NILUS LAB TECHNICAL REPORT
Internal Preclinical Target Discovery & Audit Report — Research Use Only (Not Peer Reviewed)
Date: October 2026
London & Swansea, United Kingdom
Official Discovery Console: https://www.niluslab.com/profile.html  |  Evidence Registry  |  Target Catalog

Multi-Cohort Human Cardiac Ageing and Heart Failure Target Discovery Across 6 Single-Cell Atlases, 3 Bulk RNA-Seq Cohorts, 3 Cardiac Proteomes, and Human Genetics

Alaa Aldeen1,*, Nilus Lab Computational Discovery Group1
1Nilus Lab LTD (UK Company No. 17427807), London & Swansea, United Kingdom
*Corresponding Author: info@niluslab.com
Abstract

Single-cohort transcriptomic screens and cardiomyocyte-only ageing clocks frequently fail in translational heart failure discovery for four structural reasons: (i) non-myocyte stromal, vascular, and immune compartments drive myocardial fibrosis and capillary rarefaction; (ii) mRNA levels decouple from protein abundance in aged post-mitotic myocardium due to impaired ubiquitin-proteasome and autophagic turnover; (iii) disease-associated gene expression conflates maladaptive drivers with compensatory protective stress responses; and (iv) candidate lists are rarely audited against knockout phenotype directionality, cross-tissue specificity, or human genetic colocalisation. Here we report an unblinded multi-dataset discovery and independent causal audit pipeline executed across 12 public datasets (11 human, 1 mouse): 6 human single-cell/single-nucleus RNA-seq atlases (Reichart 2022, PERIHEART 2026, CAREBANK 2025, Kanemaru 2023, Koenig 2022, and Litviňuková 2020), 3 bulk RNA-seq cohorts (MAGNet 166 DCM explants and 166 non-failing donor LV, GTEx v8/v10 LV n = 431, and GTEx AA n = 429), 3 mass-spectrometry cardiac proteomes (Malecki 2025 human LV: 12 donors aged 14–25 vs 13 aged 51–65, 25 hearts, 3,431 proteins tested; GTEx human heart TMT proteomics: 7 LV + 11 AA donors; and Takasugi 2024 mouse heart proteomics, 30 vs 6 months), and Open Targets v4 human genetics. Without any manual gene steering, unblinded decomposition of dilated cardiomyopathy (DCM) recovered positive control ATP2A2 (SERCA2a) blind at rank 76 of 5,677 genes (3rd of the 11 genes left after all filters; ZDCM = −7.994, Malecki human LV protein log2FC = −0.479, p = 0.00039, q = 0.066), alongside candidate TMEM65 (rank 74 of 5,677; ZDCM = −8.008; Malecki protein p = 0.86, no change; mouse heart protein −0.42, q = 0.00014; published cardiac knockdown in mice causes dilated cardiomyopathy; Teng et al., Nat Commun 2022). Systematic auditing eliminated three compensatory stress brakes (ARHGAP1, NOL3, CIRBP), removed systemic non-cardiac ageing markers (ZMAT3 elevated in 9/10 non-cardiac tissues; LTBP2 in 6/10), removed COL18A1 and CBR4, and removed ZSWIM8 (not measurable in human heart proteomes; knockout mice have heart defects). Three computational candidates remain: SYPL2 (lower in DCM cardiomyocytes, Reichart q = 2.0×10−5, MAGNet q = 0.19, protein measurable in only 4 of 25 hearts; note its H4 = 0.973 is a liver eQTL at a 44-gene locus, not heart evidence), PPP1R12C (Malecki human LV protein +0.534, p = 1.1×10−4, q = 0.037 in main test, q = 0.12 in two alternative tests; MAGNet DCM q = 2.4×10−5; Reichart q = 0.13; first intron holds the AAVS1 locus; Bhagwan et al., F1000Research 2020), and TMEM65. None of the four targets changes with age in GTEx left ventricle (n = 431). All candidates are at computational identification; none has been tested in the lab.

Keywords: Heart Failure Target Discovery, Cardiac Ageing, Single-Nucleus RNA-Seq, Cardiac Proteomics, SERCA2a (ATP2A2), PPP1R12C, TMEM65, SYPL2, Causal Direction Audit.

1.0 Introduction & Rationale

Advanced heart failure is characterized by progressive contractile failure, sarcoplasmic reticulum (SR) Ca2+ cycling collapse, mitochondrial bioenergetic exhaustion, microvascular rarefaction, and interstitial extracellular matrix (ECM) stiffening (1, 2). Historically, restoring SR Ca2+-ATPase (SERCA2a / ATP2A2) activity or relieving its inhibition by phospholamban (PLN) via protein phosphatase 1 (PP1) regulation (e.g., Inhibitor-1 / PPP1R1A) has provided a foundational biological paradigm for myocardial functional rescue (3). However, translating transcriptomic signatures from human myocardium into actionable therapeutic targets faces two systematic failure modes:

Furthermore, in our internal audit of single-cohort transcriptomic aging clocks on the 54-donor PERIHEART right-atrial cohort (392,819 nuclei), leave-one-donor-out Ridge regression achieved MAE 6.97 yr, r = 0.4606 (recorded in the model file; could not be re-derived from raw data). Similarly, a multi-tissue audit demonstrated that generic senescence markers (PTCHD4, EDA2R, LINC02388) track age systemically across non-cardiac tissues rather than specifically in myocardium. We therefore replaced single-clock scoring with a 4-angle multi-cohort discovery and 4-gate causal audit architecture.

2.0 Four-Angle Discovery Architecture

Every number traces to a named public dataset and an analysis script. The full audit trail is available to partners on request.

2.1 Angle 1: Non-Myocyte Single-Cell Ageing

We computed donor-level pseudobulk OLS age slopes adjusted for donor sex across four non-myocyte lineages — Cardiac Fibroblast, Endothelial, Macrophage/Myeloid, and Pericyte/SMC — across 3 single-cell cohorts (PERIHEART and CAREBANK, both right atrium, and Kanemaru 2023). Cohort-specific statistics were combined via inverse-variance fixed-effects meta-analysis:

βmeta = (∑ wk βk) / ∑ wk,   Zmeta = βmeta √(∑ wk),   wk = 1 / SEk2

To eliminate ambient cardiomyocyte RNA contamination (e.g., TTN, RYR2, MYH6 soup), every candidate had to pass the Intrinsic Cell-Type Expression Gate inside raw .h5ad count matrices: detection in ≥10% of cells in the target lineage and mean expression ≥0.5× its mean expression in ventricular cardiomyocytes.

2.2 Angle 2: Dual-Species Cardiac Proteomics

We intersected Malecki et al. 2025 (PRIDE PXD050610; 25 human left-ventricle hearts: 12 donors aged 14–25 vs 13 aged 51–65, 3,431 proteins tested), GTEx human heart tandem-mass-tag (TMT) proteomics (7 left-ventricle + 11 atrial-appendage donors), and Takasugi et al. 2024 mouse heart aging proteomics (30 vs 6 months) against our multi-cohort human RNA age meta-analysis to identify two distinct classes:

2.3 Angle 3: DCM vs. Ageing Decomposition

Across 5,677 genes evaluated in both Reichart 2022 single-nucleus LV (CELLxGENE f7995301: 52 DCM vs. 18 healthy donors) and MAGNet bulk LV (166 DCM explants and 166 non-failing donor LV), 451 genes achieved Benjamini–Hochberg FDR q < 0.05 in both cohorts with concordant disease direction. Cross-tabulating against our chronological ageing analysis and Malecki 2025 human LV proteomics isolated candidate targets and benchmark controls.

2.4 Angle 4: Human Genetics-First Colocalisation

Using live queries to the Open Targets Platform v4 GraphQL API plus local GWAS colocalisation tables, we required human genetic association or fine-mapped colocalisation posterior probability H4 ≥ 0.80 for cardiovascular traits, replicated DCM dysregulation, and confirmed small-molecule or antibody tractability.

=================================================================================================================================== MASTER AUDIT SUMMARY (12 PUBLIC DATASETS · HAGR GENAGE/CELLAGE · EUROPEPMC · IMPC MOUSE KO DIRECTIONALITY) =================================================================================================================================== • Unblinded Benchmark Recovery: ATP2A2 (SERCA2a) recovered blind (rank 76 of 5,677; 3rd of 11 after all filters; Z_DCM = -7.994) • Database Novelty Check: 0 / 15 candidate genes appear in HAGR GenAge (Human, 307 genes), GenAge (Models), or CellAge • Causal Direction Gate: Eliminated 3 compensatory brakes (ARHGAP1, NOL3/ARC, CIRBP); removed ZMAT3, LTBP2, COL18A1, CBR4, ZSWIM8 ===================================================================================================================================
Gene Discovery Angle & Lineage Multi-Cohort Human Effect Size Human LV Proteome (Malecki, 25 hearts) GenAge / CellAge Literature / Alias Context Genetics & Knockout Causal Audit Verdict
SYPL2
(MG29 · Candidate)		 Angle 3 · Cardiomyocyte Reichart SC CM q = 2.0×10−5
MAGNet DCM q = 0.19		 Measurable in only 4/25 hearts
(GTEx LV p = 0.25, n = 7)		 0 / 0 / 0 Literature under MG29 (1999) CANDIDATE (RESTORE): Lower in DCM cardiomyocytes. Note: H4 = 0.973 is a liver eQTL at a locus shared with 44 genes, not heart evidence. Striated muscle dyad tether (Nishi et al. 1999). No change with age in GTEx LV (n = 431). Computational identification only.
PPP1R12C
(MBS85 · Candidate)		 Angle 1+2 · Mural (1.73×) & CM MAGNet DCM q = 2.4×10−5
Reichart DCM q = 0.13		 log2FC = +0.534
(p = 1.1×10−4, q = 0.037 main test)		 0 / 0 / 0 AAVS1 locus in Intron 1
(Bhagwan et al. 2020)		 CANDIDATE (INHIBIT): Malecki human LV protein q = 0.037 in main test (q = 0.12 in two alternative tests). Scaffolds PP1 holoenzyme. Contains AAVS1 safe-harbour locus in intron 1 (Bhagwan et al., F1000Research 2020); requires unedited iPSC lines. Computational identification only.
TMEM65
(Candidate, Rank 74)		 Angle 2 · Cardiomyocyte Rank 74 of 5,677 genes
ZDCM = −8.008		 Mouse Prot −0.42 (q = 0.00014)
Human Prot p = 0.86 (no change)		 0 / 0 / 0 11 / 4 CANDIDATE (RESTORE): Inner mitochondrial membrane / intercalated disc protein. Cardiac knockdown in mice causes dilated cardiomyopathy (Teng et al., Nat Commun 2022). No change with age in GTEx LV (n = 431). Computational identification only.
ATP2A2
(SERCA2a Benchmark)		 Positive Control Rank 76 of 5,677 genes (3rd of 11)
ZDCM = −7.994		 log2FC = −0.479
(p = 0.00039, q = 0.066)		 0 / 0 / 0 91 / 18 POSITIVE CONTROL BENCHMARK (RESTORE): Completely unsteered blind recovery (zero script hardcoding). Sarcoplasmic reticulum Ca2+-ATPase; CUPID gene therapy benchmark. No change with age in GTEx LV (n = 431).
ZSWIM8
(Removed)		 Removed from Portfolio ZRNA = +0.58 (flat)
Not measurable in human LV MS		 Not measurable
(quantified in only 7/25 hearts)		 0 / 0 / 0 0 / 2 REMOVED (NOT MEASURABLE / KO LETHAL): Not measurable in either human heart proteome (Malecki or GTEx LV). Zswim8 knockout mice have severe heart defects and die as newborns. Recommendation to inhibit withdrawn.
COL18A1
(Endostatin · Removed)		 Removed from Portfolio GTEx LV age: p = 0.991
(slope = −0.000034)		 log2FC = +0.559 (p = 0.048)
GTEx LV p = 0.027 (qBH = 1.0)		 0 / 0 / 0 16 / 12 REMOVED (NO LV AGEING): Removed from portfolio: does not change with age in GTEx left ventricle (p = 0.991, n = 431).
ZMAT3
(WIG-1)		 Angle 1 · Fibroblast (0.867×) FB SC Age Z = +4.808 (q = 0.0056)
Bulk Age Meta-Z = +7.94 (inflated)		 Not detected in human LV hearts (0/25) 0 / 0 / 0 2 / 1 DEMOTED (UNIVERSAL SYSTEMIC AGEING): Pre-registered SAP v3 GTEx 11-tissue model demonstrates elevated expression in 9 of 10 non-cardiac tissues (ratio 0.41×). Universal p53 ageing marker, not a cardiac-specific target.
LTBP2 Angle 1 · Fibroblast (3.1×) FB SC Age Z = +2.762
Bulk Age Meta-Z = +6.89 (inflated)		 log2FC = +0.585
(p = 0.418, not significant)		 0 / 0 / 0 47 / 11 DEMOTED (UNIVERSAL SYSTEMIC FIBROSIS): Pre-registered SAP v3 model demonstrates elevated expression in 6 of 10 non-cardiac tissues. General fibrotic marker rather than heart-specific driver.
CBR4 Angle 2 · Cardiomyocyte (1.5×) ZDCM = −7.230
Bulk Age p = 0.127 (null)		 log2FC = +0.269
(p = 0.0497 — sign flip)		 0 / 0 / 0 1 / 0 DEMOTED (DISCORDANT PROTEIN DIRECTION): Human LV protein increases (+0.269), directly contradicting the "decreases with age → restore" hypothesis. Bulk cardiac ageing is null (p = 0.127).
LGR4 Angle 3 · Genetics-First (H4 = 0.910) ZDCM = −7.568
FB SC Age Z = +2.04		 Low-abundance GPCR 0 / 0 / 0 17 / 7 HOLD (LINEAGE PLEIOTROPY CONFLICT): Cardiomyocyte vs. macrophage knockout models show opposing post-MI remodeling phenotypes; held pending lineage-restricted assays.
ARHGAP1 · NOL3 · CIRBP Controls & Filters Elevated in Ageing & Stress Elevated in Proteomics ARHGAP1: Yes (#249/#13) Multiple REJECTED (COMPENSATORY STRESS BRAKES): Loss-of-function phenotypes in mouse models cause premature ageing (Arhgap1), accelerated dilated LV failure (Nol3), or ferroptotic injury (Cirbp).

3.0 Four-Gate Causal Direction Audit

To ensure that no candidate represents a known aging database entry or a compensatory stress brake, our audit enforced four sequential gates. Every number traces to a named public dataset and an analysis script. The full audit trail is available to partners on request:

  1. HAGR Database Audit: Checked all candidates against the official Human Ageing Genomic Resources releases (GenAge Human Build 21, 307 genes; GenAge Models, 2,205 entries; and CellAge Build 3, 948 senescence genes). Zero of the 15 candidate genes appeared in any HAGR database, whereas the control ARHGAP1 was immediately flagged in both GenAge Human (#249) and GenAge Models (#13).
  2. EuropePMC Literature Audit: Queried live EuropePMC Title/Abstract co-occurrence with cardiac disease and aging terms to stratify unexplored targets (≤5 cardiac papers) from translational benchmarks.
  3. IMPC Mouse Knockout Phenotype Gate: Queried live Open Targets v4 mousePhenotypes (Mammalian Phenotype ontology MP:0005385 cardiovascular system) and published mouse knockout studies to verify whether loss-of-function mimics or rescues pathology.
  4. Human Genetic Directionality Gate: Verified directional concordance against Open Targets v4 locus-to-gene and eQTL/pQTL colocalisation records.

4.0 Translational Alignment with Cardiac Gene Therapy

Our unblinded genome-wide findings converge directly onto four mechanistic pillars of clinical heart failure gene and RNA therapy:

5.0 Data & Code Availability

References

  1. Litviňuková, M. et al. (2020). Cells of the adult human heart. Nature, 588, 466–472.
  2. Reichart, D. et al. (2022). Pathogenic variants damage cell composition and single-cell transcription in cardiomyopathies. Science, 377(6606).
  3. Jessup, M. et al. (2011). Calcium Upregulation by Percutaneous Administration of Gene Therapy in Cardiac Disease (CUPID). Circulation, 124(3), 304–313.
  4. Koenig, A. L. et al. (2022). Single-cell transcriptomics reveals cell-type-specific diversification in human heart failure. Nature Cardiovascular Research, 1, 263–280.
  5. Kanemaru, K. et al. (2023). Spatially resolved multiomics of human cardiac niches. Nature, 619, 801–810.
  6. Malecki, C. et al. (2025). Quantitative proteomic profiling of human left ventricular aging across the adult lifespan. PRIDE / ProteomeXchange (25 human left-ventricle hearts).
  7. Takasugi, M. et al. (2024). Multi-tissue proteomic atlas of mammalian aging reveals post-transcriptional decoupling. Nature Aging (mouse heart, 30 vs 6 months).
  8. de Magalhães, J. P. et al. (2024). Human Ageing Genomic Resources (HAGR): GenAge Build 21 & CellAge Build 3. Nucleic Acids Research, 52(D1), D900–D908.
  9. Buniello, A. et al. (2025). Open Targets Platform: facilitating therapeutic hypotheses building in drug discovery. Nucleic Acids Research, 53(D1).