rapid metabolic evolution in human prefrontal cortex · human evolution is characterized by the...
TRANSCRIPT
Rapid metabolic evolution in human prefrontal cortexXing Fua,1, Patrick Giavaliscob,1, Xiling Liua,1, Gareth Catchpoleb, Ning Fuc, Zhi-Bin Ningc, Song Guoa, Zheng Yana,Mehmet Somela,d, Svante Pääbod, Rong Zengc,2, Lothar Willmitzerb,2, and Philipp Khaitovicha,d,2
aKey Laboratory of Computational Biology, CAS-MPG Partner Institute for Computational Biology, Chinese Academy of Sciences, 200031 Shanghai, China;bMax Planck Institute for Molecular Plant Physiology, 14476 Potsdam, Germany; cKey Laboratory for Systems Biology, Chinese Academy of Sciences, 200031Shanghai, China; and dMax Planck Institute for Evolutionary Anthropology, 04103 Leipzig, Germany
Edited by Trudy F. C. Mackay, North Carolina State University, Raleigh, NC, and approved February 28, 2011 (received for review December 23, 2010)
Human evolution is characterized by the rapid expansion of brainsize and drastic increase in cognitive capabilities. It has long beensuggested that these changes were accompanied by modificationsof brain metabolism. Indeed, human-specific changes on gene ex-pression or amino acid sequence were reported for a number ofmetabolic genes, but actual metabolite measurements in humansand apes have remained scarce. Here, we investigate concentra-tions of more than 100 metabolites in the prefrontal and cerebellarcortex in 49 humans, 11 chimpanzees, and 45 rhesus macaques ofdifferent ages using gas chromatography–mass spectrometry (GC-MS). We show that the brain metabolome undergoes substantialchanges, both ontogenetically and evolutionarily: 88% of detectedmetabolites show significant concentration changes with age,whereas 77% of these metabolic changes differ significantlyamong species. Although overall metabolic divergence reflectsphylogenetic relationships among species, we found a fourfoldacceleration of metabolic changes in prefrontal cortex comparedwith cerebellum in the human lineage. These human-specific met-abolic changes are paralleled by changes in expression patterns ofthe corresponding enzymes, and affect pathways involved in syn-aptic transmission, memory, and learning.
cognition | glutamate | development
Human evolution is characterized by rapid expansion of brainsize and increase in cognitive capabilities, leading to the
emergence of unique and complex cognitive skills. These changeshave long been associated with changes in brain metabolism, inparticular with respect to increased energy demand (1). Largebrains are metabolically costly. Thus, humans allocate approxi-mately 20% of their total energy to the brain, compared with 11–13% for apes and 2–8% for other mammalian species (2). Thisincreased metabolic demand has been associated with elevatedexpression of genes involved in neuronal functions and energymetabolism (3, 4). These changes may have been evolutionaryadvantageous, as indicated by signatures of positive selectionreported for the amino acid changes, which occur in the mito-chondrial electron-transport chain proteins, in anthropoid pri-mates and humans, as well as elevated expression of the energymetabolism pathways in the human brain (5, 6). On the histo-logical level, human brains show the largest density of glia cellsrelative to neurons in the prefrontal cortex, which provides anindirect indication of increased neuronal metabolic demand (7).Changes in brain metabolism are also implicated in neuro-
psychiatric disorders, such as schizophrenia, which affect some ofthe human-specific cognitive abilities (8, 9). Furthermore, directmeasurements of 21 metabolites in the prefrontal cortex of adulthuman controls compared with human schizophrenia patients,chimpanzees, and rhesus macaques, carried out using protonNMR spectroscopy have shown significant overlap between hu-man-specific evolutionary changes and metabolic differencesbetween controls and schizophrenia patients (10). Notably, thisstudy has identified not only energy metabolites, such as lactateand creatine, but also metabolites related to neurotransmission,such as choline and glycine, as being altered in both humanevolution and in disease. Taken together, these observationssuggest that diverse metabolic changes might have been impor-
tant to the establishment and maintenance of the human-specificcognitive abilities.Here, we used a combination of gas chromatography–mass
spectrometry (GC-MS) metabolomics and quantitative label-freeproteomics to measure concentrations of more than 100 metab-olites and ∼2,000 proteins in brains of humans, chimpanzees,and rhesus macaques. Application of GC-MS methodology tometabolic profiling in the human tissues, including brain, wassuccessfully introduced almost 30 y ago (11). This methodologyprovides good resolution for relatively small, thermally stablecompounds that can be made volatile through derivatization (12).It must be noted, however, that although GC-MS methodologycan potentially resolve thousands of metabolites, compound de-rivatization and fragmentation during the procedure precludedirect metabolite identification. Instead, the methodology relieson using pure compounds as standards, thus leaving a number ofmetabolites detected in the procedure unidentified.Using GC-MS methodology, we studied metabolic differences
among humans, chimpanzees, and macaques at different ages intwo functionally and evolutionary distinct brain regions: superiorfrontal gyrus of the prefrontal cortex (PFC) and lateral part of thecerebellar cortex (CBC). PFC is a brain region that emerged re-cently during primate evolution. It is implicated in complex as-sociative functions including reasoning, planning, social behavior,and general intelligence (13–15). Furthermore, the PFC has beenshown to undergo greater expansion at the gross anatomy level onthe human evolutionary lineage than the CBC (16). AlthoughCBC has also evolved considerably in primates (17, 18), its in-volvement in processing and execution of human-specific cogni-tive tasks is not yet well understood.
ResultsEstimation of Metabolic Variation in the Primate Brain. To directlyassess the extent of metabolic changes between human brain andbrains of other primate species, we measured metabolite levels inpostmortem PFC and CBC samples from 49 humans, 11 chim-panzees, and 45 rhesus macaques, using GC-MS (Table S1).Manyphenotypic parameters, such as maximal life span, age of sexualmaturity, and brain growth curves differ between humans, chim-panzees, and rhesus macaques (19, 20). This makes it difficult tomatch samples with respect to age among these three species. Toovercome this problem and to estimate changes in metabolitelevels during brain development, maturation and aging, we sam-
Author contributions: S.P., R.Z., L.W., and P.K. designed research; P.G., N.F., Z.-B.N., andZ.Y. performed research; S.P., R.Z., L.W., and P.K. contributed new reagents/analytic tools;X.F., P.G., X.L., G.C., N.F., Z.-B.N., S.G., and M.S. analyzed data; and X.F., P.G., X.L., M.S.,S.P., R.Z., L.W., and P.K. wrote the paper.
The authors declare no conflict of interest.
Freely available online through the PNAS open access option.
This article is a PNAS Direct Submission.1X.F., P.G., and X.L. contributed equally to this work.2To whom correspondence may be addressed. E-mail: [email protected], [email protected], or [email protected].
This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1019164108/-/DCSupplemental.
www.pnas.org/cgi/doi/10.1073/pnas.1019164108 PNAS | April 12, 2011 | vol. 108 | no. 15 | 6181–6186
EVOLU
TION
pled individuals of different age, covering most of the species’lifespan: humans 0–98 y, chimpanzees 0–40 y, and macaques0–28 y (Fig. S1 and Table S1). Tominimize sampling variation, forall individuals we took care to sample gray matter only. To min-imize technical variation, we measured each sample twice.Using GC-MS methodology, we detected a total of 232
metabolites, some present in a single individual. Out of these, wefocused on 118 metabolites, 61 annotated and 57 unknown,detected in more than 80% of samples in at least one brain regionin one or more of the three species (Table S2). For these metab-olites we found good agreement in metabolite measurements be-tween replicates in all three species and all brain regions (medianPearson r> 0.95, P< 0.0001), indicating high reproducibility of theGC-MS measurements (Fig. S2).Besides technical reproducibility, another concern is the rele-
vance of metabolite measurements made in postmortem braintissue to in vivo metabolite concentrations. Previous studies thathave used proton magnetic resonance spectroscopy in human andrat brain biopsy samples have shown that concentrations ofmetabolites associated with anaerobic glycolysis, such as glucose,alanine, glutathione and lactate, change after death (21–24). Inrat brain, postmortem delay also leads to concentration changesof dopamine, norepinephrine, and serotonin in the occipitalcortex. This effect, however, was not shared by all brain regions,as serotonin levels remained stable over an 18-h postmorteminterval in the cingulate cortex (25). Furthermore, concentrationsof many metabolites, such as myo-inositol, creatine, glutamine,glutamate, N-acetylaspartate, and taurine were shown to remainstable in the postmortem brain tissue over long time intervals (22,23, 26, 27). Similarly, experiments quantifying brain polyamines,such as spermine, spermidine, and putrescine, in postmortemhuman brain using GC-MS did not report any significant re-lationship between these metabolite concentrations and post-mortem interval (28). To estimate the effect of postmortem delayon metabolite concentrations in our study, we took advantage ofthe fact that, with the exception of one sample with approximatelya 5-h postmortem delay, all rhesus macaque brain tissues werecollected and frozen within 20 min after death (Table S1). Byexamining the profile of each metabolite among the macaquesamples, we identified 26metabolites with significant differences inconcentration between replicate measurements taken from themacaque individual with prolonged postmortem delay and meas-urements taken from all other samples in at least one brain region(Fig. S3). In agreement with previous studies, many of thesemetabolites were involved in glucose metabolism (Table S3). Con-sequently, we excluded these metabolites from all further analyses.Some of the human and chimpanzee samples used in our study
have a postmortem delay exceeding the 5-h interval tested usingmacaque samples. Still, in macaques, the vast majority of metab-olites showed no indication of postmortem-induced concentrationchange after 5 h, demonstrating their long-term stability (Fig. S4).Furthermore, our assignment of metabolic changes to the evolu-tionary lineages requires similarity between macaque metabolitemeasurements over the lifespan and metabolic measurements ineither human or chimpanzee brains. As macaque data are notaffected by postmortem delay, and as human and chimpanzeesamples have no postmortem interval bias, we can largely excludepostmortem delay as a confounding factor in this study.To estimate the proportions of the total metabolic variance
explained by species, age, and brain region identity, we used prin-cipal variance component analysis (29). The differences betweenspecies explained 49% of the total metabolic variation, differenceswith age 17%, and differences between two brain regions 9% (Fig.S5). Accordingly, metabolic differences across species, as well asdifferences with age, can be visualized using principal componentanalysis in both brain regions (Fig. 1 A and B).
Species-Specific Metabolic Changes.We next identified metabolitesshowing significant concentration changes with age in each spe-cies using polynomial regression (30). At a false discovery rate(FDR) <10% (permutation test; SI Appendix), we identified 48and 47 metabolites in the CBC and 53 and 26 in the PFC ofhumans and macaques, respectively (Table S4). In chimpanzees,due to the smaller sample size, we found only two and five age-related metabolites in CBC and PFC, respectively. Among themetabolites that change with age in our study, taurine was pre-viously reported to decrease in concentration during postnatalbrain development in the human and macaque occipital brainlobe (31). Similarly, in our data we found an approximately lineardecrease with age in taurine concentration in both human andrhesus macaque PFC and CBC (human PFC: r= −0.37, P < 0.01;human CBC: r = −0.53, P < 10−5; macaque PFC: r = −0.28, P <0.05; macaque CBC: r = −0.51, P < 0.001) (Fig. S6). Other thantaurine, we found no reported primate brain time-series data thatinvolved age-related metabolites as identified in our study.We next tested whether metabolites change differently with
age among the three species. Here, we focused on metabolitesshowing different concentration profiles over the species’ life-span, rather than on mean concentration differences alone. Giventhe large and comparable numbers of human and rhesus ma-caque samples, we first identified metabolites with significantdifference in concentration profiles between these two speciesin each brain region. Using analysis of covariance (ANCOVA) withlinear, quadratic, and cubic models, we found 49 metabolites inPFC and 43 in CBC at FDR < 1% (permutation test; SI Appendix).Of these metabolites, 30 were different between humans andmacaques in both brain regions (Fig. S7). Thus, both PFC andCBC metabolome show substantial divergence between humansand macaques, with a large proportion of the differences sharedbetween the two brain regions.We then used chimpanzee metabolite measurements to sort the
identified human-macaque metabolic differences into three cate-gories: (i) human-specific changes, (ii) macaque-specific changes,and (iii) uncategorized changes. Human-specific changes weredefined based on significant difference between human metabolicprofiles and metabolic profiles of (but not between) both chim-panzees and macaques (Wilcoxon test, P < 0.01, FDR < 1%).Macaque-specific changes were defined analogously, and there-fore include changes on both the rhesus macaque evolutionarylineage, as well as the lineage between the common ancestor ofOld World primates and the common ancestor of humans andchimpanzees. Uncategorized changes included metabolite pro-files that are not described by either of the other two groups. Wefound six human-specific, 33 macaque-specific, and four unchar-acterized metabolic changes in CBC. By contrast, in PFC, wefound 24 human-specific, 20 macaque-specific, and five unchar-acterized metabolic changes. Thus, there is a fourfold excess ofhuman-specific metabolic changes in the human PFC comparedwith CBC (Figs. 1C, F, andG and 2 and Table S5). This result wasrobust with respect to data normalization procedures (Fig. S10).The additional analyses confirmed the robustness of this re-
sult. Excluding unknown metabolites, we found three metabo-lites with human-specific profiles in CBC and 11 in PFC (Fig.S8). Similarly, limiting our analysis to 30 metabolites with sig-nificant concentration difference between humans and macaquein both PFC and CBC, we again found more human-specificchanges in PFC (Fig. 1D). Thus, since the separation from thelast common ancestor of humans and chimpanzees, many moremetabolic changes have taken place on the human evolutionarylinage in PFC than in CBC. By contrast, we found no indicationfor faster PFC metabolome evolution, compared with CBC, oneither the macaque evolutionary lineage or the lineage betweenthe common ancestor of Old World primates and the commonancestor of humans and chimpanzees (Fig. 1C).
6182 | www.pnas.org/cgi/doi/10.1073/pnas.1019164108 Fu et al.
Furthermore, we repeated the linage assignment analysis usingequal numbers (n = 11) of age-matched individuals across thethree species (Fig. S1). Equalizing sample size across the threespecies allowed us to assign metabolic differences to the humanand the chimpanzee evolutionary linage with equal confidence.To avoid artifacts caused by differences in lifespan duration, weused the following two approaches to match individuals’ ageacross species: (i) stage-of-life approach: scaling ages linearlywithin each species to the same maximal lifespan, using 120 y forhumans, 60 for chimpanzees, and 40 for rhesus macaques asa reference (19); and (ii) chronological approach: using calendarage directly. Using both approaches, we again found an excess ofhuman-specific metabolic changes in PFC and not in CBC (Fig.1E, Fig. S9, and Table S5). Thus, an acceleration of metabolicchanges in the PFC is specific to the human evolutionary linageand is not observed in CBC.
Protein Expression and Metabolic Pathway Analyses. To further testthe validity of the identified human-specific metabolic changesand to assess their functional significance, we analyzed the ex-pression of the corresponding metabolic enzymes in 12 human,12 chimpanzee, and 12 rhesus macaque PFC samples using label-free quantitative mass spectrometry.In humans, chimpanzees and rhesus macaques, we identified a to-
tal of 1301042, 1004360 and 1151143 peptides, which were mappedon 10769, 7945, and 9339 IPI peptide sequences, respectively. Re-quiring at least 10 peptide counts per protein, we detected 2,747proteins in humans, 2,343 proteins in chimpanzees, and 2,842 pro-
teins in rhesusmacaques.The1,951genesdetected inall three species(proteinwise FDR < 5%) were used in further analyses.Using these data, we first tested whether metabolic changes
with age found in the human PFC correlate with expressionchanges of enzymes directly interacting with these metabolites(i.e., enzymes connected to metabolites by a single edge in theKEGG pathway annotation). In total, we found 47 such metab-olite-enzyme pairs, 21 of them showing significant correlationbetween metabolite and enzyme concentration changes with age(Pearson correlation, P < 0.0001, |r|>0.5). Finding such a highproportion of correlated changes is unexpected (P = 0.001), asestimated by randomly substituting expression profiles of theenzymes by those of other proteins 1,000 times. The result wasalso robust at different correlation coefficient cut-offs (Fig. S11).Second, we tested whether enzymes directly associated with
metabolites with human-specific concentration profiles (for sim-plicity: human-specific metabolites) show more expression dif-ferences between human and chimpanzee PFC than enzymesdirectly associated with metabolites that change with age in thehuman PFC but are not human specific. We found a greater ex-pression divergence for 13 such enzymes associated with human-specific metabolites (one-sided Wilcoxon test, P = 0.003) com-pared with 24 enzymes associated with non–human-specificmetabolites (Fig. S12A). Finally, we tested whether enzymes di-rectly associated with human-specific metabolites show morehuman-specific expression than enzymes directly associated withnon–human-specific metabolites, by estimating the ratio betweentheir human–macaque and chimpanzee–macaque expression di-vergence. We found that for enzymes associated with human-
A B F
C D E
G
Fig. 1. Metabolite variation among species. (A and B) Principal component analysis of CBC and PFC metabolomes of the three species based on 92 detectedmetabolites. Each circle represents an individual. Size of circles is proportional to the individuals’ ages, with larger circles corresponding to older individuals.Colors represent species: red, human; purple, chimpanzee; blue, rhesus macaque. (C–E) Proportions of human-specific (red), macaque-specific (gray), chim-panzee-specific (blue), and uncategorized metabolites (white) among the following: (C) the 43 and 49 metabolites with significant difference in concen-tration profiles between humans and rhesus macaques in CBC and PFC, respectively, identified using the full set of individuals; (D) the 30 metabolites withsignificant difference in concentration profiles between humans and rhesus macaques in both CBC and PFC; (E) the 25 and 17 metabolites with significantdifference in concentration profiles among humans, chimpanzees, and rhesus macaques identified in CBC and PFC, respectively, using the subsets of 11individuals per species matched using stage-of-life approach (SI Appendix). Numbers indicate numbers of metabolites in corresponding category. (F and G)Hierarchical clustering based on concentration profiles of 24 metabolites classified as human specific in PFC and six metabolites classified as human specific inCBC. Column headers indicate species: Ch, chimpanzee; Hu, human; Ma, rhesus macaque. Red and blue color intensities indicate metabolite concentrationlevels at different age normalized to mean equal a value of 0 and SD equal a value of 1 within each brain region.
Fu et al. PNAS | April 12, 2011 | vol. 108 | no. 15 | 6183
EVOLU
TION
specific metabolites, this ratio was significantly greater (one-sidedWilcoxon test, P = 0.004) (Fig. S12B). Thus, we observed signif-icant agreement between human-specific metabolic profiles in thePFC and expression of the corresponding enzymes in terms ofboth ontogenetic changes and evolutionary divergence.To determine which pathways are associated with human-
specific metabolic changes, we first mapped the detected pro-teins onto the KEGG pathways containing at least one human-specific metabolite and more than 10 proteins. This resulted in13 KEGG pathways that include five metabolites (glutamate,histidine, spermidine, oxoproline, and hexadecanoic acid) and209 proteins. Four pathways had greater than expected human-specific protein expression divergence: long-term potentiation(P = 0.039), neuroactive ligand-receptor interaction (P = 0.015),alanine, aspartate, and glutamate metabolism (P = 0.022), andβ-alanine metabolism” (P= 0.012) (Fig. 3A and Fig. S13). For allfour pathways, Q values were <0.02, indicating significance ofresults given the number of pathways tested.These pathways are not independent, but share common genes
and a common metabolite, i.e., glutamate, which is present atlower concentrations in humans than in chimpanzees or mac-aques, particularly at the first years of life (Figs. 2B and 3B).Besides glutamate, two other metabolites with human-specificprofiles in PFC, histidine, and spermidine, are involved in theβ-alanine metabolism pathway.
DiscussionOur results show that human, chimpanzee, and macaque brainmetabolomes change substantially with age: Specifically, 88% ofthe metabolites change their concentration significantly over thecourse of lifespan in at least one species or one brain region(Table S6). Many of these metabolic changes with age differsignificantly among the three species (53% and 47% in PFC andCBC, respectively). Considering the two brain regions together,77% of metabolic changes with age differed among species(Table S6). Large numbers of significant metabolic changes withage and differences among species are in line with substantialproportion of the total metabolic variance explained by thesefactors (17% and 49%, respectively; Fig. S5).It must be noted, however, that most of the identified meta-
bolic differences between species are not due to a difference inthe pattern of concentration changes with age. Instead, 67% ofmetabolic changes in PFC and 82% in CBC reflect differences inthe rate of age-related changes and/or differences in the meanmetabolite concentration among species (Table S9). Thus, themajority of metabolic changes with age, especially in CBC, followthe same trajectory in all three species.Our main result is the discovery of an approximately fourfold
excess of metabolic changes in the PFC on the human evolutionarylineage. By contrast, CBC shows no such acceleration. This resultis robust to choice of the normalization procedure, as well aschronological and stage-of-life age-matching subsets. Further-more, we observed significant agreement between concentration
A
B
Fig. 2. Metabolites with human-specific concentration profiles. Shown are the six metabolites with human-specific concentration profiles in CBC (A) and 24metabolites with human-specific concentration profiles in PFC (B). Points show metabolite concentrations in each individual. Colors represent species: red,humans; purple, chimpanzees; and blue, rhesus macaques. Lines are spline curves fitted to data points with 3 df. The x axis shows individuals’ ages in years.The y axis shows normalized GC-MS measurements representing metabolite concentrations. Titles show metabolite annotation. Unannotated metabolites arelabeled “unknown.”
6184 | www.pnas.org/cgi/doi/10.1073/pnas.1019164108 Fu et al.
profiles of metabolites in the human PFC and expression profilesof the corresponding metabolic enzymes. As protein expressionmeasurements were carried out using independently dissectedtissue samples and different methodologies, significant corre-spondence between protein and metabolic changes underscoresthe validity of the results.Importantly, PFC and CBC samples used in our study come
from the same individuals. This fact largely excludes individualdemographic variables, such as postmortem delay, cause of death,medication, and so on, as causative factors for the difference innumbers of human-specific metabolic changes between PFC andCBC. Still, if environmental factors had a different effect on PFCand CBC metabolism, this could result in observed differencesbetween the two brain regions.Some aspects of environment and lifestyle, such as diet and life in
captivity, could be argued to be more similar between chimpanzeesand macaques, consequently making humans more distinct. Toestimate the extent to which the species-specific metabolic changesmight be caused by these environmental factors, we collected 37metabolites reported to be affected by diet and 71 by exercise(Table S7). Using Fisher’s exact test, we did not find significantoverrepresentation of diet- or exercise-affected metabolites amongmetabolites with species-specific concentration profiles (P > 0.2 inall tests). Furthermore, reported diet and exercise effects were notcompatible with the species-specific metabolic pattern that we ob-served (SI Appendix). Thus, our results provide no indication thatexcess of human-specific metabolic changes found in PFC could beexplained by the influence of environmental factors.Glutamate is one of the human-specific metabolites found in
the PFC in this study. Glutamate release and recycling is a majormetabolic pathway consuming 60–80% of the energy supplied bythe glucose oxidation in the human cerebral cortex (32). Im-portantly, in the human brain, glutamate pathways play a centralrole in both energy metabolism and neurotransmission. Evolu-tionary changes in glutamate metabolism were previously in-
dicated by the finding that glutamate dehydrogenase, an enzymecentral to the glutamate and energy metabolism of the cell, hasundergone duplication and subfunctionalization in the commonancestor of apes and humans, resulting in appearance of an ape-and human-specific GLUD2 gene (33). Our findings show thatadditional changes in glutamate metabolism took place in PFC,but not in CBC, on the human evolutionary lineage.Specifically, we find that glutamate concentrations were lower
in humans, and particularly in human newborns, compared withchimpanzees and macaques. Previous studies indicated thatGLUD2 is optimized to rapidly metabolize brain glutamate (34,35). It is therefore appealing to speculate that low glutamateconcentrations observed in human infants might represent anevolutionary continuation of the rapid glutamate turnover trend.Indeed, based on previously published mRNA data (30), we ob-served higher expression levels of GLUD2 in human newbornscompared with chimpanzee newborns (Fig. S14). This observationsupports our notion that low glutamate levels in human newbornsare caused by higher glutamate turnover. Functional interpretationof this change, however, is meaningful only in the context of thelarge metabolic flux of glutamate release and recycling takingplace in glia and neurons (36, 37). Furthermore, it must be notedthat glutamate concentrations in tissues other than brain werepreviously shown to be affected by diet and exercise (38, 39). Al-though such an environmental effect cannot be fully excluded inour study, human-specific changes in glutamate metabolism areintriguing, and might be an interesting and important subject offurther research.We must note, however, that even though our analysis covers
more metabolites than other studies conducted to date, it is farfrom being comprehensive. Thus, although our findings indicatethat changes in brain metabolism have contributed to the evo-lution of human cognition, many important metabolic pathwaysmight have been missed in our analysis. Further studies are
Gap junction
Aminoacyl tRNA biosynthesis
Huntington's disease
Glutathione metabolism
Arginine and proline metabolism
Long term depression
Fatty acid metabolism
Amyotrophic lateral sclerosis (ALS)
Butanoate metabolism
Long term potentiation
Alanine, aspartate and glutamate metabolism
Neuroactive ligand receptor interaction
beta Alanine metabolism
0.0 0.5 1.0 1.5 2.0Hu Ma Euclidian distanceCh Ma Euclidian distance
human specific metabolitehuman specific proteinhuman chimpanzee different proteinprotein metabolite correlated in humandetected protein
Alanine, aspartate and glutamate metabolismLong term potentiationNeuroactive ligand receptor interactionbeta Alanine metabolism
A B
Fig. 3. Pathway analysis of human-specific metabolites. (A) Shaded bars show median value of the human-specific protein expression divergence calculatedas ratio of Euclidian distances between human–macaque and chimpanzee–macaque expression profiles of protein detected in the 13 KEGG pathways. Eachpathway contains at least one human-specific metabolite and more than 10 detected proteins. Hatched bars show median human-specific divergence valuesexpected by chance. Chance expectations were calculated by randomly assigning the same number of detected proteins to a given pathway, 1,000 times. Errorbars show SD of the chance human-specific divergence estimates. (B) Network diagram showing four KEGG pathways with significantly greater human-specific protein expression divergence and three associated human-specific metabolites: glutamate, histidine, and spermidine. Colors indicate the following:bright red, proteins with significant human-specific expression; light red, proteins with significant human-chimpanzee expression divergence; pink, proteinswith significant correlation to the corresponding metabolite profile; gray, detected proteins.
Fu et al. PNAS | April 12, 2011 | vol. 108 | no. 15 | 6185
EVOLU
TION
needed to elucidate the entire scope of metabolic changes thattook place on the human evolutionary lineage.
Materials and MethodsDetails onmaterials andmethods are provided in SIAppendix. Briefly, sampleswere dissected from the frozen postmortem tissue from healthy individuals.PFC dissections weremade from the frontal part of the superior frontal gyrus.CBC dissections were taken from the lateral part of the cerebellar hemi-spheres. For metabolite data, metabolites were extracted from ∼100 mg tis-sue, using GC-MS measurements as detailed elsewhere (40). All samplereplicates were processed in a completely randomized order to minimize anypossible batch effects. For protein data, proteins were extracted from 100mgtissue, using label-free quantitative mass spectrometry according to a pre-viously published method (41). Analysis was performed on an LTQ massspectrometer equipped with a metal needle electrospray interface massspectrometer (ThermoFinnigan) in a data-dependent collection model (eachfull scan followed by 10 MS/MS scans of the most intense ions).
ACKNOWLEDGMENTS. We thank the National Institute of Child Health andHuman Development Brain and Tissue Bank for Developmental Disorders,the Netherlands Brain Bank, the Chinese Brain Bank Center, and Dr. H. R.Zielke, Dr. J. P. Dai, and Dr. Y. Sun in particular for providing the humansamples; the Yerkes Primate Center, the Biomedical Primate ResearchCentre, the Anthropological Institute, and Museum of the University ofZurich for chimpanzee samples, and Dr. R. Martin and Dr. W. Scheffran inparticular for providing the chimpanzee samples; Suzhou Drug SafetyEvaluation and Research Center, and C. Lian, H. Cai, and X. Zheng inparticular for providing the macaque samples; J. Dent for editing themanuscript; E. Lizano and F. Xue for assistance; and all members of theComparative Biology Group in Shanghai for helpful discussions and sugges-tions. We thank the following for financial support: the Ministry of Scienceand Technology of the People’s Republic of China (Grants 2007CB947004,2006CB910700, and 2011CB910200), the Max Planck Project “Age-relatedchanges in human-chimpanzee and rhesus macaque brain metabolism,”theChinese Academy of Sciences (Grants KSCX2-YW-R-094 and KSCX2-YW-R-251), the Shanghai Institutes for Biological Sciences (Grant 2008KIT104),the European Union Proteomage program, the Max Planck Society, andthe Bundesministerum fuer Bildung und Forschung.
1. Aiello L, Wheeler P (1995) The expensive-tissue hypothesis: The brain and thedigestive system in human and primate evolution. Curr Anthropol 36:199–221.
2. Mink JW, Blumenschine RJ, Adams DB (1981) Ratio of central nervous system to bodymetabolism in vertebrates: Its constancy and functional basis. Am J Physiol 241:R203–R212.
3. Caceres M, et al. (2003) Elevated gene expression levels distinguish human from non-human primate brains. Proc Natl Acad Sci USA 100:13030–13035.
4. Uddin M, et al. (2004) Sister grouping of chimpanzees and humans as revealed bygenome-wide phylogenetic analysis of brain gene expression profiles. Proc Natl AcadSci USA 101:2957–2962.
5. Grossman LI, Wildman DE, Schmidt TR, Goodman M (2004) Accelerated evolution ofthe electron transport chain in anthropoid primates. Trends Genet 20:578–585.
6. Khaitovich P, et al. (2006) Positive selection on gene expression in the human brain.Curr Biol 16:R356–R358.
7. Sherwood CC, et al. (2006) Evolution of increased glia-neuron ratios in the humanfrontal cortex. Proc Natl Acad Sci USA 103:13606–13611.
8. Abbott C, Bustillo J (2006) What have we learned from proton magnetic resonancespectroscopy about schizophrenia? A critical update. Curr Opin Psychiatry 19:135–139.
9. Dwyer DS, Bradley RJ, Kablinger AS, Freeman AM, 3rd (2001) Glucose metabolism inrelation to schizophrenia and antipsychotic drug treatment. Ann Clin Psychiatry 13:103–113.
10. Khaitovich P, et al. (2008) Metabolic changes in schizophrenia and human brainevolution. Genome Biol 9:R124.
11. Jellum E, Bjornson I, Nesbakken R, Johansson E, Wold S (1981) Classification of humancancer cells by means of capillary gas chromatography and pattern recognition analysis.J Chromatogr A 217:231–237.
12. Issaq HJ, Van QN, Waybright TJ, Muschik GM, Veenstra TD (2009) Analytical andstatistical approaches to metabolomics research. J Sep Sci 32:2183–2199.
13. Barbey AK, Krueger F, Grafman J (2009) An evolutionarily adaptive neuralarchitecture for social reasoning. Trends Neurosci 32:603–610.
14. Duncan J, et al. (2000) A neural basis for general intelligence. Science 289:457–460.15. Winterer G, Goldman D (2003) Genetics of human prefrontal function. Brain Res Brain
Res Rev 43:134–163.16. Clark DA, Mitra PP, Wang SS (2001) Scalable architecture in mammalian brains.
Nature 411:189–193.17. MacLeod CE, Zilles K, Schleicher A, Rilling JK, Gibson KR (2003) Expansion of the
neocerebellum in Hominoidea. J Hum Evol 44:401–429.18. Weaver AH (2005) Reciprocal evolution of the cerebellum and neocortex in fossil
humans. Proc Natl Acad Sci USA 102:3576–3580.19. de Magalhaes JP, Costa J (2009) A database of vertebrate longevity records and their
relation to other life-history traits. J Evol Biol 22:1770–1774.20. Leigh SR (2004) Brain growth, life history, and cognition in primate and human
evolution. Am J Primatol 62:139–164.21. Opstad KS, Bell BA, Griffiths JR, Howe FA (2008) An assessment of the effects of
sample ischaemia and spinning time on the metabolic profile of brain tumour biopsyspecimens as determined by high-resolution magic angle spinning (1)H NMR. NMRBiomed 21:1138–1147.
22. Opstad KS, Wright AJ, Bell BA, Griffiths JR, Howe FA (2010) Correlations between invivo (1)H MRS and ex vivo (1)H HRMAS metabolite measurements in adult humangliomas. J Magn Reson Imaging 31:289–297.
23. Perry TL, Hansen S, Gandham SS (1981) Postmortem changes of amino compounds inhuman and rat brain. J Neurochem 36:406–410.
24. Shank RP, Aprison MH (1971) Post mortem changes in the content and specificradioactivity of several amino acids in four areas of the rat brain. J Neurobiol 2:145–151.
25. Kontur PJ, al-Tikriti M, Innis RB, Roth RH (1994) Postmortem stability of monoamines,their metabolites, and receptor binding in rat brain regions. J Neurochem 62:282–290.
26. Michaelis T, Helms G, Frahm J (1996) Metabolic alterations in brain autopsies: ProtonNMR identification of free glycerol. NMR Biomed 9:121–124.
27. Petroff OA, Ogino T, Alger JR (1988) High-resolution proton magnetic resonancespectroscopy of rabbit brain: Regional metabolite levels and postmortem changes. JNeurochem 51:163–171.
28. Chen GG, Turecki G, Mamer OA (2009) A quantitative GC-MS method for three majorpolyamines in postmortem brain cortex. J Mass Spectrom 44:1203–1210.
29. Li J, Bushel PR, Chu T-M, Wolfinger RD (2009) Principal variance components analysis:Estimating batch effects in microarray gene expression data. Batch Effects and Noisein Microarray Experiments: Sources and Solutions, ed Scherer A (Wiley, Chichester) pp
141–154.30. Somel M, et al. (2009) Transcriptional neoteny in the human brain. Proc Natl Acad Sci
USA 106:5743–5748.31. Sturman JA, Gaull GE (1975) Taurine in the brain and liver of the developing human
and monkey. J Neurochem 25:831–835.32. Rothman DL, Behar KL, Hyder F, Shulman RG (2003) In vivo NMR studies of the glu-
tamate neurotransmitter flux and neuroenergetics: Implications for brain function.Annu Rev Physiol 65:401–427.
33. Burki F, Kaessmann H (2004) Birth and adaptive evolution of a hominoid gene thatsupports high neurotransmitter flux. Nat Genet 36:1061–1063.
34. Plaitakis A, Spanaki C, Mastorodemos V, Zaganas I (2003) Study of structure-functionrelationships in human glutamate dehydrogenases reveals novel molecular mechanisms
for the regulation of the nerve tissue-specific (GLUD2) isoenzyme. Neurochem Int 43:401–410.
35. Shashidharan P, et al. (1994) Novel human glutamate dehydrogenase expressed in
neural and testicular tissues and encoded by an X-linked intronless gene. J Biol Chem269:16971–16976.
36. Benjamin AM, Quastel JH (1972) Locations of amino acids in brain slices from the rat.Tetrodotoxin-sensitive release of amino acids. Biochem J 128:631–646.
37. van den Berg CJ, Garfinkel D (1971) A stimulation study of brain compartments.
Metabolism of glutamate and related substances in mouse brain. Biochem J 123:211–218.
38. Howarth KR, LeBlanc PJ, Heigenhauser GJ, Gibala MJ (2004) Effect of endurancetraining on muscle TCA cycle metabolism during exercise in humans. J Appl Physiol 97:579–584.
39. Li LO, et al. (2010) Early hepatic insulin resistance in mice: A metabolomics analysis.Mol Endocrinol 24:657–666.
40. Lisec J, Schauer N, Kopka J, Willmitzer L, Fernie AR (2006) Gas chromatography massspectrometry-based metabolite profiling in plants. Nat Protoc 1:387–396.
41. Fu X, et al. (2009) Estimating accuracy of RNA-Seq and microarrays with proteomics.
BMC Genomics 10:161.
6186 | www.pnas.org/cgi/doi/10.1073/pnas.1019164108 Fu et al.
Rapid metabolic evolution in human prefrontal cortex -- Supplementary
Information
Supplementary Analysis on Diet and Exercise Effects
To estimate the extent to which the species-specific metabolic changes might be caused
by environmental factors, we screened the literature for the studies of diet and exercise
effects on metabolite levels. We identified two studies that reported significant dietary
effect in humans and mice (1, 2) and five studies that reported significant exercise effects
in humans and rats (3-7). In total, 37 metabolites were reported to be affected by diet and
71 by exercise. An additional 28 metabolites were found to be unaffected by exercise.
Out of these, 8, 14 and 9, respectively, were detected in our dataset (Table S7). Out of the
8 diet-related metabolites, 5 overlapped with metabolites showing species-specific
profiles in PFC and 4 overlapped with metabolites showing species-specific profiles in
CBC (Table S8). Out of the 14 exercise-related metabolites, 7 overlapped with
metabolites showing species-specific profiles in PFC and 8 in CBC, with 5 found in both
brain regions. Similar extent of overlap could be found for metabolites shown not to be
affected by exercise.
To test any influence of diet or exercise on metabolite differences among species, we
compared metabolite sets identified in our study and the above mentioned studies. Using
Fisher’s exact test, we did not find significant overrepresentation of diet- or exercise-
affected metabolites among metabolites with species-specific concentration profiles
(p>0.2 in all tests).
Among metabolites showing human-specific changes in our study, glutamate in PFC and
taurine and ascorbic acid in CBC were previously reported to be affected by exercise.
Further, glutamate in PFC and taurine in CBC were previously reported to be affected by
diet. Differences between species detected in our study, however, did not display patterns
consistent with previously reported environmental effects. For example, glutamate has
been shown to decrease in concentration with increased exercise (4) or due to high-fat
diet (1). In our data, however, the main difference between human and non-human
primate glutamate concentration profiles occurs at the time of birth, with glutamate
concentration being lower in human newborns than that in the chimpanzee and rhesus
macaque newborns. Human newborns are not expected to be more physically active than
chimpanzee and macaque babies, considering the motor development in chimpanzees and
macaques occurs earlier than in humans (8-11). Further, the diet in all three species
before weaning does not differ markedly with respect to fat content (12, 13). Therefore,
these environmental effects cannot easily explain the difference in glutamate
concentration between human babies and chimpanzee or macaque babies.
Finally, we note that a previous study testing effects of human and chimpanzee diets on
gene expression profiles in mice found no detectable changes in mouse brain cortex,
while significant effects were observed in liver (14). This implies that at least on the gene
expression level, differences between human and chimpanzee diet appear to have limited
effect on brain. Thus, the environmental effects on metabolite concentration observed in
the above mentioned seven studies in human urine, muscle, mouse plasma, or rat liver
might actually be less prominent or absent in the brain tissue. Taken together, these
results provide no indication that excess of human-specific metabolic changes found in
PFC could be explained by the influence of environmental factors.
Supplementary Materials and Methods
Samples
Human samples were obtained from the NICHD Brain and Tissue Bank for
Developmental Disorders at the University of Maryland, the Netherlands Brain Bank, and
the Chinese Brain Bank Center. Informed consent for use of the human tissues for
research was obtained in writing from all donors or their next of kin. All subjects were
defined as normal controls by forensic pathologists at the corresponding brain bank. All
subjects suffered sudden death with no prolonged agonal state. Chimpanzee samples were
obtained from the Yerkes Primate Center, GA, USA, the Anthropological Institute &
Museum of the University of Zürich-Irchel, Switzerland, and the Biomedical Primate
Research Centre, the Netherlands (eight Western chimpanzees, one Central/Eastern, and
three of unknown origin). Rhesus macaque samples were obtained from the Suzhou
Experimental Animal Center, China. All non-human primates used in this study suffered
sudden deaths for reasons other their participation in this study and without any relation
to the tissue used.
For the metabolite analysis, approximately 100 mg of brain tissue was dissected from the
frozen postmortem brain tissue on dry ice. For prefrontal cortex samples, tissue
dissections were taken from the frontal part of the superior frontal gyrus, a cortical region
approximately corresponding to Brodmann area 10. For cerebellar cortex samples, tissue
dissections were taken from the lateral part of the cerebellar hemispheres. For all samples
we took care to dissect gray matter only. The tissue samples were powdered using mortar
and pestle cooled by dry ice prior to the metabolite extraction procedure. For protein
analysis, an adjacent 100 mg of brain tissue were dissected from a subset of human and
rhesus macaque individuals. For chimpanzees, all 11 individuals, plus an additional one,
were used in the protein analysis (Table S1).
GC-MS experiment
Metabolites were extracted from the frozen tissue powder by a
methanol:water:chloroform (2.5:1:1 (v/v/v)) extraction. In brief 100 mg of frozen
powdered tissue material was re-suspended 1 mL extraction solution containing 0.1 µg
mL-1 of U-13C6-sorbitol. The samples were incubated for 10 min at 4°C on an orbital
shaker. This step was followed by ultrasonication in a bath-type sonicator for 10 min at
room temperature. Finally the unsoluble tissue material was pelleted by a centrifugation
step (5 min; 14,000 g) and the supernatant transferred to a fresh 2 mL Eppendorf tube. To
separate the organic from the aqueous phase 300 µL H2O and 300 µL chloroform were
added to the supernatant, vortexed and centrifuged (2 min; 14,000 g). Subsequently, 200
µL of the upper, aqueous phase were collected and concentrated to complete dryness in a
speed vacuum at room temperature. Extract derivatization and GC-MS measurements
was performed according to (15).
The obtained metabolite concentration values (apex height of the quantitative compound
identifier mass) were normalized within each sample to the abundance of an internal
standard (13C sorbitol) and log2 transformed. To avoid negative values,
metabolite/standard ratios were scaled up by factor 10,000 prior to log2 transformation.
Next, for each species and each brain region, we filtered metabolites containing more
than 20% missing values. For the remaining metabolites, we estimated missing values
using k-th neighbors algorithm implemented in “knn” function in R package “EMV”. The
number of neighboring rows to estimate the missing values was set to five.
In addition to 13C sorbitol, we used two metabolites (unknown 22 and unknown 85) with
the lowest concentration variance across the full range of GC-MS profiles to
independently normalize the dataset. The methods of filtering metabolites and missing-
value estimation were the same as 13C sorbitol normalization.
Protein expression profiling
Proteins were extracted from 100 mg of frozen brain tissue from 12 human, 12
chimpanzee, and 12 rhesus brain samples. For each species, samples were processed in
two batches with similar age distribution between the batches (Table S1). The protein
extraction, identification and quantification procedure was carried out as described
previously (16, 17). Briefly, tissue samples were homogenized in ice-cold lysis buffer (8
M urea, 4% CHAPS, 65 mM DTT, 40 mM Tris, cocktail protease inhibitor, 100 mg of
tissue/1 ml) using an electric homogenizer. The resulting protein solutions were sonicated
on ice for a total of 3 minutes and then centrifugated at 25,000g for 1 hour at 4°C to
remove DNA, RNA and other cell debris. Next, the protein supernatants were
precipitated using 5× volumes of precipitation solution (ethanol: acetone: acetic acid =
50:50:0.1, volume ratio) at 4°C overnight, centrifugated and resolubilized in denaturing
buffer [6 M guanidine hydrochloride, 100mM Tris, cocktail protease inhibitor,
phosphatase inhibitors (1mM sodium orthovanadate and 1mM sodium fluoride), pH 8.3].
Protein concentration was determined using the Bradford assay. Next, 600µg of protein
from each sample was reduced with DTT (100µg / 1µl 1M DTT), alkylated with IAA
(100µg / 2µl 1M IAA), and precipitated again at 4°C overnight. After centrifugation, the
resulting precipitates were resolubilized in digestion buffer (100mM ammonium
bicarbonate) and incubated with Trypsin (enzyme:protein = 1:40, mass ratio) at 37°C for
20 hours, followed by ultrafiltration and lyophilization.
Each peptide sample was resolubilized in 50µl SCX loading buffer, loaded on a SCX
(Strong Cation Exchange) column (Column Technology Inc., CA, USA) and eluted using
a pH continuous gradient buffer (from pH 2.5-8.5), resulting in 10 fractions. Each of
these fractions was then automatically loaded on one of two RP (Reversed Phase)
alternative trap columns, by switching to the other RP column every 3 hours using a pH
continuous gradient buffer. Analysis was performed on the LTQ mass spectrometer
equipped with a metal needle electrospray interface mass spectrometer (ThermoFinnigan,
San Jose, CA, USA) in a data-dependent collection model (each full scan followed by ten
MS/MS scans of most intense ions).
We searched the mass spectra data of each sample. The peptides were identified by
searching against an IPI human database (IPI human v3.61) and a decoy database
containing reversed protein sequences separately, using the SEQUEST program in
BioWorks™ 3.2 software suite. A mass tolerance of 3.0 Da and one missed cleavage site
of trypsin were allowed. Cysteine carboxyamidomethylation was set as static
modification and methionine oxidation was set as variable modification. The output of
SEQUEST was filtered using the following cutoffs: Xcorr > 1.9 for charge +1; Xcorr >
2.2 for charge +2; Xcorr > 3.75 for charge +3 and delta CN above 0.1 for all the charges.
All output results were filtered and integrated to proteins by an in-house software
“BuildSummary”. Using the proteins from the decoy database as the estimation of the
false discovery in the real database, at a cutoff of a minimum total of 20 peptide counts
among all 12 samples to filter the proteins, the false discovery rate (FDR) of protein
identification was estimated to be 3%. We quantified the protein expression by counting
the number of identified peptides. Then, IPI peptide IDs were mapped to Ensembl gene
IDs using the Ensembl database (v56). Peptides mapping to multiple genes were
excluded. For Ensembl genes with multiple IPI peptide IDs, we calculated expression
levels as the mean of all peptides assigned to this gene. The expression values were then
log2 transformed and quantile normalized. Only proteins with total peptide counts ≥10
were used. The processed protein expression dataset is available at
http://www.picb.ac.cn/Comparative/data_methods/.
Identification of metabolites affected by postmortem delay
Here, we compared the abundance of each metabolite in the CBC and PFC between the
rhesus macaque that was dissected and frozen 5 hours after death and other macaques that
all had ≈20 minutes postmortem interval. To incorporate metabolic changes with age, we
first fitted spline curves with three degrees of freedom to each metabolite’s data in a
given brain region, excluding the macaque with prolonged postmortem delay. If, for both
replicates, the absolute value of residuals calculated based on the fitted curve were
greater for the macaque with prolonged postmortem delay than 1.96*standard deviation
of the residuals of all other individuals, we classified this metabolite as postmortem-delay
affected metabolite (Figure S3). For the metabolites not affected by postmortem delay,
we fitted spline curves with three degrees of freedom to the concentration profiles in all
individuals, including the macaque with prolonged postmortem delay. The residuals of
metabolite concentration measurements were calculated by subtracting the value
interpolated from the fitted curve from the actual measurement. The variance of residuals
distribution in a macaque with long postmortem delay and other macaques were
compared using F-test (Figure S4).
PCA analysis
For the PCA analysis, we merged the replicate measurements by taking the average
(mean). The PCA analysis was performed using the “prcomp” function in the R package
“stats”.
Method to calculate variation explained by each factor
We used Principle Variance Component Analysis (PVCA) to calculate the variance
explained by species, age, brain region, and individual (18, 19). In brief, principal
components with substantial contribution to the total variance greater than 5% were used
as response variables to fit a mixed linear model with different source of variation (age,
species, brain region, and individual) as random effects, using “lmer” function in the R
package “lme4”. The model fit via restricted maximum likelihood (REML) to get the
variance component estimates. The weighted average variance was then calculated based
on the eigenvalues retained from the principal component analysis. In total, the first 5
principal components with combined contribution to the total variance greater than 52%
were used.
Identification of age-related metabolites/proteins
We tested the effect of age on expression levels of 92 metabolites and 1,951 proteins
using polynomial regression models, as described previously (17, 20). Briefly, for each
metabolite/protein, we chose the best polynomial regression with age as predictor and
expression level as response, using families of polynomial regression models and the
“adjusted r2” criterion (21). The significance of the chosen regression model was
estimated using the F-test. The false discovery rate was calculated by random
permutation of age.
Identification of pair-wise differentially expressed metabolites/proteins
The test for differential expression between a pair of species is based on analysis of
covariance, or ANCOVA (21). For each of 63 and 66 age-related metabolites in CBC and
PFC, respectively and, 1,271 age-related proteins in PFC identified using the method
described above, we used the age model chosen in the above-described age-test and then
tested if adding species-specific parameters to the model significantly improved its fit to
the expression levels in the other species. The null model (no species-specific
parameters) and alternative models were compared using the F-test. In order to preserve
the age-structure in the data, the permutation test was performed by dividing the age-
range into 5 sections, and randomly permuting species’ identifiers among samples within
each section. The differential expression test was performed twice on each species pair,
using either species as a reference (for choosing the age-test model). For each
metabolite/protein, if both of the two tests were significant at a defined cutoff
(FDR<1%), we considered it as differentially expressed between the two species. The
same methods were applied to the full dataset and a subset of 11 individuals per species.
Identification of species-specific metabolites/proteins
For the metabolite data analysis based on the full set of individuals, we used chimpanzee
metabolite measurements to assigned metabolic changes for the 43 and 49 metabolites
with significant differences between humans and macaques in CBC and PFC,
respectively, to the evolutionary lineages: the human lineage and the lineage between
rhesus macaques and the common ancestor of humans and chimpanzees. To do so, we
fitted the metabolite concentration data using spline curves with three degrees of freedom
and interpolated metabolite levels for 103 age points, based on the union of actual age
points in the three species. The distance between a pair of species was defined as the
absolute value of the difference between two vectors of 103 predicted metabolite levels.
We then used a one-sided Wilcoxon test to compare human-chimpanzee difference to
chimpanzee-macaque difference. If the human-chimpanzee difference was significantly
greater than chimpanzee-macaque difference, we classified these metabolites as human-
specific. If the chimpanzee-macaque was significantly greater than human-chimpanzee
difference, we classified these metabolites as macaque-specific. The rest of age-related
metabolites with significant difference between humans and macaques were classified as
uncategorized.
For the two 11-individuals-per-species datasets, we tested the human-chimpanzee,
macaque-chimpanzee and human-macaque differences in the pair-wise comparisons as
described above. If both the human-chimpanzee and human-macaque differences were
significant, but the macaque-chimpanzee difference was not, we classified these
metabolites as human-specific. If both the human-chimpanzee and macaque-chimpanzee
differences were significant, but the human-macaque difference was not, we classified
these metabolites as chimpanzee-specific. If both the human-macaque and macaque-
chimpanzee differences were significant, but the human-chimpanzee difference was not,
we classified these metabolites as macaque-specific. The rest of age-related metabolites
were classified as uncategorized.
For the proteins data, we used chimpanzee protein measurements to sort the 288 human-
macaque protein differences in PFC, identified by the method mentioned above into four
categories. We tested the human-chimpanzee and macaque-chimpanzee differential
expression with the same method we used for the human-macaque test. Those proteins
whose profiles were significantly different (p<0.005) in humans from the profiles in both
chimpanzee and macaque, but not different between chimpanzee and macaque, were
classified as human-specific proteins. Similar methods were used to classify the
chimpanzee-specific and macaque-specific proteins. The rest of proteins were classified
as uncategorized.
Types of metabolic differences among species
The differential expressed metabolites were grouped into different categories by the
following two steps: (1) The ages of human and rhesus macaque were normalized to the
chimpanzee lifespan-scale by dividing the sample age by the maximum lifespan of the
corresponding species (human: 120 years, rhesus macaque: 40 years) and multiplying by
the maximum lifespan of chimpanzee (60 years) (22). Normalized ages were used to redo
the analysis to identify differential metabolites. The metabolites that did not show
significant difference after lifespan normalization were classified as different due to the
lifespan difference. (2) The data after removing the mean, by standardizing metabolite
concentration level in each species to mean = 0 and standard deviation = 1, was used to
identify differential metabolites. The metabolites that did not show significant difference
after mean and variance normalization were classified as different due to the mean
concentration differences. The remaining metabolites were classified as showing different
patterns of metabolic change with age among species.
Heatmap and hierarchical clustering
For each of the 24 and 6 human-specific metabolites identified as described above in PFC
and CBC, respectively, we interpolated metabolite concentration values at 50 time points
uniformly distributed over the species’ lifespan, based on the spline curves with three
degrees of freedom, fitted to the metabolite measurements in each species and each brain
region. All species’ curves were extrapolated to the minimum and maximum age of the
dataset (0 and 98 years). In a given brain region, the fitted 150 concentration values for
each metabolite were normalized to mean equal zero and standard deviation equal one
before clustering. The heatmap and hierarchical clustering dendrogram were drawn by
the “heatmap.2” function in the R package “gplots”. The clustering was based on 1-
Pearson correlation coefficient, as the distance measure. For the human-specific
metabolites identified in PFC, the clustering was done based on the PFC metabolite
concentrations. For the human-specific metabolites identified in CBC, the hierarchical
clustering was done based on the CBC metabolite concentrations.
Test relationship between metabolites and proteins
The proteins connected by a single edge to the detected metabolites in the pathways were
manually collected from the KEGG database (http://www.kegg.com). We found 38 such
proteins for 8 human-specific metabolites and 106 proteins for 23 non-human-specific
metabolites. The 11 proteins that were shared among human-specific metabolites and
non-human-specific metabolites were assigned to human specific metabolites. The
Pearson correlation between profiles of metabolite-protein pairs were calculated using the
50 points interpolated from spline curves fitted to the metabolite and protein data, with
three degrees of freedom. Three Pearson correlation coefficient cutoffs (|r|>0.3, |r|>0.5,
|r|>0.7) were applied to define correlated metabolite-protein pairs. The excess of
observed correlated metabolite-protein pairs was tested by random substitutions of
proteins’ expression profiles by expression profiles of other detected proteins 1,000
times. The p-value was calculated as the frequency of the event when the number of
random pairs identified using a given correlation coefficient cutoff was greater or equal
to the number of pairs found in the actual data. The protein expression divergence
between human and chimpanzee was calculated as the Euclidian distance between the
two vectors of 50 points interpolated from spline curves fitted to the protein expression
data with three degrees of freedom. The human-specific protein expression divergence
was calculated as the ratio between human-macaque expression divergence and
chimpanzee-macaque expression divergence. We used a one-sided Wilcoxon test to
compare both the human-chimpanzee divergence measurements and human-specific
divergence measurements between proteins associated with the human-specific
metabolites and proteins associated with non-human-specific metabolites.
KEGG pathway analysis
The xml files of human KEGG pathways were downloaded from ftp://ftp.genome.jp
/pub/kegg/xml/kgml and parsed by R package “KEGGgraph”. The human-specific
metabolites and detected proteins were mapped to the KEGG pathways. We only
considered the pathways containing at least one human-specific metabolite and more than
10 detected proteins. Using these criteria, we obtained 13 such KEGG pathways
including 5 human-specific metabolites and 209 detected proteins. Then, using the
proteins detected in each pathway, we calculated the human-specific protein expression
divergence as described above. We then estimated the significance of these estimates by
randomly assigning the same number of detected proteins to each pathway and
calculating the human-specific protein expression divergence 1,000 times. The p-value
for each pathway was calculated as the frequency of the event when the median value of
the human-specific divergence in a random sub-sampling was greater or equal to the one
found in the real pathway. To estimate FDR when the particular test is called significant,
we applied the Storey and Tibshirani approach to calculate the q-value among the p-
values from the 13 KEGG pathways, using R package “qvalue” (23). This resulted in
three human-specific metabolites associated with four pathways showing significant
excess of the human-specific protein expression divergence. The Pearson correlations
between human-specific metabolites and proteins in these four pathways were calculated
using the 50 points interpolated from spline curves, fitted to the metabolite and protein
data, with three degrees of freedom. The Pearson correlation coefficient cutoff was set to
|r|>0.7.
Supplementary Figures
Figure S1. The sample age distributions in the three species.
A. The actual age distribution of sampled individuals in the three species (Ch-
chimpanzee, Hu-human, Ma-rhesus macaque). The circles represent individuals
(purple-chimpanzee, red-human, green-rhesus macaque). The y-axis shows individuals’
age on the scale: (age in days)^0.25. We used this scale because it provides a nearly
uniform distribution of sample ages across the lifespan. Using non-transformed age or
logarithm-transformed age leads to diminished resolution at either young or old age. B.
Subset of 11 human and 11 rhesus macaque individuals sampled using the stage-of-life
matching strategy (indicated by arrows): scaling ages linearly within each species to the
same maximal lifespan, using 120 years for humans, 60 for chimpanzees and 40 for
rhesus macaques as a reference (22). In this approach, 20 years old macaque would match
60 years old human. C. Subset of 11 human and 11 rhesus macaque individuals sampled
using the chronological matching strategy (indicated by arrows): using calendar age
directly, i.e. 20 years old macaque matching 20 years old human.
Figure S2. Correlation between technical replicates.
Sown are distributions of Pearson correlation coefficients (PCC) based on metabolite
concentration measurements between technical replicates in prefrontal cortex (PFC) and
cerebellar cortex (CBC) of humans (Hu), rhesus macaques (Ma) and chimpanzees (Ch).
The bottom and top of each box are the 25th and 75th percentiles of the PCC distribution,
and the line within the box is the median. The whiskers show 1.5 interquartile range of
the lower and the upper quartiles. The data points not included between the whiskers are
plotted as circles.
Figure S3. Concentration profiles of metabolites affected by postmortem delay.
The points represent metabolite concentrations in CBC (odd panels) and PFC (even
panels) of macaque individuals with short (<20 minutes) postmortem delay (gray) and
one macaque individual with long (≈5 hours) postmortem delay (red). The replicate
measurements for each individual are shown separately. The lines are the spline curves
fitted to the data points excluding the long postmortem delay individual with three
degrees of freedom. The x-axis shows individual’s ages on the scale: (age in days)^0.25.
The y-axis shows the normalized GC-MS measurements representing the metabolite
concentrations. The titles show metabolite annotation. “Unknown” stands for
unannotated metabolites. The brain region identity is displayed below the x-axis labels.
Figure S4. Variation among metabolites not affected by postmortem delay.
Shown are the distributions of residuals of metabolite concentrations measured for
metabolites not affected by postmortem delay in a macaque with long postmortem delay
(red) and other macaques (gray) in CBC (A) and in PFC (B). Residuals were calculated
residuals to the spline curve fitted to the full metabolite data series with three degrees of
freedom. The variance of the red and gray distributions was compared by the F-test.
There were no significant differences in residuals distributions: the p-value equaled 0.254
in CBC and 0.567 in PFC. Thus, metabolites not affected by postmortem delay according
to our criteria (see Materials and Methods) did not display greater variation in a macaque
with 5 hours postmortem delay than in other macaques with postmortem delay less than
20 minutes (see Materials and Methods)
Figure S5. The proportion of variance explained by species, age, brain region, and
individual effects in metabolite data.
The normalized metabolite data with 92 detected metabolites from the three species were
used in the principal component analysis. Principal components with substantial
contribution to the total variance greater than 5% were used as response variables to fit a
mixed linear model with different source of variability (age, species, brain region and
individual) as random effects (18, 19). The model was fitted via restricted maximum
likelihood (REML) and was used to obtain the variance component estimates. The
weighted average variance was then calculated based on the eigenvalues retained from
the principal component analysis. In total, the first 5 principal components with combined
contribution to the total variance >52% were used.
Figure S6. Concentration profiles of taurine.
The concentration profiles of taurine in PFC (A) and CBC (B) in humans (red) and rhesus
macaques (blue). The points represent metabolite concentration in each individual. The
lines represent fitted linear models. The x-axis shows individual’s ages on the scale: (age
in days)^0.25. The y-axis shows the normalized GC-MS measurements representing
metabolite concentrations. The concentration of taurine shows significant decrease with
age (human PFC: r=-0.37, p<0.01; human CBC: r=-0.53, p<10-5; macaque PFC: r=-0.28,
p<0.05; macaque CBC: r=-0.51, p<0.001).
Figure S7. Concentration profiles of metabolites that differ significantly between humans
and rhesus macaques in both CBC and PFC.
The profiles of 30 metabolites in CBC (A) and PFC (B) classified as significantly
different between humans (red) and rhesus macaques (blue) in both brain regions. The
points represent metabolite concentration in each individual. The lines are the spline
curves fitted to the data points with three degrees of freedom. The x-axis shows
individual’s ages on the scale: (age in days)^0.25. The y-axis shows the normalized GC-
MS measurements representing metabolite concentrations. The titles show metabolite
annotation. “Unknown” stands for unannotated metabolites.
Figure S8. Proportions of annotated metabolites in the three phylogenetic categories.
The colors show the three phylogenetic categories: human-specific (red), macaque-
specific (gray) and uncategorized metabolites (white). Shown are the 23 and 29 annotated
metabolites with significant difference in concentration profiles between humans and
rhesus macaques in CBC and PFC, respectively, identified using the full set of
individuals. The numbers show numbers of metabolites in the corresponding category.
Figure S9. Proportions of metabolites in the four phylogenetic categories calculated using
11 chimpanzees and subsets of 11 human and 11 rhesus macaque individuals.
A. The distribution of metabolites in the four phylogenetic categories calculated using the
subset of 11 stage-of-life matched individuals per species. B. The distribution of
metabolites in the four phylogenetic categories calculated using the subset of 11
chronological matched individuals per species. C. The overlap of between the stage-of-
life matching strategy and chronological matching strategy. The colors represent the four
phylogenetic categories: human-specific (red), macaque-specific (gray), chimpanzee-
specific (sky-blue) and uncategorized metabolites (white). The numbers indicate numbers
of metabolites in the corresponding category.
Figure S10. Proportions of metabolites in the three phylogenetic categories calculated
using different normalization strategies.
A. The analysis results based on 13C sorbitol normalization. B. The analysis results based
on “unknown 22” normalization. C. The analysis results based on “unknown 85”
normalization. D. Overlap among the results from the three normalization methods.
On each panel, shown are: (left) the proportions of metabolites with significant difference
in concentration profiles between humans and macaques identified in each brain region;
(middle) the proportions of metabolites with significant difference in concentration
profiles between humans and macaques identified in both brain regions; (right) the
proportions of annotated metabolites with significant difference in concentration profiles
between humans and macaques identified in each brain region.
The colors represent the three phylogenetic categories: human-specific (red), macaque-
specific (gray) and uncategorized metabolites (white). The numbers show numbers of
metabolites in the corresponding category. The calculations were based on the full set of
individuals.
Figure S11. The numbers of metabolite-protein pairs with correlated expression profiles.
Shown are the distributions of the numbers of metabolite-protein pairs that pass a given
correlation cutoff for the randomly chosen pairs (gray) and the actual number of
correlated metabolite-protein pairs containing metabolites with human-specific
concentration profiles (red). The Pearson correlation coefficient was calculated for 47
metabolite-enzyme pairs containing metabolites with human-specific concentration
profiles and for the same number of random metabolite-protein pairs 1,000 times. The p-
value was calculated as the frequency that the number of correlated random pairs was
greater or equal to that of true pairs correlation coefficient cutoffs (|r|) equal 0.3 (A), 0.5
(B), and 0.7 (C). The numbers of real metabolite-protein pairs passing these cutoffs were
35, 21, and 9, respectively. The corresponding p-values were 0.001, 0.001, and 0.003.
Figure S12. Expression divergence of proteins associated with human-specific and non-
human-specific metabolites.
The protein’s association with metabolites is based on KEGG annotation. The proteins
are divided into two categories: 13 proteins associated with human-specific metabolites
by a single network edge (HS) and 24 proteins associated with non-human-specific
metabolites only (non-HS). A. The expression divergence between humans and
chimpanzees. The divergence was calculated as Euclidian distance between protein
expression profiles in humans and chimpanzees. B. The ratio of human-macaque and
chimpanzee-macaque expression divergence. The expression divergence was calculated
as Euclidian distance between protein expression profiles in the two species.
Figure S13. The four KEGG pathways with significantly greater human-specific protein
expression divergence associated with three metabolites with human-specific
concentration profiles.
The small red circles represent metabolites with human-specific concentration profiles:
glutamate, histidine and spermidine. The boxes in the figures represent the genes in the
pathways. The colors of the boxes indicate: proteins with significant human-specific
expression (red), proteins with significant human-chimpanzee expression divergence
(orange), proteins with significant correlation to the corresponding metabolite profile
(pink), detected proteins (gray). Two type of colors (fill and border/text of boxes) are
used when two type of proteins are included in the nodes. The pathways are drawn by the
KEGG Mapper (http://www.genome.jp/kegg/tool/color_pathway.html).
A B
C D
Figure S14. Expression profiles of GLUD2 mRNA and glutamate in the primate
prefrontal cortex.
Colors represent species (human-red; chimpanzee-purple; macaque-blue). The points
represent the expression levels in each sample. The lines are the spline curves fitted to the
data points. The x-axis shows sample’s ages on scale: (age in days)^(1/4).
Supplementary Tables
Table S1. Sample information.
Sample ID Species Gender Age (day)
Postmortem delay (hour) Experiment* Batch
C1 Chimpanzee male 0 NA M; P 2 C2 Chimpanzee female 1 NA M; P 2 C3 Chimpanzee male 7 NA M; P 1 C4 Chimpanzee male 8 NA M; P 1 C5 Chimpanzee male 39 NA M; P 2 C6 Chimpanzee female 45 NA M; P 1 C7 Chimpanzee female 522 NA M; P 2 C8 Chimpanzee female 2920 NA M; P 2 C9 Chimpanzee male 4380 NA M; P 1 C10 Chimpanzee male 4380 NA M; P 2 C11 Chimpanzee male 14600 NA M; P 1 C12 Chimpanzee female 16060 NA P 1 M1 Macaque male 16 0 M; P 1 M2 Macaque male 20 0 M; P 2 M3 Macaque male 22 0 M M4 Macaque male 23 0 M M5 Macaque male 24 0 M M6 Macaque male 151 0 M M7 Macaque male 153 0 M; P 1 M8 Macaque male 179 0 M M9 Macaque male 182 0 M
M10 Macaque male 207 0 P 2 M11 Macaque male 215 0 M M12 Macaque male 237 0 M M14 Macaque male 310 0 P 2 M15 Macaque male 353 0 M M16 Macaque male 445 0 M M17 Macaque male 449 0 M M18 Macaque male 462 0 M M19 Macaque male 471 0 M M20 Macaque male 479 0 M M21 Macaque male 535 0 M M22 Macaque male 607 0 M M23 Macaque male 647 0 M M24 Macaque male 659 0 M M25 Macaque male 739 0 M; P 2 M26 Macaque male 831 0 M M27 Macaque male 1135 0 M M28 Macaque male 1175 0 M M30 Macaque male 1487 0 M; P 1 M31 Macaque male 2355 0 M
M32 Macaque male 2570 0 M M33 Macaque male 2936 0 M M34 Macaque male 3322 0 M M35 Macaque male 3389 0 M; P 2 M36 Macaque male 3978 0 M M37 Macaque male 4361 0 M M38 Macaque male 4762 0 M M39 Macaque male 5131 0 M M40 Macaque male 5459 0 M M41 Macaque male 5478 0 M M42 Macaque male 6227 0 M M43 Macaque male 7391 0 M; P 1 M44 Macaque male 7673 0 M M45 Macaque male 8104 0 M; P 2 M46 Macaque female 9291 0 M M47 Macaque male 9518 0 M; P 1 M48 Macaque female 9490 5 M M49 Macaque female 10220 0 M; P 2 H1 Human male 1 NA M; P 1 H2 Human male 4 NA M; P 2 H3 Human female 19 NA M H4 Human male 34 NA M; P 1 H5 Human male 94 NA M H6 Human female 186 NA M H7 Human female 196 NA M H8 Human male 204 NA M; P 2 H9 Human female 328 NA M
H10 Human male 443 19 M H11 Human male 2827 12 M H12 Human male 2861 18 M H13 Human male 2922 5 M; P 1 H14 Human male 4844 5 M H15 Human male 5105 13 M; P 2 H16 Human male 5170 16 M H17 Human male 5308 16 M H18 Human male 6111 15 M H19 Human male 7555 12 M H20 Human male 8006 13 M H21 Human male 8364 4 M H22 Human male 9277 19 M; P 1 H23 Human male 9743 18 M H24 Human male 10885 12 M H25 Human male 10948 19 M H26 Human male 13587 12 M H27 Human male 14101 5 M H29 Human male 14309 12 M
H30 Human male 15206 15 M H31 Human male 15519 19 M H32 Human male 18406 8 M H33 Human male 18615 NA M H34 Human male 19457 17 M; P 2 H35 Human male 19566 19 M H36 Human male 19930 17 M H37 Human male 20030 19 M H38 Human male 21050 16 M H39 Human male 21204 9 M H40 Human male 21365 17 M H41 Human male 22452 6 M H42 Human female 22265 NA M H43 Human male 24090 NA M; P 1 H44 Human male 25786 28 M H45 Human male 26829 21 M H46 Human female 26645 NA M H47 Human male 29200 NA M; P 2 H48 Human male 32120 NA M; P 1 H49 Human female 32850 NA M H50 Human male 35770 NA M; P 2
* M: metabolite --- GC-MS experiment; P: proteomics --- label free mass-spectrometry
Table S2. List of detected metabolites.
The 1-labels indicate metabolites detected in at least 80% of samples in a corresponding tissue in a given species. The 0-labels indicate metabolites not passing the detection cutoff.
CBC PFC chimpanzee human macaque chimpanzee human macaque
3-hydroxybutyric acid 1 1 1 1 1 1 glycerol 1 1 1 1 1 1 leucine 1 1 1 1 1 1
isoleucine 1 1 1 1 1 1 glycine 1 1 1 1 1 1
benzoic acid 1 1 1 1 1 1 serine 1 1 1 1 1 1
unknown16 1 1 1 1 1 1 succinic acid 1 1 1 1 1 1
threonine 1 1 1 1 1 1 unknown138 1 1 1 1 1 1 fumaric acid 1 1 1 1 1 1 unknown20 1 1 1 1 1 1 unknown23 1 1 1 1 1 1
erythronic acid 1 1 1 1 1 1 4-hydroxyproline 1 1 1 1 1 1
gaba 1 1 1 1 1 1 threonic acid 1 1 1 1 1 1 methionine 1 1 1 1 1 1 creatinine 1 1 1 1 1 1 oxoproline 1 1 1 1 1 1 glutamate 1 1 1 1 1 1
nicotinamide 1 1 1 1 1 1 putrescine 1 1 1 1 1 1
dodecanoic acid 1 1 1 1 1 1 glutamine 1 1 1 1 1 1
phenylalanine 1 1 1 1 1 1 unknown38 1 1 1 1 1 1 unknown45 1 1 1 1 1 1 unknown48 1 1 1 1 1 1
glucose-1-phosphate 1 1 1 1 1 1 ornithine 1 1 1 1 1 1
aspartic acid 1 1 1 1 1 1 glycerol-3-phosphate 1 1 1 1 1 1
glucose 1 1 1 1 1 1 citric acid 1 1 1 1 1 1
unknown55 1 1 1 1 1 1 dehydroascorbic acid 1 1 1 1 1 1
unknown62 1 1 1 1 1 1 histidine 1 1 1 1 1 1
pantothenic acid 1 1 1 1 1 1 unknown71 1 1 1 1 1 1
hexadecanoic acid 1 1 1 1 1 1 unknown135 1 1 1 1 1 1 unknown128 1 1 1 1 1 1
octadecenoic acid 1 1 1 1 1 1 6-phosphogluconic acid 1 1 1 1 1 1
tryptophan 1 1 1 1 1 1 uridine 1 1 1 1 1 1 alanine 1 1 1 1 1 1
butyric acid 1 1 1 1 1 1 unknown40 1 1 1 1 1 1
tetradecanoic acid 1 1 1 1 1 1 unknown91 1 1 1 1 1 1 cholesterol 1 1 1 1 1 1
hydroxypyridine 1 1 1 1 1 1 unknown5 1 1 1 1 1 1 unknown7 1 1 1 1 1 1
3-hydroxypyridine 1 1 1 1 1 1 ribitol 1 1 1 1 1 1
fructose 1 1 1 1 1 1 unknown66 1 1 1 1 1 1
unknown137 1 1 1 1 1 1 unknown36 1 1 1 1 1 1 unknown8 1 1 1 1 1 1
glycerol-2-phosphate 1 1 1 1 1 1 tyrosine 1 1 1 1 1 1
unknown79 1 1 1 1 1 1 pyruvic acid 1 1 1 1 1 1 unknown130 1 1 1 1 1 1 unknown27 1 1 1 1 1 1
unknown136 1 1 1 1 1 1 heptadecanoic acid 1 1 1 1 1 1
unknown1 1 1 1 1 1 1 unknown19 1 1 1 1 1 1
valine 1 1 1 1 1 1 unknown108 1 1 1 1 1 1 oxalic acid 1 1 1 1 1 1
decanoic acid 1 1 1 1 1 1 unknown93 1 1 1 1 1 1
unknown132 1 1 1 1 1 1 unknown3 1 1 1 1 1 1
taurine 1 1 1 1 1 1 unknown64 1 1 1 1 1 1
tetracosanoic acid 1 1 1 1 1 1 glycolic acid 1 1 1 1 1 1 unknown126 1 1 1 1 1 1 unknown85 1 1 1 1 1 1 unknown6 1 1 1 1 1 1 unknown33 1 1 1 1 1 1 unknown49 1 1 1 1 1 1 unknown2 1 1 1 1 1 1 unknown22 1 1 1 1 1 1
unknown30 1 1 1 1 1 1 unknown4 1 1 1 1 1 0 unknown35 1 1 1 1 0 1
2-amino adipic acid 1 1 1 0 1 1 unknown133 1 0 1 1 1 1 unknown43 1 0 1 1 1 1 unknown99 1 1 0 1 1 1 spermidine 1 1 1 1 1 0 unknown21 1 1 0 1 1 0
unknown102 1 1 1 0 0 1 unknown74 1 1 1 1 0 0
fructose2 1 0 1 1 1 0 glucose-6-phosphate2 0 1 1 0 1 1
unknown94 1 0 1 0 1 1 unknown12 1 1 1 0 1 0
unknown134 1 1 0 1 0 1 unknown42 1 1 1 0 0 0 unknown83 1 1 0 0 1 0 ascorbic acid 1 0 1 1 0 0 unknown131 1 0 0 1 0 0 adipic acid 0 1 0 0 1 0 unknown18 1 0 1 0 0 0 unknown10 1 0 0 0 0 1
unknown139 0 0 1 0 0 0 unknown51 0 0 1 0 0 0
Table S3. The 26 metabolites affected by postmortem delay in CBC and PFC.
CBC PFC 2-amino adipic acid yes no 4-hydroxyproline yes no
6-phosphogluconic acid no yes alanine no yes
butyric acid no yes citric acid yes no
unknown135 yes no fructose yes yes fructose2 yes yes
gaba yes no glucose yes yes
glucose-1-phosphate no yes glucose-6-phosphate2 yes yes
glycerol yes no glycerol-2-phosphate yes no glycerol-3-phosphate yes yes
pantothenic acid yes no unknown10 yes no
unknown126 yes yes unknown130 yes no unknown23 no yes unknown36 yes no unknown55 no yes unknown62 yes no unknown64 yes no unknown93 yes no
Table S4. The age-related metabolites.
The 1-labels indicate metabolites with significant concentration changes with age. The 0-labels indicate metabolites not passing the significance cutoff.
CBC PFC chimpanzee human macaque chimpanzee human macaque
3-hydroxybutyric acid 0 0 1 0 0 1 octadecenoic acid 0 0 0 0 1 0
decanoic acid 0 0 0 0 1 0 ascorbic acid 1 1 1 1 1 1 aspartic acid 0 1 1 0 1 1 benzoic acid 0 1 1 0 0 0 cholesterol 0 1 1 0 1 1 creatinine 0 1 0 0 1 1
dehydroascorbic acid 0 1 1 1 1 0 erythronic acid 0 0 1 0 0 0
fumaric acid 0 0 0 0 1 0 glutamate 0 1 0 0 1 0 glutamine 0 1 1 0 0 1
glycine 0 1 1 0 1 0 glycolic acid 0 1 1 0 1 1
hexadecanoic acid 0 0 0 0 1 0 histidine 0 1 1 0 1 1
hydroxypyridine 0 0 1 0 0 1 isoleucine 0 1 0 0 1 0
leucine 0 1 1 0 1 0 methionine 0 1 1 0 1 0
nicotinamide 0 0 0 0 1 1 ornithine 0 1 1 0 1 0
oxalic acid 0 0 1 0 0 1 oxoproline 0 1 1 0 1 0
phenylalanine 0 1 1 0 1 0 putrescine 0 1 0 0 1 0
pyruvic acid 0 0 0 0 1 1 ribitol 0 1 0 0 1 0 serine 0 1 1 0 1 0
spermidine 0 0 1 0 0 1 succinic acid 0 1 1 0 0 1
taurine 0 1 1 1 1 0 tetracosanoic acid 0 0 0 0 1 0 tetradecanoic acid 0 0 0 0 1 0
threonic acid 0 1 1 0 1 0 threonine 0 1 1 0 1 0
tryptophan 0 1 1 0 1 0 tyrosine 0 1 1 0 1 0 uridine 0 0 0 0 1 0 valine 0 1 0 0 1 0
unknown1 0 0 1 0 0 0 unknown102 0 1 0 0 0 0
unknown108 0 0 1 0 0 1 unknown12 0 1 1 0 0 0 unknown128 0 1 0 0 0 1 unknown131 0 0 1 0 0 0 unknown132 0 1 1 1 1 1 unknown133 0 1 0 0 1 0 unknown134 0 1 0 0 0 0 unknown136 0 1 0 0 1 0 unknown137 0 1 0 0 1 0 unknown138 0 1 1 0 1 1 unknown18 0 1 1 0 0 0 unknown19 0 0 1 0 0 0 unknown2 0 1 1 0 0 1
unknown20 0 0 0 0 1 0 unknown22 0 0 0 0 1 0 unknown27 0 1 1 0 1 0 unknown3 0 1 0 0 1 0
unknown30 0 0 0 0 1 0 unknown35 0 0 1 0 0 0 unknown4 0 1 1 0 1 0
unknown40 0 0 1 0 0 0 unknown42 0 1 1 0 0 0 unknown43 0 0 1 0 0 0 unknown45 0 0 1 0 1 1 unknown48 1 1 1 1 1 1 unknown49 0 1 0 0 1 0 unknown5 0 0 0 0 1 0 unknown6 0 0 0 0 1 0
unknown66 0 1 0 0 1 1 unknown7 0 0 0 0 0 1
unknown71 0 1 1 0 0 0 unknown74 0 0 1 0 0 1 unknown79 0 1 0 0 0 1 unknown83 0 0 0 0 1 0 unknown85 0 0 0 0 1 0 unknown91 0 1 1 0 0 1 unknown94 0 0 0 0 1 0 unknown99 0 0 1 0 0 0
Table S5. The species-specific metabolites.
CBC PFC
human macaque chimp. human† macaque chimp.
F* SS* SC F SS SC SS SC F SS SC F SS SC SS SC
hexadecanoic acid 0 0 0 0 0 0 0 0 1 1 1 0 0 0 0 0
Octadecenoic acid 0 0 0 0 0 0 0 0 1 1 1 0 0 0 0 0
unknown128 0 0 0 1 0 0 0 0 1 1 1 0 0 0 0 0
glycolic acid 1 0 0 0 0 0 0 0 1 1 0 0 0 0 0 0
spermidine 0 0 0 0 0 0 0 0 1 1 0 0 0 0 0 0
unknown20 0 0 0 0 0 0 0 0 1 1 0 0 0 0 0 0
unknown6 0 0 0 0 0 0 0 0 1 1 0 0 0 0 0 0
glutamate 0 0 0 0 0 0 0 0 1 0 1 0 0 0 0 0
ribitol 0 0 0 1 0 0 0 0 1 0 1 0 0 0 0 0
tetracosanoic acid 0 0 0 0 0 0 0 0 1 0 1 0 0 0 0 0
ascorbic acid 1 0 0 0 0 0 0 0 0 1 1 1 0 0 0 0
Decanoic acid 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0
histidine 0 0 0 1 1 1 0 0 1 0 0 0 0 0 0 0
hydroxypyridine 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0
oxoproline 0 0 0 0 0 0 1 0 1 0 0 0 0 0 0 0
unknown2 0 0 0 1 1 1 0 0 1 0 0 0 0 0 0 0
unknown22 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0
unknown27 0 0 0 1 0 0 0 0 1 0 0 0 0 0 0 0
unknown30 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0
unknown4 0 0 0 1 1 0 0 0 1 0 0 0 0 0 0 0
unknown49 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0
unknown5 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0
unknown79 0 0 0 0 0 0 1 1 1 0 0 0 0 0 0 0
unknown83 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0
unknown85 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0
taurine 1 1 0 0 0 0 0 0 0 1 0 0 0 0 0 0
unknown45 0 0 0 1 0 0 0 0 0 1 0 1 0 0 0 0
cholesterol 0 1 1 0 0 0 0 0 0 0 1 1 0 0 0 0 3-hydroxybutyric
acid 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1
aspartic acid 0 0 0 1 1 1 0 0 0 0 0 1 0 0 1 0
benzoic acid 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 dehydroascorbic
acid 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0
erythronic acid 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0
fumaric acid 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0
glutamine 0 0 0 1 0 0 1 1 0 0 0 0 0 0 0 0
glycine 0 0 0 1 1 0 0 0 0 0 0 1 0 0 0 0
isoleucine 0 0 0 1 1 0 0 0 0 0 0 1 1 1 0 0
leucine 0 0 0 1 1 0 0 0 0 0 0 1 0 0 0 0
methionine 0 0 0 1 1 0 0 0 0 0 0 1 0 0 0 0
nicotinamide 0 0 0 0 0 1 0 0 0 0 0 1 0 0 0 0
ornithine 0 0 0 1 0 0 0 0 0 0 0 1 0 0 0 0
oxalic acid 0 0 0 1 0 0 0 0 0 0 0 1 0 0 0 0
phenylalanine 0 0 0 1 1 1 0 0 0 0 0 1 0 0 0 0
putrescine 0 1 1 0 0 0 0 0 0 0 0 1 0 0 0 0
unknown138 0 0 0 1 0 1 0 0 0 0 0 1 0 0 0 0
serine 0 0 0 1 1 0 0 0 0 0 0 0 0 0 0 0
threonic acid 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0
threonine 0 0 0 1 1 1 0 0 0 0 0 1 1 1 0 0
tryptophan 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0
tyrosine 0 0 0 1 1 1 0 0 0 0 0 0 0 0 0 0
unknown1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
unknown108 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0
unknown12 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
unknown132 0 0 0 1 0 0 0 0 0 0 0 0 0 1 0 0
unknown134 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1
unknown137 0 0 0 1 1 0 0 0 0 0 0 1 0 0 0 0
unknown18 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
unknown19 0 0 0 1 0 1 0 0 0 0 0 0 0 0 0 0
unknown21 0 0 0 0 1 0 0 0 0 0 0 0 0 1 0 0
unknown40 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0
unknown42 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0
unknown48 0 0 0 1 0 0 0 0 0 0 0 1 1 1 0 0
unknown66 0 0 0 1 1 0 0 0 0 0 0 0 0 0 0 0
unknown71 0 0 0 1 0 0 0 1 0 0 0 0 0 0 0 0
unknown74 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0
unknown91 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0
valine 0 0 0 1 1 0 0 0 0 0 0 1 1 1 0 0
* F: full dataset; SS: 11 individual subset with stage-of-life approach; SC: 11 individual
subset with chronological approach.
† The overlap between 24 and 10 human specific metabolites identified in PFC using full
set or 11-individuals subset with stage-of-life approach is 7. To estimate the possibility
that the overlap is observed by chance, we randomly choose 10 metabolites from 92
metabolites 1,000 times and calculate the p-value which is defined as the frequency of the
event in which the overlap between randomly chosen metabolites and 24 human specific
is not less than 7. The calculated p-value equals 0.005. Accordingly, the overlaps of
human specific metabolite in PFC between full set and 11-individual subset with
chronological approach or between both subsets are more than expected. There are 6
shared between full set and 11-individual subset with chronological approach with p-
value equals 0.004 and 4 between both subsets with p-value equals 0.001.
Table S6. Summary of the 118 metabolites.
annotated hypothetical total detected 61 57 118 postmortem 17 9 26 used in analyses 44 48 92
CBC* 32 31 63 PFC* 39 27 66
age-related in 3 species
Union** 41 40 81 CBC* 27 16 43 PFC* 29 20 49
human-macaque difference
Union** 36 26 62 CBC* 3 3 6 PFC* 11 13 24
human-specific ***
Union** 13 16 29 * Union of age-related metabolic changes in three species (FDR<10% in each species), analysis based on
the full set of individuals
** Union of the two brain regions
*** Based on the full-set of individuals
Table S7. Human metabolites affected by diet or exercise.
Factor Tissue Organism Metabolite Detected Category Reference
diet urine human creatinine yes
diet urine human taurine yes
diet urine human carnitine no
diet urine human trimethylamine-N-oxide no
diet urine human methylhistidine no
diet urine human acetylcarnitine no
diet urine human glutamine yes
increased in high-meat
diet urine human p-hydroxyphenylacetate no
diet urine human N-acetylglutamate no increased in vegetarian
diet urine human N,N,N trimethyllysine no increased in low-meat
(2)
exercise urine human 2-hydroxyisovalerate no
exercise urine human 2-oxoisocaproate no
exercise urine human 3-hydroxyisobutyrate no
exercise urine human 3-methyl-2-oxovalerate no
exercise urine human 2-oxoisovalerate no
exercise urine human 2-hydroxybutyrate no
exercise urine human lactate no
exercise urine human alanine no
exercise urine human pyruvate (pyruvic acid) yes
exercise urine human 2-oxovalerate no
exercise urine human inosine no
exercise urine human fumarate (fumaric acid) yes
exercise urine human hypoxanthine no
increased in exercise
exercise urine human citrate^ no
exercise urine human trimethylamine N-oxide no
exercise urine human taurine yes
exercise urine human glycine^ yes
decreased in exercise
(7)
exercise urine human allantoin no
exercise urine human phenylalanine yes
exercise urine human hippurate no
exercise urine human tryptophan yes
exercise urine human formate no
exercise urine human leucine yes
exercise urine human valine yes
exercise urine human isoleucine yes
exercise urine human 3-hydroxybutyrate (3-hydroxybutyric acid) yes
exercise urine human 2-hydroxyisobutyrate no
exercise urine human acetate no
exercise urine human acetoacetate no
exercise urine human succinate (succinic acid)* yes
exercise urine human dimethylamine no
exercise urine human creatinine* yes
exercise urine human cis-aconitate no
exercise urine human malonate no
exercise urine human carnitine* no
exercise urine human N-methylnicotinamide no
exercise urine human glucuronate no
exercise urine human allantoate no
exercise urine human trans-aconitate no
exercise urine human tyrosine yes
exercise urine human histidine yes
exercise urine human 1-methylhistidine no
exercise urine human 3-methylhistidine no
no significant change in exercise
exercise muscle human glutamate yes
exercise muscle human phosphocreatine no
exercise muscle human glycogen no
exercise muscle human carnitine no
decreased in exercise
(4)
exercise muscle human alanine no
exercise muscle human creatine no
exercise muscle human lactate no
exercise muscle human pyruvate (pyruvic acid) yes
exercise muscle human malate no
exercise muscle human fumarate (fumaric acid) yes
exercise muscle human citrate^ no
exercise muscle human isocitrate no
exercise muscle human acetylcarnitine no
increased in exercise
exercise muscle human ATP no
exercise muscle human free CoA no
no significant change in exercise
exercise muscle human creatine no
exercise muscle human ATP^ no increased in exercise
exercise muscle human phosphocreatine* no no significant change in exercise
(3)
diet plasma mouse Histamine no
diet plasma mouse α-Hydroxyisocaproate no
diet plasma mouse Linoleate no
diet plasma mouse Arachidonate no
diet plasma mouse adenosine 5'-Monophosphate no
diet plasma mouse Nicotinamide yes
diet plasma mouse α-Tocopherol no
diet plasma mouse EDTA no
increased in
high-fat food
diet plasma mouse Kynurenate no
diet plasma mouse Glutamine yes
diet plasma mouse Glutamate yes
diet plasma mouse Ornithine yes
diet plasma mouse Arginine no
diet plasma mouse 4-Guanidinobutanoate no
diet plasma mouse Asparagine no
diet plasma mouse 4-Methyl-2-oxopentanoate no
decreased in
high-fat food
(1)
diet plasma mouse Glycine yes
diet plasma mouse 1,5-Anhydroglucitol no
diet plasma mouse malate no
diet plasma mouse Citrate no
diet plasma mouse fumarate (fumaric acid) yes
diet plasma mouse Myo-inositol no
diet plasma mouse Myristate no
diet plasma mouse Acetyl-carnitine no
diet plasma mouse Hexanoyl-carnitine no
diet plasma mouse Carnitine no
diet plasma mouse Glycerol no
diet plasma mouse γ-Glutamyl-tyrosine no
diet plasma mouse Hippurate no
exercise liver rat ATP^ no
exercise liver rat TAN no
decreased in
exercise
exercise liver rat AMP no
exercise liver rat Adenosine^ no
exercise liver rat Hypoxanthine no
exercise liver rat Uric acid no
increased in
exercise
exercise liver rat ADP no
exercise liver rat IMP no
exercise liver rat Inosine* no
no change in
exercise
(6)
exercise liver rat Alanine no
exercise liver rat Glycine^ yes
exercise liver rat Threonine yes
exercise liver rat Cysteine no
exercise liver rat Glutamine yes
exercise liver rat Ornithine yes
exercise liver rat Succinate (succinic acid) yes
exercise liver rat fumarate (fumaric acid) yes
exercise liver rat Malate no
increased in
exercise
exercise liver rat Dihydroxybutyric acidaba no
exercise liver rat Oleic acid no
exercise liver rat Linoleic acid no
exercise liver rat Stearic acid no
exercise liver rat Ribofuranose no
exercise liver rat α-Ribopyranose no
exercise liver rat Gluconic acid no
exercise liver rat Hypoxanthine no
exercise liver rat Xanthine no
exercise liver rat Phosphoric acid no
exercise liver rat Glycerol phosphate no
exercise liver rat Glucose phosphate no
exercise liver rat Urea no
exercise liver rat β-Aminoisobutyric acid no
exercise liver rat Aminomalonic acid no
exercise liver rat Creatinine yes
exercise liver rat Pentanoic acid no
exercise liver rat Ascorbic acid yes
exercise liver rat Stearic acid glycerol no
exercise liver rat α-Glucose no
exercise liver rat Glucitol no
exercise liver rat β-Glucose no
exercise liver rat Saccharide 2 no
exercise liver rat Saccharide 3 no
exercise liver rat Saccharide 4 no
exercise liver rat Uridine yes
exercise liver rat Adenosine^ no
exercise liver rat Diglycerol phosphate no
decreased in
exercise
exercise liver rat 2-Hydroxybutyric acid no
exercise liver rat Lactate no
exercise liver rat Arachidonic acidbca no
changed in
(5)
exercise liver rat β-Ribopyranose no
exercise liver rat Saccharide 1 no
exercise liver rat Uric acid no
exercise#
exercise liver rat Valine yes
exercise liver rat Leucine yes
exercise liver rat Isoleucine yes
exercise liver rat Serine yes
exercise liver rat Methionine yes
exercise liver rat Aspartic acid yes
exercise liver rat Tyrosine yes
exercise liver rat Palmitic acid glycerol no
exercise liver rat Niacinamide no
exercise liver rat Inositol no
exercise liver rat Cholesterol yes
no change in
exercise
* There are 6 metabolites (succinic acid, creatinine, carnitine, ATP, phosphocreatine, and inosine) that identified as changed in exercise in some studies but not changed in other studies are considered as changed in exercise in this study.
# There are 6 metabolites (2-Hydroxybutyric acid, Lactate, Arachidonic acidbca, β-Ribopyranose, Saccharide 1, Uric acid) showing different response in exhaustive exercise and endurance training compared with controls are considered as changed in exercise in this study.
^Thereare4metabolites (citrate, ATP, adenosine, glycine) showing opposite responses to exercise among different studies will also considered as changed in corresponding factor in this study.
Table S8. Overlap between the species-specific metabolites and the diet/exercise affected metabolites.
Species specific Brain
region Species Number * Affected by diet^ Affected by exercise Not affected by
exercise
human 24 (11) glutamate# glutamate# histidine PFC
macaque
20 (15)
ornithine; glycine; fumaric acid; nicotinamide
fumaric acid; glycine; phenylalanine;
threonine; ornithine; ascorbic acid
leucine; valine; isoleucine; methionine;
aspartic acid
human 6 (3) taurine taurine; ascorbic acid -
CBC macaque 33 (20) glutamine; ornithine; glycine; phenylalanine;
leucine; valine; isoleucine;
ornithine
tryptophan; threonine; glutamine;
ornithine
tyrosine; histidine; serine;
methionine; aspartic acid
* Numbers in parentheses were the amount of metabolites with known annotation;
^ Metabolites in italic were those detected as affected by diet, exercise, or not effected by exercisebased on the data from rat or mouse.
# Glutamate has been shown to decreased in high-fat diet in mouse plasma (1) and decrease in concentration with increased exercise (4).
Table S9. Types of metabolic differences among species.
Differential metabolites Potential cause Region Type Number Life-span Mean Pattern
human-specific 24 0 17 7
macaque-specific 20 1 11 8 rest 5 4 0 1
PFC
total 49 5 (10%) 28 (57%) 16 (32%) human-specific 6 1 2 3
macaque-specific 33 2 26 5 rest 4 2 2 0
CBC total 43 5 (12%) 30 (70%) 8 (19%)
References
1. Li LO, et al. (2010) Early hepatic insulin resistance in mice: a metabolomics analysis. Mol Endocrinol 24:657-666.
2. Stella C, et al. (2006) Susceptibility of human metabolic phenotypes to dietary modulation. J Proteome Res 5:2780-2788.
3. Burgomaster KA, Heigenhauser GJ,Gibala MJ (2006) Effect of short-term sprint interval training on human skeletal muscle carbohydrate metabolism during exercise and time-trial performance. J Appl Physiol 100:2041-2047.
4. Howarth KR, LeBlanc PJ, Heigenhauser GJ,Gibala MJ (2004) Effect of endurance training on muscle TCA cycle metabolism during exercise in humans. J Appl Physiol 97:579-584.
5. Huang CC, Lin WT, Hsu FL, Tsai PW,Hou CC (2010) Metabolomics investigation of exercise-modulated changes in metabolism in rat liver after exhaustive and endurance exercises. Eur J Appl Physiol 108:557-566.
6. Mikami T,Kitagawa J (2006) Intense exercise induces the degradation of adenine nucleotide and purine nucleotide synthesis via de novo pathway in the rat liver. Eur J Appl Physiol 96:543-550.
7. Pechlivanis A, et al. (2010) (1)H NMR-Based Metabonomic Investigation of the Effect of Two Different Exercise Sessions on the Metabolic Fingerprint of Human Urine. J Proteome Res DOI: 10.1021/pr100684t.
8. Antinucci F (1989) Systematic comparison of early sensorimotor development, in Cognitive structure and development in nonhuman primates, ed Antinucci F (Erlbaum, Hillsdale (New Jersey)) pp 67–85.
9. Dienske H (1986) A comparative approach to the question of why human infants develop so slowly, in Primate ontogeny, cognition, and social behaviour, eds Else JG and Lee PC (Cambridge Univ. Press, New York) pp 147-154.
10. Ferrari PF, et al. (2006) Neonatal imitation in rhesus macaques. PLoS Biol 4:e302. 11. Watts E (1990) Evolutionary trends in primate growth and development, in
Primate life history and evolution, ed DeRousseau CJ (Wiley-Liss, New York) pp 89–104.
12. Ben Shaul DM (1963) The composition of the milk of wild animals, in The International Zoo Year Book, eds Jarvis C and Morris D (Hutchinson and Company Ltd., London,) pp 333-342.
13. Rosenberg K,Trevathan W (2002) Birth, obstetrics and human evolution. Bjog 109:1199-1206.
14. Somel M, et al. (2008) Human and chimpanzee gene expression differences replicated in mice fed different diets. PLoS One 3:e1504.
15. Lisec J, Schauer N, Kopka J, Willmitzer L,Fernie AR (2006) Gas chromatography mass spectrometry-based metabolite profiling in plants. Nat Protoc 1:387-396.
16. Fu X, et al. (2009) Estimating accuracy of RNA-Seq and microarrays with proteomics. BMC Genomics 10:161.
17. Somel M, et al. (2010) MicroRNA, mRNA, and protein expression link development and aging in human and macaque brain. Genome Res 20:1207-1218.
18. Boedigheimer MJ, et al. (2008) Sources of variation in baseline gene expression levels from toxicogenomics study control animals across multiple laboratories. BMC Genomics 9:285.
19. Li J , Bushel PR, Chu T-M,Wolfinger RD (2009) Principal Variance Components Analysis: Estimating Batch Effects in Microarray Gene Expression Data, in Batch Effects and Noise in Microarray Experiments: Sources and Solutions, ed Scherer A (John Wiley & Sons, Ltd, Chichester) doi: 10.1002/9780470685983.ch12.
20. Somel M, et al. (2009) Transcriptional neoteny in the human brain. Proc Natl Acad Sci U S A 106:5743-5748.
21. Faraway JJ (2002) Practical Regression and ANOVA Using R. Available at: http://cran.r-project.org/doc/contrib/Faraway-PRA.pdf.
22. de Magalhaes JP,Costa J (2009) A database of vertebrate longevity records and their relation to other life-history traits. J Evol Biol 22:1770-1774.
23. Storey JD,Tibshirani R (2003) Statistical significance for genomewide studies. Proc Natl Acad Sci U S A 100:9440-9445.