Protein aggregate formation permits millennium-old brain preservation

1. Introduction

Understanding the mechanisms of preservation of proteins is relevant at the interface between biomedical applications [1], protein biomarker research [2] and medicine [3].

In free nature, protein preservation is a conundrum because spontaneous decomposition is a feature of all biological macromolecules caused by simple chemical processes [4,5]. Therefore, the yellowish-brown mass seen through the foramen magnum of an Iron Age human skull (cranium OxA-20677 collagen radiocarbon dating 673-482 BC [6]) from archaeological excavations in Heslington, York, UK, offers a unique opportunity to use molecular tools [7] to investigate the preservation of human brain proteins.

The preservation of this human brain tissue remains a mystery as decomposition and autolysis starts within minutes after death [8,9]. Compared to other body parts such as bones [10], autolysis is particularly rapid in the brain, which consists of 80% water [9,11]. Biochemically, autolysis is caused by the massive activation of proteases and phospholipases, destroying the molecular structure of lipids and proteins [8,12]. Within 36–72 h, putrefaction starts and complete skeletonization results within 5–10 years [12,13]. In conclusion, the preservation of human brain proteins at ambient temperature should not be possible for millennia in free nature [4,5,7,8,11–13]. For this reason, taxonomic studies of the brain of human evolutionary descent relied until recently mainly on the discussion of skull fragments and teeth [10,11]. New data on the stability and binding affinities of proteins have, however, advanced the development of evolutionary models [14]. These mainly sequence-based data [14] are now mainly enriched by protein biophysics data [15]. The thermodynamic stability of a protein is related to its structure [16,17]. It is more challenging to unfold a very stable protein compared to an unstable protein [15]. Protein stability provides indirect evidence for protein function and is driven by mutations influencing the protein structure [14,18].

The human brain is in particular need of structural stability because the key components, neurons and axons do not normally regenerate. The cellular scaffold of neurons and their supporting cells, astrocytes, consists of intermediate filaments (IFs). These IFs represent a group of protein polymers which are classified by a fibre diameter (8–12 nm) that is intermediate between the smaller microfilaments (7 nm) and the larger microtubules (≈25 nm) [19]. The IFs of neurons/axons are neurofilaments (Nfs) [20,21] and the IFs of astrocytes are glial fibrillary acidic protein (GFAP) [22,23]. This cellular specificity has been of advantage for the development of GFAP and Nf proteins as body fluid biomarkers for disease, among many other proteins [2,21,24,25]. In addition, IFs have a number of unusual protein features because they are intrinsically unstructured, have large polyanion tails, multiple phosphorylation sites, are able to self-assemble into polymers and form intra- and extracellular aggregates in pathology [26–33]. Particularly, the ability to form protein aggregate is thought to be relevant for the progression of neurodegeneration [26,34].

Taken together the data presented in this study on protein stability from the unique find of a preserved prehistoric human brain [6] is of mutual benefit to the fields of protein biomarker research [2,20,24,25,35], medicine [3,26,34], structural and functional proteomics [14,18], biomedical applications [1] and archaeology [6,10,11].

2. Results

The data of this translational study are presented by moving from archaeology to biochemistry and from macro-structure over micro-structure to proteomics.

2.1. Preservation of the macro-structure of a 2600-year-old human brain

Mummification permits long-term preservation of soft tissues. Natural mummification was responsible for the preservation of the brain of the iceman, who was preserved in a glacier [36]. In contrast to the iceman and most other previously reported finds of preserved human brains, there was no sign of hair, skin or any other soft tissue associated with the ancient brain subject to the present study (figure 1a) [6]. A yellowish-brown mass (figure 1b) was seen through the foramen magnum of a human skull in Heslington, York, UK excavated by the York Archaeological Trust in August 2008. The identification of this putative brain tissue was strengthened following craniectomy (figure 1c). Although covered by sediment (figure 1d), individual brain gyri became discernible after cleaning (figure 1e). Collagen from the bone was radiocarbon dated (OxA-20677) to 673–482 BC [6]. There was no evidence for tannins or artificial preservation techniques [6].

• Download figure
• Open in new tab
• Download powerPoint

The preservation of the ancient brain tissue remains enigmatic because of rapid decomposition and autolysis after death [8,9].

2.2. Ancient brain histology

Consistent with the macro-structure (figure 1e), the micro-structure of the ancient brain resembled remnants of brain tissue axons (figure 2a). The electron microscopy (EM) images showed 5–10 μm long and 0.2–0.6 μm thick filamentous, electrodense structures (figure 2a). In axial sections, the diameter of these axonal structures ranged from 0.2 to 4 μm. These ancient axons were more densely packed than axons from a contemporary human brain sample (figure 2b). Contemporary axons were longer (up to 17.3 μm) and of a larger diameter (0.8–7.5 μm). These data suggest diagenetic alterations.

• Download figure
• Open in new tab
• Download powerPoint

Next, a monoclonal antibody was used to test the hypothesis that the exceptional preservation of axonal structures did contain highly specific axonal proteins. Nf proteins represent a family of differently sized isoforms expressed in neuronal tissue [19,20]. Specific immuno-EM to the Nf heavy chain (NfH) showed strong binding to the filamentous structures present in the suspected shrunken axons (figure 2c). At higher magnification, the gold-labelled anti-NfH antibodies lined up exclusively alongside the compact filamentous structures elongated inside the axons (figure 2d). The exclusive binding of the gold particles to the axonal structures confirms the morphological EM observations.

Further verification that the preserved filamentous axonal structures contain ancient, but intact protein epitopes can be obtained by the reversal of the experimental set-up. No detailed studies have yet been conducted on the suspected preservation of immunogenic ancient protein epitopes. We, therefore, proceeded with the generation of polyclonal antiserum against the Heslington brain.

2.3. Antibody generation

To screen for the potential preservation of antigenic epitopes from other human brain proteins, antibodies against the ancient brain were generated in mice. Both white matter (WM) and grey matter (GM) of the ancient brain tissue produced a robust immune response (titres 1:2000) to GFAP and myelin basic protein (MBP), as shown by staining against E21 rat cortical neuron cultures (figure 2e–j). Elements of the mouse serum staining pattern resembled oligodendrocyte (figure 2e) and astrocytic cells (figure 2h). Double label with antibodies to MBP (figure 2f) showed the exact overlap in cells with an oligodendrocyte morphology, while cells with an astrocytic morphology (figure 2h) showed perfect overlap with GFAP antibody (figure 2i,j). Western blotting on pure MBP and pure GFAP showed that mouse serum stained pure MBP (figure 2k, lane 1) as well as a bona fide MBP antibody (lane 2) and that it stained pure GFAP (lane 3) as well as a bona fide GFAP antibody (lane 4). For Nf, a weaker immune response was found which did not mature, and bands were less distinct on the immunoblot. There was no hint of this pattern of staining from the mice blood samples taken prior to injection (1:100). In conclusion, these experiments demonstrated strong immunogenicity for ancient GFAP and MBP, and to a lesser extent for Nf.

2.4. Ancient brain homogenate contains high molecular weight protein aggregates

To further investigate these protein candidates, GM and WM samples were taken from the ancient and a modern human control brain. The median wet weight of the ancient brain samples (0.076 g, n = 10) was comparable to the control tissue samples (0.104 g, n = 10, p = 0.13). Samples were homogenized, centrifuged and the supernatant taken for the analysis of the soluble protein fraction. Much of the ancient brain material remained insoluble being visible as small brown particles. The total protein concentration of the ancient brain was significantly lower (median 0.359 g l⁻¹) compared to the control brain tissue (1.302 g l⁻¹, p < 0.01). There was good correlation of the wet weight with the total protein concentration (R = 0.72, p = 0.018, figure 3a) indicating that the proportion of soluble protein from all samples was comparable despite the remaining insoluble fraction. Gel electrophoresis of the ancient brain showed multiple agrophil bands (figure 3b). The band pattern was consistent between samples and increased in density with higher protein load. Immuno-blotting showed faint binding for hyperphosphorylated Nf (NfH^SMI34) in the higher molecular range (420 kDa, figure 3c). This was about two times the expected gel migration around 220 kDa [20,31,33,37]. It seemed, therefore, likely that protein aggregate formation had occurred [38]. The aggregates from the ancient brain were larger and considerably more resistant to incubation with urea as those from recent human or animal brains [15,20,31,37–40].

• Download figure
• Open in new tab
• Download powerPoint

2.5. IF iso- and phosphoforms discriminate brain structures and show pattern reversal between ancient and modern control tissues

Taken together, the qualitative histological and immunological data suggested the presence of quantifiable amounts of soluble Nf, GFAP and MBP in the ancient brain. We hypothesized that if the preservation occurred quickly enough, it should still be possible to discern GM from WM samples. These proteins are present in a higher concentration in the WM compared to the GM [39,41,42]. Surprisingly, the GM and WM protein concentrations from the ancient brain were inverse to the control brain (figure 3d–h) and the data from the literature [39,41,42]. Consistent with the well-known higher susceptibility of non-phosphorylated NfH to proteases compared to the phosphorylated NfH [43,44], absolute levels of the non-phosphorylated NfH^SMI38 were lower in the control brain compared to the phosphorylated NfH^SMI34 and NfH^SMI35. Additionally, the Nf light chain which is known to be less stable than the phosphorylated heavy chain [43–45] was present at higher absolute concentrations in the ancient brain than NfH (figure 3g). Combined, the data suggest that the proteases of the ancient brain might have been inhibited by an unknown compound which had diffused from the outside of the brain to the deeper structures.

Chemically, preservation might have been possible by an acidic compound similar to what is known from the soft tissue preservation of the bog bodies with a pH about 3.5 from the beautifully preserved Danish examples [46]. The tissue pH of the ancient brain was 4.6, but we would be hesitant to use this as a proxy for the burial environment. The pH of a burial environment can vary spatially over very small distances and changes with time due to the deterioration of material in the deposit, changes in aeration, water saturation and water movement or land use. Furthermore, bio-apatite, which is depleted in an acidic environment, was present, while collagen was unexpectedly poorly preserved [46]. Finally, there was no evidence for tannins in either the brain tissue or the immediate burial environment.

2.6. Modern brain tissue shows short-term protein decay

Still, any speculative compound will only have been able to a preserve tissue during a limited, unknown time frame after death. Therefore, a 1-year stability experiment was performed. The summary data are shown in figure 4. The hypothesized outcome was loss of protein from both the modern control and the ancient brain tissue samples (dark blue-shaded areas in figure 4a–e). Indeed, for the modern brain, there was loss of protein as expected by decay [5]. The standard deviation for separate measurements was most narrow for GFAP (figure 4e, triangles). Compared to GFAP, the protein concentrations decreased more rapidly for non-phosphorylated NfH (figure 4a,c, dots) and phosphorylated NfH (figure 4b,d, dots). From these data, it can be deduced that to preserve GFAP and NfH phosphoforms after death, a speculative preservative compound should have entered the ancient brain within about three months after death.

• Download figure
• Open in new tab
• Download powerPoint

2.7. Ancient brain tissue shows long-term protein release from aggregates

Unexpectedly, the stability experiment revealed yet another unusual finding for the ancient brain proteins (figure 4a–e, dots). There was a consistent increase in the protein concentration for GFAP and NfH phosphoforms for all measurements from the ancient brain (light blue-shaded areas in figure 4a–e, dots). Notably, this observation would have been missed with the routine time frame used for biomarker stability experiments in the field [1,45]. The large data scatter during the first three weeks might have been readily dismissed as the measurement noise (dots in figure 4a–e). The trend for an increase in the concentration of the soluble IFs only became clear after two months of observation. The steep increase of the curves after 1 year implies that this trend will continue for a yet unknown amount of time. In contrast to GFAP and NfH phosphoforms, there was no systematic trend over time for MBP (see electronic supplementary material, figure). Of the several possible interpretations of the data, we favour that antibody-binding epitopes of intrinsically unstructured regions on IF iso- and phosphoforms [27] were partly exposed and masked at subsequent intermediate stages of a complex un-/folding process from densely packed protein aggregates [47,48].

2.8. The stability pattern of IF (GFAP > phospho-NfH > non-phospho-NfH) from control tissue was lost from the ancient brain tissue

Long-term biomarker stability was best for GFAP (green-shaded area, figure 4f). This is illustrated by accounting for the different longitudinal profiles of all IFs over time in the form of a ratio of modern to ancient proteins. For NfH, there was an overlap of the 95% confidence curves of the non-parametric regression analyses for all phosphoforms (figure 4f). The individual regression lines suggested phosphorylated NfH (red and blue lines) was more stable compared to non-phosphorylated NfH (magenta and grey lines). This interpretation was, however, only true for the modern brain if the most stable IF, GFAP, was taken as a ‘housekeeping protein’ (figure 4g). Non-phosphorylated NfH (yellow shaded area) degraded quicker compared to phosphorylated NfH in the modern brain tissue as expected from the literature [43,44]. This pattern was completely lost for ancient NfH phosphoforms.

2.9. Mass spectrometry

Samples were analysed by SDS–PAGE followed by in-gel tryptic digestion and LC-MS on a high-resolution Orbitrap mass spectrometer operated in the data-dependent mode. Proteins were identified by automatic de novo peptide sequencing and database searching [49]. As extensive protein degradation can be expected, the database search considered semi-tryptic peptides, i.e. peptides with tryptic cleavage at one end and non-tryptic at the other. N-terminal acetylation is a frequently observed modification in proteomic analysis, as result of post-translational modification but can also be introduced during sample preparation, and was, therefore, included as variable PTM in the database search. The identification results were validated using the target-decoy approach, and only peptide matches above the score threshold corresponding to <1% false-discovery rate (FDR) were used for protein identification. Furthermore, for a protein to be considered identified required at least one unique peptide, i.e. a peptide not present in any other protein in the database. In total, 881 proteins were identified; 671 in cortex, 759 in WM and 531 in both tissues (electronic supplementary material, table S1). To distinguish ancient human proteins in the samples from such contaminants, a blank gel sample was excised from empty lanes and analysed. In total, 855 proteins were identified in the Helsington brain samples and 114 proteins in the gel blank, of which 37 were keratins, abundant in skin and hair. After subtracting proteins present in the blank, 783 proteins remained (electronic supplementary material, table S1). While some of the proteins were keratins that could result from ancient or recent sample contamination, the majority of our proteins are expected to be found in the brain. GFAP (seven peptides) and NfL were among the identified proteins (three peptides) and MBP, (three peptides), confirming the presence of these proteins as detected by the enzyme-linked immunosorbent assay (ELISA) (figure 5). For GFAP (figure 5a), the amino acid sequences of six of the matched tryptic peptides were identical to the sequences found in several forms of keratin type II cytoskeletal, rendering their assignment to GFAP ambiguous; however, one peptide (DQLTANSAR, aa 128–136) identified with a score above the chosen 1% FDR threshold was unique to GFAP, thus providing strong evidence for the presence of GFAP in the sample. The annotated fragment ion spectrum for this peptide is shown in figure 5c. For NfL, there was also a single unique peptide (YEEEVLSR, aa 178–185) identified with a score above the significance threshold.

• Download figure
• Open in new tab
• Download powerPoint

2.10. Neurological disease analysis

There are a number of well-known mutations to brain proteins which can promote protein aggregate formation and which are related to human disease [27,32,50,51]. Some of these change human behaviour profoundly [52,53]. Other neurological conditions, such as kuru or Creutzfeldt–Jakob disease, are transmissible [54–56]. There was, however, no evidence for pathological human prion proteins with highly sensitive immuno tests [57]. Likewise, DNA extraction only revealed small fragments. Analyses of these DNA fragments with the Illumina kit for formalin-fixed, paraffin-embedded sample DNA did not provide useful sequence data (not shown).

3. Discussion

Taken together with the excellent preservation of intracranial soft tissue from the Heslington find [6] enabled us to perform multiple independent molecular analyses to demonstrate the exceptional preservation of ancient human brain proteins. Our approach to using a range of complementary analytical techniques [2,20,24,25,35] shows not only that protein aggregate formation contributed to protein stability, but also that protein epitopes remain highly immunogenic after 2600 years of exposure to ambient temperature in nature.

The archaeological and macroscopic data reveal the only surviving human tissue from this find to be the brain tissue. The macroscopic findings were corroborated on a microscopic level by histology and immuno-histochemistry. The filamentous structure resembled axons packed with Nf proteins [2,19,37]. The Heslington brain is not unique and other brains have been found in buried skeletons elsewhere (see references in [6]). But no other brains are known from this period. Only if one extends from natural mummification to deliberately preserved bodies, there is one example of preservation of the suspected brain tissue (and other soft tissue) dating around the same period. The ‘Boscastle skull’ had recently been dated to 361–112 BC (95.4% confidence interval OxCsal v4.17), the Ptolemaic period [58]. In the same paper, Smith et al. [58] show CT brain images of two Egyptian mummies, the ‘Hetep-Bastet’ and ‘Lady Hudson’. But, these were deliberately preserved brains and not brain tissue surviving because of natural mummification. The preservation of macroscopic structures and imaging studies of the Heslington brain enabled targeted sampling from superficial and deep brain structures for the series of tests.

First, by capitalizing on the specificity and sensitivity of the immune response, which has been used successfully to characterize brain proteins [59–61], we show that antibodies raised against ancient brain proteins react specifically with contemporary brain proteins in situ experiments. The binding characteristics are indistinguishable from antiserum raised against the contemporary brain tissues [59]. Double labelling demonstrated binding to the brain cells of glial lineage, oligodendrocytes (MBP) and astrocytes (GFAP). The signal for neurons (Nf) was weaker than what can be observed using contemporary tissue [60]. The overall extent of epitope preservation over this amount of time will be of interest to biomedical applications aimed at improving long-term protein storage [1].

Second, highly sensitive immunoassays permitted for the quantification of Nf protein isoforms and GFAP. The pattern distribution of these proteins was inverted to what can be seen in a contemporary human control brain. The topographic distribution of Nf isoform concentrations has been related to axonal density in tissue as the main source for Nf proteins [39,41,42]. Outwards-in staged inhibition of proteases as the main contributor to cellular autolysis [8,12] provides one of several possible explanations for this topographic finding. An alternative could be the pathological protein aggregate formation during life [26–33]. The presence of pathological prion proteins and their aggregates would have corroborated genetic data on prehistoric kuru-like epidemics [62]. There was no evidence for pathological prion proteins in the ancient brain tissue using state-of-the art methods [57]. The quality of the DNA was too low to permit for screening of other human diseases of pathological protein aggregation or neurodegeneration [26–33]. These data are consistent with the notion that DNA degrades about 10 times faster than proteins [63,64].

Third, proteomic analysis by LC-MS of samples digested with trypsin identified over 800 proteins (electronic supplementary material). These data confirmed the imaging and immunological data. Notably, there was excellent preservation of GFAP. For the larger Nf, LC-MS analysis was less successful. Most likely, this is explained by the challenge to get the large Nf aggregates shown in the immunoblots to fly in our set-up. This is a recognized limitation of the method [65].

In the proteomic analysis of ancient samples, it is important to consider the risk of adventitious protein identifications stemming from sample contamination that may occur prior to or during excavation, during curation and handling in museums or in the laboratory in conjunction with the analysis [66,67]. The contamination may come from the laboratory environment or reagents and may occur even if high-purity reagents are used and care is taken to clean all equipment that comes into contact with the samples. Also, proteins from bacteria that colonize the specimen may lead to peptide identifications that are mistakenly matched to human proteins. To counteract the risk of such adventitious protein identifications in our data, a contaminant sequence database was compiled from all proteins identified in blank gel and LC runs, the common Repository of Adventitious Proteins (cRAP; a public database containing known laboratory contaminant proteins; https://www.thegpm.org/crap/) and the microbial subset of the UniProt–SwissProt sequence database. The contaminant database was used to filter the identification results within the PEAKS software, such that peptide matches in the contaminant database do not contribute to protein identifications. For GFAP, its high sequence homology to several keratin forms rendered identification difficult; only a single unique peptide was identified. This is likely to be an issue not only in this, but also in other, future studies investigating IFs.

Some of the many other proteins recognized are likely to be contaminants from other tissues, notably the many keratins typically found in the skin tissue which were absent from the Heslington find. By contrast, the large number of brain-derived proteins were found in the analysis of fresh brain tissues. Because of the large amount of preserved sample available, this will enable future studies into the structure and function of these ancient proteins.

Fourth, the combined interpretation of the presence of protein aggregates in the immunoblots and the slow release of GFAP, Nf iso- and phosphoforms from these aggregates in solution over a 12-month-period is important. Protein aggregate formation does provide a formidable strategy to minimize surface protein exposure and thereby maximize long-term stability [18,68,69]. Yet, amyloid formation is avoided, epitopes are preserved and immunogenicity is maintained. This suggests that on a funnel plot energy landscape refolding from an extremely low energy state remains a distinct possibility for biomedical applications [1].

Likewise, protein aggregate formation seems a more plausible explanation for the preservation of beta-keratin in fossil feathers [70] rather than a simple mechanical compression. IFs share some structural properties with collagens which have been shown to be among the most resistant proteins, permitting detection from fossilized bones of 68 Myr Tyrannosaurus rex. Collagens were shown to undergo a range of post-translational modifications in mummified tissues [71,72]. Of these, the involvement of lysine is dominant leading to a series of rearrangements, dehydration and fragmentation reactions which end up as complex, cross-linked structures [65,71]. Expression of lysine is high in Nf proteins and GFAP [20,73]. It seems possible that processes similar to what has been observed for collagen, contributed to the preservation of the ancient IF proteins. While the body of emerging evidence in favour of the survival of proteins into deep time is seductive, the mechanisms involved are poorly understood [74].

Human species differential expression of collagen composition recently enabled Neanderthal proteins to be distinguished from Homo sapiens proteins [75]. This has not yet been achieved for the most exciting of all human organs, the brain. The present data open the possibility for future studies to investigate if, between human species, there is evidence for preservation or micro-heterogeneity of brain proteins. Such studies may not have to be restricted to preserved brain tissue from human skulls, but venture to an investigation of protein sequences bound to mineral surfaces [74].

In this context, a promising feature of Nf proteins is the high binding affinity of serine to silver [76,77]. Silver-containing minerals are not overly abundant, which increases the value to humans. But precisely, this may enhance the chance to discover silver-containing material in human settlements [78]. The mechanisms proposed to enhance ancient protein survival is lowering their configuration energy [74]. Another advantage of Nf and GFAP is that they belong to the group of intrinsically disorganized proteins [79]. These proteins can adapt many folding states and have a wide spectrum of binding abilities [80].

In summary, our data provided multiple lines of experimental evidence for the long-term preservation of human brain proteins. The data suggest that protein aggregate formation is one strategy for protein preservation.

4. Material and methods

4.1. Sample acquisition and preparation

Brain samples were processed as described [39]. In brief, 5 mm³ cubes of brain samples were cut from the surface and depth were submerged in 2 ml of barbitone-EDTA buffer (pH 8.6) and snap-frozen in liquid nitrogen and stored at − 80°C. For control, brain tissue from a non-neurological control patient was taken.

Samples were defrosted and 20 μl of a protease inhibitor cocktail (Sigma, P8340) was added (=1:100). Samples were sonicated on ice. Samples were then diluted 1:1 (=2 ml added) with barbitone-EDTA buffer containing 16 M urea and 4% CHAPS and mixed at 4°C for 24 h, centrifuged at 1000g for 30 min at 4°C. The supernatant was stored in aliquots at − 80°C.

4.2. Antibody generation

Antibodies were generated using standard procedures [60]. In brief, two different samples of 30 mg brain tissue, one derived from GM and the other from WM were each dissolved in 200 μl of 6 M urea, vortexed, sonicated and put on an orbital shaker overnight at 4°C. Each preparation was then mixed with an equal volume of Freund’s complete adjuvant and 100 μl of the mix was injected into female Balb/c mice. These mice were three months of age and blood samples were taken by tail bleeding prior to injection. Three weeks later, each mouse was injected with 80 μl of the respective mix and 10 days later about 100 μl of blood was taken from the tail of each mouse. Each mouse was boosted with 80 μl of the respective mix 25 days later and another blood sample was taken 9 days later. The mice were boosted a final time with 80 μl of the respective immunogen 31 days later and a final blood sample was taken 10 days later.

4.3. Long-term stability experiment

Twenty pairs of 250 μl aliquots of the samples from the Heslington brain and modern control brain (1:1 diluted in BarbEDTA buffer containing 0.001% azide and 0.2% bovine serum albumin) were stored light protected at room temperature for up to 1 year. One pair of aliquots was transferred to − 80°C at each of the following time points (1 day, 3 days, one week, two weeks, three weeks, four weeks (=one month), two months and so for up to 12 months (=1 year)). Samples were then batch analysed by ELISA in a randomized order with the analyst being blinded to randomization.

4.4. Brain-specific protein analysis

Brain-specific proteins were quantified using in-house developed ELISAs [60,81–83]. Total protein was determined using the Bio-Rad Protein assay (Bio-Rad, Hemel Hempstead, UK).

4.5. Gel electrophoresis and immunoblot

Gel electrophoresis and immunoblot were performed as described [39,57,60].

4.6. Mass spectrometry

Protein extracts from the cortex and WM were analysed by SDS–PAGE followed by LC-MS. Five microlitres of 4 × LDS buffer (5 μl) were added to protein extracts (15 μl) and the samples were incubated at 70°C for 15 min and analysed by SDS–PAGE. To preclude contamination of the samples with proteins from the laboratory environment, the gel chamber was cleaned prior to use, and pre-cast gels (NuPAGE, Thermo Fischer) were used that were not exposed to any other samples. Following Coomassie blue staining, the gel lanes were sliced in eight equal-sized pieces, which were subjected to the reduction and alkylation of cysteine-disulfides and tryptic digestion. The samples were reconstituted in 6 μl 2% acetonitrile and 0.1% TFA. Aliquots of 5 μl were loaded on a nanoflow-LC (RSLC nano, Thermo Scientific) equipped with a C18 trap column (PepMap Acclaim 75 μm × 20 mm, Thermo Scientific) and a C18 separation column (PepMap Acclaim 75 μm × 500 mm, Thermo Scientific), coupled to a Q-Exactive electrospray ionization mass spectrometer (Thermo Scientific), fitted with a FlexiSpray ion source. Prior to analysing the samples, several blank LC injections (loading buffer) were performed to ensure the absence of contaminating proteins from previous analyses. The loading buffer was 2% acetonitrile, 0.05% TFA; Buffer A was 0.1% formic acid; and Buffer B was 84% ACN, 0.1% formic acid. The following gradient was used: t = 0 min, $B=3\text{\%}$; 140 min, $B=30\text{\%}$; 160 min, $B=45\text{\%}$; 165 min, $B=80\text{\%}$. The mass spectrometer was operated in the positive ion mode. The data-dependent acquisition was used, acquiring one full MS scan (R 140 k, AGC target 3e6, max IT 250 ms, scan range 400–1600 m/z) and up to 10 consecutive HCD MS/MS scans (R = 70 k, AGC target = 1e6, max IT = 250 ms, isolation window 1.2 m/z, NCE 32.0, charge exclusion: unassigned, >6). Samples were analysed in duplicate. Protein identification was performed using the software PEAKS Studio X, which is based on automatic de novo peptide sequencing. De novo settings were: Parent Mass Error Tolerance: 15 ppm; Fragment Mass Error Tolerance: 0.05 Da; Enzyme: None; Fixed Modifications: Carbamidomethylation; Variable Modifications: Acetylation (N-term); Max Variable PTM Per Peptide: 1; Report # Peptides: 5. The PEAKS database search settings were: Parent Mass Error Tolerance: 15 ppm; Fragment Mass Error Tolerance: 0.05 Da; Enzyme: Trypsin; Max Missed Cleavages: 3; Non-specific Cleavage: one; Fixed Modification: Carbamidomethylation; Variable Modifications: Acetylation (N-term); Max Variable PTM Per Peptide: 1; Database: UniProt SProt; Taxon: Homo sapiens; Contaminant database: compilation of all sequences from UniProt Bacteria, cRAP, and proteins identified in blank SDS–PAGE samples and blank LC-MS runs. Validation of peptide identifications was performed using the target-decoy approach. Proteins are reported that had at least one unique peptide hit with a score corresponding to <1% FDR. The uniqueness of identified peptides was evaluated using the neXtProt peptide uniqueness checker [84].

4.7. Statistical analysis

Statistical analyses were performed using SAS software (V9.4). Independent variables were compared using the non-parametric two-sample exact Wilcoxon rank-sum test for two variables and a two-way unbalanced ANOVA (general linear model) for more than two variables. The linear relationship between continuous variables was evaluated using the Spearman correlation coefficient. Non-parametric regression analyses were performed using LOESS and 95% confidence curves were calculated. The level of significance for the multiple correlations was corrected using the Bonferroni method. Two-tailed tests were used throughout and p-values of <0.05 were accepted as significant.

4.8. Exclusion of sample contamination

Methods for the exclusion of sample contamination were applied on several levels. First, samples were independently analysed in three laboratories: London, Gothenburg and Gainsville. Next, a range of complementary methods was used in these laboratories which taken together provide several independent lines of evidence for the preservation of brain proteins. The London and Gothenburg laboratories are both accredited to ISO 15189 quality standard.

Most importantly, we excluded all peptides and proteins reported in the literature as problematic regarding sample contamination [66,67]. Finally, we have made our raw data available for independent analyses through the PRIDE repository (https://www.ebi.ac.uk/pride/archive/). Data are available via ProteomeXchange with identifier PXD014178.

For the proteomic analyses, there are additional sources of protein contamination to consider. One being carryover from previous samples analysed using the same laboratory equipment. To minimize this risk, we carefully cleaned the gel electrophoresis equipment prior to analysis. In addition, we have used new HPLC columns for these experiments. This eliminates the possibility for carryover from columns which have previously exposed to unknown proteins. Finally, prior to analysing the brain samples by LC-MS, blank samples were analysed, including both gel blanks and LC blanks, to ensure the absence of contaminating proteins.

Ethics

Ethical approval for this study was given (centre number 875KLH, study number 03/N101, UK).

Data accessibility

Data will be shared upon reasonable request and are also available through the PRIDE repository (https://www.ebi.ac.uk/pride/archive/), identifier PXD014178.

Authors' contributions

A.P. conceived of the study, designed and performed experiments and wrote the manuscript. C.-H.L., G.S., J.G. and M.G. contributed to the laboratory work. G.S. produced the antibodies against the ancient brain proteins. All co-authors revised the manuscript.

Competing interests

G.S. is a founder of EnCor Biotechnology Inc. in Gainsville, USA. H.Z. is a co-founder of Brain Biomarker Solutions in Gothenburg AB, a GU Ventures-based platform company at the University of Gothenburg. H.Z. has served on advisory boards for Roche Diagnostics, Eli Lilly, and Pharmasum Therapeutics.

Funding

The Heslington brain project which was co-funded by the University of York and English Heritage. S.O. is the principal investigator of the project for the York Archaeological Trust.

Footnotes

Electronic supplementary material is available online at http://dx.doi.org/10.6084/m9.figshare.c.4782597.