Integrating SMRT and Bulk RNA Sequencing With Metabolic Phenotyping to Examine Reduced Skeletal Muscle Mitochondrial Respiration in Type 2 Diabetes

Abstract

Recent advances in RNA sequencing (RNA-seq) techniques allow the identification of tissue-specific alternative splicing and can thereby provide new insights into molecular mechanisms of energy metabolism. Full-length transcriptomics based on single-molecule real-time sequencing (SMRT-seq) enable precise detection of isoforms with 99% accuracy in an unbiased manner. In this proof-of-concept study, we integrated SMRT-seq, bulk RNA-seq, and comprehensive metabolic phenotyping to investigate reduced mitochondrial function in the skeletal muscle of individuals with type 2 diabetes. Muscle biopsies were taken from nine individuals with type 2 diabetes and nine age- and BMI-matched glucose-tolerant men. Whole-body insulin sensitivity (WBIS) was assessed by hyperinsulinemic-euglycemic clamps, and muscle mitochondrial respiration was assessed by high-resolution respirometry. In muscle samples, SMRT-seq was used to create full-length reads and isoforms, which were mapped to the genome. Short-read sequencing was used to compare isoform expression between the groups. Participants with diabetes exhibited lower WBIS and fatty acid–driven and complex I–linked respiration compared with control participants. SMRT-seq revealed ∼67,000 isoforms originating from ∼14,000 unique genes. Although isoform numbers per gene did not differ, SMRT-seq–based mapping enabled refined data set clustering compared with conventional short-read sequencing and identified four splicing variants of the ATP5F1A gene encoding a subunit for ATP synthase. Among these, two novel transcripts were expressed exclusively in control participants. This study identified splicing variants of ATP synthase that were differentially expressed between participants with type 2 diabetes and those with normal glucose tolerance, which may contribute to the reduced fatty acid oxidation in diabetes.

Article Highlights

• In our study, we developed a pipeline to integrate single-molecule real-time sequencing (SMRT-seq) with comprehensive metabolic phenotyping to examine reduced mitochondrial respiration in the skeletal muscle of individuals with type 2 diabetes.

• SMRT-seq revealed ∼67,000 isoforms originating from ∼14,000 unique genes; the isoform numbers per gene did not differ between participants with diabetes and matched control participants.

• Our data identified novel alternative splicing events, including two variants of the ATP5F1A gene encoding a subunit for ATP synthase. Among these, two novel transcripts were expressed exclusively in control participants.

• Our findings link transcriptomic changes to impaired mitochondrial respiration in type 2 diabetes, with the potential of providing novel therapeutic targets to improve metabolic health.

Introduction

Mitochondrial biogenesis and respiratory function (1,2) are reduced in the skeletal muscle of individuals with type 2 diabetes. Previous studies have demonstrated that ATP synthesis in muscle is diminished under both basal and insulin-stimulated conditions (3). Impaired β-oxidation contributes to incomplete degradation of fatty acids, leading to the accumulation of lipid intermediates and reduced ATP synthesis (4). The accumulated intramyocellular lipid species disrupt muscle insulin signaling, which can already be present in prediabetes (5) and recent-onset type 2 diabetes (6,7). Collectively, these findings suggest that reduced mitochondrial respiration is an early hallmark of muscle insulin resistance (1,4). However, the precise molecular mechanisms underlying this dysfunction, as well as the specific site of the defect, remain unclear.

Recent advances in RNA sequencing (RNA-seq) techniques have demonstrated the potential to uncover tissue-specific alternative splicing and gene expression, providing novel insights into the molecular mechanisms underlying impaired energy homeostasis across various tissues, including islet cells (8), the liver (9), and the myocardium (10). Alternative splicing plays a critical role in regulating gene expression (11), and ∼50–60% of mutations associated with chronic diseases are linked to splice defects (12). Specifically in type 2 diabetes, recent studies have suggested that alternative splicing variants affect insulin secretion from pancreatic β-cells, including suppression of the hepatocyte nuclear factor-1α–A1CF transcription-splicing axis (13) and loss of SRY-box transcription factor 9–dependent splicing control (14). Both pathways were shown to disrupt splice programs essential to β-cell function, leading to a critical decrease in insulin secretion in type 2 diabetes (13,14). However, classical RNA-seq generates a library of fragmented short reads, which relies on the assembly of sequences based on the reference whole transcriptome and is therefore unable to resolve complex isoforms (15). Bulk RNA-seq remains the gold standard for quantifying gene expression levels across the transcriptome, but it lacks the ability to distinguish full-length isoforms, which limits its utility in studying alternative splicing events. Consequently, conventional RNA-seq lacks the resolution necessary to accurately identify complex isoforms, limiting its applicability in organ-specific transcriptomic analyses.

Previous studies have suggested that gene loci exhibit complex regulatory mechanisms, where different promotor regions can encode alternative starting exons, leading to the formation of distinct splicing variants (16). Such regulatory complexity may be particularly relevant in metabolically active tissues, where transcript isoforms influence mitochondrial function, oxidative phosphorylation (OXPHOS), and/or insulin signaling. Differential use of transcript isoforms can result in diverse biological functions, even among genes with morphological similarities, depending on the expressed isoform (16).

Advances in RNA-seq techniques now enable the unbiased and highly accurate detection of alternative splicing variants. Bulk RNA-seq provides robust exon-junction counts and thus reliable quantification for abundant transcripts; however, it cannot always resolve structures of full-length transcripts when multiple alternative splicing events co-occur (17). In contrast, single-molecule real-time sequencing (SMRT-seq), a long-read sequencing technology, reads entire cDNAs and therefore allows direct identification of complex isoforms as transcripts that combine two or more alternative events (e.g., exon skipping, retained introns, mutually exclusive exons, or read-through transcripts). Furthermore, isoform numbers and their building mechanisms of alternative splicing events vary largely among genes and conditions, also reflected by the term complex isoform. SMRT-seq in turn overcomes the limitations of conventional short-read sequencing by providing full-length transcript data, allowing the precise characterization of alternative splicing and novel isoforms (17,18). Depending on the tissue and depth, recent long-read data sets of large scale reported that ∼40–60% of detected isoforms are full-length protein-coding isoforms, whereas the rest are truncated, partial, or noncoding (18). Using SMRT-seq, our group recently determined the myocardium-specific full-length transcriptome, identifying ∼66,000 unique isoforms originating from ∼15,000 genes (10). Notably, among these unique myocardium-specific isoforms, we identified 12 previously undescribed isoforms of a key regulator of mitochondrial biogenesis (19,20), PPARGC1A (10). Given the central role of mitochondria in skeletal muscle metabolism, it seems plausible that similar transcriptomic complexity exists in muscle and may contribute to impaired energy homeostasis in type 2 diabetes.

In the current proof-of-concept study, we aimed to 1) investigate the skeletal muscle–specific transcriptome using highly accurate SMRT-seq, 2) compare mRNA expression patterns between individuals with type 2 diabetes and matched glucose-tolerant control participants, and 3) analyze the connection between in vivo metabolic phenotypes and splicing variants of skeletal muscle genes involved in mitochondrial respiration and insulin signaling. Our underlying hypothesis was that SMRT-seq would uncover differences in the expression and alternative splicing of genes implicated in mitochondrial respiration that are altered in type 2 diabetes and remain undetectable with conventional transcriptomics. By integrating SMRT-seq with bulk RNA-seq, we aimed to leverage the strengths of both approaches: the ability of bulk RNA-seq to provide quantitative gene expression data, and the capacity of SMRT-seq to resolve full-length isoforms. In particular, we used SMRT-seq to generate a disease- and tissue-specific reference transcriptome, which then served as an improved mapping backbone for bulk RNA-seq data, thereby reducing information loss inherent in classical reference-based mapping. Importantly, the integration of transcriptomic findings with clinical phenotypes and functional analyses is essential to directly assess their relevance. This approach, which requires advanced expertise, has been underexplored to date and holds considerable potential to bridge the gap between molecular discoveries and their translational effect.

Research Design and Methods

Study Population

Nine men with type 2 diabetes (known diabetes duration 1.5–12 years) and nine glucose-tolerant men (control participants) matched for age and BMI were recruited from the German Diabetes Study (GDS). Control participants had no first-degree relatives with diabetes and underwent a 75-g oral glucose tolerance test (AccuCheck Dextro O.G.-T.; Roche, Basel, Switzerland) to confirm normoglycemia. The oral glucose tolerance test was performed on a different day than the modifed Botnia protocol. The GDS is an observational prospective cohort study examining the natural course of recent-onset diabetes and associated comorbidities as previously described (21). The GDS was approved by the ethics committee of the Medical Faculty of the Heinrich Heine University of Düsseldorf, Düsseldorf, Germany (reference no. 4508); it is registered in Clinicaltrials.gov (NCT01055093) and performed according to the Declaration of Helsinki. All participants provided signed informed consent on inclusion in the study.

Three days before the study visit, participants were asked to discontinue any glucose-lowering medication, refrain from strenuous physical activity, and adhere to a balanced isocaloric diet. All metabolic tests were performed starting at 7:00 a.m. after a 10-h overnight fast. Skeletal muscle biopsies and metabolic phenotyping were conducted on separate days, with a maximum interval of 6 months.

Skeletal Muscle Biopsy

Biopsies of skeletal muscle were performed in the region above the vastus lateralis muscle, after the tissue was anesthetized by subcutaneous injection of 15 mL 2% lidocaine. Consequently, ∼200–300 mg tissue was obtained using a modified Bergström needle with suction, immediately snap frozen in liquid nitrogen, and stored at −80°C.

Metabolic Phenotyping

All participants underwent the modified Botnia protocol, which combines an intravenous glucose tolerance test to assess insulin secretion along with a subsequent hyperinsulinemic-euglycemic clamp test to assess insulin sensitivity (21). The intravenous glucose tolerance test was started with an administration of a bolus with 30% glucose infusion (1 mg/kg body weight), with frequent blood sampling up to 1 h to measure blood glucose, C-peptide, and insulin. The clamp started with a priming dose (10 mU ∗ kg body weight⁻¹ ∗ min⁻¹) of human insulin (Insuman Rapid; Sanofi, Frankfurt, Germany) for 10 min, followed by constant infusion of short-acting insulin at 1.5 mU ∗ kg body weight⁻¹ ∗ min⁻¹ for 3 h. The concentration of blood glucose was maintained at 90 mg/dL, and the mean infusion rates of 20% glucose with space correction during the last 30 min of the clamp (steady-state period) were used to calculate whole-body insulin sensitivity (WBIS; M value). Respiratory quotient and resting energy expenditure were assessed by indirect calorimetry using the canopy mode and the Vmax Encore 29n (CareFusion, Höchberg, Germany). Percentage of body fat and fat-free mass were assessed with air-displacement plethysmography (Cosmed, Concord, CA) as previously validated (22) and shown to correlate closely with body composition measures derived from magnetic resonance imaging (23). An electric hand dynamometer was used as a standardized measure of muscle strength (24). Analyses of routine laboratory parameters were performed as previously described (21). The Baecke questionnaire was used to report habitual physical activity during the last 12 months (25).

Mitochondrial Respiration

Mitochondrial respiration was assessed in duplicate in saponin-permeabilized fibers from the vastus lateralis muscle using the Oxygraph-2k respirometer (Oroboros Instruments, Innsbruck, Austria) as previously described (26). Two substrate–uncoupler–inhibitor titration protocols were used. The first protocol investigated changes in mitochondrial respiration with the addition of fatty acid–, NADH-, and succinate-linked substrates (FNS pathway), and the second protocol investigated changes in mitochondrial respiration with the addition of the NS pathway). For both protocols, data are presented as mitochondrial respiratory capacity (pmol O₂ s⁻¹ mg⁻¹ wet weight) and intrinsic mitochondrial respiratory capacity (pmol O₂ s⁻¹ mg⁻¹ wet weight/citrate synthase [CS] activity).

The first protocol investigated changes in mitochondrial respiration with the addition of the FNS pathway as follows: 0.2 mmol/L octanoyl-carnitine and 2 mmol/L malate to determine leak respiration (L) with electron input through electron-transferring flavoprotein (ETF; [ETF]L); 2.5 mmol/L ADP to determine OXPHOS capacity (P) at nonsaturating concentrations of ADP with electron input through ETF ([ETF]P); 10 mmol/L glutamate to determine P at nonsaturating concentrations of ADP with convergent electron input through ETF and complex I (CI; [ETF + CI]P); 10 mmol/L succinate to determine P at nonsaturating concentrations of ADP with convergent electron input through ETF, CI, and CII ([ETF + CI + CII]P); 5 mmol/L ADP to determine maximal P at saturating ADP concentrations with convergent electron input through ETF, CI, and CII ([ETF + CI + CII]P); 6 μmol/L cytochrome c to test the outer mitochondrial membrane integrity (samples with an increase in mitochondrial respiration >15% were excluded [1]); 5 nmol/L oligomycin to test L with convergent electron input through ETF, CI, and CII ([ETF + CI + CII]L); 0.75–1.5 μmol/L carbonyl cyanide 4-(trifluoromethoxy) phenylhydrazone (FCCP) via stepwise titration to determine maximal electron transport chain capacity (E) at saturating levels of ADP with convergent electron input through ETF, CI, and CII ([ETF + CI + II]E); and 10 mmol/L malonic acid to determine E at saturating levels of ADP with convergent electron input through ETF and CI ([ETF + CI]E).

The second protocol investigated changes in mitochondrial respiration with the addition of the NS pathway as follows: 2 mmol/L malate and 5 mmol/L pyruvate to determine L with electron input through CI ([CI]_L-PM); 2.5 mmol/L ADP to determine P at nonsaturating concentrations of ADP with electron input through CI ([CI]_P-PM); 10 mmol/L glutamate to determine P at nonsaturating concentrations of ADP with electron input through CI ([CI]_P-PMG); 10 mmol/L succinate determine P at nonsaturating concentrations of ADP with convergent electron input through CI and CII ([CI + CII]_{D-2.5 mmol/L}); 5 mmol/L ADP to determine maximal P at saturating ADP concentrations ([CI + CII]_P); 6 μmol/L cytochrome c to test the outer mitochondrial membrane integrity (samples with an increase in mitochondrial respiration >15% were excluded [1]); 5 nmol/L oligomycin to test L with convergent electron input through CI and CII ([CI + CII]_L); 0.75–1.5 μmol/L FCCP via stepwise titration to determine maximal E at saturating levels of ADP with convergent electron input through CI + CII ([CI + II]_E); and 10 mmol/L malonic acid to determine E at saturating levels of ADP with electron input through CI ([CI]_E).

CS activity was used as a surrogate biomarker of mitochondrial content in skeletal muscle (27) using the CS Assay Kit (Sigma-Aldrich). In a subgroup (n = 11 [type 2 diabetes n = 7; control n = 4), transmission electron microscopy (TEM) along with stereological principles with a 144-point grid overlaid on micrographs was used to quantify mitochondrial density and number. In a blinded fashion, 15 micrographs of longitudinally sectioned muscle tissue per biopsy (grid density 9 × 9) were analyzed (28).

RNA Isolation

Total RNA was isolated using the Fibrous Tissue RNeasy Mini Kit (Qiagen) according to the manufacturer’s protocol. The homogenate of tissue and RLT buffer was treated with proteinase K for 20 min before RNA was extracted on Qiagen RNeasy columns. A DNAse digestion took place on column. The quality of eluted RNA was assessed by microvolume spectrophotometer analysis (Nanodrop; Thermo Fisher Scientific) and the 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA). Only samples with an RNA integrity number >9 were used.

SMRT-Seq

Total RNA from human skeletal muscle samples was isolated as described above and used for library preparation according to the manufacturer’s protocol (Iso-Seq Template Preparation for Sequel Systems) without size selection. Four samples from both type 2 diabetes and control groups were used as reference libraries for additional steps. SMRT-seq was performed at the Genomics & Transcriptomics Laboratory of the Biological-Medical Research Center (Heinrich Heine University, Düsseldorf, Germany) using the Sequel I System (Pacific Biosciences) according to the manufacturer’s protocol. For each pooled sample (n = 3 per group) and an additional sample per group, the 8M Chip SMRT Cell was used (8 × 106 zero-mode waveguide holes per sample).

Bioinformatic Analysis of SMRT-Seq Data

A modified IsoSeq3 pipeline, followed by our own pipeline of open-source tools and scripts, was used for bioinformatic analysis as described previously (10). For the primary analysis, the Isoseq3 pipeline from Pacific Biosciences was used, followed by a secondary analysis pipeline with open-source tools and modified scripts. This included mapping to the genome with Minimap2 (29), collapsing transcripts with cDNA Cupcake, and performing annotation with SQANTI3 (30). A semiquantitative analysis of the SMRT-seq data together with graphical illustrations of the underlying alternative splicing (exon–intron skipping) was performed using a modified TALON-SWAN pipeline (31).

Short-Read RNA-Seq

DNase-digested total RNA samples used for transcriptomic analyses were quantified (Qubit RNA HS Assay; Thermo Fisher Scientific), and quality was measured by capillary electrophoresis using the Fragment Analyzer and Total RNA Standard Sensitivity Assay (Agilent Technologies). All samples in this study showed high RNA quality numbers (mean ± SD 9.6 ± 0.22). The library preparation was performed according to the manufacturer’s protocol using the VAHTS Universal RNA-Seq Library Prep Kit for Illumina V6 (San Diego, CA) with the mRNA capture module. Briefly, 650 ng total RNA were used for mRNA capture, fragmentation, cDNA synthesis, adapter ligation, and library amplification. Bead-purified libraries were normalized and finally sequenced on the HiSeq 3000/4000 system (Illumina), with a read setup of single reads 1 × 150 bp. The bcl2fastq tool was used for converting bcl files to fastq files as well for adapter trimming and demultiplexing.

SMRT-Seq and Short-Read Pipeline and Statistical Analysis of Short-Read Data

Transcriptomic long-read data were preprocessed to build up an organ- and disease state–specific map of the human transcriptome using the pipeline above. Afterward, transcriptomic short-read data were mapped against the novel reference transcriptome built on the SMRT-seq data set. This integrative approach allowed us to capture the quantitative strength of short-read sequencing while using an SMRT-seq–based reference to preserve novel and muscle-specific isoforms that would otherwise have been lost mapping solely to standard genome builds. To compare the effect of this mapping, the reads were alternatively also mapped against the classical reference transcriptome GRCh38. To compare both data sets sufficiently, the mapped reads for each individual per mapping method were analyzed in a principal component analysis (PCA), including clinical variables available for each individual.

Initial data analyses on fastq files were conducted with the CLC Genomics Workbench (version 21.0.4; QIAGEN, Venlo, the Netherlands). The reads of all probes were adapter (Illumina TruSeq) and quality trimmed (using the default parameters; i.e., bases below Q13 were trimmed from the ends of the reads, with a maximum of two ambiguous nucleotides).

Scripts for SMRT Analysis

The following scripts were used for analysis of SMRT data on the high-performance cluster using BioConda-Environment. As coding editor, Sublime Text for Windows (version build 4113) was used.

Transcriptomic SMRT-seq data were preprocessed for further use through a modified pipeline using the TALON tool by Wyman et al. (32). This tool enables a technology-agnostic long-read analysis pipeline for transcriptome discovery and quantification. Then, using the python library SWAN (31), a library for the analysis and visualization of long-read transcriptomes, a reference transcriptome was added, and abundance information from the initial SMRT analysis was used to create graphs showing the differential expression between novel and known isoforms within the data set. Therefore, we defined groups according to the experimental setup and used the transcript information to process the corresponding transcript path.

The running environment of all scripts was CentOS 7.7.1908 based at the high-performance cluster (HILBERT; Centre for Information and Media Technology, Heinrich Heine University, Düsseldorf, Germany).

Data and Resource Availability

To ensure the data privacy of the participants, the generated data sets of the currently still ongoing GDS are not publicly available, because they are subject to national data protection laws and restrictions by the ethics committee. However, pseudonymized data and complete sequencing data can be requested through an individual project agreement with the GDS.

Results

Characteristics of the Study Population

By design, groups were matched for age and BMI (Table 1). As expected, participants with type 2 diabetes presented with hypeglycemia and 2.5-fold lower WBIS (all P < 0.001) compared with glucose-tolerant participants (Table 1). There were no differences in body fat, fat-free mass, HDL cholesterol, triglycerides, habitual physical activity, resting energy expenditure, or respiratory quotient (all P > 0.050). Glucose-tolerant control participants had higher total and LDL cholesterol (Table 1). Of note, five of nine participants with type 2 diabetes were taking lipid-lowering medication; however, none were taking such medication in the control group, despite meeting the indication to initiate statin therapy according to the current cholesterol management guidelines (33) (Supplementary Table 1).

Table 1.

Characteristics of the study population

	Participant group		P *
	Type 2 diabetes (n = 9)	Glucose tolerant (n = 9)
Age, years	58.6 ± 5.6	53.7 ± 5.4	0.078
Sex, n
Male	9	9
Female	0	0
Known diabetes duration, years	7.7 ± 3.3
BMI, kg/m2	29.9 ± 2.1	29.2 ± 3.3	0.636
Body fat, %	30.7 ± 6.3	31.7 ± 7.5	0.776
Fat-free mass, %	66.4 ± 7.3	67.6 ± 7.7	0.750
Plasma glucose, mg/dL	170 ± 42	87 ± 6	<0.001
2-h plasma glucose (OGTT), mg/dL		85 ± 22
HbA1c (NGSP), %	7.5 ± 1.1	5.6 ± 0.4	<0.001
HbA1c, mmol/mol	58.4 ± 11.6	37.3 ± 4.0	<0.001
M value, mg ∗ kg−1 ∗ min−1	4.5 ± 1.6	11.1 ± 2.9	<0.001
Total cholesterol, mg/dL	173 ± 32	223 ± 47	0.019
LDL cholesterol, mg/dL	113 ± 38	160 ± 37	0.017
HDL cholesterol, mg/dL	57 ± 10	62 ± 10	0.279
Triglycerides, mg/dL	117 ± 77	107 ± 55	0.739
Habitual physical activity (score)	9.1 ± 1.3	8.8 ± 1.4	0.683
Muscle strength, kg	46.8 ± 7.9	47.0 ± 9.1	0.963
REE, kcal/day	1,831 ± 159	1,649 ± 223	0.064
RQ, a.u.	0.81 ± 0.04	0.84 ± 0.04	0.124

Blood sampling was performed after overnight fasting. Data are shown as mean ± SD unless otherwise indicated.

a.u., arbitrary units; NGSP, National Glycohemoglobin Standardization Program; OGTT, oral glucose tolerance test; REE, resting energy expenditure; RQ, respiratory quotient.

*P values are based on unpaired t test.

Mitochondrial Respiration

Compared with control participants, those with type 2 diabetes presented with 38% lower fatty acid oxidation (FAO)–linked mitochondrial respiration ([ETF]_P; P = 0.022) (Fig. 1A), 37% lower (ETF)_L (P = 0.012) (Fig. 1B), and 29% decrease in (ETF + CI)_E (P = 0.012) (Fig. 1C). No differences were observed in other respiratory states for either protocol assessed (Supplementary Figs. 1 and 2). Group differences in FAO-linked mitochondrial respiration were confirmed by a −36% decrease in substrate control ratio ([ETF]_P/[ETF + CI + CII]_P; P = 0.036 vs. control) (Fig. 1D). To account for possible changes in mitochondrial density, we performed TEM, the gold-standard technique for the assessment of mitochondrial content, which revealed no differences between the groups (P = 0.261) (Fig. 1E and Supplementary Fig. 3). We further confirmed these results in the complete cohort by assessing CS activity, a biomarker of mitochondrial content (27), which also did not differ between the groups (P = 0.082) (Fig. 1F). Finally, intrinsic mitochondrial respiratory capacity, obtained by normalizing mitochondrial respiration by CS activity, was 67% lower for FAO-linked OXPHOS capacity ([ETF]_P/CS activity; P = 0.035) (Fig. 1G). We also observed lower intrinsic mitochondrial respiratory states in those with type 2 diabetes compared with control participants for (ETF)_L, (ETF + CI)_P, (ETF + CI + CII)_L, (ETF + CI + CII)_E, (ETF + CI)_E, (CI)_L-PM, (CI + CII)_E, and (CI)_E, with a trend also for (ETF + CI + CII)_P, (CI + CII)_{D-2.5 mmol/L}, (CI + CII)_P, and (CI + CII)_E (Supplementary Figs. 1 and 2). Respiratory flux control ratio for (ETF)_P/(ETF + CI + CII)_P was also lower in those with type 2 diabetes compared with control participants. All the respiratory flux control ratios calculated for both protocols are listed in Supplementary Table 2.

Figure 1.

Skeletal muscle mitochondrial respiration and content in type 2 diabetes (T2D) and control (CON) groups. Mitochondrial respiration with electron input through ETF ([ETF]_P) (A), leak respiration ([ETF]_L) (B), maximal uncoupled respiration ([ETF + CI]_E) (C), substrate control ratio ([ETF]_P/[ETF + CI + CII]_P) (D), mitochondrial density assessed by transmission electron microscopy (E), mitochondrial content assessed by CS activity (F), and intrinsic mitochondrial respiratory capacity, obtained by normalizing mitochondrial respiration by CS activity, with electron input through ETF ([ETF]_P/CS activity) (G). Data are expressed as mean ± SEM. *P < 0.05.

Full-Length Transcriptome in Human Skeletal Muscle

A comprehensive full-length transcriptome map of human skeletal muscle was generated using mRNA SMRT-seq in both participants with type 2 diabetes and control participants. The sequencing strategy is depicted in Supplementary Fig. 4. We identified 66,702 unique isoforms with 99% predicted accuracy, originating from 13,955 genes in those with type 2 diabetes (Fig. 2A–C) as well as 55,544 unique isoforms originating from 11,129 genes in control participants (Fig. 2D–F). Exploring the underlying mechanism of transcript formation, we identified 76,966 resp. 72,816 known canonical (type 2 diabetes 84% vs. control 88%), 55 resp. 46 known noncanonical (type 2 diabetes and control each 0.06%), 11,162 resp. 7,230 novel canonical (type 2 diabetes 12% vs. control 9%), and 3,462 resp. 2,549 novel noncanonical (type 2 diabetes 3.8% vs. control 3.1%) splice junctions (Fig. 2A and D).

Figure 2.

Quality control analysis of SMRT-seq in type 2 diabetes and control groups. Unique genes and isoforms and their characterization from automated analysis of wild-type (A–C) and mutated (D–F) data sets. Numeric classification of found genes and isoforms (A and D), number of isoforms per gene (B and E), and length distribution of transcripts (C and F) are shown. FSM, full splice match; ISM, incomplete splice match; NIC, novel in catalog; NNC, not novel in catalog; SJ, splice junction.

A majority of all isoforms found in both data sets (type 2 diabetes n = 27,993; control n = 25,891) used junctions and corresponding exons matching the annotation (full splice match). Respectively, 12,886 resp. 10,688 isoforms used known splice junctions in consecutive order, with some parts missing (e.g., last part of a transcript; incomplete splice match). A smaller number of all isoforms were either novel in catalog (type 2 diabetes n = 10,166; control n = 8,583), using known splice junctions but resulting in different transcripts, or novel not in catalog (type 2 diabetes n = 10,924; control n = 7,946), using new splice junctions and resulting in transcripts not previously annotated.

Comparison of SMRT-Seq–Based Transcriptomic Mapping With Short Read–Based Transcriptomics for Data Set Clustering in Skeletal Muscle

To compare the effect of the novel SMRT-seq–based workflow against the conventional reference-based workflow, we mapped short-read transcriptomic reads in skeletal muscle by either conventional (Supplementary Figs. 5A and 6) or SMRT-seq–based reference transcriptome (Supplementary Fig. 5B). We thereby compared the two workflows by PCA in both type 2 diabetes (Supplementary Fig. 5, blue dots) and control groups (Supplementary Fig. 5, red dots). In the conventional approach using reference transcriptomic information published previously as background, the short-read transcriptomic analysis did not sufficiently separate the data set with regard to phenotypic clusters (Supplementary Fig. 5A). In contrast, using the SMRT-seq–derived transcriptome as reference for mapping the short-read transcriptomic information of each individual, distinct phenotypic properties could be associated and revealed, such as lowest WBIS (Supplementary Fig. 5B, upper left corner), highest level of total cholesterol (Supplementary Fig. 5B, upper right corner), or oldest age (Supplementary Fig. 5B, bottom).

Comparative Transcriptomics Between Transcripts in Muscle of Those With Type 2 Diabetes and Control Participants

Comparative transcriptomics of muscle transcripts between participants with type 2 diabetes and control participants revealed notable differences when using long read–based transcriptome as a reference. Gene Ontology analysis, performed with the 2023 Gene Ontology knowledgebase (34), identified upregulated pathways in type 2 diabetes, including those related to the mitochondrial respiratory chain and ATP synthesis, such as the aerobic electron transport chain and ATP metabolic process (Supplementary Fig. 7). Interestingly, the same analysis using the conventional transcriptome reference, although we could observe gene expression differences in single genes between type 2 diabetes and control groups (Supplementary Fig. 6), did not reveal any significantly up- or downregulated functional pathways, highlighting the enhanced data information of long-read sequencing. Among the most frequently expressed genes in type 2 diabetes (Supplementary Fig. 8) were mitochondria-encoded genes (e.g., MT-ND5) involved in mitochondrial respiration, as well as skeletal muscle–specific genes, including CKM and ACTA1. These findings highlight the advantages of SMRT-seq–based transcriptomics in detecting metabolic pathway alterations and gene expression differences that remain undetectable with conventional short-read methods.

Detection of Alternative Splicing Events Through SMRT-Seq

Focusing on genes involved in mitochondrial respiratory function and insulin signaling, SMRT-seq–based mapping revealed differential expression patterns between participants with type 2 diabetes and control participants. Figure 3 presents heatmaps comparing expression levels derived from conventional transcriptomic analysis (left heatmap) and SMRT-seq–based transcriptomics (right heatmap). Notably, the expression patterns between the two approaches differed significantly, with distinct logistical expression changes observed within type 2 diabetes and control groups. Genes such as SIRT and ESRRA were detectable exclusively using the novel SMRT-seq approach, highlighting its superior sensitivity compared with conventional methods. Additionally, the novel technique allowed the detection of splicing variants of ATP5AF1, a gene encoding for the F1 subunit α of ATP synthase (Fig. 3). Using the SMRT-seq reference, splicing variants could be clustered into high- and low-expression groups, revealing differences between type 2 diabetes and control groups. Furthermore, use of the TALON-SWAN pipeline enabled a detailed analysis of ATP5AF1 gene expression in type 2 diabetes and control groups. This approach differentiated between known and novel alternative isoforms (splice variants) and provided a semiquantitative analysis of isoform use between type 2 diabetes and control groups (Fig. 3C). These findings underscore the unique advantages of SMRT-seq for identifying and quantifying alternative splicing events, which are crucial for understanding changes in gene expression associated with type 2 diabetes.

Figure 3.

Expression of selected genes involved in mitochondrial respiration and biogenesis and insulin signaling. A: List of selected genes essential for mitochondrial respiratory function and insulin signaling. B: Heatmaps showing logistic expression based on the conventional (left) or long read–based transcriptomic analysis (right). Values for each sample and mean expression for both type 2 diabetes (T2D) and control (CON) groups are shown. C: Semiquantitative expression of variants of ATP5AF1, a gene encoding for the F1 subunit α ATP synthase, using the novel reference with TALON-SWAN pipeline. Next to the percentage of isoform use in each group, the transcript model as transcript path (intron–exon junctions; right) is also shown. NIC, novel in catalog; NNC, not novel in catalog.

Discussion

In this proof-of-concept study, we applied the innovative SMRT-seq technique to create a highly accurate and tissue-specific full-length transcriptome map of skeletal muscle in individuals with type 2 diabetes and glucose-tolerant control participants. Although the number of isoforms per gene and structural categories by transcript length did not differ between the groups, comparative transcriptomics revealed significant differences in mRNA expression of genes regulating mitochondrial respiratory function and insulin signaling in type 2 diabetes. These findings provide a direct link between altered transcriptomic profiles and reduced skeletal muscle mitochondrial respiration, particularly in FAO, as observed in vivo from metabolic phenotyping.

Short-read sequencing approaches are limited in the detection of the full spectrum of transcripts, particularly novel splice variants, because of their reliance on reference genomes from Ensembl (35,36). The fragmentation of transcripts into short reads frequently results in incomplete assembly and loss of critical splicing information (35,37). Long-read sequencing technologies, such as SMRT-seq and Oxford Nanopore, overcome these limitations by sequencing entire transcripts, providing a comprehensive and accurate view of transcript diversity (36,38). The strength of our study lies in integrating these two approaches, because short-read RNA-seq provided robust quantitative gene expression data, and SMRT-seq enabled the creation of a skeletal-muscle–specific reference transcriptome and detection of novel isoforms. Mapping the short-read data against this SMRT-seq–based reference improved resolution and reduced the systematic loss of information that occurs when relying solely on generic reference genomes. Using this approach in our cohort, SMRT-seq identified 66,702 unique isoforms from 13,955 genes in participants with type 2 diabetes and 55,544 unique isoforms from 11,129 genes in glucose-tolerant control participants with 99% predicted sequence accuracy. These results highlight the superior capacity of SMRT-seq in detecting transcript diversity. Our findings underscore the benefits of mapping short-read data onto an SMRT-seq–based reference transcriptome, enabling the detection of unique metabolic traits as well as identifying critical genes involved in mitochondrial biogenesis and energy homeostasis, such as SIRT and ESRRA (39). These results align with previous studies demonstrating the advantages of SMRT-seq in revealing transcriptomic complexity, especially in the context of metabolic diseases, where alternative splicing plays a significant role (38,40,41).

One of the key findings in our study was the identification of alternative splicing variants of ATP5AF, a gene encoding the F1 subunit of ATP synthase, a central component of mitochondrial OXPHOS (42). SMRT-seq allowed clustering of high- and low-expressed splicing variants of ATP5AF1, two of which were exclusively expressed in glucose-tolerant control participants, but absent in those with type 2 diabetes. Although intriguing, this observation is associative and should be considered hypothesis generating rather than mechanistically proven. These findings align with previous research suggesting that alternative splicing may significantly influence mitochondrial function, especially in metabolic disorders (43). At the same time, our analysis also identified numerous isoforms not directly related to mitochondrial pathways. In turn, differences in transcript isoforms could plausibly contribute to the mitochondrial respiration deficits observed in type 2 diabetes, potentially driving the upregulation of OXPHOS-related pathways as a compensatory response to impaired mitochondrial efficiency. However, mechanistic studies will be required to establish causality. Using the TALON-SWAN pipeline, we differentiated novel isoforms from known ones and quantified their expression levels between groups. This suggests that alternative splicing may contribute to the assembly of functionally distinct protein complexes in mitochondria, potentially altering respiratory efficiency and contributing to the observed impairment in FAO-linked respiration in type 2 diabetes. This finding further supports the notion that SMRT-seq provides a more complete and accurate transcriptomic landscape, particularly in disease contexts where transcriptomic complexity is increased (44). These transcriptomic alterations correlated with a 38% reduction in FAO-linked mitochondrial respiration in participants with type 2 diabetes compared with glucose-tolerant control participants, as well as impaired ETF and CI-linked respiratory states. Importantly, although mitochondrial content (assessed via CS activity and TEM) did not differ between groups, mitochondria-specific respiration normalized to CS activity was significantly lower in those with type 2 diabetes. Interestingly, despite the observed reduction in mitochondrial respiration, we found an upregulation of ATP-related pathways in type 2 diabetes. This could reflect a compensatory response to impaired OXPHOS efficiency, potentially driven by increased reliance on glycolytic ATP production or altered ATP turnover dynamics. Additional studies are required to dissect whether these transcriptomic changes translate into functional metabolic adaptations or indicate mitochondrial stress responses. Our findings propose a link between altered splicing of ATP5AF1 and reduced mitochondrial efficiency in type 2 diabetes, contributing to the inability to effectively use fatty acids and other substrates, a hallmark of insulin resistance (3,4,45).

The strengths of this study include the integration of a cutting-edge SMRT-seq technique with in vivo metabolic phenotyping, enabling the identification of transcriptomic changes linked to mitochondrial dysfunction. By using state-of-the-art methodology including SMRT-seq, high-resolution respirometry, electron transmission microscopy, and Botnia clamp, we generated a robust data set that bridges molecular findings with clinical phenotypes. Moreover, our development of a pipeline combining SMRT-seq and short-read data provides a framework for future research in transcriptomics. However, certain limitations should be acknowledged. Our cohort consisted exclusively of Caucasian men with controlled type 2 diabetes and various diabetes duration, which limits the generalizability of our findings not only to broader populations or other ethnicities but also to women. Additionally, the study design lacks a direct measure of muscle mass, and the relatively small number of participants necessitates further validation in larger cohorts. Furthermore, five of nine participants with type 2 diabetes were taking statins, whereas none of the control participants were, despite some meeting guideline-based criteria for statin therapy. The study was not powered to evaluate the effect of statin use on PCA clustering or other transcriptomic findings. Therefore, we cannot exclude that statin therapy may contribute to some of the observed differences, and this should be considered a limitation.

In conclusion, the combination of advanced sequencing technologies with detailed metabolic phenotyping offers a powerful approach to unravel the mechanisms underlying reduced mitochondrial respiration in type 2 diabetes. Our data highlight mitochondria as key regulators in type 2 diabetes pathophysiology but also point toward novel therapeutic targets to improve mitochondrial function and combat insulin resistance.

This article contains supplementary material online at https://doi.org/10.2337/figshare.30569639.

Appendix

German Diabetes Study Group: Michael Roden, Hadi Al-Hasani, Bengt-Frederik Belgardt, Gidon J. Bönhof, Gerd Geerling, Christian Herder, Andrea Icks, Karin Jandeleit-Dahm, Jörg Kotzka, Oliver Kuß, Eckhard Lammert, Wolfgang Rathmann, Sabrina Schlesinger, Vera Schrauwen-Hinderling, Julia Szendroedi, Sandra Trenkamp, and Robert Wagner and their coworkers who are responsible for the design and conduct of the GDS.

Funding Statement

The GDS was initiated and financed by the German Diabetes Center, which is funded by the German Federal Ministry of Health (Berlin, Germany), the Ministry of Culture and Science of the State of North Rhine-Westphalia (Düsseldorf, Germany), the German Federal Ministry of Education and Research to the German Center for Diabetes Research, the European Community (HORIZON-HLTH-2022-STAYHLTH-02-01 Panel A) to the INTERCEPT–Type 2 Diabetes Consortium, the German Research Foundation (GRK 2576), and the Schmutzler Stiftung, and is receiving funding from the Profilbildung 2020 program, an initiative of the Ministry of Culture and Science of the State of North Rhine-Westphalia. This project was also funded by the Deutsche Forschungsgemeinschaft (German Research Foundation; 493659010) as well as by grants from the Deutsches Zentrum für Diabetesforschung. The sole responsibility for the content of this publication lies with the authors. The funding sources had no influence on the design or conduct of this study; collection, analysis, or interpretation of the data; or preparation, review, or approval of this article.