h. After treatment, 0.5 μg/mL 3-(4,5-cimethylthiazol-2-yl)-2,5-diphenyl tetrazolium bromide (MTT) was added to each well and cultured for 2 h at 37 ◦C. After incubation, the media were removed from the wells. The formazan crystals formed were then solubilized in DMSO at room temperature for 10 minutes, and then the absorbance was measured in a multimode microplate reader at 570 nm.

#### *2.6. Mitochondrial Membrane Potential Assay*

We employed 5, 5', 6, 6'-tetrachloro-l, 1', 3, 3'-tetraethylbenzimidazolcarbocyanine iodide (JC-1), which was obtained from Cayman Chemical Co., to analyze the mitochondrial membrane potential. The cells were seeded in 96-well black plates at a density of 7000 cells/well and cultured overnight. After treatment, JC-1 staining solution was added to each well and incubated at 37 ◦C for 15–30 minutes in the dark. The plates were obtained by centrifuged at 400 × *g* at room temperature for 5 minutes, and the supernatant was discarded. Then, JC-1 assay buffer was added to each well, followed by centrifugation at 400 × *g* at room temperature for 5 minutes, after which the supernatant was discarded. Finally, JC-1 assay buffer was added again to each well for fluorescent analysis using a fluorescent plate reader.

#### *2.7. Annexin V and Propidium Iodide (PI) Double Staining by Flow Cytometry*

The Huh7 were incubated with various concentrations of conventional and formulated curcumin for 24 h. Annexin V/PI staining was performed to quantify cell apoptosis using an Annexin V-fluorescein isothiocyanate (FITC) Apoptosis Detection Kit (BioVision, Inc., Milpitas, CA, USA) according to the manufacturer's protocol. Annexin V-FITC was then added followed by incubation for 15 minutes in the dark in a 100 μL cell suspension. PI was then spiked into 400 μL Annexin V binding buffer and added immediately to the cell suspension, and subsequently analyzed on a FACScan flow cytometer (BD Biosciences, USA).

#### *2.8. Western Blot*

The cells were incubated with various treatment (conventional and formulated curcumin or combination of sorafenib and conventional and formulated curcumin), and then collected for western blot. Aliquots of cell lysates containing 20–50 μg of protein were separated by SDS-polyacrylamide gel electrophoresis (SDS-PAGE), transferred onto a polyvinylidene difluoride (PVDF) membrane and detected using specific primary and secondary antibodies. The protein bands were visualized by an enhanced chemiluminescence (ECL) detection kit (ImmobilonTM western, Millipore). The membranes were reprobed for β-actin as a loading control. All western blots were carried out at least three times for each experiment. The data were normalized to β-actin. The following primary antibodies were used: anti-extracellular regulated protein kinases (ERK), anti-caspase-3, anti-poly (ADP-ribose) polymerase (PARP) (all from Cell Signaling Technology) and anti-aurora kinase A (AURKA) (BD Biosciences). All antibodies were used at a 1:1000 dilution.

#### *2.9. Cell Cycle Analysis*

After treatment, the cells were collected by trypsinization and fixed in precooled 70% ethanol overnight. The cells were then incubated with PI in the presence of RNase A. The DNA content was analyzed by a FACSCalibur, and the data were analyzed by Flowjo software. The percentage of cells in the sub-G1 was used to indicate the apoptosis rate.

#### *2.10. Analysis the Similar Mechanism of Gene Expression Profiles of Conventional and Formulated Curcumin Using the L1000 Microarray*

Human HCC (HepG2) and human colorectal cancer (HT29) cell lines (ATCC) were treated in triplicate with 20 μM of conventional curcumin or 2 μM of formulated curcumin. Briefly, for experiments using formulated curcumin, 2 μM of curcumin formulation and curcumin excipient were dissolved in DMSO and incubated with HepG2 and HT29 cell lines. The samples were submitted to Genometry, Inc. (Cambridge, MA, USA) for L1000 microarray analysis and to obtain the gene expression profiles of formulated curcumin in HepG2 and HT29. Each set of gene expression profiles consisted of upand down-regulated gene signatures. Subsequently, CLUE was used to decipher the gene signatures in order to uncover potential mechanisms via mapping to compounds with known MOAs. To filter output data, we used a score > 90 for compounds and a score > 70 for PCLs.

#### *2.11. Statistical Analysis*

All values were expressed as the mean ± SD. The data were analyzed using a two-tailed student's *t* test. A *p* < 0.05 was considered as statistically significant.

#### **3. Result**

#### *3.1. Preparation and Evaluation Formulated Curcumin*

First, we prepared the formulated curcumin from powder to pellets with a solid dispersion technique as described in detail in the Materials and Methods section. The pellet size of formulated curcumin estimated by sieving was distributed in the range of 830–1000 μm. Curcumin content quantified by HPLC was 3.3%. Z-average (nm), PDI, and ZP (mV) of curcumin particles after re-dispersion in water were 141.9 ± 5.1, 0.308 ± 0.029 and −2.49 ± 0.39, respectively (Table 1). The dissolution rates of conventional curcumin were 0.00%, 9.80%, 13.60%, 14.05%, 17.28%, 19.41%, 24.06%, and 26.28%, while those of the formulated curcumin were 0.00%, 59.03%, 87.50%, 98.78%, 100.22%, 101.20%, 102.29%, and 103.65% at 0, 5, 10, 15, 20, 30, 45, and 60 minutes, respectively (Supplementary Figure S1). In the 1% SDS medium, more than 85% of curcumin was almost immediately released from the pellets after 10 minutes. Additionally, when compared to the US Pharmacopeia specifications for dissolution of curcuminoid capsules or curcuminoid tablets, the dissolution of the formulated curcumin was not lower than 75% after 60 minutes, suggesting that the formulation greatly increased curcumin's solubility.



*3.2. Oral Administration of Formulated Curcumin Shows an Increase in Bioavailability over Conventional Curcumin via Pharmacokinetic Analysis*

To determine the actual amount of curcumin that was released and existed in the formulated curcumin, a high-performance liquid chromatography assay and LC-MS method was adopted for the quantification of curcumin. To investigate whether the bioavailability of curcumin was increased after formulation, 60 mg/kg of formulated curcumin was orally administered in a rat model and the plasma samples were subjected to chromatography. To confirm the reliability of the method for analyzing curcumin in plasma samples, a method validation was performed. The retention times of curcumin were about 5.14 minutes, with no visible interference peak in the blank plasma chromatogram (data not shown). To perform a pharmacokinetic analysis, the curcumin concentrations in rat plasma at different time points following oral administration of 500 mg/kg of curcumin and 60 mg/kg of formulated curcumin were compared (Figure 2). However, administration of 500 mg/kg of curcumin had very low amount of curcumin in rat plasma and resulted to a huge increase of bioavailability. Therefore, the pharmacokinetic parameters of curcumin represented an estimated number and will be compared with others from literatures (see later in discussion). AUC represents the total drug exposure over time. Based on the pharmacokinetic parameters (Table 2), the AUC represents the total drug exposure over time. The AUC normalized by dose of curcumin was increased from 0.0021 to 1.864, which is an 887.6-fold increase after formulation. Overall, this result showed that oral administration of formulated curcumin significantly increased the oral bioavailability of curcumin compared with conventional curcumin.

#### *3.3. Gene Expression Analysis of Formulated Curcumin and Prediction of Highly Correlated Pathways*

We queried CLUE with regard to the analyzed group of genes in order to identify potential biological functions of formulated curcumin. The top 30 compounds (Figure 3C) and PCLs (Figure 3A) with the highest scores were obtained from CLUE. Intersection of results from both cell lines revealed that seven compounds (e.g., menadione and angiogenesis inhibitor; Figure 3D) and two PCLs (e.g., NF-κB pathway inhibitors; Figure 3B) shared common functions with conventional curcumin. In summary, formulated curcumin was similar to conventional curcumin, and both functioned as, for example, NF-κB pathway inhibitors, which is consistent with previous studies [33]. We intersected two sample groups to identify common PCL/compound classes between formulated curcumin treatment of HT29 cells and

formulated curcumin treatment of HepG2 cells. Consequently, we identified two PCLs (including NF-κB pathway inhibitors and vesicular transport loss of function (LOF; Figure 3B), and seven compounds belonging to the NF-κB pathway inhibitor PCL class were common among all groups.

In fact, CLUE revealed that only the MOA of compounds or shRNAs was similar to that of formulated curcumin; CLUE did not indicate the target of formulated curcumin. Therefore, to obtain more information about formulated curcumin, we used another database for assistance pathway analysis, ConsensusPathDB (CPDB). CPDB comprises interactions among different types of various intracellular information, such as genes, RNA, proteins, and metabolites, to predict a relatively comprehensive and unbiased cellular biology signal result. We used the Venny website to intersect our two sets of PCL results (Figure 3B) and selected their targets and members genes to predict potential pathways of formulated curcumin (Figure 4). According to *q* < 0.001, we listed the top 20 pathways at the bottom of Figure 4, and the details are in Supplementary Figure S2. In addition, we showed similar effects in L1000 microarray profiles between conventional and formulated curcumin in heatmaps (Figure 5). These data suggested that formulated curcumin exhibits similar biological functions as conventional curcumin.


**Table 2.** Pharmacokinetic parameters of curcumin in rat plasma following oral administration.

The data are expressed as the mean ± SD. Cmax: the maximum plasma concentration; Tmax: the time at which *Cmax* is observed; AUC0-t: area under the concentration-time curve from the time of drug administration to the last quantifiable concentration. \*: Conventional curcumin in some rat plasma samples was very low and was assigned to 0 for calculations. We were unable to detect curcumin from the plasma of one rat during the experimental periods, and thus data from five rats were used for calculations.

**Figure 2.** Mean plasma concentration-time profiles of curcumin in male SD rats following 500 mg/kg conventional curcumin (**A**) and 60 mg/ kg formulated curcumin (**B**) after single dose oral gavage (P.O.) linear ordinate. The data are expressed as mean ± SD, *n* = 5 for curcumin and *n* = 6 for formulated curcumin.

#### (A)


**(B)**

**Figure 3.** *Cont*.


**(C)** 

**(D)** 

84

in both HT29 and HepG2 are shown at the bottom (score > 70). (**B**) PCLs list a score > 70 and are intersected by Venny website. Two common PLCs, NF-κB pathway inhibitor and vesicular transport LOF, are shown in the diagram. To avoid missing possible predicted functions of formulated curcumin, we used an intersection-driven approach to broadly cover these PCLs (score > 70). (**C**) The top 30 compounds (CP) are representative, while the complete list is provided in the Supplementary Information (score > 90). Details are in Supplementary Figure S3. (**D**) Intersection compounds using L1000 array analysis of formulated curcumin by CLUE. The connectivity score is based on the Kolmogorov–Smirnov enrichment statistical evaluation of each gene expression profile. The results provided from CLUE are expressed as a comprehensive connectivity score, showing that the same drug has a similar MOA on different cancer cells in CLUE, Connectivity Map and Library of Integrated Network-Based Cellular Signatures (LINCS) unified environment; PCL, perturbagen class; MOA, mechanism of action.

**Figure 4.** Possible pathways of formulated curcumin predicted by CPDB. Genes in two datasets were used to query CPDB in order to predict the pathways in which these genes were likely participating. The Venn diagram shows the intersecting PCLs (top, right). We focused on intersection results indicated by red circles. The results contained two PCLs, NF-κB pathway inhibitors and vesicular transport LOF and used their targets and member gene lists (total 22 genes) to query CPDB in order to analyze interaction network modules, biochemical pathways, and functional information. Top 50 prediction pathways are listed in Supplementary Figure S2, and top 20 pathways identified by CPDB analysis (*q* < 0.001) are shown at the bottom of the figure. The size of each dot denotes the entity number of genes in the pathway. The line between two dots was calculated by the function of these two pathways to indicate the number of genes overlapping said pathways. The breadth of the line denotes the strength of the correlation between two dots. The apoptosis was analyzed in this study (highlighted in yellow, Supplementary Figure S2). CPDB, ConsensusPathDB; PCL, perturbagen class; shRNA, short hairpin RNA.

**Figure 5.** Heatmaps showing the top 50 up/down L1000 probes with similar expression patterns between conventional and formulated curcumin in HepG2 and HT29 cells, respectively. The horizontal axis denotes the treatments in three groups: (1) DMSO, (2) conventional curcumin, and (3) formulated curcumin. The vertical axis denotes the L1000 probe IDs and their corresponding gene names.

#### *3.4. Formulated Curcumin Displays Stronger Inhibition on Population Growth of Huh7 Cells*

To determine whether formulated curcumin retains the biological functions of conventional curcumin, we performed MTT assay to assess the cell proliferation and cell viability of conventional curcumin and formulated curcumin-treated Huh7 cells. As shown in Supplementary Figure S4, when the concentration of conventional or formulated curcumin increased, cell viability decreased after treatment with curcumin for 24–120 h. We also observed cytotoxicity of curcumin in Huh7 cells. The 50% inhibitory concentration (IC50) value for conventional curcumin at 24, 48, 72, 96, and 120 h

was around 90, 90, 60, 45, and 30 μM, respectively, while the IC50 value for formulated curcumin was around 60, 60, 30, 10, and 10 μM, respectively.

#### *3.5. Cytotoxic E*ff*ect of Curcumin on Other HCC Cell Lines*

To determine whether conventional and formulated curcumin could mediate the survival of other HCC cell lines, we first examined the effect of conventional and formulated curcumin on the viability of Huh7, Mahlavu, and PLC5 cells using MTT assay. While Huh7 cells are well differentiated, Mahlavu and PLC5 cells are poorly differentiated and carry *p53* mutations. To explore the cytotoxic activity of conventional and formulated curcumin against these HCC cell lines, we initiated an *in vitro* study by treating Huh7, Mahlavu, and PLC5 cells each with increasing dosages of conventional and formulated curcumin (0, 1, 3, 10, 30, 60, and 90 μM) for 72 h. MTT assay results indicated that both conventional and formulated curcumin significantly inhibit the viability of Huh7 (Figure 6A), Mahlavu (Figure 6B), and PLC5 (Figure 6C) cells. After 72 h post-treatment, formulated curcumin caused cytotoxicity in Huh7, Mahlavu, and PLC5 cells with an IC50 value of 38.1, 10.1, and 60.9 μM, respectively, while the IC50 values of conventional curcumin were 53.3, 35.4, and 82.5 μM, respectively (Figure 6A–C). Taken together, the results showed that both conventional and formulated curcumin inhibited the survival and proliferation of HCC cell lines in a dose-dependent manner. In addition, formulated curcumin was more effective than conventional curcumin.

**Figure 6.** Cytotoxic effect of curcumin and formulated curcumin on other HCC cell lines (Huh7, Mahlavu, and PLC5). After separate treatment with 1, 3, 10, 30, 60, and 90 μM of conventional and formulated curcumin for 72 h, cell viability was determined by MTT assay and expressed as a percentage relative to the control group. Formulated curcumin was more effective than conventional curcumin on the basis of cell viability in the three cell lines tested. Formulated curcumin has higher cytotoxicity in (**A**) Huh7, (**B**) Mahlavu, and (**C**) PLC5 compared to conventional curcumin. #*p* < 0.05; ##*p* < 0.01; ###*p* < 0.005 compared to the control group (formulated curcumin). \**p* < 0.05; \*\**p* < 0.01; \*\*\**p* < 0.005 compared to the control group (conventional curcumin). MTT, 3-(4,5-cimethylthiazol-2-yl)-2,5-diphenyl tetrazolium bromide.

#### *3.6. E*ff*ects of Curcumin on Apoptosis-Related Protein Expression*

MTT assay indicated that both conventional and formulated curcumin suppress cell viability. To determine whether cell viability inhibition is due to apoptosis, we examined the degree of apoptosis by PI and annexin V staining using flow cytometry. The results clearly demonstrated that 24 h post-treatment with 10–60 μM of conventional and formulated curcumin, the percentage of cells undergoing early apoptosis (annexin V+/PI–) and late apoptosis (annexin V+/PI+) increased, while the percentage of viable cells decreased (annexin V–/PI–) in a dose-dependent manner (Figure 7A). Comparison of effectiveness between conventional and formulated curcumin showed no significant statistical difference.

Caspase-3 activation is crucial for mitochondrial-dependent and mitochondrial-independent apoptotic pathways. Therefore, we examined the activity of caspase-3 by observing its active form (cleaved form) by western blotting in Huh7 cells (Figure 7B). Both conventional and formulated curcumin increased caspase-3 activity in Huh7 cells in a dose-dependent manner (Figure 7B). Caspase-3 activation led to cleavage of several substrates, including PARP. PARP cleavage was also determined by western blotting (Figure 7B). Therefore, conventional and formulated curcumin separately induced apoptosis by activating caspase-3 (Figure 7B).

Many antineoplastic drugs induce apoptosis of cancer cells via mitochondrial apoptotic pathways [34–36]. A hallmark of apoptosis induction via these pathways is a rapid, early breakdown of the mitochondrial membrane potential. Therefore, in this study, we also analyzed the mitochondrial function integrity posttreatment with conventional and formulated curcumin. Both conventional and formulated curcumin significantly induced mitochondrial membrane potential breakdown (ΔΨm) in a concentration-dependent manner (Figure 7C), as determined by ELISA using the potential-sensitive dye JC-1.

Previous studies have shown that curcumin inhibits cancer cell proliferation via suppression of the ERK signaling pathway [37–39]. To investigate whether the ERK signaling pathway is involved in curcumin-induced apoptosis, ERK activation was evaluated by detecting ERK phosphorylation. Huh7 cells were exposed to 10–60 μM of conventional curcumin and formulated curcumin separately for 48 h, and ERK activation was determined by western blotting. As shown in Figure 7D, both conventional and formulated curcumin-induced phosphor-ERK down-regulation but with little change in the total ERK protein in Huh7 cells.

Recent studies have suggested that a curcumin-induced mitotic spindle defect and cell cycle arrest in human cancer cells occur through Aurora kinase inhibition [40–42]. To determine whether conventional and formulated curcumin inhibit Huh7 cell proliferation via down-regulation of Aurora-A expression, Huh7 cells were incubated with 0–60 μM of conventional and formulated curcumin separately for 48 h. We observed a significant decrease in the level of Aurora-A by western blotting (Figure 7D). These data suggested that the biological function of both conventional and formulated curcumin involves caspase-3 activation and ERK and Aurora-A down-regulation.

**Figure 7.** Both conventional and formulated curcumin caused apoptosis of Huh7 cells. (**A**) The cells were incubated and treated with conventional and formulated curcumin. After 24 h, the cells were subjected to annexin V/PI staining and analyzed by flow cytometry. Quantitative analysis of PI- or annexin V-positive cells is shown (*n* = 3). \*\**p* < 0.01 and \*\*\**p* < 0.005 compared to the live-cell group without treatment. ##*p* < 0.01 and ###*p* < 0.005 compared to the apoptotic cell group without treatment. (**B**) Huh7 cell lysates treated with conventional and formulated curcumin separately were subjected to immunoblot analysis. The expression levels of cleaved caspase-3, caspase-9, and PARP increased, demonstrating that the cells had undergone apoptosis. (**C**) The mitochondrial membrane potential (ΔΨm) in Huh7 cells was analyzed using the JC-1 mitochondrial membrane potential assay. ΔΨm was lower in HCC cell lines treated with different concentrations (10, 30, and 60 μM) of conventional or formulated curcumin compared to control HCC cell lines (n = 3). \**p* < 0.05; \*\**p* < 0.01; and \*\*\**p* < 0.005 compared to the group without conventional curcumin treatment. #*p* < 0.05; ##*p* < 0.01; and ###*p* < 0.005 compared to the group without formulated curcumin treatment. (**D**) P-ERK and AURKA expression was down-regulated by conventional and formulated curcumin in a concentration-dependent manner, as shown by immunoblot analysis results with anti-ERK, anti-P-ERK, and anti-AURKA antibodies. PI, propidium iodide; PARP, poly (ADP-ribose) polymerase; ERK, extracellular regulated protein kinase; AURKA, aurora kinase A.

#### *3.7. E*ff*ects of Combination of Curcumin and Sorafenib*

Sorafenib and curcumin both inhibited cell viability of two HCC cell lines (Huh7 and Hep3B) in a dose-dependent manner (Supplementary Figures S5 and S6). Therefore, we explored the potential effects of sorafenib in combination with conventional or formulated curcumin on Huh7, Mahlavu, and Hep3B cell proliferation. The cells were treated with various concentrations of sorafenib in combination with conventional or formulated curcumin for 48 h (Supplementary Figures S5A and S6). The results indicated that the combination of conventional or formulated curcumin with sorafenib has a stronger inhibitory effect on the population growth of HCC cell lines.

In this study, we demonstrated that both conventional and formulated curcumin inhibit Aurora kinase, preferentially suppress proliferation, and induce apoptosis of HCC cell lines. A previous study also indicated that curcumin induces apoptosis-associated autophagy [43,44]. Therefore, we investigated whether the combination of sorafenib and conventional or formulated curcumin can increase apoptosis of HCC cell lines. Cell cycle analysis revealed that adding conventional curcumin for 48 h leads to an apoptosis rate of 13.5% (sub-G1 population), similar to the 10.9% rate induced by sorafenib. We further examined whether conventional curcumin induces apoptosis of Hep3B cells. We found that the levels of cleaved caspase-3 proteins increased in conventional curcumin-treated, sorafenib-treated, and combination-treated cells (Supplementary Figure S5C). There was no notable difference between the three groups. Conventional curcumin and its combination with sorafenib both significantly induced accumulation of LC3-II (Supplementary Figure S5D), a lipidated form of LC3 that is considered an autophagosomal marker in mammals. Similar results were found with formulated curcumin treatment (data not shown). These results suggested that the combination of conventional curcumin and formulated curcumin separately with sorafenib induces apoptosis-associated autophagy in HCC cell lines.

#### **4. Discussion**

In this study, we reported that (i) both conventional and formulated curcumin induce apoptosis of HCC cell lines and down-regulate Aurora-A and (ii) a combination of conventional or formulated curcumin with sorafenib has a stronger inhibitory effect on HCC cell viability, demonstrating a possible prevention and therapeutic application for formulated curcumin to be used as a food supplement and medical food.

HCC most often develops and progresses with a lot of oxidative stress and inflammation. Phytochemicals, such as dietary polyphenols, endowed with potent antioxidant and anti-inflammatory properties, provide a suitable alternative for alleviation of HCC. Previous studies have reported that systemic bioavailability of curcumin in humans is very poor [45,46]. Solid dispersion manufacturing of poorly soluble drugs by spray-drying is a practical commercialization strategy to improve solubility and dissolution rates because of the reasonable cost of required materials and ease of scale-up [47].

Several methods have been developed to solve issue of low oral bioavailability of curcumin. One method is the use of a natural enhancer, such as alkaloid piperine. When 20 mg of piperine was given concomitantly with 2 g of curcumin, the serum curcumin increased by 20 times in humans and 1.56 times in rats. However, because piperine is a relatively selective CYP3A inhibitor [14], drug and food interactions with piperine are a concern. Some studies have used phospholipids as a delivery vehicle for curcumin. The oral bioavailability of curcumin-silica-coated flexible liposomes (curcumin-SLs) and curcumin-flexible liposomes (curcumin-FLs) was found to be 7.76- and 2.35-fold higher, respectively, compared to curcumin suspensions [20]. However, large-scale manufacturing will incur a substantial cost.

In the pharmaceutical industry, solid dispersion pellets are one of the drug delivery solutions for poorly soluble drugs. Solid dispersion pellets increase the solubility and release rate of a drug. For example, itraconazole is a triazole antifungal agent with low solubility. In some itraconazole solid dispersion pellets, the formulated itraconazole showed around 30- and 70-fold increase in the dissolution rate compared to pure drug [48]. Tanshinone IIA (TA), one of the liposoluble bioactive

constituents extracted from the root of *Salvia miltiorrhiza* Bunge, has positive cardiovascular functions such as vasorelaxation and cardioprotective effects. The oral bioavailability of TA tSD pellets (a solid dispersion of a combination of PVP and poloxamer 188) in rabbit increased by 5.4 times compared to TA [49]. In this study, the oral bioavailability of formulated curcumin (solid dispersion pellets) increased significantly compared to conventional curcumin. Except for curcumin itself, all the materials used are inert and safe. In addition, because of the simple manufacturing process, solid dispersion pellets are also suitable for continuous and large-scale manufacturing, and the formulated curcumin can be used as a food supplement and medical food.

As mentioned in Section 3.1, 830–1000 μm is the size distribution of the pellets (formulated curcumin) determined by the sieving method. These pellets were formed by spraying ethanol solution of curcumin and PVP-K30 onto sugar spheres in a fluid-bed granulator machine. After re-dispersion in water, they were disintegrated to release free curcumin into the water as nanoparticles. The size of the curcumin particles (Z-average) measured using Zetasizer Nano ZS90 was determined to be 141.9 ± 5.1 nm, which could be filtered by a 0.45-μm pore filter. The nano size of the curcumin particles made it possible for them to be dissolved quickly in the dissolution medium. This was compatible with the data of the dissolution test.

In addition, using 1% SDS solution as the media for the dissolution test is mentioned in the US Pharmacopeia for curcuminoid tablets or curcuminoid capsules. Therefore, we used SDS in the medium to evaluate how well our newly formulated curcumin was improved in its dissolution by comparison with conventional curcumin. The results showed that dissolution of conventional curcumin in 1% SDS solution was only about 20% after 60 minutes, whereas that of the formulated curcumin was more than 85% after 10 minutes.

Supplementary Table S1 summarizes several studies that have reported techniques for enhancing curcumin oral bioavailability. While the formulated curcumin increases bioavailability over 800-fold, only approximately 5–100-fold increases have been achieved by other formulations. To clarify this difference, we investigated several pharmacokinetic parameters of conventional curcumin from these studies (supplementary Table S2) ([50–52]). In our study, administration of 500 mg/kg of curcumin resulted in a very low amount of curcumin in rat plasma: AUC0-t (h × ng/mL) was 1.1, whereas in other studies the AUC0-t (h × ng/mL) were 8.76, 80, and 60 after doses of 50, 340, and 500 mg/kg, respectively, were administered. If we used these AUC0-t (h × ng/mL) from un-formulated curcumin to calculate the AUC for the formulated curcumin, an approximately 8–16-fold increase in bioavailability was obtained. In short, this study showed that oral administration of a novel solid dispersion of curcumin significantly increased its oral bioavailability compared with that of conventional curcumin. Although the Cmax of the formulated curcumin is far below the effective concentrations of our cell culture experiments, administration of conventional curcumin (200 mg/kg) for days or weeks has been reported to exhibit significant biological activity against both chemically induced and xenograft hepatocarcinogenesis [53]. Because the formulated curcumin (single administration of 60 mg/kg) significantly increased the oral bioavailability compared with conventional curcumin (single administration of 500 mg/kg), repeating administration of the formulated curcumin at a higher dosage for a longer period of time can be expected to achieve a much higher Cmax, and especially AUC, to display the biological effect of curcumin observed *in vitro*. It is noteworthy that the formulated curcumin, as expected, retains similar characteristics to conventional curcumin at the cellular level.

It is well known that the low oral bioavailability of curcumin is due to its poor solubility, poor intestinal permeability and extensive metabolism. In another study [52], authors have used phospholipid to formulate liposome of curcumin in order to enhance intestinal absorption of curcumin. In the fasted rat model (rats were fasted overnight to avoid interference by food), they proved that the AUC for liposome of curcumin increased about 5.5-fold compared with that for the conventional curcumin after administration of 340 mg/kg dose; the AUC 26.7 μg × min/mL is equivalent to 445 h × ng/mL. In our study, rats had free access to food and water. After administrating 60 mg/kg of the formulated curcumin, the AUC recorded was approximately 111 h × ng/mL. In addition, AUC/dose

calculated in our study was higher than that in the previous study. Although we enhanced solubility of curcumin, we suggest that after administration of our formulation, curcumin was quickly released as nano particles then dissolved in gastric fluid or tiny oil droplet in food. Therefore, a part of curcumin was absorbed as free form and the rest was absorbed by pathway of oil absorption supported by bile salts. This was one of the advantages of the formulated curcumin which can be prepared easily and scaled up in industry. In future, we will conduct more experiments to verify this conclusion.

Both conventional and formulated curcumin were effective in decreasing proliferation and viability of HCC cell lines in a dose-dependent manner and induced apoptosis of HCC cell lines via mitochondria dysfunction *in vitro*. In the cell culture study, both forms of curcumin were dissolved completely before experiments, thus the benefits of formulated curcumin with improved solubility and dissolution rates could not be displayed. The *in vitro* study was carried out mainly to determine if the formulated curcumin retained similar characteristics of conventional curcumin at the cellular level. Mitochondrial hyperpolarization is a prerequisite for curcumin-induced apoptosis, and mitochondrial DNA (mtDNA) damage might be a probable mechanism for curcumin-induced apoptosis of HepG2 cells and might serve as the initial event triggering a chain of events leading to apoptosis [12]. Aurora kinases, such as Aurora-A, Aurora-B, and Aurora-C, comprise a family of centrosome-associated serine/threonine kinases that are overexpressed in various cancers and are potentially correlated with chemoresistance [54–56]. Curcumin administration [57] or Aurora-A inhibition by short interfering RNA (siRNA) [58] induces apoptosis. Curcumin has also been shown to down-regulate Notch1, the janus kinase (JAK)/signal transducer and activator of transcription (STAT) pathway, and multidrug resistance protein 1 (MDR1) expression and to inhibit histone deacetylase 1 (HDAC1) activity [12,13,42,59,60]. Sorafenib is the only chemotherapeutic drug that has been shown to be effective in prolonging the survival of HCC patients. However, the low rate of tolerance to sorafenib among HCC patients limits its use [31,32]. Recent preclinical studies have reported that combining sorafenib with other chemotherapeutic agents exerts synergistic effects [61,62], which could provide a promising strategy for the treatment of advanced HCC. In this study, conventional or formulated curcumin in combination with sorafenib inhibited the proliferation of HCC cell lines and exhibited a stronger inhibitory effect on HCC cell lines. Taken together, the formulated curcumin appeared to have properties similar to conventional curcumin, raising the possibility that our formulated curcumin could enhance cytotoxicity against HCC.

In addition, our bioinformatics screening platform effectively identified the molecular mechanisms of a phytochemical via gene expression profiles. C-Map can be queried to identify specific gene signatures from small molecules, including FDA-approved drugs. CLUE is similar to C-Map but considerably larger, with > 1.1 million L1000 profiles; therefore, similarity scores for compounds in CLUE can be obtained to identify their molecular actions. Similar results for conventional and formulated curcumin suggested that only solubility and oral bioavailability have been altered in formulated curcumin. Curcumin is known to exert strong anti-inflammatory effects by interrupting NF-κB signaling at multiple levels. Based on CLUE analysis, formulated curcumin can be predicted to have a similar action as NF-κB pathway inhibitors. Furthermore, analysis via CPDB suggests that curcumin can be linked to TNF related weak inducer of apoptosis as well as TNF mediated NF-κB pathway. Therefore, our screening platform not only confirms that the formulated curcumin has similar mechanism with unformulated curcumin, but it also predicts the novel mechanism of formulated curcumin, such as suppression of HMGB1 mediated inflammation by THBD (label in red in Supplementary Figure S2).

In conclusion, our curcumin was formulated in pellet form, which not only improved its oral bioavailability but also provided high flexibility of use. For example, it would be easy to adjust the dosage and combine the pellets with other ingredients. In addition, as a functional food and alternative medicine, it would be suitable for those who cannot swallow tablets or capsules.

*Nutrients* **2019**, *11*, 2982

**Supplementary Materials:** The following are available online: http://www.mdpi.com/2072-6643/11/12/2982/s1. Figure S1. The dissolution rates of conventional and formulated curcumin in 1% sodium dodecyl sulfate medium, Figure S2. Prediction of highly correlated pathways, Figure S3. The output data of compounds (CP, score ≥ 90) was analyzed via CLUE, and detected their similarity among these gene expression profiles, Figure S4. Curcumin and the formulated curcumin affect the cell viability of Huh7 cells, Figure S5. Inhibitory effect of the combination of sorafenib and conventional or formulated curcumin on HCC cell lines, Figure S6. Effect of sorafenib with or without curcumin/formulated curcumin on Mahlavu and Hep3B cell viability, Table S1. Comparison between our study and previous studies on the improvement of solubility/oral bioavailability of curcumin formulation, Table S2. Comparison of pharmacokinetic parameters of curcumin in rat plasma following oral administration.

**Author Contributions:** S.-C.L. and W.-H.H. initiated the studies, performed and analyzed the majority of the experiments. C.-L.S., T.-H.T., and C.-Y.F.H. conceived and designed the experiments. K.-T.L., Z.-Y.H., K.-L.C., C.-L.T., and A.-H.D. performed the experiments and analyzed data. All authors contributed his/her efforts to write the manuscript.

**Funding:** This work was supported by the grants from the Yang-Ming University collaboration project (YM106C024). This work was also supported by grants from the National Science Council, Taiwan (NSC 101-2313-B-003-002-MY3) and the Ministry of Science and Technology, Taiwan (MOST 106-2320-B-003-006-MY3, MOST 107-2320-B-003-002-MY2, and MOST 107-2320-B-010-040-MY3).

**Acknowledgments:** We thank the Everest Pharm. Industrial Co., LTD., Chiayi, Taiwan, for the experiment drug supply and Rosetta Pharmamate Co., Ltd. for helping with animal pharmacokinetic study.

**Conflicts of Interest:** No potential conflicts of interest were disclosed. Competing financial interests statement: The authors declare that there is no conflict of interest.

#### **Abbreviations**


#### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

*Article*

### **Anti-Atherosclerotic Properties of Wild Rice in Low-Density Lipoprotein Receptor Knockout Mice: The Gut Microbiome, Cytokines, and Metabolomics Study**

**Mohammed H. Moghadasian 1,2,\*, Ramandeep Kaur 1,2, Kayla Kostal 1,2, Akhila A. Joshi 1,2, Mahboubeh Molaei 2, Khuong Le 2, Gabor Fischer 3, Francesca Bonomini <sup>4</sup> , Gaia Favero <sup>4</sup> , Rita Rezzani <sup>4</sup> , Branden S. J. Gregorchuk 5, Vanessa Leung-Shing 5, Michelle Wuzinski 5, Andy I. Seo <sup>5</sup> and Denice C. Bay <sup>5</sup>**


Received: 22 October 2019; Accepted: 16 November 2019; Published: 28 November 2019

**Abstract:** Background and aim: We previously reported the anti-atherogenic properties of wild rice in low-density lipoprotein receptor knockout (LDL-r-KO) mice. The present study aimed to discover the mechanism of action for such effects. Materials: Fecal and plasma samples from the wild rice treated and control mice were used. Fecal bacterial population was estimated while using 16S rDNA technology. The plasma samples were used to estimate the levels of 35 inflammatory markers and metabolomics, while using Meso Scale multiplex assay and liquid chromatography-mass spectrometry (LC-MS/MS) techniques. Results: Many bacteria, particularly *Anaeroplasma sp*., *Acetatifactor sp*., and *Prophyromonadaceae sp*., were found in higher quantities in the feces of wild rice fed mice as compared to the controls. Cytokine profiles were significantly different between the plasma of treated and control mice. Among them, an increase in the level of IL-10 and erythropoietin (EPO) could explain the anti-atherogenic properties of wild rice. Among many metabolites tested in plasma of these animals, surprisingly, we found an approximately 60% increase in the levels of glucose in the wild rice fed mice as compared to that in the control mice. Conclusion: Additional studies warrant further investigation of the interplay among gut microbiome, inflammatory status, and macronutrient metabolism.

**Keywords:** wild rice; microbiome; metabolomics; atherosclerosis; LDL-r-KO mice; cytokines; 16S rDNA; plasma; feces; proteins; carbohydrates; functional food

#### **1. Introduction**

Appropriate types of diets and levels of physical activities are believed to be major determinants of maintaining optimal health [1,2]. Many studies have reported that regular consumption of certain

foods, particularly plant-based foods, such as whole grains, fruits, and vegetables, as well as fish, are associated with decreased prevalence of chronic diseases, specifically cardiovascular disease [3,4]. Phytochemicals that are contained within these foods are believed to mediate these health benefits and include phytosterols, dietary fiber, dietary antioxidants, oleic acid, and docosahexaenoic acid (DHA). On the other hand, food ingredients, such as saturated fat, heavy metals, and other contaminants, may increase the risk of cardiovascular disease [5,6]. One of the common chronic diseases with a significant negative impact on the quality of life is atherosclerotic vascular disease, which remains the main cause of global morbidity and mortality [7]. A fundamental contributor in the pathogenesis of atherosclerosis is the oxidation of low-density lipoprotein (LDL) particles, which are taken up by macrophages, initiating foam cell formation in the arterial wall [8]. Therefore, foods with an ability to lower LDL cholesterol and prevent LDL oxidation have been at the center of atherosclerosis prevention [3].

Wild rice has many health benefits when consumed, as noted in historical documents of the indigenous peoples of North America for centuries, as well as other nations, including Chinese and Europeans [9]. Although it is not a grain, wild rice is recognized as a 'whole grain' [10]. Unlike conventional rice, wild rice is usually consumed unprocessed, meaning that wild rice maintains its natural outer layers and contains significantly higher amounts of dietary fiber, micronutrients, and phytochemical compounds. Another important difference between wild rice and conventional white rice is the type of starch they produce [9]. Wild rice contains resistant starch, being often considered to act like a prebiotic; prebiotics are compounds within foods that beneficially affect gut bacterial population and diversity [11]. Gut bacteria produce many metabolites that can either benefit or harm the cardiovascular system [12].

We have previously reported cholesterol-lowering effects and anti-atherosclerotic properties of plant sterols in apolipoprotein E knockout (apo E-KO) mice [13,14]. Over the past few years, we also tested the potential anti-atherosclerotic effects of wild rice in LDL receptor knockout (LDL-r-KO) mice [15,16]. In these studies, we observed significant anti-atherogenic effects of wild rice; however, we were not able to identify a mechanism of action. Atherosclerosis is a multi-factorial disease, in which alterations in inflammatory pathways and oxidative stress, including LDL particle oxidation, play a major role [17]. Furthermore, recent studies reported an association between gut microbiome biology and atherogenesis [18]. Therefore, this study aimed to investigate the impact of wild rice on bacterial species abundance and diversity from 16S rDNA data analysis collected from mouse feces and monitor the metabolic products from the feces and plasma of LDL-r-KO mice.

#### **2. Materials and Methods**

#### *2.1. Animals and Diets*

Sixteen male, four week old LDL-r-KO mice were purchased from the Jackson Laboratory, USA. The animals were kept in pairs while using standard cages and fed regular mouse chow in a controlled environment for one week. After a week of chow adaptation, fasting blood samples were taken from the jugular vein under light anesthesia; body weight was also recorded. Plasma total cholesterol was estimated, and the animals were divided into two groups of treated (*n* = 8) and controls (*n* = 8), as previously reported [15]. The treated group was fed an atherogenic diet that contained 60% (*w*/*w*) wild rice powder, whereas the control group received the same atherogenic diet without wild rice powder, as previously reported [15]. Briefly, the mouse chow diet contained 9% fat that was purchased from Ren's Feed & Supplies Ltd. (Whitby, ON, Canada). This diet was supplemented with 0.06% (*w*/*w*) cholesterol to make it atherogenic; the atherogenic diet was further supplemented with or without 60% (*w*/*w*) wild rice powder and then used for this study. This supplementation was performed by replacing the atherogenic diet by the ground wild rice at 60%. Therefore, the amounts and types of dietary fiber in the control diet and the wild rice diet were not identical. The experiments lasted for 24 weeks.

#### *2.2. Sample Collection*

The blood samples were taken every four weeks. Fecal samples were collected and stored at −80◦C until analysis. At autopsy, final blood samples were taken from the hearts and animals were euthanized while using CO2 gas followed by cardiac puncture [15]. The hearts and aortae were collected for the assessment of atherosclerotic lesion development [15]. The Animal Care Committee approved the study at the University of Manitoba, Winnipeg, Canada; refer to Protocol number 18-048 [15].

#### *2.3. Plasma Cytokine Levels*

Plasma samples that were taken at week 16 of the experiments were used for the estimation of 35 inflammatory biomarkers, using Meso Scale Discovery U-PLEX multiplex assay kit for a mouse (Meso Scale Diagnostics, Rockville, MD 20850-3173, USA) [19]. These markers include interleukins (IL-2, IL-4, IL-9, IL-10, IL-13, IL-17A, IL-17E/IL 25, IL-17F, IL-21, IL-22), tumor necrosis factor-alpha (TNF-α)), TH1/TH2 Combo (IL-1β, IL-5, and IL-12p70A), TH17 Combo 1 (IL-17C, IL-23, and IL-33),TH17 Combo 2 (IL-6, erythropoietin (EPO), IL-27p28/IL-30, vascular endothelial growth factor A (VEGF-A), IL-15, IL-16, and IL-17A/F), interferon gamma-induced protein-10 (IP-10), growth regulated oncogenes (KC/GRO), monocyte chemo-attractant protein-1 (MCP-1), macrophage inflammatory proteins (MIP-1α, MIP-1β, MIP 2, and MIP-3α), granulocyte-macrophage colony-stimulating factor (GM-CSF), and interferon-gamma (IFN-γ). This cytokine analysis was performed on the pooled samples (*n* = 4). MSD SI2400 Imager device and MSD Workbench 3.0 software were used to detect and analyze the standard curves and intensity of the cytokines. The intensity for each biomarker was included in statistical analysis and then reported herein.

#### *2.4. Fecal Microbiome Analysis*

Microbial diversity and species changes in mice that were fed wild rice as compared to controls were estimated based on extracted 16S rDNA from fecal samples that were collected from pairs of mice at weeks, 0, 4, 16, and 24 during the study. Feces from four cages, where each cage contained two mice (eight mice total), were collected (*n* = 4) for each experimental diet group and stored at −80 ◦C. Microbial genomic DNA from each thawed fecal sample were extracted with a QIAamp Fast DNA Stool Mini kit (51604, QIAGEN Inc., Germantown, MD, USA), according to its recommended DNA extraction procedures. Fecal DNA was resuspended in nuclease-free water, where the DNA quantity and quality were assessed while using a Qubit™ dsDNA BR Assay Kit (Q32853, Life Technologies, Carlsbad, CA, USA). Fecal DNA samples were stored at −20 ◦C until they were shipped on dry ice to LC Sciences, LLC (Houston, TX, USA) for 16S rDNA sequencing services. The sequencing methodology that was used by this service and for this study was described previously [20]. Briefly, 16S rDNA sequencing with an Illumina MiSeq platform was carried out, using 338F/806R primers. Further amplification of V3 and V4 regions (around 469 bp in length) was performed by the polymerase chain reaction (PCR). Bioinformatics analysis of 16S rDNA sequence data was assisted by LC Sciences LLC (Houston, TX, USA). Briefly, QIIME software 1.9.1 was used to analyze paired-end reads that were merged into single tags, according to the overlapped region between pairs. The tags were filtered based on their Phred quality score (Q20 and Q30). Chimera sequences that were generated during PCR amplification of the 16S rDNA gene were also excluded, resulting in the final dataset for analysis. This 16S rDNA sequences in the dataset were mapped to the ribosome database project (RDP; http://rdp.cme.msu.edu/) and NCBI 16S rDNA Microbial databases (NT-16S; ftp://ftp.ncbi.nlm.nih.gov/blast/db/nt.gz; as of August 2018) to produce taxonomically annotated sequences, which are referred to as operational taxonomic units (OTUs), described herein. The sequence dataset was grouped using the UCLUST algorithm program. A minimum sequence identity of 99% was used to align the most abundant sequences within each OTU against the reference database sequences, and the hypervariable regions were removed and used to classify the OTUs.

#### *2.5. Metabolomics Studies*

Metabolites from fecal and plasma samples from week 18 of the study were analyzed by a previously described the liquid chromatography (LC)-mass spectrometry (MS/MS) analysis method [21]. This method combines derivatization and extraction of analytes from the samples, and the selective mass-spectrometric detection using multiple reaction monitoring pairs. The isotope-labeled internal standards were used for metabolite quantification. A total of 133 metabolites were included in the full panel. This analysis was performed through a service contract with The Metabolomics Innovation Centre (TMIC) at the University of Alberta, Edmonton, Canada. It is acknowledged that the use of fecal samples from two mice that were housed in one cage is a limitation for microbiome studies as each mouse can behave as a single ecosystem; however, the average changes among multiple mice were the objective of this study.

#### *2.6. Atherosclerotic Lesion Assessment*

Sections from the beginning of the aortae were cut and processed for morphological evaluation of the atherosclerotic lesions, as previously described [15]. The sections were stained with hematoxylin and eosin (H&E) and trichrome. Light microscopy techniques were used for semi-quantitative analysis of atherosclerotic lesions in the wild rice treated and control mice [15].

#### *2.7. Statistical Analysis*

Non-parametric Mann–Whitney tests (also known as the Wilcoxon rank-sum test) and Kruskal–Wallis rank-sum tests were used to calculate the *p*-values and identify significant differences between the two groups of wild rice fed and control mice with an *n* = 4. These statistical analyses were also used to identify significant differences between time course measurements for each animal group when appropriate. Statistical analyses of fecal microbial composition differences were assessed by non-parametric tests, as described by White et al. 2009 [22]. The Venn diagrams of OTUs determined from these analyses were generated while using 'R' statistics software (version 3.6.1, https://www.r-project.org/) 'Venn Diagram' package to show the number of common OTUs in feces of control and wild rice diet groups. Data are presented as means and standard deviations, where *p*-values ≤ 0.05 were deemed to be significantly different based on the degrees of freedom for each sample group. All of the statistical analyses were performed, while using either Microsoft Office Excel (365, Microsoft, Redmond, USA) or the comprehensive 'R' Archive Network (CRAN) statistics software (version 3.6.1, https://www.r-project.org/), with the 'PMCMR' analysis package, using 'kruskal.test' and 'wilcox.test' functions.

#### **3. Results**

#### *3.1. Consumption of Wild Rice Was Associated with Changes in Fecal Bacterial Species Populations*

Insights into microbial taxonomic alterations could only be confidently determined for high abundance OTUs due to the small number of fecal samples (*n* = 4) examined in this analysis. Microbial 16S rDNA analysis identified more than 200,000 bacterial species (OTUs) in the mouse fecal samples. Figure 1A shows a Venn diagram comparing similar OTUs that were observed between the control and wild rice fed fecal samples collected at various weeks 0, 4, 16, and 24. The majority of all OTUs (732 total) shown in the center of the Venn diagram were identical among all diet treatments, as would be expected in a study involving similar mouse breeds and housing conditions. The wild rice diet fecal samples showed a decrease in the number of unique OTUs over time, where 135 unique OTUs at week 0 reduced to 73 OTUs by week 24. The control samples showed no differences in unique OTUs over time, suggesting that the introduction of the wild rice diet reduced species diversity as compared to the control diets.

**Figure 1.** (**A**) Venn diagram comparing 16S rDNA operational taxonomic units (OTUs) from the feces of wild rice and control diet fed mice at weeks 0, 4, 16, and 24. The red curved arrow highlights the decrease in the unique total OTUs in the wild rice fed mice from week 0 to week 24 of the study. (**B**) The relative abundance of the top 20 most abundant bacterial OTUs identified from wild rice or control diet fecal samples at weeks 0 to 24. Identified OTUs are listed according to their order (bottom to top) and color within the bar chart.

Microbial composition changes were further investigated by performing 16S rDNA sequence clustering, where the top 20 most abundant OTUs determined from each fecal sample are shown as a stacked bar chart (Figure 1B). No significant differences were detected between the two major taxa, unclassified Porphyromonadales and Lachnospirales, over time (weeks 0–16 or weeks 0–24) between the control and wild rice diet fecal samples (Figure 1B). Wild rice diets significantly increased (*p* < 0.05) the proportion and appearance of a number of major OTUs when comparing week 0 to week 24 fecal samples; specifically, uncultured *Anaeroplasma sp*. (8.8-fold increase), *Acetatifactor muris* (4.4-fold increase), uncultured *Lactobacillus* sp. (3-fold increase), uncultured *Oscillospira* sp. (3-fold increase)*,* and *Dubosiella newyorkensis* (0.07% appearance) increased (Figure 1B). Losses or significant reductions (*p* < 0.05) in OTUs within the fecal wild rice diet microbiomes after comparing them to the control diet microbiomes were also noted over time (weeks 4 and/or 24). Specifically, reductions in unclassified *Barnesiella* sp. (2-fold reduction), uncultured *Butyrivibrio* sp. (2-fold reduction), and unclassified *Oscillibacter* sp. (2-fold reduction;) were detected. *Bifidobacterium choerinium* was also undetectable in the wild rice diet samples at weeks 4 and 24 as compared to control diet (Figure 1B). Altered proportions of OTUs were also noted within the control diet fecal samples over time (weeks 0 to 16); significant (*p* < 0.05) reductions in uncultured *Anaeroplasma* sp. (undetectable at week 16), uncultured *Ruminococcus* sp. (2-fold reduction), and uncultured *Filifactor* sp. (undetectable at week 16) were noted, as well as

significant increases (*p* < 0.05)in *Ileibacterium valens* (2–5% appearance) and *Bifidobacterium choerinum* (43-fold increase). It is noteworthy that the control diet OTUs, as mentioned above, were either low or completely absent in the wild rice diet fecal samples (Figure 1B). Overall, fecal microbiome analyses indicate that the wild rice diet significantly alters many high abundance bacterial species.

Figure 2 shows values for three OTUs that reached statistically significant differences (*p* < 0.05) between the treated and control animals. The abundance of unclassified *Prophyromonadaceae sp*. and uncultured *Anaeroplasma sp*. in wild rice fed mice were approximately 5000 and 1000, respectively, more than those in the control group.

**Figure 2.** The abundance of OTUs for selected species identified from 16S rDNA analysis of wild rice fed and control fecal samples. All samples were collected at week 16 of the study with an *n* = 4 per group. OTU 4: *Acetatifactor sp.* unclassified; OTU 1064: *Porphyromonadaceae sp.* unclassified; OTU 104999: uncultured *Anaeroplasma sp.* \*: *p* < 0.05 as compared with the controls.

#### *3.2. Wild Rice Consumption Is Associated with Changes in Plasma Inflammatory Markers*

Our analysis included an estimation of 35 different markers in inflammatory pathways. Statistical analyses between data from the wild rice fed and control groups only identified five markers with a significant change in their mean values (Table 1). The levels of EPO and interleukin 10 (IL-10) increased by approximately 109% and 130%, respectively, in the wild rice diet mice. In contrast, wild rice diet mice had reduced markers of approximately 18%, 18%, and 35% of tumor necrosis factor-alpha (TNF-α), vascular endothelial growth factor (VEGF), and interleukin-16 (IL-16), respectively, as compared with the control animals.

**Table 1.** Plasma cytokine intensity from the wild rice fed and control groups.


Data are presented as means ± standard deviation. Statistical analyses were performed using the Mann Whitney test; \*: *p* < 0.05 as compared with the controls. EPO: erythropoietin, TNF-α: tumor necrosis factor-α, VEGF: vascular endothelial growth factor, IL-16: interleukin-16. ↓: Decrease. ↑: Increase

#### *3.3. Wild Rice Diets Show Di*ff*erences in Fecal and Plasma Metabolites*

LC-MS/MS analysis of metabolites that were extracted from wild rice and control diet fecal and plasma samples identified a total of 133 metabolites. We performed a Mann-Whitney rank-sum test on metabolite values between the sample groups to improve the confidence in metabolite analyses due to our lower sampling numbers (*n* = 4). We focused our results on metabolites with significant differences from the control diet group (*p* < 0.05). Table 2 show significant changes in the levels of 11 plasma metabolites that were differentially detected. Glucose increased in wild rice fed mice by approximately 61%, whereas 10 metabolites, including short-chain fatty acids (C8, C10, and C12), medium-chain fatty acids (C14:1, and C16), and long-chain fatty acids (C18 and C18:1) decreased by 17–48% in the wild rice diet plasma samples as compared to those in the controls.


**Table 2.** Metabolomics data from plasma samples of mice fed wild rice and control diets.

Data are presented as means ± standard deviation. Statistical analyses were performed using the Mann-Whitney test; \*, *p* < 0.05 as compared with controls. ↓: Decrease. ↑: Increase

Among 24 fecal metabolites listed in Table 3, only four metabolites, butyric acid and three phospholipids increased by 51–323%. The remaining metabolites, which included amino acids, short-chain fatty acids (except butyric acid), and long-chain fatty acids, showed a decrease of 30–70% by wild rice fed fecal samples as compared to those in the control group (Table 3).



Data are presented as means ± standard deviation; each fecal sample represents feces from 2 caged mice, eight mice total and *n* = *4 fecal* samples per experimental group. Statistical analyses were performed using the Kruskal Wallis test; \*, *p* < 0.05 as compared to the controls. ↓: Decrease. ↑: Increase

#### *3.4. Wild Rice Consumption Prevents Atherogenesis*

In agreement with our previous findings [15,16], we report that the mice fed with wild rice had much smaller atherosclerotic lesions in their aortae as compared to that in the control animals. Figure 3 illustrates advanced atherosclerotic lesions at the beginning of aortae in the control animals (arrows), but similar lesions were absent or minimal in the similar anatomical region of aortae of the wild rice fed mice.

**Figure 3.** Representative photomicrographs were taken at the beginning of aorta from one control mouse (**A**,**B**) and one wild rice fed mouse (**C**,**D**) illustrating atherosclerotic lesions (arrows). As it is seen in (**A**,**B**), atherosclerotic lesions are large and well established in the control mouse (arrows), while such advanced lesions are missing in the wild rice fed mouse (**C**,**D**). H&E staining (**A**,**C**); trichrome staining (**B**,**D**).

#### **4. Discussion**

We have previously shown that wild rice consumption is associated with the prevention of atherosclerotic vascular disease in LDL-r-KO mice [15,16]. This effect could be related to reductions in plasma cholesterol levels. We have shown alterations in LDL-r-KO mice microbiomes may influence the detection of inflammatory markers, and alter concentrations of metabolites when fed a diet rich in wild rice based on the results of our study. LDL-r-KO mice exhibit atherosclerosis, which is known to be an inflammatory disease [23]. Therefore, treatment with agents that possess pro-inflammatory properties are expected to increase the risk for this disease and anti-inflammatory states should prevent atherosclerosis [24]. LDL-r-KO mice fed a wild rice diet had approximately 75% lower atherosclerotic lesions (0.46 <sup>±</sup> 0.11 vs. 1.95 <sup>±</sup> 0.16 mm2) in their aortic roots as compared to the control diet mice [15].

The results from the current study have identified that wild rice feeding is associated with a 130% change increase in IL-10; IL-10 was shown in previous studies to possess anti-atherogenic activities [25]. Another interesting observation was a 109% increase in the levels of EPO in plasma of wild rice fed mice. Recent studies have shown the anti-atherogenic properties for EPO [26]. The mechanism by which IL-10 and EPO levels were increased in wild rice treated animals is not presently understood, but it may be associated with changes in gut microbiome composition. We have reported a beneficial change in the inflammatory pathways of mice that were fed either wild rice or Saskatoon berries [27,28]. Additionally, a recent study monitoring dietary changes in mice demonstrated that specific microbes can alter gut T-cell responses [29]. It is possible that the changes in cytokine concentrations that we

observed in wild rice fed LDL-r-KO mice may indirectly influence plasma when phytochemicals produced by altered gut microbiome species reach the blood.

We also reported that starch from wild rice is different in nature from the starch found in conventional white rice; wild rice also contains a significant amount of dietary fiber [9,16]. These forms of carbohydrates may act as prebiotics, which thereby alters the diversity and population of the gut microbiome [11,12]. In the present study, we observed that unclassified Prophyromonadaceae decreased in the fecal samples of wild rice fed mice (Figure 1B). When we examined different OTUs associated with unclassified Prophyromonadaceae, we observed that many OTUs increased in wild rice group as compared to those in the control (Figure 2). This suggests that specific Prophyromonadaceae, such as unclassified *Barnesiella* sp., differ between control and wild rice fed mice (Figure 1B). Previous studies examining changes in mouse gut microbial species showed that mice that were fed with polysaccharides from the mushroom *Auricularia auricular* altered quantities of Prophyromonadaceae in their intestine as compared to control diet animals [30]. This study highlights the importance of carbohydrates on microbial species diversity. In the same mushroom study, the treated animals showed higher serum IgA and IgG, indicating changes in gut microbiome due to mushroom carbohydrate consumption also modulated the immune system of the mice [30]. Another noteworthy observation was the difference in *Anaeroplasma* sp. between the wild rice fed and control fecal samples. A study by Zeng et al. [31] reported an increased abundance of *Anaeroplasma* species in the intestines of wild type mice that were fed a high-fat diet. These authors concluded that high-fat diets promoted colonic aberrant crypt formation accompanied by an increase in the abundance of opportunistic pathogens, such as *Anaeroplasma* sp. in the colon of C57BL/6 mice [31]. *Acetatifactor* sp. was also identified in high abundance over time within the wild rice fed fecal samples. Although not much is known about *Acetatifactor* species' influence on murine microbiomes, Pfeiffer et al. [32] suggested the name *Acetatifactor muris* due to its isolation from the cecum of mice fed a high-fat diet, which we also observed in our study only among control diet LDL-r-KO mice (Figure 1B). *Acetatifactor* species are not known to metabolize glucose and they are associated with higher phenylalanine arylamidase activities. In our study, we identified that wild rice diets reduced the *Acetatifactor muris* levels, suggesting that this species, might indirectly influence atherosclerosis in an LDL-r-KO mouse model.

Analysis of fecal metabolic compounds revealed that wild rice consumption was associated with altered metabolite abundances, particularly metabolites that are associated with amino acids, carbohydrates, and fats. All amino acids that were detected in fecal samples were significantly reduced from mice fed wild rice diets. Wild rice diets may promote the growth and predominance of these amino acid utilizing species to catabolize more amino acids, since the most abundant bacterial species *Anaeroplasma sp*., *Acetatifactor sp*., and *Prophyromonadaceae sp*. significantly differed in wild rice as compared to the control group. For example, *Acetatifactor muris* may be a species promoting greater amino acid usage, as it possesses phenylalanine arylamidase, which breaks down L-phenylalanine from peptides [32]. This might suggest that a significant reduction in the concentrations of several amino acids could be associated with bacteria containing this and other relevant enzymes.

Among several short-chain fatty acids and their derivatives, butyric acid was found in the fecal materials from the wild rice fed mice 81% more than that in the control animals. This finding coincides with specific and significant increases in butyric acid-producing uncultured *Butyrivibrio* species identified from 16S rDNA wild rice fed fecal microbiomes (Figure 1B). *Butyrivibrio* sp. is commonly enriched in the guts of ruminant animals, where they produce butyrate from the breakdown of plant fibers and structural carbohydrates, specifically hemicellulose [33,34]. Many studies have reported the metabolic benefits of short-chain fatty acids [35,36]. Analysis of blood plasma samples did not correlate well to metabolic and microbiome changes despite increases in butyrate in the intestine of wild rice fed mice. Plasma concentrations of other short-chain fatty acids, such as caproic acid and caprylic acid, were significantly lower in the wild rice fed mice as compared to those in the controls. However, there was an association between the levels of long-chain fatty acids in the fecal and plasma samples, as these levels were increased in both samples from the wild rice fed mice as compared to those in the controls. Another observation was a significant increase in plasma glucose concentrations in the wild rice fed mice as compared to that in the control group. Increased plasma glucose levels are seen during diabetes or insulin resistance in animals as well as humans [37,38]. However, the consumption of high fiber diets is generally recommended to combat complications that are caused by diabetes [39,40]. Wild rice is a rich source of dietary fiber; therefore, this observation seems to be in contrast with our general knowledge and it certainly begs more investigation. To conclude, additional investigations examining the interplay between changes in intestinal microflora, inflammatory response, and metabolic biomarkers are warranted, as they may play a role in the pathogenesis of chronic diseases, like atherosclerosis. Overall, it should be mentioned that a low number of animals, pooled fecal, and plasma samples, as well as a lack of different doses of wild rice, could be viewed as limitations of the present study.

#### **5. Conclusions**

In conclusion, we hereby report that the long term consumption of wild rice at 60% (*w*/*w*) in LDL-r-KO mice is associated with the prevention of atherosclerosis. This effect was accompanied by significant alterations in the fecal bacterial population and diversity, as well as significant changes in several inflammatory and metabolic biomarkers. Of particular interest was an increase in the plasma glucose levels in the wild rice fed mice; currently, we have no explanation for this finding. Other findings that can support anti-atherogenic properties of wild rice are increases in the plasma levels of anti-inflammatory marker IL-10 and EPO. Altogether, this study provides preliminary evidence in support of additional studies on this animal model and others to improve our understanding of how gut bacterial species, plasma inflammatory markers, and metabolic biomarkers may prevent atherosclerosis. Furthermore, a dose-response study can help to establish whether lower doses of wild rice can result in similar findings in this animal model.

**Author Contributions:** M.H.M., R.K., K.K., M.M., and K.L. contributed to the preparation of manuscripts, A.A.J., B.S.J.G., V.L.-S., A.I.S., M.W., and D.C.B. contributed to microbiome study, D.C.B. and R.K. performed statistical analysis, G.F. (Gabor Fischer) contributed to atherosclerosis data, F.B., G.F. (Gaia Favero), and R.R. contributed to overall experimental design.

**Funding:** This study was supported by a grant from the Natural Sciences and Engineering Research Council (NSERC) of Canada to MHM (Grant number 298450). RK is a recipient of the University of Manitoba Graduate Fellowship.

**Acknowledgments:** Supports from St. Boniface Hospital Foundation for the provision of facilities needed to carry out this research are greatly appreciated.

**Conflicts of Interest:** Authors claim no conflict of interest.

#### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

*Article*

### **Intervention Study on the E**ffi**cacy and Safety of** *Platycodon grandiflorus* **Ethanol Extract in Overweight or Moderately Obese Adults: A Single-Center, Randomized, Double-Blind, Placebo-Controlled Trial**

**Ye Jin Kim 1, Eun-Young Kwon 2,3, Ji-Won Kim 2,3, Youngmi Lee 2,3, Ri Ryu 4, Jongbok Yun 5, Manheun Kim <sup>5</sup> and Myung-Sook Choi 2,3,\***


Received: 9 September 2019; Accepted: 5 October 2019; Published: 14 October 2019

**Abstract:** *Platycodon grandiflorus* root extract (PGE) has shown various properties, such as anti-hyperlipidemia, anti-diabetic, and anti-obesity, but mostly in animal studies. Therefore, we conducted a preliminary study on the anti-obesity effect of PGE in 108 Korean adults (aged 20–60 years, 30 kg/m2 <sup>≥</sup> body mass index <sup>≥</sup> 23 kg/m2). The participants were randomly assigned to four groups and were administered the placebo, PGE571 (571 mg as PGE), PGE1142 (1142 mg as PGE), and PGE2855 (2855 mg as PGE), independently, for 12 weeks. Body composition, nutrient intake, computed tomography scan, and plasma adipokines, as well as hepatic/renal function markers, were assessed. The PGE571 group revealed a significant decrease in body fat mass and body fat percentage when compared with the placebo group. Moreover, the total abdominal and subcutaneous fat areas were significantly decreased following PGE (PGE2855 group) supplementation. These results provide useful information on the anti-obesity effect of PGE for overweight and obese adult humans.

**Keywords:** *Platycodon grandiflorus* root; BMI; body fat mass; abdominal fat area

#### **1. Introduction**

Obesity is associated with the morbidity and mortality of diabetes and cardiovascular disease, and it is a major public health problem that is increasing in prevalence worldwide [1]. Obesity increases the risk of metabolic abnormalities, associated insulin resistance, hyperglycemia, type 2 diabetes, and dyslipidemia, which generate high medical costs [2,3]. Obesity leads to the accumulation of body fat mass, as well as the loss of skeletal muscle mass. Recently, a new concept of sarcopenic obesity has emerged, reflecting a combination of sarcopenia and obesity, which describes the process of muscle loss combined with increased body fat as people age—a condition associated with the loss of muscle strength and function, reduced quality of life, and even mortality [4,5].

*Platycodon grandiflorus* (PG) root is widely used in traditional Chinese medicine. It contains many active constituents, such as steroidal saponins, flavonoids, phenolic acids, and sterols, among which, the saponins are regarded as the major active compounds [6]. Numerous studies have proven that the saponins of *P. grandiflorus* exhibit diverse pharmacological activities, such as antioxidant [7], anti-inflammatory [8], and anti-apoptosis effects [9]. These saponins are also believed to have protective effects against some chemically-induced hepatotoxic reactions. Indeed, platycodin D, the major triterpenoid saponin found in PG, was recently suggested for protecting against liver damage caused by ethanol and carbon tetrachloride [10]. In addition, PG is considered to be a legal medicine and dietary supplement; it is also frequently used as an ingredient in health foods and vegetable dishes [11].

Many studies have reported the effects of PG fractions or extracts on lipid metabolism regulation and insulin resistance improvement, but mostly in animal models [11–13]. Therefore, we conducted a preliminary study on the anti-obesity effect of PG extract (PGE) in overweight and obese adult humans. This is the first study to investigate the body fat loss effects of PGE in humans. In particular, this preliminary study provides information on the approximate daily dosage of PGE for the upcoming main study, which will be performed using dual-energy X-ray absorptiometry (DEXA) equipment for body fat measurement.

#### **2. Materials and Methods**

#### *2.1. Participants*

Volunteers (20–60 years old) were recruited from Daegu and among employees of Kyungpook National University in the Republic of Korea, in March 2017. After an initial screening, 108 participants with a body mass index (BMI) of 23–30 kg/m<sup>2</sup> were selected. The exclusion criteria were as follows: (1) taking diuretics for hypertension; (2) taking oral hypoglycemic agents or insulin injection; (3) serious cardiac, renal, hepatic, thyroid, or cerebrovascular disease; (4) serious cystic or gastrointestinal disease, gout, or porphyria; (5) psychiatric problems, such as depressive disorder, schizophrenia, alcoholism, and drug intoxication; (6) cancer diagnosis and treatment; (7) asthma or other allergies; (8) a history of surgery within the past 6 months; (9) pregnant or in lactation period. The study was approved by the Kyungpook National University Human Research Committee (KNU 2017-0113). All participants gave their written informed consent for inclusion before they participated in the study.

#### *2.2. Sample Size*

In order to eliminate the effects of gender differences, the gender of the participants was assigned by replicated randomized complete block design. The sample size was estimated using G\* Power 3.1.9.2. Assuming a 95% statistical power, 0.05 significance level, and 0.40 effect size (Cohen's standard for a large effect), it was estimated that at least 120 participants were required to show a statistically significant difference in biomarkers of body fat among four groups.

#### *2.3. Design*

This single institution, randomized, double-blinded, and placebo-controlled study was conducted to confirm the effect of PGE supplementation on body fat loss in obese or overweight participants. The random assignment code was generated using the permed-block randomization method with the assistance of SAS Proc Plan (SAS Institute, Cary, NC, USA). All participants were randomly assigned in into four groups in a 1:1:1:1 ratio: placebo, PGE571 (571 mg as PGE), PGE1142 (1142 mg as PGE), and PGE2855 (2855 mg as PGE). The total content of test product for one tablet was 900 mg (including 571 mg of *Platycodon grandifloras* extract); all participants received two pouches per day of the placebo (including 666 mg of crystalline cellulose) and each PGE (PGE571, PGE1142, PGE2855), according to their assigned group, which were consumed from baseline (0 days) to the end of the 12-week experiment, at 30 min after breakfast and dinner. The placebo and PGEs were supplied by

GC WellBeing (Seongnam, Korea). All participants were instructed to maintain their routine food intake and physical activity during the study. Moreover, during the study period, we monitored the participants' compliance with the nutritional intervention and capsule consumption every week by telephone. At the end of the study, the participants were asked to return any pouches not consumed. The doses of PGE were selected by extrapolated calculations on the basis of a previous animal study [14].

#### *2.4. Anthropometric and Biochemical Analyses*

For anthropometric and physiological measurements at baseline and 4, 8, and 12 weeks post-test material supplementation, the participants visited the Science Research Center Laboratory at Kyungpook National University between 07:00 and 11:00 h after a 12 h overnight fast. The BMI, height, weight, and body composition were measured using an X-Scan Plus II body composition analyzer (Jawon Medical Co., Daejeon, Korea). Abdominal computed tomography (CT; Brivo CT385, GE Healthcare, Chicago, IL, USA) scans that included the lumbar spine were acquired at baseline and during follow-up [2,4]. The CT scans were taken at the Doctors Radiology Clinic, located in Daegu city (Korea). The waist and hip circumferences were measured with an anthropometric tape. The waist circumference was measured as the minimum circumference between the iliac crest and rib cage, and the hip circumference was measured as the maximum width over the greater trochanters. The waist-to-hip ratio was calculated by dividing the waist measurement by the hip measurement. Blood samples were collected in ethylenediaminetetraacetic acid-coated tubes and centrifuged at 1000× *g* for 15 min at 4 ◦C for plasma assays. To determine the dietary intake, 24 h dietary recalls were administered in face-to-face interviews at the participants' homes before and during the preliminary trial by dieticians. Three day dietary recalls were performed twice at baseline and during follow-up. We presented the mean of the 3 day dietary intake at each point. During the interview, the participants were asked what kinds of food they ate and drank on their dietary recall sheet. Food replicas were provided to help the participants estimate their dietary intakes and exact portions. Nutritional analysis was performed using CAN-Pro 3.0 software (The Korean Nutrition Society, Seoul, Korea), which provides a comprehensive database for the nutritional content of general foods and specific Korean foods.

#### *2.5. Biochemical Analyses*

Before and after the test, fasting blood was collected and analyzed by Seegene Co. Ltd. (Daegu, Korea). The levels of plasma adiponectin, leptin, interleukin-6, tumor necrosis factor-alpha, and monocyte chemoattractant protein-1 were determined using a Cobas 8000 analyzer (Roche Diagnostics, Mannheim, Germany). For the safety evaluation of the test material, the liver and renal function markers (albumin, blood urea nitrogen, total bilirubin, aspartate transaminase, alanine transaminase, and alkaline phosphatase (ALP)) were measured using a Hitachi LABOSPECT 008 AS (Hitachi, Tokyo, Japan).

#### *2.6. Statistical Analysis*

Data were analyzed using SPSS (IBM SPSS, version 21) and expressed as mean ± standard deviation. Statistical analysis was performed using the analysis of covariance (ANCOVA) test for comparison between the placebo control group (placebo) and the test groups. In the questionnaire, it was confirmed that there was a large difference in physical activity and drinking among the participants. Given that there is a high correlation between alcohol consumption and obesity rate [15], participants who frequently consumed alcohol (more than two bottles per week) and reported a very low physical activity (sedentary behavior) were excluded (Table 1). Paired *t*-test was used to verify the difference before and after ingestion at *p* < 0.05. Statistical analysis was also performed on all participants (23 kg/m<sup>2</sup> <sup>≤</sup> BMI <sup>≤</sup> 30 kg/m2) without any discrimination of physical activity and alcohol consumption for safety assessment.


**Table 1.** Survey of physical activity and alcohol drinking status of subjects.

#### **3. Results**

#### *3.1. Study Flow*

This study had an initial 130 candidates, and had 108 eligible individuals who were enrolled in this study through the eligibility tests. After 12 weeks, eight people dropped out for personal reasons. Participants with factors affecting obesity, very low physical activity, and frequent alcohol consumption were excluded. Physical activity and alcohol consumption were assessed through the questionnaire (Table 1). Thus, data from 72 participants were analyzed for evaluating the efficacy of PGE supplementation. Serious adverse effects were not reported by the participants consuming the PGEs or placebo supplements.

#### *3.2. Baseline Clinical Characteristics and Nutrient Intake*

The baseline characteristics of volunteers who completed the randomized controlled trial are shown in Table 2. There were no significant differences among all groups in terms of age, height, systolic blood pressure, diastolic blood pressure, and fasting blood glucose. Nutritional intake of the participants before and after the test food intake was measured six times (three times before the test, three times after the test) using the 24 h recall method. The energy and carbohydrate intake of the PGE1142 and PGE2855 groups were significantly higher than that of the placebo group. In other nutrients, no significant difference was found among all groups (Table 3).

**Table 2.** Baseline characteristics of 4 groups with overweight or obesity subjects who participated in efficacy test of *Platycodon grandiflorus* root extract (PGE).


Values are the mean ± SD; PGE571: 571 mg administered as PGE, PGE1142: 1142 mg administered as PGE, PGE2855: 2855 mg administered as PGE, BMI: body mass index, WHR: waist hip ratio, BP: blood pressure, M: male, F: female.


**Table 3.** Comparison of physical activity and nutrients intake in four groups with overweight or obesity by 24 h dietary recall performed before and in a follow-up of the trial.

*p*-Value: Analysis of covariance (ANCOVA) model with independent variable as baseline and treatment; \* *p* < 0.05, \*\* *p* < 0.01 derived from paired *t*-tests performed for values obtained before and after the trial. CFB: changes from baseline.

#### *3.3. Body Composition*

Body composition, such as body weight, BMI, body fat mass, body fat percentage (BFP), and muscle weight, were analyzed by using data that considered drinking and exercise activities (*n* = 72) (Table 4). Body fat (ANCOVA, *p* < 0.028) and BFP (ANCOVA, *p* < 0.001) were significantly decreased in the PGE571 group compared with the placebo. Also, supplementation of PGE2855 led to a decrease in the body fat mass (ANCOVA, *p* < 0.036) and BPF (ANCOVA, *p* < 0.035). In contrast, muscle mass was significantly increased in the PGE571 (ANCOVA, *p* < 0.002)-supplemented group compared with the control (placebo) group (Table 4).

Moreover, we performed CT scans at baseline and after (follow-up) to observe the effects of PGE supplementation on the participants' abdominal fat area. The high-dose PGE group (PGE2855) was observed to have a significant effect when compared with the placebo group. In particular, the PGE2855 led to a significant decrease in the L4 total abdominal fat area (ANCOVA, *p* = 0.029) and subcutaneous fat area (ANCOVA, *p* = 0.035) (Table 5).


**Table 4.** Effect of PGE supplementation for 12 weeks on change of body composition measured by bioelectrical impedance analysis (BIA), WHR-related body measurements, and blood pressure in groups with overweight or obesity.

*p*-Value: ANCOVA model with independent variable as baseline and treatment; \* *p* < 0.05, \*\* *p* < 0.01, \*\*\* *p* < 0.001 derived from paired *t*-tests performed for values obtained before and after the trial; CFB: changes from baseline; BFP: body fat percentage.

**Table 5.** Effect of PGE supplementation for 12 weeks on change of abdominal fat area assessed by computed tomography (CT) in subjects with overweight or obesity.


*p*-Value: ANCOVA model with independent variable as baseline and treatment; \* *p* < 0.05, \*\* *p* < 0.01 derived from paired *t*-tests performed for values obtained before and after the trial; CFB: changes from baseline.

#### *3.4. Plasma Adipokines*

Table 6 shows the levels of plasma adipokines. When ANCOVA was used, there were no significant differences in adipokine levels among the four groups. However, PGE571 supplementation significantly reduced leptin levels after 12 weeks from baseline. Also, the leptin:adiponectin (L:A) ratio was significantly lower after supplementation with PGE2855 compared with the baseline (before supplementation) measurement. In comparison to the baseline, PGE supplementation led to a significant decrease in leptin levels in the PGE571 (paired *t*-test, *p* < 0.01) and the L:A ratio in the PGE2855 (paired *t*-test, *p* < 0.05) after the 12 week experiment. In addition, when all participants were analyzed (*n* = 100), adiponectin levels (ANCOVA, *p* < 0.059) tended to decrease in the PGE571 group compared with the placebo (data not shown).


**Table 6.** Effect of supplementation of PGE for 12 weeks on changes of serum adipokine levels in subjects with overweight or obesity.

*p*-Value: ANCOVA model with independent variable as baseline and treatment; \* *p* < 0.05, \*\* *p* < 0.01 derived from paired *t*-tests performed for values obtained before and after the trial; CFB: changes from baseline, L:A: leptin:adiponectin.

#### *3.5. Lipids Metabolism*

Table 7 shows the changes in the indirect markers of lipids metabolism. There were no significant changes in lipids metabolism. When all participants were analyzed, there were no significant changes among the four groups in terms of cholesterol, triglyceride, high density lipoprotein cholesterol (HDL-C), low density lipoprotein cholesterol (LDL-C), and phospholipid. The changes were within the normal range and were considered to have no clinical implications in terms of the safety assessment.


**Table 7.** Effect of supplementation of PGE for 12 weeks on changes of lipids metabolism in subjects with overweight or obesity.

*p*-Value: ANCOVA model with independent variable as baseline and treatment; \* *p* < 0.05 derived from paired *t*-tests performed for values obtained before and after the trial; CFB: change from baseline; TG: triglyceride; HDL-C: high density lipoprotein cholesterol; LDL-C: low density lipoprotein cholesterol.

#### *3.6. Safety Assessment and Indirect Markers of Hepatic and Renal Function Test*

Hepatic and renal function markers in plasma were analyzed for safety assessment. When all participants were analyzed (*n* = 100), there were no significant changes in the safety-related markers of liver or renal function among the four groups. Table 8 shows the changes in the indirect markers of liver and renal function due to PGE supplementation. PGE1142 supplementation was observed to lower the glutamic-pyruvic transaminase (GPT) (ANCOVA, *p* < 0.006) levels more than the placebo. The changes were within the normal range and were considered to have no clinical implications.


**Table 8.** Effect of PGE supplementation for 12 weeks on plasma albumin, total bilirubin, alanine aminotransferase (ALP), glutamic oxalacetic transaminase (GOT), glutamic pyruvate transaminase (GPT), blood urea nitrogen (BUN), and creatinine levels in subjects with overweight or obesity (*n* = 100).

*p*-Value: ANCOVA model with independent variable as baseline and treatment; \* *p* < 0.05, \*\* *p* < 0.01, \*\*\* *p* < 0.001 derived from paired *t*-tests performed for values obtained before and after the trial; ALB: albumin, ALP: alanine aminotransferase, GOT: glutamic oxalacetic transaminase, GPT: glutamic pyruvate transaminase, BUN: blood urea nitrogen, CFB: changes from baseline.

#### **4. Discussion**

Many studies have reported on the health benefits of the extracts and saponin fractions of *P. grandiflorus*, such as anti-obesity and anti-lipid metabolism, but most of them have demonstrated efficacy in animal models. Our preliminary study also confirmed the weight loss and body fat loss efficacy of PGE in animal models of obesity. Indeed, PG extract increased thermogenic gene expression (such as SIRT1, PPARα, UCP1, PGC1α), thereby inducing the browning of white adipose tissue, resulting in an anti-obesity effect. Thus, this study was conducted to verify the effect of PGE in humans. Through the questionnaire, we found that there was a significant difference in the level of alcohol

drinking and physical activity among the participants. Accordingly, the following statistics were used in this study: an analysis, excluding dropouts, but including the participants with very low physical activity and frequent drinking activities (more than two bottles per week).

This study was a small-scale experiment to set the effective dose of PGE in humans by confirming the anti-obesity effect. We conducted bioelectrical impedance analysis (BIA), which is generally highly correlated with body weight and body fat mass, although we did not measure body fat mass by DEXA.

In particular, we screened participants who frequently consumed alcohol and had a very low physical activity, two factors closely associated with obesity, to confirm the anti-obesity effect of PGE. Energy and carbohydrates were significantly higher in the PGE1142 and PGE2855 supplement groups compared with the placebo. Body composition changes induced by PGE revealed that supplementation of PGE571 and PGE2855 significantly reduced body fat mass and BFP, but also led to an increase in muscle mass in the PGE571 group. In our previous study [14], we found increased muscle mass associated with increased energy expenditure following supplementation of PGE in animal models of obesity. Similarly, in this study, PGE supplementation significantly increased the muscle mass of the participants. Therefore, this increase is considered to be an important candidate in obesity treatment or prevention. We performed abdominal CT, as well as BIA, to confirm the effect of PGE on body fat loss. Consistent with the BIA results, PGE2855 supplementation significantly lowered the subcutaneous fat and abdominal fat area of L4 when compared with the placebo, suggesting the association of BFP and fat mass reduction by PGE2855 supplementation. Although there was no change in the abdominal fat area in the PGE571 group, this may have been due to the relatively high baseline abdominal fat area, as the PGE571 group showed a high level of leptin in the blood, which is proportional to the amount of fat [16]. PGE571 not only significantly decreased body fat and BFP, but also significantly increased muscle mass when BIA was performed, regardless of the statistical method. This was similar to our previous studies that showed increased muscle mass by PGE, which is closely associated with increased energy expenditure [17]. When all participants were analyzed in terms of lipids metabolism and markers of hepatic and renal function test, there were no significant changes among the four groups in terms of cholesterol, triglyceride, HDL-C, LDL-C, phospholipid, and markers of liver and renal function. The changes were within the normal range and were considered to have no clinical implications for the safety assessment. In this context, it is considered that the increase of muscle mass can act as an important marker in obesity (or sarcopenic obesity) research. In this study, DEXA, which is highly correlated with BIA [18] and provides more accurate levels of body fat mass, was not measured. Therefore, it is presumed that DEXA measurement may be able to confirm the change of abdominal fat area due to PGE571 supplementation in future studies.

The limitations of this trial should be emphasized. Given that the participants of this study were restricted to volunteers, it was difficult to represent the general population with this small sample size. As the study was conducted on overweight or moderately obese adults who had BMI measurements ranging between 23 and 30, part of the results that we obtained via clinical trial were difficult to identify and generate statistically meaningful data. Hence, to have more statistical meaningful data, it would be prudent to conduct the trial in obese adults. This study was a clinical trial to determine whether PGE could improve obesity and estimate the appropriate daily dose of PGE for overweight or obese adult humans. In addition, we used CT scans to improve the accuracy of the body composition measurements, but this technique only measured the abdominal fat area. Therefore, a further experiment is underway to confirm the body fat reduction function of PGE571 using DEXA.

#### **5. Conclusions**

In conclusion, despite some limitations, we demonstrated that PGE is able to reduce body fat mass and body fat percentage, which is an anti-obesity marker, in overweight or obese adult humans. PGE supplementation significantly increased the muscle mass when compared with the placebo, indicating an excellent anti-obesity effect.

**Author Contributions:** Y.J.K. performed the experiments, analyzed the data, and wrote/edited the manuscript. E.-Y.K. performed the experiments and reviewed the manuscript. J.-W.K., Y.L., and R.R. performed the experiments. GC WellBeing (J.Y. and M.K.) contributed to the preparation of *Platycodon grandiflorus* ethanol extract for this experiment. M.-S.C. supervised this work and had full access to all the data and, therefore, takes full responsibility for the integrity of the results and accuracy of the data analysis.

**Funding:** This work was supported by the Support Research and Commercialization Project grant number (NRF-20170701C4048818) of the Ministry of Science, ICT, and Future Planning through the National Research Foundation of Korea.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

*Article*

### **Comparative Investigation of Frankincense Nutraceuticals: Correlation of Boswellic and Lupeolic Acid Contents with Cytokine Release Inhibition and Toxicity against Triple-Negative Breast Cancer Cells**

**Michael Schmiech <sup>1</sup> , Sophia J. Lang 1, Judith Ulrich 1, Katharina Werner 1, Luay J. Rashan 2, Tatiana Syrovets 1,\* and Thomas Simmet 1,\***


Received: 1 September 2019; Accepted: 26 September 2019; Published: 2 October 2019

**Abstract:** For centuries, frankincense extracts have been commonly used in traditional medicine, and more recently, in complementary medicine. Therefore, frankincense constituents such as boswellic and lupeolic acids are of considerable therapeutic interest. Sixteen frankincense nutraceuticals were characterized by high-performance liquid chromatography with tandem mass spectrometry (HPLC-MS/MS), revealing major differences in boswellic and lupeolic acid compositions and total contents, which varied from 0.4% to 35.7%. Frankincense nutraceuticals significantly inhibited the release of proinflammatory cytokines, such as TNF-α, IL-6, and IL-8, by LPS-stimulated peripheral blood mononuclear cells (PBMC) and whole blood. Moreover, boswellic and lupeolic acid contents correlated with TNF-α, IL-1β, IL-6, IL-8, and IL-10 inhibition. The nutraceuticals also exhibited toxicity against the human triple-negative breast cancer cell lines MDA-MB-231, MDA-MB-453, and CAL-51 in vitro. Nutraceuticals with total contents of boswellic and lupeolic acids >30% were the most active ones against MDA-MB-231 with a half maximal inhibitory concentration (IC50) ≤ 7.0 μg/mL. Moreover, a frankincense nutraceutical inhibited tumor growth and induced apoptosis in vivo in breast cancer xenografts grown on the chick chorioallantoic membrane (CAM). Among eight different boswellic and lupeolic acids tested, β-ABA exhibited the highest cytotoxicity against MDA-MB-231 with an IC50 = 5.9 μM, inhibited growth of cancer xenografts in vivo, and released proinflammatory cytokines. Its content in nutraceuticals correlated strongly with TNF-α, IL-6, and IL-8 release inhibition.

**Keywords:** frankincense; *Boswellia*; boswellic acid; lupeolic acid; AKBA; cytokine; breast cancer; pentacyclic triterpenic acid; triterpenoid; chorioallantoic membrane assay

#### **1. Introduction**

Frankincense is an oleogum resin from trees of the genus *Boswellia* Roxb. ex Colebr., which belong to the *Burseraceae* family. Herbal preparations from frankincense have been used for centuries in traditional Ayurvedic, African, Arab, and Chinese medicine for the treatment of skin ailments, infectious diseases, and other conditions, which today could be assigned to various chronic inflammatory diseases and cancer [1,2]. Likewise, modern medicine has rediscovered frankincense and its pain-relieving, sedative, anti-inflammatory, antimicrobial effects, and even potential anticancer properties [3–5].

*Boswellia* trees grow mainly in dry areas in India, the Arabian Peninsula, and the Horn of Africa and often have a shrubby appearance with an average high of 2–6 m [6,7], whereby, the main representative

species are *Boswellia sacra* (Oman and Yemen), *Boswellia carterii* (Somalia), and *Boswellia serrata* (India). However, there are about 25 *Boswellia* species reported, but it is still not clear whether this number contains some double-counted species [8]. Frankincense, the oleogum resin of *Boswellia* trees, which contains potential biologically-active compounds, emerges from cuts in trunks and branches as a sticky-milky liquid, which dries quickly in the air, yielding an oleogum resin containing 15–20% boswellic acids and lupeolic acids [9] (Figure 1). These pentacyclic triterpenic acids are believed to be effective in the prevention and treatment of chronic inflammatory diseases and cancer [3,5]. Previous studies have demonstrated that boswellic acids inhibit essential pathways of inflammatory responses by interaction with IκB kinases and therefore inhibition of proinflammatory gene expression or by inhibition of 5-lipoxygenase and leukotrienes biosynthesis [10,11]. Moreover, also recent clinical pilot studies claimed therapeutic efficacy of extracts containing boswellic acids in the treatment of chronic inflammatory diseases like asthma, rheumatoid arthritis, Crohn's disease, osteoarthritis, collagenous colitis, or colitis ulcerosa [12,13]. Furthermore, *Boswellia serrata* extracts as well as boswellic and lupeolic acids induce apoptosis in various cancer cell lines, such as leukemia, brain, and prostate [5,14–17]. There have also been reports indicating the efficacy of boswellic acids against breast cancer [18,19].

**Figure 1.** Production of frankincense nutraceuticals. *Boswellia* tree grown in Somalia (**a**), harvesting of the frankincense oleogum resin by bark incisions (**b**), commercial frankincense nutraceuticals (**c**). Pictures with permission from Georg Huber [20].

With 24.2%, breast cancer is the most common cancer of all female cancer cases, causing 15.0% of all female cancer deaths in 2018 [21]. Particularly, triple-negative breast cancer (TNBC) is an aggressive, highly metastatic breast cancer subtype affecting mainly younger women. TNBC is characterized by the lack of expression of hormone receptors (estrogen and progesterone) and the human epidermal growth factor receptor 2 (HER2), which countervails targeted therapy [22]. First, in 2011, Suhail et al. described the induction of apoptosis in human breast cancer cells in vitro by essential oil derived from *Boswellia sacra* [19]. In a later study, we demonstrated that mainly the acidic components of frankincense, i.e., boswellic and lupeolic acids, exhibit cytotoxic efficacy against the TNBC cell line MDA-MB-231 [18].

The global market size of dietary supplements and nutraceuticals was estimated to be worth USD 115 billion in 2018 with an expected compound annual growth rate (CAGR) of 7.8% till 2025 [23]. This demonstrates the strong aptness of the general public to use natural products to maintain good health and to prevent or treat diseases. However, due to insufficient regulation regarding quality and lack of standardization of dietary supplements and nutraceuticals, several questionable or even harmful products are on the market [24].

The aim of this study is to compare frankincense nutraceuticals (FNs) regarding their chemical composition, their efficacies in inhibiting production of inflammatory cytokines, and their cytotoxic efficacy against metastatic TNBC cells.

#### **2. Materials and Methods**

#### *2.1. Materials*

All solvents and chemicals were of analytical reagent grade. Dimethyl sulfoxide (DMSO) for sample preparation was purchased from Invitrogen (Thermo Fisher Scientific, Waltham, MA, USA). The solvents used for HPLC-MS/MS analysis were methanol, acetic acid (both HiPerSolv Chromanorm, VWR chemicals, Fontenay-sous-Bois, France) and ultrapure water (reverse-osmosis type water (pureAqua, Schnaitsee, Germany) coupled to a Milli-Q station (Millipore, Eschborn, Germany). The reference substances, acetyl-α-boswellic acid (α-ABA), acetyl-β-boswellic acid (β-ABA), α-boswellic acid (α-BA), β-boswellic acid (β-BA), acetyl-11-keto-β-boswellic acid (AKBA), and 11-keto-β-boswellic acid (KBA) were purchased from Extrasynthese (Genay Cedex, France). Acetyl-lupeolic acid (ALA) and lupeolic acid (LA) were isolated and characterized as previously published [9,25]. Samples of frankincense nutraceuticals (FNs) were purchased from the respective distributors (see Table A1). Stock solutions of FNs and pure boswellic (BAs) and lupeolic acids (LAs) were prepared in DMSO and further diluted in RPMI 1640 medium (Gibco, Thermo Fisher Scientific, Waltham, MA, USA). In all experiments, the final DMSO concentration did not exceed 0.5%.

#### *2.2. Quantification of Boswellic and Lupeolic Acids by HPLC-MS*/*MS Analysis*

The method development, chromatographic separation, mass spectrometry detection, and method validation for quantification of boswellic and lupeolic acids by HPLC-MS/MS have been previously published by us [18]. The quantification by HPLC-MS/MS analysis was performed on an Agilent 1260 Infinity system (Agilent, Santa Clara, CA, USA) coupled with an AB API 2000 triple quadrupole mass spectrometer (Applied Biosystems, Foster City, CA, USA). For chromatographic separation, an analytical reversed-phase HPLC column (Dr. Maisch ReproSil-Pur Basic-C18 HD, 3 μm, 125 × 3 mm; Dr. Maisch GmbH, Ammerbruch, Germany) with a precolumn (Dr. Maisch ReproSil Universal RP, 5 μm, 10 × 4 mm) were used. For sample preparation, contents were removed from the capsules, weighted, and dissolved in DMSO (β = 1 mg/mL, *w*/*v*), whereas pills were beforehand grounded. Samples were filtered through a 0.45 μm regenerated cellulose filter before injection. For quantification, three different capsules or pills were analyzed per product, each in duplicates.

#### *2.3. Analysis of Cytokine Release*

Whole human blood was collected from the antecubital vein of healthy male donors and incubated with FNs in concentrations of 30 μg/mL for 20 min followed by a stimulation with LPS (10 ng/mL) for 18 h. Plasma was separated by centrifugation and analyzed for IL-10 and TNF-α using ELISA Duo-Set Human from R&D Systems (Minneapolis, MN, USA). Peripheral blood mononuclear cells (PBMC) were isolated from whole venous blood from healthy male donors via density gradient centrifugation using Biocoll (Biochrom GmbH, Berlin, Germany). The collection and analysis of whole blood and peripheral blood mononuclear cells used in this study were approved by the Institutional Ethics Committee (# 177/18). The participating volunteers provided written informed consent to participate in this study. Cells were seeded into a 96-well plate (4 <sup>×</sup> <sup>10</sup><sup>5</sup> cells in 200 <sup>μ</sup>L RPMI 1640 supplemented with 1% FCS and 100 U/mL penicillin, 100 μg/mL streptomycin (all from Gibco)). Cells were treated with FNs in concentrations of 10 μg/mL and pure BAs or LAs in final concentrations of 3 μg/mL for 20 min followed by a stimulation with LPS (10 ng/mL) and an 18 h incubation at 37 ◦C and in a 5% CO2 atmosphere. After incubation, cells were centrifuged and supernatants were analyzed. The amounts of IL-12p70, TNF-α, IL-10, IL-6, IL-1β, and IL-8 were quantified by flow cytometry using Cytometric Bead Array (CBA) from Becton Dickinson (Franklin Lakes, NJ, USA) according to manufacturer's instructions.

#### *2.4. Analysis of Antiproliferative and Cytotoxic E*ff*ects In Vitro*

To verify that the concentrations of the FNs and pure BAs and LAs used, are not cytotoxic to PBMC but cytotoxic to cancer cells, analysis of lactate dehydrogenase (LDH) release (Roche, Basel, Switzerland), XTT assay of cell viability and proliferation (Roche), and analysis of cell membrane integrity by propidium iodide staining [26] were carried out. Triple-negative human breast cancer cells MDA-MB-231 (ATCC, Rockville, MD, USA), CAL-51 and MDA-MB-453 (both from DSMZ, Braunschweig, Germany), and PBMC were analyzed. MDA-MB-231 cells were cultured in Dulbecco's Modified Eagle Medium (4.5 g/L glucose, GlutaMax; Life Technologies, Carlsbad, CA, USA) supplemented with 10% FCS, 0.1 mM MEM non-essential amino acids, 100 U/mL penicillin, and 100 mg/mL streptomycin. CAL-51 cells were cultured in Dulbecco's Modified Eagle Medium (4.5 g/L glucose; Life Technologies) supplemented with 10% FCS, 100 U/mL penicillin, and 100 mg/mL streptomycin. The MDA-MB-453 cells were cultured in Leibovitz's L-15 medium (Life Technologies) containing 10% FCS, 100 U/mL penicillin, and 100 mg/mL streptomycin and kept at atmospheric CO2. When cancer cells reached 80% confluence, they were subcultured according to the supplier's recommendations. MDA-MB-231 and CAL-51 cells were treated 24 h after seeding. MDA-MB-453 cells were allowed to adhere for 72 h, because of slow adhesion. Subsequently, treatment with the respective compounds for 72 h followed. Different concentrations of the respective samples were applied using a Tecan D300e Digital Dispenser (Tecan, Männedorf, Switzerland). Absorbance was measured using an Infinite M1000 PRO Tecan plate reader. For analysis of viability, the blank values containing the respective compounds in the according concentration were subtracted and the percentage of viable cell was calculated by normalization to the vehicle control.

#### *2.5. Human Tumor Xenografts on the Chick Chorioallantoic Membrane*

TNBC xenografts were established by seeding 0.7 <sup>×</sup> 10<sup>6</sup> MDA-MB-231 cells in medium/matrigel (1:1) onto the chick chorioallantoic membrane (CAM) of fertilized chick eggs 7 days after fertilization. For the next 3 consecutive days, cells were treated topically with 20 μL of the respective FN or pure compounds dissolved in 0.9% NaCl (vehicle control: 0.5% DMSO). FN16 was used in concentrations of 5 and 10 μg/mL; AKBA and β-ABA were used in concentrations of 5 μg/mL (10 μM); and doxorubicin was used in a concentration of 1 μM. On day 4 after treatment initiation, tumors were collected, imaged, fixed, and embedded in paraffin for analysis by immunohistochemistry. Tumor volume (mm3) was calculated with the formula length (mm) <sup>×</sup> width2 (mm) <sup>×</sup> <sup>π</sup>/6. For immunohistochemical analysis, 5-μm slices of the collected tumors were stained using antibodies against the proliferation marker Ki-67. For analysis of apoptosis in vivo, DNA strand breaks were visualized by deoxynucleotidyl transferase dUTP nick end labeling (TUNEL) according to the manufacturer's recommendations (Roche). Images were recorded with an Axio Lab.A1 microscope (Carl Zeiss, Göttingen, Germany) and a Zeiss 2/3" CMOS camera using Progres Gryphax software (Carl Zeiss).

#### *2.6. Statistical Analysis*

Each experiment was repeated at least three times and the data are expressed as mean ± standard deviation (SD) or standard error of the mean (SEM) as indicated. Statistical analysis was performed using Minitab 18 software (Minitab, Munich, Germany) and SigmaPlot 14.0 (Systat Software Inc., San Jose, CA, USA). All data were tested for normal distribution by the Anderson-Darling test and equality of variances by Levene's test. Sample groups were compared by one-way analysis of variance (ANOVA) and post-hoc by Dunett's test for parametric data. Non-parametric data were either transformed prior to analysis by Box-Cox transformation or compared directly by Kruskal-Wallis one-way ANOVA on ranks and post-hoc by Dunn's test. Spearman's rank correlation was used to investigate correlations of non-parametric data. Cluster analysis was performed with hierarchical agglomerative clustering, complete-linkage, and Euclidean distances. Principal component analysis (PCA) was derived from a covariance matrix of the data.

#### **3. Results**

#### *3.1. Compositions of Boswellic and Lupeolic Acids in Frankincense Nutraceuticals*

Quantification of eight BAs and LAs including α-boswellic acid (α-BA), acetyl-α-boswellic acid (α-ABA), β-boswellic acid (β-BA), acetyl-β-boswellic acid (β-ABA), 11-keto-β-boswellic acid (KBA), acetyl-11-keto-β-boswellic acid (AKBA), lupeolic acid (LA), and acetyl-lupeolic acid (ALA), in 16 frankincense nutraceuticals (FNs) was performed by HPLC-MS/MS. The total contents (*w*/*w*) of BAs and LAs varied from 0.4% to 35.7% (Table 1). Related to the mass of a capsule or a pill, the total BA and LA contents varied from 3.5 to 157.3 mg/unit (Table S1). Many manufacturers and distributers promote their products as those with boswellic acids contents over 80%. Interestingly, these concentrations could not be confirmed. Manufacturers usually determine the boswellic acid content of their products by non-aqueous titration, for example, using 0.1 N potassium methoxide and 0.3% (*w*/*v*) thymol blue as the indicator [27]. These methods estimate unselectively all acidic compounds, not only boswellic acids. Here, a highly selective and accurate quantification method for analysis of BAs and LAs by HPLC-MS/MS was applied (Figure 2a,b). Consequently, the precise contents of the individual BAs and LAs in nutraceuticals could be determined (Table 1 and Table S1).

**Table 1.** Contents of boswellic and lupeolic acids in frankincense nutraceuticals (FN) quantified by HPLC-MS/MS analysis.


Samples are arranged by descending total contents of BAs and LAs (Σ in percent, *w*/*w*). KBA, 11-keto-β-boswellic acid; LA, lupeolic acid; α-BA, α-boswellic acid; β-BA, β-boswellic acid; AKBA, acetyl-11-keto-β-boswellic acid; ALA, acetyl-lupeolic acid; α-ABA, acetyl-α-boswellic acid; β-ABA, acetyl-β-boswellic acid.

In a previous study, we showed that oleogum resins of the species *B. serrata* are characterized by higher proportions of deacetylated compounds compared to acetylated compounds [18]. This pattern was also observed in the FN samples analyzed here, which contained mainly extracts from *B. serrata*. Unusually, particularly for *B. serrata*, sample FN8 exhibited an exceptionally high content of AKBA. Especially, the concurrence of the high content of AKBA and very low contents of the other BAs and LAs indicated a special enrichment procedure putting FN8 apart from the other samples. Meins et al. observed the same uncommon ratio between AKBA content and total BAs content in a product containing the same 5-Loxin® extract [28]. FN5 and FN7 showed the lowest contents of BAs and LAs of all investigated samples, with total BAs and LAs contents of 2.9% for FN5 and only 0.9% for FN7. In addition to the exceptionally low BAs and LAs contents, the large difference between the stated

extract content of 400 mg for FN7 and the determined pill mass of 809.8 ± 5.0 mg points also to an oddly high amount of additives in FN7 (Table 1 and Table S1). According to the manufacturer, the AKBA content of FN7 is 52 mg per pill. However, analysis revealed an AKBA content <1 mg per pill. Hence, it can be assumed that either less *Boswellia* extract than stated was processed or other *Boswellia* oleogum resins containing little or no BAs and LAs were used like, for example, *B. frereana* [18,29].

**Figure 2.** Chromatograms of HPLC-DAD-MS/MS analysis of sample FN9 and multivariate statistical analysis of BAs and LAs contents in different FNs. (**a**) Total wavelength chromatogram with detection at 210 nm, 254 nm, and 280 nm. (**b**) Multiple-reaction monitoring chromatogram and structures of BAs and LAs present in FNs. (**c**) The dendrogram shows FNs assigned to three different clusters according to their BAs and LAs composition: Cluster A >30%, Cluster B with 15–30%, and Cluster C <15% of total BAs and LAs. (**d**) Samples within clusters exhibit similarity in their individual BAs and LAs compositions. Biplot of principal component analysis is shown. Sample FN8 shows the highest deviation due to its unusually high AKBA content concomitantly with very low levels of all other BAs and LAs. Numbered scores visualize the individual FN1-16 in the biplot and dashed lines demonstrating the loadings of the individual BAs and LAs.

Multivariate statistical methods, cluster analysis and principal component analysis (PCA) were applied to the datasets to visualize the data and discover patterns of the BAs and LAs compositions in different FNs. Cluster analysis assigned the samples to three different clusters (Figure 2c). Cluster A contained samples FN9, FN13, and FN16 with the highest total BAs and LAs contents of over 30%. Furthermore, these FNs showed high similarity regarding the composition of the individual components (Figure 2d). Cluster B (samples FN2, FN4, FN6, and FN14) comprised samples with total BAs and LAs contents between 15% and 30%, and Cluster C (samples FN1, FN3, FN4, FN5, FN7, FN8, FN10, FN11, and FN12) contained less than 15% BAs and LAs. FN8 showed the lowest similarity to Cluster C (<50%) due to its unusual chemical composition with high AKBA concentration. Moreover, samples FN10, FN11, and FN12, which contained non-extracted oleogum resin, formed an additional subcluster, due to their similarity. Interestingly, these three samples contained oleogum resins of different *Boswellia* species. According to the manufacturers, sample FN10 was made of *B. sacra*, FN11 of *B. serrata*, and FN12 of *B. carterii*. Comparing with the BAs and LAs compositions determined in our previous studies [9,18], FN11 exhibited indeed general BAs and LAs patterns typical for *B. serrata*

resins, with a higher proportion of deacetylated BAs and LAs than acetylated ones. However, FN10 and FN12 showed BAs and LAs patterns untypical for *B. sacra* or *B. carterii* resins, but more similar to *B. serrata*, too. Because these samples contain no extracts but untreated oleogum resin, these differences could not be caused by an extraction procedure. Therefore, it cannot be excluded that instead of *B. sacra* or *B. carterii*, possibly *B. serrata* oleogum resin might have been used for manufacturing of FN10 and FN12.

#### *3.2. Inhibition of Proinflammatory Cytokine Release by Frankincense Nutraceuticals*

For comparative investigation of the immunomodulatory activities of the FNs, whole blood of healthy donors was treated with lipopolysaccharide (LPS) to initiate an acute inflammatory response and to induce the release of proinflammatory cytokines. Pretreatment with FNs significantly decreased the expression of the proinflammatory cytokine TNF-α compared to the LPS-treated control group (Figure 3a). Interestingly, the production of the anti-inflammatory cytokine IL-10 was inhibited by the FNs, too. IL-10 acts to limit inflammatory responses by damping the uncontrolled production of proinflammatory cytokines including TNF-α [30]. However, also TNF-α regulates the production of IL-10. Thereby, the chronological sequence of cytokine expression during inflammation is essential. While high levels of proinflammatory cytokines IL-1, IL-6, IL-8, and TNF-α are released already 4–8 h after LPS stimulation, maximal levels of the anti-inflammatory IL-10 are observed only 24–48 h after stimulation [31]. Moreover, about 50–75% of IL-10 released by LPS-stimulated monocytes can be inhibited by anti-TNF-α [32], indicating that TNF-α is responsible for the majority of the released IL-10. Therefore, the decreased production of TNF-α by blood treated with FNs would decrease the later IL-10 release (Figure 3a).

The strong albumin-binding affinity of BAs complicates an accurate analysis of their immunomodulatory activity in whole blood [33]. In addition, blood contains varying amounts of different lipids aiding samples' solubility. Hence, analysis of variance (ANOVA) exhibited no significant differences between the samples or sample groups when analyzed in whole blood. As monocytes are the major producers of TNF-α and IL-10 in blood [34], for further analysis, peripheral blood mononuclear cells (PMBC) were used and the effects of the FNs on the production of a panel of cytokines including TNF-α, IL-1β, IL-6, IL-8, IL-10, and IL-12p70, were analyzed. Contents of IL-12p70 were too low for an adequate evaluation and are not presented. FNs were used at a concentration of 10 μg/mL, which did not affect cell viability as analyzed by XTT and lactate dehydrogenase release (LDH) assays and by propidium iodide staining.

FNs with BAs and LAs contents >30% (Cluster A: FN9, FN13, and FN16) exhibited a significant inhibition of TNF-α, IL-6, and IL-8 expression compared to the control LPS group (Figure 3b–f and Table S2). Moreover, FN4 and FN6 from Cluster B showed considerable inhibitory activity towards TNF-α, IL-6, and IL-8, too. The most potent five nutraceuticals, FN4, FN6, FN9, FN13, and FN16, decreased the release of proinflammatory cytokines to 20.7% for TNF-α, 9.2% for IL-6, and 16.8% for IL-8 (average values for all five FNs). Interestingly, these five samples also had the highest concentrations of ALA, α-ABA, and β-ABA. The proteasome inhibitor MG132 used as a positive control decreased cytokine release to 15.5% ± 7.2% for TNF-α, 18.5% ± 7.7% for IL-1β, 5.2% ± 3.5% for IL-6, 20.2 ± 7.2% for IL-8, and 37.5% ± 14.8% for IL-10 compared to the control group. Differently, samples FN5 and FN7, that were previously shown to contain the lowest BAs and LAs contents, exhibited no inhibitory effects on cytokine production at all but even enhanced the expression of the proinflammatory cytokines TNF-α, IL-6, and IL-8 with average values of 142.4% for FN5 and 210.7% for FN7 compared to the control group. The reason for the increased secretion of cytokines might be contamination of these FNs with bacterial products.

**Figure 3.** FNs inhibit cytokine production by LPS-stimulated blood and PBMC. (**a**) Inhibition of TNF-α and IL-10 expression by FNs (each at 30 μg/mL) in LPS-stimulated whole blood. Data are mean ± SEM of 16 FNs; each FN was analyzed in five independent experiments/donors, Wilcoxon test. (**b**–**f**) Inhibition of TNF-α, IL-1β, IL-6, IL-8, and IL-10 production by LPS-stimulated PBMC by the respective FN (each at 10 mg/mL). Samples are arranged according to β-ABA content in a descending order. Clusters are from cluster analysis of BAs and LAs compositions in Figure 2c,d: Cluster A (light red) >30%, Cluster B 30–15% (yellow), and Cluster C (green) <15% total contents of BAs and LAs. Box-Cox transformation followed by one-way ANOVA and post-hoc Dunnett's test. All data are mean ± SEM. *n* = 7 donors; \* *p* < 0.05, \*\* *p* < 0.01, \*\*\* *p* < 0.001 compared to LPS-stimulated control groups (Ctrl).

Investigation of the correlation between the chemical composition and cytokine expression revealed that the total amount of BAs and LAs correlate positively with the inhibition of TNF-α, IL-6, IL-8, and IL-10 (Table 2 and Figure S1). Especially, the acetylated compounds ALA, α-ABA, and β-ABA exhibit the highest correlation coefficients, while the acetylated compound with keto group AKBA shows the lowest correlation. However, when FN8 with the extraordinarily high AKBA content was removed as an outlier (Grubbs test), AKBA contents in FNs exhibited significant correlation with all cytokines with *p* = 0.007 for TNF-α, *p* < 0.001 for IL-1β, *p* = 0.011 for IL-6, *p* = 0.006 for IL-8, and *p* = 0.001 for IL-10. Inhibition of IL-1β correlated only with AKBA, ALA, and α-ABA contents in FNs. Moreover, the inhibition of all five investigated cytokines correlated with each other. Whereby, IL-1β, the only cytokine in the panel activated by the NLRP3 inflammasome [35], showed the lowest correlation (Table 2 and Figure S1).


**Table 2.** Correlation between BAs and LAs contents and inhibition of cytokine production by LPS-stimulated PBMC.

Spearman's rank correlation; R, correlation coefficient, \* *p* < 0.05, \*\* *p* < 0.01, and \*\*\* *p* < 0.001, *n* = 16. PBMC, peripheral blood mononuclear cells.

#### *3.3. Cytotoxic E*ffi*cacy of Frankincense Nutraceuticals Against Triple-Negative Breast Cancer Cells In Vitro*

FN16, one of the most potent nutraceuticals with respect to cytokine release, concentration-dependently inhibited the viability of the highly metastatic, treatment-resistant breast cancer cell line MDA-MB-231 in vitro. Interestingly, cancer cells were more sensitive to FN16 compared to normal human PBMC, providing evidence for selectivity against cancer cells (Figure 4a). Also, FN1 with low contents of BAs and LAs, exhibited higher toxicity against MDA-MB-231 cells compared to PBMCs (Figure S2a), though the difference was not as strong as for FN16. Therefore, all FNs were investigated for their toxicity against MDA-MB-231 breast cancer cells and the half maximal inhibitory concentrations (IC50) were determined (Table 3 and Figure 4b). The FNs exhibited cytotoxicity against MDA-MB-231 cells with an average IC50 of 15.9 μg/mL. The samples FN9, FN13, and FN16 (Cluster A) with the highest BAs and LAs contents exhibited also significantly higher cytotoxicity with IC50 values for FN9, FN13, and FN16 of 7.0 μg/mL, 6.6 μg/mL, and 6.0 μg/mL, respectively, compared to the average IC50 value of 15.9 μg/mL (Figure 4b). Samples FN5 and FN7 (Cluster C) demonstrated the lowest cytotoxicity against MDA-MB-231 with an IC50 = 44.3 μg/mL for FN5 and an IC50 = 30.7 μg/mL for FN7. Thus, FN5 and FN7 not only exhibited the lowest contents of BAs and LAs and the poorest cytokine modulatory properties, but also the cytotoxic efficacies against MDA-MB-231 were significantly below average. Hence, this points to a correlation between BAs and LAs compositions and cytotoxicity against breast cancer cells. A combination of principal component analysis (PCA) of the BAs and LAs concentrations with a contour plot that visualizes the IC50 values revealed correlation between the BAs and LAs contents and cytotoxic efficacy (Figure 4c). Interestingly, in addition to similar total BAs and LAs contents and similar effects on cytokine release, the compositions of the individual BAs and LAs in the most potent nutraceuticals FN9, FN13, and FN16 were highly alike. Spearman's rank correlation analysis revealed that contents of BAs and LAs without a keto group exhibit the highest correlation to FN cytotoxic efficacy (Figure S1). In this regard, the content of the boswellic acid β-ABA shows the highest correlation to cytotoxicity against MDA-MB-231.

**Figure 4.** FNs are toxic to a TNBC cell line. (**a**) MDA-MB-231 cancer cells and peripheral blood mononuclear cells (PBMC) were treated for 72 h with FN16, and cell viability was analyzed by XTT assay (*n* = 3). (**b**) Half maximal inhibitory concentrations (IC50) of FNs against MDA-MB-231 cell line. Comparison of IC50 values of individual FNs with the average IC50 value for all FNs (15.9 μg/mL) by Box-Cox transformation, one-way ANOVA, and post-hoc Dunnett's test. All data are mean ± SEM, *n* = 3 for the respective FN, *n* = 16 for the average value, \* *p* < 0.05, \*\* *p* < 0.01, and \*\*\* *p* < 0.001. (**c**) Principal component analysis (PCA) and contour plot visualizing the correlation between toxicity of different FNs and their BAs and LAs contents.

Furthermore, a principal component regression analysis (PCR) based on known BAs and LAs contents in FNs was used, to avoid biases caused by multicollinearity. Sample FN5 was identified as an outlier by the Grubbs test and excluded from the regression analysis. The resulting regression model exhibited a *R*<sup>2</sup> of 96.5% and a significance of correlation of *p* < 0.001. The regression analysis yielded the following equations:

$$\text{IC}\_{50}(\mu\text{g/mL}) = 31.6 - 0.223 \,\text{PCl} - 0.668 \,\text{PCl} \\ 2 + 0.000586 \,\text{PCl}^2 + 0.00737 \,\text{PCl}^2 + 0.00166 \,\text{PCl} \times \text{PCl} \,\text{2} \qquad \text{(1)}$$

with

$$\begin{aligned} \text{PCl} &= 0.289 \left[ \text{KBA} \right] + 0.115 \left[ \text{LA} \right] + 0.316 \left[ \alpha \text{-BA} \right] + 0.830 \left[ \beta \text{-BA} \right] - 0.026 \left[ \text{AKBA} \right] + 0.056 \left[ \text{ALA} \right] \\ &+ 0.103 \left[ \alpha \text{-ABA} \right] + 0.316 \left[ \beta \text{-ABA} \right] \end{aligned} \tag{2}$$

and

$$\begin{aligned} \text{PC2 } = -0.115 \left[ \text{KBA} \right] - 0.002 \left[ \text{LA} \right] + 0.008 \left[ a \text{-BA} \right] - 0.064 \left[ \text{ $\beta$ -BA} \right] + 0.936 \left[ \text{AKBA} \right] + 0.088 \left[ \text{ALA} \right] \\ + 0.136 \left[ a \text{-} ABA \right] + 0.285 \left[ \text{ $\beta$ -ABA} \right] \end{aligned} \tag{3}$$

with concentrations of corresponding BA or LA in μg/mg FN given in square brackets.

The IC50 values calculated by the regression model exhibited an average absolute residue of only ± 1.0 μg/mL compared to the experimentally-derived values (Table 3). Hence, the regression equation could be used as a tool to predict cytotoxic efficacy of FNs and other frankincense herbal preparations with known BAs and LAs content. By expanding the data set, one could achieve an even more accurate prediction.


**Table 3.** Cytotoxic efficacy of FNs against MDA-MB-231 cell line.

XTT, IC50 values were determined experimentally by XTT assay (72 h, *n* = 3). Regression, IC50 values were determined by principal component regression (PCR). Data are mean ± standard error of the mean (SEM) or standard error of the regression model (SE). Sample FN5 as an outlier (Grubbs test) was excluded from the regression analysis. Absolute residue is the difference between experimentally determined values and those determined by regression analysis. Samples are arranged by cytotoxicity (XTT) in a descending order.

Further, the three FNs, FN9, FN13, and FN16, showing the highest cytotoxicity against MDA-MB-231 cells, were investigated for their cytotoxic efficacy against two additional TNBC cell lines, MDA-MB-453 and Cal-51. Likewise, the investigated samples exhibited considerable cytotoxicity against MDA-MB-453 with IC50 values between 12.9 μg/mL and 17.3 μg/mL, and against Cal-51 with even lower IC50 values between 3.8 μg/mL and 4.0 μg/mL (Table 4). This indicates that FNs exhibit cytotoxic efficacy against different triple-negative treatment-resistant breast cancer cells.

The non-halogenated anthracycline doxorubicin, a chemotherapeutic agent used to treat patients with breast cancer, was analyzed as a positive control for FNs. Doxorubicin exhibited an IC50 = 0.41 ± 0.03 μg/mL for MDA-MB-231, an IC50 = 0.26 ± 0.02 μg/mL for MDA-MB-453, and an IC50 = 0.033 ± 0.004 μg/mL for Cal-51. However, for doxorubicin, severe adverse effects including cardio- and nephrotoxicity have been reported [36,37]. Differently, for frankincense extracts, only mild adverse effects, like heartburn or nausea, have been described [38].


#### *3.4. Boswellic and Lupeolic Acids Inhibit Cytokine Release and Are Cytotoxic for TNBC Cells*

Cytotoxicity of FNs against MDA-MB-231 cells correlated significantly to the total contents of BAs and LAs as well as to individual BAs and LAs contents, except for AKBA, in FNs (Spearman's rank correlation, *p* < 0.001) (Figure S1). The highest correlation coefficient exhibited β-ABA, whereas no significant correlation was detected for AKBA. The poor correlation of AKBA was caused mainly by the unusual chemical composition of the sample FN8 with unexpectedly high AKBA content concomitantly with very low contents of other BAs and LAs. After rejecting sample FN8 from the statistical analysis, the AKBA contents of the remaining samples correlated significantly with the cytotoxicity against MD-MB-231 cells (*p* = 0.004). Hence, we further investigated the cytotoxic efficacies of the pure individual BAs and LAs against MDA-MB-231 (Table 5). The acetylated forms of boswellic acids were more effective against MDA-MB-231 cells than their deacetylated forms; and the most cytotoxic compounds were AKBA with IC50 = 6.6 μM, α-ABA with IC50 = 7.2 μM, and β-ABA with IC50 = 5.9 μM. Differently, for lupeolic acids, LA was more cytotoxic compared to ALA. We have previously shown that acetylated BAs, similar to anticancer drugs such as camptothecins and podophyllotoxins, inhibit activities of human topoisomerases [39]. Acetylated BAs also inhibit the activation of transcriptional factor NF-κB which regulates the synthesis of antiapoptotic proteins [11,17], whereas ALA is an inhibitor of the AKT kinase [16]. These molecular targets might explain the high cytotoxicity of BAs and LAs against proliferating cancer cells.


**Table 5.** Cytotoxic efficacies of the individual pure BAs and LAs against MDA-MB-231.

XTT assay, 72 h, *n* = 3. Acetylated boswellic acids (AKBA, α-ABA, and β-ABA) are more potent compared to their deacetylated forms (KBA, α-BA, and β-BA). Differently, LA is more potent than ALA.

Interestingly, similar to FN1 and FN16, AKBA, and in particular, β-ABA, showed higher toxicity against cancer cells compared to PBMC (Figure S2b,c), suggesting selectivity against cancer cells.

Moreover, toxicity against breast cancer cells strongly correlated with cytokine inhibition with *p* = 0.023 for IL-1β and *p* < 0.001 for TNF-α, IL-6, IL-8, and IL-10 (Figure S1). Cluster analysis of variables confirmed the correlation between toxicity against breast cancer cells and cytokine inhibition (Figure 5). In addition to the activation of genes coding for antiapoptotic proteins and for those necessary for continuous cell proliferation, NF-κB activation is central to the activation of cytokine and chemokine gene expression [40]. In previous studies, we have shown that acetylated boswellic acids AKBA and β-ABA specifically inhibited the activities of human IκB kinases (IKK) and the subsequent release of

proinflammatory cytokines by human monocytes [11]. Moreover, intercepting the IKK activity by AKBA and β-ABA inhibited proliferation and promoted apoptosis in androgen-independent PC-3 prostate cancer cells in vitro and in vivo [17]. Hence, inhibition of NF-κB by boswellic acids might represent a clue as to why these compounds inhibit both inflammatory processes and cancer growth.

**Figure 5.** Heatmap visualizing correlations between BAs and LAs compositions in FNs and FN efficacies with respect to inhibition of cytokine release and cancer cell toxicity. Cluster analysis of variables (BAs, LAs, cytokines, IC50) was performed with distances of correlation coefficients and complete linkage. Cluster analysis of objects (FN1-16) was performed with standardized variables, Euclidean distances, and complete linkage. All data were standardized and efficacy values were additionally inverted. +1 indicates high contents of BAs and LAs in FNs or high efficacies of FNs, −1 indicates low contents or efficacies. BAs and LAs contents were analyzed by HPLC-MS/MS; IC50 toxicity against MDA-MB-231 breast cancer cells was determined by XTT (72 h); and inhibition of cytokine release by PBMC was analyzed after 18 h by flow cytometry.

Interestingly, chronic inflammation is known to increase cancer risk. Particularly, activation of NF-κB and production of inflammatory cytokines by tumor-associated immune cells are important components aiding tumor initiation, growth, malignant transformation, invasion, and metastasis [41]. Hence, inhibition of inflammatory mediators prevents the development of experimental cancers, and treatment with anti-inflammatory drugs reduces cancer risks and progression [42]. Hence, inhibition of inflammation by frankincense constituents could benefit anticancer therapy.

Similar to FNs, pure BAs and LAs, in concentrations which do not affect PBMC viability, inhibited cytokine production by LPS-activated PBMC, but not as efficiently as FNs (Table S3). Although BAs and LAs are identified here as the active principle of FNs, frankincense extracts also contain a rather complex lipidome including a large number of different fatty acids [43]. The lipids are extracted from frankincense during extract preparation and are persistent components of FNs. Such lipids can act as a natural emulsifier for nonpolar compounds like BAs and LAs and increase their bioavailability [44]. Accordingly, the intake of a FN in combination with a high-fat meal improved the bioavailability of BAs in humans [45]. Hence, the intake of pure BAs and LAs in the absence of solubility enhancers might decrease their bioavailability and worsen the therapeutic response. This should be considered in the development of drugs based on BAs and LAs. Formulations with lecithin for oral administration and complexation with cyclodextrins or liposomes for injections can be beneficial [46,47].

#### *3.5. Inhibition of Proliferation and Induction of Apoptosis in Breast Cancer Xenografts*

The antitumor activity of FNs and BAs was further verified in vivo, in MDA-MB-231 breast cancer xenografts grown on the chorioallantoic membrane (CAM) of fertilized chick eggs. Xenografts were treated with FN16, because it contains high amounts of BAs and LAs and exhibited the highest cytotoxicity against MDA-MB-231 cells in vitro. Also, two acetylated BAs, AKBA and β-ABA, which exhibited the highest cytotoxic efficacy in vitro, were tested in an in vivo model.

Treatment of cancer xenografts with FN16 reduced dose-dependently and significantly the tumor volume compared to the control group (Figure 6a,b). Also, treatment with β-ABA significantly reduced the tumor volume. AKBA-treated xenografts were only non-significantly smaller than control. Immunohistochemical analysis of the tumors revealed that FN16 had a dose-dependent inhibitory effect on cancer cell proliferation (Figure 6c). Also, AKBA and β-ABA significantly inhibited cancer cell proliferation. FN16, AKBA, and β-ABA all induced apoptosis in TNBC xenografts as shown by analysis of apoptosis by using the TUNEL technique (Figure 6d). Whilst the chemotherapeutic drug doxorubicin exhibited eminent cytotoxicity against MDA-MB-231 cells in vitro, in vivo its efficacy was not sufficient to induce a significant reduction of tumor size or to induce apoptosis during the treatment period. Notably, no systemic toxicity of the FN16 or BAs on the chick embryos was observed.

**Figure 6.** FNs and BAs inhibit growth and induce apoptosis in TNBC breast cancer xenografts in vivo. MDA-MB-231 cells were grafted on the chorioallantoic membrane (CAM) of fertilized chick eggs and treated for 3 consecutive days with either FN16 (5 mg/mL and 10 mg/mL), the boswellic acids AKBA or β-ABA (each 10 μM), doxorubicin (1 μM), or DMSO (0.5%) as control. (**a**) First row: tumor photographs immediately after extraction (original magnification 50×). Second row: hematoxylin and eosin staining. Third row: staining for proliferation marker Ki-67 (red-brown nuclear stain, original magnification 200×). Fourth row: staining for the apoptosis marker TUNEL (brown, original magnification 200×). (**b**) FN16 and β-ABA inhibit the tumor growth. (**c**) FN16, AKBA, and β-ABA inhibit cancer cell proliferation. (**d**) FN16, AKBA, and β-ABA induce apoptosis in cancer xenografts. All data are mean ± SEM, *n* = 4. Comparison with control by Kruskal-Wallis one-way ANOVA on ranks and post-hoc by Dunn's test with \* *p* < 0.05, \*\* *p* < 0.01, and \*\*\* *p* < 0.001.

#### **4. Conclusions**

A comparative analysis of 16 frankincense nutraceuticals (FNs) revealed major differences in chemical compositions, cytokine modulatory properties, and cytotoxicities against triple-negative breast cancer cells. FNs with total boswellic and lupeolic acids (BAs and LAs) contents over 30% and/or β-ABA contents over 36 μg/mg significantly inhibited the expression of the proinflammatory cytokines TNF-α, IL-6, and IL-8. Moreover, FNs exhibited cytotoxic efficacy against triple-negative breast cancer cell lines MDA-MB-231, MDA-MB-453, and CAL-51 in vitro. In this regard, FNs with BAs and LAs contents over 30% and β-ABA contents over 50 μg/mg proved to be the most potent ones. A FN from this group and pure β-ABA inhibited growth and induced apoptosis in breast cancer xenografts in vivo. Moreover, a remarkable correlation between BAs and LAs contents, cytokine inhibition, and cytotoxicity against breast cancer cells could be observed. The contents of β-ABA exhibited the highest average correlation to inhibition of TNF-α, IL-6, and IL-8 cytokine release as well as cytotoxicity against breast cancer cells. Furthermore, pure β-ABA showed high cytotoxic efficacy against breast cancer cells in vitro and in vivo. Therefore, β-ABA should be considered as a compound for standardization of frankincense nutraceuticals and herbal preparations and it deserves further studies aiming at the development of new anticancer drugs.

**Supplementary Materials:** The following are available online at http://www.mdpi.com/2072-6643/11/10/2341/s1, Table S1: Frankincense nutraceutical (FN) capsule/pill contents and concentrations of boswellic and lupeolic acids per capsule/pill.; Table S2: Frankincense nutraceuticals (FNs) inhibit cytokine production by LPS-activated whole blood and isolated PBMC; Table S3: Boswellic and lupeolic acids inhibit cytokine production by LPS-activated PBMC; Figure S1: Correlations between contents of individual boswellic and lupeolic acids in frankincense nutraceuticals (FNs), FN cytokine inhibitory efficacies, and FN cytotoxicity against the MDA-MB-231 triple-negative breast cancer cell line (IC50); Figure S2: Preferential toxicity of FN1, AKBA, and β-ABA against MDA-MB-231 cells compared to PBMC. (**a**) FN1, (**b**) AKBA, and (**c**) β-ABA. XTT assay, 72 h.

**Author Contributions:** M.S., T.S. (Tatiana Syrovets), T.S. (Thomas Simmet), and L.J.R. conceived and designed the experiments; M.S., S.J.L., J.U. and K.W. performed the experiments; M.S. analyzed the data; M.S. and S.J.L. wrote the original draft; T.S. (Thomas Simmet) and T.S. (Tatiana Syrovets) reviewed and edited the draft; T.S. (Thomas Simmet) and T.S. (Tatiana Syrovets) supervised and administrated the project.

**Funding:** This work was partially supported by the Academic Center for Complementary and Integrative Medicine (AZKIM), State Ministry of Baden-Württemberg for Science, Research, and Arts.

**Acknowledgments:** We thank Felicitas Genze and Eva Winkler for expert technical assistance and Georg Huber (https://weihrauch-blog.de/bilder) for providing photo material.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **Appendix A**


**Table A1.** Frankincense nutraceuticals (FNs) analyzed in the study.


**Table A1.** *Cont.*

<sup>1</sup> Recommended daily allowance; <sup>2</sup> Best-before date.

#### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

*Article*
