Fig 1.
Titration of MBNL1 using an inducible tet-on system.
(A) HEK293 cells were stably transfected to integrate a tetracycline inducible HA-MBNL1 expression construct into the genome. Addition of the tetracycline analog, doxycycline, de-represses inhibition of transcription by binding to the tet Repressor resulting in expression of full length HA-MBNL1 mRNA. (B) MBNL1 immuno-blot showing MBNL1 protein gradient resulting from dox (ng/ml) titrations. GAPDH serves as a loading control. The first lane represents treatment with control siRNA and the second lane represents treatment with a pool of siRNA against MBNL1. (C) Quantification of the MBNL1 immunoblot (in triplicate) plotted against log[dox]. (D) Schematic showing a theoretical example of the dose-dependent relationship between log[MBNL1] and percent spliced in (psi, Ψ), where MBNL1 levels are determined by Western relative to GAPDH. Curve fitting parameters EC50, slope, Ψmin, and Ψmax are illustrated.
Fig 2.
(A-G) Dose-response curves and splicing gels of individual events (A) MBNL2 (B) MBNL1 (C) ATP2A1 (D) CLASP1 (E) FN1 (F) INSR (G) NFIX. Dox (ng/ml) was titrated to induce MBNL1 expression in HEK293 cells or treated with control siRNA or siRNA against MBNL1. siRNA treatment is not shown for events with transcripts that would be targeted by knock-down (MBNL1/2). RNA was isolated from the cells, RT-PCR was performed and DNA products were resolved on a native gel and isoform ratios quantified. Ψ values were plotted against the log(HA-MBNL1) treatment and fitted to a four-parameter dose-response curve. Splicing assays and western blots for MBNL1 protein quantification were performed in triplicate. (H) Curve fitting parameters are shown in the table.
Fig 3.
The composition of MBNL1 binding sites mediates dose-dependency of MBNL1 exon 5.
(A) Mini-gene reporter sequence (exons 3-4-5 and intervening introns) of MBNL1 pre-mRNA showing the 3' region of the intronic, ultraconserved sequence between exon 4 and the regulated, or alternative, exon 5. YGCYs are indicated by colored boxes. The previously mapped distant branch point adenosine is indicated with an A [24]. Asterisks indicate C-T mutations from C2C12 cells (CLIP data, [29] (B-G) Mini-gene reporter dose-response curves were plotted (Ψ verses the log([HA-MBNL1]), determined by western blot) for (B) del1 (C) del2 (D) del3 (E) del4 (F) del5 (G) 3M in triplicate. The WT (blue) dose-response curve is included with each deletion mutant (red) for comparison. Quantification of the upper band (exons 4-5-6) and lower band (exons 4–6) were used to determine Ψ. A transiently transfected HA-MBNL1 plasmid was used to achieve the highest MBNL1 dose in this experiment and CTG960 transient transfection was used to achieve the lowest levels of functional MBNL proteins through sequestration. Representative splicing gels are shown.
Table 1.
Curve fitting parameters of MBNL1 exon 5 deletion mutant mini-genes.
Parameters for EC50, slope, and R2 for the MBNL1 deletion mutants were derived from curve fitting using a four-parameter dose-response curve.
Fig 4.
Inferring MBNL concentration using Ψ in HEK293 and DM1 tibialis anterior.
(A) [MBNL] inferred using Bayesian estimation highly correlates to measurements of MBNL protein relative to GAPDH as assessed by Western blot in the HEK293 system. (B) Total splicing dysregulation, mean ΔΨ, correlates strongly with measurements of MBNL protein relative to GAPDH in HEK293. (C) Bayesian estimation was used to infer [MBNL] in 55 tibialis anterior biopsies; [MBNL]inferred values also strongly correlate with total splicing dysregulation, mean ΔΨ, in tibialis. Non-DM1 individuals are shown in green. (D) Heat map of normalized Ψ (each event was set from 0 to 1, to aid visualization) for forty-six splicing events in forty-four DM1 patients and eleven healthy controls. Similar, non-normalized data is shown in S4 Fig. Ψmin, Ψmax, EC50, and slope values were inferred simultaneously with [MBNL] using Bayesian estimation (S3 Table). Samples are sorted by [MBNL]inferred across the horizontal axis, and splicing events are sorted by EC50 along the vertical axis. The |slope| for each event is indicated on the right-hand vertical axis. White boxes with slashes denote samples with insufficient read coverage to infer Ψ. Events studied in HEK293 are marked with an asterisk.
Fig 5.
Splicing events as biomarkers to measure functional MBNL concentration in DM1 muscle and response to therapeutics.
(A) Ψmin, Ψmax, EC50, and slope, as well as [MBNL], were estimated for each splicing event and each sample, using 70% of the tibialis biopsies using a Bayesian inference framework. Sigmoid curves with 95% confidence intervals for NFIX and CLASP1, as estimated from 70% of the data, are shown, along with the Ψ values used to derive the curves (black points). Posterior distributions for [MBNL] were derived for 30% of the data (red points), and plotted for 3 specific samples (blue, green, and orange points). These distributions are also plotted on the right, along with the “gold standard” [MBNL] as estimated using 100% of the data (blue, green, and orange vertical lines). (B) The mean predictive power of each splicing event to predict “true” [MBNL] was calculated across 120 random subsets of training and test sets, where 70% of samples were used for training, and 30% for testing. Predictive power was defined as the posterior probability estimate at “true” [MBNL] assessed using all the data. The mean predictive power was computed separately across three patient subgroups–severe ([MBNL] < 0.33), moderate (0.33 < [MBNL] < 0.66), and mild ([MBNL] > 0.66) DM1, as well as across the entire patient cohort. Mean predictive power for each patient subgroup was plotted versus mean predictive power across the entire cohort, or all patients; splicing events that perform better in specific subgroups relative to the entire cohort are labeled. (C) The ability to estimate changes in [MBNL] depends on the splicing event used to infer [MBNL], as well as the disease severity at which treatment is initiated. Hypothetical changes in [MBNL] of +0.4 and +0.3 during a therapeutic trial are illustrated as purple and brown filled and outlined points, respectively in left panels. The posterior probability estimates of [MBNL] are illustrated in right panels. (D) A greater number of biomarkers can improve predictive power of estimates of [MBNL]. The posterior probability distribution for [MBNL] is shown when using 1, 2, 5, or 10 biomarkers for severely affected tibialis (top panel) or moderately affected tibialis (bottom panel). The value of [MBNL] obtained when using 100% of samples is shown as a dotted line; the posterior probability at this value of [MBNL] increases as the number of biomarkers increases. (E) Mean predictive power increases when using more biomarkers, up to a point. The best possible combination of biomarkers was chosen for each cross-validation trial, and predictive power was averaged across 120 cross-validation trials.
Fig 6.
(A) In our HEK293 inducible model, free [MBNL1] can be titrated by administration of doxycycline (top panel). In myotonic dystrophy, free [MBNL] proteins are titrated by sequestration to toxic CUG RNA (bottom panel). (B) Deletion of YGCY sites within the ultraconserved intron upstream of MBNL1 exon 5 influences dose-dependent curve parameters EC50, slope, both, or neither. (C) Biomarkers are most informative in the regions with a non-zero slope (red parts of curve). The best biomarkers also tend to have a broad dynamic range (Ψmax- Ψmin).