Insights from computational modeling in inflammation and acute rejection in limb transplantation

Abstract

Acute skin rejection in vascularized composite allotransplantation (VCA) is the major obstacle for wider adoption in clinical practice. This study utilized computational modeling to identify biomarkers for diagnosis and targets for treatment of skin rejection. Protein levels of 14 inflammatory mediators in skin and muscle biopsies from syngeneic grafts [n = 10], allogeneic transplants without immunosuppression [n = 10] and allografts treated with tacrolimus [n = 10] were assessed by multiplexed analysis technology. Hierarchical Clustering Analysis, Principal Component Analysis, Random Forest Classification and Multinomial Logistic Regression models were used to segregate experimental groups. Based on Random Forest Classification, Multinomial Logistic Regression and Hierarchical Clustering Analysis models, IL-4, TNF-α and IL-12p70 were the best predictors of skin rejection and identified rejection well in advance of histopathological alterations. TNF-α and IL-12p70 were the best predictors of muscle rejection and also preceded histopathological alterations. Principal Component Analysis identified IL-1α, IL-18, IL-1β, and IL-4 as principal drivers of transplant rejection. Thus, inflammatory patterns associated with rejection are specific for the individual tissue and may be superior for early detection and targeted treatment of rejection.

Figure 1. Histological evaluation of skin and muscle rejection in the early postoperative phase (POD 3 and 5).

(A–C) Skin sample taken on POD 3 from an allograft without immunosuppression (A), from an isograft (B) and an allograft under TAC (C) showing no/rare inflammatory response (Grade 0 rejection). (D–F) Skin biopsies taken on POD 5 from a rejecting animal (D) displaying Grade 1 rejection, from an isograft (E) showing no/rare inflammatory response and a TAC treated allograft (F) characterized by a mild inflammatory response (Grade 0-I) in the deep dermis. (a–c) Muscle sample taken on POD 3 from an allograft without immunosuppression (a), from an isograft (b) and an allograft under TAC (c) showing no/rare inflammatory response (Grade 0 rejection). (d–f) Skin biopsies taken on POD 5 from a rejecting animal (d) displaying a mild inflammatory response (Grad 0-I rejection), from an isograft (E) and a TAC treated allograft (f) showing no/rare inflammatory response.

Figure 2. Distribution of inflammatory mediator levels at the earliest postoperative measurements (POD 3) in rat limb transplantation models.

Adjusted p-values from Wilcoxon rank-sum test between the rejection group (ATC) vs. Tacrolimus treated group (TAC) for selected inflammatory mediators, which were tested to be included in a prediction model by multinomial logistic regression analysis, are presented.

Figure 3. Similarity of sample groups and association between inflammatory mediators in rat limb transplantation models based on their profiles of mean levels in each condition.

Heatmap as a result of complete linkage hierarchical clustering on log2-transformed and mean centered data. Log2-fold differences against the respective mean levels of each inflammatory mediator are color coded (red means higher inflammatory mediator levels and blue means lower inflammatory mediator levels than the mean levels of the respective inflammatory mediator according to the color scheme at the top). Dendrograms (trees) show similarity between different conditions and different inflammatory mediator profiles, respectively.

Figure 4. Results from Random Forest classification of the different rejection groups (ISO, TAC, ATC) using over the whole time course (POD 3, POD 5, POD 7, POD 9) measured inflammatory mediators in skin (A) and muscle (B) samples of rat limb transplantation models.

Most important mediators for the decision trees based classification approach are evident by ranked mean decrease accuracy. Performances of the classifiers are indicated by the confusion table and the out-of-bag (OOB) error rate.

Figure 5. Most variable mediators identified by principal component analysis (PCA) suggesting new potential targets for therapeutic interventions to suppress limb transplant rejection.

PCA scores (loadings) for the first four principal components (PCs), which represent more than 70% of information within the data, are displayed in a stacked bar plot for all inflammatory mediators (ranked by the overall PCA score of the 4 PCs).