Figure 1.
Schematic of the study design. A.
Paired microarray of normal and lymphedematous skin derived from 27 subjects with lymphedema of one or more limbs. RNA was isolated from the whole tissue. Pathway analysis was performed to identify the final targets for protein analysis. B. High throughput assay of identified targets was performed using the Luminex 51-plex bead assay. Assays were performed on the original 27 lymphedema subjects and on 12 healthy normal controls. Logistic regression modeling was performed to identify the final targets for prospective analysis. C. Prospective assay for the 6 target proteins was performed on a distinct cohort of 36 lymphedema subjects and 15 normal controls. The results were analyzed through plotting a receiver operating characteristic (ROC) curve.
Figure 2.
Representative histological findings in paired biopsy specimens of lymphedematous and normal skin.
The specimens are derived from a representative study subject and reflect paired specimens from lymphedematous and normal limbs. A. Lymphedema skin H&E staining demonstrates an overall increase in the cellularity of the specimen that is particularly prominent in the epidermis and dermal-epidermal junction. There are prominent perivascular inflammatory infiltrates and there is obliteration of the dermis by dense eosinophilic material. Numerous dilated microvascular structures are seen in the upper dermis. B. Normal skin H&E shows normal cellularity, absence of inflammation and no notable microvascular changes. C. Lymphedema skin LYVE-1 staining demonstrates that the endothelial-lined microvascular structures seen on standard histology are lymphatic. There is evidence of positive microvascular lymphatic remodeling, as we have previously noted in the murine experimental model [10], [23], with an increase in the size and number of identified LYVE-1-positive structures. D. Normal skin LYVE-1 staining discloses scant-to-absent lymphatic structures in the dermis.
Table 1.
Lymphedema etiology.
Table 2.
Pathway analysis of differentially expressed genes by microarray.
Table 3.
Secreted protein targets identified by paired microarray.
Table 4.
The Luminex 51-plex assay.
Table 5.
Logistic regression modeling of high throughput protein assay.
Figure 3.
Receiver operating characteristic curve.
Based on the L1-regularized logistic regression model with six proteins, a receiver operating characteristic (ROC) curve yields an area under the curve (AUC) of 0.87. The color at a position along the curve is indicative of the specificity, and sensitivity can be gauged by looking at the color at the corresponding height along the left, vertical axis.