Introduction

Breastfeeding is one of the most important factors for the survival of mammalian infants, because breast milk provides immunological benefits and precious nutrients1. Infants that grow up without breastfeeding are statistically prone to higher mortality2. Therefore, accurate estimation of breastfeeding status is important to investigate the cause of infant mortality in mammals. Although behavioral observation and doubly labeled water methods are used to estimate breastfeeding status of living individuals3, these methods cannot be applied to postmortem estimation.

Stable isotope analysis has been used for the estimation of breastfeeding status in modern and ancient biological tissues from several mammalian species4. This method, however, cannot be reliably applied to identify the breastfeeding status of neonates, because several weeks or months are required for given tissues to reflect the isotopic signal of the diet. Accordingly, the isotopic signal of breastfeeding is hardly detectable in neonates who die soon after birth. Furthermore, nitrogen isotope ratios are affected by other factors than breastfeeding and weaning5, and the increased nitrogen isotope ratios in infants are not necessarily direct evidence of breast milk consumption.

To overcome these limits, we attempted reconstruction of the breastfeeding status in neonates, by directly detecting milk-specific proteins still associated with archaeological neonatal skeletal remains. We hypothesized that, in a neonate who died after ingesting milk through breastfeeding, milk proteins would diffuse from the digestive system into the surrounding bones during decomposition. Consequently, we tested whether these milk proteins can be confidently identified by tandem mass spectrometry (MS)-based sequencing, assuming the infant remains did not experience remarkable postmortem disturbances.

Recently, palaeoproteomic analysis has enabled comprehensive identification of even low amounts of proteins and peptides in ancient biological tissues6. Proteomic analysis by liquid chromatography coupled to tandem mass spectrometry (LC-MS/MS) has been used in bioarchaeological and evolutionary studies, such as reconstruction of protein expression profiles7, phylogeny8,9, diet10,11, and physiological status12,13, as well as species14,15 and sex16 identification. Palaeoproteomics also allowed detection of dinosaur proteins17, even though some caution has been raised toward these results18. While dietary reconstruction by proteomic analysis has targeted food remains11,19 or dental calculus10,20, we report the detection of breast milk-specific proteins from ancient neonatal dog bone remains.

Materials and Methods

Hamanaka 2 site and the dog neonate

A neonatal dog (Canis lupus familiaris) skeleton (2017HA1016) was excavated from the Nakatani location of the Hamanaka 2 site, Rebun Island, Hokkaido, Japan (Supplementary Fig. 1). Hamanaka 2 site is a multi-component shell midden, spanning from final Jomon (3000−2300 years BP) to historical Ainu (400 years BP) periods. The cool temperature of Rebun Island (mean annual temperature in the period 1978–2002 was 6.6 °C21) and the presence of shell promotes good preservation of organic materials at the Hamanaka 2 site. Since 2011, archaeological excavations have yielded human remains, rich faunal remains, lithics and pottery fragments22,23,24,25.

The 2017HA1016 remains, originating from the Kokumon subperiod of the Okhotsk cultural layer (430–960 cal AD22) and partially retaining their anatomical articulation, were found in 2017. Excavation and collection of the 2017HA1016 skeletal remains, as well as of an adjacent fish bone from the same context, was done wearing a face mask and nitrile gloves. Additional bones were recovered using a 4 mm mesh. Some of these specimens were frozen within one hour after excavation (see Table 1). The age at death of 2017HA1016 was estimated using the reference chart of tooth development and eruption for modern Japanese dogs26.

Table 1 Detail of the analyzed samples.
Full size table

Palaeoproteomic analysis

Ancient proteins were extracted from an entire 2017HA1016 vertebra body and three rib fragments. As negative controls, a fish bone and a soil specimen, collected less than 10 cm away from the 2017HA1016 remains, were also processed using the same experimental workflow (Table 1). Details of the proteomic methodology are reported as Supplementary Information. Briefly: bones were decalcified with EDTA solution and the proteins in the EDTA supernatant and in the collagenous pellet were separately denatured27, reduced, and alkylated (adapted from28). Protein solutions were then digested using trypsin overnight at 37 °C. Tryptic peptides from both fractions of each bone extract were purified using Stage Tips with C18 membrane29 and analyzed separately, unless otherwise specified, by nanoflow liquid chromatography tandem mass spectrometry (nLC-MS/MS), using an EASY-nLC 1200 connected to a Q-Exactive HF-X (ThermoFisher, Bremen - Germany).

RAW data files generated by LC-MS/MS were searched against a Canis lupus familiaris proteome database downloaded from Uniprot, and a common laboratory contaminant database, with the MaxQuant software version 1.5.3.3030. Protein groups having at least 2 different non-overlapping peptides were considered confidently identified, unless otherwise indicated. All protein hits that could be considered possible contamination products were excluded from further analysis. Deamidation rates for individual samples were calculated with a Python script15. Detected proteins were classified using PANTHER database version 13.131.

Results

Morphology of the dog neonate and estimation of its age at death

As the deciduous first molar does not erupt while the deciduous canine and deciduous second molar are erupting in the maxilla, the age at death of 2017HA1016 was estimated at 2 weeks after birth, based on the Mori’s reference chart26 (Fig. 1). Most elements of the skeleton were preserved, and the recovered bones of the skeleton are described in Supplementary Table 1.

Figure 1
figure 1

Photo of the main bones of neonatal dog skeleton (2017HA1016): (1) maxilla and frontal; (2) parietal; (3) scapula R; (4) humerus L; (5) radius R; (6) tibia R; (7) femur R.

Full size image

Detected proteins

Only collagen α-1(I) fragments (COL1A1) were detected in the fish bone by searching against the dog proteome, and only actin cytoplasmic 1 (ACTB) and COL1A1 were detected with ≥2 non-overlapping peptides in the soil sample. The following results and discussion are therefore limited to the data obtained from the analysis of the bone fragments of the 2017HA1016 dog. Protein groups identified in blanks but not in the dog bones with ≥2 peptides were not considered.

A total of 200 protein groups were detected across the EDTA and pellet fractions of the rib and vertebra bones (Supplementary Table 2). A total of 143 protein groups (71.5%) were detected in both the EDTA and the pellet fractions, and 154 protein groups (77.0%) were retrieved from both rib and vertebra. Recovered proteomes can differ in multiple experimental fractions extracted from a single sample32 and from different bone elements from a single individual33. PANTHER analysis indicated that 51.6% of protein groups are classified as an extracellular component (Supplementary Table 3) and 41.3% are related with binding function (Supplementary Table 4). Mean deamidation rates of samples were 36.3 ± 4.8% for asparagine (N) and 11.0 ± 2.1% for glutamine (Q) (n = 8; Supplementary Table 5). Deamidation of glutamine and asparagine residues is a non-enzymatic modification that occurs over time, resulting in a + 0.98402 Da mass shift caused by the direct or indirect hydrolysis of the N and Q side-chain amide group34.

Biological marker proteins

Protein groups characteristic of fetuses or newborns were detected from bones of 2017HA1016 (Table 2). Alpha-fetoprotein (AFP), a major plasma protein expressed in yolk sac and fetal liver and only detected in fetuses or young neonates35, was found in all of the four bones. AFP may play a role comparable to that of serum albumin in the adult, and its concentration is at its highest just after delivery (14080 ± 5944 μg/mL) and rapidly decreases during the first 2 weeks after birth (70.21 ± 52.92 μg/mL) in dogs35.

Table 2 Characteristic protein groups for fetus or growing infant and milk that were detected in 2017HA1016.
Full size table

Several protein groups that are related with endochondral bone formation were detected (Table 2). Most bones, including ribs and vertebrae, develop via endochondral bone formation; endochondral cartilage templates are then replaced by calcified bone matrix36. Collagen type X, alpha-1 (COL10A1) has functions in bone formation, and is only expressed in endochondral cartilage that will be replaced by mature bone tissue in humans37. Epiphycan (EPYC) is a small leucine-rich proteoglycan that is mostly found in fetal and neonatal epiphyseal cartilage in mice, bovines, and chickens38. EPYC has important roles in cartilage development and its maintenance39. Detection of these proteins (Table 2) is consistent with the age at death of the 2017HA1016 individual and it confirms that palaeoproteomics can retrieve growth- or developmental stage-specific proteins from archaeological bone remain28,40.

In addition, two milk proteins were detected from the EDTA fractions of two distinct subsamples, i.e. the distal and proximal ends of a single rib. Two peptides of beta-lactoglobulin-1 (LGB1) were detected from the EDTA fraction of the distal part of 2017HA1016’s rib (Fig. 2, Table 3). LGB1 is a major whey protein, absent in some mammalian species, including humans41. Although its exact physiological function is not determined yet, LGB1 binds several hydrophobic ligands and thus may act as specific transporters41.

Figure 2
figure 2

MS2 spectra of peptides assigned to LGB1.

Full size image
Table 3 Matched peptides for milk proteins.
Full size table

Protein BLAST searches indicated that the two detected peptides of LGB1 sequences were identical to those of dogs with the highest score. One of the detected peptide sequences (TLEVDNEVMEK) also matches the sequence of glycodelin of Vulpes vulpes (red fox). The identification of the other sequence (TMEDLDLQK) is supported by a spectrum including the complete y-ion series (Fig. 2), however, it’s  quality is lower, most probably due to random co-fragmentation of the precursor with another peptide. This spectrum could also be assigned equally confidently to the deamidated (N → D) peptide sequence of LGB from Callorhinus ursinus (northern fur seal), Odobenus rosmarus divergens (Pacific walrus), and Leptonychotes weddellii (Weddell seal). However, its assignment to dog LGB1 represents the most parsimonious interpretation.

In addition, two peptides of whey acidic protein (WAP) were detected from EDTA fraction of the proximal part of 2017HA1016’s rib (Fig. 3, Table 3). WAP is a major whey protein present in dog milk, which seems to play important roles in regulating the proliferation of mammary epithelial cells42. The detected peptides of WAP perfectly match with the dog WAP sequence reported by Seki and colleagues42. Protein BLAST searches indicated that the sequences recovered uniquely match dog WAP. Unassigned higher peaks in MS2 spectrum of CCLSVCAMR (Fig. 3) were mostly originated from neutral losses (Supplementary Fig. 2).

Figure 3
figure 3

MS2 spectra of peptides assigned to WAP.

Full size image

Since the neonatal dog is too young to secrete breast milk, the detected peptides of LGB1 and WAP would most probably originate from its mother. The alternative origin of LGB1 and WAP from the above-mentioned non-dog species is archaeologically and ecologically highly unlikely. The possibility of contamination is excluded because no LGB1 and WAP peptides were detected in the negative controls. Finally, the only species that matches all observed peptides is Canis lupus familiaris.

We also detected protein groups that are expressed in milk, but also in other tissues. Milk fat globule-EGF factor 8 (MFGE8) facilitates scavenging of the dying cells from the tissue and is an essential factor for controling the progression of various inflammatory diseases43. MFGE8 is a component of milk fat globule membrane protein, but it is also expressed ubiquitously in other cells and tissues43. Fourteen peptide-spectrum matches of MFGE8 were obtained from the dog bone samples.

Discussion

Authenticity of the recovered proteome

The detection of specific proteins exclusively expressed in neonates14 in 2017HA1016 (Table 2) is consistent with the morphological observation that confidently established this individual died approximately two weeks after birth. In particular, AFP35 and EPYC38,39 are expressed in fetal liver and endochondral cartilage, respectively. COL10A1 is present only in ossifying endochondral cartilage and will be replaced in mature bone tissue37. This evidence supports the authentic endogenous origin of the palaeoproteome retrieved.

We also measured relatively high rates of asparagine (36.3 ± 4.8%) and glutamine (11.0 ± 2.1%) deamidation (n = 8; Supplementary Table 5), compared to modern similar samples34. This observation is compatible with the endogenous ancient origin of the peptides we detected in the dog neonate, as high deamidation rates are generally correlated, in archaeological samples, with prolonged postmortem protein degradation34,44 (but see45).

Detection of breast milk proteins

Although the original articulation of the 2017HA1016 skeleton was retained only partially, there was no evidence indicating the burial was affected by diagenetic or ritual manipulation. No other dog remains were found around 2017HA1016. Similarly, no archaeological or ethnographic evidence of human exploitation and consumption of domesticated dog milk is documented in Hokkaido. Additionally, the negative controls processed in parallel to the dog bone samples did not show any evidence of canine milk proteins. Therefore, it is reasonable to conclude that the detected LGB1 and WAP peptides do not represent contamination products derived from anthropogenic or postmortem processes.

After excluding the possibility that the dog-specific LGB1 and WAP peptides originate from contamination, as well as anthropogenic or diagenetic activities, we conclude they originated from the breast milk the neonate ingested just before dying. During the following decomposition, the milk eventually diffused from the digestive system and was absorbed into the bone tissue. Neonatal dogs consume a large amount of breast milk, with consumption reaching its peak at 3–4 weeks after birth46. At 19 and 26 days after birth, breast milk represents respectively 17.0% and 14.6% of the animal body mass46. It is highly plausible that 2017HA1016 consumed such large amounts of breast milk just before its death, and consequently that some protein residues from such a relatively large amount of undigested milk diffused inside the body of the decomposing neonate dog soon after its death. Those protein residues eventually reached some districts of the dog skeleton where, by complexing with the mineral bone matrix, they were preserved until detected by MS analysis. It has been repeatedly observed that ancient protein preservation is higher for those residues tightly bound to the mineral matrix of bones, dental enamel, or eggshells10,20,47,48. This mechanism seems to represent a key factor in reducing spontaneous ancient protein backbone hydrolysis over extremely long time intervals47,48. Being the milk protein most frequently retrieved in archaeological human dental calculus20 and pottery matrix47, LGB1 is consistently preserved relatively better than other proteins in archaeological samples. Interestingly, the ruminant homologs of one of the LGB1 peptides (TLEVDNEVMEK) detected in this study, such as TPEVDDEALEK in bovine LGB, represent the most frequently detected LGB peptides (n = 70/135) in Bronze Age human dental calculus49. Both the TLEVDNEVMEK and the TPEVDDEALEK peptides are particularly rich in acidic amino acids, i.e. D, E and deamidated N, which are known to directly bind biominerals, enabling ancient peptide recovery even after millions of years in climatically adverse environments48. Most probably such a mechanism favoured the preservation of the ancient breast milk peptides we detected in association with the 2017HA1016 skeletal remains.

The MFGE8 protein was also detected in several samples. Studies with human infants show that there is a much higher concentration of this protein in the gut of those neonates fed with maternal milk compared to bottle-fed ones50, indicating that there is a definite transfer of this protein from the mother through her breast milk. MFGE8, however, has also been detected in archaeological adult human28 and mammoth7 bones. Detection of proteins not exclusively expressed in breast milk is therefore indicative, but not conclusive, evidence of milk consumption.

Proteomic reconstruction of breastfeeding status

To the best of our knowledge, this is the first study that reports the survival of ingested breast milk proteins in ancient skeletal material. The results of this study demonstrate that ingested maternal breast milk can be absorbed by the bones during postmortem decomposition and detected using proteomic analysis.

This study shows that the breastfeeding status of a neonate can be reconstructed by applying proteomic analysis to its archaeological remains. Proteomic estimation of breastfeeding status has three advantages compared to isotopic and observational analyses: (i) it can be applied to dead individuals whose behavior cannot be observed, (ii) it offers direct evidence of breast milk consumption, and (iii) it can provide species-specific identification of the milk source. This method is not only applicable to archaeological samples but has the potential to be used in forensic and wildlife studies.

Proteomics-based estimation of breastfeeding status is associated with some limitations as well. First, milk proteins might not be detected from the remains of biologically older individuals who, despite dying while they were breastfed, consumed proportionally smaller amounts of milk in relation to their body mass. For example, in a previous proteomic study, no milk protein was detected in a rib bone from an archaeological human skeleton of a nine months old individual28, despite the cessation of breastfeeding was reconstructed to occur most probably at the age of 3.1 years in that population51. Second, this method is harder to apply to remains from mammalian species whose genome and/or reference proteome is not publicly available yet.

Conclusions

Dog milk (LGB1 and WAP) and fetal/infant marker (AFP, COL10A1, and EPYC) proteins were detected in rib and vertebra bones of an ancient (430–960 cal AD) dog neonate (2017HA1016) that died 2 weeks after birth. The dog milk proteins most probably originated from the mother’s breast milk. This is the first study that reports the survival of ingested breast milk proteins in an ancient mammalian skeleton.