How similar is similar enough? A sufficient similarity case study with Ginkgo biloba extract

Abstract

Botanical dietary supplements are complex mixtures that can be highly variable in composition and quality, making safety evaluation difficult. A key challenge is determining how diverse products in the marketplace relate to chemically and toxicologically characterized reference samples (i.e., how similar must a product be in order to be well-represented by the tested reference sample?). Ginkgo biloba extract (GBE) was used as a case study to develop and evaluate approaches for determining sufficient similarity. Multiple GBE extracts were evaluated for chemical and biological-response similarity. Chemical similarity was assessed using untargeted and targeted chemistry approaches. Biological similarity was evaluated using in vitro liver models and short-term rodent studies. Statistical and data visualization methods were then used to make decisions about the similarity of products to the reference sample. A majority of the 26 GBE samples tested (62%) were consistently determined to be sufficiently similar to the reference sample, while 27% were different from the reference GBE, and 12% were either similar or different depending on the method used. This case study demonstrated that approaches to evaluate sufficient similarity allow for critical evaluation of complex mixtures so that safety data from the tested reference can be applied to untested materials.

Figure 1

Framework for determining sufficient similarity of mixture(s)-of-interest to a reference mixture that has been chemically and toxicologically characterized.

Figure 2

Illustration of hierarchical clustering-based categorization of samples as “similar” to, “maybe similar” to, and “different” from the GBE reference (sample 1). According to rule 1, sample A is considered to be “similar” to sample 1 because they are in the “most similar” cluster. According to rule 2, samples C, D, and E are “different” from sample 1 because they are in the “most different” clusters. According to rule 3, sample B is “maybe similar” to sample 1 because it does not belong to either of the two categories described above.

Figure 3

Example of the process for determining similarity within a data stream (i.e., primary human hepatocyte gene expression data). The Ginkgo biloba extract samples separated into three clusters, which are represented in the dendrogram (left) and constellation plots (right) by red, green, and blue. Similarity groupings are identified by color and line style in the constellation plot: “similar” to the reference sample (green - dashed), “maybe similar” (yellow - solid), and “different” (red - dotted).

Figure 4

Line plot comparison of Ginkgo biloba samples across chemical (A) and biological-response data from in vitro primary human hepatocytes (PHH; B) and in vivo data from the 5-day rat study (C). A similarity intersect was drawn based on consideration of chemical and biological-response data. In (A), (B), and (C), each circle represents the data for one of the GBE samples, and the position of the circles on the line represents the chemical similarity of the samples to the reference (sample 1) with increasing chemical difference indicated by increasing distance from sample 1 (e.g., sample H is the most chemically different from sample 1). In (B), the circles represent the chemistry data and the green, yellow, and red colors represent the similarity determination (similar, maybe similar, not similar, respectively). A subset of in vivo endpoints are included in (C) including liver weight and gene expression measures indicating the degree of change in genes associated with lipid accumulation (average lipid accumulation score) and the most sensitive gene response associated with any set of genes associated with a biological process or pathway (biological process sensitivity). In (C), the magnitude of biological effect is reflected by the size of the dot, blue represents a significant change from control and gray represents no significant difference from control.