Comprehensive coverage of human last meal components revealed by a forensic DNA metabarcoding approach
In this study we successfully applied a DNA metabarcoding approach to identify consumed food items of plant and animal origin in human stomach content samples, even when digestion was advanced and macroscopic inspection no longer possible. A wide panel of common and less common edible food items were found, including meat, fish, legumes, cereals, nuts, fruits and spices. So far, gastric content analyses in a forensic context are typically based on microscopic and macroscopic identification of food items (reviewed e.g. in1). However, this approach is characterised by low taxonomic resolution, low sensitivity, and proves ineffective when meal leftovers are rendered unidentifiable due to chewing and digestive processes. In the field of molecular ecology, studies on animals have shown that morphological identification of prey items in the stomach underestimates prey diversity, which is particularly true when digestion is advanced (e.g.25). The only study to date applying DNA metabarcoding to infer human diet was based on faecal samples and did not assess any animal components of diet, although including a controlled feeding trial of an animal-based diet12. The comparison of the obtained plant DNA sequences to self-reporting indicated that, while some items were not reported but detected by DNA metabarcoding, all but one self-reported items were detected (the only exception being coffee), thus highlighting the sensitivity of the method. The present study, based on a random sampling of 48 human stomach contents collected during routine autopsies, includes a higher number of vegetal items and shows for the first time the successful detection of dietary items of animal origin. We found no correlation between the diversity of species detected and the time since death or digestion degree, which advocates for the utility of this methodology. The Vert01 primer set, highly specific to vertebrates, enables to distinguish between commonly eaten animal taxa and is clearly advantageous over morphological identification. In line with regional eating habits and previously published diet surveys26, we found within the 48 samples mainly pig, cattle/dairy and OTUs assigned to the plant families Poaceae, Rosaceae and Asteraceae (likely cereals, fruits, lettuces; Fig. 1). We did not detect coffee (Coffea spp.) in any of the stomach content samples, in line with12, which might be due to a degrading effect of roasting procedures on DNA, the absence of this popular beverage in all of the stomach samples being unlikely. Similarly, although common in Swiss eating habits, we also did not detect potato, which is usually eaten boiled or baked. Note that additional edible plant species, not listed in Fig. 2 since not constituting at least 10% of RRA but with 100% match with the database, were also detected (e.g. buckwheat, citrus fruits, flax, mangoes, sesame; Supplementary Table S2). Because we could obviously not compare our results to self-reported diets, we applied very stringent filtering parameters to avoid the occurrence of false positives (see Bioinformatic data treatment). It is beyond the approach of this study to distinguish between the animal source and a final processed food item (e.g. dairy or egg products) based on the obtained DNA sequences. However, this could be achieved by complementing the primer set with a bacterial marker (to e.g. identify the presence of a particular cheese27) or using proteomics (see below).Overall, the Vert01 metabarcode is able to discriminate well among commonly eaten genera. However, owing to its limited taxonomic resolution (72.4% at the species level, based on in silico testing11), species-level distinction is not always possible (e.g. between perch and pikeperch) or between potentially-eaten wild species and their conspecific domestic counterparts (e.g. wild boar and pig). In Fig. 2, we present the taxonomical assignation done using ObiTools together with a common name, selected after manually inspecting each sequence using BLAST and only considering 100% matches with edible species. In some cases, the common name refers to a group of species because the barcode was not specific enough to distinguish between genera or species. This is more relevant concerning plants, as the Sper01 metabarcode length ranges from 10 to 220 bp, implying that some items with shorter metabarcode and/or closely related phylogenetically could not be distinguished to genus or species level due to limited resolutive power. This is related to the nature of this universal plant marker, which has been designed to target a region of the trnL intron of chloroplast DNA which lacks taxonomic resolution within several plant families (only 21.5% resolution at the species level9,11) but has wide taxonomic coverage. This trade-off meant for our study that we could genetically not distinguish between some close species which are clearly different morphologically (e.g. stone fruits, cucurbits). To overcome this issue and increase the taxonomic resolution of the results, it is possible to envisage multiplexing within the same PCR of additional primers specifically targeting groups of species that cannot be identified at the species level by the P6 loop of the trnL intron. Such a strategy has already been implemented to distinguish between Carpinus betulus and Corylus avellana in bison diet28. Furthermore, it must be outlined that by using these primer sets only, diet assessment is not comprehensive as it does not target all possibly present food products. Even so-called universal primers may result in preferential amplification of some taxa over others and non-amplification of target taxa29,30. For this pilot study, we chose to use two universal PCR primer pairs with wide taxonomic coverage but limited specific resolution, in order to detect a broad range of items. To gain resolution for specific vertebrate or plant taxonomic groups (e.g. fish, birds, cereals) or target taxa not covered by these primers and which could be of forensic interest (e.g. marine crustaceans and molluscs, algae, fungi), it is possible to complement Vert01 and Sper01 with additional, taxonomically-restricted PCR metabarcoding primers described in the literature (e.g.31; examples reviewed in11). Taxonomic assignation of an unknown DNA sequence strongly depends on the exhaustiveness and quality of a reference database, either public as e.g. GenBank or custom-made/local (reviewed in32). In case of a priori knowledge of the overall consumed diet in samples, local databases may be restrained to the expected DNA sequences, which subsequently improves taxonomic assignment. For this study we in silico compiled databases containing all possible sequences amplified by our markers, but restricted these to vertebrates and spermatophytes (i.e. seed plants), respectively.The duration of stomach emptying has been estimated by the percentage of a meal present in a stomach3, but this process is influenced by several variables including the type and volume of consumed food, lifestyle and health, and can therefore last from few hours to days2. While one could argue that plant items usually remain longer in the stomach, our findings do not allow to draw robust conclusions about correlations of certain food items and digestion times. In order to establish hypotheses useful for time-frame estimations, additional experiments are necessary. In a controversial case of death, MS-based proteomics provided additional information through the analysis of food-derived proteins and peptides in the gastric content sampled at autopsy, indicating a last breakfast of milk and bread. While this method is certainly promising, it might reveal difficult if digestion is in an advanced stage, and has a less comprehensive scope than a DNA metabarcoding assay33. Furthermore, the effect of food processing techniques on DNA quality must be taken into account since cooking denatures e.g. proteins which in turn renders DNA amplification preferential to immunological approaches1. Different cooking treatments (variable duration of boiling, frying, baking) of tomato seeds showed that DNA extraction yielded in good quality DNA only for fresh seeds34, while digestion did not destroy DNA21. Hence, there might be an implicit bias of DNA metabarcoding to preferentially detect non-processed food (i.e. raw versus cooked). Another issue of environmental DNA-based methods is that it is not possible to distinguish between different states of food products based on DNA sequences. As mentioned before, we could not discriminate between e.g. grapes/wine, fruits/juices, beef meat/dairy products or chicken meat/eggs, since the DNA sequence of a derived product is identical to the DNA sequence of its source. While it is less common to encounter such biases for plants, mainly in cereal-derived products, it has to be taken into account when extrapolating diet patterns from DNA metabarcoding results.Stomach content sampling is invasive, but advantageous or even required with certain animal species and in particular circumstances, including definitely the human forensic context. An advantage of stomach content over faecal samples is that food is in an early stage of digestion before passing through the pyloric sphincter into the intestines, thus the effects of inhibition by bacteria or enzymes and degradation of DNA are less significant11,18. While some food particles such as seeds sometimes remain identifiable, even morphologically, after passing through the digestive system21, others do not and the same applies to DNA which is degraded by the digestive processes taking place in the intestinal tract. In a controlled feeding experiment on insects, the detectability of food DNA in different types of dietary samples showed that regurgitates and entire animals (including stomach content) outperformed faeces regarding detectability of prey DNA13. While food journals in dietary surveys may contain errors or deliberate omissions12, they are a comprehensive and easily accessible method of human diet assessment. However, in case of deceased persons that option is no longer available.Stomach content analyses provided crucial information for criminal investigations about cases of sudden and unexplained death on numerous occasions in recent years, enabling investigators to interpret perimortem events in detail (case examples reviewed in2). The results of this pilot study show that human stomach content analyses by DNA metabarcoding can be used as a complementary tool to traditional forensic macro- and microscopic approaches, with clear advantages such as an almost unlimited flexibility in terms of nature and range of taxa targeted, as well as high sensitivity and taxonomic resolution. Consequently, information that might otherwise remain undetected can be revealed, highlighting timings and circumstances surrounding the last hours of a person and his/her food intake. In a broader perspective, taking into account the potential improvements and refinements described above, and the growing amount of research literature available for wildlife species (i.e. environmental DNA-based studies), our results open up promising and novel prospects in the broader framework of human biomedical investigations of dietary patterns, based on partially or fully digested food found in the gastrointestinal tract or in faecal samples. More
