Anatomic demarcation by positional variation in fibroblast gene expression programs

Abstract

Fibroblasts are ubiquitous mesenchymal cells with many vital functions during development, tissue repair, and disease. Fibroblasts from different anatomic sites have distinct and characteristic gene expression patterns, but the principles that govern their molecular specialization are poorly understood. Spatial organization of cellular differentiation may be achieved by unique specification of each cell type; alternatively, organization may arise by cells interpreting their position along a coordinate system. Here we test these models by analyzing the genome-wide gene expression profiles of primary fibroblast populations from 43 unique anatomical sites spanning the human body. Large-scale differences in the gene expression programs were related to three anatomic divisions: anterior-posterior (rostral-caudal), proximal-distal, and dermal versus nondermal. A set of 337 genes that varied according to these positional divisions was able to group all 47 samples by their anatomic sites of origin. Genes involved in pattern formation, cell-cell signaling, and matrix remodeling were enriched among this minimal set of positional identifier genes. Many important features of the embryonic pattern of HOX gene expression were retained in fibroblasts and were confirmed both in vitro and in vivo. Together, these findings suggest that site-specific variations in fibroblast gene expression programs are not idiosyncratic but rather are systematically related to their positional identities relative to major anatomic axes.

Figure 1. Diversity of Gene Expression in Human Fibroblasts

(A) Forty-seven primary fibroblast populations, from 43 unique anatomic sites, were obtained from arm (blue circles), leg (turquoise circles), trunk (pink circles), foreskin (yellow circles), and internal organs (red circles). The cells were derived from 20 different human autopsy or surgery donors identified by the letters A through T in parenthesis. (B) Diversity of gene expression programs in 47 fibroblast populations. Each row represents a gene; each column represents a fibroblast population. The expression level of each gene is represented relative to the median value in all samples; expression levels above and below the global median are denoted by shades of red or green, respectively. The color scale encompasses a range from 32- to 0.03-fold relative to global median transcript level for each gene (+5 to −5 logs on log base 2 scale). Black represents the median expression value; gray represents missing data. (C) Similarity in the global gene expression profiles of 47 fibroblast samples. We used unsupervised hierarchical clustering of 7,580 genes that were reliably measured in 70% of the samples and varied by 3-fold above the global median in at least five (approximately 10%) of the samples. Thirty-five of the 47 samples were placed in clusters composed predominantly of cells from the same anatomic origins (arm, leg, trunk, foreskin, or nondermal).

Figure 2. A Gene Expression Signature Divides Fibroblasts from Anterior and Posterior Sites of the Human Body

(A) Unsupervised hierarchical clustering separated fibroblasts from above and below the umbilicus. Blue bars indicate anterior (rostral) samples; green bars indicate posterior (caudal) samples. (B) Supervised analysis with SAM using all samples identifies two distinct gene expression profiles corresponding to the origin of most samples being anterior and posterior to the umbilicus (FDR = 1%).

Figure 3. Fibroblasts from Limbs Exhibit Binary Gene Expression Signatures that Demarcate Proximal-Distal Position

(A) Unsupervised hierarchical clustering of all upper limb-derived fibroblasts (left) demonstrates differential gene expression between the proximal and distal upper limb fibroblasts. Supervised analysis with SAM (FDR = 5%) revealed two strong gene expression patterns that separate all proximal and distal samples (middle). A three-class analysis (right) did not reveal a third gene expression pattern. (B) Unsupervised hierarchical clustering (left) of lower limb-derived fibroblasts also exhibited a division of proximal and distal lower limb fibroblasts. Supervised analysis with SAM (FDR = 5%) revealed a binary expression pattern separating the proximal and distal samples (middle). Organizing the leg into three segments and using a corresponding multiclass analysis failed to reveal a third gene expression pattern (right).

Figure 4. Gene Expression Signatures Related to Anatomic Divisions Recur throughout the Human Body

(A) Hand and foot fibroblasts share a distal-specific gene expression signature. Hierarchical clustering of arm and leg samples together, using genes that are differentially expressed between the proximal and distal regions of the arm and leg; 98 genes that are relatively induced in distal-derived fibroblasts are highlighted by the white box. (B) Hierarchical clustering of all 47 samples using the 98 distal genes revealed that foreskin (frsk) samples were most closely related to the feet (ft) and finger (fng) samples, indicating that the “distal” gene expression pattern of these genes is repeated in the feet, fingers, and foreskin. (C) Distinction of arm and leg fibroblasts by an anterior gene centroid. We created an averaged gene expression pattern, termed a centroid (methods), from the gene expression profiles of anterior samples from head, chest, and trunk. The anterior centroid is then compared to the expression profile of each arm and leg sample for similarity (using noncentered Pearson correlation). The anterior centroid was positively correlated with all except two profiles of arm fibroblasts and negatively correlated with all leg fibroblasts. We also calculated a posterior centroid from trunk samples below the umbilicus, and it negatively correlated with all arm samples and positively correlated with all leg samples (data not shown).

Figure 5. Genes that Vary According to Three Binary Anatomical Divisions Organize Fibroblasts by Their Anatomic Sites of Origin throughout the Body

(A) A positional model where each fibroblast sample belonged to one segment. A multiclass analysis of the 47 samples, classified according to an Anterior (blue), Posterior (green), Proximal (yellow), Distal (red), and nondermal (not shown). (B) Heat map of the 337 genes that overlapped between the model in (A) and in one or more of the comparisons within local anatomic structures in Figures 2 to 4. These 337 genes contained many transcription factors (gene names in blue), components and enzymatic modifiers of the extracellular matrix (gene names black), signaling proteins in development (gene names in pink), and guidance molecules in cell migration (gene names in orange). (C) The selected 337 genes grouped 47 of 47 of the fibroblast samples within clusters composed exclusively of samples from the same anatomical regions.

Figure 6. Uniqueness of Gene Selection and General Applicability of a Model of Fibroblast Differentiation Based on Positional Identity along Three Anatomic Divisions

(A) Statistical significance of grouping fibroblasts by their anatomic origin using the minimal set of 337 positional identifier genes. One hundred random sets of 337 genes from 7,580 well-measured and variably expressed genes were tested for their ability to organize fibroblasts by anatomic origin using hierarchical clustering; the results are shown as a histogram. We scored the minimum number of branches in the dendrogram that have to be moved to place all the samples into clusters that composed exclusively of samples from each anatomical region. The median number of correctly characterized samples by random gene sets (gray bars) was 35 (12 branch movements), and no random gene set matched the performance of the 337 genes representing positional segments (red dashed line, 47 of 47 correct). The observed result with the 337 genes is seven standard deviations away from the median performance of random gene sets. (B) The 337 genes organized an independent set of fibroblasts by anatomic origin. Using hierarchical clustering, expression patterns of the 337 genes correctly grouped ten of ten new fibroblast samples according to their location in the abdomen (Abd), gum, toe, and fetal buttocks (Fetal Bt).

Figure 7. Features of the Embryonic HOX Code Are Maintained in Adult Fibroblasts

(A) Expression patterns of 43 homeodomain transcription factors, including 12 HOX genes, organized 43 of 47 the samples by their anatomic origin. HOXA13 (yellow boxes) is expressed in fibroblasts from distal sites: feet, fingers, foreskin, and prostate. HOXB2-HOXB9 genes are expressed exclusively in fibroblasts from trunk (pink boxes) and nondermal sites (white box), echoing their expression pattern during early embryogenesis. HOXD genes are expressed in the trunk (and proximal leg) but not in the nondermal samples (top pink box). (B) Validation of HOXA13 mRNA expression in fibroblasts from distal foot (culture 1), but not fibroblasts from thigh (culture 6). Non-RT negative controls (lanes 3 and 4) and ribosomal protein S15 loading control (lanes 5 and 6) are also shown. (C) Immunoblotting confirmed HoxB4 protein expression in fetal lung fibroblasts but not in foreskin fibroblasts. Raw microarray images shown above indicate RNA expression level. (D) Immunoblotting confirmed HoxA13 protein expression in foot fibroblasts and total foreskin tissue, but not in thigh fibroblasts. Raw microarray images shown above indicate RNA expression level. (E) Immunofluorescence of HoxA13 protein expression in foreskin tissue sections. Nuclei are highlighted by DAPI staining.