Cadavers sent to the University of Utah by owners were necropsied as described previously [1]. Anatomical and pathological evaluation included 53 whole body or organ weights or dimensions, and one-to-several histological studies of each of 27 tissues. Additional samples were collected where gross lesions (e.g. tumors) were visible in non-designated tissues. Common histological observations were scored for each tissue on a severity scale of 0 (none), 1 (mild), 2 (moderate), 3 (severe). Summary scores for pathologies observed in multiple tissues (e.g. Fibrosis) were calculated for each dog by averaging the severity score over all affected tissues.

Weights of cleaned skulls were measured without the mandible. 48 standardized points were measured using a 3D digitizer. All pairwise distances between these points were calculated and used for principal component analysis. The first principal component of these distances, explaining 64% of the total variation, was used as a measure of skull size. Skull weight was regressed onto skull size using the “lm” function of R [7]. The residuals from this regression were used as a measure of skull robustness.

Samples of collagen were purified from bone (skull vault) and canine teeth (dentine) following standard procedures [8]. Briefly, the calcified bone matrix was dissolved with HCL and the residual collagen matrix subsequently purified and dried. Teeth were crushed into small pieces decalcified and purified in a similar manner to the bone sample, following standard protocols [2]. Dry material was separated for isotope analyses. Skull vault and dentine collagen samples from three dogs were analyzed for purity by mass spectrometry. All samples yielded peptide fingerprints identifying the protein samples as collagen.

Sample analyses. Collagen samples were analyzed using an isotope ratio mass spectrometer operated in continuous flow (CF-IRMS) mode in the Stable Isotope Ratio Facility for Environmental Research (SIRFER) at the University of Utah. For δ¹³C and δ¹⁵N analyses, 1 mg (± 10%) of material was used; for δ³⁴S, 9 to 12 mg (±10%) were used. The collagen material was placed into small tin capsules that were then loaded into a zero-blank autosampler interfaced with an elemental analyzer (Carlo Erba) where they were combusted to produce N₂ and CO₂ or SO₂. These gases were chromatographically separated and carried to the CF-IRMS (Finnigan MAT Delta S). All samples were analyzed alongside a set of internal laboratory reference materials that had been previously calibrated against international standards. Results for δ¹³C values are presented on the Vienna Pee Dee Belemnite (VPDB) scale, those for δ¹⁵N values on the AIR scale, and for δ³⁴S values on the Vienna Canyon Diablo Troilite (VCDT) scale. The analytical precision (1), based on long-term measurements of internal laboratory reference materials for δ¹³C, δ¹⁵N and δ³⁴S values, was 0.1‰, 0.2‰ and 0.4‰, respectively. Stable isotope values are reported using the standard δ-notation relative to an international standard in units “per mil” (‰) as follows: δ = 1000 (R_sample/R_standard − 1), where R_sample and R_standard are the molar ratios of the heavy to light isotopes of the sample and standard, respectively.

Correlations and permutations. Pearson product correlations “r” between two trait vectors (t1 and t2) were calculated with the “cor (t1, t2, use = “pair”)” function of R [7]. Significance was established using 100,000 - 1,000,000 permutations where one trait was randomized with respect to the other. P-values are reported as the fraction of permutation trials for which the absolute value of r was greater than or equal to the non-permuted absolute value of r.

Heritability. Heritability was estimated using a generalized linear mixed model from the MCMCglmm package [9] of R [7]. The MCMCglmm function was run using the marker validated pedigree for the Portuguese Water dog population and the following settings: nitt = 1,000,000, thin = 195, burnin = 25,000.

Skull weights from PW dogs were regressed on skull size (figure 1(a)), and values of residuals (robustness) determined (figure 1(b)). Based on pedigree and molecular (DNA marker) relatedness of the dogs (see Methods) we have measured the heritability of this trait. Although the range of residual values is large, we were unable to establish heritability suggesting that heritability, if it exists, is less than 23%. Correlating residuals with specific pathologies suggested that specific correlations might exist, but that correlations were low (~20%) and therefore the amount of variation involved was small, ~5%. Although analysis of histopathological change was extensive (see Methods and [1]), only a few correlations were potentially significant. Table 1 presents the most significant correlations observed. Not unexpectedly, less robust skulls were correlated significantly with several pathologies. An unexpected result was the small, but significant correlation of more robust skulls with the frequency of sarcomas.

Figure 2 presents isotope ratios for C, N and S from