Genetic and phenotypic landscape of peripartum cardiomyopathy

Abstract

Background:

Peripartum cardiomyopathy (PPCM) occurs in approximately 1:2000 deliveries in the US and worldwide. The genetic underpinnings of PPCM remain poorly defined. Approximately 10% of women with PPCM harbor truncating variants in TTN (TTNtvs). Whether mutations in other genes can predispose to PPCM is not known. It is also not known if the presence of TTNtvs predicts clinical presentation or outcomes. Nor is it known if the prevalence of TTNtvs differs in women with PPCM and preeclampsia, the strongest risk factor for PPCM.

Methods:

Women with PPCM were retrospectively identified from several US and international academic centers, and clinical information and DNA samples were acquired. Next-generation sequencing was performed on 67 genes, including TTN, and evaluated for burden of truncating and missense variants. The impact of TTNtvs on severity of clinical presentation, and on clinical outcomes, was evaluated.

Results:

469 women met inclusion criteria. 10.4% of women with PPCM bore TTNtvs (Odds ration [OR]=9.4 compared with 1.2% in reference population; Bonferroni-corrected P [P*] =1.2x10⁻⁴⁶). We additionally identified overrepresentation of truncating variants in FLNC (OR=24.8, P*=7.0x10⁻⁸), DSP (OR=14.9, P*=1.0x10⁻⁸), and BAG3 (OR=53.1, P*=0.02), genes not previously associated with PPCM. This profile is highly similar to that found in non-ischemic dilated cardiomyopathy (DCM). Women with TTNtvs had lower left ventricular ejection fraction (LVEF) on presentation than did women without TTNtvs (23.5% vs 29%, P=2.5x10⁻⁴), but did not differ significantly in timing of presentation after delivery, in prevalence of preeclampsia, or in rates of clinical recovery.

Conclusions:

This study provides the first extensive genetic and phenotypic landscape of PPCM, and demonstrates that predisposition to heart failure is an important risk factor for PPCM. The work reveals a degree of genetic similarity between PPCM and DCM, suggesting that gene-specific therapeutic approaches being developed for DCM may also apply to PPCM, and that approaches to genetic testing in PPCM should mirror those taken in DCM. Finally, the clarification of genotype/phenotype associations has important implications for genetic counseling.

Background:

Peripartum cardiomyopathy (PPCM) is a rare disease manifesting as heart failure with reduced left ventricular ejection fraction (LVEF) that develops during the late peripartum or postpartum period^{1, 2}. Women exhibit a range of presentations, including cardiogenic shock, and a range of outcomes, including the need for mechanical circulatory support or heart transplantation. The pathobiology of PPCM remains unclear, although there is a strong epidemiological relationship with preeclampsia³, and evidence in rodents and clinical studies has implicated vasculo-hormonal pathways of late gestation, including prolactin, vascular endothelial growth factor, and soluble Fms-like tyrosine kinase 1, in the pathogenesis of PPCM^{4, 5}. Genetic variation has been linked to the development of PPCM. PPCM incidence is higher in women of African American descent, and strikingly higher in populations from Haiti and Nigeria, suggesting a possible genetic contribution^{2, 6}. A number of studies have reported familial clustering of PPCM, as well as co-occurrence with non-ischemic dilated cardiomyopathy (DCM), a phenotypically similar disease, and which itself is linked to genetic variation². Finally, we showed previously that, among 172 women with PPCM, 15% bore truncating loss-of-function (LOF) variants (i.e. frameshift, nonsense and splice-altering) in the gene TTN, a prevalence significantly higher than in a reference population (p=1.3x10⁻⁷), and similar to a cohort of patients with DCM⁷. At least a subset of PPCM thus has a genetic etiology.

At least two critical questions were highlighted by previous studies. First, do LOF variants in genes other than TTN also predispose to PPCM? To date PPCM cohorts have not been of sufficient size to allow assessment of the contribution of other DCM genes. And second, are there genotype/phenotype correlations between women with and without TTN truncating variants (TTNtvs)? To date, the largest PPCM group with both genotype and phenotype information was our post-hoc analysis of 85 women from the Investigations in Pregnancy Associated Cardiomyopathy (IPAC) cohort⁷. In this small group, cardiac function at presentation was similar between women with and without TTNtvs, but at a one year follow up, women carrying TTNtvs had worse cardiac function. By contrast, however, in the Intervention in Myocarditis and Acute Cardiomyopathy 2 (IMAC-2) group, no differences in were observed in ejection fraction at enrollment or follow-up ejection fraction⁷. Thus it remains unclear if TTNtvs in women with PPCM are associated with any aspects of clinical presentation, or predict any clinical outcomes.

We therefore enrolled and examined a large, international cohort of 654 women with PPCM, who provided clinical information and/or DNA. Of these, 469 satisfied clinical and sequencing data quality metrics. We report here the identification of novel genetic determinants of PPCM, and genotypic/phenotypic correlations.

Methods:

The authors declare that all supporting data are available within this article and its online supplementary files.

Patient Recruitment and Exclusion

Eligible women with diagnosis of peripartum cardiomyopathy were retrospectively recruited from international institutions, using the internationally accepted definition of PPCM, most recently updated by the Heart Failure Association of the European Society of Cardiology Working Group, as the development of heart failure and LVEF<=45% towards the end of pregnancy or in the months following delivery, where no other cause is identified.⁸ In the US, patients carrying ICD 9 code of 674.54 or ICD 10 code of O90.3 were recruited from the identified academic centers. Patients were excluded for lack of baseline (i.e. at time of diagnosis) LVEF data or a baseline LVEF greater than 45%, as well as those with prior evidence of congenital heart disease or pre-existing valvular heart disease. Clinical and genetic data from these recruits were combined with data from the previously described cohort of 172 PPCM patients⁷. Chart reviews were performed for pertinent medical history including race and ethnicity, age at diagnosis, delivery date, mode of delivery, time from delivery to PPCM diagnosis, and history of gestational hypertension/pre-eclampsia/superimposed pre-eclampsia. Ancestry was established by patient self-report or by chart review. Cardiac data was reviewed for mode of PPCM diagnosis, ejection fraction (EF), left ventricle end-systolic dimension (LVESD), and left ventricle end-diastolic dimension (LVEDD) at time of diagnosis, presence of arrythmias, use of cardiac medications, follow up cardiac testing and endpoint of PPCM diagnosis including but not limited to recovery, persistent dysfunction, left ventricular assist device (LVAD), transplant, and death. The study was approved by the institutional review board at each study center. All the patients provided written informed consent.

DNA Capture, Sequencing, and Analysis

DNA was extracted using Qiagen’s Gentra Puregene kit. 1μg of DNA was sheared with a Diagenode BioRuptor for thirty minutes set on high with thirty seconds on/off. Sheared DNA underwent library prep using Agilent’s SureSelect XT kit. Barcoded libraries were enriched for DCM-associated genes using a custom Agilent SureSelect bait panel (Seidman DCMv5)⁷, which includes all 364 exons of TTN. Libraries were run on an Illumina NextSeq and variants were called using Agilent’s SureCall.

Reference Population and Rare Variant Definition

The MyCode Community Health Initiative at Geisinger in Pennsylvania and New Jersey is a precision medicine initiative that enrolls patient-participants in a system-wide biobank designed to store blood and other samples for research use. We previously described TTN genotyping on the first 92,455 participants who underwent exomic sequencing⁹, and use this population here as a control group of patients with TTN truncating variants but without a documented diagnosis of cardiomyopathy. The Genome Aggregation Database, GnomAD v2.1.1, was used as a reference population for pooled genetic analysis of the non TTN genes. The data set spans 125,748 exome sequences and 15,708 whole-genome sequences from unrelated individuals. For all comparative analyses, we compared our sequencing data against the GnomAD v2 exomes for homogeneity of read depth and coverage, and used a minor allele frequency (MAF) cutoff at less than 1x10⁻⁴ for rare variants, as used in prior studies in dilated cardiomyopathy (DCM) and hypertrophic cardiomyopathy (HCM)¹⁰.

Variant Analysis and Burden Testing

Protein-altering variants were defined as truncating (nonsense, frameshift and splice site altering) or non-truncating (missense). We only included variants that mapped to regions in the GnomAD dataframe with adequate coverage (defined as read depth >= 10 in at least 90% of individuals sequenced). All identified variants were passed through a filter for MAF < 1x⁻⁴ within the GnomAD database, as well as GnomAD flags including multinucleotide variant, low complexity region, low confidence predicted loss of function and LOFTEE warning. The NCBI ClinVar database was referenced for known pathogenic, unknown or benign associations with cardiomyopathy listed in Supplemental Tables 2 and 3.

Statistical Analysis

Comparisons of clinical and genetic characteristics were performed using Chi-squared and 2-sided Student’s t-tests. Ejection fraction compared against time to diagnosis was evaluated using analysis of variance (ANOVA) and Tukey’s method for multiple comparisons. Variant burden testing was performed for each gene and case excess was defined by the difference in frequency between the PPCM cohort and GnomAD database. An odds ratio with 95% confidence interval was calculated using a presumed case and control ratio inferred by variant frequency of each gene in GnomAD. P-value were adjusted by Bonferroni correction. All analyses were performed using Microsoft excel and R studio software.

Results:

Study Patients and DNA sequencing

654 women with PPCM were identified and recruited from several US and international academic centers (Figure 1). Next-generation sequencing was performed on 67 genes, including TTN (Table I in the supplement). 57 patients were excluded due to insufficient genetic data. Of the remaining 597 women, complete baseline cardiac function information was unavailable in 128, leaving 469 women with adequate genetic and clinical data. Demographic data (Table 1) indicated that 41% of US subjects were of African descent (compared with 13.4% of the US population¹¹, OR= 4.43 [3.51-5.60], p=<0.00001), consistent with previous reports of elevated prevalence of PPCM in subjects of African descent. Of the total cases, 36.5% were accompanied by gestational hypertension or preeclampsia (compared with estimated 5-8% worldwide prevalence¹², OR= 6.6-10.9 [5.4-13.4], p=<0.00001), again consistent with prior reports³. No significant differences were noted in LVEF at presentation or presence of hypertensive disorder between US and non-US centers (Table 1). There was a higher prevalence of patients of African descent in US centers.

Fig. 1:

Inclusion Criteria.

Table 1:

Demographic and clinical characteristics, and burden of rare protein-truncating variants, in each cohort of the current study.

	University of Pennsylvania	Japan	IPAC	IMAC	Germany	Chicago	Pitt	Hopkins	WashU	Yale	Israel	Denmark	Total	US	Non-US	P-value**
Number of Patients	106	8	82	31	108	4	25	19	19	4	41	22	469	290	179
Age -yr	30.49+/−6.7	30.63+/−3.5	29.89+/−6.3	31.29+/−6.8	33.85+/−5.2	23.50+/−8.6	29.67+/−5.6	32.47+/−4.8	27.67+/−6.9	27.75+/−5.9	32.15+/−6.3	30.59+/−6.7	30.94+/−6.4	30.13	32.64	1.3x10−5
Caucasian Descent	48 (46.1%)	0	52 (63.4%)	20 (63.5%)	47 (90.4%)	0	16 (64%)	12 (63.2%)	10 (55.6%)	1 (25%)	35 (87.5%)	22 (100%)	263 (64.1%)	159 (55.21%)	62 (82.67%)	2.62x10−5
African Descent	53 (51.0%)	0	26 (31.7%)	11 (35.5%)	3 (5.8%)	3 (75%)	9 (36%)	5 (26.3%)	8 (44.4%)	2 (50%)	5 (12.5%)	0	126 (30.7%)	118 (40.97%)	5 (6.67%)	1.87x10−19
Other Race	3 (2.9%)	8	4 (4.9%)	0	2 (3.8%)	1 (25%)	0	2 (10.5%)	0	1 (25%)	0	0	21 (5.1%)	11 (3.82%)	8 (10.67%)
Left Ventricular EF	27.2+/−12.4	29.1+/−10.1	29.5+/−9.2	27.2+/−7.4	26.7+/−9.0	32.9+/−6.0	27.8+/−10.2	27.8+/−10.6	31.3+/−8.3	27.0+/−12.4	35+/−8.1	25.9+/−9.5	28.5	28.43	28.6	0.85
Gestational HTN or Pre-eclampsia	42/104 (40.4%)*	6/8 (75%)	32/82 (39%)	NA	24/64 (37.5%)*	1/3 (33.3%)	8/25 (32%)	3/18 (16.7%)*	9/19 (47.4%)	1/3 (33.3%)	6/41 (14.6%)	10/22 (45.5%)	142/389 (36.5%)	96/254 (37.8%)	46/135 (34.1%)	0.51
All truncating variants	15	1	12	4	18	0	3	4	2	0	7	4	70	40 (13.8%)	30 (17.1%)	0.46
TTN truncating variants	9	1	10	4	13	0	0	2	0	0	6	4	49	25 (8.6%)	24 (13.4%)	0.14

^*some patients missing data on hypertensive disorder.

^**P-value US vs non-US by Student’s t-test for continuous variables and by Chi-squared test for discrete variables.

Genetic Variant Burden in PPCM

Variants were identified by next-generation sequencing of 67 genes previously associated with DCM, using target capture methodology (see Methods). These sequence variants were filtered to include those with high sequence quality (read depth of >=10 in 90% of sequenced individuals) and were rare in the reference GnomAD dataset (mean allele frequency (MAF)< 1x10⁻⁴). We focused on truncating variants, including nonsense, frameshift, and canonical splicing variants, because these variants are predicted to have a strong impact on protein function. Among the cohort of 469 women with PPCM, we identified 69 women who bore 70 different rare heterozygous truncating variants in 12 different genes (Table 2 and Table II in the supplement). There were no women bearing homozygous or compound heterozygous truncating variants. One woman bore a truncating variant in both MYH7 and FLCN (P=0.81 versus gnomAD, assuming a random distribution of truncating variants). The prevalence of 70 truncating variants in 469 women with PPCM (14.9%) demonstrates a 4.1 fold increased risk of disease compared with the GnomAD reference population (3.6%, Bonferroni-corrected P [P*] =3.5x10⁻⁸).

Table 2: Prevalence of rare protein truncating variants in peripartum cardiomyopathy (PPCM).

The prevalence of truncating variants in patients with PPCM is compared with controls from the Genome Aggregation Database (gnomAD).

Genes	PPCM Cohort Variants	Cohort Frequency	Controls Gnomad Variants	Gnomad Frequency	Variant Excess	Odds Ratio	Nominal P-Value	Corrected P-Value*
BAG3	1	0.21%	5	0.004%	0.209%	53.05(6.19-454.96)	2.92x10−4	1.96x10−2
DES	1	0.21%	28	0.025%	0.189%	8.72(1.18-64.2)	3.36x10−2	1
DSP	6	1.28%	108	0.087%	1.192%	14.9(6.51-34.06)	1.55x10−10	1.04x10−8
FKTN	1	0.21%	57	0.048%	0.165%	4.46(0.62-32.28)	0.14	1
FLNC	4	0.85%	41	0.035%	0.818%	24.75(8.83-69.38)	1.05x10−9	7.02x10−8
ILK	1	0.21%	59	0.048%	0.166%	4.48(0.62-32.4)	0.14	1
MYH6	2	0.43%	153	0.123%	0.303%	3.47(0.86-14.05)	0.08	1
MYH7	2	0.43%	94	0.075%	0.351%	5.7(1.4-23.21)	1.50x10−2	1
PLEC	1	0.21%	169	0.157%	0.056%	1.36(0.19-9.73)	0.76	1
TMPO	1	0.21%	87	0.069%	0.144%	3.08(0.43-22.13)	0.26	1
VCL	1	0.21%	59	0.047%	0.166%	4.54(0.63-32.84)	0.13	1
TTN	49	10.45%	1453	1.224%	9.224%	9.42(6.97-12.71)	1.81x10−48	1.21x10−46
Summary	70	14.93%	5103	3.619%	11.306%	2.25(1.74-2.9)	5.18x10−10	3.47x10−8

^*Nominal P values are from Odds Ratios, and Corrected P values are adjusted for testing 67 genes with the Bonferroni method.

Of the 70 truncating variants, 49 (70%) were in TTN, and are discussed further below. Six truncating variants were in the Desmoplakin (DSP) gene (odds ratio [OR]=14.9, compared with GnomAD, P*=1.0x10⁻⁸), 4 were in Filamin C (FLNC, OR=24.8, P*=7.0x10⁻⁸), 1 was in BAG3 (OR=53.1, P*=0.02), 2 were in each Myosin Heavy Chain 6 and 7 (MYH6 and MYH7, P*>0.05), and one was in VCL (P*>0.05). 53 of these variants were absent from the more than 125,000 exomes in GnomAD v2.1.1. This profile of genes with significant rare truncating variant enrichment is highly similar to that reported in a large cohort of patients with DCM¹⁰, which identified significant variant enrichment in BAG3, DSP, LMNA, TTN, and VCL (Figure 2). Excess variant burden in FLNC was not reported in DCM, because FLNC is not included in most panels, including in that report. Of the 70 truncating variants identified in women with PPCM, 17 were identical to variants annotated as pathogenic or likely pathogenic for DCM in the ClinVar database (www.ncbi.nlm.nih.gov/clinvar).

Fig. 2: Comparison of the prevalence of rare truncating variants between peripartum cardiomyopathy (PPCM) and the Genome Aggregation Database (gnomAD, a), and between PPCM and non-ischemic dilated cardiomyopathy (DCM, b).

Variant frequencies for DCM are from Mazarrotto et al. 2019, except for FLNC, which is from Ader et al. 2019. Diagonal line represents the equivalence line.

The prevalence of missense variants in each of the 67 genes among the 469 women with PPCM was not significantly different from the burden found in GnomAD, after Bonferroni correction (Table III–IV in the supplement). ANKRD1, DSP, LAMA2, and MYH7 reached nominal statistical significance. Excess burden of missense variants in MYH7 has been reported in DCM¹⁰. Two of the identified missense variants in MYH7 were associated with dilated and hypertrophic cardiomyopathy in the ClinVar database.

There have been recent improvements in the prediction of non-canonical splicing variants that cause cryptic splicing, including the machine learning algorithm SpliceAI¹³ with a positive predictive value (PPV) of ~85%.¹⁴ Application of SpliceAI to our cohort revealed no splicing acceptor loss (AL) or donor loss (DL) variants in TTN, and one DL variant in FLNC (P=0.022 compared with prevalence in gnomAD). SpliceAI also identified two variants, one AL and one DL, in DMD (P=0.008 compared with gnomAD), in which we previously also noted a truncating variant in a woman with PPCM⁷.

Increased risk of PPCM in subjects with TTN truncating Variant (TTNtv)

49 truncating variants were identified in TTN (10.45% of the cohort), OR=9.4 compared with 1.2% in GnomAD (P*=1.2x⁻⁴⁶), and OR=14.5 compared with 0.8% in a collection of >60,000 exomes from volunteer participants in an outpatient clinical setting (Geisinger MyCode Community Health, ref ⁹, P*=1.1x10⁻⁴²). 19, 21, and 9 truncating variants were frameshift, nonsense, and canonical splice site variants, respectively, a distribution similar to that found in GnomAD, in the Geisinger cohort, and in various DCM cohorts (Table V in the supplement and refs^{9, 10, 15}). Eighteen variants have previously been described, while 31 (63%) are new variants. 15 of the variants are identical to variants annotated as pathogenic or likely pathogenic for DCM in the ClinVar database.

TTN encodes 364 exons, only a subset of which are constitutively expressed in adult hearts^{9, 16}. Exons expressed in a high percentage (>90%) of TTN transcripts are termed high Percent Spliced In (hiPSI); mutations in these hiPSI exons are likely to cause DCM¹⁶. Of the 49 TTNtvs identified in women with PPCM, 46 (94%) were in hiPSI exons, a fraction similar to that found in the University of Pennsylvania Medical Biobank (PMBB), representing a tertiary care center referral population enriched for disease prevalence (85%), but strongly enriched compared to the outpatient community Geisinger cohort⁹ and GnomAD reference cohorts (58%, P=5.8x10⁻⁶, and 51.2%, P=5.1x10⁻⁸, respectively). The burden of hiPSI TTNtvs was similar between geographic regions (Table 1), and between racial background (Table 3).

Table 3:

Comparison of clinical characteristics between peripartum cardiomyopathy (PPCM) patients with high-percentage spliced-in (hiPSI) TTN truncating variants (TTNtvs), and those without.

Characteristic	TTN Truncating Variant Present (N=45)	TTN Truncating Variant Absent (N=424)	P Value
Age at Diagnosis -- yr	30.40 +/− 7.04	31.03 +/− 6.28	0.41
Ejection Fraction -- %	23.51 +/− 8.99	29.08 +/− 10.02	2.52x10−4
Race – % (no./total no.)			0.53
Black	24.32 (9/37)	31.38 (116/372)
Caucasian	67.57 (25/37)	63.98 (238/372)
Other	8.11 (3/37)	4.84 (18/372)
Gestational Hypertension** – % (no./total no.)	25 (9/36)	37.5 (132/352)	0.14
LV End Diastolic Diameter*** -- cm	5.71 +/− 1.33	5.68 +/− 0.77	0.93
LV End Systolic Diameter*** -- cm	4.97 +/− 1.34	4.70 +/− 0.90	0.34
Recovery -- % (no./total no.)	67.5 (27/40)	59.6 (211/354)	0.33
Negative Outcome# -- % (no./total no.)	21.74 (5/23)	17.42 (27/155)	0.59

^**Includes patients with gestational hypertension, mild and severe pre-eclampsia, and superimposed pre-eclampsia.

^***Measured on TTE from initial presentation.

^#Negative outcome counted as EF<=35 at 12 months follow up, LVAD, heart transplantation, or death.

P values are from Student’s t-test for continuous variables and by Chi-squared test for discrete variables.

The location along the TTN gene of the 49 TTNtvs identified in women with PPCM is shown in Figure 3, and compared to the locations of TTNtvs identified in a primary outpatient cohort of 1040 DCM patients from London and Singapore¹⁰, and two reference cohorts: the University of Pennsylvania Medical Biobank, and the Geisinger cohort (the London/Singapore group only reported hiPSI TTNtvs). The majority of TTNtvs in women with PPCM mapped to the A band region of TTN, a distribution akin to that observed in the London/Singapore DCM cohort, as well as among patients diagnosed with DCM within the disease-enriched PMBB cohort (dark red bars).

Fig. 3: Genomic characteristics of truncating variants in TTN, and associated clinical phenotypes, in peripartum cardiomyopathy patients. a.

Top: Titin is an integral part of the cardiac sarcomere, spanning from Z-disk to M-line. Bottom: Regions of the TTN gene and protein are designated according to their spatial distribution in the sarcomere: Z-disk, I-band, A-band, M-band. The 364 exons of the TTN gene and their associated proportion spliced in (PSI) are shown, followed by the distribution of truncating variants in those exons identified in the Geisinger cohort (a general outpatient population), in the Penn Medicine Biobank (PMBB) cohort (a tertiary care referral population), the London/Singapore cohort of patients with idiopathic dilated cardiomyopathy (DCM) (*only hiPSI TTNtvs are reported in this cohort), and the current PPCM cohort. Variants found in patients with DCM in the reference cohorts are designated in dark red. b-c, Lack of association between location of TTNtv and either left ventricle ejection fraction (b) or left time to diagnosis (c). Dark and dotted lines indicate linear regression and 95% confidence intervals, respectively. P=0.41 and 0.84 for difference from slope of zero in a and b, respectively.

Clinical Characteristics of PPCM Patients with TTNtv

PPCM patients with hiPSI (i.e. likely pathogenic) TTNtvs presented at a similar age as did PPCM patients without TTNtvs (Table 3). The prevalence of women of African descent also did not differ between the two groups (24% and 31%, P=0.5), nor did the prevalence of TTNtvs in patients of African descent versus other descent (8.8% and 11%, P=0.58). In the reference GnomAD population, TTNtvs were found in 2.1% of persons of African descent (P = 1.3×10⁻³) and in 1.1% of those of European descent (P=5.4×10⁻¹⁶). PPCM patients with hiPSI TTNtv did, however, have significantly lower LVEF at presentation than PPCM patients without TTNtv (23.5% vs 29%, P=2.5x10⁻⁴, Table 3 and Figure 4a). We reported previously that LVEF tends to be lower with increasing time between delivery and diagnosis¹⁷, and observe the same trend among the current cohort (Figure 4b, correlation coefficient−0.21, R² = 0.04, P = 4.1x10⁻⁵). The difference in presenting LVEF between PPCM patients with and without likely pathogenic TTNtvs was most pronounced among patients presenting in the first week after delivery (20.9% vs 31.3%, P=2.3x10⁻⁵, Figure 4b), compared to the subsequent 3 weeks (20.2% vs. 31.3% p=0.035), or thereafter (23.8% vs. 25.2% p=0.58). The prevalence of TTNtvs did not correlate with time to diagnosis, despite the lower average LVEF in women presenting late after pregnancy.

Fig. 4: Impact of TTN truncating variants on left ventricle ejection fraction (LVEF) at presentation, and time to diagnosis after delivery, in patients with peripartum cardiomyopathy (PPCM). a.

Comparison of LVEF at time of presentation between TTNtv positive and negative patients. P = 2.5x10⁻⁴ by Student’s t-test. b, Comparison of LVEF with time to diagnosis after delivery. TTNtv positive cases are indicated in red.

We previously reported preliminary findings, in a limited post-hoc analysis of a subset of 85 women with PPCM, that the presence of hiPSI, likely pathogenic TTNtvs was associated with a lower prevalence of gestational hypertension and preeclampsia, and with worse clinical outcomes⁷. In the current much larger cohort, these associations were not observed. Gestational hypertension and preeclampsia showed only a mild trend for lower prevalence among women with and without TTNtv (27% vs 38%, P=0.21). Recovery of LVEF (defined as >50%) was achieved in 68% and 60% of women with and without TTNtv, respectively (P=0.28), and adverse outcomes (death, advanced therapies including left ventricular assist device placement or heart transplantation, or persistent dysfunction with an LVEF <= 35) occurred in 21% and 17%, respectively (P=0.68). Recovery of LVEF was overall less frequent in women of African versus Caucasian descent (53% versus 64%, P=0.041), but in neither group did TTNtv status affect rates of recovery (P=0.73 and 0.17, respectively). There was no significant difference in prevalence of hiPSI TTNtv between patients with follow-up and those lost to follow-up (Table VI in the supplement).

Conclusions

We present here genotypic and clinical features in 469 PPCM patients, the largest such cohort, and use this information to extend our understanding of genotype-phenotype relationships in this disease. We confirm our previous observation that approximately 10% of PPCM patients carry a TTNtv and further identify a statistically significant excess burden of loss-of-function variants in three genes previously associated with DCM, but not PPCM: DSP, FLNC, and BAG3.

The proteins encoded by these four genes have significantly different role in the cardiomyocyte. TTN is a major component of the sarcomere. DSP encodes desmoplakin, the primary force transducer between cardiac desmosomes and intermediate filaments. Truncating variants in DSP are associated with arrhythmogenic left ventricular cardiomyopath¹⁸. FLNC encodes for filamin C, a cytoskeletal protein expressed in the intercalated discs of cardiac cells. Truncating variants in FLNC have only recently been reported in DCM^19–22, and thus FLNC is still absent from most clinical genetic test panels. BAG3 encodes a chaperone protein critical for chaperone-assisted selective autophagy²³. Despite these differences, the profile of identified LOF variants in PPCM is remarkably similar to that seen in cohorts with DCM. The recent extensive analysis of a primary outpatient cohort of 1040 DCM patients from London and Singapore revealed significant LOF burdens in DSP, BAG3, VCL, LMNA, and MYH7, in addition to the well-established prevalence of TTNtvs¹⁰. Excess LOF burdens in VCL and MYH7 were also seen in our PPCM cohort, but did not reach statistical significance after correction for multiple hypothesis testing, which may simply reflect insufficient statistical power. Moreover, the frequencies of LOF variants in each of these genes range widely (0.2% to 10%) and are remarkably similar between PPCM and DCM (figure 2). There is thus a high degree of genetic similarity between PPCM and DCM, both in which genes bear LOFs, and in the frequency with which LOFs are found in each gene.

A simple model to explain these observations invokes a “two-hit” mechanism, whereby a genetic background, e.g. a TTNtv, predisposes to DCM in response to an additional cardiac stressor, e.g. pregnancy in the case of PPCM. In keeping with this notion, a high burden of TTNtvs has also been noted in cardiomyopathy induced by alcohol and cancer therapy, two other prominent cardiac stressors^{24, 25}. Also in keeping with this notion, the penetrance of most LOF variants is low, suggesting that in the absence of additional cardiac stressors, cardiac function is preserved even in the presence of a genetic predisposition. For example, 95% of subjects in the outpatient Geisinger cohort who were identified to carry hiPSI TTNtvs did not have DCM, or significant cardiac dysfunction on echocardiography⁹. In the context of PPCM, pregnancy is clearly the predominant stressor, often accentuated by preeclampsia, which is associated with 30-50% of cases of PPCM. We have shown previously that pregnancy triggers a vascular stress on the heart, which is accentuated in preeclampsia, and likely contributes to the pathogenesis of PPCM^{5, 26}.

The genetic similarities between PPCM and DCM have implications for the care of women with PPCM. There are currently no recommendations for genetic testing in women with PPCM, and such testing is not reimbursed. In contrast, genetic testing for DCM is widely performed, and usually reimbursed. Current recommendations for genetic testing in DCM vary, but in general, it is recommended that probands and affected family members undergo genetic counseling and genetic testing, especially if there is evidence of conduction disease or family history of sudden death.^27–29 The presence of truncating variants in FLNC, for example, has been associated with malignant arrhythmias, and may warrant more aggressive anti-arrhythmic management (e.g. AICD).^{19, 30} The American College of Medical Genetics (ACGM) also recommends cascade testing of at-risk family members for pathogenic and likely pathogenic variants.²⁸ Our study, which indicates that PPCM and DCM are often manifestations of similar underlying genetic predispositions, therefore suggests that approaches taken to genetic testing in PPCM should mirror those taken in DCM.

Because TTNtvs were found with high frequency in PPCM subjects (~10%), we were able to assess if clinical features of PPCM differ between patients with and without TTNtvs. Left ventricular ejection fraction at presentation was significantly lower in patients with TTNtvs, suggesting that TTNtvs may confer an initially more aggressive disease than other causes of PPCM. This observation is similar to that seen in DCM. For example, among DCM patients in the University of Pennsylvania Biobank, we found LVEF to be significantly lower in patients with TTNtvs compared to those without TTNtvs⁹. LVEF is the best known predictor of clinical outcomes in PPCM², but, surprisingly, despite the lower ventricular function on presentation, the presence of TTNtvs in women with PPCM did not correlate with worse rates of recovery; with lower EF at one year; or with adverse outcomes. The apparent equivalent final recovery rates in the context of lower LVEF at presentation could reflect a higher rate of recovery in TTNtv carriers, as has been suggested in DCM^31–33. These observations differ from those we made, post-hoc, in the much smaller group of 85 women from the IPAC study alone⁷, underscoring the importance of recruiting large cohorts such as the current one, in order to avoid spurious conclusions.

The prevalence of PPCM is at least 4-fold higher in women of African descent than Caucasian descent^{2, 34, 35}. However, the prevalence of TTNtvs among PPCM subjects was similar in these two groups, and the prevalence of TTNtvs in the GnomAD reference population is also similar in these two groups, indicating that TTNtvs cannot explain the higher predisposition to PPCM in women of African descent. This difference could reflect biologic differences inherent to pregnancy as a cardiac stressor, or environmental differences, including access to health care and social determinants of health during pregnancy.

Preeclampsia and hypertension strongly predispose to PPCM³, but the prevalence of TTNtvs also was equivalent in PPCM cases with or without a hypertensive disorder. In the context of the “two-hit” hypothesis described above, these observations suggest that pregnancy as a cardiac stressor affects TTNtv carriers and non-carriers equivalently, regardless of race or presence of hypertension. In addition, the findings indicate that the mechanisms by which gene mutations and hypertensive disorders stress the heart are distinct.

The mechanism by which TTNtvs confer risk for PPCM or DCM remains unclear. We find that the location of truncation within the gene, and thus the predicted protein size, does not correlate with either LVEF at presentation, or time from delivery to diagnosis, indicating that location of the TTNtv does not influence severity of disease. This observation suggests a pathogenic mechanism that is independent of truncated titin protein expression, e.g. haploinsufficiency³⁶, or, if integration of the truncated protein into the sarcomere contributes to disease, that integration of the N-terminus is important, and that the loss of structures associated with the M-line is sufficient to cause disease.

Our study does have limitations. Because PPCM is a rare disease, our cohort size is necessarily limited, and lacks power to detect more subtle genetic predispositions. For example, we detected no significant increase in burden of missense variants in any gene tested, in contrast to much larger studies in DCM¹⁰. Increasing the number of PPCM cases studies will require a large international consortium. Our analyses are also limited to the panel of genes in our DNA capture platform. Whole exome or genome sequencing (WES or WGS) may elicit further genetic predispositions to PPCM, but such analyses will invariably suffer even further from power limitations. In addition, it is important to note that our conclusion of genetic similarity between PPCM and DCM is necessarily based on this limited panel of genes, previously associated with DCM, and thus genetic differences between DCM and PPCM may be uncovered with more comprehensive WES or WGS. Our study is retrospective, and sampling bias may have occurred, although, in light of the absence of dramatic phenotypic differences between TTNtv carriers and non-carriers for example, it is unlikely that referral would have led to a bias with respect to genetic profile. Finally, it is also worth noting that PPCM is a heterogeneous disease, and some women may have had some level of cardiac dysfunction prior to pregnancy.

In summary, we provide the first extensive genetic and phenotypic landscape of PPCM. We identify new truncating variants in FLNC, DSP, and BAG3 with robust evidence of disease association. We report detailed associations between the highly prevalent truncating variants in TTN and clinical presentation and outcomes of PPCM. And we show that PPCM and non-ischemic dilated cardiomyopathy closely share profiles of genetic predispositions, suggesting that approaches to genetic testing in PPCM should mirror those taken with DCM, and that pharmacologic agents that are being developed for the treatment of DCM associated with mutations in these genes will also protect women with genetic PPCM.

Clinical Perspective:

What is new?

• Women with peripartum cardiomyopathy (PPCM) bear a significantly high burden of loss-of-function variants in a number of genes, including TTN, FLNC, DSP, and BAG3.

• The identity and relative abundance of these variants is remarkably similar to that seen in idiopathic dilated cardiomyopathy (DCM), indicating that the genetic predisposition to PPCM and DCM may be the same.

• While PPCM patients with TTN truncating variants present with lower ejection fraction, no significant differences in rates of recovery were seen.

What are the clinical implications?

• Genetic counseling and testing should be considered for women with PPCM, following guidelines developed for DCM.

• Gene-specific therapies established for DCM, such as low threshold for defibrillator devices in patients with DSP or FLNC variants, should be considered.

Nonstandard Abbreviations

PPCM

Peripartum cardiomyopathy

LVEF

left ventricular ejection fraction

DCM

dilated cardiomyopathy

LOF

loss-of-function

TTNtv

TTN truncating variant

IPAC

Investigations in Pregnancy Associated Cardiomyopathy

IMAC-2

Intervention in Myocarditis and Acute Cardiomyopathy 2

EF

ejection fraction

LVESD

left ventricle end-systolic dimension

LVEDD

left ventricle end-diastolic dimension

LVAD

Left ventricular assist device

GnomAD

Genome Aggregation Database

MAF

minor allele frequency

HCM

hypertrophic cardiomyopathy

ANOVA

analysis of variance

OR

odd ratio

PPV

positive predictive value

AL

acceptor loss

DL

donor loss

hiPSI

high Percent Spliced In

PMBB

Pennsylvania Medical Biobank\

ACGM

American College of Medical Genetics

AICD

Automatic Implantable Cardioverter Defibrillator

WES

Whole exome sequencing

WGS

Whole genome sequencing