Evaluation of global gene expression in regenerate tissues during Masquelet treatment

Abstract

The Masquelet induced-membrane technique is indicated for large segmental bone defects. Attributes of the induced-membrane and local milieu that contribute to graft-to-bone union are unknown. Using a rat model, we compared global gene expression profiles in critically-sized femoral osteotomies managed using a cement spacer as per Masquelet to those left empty. At the end of the experiment, induced-membrane and bone adjacent to the spacer were collected from the Masquelet side. Non-union tissue in the defect and bone next to the empty defect were collected from the contralateral side. Tissues were subjected to RNA isolation, sequencing and differential expression analysis. Cell Type Enrichment Analysis suggested the induced-membrane and the bone next to the PMMA spacer were comparatively enriched for osteoblastic genes. The non-union environment was comparatively enriched for innate and adaptive immune cell markers, but only macrophages were evident in the Masquelet context. iPathway Guide was utilized to identify cell signaling pathways and protein interaction networks enriched in the Masquelet environment. For induced-membrane vs. non-union, false-discovery rate correction of p-values rendered overall pathway differences non-significant, and so only protein interaction networks are presented. For the bone comparison, substantial enrichment of pathways and networks known to contribute to osteogenic mechanisms were revealed. Our results suggest that the PMMA spacer affects the cut bone ends that are in contact with it and at the same time induces the foreign body reaction and formation of the induced-membrane. B-cells in the empty defect suggest a chronic inflammatory response to a large segmental osteotomy.

Introduction

Successful repair of critically sized (>5 cm) segmental bone injuries and established non-unions remains a significant surgical challenge ¹. Available treatment options incur substantial patient burden due to the need for multiple, staged surgeries and the length of time required for successful healing. A promising treatment option is the Masquelet induced-membrane technique ². This procedure involves two surgeries. First, the segmental defect is debrided of contamination and non-viable tissue. A PMMA cement spacer is then placed into the defect. Over the next month, a membrane forms around the cement spacer via the foreign body reaction 3. When medically and surgically appropriate, the patient is operated on a second time. The induced-membrane is carefully opened, the spacer is removed, the membrane is filled with bone graft and repaired such that it encases the graft material. Over 4–12 months, the graft consolidates and becomes functionally integrated with the patient’s skeleton^{4; 5}.

Despite relatively high union rates ^{5; 6} and increasing popularity of the Masquelet technique among surgeons, it is still a staged-surgical procedure with significant variation. For example, emerging evidence suggests impaired membrane development might contribute to Masquelet failure⁷, but optimum biological characteristics of the induced-membrane and how they might impact consolidation at the second stage are not well understood. In rodent models of the Masquelet technique, researchers allow the membrane to develop for one month because early evidence suggested that induced-membrane BMP2 and TGFβ content peak at this time ⁸. Allowing the membrane to develop for longer intervals leads to ECM accumulation and thickening of the membrane in human patients ³. The negative implications of a fibrous induced-membrane are not well-described. While the membrane_does serve as a barrier to contain and protect bone graft within the osteotomy site, efforts to replace the induced-membrane with synthetic barrier membranes have not been fruitful ⁹. An induced-membrane that forms around a spacer placed into a bone defect contains osteoblasts and osteoclasts, thus reflecting the local environment ^10–13. Conversely, the induced-membrane that develops around a subcutaneously placed spacer does not contain osteoblasts or osteoclasts ¹⁴. This observation suggests that the cells present in bone contribute to the ability of the Masquelet induced-membrane to promote engraftment.

The purpose of this study was to assess the global gene expression landscape and cellular make up of osteotomies treated using Masquelet to those of a critical sized osteotomies left empty. In addition to the induced-membrane and non-union tissue, we also compared bone that was proximal to the PMMA spacer during membrane development to that next to the empty osteotomy and developing non-union. We hypothesized that the PMMA spacer would elicit local changes in gene expression that were distinct from those associated with a developing non-union. We used an unbiased RNA sequencing approach to address the fact that very little is known about the composition of an induced membrane that develops in a skeletal site. Our results suggest that the induced-membrane contains osteoblast lineage cells, osteoclasts and macrophages. The non-union tissue contained a variety of cells indicative of innate and adaptive immune responses to a chronic segmental bone defect. In addition, the PMMA spacer itself evoked changes in gene expression consistent with activation of osteogenic signaling pathways and ECM formation in adjacent bone cells.

Methods

Masquelet Surgery

The University of Michigan Institutional Animal Care and Use Committee (IACUC) approved all procedures and all animals were treated humanely according to national guidelines. We adapted a critical sized femoral defect model previously established in our lab ¹⁵ into a two-stage procedure similar to published rat Masquelet protocols ^14,16. Our model employs 6-month old male retired breeder Sprague-Dawley rats weighing 460–540 grams (Envigo, Indianapolis, IN), and at this age a 5 mm osteotomy progresses to a non-union in virtually all cases ¹⁵. Sample size was selected based on the UM bioinformatics Core recommendation that we use a minimum of 4 biological replicates per group for RNAseq analysis. Surgeries (N=5 rats) were performed in the morning. Anesthesia was obtained using isoflurane. The femur was stabilized with a custom designed PEEK fracture fixation plate containing a 5 mm notch that serves as a guide for creating the osteotomy. Under continuous irrigation, a Hall Micro 100 oscillating saw fit with 2 blades spaced 5 mm apart was used to make the osteotomy. For the Masquelet side, a pre-formed polymethyl methacrylate spacer (PMMA Palacos-R+G (Zimmer-Biomet, Warsaw Indiana)) was placed into the defect. On the control side, the defect was left empty. With each surgery, we alternated placing the spacer into the right or left limb. Both limbs were stabilized with PEEK plates. Buprenorphine (0.05 mg/kg) was given pre-operatively and again 8 hours later. Carprofen (5 mg/kg) was given immediately post-operatively and again 24 hours later. Rats were housed in pairs under SPF conditions with a 12-hour light/dark cycle and they were monitored twice daily for 72 hours following the surgery and then daily for the 4-week healing period. Climbing implements were removed from the cages in order to prevent injury to the healing limbs.

Fracture fixation was maintained and no adverse consequences were noted during the study. All 5 rats were euthanized 4 weeks after surgery. The induced-membrane and the wound-bed tissue on the empty defect side (non-union tissue), as well as bone just proximal to the PMMA spacer and just proximal to the empty defect were collected into RNAlater. To collect the tissue, the whole femur was dissected with most of the muscle attached. Tissue was cleaned from the PEEK plate and the femur was rotated in order to visualize the defect set below the notch in the plate. A thin, pink membrane was evident around the PMMA spacer. This induced-membrane was collected by gently elevating it above the spacer using a fine forceps and cutting it away with a #15 scalpel blade. We avoided the muscle below the defect on the medial side of the femur. Similarly, the fibrous tissue that had formed between the cut bone ends was removed using forceps and a thin metal spatula. There was also fluid that had accumulated in the empty defect and this was not avoided. Approximately 1 mm of bone just proximal to the defect or the PMMA spacer was cut with a #10 scalpel blade.

Global gene expression analysis using RNA sequencing

IM and non-union tissue, as well as bone next to the PMMA and next to the empty defect were weighed, homogenized in Trizol and stored at -80°C before extraction of RNA and DNase treatment, as we have described previously ¹⁷. The RNA was subject to ammonium acetate precipitation to remove residual Trizol. RNA from 4 of the 5 animals was of sufficient quantity for RNA sequencing and these were submitted to the NextGen sequencing core at the University of Michigan for quality analysis. Table S1 summarizes RNA metrics for each tissue sample. PolyA-selected mRNA libraries were generated and 50 base, paired-end reads were conducted using the Illumina HS 4000 Platform. Data processing and analysis were conducted by the University of Michigan Bioinformatics core. Raw sequences were subject to an initial quality control protocol using FastQC v 0.113. Sequences were aligned to the Sprague-Dawley reference sequence and transcripts expressed differentially between induced-membrane vs. non-union tissue and bone next to PMMA vs. bone next to an empty defect were identified. Deseq2 (Bioconductor ¹⁸) was used for expression quantitation, normalization, and differential expression analysis. Genes with both fold-change ± 1.5 and an FDR corrected p value ≤ 0.05 were considered significantly different.

Cell type enrichment analysis

To estimate the cellular composition of each tissue type, we used CTen ¹⁹ (http://www.influenzax.org/~jshoemaker/cten). All genes detected by RNAseq were separated into 3 groups: those with increased expression in the Masquelet environment, those with decreased expression in the Masquelet environment, and those with similar expression levels in the Masquelet and non-union environments. The lists were imported into CTen to determine the dominant cell types enriched in each tissue type ²⁰. To visualize the results, CTen enrichment scores (-log10(FDR corrected p-values)), were imported into Morpheus (https://software.broadinstitute.org/morpheus) to generate heat maps for each biologically relevant cell type identified. On the color scale, green indicates a particular cell type is relatively highly enriched in that tissue (figure 2). As per CTen recommendations, cell types with enrichment scores ≤2 were considered not present. In addition, since the CTen database contains global gene expression profiles from immune cells subjected to a variety of treatments, we present categories of immune cells (e.g. B cells). Overlap between our gene expression profiles and those of irrelevant cell lines or tissue-types are not presented.

Figure 2. The Masquelet and non-union healing environments have distinct cellular compositions.

Heat maps of CTen enrichment scores for contextually relevant cells and tissues. Results from regenerate tissues (A) and from adjacent bone (B) are shown. Induced-membrane and bone next to the PMMA spacer were enriched for osteoblastic genes (blue boxes). Non-union tissue and bone next to the empty defect were enriched for a variety of immune cell markers (dashed boxes). The only immune cell genes significantly enriched in induced-membrane were macrophages (black box in panel A).

iPathway Guide

RNA sequences were subject to Advaita Bio's iPathwayGuide (https://www.advaitabio.com/ipathwayguide) to identify both canonical pathways and protein-protein interaction networks enriched in the Masquelet environment. First, iPathwayGuide was used to generate summary volcano plots of all differentially expressed genes for induced-membrane vs. non-union and for bone next to PMMA vs. bone next to an empty defect. Next, the “pathways” function of iPathway Guide was applied to the entire list of differentially expressed genes for each comparison. iPathway Guide identifies significantly enriched pathways using the Impact Analysis method, which is based on both over-representation of genes and their calculated cumulative effect on signaling in that pathway ²¹. Finally, the “networks” function of iPathway Guide ²² was applied to both the entire list of differentially expressed genes and to the components of significantly enriched signaling pathways. Protein interaction networks were exported from iPathwayGuide into Cytoscape v3.7.2 to generate diagrams with heat map scales representing log₂fold-changes.

Results

Identification of differentially expressed genes

Post alignment QC plots showed that each tissue sample provided between ~20 million and 35 million quality aligned mRNA reads with alignment rates between ~45% and 60%. 13,204 genes were detected in induced-membrane and non-union. Of these, 271 were increased (table S2) and 195 were decreased (table S3) in induced-membrane compared to non-union (Fig. 1A). 12,939 genes were detected in the bone samples taken just proximal to the defect sites. Of these, 874 (Table S4) were increased and 489 were decreased (Table S5) in bone next to PMMA compared to that next to the empty defect (Fig. 1B).

Figure 1: The Masquelet and non-union healing environment display global differences in gene expression.

Logarithmic volcano plots show FDR corrected p-value vs. fold change for differentially expressed genes identified in RNAseq analysis. A) Induced-membrane vs. Non-Union Tissue. B) Bone next to spacer vs. Bone next to empty defect.

CTen analysis reveals cellular make-up of regenerate tissues

RNAseq analysis can be used to discriminate cell types present in a tissue sample based on global gene expression profiles. To understand the cellular makeup of the defect site and tissues surrounding it, all genes detected by RNAseq were assigned to one of three groups: 1.) Enriched in Masquelet tissues 2.) Enriched in non-union tissues or 3.) Expressed equally in both injury environments. These lists were then imported into CTen8 (Cell Type Enrichment Analysis) and the top 3 most highly enriched cell/tissue types for each sample type were noted. Both the induced-membrane and the bone exposed to the PMMA spacer are highly enriched for osteoblast marker genes (table 1). The non-union tissue and the bone exposed to the empty defect are enriched for bone, bone marrow and B cell genes (table 2). Enrichment scores for cell/tissue types common to both osteotomy environments were relatively low (table 3).

Table 1: Top cells/tissues identified in the Masquelet environment.

Values are enrichment scores (Log₁₀-FDR corrected p-values) obtained in CTen.

a.) Top 3 enrichment scores in IM
	IM	Non-union	Common
Osteoblast Day 21	37.11	0.11	3.22
Osteoblast Day 14	37.72	0.44	2.68
Osteoblast Day 5	27.19	0.87	3.14
b.) Top 3 enrichment scores in bone next to PMMA
	PMMA	Empty	Common
Osteoblast Day 14	146.46	0.00	0.00
Osteoblast Day 21	139.11	0.00	0.00
Osteoblast Day 5	122.62	0.01	0.00

Table 2: Top cells/tissues identified in the non-union environment.

Values are enrichment scores (Log₁₀-FDR corrected p values) obtained in CTen.

a.) Top 3 enrichment scores in non-union
	Non-union	IM	Common
Bone	17.71	0.86	0.05
B Cells Marginal Zone	14.46	0.02	0.00
Bone Marrow	12.63	0.05	0.76
b.) Top 3 enrichment scores in bone next to non-union
	Empty	PMMA	Common
Follicular B Cells	73.17	0.00	0.00
Bone Marrow	68.74	0.00	0.00
Bone	62.54	0.04	0.00

Table 3: Top cells/tissues common to both osteotomy environments.

Values are enrichment scores (Log₁₀-FDR corrected p values) obtained in CTen.

a.) Top 3 enrichment scores common to IM and non-union
	Common	Non-union	IM
Microglia	8.63	0.31	2.28
Macrophage Bone marrow LPS 2 hours	8.48	0.04	1.56
Macrophage Bone marrow LPS 6 hours	7.99	0.42	0.79
b.) Top 3 enrichment scores common to bones ends adjacent to osteotomies
	Common	Empty	PMMA
Macrophage Bone marrow LPS 6 hours	3.77	1.05	1.27
Microglia	3.26	3.95	3.63
Skeletal Muscle	3.17	1.05	6.15

Next, the CTen data were queried further to estimate additional, less abundant, yet biologically relevant cell types that might also be present in each osteotomy milieu. The list of cell/tissue types was manually filtered to remove cells and tissues with no known relevance to the critical-sized defect model, as well as those with enrichment scores ≤2.0. The select cell/tissue types and their corresponding enrichment scores were imported into Morpheus to create heat maps (Fig. 2). Again, Masquelet tissues were enriched with genes associated with osteoblast differentiation (Fig. 2a and b, blue boxes). The only immune cell category enriched in the induced-membrane was macrophages (Fig. 2a, solid black box). In contrast, all non-union tissues contained a variety of immune cell types (Fig. 2a and b dotted black boxes). The induced-membrane and non-union (Fig. 2A), as well as the bone next to the non-union (Fig. 2B) were enriched for macrophage genes, but the bone next to the PMMA spacer was not. Finally, genes associated with osteoclasts were equally represented in induced-membrane and non-union (Fig. 2a, last column), but enriched in bone next to a non-union (Fig. 2b, last column). Enrichment scores for each cell/tissue type shown in figure 2 are shown in tables 4 and 5.

Table 4.

Enrichment scores corresponding to heat map shown in figure 2A.

	Non-union	IM	Common
Bone	17.71	0.86	0.05
Bone Marrow	12.63	0.05	0.76
Osteoblast Day 5	0.87	27.19	3.14
Osteoblast Day 14	0.44	32.72	2.68
Osteoblast Day 21	0.11	37.11	3.22
Skeletal Muscle	3.24	0.77	6.79
Microglia	0.31	2.28	8.63
Hematopoietic Stem Cell	7.21	0.36	0.00
Common Myeloid Progenitor	4.97	0.28	0.02
Megaerythrocyte progenitor	2.35	0.09	0.69
Mast Cells	5.22	1.5	2.31
Granulocytes	3.02	0.19	0.00
Lymph Nodes	8.62	3.03	0.11
Spleen	10.44	0.99	0.02
NK Cells	3.08	0.01	0.00
B cells Marginal Zone	14.46	0.02	0.00
Follicular B cells	9.51	0.03	0.00
CD8+ T cells	2.16	0.01	0.02
Macrophage Bone Marrow	0.11	1.42	5.04
Macrophage Peri	0.50	1.64	5.90
Osteoclasts	0.41	1.13	4.90

Table 5.

Enrichment scores corresponding to heat map shown in figure 2B.

	Bone adjacent to non-union	Bone adjacent to PMMA	Common
Bone	62.54	0.04	0.00
Bone Marrow	68.74	0.00	0.00
Osteoblast Day 5	0.01	122.62	0.00
Osteoblast Day 14	0.00	146.46	0.00
Osteoblast Day 21	0.00	139.11	0.00
Skeletal Muscle	1.05	6.15	3.17
Hematopoietic Stem Cell	38.28	0.00	0.00
Common Myeloid Progenitor	33.19	0.00	0.00
Megaerythrocyte Progenitor	9.38	0.00	0.28
Mast Cells	17.56	0.80	0.35
Granulocytes	31.49	0.01	0.00
Lymph Nodes	49.72	0.97	0.00
Spleen	51.83	0.06	0.00
NK cells	24.66	0.00	0.00
B cells Marginal Zone	56.13	0.00	0.00
Follicular B cells	73.17	0.00	0.00
CD4+ T cells	27.97	0.00	0.00
CD8+ T cells	32.42	0.00	0.00
Macrophage Bone Marrow	8.62	1.58	1.04
Macrophage Peri	3.45	2.70	2.33
Osteoclast	8.21	1.04	1.36

iPathway Guide Analysis

RNA sequences were analyzed using iPathwayGuide to identify cell signaling pathways and protein interaction networks enriched in the Masquelet environment. For the induced-membrane vs. non-union comparison, the top 3 canonical pathways identified in iPathwayGuide were “Staphylococcus Aureus Infection,” “Systemic Lupus Erythematosus,” and “Complement and Coagulation Cascades.” However, false-discovery rate (FDR) correction of p-values rendered pathway enrichment differences between the induced-membrane and the nonunion non-significant. We examined each of these pathways manually and noted that a subset of 5 differentially expressed complement proteins was common to all three pathways (table 6). Since no canonical pathways were significantly enriched in induced-membrane vs. non-union, we examined all the protein-protein interaction networks identified by iPathway Guide. We focused on networks that displayed unidirectional changes in gene expression in induced-membrane compared to non-union, since we are comparing different tissue types rather than experimentally induced changes in one system. Figure 3 summarizes networks that were increased in induced-membrane compared to non-union. A small network of complement proteins was identified (Fig. 3a). Select AP1 transcription factor complex members were enriched in induced-membrane (Fig. 3b). The remaining AP1 complex members (c-fos, fra1, fra2 and cJun) were not detected in induced-membrane or non-union. A candidate network between a potassium gated ion channel (kcnj15), CXCR-7 (Ackr3) and intracellular guanine nucleotide binding proteins is shown in figure 3c. A candidate interaction between the GPI-anchored cell surface protein mesothelin (msln) and a fibronectin-based ECM network is shown in figure 3d. Twenty-five different collagen genes were sequenced in the induced-membrane vs. non-union experiment. Only the α3 chain of type V collagen was differentially expressed in the induced-membrane (Fig. 3e). Collagen α3(V) homotrimers are found in the pericellular space where they interact directly with cells under restricted conditions including cancer progression ²³. No other components of type V collagen were elevated in induced-membrane. Finally, TNFα induced protein 6 (Tnfaip6) is consistent with activation of the innate immune system by the presence of the PMMA spacer (Fig. 3f).

Table 6.

Complement components differentially expressed in IM vs. non-union

Symbol	Gene Description	Log2 Fold Change	Adjusted P Value
Cr2	Complement C3d receptor 2	−2.99	4.17E-05
C1r	Complement C1r	0.95	3.48E-02
C1qb	Complement C1q B chain	1.06	4.41E-03
C1qa	Complement C1q A chain	1.19	5.69E-03
C1qc	Complement C1q C chain	1.24	3.87-E03

Figure 3. Network summaries for induced-membrane.

Biologically relevant protein-protein interaction networks identified in iPathway Guide as being enriched in induced membrane compared to non-union are shown. The heat map scale represents log₂ fold changes.

In contrast to the induced-membrane vs. non-union comparison, there were significantly enriched canonical pathways discovered in the bone comparison. The top three most highly enriched cell signaling pathways in bone next to the spacer relative to bone next to the empty defect were “Pathways in Cancer” (p=0.0003) “ECM-receptor interaction” (p=0.0005) and “Hippo signaling pathway” (p=0.0007). Next, we examined the protein-protein interaction networks under each of these canonical pathways. Figure 4 summarizes the interaction networks identified under “Pathways in Cancer.” Genes marked with an asterisk were identified in the interaction networks under both “Pathways in Cancer” and” Hippo Signaling Pathway”. Note that genes in these networks are also known modulators of osteogenesis, including Wnt, TGFbeta, Hedgehog, and Notch signaling components. Finally, we examined all of the protein-protein interaction networks identified by iPathway Guide similar to the strategy we used for the induced-membrane vs. non-union comparison. This revealed enrichment of a relatively large predicted network of ECM constituents in bone next to the PMMA spacer (Fig. 5).

Figure 4. Networks under “Pathways in Cancer” that were enriched in Masquelet bone.

“Pathways in Cancer” was the most significantly enriched pathway identified in iPathway Guide for the cut bone ends comparison. Biologically relevant protein-protein interactions that are part of this pathway are shown. * marks genes that are also part of the “Hippo Signaling” Pathway. The heat map scale represents log₂ fold changes.

Figure 5. ECM protein-protein interaction networks suggest activation of ECM formation in Masquelet bone.

Large and biologically relevant protein-protein interaction networks identified using the “networks” function in iPathway Guide are shown. The heat map scale represents log₂ fold changes.

Networks that were uniformly decreased in the Masquelet environment are summarized in figure 6. We reason that these are candidate players in the non-union microenvironment. Both the non-union and the bone next to it appear to express a network of muscle proteins. Two small networks of immune cell markers were also identified.

Figure 6. Candidate mechanisms at play in the non-union environment.

Protein-protein interactions networks that were uniformly decreased in Masquelet compared to non-union are shown. The heat map scale represents log₂ fold changes.

Discussion

The Masquelet technique is a promising surgical approach to address volumetric bone loss due to acute trauma ²⁴ and chronic fracture non-union ². Even though median success rates are high, there is a significant amount of variation and the outcome for failed union after Masquelet treatment is often amputation. Infection is the most commonly reported complication in retrospective case series ^{24; 25} and the non-union rate is slightly higher in infected bones ²⁶. Emerging data suggest that the ability to form an induced membrane varies among patients and that this correlates with successful graft-to-bone union after the second surgery ^{7; 25}. Thus, identifying strategies to improve Masquelet success is vital to patients experiencing segmental bone loss. Here, we begin to address this clinical need by exploring the local milieu of two different osteotomies with the long term goal of identifying patterns of gene expression or cell populations in the induced membrane that help drive successful graft to bone consolidation. We used unbiased approaches and the cell types and genes identified help confirm existing literature ^10–13 and also identify potential novel mechanisms and cell profiles.

Here we compared local gene expression patterns in the Masquelet environment to those in a developing non-union. In 6-month old male rats, a 5 mm osteotomy that is left empty will develop into a non-union in virtually 100% of cases¹⁵. A visible induced-membrane that could be dissected off the PMMA spacer formed in all 5 animals. Fibrous tissue had also formed between the cut bone ends in the contralateral limbs, but RNA yield for one non-union was too low for RNA sequencing and so tissue from this rat was not included in the study. After completion of RNA sequencing, the first issue we addressed was the cellular composition of each tissue type. Cten revealed that the Masquelet and non-union wound environments had very little overlap in predicted cellular composition. This observation was true for both the “soft-tissue” components and the bone samples taken immediately proximal to the spacer and non-union. Induced-membrane was unique from all other tissues examined in that osteoblasts were the most significantly enriched cell type and macrophages were the only immune cell type identified. Enrichment scores for cells associated with adaptive and innate immunity were relatively high in bone next to the non-union compared to that next to the PMMA spacer. In fact, with the exception of “macrophage peri,” a cell type that originates in the spleen, enrichment scores for all immune cell types were less than 2.0 in bone next to the spacer. Importantly, Cten analysis does not distinguish between different macrophage subtypes, and so differences in macrophage composition cannot be discerned from this analysis. Similarly, the relative contributions of resident bone macrophages (osteomacs) versus inflammatory macrophages to induced-membrane development and regulation of bone anabolic processes in the context of Masquelet remain to be elucidated. Our results confirm and extend previous studies suggesting the presence of macrophages, osteoblasts and osteoclasts in the induced membrane ^{10–14; 27}.

The foreign body reaction to materials implanted subcutaneously is very well characterized (reviewed in²⁸). Adsorption of serum proteins and neutrophils occur immediately, and these cells recruit additional innate effector cells including macrophages. Failure to clear the material triggers fusion of macrophages into foreign body giant cells and recruitment of fibroblasts, which deposit extracellular matrix and facilitate vascularization. The final outcome is a fibrous induced-membrane. Although they were not enriched in our Cten analysis of the induced-membrane, cells of the adaptive immune system are also recruited and activated during a typical foreign body reaction. This result may reflect the timing of our analysis, when the induced-membrane had all ready developed. Our data are also consistent with the premise that the induced-membrane that forms around an object placed into the bone microenvironment is a unique tissue relative to that which forms around a foreign-body placed subcutaneously.

In contrast to the variety of immune cell markers in the non-union environment, CTen analysis suggested that macrophages are the only immune cell type with significant presence in the induced-membrane. Macrophages are required for normal fracture healing and they impact MSC fate decisions and osteoblast differentiation in context dependent ways (reviewed in²⁹). They make essential contributions to normal endochondral ³⁰ and intramembraneous ³¹ fracture healing. In the context of endochondral healing, localized killing of macrophages (MAFIA mouse) at the time of fracture lead to a complete failure of hematoma maturation and soft callus formation ³². Systemic macrophage depletion 5 days after fracture, when the hematoma has formed completely, also lead to reductions in soft callus size that correlated with the degree of macrophage depletion³². Conversely, more robust cartilaginous callus formation occurred when macrophage numbers at the fracture site were enhanced ³². Thus, macrophages might bolster the osteogenic capacity of the induced membrane. In addition, macrophages might enhance the osteogenic capacity of graft placed into the defect at the second Masquelet surgery.

Others have compared the membrane that forms around PMMA placed into an osteotomy to the one that develops with subcutaneous placement^{13; 14}, as well as to periosteum³³ and mineralizing osteoblast cultures ¹². To our knowledge, we are the first the compare the induced membrane to a developing non-union. Our motivation for using this approach is the potential to not only identify attributes of the membrane that drive successful graft-to-bone union but also to identify characteristics of a non-union that thwart healing. Genes associated with systemic activation of B cell pools (follicular B cells, lymph nodes and spleen) were enriched in the non-union tissues. Accordingly, recombination activating genes 1 and 2 (Rag 1 and 2) were among the top 5 most substantially decreased genes in induced-membrane compared to non-union (table S-2). This observation suggests that the inability to mount an effective healing response and adequate new bone formation might lead to a chronic adaptive immune response in the developing non-union. B cells and T cells are present in the hematoma and in the cartilaginous callus, but Rag1-/- mice display accelerated fracture healing and return to mechanical competence after fracture ³⁴. If a chronic non-union represents an adaptive immune environment, one possible future outcome of our data would be the identification of indicators of non-union that could be monitored in at risk patients with any type of fracture. Indeed, systemic levels of inflammatory cytokines remain elevated up to 12 weeks after establishment of a non-union using the rat critical sized defect model ³⁵.

Genes associated with CD4+ T cells were detected in bones next to both osteotomy environments, but they were not detected in induced-membrane or non-union (enrichment score ≤2). Cten analysis also suggested that CD8+ T cell were enriched in the non-union environment compared to the Masquelet environment. Patients with high intrinsic numbers of terminally differentiated memory T cells (CD8+) suffer delayed fracture healing ³⁶. Accordingly, in a mouse model of anti-CD8 mediated T cell depletion, fracture healing was enhanced, while adoptive transfer of T cells thwarted bone regeneration ³⁶. Similarly, genes expressed by CD8+ T cells (and macrophages) were enriched in fracture callus of geriatric rodents, which display impaired fracture healing ²⁰.

Despite their evidently distinct cellular compositions, after false-discovery rate correction of p values, no canonical cell signaling pathways were significantly enriched in the induced membrane compared to non-union tissue. One possible explanation for this finding is that only a relatively small number of genes were enriched in each of the top three pathways identified and iPathway Guide predicted only modest impact of these genes on propagation of signaling. In addition, enriched genes are likely expressed by more than one cell type. Complement proteins, for example, were common to the top 3 pathways most enriched in induced membrane. These included all three chains of C1q, which activates the classical complement cascade ³⁷. In addition to activating complement pathways, C1q also contributes to the resolution of inflammation and clearance of apoptotic cells ³⁷. C1q can exacerbate inflammation in certain contexts. Complement proteins also contribute to osteoblast and osteoclast differentiation ³⁸, fracture healing and the foreign body response. As such, complement proteins are likely at play in the chronic non-union as well as the developing induced-membrane. Indeed, an additional ²⁷ complement components were present at equal levels in both induced-membrane and non-union tissue. The role of complement in membrane development is worth considering in future studies. The presence of complement components in the 4-week old induced-membrane might reflect the initial innate immune reaction to the PMMA spacer and activation of inflammatory cascades. S. Aureus and other pathogens that cause osteomyelitis also engage the complement system. The extent to which pre-activation of complement in a chronically infected non-union might influence induced-membrane formation during Masquelet treatment is an important clinical consideration.

To our surprise, changes in gene expression evoked in bone by the PMMA spacer were unique from and more substantial than those identified in the induced-membrane vs. non-union comparison. iPathway Guide analysis revealed “pathways in cancer”, “cell-ECM interactions” and “activation of Hippo signaling” in bone adjacent to the PMMA spacer. Many of the protein interaction networks identified under “pathways in cancer” are also known modulators of osteogenesis and bone healing. These included Wnt, Hippo, TGF-b and notch-delta signaling pathways. A rigid surface is known to promote osteoblast lineage progression of MSC ³⁹, and so the rigid surface afforded by the PMMA spacer may provide pro-osteogenic cues to MSC present in the cut the bone ends. Alternatively, the spacer may promote more active bone healing and cortical vitality as a result of the mechanical pressure put onto the cut bone surfaces by the spacer. Indeed, during the second phase of Masquelet, bone formation appears to initiate from the cut bone ends and not from the center induced membrane ⁴⁰. Thus, our results may suggest that the bone ends in the Masquelet are ‘primed’ for regeneration, and thus able to respond favorably to autograft when placed during the second stage. The PMMA spacer alone is not sufficient for consolidation after the second surgery, however, because dissecting away the induced membrane results in poorer healing ⁴¹.

Finally, in order to identify potential biological mechanisms at play in the non-union, we examined networks that were uniformly decreased in the Masquelet environment. We reasoned that these genes might be elevated in a non-union. The number of interactions identified using this approach was small, but the genes involved were consistent with the possibility of a chronic immune response to an un-resolving boney injury. A pre-B cell receptor (Igll1 also called CD179b) and a pre-B cell markers (Vpreb2 also called CD179a) were less abundant in the Masquelet environment. Mast Cell Expressed protein 1 (Mcemp1) is associated with innate immunity. CD177 is a GPI anchored cell surface protein that plays a role in neutrophil activation and extravasation. In addition, a small network of muscle genes (isoforms of troponin, myosin and actin) was less abundant in induced-membrane and bone next to the PMMA spacer. Although we were careful to avoid it, we cannot rule out the possibility that surrounding muscle tissue had collapsed into the osteotomy site and that we had collected some into our “non-union” samples.

Our gene expression analyses support and extend previous studies that sought to characterize the induced-membrane and its contribution to successful healing of critical-sized osteotomies. Osteoblasts are a component of an induced-membrane that forms around PMMA placed in bone, but not the membrane that forms around PMMA placed in a subcutaneous pocket ¹⁴. Accordingly, the induced-membrane contains markers of osteoblasts and bone regeneration including BMP2, TGFβ, RUNX2, VEGF, IL-6 and Stro-1 ^{11; 12; 14; 40}. The membrane also contains osteoclasts ¹⁰. Islands of calcification have been documented in the membrane induced in rodent models of Masquelet ^{10; 11; 40; 42} and in biopsies of induced-membrane harvested from human patients ³. Cellular and growth factor content of the induced-membrane diminish with time as it becomes more ECM-rich and fibrotic, however, the impact of this decrease on outcomes post stage second surgery is not known ^{3; 14}.

Our study is limited primarily by small sample size and the fact that bulk RNA sequence data was used to estimate the cell types present and the signaling pathways at play in the Masquelet and non-union contexts. Validation and extension of our results at the protein level and by using more spatially informative methodologies like single cell RNA sequencing, as well as in situ hybridization and immunohistochemistry of induced membranes are essential future experiments.

Together our data suggest that the PMMA spacer itself primes an osteogenic response during Masquelet surgery in both the bone and the membrane, and that activation of the adaptive immune system contributes to failed healing in a segmental defect that progresses to a chronic non-union.