Integrative Proteogenomics for Differential Expression and Splicing Variation in a DM1 Mouse Model

Abstract

Dysregulated mRNA splicing is involved in the pathogenesis of many diseases including cancer, neurodegenerative diseases, and muscular dystrophies such as myotonic dystrophy type 1 (DM1). Comprehensive assessment of dysregulated splicing on the transcriptome and proteome level has been methodologically challenging, and thus investigations have often been targeting only few genes. Here, we performed a large-scale coordinated transcriptomic and proteomic analysis to characterize a DM1 mouse model (HSA^LR) in comparison to wild type. Our integrative proteogenomics approach comprised gene- and splicing-level assessments for mRNAs and proteins. It recapitulated many known instances of aberrant mRNA splicing in DM1 and identified new ones. It enabled the design and targeting of splicing-specific peptides and confirmed the translation of known instances of aberrantly spliced disease-related genes (e.g., Atp2a1, Bin1, Ryr1), complemented by novel findings (Flnc and Ywhae). Comparative analysis of large-scale mRNA and protein expression data showed quantitative agreement of differentially expressed genes and splicing patterns between disease and wild type. We hence propose this work as a suitable blueprint for a robust and scalable integrative proteogenomic strategy geared toward advancing our understanding of splicing-based disorders. With such a strategy, splicing-based biomarker candidates emerge as an attractive and accessible option, as they can be efficiently asserted on the mRNA and protein level in coordinated fashion.

Graphical Abstract

Highlights

• •Performed large-scale proteogenomic analysis of DM1 mouse model (HSA^LR).
• •Identified and confirmed known and novel instances of aberrant mRNA splicing in DM1.
• •Showed agreement of dysregulated splicing patterns on transcript and protein level.

In Brief

Dysregulated mRNA splicing is involved in many diseases including myotonic dystrophy type 1 (DM1). Our proteogenomics analysis of DM1 mouse model (HSA^LR) comprised gene- and splicing-level assessments for mRNAs and proteins. It enabled the design and targeting of splicing-specific peptides and confirmed the translation of known instances of aberrantly spliced disease-related genes (e.g., Atp2a1, Bin1, Ryr1), complemented by novel findings (Flnc and Ywhae). Comparative analysis between transcriptomics and proteomics showed good quantitative agreement of differentially expressed genes and splicing patterns.

The molecular mechanism of mRNA splicing is essential for creating functional mRNA molecules from disjointly encoded genomic predecessors (exons) in eukaryotes. Alternative splicing (AS) is key for extending, varying, and tuning the arsenal of mRNAs and proteins in the context of developmental or tissue-specific functions. Dysregulation of splicing, however, is increasingly being recognized as a hallmark of disease, notably cancer (1), and also degenerative diseases and aging, with interesting implications for shared biological processes (2, 3, 4, 5, 6). On the transcript and protein level, such dysregulation manifests itself in splicing patterns altered in disease compared to healthy controls. This differential is measured with dedicated methods applied to high-throughput transcriptomics (Tx) and proteomics (Px) data. We are hence using the term differential alternative splicing (DAS) to describe instances of altered splicing between two comparison groups.

Modern high-throughput mRNA-seq has become a standard technology for assessing whole-transcriptome gene expression (GE) in biological samples and differential gene expression (DGE). For this work, to extend differential RNA-seq analyses to instances of altered splicing, or DAS, we specifically employed the LeafCutter software (https://github.com/davidaknowles/leafcutter) (7). This tool analyzes the exon junction-spanning reads of a genome reference-aligned RNA-seq dataset to identify all sets of splicing alternatives interconnected via one or more shared exons, the so-called “intron clusters.”

Proteome-scale investigations of cellular or tissue-based protein content have become a powerful tool for many biological and medical studies (8). Today, mass spectrometry (MS)-based shotgun Px allows identification of thousands of proteins within several hours (9, 10). In homogeneous samples from cell lines, the achieved comprehensive coverage of a sample’s protein content is approaching the routine performance of RNA-seq (11). However, in complex samples such as tissue, deep coverage is much harder to achieve, as tissue or biofluid proteomes are often dominated by a few highly abundant species. Muscle tissue in particular poses additional challenges due to the complexity and resilience of its tissue architecture (12, 13).

In MS-based Px, data-dependent acquisition (DDA) is a method where ionized peptides (precursors) are automatically selected and subjected to fragmentation (tandem mass spectrometry). DDA is a powerful discovery modality for profiling general protein composition and abundances. However, most of the proteins represented in DDA results are identified by only few tryptic peptides (14). Moreover, the overlap between peptides identified under different instrument settings is often modest due to automatic precursor ion selection. As a consequence, DDA is generally underpowered to systematically investigate a specific set of protein isoforms, although a few methods for AS analysis at the protein level using DDA were described recently (15, 16, 17).

As an alternative, targeted approaches (e.g., parallel reaction monitoring, PRM) can reliably quantify peptides with low abundance in a complex mixture. Synthesized analogues of predefined peptides of interest distinguishable via heavy isotope labeling are added to a sample and monitored during its retention time window. These techniques then permit to calculate with high accuracy the ratios of labeled peptides to their unlabeled counterparts naturally present in a sample based on fragment intensities (18). Targeted approaches have greater sensitivity and reproducibility than DDA and allow monitoring of up to hundreds of predefined peptides in one run. Moreover, peptide design can be custom-tailored to interrogate specific splice events (19), informed by sequence database annotation or, as demonstrated here, by adequately processed transcriptomic data.

DM1 (myotonic dystrophy type 1) is a degenerative disease without a cure. It is one of the most common inherited muscle disorder in adults, with a prevalence commonly reported to vary around 1/8000 (20), although a recent study ascertained a nearly 4-fold higher frequency, pointing to a significant potential for underdiagnosis (21). DM1 is a multisystem disorder that affects skeletal and smooth muscle as well as heart, eye, and the central nervous system. Clinical symptoms involve myotonia (impaired muscle relaxation, a hallmark of DM1), progressive distal muscle weakness and atrophy, cataracts, and cardiac conduction abnormalities. DM1 results from a trinucleotide (CTG) repeat expansion in the DMPK gene that is transcribed into RNA, with disease severity positively correlating with the length of the repeats. The excessive noncoding CUG repeats in the DMPK transcript bind proteins important for mRNA splicing. The resulting patterns of splicing abnormalities have been detected in many transcripts coding for proteins important for muscle function (22, 23, 24), among them chloride channels like CLCN1 (25, 26) and calcium channels and pumps like RYR1 (27), ATP2A1 (28), and CACNA1S (29).

The significant unmet medical need presented by DM1 underscores the importance of a thorough understanding of animal models used to study this disease. The Tg HSA^LR mouse model used in our study is characterized by a human ACTA1 (human skeletal actin) transgene with more than 250 CTG repeats (30). Posttranscriptionally, the RNA accumulates in the nucleus, where the CUG repeats form extended hairpin structures, which sequester AS modulators from the “muscleblind” (Mbnl) family (31), impacting splicing patterns (32, 33, 34). Similar to a more recently described inducible DM1 mouse model (35), the CUG repeat expansion is restricted to skeletal muscle. However, unlike human DM1 disease in this model the repeats are not associated with the DMPK gene. A better understanding of the splicing alterations in this DM1 mouse model and, ultimately, their similarities and differences to human disease are hence of prime interest.

To date, mostly only isolated individual proteins and protein complexes have been investigated in the context of DM1 disease (36, 37, 38, 39) with some exceptions specifically focused on the global proteome changes in neurological context (40, 41). This leaves a notable gap in our understanding of protein and specifically protein isoform changes in this pathological condition. Although the genetic cause of DM1 is known, a better understanding of the molecular disease processes in human and animal model systems is urgently needed to aid progress toward a cure.

To obtain a comprehensive picture of global GE in a biological system, the integration of available genomic, Tx, and Px data is key (42), reflecting an evolved meaning of the term “proteogenomics” since its original introduction (43). In this work, using an in vivo murine DM1 disease model, we describe in detail a highly integrative Tx and Px analysis with dedicated focus on DAS. Moving beyond simplistic assumptions, we demonstrate that the correspondence between the two technologies is good or even excellent, given the proper experimental constraints and statistical analyses. Our proteogenomic approach not only robustly recapitulates well-known splicing changes in DM1, but also identifies and confirms novel ones. It thus provides a viable path to establishing robust splicing-based transcript- and protein-level biomarker and drug target candidates in an integrative setting and is easily generalizable an actionable blueprint for investigations into other splicing-related diseases.

Experimental Procedures

Animals

Animal experiments were carried out in accordance with the guidelines of the Swiss Federal and Cantonal veterinary offices for care and use of laboratory animals. Described studies were approved by the Swiss Cantonal Veterinary Authority of Basel City, Switzerland and performed according to animal license number BS-2885.

Transgenic mice in HSA^LR line 20b were maintained as homozygotes on an FVB inbred background (30), supplied by Novartis Pharma AG, mouse colony MB828. FVB mice were used as wild-type (WT) animals and were purchased from Janvier Labs (Le Genest-Saint-Isle). The mice were housed at 25 °C with a 12:12 h light-dark cycle in groups of 2 to 4 animals and acclimated to the facility for 7 days. Food and water were provided ad libitum.

Experimental Design and Statistical Rationale

For all Tx and Px analyses (including qRT-PCR and Western blots), the right and left hind limb gastrocnemius muscles of five HSA^LR and five WT, 10 to 12-week-old male mice were dissected, snap-frozen, and stored at −80 °C until RNA and protein extraction. The same biological samples were used for transcriptomic and for both tandem mass tag (TMT)-based and targeted proteomic analyses. Five biological (n = 5) and no technical replicates for control (WT) and diseased (HSA^LR) groups were used for RNA-seq and DDA Px, since the biological variation between different animals is expected to greatly surpass the technical variability of these approaches. To improve peptide detectability in targeted Px experiments, two technical replicates with different gradients were generated in addition. Differential gene and protein expression analysis was performed using moderated t-statistics (limma test) with Benjamini–Hochberg correction for multiple testing. For the significance cut-off (adjusted p-value and fold changes) used, see the corresponding section.

Short-Read RNA-Seq

Next generation sequencing libraries were prepared with the TruSeq Stranded mRNA Sample Preparation kit (Illumina) from 350 ng of input RNA using IDT for Illumina TruSeq UD Indexes (IDT). Adapter-ligated fragments were amplified using 12 rounds of PCR. The resulting libraries were pooled and loaded on two lanes of a HiSeq 4000 (Illumina) for paired-end 76 bp sequencing, generating 71 million reads per sample on average. Reads were aligned for each sample to mouse genomic reference GRCm38/mm10 using STAR, v2.5.2a (https://github.com/alexdobin/STAR) (44) with success rates between 90% and 95%, resulting in 66 million mapped reads per sample on average.

Bioinformatics: Short-Read RNA-Seq

Analysis of alignment BAM files on the gene level was performed in R/Bioconductor, using the edgeR (45) and limma (46) packages. The tables ReadsPerGene.out.tab generated by STAR were filtered before statistical analysis, so that only genes with counts per million value more than 1 in at least three samples were considered as expressed. Differential expression analysis was performed via linear modeling, accounting for mean-variance relationship. The p-values were adjusted using Benjamini–Hochberg correction for multiple testing.

Analysis of alignment BAM files for DAS was performed using LeafCutter v0.2.7 (7), requiring at least 20 reads per cluster (-m 20), a maximum intron length of 500 kb (-l 500000), and maximum level of consistency within groups (-i 5 -g 5). Intron clusters were annotated for gene of origin and junction novelty status based on a RefSeq transcriptome (release 90) representing 36,000 annotated genes and more than 100,000 transcripts.

RNA Extraction and qRT-PCR

For total RNA, muscles were pulverized, and 15 to 20 mg was used to extract RNA using a combination of Trizol and RNeasy Micro Kit (Qiagen), followed by a DNase removal step according to the manufacturer.

OneStep qRT-PCR was run in duplicate on a 7900 Fast Real-Time PCR System (Thermo Fisher Scientific). 250 nl (2.5 ng) of RNA was transferred to 384-well plates using an ECHO liquid handler, followed by the addition of 2.25 μl qRT-PCR Mix (Quantitect Multiplex RT PCR Kit: Qiagen #204645, TaqMan probes 20x). Data analysis was done using the 2-ΔΔCT method, with Taok1 used as house-keeping gene. The following splicing-specific probes were used:

Gene	Splice event	Forward primer sequence	Reverse primer sequence	Reporter sequence
Bin1	exon 7 inclusion	TGAGTCTCTTCAAACCGCCAAAA	GGGCGGCTTTCTCAAGCA	ATTGCCAAGCCTGTCTCG
Bin1	exon 7 exclusion	TGAGTCTCTTCAAACCGCCAAAA	CGAACACCTTCTGGGCTTTGAT	TTCTGCCTTGGCAATTT
Atp2a1	exon 22 inclusion	CCCCTCCTCCATGTCTTTGAA	GCTGGTTACTTCCTTCTTTCGTCTT	CCGTGTCACAGATCCAG
Atp2a1	exon 22 exclusion	GGCTGGATGAGCTTCTCAAGTTC	CACAAGGGCTGGTTACTTCCT	TCGTCTTCTGGATCCTCC

Proteomics

Tissue Lysis and Digestion, Protein Extraction

In the light of known challenges to effective protein extraction in skeletal muscle tissue (47), we designed the following extraction protocol. Laboratory chemicals in analytical quality were obtained from Sigma unless stated otherwise. Aliquots of 10 mg frozen muscle tissue powder were resuspended in 250 μl of 50 mM Tris buffer pH 8.5, containing 1% of SDS. Tissue lysis was performed by tip sonication with a Branson Digital Sonifier for 15 s at 25% amplitude on ice for 3 to 5 times, with 1 min breaks to allow sample cooling. Proteins were then precipitated using chloroform-methanol extraction (48). After precipitation, protein pellets were dried in a stream of N₂ and dissolved in 200 μl 50 mM Tris buffer 8.5 pH with 8 M Urea using a sonication bath (Sonorex Digital 10P) (2 × 3 min). Protein extracts were reduced in 5 mM DTT at 56 °C for 40 min and alkylated in 15 mM iodoacetamide at room temperature for 30 min in the dark. Finally, to quench iodoacetamide excess and prevent over alkylation, DTT was added to the solution to 15 mM concentration. For digestion, samples were diluted three times with 50 mM Tris buffer (down to 2 M final urea concentration) and digested overnight at 37 °C using a Trypsin/Lys-C mixture (Promega) at a ratio 1:100 w/w. To reduce the number of missed cleavages, an additional digestion step was performed for 3 h with the protease mixture at the ratio 1:200 w/w. Enzymatic digestion was terminated by adding trifluoroacetic acid to 1%. After the reaction was stopped, the sample was centrifuged at 10,000×g for 5 min, followed by supernatant cleanup on a C18 Sep-Pak cartridge (Waters) according to manufacturer protocol. Finally, the samples were dried in a SpeedVac (Thermo Fisher Scientific). The total peptide concentration was measured by UV absorbance at 210 nm using an LC system with UV detector (Agilent), with roughly 10 mg tissue yielding 500 to 750 μg peptide mixture.

TMT-Based Px

For TMT-labeling, 100 μg of peptides from each sample were processed according to manufacturer instructions (TMT 10plex, Thermo Fisher Scientific). After labeling, all samples were mixed and desalted on C18 Sep-Pak cartridges (Waters) according to manufacturer protocol, followed by dissolving samples in 10 mM ammonium formate buffer for high-pH reverse phase fractionation. An Agilent1200 Series HPLC system with a Waters XBridge column (C18 3.5 μm, 150 × 2.1 mm) was used to fractionate samples into 72 fractions during a one-hour linear gradient from 0% to 50% solvent B (acetonitrile [ACN] with 20 mM ammonium formate pH 10). The 72 fractions were concatenated into 24 fractions, dried, and dissolved to 0.3 μg/μl concentration (2% ACN with 0.1% formic acid [FA]) for further analysis.

LC-MS/MS/MS analysis was performed using an Orbitrap Fusion Lumos mass spectrometer (Thermo Fisher Scientific) coupled with a Thermo Easy-nLC system. From each high pH fraction, 5 μl of peptides were injected and separated on a heated (55 °C) Aurora nanoZero C18 column (1.6 μm,75 μm i.d. × 250 mm, 120 Å) (Ion Opticks). Mobile phases were as follows: (A) 0.1% FA in water; (B) 90% ACN, 0.1% FA in water. Peptides were eluted using a linear gradient from 2% B to 6% B for 50 min, followed by a linear gradient to 15% B for 50 min and to 30% B for 68 min at a flow rate of 400 nl/min. The column was washed at 95% B for 20 min and equilibrated to the start concentration of mobile phase B. MS measurements were performed using DDA mode (Top Speed, 3 s/cycle), using an MS3 method for accurate TMT quantification (49). MS1 scans were measured in the Orbitrap mass analyzer with following settings: mass range from 350 m/z to 1500 m/z, resolving power of 120 K, maximum injection time (IT) set to 50 ms, automatic gain control (AGC) for MS1 was 2.0e5, dynamic exclusion set to 40 s. Precursor ions were isolated using a 0.7 Th window, followed by fragmentation using collision-induced dissociation at normalized collision energy (NCE) of 35%. Fragment ions were measured in the ion trap with maximum IT of 50 ms and AGC value of 1.0e4. After tandem mass spectrometry, 5 notches were isolated with 2 Th window for high-collision dissociation at NCE 55%. The MS3 scans were measured in the Orbitrap mass analyzer in the mass range from 100 m/z to 500 m/z with resolving power of 50 K, maximum IT 50 ms and AGC value 1.0e5.

Peptide Design for Targeted Px

For the selection of DAS events to be targeted, we evaluated the most significant (by adjusted p-values) clusters reported by LeafCutter. The following criteria were applied: (i) the cluster is simple (the number of arc/junctions <6, preferably 3) and therefore a specific splicing event can be extracted (e.g., exon inclusion/exclusion), (ii) the tryptic event–specific peptides are 6 to 21 amino acid long, (iii) the identified changes are not leading to annotated noncoding transcripts, and (iv) the main changes occur in junctions with high usage (>0.1); see supplemental Table S5 for individual cluster assessments. In addition to 27 clusters selected in this way, we chose three clusters for genes of special interests due to their roles in DM1 and/or muscle contraction (Mef2d, Tnnt2, Svil).

For the target tryptic peptides, we adhered to the following design principles: each intron cluster was targeted by ideally at least one peptide specific to each DAS event (e.g., an exon inclusion and an exon exclusion–specific peptide in case of an alternatively spliced cassette exon). Where impossible, interrogating only one alternative was acceptable. In addition to splice event-specific peptides, one “normalizing peptide” per gene was chosen, which, based on available transcript annotations, is common to all known isoforms and unique to this gene.

For each chosen DAS event, amino acid sequences were translated from RefSeq transcripts representing the event or, for exonic sequences without matching RefSeq transcript, from the genomic sequence in the inferred reading frame. For each gene, a normalizing peptide was chosen according to three rules: (i) it corresponds to a region common for all known transcripts, (ii) it has a high potential detectability according to Peptide Atlas (50, 51), and (iii) it has a minimal number of amino acids with possible modifications, such as asparagine and glutamine (deamidation), methionine (oxidation), or known phosphorylation sites. In case of highly probable missed cleavages (close arginine and/or lysine), both peptides were considered.

Based on LeafCutter DAS results, 30 genes were chosen for PRM analysis. In total 98 SpikeTides L peptides with heavy labeled arginine or lysine were synthesized (JPT) with an approximate amount of 10 nmol per peptide. The developed methods correspond to tier 3 measurement according to the guidelines for targeted analysis. The stability of the heavy peptide standards was not a concern since the targeted analysis was performed for only ten samples (in two replicates), which required less than 2 days of analysis.

Targeted Px

For PRM analysis, samples were dissolved in 2% ACN with 0.1% FA, spiked with heavy labeled peptides. 1.5 μg of peptide mixture was injected for analysis. The amount of spiked heavy labeled peptides was adjusted to obtain similar ion current (the final amount varied in the range of 10–100 fmol per sample). The exact concentrations of the heavy peptide standards were not assessed since the analysis performed in the paper is relative (WT versus DM1) and does not require the absolute concentration values. The analysis was performed using an Orbitrap QExactive HF-X mass spectrometer (Thermo Fisher Scientific) coupled with a Thermo Easy-nLC system. For each sample, the analysis was conducted twice using two different chromatographic conditions. Under the first condition, a 90-min gradient (from 3 to 40% solvent B) and a two-column set up with a C18 PepMap100 trap-column (5 μm, 300 μm i.d.× 5 mm, 100 Å, Thermo Fisher Scientific) and a 25 cm analytical column (EASY-Spray PepMap C18, 2 μm, 75 μm i.d, 100 Å, Thermo Fisher Scientific) were used. This condition is more stable and appropriate for a scheduled PRM analysis with a large number of monitored peptides, but the depth of separation is not sufficient to analyze the low-abundance protein isoforms, especially in the case of muscle tissues. To overcome this obstacle, a 150 min gradient with a longer 50 cm analytical column (EASY-Spray PepMap C18, 2 μm, 75 μm i.d, 100 Å, Thermo Fisher Scientific) was employed as a second condition. MS measurements were performed using targeted PRM mode with the following settings: mass range from 340 m/z to 1200 m/z, resolving power of 60 K, maximum IT set to 50 ms, AGC for MS1 was 1.0e6, dynamic exclusion set to 50 s. Precursor ions were isolated using a 0.7 Th window, followed by their fragmentation using high-collision dissociation at NCE of 27%. Fragment ions were measured with maximum IT of 100 ms, AGC value of 2.0e5, and resolution 60 K.

Px Data Analysis

For TMT-based experiments, database searching was performed using Thermo Proteome Discoverer (version 2.1.0.81) with Sequest HT search engine and Percolator postprocessing against a RefSeq mouse database (RefSeq release 91, 77,337 entries, 58,517 nonredundant protein sequences). Search parameters were as follows: 10 ppm for precursor mass tolerance; 0.6 Da for fragment mass tolerance; protease: trypsin with maximally two missed cleavage sites; fixed carbamidomethylated cysteine, TMT-labeled N terminus and lysine; potential methionine oxidation. The identified peptides were filtered to 1% false discovery rate. The threshold for precursor contamination was set to 50%, and the minimal average reporter S/N to 10; all missing values and shared peptides were excluded from quantitation analysis. The TMT intensities were normalized inside each sample to the total intensity as well as to the average value across the WT group. The differential abundance analysis was done using moderated t-statistics as implemented in the R package applying limma test with Benjamini–Hochberg correction for multiple testing. The spectra for proteins identified based on one peptide and demonstrating differential expressions can be found in the supplemental File S1. For comparing the number of identifications obtained with different reference databases, an additional search against a canonical UniProt database (25,021 entries, 23,445 nonredundant protein sequences) was performed with the same parameters.

The Skyline software (4.2.0.19009; https://skyline.ms/project/home/software/Skyline/begin.view) (52, 53) was used for PRM analyses. The library was built based on Mascot 2.4 (Matrix Science) (54) identifications of LC-MS/MS runs with the excess heavy labeled peptides spiked into a mixture of samples. The fragments from the second to the penultimate ion were used for quantification to increase specificity. The total intensities were calculated as the sum of area under curve for corresponding fragments. The identifications with “rdotp” and “dotp” indexes lower than 0.7 and 0.9, respectively, were excluded. All matches with an “rdotp” index lower than 0.9 were manually checked for fragment coelution and chromatographic peak shape. Only fragments represented in all samples were considered, others were excluded manually. For all further analyses, custom Python scripts were used.

Protein Extraction and Western Blot

For the Western blot experiments, approximately 30 mg frozen muscle tissue powder as resuspended in 400 μl lysis buffer (1 mM EDTA, 0.5% Triton X-100, 5 mM NaF, 6 M Urea in PBS, pH 7.2–7.4). Lysis was performed using FastPrep instrument (MP Biomedicals) (4 °C, 3× 20 s with 30 s break in between), followed by sonication in an ice-cold water bath (3 × 1 min, with 1 min break). The sample was then centrifuged for 15 min (4 °C, 20,000 rcf), and supernatant was transferred to a new tube. Western blot analysis was performed using Jess automated system (ProteinSimple) with 25-capillary cartridge, 12 to 230 kDa (Proteinsimple #SM W003 and #PS-ST01EZ). Supernatants were diluted to 1 mg/ml and 0.2 mg/ml using Sample Buffer (Jess Sample Kit) for Bin1 and Atp2a1 (Serca1), respectively. Primary antibodies were obtained from Cell Signaling—Rb anti-BIN1 mAB (#51844) and Rb anti-SERCA1 mAb (#12293). Biotinylated conjugate anti-Rb was used for secondary antibody (Jess Anti Rabbit detection module DM-001). Detection was performed by Streptavidin HRP and Luminol Peroxide (Jess anti-Rabbit detection module).

Results

Differential GE and DAS Analysis of RNA-Seq Data

To investigate transcriptomic changes associated with DM1 disease, short-read RNA-seq analysis was performed on two groups: gastrocnemius muscle tissue obtained from (1) five HSA^LR mice (DM1 disease model) and (2) five healthy (WT) mice. After data processing (see Experimental Procedures), 12,861 genes were quantified and compared between the two groups using moderated t-statistics as implemented in the R package limma (46). Three hundred seventy-four genes showed significant DGE of at least 2-fold and p-values lower than 0.01 after multiple-testing correction (Figs. 1B, 2A, and supplemental Table S1).

Fig. 1.

Gene sets showing differential gene expression and differential alternative splicing and overlaps.A, adjusted p-value cut-off p < 0.01 for both DGE and DAS. B, adjusted p-value cut-off p < 0.01 for both DGE and DAS and an additionally for DGE, an absolute abundance fold change of >2. DAS, differential alternative splicing; DGE, differential gene expression.

Fig. 2.

Differential analysis of transcripts and proteins, Differential analysis of (A) transcripts and (B) proteins identified in WT compared to DM1 mice (volcano plots). The coordinates are log-transformed and correspond to adjusted p-values and abundance fold change calculated by limma test with Benjamini–Hochberg correction. The red dots represent differentially expressed transcripts and proteins (p-value <0.01 and absolute abundance fold change >2). C, the correlation between observed transcript and protein abundance fold changes in log-transformed coordinates. D, the intersection between gene sets with significant differential gene expression (DGE) on transcript and protein levels (note that 65 differential expression proteins correspond to 64 genes) and (E) their intersection with genes showing differential alternative splicing (DAS). DM1, myotonic dystrophy, type 1.

We used the LeafCutter software (7) to identify intron clusters (i.e., sets of splicing alternatives sharing at least one exon). In a second step, LeafCutter statistically evaluates the relative differential usage of the exon junctions (arcs) in each intron cluster between any two comparison groups of interest within the dataset. We equate statistically significant intron clusters with instances of DAS, and we use the term “splice event” to denote a specific splicing variant or noncircular path of arcs, traversing an intron cluster (see Box 1).

Box 1. Three examples of DAS instances shown as LeafCutter-identified intron clusters with different complexity and graphical explanation of terminology in the first example. DAS, differential alternative splicing.

We identified 694 instances of DAS in 591 genes based on a multiple-testing–adjusted p-value cut-off of p < 0.01 (supplemental Table S2). Of these, 268 (39%) intron clusters contain exactly three arcs (exon–exon junctions), often corresponding to an alternative inclusion or exclusion of a single cassette exon flanked by two constitutive exons. The remaining intron clusters have either only two arcs (typically due to alternative starts or exons with a long and a short alternative, 13%) or a more complex structure (multiple cassette exons and/or variable length exons, 49%). Interestingly, more than half (392) of all identified intron clusters have at least one exon–exon junction not present in the RefSeq transcriptome annotation used. These junctions include unannotated exon pairings or 5′ or 3′ exon boundaries or combinations. Out of a total of 3161 arcs, 926 (29%) were novel, representing splice junctions absent from the RefSeq annotation. Some of these unannotated events reflect simple omissions in the transcriptome reference or are likely due to errors in the genome reference or sequencing data, but far from all instances can be explained in this fashion. For example, we identified a novel exon in an intron cluster in the Flnc gene and its corresponding protein product (see below), with no obvious prior evidence in any of the common annotation sources. Interestingly, this exon is highly conserved in mammals except for primates (where a shorter in-frame version of this exon occurs), which provides additional support for the identified novel exon. We similarly identified and confirmed the existence of a novel exon for Ywhae and a novel junction (exon exclusion) for Ttn (supplemental Table S3). This demonstrates that at least some splicing novelty we observe is real and can be robustly discovered with our approach.

Genes with known differential splicing patterns in DM1 constitute an intrinsic positive control in this experiment. For example, our data recapitulates the DM1-specifc exclusion of Atp2a1 exon 22 (p.adj = 6.48∗10⁻²³) described for human muscle biopsies from DM1 patients (28); for more examples see Figure 4, Discussion and Table 2.

Fig. 4.

Comparison of instances of differential alternative splicing at transcriptional and protein level for Atp2a1 and Bin1 genes. Panels on the left show LeafCutter AS intron clusters with relative exon usage in the DM1 and WT control groups. Panels in the middle show qRT-PCR results with the splicing-specific probes (see Experimental Procedures) covering the corresponding exon inclusion (gray) and exon exclusion (blue). Panels on the right show the corresponding relative abundances of peptides specific to exon inclusion or exclusion in the same sample groups. See also supplemental Fig. S4 for a UCSC genome browser-based display of these LeafCutter AS intron clusters. AS, alternative splicing; DM1, myotonic dystrophy, type 1.

Table 2.

Overview of select genes of known importance to DM1, and summary of their performance in our work, summarized in columns (3) to (6)

(1) Gene symbol	(2) Gene relevant in DM1 [reference]	(3) DAS (Tx) observed via RNA-seq	(4) Targeted Px (PRM) confirms DAS (Tx), with exon number and coordinates (mm10)	(5) DE (Tx) observed via RNA-seq	(6) DE (Px) observed via DDA
Atp2a1	[1,2, 3,4]	+	+ (exon 22, chr7:126,446,545–126,446,586)
Bin1	[1,2,3,4]	+	+ (exon 7, chr18:32,414,945–32,415,037)		+
Ldb3	[1,2,3,4]	+	(exon 8 – alternative stop, chr14:34,561,254–34,561,880)b
Ttn	[1,3,4]	+	+ (exon “Mex5”, chr2:76,705,670–76,705,972)h
Ryr1	[1,3]	+	+ (exon 70, chr7:29,056,234–29,056,248)f
Nfix	[3,4]	+	+ (exon 7, chr8:84,723,728–84,723,850)f
Pdlim3	[2,4]	+a	(exon 4, chr8:45,908,469–45,908,657)i		+
Pdlim7	[2]	+	+ (exon 5, chr13:55,508,208–55,508,326)	+	+e
Svil	[2]	+g	+ (exon 14a, chr18:5,077,371–5,077,418g
Ywhae	[2]	+g	+ (unannotated exon, chr11:75,762,698–75,762,742)g
Fermt2	[2]	+	(exon 5, chr14:45,476,418–45,476,450)b
Neb	[3]	+	+ (exon 138, chr2:52,165,167–52,165,277; exon 140, chr2:52,163,937–52,164,050)		+
Phka1	[4]	+	+ (exon 19, chrX:102,557,095–102,557,272)		+
Ppp1r12b		+	+ (exon 24, chr1:134,765,897–134,766,077)
Camk2g		+	+ (exon 13, chr14:20,757,666–20,757,728)
Flnc		+g	+ (unannotated exon, chr6:29,441,974–29,442,036)g		+
Mbnl1	[1,2,3,4]	+	n/ac
Clcn1	[1,3,4]	+	n/ac		+
Gfpt1	[2,3,4]	+	n/ad		+

[1] (77), as per http://diseases.jensenlab.org/, top 50 human genes for “Myotonic dystrophy type 1 [DOID:11,722]”, [2] (78), [3] (24), [4] (79), + = yes, (empty) = no, n/a = not applicable.

Boldface text indicates novel exons.

DE, differential expression; DM1, myotonic dystrophy, type 1; PRM, parallel reaction monitoring; Px, proteomics; Tx, transcriptomics.

^aSmall differential.

^bConfirms (+) by trend, but not statistically significant.

^cTargeted peptide design not possible.

^dNot targeted.

^eConfirms DE observed by RNA-seq.

^fSplice event–specific peptide not identified in one of the two comparison groups.

^gIncompletely RefSeq-annotated intron cluster, indicating potential novelty.

^hAccording to (80).

ⁱNo change observed.

We found the overlap between genes exhibiting DGE and DAS to be very modest—fewer than 10% of genes with DGE show DAS, and conversely, fewer than 20% of genes with DAS surface in the DGE analysis (Fig. 1A). The dearth of genes in common between DGE and DAS is even more pronounced when imposing a commonly used 2-fold change cutoff on DGE (Fig. 1B). This reinforces the rationale for separate, dedicated analysis approaches for DGE and DAS, especially in settings like here, where altered splicing is known to be biologically important. We furthermore investigated if these transcriptomic changes actually translate to altered proteins.

Large-Scale DDA Proteome Analysis; GE at Protein and Transcript Levels

To study proteomic changes in DM1 and their correlation with our findings at the transcriptome level, we conducted quantitative MS-based Px analyses on the same ten muscle samples analyzed by RNA-seq. Tryptic peptide mixtures were prepared and labeled using the TMT approach, fractionated to 24 fractions using high-pH reversed-phase chromatography and analyzed using high-resolution MS, coupled with HPLC in a data-dependent manner. To improve the analysis depth, we used extensive prefractionation, addressing the wide dynamic range due to few highly abundant proteins (actin, myosin). Overall, across a total of 24 fractions, more than 53,000 peptides were identified based on a RefSeq protein database, which resulted in 5832 identified and quantified protein groups (supplemental Table S4).

Based on this deep coverage of the muscle tissue proteomes, we identified 65 proteins from 64 genes with significantly different expression between DM1 and control animals (supplemental Table S1). Of note, applying the same conventional p-value and abundance fold change thresholds resulted in almost six times more differentially expressed genes on the transcriptional level (Fig. 2A) than on the protein level (Fig. 2B). This considerable difference (summarized in Fig. 2D) has likely both technical and biological reasons. An illustrative example of the latter presents itself in our data with Clcn1, for which we observe significantly lowered expression in DM1 on the protein (supplemental Table S1), but not on the transcript level, where however DAS was evident (Fig. 2E, “DAS” and “DGE, protein level” intersection, and supplemental Fig. S7). This observation is consistent with the known preferential inclusion of a 79 bp exon in DM1 over WT, which leads to a frameshift and premature stop codon. Consequently, these transcripts are subject to elimination via nonsense-mediated decay, abrogating translation (25).

The overall correlation between DM1-WT abundance fold changes of all genes identified on both the transcriptional and the protein level is low (Pearson correlation = 0.384) (Fig. 2C), consistent with general findings from previous proteogenomics studies (55, 56). The intersection between these two sets is modest (Fig. 2D): Even if only genes identified in both the Tx and Px experiment are considered (Fig. 2D under “identified in both analyses”), only 30% of genes with differential expression on the transcriptional level demonstrate differential expression on the protein level. Moreover, differentially expressed proteins overlap modestly with both DGE and DAS (Fig. 2E).

Importantly, the choice of search database (full RefSeq mouse, 58,517 unique sequences vs. canonical database, UniProt mouse, 23,445 unique sequences, solely for comparison purposes) only minimally impacted the total number of identified proteins. However, 11 of the 31 genes with both DAS and DGE at the protein level were identified only when searching against the full RefSeq database (supplemental Fig. S1).

The correlation between Tx- and Px-based DGE (0.384 for all genes) markedly increases, when the number of genes is restricted by cut-offs on fold change and/or p-value, even if these cut-offs are very permissive. For example, applying a p-value threshold of p-val <0.1 without any fold change constraints already results in a respectable correlation of 0.72 (Fig. 3A) for a gene set ten times smaller (526) than the total number of genes in common between Tx and Px (5430) (Fig. 3B). Imposing commonly used cutoff values (|FC|>2 and p-val <0.01) yields 22 genes with a correlation of 0.91, and for the five most significantly changing genes, the correlation is almost perfect. This illustrates that the low correlation observed for the complete dataset is predominantly determined by noise, while there is substantial agreement between the two technologies for differentially expressed genes beyond standard significance and fold change cutoffs.

Fig. 3.

Correlation of fold changes on transcript and protein levels.A, the Pearson correlation of abundance fold changes and (B) the number of common differential expression genes at transcript and protein levels. Each heatmap cell corresponds to different significance thresholds, the colour represents (A) the value of Pearson correlation and (B) the number of common genes.

Design of Splice Event–specific Targeted Peptides

For further focused investigation of particular DAS instances and their effect on proteins, we used a targeted approach. We applied PRM, allowing precise detection of peptides using heavily labeled synthetic analogues. PRM allows for precise detection of local changes in protein sequence, even when such changes are extremely small, thus comparing favorably to Western blot analysis (57). For example, Atp2a1 exon 22 inclusion/exclusion leads to a less than 1 kDa molecular weight difference between the resulting proteins, challenging to resolve by Western blot (supplemental Fig. S8, C and D). Bigger differences are distinguishable per se, but interpretation may still be confounded by underlying splicing complexity (Bin1, supplemental Fig. S8, A and B).

We used PRM to compare the abundance of specific peptides corresponding to parts of proteins predicted to be differentially alternatively spliced between DM1 and WT samples by RNA-seq data analyzed for DAS by LeafCutter. In addition to splice event-specific peptides, one “normalizing peptide” per gene, common to all known isoforms, was chosen. For detailed design principles and the DAS event selection criteria, see Experimental Procedures (Peptide design for targeted Px) and supplemental Table S5. Applying these criteria, we designed splice event specific peptides for intron clusters from 30 genes, with four genes containing two separate clusters each. This amounted to 45 distinct splice events, for which 97 heavy peptides were synthesized and used for PRM analysis of the ten samples (same as used for Tx and Px analyses described above). PRM analysis was conducted twice per sample with different instrument settings (see Materials and Methods). Stages of the analysis were marked by distinct success rates (Table 1), the most prominent obstacle being the detection of the splice event specific peptides in the biological samples (66 out of 87 peptides). As expected, we observed a correspondence between a gene’s abundance (fragments per kilobase of transcript per million mapped reads values from the RNA-seq analysis) and the recoverability of its "normalizing" and splice event specific peptides in targeted Px (supplemental Fig. S2). Altogether, we reliably quantified 21 events in 16 genes. Among these, for eight events with splice event specific peptides identified in only one group (WT or DM1), results qualitatively correspond to the Tx results (supplemental Table S3 and supplemental Fig. S3). For three of them, there were other peptides covering the same event, which all together sum up to 16 splicing events in 14 genes quantified in both conditions.

Table 1.

The number of successful peptides and corresponding genes at different stages of PRM analysis

PRM analysis stage	# Peptides	# Corresponding genes/splice events
Synthesized peptides	97	30/45
Heavy peptides detected	87	25/40
Splice event–specific peptides detected in samples	66	20/27
Normalizing peptides detected in samples	57	16/21
Splice event–specific and normalizing peptides, same gene	47	16/21
Splice event–specific and normalizing peptides, same gene, in WT and DM1 condition	37	14/16

Number of peptides having “normalizing” peptides represents the total number of peptides used to calculate isoform ratios.

DM1, myotonic dystrophy, type 1; PRM, parallel reaction monitoring.

Quantitative Analysis of DAS in Proteins and Transcripts

A comparison between targeted proteomic results and DAS transcriptomic results shows a good qualitative correspondence. Additionally, for Atp2a1 and Bin1, two genes well known in the context of DM1, we performed qRT-PCR using splicing-specific probes, which further confirmed the splicing pattern differences observed by RNA-seq. Figure 4 shows side-by-side splicing data for these two genes from RNA-seq (LeafCutter clusters), targeted Px, and qRT-PCR data—details:

• •Skipping of Atp2a1 exon 22 (based on transcript isoform NM_007504.2), its penultimate exon, is clearly observed near exclusively in the DM1 condition in short-read RNA-seq data analyzed for DAS and by targeted Px via a peptide-specific to exon 22 exclusion. Unfortunately, it was not possible to design an exon inclusion–specific peptide for this splice event, since it results in an alternative early stop codon and the corresponding peptide contains only four amino acids. Importantly, this observation recapitulates findings described in human muscle biopsies from DM1 patients (58).
• •Inclusion of Bin1 exon 7 (based on transcript isoform NM_009668.2) is observed at a very low rate in the WT group (about 0.5%). By contrast, in the DM1 group more than 10-fold increase of the inclusion rate is observed by both RNA-seq and qRT-PCR. Concordantly, the peptide targeting exon 7 inclusion has significantly higher normalized intensity in the DM1 group than WT group. The same event has been observed in multiple studies in the context of myotonic dystrophy (see Table 2).

Protein- and mRNA-based splice event ratios (Box 2, Equations 2, and 3) resemble the intensity fold changes in a common differential expression analysis, but with a focus on local AS.

Box 2. To quantitatively describe splice events and DAS instances observed between DM1 and WT in our PRM analyses, and to furthermore compare them with their counterparts originating from LeafCutter-processed RNA-seq data, we used the following definitions.

The relative peptide intensity in a PRM analysis is calculated as the ratio of the abundances of the light peptide and the heavy (labeled) peptide: $I^{light}/I^{heavy}$.

We then quantified the relative intensity of a local splice event (variant) as the ratio between the abundances of the splice event–specific peptide and the normalizing peptide (Equation 1),

$$\begin{array}{c}{I}_{spliceevent}=\frac{I_{SES}^{light}/I_{SES}^{heavy}}{I_{Norm}^{light}/I_{Norm}^{heavy}}\text{,}\end{array}$$ (1)

where $I_{SES}$ is the intensity of the splice event–specific peptide, I_Norm is the intensity of the normalizing peptide, and the indexes light and heavy indicate light and labeled peptides, respectively.

Protein-based splice event ratios were calculated as follows for each splice event-specific peptide identified in either group (DM1 and WT):

$$\begin{array}{c}{I}_{spliceeventratio}^{Proteins}=\frac{{⟨I_{spliceevent}⟩}^{DM1}}{{⟨I_{spliceevent}⟩}^{WT}}\text{,}\end{array}$$ (2)

where ⟨ ⟩ designates the mean value across the group and I_{splice event} is calculated according to Equation 1.

Correspondingly, mRNA-based splice event ratios were obtained from our RNA-seq data as follows:

$$\begin{array}{c}{I}_{spliceeventratio}^{Transcripts}=\frac{{⟨\frac{N_{SES}}{N_{Norm}}⟩}^{DM1}}{{⟨\frac{N_{SES}}{N_{Norm}}⟩}^{WT}}\text{,}\end{array}$$ (3)

where N_SES and N_Norm stand for the number of reads covering the genome coordinate range corresponding to the equivalent splice event–specific and normalizing peptide, respectively. For exon–exon junction-spanning peptides the number of spliced reads covering this junction was used.

The relative intensity of a local splicing variant (Equation 1) is similar to the exon usage value calculated by LeafCutter based on short-read RNA-seq data.

Notably, for eight events, we were not able to calculate this ratio, since the corresponding peptides were not detected in one of the groups (e.g., the exon inclusion–specific peptide in the Flnc gene was not detected in DM1, Fig. 4), but all of them qualitatively agree with the RNA-seq-based results (supplemental Fig. S3). For 14 genes (16 splice events, with three events targeted by two peptides), splice event ratios were calculated on the protein and transcriptional level (Fig. 5).

Fig. 5.

The correlation between log-transformed splice event ratio observed in targeted Px (Equation 2andBox 2) and Tx (Equation 3andBox 2) for 14 genes, corresponding to 16 splice events and 19 peptide pairs detected in DM1 and WT. The superscript numbers correspond to different clusters/splice events in the same gene (e.g., Neb¹ is an inclusion event in cluster 5321, Neb^2&3 correspond to different peptides covering the same inclusion event in cluster 5320, and Neb⁴ is an exclusion event in that same cluster 5320. For all other events, see supplemental Table S3, “AS event ratio” list). The Pearson correlation for all examined events is 0.95 (0.98 without two outliers, Atp2a1 and Pdlim7). AS, alternative splicing; DM1, myotonic dystrophy, type 1; Px, proteomics; Tx, transcriptomics.

The very high correlation (Pearson’s r = 0.95) observed between protein and transcriptional splice event ratios demonstrates that DAS events often faithfully translate from mRNA to protein in terms of the relative frequencies of the alternatives in the two sample groups compared (DM1 and WT). Atp2a1 and Pdlim7 deviate notably from the diagonal (Fig. 5), likely as a consequence of limitations specific to the splice events targeted in these two genes. The transcriptional splice event ratio for Atp2a1 is the highest (>2000) among all events considered here. But in our targeted Px experiment, the same concentration of heavy peptides was used for all samples across both groups. Therefore, the nonlinear dependency of the peptide ion signal on the concentration may cause inaccurate results for the peptide with the high dynamic concentration range. The Pdlim7 gene has the most complex of all intron clusters considered in this study: it includes two alternatively spliced exons and an alternative length flanking exon, resulting in seven arcs. This complexity interferes with the ability to accurately calculate the targeted Pdlim7 splice event intensity. Without these two explicable outliers, the correlation of splice event ratios obtained by Tx and Px is even higher (Pearson’s r = 0.98) and demonstrates an outstanding agreement of the results.

Our data-driven approach enabled us to find novel DAS events, which are not represented in transcript annotations, such as the Flnc gene (supplemental Fig. S3). A cryptic exon (63 nucleotides, “exon 8a” between exons 8 and 9, NM_001081185.2) is included at low levels in WT samples according to transcriptomic data (∼10% of exon 8 and 9 abundance), whereas its inclusion in the DM1 condition appears negligible (<1% of exon 8 and 9 abundance). Remarkably, we identified the peptide designed to target the 3′ junction of this hypothetical exon in WT samples only. The high correlation between fragmentation spectra of the peptide detected in the sample and the corresponding heavy peptide proves the identification of the targeted peptide, confirming the existence and differential expression of a cryptic exon (supplemental Fig. S5).

We investigated, if the DDA methodology’s generally low peptide identification rate would indeed result in low recovery of the specific peptides targeted in our PRM experiment, as suggested by conventional wisdom. We identified 14 peptides pairs corresponding to 12 targeted DAS events in our DDA data. Moreover, five peptide pairs (from three genes) were identified exclusively in DDA data, including one indicative of the well-characterized DM1-specific exclusion of Bin1 exon 11 (59). Although our DDA data indicate the known effect of ratio compression in TMT-based experiments (60, 61) compared to PRM data, the correlation between log-transformed splice event ratios for Px DDA and Tx data (Pearson correlation = 0.91, supplemental Fig. S6B) is close to the one obtained for PRM, and the correlation between the two alternative Px splice event ratios is excellent (supplemental Fig. S6C). This demonstrates that pursuing even targeted splicing analyses using DDA can be feasible under certain favourable conditions.

Discussion

Mouse models are indispensable tools for investigating, deciphering, and isolating underlying disease mechanisms. An array of murine DM1 model systems have been devised over the years, and they are necessarily approximations of the human disease or aspects thereof, each characterized by their own strengths and limitations (62). Understanding these not only on a phenotypic, but also on a molecular level is vital for our ability to use these models productively for disease research and drug discovery. The widely used HSA^LR model is one of the earliest introduced, with disease manifestations restricted to skeletal muscle. It is thus well suited for studying myotonia, RNA foci accumulation, sequestration of MBNL1, and splicing alterations, informing its choice for this study (63).

We are hopeful that our congruent generation of comparative (WT/diseased) in vivo transcriptome and proteome datasets with an integrated analysis on multiple levels (gene and protein expression, splicing variation; (Figs. 1, 2, supplemental Tables S1, and S2)) will enable future enquiries into DM1 aided by the HSA^LR model and encourage similar integrated approaches in other model systems.

In recent years, major advances have been published regarding the transcriptomic landscape of DM1 (64) and how alternative mRNA splicing affects the proteome either generally (65, 66) or specifically in muscle tissue (67). A direct correspondence between Tx and Px readouts cannot be universally assumed or expected, for diverse technical and biological reasons. However, our data illustrate how one major driver of poor interdataset correspondence is noise affecting genes with no actual differential expression, whereas significantly changing genes trend similarly in Tx and Px data (Fig. 3). Furthermore, for the specific instances of DAS identified using both platforms, our data show an excellent correspondence between Tx and Px data (Figs. 4, 5, supplemental Fig. S3, Tables 1, 2, and supplemental Table S3). Although this cannot be assumed true for all DAS transcripts (see the case of Clcn1 in the Results section), such agreement is generally plausible since the constraints for translating splice events from mRNA to protein are higher than for simple expression level differences.

The impact of splicing diversity on proteome complexity is a matter of ongoing debate, and so is the degree to which it is detectable in large scale MS-based Px studies (68, 69). Limited protein coverage and sampling depth are considered key factors in suppressing protein-level evidence for splicing complexity and can in principle be overcome using exhaustive experimental conditions (e.g., deep fractionation as used here). Our findings echo the insights from other recent studies specifically designed for isoform detection (16) as well as large-scale deep proteome analysis (66).

For a sizable subset of our top DAS events, it was not possible to generate suitable tryptic peptides (supplemental Table S5). In general, the increased probability of arginine or lysine being encoded at exon boundaries (70) is limiting the number of detectable splice-events using trypsin for protein digest (66). This problem cannot be easily mitigated without resorting to less efficient and established enzymes (71).

Following common practice, we used DDA Px initially for establishing the proteomic differential expression landscape in the DM1-WT context. However, in addition, we also found that an analysis of DDA-Px data dedicated to the discovery of splicing events can yield results comparable to targeted approaches. Using a RefSeq-based protein database comprising annotated isoforms allowed us to discover splicing-specific peptides, but did not markedly affect the total number of identifications compared to a “canonical” UniProt database (supplemental Fig. S1). Thus, augmenting the search space to systematically include protein isoforms does not lead to decreased detection sensitivity. Moreover, this approach also represents a powerful means of improving Tx/Px data integration, as transcripts and their derived proteins are consistently and systematically linked. The concept is aptly illustrated by our identification, in DDA data, of Bin1 exon 11 exclusion as a DM1-specific event (59): Similar to targeted data (Fig. 5), Tx- and Px-based splice event ratios are correlating extremely well (supplemental Fig. S6B).

Taken together, these insights suggest an optimized proteogenomics strategy to assess splicing diversity as follows: short-read RNA-seq data are simultaneously analyzed for gene-level and splicing changes (preferably using annotation-agnostic tools). For these, we show that good tractability and validation rates on the protein level can be expected, when candidate choice is prioritized by high GE levels (supplemental Fig. S2). Px DDA profiling experiments may be conducted and analyzed concurrently with RNA-seq, but ideally before targeted approaches are pursued, so that the former can not only complement but also inform the latter, in conjunction with the transcriptomic results. Conceivably then, these aggregated and integrated results may guide follow-on long-read RNA-seq of select genes of interest in order to converge on actual full-length isoform quantitation. The insights from such long-read data could help constrain possible interpretation of the otherwise local splice event data even on the protein level, where measuring isoforms remains technologically elusive.

We observed a surprisingly high frequency of intron clusters in our Tx data with one or more unannotated splicing connections, when using a complete RefSeq transcriptome as annotation reference (supplemental Table S2). This was unexpected, given that mouse genome assembly and annotations are considered generally mature today, and it illustrates that uncritically relying on existing standard transcriptome annotations is still premature. Our cursory investigation showed that many (but not all) novel splicing events have direct or indirect supporting evidence (complementary Ensembl annotations, sequencing data, cross-species sequence conservation) not reflected in RefSeq. Importantly, two potentially novel identified events were subjected to targeted Px and were confirmed. It stands to reason that such novel events may be enriched in cases that selectively manifest only in specialized (for example disease or tissue) contexts, possibly as lower expression variants. Indeed, the novel Flnc variant we found and confirmed was WT specific in our comparison and may represent a minor muscle-specific variant lost in DM1 disease.

The high degree of correspondence found underscores the potential of AS events as robust disease differentiators. This concept is illustrated in Table 2, which shows a selection of known DM1-relevant genes as well as newly identified muscle-related genes demonstrating significant DAS on the transcript and protein level. The first 16 genes in this table are those for which we were able to investigate the Tx-based DAS hypothesis using targeted Px (PRM), fully confirming it in 13 cases and confirming by trend without statistical significance in two cases. In one case (Pdlim3), the weak but significant DAS event did not correspond to any discernible difference in PRM data.

DM1 muscles show perturbed calcium homeostasis and abnormal excitation-contraction coupling processes. This involves key regulatory proteins like ATP2A1, RYR1, and CAMK2G, for which we found and confirmed DAS. Several genes listed in Table 2 code for proteins which play a major role in assembly and function of the sarcomere, the basic contractile unit of the striated muscle fibers (e.g., TTN and NEB), or are involved in sarcomere protein organization (BIN1, LDB3). Yet other proteins (e.g., FLNC, ACTB, SVIL, ACTG1) localize to the Z-disc and are responsible for sarcomere maintenance, integrity, and force transduction. FLNC is particularly intriguing due to the novel WT-specific exon we found. Given its described essential role for myogenesis, it is tempting to speculate about a narrowing of its functional role under disease conditions. Nfix encodes a transcription factor involved in muscle development and regeneration, where it controls AS. MBNL1, another RNA AS factor known for its prominent role in DM1 pathology, also shows DAS in our data. Chloride ion influx stabilizes the electrical charge of the cell, which prevents muscles from contracting abnormally. We also recovered described DAS for the Clcn1 gene (25) encoding the chloride channel 1, which controls the flow of chloride ions into and out of muscle cells and thus plays a key role in repolarization of the muscle after contraction. Our work also revealed significant DAS for PPP1R12B, a key regulatory enzyme involved in glycogen metabolism, which supplies the muscle with glucose during contraction.

Overall, we have demonstrated that differential splicing events between muscles from DM1 and WT mice can be robustly and congruently detected in both Tx and Px data in an integrative framework. Furthermore, many of the genes found to be alternatively spliced in this study have been described as playing a role in animal models of DM1 and the human disease itself or at least have a plausibly relevant function. Without question, additional experiments are needed to assess the functional role of the identified splice variants and their relevance to disease. However, generalizing considerations by (72) and extending them from mRNA- to protein-based measurements, we believe that aberrant splicing events could be used as robust disease biomarker and, possibly, target candidates tractable within the scope of splicing-related disorders such as dystrophies more generally (73, 74), in neurological disease such as Alzheimer’s (5), or in the context of aging-related disorders. Indeed, DM1 is considered a disease of premature aging (2,4), and the parallels described between DM1 and aging are strikingly echoed in this work and our related investigations into muscle senescence (6). We believe it is remarkable that many of the same genes, with the same splicing events, are affected similarly in an aging model (old rats compared to young) and in the murine disease versus normal context we describe here. To immediately further these inquiries, validated peptides from this study could be used directly, after species-specific adjustments as needed, for rising splicing event-specific antibodies for diagnostic research.

Data Availability

The RNA-seq FASTQ files for this study have been deposited to the European Nucleotide Archive (ENA) via ArrayExpress (https://www.ebi.ac.uk/arrayexpress/) (75) under accession number E-MTAB-10842.

The MS Px data have been deposited to the ProteomeXchange Consortium via the PRIDE (https://www.ebi.ac.uk/pride) (76) partner repository with the dataset identifier PXD025589. The Skyline files with corresponding transition and chromatograms are deposited to the Panorama Repository (https://panoramaweb.org/dm1_splicing.url).

Supplemental data

This article contains supplemental data.

Conflict of interest

All authors are employees of Novartis and some hold Novartis stock.