The mononuclear phagocyte system obscures the accurate diagnosis of infected joint replacements

Abstract

Introduction: Diagnosing infected joint replacements relies heavily on assessing the neutrophil response to bacteria. Bacteria form biofilms on joint replacements. Biofilms are sessile bacterial communities encased in a protective extracellular matrix, making them notoriously difficult to culture, remarkably tolerant to antibiotics, and able to evade phagocytosis. Phagocytized bacteria dramatically alter cytokine production and compromise macrophage antigen presentation. We hypothesize that a subset of joint replacements have a dormant infection that suppresses the neutrophil response to bacteria but can be distinguished from uninfected joint replacements by the response of the mononuclear phagocyte system (MPS) within periarticular tissue, synovial fluid, and circulating plasma.

Methods: Single cell RNASeq transcriptomic and OLink proteomic profiling was performed on matched whole blood, synovial fluid, and periarticular tissue samples collected from 4 joint replacements with an active infection and 3 joint replacements without infection as well as 6 joint replacements with a prior infection deemed "infection-free" by the 2018 Musculoskeletal Infection Society criteria (follow-up of 26 ± 3 months).

Results: The MPS and neutrophil responses differ by infected state; the cellular distribution of the MPS response in the subset of joints with dormant infections resembled actively infected joints (p = 0.843, Chi-square test) but was significantly different from uninfected joints (p < 0.001, Chi-square test) despite the absence of systemic acute phase reactants and recruitment of neutrophils (p < 0.001, t-test). When compared to no infection, the cellular composition of dormant infection was distinct. There was reduction in classically activated M1 macrophages (p < 0.001, Fischer's test) and alternatively activated M2 macrophages coupled with an increase in classical monocytes (p < 0.001, Fischer's test), myeloid dendritic cells (p < 0.001, Fischer's test), regulatory T-cells (p < 0.001, Fischer's test), natural killer cells (p = 0.009, Fischer's test), and plasmacytoid dendritic cells (p = 0.005, Fischer's test). Hierarchical cluster analysis and single-cell gene expression revealed that classically M1 and alternatively M2 activated macrophages as well as myeloid dendritic cells can independently distinguish the dormant and uninfected patient populations suggesting that a process that modulates neutrophil recruitment (C1QA, C1QB, LY86, SELL, CXCL5, CCL20, CD14, ITGAM), macrophage polarization (FOSB, JUN), immune checkpoint regulation (IFITM2, IFITM3, CST7, THBS1), and T-cell response (VISIG4, CD28, FYN, LAT2, FCGR3A, CD52) was occurring during dormant infection. Gene set variation analysis suggested that activation of the TNF (FDR < 0.01) and IL17 (FDR < 0.01) pathways may distinguish dormant infections from the active and uninfected populations, while an inactivation of neutrophil extracellular traps (NETs) may be involved in the lack of a clinical response to a dormant infection using established diagnostic criteria. Synovial inflammatory proteomics show an increase in synovial CXCL5 associated with dormant infection (p = 0.011, t-test), suggesting the establishment of a chronic inflammatory state by the MPS during a dormant infection involved in neutrophil inhibition. Plasma inflammatory proteomics also support a chronic inflammatory state (EGF, GZMN, FGF2, PTN, MMP12) during dormant infection that involves a reduction in neutrophil recruitment (CXCL5, p = 0.006, t-test), antigen presentation (LAMP3, p = 0.047, t-test), and T-cell function (CD28, p = 0.045, t-test; CD70, p = 0.002, t-test) that are also seen during the development of bacterial tolerance.

Discussion: All current diagnostic criteria assume each patient can mount the same neutrophil response to an implant-associated infection. However, the state of the MPS is of critical importance to accurate diagnosis of an implant-associated infection. A reduction in neutrophil recruitment and function mediated by the MPS may allow joint replacements with a dormant infection to be mischaracterized as uninfected, thus limiting the prognostic capabilities of all current diagnostic tests.

Fig. 1

A The local and systemic polymorphonuclear neutrophil (PMN) response is used to diagnose active infection and drive surgical management. B The PMN response is part of a more complex innate immune response to bacteria regulated by the mononuclear phagocyte system (MPS). We hypothesize that a subset of joint replacements harbors a dormant infection that can be distinguished from uninfected joint replacements by the MPS response to bacteria

Fig. 2

The Musculoskeletal Infection Society (MSIS) criteria were used to subdivide clinical tissue samples from patients with (red) and without (blue) an active infection (top panel). The uninfected samples included cases with and without a previous infection and are both marked in blue. Single-cell transcriptomic expression from synovial tissue was used to re-characterize each clinical tissue sample (bottom panel). Samples previously deemed “infection-free” after a prior infection using the polymorphonuclear neutrophil (PMN) response, were able to be subdivided into two subpopulations based on their immunologic phenotype. One group was statistically similar to the uninfected cases (p = 0.902, Chi-squared) and remains marked in blue, while another group, dormant infections, was statistically similar to an active infection (green, p = 0.843, Chi-squared) and is now marked in green

Fig. 3

Hierarchical clustering of the single-cell gene expression from classically activated M1 and alternatively M2 activated macrophages as well as myeloid dendritic cells each independently distinguish the uninfected tissue samples (blue circle) from the infected (red circle) and dormant infection (green) samples for which the 95% confidence intervals tend to overlap demonstrating their similarity with respect to principal components (PC) 1 and 2. Loadings plot is overlayed to visualize features defining PC1 and PC2 and directionality relative to the localization of uninfected, infected, and dormant infection groups along these components

Fig. 4

Volcano plot highlighting the significant transcriptomic changes (red dot with the top 20 transcripts labeled and the remainder of the significant transcripts are listed in Supplemental Tables 1–3; the green and grey dots represent transcripts that did not reach significance) in A classically activate M1 macrophages, B alternatively M2 activated macrophages, C myeloid dendritic cells, and D in the tissue via pseudobulk analysis. Arrows indicate genes that exceeded the presented axes and are listed in Supplemental Tables 1 and 2

Fig. 5

M1 (top panel) and M2 (bottom panel) polarization of M1 macrophages, M2 macrophages, and myeloid dendritic cells displayed as median and interquartile range in dormant infection (green), active infection (red), and uninfected samples (blue). The symbols in the plots represent statistical significance as follows: non-significant (ns), FDR < 0.001 (*), FDR < 0.0001 (**), and FDR < 0.0001 (***)

Fig. 6

A The Venn diagram looks at unique and overlapping gene expressions after bulk transcriptomic analysis, comparing dormant infection (green) to active (red) and no infection (blue). B Volcano plot highlighting the transcriptomic changes of the 13 unique (white boxes) and 1 overlapping (grey box) genes expressed during dormant infection (red dot with transcript labeled; the green and grey dots represent transcripts that did not reach significance) when compared to samples with and without an infection

Fig. 7

Heatmap of gene set variation analysis (GSVA) using gene ontology (GO) terms and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways comparing dormant infection to samples A with and B without an infection based on log₂ fold change (FC) as well as false discovery rate (FDR). C) Hieratical clustering of GSVA demonstrating comparing dormant infection to active infection (left panel) supports their similarity yielding a normally distributed inflammatory response to infection. However, comparing dormant infection to uninfected controls (right panel) suggest the emergence of a distinct immune signature during dormant infection

Fig. 8

Statistically significant (Welch two-tailed t-test) proteomic changes displayed as median and interquartile range of normalized protein expression (NPX, in log₂ units) as measured by proximity extension assay on in A) synovial fluid and B) plasma comparing dormant infection (green) to uninfected samples (blue)

Fig. 9

A Schematic understanding of dormant infection (green) as an MPS-mediated IL17 response to biofilm inhibiting PMN recruitment and function when compared to patients with (red) or without (blue) and active infection. B Schematic understanding of how focusing on the MPS-mediated response as a biomarker of dormant infection (green) enables the screening, diagnosis, and prognosis of dormant infection and identifies a novel paradigm of immuno-prophylactic and immuno-modulation