Nucleus of the heme-iron cluster — the canonical home for heme iron as a within-red-meat
attribute and candidate shared causal channel. Facet pages (per-outcome red-meat pages, the
dose-response matrix) link up here for the heme-specific attribution; this page owns the cross-outcome
question. — this is the wiki’s own graph-structuring note, not a source claim.
The emergent question this page opens
Red meat is associated with several bad outcomes, but red meat is a bundle of attributes (-> Is the Food Category Doing Any Work). Heme iron — the iron bound in the porphyrin ring of myoglobin/haemoglobin, ~10-fold more concentrated in red than white meat and absorbed far more efficiently than non-heme iron — is one candidate component doing some of the work. If a single attribute drives risk across multiple outcomes through a shared mechanism, then reduce red meat is a multi-outcome lever rather than a coincidence of outcome-by-outcome associations, and the component (not the food label) is what a substitution should target. This page assembles the isolated-heme evidence across outcomes to test that. — this section is the wiki’s own framing of the cross-outcome question; the heme-concentration and absorption background is general nutrition knowledge, not from Zhao, whose own contribution (the dietary-heme->T2D and ferritin->T2D estimates) is extracted in Leg 1 below.
Status: banked cross-source synthesis (type-A/C), two legs extracted here. The heme->CHD leg
(Yang 2013) is now extracted alongside the heme->T2D leg (Zhao 2012), so the shared-channel question is
answered across sources, not opened by one. The mortality and colorectal-cancer legs remain held
elsewhere and named below as cross-links. The unifying shared channel claim stays and
confidence: low even so — a second extracted leg raised coverage but not independence (the two
extracted legs re-use the same flagship cohorts and FFQ instrument; see the Synthesis parameter table),
and no natural experiment isolates heme. — the wiki’s own status assessment.
Update 2026-09-03 — the iron-status natural experiment (MR) has landed and SPLITS the thesis by outcome. An MR on genetically-determined systemic iron status (Liu 2024, Leg 5 below) is now held. It does not support a single shared harmful iron channel across outcomes: genetically-predicted iron stores are protective for CAD and adverse for T2D, with no effect on ischemic stroke or heart failure. So the T2D leg is corroborated by an independent method (at the iron-burden level), while the CHD leg’s harmful-direction is challenged — the natural experiment gives no clean support for iron burden being a harmful causal channel for coronary disease. The same-quantity bound (systemic iron != dietary heme) means this still does not close the heme-isolation gap..
Leg 1 — Heme iron -> incident type 2 diabetes (EXTRACTED, this page’s source)
(Zhao et al., 2012) — gold SR+MA, PLoS ONE 2012. The paper isolates heme iron on T2D and separates two exposures that must not be conflated:
- Dietary heme-iron intake (self-reported, FFQ): pooled RR 1.31 (95% CI 1.21-1.43), highest vs lowest category, 4 prospective cohorts (9,246 cases / 179,689 controls), I2 = 0%, no publication bias. The authors call these «convincing combined results» given 4/5 well-designed prospective cohorts (Zhao et al., 2012). This is the exposure the heme channel thesis needs — a dietary signal, not just a biomarker.
- Body iron stores (ferritin, a measured biomarker): pooled RR 1.66 (95% CI 1.15-2.39) prospective; 2.29 (1.48-3.54) cross-sectional (Zhao et al., 2012). Higher and more heterogeneous, but more confounded (see caveat).
Ferritin is not a clean iron signal — it tracks inflammation. Adjusting for metabolic factors attenuated the prospective ferritin RR to non-significance (1.49, 0.90-2.46; vs 1.90, 1.33-2.73 unadjusted) (Zhao et al., 2012), and the authors concede they «couldn’t completely rule out the possibility that ferritin may only serve as a mediator for metabolic abnormalities» (Zhao et al., 2012). So the ferritin arm is best read as supportive of an iron-status gradient but partly confounded; the dietary heme intake arm (I2=0%, no pub bias) is the cleaner exposure for a decision.
Reverse causation was checked, not assumed — cross-sectional > prospective RR raised the flag, but the difference was non-significant and the prospective RR strong, so the authors judged reverse causality unlikely (Zhao et al., 2012) (the gate-1 U/J machinery, run and cleared -> Measurement Error in Dietary Assessment governs the opposite risk: FFQ error attenuates the true gradient toward null).
Leg 2 — Heme iron -> coronary heart disease (EXTRACTED, this page’s source)
(Yang et al., 2013) — gold SR+MA, Eur J Nutr 2013. Six prospective cohorts, 131,553 participants, 2,459 CHD cases.
- Highest vs lowest dietary heme-iron intake: pooled RR 1.31 (95% CI 1.04-1.67), random-effects, with significant heterogeneity (I2 = 55.0%, Pheterogeneity = 0.05) and no significant publication bias (PEgger = 0.23) (Yang et al., 2013).
- Heterogeneity resolves to one outlier cohort. Omitting the single Japanese cohort (JACC) gave RR 1.46 (95% CI 1.21-1.76), I2 = 0.0% — the association is stronger and homogeneous once Japan is excluded (Yang et al., 2013). Yang attributes the outlier to Japanese heme coming mainly from fish/shellfish (carrying protective n-3 / vitamin D) at intakes far below Western levels, not to a reversed effect — so the Western estimate is the transportable one for a red-meat-heme stratum.
- Dose-response: a 1 mg/day increment in heme iron gave RR 1.27 (95% CI 1.10-1.47), I2 = 25.8% (Yang et al., 2013) — a monotone gradient over the studied FFQ-heme range (roughly 0.06-2.8 mg/day across the six cohorts’ category medians, Table 1); no knee is located and none is claimed. CHD is a patient-important outcome, not a surrogate.
Confounding by co-travelling meat components was probed, not assumed. Because saturated fat and cholesterol correlate with heme intake, Yang re-ran the pool excluding the two cohorts that did not adjust for them: «Repeating the analysis by excluding two studies [7, 11] that did not control for these nutrients obtained a summary RR of 1.55 (95 %CI 1.27-1.90), with no heterogeneity» (Yang et al., 2013) — the estimate did not attenuate, so the authors conclude «the observed relation appears to be independent of potential confounding factors, including saturated fat and cholesterol intakes» (Yang et al., 2013). — this addresses measured co-nutrients only; it does not separate heme from red meat as a whole, since heme is partly computed as a fixed fraction of meat iron (see Synthesis).
Measurement error runs toward the null here. All six cohorts used a single baseline FFQ — «most included studies measured heme iron intake only once at baseline rather than updating diet information, which may have led to ‘regression dilution bias’» (Yang et al., 2013) — and Yang notes both regression dilution and non-differential FFQ misclassification «would result in an underestimation of summary risk estimates» (Yang et al., 2013). So the ~1.3-1.5 RRs are, if anything, conservative for the true FFQ-heme gradient -> Measurement Error in Dietary Assessment.
Mechanism (source’s framing, mechanism-not-outcome). Yang frames heme as contributing «to the development of atherosclerosis by catalyzing production of hydroxyl-free radicals and promoting low-density lipoprotein oxidation» (Yang et al., 2013), plus an association with inflammation markers. — this is a proposed pathway offered by the discussion, not an outcome the MA measured; it informs direction, and must not be read as establishing that oxidation is the operative cause of the observed CHD association.
Legs 3-4 — held elsewhere, NAMED not re-extracted
These outcome-legs are already in the fabric on their own pages; they are cross-links here, not this page’s extractions. — the effect figures below are extracted on the linked pages (Etemadi, Bastide), not from Zhao or Yang; this section only names them as sibling legs of the cross-outcome question.
- Heme iron -> all-cause + cause-specific mortality — held on Red and Processed Meat and Cancer and the mortality row of Food Groups and Health Outcomes - A Dose-Response Matrix (Etemadi/NIH-AARP, with a mediation model attributing ~20-24% of the red-meat->mortality association to heme iron).
- Heme iron -> colorectal (colon) cancer — held on Red and Processed Meat and Cancer (Bastide 2011 gold MA + mechanism review; heme as catalyst of endogenous N-nitroso formation and lipid peroxidation). That page also carries the crucial bound: in cohorts heme is partly a red-meat proxy (some studies compute it as a fixed factor of meat iron), so heme is not cleanly separated from red meat by the observational data alone.
Leg 5 — the iron-status natural experiment (MR): the shared channel splits by outcome (EXTRACTED)
(Liu et al., 2024) — MR study,
J Am Heart Assoc 2024 (tier: high); UK Biobank (368,406 observational / 331,964 genetic) plus global
GWAS consortia (CAD n=181,522 cases; HF n=115,150; IS n=62,100; T2D n=80,154). It instruments systemic
iron status — transferrin saturation (TSAT), serum iron, ferritin, TIBC — and hemoglobin with
genetic variants, and compares the observational against the MR association for CAD, HF, IS and T2D.
The same-quantity bound — carry it on every causal claim below. This MR instruments systemic iron burden / hemoglobin, NOT dietary heme intake. It is the very distinction Leg 1 draws between the dietary heme signal and the inflammation-confounded ferritin biomarker. So every causal verdict here bears on the downstream iron-burden mechanism, and answers “does body iron cause these outcomes?” — it does not isolate dietary heme from red meat as the route into that iron burden. The heme-isolation gap is narrowed, not closed.
What the MR found, by outcome (verified figures):
- CAD — PROTECTIVE for iron stores. Genetically-predicted higher iron biomarkers gave modest inverse ORs per 1 SD: «There were modest inverse associations of genetically predicted higher iron status biomarkers with CAD, with ORs of 0.93 (95% CI, 0.88-0.98) for TSAT, 0.91 (95% CI, 0.83-0.99) for serum iron, 0.86 (95% CI, 0.77-0.96) for (log-transformed) ferritin, and 1.04 (95% CI, 0.96-1.12) for TIBC (reflecting lower systemic iron)» (Liu et al., 2024). That is a 7-14% lower CAD risk per 1 SD higher iron (0.86 = 14% lower; TIBC runs inverse to iron, so its OR>1 is concordant). The abstract states it as «modest protective effects of iron biomarkers for CAD (7%-14% lower risk for 1 SD higher levels of iron biomarkers)» (Liu et al., 2024).
- T2D — ADVERSE. Same source, opposite sign: «we found adverse effects of higher levels of iron status with 7% higher risks of T2D per 1 SD higher level of TSAT but not for other iron status biomarkers» (Liu et al., 2024); and genetically-predicted hemoglobin gave «10% to 13% for diabetes» higher risk per 1 SD (Liu et al., 2024). The T2D-adverse signal among iron biomarkers is TSAT-specific (the others were null for T2D) — modest, not uniform across markers.
- Ischemic stroke and heart failure — NULL. «There was no evidence of associations of iron status markers with HF or IS»; hemoglobin too was positive for CAD and diabetes «but not with IS or HF in UK Biobank» (Liu et al., 2024).
- The observational J/U shapes did not survive the genetic check. «The observational analyses in Figure 1 demonstrated U-shaped associations of hemoglobin levels with CAD, wherein both lower and higher levels of hemoglobin were each associated with higher risks of CAD (reference level 14 mg/dL). However, there was no evidence of nonlinearity in the MR analyses (Cochran Q P=0.853, quadratic test P=0.703)» (Liu et al., 2024) — a clean instance of an observational U/J arm that an MR can test -> The U-Shaped Association Artifact.
- An internal discordance the source leaves unexplained. Higher genetically-predicted hemoglobin was ADVERSE for CAD (+8% per 1 SD, men; OR 1.08 [1.04-1.13]) while higher genetically-predicted iron biomarkers were PROTECTIVE — «The discrepant results of protective effects of higher iron status biomarkers for CAD, but adverse effects of higher hemoglobin, are unexplained» (Liu et al., 2024). So the MR’s CAD picture is itself mixed — the iron-store instruments point protective, the hemoglobin instrument adverse — which is a further reason not to read a clean harmful iron channel into CAD.
Adjudication — observational dietary-heme->CHD vs MR systemic-iron->CAD (parameter table, BLOCKING)
The observational heme->CHD leg (Yang, positive/harmful) and this MR (iron stores protective for CAD) run in opposite directions. Before calling that a contradiction, the exposures must be the same quantity. They are not:
| Parameter | Observational dietary-heme->CHD (Yang 2013) | This MR: systemic-iron/Hb->CAD (Liu 2024) | Same quantity? |
|---|---|---|---|
| Exposure | dietary heme-iron intake, FFQ-estimated (as a fraction of meat iron), highest vs lowest (Yang et al., 2013) | genetic instruments for systemic iron status (TSAT / serum iron / ferritin / TIBC) + hemoglobin, per 1 SD (Liu et al., 2024) | NO — dietary intake vs body-iron status; the Leg-1 dietary-vs-biomarker split |
| Outcome | incident CHD | CAD (CARDIOGRAMplusC4D) | ~YES — same coronary endpoint family |
| Design | prospective observational cohorts, single baseline FFQ | Mendelian randomization (genetic natural experiment), 2-sample | NO — MR is immune to reverse causation + dietary confounding |
| Effect / direction | RR 1.31 (1.04-1.67) high-vs-low, POSITIVE (harmful) (Yang et al., 2013) | iron stores OR 0.86-0.93 per SD PROTECTIVE, hemoglobin +8% adverse in men (Liu et al., 2024) | NO — opposite for iron stores, MR internally discordant |
| Confounding structure | red meat / SFA / cholesterol / food-matrix co-travel with FFQ heme | genetic instruments; no dietary confounding (pleiotropy tested, sensitivity concordant) | NO — orthogonal error/confounding structures |
Verdict — this is NOT a joined type-D tension; it is type-F refinement + a G-gap. The
exposures are different quantities (not-joined check (ii): different unit/exposure, consistent once
matched), so a direction reversal is not a contradiction to file as a [[tension]]. The natural
experiment instead bounds the heme nucleus by outcome (type-F): it cannot be the case that systemic
iron is a shared harmful channel across T2D and CAD, because iron stores are adverse for T2D but
protective for CAD. Two readings survive, and the source cannot separate them:
- (a) the observational dietary-heme->CHD signal is confounded by red meat / SFA / the food matrix, and the true iron-mechanism effect on coronary disease is protective or null (the MR estimate). Prior observational iron-status evidence is consistent with this: a prior meta-analysis of 17 studies before 2014, «involving 9236 cases of coronary heart disease and 156 427 participants, demonstrated an inverse association of TSAT and coronary heart dis- ease, and no such associations were found for serum iron, ferritin, and TIBC» (Liu et al., 2024) — so iron status looked protective for CHD even observationally, unlike dietary heme intake.
- (b) dietary heme and systemic iron genuinely differ in their coronary effect — a local pro-oxidant heme effect (LDL oxidation at the vessel wall) distinct from circulating iron stores.
The decision-change is the same under either reading: the natural experiment does not support reduce red meat to cut CHD via the iron/heme channel. If red-meat reduction lowers coronary risk, it most likely runs through a non-iron channel (saturated fat, the food matrix, TMAO, sodium in processed forms) rather than through iron burden. The iron-burden mechanism that the harmful-channel thesis needed for the coronary leg is contradicted by the store instruments and only equivocally supported by the hemoglobin instrument.
Synthesis — the candidate shared channel (INFERRED, low confidence)
Across four outcomes now — T2D (Zhao, extracted) and CHD (Yang, extracted) here; all-cause/cause-specific mortality and colorectal cancer held elsewhere — the same within-red-meat attribute (heme iron) shows a positive dietary association, each with a proposed oxidative/catalytic mechanism. That convergence is consistent with heme iron being a shared channel by which red meat raises risk on several axes at once. This is a type-A emergent synthesis (a cross-outcome channel no single leg asserts), not type-E independent corroboration of one effect — the parameter table below is what enforces that distinction.
Parameter table — the two EXTRACTED legs (BLOCKING precondition for any cross-leg claim).
| Parameter | Zhao — T2D leg (quoted + locus) | Yang — CHD leg (quoted + locus) | Same quantity? |
|---|---|---|---|
| Exposure | dietary heme-iron intake, self-reported FFQ, highest vs lowest category (Zhao et al., 2012) | dietary heme-iron intake, self-reported FFQ, highest vs lowest category (Yang et al., 2013) | YES — same exposure construct + instrument family |
| Outcome | incident type 2 diabetes | incident coronary heart disease | NO — different endpoints |
| Pooled RR (high vs low) | RR 1.31 (95% CI 1.21-1.43), I2 0% (Zhao et al., 2012) | RR 1.31 (95% CI 1.04-1.67), I2 55% (Yang et al., 2013) | NO — identical point estimate, DIFFERENT outcome; a coincidence, not a replication |
| Constituent cohorts | HPFS (Jiang 2004), NHS (Rajpathak 2006), WHS (Song 2004), Iowa-WHS (Lee 2004) (Zhao et al., 2012) | HPFS (Ascherio 1994), NHS-T2D subset (Qi 2007), Rotterdam, Dutch-EPIC, Italian, JACC (Yang et al., 2013) | PARTIAL OVERLAP — HPFS and NHS feed BOTH legs; NOT independent |
| Measurement | FFQ heme, estimated as a fraction of meat iron | FFQ heme, single baseline, regression-dilution toward null | YES — shared self-report + estimation error structure |
What the table forces (the two traps this ingest was most exposed to).
- The RR ~1.31 coincidence is across DIFFERENT outcomes (T2D vs CHD), so it is not two independent measurements of one effect that happen to agree — it is one number appearing on two endpoints. It is suggestive of a shared upstream channel and nothing more; it does not raise confidence in any single estimate..
- The two extracted legs are NOT type-E independent backing. The Harvard flagship cohorts — HPFS and
NHS — and the Willett-lineage FFQ heme instrument feed both the T2D pool (Jiang 2004 = HPFS;
Rajpathak 2006 = NHS) and the CHD pool (Ascherio 1994 = HPFS; Qi 2007 = NHS). Shared participants +
shared measurement instrument means shared confounding and shared measurement-error structure, so the
agreement between legs is volume, not independence — it must not be tokened
[E-independent]. — established by diffing the two MAs’ included-study lists.
This is a hypothesis to test, not an established unifying cause. The original four guards still bound it hard:
- The downstream mechanisms differ by outcome — beta-cell oxidative apoptosis + insulin resistance for T2D (Zhao et al., 2012); LDL oxidation + hydroxyl-radical catalysis driving atherosclerosis for CHD (Yang et al., 2013); endogenous N-nitroso catalysis + lipid peroxidation for colorectal cancer (held elsewhere). The oxidative motif recurs, but a common exposure is not a common pathway; shared channel means shared upstream agent, not one mechanism.
- Heme is not cleanly separable from red meat in any of the cohorts (partly computed as a meat-iron proxy), so some of the heme signal may just be the red-meat signal relabelled -> Is the Food Category Doing Any Work. Yang’s SFA/cholesterol-adjustment sensitivity (RR held at 1.55) rules out those measured co-nutrients, not the meat matrix as a whole.
- The exposure carries compounded measurement error. Heme intake is self-reported (FFQ) AND estimated (heme taken as roughly a fixed fraction of meat iron), not directly measured — two error layers stacked. Both extracted MAs note this attenuates toward the null, so the RRs are conservative, but it also means the heme-specific signal is only as clean as the fixed-fraction assumption -> Measurement Error in Dietary Assessment..
- No natural experiment isolates heme — no Mendelian-randomization or feeding-trial pins heme (vs total iron, vs red meat) to any of these outcomes; all legs are observational with self-reported intake. This is the standing decider the fabric names for the whole cluster.
- The ferritin arm is inflammation-confounded (above), so the biomarker cannot be treated as a clean proxy for the dietary exposure either.
So the decision-relevant reading is modest: heme iron is a plausible, non-negligible component-level lever with a consistent-direction observational signal now across four outcomes (T2D, CHD, mortality, colorectal cancer) — enough to make it a named target for the red-meat substitution question, not enough to claim it is the cause on any axis. The two cleanest extracted legs both sit at RR ~1.31 highest-vs-lowest (T2D 1.31, CHD 1.31; CHD 1.46 in Western-only cohorts), modest relative effects whose absolute size needs a per-stratum baseline the sources do not supply — and whose apparent mutual corroboration is discounted by the shared-cohort overlap above.
MR update 2026-09-03 — the shared-channel synthesis fragments, and the T2D leg gains independent backing (Leg 5). The natural experiment (Liu 2024) revises this synthesis in two directions at once, and the net effect is to lower confidence in a single cross-outcome harmful channel:
- T2D leg — independent-method corroboration, at the iron-burden level
[E-independent]. The observational dietary-heme/ferritin->T2D signal (Zhao, RR 1.31) and the genetic systemic-iron->T2D signal (Liu; TSAT +7%/SD, hemoglobin +10-13%/SD) are reached by genuinely different methods with orthogonal confounding structures — an FFQ/biomarker cohort and a genetic natural experiment immune to reverse causation — arriving separately at higher iron burden raises T2D risk. This is a genuine E on the narrow claim that systemic iron burden is causally adverse for T2D, and it also answers Zhao’s own unresolved worry that ferritin might be «only … a mediator for metabolic abnormalities»: the MR removes the reverse-causation/inflammation confound and the adverse direction holds. Corroborated by Liu MR (systemic iron -> T2D). But the same-quantity bound blocks lifting this to dietary heme: the E backs the iron-burden mechanism, NOT the claim that dietary heme from red meat is the route in. So it narrows the heme gap (body iron is causal for T2D) without closing it. - CHD leg — the shared-harmful-channel reading is contradicted for coronary disease. Iron stores are protective for CAD in the MR, so a single iron channel cannot be harmful for both T2D and CHD. Whatever drives the observational dietary-heme->CHD association, the natural experiment says it is not a harmful iron-burden effect (adjudication table in Leg 5).
— the composite reading is the wiki’s own; the per-leg
figures are extracted on the respective legs. So the honest post-MR position: heme/iron is a causally
supported adverse lever for T2D (at the iron-burden level, dietary-route unproven) and a
not-supported / likely-confounded lever for CHD — the cross-outcome shared harmful channel is no
longer even directionally uniform, and confidence: low stands.
Gap (type-G) — the hook the next sources fill
- Heme iron -> coronary heart disease — CASHED (Yang 2013, extracted as Leg 2 above); the coronary channel is now held.
- An iron-STATUS natural experiment is now HELD (Liu 2024, Leg 5) — the systemic-iron mechanism is
causally tested, but the dietary-heme-isolating natural experiment is STILL ABSENT. The MR causally
implicates systemic iron burden (adverse for T2D, protective for CAD, null for IS/HF), which narrows
the gap; but by the same-quantity bound it instruments body iron / hemoglobin, not dietary heme
intake, so it does not isolate dietary heme from red meat as the route into that iron burden. The
binding residual gap is therefore an exposure-specific one: a natural experiment (feeding design, or
an MR on a heme-absorption / dietary-heme-relevant locus) that isolates DIETARY heme — distinct from
systemic-iron MR — remains unheld.
G (needs a dietary-heme-specific genetic/feeding test). A second independent-instrument systemic-iron->T2D MR is still identified but not held: — would add independence to the T2D leg, not close the dietary-heme gap. — the residual-gap framing is this page’s reasoning from the Leg-1 dietary-vs-biomarker split and the Leg-5 same-quantity bound. - Absolute risk / substitution: every leg is a highest-vs-lowest addition contrast; the Layer-3 question (what replaces the heme — white meat? plant protein?) and per-stratum absolute risk are unanswered -> Should Adults Reduce Red and Processed Meat.