Expression Profiling in the Muscular Dystrophies: Identification of Novel Aspects of Molecular Pathophysiology

The dystrophin–glycoprotein complex of muscle fibers has emerged as a critical multiprotein complex that imparts structural integrity to the muscle fiber plasma membrane during the contraction of skeletal muscle. Identification of a principle component, dystrophin (Hoffman et al. 1987a), led to the identification of additional protein components of the protein complex (Ibraghimov-Beskrovnaya et al. 1992; Roberds et al. 1993; Bonnemann et al. 1995; Noguchi et al. 1995; McNally et al. 1996; Nigro et al. 1996). A major role of these proteins is to provide physical connections between intracellular cytoskeletons (actin filaments) and extracellular basal lamina (laminins). In brief, dystrophin attaches to the plasma membrane via the transmembrane protein β-dystroglycan using a cysteine-rich COOH-terminal domain. Physical connections between the plasma membrane and basal lamina are mediated by a multimeric chain involving dystrophin, β-dystroglycan, α-dystroglycan, and laminin α-2. The interaction of dystrophin with β-dystroglycan is stabilized by the heterotetrameric sarcoglycan complex (α, β, δ, and γ subunits) (Ervasti and Campbell 1991; Cullen et al. 1996; Duclos et al. 1998). Interactions between dystrophin and intracellular actin filaments is mediated by a series of actin-binding sites, many of which are clustered in the NH₂-terminal 200 amino acids. Importantly, loss of dystrophin leads to loss of all actin–filament association with the plasma membrane (Rybakova et al. 2000). Utrophin is a paralogue of dystrophin, which can in part functionally replace dystrophin when overexpressed in muscle, though it is not able to restore actin–filament association with the membrane (Tinsley et al. 1998; Rybakova et al. 2000). The dystrophin/dystroglycan/laminin complex is partially redundant with the vinculin/integrin/laminin complex, though the dystrophin-based complex seems more important for homeostasis during muscle contraction, whereas the integrin-based cytoskeleton appears more important for the appropriate development of muscle (Hayashi et al. 1998; Burkin and Kaufman 1999; Kaariainen et al. 2000).

In addition to the structural role for dystrophin and the associated proteins, there is clear evidence for signal transduction roles. The COOH-terminal domains of dystrophin bind a series of proteins, both directly and indirectly, each of which has been shown to have a role in muscle development or homeostasis. The dystrophin-binding syntrophins are a series of proteins with PDZ domains that in turn bind nNOS, ERK6, and other proteins (Yang et al. 1995; Chao et al. 1996; Hasegawa et al. 1999). Loss of dystrophin leads to the secondary loss of nNOS at the plasma membrane and defects in local blood flow regulation during exercise (Thomas et al. 1998). There are likely many additional directly or indirectly interacting proteins involved in signal transduction, and the delineation of these pathways is critical for understanding the pathophysiology of dystrophin deficiency.

Genetic abnormalities of dystrophin are the cause of the most common type of muscular dystrophies, Duchenne and Becker muscular dystrophy. The gene encoding dystrophin has a very high mutation rate (1 in 10,000 germline cells), due to the large size of the gene (2.5 million bp) and the presence of certain hotspots for rearrangements within introns (Gorospe and Hoffman 1992; Hamed et al. 2001). As a result, the disease has a high incidence in all world populations, as well as in animals (Williams et al. 1983; Bulfield et al. 1984; Cooper et al. 1988; Carpenter et al. 1989; Roddie and Bundey 1992). The presentation and progression of dystrophin deficiency varies dramatically between species, despite the shared biochemical feature of complete deficiency of the protein in muscle and heart from fetal life onward. Humans show evidence of myofiber membrane instability (high serum creatine kinase levels) from birth onwards, but do not show obvious clinical weakness until ∼4 yr of age. The disease is then progressive, with patients gradually losing most of their skeletal muscle by 16 yr of age (Hoffman and Schwartz 1991). Cats and mice lacking dystrophin show muscle hypertrophy as the predominant symptom, with little clinical evidence of weakness or muscle loss (Bulfield et al. 1984; Carpenter et al. 1989; Gaschen et al. 1992). Dogs lacking dystrophin show an even more rapid onset and progression than humans, though this can be variable from dog to dog (Cooper et al. 1988; Kornegay et al. 1988). Primary deficiencies of each of the sarcoglycan proteins have been documented in human patients, and the clinical symptoms are indistinguishable from primary dystrophinopathies (Duggan et al. 1997). Rodent models have been described for each of the four sarcoglycans, and, like dystrophin deficiency, these mouse and hamster models show muscle hypertrophy as the primary clinical symptom (Mizuno et al. 1995; Duclos et al. 1998; Hack et al. 1998; Durbeej et al. 2000).

The dramatic progressive nature of dystrophin deficiency and sarcoglycan deficiency in humans and dogs, but not cats and rodents, implies that there are important secondary, downstream effectors of muscle wasting and weakness. The clinical differences between the human diseases and their animal orthologues also support this hypothesis. Considerable data has recently accumulated concerning biochemical and environmental effectors of “acute” myofiber damage and necrosis, particularly in regards to membrane damage (plasma membrane fragility) and functional ischemia (nNOS deficiency) (Thomas et al. 1998). These “acute” features appear to be shared between dystrophin and α-sarcoglycan–deficient fibers of all species. Less well studied are the secondary “chronic” aspects of these disorders, which is particularly important as the downstream events appear responsible for the progressive weakness and early death in human patients. Histopathological correlations have provided descriptive associations with the progressive aspects of the human disease, such as proliferation of connective tissue (fibrosis), failure of myofiber regeneration, and infiltration of mast cells and other inflammatory cells (Hoffman and Schwartz 1991; Gorospe et al. 1994a). However, no clear pathophysiological cascades have yet been defined and no molecular basis identified for any of the histological correlates.

We hypothesized that genome-wide expression profiling would provide insight into the downstream pathophysiological cascades in the muscular dystrophies. We felt that the muscular dystrophies would be an excellent disease system in which to apply emerging microarray technology for the following reasons. First, muscle is a relatively simple tissue, with the myofiber being the predominant cell type. Second, muscle is routinely biopsied from human patients and rapidly frozen, making tissue easily available for expression array studies. Third, the primary molecular defect in many cases of muscular dystrophy is known, such that it is possible to study groups of genetically homogeneous patients. To test this, we present expression profiling using Affymetrix HuGeneFL high-density oligonucleotide arrays (∼6,000 full-length genes) on human muscle biopsies from dystrophin deficiency, α-sarcoglycan deficiency (α-SGD), and normal controls.

We used muscle biopsies from primary dystrophinopathy (Duchenne muscular dystrophy [DMD] and female manifesting carriers), other dystrophies (α-SGD, calpain deficiency, and merosin deficiency), and normal controls. All patients had documented mutations of the corresponding genes (α-sarcoglycan, merosin, and calpain III; see below) or showed primary biochemical deficiency at the protein level (dystrophinopathies: DMD, and female manifesting carriers).

For expression profiling, frozen muscle biopsies from five male patients with dystrophin deficiency (age 6–9 yr), four patients with α-SGD (two males and two females; age 8–11 yr), and five male controls (age 6–9 yr) were used. For dystrophin deficiency, all biopsies were shown to have marked dystrophin deficiency by immunoflouresence (60 kD and d10 polyclonal antibodies [Hoffman et al. 1987b; Koenig et al. 1988] and/or dys III COOH-terminal antibody monoclonal antibodies [Novocastro] [Nicholson et al. 1989]), immunoblot (30 kD [Hoffman et al. 1987b] or dys II rod domain [Novocastro] [Nicholson et al. 1989]), or both. All patients showed clinical symptoms consistent with the diagnosis of DMD. All patients showed a dystrophic myopathy by hematoxylin/eosin staining and dramatic elevations of serum creatine kinase level. For α-SGD, all biopsies were shown to have normal dystrophin by immunostaining and immunoblot, but complete α-SGD by immunostaining. Patients used for expression array experiments had the following previously reported gene mutations in the α-sarcoglycan gene: I103T/L173P compound heterozygote, I124T, where the second allele was not expressed at the RNA level, R77C/D97G compound heterozygote, and ΔC166 homozygote (Duggan et al. 1997). All patients showed a dystrophic myopathy by hematoxylin/eosin staining and elevation of serum creatine kinase levels. Control biopsies were from patients sent for diagnosis of a possible myopathy, but who showed normal muscle histology by hematoxylin/eosin staining, normal myofiber structure by electron microscopy, and normal mitochondria enzyme activity.

For immunostaining experiments, muscle biopsies from normal controls (7 males and 1 female; age fetus to 26 yr), DMD (17 males; age fetus to 9 yr), α-SGD (2 females, age 8 and 12 yr), female-manifesting carriers of Duchenne dystrophy (mosaic for dystrophin) (2 females, age 4 and 7 yr), calpain III deficiency (female, 15 yr), and partial-merosin deficiency (female, 23 yr) were studied. α-Sarcoglycan patients used for immunostaining had the following mutations: compound heterozygote I103T/L173P, and compound heterozygote R77C/A93V (Duggan et al. 1997). For calpain III deficiency and partial-merosin deficiency, biopsies were shown to have normal dystrophin by immunofluorecence and immunoblot. The partial-merosin deficient patient showed partial-merosin deficiency, by immunofluorecence, and a C862R missense mutation, with the second allele not expressed at the RNA level (Tezak et al. 2000). The calpain III–deficient patient showed a R572Q homozygous/hemizygous mutation in the calpain III gene as previously reported (Chou et al. 1999).

Five dystrophin deficient, four α-sarcoglycan deficient, and five control muscle biopsies were used to extract total RNA by using TRI_ZOL^® reagent (GIBCO BRL). Each biopsy was divided into two fragments, and RNA was isolated from each fragment independently (28 biopsy fragments total). 10 μg of total RNA from each biopsy fragment (28 muscles total) was converted into double-stranded cDNA by using SuperScript Choice system (GIBCO BRL) with an oligo-dT primer containing T7 RNA polymerase promoter (Genset). The double-stranded cDNA was purified by phenol/chloroform extraction, and then used for in vitro transcription using ENZO BioArray RNA transcript labeling kit (Affymetrix). Biotin-labeled cRNA was purified by RNeasy kit (QIAGEN), and fragmented randomly to ∼200 bp (200 mM Tris-acetate, pH 8.2, 500 mM KOAc, 150 mM MgOAc). cRNA samples of each group were pooled in duplicate before hybridizing to Affymetrix HuGeneFL microarray for 16 h. The microarray was washed and stained on the Affymetrix Fluidics Station 400, using instructions and reagents provided by Affymetrix. This involves removal of nonhybridized material, and then incubation with phycoerythrin–streptavidin to detect bound cRNA. The signal intensity was amplified by second staining with biotin-labeled anti-streptavidin antibody and followed phycoerythrin–streptavidin staining. Fluorescent images were read using the Hewlett-Packard G2500A Gene Array Scanner.

Data analysis of Affymetrix microarrays was done using GeneChip^® software (version 3.3), as described previously (Lockhart et al. 1996). In brief, each gene is queried by 20 perfect match (PM) and 20 mismatch (MM) 25-base probes; the latter has a single base change in the center of the 25-bp probe. Comparison of the hybridization signal from the PM and MM probes allows a specificity measure of signal intensity, and elimination of most nonspecific cross-hybridization from the data analysis. Values of intensity difference, as well as ratios of each probe pair, are used for determining whether a gene is called “present” or “absent.” For comparison of different datasets (e.g., dystrophin deficiency versus normal control), each probe pair in an experimental GeneChip^® assay is compared with control groups, and four matrixes were used to determine the difference calls that indicate whether transcription level of a gene is changed.

Iterative comparisons of different datasets were done by spreadsheet analysis (Microsoft Excel). In brief, each dystrophin deficiency chip (n = 2) and α-SGD chip (n = 2) was compared with each control chip (n = 2) to determine the expression difference between each muscular dystrophy and the control. Difference calls that showed consistent results in all four pairwise comparisons of each disease were extracted for further analysis.

Polyclonal antibodies against complement component 3 (C3) and thrombospondin-4 were provided by Dr. Fernando Vivanco (Fundacion Jimenez Diaz, Madrid, Spain) (Alberti et al. 1996) and Dr. Jack Lawler (Beth Israel Deaconess Medical Center and Harvard Medical School, Boston, MA) (Lawler et al. 1995). Sheep-anti–human factor XIIIa polyclonal antibody was from Cedarlane. Monoclonal antibody against HLA-DR was from Biomeda. Monoclonal antibody against secreted phospholipase A2 was from Cayman Chemical. Monoclonal antibodies against secreted protein, acidic and rich in cysteine (SPARC)/osteonectin and versican were from USBiological. A monoclonal antibody against α-cardiac actin was from Maine Biotechnology Services. All secondary antibodies were purchased from Jackson ImmunoResearch Laboratories, including FITC-conjugated donkey anti–mouse IgG, Cy3-conjugated goat anti–mouse IgG, Cy3-conjugated donkey anti–sheep IgG, and Cy3 conjugated donkey anti–rabbit IgG.

Serial 4-μm-thick frozen muscle sections were cut with an IEC Minotome cryostat, mounted to Superfrost Plus Slides (Fisher Scientific,), and fixed in cold anhydrous acetone. Sections were then blocked for 30 min in 10% horse serum and 1× PBS, and incubated with primary antibody for 3 h at room temperature. Antibody dilutions were as follows: (a) 1:500 for C3, thrombospondin-4, and factor XIIIa, (b) 1:200 for PLA2, (c) 1:10,000 for SPARC/osteonectin (d) 1:2,000 for versican, (e) 1:1,000 for embryonic myosin heavy chain, (f) 1:20 for HLA-DR, and (g) 1:10 for α-cardiac actin. Washes were done with 10% horse serum and 1X PBS, and sections then incubated with secondary antibody for 1 hour. FITC-conjugated donkey anti-mouse IgG was diluted 1:100. All other Cy3-conjugated secondary antibodies were diluted 1:500.

Affymetrix image files for the six chip hybridizations and the absolute analysis results of each chip are available.

The goal of this study was to determine downstream gene expression changes resulting from known primary biochemical defects in muscle. However, other sources of gene expression changes include variability in cell-type content of patient muscle biopsies and genetic background differences between individuals. These variables can complicate interpretation. To minimize the effect of these variables, we used the following experimental strategy (Fig. 1). First, each patient muscle biopsy to be studied was split and processed in duplicate. The duplication of each biopsy sample would be expected to control for all sources of both tissue and experimental variability, including cell-type heterogeneity within the tissue, variables in RNA isolation and biotinylated cRNA production, and variability in hybridization to GeneChip^® microarrays. Second, to minimize genetic polymorphic variation in expression patterns between different individuals, we studied four or five patient biopsies simultaneously, with equal amounts of cRNA mixed for each of the groups, and the resulting cRNA pools were then hybridized to a single GeneChip^® array. Polymorphic variations in expression profiles should be normalized by this approach, whereas gene expression changes correlating with the primary biochemical defect should be retained. The mixing protocol also reduced the cost of the analysis, requiring substantially less microarrays to carry out the experiments. This protocol resulted in six datasets (normal1, normal2, DMD1, DMD2, α-SGD1, and α-SGD2) (Fig. 1). An example of raw image data showing hybridization of cRNA to 20 probe pairs of a single gene is shown in Fig. 2 A.

Description of genes tested on the GeneChip^® HuGeneFL array is listed on our site (http://www.cnmcresearch.org; link to microarray). Among the 7,095 probe sets (∼280,000 oligonucleotide features) on the Affymetrix HuGeneFL microarray, we found a consistent number of “present” calls for each of the six cRNA pools tested (control 32 and 37%; DMD 36 and 32%; α-SGD 30 and 36%). Data from each experiment is posted on a web site for public access, and comparison to other HuGeneFL datasets (http://microarray.CNMCResearch.org/resources.htm; link to “muscle, human”). Included on the web site is the raw image files for each of the six microarrays, text files containing absolute analyses of each chip (“present” calls; GeneChip^® software output), and comparison analyses between different chips (difference calls; GeneChip^® software output).

There was high concordance (88%) of “present” calls between duplicated datasets. However, the level of RNA found for each gene showed some variability between datasets; consistent with most microarray data published to date, the highest variability in levels was in genes showing low levels of cRNA hybridization (Fig. 2 B). The variability in levels of specific genes is likely a combination of tissue variability and experimental variability.

Genes that are consistently increased or decreased in all four possible iterative comparisons were determined (e.g., control 1 versus DMD1, control 1 versus DMD2, control 2 versus DMD1, and control 2 versus DMD2) (Fig. 3). Only ∼40–60% of difference calls from a single comparison typically survived all four iterative comparisons of datasets. For this analysis, “marginal” difference calls assigned by GeneChip^® software analysis were retained in the data sets. The cutoff used for difference calls was a twofold change in expression (either increase or decrease). The four iterative comparisons gave four values for “fold change,” which were then averaged.

It is important to note that we focused on difference calls that satisfied all four iterative comparisons of data. This can be considered a very stringent selection of data, as genes that showed significant changes in expression in three data comparisons, but not the fourth, would be excluded from further study. We studied a series of genes that showed difference calls in three comparisons, but not the fourth; in each case the fold changes were close to the twofold cutoff used (data not shown). All data is presented on the web site, with selected data presented in Table and Table.

From the four pairwise comparisons for each disease to normal controls, we identified 275 differentially regulated genes for dystrophin deficiency, and 233 differentially regulated genes for α-SGD. Thus, ∼30% of probe sets tested were expressed in muscle, and ∼10% of these showed differential regulation in dystrophin deficiency and/or α-SG deficiency. Expression of 138 genes was upregulated, and 137 genes downregulated in dystrophin deficiency versus control, 90 were upregulated and 143 genes downregulated in α-SGD versus control (Fig. 3). These data are also presented as a log scale graph of fold changes, with and without “tilda” values (Fig. 4). As explained in the figure legend, “tildas” are assigned when the denominator approaches zero (e.g., “absent” call), leading to possible exaggeration of the resulting ratio (Fig. 4).

We expected many pathological processes to be involved in muscular dystrophy patient muscle, including degeneration cascades, regeneration programs, and fibrotic proliferation genes. We expected these changes to be shared between the two closely related primary biochemical defects. Consistent with this, we found 144 genes to show greater than twofold up and downregulation in both dystrophin deficiency and α-SGD. We then clustered these genes by pathological processes (Table). The largest functional group of upregulated genes were genes of cell surface and extracellular proteins (42%). Additional functional groups included genes involved in immune responses (20%) and cell growth, differentiation, and signaling (15%).

Among 80 downregulated genes, 36% of them were involved in mitochondria function and energy metabolism (Table). Importantly, this data suggests that there is a widespread disorder of both aerobic and anaerobic energy metabolism in patient muscle. Approximately 12% of downregulated genes were involved in cell growth, differentiation, and signaling. Specific genes were selected for verification of expression changes by immunostaining of patient muscle biopsies. As described below, all tested expression changes were confirmed by immunostaining data.

We then tested for gene expression changes that were specific for either dystrophin deficiency or α-SG deficiency. We expected that this analysis would provide valuable transcriptional information regarding well documented secondary protein deficiencies in each disorder, and would help determine whether these secondary biochemical abnormalities were due to reduced RNA levels or protein instability. In addition, we hypothesized that such disease-specific changes might point to genes or proteins that could have a functionally significant association with dystrophin or α-sarcoglycan at the protein level or gene transcription level. Such biochemical or genetic partners could provide insights into novel pathways or pathophysiological cascades.

For dystrophin deficiency, 131 genes showed significant expression changes relative to normal muscle, yet were assigned as “no change” in α-sarcoglycan deficient muscle. Similarly, α-sarcoglycan deficient muscle showed 89 genes with difference calls that were not seen in dystrophin deficiency. However, the large majority of these genes, in both dystrophin deficiency and α-SGD, showed relatively small changes near the twofold cutoff for statistical acceptance of a difference call. Thus, we felt it was likely that many, if not most, of these potential disease-specific expression changes represented experimental and biological “noise” and not biochemical or genetic “partners” for dystrophin or α-sarcoglycan.

To focus on those gene expression changes most likely to represent biochemical or genetic “partners,” we selected only those genes showing fivefold or greater changes specifically for either dystrophin deficiency or α-SGD (Table). Using these criteria, only a relatively few genes showed disease-specific changes in gene expression (12 upregulated and 4 downregulated genes specific for dystrophin deficiency, and 11 upregulated and 9 downregulated genes specific for α-SGD).

Dystrophin-deficient patient biopsies showed a specific decrease in dystrophin mRNA (fourfold) that was not seen in α-SGD, suggesting that primary genetic defects can potentially be identified by expression profiling. However, there were other genes showing similar or greater disease-specific differences. Two disease-specific changes were an extracellular signal regulated kinase (ERK6; 10-fold decrease) and a protein tyrosine phosphatase (8-fold decrease). For α-SGD, nine genes showed disease-specific decreases. For example, distinct sets of probes for the uncoupling protein3 gene detected nine- and sevenfold decreases in α-sarcoglycan-deficient, but not dystrophin-deficient patients.

Importantly, there were no significant gene expression changes of well documented secondary protein deficiencies, such as dystroglycan, sarcoglycans (β, δ, and γ), or nNOS. Also, there was not a change in utrophin RNA levels in either Duchenne or α-sarcoglycan dystrophies. On the other hand, another dystrophin-binding protein, α1-syntrophin, showed dramatic reductions in RNA levels in both dystrophin deficiency (14-fold) and α-SGD (6-fold), and α-sarcoglycan showed 3–5-fold reductions in both diseases as well.

A recent report has shown that ERK6 associates with α1-syntrophin, and it is also known that the PDZ domain of α1-syntrophin interacts directly the COOH terminus of dystrophin (Hasegawa et al. 1999). Thus, some of the biochemical partners of dystrophin (α1-syntrophin and ERK6) are also seen to be part of a coordinately regulated transcriptional group, as all three genes show dramatically reduced levels of RNA. Our data suggests this coordinately regulated gene cluster may also include a protein tyrosine phosphatase as a similar disease-specific mRNA reduction is seen.

To confirm our expression array findings, we chose a series of differentially regulated genes to study by immunostaining patient muscle biopsies. This approach also allowed us to identify the localization of the differentially regulated gene product.

Serial 4-μm-thick frozen muscle sections were processed for immunostaining with antibodies against 15 protein products of differentially expressed genes. All were tested on both unfixed and acetone-fixed sections of normal controls, DMD, and α-sarcoglycan–deficient patient muscles. A subset of antibodies were also tested on female manifesting carriers of DMD (somatic mosaic for dystrophin expression in muscle) and unrelated muscular dystrophies of known causes (calpain III deficiency and partial merosin deficiency). Of the 15 antibodies tested, 9 provided adequate signal/noise ratios for interpretation of protein localization and amount. Results of antibody studies are summarized in Table.

Factor XIIIa is a protein known to be involved in blood coagulation as a fibrin cross-linker. We found upregulation of this gene by 11–26-fold in both muscular dystrophies, though the expression in normal muscle was undetectable by GeneChip^® array studies, hence possibly exaggerating the extent of upregulation. Immunolocalization of this protein showed positive cells in both epimysial and endomysial connective tissue in dystrophic muscle (Fig. 5). Double immunostaining with a marker for endothelium (laminin α1) showed that the staining for factor XIIIa did not colocalize with this protein, though the factor XIIIa–positive cells were often in close proximity to blood vessels (data not shown).

HLA-DR is a histocompatibility antigen highly expressed in antigen presenting cells. HLA-DRα was upregulated in both dystrophin-deficient (threefold) and α-sarcoglycan–deficient (threefold) patient muscle. Immunostaining for HLA-DRα showed strong immunolocalization to a subset of cells that resembled those immunostained by factorXIIIa. Indeed, double immunostaining for both factor XIIIa and HLA-DRα showed that most positively stained cells coexpressed these two proteins (Fig. 5).

This data suggested that these cells represented tissue dendritic cells (Sueki et al. 1993). To test this, immunostaining was carried out with markers for circulating dendritic cell subtypes (CD1a, CD1b, and CD1c). Many of the factorXIIIa/HLADR-positive cells also stained with CD1a, and less frequently with CD1b and CD1c (data not shown). The data suggests that the infiltrating cells responsible for expression of factorXIIIa and HLA-DRα are related to tissue (dermal) dendritic cells. This is the first report of factor XIIIa+ and HLA-DR+ dendritic cell infiltration in dystrophic muscles.

Thrombospondin 4 is an extracellular matrix calcium-binding protein particularly abundant in tendon and early osteogenic tissues. It has been shown to be upregulated in denervated muscle, though its function is poorly understood. Thrombospondin 4 was 15-fold increased in dystrophin deficiency, and 23-fold increased in α-SGD. Immunostaining of dystrophic patient muscle biopsies showed thrombospondin 4 to be localized to areas of macrophage infiltration, though the areas showing very strong staining for thrombospondin 4 extended beyond frankly necrotic regions (Fig. 6, D–G). This data suggests that thrombospondin 4 is expressed by interstitial cells in response to macrophage infiltration, denervation, and/or cellular damage of neighboring myofibers (Arber and Caroni 1995).

SPARC/osteonectin is an extracellular glycoprotein that is strongly expressed during development and tissue regeneration, where it functions to mediate connections between cells and the extracellular matrix (Lane and Sage 1994). SPARC showed fivefold upregulation in dystrophin deficiency, and fourfold elevation in α-SGD. Immunolocalization showed punctate immunostaining that was dramatically increased in the endomysial and perimysial connective tissue (data not shown).

Versican is a chondroitin sulfate proteoglycan that, like SPARC and thrombospondin, is prevalent in myogenesis of muscle (Carrino et al. 1999). This gene showed eightfold upregulation in both dystrophies. Immunolocalization identified diffusely increased amounts of the protein in endomysial, but not perimysial, connective tissue of dystrophic muscle (Fig. 6, A–C).

Both α-cardiac actin and embryonic myosin heavy chain are specific isoforms of proteins that are transiently expressed during normal muscle development and regeneration (Whalen et al. 1979; Toyofuko et al. 1992). Both of these proteins showed upregulation in dystrophic muscle: α-cardiac actin was increased 7–9-fold and embryonic myosin 124–140-fold. Immunolocalization of these proteins in muscle biopsies, similar to those used for expression profiling, showed high-level expression of embryonic myosin in ∼20% of dystrophic myofibers, and of α-cardiac actin in ∼80% of fibers (Fig. 7).

Overt degeneration/regeneration of myofibers is a relatively rare event by histological assays (Fig. 6), and the large proportion of myofibers positive for these developmentally specific isoforms did not seem to be justified by the limited amount of regeneration in the dystrophic muscle biopsies. To test the association of myofiber regeneration with the amount of α-cardiac–actin positive myofibers, we studied both female mosaics for dystrophin deficiency (Fig. 7D and Fig. E) and a series of patient muscle biopsies from DMD patients (fetal, neonate, 3-, 5-, and 8-yr old), normal controls, merosin deficiency, and calpain deficiency (Fig. 7 F). In the manifesting carrier of dystrophin deficiency (Fig. 7D and Fig. E), both dystrophin-positive and dystrophin-negative myofibers were strongly positive for α-cardiac actin, and most appeared to be fully developed myofibers, suggesting that α-cardiac–actin expression persisted beyond the point of complete myofiber regeneration.

Immunostaining of α-cardiac actin in both normal and dystrophin-deficient fetal muscle (∼18–22-wk gestation) showed 100% of myofibers to be positive for α-cardiac actin. By birth, the proportion of α-cardiac–actin positive myofibers in normal muscle declined to 0%, whereas in dystrophin-deficient muscle, ∼60% of fibers remained strongly positive (Fig. 7 F). This high level was maintained throughout the disease process. Two unrelated dystrophic controls, partial merosin deficiency and calpain III deficiency, showed lower levels of α-cardiac actin (∼15–20% positive fibers). As dystrophin-deficient neonatal muscle shows relatively little evidence of degeneration/regeneration by histopathology, we conclude that the overexpression of α-cardiac actin shows persistent expression beyond the normal windows of development and regeneration.

Here, we report expression profiling as an experimental approach to define the biochemical cascades underlying the progressive pathophysiology of the inherited muscular dystrophies. A critical aspect of expression profiling studies is to limit the number of variables under study. This becomes highly problematic in the study of human pathological tissue in that there is considerable heterogeneity within tissue biopsies, differences in the genetic background between unrelated patients (genetic noise), and many other uncontrolled variables. One method to control for these variables is to study large numbers of patient samples, and then use elaborate statistical analyses to parse specific variables. This approach requires large numbers of analysis, which is costly and can be subject to statistical artifacts.

We present a novel approach for expression profiling of human patient pathological tissues, which we show is successful in reducing the effect of many uncontrolled variables on the resulting data, and thereby uncovering significant gene expression changes. This method employs tissue biopsies from groups of patients with known primary genetic defects that are age matched; the primary biochemical defect is held constant, and, thus, the major variable under study is controlled. All sources of experimental variability, including genetic background differences and tissue heterogeneity, were attenuated through the use of two different regions of each biopsy and mixing equal amounts of target cRNA from unrelated patients (Fig. 1). We show that it is critical to have iterative comparisons of data replicates; ∼50% of expression changes identified by one pairwise comparison were not consistently detected in comparisons of multiple data sets. We showed this iterative method to be accurate in detecting differentially expressed genes; all differentially expressed genes tested by protein immunostaining of patient tissues validated the expression profiling data.

We also present a novel application of expression profiling where two biochemically related primary genetic defects are compared to identify potential members of transcriptional pathways. In this case, patients with primary genetic defects of two associated proteins, dystrophin and α-sarcoglycan, were compared; these disorders show nearly identical clinical, histological, and functional defects. Some of the differentially expressed genes are indeed biochemical partners, and our data shows that these corresponding genes are associated in a transcriptional regulatory pathway (ERK6, α-syntrophin, and dystrophin). Moreover, our data provides the first proof of principle that primary genetic defects may be able to be identified by expression profiling. We found a specific fourfold decrease in dystrophin RNA in Duchenne dystrophy biopsies. We did not find a similar specific decrease in α-sarcoglycan gene expression in α-sarcoglycan–deficient patients, however, this is likely due to the fact that these patient had missense mutations that typically do not show nonsense-mediated decay. On the other hand, nearly all Duchenne dystrophy patients show frameshift mutations or stop codons that result in nonsense-mediated decay of the dystrophin mRNA (Hamed et al., 2000).

We provide insights into the disease process of these related muscular dystrophies at multiple levels: (a) a general metabolic crisis in patient muscle, (b) novel proteins involved in early stages of myofiber necrosis, (c) novel cells and proteins involved in local inflammatory and bystander responses, and (d) persistent expression of developmentally regulated genes suggesting that muscle may assume a chronically dedifferentiated state. Whereas expression profiling is widely felt to provide the parallel data generation needed to understand many complex biological processes, previous reports of expression profiling have been limited to physiological responses of microorganisms to environmental change and the subclassification of human cancers (Alon et al. 1999; Golub et al. 1999; Jelinsky and Samson 1999). To our knowledge, we present the first report of expression profiling in patient tissues with monogenic defects. Below, we briefly discuss some of the more dramatic changes we identified in the dystrophies, focusing on those that are cell autonomous and those that involve tissue microenvironment (non-cell autonomous). Many additional interpretations of the data obtained are possible, however, all data is deposited on a web site for public access for further analyses (http://microarray. CNMCResearch.org/resources.htm).

Many genes involved in mitochondrial function and energy metabolism were found to be reduced by twofold or more (26 genes), suggesting generalized mitochondrial dysfunction and metabolic crisis. Mitochondrial dysfunction has been reported previously, using a variety of assays in both human dystrophy patients and animal models, including 31P-NMR studies (Barbiroli et al. 1992) and biochemical assays (Gannoun-Zaki et al. 1995; Kuznetsov et al. 1998). It is widely felt that the chronic calcium influx due to the poor integrity of the plasma membranes of dystrophic muscle leads to calcium overloading of the mitochondrial matrix, as well as decreased mitochondrial function. Our data shows that much of this decreased function is due to reduced transcription of nuclear-encoded mitochondrial genes. The finding of a metabolic defect agrees with the beneficial effect of performance-enhancing metabolic agents such as creatine, coenzyme Q, carnitine, and others on dystrophic muscle function (Granchelli et al. 2000) (J.A. Granchelli, personal communication).

We noted that many important calcium-regulated signaling molecules were downregulated at the transcriptional level, such as adenylate cyclase (sixfold), calmodulin-dependent protein kinase (fivefold), and phospholipase C (twofold). Presumably, this reflects a negative-feedback loop where the chronic calcium influx from membrane instability over stimulates calcium-sensitive pathways, leading to a compensatory reduction in transcription of the corresponding genes. The downregulation of these genes might also downregulate down stream protein kinase A and C that are known to regulate expression of mitochondria genes and energy metabolism.

We found very high expression of many developmentally regulated genes, including α-cardiac actin (8-fold), embryonic myosin heavy chain (100-fold), versican (8-fold), perinatal myosin heavy chain (80-fold), acetylcholine receptor α 1 (12-fold), embryonic myosin light chain (10-fold), and others (Fig. 4). At first, we suspected that these simply reflected regenerative processes in the muscle, as it is well known that myofiber regeneration recapitulates much of the developmental program of muscle fibers. However, a variety of observations suggested that expression of these developmentally regulated genes was not limited to actively regenerating myofibers, and, instead, reflected persistent overexpression of these genes. First, overexpression of α-cardiac actin was present in 60–80% of DMD muscle fibers, whereas only a small subset of these showed histological evidence of regeneration (Fig. 7). Second, neonatal myofibers from normal muscle showed no α-cardiac actin, whereas 60–80% of DMD myofibers were positive. Third, mature dystrophin-positive myofibers in female mosaics also showed high expression of α-cardiac actin, suggesting that the protein was diffusing from adjascent dystrophin negative or regenerating regions, or that signals resulting in overexpression were tramsmitted within mosaic fibers. A previous study of α-cardiac–actin gene transcription in rat muscle showed that denervated muscle induced α-cardiac actin mRNA to levels which were sixfold higher than normal, and, thereafter, maintained these levels at a constant level (Toyofuko et al. 1992). Regenerating muscle from an autograft resulted in mRNA levels that were initially increased by 40-fold, but then dropped to normal levels within 40 d. Taking these observation together, it is possible that degeneration/regeneration cycles in dystrophic muscle leads to high levels of α-cardiac actin mRNA and protein that diffuse within syncytial myofibers. However, our finding of the majority of neonatal fibers from preclinical DMD patient biopsies to be positive for α-cardiac actin suggests that these may be a persistent overexpression of this protein in dystrophin-deficient and α-sarcoglycan–deficient myofibers. We propose a model where this persistent expression is due to a chronic state of undifferentiation induced by altered Ca²+ signaling within the myofiber. Consistent with this model, it has been recently shown that altering the pattern of Ca²+ influx in differentiating myogenic cells can determine the developmental pathway that the cells will take (Naya et al. 2000).

A second result suggesting dedifferentiation of dystrophic muscle was the upregulation of versican (eightfold). Versican is one of the large chondroitin sulfate proteoglycans that has been shown to be important during the early development of muscle (Carrino et al. 1999). Muscle tissue switches from large proteoglycans, early in development, to small proteoglycans later. Versican is expressed during regeneration, but our immunostaining results showed high level, persistent expression of versican protein throughout the endomysial connective tissue in both dystrophin deficiency and α-sarcoglycan–deficient muscle, consistent with the dramatic upregulation of RNA levels shown by the expression array studies (Fig. 6). Importantly, versican has been shown to stimulate the proliferation of chondrocytes (Zhang et al. 1999); we also found upregulation of other chondrocyte- and bone-related transcripts, such as OSF-2 (4-fold), matrix Gla protein (7-fold), osteopontin (13-fold), and serine protease 11 (3-fold). Our results suggest that persistent expression of developmentally regulated genes, coupled with chronic calcium influx in dystrophin-deficient myofibers, leads to altered development and regeneration of myofibers in dystrophic muscles.

The acetylcholine receptor α1 subunit was also dramatically upregulated in patient muscle (19-fold). This protein is known to increase at extrajunctional sites during both normal development and denervation atrophy (Fischer et al. 1999), however, similar to α-cardiac actin, the extent of induction of the acetylcholine receptor seemed too large to explain by the small amount of regenerating fibers and the few atrophic fibers seen in the biopsies. Several studies have demonstrated that deficiency of dystrophin and the sarcoglycan complex disturb the muscle fiber–laminin interaction and the stability of neuromuscular synapse (Brown et al. 1999; Grady et al. 2000), and that this leads to morphological changes of the junction (Zaccaria et al. 2000). These findings are consistent with two possible models: either myofibers are in a persistent undifferentiated state, or motor neurons are unable to maintain strong attachments to myofibers, leading to a constant stimulation of denervation signal pathways.

This is the first report of factor XIIIa+ and HLA-DR+ dendritic cell infiltration in dystrophic muscles. MHC class II+ dendritic cells have been reported at low level in normal and regenerating rodent muscles (Pimorady-Esfahani et al. 1997), and muscle dendritic cells have been shown to be important in immune aspects of gene delivery to muscle (Jooss et al. 1998). However, the expression profiling and immunophenotyping we present here shows that the dendritic cell infiltrate in dystrophic muscle is that of activated dermal dendritic cells. In skin, similar subpopulations of XIIIa+ and HLA-DR+ dermal dendritic cells have been shown to be closely associated with tissue mast cells, and increased factor XIIIa protein expression in dendritic cells has been observed in response to mast cell degranulation (Sueki et al. 1993). We have shown previously that in muscle, extensive mast cell proliferation and degranulation in dystrophin-deficient human, dog, and mouse muscle (Gorospe et al. 1994a,Gorospe et al. 1994b). Also, we have shown that dystrophin-deficient myofibers are acutely sensitive to mast cell inflammatory mediators and proteases, showing widespread grouped necrosis when exposed to mast cell granules; this is presumably due to exacerbation of the membrane defect by the mast cell mediators (Gorospe et al. 1996). Our new results suggest that dendritic cells and mast cells may act coordinately to mediate chronic and acute microenvironmental changes in dystrophic muscles. These findings may have relevance to current treatment of Duchenne dystrophy patients with steroids. Prednisone is well documented to slow the progression of DMD and α-SGD, yet the inflammatory cell target of prednisone has not been identified, as drugs which inhibit T and B cell function do not seem to benefit DMD patients (Griggs et al. 1993; Connolly et al. 1998). Prednisone recently has been shown to prevent activation of dendritic cells (Matasic et al. 1999). Thus, our findings suggest that prednisone may improve patient muscle function by acting on dendritic cell pathways.

We identified a series of novel upregulated proteins in the endomysial connective tissue of dystrophic muscle that are traditionally considered developmentally specific isoforms, namely thrombospondin IV, SPARC, and versican. Each of these proteins are also known to be important in regeneration and/or denervation of muscle. However, we found that expression of these proteins was not limited to regions of regenerating myofibers, suggesting that these, like α-cardiac actin and acetylcholine receptor α1, show a persistent and/or chronic overexpression in dystrophic muscle. Importantly, each of these three proteins have been found to have a role in tissue remodeling and cell migration; these microenvironmental changes could exacerbate the effects of dystrophin deficiency in muscle. The thrombospondins are a family of extracellular calcium-binding proteins that are involved in cell proliferation, adhesion, and migration (Arber and Caroni 1995; Newton et al. 1999). SPARC/osteonectin is a secreted glycoprotein that contains modular domains that can function independently to bind cells and matrix components. Because SPARC can selectively disrupt cellular contacts with matrix and, thereby, effect changes in cell shape, it has been referred to as an anti-adhesin (Lane and Sage 1994). Versican is involved in stimulating cell proliferation and migration in both smooth muscle and chondrocytes (Evanko et al. 1999; Zhang et al. 1999).

The global gene expression analyses presented here add critical new information to existing pathophysiological models of dystrophin and α-SGD (Fig. 8). Our results suggest that developmental reprogramming, both in the form of failure to deactivate developmentally regulated genes and the persistent activation of proteins involved in development and regeneration, leads to a series of both cell autonomous, and non-cell autonomous (microenvironmental) changes. We hypothesize that these changes lead to the progressive aspects of the human disease and phenotypic differences in the orthologous animal models. Future expression profiling experiments will focus on the correlation of specific pathological cascades with patient age and the degree of disability, as well as cross-species experiments in animal models. These will help distinguish between regeneration cascades, chronic undifferentiated status, and abnormal differentiation pathways in dystrophic muscle.

Our results also show that it is possible to identify primary genetic defects in patient tissue through analysis of carefully controlled expression profiling experiments. In this instance, we were able to show a disease-specific reduction of dystrophin mRNA in Duchenne dystrophy patient muscle (fourfold reduction) compared with α-sarcoglycan deficient muscle. In the future, molecular diagnostics is likely to be possible with expression profile “fingerprints” of patient muscle. These profiles would reflect both generic responses to pathological processes in muscle and specific changes reflecting the primary biochemical defect. Critical in the accuracy and sensitivity of such testing will be the public availability of large numbers of expression profiles of patients with known genetic defects.

The authors are indebted to the Affymetrix Academic User Center for assistant with initial expression profiling (Drs. Chris Harrington, Gene Tanimoto, Uyen Truong, and Sumathi Venkatapathy), and to Dr. Roma Chandra, Department of Pathology (CNMC), for assistance with control biopsies. The authors thank Drs. Louise Anderson, Jack Lawler, and Fernando Vivanco for kind provision of dystrophin, thrombospondin 4, and C3 antibodies, and Sheila Caldwell for assistance with interpretation of dendritic cell data. The authors also thank Steve Engratt for establishment of the microarray. CNMCResearch.org web site. Kevein Kathrotia assisted with analysis of interative data comparisons.

Dr. Chen is supported by a Duchenne Muscular Dystrophy Research Center postdoctoral fellowship from Stichting Porticus. Supported in part by a grant from the National Institutes of Health (NIH)/National Institute of Neurological Disorders and Stroke (3RO1 NS29525-09). The Affymetrix Academic User Center was supported in part by a grant from NIH (PO1-HG0132).