Integrated Quantitative Systems Pharmacology model of peripartum cardiomyopathy, built around the prolactin-cleavage / anti-angiogenic mechanism: peripartum haemodynamic and hormonal load on a STAT3/MnSOD-deficient myocardium → oxidative stress → cathepsin D activation → cleavage of 23-kDa prolactin into 16-kDa prolactin (vasoinhibin) → together with placental sFlt-1 a two-hit anti-angiogenic milieu → coronary microvascular capillary rarefaction, plus NF-κB-driven endothelial miR-146a-5p delivered by exosomes to cardiomyocytes → loss of Erbb4/Nras/Notch1 survival signalling → myocyte loss, fibrosis, LV dilation and systolic failure. The pharmacology layer covers bromocriptine (the disease-specific intervention) and the full guideline heart-failure stack under the antepartum fetotoxicity gate that defines prescribing in this disease.

Peripartum cardiomyopathy is new-onset systolic heart failure (LVEF <45%, no other identifiable cause) arising in the last month of pregnancy or within about five months of delivery. It is not simply heart failure that happens to occur around childbirth: it has a specific, largely reversible pathobiology. In late gestation the maternal circulation carries a 30–50% higher cardiac output and a 40–50% larger plasma volume, prolactin rises roughly ten-fold, and the placenta releases the VEGF-trapping decoy receptor sFlt-1. In a myocardium whose antioxidant reserve is compromised — reduced cardiomyocyte STAT3 and hence reduced MnSOD, frequently on a background of a TTN truncating variant that leaves the sarcomere with less reserve — this load generates oxidative stress that activates lysosomal cathepsin D. Cathepsin D cleaves the abundant 23-kDa prolactin into a 16-kDa fragment, vasoinhibin, which is strongly anti-angiogenic and pro-apoptotic for endothelium. Acting together with placental sFlt-1 (the two-hit model), it causes coronary microvascular endothelial apoptosis and capillary rarefaction, and it drives endothelial NF-κB to produce miR-146a-5p, which is packaged into exosomes, transferred to cardiomyocytes, and knocks down Erbb4, Nras and Notch1 — the myocyte's own survival and metabolic signalling. The result is myocyte loss, replacement fibrosis, chamber dilation and systolic failure, amplified by RAAS and sympathetic activation. The mechanism is also the therapeutic target: the dopamine D2 agonist bromocriptine removes the prolactin substrate, and this is the only disease-specific therapy PPCM has.

1. Genetic substrate and predisposing risk factors — TTN truncating variant (~15%), BAG3, DSP, MYH7/TNNT2, LMNA, FLNC; shared architecture with idiopathic DCM; African ancestry; multiparity; advanced maternal age; twin gestation; preeclampsia (~22%); prolonged β-agonist tocolysis; anaemia; selenium deficiency; delayed care access
2. Late-pregnancy and peripartum haemodynamic / hormonal load — plasma-volume expansion, cardiac-output rise, falling systemic vascular resistance, physiological eccentric hypertrophy, relaxin, oestrogen/progesterone, the delivery event, postpartum uterine autotransfusion, lactation demand, suckling-induced prolactin surge
3. Prolactin axis and cathepsin D cleavage — hypothalamic dopamine, lactotroph D2 receptor, 23-kDa prolactin, PRLR/JAK2/STAT5 pro-survival signalling, pro- and active cathepsin D, the Ser–Asn cleavage site, 16-kDa prolactin (vasoinhibin)
4. Oxidative stress · STAT3 · MnSOD axis — STAT3-driven MnSOD induction, NOX2/NOX4, peroxynitrite, uncoupled eNOS, Nrf2, lysosomal destabilisation, the mitochondrial ROS feed-forward loop, 8-isoprostane/oxLDL
5. Anti-angiogenic milieu — placental sFlt-1, soluble endoglin, low PlGF, cardiac PGC-1α, myocardial VEGF production, free VEGF, VEGFR-2 signalling, the sFlt-1/PlGF ratio, the two-hit node
6. Coronary microvascular endothelium — endothelial apoptosis, capillary density, NO bioavailability, endothelin-1, coronary flow reserve, microvascular ischaemia, pericyte detachment, permeability
7. 16K-PRL → endothelial exosome → miR-146a cardiomyocyte axis — endothelial NF-κB, miR-146a-5p, exosome packaging and transfer, serum miR-146a as a PPCM-specific biomarker, Erbb4/Nras/Notch1/Irak1, neuregulin-1 signalling
8. Cardiomyocyte injury, energetics and structural remodelling — mitochondrial dysfunction, ATP and PCr:ATP, fatty-acid oxidation, substrate shift, sarcomere dysfunction, Ca²⁺ handling (SERCA2a, RyR2), apoptosis, hypertrophy, autophagy, viable contractile mass, fibroblast activation, TGF-β1/SMAD, collagen, MMP:TIMP, fibrosis, galectin-3
9. Immune / inflammatory and viral contribution — IFN-γ, TNF-α, IL-6, sFas/Fas, hs-CRP, macrophage infiltration, anti-cardiac autoantibodies, fetal microchimerism, parvovirus B19/HHV-6, myocarditis overlap
10. LV dysfunction, chamber remodelling and haemodynamics — contractility, LVEF, LVEDD/LVEDV, LVESV, stroke volume, cardiac output, filling pressure, functional mitral regurgitation, RV dysfunction (~30%), congestion, global longitudinal strain, wall stress, reverse remodelling
11. Neurohormonal activation — renin–angiotensin–aldosterone, AT1 signalling, sympathetic activation, β1 downregulation, vasopressin, sodium/water retention, cardiorenal interaction, ANP/BNP, neprilysin, NT-proBNP, cGMP counter-regulation
12. Bromocriptine PK/PD (disease-specific, BOARD concept) — absorption with high first-pass, CYP3A4 metabolism, pituitary D2 agonism, prolactin suppression, lactation suppression, the 1-week and 8-week regimens, prothrombotic risk mandating anticoagulation, adverse effects
13. Guideline heart-failure pharmacotherapy — β-blocker, ACEi/ARB with its antepartum fetotoxicity block, ARNI, MRA, SGLT2 inhibitor, loop diuretic, antepartum-safe hydralazine + nitrate, ivabradine, digoxin, anticoagulation, levosimendan and IVIG/pentoxifylline (mapped but unproven), the breastfeeding-compatibility filter
14. Device, mechanical support and obstetric management — wearable cardioverter defibrillator, ICD deferral pending recovery, IABP/VA-ECMO, LVAD, transplantation, delivery timing, preterm delivery, CRT, contraception and subsequent-pregnancy counselling
15. Clinical endpoints, complications and outcomes — NYHA class, 6-minute walk, LVEF recovery ≥50%, persistent dysfunction, LV thrombus, thromboembolism, ventricular arrhythmia/SCD, cardiogenic shock, rehospitalisation, mortality, MACE, subsequent-pregnancy relapse, quality of life, lactation cessation and maternal–infant bonding, peripartum depression

30 ODE compartments — 11 drug PK/exposure states and 19 disease states.

The cathepsin D threshold is the disease switch. 16-kDa prolactin is generated only when cathepsin D activity exceeds its physiological baseline (CTSD_TOL). This is what makes the model clinically honest: a normal lactating woman has abundant 23-kDa prolactin and no disease, so breastfeeding alone is not pathogenic here — consistent with the absence of convincing evidence that breastfeeding worsens PPCM outcome. Disease requires the prolactin substrate and an oxidatively stressed myocardium.

Recovery has a path-dependent ceiling. Injury above a physiological turnover threshold accumulates as irreversible SCAR, which caps how far CM can recover. Damage prevented early is therefore worth more than damage treated late — this is the mechanism by which a time-limited intervention produces a permanent difference in outcome.

Lactation involution is absorbing. LACT is an autocatalytic supply-and-demand state; once bromocriptine drives it below LACT_MIN it cannot restart. Milk supply, once involuted, does not spontaneously return.

Both are enforced in the model rather than left to the user, because they are the defining prescribing constraints of PPCM:

• ACEi/ARB/ARNI effect is zero while t < T_DELIVERY (antepartum fetotoxicity: fetal renal failure, oligohydramnios). Dosing records may be present; the effect simply does not exist until delivery. Hydralazine + nitrate is the antepartum-safe substitute.
• Bromocriptine effect is zero while t < T_DELIVERY — it is a postpartum intervention.

The [MAIN]/[ODE]/[TABLE] blocks were compiled verbatim into a standalone integrator and run before release. From a presenting LVEF of 29.5% (SEV = 0.72), delivered on day −7, patient breastfeeding:

Also verified:

• the healthy control has a stable normal attractor (LVEF 64.5% → 63.9% at one year, LVEDD 4.6 cm, negligible scar) — the model does not simply decay;
• the untreated arm reaches a genuine non-recovery attractor (LVEDD 6.2 cm, NT-proBNP ~7400 pg/mL, NYHA 3, scar fraction 0.69);
• the antepartum arm nadirs at 21.7% before delivery, while ACE inhibition is gated off and only hydralazine/nitrate is available, then recovers once the placenta and its sFlt-1 are gone;
• anticoagulation leaves LVEF completely unchanged and cuts peak LV thrombus propensity from 0.40 to 0.21 au — a safety intervention cleanly separated from an efficacy one, which is exactly why bromocriptine should not be given without it.

PPCM_simulate_scenarios() runs: (1) no therapy, (2) standard HF therapy, (3) + bromocriptine 1 week + LMWH, (4) BOARD with bromocriptine 8 weeks, (5) BOARD + SGLT2 inhibitor, (6) TTN-variant carrier on BOARD, (7) antepartum preeclamptic presentation with ACE inhibition gated off, and (8)/(9) a matched severe pair with and without anticoagulation that isolates thrombus safety from LVEF efficacy.

Eight tabs: patient profile · drug PK/exposure · prolactin axis · microvascular and miR-146a · LV function · clinical endpoints · scenario comparison · biomarker panel. The sidebar exposes susceptibility (severity, TTN variant, preeclampsia, STAT3 deficit), delivery timing, the bromocriptine regimen and each HF agent independently. When the patient is antepartum the app annotates the delivery day and reports the ACEi/bromocriptine gates as dosed but gated off, so the constraint is visible rather than implicit.

shiny::runApp("ppcm_shiny_app.R")

• The bromocriptine duration effect is exaggerated. The model reproduces the direction of the German trial's 1-week versus 8-week result and offers a mechanism for it (coverage of the early high-oxidative-stress window, with early damage locked in as scar), but the mean-LVEF gap it produces (47.6% versus 56.5% at 6 months) is larger than the trial's (0.49 versus 0.51, even though the recovery-rate difference there was 16 points). Read the duration comparison as a hypothesis, not a prediction.
• Bromocriptine efficacy itself is contested — the BRO-HF retrospective cohort did not find an independent benefit. The model encodes the mechanistic hypothesis, not a settled fact.
• Drug exposures are dose-proportional arbitrary units, not validated plasma PK; only bromocriptine has a structured depot-plus-two-compartment model.
• Fibrosis may be over-weighted as a recovery brake. Late gadolinium enhancement is in fact uncommon in PPCM; the reversible-FIB/irreversible-SCAR split respects that, but the balance between the two is not identified from data.
• No mortality or competing-risk model. MACE_hazard_yr is an illustrative algebraic surrogate; no patient leaves the simulation.
• Deterministic single-patient simulation — there is no $OMEGA/$SIGMA variability block, so outputs are typical-patient trajectories. Comparing a model LVEF value against a published recovery percentage requires that caveat.

63 annotated references in ppcm_references.md, grouped by the role each plays in the model. The mechanistic backbone rests on Hilfiker-Kleiner 2007 Cell (STAT3/cathepsin D/16K-prolactin), Patten 2012 Nature (PGC-1α/VEGF and placental sFlt-1 as the second hit) and Halkein 2013 J Clin Invest (the miR-146a exosome axis); the quantitative anchors are IPAC (McNamara 2015 JACC), the German bromocriptine trial (Hilfiker-Kleiner 2017 Eur Heart J), Sliwa 2010 Circulation, the ESC EORP registry (Sliwa 2020 Eur Heart J) and Ware 2016 NEJM.

Part of the QSP Disease Model Library. Research, education and hypothesis generation only — not for clinical decision-making or regulatory use.