Proteomic Analyses of Plasma from Patients with Fracture Related Infection Reveals Systemic Activation of the Complement and Coagulation Cascades

Abstract

OBJECTIVES:

To compare plasma proteomes of patients with confirmed fracture related infections (FRIs) matched to noninfected controls using liquid chromatography-mass spectrometry (LC-MS)

METHODS:

Design:

Prospective Case-Control Study

Setting:

Single, Academic, Level 1 Trauma Center

Patient Selection Criteria:

Patients meeting confirmatory FRI criteria were matched to controls without infection based on fracture region, age, and time after surgery from June 2019 to January 2022. Tandem Mass Tag LC-MS analysis of patient plasma samples was performed.

Outcome Measures and Comparisons:

Protein abundance ratios in plasma for FRI patients compared to matched controls without infection were calculated.

RESULTS:

Twenty-seven patients meeting confirmatory FRI criteria were matched to 27 controls. Abundance ratios for over 1,000 proteins were measured in the 54 plasma samples. Seventy-three proteins were found to be increased or decreased in FRI patients compared to the matched controls (unadjusted t-test p<0.05). Thirty-two of these proteins were found in all 54 patient samples and underwent subsequent principal component (PC) analysis (PCA) to reduce the dimensionality of the large proteomics data set. A three component PCA accounted for 45.7% of the variation in the data set and had 88.9% specificity for the diagnosis of FRI. STRING protein-protein interaction network analysis of these three PCs revealed activation of the complement and coagulation cascades via the Reactome pathway database (false discovery rates<0.05).

CONCLUSIONS:

Proteomic analyses of plasma from FRI patients demonstrates systemic activation of the complement and coagulation cascades. Further investigation along these lines may help to better understand the systemic response to FRI and improve diagnostic strategies using proteomics.

Introduction

The incidence of fracture related infection (FRI) varies widely depending on type, location, severity of fracture,¹ and the treatment of FRI is costly.² Financial burden is estimated to exceed $23,000 per infection.³ Despite the significant socioeconomic impact, our ability to accurately diagnose FRI remains hindered, especially prior to reoperation. The infection work-up is largely based upon the history and physical exam, white blood cell count (WBC), erythrocyte sedimentation rate (ESR), C-reactive protein (CRP), radiographs, and occasionally advanced imaging. However, WBC, ESR, and CRP have a limited predictive value for FRI as studies have shown both FRI and control patients have overlap in values for these inflammatory markers.^4–6 Quantitative histology and tissue culture from intra-operative tissue samples can be a very useful tool to diagnose FRI but are invasive, dependent on sample quality, and the results aren’t available until after the surgery has been performed. Having a definitive diagnosis of FRI prior to reoperation would aide surgeons in operative planning and patient counseling.

Technical advances in mass spectrometry sample preparation and instrumentation make this method a powerful tool for global unbiased profiling of proteomes in a variety of biofluids, including serum, plasma, and whole blood, for diagnostic purposes. Many systemic biomarkers are nonspecific, and their raised levels in peripheral blood may be caused by pathological or physiological changes besides infection. It was hypothesized that plasma proteomic profiling of patients with confirmed FRI would identify systemically activated biologic pathways, which may find eventual use as a diagnostic biomarker strategy. This hypothesis was tested by dimensionality reduction of the proteomics data using principal component (PC) analysis (PCA), followed by bioinformatics of protein-protein interaction network analysis using STRING and Reactome pathway database results.

Methods

Patient Selection

The study was approved by the local institutional review board (IRB), and all patients underwent informed consent. The FRI group includes patients 18 to 85 years old with an extremity, pelvic ring, or acetabulum fracture who were surgically treated with retained orthopaedic implant and went on to meet confirmatory FRI criteria.^1,7 FRI criteria consisted of confirmatory and suggestive categories.^1,8 Confirmatory criteria were the following: 1) fistula, sinus tract, or wound breakdown proximate to the fracture; 2) presence of pus at the fracture site; 3) visible microorganisms on histologic analysis (e.g., gram stain); and 4) phenotypically indistinguishable organisms identified from ≥2 deep tissue specimens from the fracture site. Suggestive criteria were as follows: 1) local or systemic signs of infection (e.g., redness, pain, swelling); 2) radiologic and/or nuclear imaging signs (e.g., failing implants, increased tracer uptake); 3) pathogen identified on only 1 deep tissue specimen sent for culture; 4) elevated serum inflammatory markers (i.e., ESR, WBC, CRP); 5) persistent, increasing, or new onset wound drainage; and 6) new onset joint effusion proximate to the fracture. Exclusion criteria were hand and spine fractures, pregnancy, prisoners, known immunosuppressive state, known presence of a separate source of infection, systemic infection, septic joint, undergoing second debridement or prior failure of infection treatment, pathologic fracture, definitive treatment including arthroplasty, definitive treatment using percutaneous Kirschner wires or external fixation pins only, and known venous thromboembolism.⁹ The control group was a set of patients who were infection free for ≥6 months after their fracture surgery. Controls and FRI patients were matched for age (±15 years), fracture region, and time after index surgery at which the FRI was diagnosed (±2 weeks). Fracture region was defined as upper extremity long bones (humerus, radius/ulna, and clavicle), lower extremity long bones (femur and tibia), or other lower extremity fractures (patella, ankle, and tarsal bones). Once an FRI case was identified, clinic schedules were searched for a patient that would match to the FRI case based on the above criteria. That patient was then approached, enrolled, and blood drawn at that time. Such a patient was then held as a putative matched control for the FRI case but did not become a fully matched control until they were at least 6 months post-operative without any suggestive or confirmed FRI criteria. When any patient was screened for study participation and approached for consent they were asked about any symptoms of current illness and underwent physical exam. Charts were also reviewed for any suggestion of alternative source of infection for one month prior to blood sample acquisition; however, our protocol did not require formal testing to rule out separate sources of infection.

Sample Collection

Blood samples were obtained from the FRI cohort at the time they were determined to meet FRI confirmatory clinical criteria or prior to reoperation if they met suggestive criteria¹⁰ and underwent surgery where the diagnosis of FRI was confirmed based on the confirmatory criteria numbers 2, 3, and/or 4 mentioned above. Blood samples for controls were obtained in the clinic at routine post-operative follow-up visits. 5mL of peripheral venous blood was collected in an EDTA added tube (BD Vacutainer, Becton). The tube was inverted 4–5 times allowing the blood to mix with the anticoagulant before it was centrifuged at 1500g for 10 minutes. Plasma was then extracted, aliquoted into 500 mL tubes, and stored at −80°C until batch analysis.

Mass Spectrometry Methods are Provided in Supplemental Digital Content 1.

Mass Spectrometry Data Analysis

Raw data were analyzed using Proteome Discoverer (v 2.5.0.400, Thermo Fisher Scientific^™). Sequest HT searches utilized a Homo sapiens UniProt reviewed FASTA and common contaminants (20417 total sequences) with full trypsin, a maximum number of 3 missed cleavages, precursor mass tolerance of 10 ppm, and fragment mass tolerance of 0.02 Da. Static modifications used for the search were carbamidomethylation (+52.021) on cysteine (C) residues and Tandem Mass Tag (TMT)duplex (+225.156) label on lysine (K) residues; Dynamic modifications used for the search were oxidation (+15.995) of methionines, deamidation (+0.984) of asparagine (N) and glutamine (Q) residues,TMT2plex label (+225.156) on the N-termini of peptides (max 3 dynamic mods). Dynamic protein terminus modifications allowed were: acetylation (N-terminus), Met-loss, or Met-loss plus acetylation (N-terminus). Percolator False Discovery Rate (FDR) with concatenated target/decoy selection and q-value based validation was set to a strict setting of 0.01 and a relaxed setting of 0.05. In the consensus workflow, the Protein FDR validator in the consensus was set to a strict 1% protein FDR cutoff and relaxed 5% protein FDR cutoff. Quantification was performed using intensity values of the TMTduplex reporter ions with lot-specific isotopic impurity corrections, a co-isolation threshold of 50%, and average reporter ion S/N cutoffs of five. Abundances were normalized to the total peptide amount; all peptides were used for normalization and modified peptides were excluded for pairwise ratios. Protein abundance-based protein ratios were calculated (protein ratios directly calculated from grouped protein abundances) with a maximum fold change of 100, no imputation and a nested study design (ratios calculated for each set of TMTduplex) was used with ANOVA (Individual proteins) hypothesis test to calculate p-values for protein abundance changes. Protein grouping applied strict parsimony principle to determine master protein accession numbers. PCA of normalized protein abundances indicated grouping by biological sample suggesting high analytical reproducibility. Normalized abundance values for each sample type, abundance ratios, log2(abundance ratio) values, and respective p-values from Proteome Discover^™ were exported to Microsoft Excel.

Select Plasma Protein Measurement by ELISA^11,12

Four proteins with unadjusted p<0.05 abundance ratios between FRI and control groups from the protein LC-MS analysis were selected to undergo ELISA testing. The selected proteins were Complement Factor H, Complement Factor B, Serum Amyloid A-1, and Platelet Factor 4, and all had readily available commercial assay kits. Isolated species from the Human Complement Panel 2 (Cat. No. HCMP2MAG-19K), Human CVD Panel 2 (Cat. No. HCVD2MAG-67K), and the Human CVD Panel 3 Acute Phase (Cat. No. HCVD3MAG-67K) 48 Plex (EMD Millipore Corporation, Burlington, MA) were used for protein multiplex ELISA using the Luminex xMAP technology (Luminex, Austin, TX) according to manufacturer’s protocol. Samples were thawed at room temperature (22°C) and spun at 10,000g for 10 minutes. All samples were run in duplicates. For samples with biomarker concentrations that were undetectable, ½ x lowest detectable value was used for analysis.¹³ Samples exceeding maximal detectable values were diluted to fall with the assay range, remeasured, and corrected for dilution. Due to insufficient sample left after proteomics analysis, matched pair 4 was eliminated from ELISA testing. CRP levels from all samples were measured by the hospital laboratory with a value >1 mg/dL considered elevated.

Statistical and Biologic Pathway Analyses

Demographic characteristics were compared between FRI and control groups using t-tests for continuous variables and Chi-square test for categorical variables. PC and discriminant analyses were performed to determine how the protein data clustered together and were able to differentiate between groups. These analyses were performed with 32 of the 73 original proteins with unadjusted p<0.05 abundance ratios between FRI and control groups from the protein LC-MS analysis, due to missing information on most proteins. Although we did check for data being missing completely at random, the data were not, and in either case, we used the analyses proposed by Yi and Latch to only use data with no missingness to avoid imputation.¹⁴ This is not surprising given the nature of TMT-based quantitative proteomics where ‘missing values’ are more likely to occur in different multiplexes than within the same multiplex. The components given by these analyses were then used to create algebraic linear combinations to graphically represent the component classifications. All analyses were performed using SAS v9.4 (SAS Institute, Cary, NC). All analytic assumptions were verified. Finally, the proteins corresponding to PC1, PC2, and PC3 were each entered into STRING (https://string-db.org/) for pathway analysis and their Reactome database (https://reactome.org/) results were recorded.

Results

Demographic Characteristics

The time period of blood sample collection was June 2019 to January 2022. No patient had documented COVID within one month leading up to, or for three months after, collection. Twenty-seven patients met confirmatory FRI criteria, having either a fistula/draining sinus/wound breakdown (n=15), purulent drainage (n=23), and/or two or more positive intraoperative cultures (n=20).⁷ These patients were matched with 27 patients free of infection for at least six months following the index surgery as described above. No control patient exhibited any suggestive or confirmatory criteria for infection at any point in the study period, and none had returned for infection at a minimum of 1 year and 10 months post-operative. Table, Supplemental Digital Content 2 summarizes demographic and clinical data for both cohorts. There was no significant difference in terms of age, sex, BMI, fracture region, category of implant used, and weeks post-operation at which blood samples were collected (p>0.05). The average standard deviation of the absolute value of pair age difference was 5.4 ± 4.9 years. Thus, although up to 15 years age difference was allowed in the matching process, the actual age difference was substantially less. There were more open fractures in the FRI cohort compared to the control (48.2% vs. 3.7%, p<0.001). There was no statistically significant difference between cohorts for a diagnosis of diabetes mellitus or use of tobacco, alcohol, or non-steroidal anti-inflammatory drugs (p>0.05).

Protein Identification

Plasma samples of the 27 matched pairs were analyzed using TMT LC-MS proteomics. Across the samples, 1,058 protein groups (1,505 proteins), and 15,658 peptide groups were identified. Further analysis revealed the presence of 73 differentially abundant proteins between FRI and matched controls (unadjusted p<0.05) (Figure 1), meaning the protein species was found to a greater or lesser amount in the FRI patients compared to controls. Of the 73, 32 were present in all 54 samples, with 24 found at higher levels in FRIs compared to controls versus eight found at lower levels in FRIs compared to controls. Table 1 summarizes these 32 distinct proteins, their genes, and their location on Figure 1.

Figure 1.

Volcano plot of all unique protein groups identified. Proteins significantly increased in FRI patients are found in the red shaded area, while proteins significantly decreased are found in green shaded area. X-axis [Log₂(abundance ratio)] and Y-axis [−Log₁₀(p-value)]. To aide interpretation, −Log₁₀(0.1) = 1, −Log₁₀(0.05) = 1.3, −Log₁₀(0.01) = 2, and −Log₁₀(0.001) = 3. Also, Log₂(1) = 0, meaning the ratio of FRI to control abundance was 1, or no difference in the plasma value for that protein between groups. A 2-, 4-, or 8-fold difference is 1, 2, or 3, respectively, on the x-axis, while a 0.5-, 0.25-, or 0.125-fold difference is a −1, −2, or −3, respectively, on the x-axis.

Table 1:

Thirty-two Distinct Proteins Present in All Samples and their Genes

Protein Name	UniProt ID	Gene	Log2(Ratio)	−Log10(p-value)	PC*
Alpha-1-acid glycoprotein 1	P02763	ORM1	0.32	1.57	1
Serum amyloid A-2 protein	P0DJI9	SAA2	2.45	2.62	1
Leucine-rich alpha-2-glycoprotein	P02750	LRG1	0.52	2.16	1
Complement factor H-related protein 5	Q9BXR6	CFHR5	0.36	3.32	1
C-reactive protein	P02741	CRP	1.74	2.35	1
Complement component C9	P02748	C9	0.2	1.5	1
Haptoglobin	P00738	HP	0.69	1.78	1
Serum amyloid A-1 protein	P0DJI8	SAA1	2.01	1.89	1
Lipopolysaccharide-binding protein	P18428	LBP	0.36	1.72	1
Peptidyl-prolyl cis-trans isomerase A	P62937	PPIA	0.57	1.56	-
Calmodulin-3	P0DP25	CALM3	0.42	1.45	-
Fructose-bisphosphate aldolase A	P04075	ALDOA	0.41	1.77	2
Complement factor B	P00751	CFB	0.2	1.5	1
Peroxiredoxin-2	P32119	PRDX2	0.54	1.42	1
Mannosyl-oligosaccharide 1,2-alpha-mannosidase IA	P33908	MAN1A1	0.12	2.43	2
Alpha-1-acid glycoprotein 2	P19652	ORM2	0.39	1.97	3
Transgelin-2	P37802	TAGLN2	0.27	1.33	1
Complement C1r subcomponent	P00736	C1R	0.09	1.77	2
Retinoic acid receptor responder protein 2	Q99969	RARRES2	0.28	2.34	2
Ceruloplasmin	P00450	CP	0.17	1.72	2
Complement C1s subcomponent	P09871	C1S	0.06	1.58	2
Carboxypeptidase N catalytic chain	P15169	CPN1	0.17	1.55	-
Alpha-2-macroglobulin	P01023	A2M	−0.29	1.72	3
Coagulation factor X	P00742	F10	0.09	1.73	3
Insulin-like growth factor-binding protein 4	P22692	IGFBP4	0.13	1.35	3
Selenoprotein P (SEPP1)	P49908	SELENOP	−0.17	1.35	3
Fibronectin	P02751	FN1	−0.15	1.34	-
Cell surface glycoprotein MUC18	P43121	MCAM	−0.06	1.39	1
Hepatocyte growth factor activator	Q04756	HGFAC	−0.1	1.31	-
Apolipoprotein A-IV	P06727	APOA4	−0.36	1.9	1
Plasma kallikrein	P03952	KLKB1	−0.16	1.99	3
Afamin	P43652	AFM	−0.17	2.3	1

^*PC = principal component. Five proteins did not fall into the first three PCs and are noted by the – in this column.

The values in columns Log2(Ratio) and −Log10(p-value) serve as coordinates for referencing on the Volcano Plot (see Figure 1). For example, the dot highest on the Y-axis has a −Log10(p-value) = 3.32 and Log2(Ratio) = 0.36, which corresponds to Complement factor H-related protein 5.

The 32 proteins subsequently underwent PCA (Figure 2) to reduce the dimensionality of the proteomics data set. A 3-factor PCA accounted for 45.7% of variation in the data for categorizing FRI versus control, having a specificity of 88.9% and a sensitivity of 35.2% for the diagnosis of FRI. Consistent with the proteomics results, ELISA testing of Complement Factor B and Serum Amyloid A-1 were significantly different between the FRI and control groups, while Complement Factor H (note, this is a separate protein species from Complement Factor H-related protein; Table, Supplemental Digital Content 3) was not statistically different. Platelet Factor 4 was discordant between proteomics and ELISA results (see Table, Supplemental Digital Content 3) when comparing FRI patients to controls, being different in the proteomic analysis (unadjusted p=0.05), but not statistically significant in the ELISA analysis (p=0.82). Of note, though Platelet Factor 4 was one of the 73 significantly different proteins on the volcano plot (Figure 1), it was not used in the PC, STRING, or Reactome analyses given it was not present in all proteomic samples.

Figure 2.

Three-factor PCA accounting for ~46% of the variation. This plot visually displays the 3-factor PCA demonstrating spatial differentiation of FRI cases from controls. Red squares represent FRIs, and green circles represent Controls. Note the general separation of red to the lower left of the plot and green to the upper right.

With respect to quantitative CRP measurement, the lower limit the hospital reports out is 0.5 mg/dL. Anything less than 0.5 mg/dL was made equal to 0.5 mg/dL. Mean ± standard deviation for CRP was 4.6 ± 6.4 mg/dL and 1.5 ± 2.3 mg/dL for FRI and control groups, respectively (p=0.02, two-sided paired t-test). Using the hospital threshold of >1 mg/dL as elevated, the specificity of CRP for FRI from the hospital lab was 74.1% and sensitivity 51.9%.

The STRING protein network analysis revealed Reactome pathways suggesting that both the complement and coagulation cascades were systemically activated in plasma from FRI patients compared to controls. Figure, Supplemental Digital Content 4 demonstrates the STRING protein-protein interaction networks of all 32 proteins amongst each other. Further inspection of the relationships indicated that certain proteins corresponded to identifiable biological processes. Using the 3-factor PCA, STRING network analysis of each of the 3 PCs was performed. PC1’s (Figure 3A) and PC2’s (Figure 3B) STRING networks indicate activation of the complement pathway, while PC3’s (Figure 3C) STRING network is associated with the coagulation cascade. Table 2 summarizes the Reactome database pathways associated to each of the PC STRING analyses. The collective results identify a systemic response to the localized FRI.

Figure 3A.

STRING protein-protein interaction network for PC1 proteins.

Figure 3B.

STRING protein-protein interaction network for PC2 proteins.

Figure 3C.

STRING protein-protein interaction network for PC3 proteins.

Table 2:

Reactome Pathways for Each Principal Component (PC)

PC *	Reactome Pathway	Description	FDR **
1	HSA-168249	Innate Immune System	2.51E-05
	HSA-166658	Complement cascade	0.00012
	HSA-977606	Regulation of Complement cascade	0.0038
2	HSA-173623	Classical antibody-mediated complement activation	0.0061
	HSA-977606	Regulation of Complement cascade	0.0497
3	HSA-140837	Intrinsic Pathway of Fibrin Clot Formation	9.02E-05
	HSA-109582	Hemostasis	0.00018
	HSA-114608	Platelet degranulation	0.0031
	HSA-9651496	Defects of contact activation system (CAS) and kallikrein/kinin system (KKS)	0.0043

^*Website for PC1 String network analysis https://version-11-5.string-db.org/cgi/network?networkId=bH0Yb0A5LfO9

Website for PC2 String network analysis https://version-11-5.string-db.org/cgi/network?networkId=bZJpnGfHjh9g

Website for PC3 String network analysis https://version-11-5.string-db.org/cgi/network?networkId=byaBCc0VATQv

^**False discovery rate

Discussion

This study builds on an initial pilot study looking at differences between FRI patients and controls based on ELISA protein measurements⁹ and further delves into proteomics changes within the plasma of patients with FRI. The main finding from this study was systemic activation of the complement and coagulation cascades in patients with FRI. This finding was determined using unbiased global quantitative TMT based LC-MS/MS proteomic profiling approach coupled to bioinformatics via STRING protein-protein interactions network that identified biologic pathways from the Reactome database. While this activation is known for other bacterial infections^15–17 and COVID-19,^18,19 it has not previously been reported for localized FRI. This study serves as a useful launching point to better understand the systemic manifestations for a localized FRI and may provide insight to potential biomarkers for improving on the diagnosis of FRI. Now that it has been demonstrated that a systemic proteomic response can be measured in patients with FRI, future research should aim to correlate the response with different infecting organisms and individual confirmatory criteria to better understand infection severity. Also, investigation of the proteomic response in patients with suggestive criteria may be informative to identify whether they will progress to confirmatory criteria.

Seventy-three differentially abundant proteins were identified between FRI and control patients, of which 32 were present in all samples. Many of the proteins have known activity with respect to inflammation or associations to other musculoskeletal/inflammatory/infectious diseases. Alpha-1-acid glycoprotein 1 (AGP) also known as orosomucoid (ORM) is an acute phase reactant that increases with systemic injury that has been associated with maintaining capillary permeability.²⁰ Serum amyloid A-2 (SAA), on the other hand, is associated with a rise in neutrophils.²¹ Leucine-rich alpha-2-glycoprotien (LRG) is also an acute phase reactant that has recently been found to be an alternative to CRP for assessing for rheumatoid arthritis inflammatory bowel disease that is generally activated when there is an increase in tumor necrosis factor (TNF)-a, interleukin (IL)-1b, IL-6, and IL-22.²² Haptoglobin and C-reactive protein are two major acute phase reactants that signify an inflammatory process within the body. When an immune response is mounted, both haptoglobin and CRP are significantly increased. They have also been associated with avascular necrosis of the femoral head.²³ Complement factor H-related protein 5 (CFHR-5) is one of the regulators of the complement pathway, part of the host defense that forms the membrane attack complex (MAC) and eventually ends in cell membrane damage.²⁴ CFHR-5 is specifically involved in controlling the binding and cleaving of C3b. CFHR-5 has been associated with complement-related nephropathies and glomerulopathies. Complement component C9 is further downstream in the complement cascade than CFHR-5 and is directly linked to MAC formation. Lack of C9 is associated with increased Neisseria meningiditis infections along with other encapsulated bacteria.²⁵

In the field of orthopaedics, a recent study by Chen et al.²⁶ performed similar protein analysis in patients with known prosthetic joint infection (PJI). The authors compared five patients with PJI to five with aseptic failure and found 435 differentially abundant proteins using label-free LC-MS, which demonstrated activation of not only the complement and coagulation cascades, but also phagocytosis and neutrophil recruitment as well. Of note, the samples in the Chen et al.²⁶ study were of bone. Of the proteins found in their analysis, complement factor C3, gamma-glutamyl hydrolase, alpha-1 acid glycoprotein 2, integrin beta-2, actin-related protein 2, matrix metalloproteinase-9, myeloperoxidase, integrin alpha-M, and haptoglobin were significantly upregulated in patients with PJI. The overlap between the proteins found by Chen et al.²⁶ and those in the current study support the use of proteomic analysis in the identification of biomarkers for FRI.

Another study from 2015 by Liu et al.²⁷ aimed to determine changes in serum proteomics of patients with Legg-Calve-Perthes disease (LCPD). Serum samples from age- and gender-matched pairs (n=20) of patients with LCPD and controls were analyzed using the isobaric tags for relative and absolute quantification (iTRAQ) technique. Significantly differentially abundant proteins found in their analysis included complement factor H, S100-A8, alpha-1-acid glycoprotein 1, haptoglobin, and apolipoprotein E. Of these proteins, complement factor H, alpha-1-acid glycoprotein 1, and haptoglobin were also significantly changed in this study between FRIs and the control group.

Multivariate analysis techniques can be used to analyze large datasets and differentiate between cohorts. One such technique that can reduce redundancy and represent the dynamics within the dataset is PCA.^28–30 PCA takes a multivariate dataset and linearizes it while preserving the information within. The PCs are created as new variables to allow preservation of variance within the dataset.³¹ The ClustVis web tool was used to statistically differentiate between proteins present in FRI and control plasma samples using a 3-factor PCA and identified significant protein-protein interactions within specific PCs. However, to interpret the biologic interactions amongst the proteins, databases such as STRING and Reactome are necessary.³² STRING can be used to interpret datasets and find “functional associations” amongst the proteins, and Reactome pathway analysis can describe common functions those proteins have. The use of STRING protein-protein interaction networks and Reactome database pathway analyses are limited in clinical orthopaedic trauma research; however, they are well-established in other fields.^33,34 Using these freely available bioinformatics platforms, this study identified the complement and coagulation pathways as being systemically activated in response to localized FRI. The use of proteomics and bioinformatics tools such as these will serve to advance understanding of the sequalae of orthopaedic trauma (e.g., FRI and nonunion). Further, they may be useful in biomarker identification.

The main limitation of this study was the small sample size and the unprecedented nature of the mass spectrometry analysis in FRI patients. As such, p-values to determine differences in protein species from LC-MS were not adjusted for multiple comparisons. To mitigate this concern, the proteomics results for several proteins were validated with ELISA and hospital-based CRP measurement. Future work will require validation of these results in larger data sets with more restrictive statistical approaches.³⁵ A well-known challenge of protein quantification from biofluids such as serum and plasma is the large dynamic range (see Table, Supplemental Digital Content 3 for ELISA-based plasma concentration range of four proteins identified in the LC-MS semi-quantitative analysis). Although CRP was identified in this and a prior experiment as significantly different between FRI and control, platelet derived growth factor (PDGF)-AB/BB and monokine induced by gamma interferon (MIG) were not detected in this mass spectrometry experiment. These proteins are not commonly identified in unbiased, discovery mass spectrometry of biofluids.^36,37 Even with depletion of abundant proteins, not all proteins will be detectable using typical data dependent mass spectrometry, and the results of this study are consistent with previous TMT-based quantitation of plasma proteins.^36–39 The field of mass spectrometry has seen significant advances in the use of data independent acquisition (DIA) based quantitation to improve the numbers of protein identifications in biofluids, and it is expected that application of this methodology will be useful in follow up studies.^36,40–42 In addition, several antibody- or aptamer-based proteomics methodologies, including Olink^® and Somascan^®, have been developed in recent years which are highly complementary to mass spectrometry-based proteomics.^43–45 Further, there were significantly more open fractures in the FRI cohort, and definitive identification of how much of the change in proteomic signature between FRI and controls is due to the added soft tissue injury of open fracture could not be determined. Finally, although inclusion and matching criteria were both strict, it is possible that there were some potential confounding variables that resulted in elevated inflammatory biomarkers. Formal testing for other infectious etiology was not required as part of the current study’s protocol, which instead relied on patient report of symptoms, physical exam, and medical record review to ascertain if there was known presence of a separate source of infection.

In summary, this study demonstrated systemic activation of the complement and coagulation cascades in patients with FRI using LC-MS proteomics and bioinformatics analyses through STRING protein-protein network and Reactome pathway database analyses. These data shed insight into the systemic response to localized FRI and provide evidence that proteomics may eventually be used to identify a biomarker signature for FRI diagnosis.