Facioscapulohumeral dystrophy transcriptome signatures correlate with different stages of disease and are marked by different MRI biomarkers

Abstract

With several therapeutic strategies for facioscapulohumeral muscular dystrophy (FSHD) entering clinical testing, outcome measures are becoming increasingly important. Considering the spatiotemporal nature of FSHD disease activity, clinical trials would benefit from non-invasive imaging-based biomarkers that can predict FSHD-associated transcriptome changes. This study investigated two FSHD-associated transcriptome signatures (DUX4 and PAX7 signatures) in FSHD skeletal muscle biopsies, and tested their correlation with a variety of disease-associated factors, including Ricci clinical severity score, disease duration, D4Z4 repeat size, muscle pathology scorings and functional outcome measures. It establishes that DUX4 and PAX7 signatures both show a sporadic expression pattern in FSHD-affected biopsies, possibly marking different stages of disease. This study analyzed two imaging-based biomarkers-Turbo Inversion Recovery Magnitude (TIRM) hyperintensity and fat fraction-and provides insights into their predictive power as non-invasive biomarkers for FSHD signature detection in clinical trials. Further insights in the heterogeneity of-and correlation between-imaging biomarkers and molecular biomarkers, as provided in this study, will provide important guidance to clinical trial design in FSHD. Finally, this study investigated the role of infiltrating non-muscle cell types in FSHD signature expression and detected potential distinct roles for two fibro-adipogenic progenitor subtypes in FSHD.

Figure 1

Representative examples of the two MRI-based imaging biomarkers used in this study; fatty infiltration and TIRM hyperintensity. (a) Axial T1-weighted and (b) TIRM image of the left upper leg of a 50-year-old FSHD patient (FSHD-09) showing marked fatty infiltration of nearly all muscles and focal hyperintensity in the VL muscle. The MRI-guided biopsy sites are marked with the yellow circles. In T1: Normal muscle is dark grey, fat infiltrated muscle is white. Note the relative sparing of the sartorius muscle and the severe fatty infiltration of the posterior compartment and quadriceps. (c) HPhlox staining of the Fat^POS/TIRM^POS biopsy from the same patient demonstrates severe dystrophic changes indicated by a marked increase in fiber size variability, increased internal nuclei, regenerating fibers and fatty infiltration. (d) HPhlox staining of the Fat^POS/TIRM^NEG biopsy from the same patient shows fiber size variability, increased internal nuclei and few regenerating fibers corresponding to mild dystrophic changes. (e) TIRM hyperintensity frequencies in muscle biopsies of each disease state. p-values depict a Fisher’s exact test result (excluding the replicate samples). (f) Fat fractions of all individual muscle biopsies grouped by disease state. p-values depict a Mann–Whitney U test results (excluding the replicate samples). The red dashed line indicates the classification threshold for Fat^POS versus Fat^NEG FSHD muscle biopsies (15% fat fraction) and the boxplots on the right depict the FSHD biopsies separated based on this classification. p-values: ns = not significant, *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001. MRI scans were exported from Agfa IMPAX (https://global.agfahealthcare.com). Histology sections were digitized using a Philips UFS (www.usa.philips.com/healthcare/resources/landing/philips-intellisite-pathology-solution) and images were assessed using 3DHISTECH’s CaseViewer (v2.3, www.3dhistech.com). All data plots are generated in R (v4.0.3, www.R-project.org) using the packages gplots (v3.1.1) and ggplot2 (v3.3.3). Figure and panel layout was further adapted in Adobe Illustrator CC 2018 (www.adobe.com).

Figure 2

DUX4 and PAX7 signature expression in FSHD and control biopsies. (a) DUX4 signature expression in FSHD and control muscle biopsies. The threshold criterium for DUX4^POS biopsy selection (cumulative normalized read count > 20) is marked with a red dashed line and the boxplots on the right depict the FSHD biopsies separated based on this classification. The two replicate FSHD samples sequenced in the two major sequence batches are highlighted in color. Both replicates are selected as DUX4^POS in both sequence batches, indicating that the absence of a DUX4 signature in the controls is not due to a sequencing bias. (b) PAX7 signature expression in FSHD and control muscle biopsies. As a classification threshold for PAX7-affected versus non-affected biopsies could not be clearly defined based on the scores in control samples, for further subgroup comparisons the ten FSHD samples with the highest PAX7 score were classified as PAX7^HIGH (i.e. likely non-affected) and the ten FSHD samples with lowest PAX7 scores were classified as PAX7^LOW (i.e. most-affected). The boxplots on the right depict the FSHD biopsies included based on this classification. The two replicate FSHD samples sequenced in the two major sequence batches are labeled and highlighted in color. (c) Receiver Operator Characteristic curve for FSHD versus control biopsy classification with either DUX4 or PAX7 signature expression. AUC; area under curve. p-values depict the results of a Student’s t-test (excluding the replicate samples). p-values: ns = not significant, *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001. All data plots are generated in R (v4.0.3, www.R-project.org) using the packages gplots (v3.1.1), ggplot2 (v3.3.3) and pROC (v1.17.0.1). Figure and panel layout was further adapted in Adobe Illustrator CC 2018 (www.adobe.com).

Figure 3

Linear correlation analysis between al metadata in this study’s dataset. (a) Pearson correlation scores r (top right values) and Spearman’s rank correlation scores r_s (bottom left) for all comparisons between the two molecular signatures (DUX4 and PAX7 signatures), the two imaging-based biomarkers (TIRM hyperintensity and fat fraction) and all metadata of our dataset (see also methods). (b) p-values for all correlations. Color-coding follows the correlation values and specific scores are indicated in each box. Grey boxes indicate that no linear correlation score could be calculated. Note that all results for non-muscle-specific metadata (i.e. age, age at onset, disease duration, group, CSS, D4Z4 repeat size, BMI, sex, 6-MWT and MFM) may be biased by duplicate samples for participants that donated a muscle biopsy from both the TA and VL muscle. The two duplicate VL muscle biopsies from participant FSHD-09 and FSHD-13 are also included in the data. This analysis indicates the strength of linear (for Pearson) and monotonic (for Spearman) correlations. For some metadata, other correlations may still apply. See also Supplementary Figs. S3 and S4 for detailed visualization of all quantitative correlations in this analysis. Pearson and Spearman’s rank correlation scores and p-values were calculated in R (v4.0.3, www.R-project.org) with the cor.test function of the stats package (v4.0.3). Data is plotted with the heatmap.2 function of R package gplots (v3.1.1). Figure and panel layout was further adapted in Adobe Illustrator CC 2018 (www.adobe.com).

Figure 4

Correlations between DUX4 and PAX7 signature expression and TIRM hyperintensity and fat fraction. (a–c) DUX4 signature expression levels in control biopsies and FSHD biopsies separated based on imaging biomarker classifications, with in (a) TIRM hyperintensity, in (b) Fat fraction (Fat^POS: > 15% Fat), an in (c) the combined biomarker (TIRM^POS and/or Fat^POS). The threshold criterium for DUX4^POS biopsy selection (cumulative normalized read count > 20) is marked with a red dashed line. The stacked bar plots on the right depict the relative frequency of DUX4^POS biopsy classification in each subgroup. The numbers indicate the true frequencies. (d–f) PAX7 scores in control biopsies and FSHD biopsies separated based on imaging biomarker classifications, with in (d) TIRM hyperintensity, in (e) Fat fraction (Fat^POS: > 15% Fat), an in (f) the combined biomarker (TIRM^POS and/or Fat^POS). All p-values for the quantitative comparison of signature expression (boxplots) depict the results of a Student’s t-test comparing imaging biomarker-positive versus imaging biomarker-negative FSHD muscle biopsies respectively. For the frequency plots, the result of a Fisher’s exact test is depicted. For reference, in (a, d) Fat^POS biopsies are highlighted (green), in (b, e) TIRM^POS biopsies are highlighted (red) and in (c, f) both TIRM^POS (red), Fat^POS (green) and TIRM^POS/Fat^POS (blue) biopsies are highlighted. p-values: ns = not significant, *p-value < 0.05, **p-value < 0.01, ***p-value < 0.001, ****p-value < 0.0001. All data plots are generated in R (v4.0.3, www.R-project.org) using the packages gplots (v.3.1.1) and ggplot2 (v3.3.3). Figure and panel layout was further adapted in Adobe Illustrator CC 2018 (www.adobe.com).

Figure 5

Significant non-muscle cell type contributions in FSHD, and their correlation with DUX4 and PAX7 signature expression. (a) Estimated relative contributions for all muscle and non-muscle cell types that show significantly different contributions in FSHD versus control muscle biopsies. See Supplemental Fig. S8 for the results of all (significant and non-significant) identified cell types. Results are based on the RNA deconvolution analysis (see “Methods” for details on PLIER analysis), based on cell types previously identified by Rubenstein AB et al. (in healthy human muscle biopsies). LV: latent vector best representing the respective cell type signature noted behind the LV number. FBN1 + FAPs: Fibrilin-1 positive fibro-adipogenic progenitors. LUM + FAPs: lumican-positive fibro-adipogenic progenitors. PCV-Endothelial cells: post-capillary venules endothelial cells. (b, c) Estimated relative contribution of the Type IIa myofiber content in controls and FSHD muscle biopsies showing a reduction in Type IIa myofiber content in both DUX4^POS versus DUX4^NEG FSHD biopsies (b) and in PAX7^LOW versus PAX7^HIGH FSHD biopsies (c). (d, e) Estimated relative contribution of the LUM⁺ FAP subtype in controls and DUX4^POS versus DUX4^NEG FSHD muscle biopsies (d) and FBN1⁺ FAP subtype in controls and PAX7^LOW versus PAX7^HIGH FSHD muscle biopsies (e). p-values in (b–e) depict the results of Mann–Whitney U tests. (f–i) Linear quantitative correlation analysis for both molecular signatures [DUX4 signature (f, h) and PAX7 score (g, i)] with each FAP subtype [FBN⁺ FAPs (f, g) LUM⁺ FAPs (h, i)], showing the strongest correlation of each molecular signature with a distinct FAP subtype. For quantitative correlations, only FSHD samples were included. Grey shadings indicate the 95%-confidence interval for the linear regression line. p-values and R² values depict the result of a Pearson correlation. p-values: ns = not significant, *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001. All data plots are generated in R (v4.0.3, www.R-project.org) using the packages gplots (v3.1.1), ggplot2 (v3.3.3) and ggpubr (v0.4.0). Figure and panel layout was further adapted in Adobe Illustrator CC 2018 (www.adobe.com).