Phylogeny and evolution of erythrocytes in mammals

Vertebrates (with a few derived exceptions) use haemoglobin (Hb), carried inside red blood cells (RBCs), for the transport of oxygen from respiratory organs to tissues, as well as for the clearing of carbon dioxide from tissues back to respiratory organs. Because of the role of blood in delivering oxygen, endotherms (mammals and birds) show higher values of haematocrit (Hct, concentration of RBCs in the blood) and concentrations of Hb than ectotherms (Hawkey et al., 1991). However, within classes, values of Hct (25.7±6% in reptiles: Hawkey et al., 1991; 45.7±5% in birds: Hawkey et al., 1991; 43.4±6% in mammals, present data) and Hb (8.5±27% in reptiles: Hawkey et al., 1991; 15.3±2% in birds: Hawkey et al., 1991; 14.6±2% in mammals, present data) are quite constant, which is thought to reflect a balance between the advantage of increased oxygen capacity and the disadvantage of increased viscosity (Baskurt and Meiselman, 2013). However, it must be added that some mammals show important variations, e.g. the high Hct in bats and diving mammals.

As Hct is given by the product of the number of RBCs and their volume (MCV, mean corpuscular volume; Table 1), and assuming Hct is constant in a class, the number of RBCs is inversely correlated with MCV. This correlation was first recognized by Wintrobe (1933), using a few species, and later confirmed using much greater samples (Hawkey et al., 1991). Besides the uniformity of Hct and Hb in blood, inter-species comparisons show a great degree of variability for MCV (and, therefore, also for the number of RBCs). As early as the 19th century, Gulliver (1875), measuring RBC diameter, noted that this scales allometrically. This was later confirmed by some researchers (Hawkey et al., 1991; Promislow, 1991), but rejected by others (Emmons, 1927; Prothero, 1980; Schmidt-Nielsen, 1984). It must be added, beside the paucity of this kind of study, that almost all were done at a time when correlations were not corrected for phylogeny. The only exception is the work by Promislow (1991), who used the phylogenetic corrections of the time, and indeed found allometric relationships denied by previous authors. To this, I would like to add that Gulliver (1875) noted that large RBCs are characteristic not only of large mammals but also of what he called ‘primitive’ species (xenarthrans, aardvark, i.e. Orycteropus afer, marsupials).

Another parameter for which allometry was investigated is blood–oxygen affinity (which is measured through P₅₀, the P_O2 at which Hb is half-saturated; note that a high P₅₀ value means low affinity). Several authors discovered, as a general trend, that body mass is directly correlated with blood–oxygen affinity (Schmidt-Nielsen and Latimer, 1958; Scott et al., 1977; Poyart et al., 1992). However, it was suggested that this correlation is poor because (besides variation in within-species measurements and technique differences) species adapted to hypoxia (burrowers and high-altitude species) have affinities decisively higher than those expected by their mass (Scott et al., 1977). Indeed, the only study that investigated the allometry of Hb–oxygen affinity using phylogenetic correction found no correlation (Promislow, 1991). Moreover, feline and ruminant species have very low affinities (Scott et al., 1977). This seems to indicate that ecological and phylogenetic features are more important, for this parameter, than body mass. The high affinity of Hb for oxygen in species living in burrows or at high altitude is generally interpreted as a clear demonstration of evolutionary adaptation to hypoxia (Storz, 2007). However, Wells (1990) rejected this strictly adaptationist interpretation, and proposed that Hbs with high oxygen affinity are not the result of adaptive selection for hypoxic environments, but rather the result of other phenomena, such as ontogenetic and metabolic constraints, phylogenetic inertia, etc. Indeed, the high affinities of Andean camelids (llamas, vicunas, etc.), which are often used as an adaptationist argument, are shared also by lowlands camelids (camels and dromedaries) (Poyart et al., 1992). So, these were a prerequisite (preadaptation) that allowed camelids to colonize the Andes (Storz, 2007), and not the result of evolutionary adaptation. Another blow to the adaptationist view was the discovery that the snow leopard (living well above 4000 m) has a very low affinity, exactly like lowland felids (Janecka et al., 2015).

If studies on the correlation between body mass and blood parameters are few (and most of them did not consider statistical non-independence of species data), investigations on the correlation between the latter and metabolic rates are really rare. Based on the fact that (in general) small mammals have higher metabolic rates, a correlation between basal metabolic rate (BMR) and MCV and number of RBCs (Dunaway and Lewis, 1965; Bartels et al., 1969) or P₅₀ (Scott et al., 1977) was suggested, but not investigated. Only Promislow (1991) studied the relationships between BMR and MCV, number of RBCs and mean corpuscular haemoglobin (MCH), finding no correlation.

The aim of this paper was to study the correlations between mass and mass-specific basal metabolic rate (MS-BMR) and haematological parameters (using modern phylogenetically corrected analyses and a dataset with 441 species), their possible biological meaning, and how erythrocytes evolved in mammals.

Data for body mass, MS-BMR and haematological parameters were taken from published articles and databanks, i.e. ANAGE (Tacutu et al., 2018) and ISIS (2002). The whole dataset is shown in Table S1. Blood–oxygen affinity was recorded from whole-blood values at the normal body temperature of the selected species, at pH 7.4, a P_CO2 equal to 40 mmHg and a physiological level (thus not degraded from storage) of 2,3-diphosphoglycerate (DPG).

Correlations between body mass, MS-BMR and haematological parameters were performed on logarithmically transformed data using phylogenetically informed analyses (explained below). When a statistically significant (P<0.05) correlation was found, analysis of residuals was conducted (Speakman, 2005).

All analyses were investigated using a phylogeny-informed generalized least squares method (Freckleton, 2012) with BayesTraits software version 3.0 (https://www.evolution.reading.ac.uk/BayesTraitsV3/BayesTraitsV3.html). The generalized least squares (GLS) method was preferred instead of the reduced major axis (RMA) one because, although GLS makes the assumption that the error variation in the x-variable is zero (which is unrealistic), RMA makes the even less realistic assumption that the error variance in both variables (x and y) is equal (Speakman, 2005).

As all the investigated variables were not discrete, I used the ‘continuous method’, which employs a Markov chain Monte Carlo (MCMC) sampling algorithm, using a generalized least squares approach that assumes a Brownian motion model of evolution. In this model, non-independence among the species is accounted for by reference to a matrix of the expected covariance among species. This matrix is derived from the phylogenetic tree. The model estimates the variance of evolutionary change (the Brownian motion parameter) and the ancestral state of traits at the root of the tree.

For each analysis, one million generations were run, sampling every 1000 generations and discarding as burn-in the first 100,000. Acceptance rates for rate change parameters were confirmed to range within 20–40% to ensure proper mixing of MCMC chains.

BayesTraits also allows the reconstruction of ancestral nodes, during the analysis of simple or multiple regressions.

In order to estimate MS-BMR values of ancestral nodes, I performed a phylogenetically informed multiple regression (with body mass and temperature as independent variables) with BayesTraits. Besides data on body mass, temperature and MS-BMR of extant species, I also used values of body mass (including fossil calibration) and temperature at ancestral nodes reconstructed by other authors (Lovegrove, 2012; Slater, 2013; Baker et al., 2015; Álvarez et al., 2017), thus giving more power to the analysis. The obtained values of MS-BMR were subsequently used for the reconstruction of haematological parameters at ancestral nodes. Thus, MCV and Hb–oxygen affinity values of ancestral nodes were estimated by performing a phylogenetically informed regression (with MS-BMR as independent variable) with BayesTraits.

When two models were compared, the marginal likelihood of each model was estimated using the stepping stone sampler (Xie et al., 2011), employing 100 stones and running each stone for 10,000 iterations. Then, the models were compared by a likelihood ratio test, using the formula Bayes Factor=2(log marginal likelihood model a−log marginal likelihood model b) (Raftery, 1996). The tree used for phylogenetic analyses is taken from Slater (2013).

As explained in Materials and Methods, all the following correlations were found used phylogenetically informed analyses. This kind of analysis takes into account variations in a parameter due to phylogeny and can reveal how the correlation under scrutiny is valid inside the different clades.

As expected (de Magalhães et al., 2007), body mass and MS-BMR were significantly correlated (Fig. S1A; r²=0.7409, P<0.0001) and this regression was performed to enable subsequent residuals analyses. As revealed by phylogenetically informed analysis, Hct (Fig. 1A) was significantly correlated with body mass (P=0.0036), although the correlation was poor (r²=0.02). The same was observed between Hb and body mass (Fig. S1B; r²=0.013, P=0.0171). RBC count was significantly correlated with body mass (Fig. 1B; r²=0.395, P<0.0001). MCHC, instead, was not correlated with body mass (Fig. 1C; r²=0.0026, P=0.957). MCH was significantly correlated with body mass (Fig. 1D; r²=0.361, P<0.0001), and MCV was also significantly correlated with body mass (Fig. 1E; r²=0.367, P<0.0001).

Phylogenetically informed analysis revealed that Hct (Fig. 2A) was significantly correlated with MS-BMR (P=0.00001), but the correlation (as seen for body mass) was poor (r²=0.059). In contrast, the correlation between Hb and MS-BMR was not significant (Fig. 2B; r²=0.012, P=0.052). RBC count was significantly correlated with MS-BMR (Fig. 2C; r²=0.356, P<0.0001). MCHC, instead, was not correlated with MS-BMR (Fig. 2D; r²=0.01, P=0.07). MCH was significantly correlated with MS-BMR (Fig. 2E; r²=0.379, P<0.0001), and MCV was also significantly correlated with MS-BMR (Fig. 2F; r²=0.383, P<0.0001).

Residuals of Hct, RBC count, MCH and MCV after correlation with body mass were all significantly correlated (P<0.0001) with residuals of MS-BMR (Fig. S2). Conversely, residuals of Hct, RBC count, MCH and MCV after correlation with MS-BMR were not correlated (P=0.081, 0.081, 0.652 and 0.319, respectively) with residuals of body mass (Fig. S2). Thus, analysis of residuals revealed that all the above correlations between body mass and haematological parameters are in reality due to indirect correlation with MS-BMR.

As the link between MCV and mass-specific metabolic rate could be the accessibility of Hb to the surface area of the RBC, I tested the correlation between MS-BMR and the MCH/surface area ratio (MCH/SA). This correlation (Fig. 3A) was statistically significant (P<0.0001) and better correlated (r²=0.411) than that between MS-BMR and MCV (r²=0.383, Fig. 2F). The fact that MS-BMR is better correlated to MCH-SA than to MCV was also confirmed by a likelihood ratio test, giving log(Bayes Factor)=17.95 (very strong evidence). Once again, residual analyses showed that this correlation was independent from body mass (P=0.3631), whereas the correlation between body mass and MCH-SA was due to residual MS-BMR (P=0.0006; Fig. S3B,C). Therefore, my hypothesis was supported.

However, also in this case, aquatic species showed values of MCH/SA higher than those predicted by their mass-specific metabolic rates (Fig. 4A). This was also the case for fossorial and, in part, altitude-dwelling species (e.g. Lama glama, Cavia porcellus, Chionomys nivalis). In contrast, Ruminantia and Feliformia showed values of MCH/SA lower than those predicted by their mass-specific metabolic rates (Fig. 4C). It is interesting to note that fossorial and altitude-dwelling species have Hbs with high oxygen affinity, while Ruminantia and Feliformia have low-affinity Hbs (Scott et al., 1977).

Fossorial and altitude-dwelling species had P₅₀ values lower than those expected from their MS-BMR (Fig. 4B). Moreover, aquatic species also had lower than expected P₅₀ values, while Ruminantia and Feliformia had higher than expected P₅₀ values (Fig. 4D).

MCH/SA was also significantly correlated with P₅₀ (Fig. 3B; r²=0.406, P<0.0001) and this correlation was independent from MS-BMR (Fig. S3G; P=0.2802). Analysis of residuals revealed that the correlation between MCH/SA and MS-BMR was in reality due to correlation with P₅₀ (Fig. S3H; P<0.0001).

In order to reconstruct the evolution of RBC parameters, first of all I reconstructed the evolution of MS-BMR (Fig. S4). The result was similar to the one obtained by Avaria-Llautureo et al. (2019). This and the subsequent reconstructions obviously have a degree of uncertainty, which increases for the most ancient nodes. However, the interesting data to discuss are those relative to more recent nodes (ancestors of feliforms, carnivores, ruminants, camelids, etc.) rather than the most ancient ones. Using MS-BMR reconstructed values, the evolutionary tree of P₅₀ was reconstructed (Fig. 5). These values represent the oxygen affinity of whole blood (at the physiological temperature of each species), i.e. the affinity of Hb for oxygen in the presence of allosteric factors (which lower the affinity), above all by DPG.

In general, it can be seen that, during evolution, the increase of body mass (and decrease of MS-BMR) was accompanied by an increase in whole-blood–oxygen affinity (Fig. 5). However, this is not the case for the intrinsic affinity of Hb, which shows a phylogenetic signal independent from body mass and MS-BMR (Figs S4 and S5A,B). This suggests that in most species, the increase of body mass (and concomitant decrease of MS-BMR) was accompanied by a decrease of intraerythrocytic DPG, and a consequent increase of whole-blood–oxygen affinity. Indeed, DPG concentrations were significantly correlated with body mass and MS-BMR (r²=0.1, P=0.007 and r²=0.116, P=0.006, respectively; Fig. S5C,D). Nonetheless, these correlations were absent in Ruminantia and Feliformia. Indeed, Bunn (1980) put forward the hypothesis (which is still valid today) that the Hbs of the ancestors of these two suborders lost the response to DPG.

Analysis of residuals revealed that correlations between body mass and haematological parameters (Hct, RBC count, MCH and MCV) are in reality due to indirect correlation with MS-BMR. This phenomenon can be exemplified by the simple observation that in Xenarthra (with their very low metabolic rates), RBCs are among the largest in mammals (Fig. 1E); for example, a 6 kg Choloepus hoffmanni has a MCV of 125 fl. Conversely, the smallest RBCs are found in Eulipotyphla, which show the highest MS-BMR (Fig. 2F). Nonetheless, Tragulus javanicus shows even smaller RBCs (5.6 fl); this issue will be dealt with later.

The link between a high Hct and a high metabolic demand seems plausible, but although statistically significant, it shows a poor correlation. The highest Hct can bee seen in seals and bats (order Carnivora and Chiroptera, respectively; Fig. 2A; Table S1). However, in many species, the spleen can store and release significant amounts of RBCs, and under stress can increase the Hct up to 150% of the resting value (Turner and Hodgetts, 1959; Cabanac et al., 1997; Udroiu, 2017). In fact, several authors demonstrated the effects of anaesthesia and restraint on Hct (Kuttner and Wiesner, 1987; López-Olvera et al., 2007), thus influencing the results of blood analysis. Therefore, further considerations on Hct will not be done. In any case, the variation in Hct is quite small (coefficient of variation, CV=13%) in comparison to RBC count (CV=51%, ranging from 2.3 to 56) and MCV (CV=44%, ranging from 5.6 to 239 fl). Indeed, RBC count and MCV are strongly inversely correlated (Fig. S1C) and their non-log-transformed data show a hyperbolic relationship (Fig. S1D). These latter findings have been known for many decades (Wintrobe, 1961; Hawkey et al., 1991; Promislow, 1991) and confirm that, among mammals, Hct can be considered virtually constant in the formula MCV=10×Hct/RBC.

MCHC showed no correlation with either body mass (Fig. 1C) or MS-BMR (Fig. 2D): the concentration of Hb inside RBCs is constant in mammals (CV=6.9%, the smallest among all haematological parameters), probably reflecting a limit imposed by its effect on intracellular viscosity. Indeed, a higher MCHC allows greater oxygen transport (and, in fact, mammals and birds have higher values than ectotherms), but cytoplasmic viscosity, which is affected by MCHC (Chien et al., 1971), increases exponentially as intracellular Hb concentration rises (Castellini et al., 2010).

As MCHC is constant, and given MCH=MCV×MCHC (Table 1), the correlations between MS-BMR and MCV (Fig. 2F) and MCH (Fig. 2E) are almost identical. The correlations of MCV and MCH with MS-BMR (which were significant within all orders) can be explained by the fact that small RBCs, having a greater surface area to volume ratio, can release oxygen more rapidly, in agreement with a higher metabolic demand (Holland and Forster, 1966; Jones, 1979). Differences between orders can be easily seen; however, these are greatly influenced by the presence or absence of aquatic species. In fact, aquatic species are all above the regression line of normoxic species (Fig. S3A). The same cannot be said for altitude-dwelling species.

As noted above, small RBCs (with a low MCH/SA ratio) should favour the rapid exchange of oxygen in mammals with a high metabolic demand. However, the surface area of RBCs is not the sole parameter governing the exchange of oxygen. Blood–oxygen affinity (measured as the inverse of whole-blood P₅₀) is equally (if not more) important (Poyart et al., 1992). Blood–oxygen affinity is significantly positively correlated with body mass (Fig. 3C; r²=0.298, P<0.0001) and inversely with MS-BMR (Fig. 3D; r²=0.337, P<0.0001). In this case as well, this last correlation is independent from body mass (P=0.79), whereas the correlation between body mass and P₅₀ is due to residual MS-BMR (P=0.0022; Fig. S3E,F). Biologically, the correlation between P₅₀ and MS-BMR can be easily explained. In a normoxic environment, Hb–oxygen affinity is not a limiting factor in the lungs, where P_O2 is 100 mmHg compared with mammalian Hb affinity of 17–40 mmHg (Udroiu, 2020). However, P₅₀ also determines the speed of oxygen unloading in the tissues (Bartels et al., 1979; Wagner, 1997). In fact, Hb–oxygen affinity is determined by its speed of dissociation and association with O₂ (Holland et al., 1985). This has been experimentally measured, showing that the time needed to desaturate from 100% to 50% is 35 ms for goat Hb (P₅₀=32.2), 50 ms for human Hb (P₅₀=27.5) and 80 ms for human Hb stripped of DPG (P₅₀=17.1) (Holland et al., 1985). At the tissue level, an elevated P₅₀ also allows Hb to offload oxygen at a higher P_O2, thus increasing the P_O2 gradient between blood and tissues and allowing a faster delivery of O₂ to the tissues (Vadapalli et al., 2002). Therefore, high P₅₀ favours faster oxygen release in tissues with high metabolic demands.

As expected, fossorial and altitude-dwelling species have P₅₀ values lower (therefore a greater affinity) than those expected by their MS-BMR (Fig. 4B). Moreover, aquatic species also have lower than expected P₅₀ values, while Ruminantia and Feliformia have higher than expected P₅₀ values (Fig. 4D).

MCH/SA is also significantly correlated with P₅₀ (Fig. 3B) and this correlation is independent from MS-BMR (Fig. S3G). However, analysis of residuals revealed that the correlation between MCH/SA and MS-BMR is in reality due to correlation with P₅₀ (Fig. S3H). In particular, in Ruminantia, in which P₅₀ values are not correlated with MS-BMR but most probably with ‘phylogenetic inertia’ (mutations of the Hb gene determining low affinity), small MCH/SA (and MCV) values are linked to low Hb–oxygen affinity and not to MS-BMR. It is reasonable to view MCH/SA (and MCV) as a more modifiable parameter (phenotypic plasticity) than Hb–oxygen affinity. MCV and Hb content are mainly determined by the length of the RBC maturation time in the bone marrow (Sankaran et al., 2012; Udroiu, 2016). When the body needs a quicker release of RBC (e.g. during anaemia), maturation time is reduced and larger RBCs are released into the blood flow (macrocytosis). In contrast, the low affinity of ruminant blood is due to the low intrinsic affinity of Hb molecules and cannot be increased by any means. Therefore, in addition to the fact that MCH/SA and P₅₀ are correlated independently from MS-BMR (as demonstrated above), I suggest that MCH/SA seems to be dictated by P₅₀.

As MCHC is constant in mammals, MCH/SA is directly proportional to MCV (Fig. S3D), and thus the correlation between P₅₀ and MCH/SA can be translated into MCV for all those mammals sharing the same biconcave RBC geometry. This last specification is important, because RBCs are elliptical in Camelidae and spherical in Tragulidae (Snyder and Weathers, 1977; Hawkey et al., 1991). Indeed, the regression line of MCV against MCH/SA for Camelidae is parallel but different from that of other mammals (Fig. S3D). Therefore, regressing MCV or MCH-SA against MS-BMR (Figs 2F and 3A) or against P₅₀ (Fig. 3B; Fig. S3I) gives similar results, except for Camelidae and Tragulidae.

So, it can be said that RBC size (MCV) is, in the final analysis, correlated with blood–oxygen affinity (P₅₀). Small mammals usually have high MS-BMR and high P₅₀ (which means low affinity) and therefore small RBCs. Large mammals usually have low MS-BMR and low P₅₀ and therefore large RBCs. Fossorial species, which are predominantly Rodentia and Eulipotyphla, have high MS-BMR, but low P₅₀, showing RBCs larger than their normoxic counterpart. For aquatic species, RBCs are even larger than their P₅₀ would predict. It could be argued that larger RBCs, because they release oxygen more slowly (Vandegriff and Olson, 1984), could be advantageous during long dives. The effect of altitude, instead, seems to be less clear. Rodents living at high altitude have RBCs larger than those of their lowland counterparts, probably because of their lower P₅₀. Among Felidae, there are no significant differences, either for P₅₀ or for MCV, between lowland and high-altitude species. The same is true for Camelidae and other Artiodactyla.

Additional evidence that MCV is correlated with P₅₀ (and that correlations with body mass and mass-specific metabolic rates are indirect) comes from an ontogenetic comparison. It is well known that mammalian fetuses and newborns (excluding felids) have lower P₅₀ than adults (Delivoria-Papadopoulos et al., 1971) and that they have larger RBCs than adults (Dallman and Siimes, 1979; Udroiu, 2016): this cannot be due to body mass (which is obviously smaller in newborns) or MS-BMR (which is higher in newborns), and therefore this strengthens the link with blood–oxygen affinity.

As stated above, in most mammals the increase/decrease of MS-BMR coevolved with an increase/decrease of DPG, decreasing/increasing blood–oxygen affinity (Fig. 5; Fig. S5D). This coevolution can be interpreted as a result of adaptive selection. However, the evolution of Hb intrinsic affinity for oxygen seems characterized by punctational events, unrelated to adaptive selection. I have already cited the example of camelids, whose high Hb–oxygen affinity is shared not only by highland but also by lowland species. Another important example is represented by Ruminantia (cervids and bovids). These, in fact, show the lowest Hb–oxygen intrinsic affinities (Fig. S4). Because, for a given Hb intrinsic affinity (genetically determined), whole-blood affinity can be increased only by decreasing DPG concentration, it cannot be increased above the level reached when DPG is absent. Moreover, Hbs of these species show very little sensitivity to DPG and have near-zero concentrations of this allosteric factor (Scott et al., 1977). Therefore, mutations in the Hb of Ruminantia ancestors caused such a lowering of their affinity that could not be compensated for by modification of the level of allosteric factors. As MCV is correlated with Hb–oxygen affinity (as I demonstrated above), this probably determined the small size of RBCs in these species. An interesting case is that of mouse deer (Tragulus), which show several primitive anatomical features (e.g. canine teeth). They also show one of the lowest Hb affinities and the smallest RBCs among all mammals, being so small that the biconcave shape is lost and they are spherical (Snyder and Weathers, 1977).

Small RBCs can be very useful in some situations. For example, they are more osmoresistant, therefore resisting osmotic stress during dehydration (Peinado et al., 1992), a situation reasonably undergone by many goats and antelopes living in arid environments. Moreover, it has been suggested that they can be better stored in the small and strongly muscularized spleen of terrestrial Artiodactyla, as a reservoir to be used during athletic sprints (Udroiu, 2017) (the spleen acts as a blood reservoir in many other species, such as Perissodactyla and Carnivora, but in these species it is less muscularized and decisively larger; interestingly, these species also have RBCs significantly larger than those of terrestrial Artiodactyla). It could be argued that these are proof of an adaptive selection of small RBCs in ruminants. However, having small RBCs is a feature of all Ruminantia, including those species that are not environmentally subject to dehydration (e.g. cervids and Tragulidae) or that are not athletic (e.g. cows and Tragulidae). In particular, there is no hint of an adaptive advantage for Tragulus of having RBCs smaller than those of Rodentia or Eulipotyphla. Therefore, it can be interpreted that the small size of ruminant RBCs (linked to Hb–oxygen affinity) was co-opted by some species to be advantageous for resistance to dehydration and/or athletic performance. Also, camelids undergo dehydration (perhaps in the most extreme way): after a long period without drinking water, camelids can ingest water equivalent to 30% of their body weight without haemolysis (Etzion et al., 1984), whereas other mammals drinking 10% of their body weight face haemolysis (Bianca, 1970). Nonetheless, as noted above, they show Hbs with high affinity for oxygen. As the latter has been shown by my analyses to be correlated with large RBCs, and given that large RBCs (being less osmoresistant; Peinado et al., 1992), would probably be lethal for camelids during dehydration/rehydration, it is tempting to speculate that natural selection has favoured an elliptical, non-biconcave geometry of camelids RBCs, because of the decrease in the surface area to volume ratio, without increasing the RBC volume (MCV). Indeed, among the species studied, camelids are the only mammals with small RBCs and high Hb affinities.

Summing up, the results of the ancestral reconstruction suggest that Hb intrinsic affinity for oxygen shows little variation during evolution. Nonetheless, in camelids, mutations happened that allowed some species to colonize hypoxic environments (preadaptation), while in other lineages (e.g. moles, fossorial and high-altitude rodents), Hb–oxygen affinity has increased as part of their adaptation to low-oxygen air (adaptive selection). In Ruminantia, there was a drastic decrease of Hb intrinsic affinity. In most species, modifying the levels of DPG acted to adapt Hb–oxygen affinity to metabolic demands (the same also happens during ontogeny; Delivoria-Papadopoulos et al., 1971; Petschow et al., 1978). In all cases, acting as a biological constraint for oxygen exchange, Hb affinity determined the size of RBCs (Fig. 6).

Therefore, the study of the evolution of RBCs in mammals gives not only a clearer picture of their characteristics and the relationships between the different measured quantities but also interesting applications of concepts such as adaptive selection, preadaptation and biological constraints.