Pilot proteomics study of three different human femur sites and autologous serum samples

Abstract

Tissue proteomics will help discover the molecular pathophysiology of bone diseases. So far, most human studies have been performed on serum/plasma samples. We present an initial pilot proteomics analysis comparing three clinically important anatomical femur sites (caput, collum, and trochanter). The specimens were obtained from osteoarthritic patients undergoing hip replacement surgery and represented predominantly cortical bone from collum and mostly trabecular bone from caput, together with a mixture from trochanter. The bone and the corresponding serum samples were analyzed by high-throughput data-independent acquisition (DIA) liquid chromatography-mass spectrometry (LC-MS/MS). The results showed that the protein composition of bone/bone marrow was substantially more complex than serum, presenting a total of 3085 and 597 detected proteins, respectively. A major part of the detected osteoblast/osteocyte, adipocyte, mesenchymal stem cell (MSC), and hematological/immunological characteristic proteins were similar between the corresponding femur sites in all donors and different from serum. Although the majority of the bone proteome was conserved across anatomical sites, exploratory analyses of the global proteomic landscape revealed site-specific protein signatures and associated functional pathways that were highly consistent across individuals. These findings suggest that anatomically adjacent femoral sites share a conserved core bone proteome while maintaining distinct molecular characteristics of potential functional relevance. With validation in larger cohorts, the results will contribute to the development of a human bone tissue proteomics atlas providing novel insight in the molecular microarchitecture and functionality relevant for bone health.

Highlights

• •Each anatomical site of the femur showed trends toward distinct protein signatures and associated functional pathways.
• •Differences between femoral site proteomes appeared greater than the variation observed between serum samples.
• •Protein profiles were highly homogeneous within sites across individuals, exceeding similarity across sites.

1. Introduction

Bone is an essential organ system providing structural support for ambulation, protection of inner organs, and serving as a reservoir for hormones and mineral metabolites. Accelerated bone loss in older adults may lead to primary osteoporosis, a worldwide common disorder, affecting more than 200 million people (Sözen et al., 2017). The accumulated disability and increased risk for mortality, especially represented by femur fracture complications, is a global burden on public health (Cauley, 2013). In addition, most chronic diseases of the kidney, gut, and malnutrition also affect the skeleton, causing bone loss and secondary osteoporosis as do many important drugs used to combat non-skeletal diseases (Panday et al., 2014). Therefore, it is vital to improve our understanding of the molecular mechanisms of bone disease.

Bone tissue consists of matrix and several different cell types, such as osteoblast, osteocyte, and osteoclast (Salgado et al., 2004). The bone matrix has two components: a mineral part constituted by calcium-hydroxyapatite which contributes 65–70% of the bone matrix and the organic part which comprises the remaining 30–35% (Feng, 2009). The major organic components are collagen and proteoglycans being essential building components of bone. Other important non-collagen proteins are growth factors, cytokines, and bone-specific functional proteins such as osteocalcin, osteonectin, and osteopontin (Sommerfeldt and Rubin, 2001). These proteins are crucial mediators of all cellular bone-dependent processes. In addition, bone contains marrow consisting of mesenchymal stem cells (MSC), adipocytes, and cells within the hematological or immunological cell lineages, all of crucial importance for bone metabolism. A comprehensive description of the total skeletal proteome will improve greatly the knowledge base to understand the functional aspects of bone in health and disease.

Long bones are organized into external cortical and inner trabecular structures at both ends. Together, they give bones their mechanical strength, but while the cortex is dense, trabecular bone is porous, and serves many other functions. The molecular compositions of cortical and trabecular bone at different skeletal sites reflect their distinct functional characteristics that are relevant for disease risk and progression. Trabecular bone shows higher rates of remodeling and is actively involved in mineral metabolism (Parfitt, 2002). The initial bone loss in osteoporosis occurs primarily in the trabecular vertebrae. During disease progression, cortical resorption takes place causing porosity, thinner trabeculae, and increased endocortical surface (Osterhoff et al., 2016). How these changes are reflected within the proteome and especially at different functional anatomical sites, are unknown and is a prerequisite to better understand the mechanism of bone diseases.

Bone-related proteomics studies have been performed in several types of human tissues such as serum or plasma (Chen et al., 2020; Huang et al., 2020), adipose tissue, and bone marrow (Dadras et al., 2020), in primary cells isolated from blood, such as peripheral blood monocytes (Zeng et al., 2016), and in cultures of stem cell-derived bone cells such as osteoclasts (Kubota et al., 2003) and osteoblasts (Salasznyk et al., 2005). Direct proteomic characterization of clinically obtained human bone tissue remains limited, with previous studies generally restricted to small numbers of specimens from a single skeletal site. For example, Alves et al. analyzed four cancellous bone samples taken from the proximal femur of four patients undergoing total hip replacement using high-resolution mass spectrometry-based methods (Alves et al., 2011). In addition, previous LC-MS/MS studies have investigated intra- and inter-skeletal proteome variation, primarily in animal models and forensic contexts (Mickleburgh et al., 2021; Procopio et al., 2017). While these studies provided important insights into skeletal-site-specific proteomic differences, they did not examine human bone tissue collected from living patients or compare bone proteomes with autologous serum samples. Therefore, our objective of the pilot study was to characterize the proteomic profiles across three anatomically distinct regions of the proximal femur and compare these profiles with the matched serum sample from the same individuals. Despite the critical role of bone proteins in skeletal biology, we have scarce knowledge regarding bone protein composition, especially at different functional skeletal sites, and how the molecular networks are orchestrated in health and diseases.

We present an initial, but detailed study comparing the proteomes at three different femur sites (caput, collum, and trochanter), and the corresponding serum profiles from three subjects who were clinically well-characterized and undergoing hip replacement surgery. We used liquid chromatography with tandem mass spectrometry (LC-MS/MS) technique to identify and quantify the proteome with high sensitivity. By performing the bioinformatics analysis on corresponding bone biopsy/serum proteomics data, we describe the protein landscape of human femur bone tissue and identify tentatively the signature proteins at each skeletal site.

2. Methods

2.1. Subjects and samples

The femoral bone samples were obtained from three osteoarthritic patients undergoing hip replacement surgery (Table 1). The patients were recruited consecutively at Lovisenberg Diaconal Hospital, Oslo, Norway, as described earlier (Gautvik et al., 2020). Their serum samples were collected in the morning before food intake. The donors had no secondary diseases and did not take medication known to affect bone metabolism. Anthropometric data, including bone mineral density, and biochemical parameters for each of the donors are shown in Table 1. Bone specimens were taken from three distinct sites as illustrated (Fig. 1) and immediately frozen on liquid nitrogen. The trochanteric samples constituted mostly trabecular bone and marrow and represented the intertrochanteric part that was removed to accommodate the femoral prosthesis. The collum sample, containing predominantly cortical bone with limited marrow admixture, was sawed off approximately 1 cm distal to the typical femoral fracture site (see Fig. 1). Caput bone samples, represented mainly trabecular bone, were carved out with a sharp spoon after removal. However, no quantitative histological or imaging-based assessment of cortical/trabecular composition or marrow content was performed. All bone samples were further cut into representative pieces and pulverized in a mortar cooled with liquid nitrogen. One hundred milligrams of each sample were weighed and transferred to a microcentrifuge tube, then stored at −80 °C until further processing. Based on the preparation process and the tissue characteristic of samples, the admixture of bone marrow was minimal in the collum sample and most in the caput sample.

Table 1.

Donor descriptions.

Person ID	subj1	subj2	subj3
Sex	Male	Male	Female
Age (years)	75	63.6	77.3
Height (m)	1.865	1.78	1.63
Weight (kg)	122.2	70	51.4
Hight loss (self-reported, cm)	0.5	1.5	3
BMD
L1 g/cm2	1.003	0.832	0.722
L1 T-score	−1.3	−2.7	−3.4
L1 Z-score	−1.4	−2.1	−1.2
L2 g/cm2	1.179	NA	0.702
L2 T-score	−0.5	NA	−4.2
L2 Z-score	−0.6	NA	−1.9
Left femoral neck total g/cm2	1.109	0.74	0.683
Left femoral neck total T-score	0.1	−2.6	−2.6
Left femoral neck total Z-score	0.6	−1.7	−0.6
Right femoral neck total g/cm2	1.067	0.755	NA
Right femoral neck total T-score	−0.2	−2.6	NA
Right femoral neck total Z-score	0.3	−1.7	NA
Head g/cm2	2.384	2.112	1.817
Total body g/cm2	1.413	1.055	0.965
Total body T-score	2.4	−2.1	−2
Total body Z-score	2.1	−1.4	0.2
Tissue (g)	114,406	67,488	48,498
Fat (g)	46,150	16,132	15,043
Lean mass (g)	68,256	51,356	33,455
BMC (g)	4524	2580	1789
Serum ionized Ca2+ (mmol/l)	1.12	1.22	1.17
Serum-Ca2+ corr. (mmol/l)	2.29	2.36	2.15
Serum-25(OH)vitD nmol/l)	78	81	63
Serum CTX-1 (ug/l)	0.3	0.49	0.73
Serum-bone spec alk phosphatase (U/i)	46	34	15.6
Serum BGLAP (nmol/l)	0.7	1.9	5.3

BMD: bone mineral density; BMC: bone mineral content; NA: not applicable; CTX-1 = carboxyterminal collagen crosslinks (bone resorption marker); Z-score = age adjusted T-score; BGLAP: bone gamma-carboxyglutamic acid-containing protein (osteocalcin)

Fig. 1.

Study design and tissues analyzed in this study.

2.1.1. Ethics

The study was approved by the Norwegian Regional Ethical Committee (REK no: 2010/2539) and conducted in accordance with the Declaration of Helsinki. Verbal and written informed consent were obtained from all participants in the study.

2.1.2. Human bone powder sample preparation for LC-MS/MS

The powdered bone samples were shipped frozen to the BGI Americas Mass Spectrometry Center for proteomic quantification using their standard protocol. Human bone powder samples were resuspended in lysis buffer (9 M Urea, 20 mM HEPES), homogenized and super-sonicated. Then the samples were centrifuged at 14,000 rpm and the supernatant was collected for BCA assay and following analysis. Fifty μg of each sample were adjusted to the same volume with lysis buffer, then reduced with dithiothreitol (DTT) and alkylated with Iodoacetamide (IAA), followed by Trypsin/LysC (Thermo Fisher Scientific) digestion at 37 °C overnight.

2.1.3. Human serum sample preparation for LC-MS/MS

Eight μL of each human serum sample were first processed with High Select™ Top14 Abundant Protein Depletion Resin (Thermo Fisher Scientific, A36372), then reduced with DTT, and alkylated with IAA, followed by Trypsin/LysC (Thermo Fisher Scientific) digestion at 37 °C overnight. After digestion, samples were acidified with 10% TFA and then cleaned up with C18 stage tips (Biotage). The peptides were dried down in SpeedVac (Thermo Fisher Scientific).

2.2. DIA-PASEF quantitative proteomics

2.2.1. LC-MS/MS analysis and DIA-PASEF data acquisition

Peptide samples were analyzed using a nano-flow liquid chromatography tandem mass spectrometry (LC-MS/MS) platform consisting of a nanoElute UHPLC system (Bruker Daltonics, Billerica, MA, USA) coupled to a timsTOF Pro 2 mass spectrometer (Bruker Daltonics). Peptides were separated on an Aurora Series UHPLC analytical column (75 μm × 25 cm, 1.6 μm particle size; IonOpticks, Australia) using a 120-min water-acetonitrile gradient at a flow rate of 0.28 μL/min. Ionization was achieved using a CaptiveSpray ion source operated at 1.7 kV.

To generate a project-specific spectral library, pooled and fractionated peptide samples were analyzed using data-dependent acquisition parallel accumulation serial fragmentation (DDA-PASEF). DDA-PASEF data were acquired over a mass range of 100–1600 m/z and an ion mobility range of 0.70–1.40 1/K0. Four PASEF MS/MS scans were acquired per cycle with a total cycle time of 0.53 s. An intensity threshold of 1000 and a target intensity of 15,000 were applied. Collision energies were ramped according to ion mobility, ranging from 27 eV at 0.85 1/K0 to 45 eV at 1.30 1/K0.

Quantitative proteomic analysis of experimental samples was subsequently performed using data-independent acquisition parallel accumulation serial fragmentation (DIA-PASEF). Full MS scans were acquired over a mass range of 100–1600 m/z and an ion mobility range of 0.70–1.40 1/K0. DIA-PASEF acquisition employed 20 mass steps per PASEF cycle with isolation windows of 50 Da. DIA windows covered a precursor mass range of 682.6–1682.6 Da. Collision energies were linearly ramped from 27 to 45 eV across the ion mobility range.

2.2.2. Spectral library generation

DDA-PASEF data acquired from fractionated samples were searched against the UniProtKB/Swiss-Prot Homo sapiens protein database (UP000005640, available at the time of analysis in December 2020) using the Andromeda search engine implemented in MaxQuant. Peptide and protein identifications were filtered at a false discovery rate (FDR) of 1%. Identified peptides passing quality-control criteria were used to generate a project-specific spectral library.

DIA-PASEF data from experimental samples were analyzed using Spectronaut™ (Biognosys, Switzerland). The DDA-derived spectral library was used for peptide identification, signal extraction, and deconvolution of DIA data. Analytical quality control was performed using the mProphet algorithm with a 1% FDR threshold at the peptide and protein levels. Protein abundances were calculated from DIA quantitative signals generated by Spectronaut. Peptide-level quality metrics were derived from the Proteome Discoverer protein-group outputs generated separately for each bone site. For proteins identified in multiple sites, the highest peptide support observed across sites was used to summarize overall protein identification confidence.

2.3. Bioinformatics analysis

Given the small sample size and feasibility purpose of the study, we did not measure the absolute protein mass. For all downstream analyses, protein mass (PM) was defined as a log2 transformation of the original DIA quantitative signals. For descriptive analyses and data visualization, log2 transformed abundances were used directly. For statistical analyses other than descriptive analyses, the PM were median-normalized sample-wise. Briefly, the median abundance of each sample was calculated, and all protein abundances within a sample were adjusted by a constant offset so that the sample median equaled the mean median across all samples. The average PM over three samples for a bone site or serum was used to estimate the protein abundance at a site. Proteins not detected in a given sample were treated as missing values and were not imputed. We refer to ‘site’ as the samples obtained from a defined femur site or serum.

2.3.1. Protein diversity and site-specific proteome analysis

The number of proteins detected in each bone and serum sample was summarized and compared across anatomical sites and subjects. Site-specific proteins were defined as proteins detected in all three biological replicates from one anatomical site but not detected in either of the other two bone sites. Similarly, serum-specific proteins were defined as proteins detected in serum but absent from all bone samples. The overlap of detected proteins among bone sites, serum, and individual subjects was visualized using Venn diagrams.

2.3.2. Clustering and correlation analysis

For the clustering of samples from bone sites, 1291 proteins commonly found in all nine bone samples were used. Principal components analyses were performed by the prcomp function in R on the median-normalized PM. The first two principal components of proteins for each sample were plotted to visualize the sample clustering. The similarity of two sites was quantified as the Pearson correlation of the site-average abundance score of all commonly found proteins in the two sites.

2.3.3. Comparison of protein abundance between bone and serum

To compare the most abundant proteins in bone and serum, we first calculated the site-average protein abundance (mean PM of three samples in each site) and ranked the proteins accordingly. The top 50 proteins with the largest site-average protein abundance were extracted for each site. The bone-abundant proteins (BAPs) were defined as the intersection of the three sets of the 50 most abundant proteins in each bone site. The serum-abundant proteins (SAPs) were defined as the top 50 proteins with the largest site-average protein abundance in serum. Protein abundance fold-change (FC) of each bone site over serum was calculated as (site-average protein abundance in a bone site/site-average protein abundance in serum). Based on the presence of proteins and its FC of bone over serum, we grouped the BAPs and SAPs into five categories: 1) proteins specific to bone: found in one or more bone sites but not in any serum sample; 2) proteins specific to serum: found in serum but not in any bone samples; 3) proteins enhanced in bone: commonly found in serum and one or more bone sites, and the FC > 1 for all bone sites vs serum; 4) proteins reduced in bone: commonly found in serum and at one or more bone sites with FC < 1 for all bone sites vs serum; 5) proteins homogenously distributed: commonly found in serum and one or more bone site(s) with FC not reaching the statistical criteria of enhancement/reduction (Table S2). We also compared the bone and serum protein abundance of proteins in the lists of classical bone protein and plasma proteins from previous publications (Alves et al., 2011; Anderson and Anderson, 2003) to check if the protein distribution pattern replicates. For the proteins that are commonly detected at all three bone sites, we performed repeated-measures mixed-effect analysis with the effect of a bone site as a fixed effect and the effect of subjects as a random effect to exploratorily assess differences in protein abundance across the three bone sites.

2.3.4. Groups of proteins characteristic of the different cell types

We compared how the proteins characteristic for bone marrow mesenchymal stem cells, adipocytes, osteoblasts/ osteocytes, and cells belonging to the hematological/immunological cell lineages distributed between the different persons and at the three femoral sites of the same person. A heatmap table was generated using the rank-normalized PM of the listed proteins. The protein lists were taken from published datasets and public databases, including some datasets derived from transcriptomic analyses in mouse models. (BMSC: Fonseca et al. (Fonseca et al., 2023); Adipocytes: Ahn et al. (Ahn et al., 2019); Osteoblasts/osteocytes: Youlten et al. (Youlten et al., 2021) and the Harmonizome portal (https://maayanlab.cloud/Harmonizome/gene_set/osteoblasts/BioGPS+Mouse+Cell+Type+and+Tissue+Gene+Expression+Profiles); hematology or immunology characteristic genes: The Human Protein Atlas (https://www.proteinatlas.org/ENSG00000005381-MPO/tissue) (Uhlén et al., 2015; Uhl et al., 2015). These analyses were intended as exploratory indicators of broad cell-type representation rather than definitive cell-specific annotation.

2.3.5. Cellular annotation of site-specific protein signatures

To provide cellular context for the site-specific proteomic signatures, site specific proteins were mapped to their corresponding gene and queried in a reference human bone marrow single-cell RNA sequencing (scRNA-seq) dataset (Bandyopadhyay et al., 2024). For each annotated cell type, the expression level and the fraction of expressing cells were calculated for every mapped gene. A cell-type association score was then computed as the product of the mean expression and the fraction of expressing cells, normalized across all cell types for each gene. Each gene was assigned to its most strongly associated cell type based on the highest association score. For each anatomical protein set, the proportions of genes assigned to each cell type were summarized, and weighted summaries based on the top three associated cell types per gene were also generated. The percentage of proteins in each input set associated with the top 12 cell types were visualized using a dot plot.

2.3.6. Gene ontology and pathway enrichment

Each LC-MS/MS identified protein was annotated to one or more gene ontology (GO) terms. We calculated a site-wide frequency of a GO term as the number of proteins annotated to the GO term divided by the total number of unique proteins measured in the site and plotted the site-wide GO frequency of caput, collum, trochanter and serum in a bar chart to compare the GO distribution of proteins. The difference in site-wide GO frequency of each bone site vs serum was tested by Pearson's chi-square test using the prot.test function in R. We considered the proteins that were detected in only one bone site as bone-site specific. For each set of site-specific (caput, collum, and trochanter) proteins, pathway enrichment analysis was carried out using Metascape (Zhou et al., 2019) with three ontology sources: KEGG Pathway, Canonical Pathways, and WikiPathways. All genes in the genome were used as the enrichment background. Terms with a p-value <0.01, a minimum count of 5, and an enrichment factor > 2.0 (the enrichment factor is the ratio between the observed counts and the counts expected by chance) were collected. Enrichment p-values were calculated based on the accumulative hypergeometric distribution, and q-values were calculated using the Benjamini-Hochberg procedure (Benjamini and Hochberg, 1995) to account for multiple testing.

3. Results

3.1. Protein diversity and abundance in bone

Through integrative data analyses, we found that the bone proteome is significantly more diverse than that of serum. From the nine bone samples (three caput, three collum, and three trochanter samples), a total of 3085 proteins were detected and quantified using LC-MS/MS. In contrast, a total of 597 proteins were quantified in the three serum samples. On average, individual subjects had 1752, 2298, and 2365 detected proteins in the caput, collum, and trochanter, respectively. Each individual showed a high similarity in the number of detected proteins in both bone and serum (Table 2). However, each femur site contained a set of unique proteins that were not found in other sites. The percentage of suggested site-specific proteins was 7.2% in caput (142/1963), 12.0% in collum (310/2575), and 11.4% in trochanter (292/2571) (Fig. 2A). Approximately one-third (218/597) of the serum proteins were not detected in bone tissue, and half (297/597) of the serum proteins were common to all three skeletal sites (Fig. 2B). The three sets of site-specific proteins shown in Fig. 2A (caput 142, collum 310, and trochanter 292) were detected in all participants (Fig. S1A-C). We further visualized the cross-subject proteome variation with a Venn diagram (Fig. S2). Compared to Fig. 2A, the number of subject-specific proteins is much smaller than the number of site-specific proteins, suggesting possible between-site variation.

Table 2.

Protein diversity in each sample.

	Caput	Collum	Trochanter	All Bone Sites	Serum	Bone+Serum
Subject 1	1907	2500	2476	2986	572	3211
Subject 2	1722	1911	2132	2536	571	2772
Subject 3	1626	2482	2487	2942	586	3165
Average	1751.7	2297.7	2365.0	2821.3	576.3	–
All Subjects	1963	2575	2571	3085	597	3303

Average: the arithmetic mean number of proteins identified in the three individuals;

All-subjects: the union of all unique proteins detected across the three individuals at each anatomical site

Fig. 2.

Protein diversity and clustering. (A) Protein distribution across three anatomical bone sites from three persons. (B) Protein distribution between the three bone sites and serum. (C) Scatter plot of first two PCs based on commonly detected proteins in three bone sites showing the sample clustering. (D) Heat map showing the pairwise correlation between three bone sites and serum based on protein abundance of commonly detected proteins in each pair.

To facilitate assessment of protein identification quality, peptide-level metrics have been added to the supplementary materials (Table S3), including the number of peptides, peptide-spectrum matches (PSMs), unique peptides, razor peptides, and sequence coverage for each identified protein. Across 3303 unique protein identifications, protein-level quality metrics were summarized using the bone site with the highest peptide support for each protein. The overall protein identifications were supported by a median of 4 peptides and 6 peptide-spectrum matches (PSMs) per protein, with 1905 proteins (57%) identified by at least two unique peptides and only 259 (8%) proteins rely on razor peptides. The median sequence coverage was 14%. Together, these metrics provide additional confidence in the protein identifications and subsequent relative abundance estimates reported in this study. Protein identifications were assigned confidence levels by Proteome Discoverer based on the combined protein q-value. High-, medium-, and low-confidence proteins corresponded to combined protein q-values of ≤0.01, 0.01–0.05, and > 0.05, respectively. As this was an exploratory pilot study, all identified proteins (maximum combined protein q-value <0.1) were retained for downstream analyses. The combined protein q-value and corresponding confidence category for each identified protein are provided in Table S3.

To investigate whether site-specific proteins are likely to be borderline or abundant, we first evaluated the relationship between protein abundance and detection frequency across all 3303 bone proteins. A strong positive correlation was observed between mean protein abundance and the number of samples in which a protein was detected (Spearman's rank correlation, ρ = 0.689, P < 2.2 × 10⁻¹⁶), indicating that higher-abundance proteins were detected in a greater number of samples. We then compared the abundance distributions of site-specific proteins (142 caput-specific, 310 collum-specific, and 292 trochanter-specific proteins) with those of proteins detected at multiple bone sites using a Wilcoxon rank-sum test. Site-specific proteins exhibited significantly lower abundance than non-site-specific proteins (P < 2.2 × 10⁻¹⁶).

We also looked into the overlap of proteins in the bone samples with tissues in Human Protein Atlas (HPA) (Uhlén et al., 2015) and found that 806 out of 3085 proteins detected in bone overlapped with the 6133 proteins detected in 55 different tissues using antibody-based immuno-histochemistry (protein atlas version 20.1). The remaining 2279 out of 3085 proteins were not detected in any of the 55 tissues. Among the 806 overlapping proteins from bone samples and any HPA available tissue, 87 were also found in our serum samples. We detected 36 out of the 38 previously reported abundant bone tissue proteins (Alves et al., 2011), and only 11 of these 36 proteins were found in the HPA database (v20.1). These results suggest that bone contains a unique source of proteins complementary to the proteome of multiple different human tissues.

The distribution of protein abundancy in all 12 samples was slightly skewed after log2 transformation (Fig. S3). The plot of the first two principal components (PCs) of the median-normalized PM revealed samples primarily clustered according to bone site rather than subject, indicating that site-specific proteomic differences were the major source of variation (Fig. 2C). Hierarchical clustering based on the median-normalized PM further supported the PCA results (Fig. S4). The different clustering patterns at three anatomically, neighboring sites may be due to different compositions of trabecular or cortical bone possibly reflecting variations in microarchitecture. We then investigated the correlation between the global proteomic profiles at the different femur sites and serum (Fig. 2D) and observed marked similarities between bone samples while the serum presented as an outlier.

3.2. Most abundant proteins in bone and serum samples

By intersecting the top 50 most abundant proteins in each bone site, we obtained 33 bone-abundant proteins (BAPs). The site-average abundance of the 33 BAPs is shown in Fig. 3A and Table S4A. Among the 33 BAPs, 4 (12.1%) were specific in bone, 18 (54.5%) were enhanced in bone, 3 (9.1%) were homogeneous in bone and serum, and 8 (24.2%) was reduced in bone. Among the 50 serum-abundant proteins (SAPs), 48 (96.0%) were reduced in bone and 2 (4.0%) were homogeneous in bone and serum (Fig. 3B and Table S4B). Ten proteins overlapped between BAP and SAP, including serum albumin, immunoglobulin (gamma 1, alpha 1, and kappa), apolipoprotein, serotransferrin, keratin, haptoglobin, alpha-2-HS-glycoprotein, and prothrombin (Table S4A, B). Excluding the 10 overlapping proteins, 86.2% (19/22) of the bone-specific BAPs were either enhanced or specific in bone (Table S4A). Analogously, all of the serum-specific SAPs were reduced in bone (Table S4B). These results suggest that the BAPs and SAPs might represent different features of functional relevance and that the bone BAP levels, compared to that in serum, might more significantly reflect certain biological conditions of the subjects.

Fig. 3.

Bone abundant proteins (BAPs) and serum abundant proteins (SAPs). (A) Site average abundance of BAPs in three bone sites and serum. For the fold-change of protein abundance in each bone site over serum, see Table S4A. (B) Site average abundance of SAPs in three bone sites and serum. For the fold-change of protein abundance in each bone site over serum, see Table S4B.

To further validate our findings, we incorporate a previously identified list of bone signature proteins from cancellous bone samples (Alves et al., 2011). The list contains 23 most abundant proteins in the cancellous bone, 11 extracellular matrix (ECM) proteins, and 8 classical bone proteins (Table S5A). Two out of the 38 proteins were not detected in our samples. Among the 36 detected proteins, 19 (52.8%) were specifically expressed in bone, 13 (36.1%) were enhanced in bone, three (8.3%) were reduced in bone and one (2.8%) were similarly expressed in bone and serum (Table S5B, Fig. S5A). Additionally, we examined in our samples the 35 most abundant serum proteins as listed in a review article (Anderson and Anderson, 2003) (the full list is shown in Table S5C), and 33 (94.3%) were detected in both bone and serum. Of these, 24 (68.6%) were reduced in bone, five (14.3%) were similarly expressed, and four (11.4%) was enhanced in bone (Table S5D, Fig. S5B). After Bonferroni correction for multiple testing (significance threshold = 0.05/1683), repeated-measures mixed-effect analysis identified five proteins with significant differences in abundance across the three bone sites: UCHL5, STOM, RPS4X, IGHG3, and H3-3 A. (Table S6).

3.3. Groups of proteins characteristic of the different cell types

The bone samples contained bone and bone marrow, thus representing a composite tissue where cells are functionally linked. We attempted to categorize how the proteins characteristic for bone marrow mesenchymal stem cells (Table S7A), adipocytes (Table S7B), osteoblasts/ osteocytes (Table S7C), and cells belonging to the hematological/immunological cell lineages (Table S7D) distributed between the different persons and at the three femoral sites of the same person. The results showed a high resemblance in the cell characteristic protein signatures between sites and between persons and as expected, distinct from serum. However, interpretation of these analyses is limited by incomplete overlap between transcriptomic and proteomic datasets, as well as potential species differences in the reference signatures used. Several of the proteins described in the literature failed to reach the analysis detection level. In contrast to the global analysis, we did not discover a clear trend pointing to noticeable differences.

To further investigate the proteins in a cellular context, we utilized single cell RNA-seq data from human bone marrow to detect expression of site-specific proteins (Bandyopadhyay et al., 2024). For each protein we looked at the corresponding transcript and identified the top three expressing cell types with weighted ranks. The cell types for each site were aggregated to produce a dot plot profile (Fig. 4). The results indicate a large fraction of site-specific proteins from bone sites are chiefly expressed by hematopoietic cell types, such as erythroblasts, and megakaryocytes. Interestingly, the serum-specific proteins showed enriched expression in bone marrow mesenchymal stem cells (MSC).

Fig. 4.

Top three weighted cell-type associations by protein set. Cell-type associations were determined by assigning each protein to its three highest-weighted cell-type signatures and summarizing the assignments within each input protein set. Columns represent proteins identified in the femoral caput (n = 140), femoral collum (n = 304), serum (n = 218), and trochanter (n = 291). Dot size and color represent the percentage of proteins in each input set associated with the indicated cell type, with larger and lighter-colored dots indicating a higher percentage.

3.4. Gene ontology distribution of proteins in bone

To elucidate the biological role of proteins in bone and serum, we compared the site-wide GO term frequency of each bone site and serum. The proteins were annotated to 17 biological process terms, 14 molecular function terms, and 17 cellular component terms. The frequency of each GO term is shown in Fig. 5A-C. The metabolic process, regulation of biological process, response to stimulus, transport, and cell organization were shortlisted for the top frequent biological process GO terms with bone tissue frequency > 40%. In 59% (10/17) biological process terms, the frequency of bone proteins and serum proteins significantly differs from each other (chi-square test p < 0.05, Fig. 5A, Table S8A). The top molecular functions included protein binding, catalytic activity, metal ion binding, and RNA binding with bone GO term frequency > 20%. In 57% (8/14) of the molecular function terms, the frequency of bone proteins and serum proteins significantly differed from each other (chi-square test p < 0.05, Fig. 5B, Table S8B). Not unexpectedly, the cellular component GO categories exhibited the most contrasting pattern. The major cell localization of bone proteins was cytoplasm, cytosol, and nucleus while serum proteins were classified as extracellular and cell surface localization. Bone proteins and serum proteins in 11 (65%) GO categories distributed differently with chi-square p-value <0.05, including cytoplasm, cytosol, nucleus, extracellular, organelle lumen, mitochondrion, etc. (Fig. 5C, Table S8C). We further investigated the important pathways represented by site-specific proteins in each bone site using MetaScape (https://metascape.org/) (Zhou et al., 2019) annotating the top five most enriched pathways (Table 3, the full list of enriched pathways can be found in Table S9). Notably, genes specific to the caput region exhibited significant enrichment in striated muscle contraction, while those specific to the trochanter region were associated with the osteoclast differentiation pathway.

Fig. 5.

Frequency of GO protein annotations at each bone site and serum: (A) Biological Process; (B) Cellular Component; (C) Molecular Function. Star (*) indicates at least one bone site is significantly different from serum in terms of annotation frequency at p < 0.05 significance level. For test statistics of all GO terms, see Table S8A—C.

Table 3.

Top 5 pathway with bone site-specific gene enrichment. “Count” is the number of genes in the user-provided lists with membership in the given ontology term. “%” is the percentage of all of the user-provided genes that are found in the given ontology term (only input genes with at least one ontology term annotation are included in the calculation). “Log10(P)” is the p-value in log base 10. “Log10(q)” is the multi-test adjusted p-value in log base 10.

GO	Category	Description	Count	%	Log10(P)	Log10(q)	Proteins
Caput
WP383	Wiki Pathways	Striated muscle contraction pathway	12	9.38	−18.95	−15.71	MYH3, ACTN3, MYBPC2, TNNC1, MYOM1, TNNI2, TNNC2, TNNT1,TNNI3, NEB, DMD, MYH8
hsa01200	KEGG Pathway	Carbon metabolism	6	4.69	−4.85	−2.09	ENO3, PGAM2, PRPS2, PFKM, ACSS2, CPS1
hsa04530	KEGG Pathway	Tight junction	5	3.91	−2.96	−0.62	MYH2,MYH3,PPP2CB,MYH8,AMOTL1
hsa04144	KEGG Pathway	Endocytosis	6	4.69	−2.91	−0.62	HSPA6,ADRBK2,GRK3,SH3GL1,STAM,CHMP3(RNF103-CHMP3)
Collum
M7	Canonical Pathways	PID FCER1 PATHWAY	7	2.49	−5.65	−2.86	HRAS,PIK3R1,MAP2K2,PXN,CBL,RASA1,FCER1G
hsa04144	KEGG Pathway	Endocytosis	13	4.63	−5.62	−2.86	HRAS,EPS15L1,SNF8,VPS36,ARAP1,ARRB1,VPS25,CHMP4A,CHMP4B,RUFY1,TSG101,CBL,RAB11A
ko03013	KEGG Pathway	RNA transport	10	3.56	−5.04	−2.4	NCBP1,EIF5B,THOC3,EIF3G,NUP37,NUP93,NUP88,SAP18,FXR1,NUP160
WP3888	Wiki Pathways	VEGFA-VEGFR2 signaling	14	4.98	−3.83	−1.77	ACOT9,CBL,FAM120A,GRSF1,HRAS,INPP5K,MAP2K2,MMP2,PDE4DIP,PIK3R1,PXN,RAB11A,RPL26,TMOD1
ko05164	KEGG Pathway	Influenza A	8	2.85	−3.44	−1.53	AGFG1,DNAJB1,HLA-DPB1,IRF3,KPNA1,MAP2K2,MX1,PIK3R1
Trochanter
ko04380	KEGG Pathway	Osteoclast differentiation	9	3.15	−5.12	−1.97	CREB1,IKBKG,LCP2,LILRB5,MAP2K1,NCF1,NOX1,PLCG2,SIRPB1
hsa05016	KEGG Pathway	Huntington's disease	9	3.15	−3.77	−1.14	AP2S1,COX7B,COX7C,CREB1,DNAH11,DNAH5,NDUFA6,NDUFS5,UQCR11
WP437	Wiki Pathways	EGF/EGFR signaling pathway	8	2.8	−3.54	−1.01	AP2S1,ASAP1,CREB1,EPN1,KIAA1033,MAP2K1,RAB5A,RPS6KA3,SH3KBP1
hsa05120	KEGG Pathway	Epithelial cell signaling in Helicobacter pylori infection	5	1.75	−3.19	−0.95	ATP6V0C,ATP6V1F,IKBKG,NOD1,PLCG2
ko04145	KEGG Pathway	Phagosome	7	2.45	−2.99	−0.83	ATP6V0C,ATP6V1F,NCF1,NOX1,RAB5A,THBS4,TUBB8

4. Discussion

In this study, we have characterized the human proteomes from surgically removed bone specimens representing composite tissue also containing bone marrow. However, bone and marrow cells are functionally closely linked already during embryological development of the skeleton and blood-forming cell lineage as well as later in life. Quantitative LC-MS/MS proteomics data were generated from femur caput, collum, and trochanter, and compared to the corresponding serum samples in three individuals. Under the current analytical workflow, the majority of the proteomes were shared across the three bone sites. The bone and bone marrow proteomes exhibited greater detected proteome diversity and distinct protein compositions compared with serum. In agreement with Procopio et al (Procopio et al., 2017), we found that protein profiles were highly consistent among individuals within each anatomical site, whereas greater proteomic differences were observed among the caput, collum, and trochanter within the same individual. These findings indicate that anatomical location is an important determinant of the detected bone proteome. We further identified proteins detected exclusively at individual anatomical sites, suggesting potential site-related proteomic specialization. However, these observations should be interpreted as exploratory because of the small sample size, the limited LC-MS/MS analytical depth, and the lack of independent orthogonal validation. Future studies with larger cohorts, deeper proteomic profiling, and complementary validation approaches will be important to confirm the anatomical localization and biological significance of these candidate site-specific proteins.

The study highlighted several methodology strengths. The tissue samples were freshly collected from patients undergoing hip replacement surgery under spinal anesthesia to maintain the quality of the bone specimen. These samples were immediately preserved on liquid nitrogen before processing to avoid protein degradation. The study proves the feasibility of the methodological approach where well-preserved bone biopsies can be used to identify and quantify bone proteins on a proteomic scale. By comparing the bone and serum proteome from the same patient, we have established a platform which may be used for the discovery of bone/bone marrow characteristic proteins possibly also related to diseases. Furthermore, we demonstrate that proteomics data from bone will be valuable to provide complementary information beyond what blood analyses have offered relevant to skeletal diseases. Although ultra-sensitive immunoassay techniques like the Proximity Extension Assay (PEA) offer higher sensitivity for protein measurement, they rely on specific antibody panels. In contrast, LC-MS/MS, the gold standard for untargeted proteomics, is better suited for discovering new proteins without biases related to antibody availability or cross-reactivity. A further methodological consideration is that the bone specimens were analyzed without decalcification before proteomic processing. This approach was chosen to minimize protein loss and preserve the native composite tissue environment, although it likely contributed to overlap between bone and bone marrow proteomes and may have increased comparability with the corresponding serum proteomes.

The specimen represented bone marrow in addition to cortical bone, especially in the collum, and trabecular bone predominantly present in the caput and trochanter samples. There is a recognized, close functional interaction between bone marrow cells, notably stromal and progenitor cells and bone-specific cells. In an attempt to further discover possible proteome differences between persons and also between femur sites of the same individual, we supplemented the global proteome analyses with a description of cell-type characteristic proteins. Tables S7A-D present the overall results. One general finding was that the detected proteins were not a complete map of the listed cell characteristic proteins, and as expected, several of the proteins belonged to more than one cell phenotype as shown e.g. in Table S7C, presenting the osteoblast/osteocyte characteristic proteins. There are also limitations regarding the interpretation of the data since the status of available knowledge accessible for comparing experimental human proteomics data is limited. For example, Ahn et al.'s (Ahn et al., 2019) lists of adipocyte characteristic proteins, did not include fatty acid synthase (Uniprot ID: P49327) which was abundantly present in all biopsies from the three persons (rank-normalized abundance range from 0.80 to 1.39). We also discovered hepatic triacylglycerol lipase/Lipase C which was also not included among adipocyte characteristic proteins in Ahn's paper (Ahn et al., 2019). Therefore, a better functional interpretation of bone proteomics data has to await further studies. The present protein map is only valid to describe the proteome from elderly people, and bone/bone marrow composition is known to change with age, e.g. red bone marrow is partly replaced by white marrow in elderly people (Tuljapurkar et al., 2011; Johnston and Buckley, 2020).

We were surprised by the results of our analysis on site-specific protein expression with single cell data. The samples derived from bone biopsy contained genes that were enriched with hematopoietic cell types, while the serum sample contained genes enriched in MSC cell types. We reason this is due to secreted proteins from MSCs having a greater relative abundance in serum as opposed to bone marrow. In bone marrow, MSCs only account for 0.3% of the total cell population far outnumbered by hematopoietic cell types (Gomariz et al., 2018). Additionally, MSCs are located on blood capillaries and are known to have paracrine and endocrine functionality (Han et al., 2022). Therefore, we suspect that hematopoietic specific proteins are enriched in bone marrow due to the preponderance of hematopoietic cells in bone marrow. Megakaryocytes and erythroblasts produce platelets and red blood cells respectively, most of the proteins in these cells will be contained within their daughter cells rather than being free in serum. Secreted proteins by MSCs however are likely to be transiently secreted and may be destroyed by target cells in the bone marrow, yet persisting longer in serum.

We chose to use serum instead of plasma and therefore may have missed certain low abundancy clotting factors whose concentrations could have been reduced. E.g. the amounts of fibrinogen are lower in serum than in the biopsies, while prothrombin has, as expected, highest levels in serum. The three donors had coxarthrosis, and one participant additionally had manifest osteoporosis while another had marked obesity, all factors that may have influenced bone/bone marrow proteomic composition. Therefore, the present protein profiles may reflect a pathological rather than physiological skeletal state, although the biopsies were collected relatively far from the affected joint region. It was inevitable that the bone sample reflected a pathological rather than a physiological status, as its acquisition involved a surgical and invasive procedure. Furthermore, the absence of a healthy control group limits the generalizability of the findings to normal bone biology. Animal models provide a valuable complementary platform for investigating skeletal proteomic variation under controlled experimental conditions and with larger sample sizes. Although animal bone proteomic datasets are increasingly available, comprehensive atlases of site-specific skeletal regions remain limited (Mullin et al., 2023). Future integration of animal and human proteomic studies may facilitate validation of candidate site-specific proteins and improve our understanding of skeletal site specificity.

This pilot study included only three participants and was therefore not powered for robust statistical inference regarding site-specific protein abundance. Nevertheless, repeated-measures mixed-effect analysis identified a small number of proteins exhibiting significant site effect after correction for multiple testing, providing preliminary evidence of molecular difference across femur sites. The exploratory pathway and clustering analyses further suggested potential site-related biological differences. Given the sample size, however, the findings should primarily be interpreted as hypothesis-generating and feasibility-supporting. Future studies with substantially larger cohorts and integration with transcriptomic and structural tissue data will be necessary to validate the observed trends. With these reservations, we observed a consistent and notable trend suggesting that the number of proteins in the caput sample was less than in collum and trochanter. If confirmed in extended studies, quantitative or qualitative differences in protein composition may reflect variations in microarchitecture and then functional demands. The detection of proteins belonging to the ‘striated muscle contraction’ pathway (Table 3) specifically in caput may hypothetically reflect increased osteocyte activity at this site, since osteocytes are dynamic cells with protruding and contracting dendritic processes exerted by similar proteins as found in muscle (Dallas and Bonewald, 2010; Paic et al., 2009). However, this interpretation remains speculative and requires validation in larger and functionally oriented studies. Some enriched pathways identified from site-specific proteins, particularly in the collum samples, are not obviously linked to established bone biology and should therefore be interpreted cautiously.

Residual blood contamination and regional differences in bone marrow composition may contribute to variation among anatomical bone sites. To evaluate this possibility, we examined the abundance or representative erythrocyte proteins hemoglobin subunit α1/ α2 (HBA1/HBA2), hemoglobin subunit β (HBB) and hemoglobin subunit δ (HBD) together with major serum proteins albumin (ALB), transferrin (TF), and haptoglobin (HP) across all bone samples. The circulating serum proteins (ALB, TF, and HP) showed modest site related difference, higher in the collum than in the caput or trochanter. But the three erythrocyte signature proteins (HBA1/A2, HBB and HBD) did not display a consistent pattern across the three bone sites (Fig. S6). These findings suggest that differences in residual blood and/or marrow-associated hematopoietic content may partially influence the observed proteomic variation. Because erythrocytes are also a normal component of bone marrow, differences in hemoglobin abundance may reflect variation in marrow content, residual blood, or intrinsic biological differences between anatomical sites, which cannot be distinguished in this study. Nevertheless, bone-associated proteins, including COL1A1, SOST, and ALPL, also exhibited site-dependent variation. To further address this potential confounding factor, we performed repeated-measures mixed-effects analyses that accounted for within-subject correlation. Several proteins remained significantly associated with anatomical site, including UCHL5, STOM, RPS4X, IGHG3, and H3-3 A (Table S6), and their abundance patterns differed from those of the erythrocyte-associated proteins (Fig. S7). Collectively, these results indicate that while differences in blood and/or marrow content may contribute to some of the observed proteomic variation, they unlikely to fully account for the observed the site-specific differences in this study.

The serum proteome coverage in this study was obtained using a single-shot DIA workflow following Top14 depletion without additional fractionation. Because serum spans an exceptionally wide dynamic range of protein abundances, this approach is expected to detect fewer low-abundance proteins than deep-proteome workflows employing extensive fractionation or multiple LC-MS acquisitions (Keshishian et al., 2017; Ward et al., 2024). Therefore, the observed differences between bone, bone marrow, and serum should be interpreted within the context of the analytical depth achieved in this study. Although single-cell approaches could provide a higher resolution of the proteome with relevance to both cell type and disease (Labib and Kelley, 2020; Schoof et al., 2021), it is still technically challenging to obtain single-cell samples from hard tissue with preserved protein quality. It has been shown that albumin is present in bone tissue and bone-related cells (Ishida et al., 2004). Information from the HPA database also indicates that hemoglobin subunit genes are highly expressed in bone marrow (Uhlén et al., 2015), but since no bone tissue data is available, it is not clear if the detection of hemoglobin in our bone samples is due to the contamination of blood-derived cells or circulation of hemoglobin subunit between bone matrix and bone marrow (Marenzana and Arnett, 2013). Recently, gene-expression deconvolution methods have been developed to estimate the cell-type gene-expression profile and the cell-type fractions simultaneously (Zaitsev et al., 2019). However, these methods require a larger sample size resulting in average numbers. Anticipating this problem, we used strict sample preparation precautions to reduce the contamination of blood. If the amount of bone marrow contained in the bone samples is not highly variable across bone sites, the relative differences observed in protein abundance across caput, collum, and trochanter samples should be valid, especially regarding proteins that do not belong to serum or the hematological/immunological cell systems. Although the biopsies were obtained using standardized anatomical sampling procedures, we did not directly quantify marrow content in each specimen. Therefore, variation in marrow admixture between sites may have contributed to some of the observed proteomic differences, and future studies may benefit from more selective isolation of mineralized bone tissue.

In conclusion, the study shows the feasibility of performing quantitative global proteomics on human bone samples. We present a preliminary pilot description of femur bone/bone marrow proteomic composition demonstrating the feasibility of quantitative proteomics in human bone biopsies. The observed trends toward site-related proteomic differences require validation in larger studies. These findings will encourage further studies and may help in the selection and preparation of tissue samples for bone-related proteomics research.

CRediT authorship contribution statement

Fangtang Yu: Writing – review & editing, Writing – original draft, Visualization, Methodology, Formal analysis. Sjur Reppe: Writing – review & editing, Writing – original draft, Resources, Methodology, Formal analysis, Data curation, Conceptualization. Vigdis T. Gautvik: Resources, Investigation, Data curation. Keller McKowen: Formal analysis, Writing – review & editing. Wei-Zhi Lin: Formal analysis, Writing – review & editing. Tor P. Utheim: Investigation, Data curation. Kaare M. Gautvik: Writing – review & editing, Supervision, Funding acquisition, Conceptualization. Yi-Hsiang Hsu: Writing – review & editing, Supervision, Funding acquisition, Conceptualization.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

Supplementary material

Table S7

A. Relative levels of proteins from mesenchymal stem cell characteristic genes

B. Relative levels of proteins from adipocyte characteristic genes

C. Relative levels of proteins from osteoblast/osteocyte characteristic genes

D. Relative levels of proteins from hematology or immunology characteristic genes

Fig. S1.

Protein diversity at three sites. (A) Protein distribution at Caput from three subjects. (B) Protein distribution at Collum from three subjects. (C) Protein distribution at Trochanter from three subjects.

Fig. S2.

Protein distribution across three subjects.

Fig. S3.

Distribution of protein abundancy in all study samples.

Fig. S4.

Heat map with the clustering of all samples and proteins.

Fig. S5.

Bone abundant proteins (BAPs) and serum abundant proteins (SAPs). (A) Site average abundance of BAPs in three bone sites and serum. For the fold-change of protein abundance in each bone site over serum, see Table S5B. (B) Site average abundance of SAPs in three bone sites and serum. For the fold-change of protein abundance in each bone site over serum, see Table S5D.

Fig. S6.

Site-specific abundance of representative blood and bone proteins across matched samples from three donors. Paired line plots show the relative abundance of representative erythrocyte-associated proteins (HBA1/HBA2, HBB, and HBD), circulating serum proteins (ALB, TF, and HP), and bone-associated proteins (COL1A1, SOST, and ALPL) in matched serum, femoral caput, collum, and trochanter samples from three individuals.

Fig. S7.

Relative abundance of proteins exhibiting significant anatomical site differences by repeated-measures mixed-effects analysis. Paired line plots show the relative abundance of proteins identified as significantly associated with anatomical site, including UCHL5, STOM, RPS4X, IGHG3, and H3-3 A, across matched serum, femoral caput, femoral collum (neck), and trochanter samples from three individuals.

Data availability

Data will be made available on request.