Ferroptosis regulates hemolysis in stored murine and human red blood cells

Key Points

• •QTL analyses identified >400 gene-metabolite associations in fresh and stored RBCs from 350 diversity outbred mice.
• •Steap3 or ferroptosis-related genes FADS1/2, EPHX2, and LPCAT3 regulate lipid oxidation, PTR in mice, hemolysis, and Hgb increments in human RBCs.

Visual Abstract

Abstract

Red blood cell (RBC) metabolism regulates hemolysis during aging in vivo and in the blood bank. However, the genetic underpinnings of RBC metabolic heterogeneity and extravascular hemolysis at population scale are incompletely understood. On the basis of the breeding of 8 founder strains with extreme genetic diversity, the Jackson Laboratory diversity outbred population can capture the impact of genetic heterogeneity in like manner to population-based studies. RBCs from 350 outbred mice, either fresh or stored for 7 days, were tested for posttransfusion recovery, as well as metabolomics and lipidomics analyses. Metabolite and lipid quantitative trait loci (QTL) mapped >400 gene-metabolite associations, which we collated into an online interactive portal. Relevant to RBC storage, we identified a QTL hotspot on chromosome 1, mapping on the region coding for the ferrireductase 6-transmembrane epithelial antigen of the prostate 3 (Steap3), a transcriptional target to p53. Steap3 regulated posttransfusion recovery, contributing to a ferroptosis-like process of lipid peroxidation, as validated via genetic manipulation in mice. Translational validation of murine findings in humans, STEAP3 polymorphisms were associated with RBC iron content, lipid peroxidation, and in vitro hemolysis in 13 091 blood donors from the Recipient Epidemiology and Donor Evaluation Study. QTL analyses in humans identified a network of gene products (fatty acid desaturases 1 and 2, epoxide hydrolase 2, lysophosphatidylcholine acetyl-transferase 3, solute carrier family 22 member 16, glucose 6-phosphate dehydrogenase, very long chain fatty acid elongase, and phospholipase A2 group VI) associated with altered levels of oxylipins. These polymorphisms were prevalent in donors of African descent and were linked to allele frequency of hemolysis-linked polymorphisms for Steap3 or p53. These genetic variants were also associated with lower hemoglobin increments in thousands of single-unit transfusion recipients from the vein-to-vein database.

During storage in blood banks, red blood cells (RBCs) accumulate oxidative damage that make them more susceptible to hemolysis, leading to poor transfusion outcomes. With a comprehensive genetic and metabolic approach, using well-phenotyped mice and human cohorts, D’Alessandro and colleagues provide new insights into the mechanism of RBC “storage lesions.” The authors identify a ferroptosis-like process modulated by Steap3 as a mechanism of red cell damage and provide a potential platform for pharmacological intervention.

Introduction

Garrod’s concept of “chemical individuality,” a founding principle of clinical biochemistry, fostered a century of research on the molecular origins of human diseases.¹ Fueled by discoveries on the genetic underpinnings of cross-individual heterogeneity in metabolism, clinical biochemistry and modern diagnostic medicine rely on the metabolic characterization of blood as a window into systems health.² Red blood cells (RBCs) represent 99% of the circulating blood cells and 83% of the 30 trillion cells in an adult human.³ As they circulate through the body from the large arteries to narrowest capillaries, RBCs exchange metabolites via >77 known active membrane transporters.⁴ As such, RBC metabolism offers a unique window on systems health, and dysregulation thereof, as a function of genetic and nongenetic factors.⁵

Owing to their central role in gas transport, each RBC is loaded with ∼250 to 270 million copies of hemoglobin.^6,7 Each hemoglobin subunit carries a heme group that coordinates a molecule of iron, resulting in up to ∼1 billion molecules of oxygen per RBC at full O₂ saturation. Cumulatively, the total RBC “organ” accounts for up to 2.5 g of iron, ∼66% of total bodily iron.⁸ Every time an oxygen molecule binds to and then dissociates from hemoglobin, there is a 1:8 chance that it strips an extra electron: this process generates reactive oxygen species through ferrous-iron–promoted Fenton and Haber-Weiss chemistry.⁹ Owing to the lack of nuclei and organelles, RBCs are incapable of replacing oxidatively damaged components through de novo protein synthesis. Therefore, in addition to their direct relevance to human health, RBC metabolism offers an excellent, simplified eukaryotic cell model to investigate metabolic responses to oxidant damage—especially that arising from iron-dependent chemistry—in the absence of confounding buffers, such as transcriptional, translational, or mitochondrial activity.

RBC transfusion is a lifesaving in-hospital medical procedure, the most common after vaccination, and blood storage is a practical necessity to meet transfusion demands. During storage, RBCs accumulate oxidative damage to proteins and lipids,¹⁰ a progressively irreversible phenomenon.¹¹ Such “storage lesion” promotes the vesiculation of damaged components, which, in turn, results in morphologically irregular,¹⁰ smaller RBCs¹² that are more susceptible to in-bag hemolysis, to intravascular hemolysis in the bloodstream or extravascular hemolysis via splenic and/or hepatic sequestration and erythrophagocytosis.¹² The latter process is also an essential regulator of the in vivo lifespan of RBCs.⁵ Population studies in twins¹³ or large blood donor cohorts, such as the ∼13 000 donors enrolled in the Recipient Epidemiology and Donor Evaluation (REDS) Study,¹⁴ are now shedding light on how genetics contribute to RBC storage quality. Current storage quality gold standards, as defined by the US Food and Drug Administration, are determined by measuring in-bag hemolysis (<1%) and posttransfusion recovery (PTR) (>75%), (ie, the percentage of transfused RBCs that still circulate at 24 hours on transfusion). Factors like donor age, sex, and ethnicity have been associated with a lower propensity to hemolyze following osmotic insults (eg, in donors of African descent),¹⁵ although a molecular mechanism for this observation has not yet been elucidated.

Investigations of the genetic underpinnings of the metabolic and hemolytic heterogeneity in fresh and stored RBCs offer a window into the mechanisms that drive RBC responses to oxidant, hemolytic insults. To mechanistically explore these systems, our group has developed murine models of blood transfusion,¹⁶ which we have leveraged to show how heterogeneous posttransfusion performances can be investigated by using cross-strain diversity in mice.¹⁷ However, such studies have been limited to a handful of strains and relatively conservative breeding strategies. The recently developed Jackson Laboratory Diversity Outbred (J:DO) mouse model encompasses far greater levels of genetic diversity than traditional mouse crosses, because of the intercrossing of 8 highly divergent founder mouse strains, followed by extensive outbreeding.¹⁸ This resource population can capture the impact of genetic heterogeneity in like manner to population-based studies,¹⁹ while still offering the benefits of murine experiments, including the standardization of relevant factors (eg, age, sex, diet, and exposures for RBC storage models) and strict control of environmental sources of variations,^18,20 enabling a next-generation investigation into how genetic variation contributes to RBC metabolism and storage lesion heterogeneity.

Materials and methods

Extensive details for this section are provided in supplemental Materials (available on the Blood website).

J:DO mouse studies: RBC storage and posttransfusion recovery

A total of 350 J:DO mice were derived from 34-generation breeding of 8 inbred founder strains that represent genetically distinct lineages of the house mouse: A/J, C57BL/6J, 129S1/SvlmJ, NOD/ShiLtJ, NZO/HILtJ, CAST/EiJ, PWK/PhJ, and WSB/EiJ (Figure 1A). All animal procedures were approved by the University of Virginia Institutional Animal Care and Use Committee (protocol number 4269). All animals were genotyped at 143 259 single-nucleotide polymorphisms (SNPs) using the GigaMUGA array.²¹ RBC storage and PTR studies were performed as previously described.²² Fresh and stored murine RBCs were tested for metabolomics,23, 24, 25 lipidomics,²⁶ and oxylipins.²⁶ The quantitative trait loci (QTL) workflow (for PTR, metabolite QTL [mQTL] and lipid QTL [lQTL]) in J:DO mice followed previously defined conventions.²⁷

Figure 1.

Genetic underpinnings of RBC metabolic and storage quality heterogeneity in J:DO mice. (A) Diagram of genetically diverse mouse experiment. Eight founder strains were extensively intercrossed to produce highly recombinant outbred progeny. RBCs were collected and stored under conditions mimicking human RBC blood bank storage for 7 days, the end of shelf-life for pRBC in humans. RBCs were thus transfused into green fluorescent protein–positive (GFP⁺) mice to determine the percentage of transfused RBCs still circulating at 24 hours after transfusion (posttransfusion recovery [PTR]). Mice were genotyped, and fresh and stored RBCs were analyzed via mass spectrometry–based metabolomics and lipidomics. A quantitative trait locus (QTL) for PTR was mapped on chromosome 1 in the region encoding the ferrireductase Steap3, shown as the genome scan (B), haplotype effects at the QTL (C), and SNP associations in the QTL region (D). (E-J) Combined Manhattan plots of peak associations for metabolites, lipids, and oxylipins in fresh and stored RBCs reveal a hot spot region on chromosome 1 associated with changes in metabolite, lipid, and oxylipin levels specific to stored samples. Metabolites, lipids, and oxylipins with strong QTL are highlighted as labels. Molecular correlates to PTR identified lipid peroxidation products as strongly associated with PTR (K), including prostaglandin G2 and D2/E2 (isobars) (L-M). FL, fatty acid; LOD, logarithm (base 10) of odds.

REDS RBC omics

Metabolomics analyses were performed on day 42 packed RBCs from 13 091 “index” donors. Donors ranking in the 5th and 95th percentile for end of storage hemolysis (n = 643) were invited to donate a second (“recalled”) unit, which was tested at storage days 10, 23, and 42 (1929 samples) for hemolytic parameters, high-throughput metabolomics,23, 24, 25 proteomics,²⁸ lipidomics,²⁶ and inductively coupled plasma mass spectrometry.²⁹

REDS RBC omics mQTL analyses of oxylipins

The workflow for the mQTL analysis of oxylipin metabolites is consistent with previously described methods.³⁰

Determination of hemoglobin and bilirubin increment via the vein-to-vein database

Association of metabolite levels with hemoglobin increments was performed by interrogating the vein-to-vein database, as described by Roubinian et al^31,32

Data analysis and statistical analyses

Data analysis and statistical analyses, including hierarchical clustering analysis, linear discriminant analysis, uniform manifold approximation and projection, correlation analyses, and Lasso regression, were performed using both MetaboAnalyst 5.0³³ and in-house–developed code in R (4.2.3; 2023-03-15).

Results

Mapping Steap3 as a key genetic driver of RBC posttransfusion recovery in J:DO mice

To map genetic variants that influence RBC storability in mice, 350 mice were selected from the 34th generation of the ongoing J:DO breeding program (Figure 1A). RBCs from these mice were stored for 7 days under conditions mimicking human RBC storage in the blood bank.¹⁶ Stored RBCs were mixed with fresh mCherry tracer RBCs and transfused into green fluorescent protein–positive mice. PTR was determined by establishing the test:tracer RBC ratio by flow cytometry on blood obtained 24 hours posttransfusion, on normalization to the pretransfusion ratio (henceforth, posttransfusion recovery or PTR³⁴; Figure 1A). Although complete blood cell counts were not captured for this J:DO cohort, no hematological abnormality (including no anemia, ie, hemoglobin <13.6 g/dL³⁵) had been noted for any of the founder strains.³⁶ Moreover, anemia is not an emergent property of diversity breeding, as other studies on J:DO populations have shown normal hemoglobin in J:DO mice (98/100 J:DO mice in the Svenson cohort³⁷ had hemoglobin levels above a mild anemia threshold of >11 g/dL, as defined by Kim et al³⁸; supplemental Figure 1). Mice were genotyped on a ∼143 000 marker array,²¹ and QTL analysis was performed using PTR as a quantitative trait (Figure 1B), which identified a strong QTL (peak logarithm [base 10] of odds [LOD] score = 36.3) on chromosome 1 (120.0 Mbp) nearby the encoded ferrireductase Steap3 (Figure 1C-D).

Genetic drivers of RBC metabolic heterogeneity in J:DO mice

Refrigerated storage has significant effects on stored murine RBC metabolism, with changes in markers of the metabolic lesion (supplemental Figure 2A-E), consistent with the literature.⁵ Genetic mapping analysis was then performed on metabolomics and lipidomics data to identify QTL (here referred to as mQTL for metabolites and lQTL for lipids) for fresh and stored RBCs (Figure 1E-J). On the basis of a stringent threshold of LOD score >8, we mapped 76 mQTL and 54 lQTL in fresh samples and 114 mQTL and 168 lQTL in stored samples. QTL hotspot regions were identified on the basis of >5 QTL comapping within a sliding window of 4 Mbp. We observed mQTL hotspots on chromosomes 1, 5, 7, 9, 12, and 14 (detailed plots of QTL density across the genome in supplemental Figure 3A-F). For lQTL, we observed a fresh-specific hotspot on chromosome 3 and stored-specific hotspots on chromosomes 1, 7, 9, 10, 11, 13, and 19.

We also imputed SNP and structural variant (SV)³⁹ genotypes and performed association analysis (in contrast to the conventional haplotype-based approach commonly used as the initial QTL mapping in J:DO data) within QTL support regions. As an internal validation of the quality of the data, we identified an SV that was more strongly associated with thymidine levels than any SNPs, suggesting it may be the actual causal variant (supplemental Figure 4A). The thymidine-associated SV was just upstream of the thymidine phosphorylase (Tymp) gene, an enzyme that catalyzes thymidine phosphorolysis. The QTL haplotype effects are highly consistent with a cis-eQTL for Tymp observed in liver tissue from another J:DO cohort⁴⁰ (supplemental Figure 4F), suggesting that J:DO mice with the B6 and PWK alleles at Tymp have low levels of Tymp protein, resulting in higher thymidine levels (supplemental Figure 4A-B).

Genetic variation at Steap3 is associated with posttransfusion recovery and lipid peroxidation

Across all sets, the top negative metabolic correlates to PTR were oxylipins, including prostaglandins, eicosanoid, and octadecadienoic hydroxy- and hydroxy-peroxides (HETEs, HODEs, and HPETEs, respectively; Figure 1K-M). Conversely, long-chain poly- and highly unsaturated fatty acids (20:3; 20:5; and 22:5) were positive correlates, suggesting that decreased storage-induced oxidation of these substrates is protective against a decline in posttransfusion performance (Figure 1K).

Looking across metabolites and lipids reveals an extensive QTL hotspot that covers the coding region for Steap3 on chromosome 1 (Figure 2A-B). We used the QTL results to define a comapping network view of the gene-metabolite pairs, which illustrates the centrality of the STEAP3 QTL hotspot (Figure 2C), as illustrated via density plots (Figure 2D-F). The haplotype effects at this QTL region are complex with alleles from B6 and WSB being strongly distinct from the others (eg, HODEs; Figure 2G). Steap3 is located centrally within the QTL region, and SNP variants with alleles shared by B6 and WSB best capture the haplotype association (Figure 2H). The genetic effects are highly consistent and specific to stored RBCs within this QTL hotspot, which is emphasized by how the relationships between metabolites, lipids, oxylipins, and PTR resolves into 2 correlated and anti-correlated groups, suggesting they are driven by the same genetic variation at STEAP3 (Figure 2I-K).

Figure 2.

Chromosome 1 QTL hotspot reveals Steap3 as a key regulator of metabolites, oxylipins, and lipids in stored RBCs. QTL for metabolites, oyxlipins, and lipids revealed comapping hotspots, including the STEAP3 locus on chromosome 1 specific to stored RBCs, depicted as hive plots (A-B), a network plot (C), and QTL density plots (D-F). Three-dimensional (3D) model of the membrane proximal oxidoreductase domain of human Steap3 (bound to NADP) is overlayed with the chromosome 1 hotspot. Other candidate gene drivers of hotspots are included as labels. Hydroxyoctadecadienoic acid (HODE) is an example of a metabolite that maps the Steap3 QTL (chromosome 1 scan and haplotype effects [G] and SNP associations [H]). The correlation structure among metabolites, lipids, oxylipins, and PTR reflects the Steap3 QTL, with fresh samples being uncoordinated across data types (I), but highly coordinated in stored samples, at both the individual-level data (J) and the QTL haplotype effects (K). STEAP3 KO also has a significant impact on RBC metabolism on storage (L), both KO and hypermorphic gain of function (GOF) resulting in decreases in PTR in mice (M). (N) However, only GOF mice but not KO show elevated levels of oxylpins in end of storage RBCs. ∗∗∗P < .001; ∗∗∗∗P < .0001.

A metabolomics and lipidomics QTL resource webtool

We have developed a publicly available online portal using our QTLViewer webtool⁴¹ for interactive exploration of QTL results and distribution of the processed data, accessible at: https://churchilllab.jax.org/qtlviewer/Zimring/RBC. Functionality includes plotting and exporting results from PTR and metabolite and lipid phenotypes, including the ability to inspect how covariates of interest (eg, sex) correlate with the traits, QTL scans, SNP association, and haplotype effect scans. We also identified strong candidate genes at some of the other QTL hotspots, which are visualized as hive plots and network graphs in Figure 2A-C (details in supplemental Figures 5-13). Other than the hotspot on chromosome 1, we identified additional metabolite-gene associations, including (i) glycolysis⁴² and pentose phosphate pathway metabolites and the region on chromosome 5 coding for CD38/BST1 (supplemental Figure 5); (ii) antioxidant metabolites and a region on chromosome 7 mapping on polymorphic hemoglobin β (Hbb1 and Hbb2; supplemental Figure 6); (iii) purine nucleosides and a region on chromosome 9 mapping on MON1 homolog A (Mon1a; supplemental Figure 7); and (iv) lysine metabolites and a region on chromosome 14 (supplemental Figure 8A-E).

Similarly, lipid-gene associations were identified for lipids extracted from fresh RBCs and a region mapping on chromosome 3 (supplemental Figure 8F-I), or lipids extracted from stored RBCs and regions on (1) free fatty acids (18, 20, and 22 C aliphatic chain length); (2) oxo-diacyl- and oxo-triacyl-glycerols and chromosome 7 (supplemental Figure 9A-D); (3) phospholipids and chromosome 9 and 10 (supplemental Figure 8E-H and supplemental Figure 10A-D, respectively); (4) sphingolipids, triacylglycerols, and additional hotspots on chromosome 10 (supplemental Figure 10E-H and supplemental Figure 11A-D); lysophospholipids and chromosome 11 (supplemental Figure 11E-H); (5) triacylglycerols and chromosome 13 (supplemental Figure 12A-D); and (6) oxolysphophospholipids and chromosome 18 (supplemental Figure 12E-H). The depth of these secondary findings highlights the value of the J:DO population as a resource for metabolomics and lipidomics studies.

Genetic manipulation of STEAP3 regulates lipid peroxidation in stored murine RBCs

Blessing and curse of careful science, our findings from the J:DO population on PTR independently validated previous work from our groups³⁴ that used an F₂ cross-generation between just 2 strains of mice: C57BL/6J mice, previously identified as good storers,¹⁷ and FVB/J mice, previously identified as poor storers.¹⁷ This analysis identified the Steap3 locus at lower resolution (3 Mb,¹⁷ here further narrowed down to 2.85 Mb after additional 6 generations of backcrossing to C57BL/6J). Steap3 activity showed strain-specific ferrozine activity (higher in poor storers), associated with higher lipid peroxidation (supplemental Figure 14A-D), and western blots confirmed higher Steap3 protein levels in poorly storing RBCs. Given that the J:DO population included C57BL/6J as founder but not FVB/J, the present study allowed us to characterize the effects of new STEAP3 alleles and potentially map additional genetic loci for PTR. Although we did not map new additional QTL for PTR, the haplotype effects at STEAP3 are complex. Bayesian modeling of the STEAP3 QTL for PTR found evidence for >2 functional alleles in the J:DO mice,⁴³ suggesting multiple genetic variants at STEAP3 affect PTR and downstream metabolites/lipids (supplemental Figure 13).

To mechanistically follow up on these observations, first we bred the hypermorphic STEAP3 from FVB/J mice onto a C57BL/6J background, which increased lipid peroxidation and decreased PTR (∼25%) in otherwise good storing mice (supplemental Figure 14E-F). We then investigated STEAP3 knockout (KO) mice, which also had a significant impact on RBC metabolism on storage (Figure 2L). Specifically, both Steap3 KO and hypermorphic gain of function (GOF) resulted in decreases in PTR in mice (∼45%), which was still 2-fold higher in Steap3 KO compared with GOF mice (Figure 2M). However, only GOF mice but not KO show elevated levels of oxylipins in end of storage RBCs (Figure 2N), demonstrating that Steap3 presence/activity in mature RBCs is required to promote lipid peroxidation and quality decline in stored RBCs, but also that loss of STEAP3 is deleterious—although to a lesser extent than GOF—to PTR, suggestive of a potential goldilocks effect.

Oxylipins, iron, and STEAP3 are associated with hemolysis in 13 000 REDS RBC omics donors

The roles of Steap3 in lipid peroxidation and decline in storage and posttransfusion quality, as identified in murine models, have not yet been validated in humans. To address this limitation, we first determined the metabolic underpinnings of hemolysis measurements in 13 091 index and 643 recalled donors from the REDS RBC Omics studies (Figure 3A). We identified oxylipins (particularly, HETEs and HODEs) as top correlates to end of storage hemolysis and vesiculation in index and recalled blood units (Figure 3B-C).

Figure 3.

Oxylipins, iron, and STEAP3 are associated with hemolysis in 13 000 REDS RBC Omics blood donors. (A) Hemolysis measurements in 13 091 donors from the REDS RBC Omics studies identified oxylipins (especially, HETEs and HODEs) as top predictors of end of storage hemolytic propensity. A second independent unit from 643 of the index donors who were contacted because of their extreme hemolytic propensity (5th vs 95th percentile). (B) Determination of vesiculation rates in the 643 units at storage day 10, 23, and 42 identified HETEs and HODEs as top markers of RBC vesiculation. (C) A total of 37 SNPs were monitored for STEAP3 in the REDS RBC Omics study. (C) Common nonsynonymous coding SNPs correlated with osmotic and oxidative hemolytic propensity and oxylipin levels. (D) Some of these SNPs were predicted to impact the ferrireductase function of the enzyme. (E) Three-dimensional (3D) uniform manifold approximation and projection (uMAP) representation of HETE levels and all nonsynonymous STEAP3 SNPs in homozygosity are shown. (F) Iron measurements via inductively coupled plasma mass spectrometry (ICP-MS) in the REDS RBC Omics recalled donor cohort showed significant association with hemolysis and oxylipins. (G) Hemolysis was significantly lower in donors carrying 2 alleles of rs17013371, the most common nonsynonymous STEAP3 coding SNP. (H) Multi-omics correlates in the same recalled donor cohort confirmed such association for the rs17013371, an SNP that is overrepresented in donors of African American descent (I). ∗∗P < .01; ∗∗∗∗P < .0001. FA, fatty acid.

Thirty-seven STEAP3 SNPs were monitored in the REDS cohort: 23 were intronic, 3 mapped in the downstream 3^’-untranslated region, while 11 were coding, of which 2 were synonymous and 9 nonsynonymous (Figure 3C; supplemental Figure 15). In stored human RBCs, STEAP3 variants were associated with hemolysis (rs17013371) or oxylipins (rs141086904 and rs147820529) (Figure 3C; supplemental Figure 15B). These common variants are located in regions neighboring the active site of the Steap3 enzyme (eg, rs17013371 results in the A184T substitution), which is predicted to impact enzyme kinetics (Figure 3D). Allele frequence for Steap3 SNPs and lipid peroxidation overlapped in 2-dimensional uniform manifold approximation and projection representations of population data (Figure 3E).

Steap3 is a ferrireductase required for transferrin-dependent iron uptake during erythropoiesis⁴⁴ and thus regulates intracellular iron content in mature RBCs. On the basis of this mechanism, we predict that elevated iron content, especially in its ferrous state as a result of Steap3 activity, would promote Fenton and Haber-Weiss chemistry in iron-loaded erythrocytes, which, in turn, would favor lipid peroxidation via ferroptosis,⁴⁵ also referred to as “death by lipid peroxidation.”⁴⁶ Consistent with our hypothesis, correlation of omics data to RBC iron levels revealed a significant association with hemolysis and oxylipins (n = 643; Figure 3F). In both index and recalled donors, carrying 1 or 2 alleles of nonsynonymous coding SNPs for STEAP3 (eg, rs17013371) was negatively associated with osmotic and storage hemolysis (Figure 3G-H), RBC heme and iron content, and oxononanoic acid (fatty acid [FA] 9:0-OH), a breakdown product of FA 18:2-derived oxylipins⁴⁷ (eg, HODEs) (Figure 3H). The rs17013371 alleles had higher prevalence in donors of African descent (Figure 3I).

Genetic factors contribute to heterogeneous lipid peroxidation in 13 091 stored human RBCs

To investigate the genetic determinants of oxylipin levels in stored human RBCs, targeted measurement of oxylipins in 13 091 day 42 index units were combined with data on 879 000 SNPs from a precision transfusion medicine array⁴⁸ (Figure 4A). Results from mQTL analyses are shown for HETE, HPETE, leukotriene A4, prostaglandin A2, dinor-prostaglandin F2α, prostaglandin D2/E2 and G2 (representative Manhattan and locusZoom plots in Figure 4A-K). The combination of oxylipin-gene hits elucidates a pathway of lipid detoxification fueled by L-carnitine,⁴⁹ NADPH, and glutathione-dependent systems (Figure 4I), a map that substantially overlaps with literature in the field of ferroptosis.⁴⁵

Figure 4.

Genetic factors contributing to heterogeneous lipid peroxidation in 13 091 human RBCs after storage for 42 days. (A) Genome-wide association studies (GWASs) were performed for lipid peroxidation products in the REDS RBC Omics index donor cohort (n = 13 091) against 870 000 SNPs from a precision transfusion medicine array. (B-H) Manhattan plots are shown for HETE, HPETE, leukotriene A4, prostaglandin A2, dinor-prostaglandin F2α, prostaglandin D2/E2, and G2. (I) A summary model of the pathway that emerges as a regulator of lipid peroxidation in stored human RBCs. (J-M) Representative locus zoom plots for selected top hits by significance (–log10 P values, y axes). ELOVL, very long chain fatty acid elongase; SLC22A16, solute carrier family 22 member 16.

Of all oxylipins, HETEs (combined isomers; Figure 4B) showed the highest number of hits, including polymorphisms in the regions coding for lysophosphatidylcholine acetyl-transferase 3 (LPCAT3; rs60015123; P = 1.54E-168); fatty acid desaturases 1 and 2 (FADS1 or 2; rs174564; P = 1.38E-53); the carnitine transporter solute carrier family 22 member 16 (SLC22A16) (rs12210538; P = 3.03E-34); glutathione synthetase (rs6087652; P = 1.88E-08); phospholipase A2 group VI (rs3827354; P = 4.82E-11); glucose 6-phosphate dehydrogenase (G6PD; rs1050828; c.202G>A; p.Val68Met, known as the “common African variant”), the rate-limiting enzyme of NADPH-generating pentose phosphate pathway; and FAM234A (associated with G6PD deficiency in population studies⁵⁰). Most of these hits were shared with other oxylipins, including HPETE, prostaglandin A2 and D2/E2, leukotriene A4 (Figure 4), and HODEs (supplemental Figure 16). Additionally, epoxide hydrolase 2 (EPHX2) emerged as the most significant hit for dinor-prostaglandin F2α (rs10090802; P < e−308; Figure 4F) and prostaglandin G2 (Figure 4H).

Similar to Steap3 SNPs, allele frequencies for LPCAT3 rs60015123 (nonsense mutation associated with increased protein decay), FADS1/2 rs174564 (upstream variant), and EPHX2 rs10090802 (regulatory region variant) were significantly associated with higher (EPHX2 and LPCAT3) and lower (FADS1/2) lipid peroxidation and hemolytic propensity on osmotic insult (Figure 5A-I). Also consistent with Steap3, breakdown of allele frequency by ethnicity showed a significant enrichment in blood donors of African descent for LPCAT3 and EPHX2, whereas FADS1/2 was underrepresented (Figure 5J). Like Steap3 (Figure 3H), SNPs in all genes associated with oxylipins in stored human RBCs are also associated with the Duffy blood group antigen protein atypical chemokine receptor 1 (ACKR1) (supplemental Figure 17A).

Figure 5.

Genetic polymorphisms in LPCAT3, FADS1/2, and EHPHX2. Analyses in the recalled donor cohort confirmed findings from the index donor studies showing a strong association between SNPs in LPCAT3, FADS1/2, and EPHX2 and lipid peroxidation products via linear discriminant analysis unadjusted or adjusted by storage day (A-C) and Spearman correlation (D-F). (G-I) Allele dosage was associated with a decrease in hemolytic propensity (on osmotic insult) in the larger index donor cohort. (J) Breakdown of allele frequency by ethnicity showed a strong enrichment in blood donors of African descent for LPCAT3 and EPHX2, whereas FADS1/2 was underrepresented in this ethnic group. ∗∗∗P < .001; ∗∗∗∗P < .0001. FA, fatty acid; ns, not significant.

TP53 is polymorphic in healthy blood donors and associates with genetic regulators of lipid peroxidation and hemolysis

A transcriptional target of tumor suppressor protein TP53, Steap3 is also known as tumor suppressor–activated pathway 6 or TSAP6.⁵¹ The role of TP53 in the regulation of ferroptosis has been extensively investigated.^45,52 Both p53 and STEAP3 are critical to erythropoiesis53, 54, 55 and polymorphic in humans: the prevalence of p53 germ line mutations is estimated ∼1:2000 births,⁵⁶ and somatic mutations accumulate with age.56, 57, 58, 59 Although the role of p53 has been extensively studied in normal and malignant hematopoiesis,⁶⁰ little is known whether p53 is polymorphic in healthy human blood donor volunteers and whether such polymorphisms affect mature RBC metabolism and hemolysis.

To address this gap, we evaluated >50 common TP53 SNPs in the 13 091 REDS RBC Omics blood donors (20 most prevalent alleles in Figure 6A). The list includes common SNPs associated with increased likelihood to develop cancers, such as rs1042522 (associated with the P72R mutation), rs150200764 (also known as the hereditary cancer predisposing syndrome, benign), and rs1800371 and rs1800372 (H178R mutation in the DNA-binding site of TP53; Figure 6A). Notably, allele copies for a noncoding SNP, rs8064946, were associated with osmotic hemolysis, and linked to allele frequencies for EPHX2, LPCAT3, G6PD, Steap3, GPX4, ACKR1, and SLC22A16 (Figure 6B). In both index and recalled donors (Figure 6C-D), rs8064946 was associated with markers of osmotic fragility (FA 22:5 and FA:6; kynurenine⁶¹) and oxidation-reduction homeostasis (pentose phosphate isomers, NADP⁺). Allele frequency as a function of ethnic breakdown revealed a lower prevalence of the potentially pathogenic rs1042522 and rs105200764, and higher prevalence of the noncoding rs8064946 in donors of African descent (Figure 6E-H), all associated with higher osmotic fragility (Figure 6I). Only 3 blood donors were homozygous for the pathogenic TP53 SNPs rs1800371 and rs1800372 alleles (H178R and R248Q mutations in the DNA-binding site of TP53 [Figure 6I; supplemental Figure 17.B]), whereas heterozygous genotypes were far more frequent and linked to higher osmotic hemolysis.

Figure 6.

TP53 is polymorphic in the healthy blood donor population and associates with hemolysis. Over 50 common TP53 SNPs were monitored in the 13 091 REDS RBC Omics blood donors (top 20 plotted [A] as a function of allele frequency), including common SNPs associated with increased likelihood to develop cancers. One of these SNPs was identified as the top TP53 SNP associated with hemolytic propensity, an SNP whose allele frequency significantly correlated with that of the alleles for all the genes contributing to lipid peroxidation from the GWAS analysis (B). In index (n = 13 091; C) and recalled donors (validation cohort; n = 643; D), the TP53 rs8064946 was associated with markers of osmotic fragility, oxylipins, and polyunsaturated fatty acids, as well as to allele frequency and protein levels for all lipid peroxidation-associated genes, when detected via proteomics (D). (E-H) Allele frequency as a function of ethnic breakdown shows a lower frequency of the potentially pathogenic mutations rs1042522 and rs105200764 and higher frequency of the noncoding rs8064946 in donors of African descent. (I) All these variants were associated with lower osmotic fragility. ∗∗∗P < .001; ∗∗∗∗P < .0001. ns, not significant.

Oxylipins and related genetic traits are associated with lower transfusion efficacy in recipients

To investigate the potential clinical relevance of our findings, we accessed the “vein-to-vein” database of the REDS program,³² and linked donor genotypes or oxylipin measurements to the hemoglobin increments in thousands of single-unit transfusion recipients of units donated by REDS donors. Unfortunately, not enough transfusion events were recorded in our database for units from donors carrying relevant STEAP3 alleles to perform these translationally relevant analyses. However, we observed that transfusion of units from donors who carried 2 copies of the rs8064946 TP53 SNP was associated with significantly lower hemoglobin increments (P < .001), particularly when the transfused unit was older than 5 weeks (n = 4636; Figure 7A-B). Similar observations were made for donors carrying 2 copies of the LPCAT3 rs60015123 allele, with significantly lower hemoglobin increments, both immediately after transfusion and at 24 hours, when transfused units were older than 3 weeks or 36 to 42 days old (n = 4473 transfusion events; Figure 7C-E). Finally, we show significant decreases in hemoglobin increments in recipients of units from donors with elevated end of storage oxylipin (HETE) levels (n = 1241; Figure 7F).

Figure 7.

Oxylipins and genetic polymorphisms that regulate them are associated with in vivo hemolysis critically ill individuals receiving nonautologous transfusion. By accessing the “vein-to-vein” database of the REDS RBC Omics program, we linked blood donor TP53 rs8064946 status (homozygous dominant [HD]; heterozygous [HET]; homozygous recessive [HR]) to adjusted hemoglobin increments in clinical recipients of single pRBC unit transfusions of different storage ages (A) or at storage day 36 to 42 (B). Similar analyses were performed for LPCAT3 rs60015123, focusing on Hgb increments immediately after transfusion (C) or at 24 hours (D) at any storage age, or for transfused units aged 36 to 42 days (E). (F) Similarly, we linked end of storage oxylipin levels (eg, hydroxyeicosatetraenoic acid [HETE]) by quartiles to hemoglobin increments on heterologous transfusion of products from the same donors to critically ill recipients requiring transfusion. ∗P < .05; ∗∗P < .01; ∗∗∗P < .001; ∗∗∗∗P < .0001.

Discussion

In the present study, we combined metabolomics and lipidomics of fresh and stored RBCs with genomics data from 350 diversity outbred mice. Results were collated in a novel, publicly available interactive portal that encompasses >400 QTL (gene-metabolite associations). We then sought to determine the genetic underpinnings of PTR, identifying SNPs in the region coding for Steap3 as the main driver of lipid peroxidation and poor PTR in stored murine RBCs. Through genetic manipulation of poor and good “storer” mouse strains,¹⁷ we show that ablation of STEAP3 or introduction of hypermorphic STEAP3 are sufficient to modulate lipid peroxidation and PTR of stored murine RBCs, demonstrating a causal role. We proceeded to establish the translational relevance of these findings in humans, by showing that common polymorphisms of orthologous human genes—especially prevalent in blood donors of African descent—are likewise associated with decreased lipid peroxidation and hemolysis.⁶² Interestingly, other alleles, like G6PD V68M (African variant, resulting in deficient G6PD activity), are more prevalent in donors of African descent, and have been previously associated with increased susceptibility to hemolysis following oxidant insults or storage.^31,63,64 As G6PD deficiency has been proposed to undergo positive selection owing to a potential protective role against malaria,⁶⁵ it is interesting to speculate that some of these compensatory SNPs may have become more prevalent in donors of genetically inferred African ancestry under similar evolutionary pressure.

In mice and humans, STEAP3 plays a central role in the regulation of iron metabolism via endosome trafficking⁶⁶ and exocytic transport via vesiculation.⁵¹ Genetic deficiency of STEAP3 is associated with anemia in humans⁶⁷ with a highly similar anemic phenotype in STEAP3 knockout mice.⁵³ However, polymorphisms predicted to cause decreased (but not absent) STEAP3 did not show an association with anemia in people of Asian descent.⁵⁷ The same effect was observed in mice wherein no anemia is observed, and RBC blood parameters are normal in mice with either hypomorphic or hypermorphic variants of the STEAP3 gene. Thus, although the absence of STEAP3 shows its importance in normal erythropoiesis in both humans and mice, milder hypomorphic or hypermorphic variants do not affect peripheral blood parameters and are thus not excluded from the population of donated blood units. This is formally demonstrated by the genetic variability of STEAP3 in our analyzed human cohort (all donors had normal peripheral blood parameters) and by the observation that neither the parental strains used to generate J:DO mice nor the outbred population shows abnormal parameters. In aggregate, these findings demonstrate that although hypomorphic or hypermorphic STEAP3 polymorphisms do not affect normal blood parameters, they do alter how RBCs store, with strong associations in both humans and mice and formal demonstration of causality in mice.

Mechanistically, our results are consistent with a role for Steap3 in promoting Fenton and Haber-Weiss chemistry in iron-loaded mature erythrocytes. Specifically, ferrous iron participates in this chemistry, becoming oxidized to its ferric state in the process with the concomitant generation of hydroxyl or hydroperoxyl radicals, which can, in turn, attack fatty acids (preferentially, poly- and highly unsaturated fatty acids, such as octadecadienoic and eicosatetraenoic acid), thus promoting lipid peroxidation. By reducing Fe³⁺ to its Fe²⁺ state, the ferrireductase Steap3 tilts the balance of this reaction by increasing the availability of the reactant, thus favoring the generation of product radicals. This cascade of events is akin to ferroptosis,⁴⁵ a process also referred to as death by lipid peroxidation,⁴⁶ which closely resembles the hallmarks of nonapoptotic RBC death or eryptosis.⁶⁸ Although some of these processes have been reported to occur during RBC aging,⁶⁹ and pharmacologic and genetic manipulation of ferroptosis pathways has been tested in cancer,⁴⁵ no study to date has pursued these strategies to alter RBC lifespan in vitro (blood bank storage) and in vivo. Studies have documented ferroptosis-like phenotypes of intracellular iron accumulation and hemolysis by oxidant stress in erythroid cells in the context of ferroportin knockouts.⁷⁰ Yet, the burgeoning literature on ferroptosis has hitherto disregarded anucleate RBCs, despite two-thirds of bodily iron being stored in erythrocytes.

The concept of RBC storage being “ferroptosis-like” has substantial additional support from the specific QTL generated in both mice and humans in the current study. First, the current findings show that TP53 is associated with altered lipid peroxidation and hemolytic propensity of stored human blood. TP53 is known to both promote ferroptosis^52,71 and drive STEAP3 expression,^72,73 linking it to our new model of RBC storage. Second, numerous other QTL from the current studies or related REDS studies on the genetic determinants of RBC hemolytic propensity^14,74 are genes already validated as key regulators of ferroptosis, including EPHX2,⁷⁵ G6PD,⁷⁶ LPCAT3,⁷⁷ GPX4,⁷⁸ FADS1/2^79,80, ELOVLs,^79,80 phospholipase A2 group VI,⁸¹ SLC22A16,⁴⁹ and GSH biosynthesis.⁸² It is noted that the genes identified in J:DO mice and human cohorts are not entirely overlapping. As always, this may reflect differences in the species. However, it is essential to consider that a genome-wide association study cannot identify a gene as being involved in a process unless the gene varies appropriately in the cohort being studied. As such, the absence of a QTL in a particular gene does not indicate that gene is not involved, unless it is clear that appropriate variants of the gene are present in sufficient numbers. Overall, the combined associations of numerous genes known to regulate ferroptosis with lipid peroxidation (a hallmark of ferroptosis) in human RBCs provides extensive evidence that the RBC storage lesion has numerous similarities to classic ferroptosis. As RBCs lack nuclei and other processes found in nucleated cells, the observation of “ferroptosis-like” mechanisms as drivers of RBC lipid peroxidation and increased extravascular hemolysis posttransfusion is novel, in that it reveals that ferroptosis or “death by iron-dependent lipid peroxidation” is a biochemically driven process that may occur independently from regulatory events that require de novo gene expression or even the presence of mitochondria.

Understanding that the storage lesion process is a ferroptotic one is not a mere matter of semantics, rather it has direct medical relevance. Multiple pharmacologic or dietary interventions can trigger (eg, erastin⁸³), mitigate, or even prevent ferroptosis in other systems, including iron chelators, inhibitors of lipid peroxidation or lipophilic antioxidant,^84,85 or dietary supplementation of omega-3 or deuterium-labeled⁸⁶ fatty acid–enriched diets. The efficacy of these drugs and interventions in the context of blood storage remains untested and holds the potential to extend the shelf-life, quality, and posttransfusion performances of packed RBC products, as well as potentially paving the way for new treatment strategies for hemolytic disorders. Finally, with the advent of the era of precision transfusion medicine, the findings reported herein can inform on genetic screening for alleles (eg, STEAP3, TP53, EPHX2, and LPCAT3) that are associated with poorer/improved storage quality, with an observed loss of potency up to 30% when units from donors carrying these polymorphisms are stored longer than 3 weeks. As such, our findings may inform blood management strategies to tailor the shelf-life of packed RBCs, to guide priority in blood unit issuing based on genetically informed prediction of storability rather than first-in/first-out strategies. Providing units known to circulate longer and thereby requiring fewer transfusions will be of critical benefit to patients for whom the number of transfusions is associated with medical sequelae (eg, alloimmunization in patients with sickle cell disease and iron overload in patients with thalassemia). Although not a formal clinical trial comparing RBC units screened by such measures, the observation from the “vein-to-vein” data of increased posttransfusion circulation of RBCs from donors with variants in the same genes identified in the presented QTL provides direct translational evidence to predict the efficacy of such an approach.

Conflict-of-interest disclosure: A.D., K.C.H., and T.N. are founders of Omix Technologies Inc. A.D., S.L.S., and T.N. are scientific advisory board (SAB) members for Hemanext Inc. A.D. is a SAB member for Macopharma Inc. S.L.S. is a SAB member for Alcor, Inc, and consultant for Tioma, Inc, and Team Conveyer Intellectual Properties, serves as executive director for Worldwide Initiative for Rh Disease Eradication and as CEO for Ferrous Wheel Consultants, LLC. J.C.Z. is a founder of Svalinn Therapeutics. B.R.S. is an inventor on patents and patent applications involving ferroptosis; cofounded and serves as a consultant to ProJenX, Inc, and Exarta Therapeutics; holds equity in Sonata Therapeutics; and serves as a consultant to Weatherwax Biotechnologies Corporation and Akin Gump Strauss Hauer & Feld LLP. The remaining authors declare no competing financial interests.