Odour-extraction methods and breadth of sampling.

(a) Number and distribution of sampled animal species (and families) across phylogeny (time-calibrated phylogeny shown; see Methods for source data). (b–d) Three different types of glass chamber were used for odour sampling: (b) enclosed chambers of varying sizes for small and/or hairless animals, (c) a chamber with a Teflon gasket to accommodate a human arm, and (d) a narrow, heated tube containing tightly packed animal hair. Hair samples were heated to approximate body temperature (38°C) during odour extraction. (e) Sample types and sources. Coloured lines represent odour samples from different taxonomic groups (left) that were sampled from hair or live animals (centre) acquired from different sources (right). Line thickness reflects the number of unique species (dark shades) or within-species replicates (light shades). (f) Number of replicate individuals sampled for each species. Colours in (e,f) represent the same taxonomic groups listed in (a).

Odour of vertebrate animals.

(a) Odour composition of one representative from each animal species sampled. The cladogram shows phylogenetic relationships (see Methods for source data). The length of each coloured bar shows the relative proportion of individual compounds in a sample. Also see Fig. S4 for an interactive plot showing all species replicates (including common names) and individual compound names. (b) Number of compounds found in each sample. (c) Frequency of compounds across samples (presence/absence). (d) Pairwise compound correlations across species for the real dataset (upper triangle) and a shuffled dataset (lower triangle; see Methods for details on shuffling). See Fig. S6 for compound labels. (e) Empirical cumulative distribution functions (ECDFs) of all pairwise compound correlations for the actual data (pink) and the median across 1000 shuffled datasets (green) (significantly different by Kolmogorov–Smirnov test, P < 10-15). For shuffled data, the narrow 95% credible interval (light-green shading) is barely visible behind the green line. Shaded density plots illustrate the difference between the two ECDFs on a 10×-magnified scale.

Comparison of vertebrate and floral odour-space.

Top row: vertebrate odour space (this study), bottom row: floral odour space (Kantsa et al. 2018). (a,b) Number of compounds found in the odour of each species. (c,d) Proportion of species in which each compound is found. N = 64 and 41 for vertebrate and flower samples, respectively. (e,f) The total number of compounds one expects to encounter across a given proportion of all species (x axis) for a given sensitivity threshold (shades of pink). Lines and shading show medians and interquartile ranges for 100 randomly sampled sets containing the given number of odour samples. Lower sensitivity thresholds allow less abundant compounds in each species to contribute to the total.

Phylogenetic signal in odour-blend composition.

(a) Time-calibrated phylogenetic tree for sampled animal species (left) shown with linkages to a hierarchical clustering of odour profiles based on Euclidean distance (right). One representative sample from each species was chosen. Nodes have been rotated to optimize vertical matching of tips between trees. (b) Pairwise Euclidean distance (in n-dimensional odour space) between odour profiles (one from each species) as a function of divergence time. Mantel test P = 0.04 for all animals; P = 0.52 for mammals only. (c) Unscaled PCA of odour blends showing all biological replicates from all animal species. Coloured dots connected by lines represent replicates of the same species. Grey dots represent species that were only sampled once. Inset: Number of replicates sampled from each species. Same as Fig. 1f, but colours now match the PCA. (d–f) Median odour distance within and between species (N = 22), families (N = 10), and orders (N = 7) (see Methods for details). P-values shown for paired t-tests.

Compounds informative for host discrimination.

(a) Odour-discrimination tasks for a generalist blood feeder: distinguishing between animals and non-animals. (b) Compound abundance (relative proportion within a sample) in target and non-target odours for a generalist. N = 64 animal samples (one per species) vs. 23 other samples. Dots and lines show medians and interquartile ranges. Points along the axes are jittered. Lines corresponding to points at the origin have been omitted for clarity. Compounds are coloured according to the scheme in Fig. 2a. (c) Volcano plot showing two metrics for each compound for the generalist task: (1) scaled fold difference between target and non-target (see Methods) and (2) significance level of that difference. P-values come from Kolmogorov–Smirnov tests followed by Benjamini–Hochberg multiple-test correction. Red compounds are those with higher abundance in the target; black for non-target. (d–f) Same as (a–c), but for a human specialist distinguishing between humans and all other odour blends. N = 12 humans vs. 63 other animals (one representative per species) and 23 other samples.

Comparison between odour extracted from live hosts vs. hair samples.

(a) Odour composition of 7 individuals spanning 3 species. Human hair samples were taken from the head, and rat and guinea-pig samples from the whole back and sides of the animal. The guinea pig urinated and defecated during the live-host extraction. (b) PCA of the odour profiles in (a). Triangles and circles denote live-host and hair extractions, respectively. Line segments connect odour blends from the same individual. Note that the live-host sample from human subject 3 is an outlier in the PCA because it contains a high level of the terpene limonene, a common component of soaps and other skin products.

Odour-analysis pipeline.

(a) Schematic of pipeline for identifying compounds present in the odour blends and creating a custom library of compounds. (b) Schematic of pipeline for deconvoluting compounds and estimating their abundance. See Methods for details.

Additional compound information.

(a) Molecular weight of compounds found in the dataset. Note that our methods undersample highly volatile (low-AMU) compounds. (b) Circle areas denote the number of compounds belonging to each class. Line widths show the number of compounds shared between two given classes. For visualization purposes in all other plots, compounds belonging to multiple classes were assigned a main class in descending order of priority as follows: nitrogenous/sulphurous, terpene/terpenoid (only if Ncarbons was a multiple of 5), aromatic, carbonyl-containing (aldehyde/ketone/ester/acid), alcohol, ether, hydrocarbon.

Interactive barplot showing odour profiles of all sampled individuals.

Static screenshot shown above; please download the file for full functionality. A time-calibrated phylogeny is shown for animal samples. Colour scheme is the same as in Fig. 2a. Compounds are listed (in order by compound class, except for the first 9 compounds) in the scrollable legend on the right. Mousing over a compound in the legend highlights that compound in all samples. Mousing over a coloured bar in the barplot also highlights the given compound across all other samples and additionally shows a tooltip with more information about the focal sample, an image of the compound structure, and the exact numerical abundance of the compound in that sample.

Compounds x samples heatmap.

Heatmap shows compound abundance across all samples (including species replicates and non-host samples). Bars on the left show frequency (presence/absence) of compounds across all samples, coloured by compound class according to the colour scheme in Fig. S3b. Column immediately to the right of bars shows the median abundance of each compound across samples, excluding samples where the compound was not detected. Compounds are clustered by their abundance across samples (complete-linkage method), shown on the right. Samples are ordered as in Fig. S4.

Pairwise compound correlations across species.

Same correlation matrix as upper left in Fig. 2d, but with compound labels. Rows and columns are ordered by hierarchical clustering of compound correlations (complete-linkage method). Note that compound order differs from that in Fig. S5 because (1) the correlation analysis includes only one representative per species and (2) it does not consider non-host samples. Coloured boxes next to compound labels denote compound class according to the scheme in Fig. 2c, S3b.

Association between phylogeny and odour profile for subclades.

Same type of analysis as in Fig. 4a, but for four smaller clades: (a) reptiles and birds, (b) ungulates, (c) carnivores, and (d) primates. P-values shown are for Mantel tests, adjusted for multiple comparisons using the Benjamini–Hochberg method.

Composition of odour blends and host discrimination.

(a–c) Overview of odour blends used in the analyses underlying Fig. 5. (a) Odour-blend composition of one representative from each animal species sampled. Same data as in Fig. 2a. (b) Odour blends from 12 individual humans. (c) Odour blends from 23 non- host extracts. The analyses in Fig. 5b,c compare the blends shown in (a) to the blends shown in (c). Analyses in Fig. 5e,f compare the human odour blends shown in (b) to the non-human blends shown in both (a) and (c). (d) Number of compounds (and percentage of the total) shared among groups of odour blends. (e) Rank change in informativeness of compounds for generalists vs. specialists. Grey bars show compounds ranked and shaded by scaled fold difference (see Fig. 5c,f) for the generalist (top) and specialist (bottom) tasks. Lines trace changes in rank between the tasks.