NERINE reveals rare variant associations in gene networks across phenotypes and implicates an SNCA-PRL-LRRK2 subnetwork in Parkinson’s disease

NERINE reveals rare variant associations in gene networks across phenotypes and implicates an SNCA-PRL-LRRK2 subnetwork in Parkinson’s disease | bioRxiv /* */ /* */ <!-- <!-- /*! * yepnope1.5.4 * (c) WTFPL, GPLv2 */ (function(a,b,c){function d(a){return"[object Function]"==o.call(a)}function e(a){return"string"==typeof a}function f(){}function g(a){return!a||"loaded"==a||"complete"==a||"uninitialized"==a}function h(){var a=p.shift();q=1,a?a.t?m(function(){("c"==a.t?B.injectCss:B.injectJs)(a.s,0,a.a,a.x,a.e,1)},0):(a(),h()):q=0}function i(a,c,d,e,f,i,j){function k(b){if(!o&&g(l.readyState)&&(u.r=o=1,!q&&h(),l.onload=l.onreadystatechange=null,b)){"img"!=a&&m(function(){t.removeChild(l)},50);for(var d in y[c])y[c].hasOwnProperty(d)&&y[c][d].onload()}}var j=j||B.errorTimeout,l=b.createElement(a),o=0,r=0,u={t:d,s:c,e:f,a:i,x:j};1===y[c]&&(r=1,y[c]=[]),"object"==a?l.data=c:(l.src=c,l.type=a),l.width=l.height="0",l.onerror=l.onload=l.onreadystatechange=function(){k.call(this,r)},p.splice(e,0,u),"img"!=a&&(r||2===y[c]?(t.insertBefore(l,s?null:n),m(k,j)):y[c].push(l))}function j(a,b,c,d,f){return q=0,b=b||"j",e(a)?i("c"==b?v:u,a,b,this.i++,c,d,f):(p.splice(this.i++,0,a),1==p.length&&h()),this}function k(){var a=B;return a.loader={load:j,i:0},a}var l=b.documentElement,m=a.setTimeout,n=b.getElementsByTagName("script")[0],o={}.toString,p=[],q=0,r="MozAppearance"in l.style,s=r&&!!b.createRange().compareNode,t=s?l:n.parentNode,l=a.opera&&"[object Opera]"==o.call(a.opera),l=!!b.attachEvent&&!l,u=r?"object":l?"script":"img",v=l?"script":u,w=Array.isArray||function(a){return"[object Array]"==o.call(a)},x=[],y={},z={timeout:function(a,b){return b.length&&(a.timeout=b[0]),a}},A,B;B=function(a){function b(a){var a=a.split("!"),b=x.length,c=a.pop(),d=a.length,c={url:c,origUrl:c,prefixes:a},e,f,g;for(f=0;f<d;f++)g=a[f].split("="),(e=z[g.shift()])&&(c=e(c,g));for(f=0;f<b;f++)c=x[f](c);return c}function g(a,e,f,g,h){var i=b(a),j=i.autoCallback;i.url.split(".").pop().split("?").shift(),i.bypass||(e&&(e=d(e)?e:e[a]||e[g]||e[a.split("/").pop().split("?")[0]]),i.instead?i.instead(a,e,f,g,h):(y[i.url]?i.noexec=!0:y[i.url]=1,f.load(i.url,i.forceCSS||!i.forceJS&&"css"==i.url.split(".").pop().split("?").shift()?"c":c,i.noexec,i.attrs,i.timeout),(d(e)||d(j))&&f.load(function(){k(),e&&e(i.origUrl,h,g),j&&j(i.origUrl,h,g),y[i.url]=2})))}function h(a,b){function c(a,c){if(a){if(e(a))c||(j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}),g(a,j,b,0,h);else if(Object(a)===a)for(n in m=function(){var b=0,c;for(c in a)a.hasOwnProperty(c)&&b++;return b}(),a)a.hasOwnProperty(n)&&(!c&&!--m&&(d(j)?j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}:j[n]=function(a){return function(){var b=[].slice.call(arguments);a&&a.apply(this,b),l()}}(k[n])),g(a[n],j,b,n,h))}else!c&&l()}var h=!!a.test,i=a.load||a.both,j=a.callback||f,k=j,l=a.complete||f,m,n;c(h?a.yep:a.nope,!!i),i&&c(i)}var i,j,l=this.yepnope.loader;if(e(a))g(a,0,l,0);else if(w(a))for(i=0;i (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0];var j=d.createElement(s);var dl=l!='dataLayer'?'&l='+l:'';j.src='//www.googletagmanager.com/gtm.js?id='+i+dl;j.type='text/javascript';j.async=true;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-M677548'); Skip to main content Home About Submit ALERTS / RSS Search for this keyword Advanced Search New Results NERINE reveals rare variant associations in gene networks across phenotypes and implicates an SNCA-PRL-LRRK2 subnetwork in Parkinson’s disease View ORCID Profile Sumaiya Nazeen , View ORCID Profile Xinyuan Wang , Autumn R. Morrow , Ronya Strom , Elizabeth Ethier , Dylan Ritter , Alexander B. H. Henderson , Jalwa Afroz , Christopher S. Cassa , View ORCID Profile Nathan O. Stitziel , Rajat M. Gupta , View ORCID Profile Kelvin C. Luk , View ORCID Profile Lorenz Studer , View ORCID Profile Vikram Khurana , View ORCID Profile Shamil R. Sunyaev doi: https://doi.org/10.1101/2025.01.07.631688 Sumaiya Nazeen 1 Department of Biomedical Informatics, Harvard Medical School , Boston, MA, USA 2 Division of Genetics, Brigham and Women’s Hospital and Harvard Medical School , Boston, MA, USA 3 American Parkinson’s Disease Association Center for Advanced Research, Harvard Biomarkers Study 2.0 and MyTrial Programs, Division of Movement Disorders, Department of Neurology, Brigham and Women’s Hospital and Harvard Medical School , Boston, MA, USA 4 Broad Institute of MIT and Harvard , Cambridge, MA, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Sumaiya Nazeen Xinyuan Wang 3 American Parkinson’s Disease Association Center for Advanced Research, Harvard Biomarkers Study 2.0 and MyTrial Programs, Division of Movement Disorders, Department of Neurology, Brigham and Women’s Hospital and Harvard Medical School , Boston, MA, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Xinyuan Wang Autumn R. Morrow 1 Department of Biomedical Informatics, Harvard Medical School , Boston, MA, USA 3 American Parkinson’s Disease Association Center for Advanced Research, Harvard Biomarkers Study 2.0 and MyTrial Programs, Division of Movement Disorders, Department of Neurology, Brigham and Women’s Hospital and Harvard Medical School , Boston, MA, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site Ronya Strom 3 American Parkinson’s Disease Association Center for Advanced Research, Harvard Biomarkers Study 2.0 and MyTrial Programs, Division of Movement Disorders, Department of Neurology, Brigham and Women’s Hospital and Harvard Medical School , Boston, MA, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site Elizabeth Ethier 3 American Parkinson’s Disease Association Center for Advanced Research, Harvard Biomarkers Study 2.0 and MyTrial Programs, Division of Movement Disorders, Department of Neurology, Brigham and Women’s Hospital and Harvard Medical School , Boston, MA, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site Dylan Ritter 5 The Center for Stem Cell Biology, Sloan-Kettering Institute for Cancer Research , New York, NY, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site Alexander B. H. Henderson 6 Department of Neurology, Sean M. Healey & AMG Center for ALS, Massachusetts General Hospital and Harvard Medical School , Boston, MA, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site Jalwa Afroz 5 The Center for Stem Cell Biology, Sloan-Kettering Institute for Cancer Research , New York, NY, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site Christopher S. Cassa 2 Division of Genetics, Brigham and Women’s Hospital and Harvard Medical School , Boston, MA, USA 4 Broad Institute of MIT and Harvard , Cambridge, MA, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site Nathan O. Stitziel 7 Cardiovascular Division, John T. Milliken Department of Medicine, Washington University School of Medicine , St. Louis, MO, USA 8 Department of Genetics, Washington University School of Medicine , St. Louis, MO, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Nathan O. Stitziel Rajat M. Gupta 2 Division of Genetics, Brigham and Women’s Hospital and Harvard Medical School , Boston, MA, USA 9 Division of Cardiovascular Medicine, Brigham and Women’s Hospital and Harvard Medical School , Boston, MA, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site Kelvin C. Luk 10 Department of Pathology and Laboratory Medicine, Perelman School of Medicine at the University of Pennsylvania , PA, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Kelvin C. Luk Lorenz Studer 5 The Center for Stem Cell Biology, Sloan-Kettering Institute for Cancer Research , New York, NY, USA 11 Aligning Science Across Parkinson’s (ASAP) Collaborative Research Network , Chevy Chase, MD, 20815 Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Lorenz Studer Vikram Khurana 3 American Parkinson’s Disease Association Center for Advanced Research, Harvard Biomarkers Study 2.0 and MyTrial Programs, Division of Movement Disorders, Department of Neurology, Brigham and Women’s Hospital and Harvard Medical School , Boston, MA, USA 4 Broad Institute of MIT and Harvard , Cambridge, MA, USA 11 Aligning Science Across Parkinson’s (ASAP) Collaborative Research Network , Chevy Chase, MD, 20815 12 Harvard Stem Cell Institute , Cambridge, MA, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Vikram Khurana For correspondence: ssunyaev{at}hms.harvard.edu khuranalab_admin{at}bwh.harvard.edu Shamil R. Sunyaev 1 Department of Biomedical Informatics, Harvard Medical School , Boston, MA, USA 2 Division of Genetics, Brigham and Women’s Hospital and Harvard Medical School , Boston, MA, USA 4 Broad Institute of MIT and Harvard , Cambridge, MA, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Shamil R. Sunyaev For correspondence: ssunyaev{at}hms.harvard.edu khuranalab_admin{at}bwh.harvard.edu Abstract Full Text Info/History Metrics Supplementary material Preview PDF Abstract There are two primary approaches to study the genetic basis of human phenotypes. Experiments in model systems generate interpretable gene networks but, in isolation, do not establish relevance to the human condition. Statistical genetics identifies relevant association signals at the variant or gene level but lacks tools to test specific mechanistic models, as existing methods do not incorporate the topology of gene-gene interactions. We bridge these two strategies by introducing a method that competitively tests network hypotheses with rare variant associations. A hierarchical model-based association test NERINE for the first time incorporates gene network topology while remaining resilient to network inaccuracies. We demonstrate NERINE’s ability to test network hypotheses derived from both canonical pathway databases and model system screens. Comprehensive database-wide search of pathway networks with NERINE uncovers compelling associations for breast cancer, cardiovascular diseases, and type II diabetes, which are undetected by single-gene tests. Testing bespoke networks from experimental screens targeting key PD pathologies: dopaminergic neuron survival and α-synuclein pathobiology, NERINE highlights rare variant burden in gene modules related to autophagy, vesicle traiicking, and protein homeostasis. Genome-scale CRISPRi-screening of α-synuclein toxicity modifiers in human neurons and NERINE converge on PRL , revealing an intraneuronal α-synuclein / prolactin stress response that may impact resilience to PD pathologies. Introduction Advancements in high-throughput screens, such as perturbation assays in model systems, have enabled the generation of vast amounts of data capturing crucial biological knowledge, often represented as gene and protein interaction networks 1 – 7 . Some networks represent well-characterized metabolic or signaling pathways, while others capture regulatory interactions, physical protein-protein interactions, or protein-DNA associations. More abstractly, genetic interactions, such as co-essentiality, reflect dependencies in gene contributions to cellular fitness and phenotypes. As a result, gene networks serve as fundamental frameworks for understanding polygenic trait variation, including complex disease mechanisms, enabling researchers to formulate testable biological hypotheses about human phenotypes 8 – 10 . However, establishing causal links between gene networks and human phenotypes requires direct evidence from human genetics. Rare protein-coding variants oier a direct means of establishing such connections, as these mutations occur within gene bodies and can directly impact gene function 11 , 12 . When tested at the level of pathways and networks, they can reveal new insights about causal disease mechanisms 13 – 16 . Despite their potential, no existing rare variant association test leverages gene networks with defined topologies to assess their association with human phenotypes 14 – 25 . They treat pathways as either a single entity (“mega-gene”) or a simple collection of genes (“bag of genes”), not accounting for the complex interactions between genes and the fact that not all genes in a pathway contribute equally to a phenotype. To address this gap, we present NERINE (NEtwork-based Rare varIaNt Enrichment), a new statistical framework for rare variant association testing that integrates information on genes and their interactions into a hierarchical model, ensuring robustness against network inaccuracies. NERINE oiers several key advantages. First, it allows for the competitive evaluation of network hypotheses, prioritizing network topologies and the experimental assays that define them by their relevance to human phenotypes. Second, it enhances statistical power by aggregating rare variants across functionally connected genes, leading to improved mechanistic insights into the directional eiects of genes on a phenotype. Third, it moves beyond broad and overlapping gene classifications, such as gene ontology, by utilizing experimentally derived relationships with defined edge geometries and signed weights, thereby increasing the biological specificity of the burden test. Importantly, fourth, it allows for inaccuracies (e.g., false positive “nodes”) in that network. This framework paves the way for integrative research strategies whereby genetic screens identify network-based hypotheses to test for association with human phenotypes. Networks demonstrating significant associations can subsequently be refined and validated through targeted experimental approaches, bridging the gap between genetic discovery and functional understanding. We demonstrate two ways of applying NERINE: first, using pre-existing pathway modules to analyze prevalent diseases—breast cancer (BRCA), type II diabetes (T2D), coronary artery disease (CAD), and myocardial infarction (MI) in UK and MGB biobanks; second, leveraging experimentally derived gene networks from model system screens to uncover rare variant associations in a complex neurodegenerative disease like Parkinson’s disease (PD), where existing gene-level rare variant association tests have been thoroughly underpowered with existing cohorts 26 , 27 . Comprehensive testing of database pathway modules in large-scale biobanks with NERINE has enabled us to identify several novel associations, including in the estrogen receptor pathway in BRCA, an adipogenesis-related network in T2D, and non-lipid gene networks (e.g., complement system) in CAD and MI. In PD, NERINE tests network hypotheses generated from experimental screens targeting two core pathologies of PD: i) alpha-synuclein (ɑS) proteotoxicity and ii) human dopaminergic (DA) neuron survival. We uncover associations in gene modules linked to autophagy regulation ( HMGB1-OPTN-USP10 module) and vesicle traiicking and protein homeostasis ( LRRK2-PRL-SNCA module)—findings that all have eluded single-gene tests. One novel finding by NERINE—an association of rare damaging missense variants at the PRL locus encoding prolactin ( Prl ) with PD risk—converges with an independent genome-scale functional screen in a CRISPRi-induced synucleinopathy cortical neuron (CiS-CN) model designed for studying physical and genetic modifiers of α-synuclein (αS) toxicity in cortical neurons 28 . Subsequent functional validation using the CiS-CN model and a well-established mouse PFF model 29 uncovers a hitherto unknown intraneuronal role of PRL in α-synuclein stress response in PD. This underscores the transformative potential of experimental screens and human statistical genetics as mutually reinforcing methodologies to reveal novel disease mechanisms. Results Modeling rare variant burden in gene networks incorporating edge geometry We present NERINE, a new statistical framework for assessing the cumulative eiect of rare variants in a gene network on a dichotomous phenotype ( Figure 1 ). NERINE uniquely incorporates information on network vertices (genes) and network edges (interactions) into a parametric model and is robust with respect to unimportant genes in the network (see Supplementary Table T1 for distinction from other methods). Many data types can represent interactions between genes and gene products (i.e., transcripts and proteins). Interactions may diier in terms of relationship types (from protein complexes to sets of co-expressed genes to protein-DNA or protein-RNA interactions) and scale (from extensive protein interaction networks to pathways of just a dozen genes). Regardless of the data source, we encode gene-gene relationships using a positive semidefinite matrix, Σ. Download figure Open in new tab Figure 1. Overview of NERINE: a rare variant association test leveraging gene network topology. NERINE’s hierarchical framework for testing rare variant burden aggregated across a gene network for binary traits. Inputs include allele counts in cases (𝑿) and controls (𝒀) from WGS or WES datasets and a gene network encoded by a symmetric positive semidefinite matrix, Σ D×D . Σ can represent diverse biological relationships (e.g., physical or genetic interactions, co-expression, co-essentiality, and pathway membership). Gene-eiects are modeled with a multivariate skew-normal distribution parameterized by Σ, network eiect 𝜃, and case-control skew 𝜈. This parameterization allows network genes to have either zero eiect or varying degrees of trait-increasing and decreasing eiects. 𝜃 is inferred using a maximum likelihood estimation (MLE) framework where the likelihood, 𝐿, is an integral over the product of two terms—(i) the product of the per-gene conditional probability of observing allele count in cases (𝑋 i ) given the allele count in the overall cohort (𝑋 i + 𝑌 i ) in the network, and (ii) the probability of observing a specific combination of gene eiects determined by 𝜃 and Σ. NERINE performs nested hypothesis testing, with log-likelihood ratio (LLR) being the test statistic, and provides an asymptotic p -value from the mixture of the delta function, 𝛿(𝑥), and a chi-square distribution with one degree of freedom 𝜒 i 2 . It also estimates the most likely gene eiects under the estimated ( Methods ). We model gene eiect sizes, denoted by vector as draws from a multivariate skew-normal distribution with univariate skew-normal marginals. This hierarchical model captures the biological expectation that functionally related genes exhibit correlated eiect sizes (when non-zero) and correlated probabilities of having no phenotypic eiect. The parameter 𝜃 reflects the importance of the gene group and is the object of inference, while the skewness parameter, ν = 𝑓(𝑁 case , 𝑁 control ) encodes the case–control imbalance ( Figure 1 , Methods ). Here, 𝑁 case and 𝑁 control represent the number of cases and controls, respectively. When case-control groups are perfectly balanced, this model reduces to a multivariate normal, with scalar normal marginals. This parameterization allows gene eiect sizes to vary in both magnitude and direction while permitting some to be zero (in case of noisy networks). NERINE infers 𝜃 using the maximum likelihood estimation (MLE) framework, where the likelihood is given by, We approximate 𝐿 as a weighted sum over multivariate quadrature points from the domain of integration achieved through a lookup table with pruning ( Methods, Supplementary Note ). To ensure computationally eiicient evaluation of the quadrature points, we make several design choices. First, we model rare variant counts in member genes in cases (𝑿) and controls (𝒀) as independent Poisson variables, with rates determined by population allele frequencies renormalized between cases and controls by This implies that the conditional probability of observed rare variant count in each gene in cases, given the total rare variant count in the cohort, follows a Binomial distribution ( Supplementary Note ). Next, we apply a custom transformation to map the unbounded univariate skew-normal marginals of to Beta distributions on [0, 1] ( Supplementary Note ), so that the transformed 𝛼 * s can be used as Binomial success probabilities. We denote the transformed gene-eiects as which incorporates skewness (𝜈) by controlling the shape parameters of the Beta distribution ( Supplementary Note ). When the network has no eiect on the trait, the expected case–control allele ratio reflects sample size proportions in each gene. When the network influences the trait, allele count ratios deviate from this null expectation in relevant genes. The approximate likelihood is given by, NERINE performs nested hypotheses testing, the test statistic being the log-likelihood ratio (LLR). We test the hypothesis 𝜃 = 0 vs. 𝜃 > 0 using the maximum likelihood estimate of 𝜃 ( Methods ). Since 𝜃 lies on the boundary of the parameter space, the test statistic, LLR, follows an asymptotic distribution arising from a weighted mixture of a point mass at zero and a chi-square distribution with one degree of freedom 30 , 31 . We confirmed this asymptotic behavior using extensive simulations with four well-studied canonical pathways ( Supplementary Figure S1 and S2 ), and artificially constructed topologies ( Supplementary Figure S3 ) under the null model (𝜃 = 0). NERINE provides asymptotic p-values for significant networks (𝜃 > 0), and estimates the maximum likelihood eiects of member genes under the inferred 𝜃. However, gene-level p-values are not provided ( Methods ). We benchmarked the performance of NERINE against existing rare variant tests in simulations under the alternative hypothesis (𝜃 > 0) ( Figure 2 , Supplementary Figures S4 and S5 , Methods ). Compared with pathway-level adaptations of such tests, including CMC-Fisher, Fisher minimum p-value, Fisher combined test, SKAT-O, and pathway-based trend test, RVTT (only trait-increasing genes case), NERINE consistently showed higher empirical power, especially in noisy networks ( Figure 2 , Supplementary Figures S4 and S5 ). Positive control experiments using UK Biobank (UKBB) data on LDL and HDL cholesterol (LDL-C and HDL-C) confirmed NERINE’s ability to detect significant rare variant burdens in key lipid-related pathways in European ancestry-specific ( Supplementary Figures S6, S7, and S8 , Methods ) as well as pan-ancestry ( Supplementary Figure S9 , Supplementary Table T2 , Methods ) analyses. Compared to SKAT-O, NERINE provided lower p-values, demonstrating its ability to leverage network connectivity for improved power ( Supplementary Figure S6C ). NERINE’s estimated eiect sizes and directions for individual genes, correctly identified PCSK9 and APOB as reducing LDL cholesterol and LDLR as increasing it, consistent with known biology 32 ( Supplementary Figure S6D ). Similarly, for HDL cholesterol, it linked ABCA1 , LCAT , and APOA1 to low levels and CETP , LIPC , LIPG , and SCARB1 to high levels. ( Supplementary Figure S6E ). Download figure Open in new tab Figure 2. NERINE outperforms existing rare variant association tests in simulations. Power was evaluated using a simulated binary trait (2,000 cases and 2,000 controls) across four canonical pathways (NOTCH pathway, WNT pathway, protein export, and EGFR signaling) with non-zero network eiect (𝜃 = 0.2). Noise was varied by the proportion of genes with non-zero eiects (10–90%), spanning highly noisy (only 10–30% genes with eiect) to highly informative (70–90% genes with eiect) networks. For each noise profile, 250 iterations were performed per network, and power was measured as the positive predictive value (PPV) at diierent significance cutois (1 × 10 -2 , 5 × 10 -3 , 1 × 10 -3 , 5 × 10 -4 , 1 × 10 -4 , 5 × 10 -5 , and 1 × 10 -5 ) ( Methods ). NERINE consistently outperformed existing rare-variant association tests, with the largest gains in noisy settings. All tests were two-sided. Selecting the most informative network topology with NERINE NERINE selects the most informative network topology when multiple alternatives exist for a given gene set, underscoring its ability to leverage the edge geometry of gene networks for more powerful inference. We demonstrated this feature in simulations with (i) four well-characterized pathways of varying sizes (NOTCH, WNT, protein export, and EGFR signaling) for a simulated binary phenotype, and (ii) significant lipid-related pathways on LDL-C and HDL-C phenotypes in UKBB. For the former, NERINE consistently achieved a higher PPV at diierent p-value cutois with “true” networks compared to randomized counterparts, especially in noisy scenarios ( Supplementary Figure S10, Methods ). Similarly, in the latter analysis, NERINE achieved higher log-likelihood ratios and lower p-values with real network topologies compared to random ones ( Supplementary Figure S11, Methods ). When gene-gene relationships in a network are available from multiple data sources, NERINE selects the optimal topology for estimating the burden of rare variants in a network, as showcased using lipid-related pathway gene networks for LDL-C and HDL-C in UKBB ( Supplementary Table T3 ; Methods). For lipid-related pathway gene sets, we applied NERINE on network topologies constructed from three data sources --- (i) high confidence experimentally derived physical and genetic interactions of proteins from canonical databases, (ii) gene-gene co-expression in liver tissue in GTEx (v8), and (iii) co-essentiality of genes in liver cell lines in DepMap3 (v2023Q2) ( Methods ). Our results revealed that the data source providing optimal topology was highly dependent on the specific gene set under investigation ( Supplementary Table T3 ). Figure 2A shows an example scenario where one data source provided more relevant topological information than the others. When comparing individuals with low HDL-C to those with high HDL-C in UKBB, liver tissue co-expression emerged as the most informative source for the core module of HDL-cholesterol-related genes (Bonf. p-value = 1.34e-73). Example scenarios where data sources, other than co-expression, captured the edge relationships of specific gene sets more accurately are shown in Supplementary Figure S12 . For instance, physical and genetic interactions best described the VLDLR internalization and degradation pathway module associated with the LDL-cholesterol phenotype in UKBB (Bonf. p-value = 2.50e-31). Meanwhile, co-essentiality in liver cell lines provided the most relevant edge geometry for the pathway of LDL, HDL, and triglyceride metabolism (Bonf. p-value = 1.58e-46). These findings underscore a crucial insight: diierent data sources encode distinct and complementary aspects of gene relationships. NERINE could eiectively exploit this phenomenon to select the most informative data source. Applying NERINE across tissue-specific co-expression networks spanning all 52 tissue types from GTEx (v8) further reinforced its ability to identify the most informative biological context. Specifically, liver tissue emerged as the most relevant context for core HDL-cholesterol-related genes, where NERINE a high estimated network eiect of 𝜃 = 0.8 with the strongest statistical support (Bonf. p-value = 1.34e-73; Figure 3B , Supplementary Table T4 ). This result highlights a key advantage of NERINE—its capacity to leverage the variability observed in edge relationships and their strengths across diierent tissue contexts for the same set of input genes, thereby selecting the optimal tissue context with biological interpretability. Download figure Open in new tab Figure 3. NERINE selects the most informative network topology. A. NERINE evaluates competing network topologies and selects the most informative one for a given gene set and phenotype. For core HDL-related genes, co-expression in liver (Bonf. 𝑝 = 1.34 × 10 - 73 ; eiect size, = 0.8) outperforms PPI and liver co-essentiality networks when testing for rare damaging variant burden in low vs. high HDL-C individuals (𝑁 case = 26,800; 𝑁 control = 27,178) in UKBB ( Methods ). Here, node colors denote the direction of gene eiects (orange: trait-increasing, and purple: trait-decreasing), with intensity reflecting magnitude. Co-expression and co-essentiality network edges indicate correlation (red: positive, and blue: negative), with edge thickness reflecting the strength of the correlation. PPI network edges represent binary relationships and are colored gray. B. Across 52 GTEx (v8) tissues, NERINE identifies liver as the most informative context for core HDL-related genes when testing for rare damaging variant association with the binarized HDL-C phenotype in UKBB, as in A ( Methods ). NERINE yields = 0.8, the highest LLR (339.32), and strongest significance (Bonf. 𝑝 = 1.34 × 10 -73 ) using the liver co-expression network. Identifying rare variant associations in common diseases across canonical pathway networks NERINE enables comprehensive rare variant burden testing in gene networks for common diseases using biobank-scale data from the UK Biobank (UKBB) and the Mass General Brigham Biobank (MGBBB) ( Figure 4A , Methods ). Applied to breast cancer (BRCA), type II diabetes (T2D), coronary artery disease (CAD), and early-onset myocardial infarction (MI)—which impact 3-13% of the population—NERINE identified significant rare loss-of-function (LoF) burdens in seven pathways for BRCA, one for T2D, two for CAD, and fourteen for MI ( Figure 4B , Supplementary Table T5, Supplementary Figures S13–S15 ). Other than LoF variants, significant burdens were also detected in the damaging and damaging missense categories ( Supplementary Table T5 ), with no enrichment of neutral or synonymous variants. Notably, networks with significant burden in the BRCA phenotype overlapped with biologically plausible pathways, such as cancer susceptibility (i.e., DNA repair, cell cycle regulation, apoptosis) and hormone signaling, while cardiovascular disease networks highlighted lipid metabolism, immune response, blood coagulation, and apoptosis ( Supplementary Figures S13–S15, Supplementary Tables T6 and T7 ), indicating true-positive findings. NERINE-predicted directional eiects of genes in significant pathways per disease phenotype are shown in Supplementary Tables T8-T11. Download figure Open in new tab Figure 4. NERINE identifies rare variant burden in canonical pathway networks across common diseases in UKBB and MGBBB. A. NERINE’s application to case-control WES cohorts in UKBB and MGBBB for rare variant association across database pathway networks. NERINE tests 306 pathway networks (derived from curated physical/genetic interactions) and six variant classes (LoF, damaging missense, damaging, missense, synonymous, neutral missense) ( Methods ). Case-control cohort sizes (𝑁 case /𝑁 control ): BRCA (UKBB: 10,648/91,886; MGBBB: 1,113/2,459), T2D (UKBB: 22,502/68,370; MGBBB: 747/2,188), CAD (UKBB: 4,561/12,321; MGBBB: 902/1,488), and MI (UKBB: 2,521/5,012; MGBBB: 326/2,068). Two-sided p -values were meta-analyzed via Fisher’s combined test and Bonferroni-corrected with Nyholt’s adjustment ( Methods ). B. Bonferroni-significant disease pathways with rare variant burden in LoF, damaging missense, and damaging categories are shown. Negative log-transformed Fisher’s combined p -values are reported; dashed gray line indicates the Bonferroni threshold of 0.05. C. Adipogenesis pathway (BIOCARTA VOBESITY PATHWAY) shows significant rare damaging variant-burden in T2D (avg. = 0.18, Bonf. 𝑝 = 2.33 × 10 -3 ) and serves as a representative example of NERINE’s output. Top: NERINE-predicted gene eiects (averaged across cohorts) across the network nodes (orange: trait-increasing; blue: trait-decreasing; intensity reflects magnitude). Bottom: Information on gene-phenotype association in germline genetics from the GWAS catalog ( https://www.ebi.ac.uk/gwas/ ), Genebass 33 , SAIGE-GENE+ 19 , and disease-specific literature 34 – 36 . Network genes, RXRA , PPARG , ADIPOQ , TNF , NR3C1 , LPL , RETN , and HSD11B1 were not identified in prior rare variant studies; only PPARG and LPL in GWAS (bottom). Five of eight genes are under investigation as T2D drug targets 37 – 43 , highlighting NERINE’s ability to identify therapeutically relevant pathways. Unlike single-gene tests, which had previously identified only a handful of rare variant associations in these diseases (Genebass 33 — BRCA: 6, T2D: 3, CAD: 1, and MI: 1), NERINE captured aggregated eiects of rare variants across gene networks, estimating directional contributions of member genes to putative disease pathways. For instance, in the T2D phenotype, we detected a significant burden of rare damaging variants (i.e., LoF and predicted damaging missense) in the adipogenesis pathway (i.e., BIOCARTA VOBESITY PATHWAY, ( Figure 4C , Supplementary Tables T5 and T9 ), highlighting the potential roles of RXRA , PPARG , ADIPOQ , TNF , NR3C1 , LPL , RETN , and HSD11B1 —genes largely unrecognized in previous rare variant studies 19 , 33 ( Figure 4C , Supplementary Table T9 ). Only PPARG and LPL were implicated in common variant GWAS studies 34 – 36 . Sensitivity analysis showed that while the inclusion of LPL was critical for the database-wide significant signal in this network, the inclusion of PPARG was not ( Supplementary Figure S16 ). Five of the genes had been studied as drug targets before 37 – 43 ( Figure 4C ). Notably, NERINE’s findings align with the fact that the deficiency of adiponectin —a protein encoded by ADIPOQ —increase insulin resistance 39 and the risk of T2D, suggesting ADIPOQ as a potential candidate for therapeutic investigation. Additionally, NERINE’s prediction of trait-decreasing roles of damaging variants in RXRA and HSD11B1 supports their inhibition as possible therapeutic strategies for T2D 40 – 43 , paralleling the example of PCSK9 in cardiovascular disease where protective LoF and damaging missense variants guided eiective therapy development 44 , 45 . However, further experiments are required to distinguish the roles of predicted damaging missenses from LoF variants contributing towards the observed trait-decreasing eiect. While NERINE identified rare damaging variant burden in TNF , NR3C1 , and RETN in the adipogenesis network, their mechanistic roles in T2D remain understudied, highlighting NERINE’s ability to suggest potential gene-disease associations for further investigation. Examples of new network-level associations in breast cancer and cardiovascular diseases NERINE not only established rare variant connections between new gene networks and phenotypes but also highlighted how rare functional variants, aggregated at the gene level, could influence the phenotype within the network context. For instance, while the estrogen receptor pathway has been clinically studied for breast cancer, rare variant evidence for member genes beyond BRCA1 has been limited 46 , 47 . NERINE identified a significant LoF burden at the network level, predicting how putative LoF variants in member genes might have exerted either trait-increasing or trait-decreasing eiects on the phenotype. As expected, rare LoF variants in BRCA1 were associated with an increased risk of BRCA ( Figure 5A : left ; However, after removing the eiect of BRCA1 LoF variants, the network still showed a nominally significant association with BRCA in UKBB, suggesting albeit small but non-zero contributions of other member genes ( Supplementary Figure S17 ). Download figure Open in new tab Figure 5. Examples of NERINE’s novel findings in breast cancer and cardiovascular diseases. A. Estrogen receptor regulation (BIOCARTA CARM ER PATHWAY) shows significant LoF burden in BRCA (avg. = 0.33, Bonf. 𝑝 = 1.04 × 10 -4 ). Top: NERINE-predicted gene eiects (averaged across biobanks). Bottom: Information on gene-phenotype association from somatic and germline genetics with functional annotations. B. Intrinsic prothrombin activation and coagulation network (BIOCARTA AMI PATHWAY) shows significant LoF burden in MI (avg. = 0.18, Bonf. 𝑝 = 3.50 × 10 -5 ). Top: NERINE-predicted gene eiects (averaged across biobanks). Bottom: Information on gene-phenotype association from germline genetics with functional annotations. C. Alternative complement system network (BIOCARTA ALTERNATIVE PATHWAY) shows significant LoF burden in CAD (Left; avg. = 0.03, Bonf. 𝑝 = 1.94 × 10 -2 ), and MI (Middle; avg. = 0.30, Bonf. 𝑝 = 1.76 × 10 - 10 ). Right: Information on gene-phenotype association from germline genetics with functional annotations. Across panels, node color represents the direction of gene eiect (orange: trait-increasing; purple: trait-decreasing; intensity reflects magnitude). Information sources: GWAS associations from GWAS catalog ( https://www.ebi.ac.uk/gwas/ ); rare variant associations from Genebass 33 , SAIGE-GENE+ 19 , and disease-specific studies 47 , 53 , 54 ; functional annotations for estrogen receptor regulation pathway members from COSMIC ( https://cancer.sanger.ac.uk/cosmic ); functional annotations of network genes from SynGO ( https://www.syngoportal.org/ ). Among the estrogen receptor pathway members, 17 were classified as neither tumor suppressor genes (TSGs) nor oncogenes ( Figure 5A : right ). Within the context of this network, NERINE newly linked rare LoF variants in putative GWAS genes, ESR1 and CCND1 , to increased BRCA risk. Additionally, it shed light on the eiects of loss-of-function of genes, which had unknown oncogenic roles and no genetic evidence. LoF variants in TBP , GRIP1 , GTF2A1 , HDAC1 , MED1 , MEF2C , and PHB2 showed trait-increasing eiects, while HDAC5 , NRIP1 , and SRA1 loss-of-function had trait-decreasing eiects ( Supplementary Table T8 ). Indeed, PHB2 ( prohibitin 2 ) has recently emerged as a key biomarker and therapeutic target in BRCA because of its potential as a tumor suppressor through several possible mechanisms, including inhibition of estrogen receptor-alpha ( ERα ), interaction with BIG3 ( brefeldin A-inhibited guanine nucleotide-exchange protein 3 ), and promotion of PIG3 ( p53-inducible gene 3 ) transcription 48 – 50 . Furthermore, NERINE’s prediction of HDAC5 loss-of-function variants decreasing the risk of BRCA aligns with the fact that the inhibition of HDAC5 induces intrinsic apoptosis in human breast cancer cells, thereby exerting an anti-neoplastic eiect 51 , 52 , providing rationale for follow-up investigations for therapeutic potential. Rare variant studies in cardiovascular diseases had been underpowered to uncover novel mechanisms beyond lipid-related pathways despite sequencing of over 100,000 subjects in multiple biobanks 33 , 53 . Here, we identified a significant enrichment for rare variants in several non-lipid pathway gene networks ( Figure 4B , Supplementary Figures S14 and S15 ) whose member genes had not been identified in single-gene burden analyses 33 . One such pathway of note in MI is the intrinsic prothrombin activation and coagulation cascade (i.e., BIOCARTA AMI PATHWAY), with a significant LoF variant burden This module consists of several genes associated with collagen and vascular basement membrane integrity ( Figure 5B : left, Supplementary Table T11 ). Among the members, extracellular matrix proteins (i.e. COL4A1, COL4A2, COL4A4 ), and clotting factor ( PLG ) had previously been linked to myocardial infarction by common variant GWAS ( Figure 5B : right ) and in vivo studies in knockout mice 5 , 54 – 57 . NERINE provided more direct rare variant evidence for this gene module in MI. Interestingly, NERINE identified an enrichment of rare LoF variants in the alternative complement pathway for both CAD Bonf. p = 1.76e-10), a connection not previously established through human genetics ( Figure 5C ). Functional studies in murine models linked complement components, C3 and C5 , to inflammation, endothelial dysfunction, and atherosclerosis 58 – 60 . In the network, we observed trait-increasing eiects for C3 LoF variants in both CAD and MI and C5 LoFs in MI. Additionally, rare LoF variants in complement factors C8A , CFB , and CFD had trait-increasing eiects on both conditions ( Supplementary Tables T10 and T11 ). These directionalities can help reconcile the growing experimental evidence demonstrating that complement activation products in human atherosclerotic plaques can exhibit both protective and pro-atherogenic properties 60 . While the complement signal in cardiovascular diseases is intriguing, the observed diierences between MGBBB and UKBB results might be because MGBBB participants had a higher frequency of cardiac events, with thrombosis and inflammation, compared to those in UKBB. Bridging genetics and experimental biology with NERINE in Parkinson’s disease NERINE provides a robust framework for integrating insights from human genetics, cell systems, and model organisms simultaneously to facilitate the study of complex diseases. We illustrate this with Parkinson’s disease (PD), a complex age-related neurodegenerative disorder for which there are no disease-modifying therapies, but for which there is a growing number of genomic and biologic datasets 2 , 16 , 28 , 61 – 63 ideally suited for testing with NERINE. While GWAS have identified hundreds of PD-associated loci 64 , 65 and linkage analyses 66 – 68 have uncovered Mendelian rare variants in key genes—including SNCA (encoding αS itself), GBA1 , and LRRK2 —existing rare-variant association tests struggle to identify additional causal loci with replication across multiple independent cohorts 26 , 27 . For idiopathic PD, which accounts for ∼80-85% of cases 69 , the role of rare variants remained poorly understood. Traditional pathway-based analyses 70 have implicated broad PD mechanisms, including vesicle traiicking, G-protein–coupled receptor (GPCR) signaling, and immune pathways. However, limited resolution and statistical power, due to aggregation of signal across many genes, prevent researchers from determining individual gene contributions or how genes interact to influence disease processes. Thus, such aggregated pathway-or ontology-level signal are less informative for generating experimentally testable hypothesis in specific cellular contexts. It is possible to assemble the same canonical gene sets into molecular networks with well-defined edge geometries and apply NERINE to pinpoint specific genes and modules for targeted experimentation. We demonstrate this by applying NERINE to gene ontology (GO) biological process modules significantly enriched in PD GWAS-associated genes 64 , 65 , 71 ( Methods , Supplementary Figure S18A , Supplementary Table T12 ). Using database edge-based topologies, NERINE identified a significant burden of rare LoF variants in the peptidyl-threonine modification co-essentiality module in two independent PD case-control cohorts from UKBB and AMP-PD Supplementary Figure 18B , Supplementary Table T13 ). Within this module, NERINE suggested trait-increasing LoF burdens in MCCC1 and DYRK1A , and trait-decreasing LoF eiects in several other genes, including FYN , USP8 , and LRRK2 ( Supplementary Table T14 ; see Supplementary Note for detailed discussion). Experimentally derived networks to capture core PD pathologies As a complementary approach, we turned to biologically informed networks constructed from large-scale genetic data in cellular and model-organism studies. These networks represent experimentally validated assemblies of genes with defined functional relationships, oiering a degree of biological specificity not captured by canonical pathways or ontologies. NERINE can pinpoint gene-and module-level eiect sizes within these networks, enabling targeted follow-up experiments. Specifically, we analyzed networks comprising genes that modulate α-synuclein toxicity and mistraiicking or support dopaminergic (DA) neuron survival, representing the two pathological hallmarks of Parkinson’s disease (PD): α-synuclein aggregation in Lewy bodies and DA neuron loss 72 , 73 ( Figure 6A ; Methods ). We assessed these networks in independent PD case-control cohorts from UKBB and AMP-PD. Download figure Open in new tab Figure 6. NERINE identifies rare-variant burden in bespoke PD networks from model-system screens. A. Key pathological features of PD—Lewy body formation and dopaminergic (DA) neuron loss—targeted via (i) DA neuron essentiality and (ii) 𝛼- synuclein (𝛼 S ) proteotoxicity screens. B. Network hypotheses generation from model system screens ( Methods ). Network hypotheses were derived from (i) DA neuron essentiality genes (10 GO BP modules) and (ii) 𝛼 S -toxicity modifier genes (17 TransposeNet stems). Topologies for DA essentiality modules were generated using PPI, substantia nigra co-expression (GTEx v8), and CNS co-essentiality (DepMap v2023Q2) data. For 𝛼 S proteotoxicity network stems, TransposeNet topologies 2 were used. C. NERINE identified screen-wide Bonferroni-significant burden of rare damaging missense variants across AMP-PD (2,117 cases and 1,095 controls) and UKBB-extreme (2,237 cases and 2,553 controls) cohorts in the HMGB1-OPTN- containing autophagy regulation module (avg. = 0.03, Bonf. 𝑝 = 1.86 × 10 -3 ), with substantia nigra co-expression as the optimal topology. D. The LRRK2-SNCA- containing vesicle tra[icking and protein homeostasis stem showed screen-wide Bonferroni-significant burden of rare damaging missense variants (avg. = 0.53, Bonf. 𝑝 = 1.28 × 10 -3 ) across AMP-PD and UKBB-sporadic cohorts (AMP-PD: 2,117/1,095; UKBB: 2,237/167,188). In C and D , node color represents the direction of gene eiect (orange: trait-increasing; purple: trait-decreasing; intensity reflects magnitude). In C , edge color indicates the sign of the correlation (red: positive; blue: negative), while edge width reflects the correlation strength. In D , TransposeNet edges represent binary relationships and are colored in gray. Rare variant burden in DA neuron essentiality genes involved in autophagy regulation Using a genome-scale CRISPR screen of 19,993 genes ( Figure 6B ; top ), we identified 693 essential genes for DA neuron survival with significant enrichment in ten GO modules, covering a wide range of biological processes, including autophagy regulation, mRNA processing, apoptosis, and cellular response to DNA damage ( Supplementary Table T15 , Methods ). We generated networks from these ontological modules by grouping semantically similar GO biological process terms enriched in essentiality genes and then extracting the edge relationships of genes within each group from PPI, co-expression, and co-essentiality databases ( Methods ). NERINE identified a significant rare damaging variant burden the gene module linked to the regulation of autophagy Bonf. p-value: 1.86e-2; Figure 6C , Supplementary Figure S19 , Supplementary Tables T16 and T17 ) in sporadic PD cases compared to controls in the AMP-PD sporadic and UKBB extreme (median age of controls at recruitment >= median age of cases) cohorts ( Methods ). NERINE predicted trait-increasing eiects for damaging missense and damaging variants in key genes such as HMGB1 and OPTN in PD ( Figure 6C , Supplementary Figure S19 , Supplementary Tables T17 ). The impairment of HMGB1 function may inhibit autophagy and promote αS accumulation 74 , while the disruption of OPTN function may result in the improper clearance of damaged mitochondria (or mitophagy), leading to neurodegeneration 75 . NERINE also identified a trait-increasing burden of rare damaging variants in USP10 ( Supplementary Figure S19 , Supplementary Table T17 ) whose inactivation may disrupt αS-containing aggresome formation 76 , exacerbating toxicity. These findings readily suggest a small number of potential candidate genes and testable hypotheses supported by orthogonal human genetic evidence, justifying in-depth future experimentation. Rare variant burden in a LRRK2-and SNCA-containing αS proteotoxicity module related to vesicle tra[icking and protein homeostasis Next, we applied NERINE on an αS proteotoxicity network assembled through “yeast-to-neuron” discovery screens 2 targeting the “αS accumulation in Lewy bodies” feature of PD ( Figure 6B : bottom ). The αS network was generated using the TransposeNet methodology that connected hits from multiple genome-scale deletion and over-expression screens against ɑS proteotoxicity in yeast into an optimal topology by transposing yeast genes into the human interactome space and adding “predicted” genetic nodes as needed. TransposeNet introduced numerous human genes, without any known homologs in yeast, including SNCA (that encodes ɑS itself) and LRRK2 , two genes that are among the most definitively connected directly to PD risk in GWAS and Mendelian linkage analyses 64 – 68 . Importantly, the network, validated in human iPSC cortical and DA neurons 2 , 16 , converged with a proteome-scale proximity labeling screen for αS in neurons 63 . It spanned many relevant pathways, including vesicle traiicking, mRNA metabolism and translation, mitophagy, oxidative metabolism, calcium/NFAT signaling, Toll-like receptor signaling, and purine metabolism ( Supplementary Tables T18 ). Among the TransposeNet αS network branches ( Supplementary Figure S20 ), we observed a screen-wide significant burden of rare damaging missense variants in the LRRK2- and SNCA -containing vesicle tra[icking and protein homeostasis -related gene module associated with PD risk in AMP-PD and UKBB Figure 6D , Supplementary Figure S21 , Supplementary Tables T18 and T19 ). None of the member genes, except LRRK2 showed an exome-wide significant rare-variant burden in previous single-gene analysis 26 , 33 , and even this association could not be replicated in multiple cohorts. The network statistically supported by NERINE showed trait-increasing eiects of rare damaging missenses in several SNCA -interactors in the stem, including VDAC1 , NEDD4 , and TOR1A , which were previously linked to PD pathology only in experimental models 61 , 62 , 77 – 81 . PRL also showed a strong trait-increasing eiect, but it was not previously causally linked to PD. Beyond direct interactors of SNCA in the module, NERINE detected non-zero eiects in genes such as NDFIP1 , PBX1 , and RNF11 with PD risk, which had been linked to PD pathobiology only in functional studies 82 – 85 but not through rare variant studies. Convergence of an unbiased functional genomics screen in CRISPRi-induced synucleinopathy model with NERINE on PRL-SNCA interaction NERINE pinpointed genetic signals within the humanized TransposeNet αS proteinopathy network, which was originally derived from yeast-based screens 2 and therefore lacked the specific human cellular context for key network interactions. Brain-related disease like PD involves a plethora of cell types and peripheral systems that can influence brain function, oiering many possibilities for follow-up experimentation. To focus on neuronal mechanisms, we sought convergence between NERINE’s findings and an independent genome-scale functional screen in a CRISPR interference (CRISPRi) induced synucleinopathy cortical neuron (CiS-CN) model 28 designed for studying genetic determinants of αS toxicity ( Methods ). Briefly, this model was established in the WTC11 iPSC genetic background 28 , and enabled—i) one-step trans-diierentiation (AAVS1-Ngn2) into cortical glutamatergic neurons that are vulnerable in Lewy body diseases, including PD; ii) flexible gene knockdown via CRISPRi (dCas9-KRAB) 86 ; iii) lentivirally-driven clonal αS expression at brain-like levels. The CiS-CN model, with two SNCA -overexpression ( SNCA -OE) clones ( SNCA- high and SNCA- intermediate) and a control clone ( SNCA- endo1), thus provided a tractable but physiologically relevant system for studying αS aggregation and toxicity in cortical neurons 28 , 87 . CRISPRi was utilized to target physical and genetic interactors of ɑS in the CiS-CN model, including all genes from the TransposeNet ɑS proteinopathy network, using a custom sgRNA library ( Methods ). The full screen will be described in a forthcoming publication. Briefly, we assessed sgRNA representation at start, DIV28, and DIV42, and looked for the dropout of sgRNAs in three comparisons: (1) SNCA -high vs. SNCA -endo1 at DIV42, (2) DIV42 vs. DIV28 in SNCA -endo1, and (3) DIV42 vs. DIV28 in SNCA -high ( Figures 7A,B ). Genes showing significant sgRNA dropout in MAGeCK-iNC analysis (FDR < 0.1) were classified as αS toxicity enhancers ( Methods ). Download figure Open in new tab Figure 7. Unbiased functional genomics screens converge on an intraneuronal SNCA-PRL stress response. A. Top: Timeline of CiS-CN diierentiation. Bottom: iPSCs were transduced to overexpress either mCherry (control) or SNCA-mCherry, generating the SNCA-endo1 and SNCA -high lines, respectively. At DIV0, neurons were transduced with the sgRNA library. Neuronal samples were harvested and sequenced at DIV3, DIV28, and DIV42 using next-generation sequencing. Comparison 1 shows the comparison of the sgRNA frequencies between DIV42 SNCA -high neurons and DIV42 SNCA- endo1 neurons. Comparison 2 compares the sgRNA frequencies in SNCA- endo1 neurons between DIV42 and DIV28. Comparison 3 compares the sgRNA frequencies in SNCA -high neurons between DIV42 and DIV28. B. Left: PRL sgRNA-containing neurons dropped out in Comparison 1, indicating PRL knockdown was toxic to SNCA -high neurons compared to SNCA- endo1 neurons. Middle: No significant drop-out in PRL sgRNA-containing neurons was observed in Comparison 2, indicating PRL knockdown was non-toxic to DIV42 SNCA -endo1 neurons compared to DIV28 SNCA- endo1 neurons. Right: PRL sgRNA-containing neurons dropped out in Comparison 3, indicating PRL knockdown was toxic to DIV42 SNCA -high neurons compared to DIV28 SNCA -high neurons. C. Left: Immunostaining data show prolactin was upregulated in DIV7 SNCA -high neurons (n=3, one-way ANOVA). Right: Representative images acquired with a Nikon Eclipse Ti microscope, with solar power set to 10% and an exposure time of 50ms for prolactin . Scale bar: 20um. D. PRL knockdown was validated with two diierent PRL sgRNAs from the screen library (n=3, one-way ANOVA). DIV0 SNCA -high neurons were transduced with either control sgRNA or PRL sgRNA at MOI=5. At DIV7, neurons were stained with prolactin antibody and Hoechst. Images were captured with a Nikon Eclipse Ti microscope. Prolactin intensity per neuron is reported here. E. Immunostaining data show prolactin -upregulation was diminished in DIV28 SNCA- high neurons (n=3, one-way ANOVA). Images were captured with a Nikon Eclipse Ti microscope, with solar power set to 30% and a 30ms exposure time for prolactin . F. Immunostaining data show prolactin -upregulation was diminished in DIV28 SNCA -high neurons treated with PFF (n=3, one-way ANOVA). Images were captured with a Nikon Eclipse Ti microscope, with solar power set to 30% and a 30ms exposure time for prolactin . G. Left: Illustration of the PFF-induced mouse model. Amygdala tissue, micro-dissected from mice and injected with 𝛼 S PFFs, was subjected to transcriptomic analysis using the NanoString Neuropathology panel ( Methods ). Right: In comparison to PBS-injected control animals, which had no pS129 positive inclusions across all brain regions, PFF-injected mice showed a 16.6-fold downregulation of Prl mRNA expression in the amygdala region ipsilateral to the site of injection (n=5 per group; 3 males/2 females). H. CellRox assay shows that, with menadione treatment, oxidative stress in DIV28 SNCA -high neurons was significantly higher than in others of the same age (n=3, two-way ANOVA). I. Top: Timeline of exogenous prolactin assay. Bottom: CellRox assay shows that exogenous prolactin significantly decreased menadione-triggered oxidative stress in DIV7 SNCA -OE neurons (n=3, two-way ANOVA). In these comparisons, PRL emerged as a particularly intriguing finding. NERINE had already pinpointed PRL within the LRRK2– and SNCA-containing branch of the same TransposeNet network (see above; Figure 6D ), where damaging missense variants showed a strong trait-increasing eiect. In the CiS-CN CRISPRi screen, sgRNAs targeting PRL significantly dropped out in SNCA -high versus SNCA -endo at DIV42 (Comparison 1) and from DIV28 to DIV42 in SNCA -high clones (Comparison 3), aligning with the NERINE result. Thus, in light of these convergent results, PRL , which previously lacked direct evidence of a causal role in PD, was selected as a key candidate for further experimental validation. A novel intraneuronal SNCA-PRL stress response identified in human neurons While Prl is predominantly expressed in neuroendocrine tissue, its unexpected identification in our neuronal functional genomics screen suggests a highly unexpected cell autonomous role in neurodegeneration. Previous studies, though conflicting, have suggested PRL expression in rodent neurons outside the pituitary under stress conditions 88 , with speculation, albeit based on a few studies, that its expression may only be detected under conditions of neuronal stress. There is also evidence that its protein levels may be post-translationally regulated despite low mRNA expression 89 . Overall, the strong literature consensus is that any eiects of Prl are likely to be through exogenous release of Prl from the pituitary gland. To investigate the possibility that Prl may be part of an intraneuronal stress response directly related to αS, we performed immunostaining in neurons of diierent age. Prl expression was significantly increased in SNCA-overexpression neurons, with ∼7-fold-higher expression in SNCA-high , and 2.5-fold-higher in SNCA-intermediate. than in SNCA- endo1 neurons at DIV7 ( Figure 7C ). This elevation was abrogated, as expected, by our sgRNAs directed to PRL ( Figure 7D ). Importantly, Prl levels were reduced over time, such that by DIV28, Prl levels in the SNCA- high lines, were only ∼1.5-fold-higher than in SNCA- endo1 neurons ( Figure 7E ). These data collectively suggest a potentially protective role of αS-induced Prl that diminishes over time. We next tested whether inducing αS aggregation, thereby reducing soluble αS levels 87 , in cellular and animal models via pre-formed αS fibrils (PFFs) 29 , 90 led to reduced Prl levels. Our CiS-CN neurons challenged with seven (7) days of exposure to PFFs showed a reduction of Prl levels as indicated by immunofluorescence ( Figure 7F ). We further considered whether this was transcriptional or post-transcriptional. mRNA expression levels of PRL were low in our models, possibly related to the immaturity of the iPSC-derived neurons. We thus turned to a well-established mouse PFF model 29 . We unilaterally injected PBS versus 5 µg of PFFs (2.5 µL volume) into the dorsal striatum of mice ( Figure 7G : left , Methods ). Mice were aged for 30 days post-injection. By this stage, as has been previously described 91 , the αS aggregation pathology “spreads” distally and reaches the cortex and the amygdala, two brain regions highly susceptible to αS pathologies in later stage PD. We harvested mRNA from the amygdala and assessed gene expression with the Nanostring Neuropathology panel. In comparison to PBS-injected control animals which showed no pS129 positive inclusions across all brain regions (not shown here), PFF-injected mice showed a 16.6-fold downregulation of Prl mRNA expression in the amygdala region ipsilateral to the site of injection (n=5 per group; 3 males/2 females; Figure 7G : right ), consistent and indeed stronger than our findings in the shorter-term human CiS neuronal model. At least in the intact rodent model, our data indicate that a component of this eiect is transcriptional. Our data indicate that in the context of aging and fibrillar αS pathologies, Prl levels drop intraneuronally, and potentially could sensitize the neurons to αS-induced cytotoxicity. To determine whether Prl directly protects against αS toxicity, we assessed its neuroprotective eiects against oxidative stress in our CiS models. By DIV28, SNCA- high neurons exhibited heightened sensitivity to menadione-induced oxidative stress 92 , measured with the fluorogenic probe CellRox (Thermo) ( Figure 7H ). In SNCA- high and SNCA- endo1 models, we pre-conditioned DIV6 CiS neurons with prolactin for 24hrs before treatment with menadione ( Figure 7I : top ). Importantly, exogenous prolactin treatment significantly decreased oxidative stress in CiS neurons ( Figure 7I : bottom ), strongly suggesting a neuroprotective eiect against αS-induced toxicity. Our data collectively indicate that aged neurons in which αS aggregates leads to reduced Prl levels and increased vulnerability to oxidative stress. Taken together, the convergence of our forward genetics iPSC screen with NERINE uncovered an unexpected αS- Prl intraneuronal stress response, suggesting that Prl may play a crucial role in neuronal resilience to αS pathology in PD. These data underscore the utility of NERINE in uncovering novel biological insights into complex disease processes. Discussion NERINE is a powerful statistical test designed to bridge the gap between human genetics and experimental biology, adding a crucial mechanistic understanding to rare variant burden analysis. Unlike traditional rare variant tests, which focus on isolated genes or regions, NERINE contextualizes rare variant eiects within biological networks, thereby enhancing our ability to interpret genetic associations. By incorporating network topologies, we provide a more structured framework for understanding how rare variants contribute to disease mechanisms. One of NERINE’s key advantages is its ability to select between competing network topologies for the same gene set—an ability no other method currently oiers. By utilizing edge geometry, NERINE gains a statistical power boost, a feature demonstrated through extensive simulations and real-world data analyses, even in cases where network structures are poorly defined. Despite its strengths, NERINE has some limitations, particularly in cases when the networks are too large, or sample sizes are exceedingly small (see Methods ). However, its ability to integrate human genetic data with functional insights from model system experiments makes it a valuable tool for hypothesis testing in disease genetics. To showcase NERINE’s utility, we applied it in two distinct contexts: first, to test network hypotheses from publicly available databases across four prevalent diseases—BRCA, T2D, CAD, and MI. NERINE provided novel rare variant signals for biologically plausible disease mechanisms, such as collagens and inflammatory response genes in cardiovascular diseases, adipogenesis pathway in T2D, and the network of estrogen receptor regulation in BRCA. The implication of collagens, key structural proteins in the extracellular matrix, in MI is consistent not only with their primary role in maintaining the stability of atherosclerotic plaques 93 but also with their indirect role in coagulation via platelet eiects and interactions with clotting factors 94 . Furthermore, the significant rare variant burden in networks linking inflammatory response genes to the risk of both CAD and MI supports the potential of anti-inflammatory therapies as an alternative to LDL-lowering drugs for treating cardiovascular diseases 95 . Second, we applied NERINE to bespoke gene networks derived from experimental screens in Parkinson’s disease (PD), a complex disorder where traditional rare-variant analyses have struggled due to limited sample sizes. Here, NERINE revealed a significant rare variant burden in gene modules related to autophagy regulation, vesicle traiicking, and protein homeostasis—signals largely undetected by single-gene tests. Our findings further highlighted two major contributions: first, we provided rare variant evidence supporting hits from previous functional studies in model systems. For example, several member genes of the LRRK2-and SNCA -containing vesicle tra[icking and protein homeostasis subnetwork (e.g., NEDD4 , NDFIP1 , VDAC1, PBX1, TOR1A , and RNF11 ) which have been previously implicated in PD only in functional studies 61 , 62 , 77 – 85 , 96 – 98 , are now corroborated with human genetics evidence by NERINE. Second, NERINE helped resolve conflicting hypotheses by providing genetic support for specific disease mechanisms. For instance, there is biological data suggesting haploinsuiiciency of DYRK1A —a member of the PD GWAS gene module involved in peptidyl-threonine modification identified by NERINE—reduces its kinase activity and exacerbates DA neuron degeneration in mice 99 . There is also data arguing the opposite through phosphorylation of ɑS 100 . By providing genetic evidence in favor of the latter direction, we oiered some resolution of the matter. Although NERINE does not provide gene-level p-values within a network, it yields informative eiect sizes and directions for member genes that often align with known biology, as demonstrated in our positive control experiments with lipid phenotypes in the UK Biobank. Importantly, NERINE captures both trait-increasing and trait-decreasing rare variant burdens at the gene level, derived from its joint modeling of network interactions and observed allele counts at each locus. While the latter may reflect gain-of-function eiects requiring experimental validation, they can be of considerable therapeutic significance. A well-established precedent is PCSK9 , where rare LoF variants are associated with reduced LDL cholesterol levels; this discovery directly enabled the development of PCSK9 inhibitors as an eiective therapeutic strategy for hypercholesterolemia with significantly improved cardiac outcomes 44 , 45 . In this context, the trait-decreasing eiects predicted by NERINE within disease-relevant networks may highlight genes of potential therapeutic interest, supporting future experimentation. NERINE’s integrative framework for unifying insights from model system screens with human genetics complements traditional pathway-based investigations in PD 70 , 101 . Traditional investigations rely on pathway-based polygenic risk scores from common variants or on rare variant aggregation tests, such as SKAT-O, applied at the pathway level 70 , 101 to implicate broad mechanisms, including endolysosomal traiicking, GPCR signaling, neuronal transmission, and immune response 70 , 101 . While informative, such approaches generally lack statistical power, and resolution to pinpoint contributions of individual genes and their specific interactions. A network-based approach like NERINE, which incorporates experimentally defined gene relationships, provides additional biological specificity, improves power to detect associations, and helps guide mechanistic hypothesis generation for follow-up studies. Among NERINE’s findings in PD, the most intriguing one is related to PRL , supported by converging evidence from NERINE and our functional CRISPRi cellular screens. PRL encodes the protein Prl , a hormone secreted by the pituitary gland with no known connection to ɑS. The original TransposeNet network 2 unexpectedly introduced Prl to a small network of genes that also included key known PD genes SNCA (encoding ɑS itself) and LRRK2 (encoding a protein with increased enzymatic kinase activity that has been implicated in PD). Mechanistically, a connection of Prl to ɑS had been elusive until this study. This is because, although not unprecedented, the presence of Prl in neurons outside the pituitary has been highly speculative 88 , 102 . While Prl showed some neuroprotective eiects against oxidative stress 88 , 103 – 105 , this eiect had been attributed exclusively to an exogenous endocrine eiect through pituitary secretion. It was thus remarkable for Prl loss-of-function to simultaneously be implicated by both NERINE from human genetic data and in a genome-scale screen for enhancers of ɑS proteotoxicity: predicted deleterious missense mutations were tied to PD risk by NERINE, and Prl knockdown enhanced ɑS toxicity within neurons. Our data suggest that the eiect results from an early, possibly post-translational, intraneuronal mechanism in which Prl is specifically elevated in response to ɑS overexpression. This response diminishes with time and with ɑS aggregation, the latter eiect being at least partly transcriptional in more chronic treatment of mice with fibrillar ɑS. This decrease is, in turn, associated with the sensitization to PRL knockdown and exogenous stress, demonstrated in this study with the oxidative stressor menadione. While we could not directly assess pituitary function in individuals carrying PRL variants in our study cohorts, literature evidence provides additional context. A recent proteomic study in the Harvard Biomarkers Study cohort found that Prl levels in the cerebrospinal fluid—likely reflecting non-pituitary sources—was the top feature in a classifier model distinguishing PD patients from controls, even after accounting for levodopa eiects 106 . Thus, these converging findings from NERINE, our CRISPRi screen, mouse models, and literature strengthen the plausibility of a role for an unexpected role of Prl in PD that originates not in the pituitary but within neurons subjected to ɑS stress. Further work will now be necessary to clarify its mechanisms and to evaluate potential pleiotropic eiects. In sum, NERINE bridges the gap between human genetics and experimental biology, providing a robust framework for attributing putative disease-causing factors to experimentally derived molecular networks. Such a method is especially significant for complex diseases like PD, where the limited availability of well-phenotyped genomic data hinders traditional approaches. As large-scale molecular interaction networks become increasingly common, we anticipate that research programs where human genetics and experimental biology mutually reinforce each other—such as those enabled by NERINE—will play a growing role in identifying disease-relevant signals, resolving conflicting evidence, and uncovering druggable targets. Methods Overview of the NERINE methodology NERINE models the “variant-gene-network” hierarchy as follows: it encodes gene-gene relationships in a network of m genes by a positive semidefinite matrix, Σ. We assume that phenotypic eiects of genes within a network, represented as vector are drawn from a multivariate skew-normal distribution Here, 𝜃 is a parameter reflecting the cumulative eiect of the gene network on a phenotype and is the object of inference ( Figure 1 ). The skewness parameter 𝜈 = 𝑓(𝑁 case , 𝑁 control ) adjusts for case-control imbalance. Here, N case and N control represent the case-and control-group sizes. For balanced cohorts (i.e., 𝜈 = 0), the model reduces to where marginals are normally distributed with zero mean. Under this model, an edge between two genes in the network implies that they have either correlated non-zero eiect sizes or correlated chances of having no phenotypic eiects. The network-eiect, 𝜃 on a dichotomous phenotype is estimated using the maximum likelihood estimation (MLE) framework, where the likelihood is modeled as an integral over two components: (i) the product of conditional probabilities of observed case mutation counts given total mutation counts in network-genes, and (ii) the probability of gene eiects given the network parameters: We approximate this integral as a weighted sum over multivariate quadrature points from the domain of integration. The weight of each multivariate quadrature point is given by the product of the corresponding univariate weights. To make the computation of the integrand tractable, NERINE makes several design choices and employs a lookup table approach with pruning ( Supplementary Note ). NERINE models rare variant counts in genes for cases and controls as two independent Poisson random variables as follows, Here, 𝜆 case i and 𝜆 control i denote the rate parameters for case-and control-allele count distributions in gene 𝑖, and correspond to population allele frequencies renormalized between cases and controls by transformed network-gene eiect, 𝛼 i . This implies that the conditional probability of observed rare variant count in each gene in cases, given the total rare variant count in the cohort, follows a Binomial distribution ( Supplementary Note ). , where 𝑝 = 𝜙(𝛼 i ) ( Supplementary Note ). Since each 𝛼 i represents unbounded univariate skew normal distributions, they cannot be directly used as Binomial success probabilities. Hence, we apply a custom variable transformation to map 𝛼 i s to Beta distributions so that they fall on the interval [0, 1] ( Supplementary Note ). This transformation ensures that the mean gene eiect is centered around 𝑁 case /(𝑁 case + 𝑁 control ), and the shape of the transformed distribution adapts with 𝜃: small 𝜃s (close to 0) concentrate density near the mean, while large 𝜃s shift density toward the extremes. Case-control skew is incorporated using the shape parameters of the Beta distribution. The transformed gene-eiects within the network are denoted ( Supplementary Note ). Thus, NERINE’s approximate likelihood takes the form: The conditional probability term is calculated using the probability density function of a standard Binomial distribution, which makes the likelihood computation fast and tractable. To calculate the probability of network-gene eiects for a given network topology (Σ) and network eiect (𝜃), we use a lookup table, where highly improbable are pruned ( Supplementary Note ). NERINE performs nested hypothesis testing; the null hypothesis being 𝐻 0 : 𝜃=0 (i.e., the gene network does not aiect the trait and expected ratio of variant counts between cases and controls is proportional to the ratio of sample sizes in each gene) and the alternative set of hypotheses being 𝐻 z : 𝜃 = 𝜃 z >0 (i.e., the gene network has an overall eiect on the trait and the ratio of variant counts between cases and controls may deviate from the null expectation in relevant member genes). The test statistic of NERINE is the log-likelihood ratio (LLR): We denote the maximum-likelihood estimate of network eiect with Since 𝜃 lies on the boundary of the parameter space, the test statistic asymptotically follows the distribution of a weighted mixture of a point mass at zero and a chi-square distribution with a degree of freedom of one (𝑑𝑜𝑓 = 1) 30 , 31 . NERINE draws its asymptotic p-values from this distribution. For significant networks, NERINE calculates the maximum likelihood gene-specific eiects under the estimated as follows: The search space for possible network-gene eiects is determined by the estimated the network structure (Σ), and the lookup table entries ( Supplementary Note ). NERINE has several limitations. Currently the test is designed to accommodate dichotomous traits only. Continuous traits need to be dichotomized before NERINE can be run on them. Covariates correction is not directly included in the test. For continuous traits, it can be done as a preprocessing step using a regression-based approach and then the trait can be dichotomized. For binary traits, NERINE currently performs ancestry-stratified analysis and combines the p-values using Fisher’s combined test post-hoc. For large networks with > 50 genes, NERINE’s test statistic falls outside the asymptotic regime, hence we recommend using NERINE with networks up to 50 genes. Simulations under the null model To evaluate the performance of NERINE under the null model (𝜃=0), we performed extensive simulations with well-studied biological pathways, namely, NOTCH signaling (m=6), WNT signaling (m=24), protein export (m=24), and EGFR signaling (m=50), from the canonical pathways database ( Supplementary Figures S1 and S2 ) and diierent simulated network architectures—(i) clique: complete graph with all nodes connected to each other; (ii) path: each node connected to two other nodes except the first and the last nodes; (iii) random: randomly generated scale-free graph of 𝑚 nodes; and (iv) isolated genes: nodes not connected with each other ( Supplementary Figure S3 ). For the canonical pathways, gene lists were extracted from MSigDB (v7.3) and high confidence physical and genetic interactions from protein-protein interaction (PPI) databases were used as network edges between pathway genes. We simulated three diierent scenarios – (i) equal sized case/control groups (𝑁 case = 1,000; 𝑁 control = 1,000) (ii) case group is larger (𝑁 case = 3,000; 𝑁 control = 1,000), and (iii) control group is larger (𝑁 case = 1,000; 𝑁 control = 3,000). Under diierent network architectures and case-control skews, we simulated allele counts in cases and controls using independent Binomial distributions under the null model. For each gene, we assumed the presence of up to five qualifying loci each with minor allele frequency (MAF) of 0.001. The binomial probabilities for case-and control-groups are adjusted according to the group sizes and MAF of variants assuming no gene-specific eiects under the null model (𝜃 = 0). For each scenario, we performed 1,000 iterations to generate the QQ-plots. We used the pchibarsq function from the emdbook (v1.3.13) in R (v4.3.2) to calculate the p-values from the mixture of chi-square distribution with one degree of freedom (𝑑𝑜𝑓 = 1) and the delta function at zero (0). We calculated 95% bootstrap confidence intervals around NERINE’s test-statistic for visualization. For evaluating the null behavior of NERINE’s test statistic in simulated networks of diierent sizes (𝑚 = 5, 10, and 25 genes) and diierent topological architectures (i.e., clique, path, random, and isolated nodes), we used equal sized case-and control-groups ( Supplementary Figure S3 ). For each scenario, we performed 1,000 iterations to generate the QQ-plots. The allele counts in cases and controls were generated from independent Binomial distributions following the same procedure as above. Performance benchmarking with simulated data We evaluated the performance of NERINE under the alternative hypothesis (𝜃>0) in two sets of simulations—(i) when genes have only trait-increasing eiects, and (ii) when genes have both trait-increasing and trait-decreasing eiects using the same database pathway network topologies used for the null simulations ( Figure 2 , Supplementary Figure S4 ). For each scenario, we simulated diierent noise profiles, i.e., varying proportions of genes within the network with eiects on the trait given the network topology. This mimics situations from having a very noisy network (i.e., ∼10-30% genes with an eiect on the trait) to a highly relevant network (i.e., ∼70-90% genes with an eiect on the trait). We simulated allele counts from cases and controls using independent Binomial distributions under the alternative model with 𝜃 = 0.2. For each gene, we assumed the presence of up to five (5) qualifying loci per gene with minor allele frequency (MAF) of 0.001. The Binomial probabilities for case-and control-groups are adjusted according to the group sizes and MAF of variants assuming possible gene-specific eiect configurations under the alternative model (i.e., 𝛼⃗ given the network topology (Σ) and network eiect, 𝜃 = 0.2). Using this setup, we simulated a cohort of 2,000 cases and 2,000 controls. Currently, there are no existing rare variant association tests that take gene network topology into account Thus, we compared the performance of NERINE with existing gene-level rare variant association tests adapted to the pathway level, namely, CMC-Fisher test 107 , Fisher minimum p-value test 108 , Fisher combined test 108 , SKAT-O 18 , and pathway-based rare variant trend test (RVTT) 15 , 16 . For each noise profile, we performed 250 iterations, resulting in 1,000 iterations per network. Empirical power of each method was measured as the positive predictive value (PPV) across iterations using diierent p-value cutois (𝑐): 1e-2, 5e-3, 1e-3, 5e-4, 1e-4, 5e-5, and 1e-5. Here, Additionally, we conducted performance benchmarking for NERINE on diierent simulated network architectures for 25 genes ( Supplementary Figure S5 ). As before, we performed two sets of simulations—(i) when genes have only trait-increasing eiects, and (ii) when genes have both trait-increasing and trait-decreasing eiects for four network topologies—clique, path, random graph, and isolated nodes. We simulated allele counts in 1,000 cases and 1,000 controls using independent binomial distributions under the alternative model with a diierent network eiect, 𝜃 = 0.5. We simulated various network-noise profiles ranging from 0-90% following the same procedure as above. Empirical power of each method was calculated as PPV across 250 iterations per noise profile per network topology, resulting in 1,250 iterations per network. Since RVTT, by design, assumes that all qualifying rare variants in a pathway have the same direction of eiects, we compared RVTT’s performance with NERINE in simulations with genes having only trait-increasing eiects. Moreover, RVTT computes a permutation-based p-value, and for 10,000 iterations its p-values cannot be less than 1e-4. Hence, we compared RVTT’s performance only at cutoi values ≥ 1e-4. Furthermore, we evaluated NERINE’s performance with true versus randomized network architectures in simulations with the same four well-studied pathways ( Supplementary Figure S10 ). We constructed the “true” network topologies for these pathways using the high-confidence protein–protein and genetic interactions from PPI databases as described above. For each gene set, we also generated 100 randomized networks by randomly introducing edges among the member genes. A binary trait was simulated in 2,000 cases and 2,000 controls with varying trait-increasing and trait-decreasing eiects under the alternative model with a network eiect, 𝜃 = 0.1. We simulated diierent noise profiles (i.e., 10-30%, 30-50%, 50-70%, and 70-90%) and performed 250 iterations per noise profile. We measured NERINE’s PPV at diierent p-value thresholds (i.e., 1e-2, 5e-3, 1e-3, 5e-4, 1e-4, 5e-5, 1e-5). Pathway database construction We created a pathway database with all canonical pathways of five to fifty genes from the BIOCARTA database along with all lipid-, DNA replication-, DNA damage repair-, and cell cycle-related pathways from the REACTOME, KEGG, PID, and Wiki pathways databases. The lists of member genes for these pathways were extracted from the Molecular Signatures Database (MSigDB v7.3; https://www.gsea-msigdb.org/gsea/msigdb ). The database contained 306 pathways with a median pathway length of 25 genes ( Supplementary Table T20 ). Due to the pleiotropy of genes, many biological pathways tend to overlap significantly with each other. Thus, we determined the eiective number of independent hypotheses (𝑡 eff ) in our pathway database of 𝑡 pathways by adapting Nyholt’s approach 109 . First, we calculated the pairwise Jaccard similarity of pathways, and encode it in a matrix, 𝑀 t×t , which is symmetric but not positive semi-definite. We then converted it to its nearest positive definite matrix using the nearPD function 110 from the Matrix package (v1.6-5) in R (v4.3.2). Next, we determined the eigenvalues of this approximate covariance matrix. The eiective number of hypotheses/pathways was computed using the following formula 109 : For our pathway database, we found the eiective number of independent hypotheses, 𝑡 eff to be 300, which was used for Bonferroni correction to determine database-wide significance. Gene-network topology extraction NERINE treats the gene-gene network as an input and can flexibly handle any symmetric pairwise relationship matrix that is positive-semidefinite. For a screen that provided the gene network topology, we used that network as is. For example, for the 𝛼-synuclein proteotoxicity screen in PD, we directly applied NERINE on the published TransposeNet humanized 𝛼-synuclein-modifier network 2 stems. In absence of the true network topology for a particular gene set, we adopted the following approach to construct one. For canonical database pathway gene sets, we constructed network topologies by extracting physical and genetic interactions from several sources: (i) high confidence physical interactions (weight >= 0.7) from STRING v11.5 111 , (ii) InWeb inBio Map database 112 , (iii) HuRI 6 , (iv) genetic interactions from the Megchelenbrink et al. study 113 , and (v) humanized 𝛼-synuclein-, 𝛽-amyloid-, and TDP-43-modifier networks 2 . Using a heuristic approach, we first generated adjacency matrices for the genes, where edges were represented as binary values (1 = presence of an edge; 0 = absence of an edge), corresponding to the non-zero oi-diagonal entries of the network. The diagonal was set to two (2) to indicate an equal prior on each gene. Because these modified adjacency matrices were not always positive semidefinite, we applied the nearPD function 110 from the Matrix package (v1.6-5) in R (v4.3.2) to compute the nearest positive-definite matrix. This matrix was input to NERINE as an approximate covariance matrix, Σ. We also constructed co-expression networks in relevant tissue types for diierent phenotypes using the bulk tissue expression data from the Genotype Tissue Expression database 114 (GTEx v8). We computed a real-valued gene-gene co-expression networks in a specific tissue where the edges between two genes represented the Pearson correlation of their expression profiles in that tissue types. For lipid-related phenotypes, we constructed co-expression networks using the bulk expression data from liver tissue. For Parkinson’s disease (PD), we used bulk expression data from the substantia nigra region of the mid-brain. To construct co-essentiality networks in relevant cell lines for diierent phenotypes, we used the gene dependency data from CRISPR knockout screens from project Achilles, as well as genomic characterization data from the Cancer Cell Lines Encyclopedia (CCLE) project from the DepMap portal 115 , 116 (release: 2023Q2). We computed a gene-gene co-essentiality networks in relevant cell lines where the edges between two genes represent the Pearson correlation of their dependency profiles in those cell lines. While for lipid-related phenotypes, we constructed co-essentiality networks using data from liver cell lines, for PD, we used data from cell lines pertaining to the central nervous system (CNS). The cell lines used in this study are listed in the Supplementary Table T21 . Performance benchmark on lipid phenotypes in the UK biobank (UKBB) Using jointly genotyped variant calls from the whole exomes of 469,589 individuals in the UK biobank (UKBB), we performed positive control experiments with NERINE in two lipid related phenotypes: direct LDL cholesterol (LDL-C; data field: 30780) and HDL cholesterol (HDL-C; data field: 30760) ( Supplementary Figures S6-S9 ). We created two dichotomous phenotypes: (i) high LDL vs. low LDL and (ii) low HDL vs. high HDL, by selecting individuals belonging to the top and bottom quartiles of the distributions for LDL-C and HDL-C measurements. European ancestry groups had the largest sample sizes: LDL-C (30,007 cases and 28,673 controls), and HDL-C (26,800 cases and 27,178 controls). NERINE was competitively applied on these two phenotypes across networks from our canonical pathways database. We extracted the edge relationships of genes from high-confidence physical and genetic interactions from protein interaction databases for this exercise. This analysis served as an ideal positive control because the genetic determinants of these phenotypes are well-annotated. Variants with a minor allele frequency (MAF) < 0.001 were considered rare for the test to not be influenced by artificial signal from common variant space propagated through linkage disequilibrium (LD). We stratified variants into six functional categories: LoF (i.e., frameshifts, insertions, deletions, and splice variants), damaging missense (i.e., missenses predicted to be damaging by in-silico tools), damaging (i.e., LoF and damaging missenses), missense, neutral (i.e., missenses predicted to be benign by in-silico tools), and synonymous. We used neutral and synonymous variants in a pathway as control to safeguard against technical biases and LD leakage. Bonferroni correction, accommodating the presence of correlated hypotheses in the database 109 , was used to control for Type I error. To ensure our results for LDL-C phenotype were not driven by LDLR and PCSK9 , which have large individual eiect sizes on LDL-C levels, we performed sensitivity analysis by removing LDLR and PCSK9 from the significant networks for the LDL phenotype. Even without LDLR and PCSK9 , the module of core LDL-related genes, along with the LDL clearance and chylomicron clearance pathways remained significant for the LDL phenotype after Bonferroni correction ( Supplementary Figure S22 ). For many disease phenotypes, the large sample sizes available for the UKBB lipid phenotypes, are simply not attainable. Thus, we performed a down-sampling experiment to evaluate the consistency of NERINE’s performance across diierent sample sizes and the robustness of NERINE’s performance with small sample sizes. We downsampled the high LDL vs. low LDL cohort at diierent case-control ratios (1/3, 1/10, and 1/60) and competitively applied NERINE across the pathway database, demonstrating NERINE’s eiectiveness even in a cohort with as few as 500 cases and 500 controls ( Supplementary Figure S23 ). Since rare variants are more susceptible to subtle eiects of population stratification than common variants, we performed stratified analysis of the LDL-C phenotype in all five major ancestry groups (i.e., European, American, African American, South Asian, and East Asian) and meta-analyzed the results using Fisher’s combined test ( Supplementary Figure S9 , Supplementary Table T2 ). Bonferroni correction was applied on the meta p-values. We further evaluated NERINE on real versus randomized network topologies for binarized LDL-C and HDL-C phenotypes ( Supplementary Figure S11 ). For each phenotype, we selected database-wide significant pathways with rare damaging variant burden and used the most significant topologies for each pathway shown in Supplementary Table T3 as “real” network topologies. We also generated 100 randomized networks per pathway by introducing random binary edges among member genes. We applied NERINE to both real and randomized networks and assessed its performance by comparing LLRs and p-values; higher LLRs and lower p-values indicated superior performance. Study cohorts for common disease phenotypes We used whole exome sequencing (WES) data from two population-scale biobanks, namely, the UK Biobank (UKBB) and the Mass General Brigham Biobank (MGBBB) and whole genome sequencing (WGS) data from the AMP-PD consortium for investigating network-level rare variant associations in complex diseases. Our analysis focused on four representative, high-prevalence complex diseases of public health relevance: breast cancer (BRCA), type II diabetes mellitus (T2D), coronary artery disease (CAD), and early-onset myocardial infarction (MI). These phenotypes were chosen from the UK Biobank because large sample sizes were available, and replication cohorts existed in the MGB Biobank. They served to showcase NERINE’s applicability in comprehensive database searches for rare variant network burden. Cohorts for BRCA, T2D, CAD, and MI were selected primarily based on the summary diagnoses recorded in the data field 41270 as ICD-10 codes. For BRCA, the case group consisted of unrelated females of European ancestry with ICD-10 code C50 and the control group consisted of unrelated European females of age 60 or above with no history of neoplasms (ICD-10 codes: C00-C97 and D00-D48). The resulting cohort had 10,648 cases and 91,886 controls. For T2D, we included unrelated European individuals with ICD-10 code E11 in the case group and unrelated European individuals with no endocrine, nutritional and metabolic diseases (ICD10 codes: E00-E90) in the control group. The resulting cohort had 22,502 cases and 68,370 controls. We created an age-stratified case-control cohort for the CAD phenotype where cases consisted of unrelated European individuals of age 65 with no diseases of the circulatory system (ICD-10 codes: I00-I99). This left us with 4,561 cases and 12,321 controls. Finally, for the MI phenotype, our cohort consisted of 2,521 cases and 5,012 controls. Cases included unrelated European individuals with ICD-10 code I 21. Only males with age <=55 and females with age = 69 who have no history of any disease of the circulatory system (ICD-10 codes: I00-I99). Parkinson’s disease (PD) was chosen as a case study to demonstrate NERINE’s ability to bridge human genetics with model-system screens. We had access to two genome-scale screens, targeting the main pathological features of the disease and providing experimental gene networks for testing. We also had our in-house iPSC-derived cortical neuron model (CiS-CN) for functional validation. We focused on sporadic disease cases as they constitute∼85% of the PD population. For sporadic PD phenotype, we created a discovery cohort from UKBB individuals by including unrelated European individuals with ICD-10 code G20 and no family history of PD as cases. The control group consisted of unrelated European individuals with no history of PD and other nervous system diseases (ICD-10 codes: G00-G99). In the original UKBB cohort, we observed notable diierences in the incidence of type 2 diabetes (T2D) between PD cases (17.5%) and controls (7.1%), which could bias downstream analyses; therefore, we excluded all individuals with a history of T2D from both groups. We termed this cohort the “UKBB-Sporadic” cohort with 2,237 cases and 167,188 controls. Controls were on average younger than cases at recruitment in this cohort ( Supplementary Table T22 ), which could introduce bias when investigating signatures of dopamine (DA) neuron degeneration. Thus, for the DA neuron essentiality screen analysis, we employed a super-control design by selecting older controls with a higher median age at recruitment than the cases ( Supplementary Table T22 ), which are more likely to capture naturally occurring signatures of neurodegeneration and less likely to have sporadic PD later in life. We termed this cohort with super controls as UKBB-Extreme (Ncase = 2,237 and Ncontrol = 2,553). For the database-wide investigations of BRCA, T2D, CAD, and MI, we created replication cohorts using the jointly genotyped WES data from 53,343 individuals in MGBBB, a biorepository of consented patient samples at Mass General Brigham (parent organization of Massachusetts General Hospital and Brigham and Women’s Hospital). Same inclusion/exclusion criteria were used for each phenotype to ensure consistency between the biobanks. The resulting cohort sizes are as follows: BRCA (Ncase = 1,113; Ncontrol = 2,459), T2D (Ncase = 747; Ncontrol = 2,188), CAD (Ncase = 902; Ncontrol =1,488), and MI (Ncase = 326; Ncontrol = 2,068). For the BRCA cohort, we used an MAF cutoi of 0.001 to select rare variants. For the other three phenotypes, variants with MAF <0.03 were considered to be “rare”. Since the cohort sizes in MGBBB were smaller compared to UKBB for T2D, CAD, and MI, we did this adjustment to the MAF cutoi and performed tests in the synonymous and neutral missense categories for each pathway to make sure that there was no LD leakage. For PD, we created a replication cohort using the WGS data from 10,418 individuals from AMP-PD (Accelerating Medicines Partnership: Parkinson’s Disease) v3 release (2022), which encompasses study participants at Mass General Brigham (i.e., Harvard Biomarkers Study 2.0). Any individual belonging to the genetic registry and genetic cohort group, as well as subjects without evidence of dopamine deficit (SWEDD) and subjects belonging to the prodromal categories and the AMP-LBD cohort were excluded from the analysis. As the AMP-PD cohort predominantly consists of individuals of European ancestry, we retained only unrelated individuals of the same ancestry group. We called the resulting cohort “AMP-PD-sporadic,” which consisted of 2,117 sporadic PD cases and 1,095 neurotypical controls. Both case and control groups had similar median ages in this cohort ( Supplementary Table 22 ). We performed both variant-and sample-level quality control (QC) steps on each dataset to ensure the study cohorts are free from technical biases as much as possible (see Supplementary Note for detailed steps). We retained only high-quality biallelic variants passing GATK best practices filters and having maximum 10% missingness for our analysis. We annotated variants with their functional consequences and gnomAD allele frequencies with VEP (v109) and dbNSFP (v4.3a) database. We used six masks to group variants into functional categories: (i) Damaging missense : missense variants predicted to be either ‘‘P’’ or ‘‘D’’ by PolyPhen2 (v2.2.3) or ‘‘deleterious’’ by SIFT (v6.2.1), (ii) LoF : variants labelled as splice donors, splice acceptors, splice region variants, stop-gained, stop-lost, start-lost, frameshifts, in-frame insertions, and in-frame deletions; (iii) Damaging : LoFs and damaging missenses, (iv) Missense , (v) Neutral : missense variants predicted to be either ‘‘B’’ by PolyPhen2 or ‘‘tolerated’’ by SIFT, and (vi) Synonymous . These masks were used in our analysis of binarized cholesterol phenotypes as well as complex diseases. Notably, NERINE is agnostic to the variant annotation tool and can test any user-defined categories of rare variants. To assess robustness, we reclassified variants using REVEL 117 (version May 2021) and AlphaMissense 118 scores (hg38 v1) within the VEP (v109) plugin and custom annotation tools in a positive control experiment with the LDL-C phenotype in UKBB. We defined three categories: (i) pathogenic missense (REVEL > 0.649 or AlphaMissense > 0.564), (ii) pathogenic (pathogenic missense plus LoF), and (iii) benign missense (REVEL < 0.29 and AlphaMissense < 0.34). Results were highly concordant with those obtained using PolyPhen2 and SIFT ( Supplementary Figure S24 ). We removed sample outliers based on Ts/Tv, Het/Hom ratios, and per-haploid SNV counts, where outliers were defined as samples that were three standard deviations away from the mean. We retained only unrelated European samples for our analyses because, at the time, individuals of European ancestry constituted the largest group in our study cohorts. Concentrated eiorts in building large biobanks with diverse participants are already underway 119 and will enable NERINE to overcome this limitation and provide more insight into the contribution of rare variants to common disease etiology across populations. Human iPSCs culture and DA neuron di_erentiation Human H9 (MSKCC stem cell core facility; WA-09) ESCs were cultured in Essential 8 (E8) medium (Gibco/Thermo Fisher Scientific; Cat. No. A1517001) on 10cm plates coated with Vitronectin (Thermo Fisher Scientific; Cat. No. A14700) diluted 1:100 in Dulbecco’s PBS (DPBS; Thermo Fisher Scientific; Cat. No. 14190136) and passaged at ∼80% confluence using StemPro Accutase Cell Dissociation Reagent (Gibco/Thermo Fisher Scientific; Cat. No. A11105-01). E8 medium was replaced every day. ESCs harboring a doxycycline-inducible Cas9 cassette in the AAVS1 safe harbor locus were diierentiated following a previously published protocol 120 . Briefly, ESCs were dissociated and plated at high density (600,000 cells/cm²) in Neurobasal medium (Gibco/Thermo Fisher Scientific; Cat. No. 21103-049) supplemented with 1X N2 supplement (Life Technologies; Cat. No. 17502048), 1X B27 supplement (Life Technologies; Cat. No. 17504044), 2 mM L-glutamine (Gibco/Thermo Fisher Scientific; Cat. No. 25030-081), penicillin-streptomycin (Gibco/Thermo Fisher Scientific; Cat. No. 15140122), 250 nM LDN193189 (Reprocell; Cat. No. 04-0074), 10 μM SB431542 (R&D Systems; Cat. No. 1614), 1 μM CHIR99021 (R&D Systems; Cat. No. 4432), 500 ng/mL SHH (R&D Systems; Cat. No. 464-SH), and 10 nM Y-27632 ROCK inhibitor (Bio-Techne; Cat. No. HY-10583). Cells were counted and plated on DMEM/F12 (Gibco/Thermo Fisher Scientific; Cat. No. 11320033) and Geltrex-coated plates (Life Technologies; Cat. No. A1413201) at high density (600,000 cells/cm 2 ), while being maintained in this medium from DIV 0 (day of plating) to DIV 3. Y-27632 ROCK inhibitor was added only at plating. Medium change was performed on DIV 4 and every three days till DIV 9. On DIV 4, the medium was supplemented with 6µM CHIR99021. On DIV 7, the medium was composition was changed to exclude SB431542 and SHH. From DIV 10, cells were transitioned to maturation medium with the following composition: Neurobasal Medium as the base, 1X B27 Supplement, 2mM L-glutamine, penicillin-streptomycin, 3µM CHIR99021, 0.2mM Ascorbic Acid (Sigma-Aldrich; Cat. No. A4034), 20ng/mL GDNF (Gibco/Thermo Fisher Scientific; Cat. No. 450-10), 1ng/mL TGFβ3 (Gibco/Thermo Fisher Scientific; Cat. No. 100-36E), 20ng/mL BDNF (R&D Systems; Cat. No. 248-BDB), and 0.2mM dbcAMP (Sigma-Aldrich; Cat. No. 4043). For DIVs 10 and 11, cells were maintained in this medium. On DIV 11, partially-diierentiated cells were released and centrifuged as above, and pelleted cells were resuspended in the same medium from DIVs 10-11 and plated at high density (800,000 cells/cm 2 ) on 15µg/mL Poly-L-Ornithine (Sigma-Aldrich; Cat. No. P3655), 2µg/mL Fibronectin (Gibco/Thermo Fisher Scientific; Cat. No. 356008), and 1µg/mL Laminin (R&D Systems; Cat. No. 3400-010-03)-coated plates in DPBS. From DIVs 12 to 15, cell were placed into maturation medium with a composition similar to DIVs 10-11 medium supplemented with 1 μM IWP-2 (Tocris Bioscience; Cat. No. 3533) and 100 ng/mL FGF-18 (PeproTech; Cat. No. 100-28). Doxycycline hydrochloride (2 μg/mL; Sigma-Aldrich; Cat. No. D3447) was added on DIVs 14 and 15 to induce Cas9 expression. On DIV 16, partially-diierentiated cells were released and centrifuged as above, and pelleted cells were replated at 1,200,000 cells/cm² and resuspended in maturation medium with a composition similar to DIVs 10-11 medium supplemented with 10 μM DAPT (R&D Systems; Cat. No. 2634) for ten days. On DIV 25, neurons were dissociated and replated at low density (200,000 cells/cm²) and maintained in maturation medium containing 15µg/mL Poly-L-Ornithine, 2µg/mL Fibronectin, and 1µg/mL Laminin-coated plates in DPBS. Genome-wide CRISPR/Cas9 screen in PSC-derived DA neurons We performed genome-wide CRISPR/Cas9 screen in DA neurons diierentiated from human WA-09 (H9) embryonic stem cells as described above. The Gattinara human CRISPR pooled knockout library 121 (Addgene; pooled library #136986) was used for this screen; this library includes two gRNAs for 19,993 genes as well as 500 non-targeting controls and 500 one-intergenic site targeting controls. Guide RNAs (gRNAs) representing 19,993 genes were transduced into H9 hESC lines at an MOI of 0.3-0.5 and 1000x representation. Transduced stem cells were selected by puromycin and diierentiated toward DA neurons until reaching the neural progenitor stage as described by Kim and colleagues 122 . Briefly, at DIVs 14-16 of the diierentiation, we induced iCas9 expression by doxycycline addition using the AAVS1 safe-harbor locus as previously described 123 , while cells were neural progenitors. Then, we waited until DIV25 when they diierentiated into DA neurons to take our initial sample to obtain gRNA representation through whole genome sequencing (DIV26). The remaining neurons were allowed to stay in the dish until our final collection time (DIV42) when neuronal cell death began. Samples were processed for library preparation and sequenced and sequencing reads were aligned to the screened library and analyzed using MAGeCK-MLE from the MAGeCKFlute 124 (v2.6.0) package in R (v4.3.2). Essentiality genes were classified as having Wald test FDR-adjusted p-value < 0.05 and beta <-0.58. Broadly essential genes 124 that are not specific to DA neurons, were removed from the list of essentiality genes. We identified 693 essentiality genes. The complete screen will be described in a forthcoming publication. Constructing ontology-based network topologies from gene sets For PD GWAS genes as well essential genes for DA neuron survival, we first performed gene set enrichment analysis using the enrichr function from the GSEApy package (v 1.1.3) in python (v 3.12.4) and identified all GO biological processes with nominal significance (p-value < 0.05). We then grouped semantically similar GO terms using REVIGO 125 ( http://revigo.irb.hr/ ) to identify GO biological process (BP) modules with minimal overlap. We only kept modules of 10 or more genes for our analysis. For PD GWAS genes, we identified six such modules ( Supplementary Table T12 ) and for DA neuron essentiality genes, we identified 10 such modules ( Supplementary Table T15 ). To impose network topology on these gene sets, we extracted edge relationships of genes in each group from three diierent data sources as described above: (i) high-confidence physical and genetics interactions from protein interaction databases, (ii) co-expression in the substantia nigra region of the mid-brain from GTEx v8, and (iii) co-essentiality in CNS cell types in DepMap v2023Q2. We also explored an alternative approach from a recent study 65 for constructing gene modules from PD GWAS loci, which involved running MAGMA 126 analysis on GWAS genes, followed by GO BP enrichment. This approach identified 21 significant conditionally independent GO terms 65 . We then imposed network topology on the overlapping genes with each GO term by extracting gene-gene relationships from the sources described above. Human iPSCs culture and induced neuron di_erentiation Human GM29371 iPSCs (Coriell Institute; GM29371*C) were cultured in Stemflex (Gibco/Thermo Fisher Scientific; Cat. No. A33493) on 6-well plates coated with Matrigel Matrix (Corning; Cat. No. 356231) diluted 1:100 in Knockout DMEM (Gibco/Thermo Fisher Scientific; Cat. No. 10829-018). Essential 8 Medium (Gibco/Thermo Fisher Scientific; Cat. No. A1517001) was replaced every day. When 80% confluent, cells were passaged with StemPro Accutase Cell Dissociation Reagent (Gibco/Thermo Fisher Scientific; Cat. No. A11105-01). Human iPSCs engineered to express NGN2 under a doxycycline-inducible system in the AAVS1 safe harbor locus were diierentiated following previously published protocol 86 . Briefly, iPSCs were released as above, centrifuged, and resuspended in N2 Pre-Diierentiation Medium containing the following: Knockout DMEM/F12 (Gibco/Thermo Fisher Scientific; Cat. No. 12660-012) as the base, 1X MEM Non-Essential Amino Acids (Gibco/Thermo Fisher Scientific; Cat. No. 11140-050), 1X N2 Supplement (Gibco/Thermo Fisher Scientific; Cat. No. 17502-048), 10ng/mL NT-3 (PeproTech; Cat. No. 450-03), 10ng/mL BDNF (PeproTech; Cat. No. 450-02), 1 μg/mL Mouse Laminin (Thermo Fisher Scientific; Cat. No. 23017-015), 10nM ROCK inhibitor (Peprotech; Cat. No. 1293823), and 2μg/mL doxycycline hydrochloride (Sigma-Aldrich; Cat. No. D3447-500MG) to induce expression of mNGN2. iPSCs were counted and plated on Matrigel-coated plates in N2 Pre-Diierentiation Medium for three days. After three days, hereafter DIV 0, pre-diierentiated cells were released and centrifuged as above, and pelleted cells were resuspended in Classic Neuronal Medium containing the following: half DMEM/F12 (Gibco/Thermo Fisher Scientific; Cat. No. 11320-033) and half Neurobasal-A (Gibco/Thermo Fisher Scientific; Cat. No. 10888-022) as the base, 1X MEM Non-Essential Amino Acids, 0.5X GlutaMAX Supplement (Gibco/Thermo Fisher Scientific; Cat. No. 35050-061), 0.5X N2 Supplment, 0.5X B27 Supplement (Gibco/Thermo Fisher Scientific; Cat. No. 17504-044), 10ng/mL NT-3, 10ng/mL BDNF, 1μg/mL Mouse Laminin, and 2μg/mL doxycycline hydrochloride. Pre-diierentiated cells were subsequently counted and plated on BioCoat Poly-D-Lysine coated plates (Corning; Cat. No. 356470) in Classic Neuronal Medium. On DIV 7 and each week after, medium change was performed without doxycycline added. In the PFF exposure experiment, a complete medium change to the medium containing 10 µg/mL synthetic PFF was performed on DIV 21 neurons, and the neurons were fixed at DIV28 for immunostaining. CRISPRi screen and analysis in the CiS-CN model A customized CRISPRi library comprising 9,852 sgRNAs targeting 1,705 physical and genetic interactors of α-synuclein (αS)—including genes from the TransposeNet αS proteinopathy network—along with negative controls, was constructed. The library was packaged into lentivirus by the Virus Core at Boston Children’s Hospital. Neurons were transduced with the viral library and cultured until the harvesting date (DIVs 3, 28, and 42). Genomic DNA was extracted from harvested neurons, and the sgRNA-encoding regions were amplified for sequencing based on previously described protocols 86 , 127 , 128 . Sequencing and annotation were conducted at Memorial Sloan Kettering Cancer Center. Data was analyzed using the MAGeCK-iNC pipeline 86 from the MAGeCK package (v0.5.9.2) using python (v2.7). Hits were classified as having an FDR-adjusted p-value < 0.1, and the gene product cutoff was selected by the MAGeCK-iNC pipeline dynamically for each comparison. Hits with significant sgRNA dropout were termed “toxicity enhancers.” Details of the functional screen in its entirety will be described in a forthcoming publication from our group. Immunostaining and microscopy imaging Neurons are fixed with 4% PFA (EM Sciences; Cat. No. 15710) for 15 minutes at room temperature, and then permeabilized with 0.5% Triton X-100 (Sigma-Aldrich; Cat. No. T8787) and blocked with 0.05% Triton X-100 and 5% BSA (Sigma-Aldrich; Cat. No. A7906) in DPBS (Thermo Fisher Scientific; Cat. No. 14040182) for 1hr at room temperature. Samples were incubated with primary antibodies (PRL: Thermo Fisher Scientific; Cat. No. MA1-10597; 1:200 dilution) at 4°C overnight, followed by incubation with secondary antibody and Hoechst 33342 (Thermo Fisher Scientific; Cat. No. H3570; 1:2000 dilution) for 1hr at room temperature. Images were captured with identical settings for parallel cultures using Nikon Eclipse Ti microscope or Nikon TiE/C2 confocal microscope. Image analysis was performed with ImageJ2 129 Macro Software ( Supplementary Note ). Prl level was determined by D (D = total prolactin intensity / total DAPI number in a given image). Oxidative stress assay DIV0 CiS neurons were seeded at a density of 40,000 cells/well of poly-L-ornithine-coated 96-well plate (solution: Sigma-Aldrich; P4957). At DIV6, the neuron media was fully changed for 100uL Prolactin -containing (Thermo Fisher Scientific; Cat. No. 100-07-10UG) media at a concentration of 0,1 or 10 nM. 24hrs post-treatment, the neuron media was fully changed for 100uL of Menadione-containing (Thermo Fisher Scientific; Cat. No. ICN10225925) media at a concentration of 0 or 100uM and incubated for an hour at 37°C. After an hour, CELLROX™ Green Reagent (Thermo Fisher Scientific; Cat. No. C10444) was added on top of the Menadione-treated media to a final concentration of 5 uM CELLROX™ for 30 minutes at 37°C. All media was then removed, and wells were washed 3 times with PBS. After the third wash, the PBS was replaced with neuron media. The plates were then taken to the Incucyte ® S3 live-cell analysis system (Sartorius) for imaging. Incucyte analysis was perform with S3 software, and CellRox = total integrated intensity/neuron was reported. Mouse model Wildtype B6C3F1 mice (The Jackson Laboratories; Stock 100010) were used for the stereotactic injection studies described. Animals were maintained on a 12-hour light/dark schedule and provided with food ad libitum. All housing, breeding, and procedures were performed according to the NIH Guide for the Care and Use of Experimental Animals and approved by the University of Pennsylvania Institutional Animal Care and Use Committee. ɑS and PFF preparation Full-length mouse ɑS was expressed in BL21 (DE3) RIL-competent E. coli cells (Agilent Technologies; Cat. No. 230245) transformed with pRK172/mSyn containing ɑS cDNA 29 . Protein purification was previously described 29 , 130 . Cultures expanded in Terrific Broth (Sigma-Aldrich; Cat. No. T9179, composition: 12 g/L of Bacto-tryptone, 24 g/L of yeast extract 4% (vol/vol) glycerol, 17 mM KH2PO4 and 72 mM K2HPO4) containing ampicillin (Fisher Scientific; Cat. No. BP1760) were harvested and sonicated in high salt buier (750 mM NaCl in 10 mM Tris, pH 7.6; NaCl: Fisher Scientific, Cat. No. S271; Tris: Fisher Scientific, Cat. No. BP152). After boiling for 15 mins, the supernatant was dialyzed against 10 mM Tris, pH 7.6, 50 mM NaCl, 1 mM EDTA (Sigma-Aldrich; Cat. No. E5134) overnight at 4°C, filtered and concentrated using Amicon Ultra-15 centrifugal filter units (Merck Millipore; Cat. No. UFC901008). Gel filtration using a Superdex 200 column (Cytiva; Cat. No. 17517501) was performed and fractions containing ɑS pooled and dialyzed in 10 mM Tris, pH 7.6, 50 mM NaCl, 1 mM EDTA overnight. The product was polished using a HiTrapQ HP column (Cytiva; Cat. No. 645932) and eluted over an ionic gradient (25 to 1,000 mM NaCl). Fractions containing ɑS were combined and dialyzed into DPBS (Invitrogen; Cat. No. 14200075), sterile filtered and concentrated to 5 mg/mL and frozen at-80°C until used. PFFs were assembled by shaking monomer at 5 mg/mL using a Thermomixer C (Eppendorf) set at 1,000 rpm for 7 days at 37°C. Fibril content was validated by sedimentation at 100,000 x g for 30 minutes and Thioflavin T fluorimetry. Stereotaxic administration of PFFs Prior to injection, PFFs were diluted to 2 mg/mL in DPBS and sonicated using a bath sonicator (Diagenode; Biorupter UCD-300) on high power for 10 cycles (30 sec on; 30 sec oi) at 10°C. Each mouse received a single unilateral injection of PFFs (5 µg of PFFs in 2.5 µL volume) into the dorsal striatum using a Hamilton syringe (33 gauge) using the following co-ordinates: AP +0.2 mm relative to Bregma; ML +2.0 mm; depth 2.6 mm beneath the dura. DPBS injected into the same region was used as a negative control. Mice were perfused transcardially with heparinized PBS at 30 d.p.i. and brains flash frozen at-80°C until use. RNA isolation and NanoString analysis The amygdala region ipsilateral to PFF-or PBS injection was microdissected from each brain and homogenized in 1 ml of TRI Reagent (Sigma-Aldrich, Cat. No. T9424) using TissueRuptor II (Qiagen, Cat. No. 9002755) with a disposable probe (Qiagen, Cat. No. 990890). RNA was then isolated using Direct-zol RNA MiniPrep kit with in-column DNaseI treatment (Zymo; Cat. No. R2050). Samples were quantitated with a NanoDrop 1000 Spectrophotometer (Thermo Fisher Scientific) and assayed for RNA integrity on a 4200 TapeStation (Agilent; Cat. No. G2991AA). NanoString hybridization of the resultant RNA was carried out for a constant 18 hours at 65° C on the Mus musculus Neuropathology panel (v1.0). Post-hybridization processing in the nCounter Prep Station used the High Sensitivity settings. The cartridge scanning parameter was set at high (555 FOV). RNA isolation and NanoString studies were performed at the Wistar Genomics core facility. Genes with hybridization counts >20 were normalized to the geometric mean of panel positive controls as recommended by the manufacturer. Diierential expression analysis was performed using a generalized linear model in the nCounter module in Rosalind (Rosalind.bio). Adjusted p-values were calculated using the Benjamini-Hochberg method using treatment (i.e. PFF vs PBS). Data and code availability Canonical pathway gene set were obtained from MSigDB 131 (v7.3; https://www.gsea-msigdb.org/gsea/msigdb/human/collections.jsp ). Human physical protein–protein interactions (PPI) were downloaded from STRING 132 (v11.5; https://string-db.org/cgi/download ), HuRI ( http://www.interactome-atlas.org/download : last accessed on Jan 2022), and inBio Map 112 ( https://www.intomics.com/inbio/map : last accessed on Jan 2022) databases. Additionally, genetic interactions data was obtained from the supplement of the Megchelenbrink et al. study 113 . TransposeNet’s humanized 𝛼-synuclein-, 𝛽-amyloid-, and TDP-43-modifier networks were obtained from our previously published study 2 . Bulk expression data in TPM format from diierent human tissue types were downloaded from the GTEx 114 (v8; https://www.gtexportal.org/ ), and gene dependency data in diierent cell lines from DepMap 116 (release: 2023Q2; https://depmap.org/portal/data_page/?tab=allData ). Whole exome sequencing (WES) and phenotypic data from UKBB 133 , available through https://ams.ukbiobank.ac.uk , were accessed via application 41250 and processed on the DNAnexus platform ( https://ukbiobank.dnanexus.com/landing ). MGBBB 134 WES and phenotypic data were accessed via https://biobankportal.partners.org/ (PI: Khurana). Access to MGBBB data was restricted to only aiiliated investigators. AMP-PD 135 whole genome sequencing and phenotypic data (v2.5; release 2022) were accessed through the AMP-PD Knowledge Platform ( https://www.amp-pd.org ). NERINE’s source code is available on github ( https://github.com/snz20/NERINE ) and zenodo (doi: 10.5281/zenodo.19209293). RVTT was run adapting the code from https://github.com/snz20/RVTT (zenodo doi: 10.5281/zenodo.10627549). CMC-Fisher, Fisher’s combined test, and SKAT-O were run on R (v4.3.2) using stats (v4.3.2), poolr (v1.2.0), and SKAT (v2.2.5) packages. MAGeCK-iNC analysis was performed using the MAGeCK 136 (v0.5.9.2) package on python (v2.7) and GO gene set enrichment analysis was performed using GSEApy (v1.1.3) on python (v3.12.4). Comprehensive information on study-related resources is listed in Supplementary Table T24 . Author contributions S.N., V.K., and S.S. conceived of the project, interpreted the results; S.N., X.W., V.K., and S.S. wrote the manuscript; S.N. and S.S. developed the novel methodology NERINE; S.N. pre-processed and analyzed the study cohorts; A.M. assisted in the quality control and the pre-processing of the sequencing data; S.N., S.S., N.S. and R.G. interpreted the results in lipid and cardiac phenotypes; C.C. helped analyze the UKBB cohorts; X.W. and V.K. designed the wet laboratory experiments to validate the NERINE’s findings in PD in neuronal models; X.W., R.S., and E.E. generated the CiS neurons and performed the experiments; X.W. and S.N. analyzed the experimental data and interpreted the results with V.K.; D.R., A.H., J.A., and L.S. designed the genome-wide CRISPR/Cas9 experiments in mid-brain DA neurons, identified the essentiality genes, and curated the GO biological process modules; K.L. designed and performed the experiments in the mouse PFF model, analyzed the data, and interpreted the results with V.K., X.W., and S.N.. All the authors read and helped edit the manuscript. Declaration of conflict-of-interest The authors have declared that no conflicts of interest exist. Acknowledgements We thank Drs. Matthew Stevens, Richard Sherwood, Benjamin Neale, and Isabel Lam for their valuable insights. S.N. is supported by the NIH grant R35GM127131, the Sudarsky Scholar Award (Brigham and Women’s Hospital, Movement Disorders Division), and the Australian Parkinson’s Mission. S.S. is supported by NIH grants U01HG012009, R35GM127131, and R01MH101244. V.K, X.W., L.S., and experiments in PD neuronal models are supported by the Aligning Science Across Parkinson’s Initiative (ASAP) award ASAP-000472 (PI: Studer, co-PI: Khurana). V.K. also acknowledges support from an APDA Center for Advanced Research grant, the NIH grant R01NS109209, the New York Stem Cell Foundation Robertson Investigator award (NYSCF-R-I49), as well as the generous support from Mrs. Nancy Black Simches and the Ocko family Parkinson’s Disease Innovation Award. X.W. additionally acknowledges support from the NIH grant T32AG000222 (PI: Yankner). We analyzed WES data from the UK Biobank (UKBB, application 41250, PI: Cassa; Mass General Brigham IRB 2020P002093; NIH R01HG010372, PI: Sunyaev), a global biomedical resource supported by the Wellcome Trust, UK Medical Research Council, the Department of Health, Scottish Government, the Northwest Regional Development Agency, British Heart Foundation, and Cancer Research UK. We also used WES and phenotypic data from ∼50k participants from the Mass General Brigham Biobank (MGBBB), a biorepository of consented patient samples at Mass General Brigham (parent organization of Massachusetts General Hospital and Brigham and Women’s Hospital), and WGS data from AMP-PD, a public-private partnership managed by FNIH and funded by Celgene, GSK, the Michael J. Fox Foundation, NINDS, Pfizer, and Verily. AMP-PD investigators did not review this work. We thank all the participants, clinical investigators, and research teams who contributed to UKBB, MGBBB, and AMP-PD. Footnotes This version contains updated affiliation of authors, Methods, acknowledgments, and data and code availability sections. Typos are fixed for main figures and legends. A supplemental table on study related resources was added. Reference 1. ↵ Horvath , S. , and Dong , J . ( 2008 ). Geometric interpretation of gene coexpression network analysis . PLoS Comput Biol 4 , e1000117 . doi: 10.1371/journal.pcbi.1000117 . OpenUrl CrossRef PubMed 2. ↵ Khurana , V. , Peng , J. , Chung , C.Y. , Auluck , P.K. , Fanning , S. , Tardii , D.F. , Bartels , T. , Koeva , M. , Eichhorn , S.W. , Benyamini , H. , et al. ( 2017 ). Genome-Scale Networks Link Neurodegenerative Disease Genes to alpha-Synuclein through Specific Molecular Pathways . Cell Syst 4 , 157 – 170 e114. doi: 10.1016/j.cels.2016.12.011 . OpenUrl CrossRef PubMed 3. Tsherniak , A. , Vazquez , F. , Montgomery , P.G. , Weir , B.A. , Kryukov , G. , Cowley , G.S. , Gill , S. , Harrington , W.F. , Pantel , S. , Krill-Burger , J.M. , et al. ( 2017 ). Defining a Cancer Dependency Map . Cell 170 , 564 – 576 e516. doi: 10.1016/j.cell.2017.06.010 . OpenUrl CrossRef PubMed 4. Sledzieski , S. , Singh , R. , Cowen , L. , and Berger , B . ( 2021 ). D-SCRIPT translates genome to phenome with sequence-based, structure-aware, genome-scale predictions of protein-protein interactions . Cell Syst 12 , 969 – 982 e966. doi: 10.1016/j.cels.2021.08.010 . OpenUrl CrossRef PubMed 5. ↵ Schnitzler , G.R. , Kang , H. , Fang , S. , Angom , R.S. , Lee-Kim , V.S. , Ma , X.R. , Zhou , R. , Zeng , T. , Guo , K. , Taylor , M.S. , et al. ( 2024 ). Convergence of coronary artery disease genes onto endothelial cell programs . Nature 626 , 799 – 807 . doi: 10.1038/s41586-024-07022-x . OpenUrl CrossRef PubMed 6. ↵ Luck , K. , Kim , D.K. , Lambourne , L. , Spirohn , K. , Begg , B.E. , Bian , W. , Brignall , R. , Cafarelli , T. , Campos-Laborie , F.J. , Charloteaux , B. , et al. ( 2020 ). A reference map of the human binary protein interactome . Nature 580 , 402 – 408 . doi: 10.1038/s41586-020-2188-x . OpenUrl CrossRef PubMed 7. ↵ Yuan , Q. , and Duren , Z . ( 2025 ). Inferring gene regulatory networks from single-cell multiome data using atlas-scale external data . Nat Biotechnol 43 , 247 – 257 . doi: 10.1038/s41587-024-02182-7 . OpenUrl CrossRef 8. ↵ Vidal , M. , Cusick , M.E. , and Barabasi , A.L . ( 2011 ). Interactome networks and human disease . Cell 144 , 986 – 998 . doi: 10.1016/j.cell.2011.02.016 . OpenUrl CrossRef PubMed Web of Science 9. Barabasi , A.L. , Gulbahce , N. , and Loscalzo , J . ( 2011 ). Network medicine: a network-based approach to human disease . Nat Rev Genet 12 , 56 – 68 . doi: 10.1038/nrg2918 . OpenUrl CrossRef PubMed Web of Science 10. ↵ Ota , M. , Spence , J.P. , Zeng , T. , Dann , E. , Marson , A. , and Pritchard , J.K . ( 2025 ). Causal modeling of gene eiects from regulators to programs to traits: integration of genetic associations and Perturb-seq . bioRxiv . doi: 10.1101/2025.01.22.634424 . OpenUrl Abstract / FREE Full Text 11. ↵ Lee , S. , Abecasis , G.R. , Boehnke , M. , and Lin , X . ( 2014 ). Rare-variant association analysis: study designs and statistical tests . Am J Hum Genet 95 , 5 – 23 . doi: 10.1016/j.ajhg.2014.06.009 . OpenUrl CrossRef PubMed 12. ↵ Chen , W. , Coombes , B.J. , and Larson , N.B . ( 2022 ). Recent advances and challenges of rare variant association analysis in the biobank sequencing era . Front Genet 13 , 1014947 . doi: 10.3389/fgene.2022.1014947 . OpenUrl CrossRef PubMed 13. ↵ Richardson , T.G. , Timpson , N.J. , Campbell , C. , and Gaunt , T.R . ( 2016 ). A pathway-centric approach to rare variant association analysis . Eur J Hum Genet 25 , 123 – 129 . doi: 10.1038/ejhg.2016.113 . OpenUrl CrossRef PubMed 14. ↵ Lee , S. , Choi , S. , Kim , Y.J. , Kim , B.J. , Consortium , T.d.-G. , Hwang , H. , and Park , T. ( 2016 ). Pathway-based approach using hierarchical components of collapsed rare variants . Bioinformatics 32 , i586 – i594 . doi: 10.1093/bioinformatics/btw425 . OpenUrl CrossRef PubMed 15. ↵ Bendapudi , P.K. , Nazeen , S. , Ryu , J. , Soylemez , O. , Robbins , A. , Rouaisnel , B. , O’Neil , J.K. , Pokhriyal , R. , Yang , M. , Colling , M. , et al. ( 2024 ). Low-frequency inherited complement receptor variants are associated with purpura fulminans . Blood 143 , 1032 – 1044 . doi: 10.1182/blood.2023021231 . OpenUrl CrossRef 16. ↵ Hallacli , E. , Kayatekin , C. , Nazeen , S. , Wang , X.H. , Sheinkopf , Z. , Sathyakumar , S. , Sarkar , S. , Jiang , X. , Dong , X. , Di Maio , R. , et al. ( 2022 ). The Parkinson’s disease protein alpha-synuclein is a modulator of processing bodies and mRNA stability . Cell 185 , 2035 – 2056 e2033. doi: 10.1016/j.cell.2022.05.008 . OpenUrl CrossRef PubMed 17. Li , X. , Quick , C. , Zhou , H. , Gaynor , S.M. , Liu , Y. , Chen , H. , Selvaraj , M.S. , Sun , R. , Dey , R. , Arnett , D.K. , et al. ( 2023 ). Powerful, scalable and resource-eiicient meta-analysis of rare variant associations in large whole genome sequencing studies . Nat Genet 55 , 154 – 164 . doi: 10.1038/s41588-022-01225-6 . OpenUrl CrossRef PubMed 18. ↵ Lee , S. , Emond , M.J. , Bamshad , M.J. , Barnes , K.C. , Rieder , M.J. , Nickerson , D.A. , Team, N.G.E.S.P.-E.L.P., Christiani , D.C. , Wurfel , M.M. , and Lin , X . ( 2012 ). Optimal unified approach for rare-variant association testing with application to small-sample case-control whole-exome sequencing studies . Am J Hum Genet 91 , 224 – 237 . doi: 10.1016/j.ajhg.2012.06.007 . OpenUrl CrossRef PubMed 19. ↵ Zhou , W. , Bi , W. , Zhao , Z. , Dey , K.K. , Jagadeesh , K.A. , Karczewski , K.J. , Daly , M.J. , Neale , B.M. , and Lee , S . ( 2022 ). SAIGE-GENE+ improves the eiiciency and accuracy of set-based rare variant association tests . Nat Genet 54 , 1466 – 1469 . doi: 10.1038/s41588-022-01178-w . OpenUrl CrossRef PubMed 20. Lee , S. , Kim , Y. , Choi , S. , Hwang , H. , and Park , T . ( 2018 ). Pathway-based approach using hierarchical components of rare variants to analyze multiple phenotypes . BMC Bioinformatics 19 , 79 . doi: 10.1186/s12859-018-2066-9 . OpenUrl CrossRef PubMed 21. Lee , S. , Kim , S. , Kim , Y. , Oh , B. , Hwang , H. , and Park , T . ( 2019 ). Pathway analysis of rare variants for the clustered phenotypes by using hierarchical structured components analysis . BMC Med Genomics 12 , 100 . doi: 10.1186/s12920-019-0517-4 . OpenUrl CrossRef PubMed 22. Guo , M.H. , Plummer , L. , Chan , Y.M. , Hirschhorn , J.N. , and Lippincott , M.F . ( 2018 ). Burden Testing of Rare Variants Identified through Exome Sequencing via Publicly Available Control Data . Am J Hum Genet 103 , 522 – 534 . doi: 10.1016/j.ajhg.2018.08.016 . OpenUrl CrossRef PubMed 23. Zhao , J. , Zhu , Y. , Boerwinkle , E. , and Xiong , M . ( 2015 ). Pathway analysis with next-generation sequencing data . Eur J Hum Genet 23 , 507 – 515 . doi: 10.1038/ejhg.2014.121 . OpenUrl CrossRef PubMed 24. Wu , G. , and Zhi , D . ( 2013 ). Pathway-based approaches for sequencing-based genome-wide association studies . Genet Epidemiol 37 , 478 – 494 . doi: 10.1002/gepi.21728 . OpenUrl CrossRef PubMed 25. ↵ Pan , W. , Kwak , I.Y. , and Wei , P . ( 2015 ). A Powerful Pathway-Based Adaptive Test for Genetic Association with Common or Rare Variants . Am J Hum Genet 97 , 86 – 98 . doi: 10.1016/j.ajhg.2015.05.018 . OpenUrl CrossRef PubMed 26. ↵ Makarious , M.B. , Lake , J. , Pitz , V. , Ye Fu , A. , Guidubaldi , J.L. , Solsberg , C.W. , Bandres-Ciga , S. , Leonard , H.L. , Kim , J.J. , Billingsley , K.J. , et al. ( 2023 ). Large-scale rare variant burden testing in Parkinson’s disease . Brain 146 , 4622 – 4632 . doi: 10.1093/brain/awad214 . OpenUrl CrossRef 27. ↵ Pitz , V. , Makarious , M.B. , Bandres-Ciga , S. , Iwaki , H. , and Me Research , T. , Singleton , A.B. , Nalls , M. , Heilbron , K. , and Blauwendraat , C. ( 2024 ). Analysis of rare Parkinson’s disease variants in millions of people . NPJ Parkinsons Dis 10 , 11 . doi: 10.1038/s41531-023-00608-8 . OpenUrl CrossRef PubMed 28. ↵ Nazeen , S. , Wang , X. , Zielinski , D. , Lam , I. , Hallacli , E. , Xu , P. , Ethier , E. , Strom , R. , Zanella , C.A. , Nithianandam , V. , et al. ( 2024 ). Deep sequencing of proteotoxicity modifier genes uncovers a Presenilin-2/beta-amyloid-actin genetic risk module shared among alpha-synucleinopathies . bioRxiv , 2024.2003.2003.583145. doi: 10.1101/2024.03.03.583145 . OpenUrl Abstract / FREE Full Text 29. ↵ Luk , K.C. , Kehm , V. , Carroll , J. , Zhang , B. , O’Brien , P. , Trojanowski , J.Q. , and Lee , V.M . ( 2012 ). Pathological alpha-synuclein transmission initiates Parkinson-like neurodegeneration in nontransgenic mice . Science 338 , 949 – 953 . doi: 10.1126/science.1227157 . OpenUrl Abstract / FREE Full Text 30. ↵ Shapiro , A . ( 1985 ). Asymptotic distribution of test statistics in the analysis of moment structures under inequality constraints . Biometrika 72 , 133 – 144 . doi: 10.1093/biomet/72.1.133 . OpenUrl CrossRef Web of Science 31. ↵ Stoel , R.D. , Garre , F.G. , Dolan , C. , and v an den Wittenboer , G. ( 2006 ). On the likelihood ratio test in structural equation modeling when parameters are subject to boundary constraints . Psychol Methods 11 , 439 – 455 . doi: 10.1037/1082-989X.11.4.439 . OpenUrl CrossRef PubMed Web of Science 32. ↵ Paththinige , C.S. , Sirisena , N.D. , and Dissanayake , V . ( 2017 ). Genetic determinants of inherited susceptibility to hypercholesterolemia - a comprehensive literature review . Lipids Health Dis 16 , 103 . doi: 10.1186/s12944-017-0488-4 . OpenUrl CrossRef PubMed 33. ↵ Karczewski , K.J. , Solomonson , M. , Chao , K.R. , Goodrich , J.K. , Tiao , G. , Lu , W. , Riley-Gillis , B.M. , Tsai , E.A. , Kim , H.I. , Zheng , X. , et al. ( 2022 ). Systematic single-variant and gene-based association testing of thousands of phenotypes in 394,841 UK Biobank exomes . Cell Genom 2 , 100168 . doi: 10.1016/j.xgen.2022.100168 . OpenUrl CrossRef PubMed 34. ↵ Vujkovic , M. , Keaton , J.M. , Lynch , J.A. , Miller , D.R. , Zhou , J. , Tcheandjieu , C. , Huiman , J.E. , Assimes , T.L. , Lorenz , K. , Zhu , X. , et al. ( 2020 ). Discovery of 318 new risk loci for type 2 diabetes and related vascular outcomes among 1.4 million participants in a multi-ancestry meta-analysis . Nat Genet 52 , 680 – 691 . doi: 10.1038/s41588-020-0637-y . OpenUrl CrossRef PubMed 35. Zeggini , E. , Scott , L.J. , Saxena , R. , Voight , B.F. , Marchini , J.L. , Hu , T. , de Bakker , P.I. , Abecasis , G.R. , Almgren , P. , Andersen , G. , et al. ( 2008 ). Meta-analysis of genome-wide association data and large-scale replication identifies additional susceptibility loci for type 2 diabetes . Nat Genet 40 , 638 – 645 . doi: 10.1038/ng.120 . OpenUrl CrossRef PubMed Web of Science 36. ↵ Zhang , W. , Zhang , L. , Xiao , C. , Wu , X. , Cui , H. , Yang , C. , Yan , P. , Tang , M. , Wang , Y. , Chen , L. , et al. ( 2024 ). Bidirectional relationship between type 2 diabetes mellitus and coronary artery disease: Prospective cohort study and genetic analyses . Chin Med J (Engl ) 137 , 577 – 587 . doi: 10.1097/CM9.0000000000002894 . OpenUrl CrossRef PubMed 37. ↵ Frkic , R.L. , Richter , K. , and Bruning , J.B . ( 2021 ). The therapeutic potential of inhibiting PPARgamma phosphorylation to treat type 2 diabetes . J Biol Chem 297 , 101030 . doi: 10.1016/j.jbc.2021.101030 . OpenUrl CrossRef PubMed 38. Geldenhuys , W.J. , Lin , L. , Darvesh , A.S. , and Sadana , P . ( 2017 ). Emerging strategies of targeting lipoprotein lipase for metabolic and cardiovascular diseases . Drug Discov Today 22 , 352 – 365 . doi: 10.1016/j.drudis.2016.10.007 . OpenUrl CrossRef PubMed 39. ↵ Li , S. , Shin , H.J. , Ding , E.L. , and van Dam , R.M. ( 2009 ). Adiponectin levels and risk of type 2 diabetes: a systematic review and meta-analysis . JAMA 302 , 179 – 188 . doi: 10.1001/jama.2009.976 . OpenUrl CrossRef PubMed Web of Science 40. ↵ Yamauchi , T. , Waki , H. , Kamon , J. , Murakami , K. , Motojima , K. , Komeda , K. , Miki , H. , Kubota , N. , Terauchi , Y. , Tsuchida , A. , et al. ( 2001 ). Inhibition of RXR and PPARgamma ameliorates diet-induced obesity and type 2 diabetes . J Clin Invest 108 , 1001 – 1013 . doi: 10.1172/JCI12864 . OpenUrl CrossRef PubMed Web of Science 41. Miyazaki , S. , Taniguchi , H. , Moritoh , Y. , Tashiro , F. , Yamamoto , T. , Yamato , E. , Ikegami , H. , Ozato , K. , and Miyazaki , J . ( 2010 ). Nuclear hormone retinoid X receptor (RXR) negatively regulates the glucose-stimulated insulin secretion of pancreatic ss-cells . Diabetes 59 , 2854 – 2861 . doi: 10.2337/db09-1897 . OpenUrl Abstract / FREE Full Text 42. Anderson , A. , and Walker , B.R . ( 2013 ). 11beta-HSD1 inhibitors for the treatment of type 2 diabetes and cardiovascular disease . Drugs 73 , 1385 – 1393 . doi: 10.1007/s40265-013-0112-5 . OpenUrl CrossRef PubMed 43. ↵ Alberts , P. , Nilsson , C. , Selen , G. , Engblom , L.O. , Edling , N.H. , Norling , S. , Klingstrom , G. , Larsson , C. , Forsgren , M. , Ashkzari , M. , et al. ( 2003 ). Selective inhibition of 11 beta-hydroxysteroid dehydrogenase type 1 improves hepatic insulin sensitivity in hyperglycemic mice strains . Endocrinology 144 , 4755 – 4762 . doi: 10.1210/en.2003-0344 . OpenUrl CrossRef PubMed Web of Science 44. ↵ Cohen , J.C. , Boerwinkle , E. , Mosley , T.H. , Jr. , and Hobbs , H.H . ( 2006 ). Sequence variations in PCSK9, low LDL, and protection against coronary heart disease . N Engl J Med 354 , 1264 – 1272 . doi: 10.1056/NEJMoa054013 . OpenUrl CrossRef PubMed Web of Science 45. ↵ Sabatine , M.S. , Giugliano , R.P. , Keech , A.C. , Honarpour , N. , Wiviott , S.D. , Murphy , S.A. , Kuder , J.F. , Wang , H. , Liu , T. , Wasserman , S.M. , et al. ( 2017 ). Evolocumab and Clinical Outcomes in Patients with Cardiovascular Disease . N Engl J Med 376 , 1713 – 1722 . doi: 10.1056/NEJMoa1615664 . OpenUrl CrossRef PubMed 46. ↵ Clusan , L. , Ferriere , F. , Flouriot , G. , and Pakdel , F . ( 2023 ). A Basic Review on Estrogen Receptor Signaling Pathways in Breast Cancer . Int J Mol Sci 24 . doi: 10.3390/ijms24076834 . OpenUrl CrossRef PubMed 47. ↵ Wilcox , N. , Dumont , M. , Gonzalez-Neira , A. , Carvalho , S. , Joly Beauparlant , C. , Crotti , M. , Luccarini , C. , Soucy , P. , Dubois , S. , Nunez-Torres , R. , et al. ( 2023 ). Exome sequencing identifies breast cancer susceptibility genes and defines the contribution of coding variants to breast cancer risk . Nat Genet 55 , 1435 – 1439 . doi: 10.1038/s41588-023-01466-z . OpenUrl CrossRef PubMed 48. ↵ Qi , A. , Lamont , L. , Liu , E. , Murray , S.D. , Meng , X. , and Yang , S . ( 2023 ). Essential Protein PHB2 and Its Regulatory Mechanisms in Cancer . Cells 12 . doi: 10.3390/cells12081211 . OpenUrl CrossRef 49. Yang , J. , Li , B. , and He , Q.Y . ( 2018 ). Significance of prohibitin domain family in tumorigenesis and its implication in cancer diagnosis and treatment . Cell Death Dis 9 , 580 . doi: 10.1038/s41419-018-0661-3 . OpenUrl CrossRef PubMed 50. ↵ Liu , M. , Wang , Z. , Li , S. , Deng , Y. , and He , N . ( 2022 ). Identification of PHB2 as a Potential Biomarker of Luminal A Breast Cancer Cells Using a Cell-Specific Aptamer . ACS Appl Mater Interfaces 14 , 51593 – 51601 . doi: 10.1021/acsami.2c12291 . OpenUrl CrossRef PubMed 51. ↵ Hsieh , T.H. , Hsu , C.Y. , Tsai , C.F. , Long , C.Y. , Wu , C.H. , Wu , D.C. , Lee , J.N. , Chang , W.C. , and Tsai , E.M . ( 2015 ). HDAC inhibitors target HDAC5, upregulate microRNA-125a-5p, and induce apoptosis in breast cancer cells . Mol Ther 23 , 656 – 666 . doi: 10.1038/mt.2014.247 . OpenUrl CrossRef PubMed 52. ↵ Oltra , S.S. , Cejalvo , J.M. , Tormo , E. , Albanell , M. , Ferrer , A. , Nacher , M. , Bermejo , B. , Hernando , C. , Chirivella , I. , Alonso , E. , et al. ( 2020 ). HDAC5 Inhibitors as a Potential Treatment in Breast Cancer Aiecting Very Young Women . Cancers (Basel ) 12 . doi: 10.3390/cancers12020412 . OpenUrl CrossRef 53. ↵ Khera , A.V. , and Kathiresan , S . ( 2017 ). Genetics of coronary artery disease: discovery, biology and clinical translation . Nat Rev Genet 18 , 331 – 344 . doi: 10.1038/nrg.2016.160 . OpenUrl CrossRef PubMed 54. ↵ Aragam , K.G. , Jiang , T. , Goel , A. , Kanoni , S. , Wolford , B.N. , Atri , D.S. , Weeks , E.M. , Wang , M. , Hindy , G. , Zhou , W. , et al. ( 2022 ). Discovery and systematic characterization of risk variants and genes for coronary artery disease in over a million participants . Nat Genet 54 , 1803 – 1815 . doi: 10.1038/s41588-022-01233-6 . OpenUrl CrossRef 55. Jeanne , M. , Jorgensen , J. , and Gould , D.B . ( 2015 ). Molecular and Genetic Analyses of Collagen Type IV Mutant Mouse Models of Spontaneous Intracerebral Hemorrhage Identify Mechanisms for Stroke Prevention . Circulation 131 , 1555 – 1565 . doi: 10.1161/CIRCULATIONAHA.114.013395 . OpenUrl Abstract / FREE Full Text 56. Turner , A.W. , Nikpay , M. , Silva , A. , Lau , P. , Martinuk , A. , Linseman , T.A. , Soubeyrand , S. , and McPherson , R . ( 2015 ). Functional interaction between COL4A1/COL4A2 and SMAD3 risk loci for coronary artery disease . Atherosclerosis 242 , 543 – 552 . doi: 10.1016/j.atherosclerosis.2015.08.008 . OpenUrl CrossRef PubMed 57. ↵ Zhang , L. , Zhou , J. , and Kong , W . ( 2025 ). Extracellular matrix in vascular homeostasis and disease . Nat Rev Cardiol . doi: 10.1038/s41569-024-01103-0 . OpenUrl CrossRef PubMed 58. ↵ Hertle , E. , Stehouwer , C.D. , and van Greevenbroek , M.M. ( 2014 ). The complement system in human cardiometabolic disease . Mol Immunol 61 , 135 – 148 . doi: 10.1016/j.molimm.2014.06.031 . OpenUrl CrossRef PubMed 59. Onat , A. , Can , G. , Rezvani , R. , and Cianflone , K. ( 2011 ). Complement C3 and cleavage products in cardiometabolic risk . Clin Chim Acta 412 , 1171 – 1179 . doi: 10.1016/j.cca.2011.03.005 . OpenUrl CrossRef PubMed 60. ↵ Kiss , M.G. , and Binder , C.J . ( 2022 ). The multifaceted impact of complement on atherosclerosis . Atherosclerosis 351 , 29 – 40 . doi: 10.1016/j.atherosclerosis.2022.03.014 . OpenUrl CrossRef PubMed 61. ↵ Tardii , D.F. , Jui , N.T. , Khurana , V. , Tambe , M.A. , Thompson , M.L. , Chung , C.Y. , Kamadurai , H.B. , Kim , H.T. , Lancaster , A.K. , Caldwell , K.A. , et al. ( 2013 ). Yeast reveal a “druggable” Rsp5/Nedd4 network that ameliorates alpha-synuclein toxicity in neurons . Science 342 , 979 – 983 . doi: 10.1126/science.1245321 . OpenUrl Abstract / FREE Full Text 62. ↵ Chung , C.Y. , Khurana , V. , Auluck , P.K. , Tardii , D.F. , Mazzulli , J.R. , Soldner , F. , Baru , V. , Lou , Y. , Freyzon , Y. , Cho , S. , et al. ( 2013 ). Identification and rescue of alpha-synuclein toxicity in Parkinson patient-derived neurons . Science 342 , 983 – 987 . doi: 10.1126/science.1245296 . OpenUrl Abstract / FREE Full Text 63. ↵ Chung , C.Y. , Khurana , V. , Yi , S. , Sahni , N. , Loh , K.H. , Auluck , P.K. , Baru , V. , Udeshi , N.D. , Freyzon , Y. , Carr , S.A. , et al. ( 2017 ). In Situ Peroxidase Labeling and Mass-Spectrometry Connects Alpha-Synuclein Directly to Endocytic Traiicking and mRNA Metabolism in Neurons . Cell Syst 4 , 242-250 e244. doi: 10.1016/j.cels.2017.01.002 . OpenUrl CrossRef PubMed 64. ↵ Nalls , M.A. , Blauwendraat , C. , Vallerga , C.L. , Heilbron , K. , Bandres-Ciga , S. , Chang , D. , Tan , M. , Kia , D.A. , Noyce , A.J. , Xue , A. , et al. ( 2019 ). Identification of novel risk loci, causal insights, and heritable risk for Parkinson’s disease: a meta-analysis of genome-wide association studies . Lancet Neurol 18 , 1091 – 1102 . doi: 10.1016/S1474-4422(19)30320-5 . OpenUrl CrossRef PubMed 65. ↵ Kim , J.J. , Vitale , D. , Otani , D.V. , Lian , M.M. , Heilbron , K. , and Me Research , T. , Iwaki , H. , Lake , J. , Solsberg , C.W. , Leonard , H. , et al. ( 2024 ). Multi-ancestry genome-wide association meta-analysis of Parkinson’s disease . Nat Genet 56 , 27 – 36 . doi: 10.1038/s41588-023-01584-8 . OpenUrl CrossRef 66. ↵ Polymeropoulos , M.H. , Lavedan , C. , Leroy , E. , Ide , S.E. , Dehejia , A. , Dutra , A. , Pike , B. , Root , H. , Rubenstein , J. , Boyer , R. , et al. ( 1997 ). Mutation in the alpha-synuclein gene identified in families with Parkinson’s disease . Science 276 , 2045 – 2047 . doi: 10.1126/science.276.5321.2045 . OpenUrl Abstract / FREE Full Text 67. Nichols , W.C. , Pankratz , N. , Hernandez , D. , Paisan-Ruiz , C. , Jain , S. , Halter , C.A. , Michaels , V.E. , Reed , T. , Rudolph , A. , Shults , C.W. , et al. ( 2005 ). Genetic screening for a single common LRRK2 mutation in familial Parkinson’s disease . Lancet 365 , 410 – 412 . doi: 10.1016/S0140-6736(05)17828-3 . OpenUrl CrossRef PubMed Web of Science 68. ↵ Funayama , M. , Nishioka , K. , Li , Y. , and Hattori , N . ( 2023 ). Molecular genetics of Parkinson’s disease: Contributions and global trends . J Hum Genet 68 , 125 – 130 . doi: 10.1038/s10038-022-01058-5 . OpenUrl CrossRef PubMed 69. ↵ Towns , C. , Richer , M. , Jasaityte , S. , Staiord , E.J. , Joubert , J. , Antar , T. , Martinez-Carrasco , A. , Makarious , M.B. , Casey , B. , Vitale , D. , et al. ( 2023 ). Defining the causes of sporadic Parkinson’s disease in the global Parkinson’s genetics program (GP2) . NPJ Parkinsons Dis 9 , 131 . doi: 10.1038/s41531-023-00533-w . OpenUrl CrossRef PubMed 70. ↵ Bandres-Ciga , S. , Saez-Atienzar , S. , Kim , J.J. , Makarious , M.B. , Faghri , F. , Diez-Fairen , M. , Iwaki , H. , Leonard , H. , Botia , J. , Ryten , M. , et al. ( 2020 ). Large-scale pathway specific polygenic risk and transcriptomic community network analysis identifies novel functional pathways in Parkinson disease . Acta Neuropathol 140 , 341 – 358 . doi: 10.1007/s00401-020-02181-3 . OpenUrl CrossRef PubMed 71. ↵ Foo , J.N. , Chew , E.G.Y. , Chung , S.J. , Peng , R. , Blauwendraat , C. , Nalls , M.A. , Mok , K.Y. , Satake , W. , Toda , T. , Chao , Y. , et al. ( 2020 ). Identification of Risk Loci for Parkinson Disease in Asians and Comparison of Risk Between Asians and Europeans: A Genome-Wide Association Study . JAMA Neurol 77 , 746 – 754 . doi: 10.1001/jamaneurol.2020.0428 . OpenUrl CrossRef 72. ↵ Maiti , P. , Manna , J. , and Dunbar , G.L . ( 2017 ). Current understanding of the molecular mechanisms in Parkinson’s disease: Targets for potential treatments . Transl Neurodegener 6 , 28 . doi: 10.1186/s40035-017-0099-z . OpenUrl CrossRef 73. ↵ Braak , H. , Del Tredici , K. , Rub , U. , de Vos , R.A. , Jansen Steur , E.N. , and Braak , E. ( 2003 ). Staging of brain pathology related to sporadic Parkinson’s disease . Neurobiol Aging 24 , 197 – 211 . doi: 10.1016/s0197-4580(02)00065-9 . OpenUrl CrossRef PubMed Web of Science 74. ↵ Tang , D. , Kang , R. , Livesey , K.M. , Cheh , C.W. , Farkas , A. , Loughran , P. , Hoppe , G. , Bianchi , M.E. , Tracey , K.J. , Zeh , H.J ., 3rd, and Lotze , M.T. ( 2010 ). Endogenous HMGB1 regulates autophagy . J Cell Biol 190 , 881 – 892 . doi: 10.1083/jcb.200911078 . OpenUrl Abstract / FREE Full Text 75. ↵ Wong , Y.C. , and Holzbaur , E.L . ( 2015 ). Temporal dynamics of PARK2/parkin and OPTN/optineurin recruitment during the mitophagy of damaged mitochondria . Autophagy 11 , 422 – 424 . doi: 10.1080/15548627.2015.1009792 . OpenUrl CrossRef PubMed 76. ↵ Takahashi , M. , Kitaura , H. , Kakita , A. , Kakihana , T. , Katsuragi , Y. , Nameta , M. , Zhang , L. , Iwakura , Y. , Nawa , H. , Higuchi , M. , et al. ( 2018 ). USP10 Is a Driver of Ubiquitinated Protein Aggregation and Aggresome Formation to Inhibit Apoptosis . iScience 9 , 433 – 450 . doi: 10.1016/j.isci.2018.11.006 . OpenUrl CrossRef PubMed 77. ↵ Ham , S.J. , Lee , D. , Yoo , H. , Jun , K. , Shin , H. , and Chung , J . ( 2020 ). Decision between mitophagy and apoptosis by Parkin via VDAC1 ubiquitination . Proc Natl Acad Sci U S A 117 , 4281 – 4291 . doi: 10.1073/pnas.1909814117 . OpenUrl Abstract / FREE Full Text 78. Chu , Y. , Goldman , J.G. , Kelly , L. , He , Y. , Waliczek , T. , and Kordower , J.H . ( 2014 ). Abnormal alpha-synuclein reduces nigral voltage-dependent anion channel 1 in sporadic and experimental Parkinson’s disease . Neurobiol Dis 69 , 1 – 14 . doi: 10.1016/j.nbd.2014.05.003 . OpenUrl CrossRef PubMed 79. He , Y. , Wang , W. , Yang , T. , Thomas , E.R. , Dai , R. , and Li , X . ( 2022 ). The Potential Role of Voltage-Dependent Anion Channel in the Treatment of Parkinson’s Disease . Oxid Med Cell Longev 2022 , 4665530 . doi: 10.1155/2022/4665530 . OpenUrl CrossRef PubMed 80. Tofaris , G.K. , Kim , H.T. , Hourez , R. , Jung , J.W. , Kim , K.P. , and Goldberg , A.L . ( 2011 ). Ubiquitin ligase Nedd4 promotes alpha-synuclein degradation by the endosomal-lysosomal pathway . Proc Natl Acad Sci U S A 108 , 17004 – 17009 . doi: 10.1073/pnas.1109356108 . OpenUrl Abstract / FREE Full Text 81. ↵ McLean , P.J. , Kawamata , H. , Sharii , S. , Hewett , J. , Sharma , N. , Ueda , K. , Breakefield , X.O. , and Hyman , B.T. ( 2002 ). TorsinA and heat shock proteins act as molecular chaperones: suppression of alpha-synuclein aggregation . J Neurochem 83 , 846 – 854 . doi: 10.1046/j.1471-4159.2002.01190.x . OpenUrl CrossRef PubMed Web of Science 82. ↵ Fu , X. , Qu , L. , Xu , H. , and Xie , J . ( 2024 ). Ndfip1 protected dopaminergic neurons via regulating mitochondrial function and ferroptosis in Parkinson’s disease . Exp Neurol 375 , 114724 . doi: 10.1016/j.expneurol.2024.114724 . OpenUrl CrossRef PubMed 83. Howitt , J. , Gysbers , A.M. , Ayton , S. , Carew-Jones , F. , Putz , U. , Finkelstein , D.I. , Halliday , G.M. , and Tan , S.S . ( 2014 ). Increased Ndfip1 in the substantia nigra of Parkinsonian brains is associated with elevated iron levels . PLoS One 9 , e87119 . doi: 10.1371/journal.pone.0087119 . OpenUrl CrossRef PubMed 84. Villaescusa , J.C. , Li , B. , Toledo , E.M. , Rivetti di Val Cervo , P. , Yang , S. , Stott , S.R. , Kaiser , K. , Islam , S. , Gyllborg , D. , Laguna-Goya , R. , et al. ( 2016 ). A PBX1 transcriptional network controls dopaminergic neuron development and is impaired in Parkinson’s disease . EMBO J 35 , 1963 – 1978 . doi: 10.15252/embj.201593725 . OpenUrl Abstract / FREE Full Text 85. ↵ Pranski , E. , Van Sanford , C.D. , Dalal , N. , Orr , A.L. , Karmali , D. , Cooper , D.S. , Gearing , M. , Lah , J.J. , Levey , A.I. , and Betarbet , R. ( 2013 ). NF-kappaB activity is inversely correlated to RNF11 expression in Parkinson’s disease . Neurosci Lett 547 , 16 – 20 . doi: 10.1016/j.neulet.2013.04.056 . OpenUrl CrossRef PubMed 86. ↵ Tian , R. , Gachechiladze , M.A. , Ludwig , C.H. , Laurie , M.T. , Hong , J.Y. , Nathaniel , D. , Prabhu , A.V. , Fernandopulle , M.S. , Patel , R. , Abshari , M. , et al. ( 2019 ). CRISPR Interference-Based Platform for Multimodal Genetic Screens in Human iPSC-Derived Neurons . Neuron 104 , 239 – 255 e212. doi: 10.1016/j.neuron.2019.07.014 . OpenUrl CrossRef PubMed 87. ↵ Lam , I. , Ndayisaba , A. , Lewis , A.J. , Fu , Y. , Sagredo , G.T. , Zaccagnini , L. , Sandoe , J. , Sanz , R.L. , Vahdatshoar , A. , Martin , T.D. , et al. ( 2022 ). Rapid iPSC inclusionopathy models shed light on formation, consequence and molecular subtype of α-synuclein inclusions . bioRxiv , 2022.2011.2008.515615. doi: 10.1101/2022.11.08.515615 . OpenUrl Abstract / FREE Full Text 88. ↵ Grattan , D.R . ( 2024 ). Does the brain make prolactin? J Neuroendocrinol 36 , e13432 . doi: 10.1111/jne.13432 . OpenUrl CrossRef PubMed 89. ↵ Featherstone , K. , White , M.R. , and Davis , J.R . ( 2012 ). The prolactin gene: a paradigm of tissue-specific gene regulation with complex temporal transcription dynamics . J Neuroendocrinol 24 , 977 – 990 . doi: 10.1111/j.1365-2826.2012.02310.x . OpenUrl CrossRef PubMed Web of Science 90. ↵ Lam , I. , Ndayisaba , A. , Lewis , A.J. , Fu , Y. , Sagredo , G.T. , Kuzkina , A. , Zaccagnini , L. , Celikag , M. , Sandoe , J. , Sanz , R.L. , et al. ( 2024 ). Rapid iPSC inclusionopathy models shed light on formation, consequence, and molecular subtype of alpha-synuclein inclusions . Neuron 112 , 2886 – 2909 e2816. doi: 10.1016/j.neuron.2024.06.002 . OpenUrl CrossRef 91. ↵ Luk , K.C. , Covell , D.J. , Kehm , V.M. , Zhang , B. , Song , I.Y. , Byrne , M.D. , Pitkin , R.M. , Decker , S.C. , Trojanowski , J.Q. , and Lee , V.M . ( 2016 ). Molecular and Biological Compatibility with Host Alpha-Synuclein Influences Fibril Pathogenicity . Cell Rep 16 , 3373 – 3387 . doi: 10.1016/j.celrep.2016.08.053 . OpenUrl CrossRef PubMed 92. ↵ Loor , G. , Kondapalli , J. , Schriewer , J.M. , Chandel , N.S. , Vanden Hoek , T.L. , and Schumacker , P.T . ( 2010 ). Menadione triggers cell death through ROS-dependent mechanisms involving PARP activation without requiring apoptosis . Free Radic Biol Med 49 , 1925 – 1936 . doi: 10.1016/j.freeradbiomed.2010.09.021 . OpenUrl CrossRef PubMed 93. ↵ Di Nubila , A. , Dilella , G. , Simone , R. , and Barbieri , S.S. ( 2024 ). Vascular Extracellular Matrix in Atherosclerosis . Int J Mol Sci 25 . doi: 10.3390/ijms252212017 . OpenUrl CrossRef 94. ↵ Manon-Jensen , T. , Kjeld , N.G. , and Karsdal , M.A . ( 2016 ). Collagen-mediated hemostasis . J Thromb Haemost 14 , 438 – 448 . doi: 10.1111/jth.13249 . OpenUrl CrossRef PubMed 95. ↵ Libby , P . ( 2021 ). Inflammation during the life cycle of the atherosclerotic plaque . Cardiovasc Res 117 , 2525 – 2536 . doi: 10.1093/cvr/cvab303 . OpenUrl CrossRef PubMed 96. ↵ Sakata , T. , Sakaguchi , H. , Tsuda , L. , Higashitani , A. , Aigaki , T. , Matsuno , K. , and Hayashi , S . ( 2004 ). Drosophila Nedd4 regulates endocytosis of notch and suppresses its ligand-independent activation . Curr Biol 14 , 2228 – 2236 . doi: 10.1016/j.cub.2004.12.028 . OpenUrl CrossRef PubMed Web of Science 97. Privman Champaloux , E. , Donelson , N. , Pyakurel , P. , Wolin , D. , Ostendorf , L. , Denno , M. , Borman , R. , Burke , C. , Short-Miller , J.C. , Yoder , M.R. , et al. ( 2021 ). Ring Finger Protein 11 (RNF11) Modulates Dopamine Release in Drosophila . Neuroscience 452 , 37 – 48 . doi: 10.1016/j.neuroscience.2020.10.021 . OpenUrl CrossRef PubMed 98. ↵ Wakabayashi-Ito , N. , Doherty , O.M. , Moriyama , H. , Breakefield , X.O. , Gusella , J.F. , O’Donnell , J.M. , and Ito , N. ( 2011 ). Dtorsin, the Drosophila ortholog of the early-onset dystonia TOR1A (DYT1), plays a novel role in dopamine metabolism . PLoS One 6 , e26183 . doi: 10.1371/journal.pone.0026183 . OpenUrl CrossRef PubMed 99. ↵ Barallobre , M.J. , Perier , C. , Bove , J. , Laguna , A. , Delabar , J.M. , Vila , M. , and Arbones , M.L . ( 2014 ). DYRK1A promotes dopaminergic neuron survival in the developing brain and in a mouse model of Parkinson’s disease . Cell Death Dis 5 , e1289 . doi: 10.1038/cddis.2014.253 . OpenUrl CrossRef PubMed 100. ↵ Yong , Y. , Wu , Q. , Meng , X. , Lu , R. , Xia , H. , Pei , F. , and Yang , X . ( 2023 ). Dyrk1a Phosphorylation of alpha-Synuclein Mediating Apoptosis of Dopaminergic Neurons in Parkinson’s Disease . Parkinsons Dis 2023 , 8848642 . doi: 10.1155/2023/8848642 . OpenUrl CrossRef PubMed 101. ↵ Ye , H. , Robak , L.A. , Yu , M. , Cykowski , M. , and Shulman , J.M . ( 2023 ). Genetics and Pathogenesis of Parkinson’s Syndrome . Annu Rev Pathol 18 , 95 – 121 . doi: 10.1146/annurev-pathmechdis-031521-034145 . OpenUrl CrossRef PubMed 102. ↵ Costa-Brito , A.R. , Goncalves , I. , and Santos , C.R.A . ( 2022 ). The brain as a source and a target of prolactin in mammals . Neural Regen Res 17 , 1695 – 1702 . doi: 10.4103/1673-5374.332124 . OpenUrl CrossRef PubMed 103. ↵ Ramos-Martinez , E. , Ramos-Martinez , I. , Molina-Salinas , G. , Zepeda-Ruiz , W.A. , and Cerbon , M . ( 2021 ). The role of prolactin in central nervous system inflammation . Rev Neurosci 32 , 323 – 340 . doi: 10.1515/revneuro-2020-0082 . OpenUrl CrossRef PubMed 104. Molina-Salinas , G. , Rodriguez-Chavez , V. , Langley , E. , and Cerbon , M . ( 2023 ). Prolactin-induced neuroprotection against excitotoxicity is mediated via PI3K/AKT and GSK3beta/NF-kappaB in primary cultures of hippocampal neurons . Peptides 166 , 171037 . doi: 10.1016/j.peptides.2023.171037 . OpenUrl CrossRef PubMed 105. ↵ Macias , F. , Ulloa , M. , Clapp , C. , Martinez de la Escalera , G. , and Arnold , E. ( 2024 ). Prolactin protects hippocampal neurons against H2O2-induced neurotoxicity by suppressing BAX and NOX4 via the NF-kappaB signaling pathway . PLoS One 19 , e0313328 . doi: 10.1371/journal.pone.0313328 . OpenUrl CrossRef PubMed 106. ↵ Karayel , O. , Virreira Winter , S. , Padmanabhan , S. , Kuras , Y.I. , Vu , D.T. , Tuncali , I. , Merchant , K. , Wills , A.M. , Scherzer , C.R. , and Mann , M . ( 2022 ). Proteome profiling of cerebrospinal fluid reveals biomarker candidates for Parkinson’s disease . Cell Rep Med 3 , 100661 . doi: 10.1016/j.xcrm.2022.100661 . OpenUrl CrossRef PubMed 107. ↵ Li , B. , and Leal , S.M . ( 2008 ). Methods for detecting associations with rare variants for common diseases: application to analysis of sequence data . Am J Hum Genet 83 , 311 – 321 . doi: 10.1016/j.ajhg.2008.06.024 . OpenUrl CrossRef PubMed Web of Science 108. ↵ Derkach , A. , Lawless , J.F. , and Sun , L . ( 2013 ). Robust and powerful tests for rare variants using Fisher’s method to combine evidence of association from two or more complementary tests . Genet Epidemiol 37 , 110 – 121 . doi: 10.1002/gepi.21689 . OpenUrl CrossRef PubMed 109. ↵ Nyholt , D.R . ( 2004 ). A simple correction for multiple testing for single-nucleotide polymorphisms in linkage disequilibrium with each other . Am J Hum Genet 74 , 765 – 769 . doi: 10.1086/383251 . OpenUrl CrossRef PubMed Web of Science 110. ↵ Higham , N.J . ( 2002 ). Computing the nearest correlation matrix—a problem from finance . IMA journal of Numerical Analysis 22 , 329 – 343 . OpenUrl CrossRef 111. ↵ Szklarczyk , D. , Kirsch , R. , Koutrouli , M. , Nastou , K. , Mehryary , F. , Hachilif , R. , Gable , A.L. , Fang , T. , Doncheva , N.T. , Pyysalo , S. , et al. ( 2023 ). The STRING database in 2023: protein-protein association networks and functional enrichment analyses for any sequenced genome of interest . Nucleic Acids Res 51 , D638 – D646 . doi: 10.1093/nar/gkac1000 . OpenUrl CrossRef PubMed 112. ↵ Li , T. , Wernersson , R. , Hansen , R.B. , Horn , H. , Mercer , J. , Slodkowicz , G. , Workman , C.T. , Rigina , O. , Rapacki , K. , Staerfeldt , H.H. , et al. ( 2017 ). A scored human protein-protein interaction network to catalyze genomic interpretation . Nat Methods 14 , 61 – 64 . doi: 10.1038/nmeth.4083 . OpenUrl CrossRef PubMed 113. ↵ Megchelenbrink , W. , Katzir , R. , Lu , X. , Ruppin , E. , and Notebaart , R.A . ( 2015 ). Synthetic dosage lethality in the human metabolic network is highly predictive of tumor growth and cancer patient survival . Proc Natl Acad Sci U S A 112 , 12217 – 12222 . doi: 10.1073/pnas.1508573112 . OpenUrl Abstract / FREE Full Text 114. ↵ Consortium , G.T . ( 2020 ). The GTEx Consortium atlas of genetic regulatory eiects across human tissues . Science 369 , 1318 – 1330 . doi: 10.1126/science.aaz1776 . OpenUrl Abstract / FREE Full Text 115. ↵ Meyers , R.M. , Bryan , J.G. , McFarland , J.M. , Weir , B.A. , Sizemore , A.E. , Xu , H. , Dharia , N.V. , Montgomery , P.G. , Cowley , G.S. , Pantel , S. , et al. ( 2017 ). Computational correction of copy number eiect improves specificity of CRISPR-Cas9 essentiality screens in cancer cells . Nat Genet 49 , 1779 – 1784 . doi: 10.1038/ng.3984 . OpenUrl CrossRef PubMed 116. ↵ Dempster , J.M. , Rossen , J. , Kazachkova , M. , Pan , J. , Kugener , G. , Root , D.E. , and Tsherniak , A . ( 2019 ). Extracting Biological Insights from the Project Achilles Genome-Scale CRISPR Screens in Cancer Cell Lines . bioRxiv , 720243 . doi: 10.1101/720243 . OpenUrl Abstract / FREE Full Text 117. ↵ Ioannidis , N.M. , Rothstein , J.H. , Pejaver , V. , Middha , S. , McDonnell , S.K. , Baheti , S. , Musolf , A. , Li , Q. , Holzinger , E. , Karyadi , D. , et al. ( 2016 ). REVEL: An Ensemble Method for Predicting the Pathogenicity of Rare Missense Variants . Am J Hum Genet 99 , 877 – 885 . doi: 10.1016/j.ajhg.2016.08.016 . OpenUrl CrossRef PubMed 118. ↵ Cheng , J. , Novati , G. , Pan , J. , Bycroft , C. , Zemgulyte , A. , Applebaum , T. , Pritzel , A. , Wong , L.H. , Zielinski , M. , Sargeant , T. , et al. ( 2023 ). Accurate proteome-wide missense variant eiect prediction with AlphaMissense . Science 381 , eadg7492. doi: 10.1126/science.adg7492 . OpenUrl CrossRef PubMed 119. ↵ All of Us Research Program Genomics, I. ( 2024 ). Genomic data in the All of Us Research Program . Nature 627 , 340 – 346 . doi: 10.1038/s41586-023-06957-x . OpenUrl CrossRef PubMed 120. ↵ Kim , T.W. , Piao , J. , Bocchi , V.D. , Koo , S.Y. , Choi , S.J. , Chaudhry , F. , Yang , D. , Cho , H.S. , Hergenreder , E. , Perera , L.R. , et al. ( 2025 ). Enhanced yield and subtype identity of hPSC-derived midbrain dopamine neuron by modulation of WNT and FGF18 signaling . bioRxiv . doi: 10.1101/2025.01.06.631400 . OpenUrl Abstract / FREE Full Text 121. ↵ DeWeirdt , P.C. , Sangree , A.K. , Hanna , R.E. , Sanson , K.R. , Hegde , M. , Strand , C. , Persky , N.S. , and Doench , J.G . ( 2020 ). Genetic screens in isogenic mammalian cell lines without single cell cloning . Nat Commun 11 , 752 . doi: 10.1038/s41467-020-14620-6 . OpenUrl CrossRef PubMed 122. ↵ Kim , T.W. , Piao , J. , Koo , S.Y. , Kriks , S. , Chung , S.Y. , Betel , D. , Socci , N.D. , Choi , S.J. , Zabierowski , S. , Dubose , B.N. , et al. ( 2021 ). Biphasic Activation of WNT Signaling Facilitates the Derivation of Midbrain Dopamine Neurons from hESCs for Translational Use . Cell Stem Cell 28 , 343 – 355 e345. doi: 10.1016/j.stem.2021.01.005 . OpenUrl CrossRef PubMed 123. ↵ Gonzalez , F. , Zhu , Z. , Shi , Z.D. , Lelli , K. , Verma , N. , Li , Q.V. , and Huangfu , D . ( 2014 ). An iCRISPR platform for rapid, multiplexable, and inducible genome editing in human pluripotent stem cells . Cell Stem Cell 15 , 215 – 226 . doi: 10.1016/j.stem.2014.05.018 . OpenUrl CrossRef PubMed 124. ↵ Wang , B. , Wang , M. , Zhang , W. , Xiao , T. , Chen , C.H. , Wu , A. , Wu , F. , Traugh , N. , Wang , X. , Li , Z. , et al. ( 2019 ). Integrative analysis of pooled CRISPR genetic screens using MAGeCKFlute . Nat Protoc 14 , 756 – 780 . doi: 10.1038/s41596-018-0113-7 . OpenUrl CrossRef PubMed 125. ↵ Supek , F. , Bosnjak , M. , Skunca , N. , and Smuc , T . ( 2011 ). REVIGO summarizes and visualizes long lists of gene ontology terms . PLoS One 6 , e21800 . doi: 10.1371/journal.pone.0021800 . OpenUrl CrossRef PubMed 126. ↵ de Leeuw , C.A. , Mooij , J.M. , Heskes , T. , and Posthuma , D . ( 2015 ). MAGMA: generalized gene-set analysis of GWAS data . PLoS Comput Biol 11 , e1004219. doi: 10.1371/journal.pcbi.1004219 . OpenUrl CrossRef PubMed 127. ↵ Gilbert , L.A. , Horlbeck , M.A. , Adamson , B. , Villalta , J.E. , Chen , Y. , Whitehead , E.H. , Guimaraes , C. , Panning , B. , Ploegh , H.L. , Bassik , M.C. , et al. ( 2014 ). Genome-Scale CRISPR-Mediated Control of Gene Repression and Activation . Cell 159 , 647 – 661 . doi: 10.1016/j.cell.2014.09.029 . OpenUrl CrossRef PubMed Web of Science 128. ↵ Kampmann , M. , Bassik , M.C. , and Weissman , J.S . ( 2014 ). Functional genomics platform for pooled screening and generation of mammalian genetic interaction maps . Nat Protoc 9 , 1825 – 1847 . doi: 10.1038/nprot.2014.103 . OpenUrl CrossRef PubMed 129. ↵ Schneider , C.A. , Rasband , W.S. , and Eliceiri , K.W . ( 2012 ). NIH Image to ImageJ: 25 years of image analysis . Nat Methods 9 , 671 – 675 . doi: 10.1038/nmeth.2089 . OpenUrl CrossRef PubMed Web of Science 130. ↵ Volpicelli-Daley , L.A. , Luk , K.C. , and Lee , V.M . ( 2014 ). Addition of exogenous alpha-synuclein preformed fibrils to primary neuronal cultures to seed recruitment of endogenous alpha-synuclein to Lewy body and Lewy neurite-like aggregates . Nat Protoc 9 , 2135 – 2146 . doi: 10.1038/nprot.2014.143 . OpenUrl CrossRef PubMed 131. ↵ Liberzon , A. , Birger , C. , Thorvaldsdottir , H. , Ghandi , M. , Mesirov , J.P. , and Tamayo , P . ( 2015 ). The Molecular Signatures Database (MSigDB) hallmark gene set collection . Cell Syst 1 , 417 – 425 . doi: 10.1016/j.cels.2015.12.004 . OpenUrl CrossRef PubMed 132. ↵ Szklarczyk , D. , Gable , A.L. , Lyon , D. , Junge , A. , Wyder , S. , Huerta-Cepas , J. , Simonovic , M. , Doncheva , N.T. , Morris , J.H. , Bork , P. , et al. ( 2019 ). STRING v11: protein-protein association networks with increased coverage, supporting functional discovery in genome-wide experimental datasets . Nucleic Acids Res 47 , D607 – D613 . doi: 10.1093/nar/gky1131 . OpenUrl CrossRef PubMed 133. ↵ Backman , J.D. , Li , A.H. , Marcketta , A. , Sun , D. , Mbatchou , J. , Kessler , M.D. , Benner , C. , Liu , D. , Locke , A.E. , Balasubramanian , S. , et al. ( 2021 ). Exome sequencing and analysis of 454,787 UK Biobank participants . Nature 599 , 628 – 634 . doi: 10.1038/s41586-021-04103-z . OpenUrl CrossRef 134. ↵ Boutin , N.T. , Schecter , S.B. , Perez , E.F. , Tchamitchian , N.S. , Cerretani , X.R. , Gainer , V.S. , Lebo , M.S. , Mahanta , L.M. , Karlson , E.W. , and Smoller , J.W . ( 2022 ). The Evolution of a Large Biobank at Mass General Brigham . J Pers Med 12 . doi: 10.3390/jpm12081323 . OpenUrl CrossRef PubMed 135. ↵ Iwaki , H. , Leonard , H.L. , Makarious , M.B. , Bookman , M. , Landin , B. , Vismer , D. , Casey , B. , Gibbs , J.R. , Hernandez , D.G. , Blauwendraat , C. , et al. ( 2021 ). Accelerating Medicines Partnership: Parkinson’s Disease. Genetic Resource . Mov Disord 36 , 1795 – 1804 . doi: 10.1002/mds.28549 . OpenUrl CrossRef PubMed 136. ↵ Li , W. , Xu , H. , Xiao , T. , Cong , L. , Love , M.I. , Zhang , F. , Irizarry , R.A. , Liu , J.S. , Brown , M. , and Liu , X.S . ( 2014 ). MAGeCK enables robust identification of essential genes from genome-scale CRISPR/Cas9 knockout screens . Genome Biol 15 , 554 . doi: 10.1186/s13059-014-0554-4 . OpenUrl CrossRef PubMed View the discussion thread. Back to top Previous Next Posted May 04, 2026. Download PDF Supplementary Material Email Thank you for your interest in spreading the word about bioRxiv. NOTE: Your email address is requested solely to identify you as the sender of this article. Your Email * Your Name * Send To * Enter multiple addresses on separate lines or separate them with commas. You are going to email the following NERINE reveals rare variant associations in gene networks across phenotypes and implicates an SNCA-PRL-LRRK2 subnetwork in Parkinson’s disease Message Subject (Your Name) has forwarded a page to you from bioRxiv Message Body (Your Name) thought you would like to see this page from the bioRxiv website. Your Personal Message CAPTCHA This question is for testing whether or not you are a human visitor and to prevent automated spam submissions. Share NERINE reveals rare variant associations in gene networks across phenotypes and implicates an SNCA-PRL-LRRK2 subnetwork in Parkinson’s disease Sumaiya Nazeen , Xinyuan Wang , Autumn R. Morrow , Ronya Strom , Elizabeth Ethier , Dylan Ritter , Alexander B. H. Henderson , Jalwa Afroz , Christopher S. Cassa , Nathan O. Stitziel , Rajat M. Gupta , Kelvin C. Luk , Lorenz Studer , Vikram Khurana , Shamil R. Sunyaev bioRxiv 2025.01.07.631688; doi: https://doi.org/10.1101/2025.01.07.631688 Share This Article: Copy Citation Tools NERINE reveals rare variant associations in gene networks across phenotypes and implicates an SNCA-PRL-LRRK2 subnetwork in Parkinson’s disease Sumaiya Nazeen , Xinyuan Wang , Autumn R. Morrow , Ronya Strom , Elizabeth Ethier , Dylan Ritter , Alexander B. H. Henderson , Jalwa Afroz , Christopher S. Cassa , Nathan O. Stitziel , Rajat M. Gupta , Kelvin C. Luk , Lorenz Studer , Vikram Khurana , Shamil R. Sunyaev bioRxiv 2025.01.07.631688; doi: https://doi.org/10.1101/2025.01.07.631688 Citation Manager Formats BibTeX Bookends EasyBib EndNote (tagged) EndNote 8 (xml) Medlars Mendeley Papers RefWorks Tagged Ref Manager RIS Zotero Tweet Widget Facebook Like Google Plus One Subject Area Genetics Subject Areas All Articles Animal Behavior and Cognition (7624) Biochemistry (17651) Bioengineering (13871) Bioinformatics (41884) Biophysics (21424) Cancer Biology (18566) Cell Biology (25463) Clinical Trials (138) Developmental Biology (13365) Ecology (19867) Epidemiology (2067) Evolutionary Biology (24290) Genetics (15590) Genomics (22477) Immunology (17714) Microbiology (40331) Molecular Biology (17148) Neuroscience (88487) Paleontology (666) Pathology (2828) Pharmacology and Toxicology (4817) Physiology (7635) Plant Biology (15114) Scientific Communication and Education (2044) Synthetic Biology (4286) Systems Biology (9815) Zoology (2268)