Deep Coverage Mouse Red Blood Cell Proteome

Mice have close genetic/physiological relationships to humans, breed rapidly, and can be genetically modified, making them the most used mammal in biomedical research. Because the red blood cell (RBC) is the sole gas transporter in vertebrates, diseases of the RBC are frequently severe; much research has therefore focused on RBC and cardiovascular disorders of mouse and humans. RBCs also host malaria parasites. Recently we presented an in-depth proteome for the human RBC. Here we present directly comparable data for the mouse RBC as membrane-only, soluble-only, and combined membrane-bound/soluble proteomes (comprising, respectively, 247, 232, and 165 proteins). All proteins were identified, validated, and categorized in terms of subcellular localization, protein family, and function, and in comparison with the human RBC, were classified as orthologs, family-related, or unique. Splice isoforms were identified, and polypeptides migrating with anomalous apparent molecular weights were grouped into putatively ubiquitinated or partially degraded complexes. Overall there was close concordance between mouse and human proteomes, confirming the unexpected RBC complexity. Several novel findings in the human proteome have been confirmed here. This comparison sheds light on several open issues in RBC biology and provides a departure point for more comprehensive understanding of RBC function.

chosen for the first mouse genome sequence (1) and for that reason has been used here. Mice share many characteristics with humans, and many of the strains and lines develop or are susceptible to diseases of the immune, endocrine, nervous, cardiovascular, and skeletal systems that have important similarities to human disease. The red blood cell (RBC) 1 is one of the most important cell populations in any mammalian organism and has therefore been an object of intensive investigation for many years. Mouse RBCs have served as models for study of hemolysis (2,3), infectious diseases (4 -8), and different forms of anemia (9 -13) since the early 1950s. In the sixties and seventies, the establishment of routine in vitro bone marrow cultures (14) opened investigation of hematopoiesis (15,16), blood cell differentiation (17)(18)(19), and abnormalities thereof (19,20). At the same time interest developed in components of the immune system (21)(22)(23), cancer-related phenomena such as erythroleukemia (24 -26), and the use of irradiated mouse models to explore the role of RBCs and other blood components in transplantation (27,28). Continuous improvement in biochemical, imaging, and genetic techniques has ensured that the mouse remains the most used biomedical animal model. The availability of high accuracy and high sensitivity mass spectrometry techniques offers, for the first time, the prospect of deriving in-depth proteomes for critical cell types. Because of the experimental importance of the mouse we felt it critical to undertake a comparison of a deep coverage mouse RBC proteome with the recently derived deep coverage human (29) RBC proteome. To our knowledge this is not only the first global analysis of mouse membrane and soluble RBC proteins but also the first comparative RBC proteomics analysis in which samples were prepared and analyzed under identical conditions and data were validated according to consistent criteria. preclude citation of all relevant studies; as might be expected findings were sometimes contradictory.
Validation by MSQuant (version 1.4.1; official release, June 3, 2007), an open source software developed in our laboratory at the Centre for Experimental Bioinformatics (University of Southern Denmark), provided a manual score and spectrum evaluation for each of the peptides that had led to the identification of a given protein. The same stringent criteria were applied as for the human RBC proteome (29). Proteins were then BLAST searched (all versus all; cutoff 95% to remove redundancy). Swiss-Prot/TrEMBL (36), Ensembl, and Gene Ontology databases (37) were used for annotation. Unique Swiss-Prot/TrEMBL/Ensembl numbers provided access to sequence, isoform, family, localization, and function data for identified proteins; in parallel IPI numbers were queried against the Gene Ontology database (37), enabling grouping by class, function, or localization and quantitation within grouping (for example within the different signal transduction pathways).
An all versus all BLAST search of validated proteins was done to eliminate redundancy, providing final membrane, soluble, and mixed fraction protein lists (supplemental material). In particular, where proteins were identified only with only one method (output for each method is typically the result of three runs of pooled RBCs prepared and analyzed using the same conditions), annotation databases and the literature were extensively used to define proteins as genuine red cell components or as probable contaminants from other blood sources.

RESULTS
In material used for analyses, two WBCs, two reticulocytes, and five platelets were maximally detected per 10 6 RBCs; granulocytes and monocytes were undetectable. Our combined purification procedures therefore reduced contaminating cell types by at least 1000-fold such that we present data for an essentially pure RBC proteome. This is confirmed by the fact that high copy number molecules CD45 (leukocytespecific) (38), transferrin receptor (reticulocytes, lost in exosomes during maturation) (39), and ferritin receptor (40) were not in our final protein list.
The study goal was to obtain the fullest possible proteome, including low abundance proteins; these were identified using strategies we first described for the human RBC proteome (29). Due to repeat detection, highly abundant proteins such as Band 3 and spectrin (for mice 25 and 27% of the membrane proteome, respectively (41)) can provide a dynamic range challenge that reduces rare protein detection. For this and other reasons, we used parallel approaches, utilizing both the LTQ-FT and QSTAR for MS. The LTQ-FT provides rapid cycling, high mass accuracy (FT), high fragmentation speed (LTQ), and an additional fragmentation step (MS 3 ) in the ion trap, whereas the QSTAR affords better quantitation statistics. For the QSTAR, exclusion lists were created. As for the human RBC proteome, LTQ-FT MS significantly enhanced the dynamic range, allowing us to identify more proteins than previously possible, including extremely low abundance extracellular RBC-binding proteins such as the complementrelated factors complement C4 precursor, complement C1Q, and subcomponent subunit A.

Comparative Analysis of Mouse and Human RBC Proteomes
Membrane Proteins-40 membrane proteins (93 including proteins in the membrane/soluble fraction) were identified on initial MS analysis of digested crude RBC membranes (Table  I). Because membrane proteins are often (partially) shielded from tryptic digestion by lipids and may occur as either integral or membrane-associated proteins, we assessed procedures to increase the number of detectable proteins. In contrast to human RBC membranes, ethanol solubilization/ precipitation to remove lipids (42) decreased the proteins found (31 proteins; 72 including proteins in the membrane/ soluble fraction). 16 proteins (47 including proteins in the membrane/soluble fraction) were found in both EtOH-treated and crude samples; G-proteins and proteasome subunits were almost exclusively present in the crude sample only.
To improve detection/differentiation between membraneassociated and integral membrane proteins, we combined EtOH treatment with sodium carbonate extraction at two concentrations (43). We compared EtOH solubilization and calcium carbonate extraction alone and combined. As for human RBC membrane samples, fewer proteins were detected using saturated carbonate procedures; loosely associated membrane proteins (e.g. Rab) disappeared with more intense carbonate extraction (mouse membrane only; supplemental material). The fewest proteins were identified using double carbonate extraction; numbers of known integral membrane proteins were stable independent of treatment suggesting that many mouse RBC membrane proteins are membraneassociated rather than integral (see Band 3, stomatin, and urea transporter in mouse membrane only; supplemental material). The abundant, tightly membrane-associated 55-kDa erythrocyte membrane protein (44) followed this trend. Consistent with our observations on human RBCs, glycosylphos-phatidylinositol-anchored proteins such as CD59 glycoprotein and ADP-ribosyltransferase 4 showed constant peptide numbers across all carbonate extractions. SDS-PAGE fractionation (gel slicing) improved detection to 133 membrane proteins (278 including proteins in the membrane/soluble fraction) (Table I) and gave a molecular size window.
Extracellular, Membrane-associated Proteins-19 membrane proteins were probably of extracellular origin. Among these serum albumin (binds also to WBCs (45)(46)(47)) and clusterin (complement inhibition (48)) were common to human and mouse data sets. Cathepsins E (49) and B (50) were found here; Cathepsin G (51) was found on human RBCs (29). RBCs interact with plasma lipoproteins through apolipoprotein B, allowing RBC membrane remodeling (52). L-apolipoproteins are thought to be absent in rodents (53); however, we detected a widely expressed (heart, endothelial cells, and bone marrow) mouse protein similar to apolipoprotein L (Swiss-Prot accession number Q8BUC6) that is believed to affect lipid movement and/or allow lipid binding to organelles. Some low abundance proteins we identified are expressed by endothelial cells (such as phosphoribosyl-pyrophosphate synthetaseassociated (54) and claudin-13 (55) and are attributed (e.g. by Swiss-Prot) to extracellular space. Although minor endothelial cell contamination cannot be excluded, RBCs express a phosphoribosyl-pyrophosphate synthetase (56). In other cells (57) these complex with phosphoribosyl-pyrophosphate synthetase-associated protein, suggesting that similar association may occur in RBC membrane. SPARC (secreted protein acidic and rich in cysteine) (which is secreted by megakaryocytes with a possible role in hematopoiesis (58)) and collagen (binds to CD36 (59) and present both on erythrocytes and platelets) were also identified as was prosaposin (secreted by liver; found in various organs including nervous system (60) and may transfer gangliosides from liposomes to erythrocyte ghost membranes (61)). In contrast to human RBC binding of covalent C3b2-IgG complexes (62,63), in mice elements of the classical complement pathway were evident.
Soluble Proteins-Soluble proteins were analyzed three times by LTQ-FT following SDS-PAGE fractionation. To maximize low abundance hits, sensitivity was increased by filling the ion trap to capacity with one of five mass ranges in turn. Although mouse RBCs have fewer membrane proteins than human RBCs, the soluble proteome is roughly the same size.
Detection and Validation, Including Isoforms and Protein Families-To confirm that multiple members of the same protein family were present, unique peptides were identified wherever possible. Peptide spectra (scoring over 30) were checked for correct attribution of amino acid sequence, leading to unequivocal attribution of splice isoforms for seven of 20 membrane proteins, six of 17 soluble proteins, and four of 11 mixed fraction proteins. For example, splice isoform 2 of Q61469 lipid phosphate phosphohydrolase 1 shared peptides NYSTNHEP and EEDPHT-TLHETASSR with isoform 1 but had peptides YPYHDSTIPSR and GFFCTDNSVK, both unique to isoform 2 ( Fig. 1, A and B). a For the mouse a distinction is made between proteins unique to the membrane fraction and proteins found in both membrane and soluble fractions.

FIG. 1. Protein isoforms in RBC membranes.
A, summary of identified splice isoforms with peptides that permitted unequivocal attribution to a specific isoform in bold and the further peptides identified in regular font. B, a partial ClustalW alignment for splice isoforms of lipid phosphate phosphohydrolase 1 (LPP1) proteins shown as an example of the way by which the presence of specific peptides unique to an isoform was ensured. *, identical amino acid residue; :, synonymous amino acid replacement; ., non-synonymous amino acid replacement; empty space, completely different amino acids.
For cytoskeleton components, isoforms/members of the main protein families were similar in human and mouse proteomes. Thus the same ankyrin and spectrin isoforms were unequivocally identified in both RBC proteomes (Table II). Similar profiles emerged for tropomyosin and actin, although tropomyosin 2 was unique to mouse RBC. Tropomyosins interact in an isoform/member-dependent manner with tropomodulin; defects in expression can result in RBC shape abnormalities (64). Actin cytoplasmic 2 (␥-actin) appears to be a mouse, but not human, mature RBC protein. Increased ␥-actin in human RBCs (induced by high circulating glucose levels) may pathologically reduce RBC deformability (65). Glucose effects on mouse RBC may consequently differ from human RBC. Adducins ␣ and ␤ occur in both proteomes; adducin ␥ was detected only in mouse RBC. ␤-Adducin knock-out mice show compensatory overexpression of ␥-adducin, leading to spherocytic hereditary elliptocytosis (66). In the analysis of the human RBC tubulin family members only ␣-3 could be identified with certainty; this is missing in the mouse RBC proteome. Due to the considerable identity in amino acid stretches between tubulin ␤-5/␤-1 and ␣-6/␣-1, peptides from these RBC proteins could not be attributed unequivocally to one or the other tubulin family member in humans, whereas attribution was possible in the mouse where tubulin ␤-5 and ␣-1 are definitely present. Interestingly ExPASy (36), Ensembl, and PRIDE (Proteomics Identifications database) (67) only contain a "similar to tubulin ␤-1" sequence that differs significantly from mouse tubulin ␤-5 (Table II).
There is debate as to whether human erythrocytes possess actin capping protein. In accord with published reports (68, 69), we detected F-actin capping protein in both human (␣ and ␤ subunits; unique to membrane fraction) and mouse (␣ subunit only; membrane and soluble fractions) RBCs. Thus, combined evidence suggests that such a protein is present in RBCs, although uncertainty concerning state of activity/role remains (70,71).
Human/Mouse Comparison: Orthologs and Family-related Proteins-Having eliminated redundancy, final human and mouse RBC protein lists were compared with identify orthologs and family-related proteins. The mouse proteome comprising uniquely membrane (MM; 247 proteins), uniquely soluble (mouse soluble; 256 proteins), and both membrane and soluble (MM&S; 165 proteins) sets was compared with the human RBC proteome comprising uniquely membrane (HM; 341 proteins) and uniquely soluble (HS; 252 proteins) protein sets (Fig. 2). Determination of commonality/uniqueness and evolutionary relationships used a BLAST homology algorithm based on National Center for Biotechnology Information (NCBI) BLAST.exe and algorithms by Bork and coworkers (72)(73)(74). As no algorithm reliably defines orthologs, we used orthology/paralogy predictions generated by the European Bioinformatic Institute through a pipeline where maximum likelihood phylogenetic gene trees (generated by an algorithm to estimate large PHYlogenies by Maximum Likelihood (PHYML)) play a central role (75)(76)(77). This reconciles gene trees with species trees (using RAP), annotating internal nodes distinguishing duplication/speciation events, to repre- Actin, ␣ skeletal muscle Actin, skeletal muscle (␣-actin1) Ortholog one2one Actin, cytoplasmic 1 Actin, cytoplasmic 1 (␤-actin1) Adducins Ortholog one2one sent an evolutionary history for gene families. Although there is a clear concordance with reciprocal best approaches in the simple case of unique orthologous genes, this approach also finds more complex one-to-many and many-to-many relationships. In this study, proteins were defined as orthologs when homology BLAST returned Ͼ80% identity and the algorithm tree of Bork and co-workers (72-74) showed a common precursor or when EBI predictions indicated that two proteins were orthologs. These combined approaches returned orthologs as follows: 56

Annotation
To render human and mouse data sets as comparable as possible the final protein lists were evaluated in depth using all available literature and databases. Because single proteins may have multiple functions/locations, little or no information may be available for a given protein, and databases are dynamic and may not recognize submitted queries (different accession numbers or protein descriptions), all annotations were manually reviewed using consistent criteria, ensuring maximal comparability between human and mouse proteomes. A Gene Ontology (37) annotation of some sort was found for 535 of 668 proteins detected; proteins were classified as involved in cellular (188 proteins), physiological (183 proteins), or regulatory (34 proteins) processes. 130 were unclassified.
Membrane Components-To enable comparison, all MM proteins and HM orthologs in the MM&S fraction were treated as a single set (subcellular localization is shown in Fig. 3). For six proteins with double annotation both localizations were attributed. Visual inspection of equally loaded SDS-PAGE gels suggested the mouse membrane RBC proteome to be less complex than the human equivalent (Fig. 4); our annotation confirms this. Extending the analysis to include MM&Sunique and mouse soluble proteins with HM orthologs does not alter the finding. In fact, most proteins in these two sets are cytoplasmic, cytoskeletal, or organellar; only five of 61 such proteins are classified as integral membrane proteins.
Mouse and human RBC membranes appear to differ substantially in the number of integral membrane proteins and the degree and tightness of association of cytosolic proteins (e.g. glycolytic enzymes and carbonic anhydrase II) to the membrane network. Fewer proteins involved in glycolysis were associated with mouse membranes (three proteins) in comparison with human membrane (eight proteins) ( Table III). As for human RBC, with the exception of glyceraldehyde dehydrogenase, cytoskeleton removal protocols not only resulted in fewer peptides from cytoskeletal and cytoskeleton-asso-

FIG. 2. Mouse/human RBC protein orthologs and localization.
Shown are Venn diagrams of mouse RBC proteins grouped on the basis of their relationship to human RBC proteins as orthologs and familyrelated and unique proteins. The human RBC proteome comprises 341 membrane and 252 soluble proteins; the mouse RBC proteome comprises 247 uniquely membrane, 256 uniquely soluble, and 165 both membrane and soluble proteins. An all versus all comparison was carried out in which orthology was established using a BLAST algorithm to determine homology/identity, different algorithms by Bork and coworkers (72)(73)(74) to determine evolutionary relationships between proteins from the two species, and a mouse-human orthology list as predicted by EBI. As a result we were able to determine which proteins in mouse membrane, soluble, or membrane and soluble fractions were orthologous or family-related to proteins that had been identified previously in specific human RBC fractions. In the diagram, the human membrane protein fraction is shown in orange, and the human cytosol fraction is shown in blue. In white circles we report the number of shared orthologs (BLAST Ͼ 85%), EBI-predicted orthologs, and common ancestor according to the algorithms by Bork and co-workers (72-74); proteins shown not circled are family-related proteins (potential paralogs, BLAST Ͼ 50%, and common tree branch (Bork and co-workers (72)(73)(74)). At the intersection the number of unique proteins in the specific mouse fraction is recorded. ciated proteins but also reduced hits for almost all glycolytic proteins. This is in accord with data from various mouse strains showing that inward facing membrane-associated enzymes function by carrying the organization of the membrane into the cell interior and that few glycolytic enzymes known to stably bind to the human RBC membrane do so in mice (78,41).
As for the human RBC proteome, most membrane proteins are involved in binding (156 proteins) and have catalytic activity (112 proteins). Many show transporter (67 proteins), signal transducer (37 proteins), or structural activity (22 proteins). Transport (Fig. 5A) and metabolic (Fig. 5B) activities are diverse. As for the human proteome, probable reticulocyte legacies such as intracellular transport proteins of the Golgi, endoplasmic reticulum, and mitochondrion were detected. This supports the concept of a scheduled RBC protein degradation process for RBC maturation and aging that is con-served among mammals. An ortholog for vesicle trafficking protein SEC22b identified in the human proteome was absent in mouse RBC. However, vesicle-associated membrane protein 5 present in the mouse-unique set is a protein with very similar characteristics: it belongs to the synaptobrevin family and has the same trafficking function (79 -81). Thus although direct human-mouse orthologs are not always present, functionally equivalent proteins of different origin may perform similar tasks in RBCs of the two species.
Although RBCs are devoid of nuclei, three proteins annotated to have transcription regulator activity (two proteins) and translation regulator activity (one protein) were identified. Closely similar sets of organellar proteins occurred in both proteomes. Some migrated at anomalously high molecular weight (elongation factor A protein 1 and eukaryotic translation initiation factor 2C 2) and may be inactive (see below). Some may have roles in mature RBCs other than so far ascribed. Ribosomal proteins S19 (MM) and S27 (HM) and 40 S ribosomal proteins S3 (mouse soluble) and S6 (HS) may provide one example. rpS3 and rpS6 interact with heat shock protein 90 (Hsp-90) thereby preventing ubiquitination and proteasome-dependent degradation (82). A further regulatory mechanism involves Hsp-70, which associates with free rpS3 promoting its degradation (83). Both chaperones occur in mature red blood cells (83) and were found in this study.
We analyzed regulation of cellular processes, physiological processes, and enzyme activities. For cellular processes, as for human RBC, the most common regulatory activity involves programmed cell death, presumably a legacy of prior development (84), followed by involvement in regulation of transport and signal transduction. 12 proteins are involved in general signal transduction, seven are linked to receptor signaling pathways, and 24 have a role in the intracellular signaling cascade. Three proteins are ascribed by databases as being part of a phosphorelay (two-component signal transduction system). Of proteins involved in surface receptor-linked signal transduction, five belong to the G-protein-coupled receptor protein signaling pathway, two belong to the acetylcholine receptor signaling muscarinic pathway, and one belongs to the integrin-mediated signaling pathway; the intracellular signaling cascade is itself divided into protein kinases (four pro- Mouse RBC Proteome: a Comparison with the Human RBC teins), second messengers (one protein), and small GTPases (19 proteins). Such a high representation of signal transduction proteins seems unlikely to be purely a reticulocyte legacy and is in line with data for the human RBC (29). The predominance of small GTPases agrees with data showing that the two ␣ subunits of G s and G i and the Rab proteins function together with an unknown G-protein (85); our data suggest this may be the ␣ inhibiting activity polypeptide 2. Further to cellular processes, we identified proteins regulating cell shape (ezrin, moesin, and phospholipid scramblase 3), cell adhesion (two proteins), complement activation (two proteins), cell volume (one protein), and cell redox homeostasis (one protein). Amyloid ␤ A4 protein, which appears to be involved in a great variety of processes, was present in both mouse and human RBCs.
Diverse mechanisms/proteins including vascular processes, regulating blood vessel size, and coagulation fall under physiological regulatory processes. Protein sets involved in these processes are very similar to those identified in the human RBC proteome, including those involved in nitric oxide  (86,87). These comprise inhibitory G-proteins (guanine nucleotide-binding protein G i and ␣ inhibiting activity polypeptide 2), which block nitric oxide-mediated ATP release from RBCs (88) (and have a further role in regulation of the AQP1 water channel (89)) and Hsp-90, which modulates nitric oxide biosynthesis. RBC nitric-oxide synthase (NOS) has been reported to be membrane-bound (P-face) (90,91), but reports disagree on the nature of the NOS. Although Kleinbongard et al. (90) described it as a 140-kDa endothelial NOS (eNOS) Bhattacharya et al. (91) described it as an insulinactivated NOS comprising heavy (135 kDa) and light chains (95 kDa). Analysis of our data for evidence of NOS expression showed low confidence-scoring peptides mapping to an inducible NOS (human, FDVVPLVLQANGR; mouse, VVF-FASMLMR) in the soluble mouse and human RBC proteomes. Initially excluded as low level contaminants, it may be that RBC eNOS diverges from eNOS as annotated in the databases and shows partial similarity to inducible NOS and that this 135-kDa protein is the heavy subunit of the insulinactivated NOS reported by Bhattacharya et al. (91). By analogy with protein-tyrosine phosphatase receptor type C-associated protein (PTPRC) shown to have a role in the insulin-mediated activation of NOS in monocytes (92) and endothelial cells (93), the low abundance RBC membranebased receptor-type tyrosine-protein phosphatase ␣ we identified in both proteomes would be the insulin-reactive 95-kDa light chain.
Soluble Components-To enable comparison, all mouse soluble proteins and HS orthologs in the MM&S fraction were treated as a single set. Of 292 unique soluble proteins identified, 228 were unequivocally annotated as cytoplasmic, 10 were unequivocally annotated as cytoplasmic and nuclear, eight were unequivocally annotated as Golgi and/or endoplasmic reticulum, and seven were unequivocally annotated as mitochondrial and/or cytoplasmic (Fig. 6). Those annotated as both cytoplasmic and organellar appeared to have trans- porter roles, accounting for the distribution. Although different numbers of total soluble proteins were identified for mouse and human RBCs (292 and 252, respectively) more proteins appear to be metabolized in the mouse (47 and 20, respectively) making the difference between the two proteomes (245 and 232, respectively) much less significant.
As for the human proteome, most soluble proteins were implicated in cellular metabolism and/or transport; prevalent functions were proteolysis (17 proteins), ATPase (eight proteins), and dehydrogenase (34 proteins) activity. Most proteins involved in metabolic processes were catabolic (27 proteins); this is not surprising given the RBC life history and is concordant with what was seen for human RBC. A further 17 proteins were involved in household metabolism (GSH, nucleotide (ATP and ADP), and glucose) and in macromolecule metabolism, whereas only a few proteins were involved in cellular biosynthesis (five proteins). The complex with most members was the proteasome (17 proteins). RBC household complexes (ubiquitin ligase complex, 6-fructokinase complex, Hb complex, and phosphopyruvate hydratase complex) were also present along with complexes probably resulting from incomplete degradation of molecules no longer functionally required (RNA complex, DNA factor A complex, nucleosome, and ribonucleoprotein complex). Although household complexes migrated at the expected molecular weight, residual complexes did not (as defined by UniProt (36)). Some residual complexes (e.g. nucleosome and ribonucleoprotein complex) were detected in both mouse and human RBCs, suggesting that common catabolic pathways and schedules for RBC maturation exist.
Not All Proteins Are Present in Their Active Form-Although our preparation methods focused on purification of mature RBCs, the protein content is likely to vary with cell age; RBCs may be considered as "in development" because degradation of organelles occurs during maturation, and chemical and enzymatic modifications occur during cellular aging. "Residual" proteins may therefore be detected, and it becomes important to know their state. This thesis, first presented in analysis of our human RBC proteome, is supported by the mouse RBC proteome data. As before, we compared expected and observed molecular weights for proteins in question to obtain information on their status (Fig. 7). Of 192 proteins identified after in-gel separation and digestion, 16 migrated faster than expected, and 38 migrated slower than expected in a gel fraction also containing ubiquitin. High apparent molecular weights may reflect ubiquitination (known for spectrin (94)) and/or oligomerization in aging RBCs (95). Several proteins migrated anomalously in both MM and HM (78-kDa glucose-regulated protein precursor, elongation factor 1-␣ 1, importin 9, 14-3-3 protein family, transforming protein p21, ATP-binding cassette subfamily B, member 6, and Hsp-90). We hypothesized (29) that the molecular weight shift may be attributable to complexes resulting from reductioninsensitive protein-protein interactions. Although the molecu-lar weight shift is often similar, indicating a common degradation plan (strategy and timing), exceptions occur, allowing us to exclude coincidental behavior due to the in-gel migration properties of the different proteins. Thus, the 78-kDa glucose-regulated protein precursor is degraded in mice (lower molecular weight) and probably ubiquitinated in humans (higher molecular weight); both 14-3-3 proteins migrate at high apparent molecular weight with ubiquitin in mice; in human RBCs one protein is probably ubiquitinated, and the other is degraded. Control proteins occurred in fractions corresponding to the expected molecular weight, whereas Band 3, spectrins, and other cytoskeletal proteins occurred at several molecular weights as reported previously (96). Of soluble proteins also analyzed for in-gel migration, 35 were unique to the mouse RBC proteome; 23 migrated slower than expected in fractions also containing ubiquitin, 19 migrated faster than expected, and 197 were found at the expected molecular weight (soluble protein not at molecular weight/soluble proteins at molecular weight; supplemental material).
Although mature RBCs are devoid of both nucleus and internal organelles, organellar proteins common to both human and mouse RBCs were identified usually in the same metabolic state. Golgi proteins (nicastrin, adaptor protein complex, and copper-transporting ATPase), ribosomal proteins (40 S ribosomal proteins S6 and S3), and some endoplasmic reticulum proteins (ERp29, reticulon 3, and proteindisulfide isomerase A3) migrate at expected molecular weights. In contrast, most nuclear (nucleosome assembly proteins 1-like 1 and 1-like 4 and elongation factor 1) and mitochondrial proteins (ATP-binding cassette subunit B, member 6; ATP synthase ␤ chain; malate dehydrogenase; and protein NipSnap 2) appear to be degraded or ubiquitinated/polyubiquitinated. However Ras-related proteins appeared predominantly degraded in the human RBC but at the expected molecular weight in the mouse, and calreticulin FIG. 7. Mouse membrane gel with molecular weight indicator. Shown is a typical mouse (Mr) membrane SDS-PAGE gel with the relative size makers (M) giving an example of a few proteins that presented an apparent molecular weight different from the expected molecular weight. appears to be polyubiquitinated in human RBC but at the expected molecular weight in mouse RBCs (supplemental material).

DISCUSSION
Advanced LC-MS technology, coupled with biochemical procedures for sample preparation and new bioinformatics tools, has been used to derive a normal mouse RBC proteome 72-96 h after blood collection. Using stringent purification procedures to isolate RBCs and detection limits of about 500 copies of protein/cell we present the most complete analysis of the mouse RBC proteome undertaken to date. To provide for comparability with our recently published in-depth analysis of the human RBC proteome (29), we ensured that the harvesting of the RBCs, their downstream processing (biochemical procedures), the type of MS analysis (MS machines, time, and HPLC gradients), and the validation parameters were fully equivalent between the two data sets. Thus, for example, we decided to maintain use of the Percoll-based reticulocyte depletion protocol used by Vettore et al. (30) in both studies rather than switch to implement a novel Percoll-based method (97) midway between the studies. Although this novel method is reported to be superior to fluorescence-activated cell sorting and magnetic cell sorting depletion, it was not compared with existing Percoll-based methodologies including that of Vettore et al. (30), and there are to date no reports of its use in proteomics study. Using this conservative approach has allowed an in-depth comparison of human and mouse RBC components, a comparison only possible when components of both proteomes can be unequivocally identified. A recent minireview (98) provides an overview of all human RBC proteins found to date by different authors. Unfortunately for the purposes of comparison between our findings and those reported in this review, this review did not incorporate accession numbers nor were the proteomics methodologies leading to protein identification described in sufficient detail to make comparison with our list feasible.
Using these conservative approaches and with the caveat that, due to the nature of the study, inclusion of a small number of false positive identifications or incorrect assignment of contaminating proteins cannot be precluded, after validation 247 solely membrane and 232 solely soluble proteins were identified; 167 proteins were in both membrane and soluble fractions. Proteins were scanned for likely physiological role(s) in RBC, for subcellular localization, and molecular function. Isoforms were critically screened for unique peptides, and all versus all BLAST removed redundancy from final lists. Proteins were subsequently assessed for probable metabolic status by comparing expected with apparent molecular weight. Anomalous migration may be expected in reticulocyte legacy proteins; migration faster than expected was taken to indicate degradation, whereas co-migration with ubiquitins at elevated molecular weight was suggestive of non-functional protein incompletely proteolyzed by protea-somes. At 60 days, the life span of mouse RBC is half that of human RBC (99), and mouse blood has a higher reticulocyte content than human blood. Taken together these observations, which are concordant with the high red blood cell turnover rate (100) and higher overall metabolism in the mouse, explain the higher content of metabolized residual proteins (at different molecular weight than expected) found in the mouse soluble RBC fraction. The higher reticulocyte content of mouse blood leads to a relatively younger population of mature RBCs in mice (101,102) than in humans. Such RBCs are likely to contain more reticulocyte legacy proteins that have not (yet) been fully degraded and, as shown here, are therefore revealed by MS.
Using algorithms to identify orthologs and family-related and unique proteins, a comparison of human and mouse RBC proteomes revealed many orthologs likely to share function. Where direct human-mouse orthologs were missing, proteins could be identified for which related functions have been reported. These may function similarly in the two RBC species.
Strikingly protein isoforms of the cytoskeletal network were frequently common to human and mouse; differences were consistent with prior literature. Although RBCs lack organelles, the same organelle proteins occur in human and mouse RBCs. Apart from rare exceptions these are in the same metabolic state, suggesting that some organellar proteins may have (short lived) relevance even beyond the reticulocyte stage and share a common plan/schedule for metabolic degradation. Our findings underpin the hypothesis (29) that maturation of RBCs from reticulocytes and aging of RBCs are ongoing processes that continue their whole life span. The RBC thus appears to be a dynamic blood component, and with these data further biochemical studies can be directed to unlocking further surprises in RBC behavior. The number of membrane-associated proteins described in this study suggests that this may be especially true in the context of their interplay with other cells and plasma proteins (103)(104)(105)(106). A detailed summary of all proteins, their metabolic form, their isoforms, and most relevant peptides found and an ortholog table is available as supplemental material; a database is also being created.