Amino Acid-level Signal-to-Noise Analysis Aids in Pathogenicity Prediction of Incidentally-identified TTN-Encoded Titin Truncating Variants

Abstract

Background -

TTN, the largest gene in the human body, encodes titin (TTN), a protein that plays key structural, developmental, and regulatory roles in skeletal and cardiac muscle. Variants in TTN, particularly truncating variants (TTNtvs), have been implicated in the pathogenicity of cardiomyopathy (CM). Despite this link, there is also a high burden of TTNtvs in the ostensibly healthy general population. This complicates the diagnostic interpretation of incidentally identified TTNtvs which are of increasing abundance given expanding clinical exome sequencing (ES).

Methods -

Incidentally identified TTNtvs were obtained from a large referral database of clinical ES (Baylor Genetics) and compared to rare population variants from gnomAD and CM-associated variants from cohort studies in the literature. A subset of TTNtv-positive children evaluated for cardiomyopathy at Texas Children’s Hospital (TCH) were retrospectively reviewed for clinical features of cardiomyopathy. Amino acid-level signal-to-noise analysis (S:N) was performed.

Results -

Pathologic hotspots were identified within the A-band and N-terminal I-band that closely correlated with regions of high percent spliced in (PSI) of exons. Incidental TTNtvs and population TTNtvs did not localize to these regions. Variants were re-classified based on current ACMG criteria with incorporation of S:N analysis among TCH cases. Those re-classified as likely pathogenic or pathogenic were more likely to have evidence of CM on echocardiography than those re-classified as variants of unknown significance.

Conclusions:

Incidentally found TTNtvs are common among clinical ES referrals. Pathologic hotspots within the A-band of TTN may be informative in determining variant pathogenicity when incorporated into current ACMG guidelines.

Introduction

Recent advances in genetic sequencing technologies provide unprecedented research and clinical diagnosis opportunities. Exome sequencing (ES), is a cost-effective tool that enables sequencing of all protein-coding regions of the genome thus facilitating detection of both coding and splice-site genetic variants. It has the unique benefit of allowing the accurate diagnosis of individuals with Mendelian disorders that may have atypical presentations or require extensive and expensive lab testing for diagnostic confirmation.¹ While ES is a powerful diagnostic tool, it has also led to the dramatic increase in identification of incidentally identified variants, also known has secondary findings, which are associated with potentially life-threatening disease yet were not suspected at the time of genetic testing. The American College of Medical Genetics and Genomics (ACMG) recognizes that new sequencing platforms have led to a high burden of these incidental variants and provides guidelines for assessing the likelihood of pathogenicity to determine whether variants should be reported during return of results.² Fifty-nine genes (ACMG-59) have been designated “reportable” should an incidental variant be identified if that variant is determined to be likely pathogenic or pathogenic (LP/P). Despite this guidance, discerning the pathogenicity of variants, particularly those related to cardiovascular disease, remains a challenge due to reduced penetrance of disease and variable expressivity.

TTN, the largest gene in the human body, encodes cardiac titin (TTN), a protein that is heavily involved in the functioning of sarcomeres, the basic contracting units of both skeletal and cardiac muscle cells. Given its key roles in the structure, development, and regulation of cardiac sarcomeres, variants in TTN have been implicated in the pathogenicity of highly morbid cardiomyopathies (CM). Indeed, TTN has been found to be one of the most commonly mutated genes among probands with dilated cardiomyopathy (DCM) who underwent genetic testing and is a rare cause of hypertrophic cardiomyopathy (HCM).³ Specifically, heterozygous truncating variants in TTN (TTNtvs), those that lead to early termination of translation, have been implicated in familial cardiomyopathies in both pediatric and adult populations.⁴

Cardiomyopathies are primary diseases of the ventricular myocardium that are not caused by congenital heart disease or abnormal loading conditions, and are classified by morphology and physiology.^5–7 In the pediatric population, DCM and HCM are the most common subtypes, and left ventricular noncompaction CM (LVNC) is less common. Despite the clear link between TTNtvs and CM, there is a high burden of TTNtvs among the ostensibly healthy population. Specifically, TTNtvs have also been identified in up to 3% of control individuals who had no clinical evidence of cardiomyopathy,⁸ which complicates the diagnostic evaluation of TTNtvs, particularly when identified incidentally. Previous studies that identified the location of pathologic TTNtvs have found an overrepresentation of pathologic variants in the A-band region and a lack of these variants in the M-band and Z-disk regions of titin.^{9, 10} Further, given the high degree of splice variability in the transcription of TTN, it is being increasingly established that TTNtvs localizing to exons that are more often spliced, or have a larger percent spliced in (PSI), are more likely to be associated with development of disease. Given the localization of both disease and population-based variation across the gene, and the increasing burden of incidentally identified TTNtvs, additional methods to clarify variant pathogenicity are needed. This is particularly important as variants in TTN are not currently included in the ACMG list of 59 genes to be reported as incidental findings, while other more rare variants linked with CM are included, likely due to this lack of clarity in pathogenicity in TTNtvs.¹¹ With a better understanding of pathologic vs. non-pathologic variation in TTN, inclusion onto this list of reportable findings would be possible.

In the current study, we used amino acid-level signal to noise (S:N) and PSI analysis to compare TTNtvs to identify discrete mutation hotspots within TTN. We then used this tool to predict pathogenicity of TTNtvs identified incidentally during clinical ES testing. Among cases that ultimately underwent evaluation, we found that S:N hotspot analysis incorporation into 2015 ACMG guidelines yielded an increased frequency of pediatric-onset CM when the variants were designated as likely pathogenic vs. uncertain significance (VUS).² We conclude that incidental TTNtvs are potentially commonplace, and S:N analysis may refine the diagnostic utility of the ACMG guidelines.

Methods

Study methods can be found in the supplemental materials file. This research study was approved by the Baylor College of Medicine and Duke University Hospital System Institutional Review Boards. The data that support the findings of this study are available from the corresponding author upon reasonable request.

Results

Burden of TTNtvs among clinical ES referrals

To determine the spectrum and frequency of incidentally identified TTNtvs among clinical ES testing, we identified these variants in a large clinical ES referral cohort. Figure 1 shows cohort derivation for all cohorts. There were 7,938 unrelated clinical ES referrals that met inclusion/exclusion criteria. The demographics of the clinical ES cohort, and variant positive individuals within the cohort, are detailed in Supplemental Table 1. Out of these referrals, 123 probands were found to carry 125 unique TTNtvs giving a prevalence of 1.6% among clinical ES referrals. Among these 125 variants 96 (1.2% of all variants found among ES referrals) were assigned LP/P assessment at time of genetic test reporting and 27 (0.3% of referrals) were VUSs. Ninety of these variants were incidental TTNtvs with valid positions within the meta transcript and were carried forward for additional analysis.

Figure 1.

Summary of cohorts and analyses performed. Flowchart diagraming study design including inclusionary and exclusionary criteria for percent-spliced in (PSI) and amino acid-level signal-to-noise (S:N) modeling and American College of Medical Genetics and Genomics (ACMG) re-classification for TTN-encoded titin truncating variants (TTNtvs). Retrospective clinical evaluation was then conducted on a subset of patients seen at Texas Children’s Hospital (TCH). MAF, minor allele frequency. CM, cardiomyopathy. ES, exome sequencing.

To determine the diagnostic value of these variants for monogenic disease, we established a CM-associated cohort and a rare population-based variant cohort for comparison. The cardiomyopathy cohort, including 3,722 patients with CM, was derived from 17 proband-based studies (Supplemental Table 2), confirmed as likely pathogenic or pathogenic by ClinVar, and included 549 probands with 564 TTNtvs, of which 489 were classified by ClinVar as pathogenic. This gave an overall prevalence of 15% and a cumulative minor allele frequency (MAF) of 0.076 among pathologic cases of CM. The rare population-based cohort was derived from the Genome Aggregation Database (gnomAD) database. From 141,456 subjects in gnomAD, 2,134 TTNtvs were identified to form this cohort. This gave a cumulative MAF of 0.008. Of these, 1,386 were rare population TTNtvs, excluding splice site variants. The yield of TTNtvs in these cohorts is summarized in Figure 2. A summary of the ES and pathologic cohort variants is available in Supplemental Tables 1 and 2. Taken together, these findings show that incidentally identified TTNtvs are found in similar proportions in ES referrals as are found in the general population while CM-associated TTNtvs had a significantly higher yield of 15% among CM patients. There were also significant differences in the proportional makeup of variants in our three cohorts (Figure 2C, Chi-Square Test, p < 0.001). The CM cohort demonstrated 44% stop gained, 13% splice site, and 43% frameshift variants, the ES cohort demonstrated 29% stop gained 27% splice site and 44% frameshift variants, while the gnomAD cohort demonstrated 31% stop gained, 35% splice site, and 34 % frameshift variants. Therefore, the burden of variants among the incidentally identified ES cohort was more similar to the rare population-based gnomAD cohort, while there was a distinctly higher burden with a different underlying composition among the pathogenic CM cohort.

Figure 2.

The yield of TTNtvs among cardiomyopathy cases, gnomAD, and exome sequencing. A) A bar graph representing the minor allele frequency (MAF) of truncating variants in TTN (TTNtv) found in our cardiomyopathy (CM, green, gnomAD (blue), and exome sequencing (ES, orange) cohorts. No significant difference was found between gnomAD and ES cohorts. ***, p = 0.00001. B) A pie chart demonstrating the burden of TTNtv found in the ES cohort (Blue = no variants, orange = 1+ variant). C) Pie charts demonstrating the relative proportion of frameshift (blue), splice site (yellow), and stop gained (gray) variants among CM, gnomAD, and ES cohorts.

Amino acid-level pathogenicity probability analysis

Given the relatively high yield of incidentally identified variants among clinical ES referrals, we next attempted to develop a method to determine the diagnostic weight of these variants. We have previously shown that amino acid-level S:N calculations, comparing frequencies of disease-associated variants with population-based variants at individual amino acid positions, can differentiate between incidentally identified and pathologic variants based on primary sequence location in genes in cardiac channelopathic disease.^{12, 13} Thus, we next applied amino acid-level S:N analysis and correlated with PSI analysis, a measurement of how efficiently sequences of interest are spliced in to transcripts within healthy individuals (Figure 3). Overall, the gnomAD cohort variants localized to 811 unique amino acid positions distributed throughout the primary sequence of TTN. When normalized against the overall frequency of rare TTNtvs found in the gnomAD cohort, we found a marked difference in the S:N values between ES and pathologic variants as well as the location of the highest S:N values seen among the two cohorts. While the highest peak S:N value for the CM cohort was located in the I-band (peak S:N = 333), near the A-band, the majority of high S:N values (peak S:N = 277 to 200) were located heterogeneously throughout the A-band (Figure 3A, 3B). There were also regions of relatively low S:N values for the CM cohort located throughout the A-band, particularly in positions ~16200 to ~16400, ~22400 to ~23200, ~29,300 to ~30,000, and ~34,000 to 35,000. Of note, while smaller peaks are distributed in the M-band (peak S:N = 200, 117), Z-band (peak S:N = 33, 25), and the near-Z region (peak S:N = 83, 67), there is a relative paucity of high S:N values in the CM cohort in the majority of the I-band, with the exception of the previously noted maximum peak near the I-band, A-band junction. The ES cohort has comparatively smaller S:N values, with heterogeneous spread throughout the protein (peak S:N = 31 to 7 outside the I-band), with noted minimum values throughout locations ~15000 to ~20000 in the A-band (peak S:N = 0), with a maximum peak located in the mid I-band (peak S:N = 43). The per position average S:N for the CM case cohort variants was 42.3 ± 2.4 compared to 4.0 ± 0.8 in the clinical ES referral cohort (p<0.05), resulting in a yield that was 10-fold higher in the cardiomyopathy cohort compared to the clinical ES referral variants (Figure 3D). Additionally, TTNtvs found in the CM cohort were more likely to be localized to regions with higher PSI (98.6 ± 0.45%) compared to variants found in the gnomAD (65.2 ± 1.17%) and clinical ES referral (57.8 ± 4.8%) cohorts (Figure 3E). Taken together, this shows that pathologic variants exhibit significantly higher frequency measured with S:N analysis and PSI, compared with incidental variants identified in the clinical ES referral cohort (p<0.05).

Figure 3.

CM variants have large signal-to-noise, large percent spliced in while ES variants have low signal to noise, low percent spliced in. A) Location of affected amino acid for each variant in the three cohorts. Corresponding linear topology of the TTN protein including the Z-disk, I-band, A-band, and M-band. B) Rolling average of the signal-to-noise of the nearest 200 variants of the CM (green) and ES (yellow) cohorts compared to gnomAD (orange) cohort plotted as a rolling average at each amino acid position along the length of TTN. Percent spliced in (PSI) for the respective exon is plotted at each amino acid position is plotted in grey for reference. C) Signal-to-noise vs. PSI for each variant in CM and ES cohorts. D, E) Bar graphs comparing the signal-to-noise of CM and ES variants to gnomAD variants and comparing PSI of CM, gnomAD, and ES variants. * = p < 0.05.

Application of pathogenicity scoring to determine likelihood of incidental variant pathogenicity

To determine the pathogenicity of incidentally identified TTNtvs in the clinical ES referral cohort, we reclassified these variants according to the 2015 ACMG pathogenicity guidelines with incorporation of localization of the variant to areas of high S:N within the criteria.² Initial classification estimated that 78% of the incidentally found TTNtvs in the ES clinical cohort were pathogenic. Re-assignment according to the ACMG guidelines with S:N analysis showed a reduction in proportion that were pathogenic, however, the majority of variants was still deemed pathogenic or likely pathogenic (41% pathogenic, 18% likely pathogenic, and 41% VUS; Figure 4A–B). After reassignment, the per position average S:N and the PSI were 4.4 ± 1.0 and 96.2 ± 2.4%, respectively, for variants reassigned as pathogenic/likely pathogenic and 2.0 ± 0.4) and 18.5 ± 4.7% for TTNtvs reassigned as VUS (Figures 4E and 4F). Overall, these findings suggest an enrichment of variants with higher S:N and PSI among the pathogenic and likely pathogenic incidentally identified variants (Figure 4D).

Figure 4.

ES variants meeting pathogenic ACMG criteria are high percent spliced in and have high signal-to-noise. A) Analysis schematic of exome sequencing (ES) cohort variants re-classified according to 2015 ACMG Pathogenicity Guidelines. B) Pie charts demonstrating proportion of pathogenic (Path, red), likely pathogenic (LP, orange), and variants of unknown significance (VUS, blue) before and after re-assignment. C) Rolling average of the signal-to-noise of the nearest 200 variants of the ES cohort compared to gnomAD database by LP/P (red) and VUS (blue) ACMG criteria after re-assignment with PSI labeled in gray. D) Signal-to-noise graphed vs percent spliced in (PSI) of the exon of the variant for each of the ES cohort variants sorted by ACMG criteria. E. F) Bar graphs comparing relative frequency and PSI of LP/P and VUS variants. * = p < 0.05.

Validation of pathogenicity prediction

To validate the predicted pathogenicity assigned above, we retrospectively reviewed the records of individuals with incidentally found TTNtvs in the clinical ES cohort who were seen at Texas Children’s Hospital. In total, out of 7,938 clinical ES referrals, 12 patients met inclusion criteria for review of records and were evaluated at Texas Children’s Hospital by cardiology, 7 of which had an echocardiogram (Figure 5), thus the final Texas Children’s Hospital clinical validation cohort included 7 patients. The majority (86%) of this cohort was male. The mean age at diagnosis of CM (if present) was 7.4 ± 3.3 years, and the mean length of follow-up was 5.5 ± 2.5 years. Among these 7 individuals, 7 unique TTNtvs were identified, 3 (43%) of which were classified as LP/P and 4 (57%) of which were classified as VUS following reassessment of pathogenicity incorporating ACMG criteria with S:N analysis. In those patients with TTNtvs classified as LP/P, 67% showed clinical evidence of cardiomyopathy, whereas in patients with TTNtvs classified as VUS, 25% showed clinical evidence of cardiomyopathy. This included diagnosis of DCM, LVNC, and limb-girdle muscular dystrophy with mildly depressed ventricular function. These findings suggest that, even among this limited cohort, S:N incorporation into ACMG criteria may have a diagnostic role.

Figure 5.

ES cohort evaluated clinically shows trend towards enrichment for cardiomyopathy in patients with variants classified as pathogenic using signal-to-noise modified ACMG criteria. A) Exome sequencing (ES) cohort was assigned pathogenicity based on 2015 ACMG criteria. A chart review was then performed on those patients seen at Texas Children’s Hospital (TCH) following exclusion of subjects with structural heart disease, mitochondrial disease, or chromosomal abnormalities, or those without echocardiogram. B) A bar graph of the percentage of variants classified as pathogenic or likely pathogenic (LP/P, N = 3) with cardiomyopathy and comparing them to those variants classified as variants of uncertain significance (VUS, N = 4).

Discussion

Pediatric cardiomyopathies are rare, with an estimated annual incidence of 1.1 to 1.5 per 100,000, however these diagnoses can carry significant morbidity and mortality. Forty percent of pediatric patients with DCM, the most common cardiomyopathy in children, will die or undergo heart transplantation within two years of presentation.⁶ Five-year transplant-free survival rates in hypertrophic, restrictive, and LVNC cardiomyopathies are 90%, 30%, and 60%, respectively.¹⁴ In conjunction with the ACMG, the Heart Failure Society of America recommends genetic testing for all pediatric patients with CM.¹⁵ However, TTN is not a gene for which the AMCG recommends reporting of incidental variants, though other more rare variants associated with CM are reportable.¹⁶ In contrast, the recent Babyseq project, which proposed a list of gene loci associated with clinically actionable, and therefore reportable, variants, did include TTN.¹⁷ This highlights that with better ability to distinguish between LP/P and benign TTNtvs, it would be beneficial to include these variants on the ACMG reportable list, where LP/P are recommend to be reported to physicians as clinically relevant and VUSs are not. In lieu of TTN’s inclusion on the ACMG-59, physicians can still find themselves needing to interpret the clinical significance of TTNtvs, as they can be found in instances such as gene panel testing ordered for inappropriate indications and as direct to consumer third party sequencing companies expand, may be reported through these as well. Interpretation of clinical significance of TTNtvs will therefore remain an important part of clinical practice as genetic testing continues to become an expanding modality of clinical assessment.

This is particularly important given the prevalence of TTN variants in the general population. In the current study, we found that 63% of the clinical ES referral cohort hosted at least one TTN variant, and 1.6% hosted a TTNtv. The prevalence of TTNtvs in the gnomAD cohort was 1.6%. In previous studies, it has been found that the prevalence is ~ 3% in the general population, and a recent meta-analysis found TTN variants of any type in up to 23% of patients with DCM.^{8, 18} Taken together, it is clear that genetic sequencing has diagnostic value,^{19, 20} however, it also introduces a large number of incidentally identified rare variants with unclear diagnostic value. This can present a dilemma for the interpreting clinician as well as possibly undue distress for families, with the occurrences of these situations only increasing as more patients undergo broad genetic testing and more genes become recommended as reportable. We hypothesize that the solution to this dilemma lies in a more robust role for S:N analysis in order to help provide clarity in variant pathogenicity.

Previous analyses of TTN variants in CM found that variant location can help evaluate for pathogenicity, with the majority of TTNtvs associated with CM localizing to the A-band coding region.^{9, 21} We also found that pathologic TTNtvs were disproportionately localized to the A band region compared with TTNtvs found in the gnomAD and clinical ES referral cohorts. The A-band of titin is thought to have many roles including acting as a binding/interacting site for several key proteins, including myosin and myosin binding protein C, thus controlling assembly and length of the thick filament. The A-band region of titin also plays an important role in biomechanical sensing and signaling and contains the titin kinase domain, which is thought to play a role in embryonic sarcomere development and structure maintenance as well as acting as a mechanical sensor to regulate muscle protein expression.^22–24 In mouse studies, TTNtvs seem to have little effect on cardiac morphology or function, however when exposed to a pharmacological or hemodynamic stress, those hearts with a TTNtv will dilate and develop systolic dysfunction accompanied by fibrotic changes.²⁵ In studies of cardiomyocytes derived from human induced pluripotent stem cells, TTNtvs caused baseline impairment of contractility and impaired cardiomyocyte response to stress.²⁶ Importantly, however, by utilizing an amino acid level approach, our work demonstrates that not all of the A band is uniformly associated with disease variant regions. Thus, amino acid level resolution allows for identification of discrete regions of the A band that are significantly more pathogenic than others, which can inform which variants are more likely to be clinically impactful.

While knowledge of TTN variant coding region can help predict the likelihood of variant pathogenicity, more refined tools are needed, as studies have also found that, while not as common, variants located in the I-band and Z-disk and can also be associated with CM.^{21, 27} As evidenced in our analysis, amino acid-level S:N calculations may help inform the diagnostic weight of incidentally identified variants; however, additional studies are needed to validate our S:N variant interpretation in TTN and to validate this method more broadly CM associated genes, in general. Indeed, replication of this finding in prospectively followed, large, independent cohorts is needed prior to widespread clinical adoption. We found that not only do LP/P TTNtvs localize to differing regions of TTN compared with benign variants, but they are also significantly more likely to localize to regions with higher PSI.

One limitation of the current study is that TTN-associated CM may present later in life. Thus we cannot exclude the possibility of adult-onset disease. However, a recent, though small, study of pediatric patients with CM found an age of presentation at 9–18 years of age for those with pathogenic TTN variants.³ Thus, among those in the clinical validation cohort with TTNtv VUS and normal echocardiogram findings, we may have captured patients prior to disease onset. Larger studies investigating age of onset of TTN-related CM are needed. An additional limitation is the small size of the clinical validation cohort. While we were able to demonstrate the utility of S:N analysis in distinguishing VUS variants from LP/P variants based on comparisons between the ES, gnomAD, and CM cohorts, a larger clinical validation cohort could strengthen this conclusion. To address this limitation, we compared the variants in the ES cohort that were classified as LP/P after S:N analysis to variants found in two recent cohort studies that identified TTNtvs associated with higher risk of clinical CM.^{28, 29} We were able to validate that ~ 8% of the TTNtvs that we deemed LP/P were also found in the larger cohort study that included a general clinical population and a referral population.²⁹ Given the rarity of each TTNtv variant, future studies are needed to further validate their pathogenicity. Larger studies are also needed to determine whether the make-up of TTNtvs (i.e. frameshift, splice site, and stop gained variants) differ between TTNtvs found among CM cases and those found in general population cohorts. While the current study determined that the proportion of variant type differed between the CM, gnomAD, and ES cohorts, relatively small variant numbers precluded direct comparison of the percentage of each variant type individually.

Conclusions

Incidentally identified TTNtvs are relatively common in the general population as well as in children undergoing exome sequencing. Based on amino acid-level signal-to-noise analysis, variants found in the clinical ES cohort were similar to background variants found in the general population, while variants found in cardiomyopathy cases had clearly different localization patterns. The majority of TTN VUS found in the ES cohort were not associated with signs of cardiomyopathy, and even the pathogenic and likely pathogenic variants in the cohort were not uniformly associated with signs of cardiomyopathy.

Nonstandard Abbreviations and Acronyms

ES

Exome sequencing

ACMG

American College of Medical Genetics and Genomics

LP/P

Likely pathogenic/pathogenic

TTN

Titin

CM

Cardiomyopathy

DCM

Dilated cardiomyopathy

HCM

Hypertrophic cardiomyopathy

TTNtv

Truncating variants in TTN

LVNC

Left ventricular noncompaction

PSI

Percent spliced in

S:N

Signal to noise

VUS

Variant(s) of unknown significance

MAF

Minor allele frequency

gnomAD

Genome Aggregation Database