Category: 3. Business

Busy mornings call for a quick and delicious breakfast, and these meals are the perfect solution. Each of these vegetarian recipes takes no more than 15 minutes to prepare, so you’ll have a satisfying meal that’s ready fast without having to resort to a smoothie. Plus, these dishes are lower in calories and high in protein and/or fiber to help support healthy weight loss. You’ll love options like our Cheesy Bean Toast and our Quinoa & Chia Oatmeal Mix for a nourishing meal that will help you meet your goals. 

Contribution by Piero Cipollone, Member of the Executive Board of the ECB, to a roundtable at Aspen Institute Italia

Rome, 19 December 2025

Money is at the heart of what central banks do.^[1] Ever since central banks have existed, their core role has been to issue money and protect its value. That mandate will not change – but the technological environment in which we deliver it is changing, and it is changing radically.

Digital payments are now the norm, and new technologies are disrupting financial services. Financial institutions have become technological entities, and tech firms have entered the spheres of payments and finance.

Central banks are no exception. If they want money to remain stable, trusted and usable in a digital world, they must help shape that world and modernise central bank money. If they fail to do so, central banks may no longer be able to provide an anchor of stability to the financial system.

In the euro area context, there are good reasons for the central bank to not just follow but take the lead in the transformation of money. As a monetary union, we share a single currency and a single monetary policy. For that to work, we must ensure the singleness of money across the euro area: one euro must always be worth one euro, no matter its form and no matter where in the euro area.

The Eurosystem – that is, the ECB and the national central banks of euro area countries – has played a key role in this respect. In just 25 years the euro has become the currency of 20 countries (soon to be 21) and the world’s second most important currency. The Eurosystem issues euro banknotes, which have become the tangible symbol of Europe’s economic unity. And we have built robust infrastructures – T2 for large-value payments, T2S for securities, TIPS for instant payments and ECMS for collateral – which allow money and assets to move safely and efficiently across the euro area.

We now need to take the next steps. Today, I will discuss the challenges we face, how public and private money can complement each other and what this means for retail, wholesale and cross-border payments. Our strategy is three-fold. First, we are getting ready for the potential issuance of a digital equivalent of cash: the digital euro. Second, starting next year, we will make it possible to settle transactions based on distributed ledger technology (DLT) in central bank money. And third, we are working on interlinking our fast payment system with those of other countries to enhance cross-border payments.

Three problems in search of a solution

But let me start with the challenge we face.

It has sometimes been suggested that digital central bank money is a solution in search of a problem. But it is increasingly acknowledged, even by those that dispute the solution, that we face a real issue in the euro area context. To paraphrase the title of Pirandello’s famous play^[2], I see three problems in search of a solution.

First, retail payments in Europe are still fragmented. The Single Euro Payments Area (SEPA) has integrated credit transfers and direct debits, but we still lack a European solution for everyday payments at the point of sale and in ecommerce that works throughout the euro area.^[3] As a result, we rely heavily on a few non-European card and wallet providers. This dependence puts our strategic autonomy at risk.

Second, the nature of money and payments is changing. Tokenisation and DLT promise more efficient capital markets.^[4] Yet without tokenised central bank money at their core, these new ecosystems would rely on fragmented pools of private settlement assets, reintroducing credit risk and fragmentation. We would be more exposed to the expansion of settlement assets denominated in foreign currencies or issued elsewhere, which would undermine our monetary sovereignty. And public money would no longer provide the anchor of stability into which all private assets can be converted.

Third, cross-border payments remain too slow, too costly and too opaque. Stablecoins offer an alternative. But stablecoins come with a number of risks for domestic currencies and financial systems.^[5] And if dollar-based stablecoins were to expand and continue dominating the market, they could erode the international role of the euro.

In this context, doing nothing is not a sound option. If central bank money were to become marginal in a digital world, we would risk having a less resilient payment system, a less stable financial system, weaker monetary sovereignty and reduced strategic autonomy. European financial institutions and infrastructures would be at a competitive disadvantage, and the euro’s role could diminish.

Building on the complementarity of public and private money

Our mandate does not allow us to ignore these risks. When the foundations of money and payments are shifting, the central bank must evolve as well. Our goal is not to crowd out private innovation, but to provide a solid public foundation on which innovation can flourish safely and at scale.

This requires a renewed public-private partnership across all payment dimensions – retail, wholesale and cross-border. Our strategy rests on three pillars: the complementarity of public and private money, a collaborative approach with market participants and strict technology neutrality.

Central bank money and private money are not rivals, they complement each other. Central bank money provides the ultimate settlement asset, free of credit and liquidity risk, and the reference that makes one euro equal to one euro across banks, instruments and technologies. The convertibility of private money into central bank money gives people confidence that a euro is a euro, whatever form it takes. 

This gives private intermediaries a solid basis on which to provide trusted and innovative services. Moreover, our infrastructures and standards provide common rails that the private sector can use across Europe. This reduces fragmentation, ensures interoperability and lowers costs in a network industry where scale and common standards matter.

We already offer digital central bank money and the associated rails for wholesale payments. And the digital euro would extend the same approach to retail payments, complementing cash with its digital equivalent and offering pan-European rails that private European providers can use to innovate and scale up their solutions. But we cannot stand still in wholesale payments either as the market explores the opportunities associated with tokenised securities, DLT-based trading and settlement, and smart contract automation.^[6] For these innovations to be scaled up safely in Europe, central bank money has clear advantages in terms of safety, scalability and liquidity management compared with private settlement assets constrained by the reserves backing them and market risk. In fact, the private sector has been clear: the absence of central bank money as a settlement asset is a major obstacle to the growth of the digital asset ecosystem.

Our approach is explicitly collaborative. We engage with all stakeholders. We test solutions with the market rather than designing them in isolation. This is what we did in 2024 when we conducted the most extensive exploratory work on wholesale DLT settlement in central bank money in the world to date.^[7] And we are following the same approach in preparing for the possible issuance of the digital euro. For instance, we collaborated with market participants to explore the digital euro’s innovative potential.^[8] And we will launch a pilot exercise that will offer banks an opportunity to gain first-hand experience in a simulated digital euro ecosystem.^[9]

In supporting this digital transformation, we remain technology neutral. While being open to new technologies, we do not pick winners. Instead, we focus on setting the conditions for a safe, integrated system that is fit for the digital age and supports innovation.

Shaping the future of money

So, in practice, what are we doing to help shape the future of money?

The digital euro

In the retail space, we are working on the potential issuance of a digital euro. Assuming that European co-legislators adopt the Regulation on the establishment of the digital euro in the course of next year, a pilot exercise and initial transactions could take place as of mid-2027, and the digital euro could be ready for first issuance in 2029.

The digital euro would be a digital form of cash. It would offer a public solution that is legal tender and can thus be used to pay wherever merchants accept digital payments, throughout the euro area, in both physical and online shops.

The digital euro would extend the benefits of physical cash to the digital sphere. At a time when the role of cash in day-to-day payments is declining, it would ensure that consumers always have a European option to pay digitally. This would increase consumers’ freedom of choice and Europe’s strategic autonomy. The digital euro would be available both online and offline, supporting resilience and privacy. And by avoiding excessive reliance on a few dominant players, it would reduce costs for merchants and ultimately prices for consumers.

The digital euro is also being designed to preserve the role of banks in financing the economy. Banks will distribute the digital euro, maintain customer relationships and manage digital euro accounts or wallets. They will be remunerated for these services. Moreover, we have included safeguards to preserve banks’ role in credit intermediation and monetary transmission:^[10] the digital euro will not bear interest, holding limits will prevent destabilising outflows and links to existing bank accounts will allow consumers to seamlessly pay amounts that exceed their digital euro holdings.

For payment service providers, including banks, the digital euro is an opportunity. A single European standard, backed by legal tender status and an unparalleled acceptance network, will make it easier to scale up European cards, wallets and value-added services. Co-badging existing solutions with the digital euro^[11] and building on common standards will lower the cost of expanding acceptance and make it easier for European initiatives to expand across the euro area.

Tokenised central bank money

In wholesale payments and capital markets, we aim to make tokenised central bank money available to support an integrated European market for digital assets.

Tokenisation can reduce reconciliation, shorten settlement chains, enable atomic delivery-versus-payment and allow near-continuous trading and settlement.^[12] But without a common, risk-free settlement asset, liquidity can splinter, assets may not be traded across platforms and the landscape could fragment along national or private lines.

Tokenisation also offers us the opportunity to design an integrated European market for digital assets – in other words, a digital capital markets union – from the outset. Providing tokenised central bank money is essential for this digital asset ecosystem to grow in Europe and not elsewhere. This will also ensure it is built on European infrastructures, euro settlement and EU-wide rules.

To this end, the ECB is pursuing a dual-track approach.^[13]

Project Pontes will connect market DLT platforms to our existing TARGET services, so that tokenised asset transactions can be settled in central bank money.^[14]

Project Appia will explore two possible approaches for an integrated digital asset ecosystem, which could potentially be combined.^[15] First, a European shared ledger that brings together central bank money, commercial bank money and other assets on a single platform where market stakeholders provide services. Second, a European network of interoperable platforms that reduces current frictions in the market.

Interlinking fast payment systems

In cross-border payments, our objective is openness with autonomy.

Today, many cross-border transactions still pass through long correspondent banking chains, making them slow, costly and opaque. One possible future would see global, dollar-based stablecoins and their platforms dominate cross-border payments, creating risks of new dependencies and currency substitution.

We want a different path.

Within Europe, TIPS already provides instant settlement in central bank money and is being extended across currencies. In the near future, TIPS could evolve into a global hub for instant cross-border payments. By interlinking TIPS with the fast payment systems of other countries, starting with India and other partners worldwide, we can cut intermediaries, shorten transaction chains and lower costs.

The digital euro, too, is being designed with potential international use in mind, in a way that respects other countries’ sovereignty and avoids unwanted currency substitution. It could in time act as a connector, adding another safe option for cross-border payments. Moreover, like TIPS, the digital euro’s design includes multi-currency enabling features that would allow non-euro area countries to use the digital euro infrastructure to offer their own digital currencies and facilitate transactions across these currencies.

Conclusion

Let me conclude.

Technological disruption is transforming money and finance. For Europe, this is both a risk and an opportunity. If we simply rely on foreign private solutions, we will import technologies, standards and dependencies and risk fragmentation and instability. If we act together, we can build an innovative, integrated and resilient digital financial system that has the euro at its core but remains open and respectful of the sovereignty of our partners.

Our strategy is clear. Central bank money must remain available and usable, also in digital form, as the anchor of trust. Public and private sectors must work together. The Eurosystem provides settlement in central bank money and common standards, thereby giving private intermediaries a sound basis for competing and innovating. And markets, not the central bank, will decide which technologies and business models succeed, within a framework that keeps money and payment systems safe and integrated.

In retail payments, the digital euro will complement cash and support a truly European market for everyday digital payments.

In wholesale markets, tokenised central bank money through projects such as Pontes and Appia will make it possible to settle digital asset transactions safely in central bank money.

In cross-border payments, interlinking fast payment systems and exploring tokenised settlement will make payments cheaper, faster and more transparent while preserving our monetary sovereignty.

The choice before us is simple: watch the future of money being shaped elsewhere, or help design it ourselves. By acting now, in partnership with the private sector, Europe can lead in the transformation of money, support its competitiveness and resilience, and deliver tangible benefits for citizens and businesses.

This article first appeared on GuruFocus.

Market participants have been positioning for further upside in precious metals as softer US inflation data has reinforced expectations that borrowing costs could move lower. After the latest core consumer price index showed the slowest pace of increase since early 2021, investors observed conditions that could continue to favor non-yielding assets. Spot gold (GLD) was trading near $4,320 an ounce during Asian hours on Friday and was on track for a second weekly gain, while silver hovered close to record levels. The inflation signal has been supportive, even as the data was partly clouded by the impact of a record six-week US government shutdown that ended last month.

The policy backdrop has remained uncertain despite the Federal Reserve delivering its third consecutive rate cut last week. Officials have offered limited guidance on how quickly further easing could unfold, with traders pricing roughly a 25% chance of another reduction in January. US President Donald Trump has continued to advocate for more aggressive rate cuts next year. In Asia, currency markets reacted to the Bank of Japan’s decision to raise interest rates to the highest level since 1995, with the yen weakening after policymakers refrained from signaling additional tightening. At the same time, geopolitical developments, including escalating US pressure on Venezuela through a blockade of sanctioned oil tankers, have added to gold’s appeal as a potential haven.

Precious metals have delivered an exceptional run this year, with both gold and silver heading toward their strongest annual performances since 1979, supported by sustained central-bank buying and inflows into bullion-backed exchange-traded funds. Platinum has also surged, rising for a seventh straight session and trading near a 17-year high above $1,980 an ounce, as signs of tightening supply in the London market and robust exports to China lifted prices. By midday in Singapore, gold edged 0.3% lower to $4,320.88 an ounce, silver slipped 0.3% to $65.32, platinum eased, and palladium was little changed, while the Bloomberg Dollar Spot Index ticked up 0.1%.

Our cross-disciplinary team is led by Cian McCourt (Corporate and M&A) and Cormac O’Donoghue (Corporate and M&A), and includes Ailish Finnerty (Tax), Colin Rooney (Technology and Innovation), David Molloy (Debt Finance), Kevin Langford (Employment), Richard Ryan and Patrick Horan (Competition and Regulated Markets), Philip Smith (Pensions and Employee Benefits) and Simon Hannigan (Real Estate) working in close collaboration with a team from Paul, Weiss, Rifkind, Wharton & Garrison LLP.

Our full Dublin team included Graham Murtagh and Maria Colom Lawlor (Corporate and M&A), Emily Tyler and Aoibhín Ní Dhubháin (Competition and Regulated Markets), Freda McCusker (Debt Finance), Grace-Ann Meghen and Gavan McLaughlin (Employment), Michael Shovlin and Alan Harney (Pensions and Employee Benefits), Lucy Byrne (Real Estate), Nicola Cavey and Rachel Coyle (Tax), and Aoife Coll and Kerry Burns (Technology and Innovation).

The team was also supported by the work of our colleagues in Belfast, Richard Armstrong and Reuben Kane (Corporate and M&A Northern Ireland), and Chris Fullerton and Madison Bowyer (Employment Northern Ireland).

For more information on the transaction, please visit the GI Partners website.

All participants for all studies provided written or verbal consent and studies were approved by the local ethics committee or institutional review board (IRB).

For the BioME study, protocols were approved by the IRB at the Icahn School of Medicine at Mount Sinai (GCO 07–0529; STUDY-11–01139) and all participants provided informed consent. For the BioVU study, all DNA samples in BioVU are de-identified and have been designated with the IRB, thus allowing the use of blood samples collected for clinical care otherwise scheduled for discard. The program has received IRB approval and was reviewed in detail by the federal Office for Human Research Protections, which agreed with the regulatory designation of the nonhuman participants. For the CathGen study, all participants provided informed consent, and the study was approved by the Duke University IRB.

CARTaGENE obtained ethics approval from the Centre Hospitalier Universitaire Sainte-Justine (reference MP-21-2011-345, 3297). The Danish analyses for CHB/DBDS were conducted within the CHB–CVDC and DBDS cohorts, which were approved by the Danish National Committee on Health Research Ethics (approval NVK-1708829 and NVK-1700407) and the Capital Region Data Protection Agency (approval P-2019-93 and P-2019-99). Participants in FinnGen provided informed consent for biobank research under the Finnish Biobank Act. Alternatively, separate research cohorts, collected before the Finnish Biobank Act came into effect (September 2013) and the start of FinnGen (August 2017), were collected on the basis of study-specific consent and later transferred to the Finnish biobanks after approval by Fimea, the National Supervisory Authority for Welfare and Health. Recruitment protocols followed the biobank protocols approved by Fimea. The Coordinating Ethics Committee of the Hospital District of Helsinki and Uusimaa (HUS) approved the FinnGen study protocol (HUS/990/2017).

The GERA study (Kaiser Permanente Research Program on Genes, Environment, and Health, RPGEH) was approved by the Grand Opportunity Project (IRB CN-09CScha-06-H). For Genes & Health, a favorable ethical opinion for the main genes and health research study was granted by NRES Committee London—South East (reference 14/LO/1240) on 16 September 2014. Queen Mary University of London is the sponsor and data controller. The analyses in HUNT have been approved by the Norwegian Data Protection Authority and the Regional Committee for Medical and Health Research Ethics (REC reference 2014/144).

Ethical approval for the Malmö Diet and Cancer study was obtained from the Lund University IRB, and all participants provided written informed consent. We acknowledge the Penn Medicine Biobank (PMBB) for providing data and thank the patient-participants of Penn Medicine who consented to participate in this research program. We also thank the PMBB team and Regeneron Genetics Center for providing genetic variant data for analysis. The PMBB is approved under IRB (protocol 813913) and supported by the Perelman School of Medicine at the University of Pennsylvania, a gift from the Smilow family, and the National Center for Advancing Translational Sciences of the National Institutes of Health under CTSA award UL1TR001878.

The Northern Swedish Health and Disease Study was approved by the Regional Ethical Review Board in Umeå (Dnr. 07-174 M, Dnr. 2014-348-32 M and Dnr. 2015-326-32 M). The analyses based on data from COSMC, SIMPLER and SMCC were approved by the Swedish Ethical Review Authority (Dnr. 2019-03986). The IUCPQ-UL study was approved by the ethics committee of IUCPQ-UL, and all participants provided written informed consent. For All of Us (AoU), written informed consent was provided in accordance with the primary IRB. AoU data analysis was facilitated through the AoU Researcher Workbench.

The Biobank Japan study was approved by the ethics committees of the RIKEN Center for Integrative Medical Sciences, the Institute of Medical Sciences and the University of Tokyo. Informed consent was obtained from all participants, all of whom were Japanese and registered in the BBJ project. The CAVS-France study was approved by the local ethics committees (CCPPRB Nantes, 404/2002; CPP Sud Méditerranée, 13.061; CCPPRB Hôtel-Dieu Paris, 0611285 and CPP Ile de France 1, 2014-juillet-13625) and all participants provided informed consent for genetic research.

The use of data from Iceland was approved by the National Bioethics Committee (NBC, VSN-15-057). All genotyped participants signed a written informed consent allowing the use of their samples and data in projects at deCODE genetics, approved by the NBC. The activities of the Estonian Biobank are regulated by the Human Genes Research Act, adopted in 2000 specifically for the Estonian Biobank. Individual-level data analysis in the Estonian Biobank was carried out under ethical approval 1.1-12/624 from the Estonian Committee on Bioethics and Human Research (Estonian Ministry of Social Affairs), using data according to release application 6-7/GI/16274 from the Estonian Biobank.

The cases included in the German GWAS were approved by the ethics committees of the University of Bonn and the Technical University of Munich (KaBI DHM). The control groups were drawn from the following biobanks, each approved by their respective local ethics committees: the Heinz Nixdorf Recall Study (University Hospital Essen), the PROCAM-2 Study (University of Münster), as well as the PopGen Biobank and the FOCUS Study (University of Schleswig-Holstein). All participants provided written informed consent.

Participants were recruited by HerediGene and Inspire studies. HerediGene is a population study, a large-scale collaboration between Intermountain Healthcare, deCODE genetics and Amgen. Inspire is Intermountain’s active registry for the collection of biological samples, clinical information, laboratory data and genetic information, from consenting patients diagnosed with any healthcare-related conditions. The Intermountain Healthcare IRB approved both studies, and all participants provided written informed consent before enrollment.

At Mass General Brigham Biobank, all participants provided written/electronic informed consent for broad biological and genetic research. The study protocol to analyze MGBB data was approved by the Mass General Brigham IRB under protocol 2018P001236. At UCLA ATLAS, all individuals provided written informed consent to participate in the study. Patient Recruitment and Sample Collection for Precision Health Activities at UCLA is an approved study by the UCLA IRB (17-001013). The TIMI trials were approved by each site’s IRB or ethics committee, including protocols for genetic analyses.

Biospecimens and associated data used in the Colorado Center for Personalized Medicine (CCPM) study were obtained from the biobank at the University of Colorado Anschutz Medical Campus (CU AMC). All samples and data were collected under IRB-approved protocol (15-0461) with appropriate informed consent from participants. Research using these materials was conducted in accordance with the ethical guidelines and regulations governing human subjects research, upholding the principles of beneficence and nonmaleficence.

Recruitment to the GENCAST study in Leicester was approved as part of the Biomedical Informatics Centre for Cardiovascular Science (BRICCS) project (REC ID 09/H0406/114). For the MVP study, all participants provided informed consent under approval from the Veterans Affairs Central IRB. The study protocols for analyzing UK Biobank (UKB) data were approved under protocol 2021P002228 and conducted under UKB application 7089. At the Center for Interdisciplinary Cardiovascular Sciences, all individuals provided written informed consent to donate valve tissue and cells for research purposes. Experimental work at the Cardiovascular Life Sciences Center is approved by the BWH IRB (2011P001703).

Study populations and phenotyping

The IAVGC comprises 30 studies. Descriptive characteristics for contributing studies are presented in Supplementary Table 1. A consistent AS phenotype was applied across all IAVGC studies (except as otherwise described in Supplementary Methods) using a previously validated definition for AS comprised of ‘International Classification of Diseases’ (ICD) and ‘Current Procedural Terminology’ codes⁷ (Supplementary Table 2). Study-level quality control thresholds are described in detail in Supplementary Methods. Most studies performed genome-wide imputation with the NHLBI Trans-Omics for Precision Medicine (TOPMed) imputation panel⁴². After quality control and imputation, participating studies performed a GWAS using either SAIGE⁴³ or REGENIE⁴⁴ for both autosomes stratified by genetic ancestry, as well as autosomes and the X chromosome, stratified by sex.

Meta-analysis

GWAS summary data were uploaded to central servers at the Broad Institute and the Digital Research Alliance of Canada and consortium-level quality control, including the removal of variants with imputation quality of ≤0.3 or minor allele count of <10, was performed independently by two authors (A.M.S. and L.D.) of this study. Summary statistics in hg19 were converted to hg38 using LiftOver (v1.04.00). LD score regression intercepts were calculated for each GWAS using LD score (v.1.0.1)⁴⁵ and corrected standard errors (SEldsc) were calculated by multiplying the standard error by the square root of the LD score regression intercept in cases where the LD score regression intercept was more than 1. Fixed-effects, inverse-variance weighted meta-analysis was performed using GWAMA (v2.2.2)⁴⁶ with SEldsc to correct for inflation. GWAS meta-analysis was performed for the entire multi-ancestry population as well as for ancestry-stratified and sex-stratified populations. X chromosome analysis was performed by meta-analyzing all sex-stratified X chromosome data. Variants with an MAF of ≥0.001 and present in only one study or with an MAF of <0.001 and present in three or fewer studies were removed from the resulting meta-analysis summary files. Genome-wide significance was defined as P < 5 × 10⁻⁸. Independent lead variants in each GWAS were established by determining the top-most significant variant within a 500-kb region. Lead variants were additionally tested for independence by establishing that each lead variant was independent (r² < 0.2) from all other lead variants in all available 1000 Genomes (1000G) populations. All variant pairs with an r² between 0.1 and 0.2 were additionally evaluated for conditional independence using European genetic ancestry individual-level data in the MVP. Variant pairs were evaluated in association with AS independently, and, in a joint model, adjusting for age², sex and principal components. For variant pairs that were not conditionally independent, the top-most significant variant of the pair was considered the lead variant. Random-effects inverse-variance weighted meta-analysis was performed as a sensitivity analysis for lead variants with significant heterogeneity (q value < 0.05/261 (total number of independent lead variants) = 0.0002). Liability-scale heritability was calculated using LD score (v.1.0.1). Percent variance explained was calculated from independent lead variants using the method described in ref. ⁴⁷.

Sex interaction

We tested for differences in AS effect sizes between males and females for lead variants identified in our multi-ancestry meta-GWAS using⁴⁸:

where ({B}_{m})/({rm{s}}.{rm{e}}{.}_{m}) refers to the β/s.e. in the AS male meta-GWAS, ({B}_{f})/({rm{s}}.{rm{e}}{.}_{f}) refers to the β/s.e. in the AS female meta-GWAS, and (r) refers to the correlation between β in the AS male and female meta-GWAS. We considered effect estimates to be significantly different by sex if the z score for the difference was greater than 3.7 (corresponding to a two-tailed P value of <0.05/252 (total number of independent lead variants in multi-ancestry or ancestry-stratified GWAS) = 2.0 × 10⁻⁴).

We also evaluated whether any lead variants identified in the male or female AS meta-GWASs were independent of those discovered in our combined meta-analysis. We considered sex-specific lead variants independent from combined GWAS lead variants if they were both greater than 500 kb from any full population lead variant and in linkage equilibrium (r² < 0.2) with all full multi-ancestry population lead variants in all 1000G populations. We then evaluated for heterogeneity between the male and female sex-stratified GWAS using a fixed-effects inverse-variance weighted meta-analysis framework for lead variants identified in sex-stratified GWAS (n = 5). Significant heterogeneity was considered for P < 0.05/5 = 0.01.

Transcriptome-wide association analysis

Transcriptomic data were previously generated from human AV samples from 484 individuals who underwent AV replacement or heart transplant at the Institut Universitaire de Cardiologie et de Pneumologie de Québec-Université Laval (IUCPQ-UL) as part of the QUEBEC-CAVS study⁹. All participants provided informed consent and the study was approved by the ethics committee of the IUCPQ-UL. Briefly, RNA sequencing was performed on a NovaSeq 6000 instrument (Illumina), targeting >50 million paired reads per sample. Read counts were generated using GENCODE Release 41 on build GRCh38. Genotyping was performed using the Illumina Global Screening Array. All transcriptomic data were from participants who self-reported as European and clustered with the 1000G Phase 3 European ancestry data. Genotypes were imputed using the TOPMed Imputation Server with the TOPMed Imputation Reference panel (version TOPMed-r2). Variants with an MAF of <0.01 or imputation quality score of <0.3 were excluded.

We used our human AV transcriptomic data to generate a gene-expression model estimating the regulatory effects of SNPs on protein-coding gene expression using the software PredictDB (v7)⁴⁹. Elastic-net models were trained using nested cross-validation from genotype and normalized gene expression data adjusted for age, sex, smoking status (current or not), the first 60 probabilistic estimation of expression residuals⁵⁰ factors, and the first five ancestry-based principal components. Variants were considered to have a regulatory effect on gene expression if they were located within 1 Mb of the transcription start site for any given gene of interest. A model testing the association between a given SNP and gene expression was considered significant when the average Pearson correlation between predicted and observed expression was greater than 0.1 and the estimated P value was less than 0.05.

A TWAS was then performed using the S-PrediXcan extension⁵¹ in MetaXcan (v0.7.4) with European genetic ancestry summary statistics from our autosomal meta-GWAS of AS (chosen to optimize population structure overlap between the AS GWAS and AV samples). The statistical significance threshold was set using Bonferroni correction for the number of genes tested (P < 0.05/10,574 = 4.73 × 10⁻⁶).

Colocalization between eQTLs in human AVs and AS risk was evaluated using COLOC (v3.2.1) for genes identified by TWAS⁵² and variants from the IAVGC AS GWAS located within 1 Mb of a gene’s transcription start or end sites. eQTLs were generated using QTLtools (v1.1)⁵³. The two signals were considered colocalized if their posterior probability of shared signal (PP4) was >0.75. The LocusCompareR package (v1.0.0) was used to validate colocalization⁵⁴.

We also compared relative gene expression between the AV and 43 GTEx⁵⁵ tissues using previously calculated ESS⁹. Briefly, ESS were calculated by dividing the median log₂(transcripts per million) value from AV tissue by the sum of the median log₂(transcripts per million) values of all 43 GTEx v8 tissues. An ESS of greater than 0.1 in the AV (corresponding to AV-specific gene expression of greater than 10% total gene expression in all examined tissues) was considered the threshold for significant AV gene expression enrichment.

eQTL colocalization

We performed eQTL colocalization for all lead variants in autosomes using AS GWAS summary statistics and cis-QTL data from GTEx v8 for relevant extravalvular tissues (heart left atrial appendage, heart left ventricle, lung, liver, skeletal muscle, whole blood, cultured fibroblasts, Epstein–Barr virus-transformed lymphocytes, subcutaneous adipose, visceral omentum adipose, coronary artery, tibial artery and aorta). For each colocalization analysis, AS summary data were subset to a region within 1 Mb around each lead variant, and these were merged with variant–QTL associations from tissue-specific GTEx data. Colocalization was performed using the COLOC (v5.1.0) package in R. We considered a PP4 >0.75 as evidence of colocalization.

Combined SNP to gene

Combined SNP to gene (cS2G) leverages seven different SNP-to-gene prioritization strategies to generate an optimal SNP–gene pair per significant independent SNP⁵⁶. We obtained cS2G annotations for all IAVGC lead variants, and restricted SNP–gene pairs with a cS2G score ≥0.5 to maximize precision/recall.

Causal gene prioritization

For each lead variant, we generated a list of prioritized genes based on the following methods: (1) nearest gene, (2) cS2G, (3) extravalvular eQTL colocalization, (4) protein-altering variation, (5) AV eQTL TWAS and eQTL colocalization and (6) AV gene expression or protein abundance data. We considered a lead variant to be prioritized by protein-altering variation if it was in significant LD (r² > 0.8) with a protein-coding variant. Coding variants were further annotated as damaging if they were missense and predicted by PolyPhen-2 (ref. ⁵⁷) to be probably damaging or by SIFT⁵⁸ to be deleterious or were protein truncating. We considered genes to be prioritized based on human AV transcriptomic data if they were both significant in TWAS (P < 4.73 × 10⁻⁶) and in eQTL colocalization (PP4 > 0.75). AV protein abundance and gene expression data were obtained from published liquid chromatography–mass-spectrometry-based proteomics and transcriptomics datasets from human AV tissue and cultured VICs^18,19. In the dataset discussed in ref. ¹⁹, nine human AV specimens from patients with severe AS were obtained and microdissected into nondiseased, fibrotic and calcific segments of the valve. Mass spectrometry proteomics (n = 9) and transcriptomics (n = 3) were performed comparing diseased and nondiseased segments. Genes were annotated based on whether the protein product was detected in bulk proteomics, whether the gene transcript was identified in bulk transcriptomics, or whether protein abundance/gene expression was differentially apparent across disease states (defined as adjusted P < 0.5 and absolute log₂(fold change) > 0.5). VICs were cultured from these samples (separately from the fibrosa and ventricularis) and subjected to either osteogenic or normal media (NM) conditions. Mass spectrometry proteomics was then performed to compare protein abundances among VICs in osteogenic and NM conditions. In the dataset discussed in ref. ¹⁸, human AV specimens were obtained from patients with severe AS and microdissected into nondiseased, fibrotic and calcific segments. Mass spectrometry proteomics was then performed comparing diseased (calcific or fibrotic) and nondiseased segments. Genes were annotated based on whether their protein products were detected in bulk proteomics or were differentially expressed across disease states (defined as adjusted P < 0.5 and absolute log₂(fold change) > 0.5).

A single causal gene was determined for all lead variants using the following criteria: (1) for lead variants with significant LD (r² > 0.8) to a damaging protein-altering variant, the altered gene was prioritized as the most likely causal gene, (2) by consensus of the greatest number of indicators (including nearest gene, cS2G, extravalvular eQTL colocalization, nondamaging protein-altering variation, AV eQTL TWAS and eQTL colocalization, detection in AV proteomics, or differential gene expression in AV transcriptomics) or (3) for variants with only one indicator or with equal numbers of indicators for more than one gene, the nearest gene was prioritized.

Gene-set enrichment

We used DEPICT (v1) to prioritize causal gene sets from our multi-ancestry AS GWAS summary statistics⁵⁹. We defined significant gene-set enrichment for results with an enrichment P value <0.05/10,968 = 4.6 × 10⁻⁶. We then created similarity matrices among gene sets using the Jaccard index and performed affinity propagation with the apcluster package (v1.4.11) in R, following the method described in ref. ⁶⁰. Exemplar gene sets were then plotted as nodes with the density of edges representing the similarity of genes between sets. We additionally annotated causal gene sets using Enrichr, a web-based software that prioritizes gene-set ontologies from a provider-designated list of genes⁶¹.

Evaluation of pleiotropy

We evaluated pleiotropic associations among all 261 independent lead variants using publicly available summary statistics from the recent PheWAS across 44.3 million genotyped variants in the MVP²⁰. We queried European genetic ancestry PheWAS summary data for our trans-ancestry meta-analysis lead variants and European genetic ancestry lead variants, and African genetic ancestry PheWAS summary data for our African genetic ancestry lead variants. We evaluated associations across all 1,854 binary and 214 quantitative traits in the MVP PheWAS and considered any association with a P value <9.3 × 10⁻⁸ to be statistically significant, which is a Bonferroni-corrected significance threshold accounting for 261 lead variants evaluated across 214 quantitative and 1,854 binary traits.

Development of an AS PRS

We developed an AS PRS using summary data from autosomal AS GWAS meta-analysis performed in a sample excluding the MGBB and UKB populations, which were used to test and validate the PRSs, respectively. PRSs were developed using LDpred2 (ref. ⁶²) and PRS-CS⁶³, both of which use Bayesian approaches using GWAS summary-level data. Autosomal Hapmap3 SNPs were extracted from multi-ancestry and population-specific AS GWAS and used as the inputs of LDpred2 and PRS-CS. LD reference panels for both LDpred2 and PRS-CS were built using 1000G data of matched populations for each population-specific AS meta-GWAS (European, African, Hispanic, East Asian and South Asian). We used a European LD reference panel for our multi-ancestry AS meta-GWAS as the majority of samples in the multi-ancestry IAVGC AS meta-GWAS were European.

For LDpred2, we generated multiple PRSs using a grid of hyperparameters, including assumed heritability of 0.7, 1.0 and 1.4 times the estimated heritability, assumed proportion of causal SNPs as a sequence of 17 values from 1 × 10⁻⁴ to 1 on a log-scale, and sparsity (true or false, representing whether some of the posterior effect size can be shrunk to zero). For PRS-CS, we used the default hyper-parameter settings indicated by the authors—a = 1, b = 0.5, φ = 1 × 10⁻⁶, 1 × 10⁻⁴, 1 × 10⁻², 1. The resulting PRSs were tested using European data from MGBB, a nonoverlapping dataset. For each set of posterior effect sizes generated using either LDpred2 or PRS-CS, we identified the PRS in the MGBB with the best predictive value (highest phenotypic variance explained by r²).

AS PRS evaluation

The best-performing PRS in the MGBB was validated using data from the UKB, UCLA ATLAS⁶⁴, and aggregate data from six TIMI clinical trials (ENGAGE AF-TIMI 48 (ref. ⁶⁵), SOLID-TIMI 52 (ref. ⁶⁶), SAVOR-TIMI 53 (ref. ⁶⁷), PEGASUS-TIMI 54 (ref. ⁶⁸), DECLARE-TIMI 58 (ref. ⁶⁹) and FOURIER (TIMI 59) (ref. ⁷⁰)), all of which are independent samples from those used for the AS GWAS. Cox proportional hazards models were used to calculate HRs in both the UKB and TIMI trials for AS against our continuous, normalized AS PRS in an analysis adjusting for age, sex, genetic ancestry principal components 1–5 and clinical risk factors including T2D, HTN, CAD, HLD, body mass index, current smoking and renal failure (eGFR < 30 ml min⁻¹ 1.73 m⁻²). Testing using TIMI trial data required that individual patient-level data were pooled from the six clinical trials. All analyses were compared to the performance of our previously published AS PRS generated using MVP data²¹. Results from the UKB and TIMI clinical trials were meta-analyzed using fixed-effects, inverse-variance weighting. Logistic regression was used to calculate ORs in UCLA ATLAS for AS against our continuous, normalized AS PRS with the same covariates included for the UKB and TIMI trial analyses. We also evaluated whether the AS PRS was associated with incident AV replacement in the UKB, using a composite outcome of surgical or transcatheter AV replacement codes, compared with a control population without any AS. We assessed AS risk prediction for genetic and clinical factors in the UKB using a Cox proportional hazards model including genetic risk categories (top 1%, 2%, 10% and 20% of genetic risk, compared to a referent of middle 40–60% genetic risk), adjusting for age (>65 years), male sex, ancestry-specific principal components, T2D, HTN, CAD, HLD, elevated body mass index (≥30 kg m⁻²), current smoking and renal failure (eGFR < 30 ml min⁻¹ 1.73 m⁻²).

Phenotyping for AS and clinical risk factors in TIMI trials have been previously described²¹. Phenotyping for AS in both the UKB and UCLA ATLAS also used our IAVGC definition as stated above. Individuals with prevalent AS in the UKB were excluded. To evaluate the relative contributions of the AS PRS and individual clinical risk factors, C-indices were calculated for the AS PRS and clinical risk factors either alone or in a full model including both. The C-index was compared across models using likelihood-ratio tests. We also calculated continuous NRIs comparing models with the AS PRS and clinical risk factors to a model with clinical risk factors alone in both the UKB and TIMI trials. Kaplan–Meier curves were drawn using UKB and TIMI clinical trial data, stratified by quintiles of genetic risk.

Isolation of human VICs

Human AV samples were collected from 11 donors undergoing AV replacement surgeries for severe AS at Brigham and Women’s Hospital after written informed consent was obtained (BWH IRB protocol 2011P001703). The AV samples were kept on ice in DMEM culture media (Thermo Fisher Scientific, 11-965-118) and then washed thrice in PBS. Human primary VICs were isolated from the AV leaflets using collagenase digestion. After cutting into 1–2-mm pieces, sections were digested using 1 mg ml⁻¹ collagenase (MilliporeSigma, C5894) in DMEM at 37 °C for 1 h with gentle mixing every 20 min. Valvular endothelial cells were washed away with DMEM and discarded. AV pieces were further digested using 1 mg ml⁻¹ collagenase for 3 h with gentle mixing every 20 min and isolated VICs were collected by centrifugation at 523g (1,500 rpm) for 5 min and plated in 75 cm² culture flasks. Isolated VICs were cultured in growth media (GM) containing DMEM supplemented with 10% FBS, 1% penicillin–streptomycin (PS; Lonza, 17-602E), and 1 mmol l⁻¹ sodium pyruvate (Thermo Fisher Scientific, 11-360-070) in a CO₂ incubator (37 °C, 5% CO₂) until the cells were >90% confluent. Then, cells were detached using 0.05% trypsin–ethylenediaminetetraacetic acid (Thermo Fisher Scientific, 25200056) and plated for subculture. VIC passages 4–7 were used for all experiments.

Gene silencing and calcification detection in human VICs

Human VICs were plated in 24-well or 48-well plates at a density of 1 × 10⁵ cells per ml using GM. After 24 h, cells were transfected with 20 nmol l⁻¹ siRNA of either LTBP4 (Horizon Discovery, L-019552-00-0005), CMKLR1 (Horizon Discovery, L-005467-00-0005), CLCA2 (Horizon Discovery, L-003813-00-0005), CERS2 (Horizon Discovery, L-010282-00-0005) or CEP120 (Horizon Discovery, L-016493-02-0005) and control (Horizon Discovery, P-001810-10-05) using DharmaFECT 1 Transfection Reagent (Horizon Discovery, T-2001-03). After 3 days, GM was replaced with NM or osteogenic media (OM), and this time point was considered as day 0. Furthermore, OM was composed of DMEM supplemented with 10% FBS, 1% PS, 1 mmol l⁻¹ sodium pyruvate, 10 nmol l⁻¹ dexamethasone, 10 mmol l⁻¹ β-glycerophosphate (MilliporeSigma, 35675-100G) and 100 μmol l⁻¹ L-ascorbic acid 2 phosphate (MilliporeSigma, A8960-5G). NM was composed of DMEM with the same concentration of FBS, PS and sodium pyruvate with GM. Media was changed every 3–4 days. siRNA transfection was performed when the media was replaced. Gene silencing by siRNA transfection was confirmed by real-time quantitative PCR (RT–qPCR).

Human VICs were suspended in 0.4-ml RNAzol (MilliporeSigma, R4533) in each well of a 24-well plate, and total RNA was extracted by following the manufacturer’s instructions. In total, 160 µl RNase-free water was added and mixed for 15 s. Samples were incubated at room temperature for 5 min, and centrifuged at 12,000g (10,000 rpm) for 15 min at 4 °C. The upper supernatant was transferred to a new 1.5-ml tube, leaving a layer of the supernatant above the DNA/protein pellet. An equal volume of isopropanol was added to precipitate mRNA, and the samples were incubated at room temperature for 10 min and centrifuged at 12,000g (10,000 rpm) for 10 min. The supernatant was removed, and RNA pellet was washed twice with 160 μl of 75% ethanol (vol/vol). Samples were then centrifuged at 4,000–8,000g for 1 min at room temperature. Alcohol solution was removed with a micropipette. The RNA pellet was solubilized without drying in 20 μl of RNase-free water by pipetting up and down about 30 times. RNA concentration was quantified using NanoDrop 2000 spectrometer (Thermo Fisher Scientific, ND-2000). Next, cDNA was prepared from the RNA sample using qScript RT (KIT F/CDNA SYNTHESIS QSCRIPT; Quanta BioSciences, 95047) as per the manufacturer’s protocol and reverse transcription was performed using Thermal Cycler at 22 °C for 5 min, 42 °C for 30 min and 85 °C for 5 min. Prepared cDNA diluted 1:5 using RNase-free water. PerfeCTa FastMix II ROX (Quantabio, 97065) was used for RT–qPCR with QuantStudio5 real-time PCR system (Thermo Fisher Scientific, A28140) following the manufacturer’s protocol. Gene-specific primers from Life Technologies were used—human GAPDH, Hs02758991_g1; human LTBP4, Hs00943217_m1; human CMKLR1, Hs01081979_s1; human CLCA2, Hs00998923_m1; human CERS2, Hs00371958_g1; human CEP120, Hs00537880_m1. Samples were normalized by endogenous human GAPDH.

Calcium deposition was detected using 2% Alizarin red staining solution (Lifeline Cell Technology, CM-0058). Human VICs were fixed with 10% formalin for 15 min and washed with distilled water. After adding Alizarin red staining solution, cells were stained for 30 min at room temperature. Excess stain was washed thrice with distilled water. Alizarin red staining was extracted using 5% formic acid and calcium content was quantified by absorbance at 450 nm. Statistical analysis was performed using Student’s t tests (two-tailed; paired) for comparison between two groups using Prism 10 (GraphPad). A P value of <0.05 was considered significant. Biological replicates were used for calcification assays and qPCR. Two technical replicates were used for each experimental condition.

Histological assessment of human AV tissues

Five donors of human AV samples were used for histological analysis. AV samples embedded into Optimum Cutting Temperature compound (OCT, Sakura Finetek) were cut into 7-μm serial sections using a cryostat (Leica, CM3050S) followed by immunohistochemical staining. Cryosections were fixed for 5 min in 4% paraformaldehyde solution and incubated for 1 h in blocking solution (PBS, 10% donkey serum, 1% BSA) at room temperature. Sections were then incubated with primary antibodies—anti-LTBP4 antibody (Invitrogen, PA5-85149) and anti-CMKLR1 antibody (Abcam, ab230442), overnight at 4 °C. After washing with PBS, sections were incubated with fluorescence-conjugated secondary antibodies, specifically donkey antigoat IgG (H + L) cross-adsorbed secondary antibody (Alexa Fluor 594, 1:100 dilution; Invitrogen, A-11058) for 45 minutes at room temperature, followed by two washes with PBS. Slides were then incubated with calcium-binding near-infrared imaging fluorescence agent, Osteosense680 (1:1,000) for 30 min and then mounted with a mounting medium containing DAPI (VECTASHIELD, H-1500). The fluorescence signal was examined with a Nikon Eclipse Confocal microscope (Nikon).

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.