Public Notices

A Polygenic Risk Score of glutamatergic SNPs associated with schizophrenia predicts attentional behavior and related brain activity in healthy humans

Description
Multiple genetic variations impact on risk for schizophrenia. Recent analyses by the Psychiatric Genomics Consortium (PGC2) identified 128 SNPs genome-wide associated with the disorder. Furthermore, attention and working memory deficits are core
Categories
Published
of 12
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
Share
Transcript
  A Polygenic Risk Score of glutamatergic SNPsassociated with schizophrenia predictsattentional behavior and related brainactivity in healthy humans Antonio Rampino a , Paolo Taurisano a , Giuseppe Fanelli a ,Mariateresa Attrotto a,b , Silvia Torretta a ,Linda Antonella Antonucci a , Grazia Miccolis a , Giulio Pergola a ,Gianluca Ursini c , Giancarlo Maddalena a,b , Raffaella R omano a ,Rita Masellis b , Pasquale Di Carlo a , Patrizia Pignataro a ,Giuseppe Blasi a,b , Alessandro Bertolino a,b, n a Department of Basic Medical Science, Neuroscience and Sense Organs   –  University of Bari  “   Aldo Moro ”  ,Piazza Giulio Cesare 11, 70124 Bari, Italy  b Psychiatry Unit  –  Bari University Hospital, Piazza Giulio Cesare 11, 70124 Bari, Italy  c Lieber Institute for Brain Development, Johns Hopkins University Medical Campus, 21205 Baltimore, MD,USA Received 6 December 2016; received in revised form 13 April 2017; accepted 10 June 2017 KEYWORDS Glutamate;Working memory;Attention;Schizophrenia;Prefrontal;Cortex Abstract Multiple genetic variations impact on risk for schizophrenia. Recent analyses by the PsychiatricGenomics Consortium (PGC2) identi 󿬁 ed 128 SNPs genome-wide associated with the disorder.Furthermore, attention and working memory de 󿬁 cits are core features of schizophrenia, areheritable and have been associated with variation in glutamatergic neurotransmission. Based onthis evidence, in a sample of healthy volunteers, we used SNPs associated with schizophrenia inPGC2 to construct a Polygenic-Risk-Score (PRS) re 󿬂 ecting the cumulative risk for schizophrenia,along with a Polygenic-Risk-Score including only SNPs related to genes implicated inglutamatergic signaling (Glu-PRS). We performed Factor Analysis for dimension reduction of indices of cognitive performance. Furthermore, both PRS and Glu-PRS were used as predictorsof cognitive functioning in the domains of Attention, Speed of Processing and Working Memory.The association of the Glu-PRS on brain activity during the Variable Attention Control (VAC) taskwas also explored. Finally, in a second independent sample of healthy volunteers we sought to www.elsevier.com/locate/euroneuro http://dx.doi.org/10.1016/j.euroneuro.2017.06.0050924-977X/ &  2017 Elsevier B.V. and ECNP. All rights reserved. n Corresponding author at: Department of Basic Medical Sciences, Neuroscience and Sensory Organs, University of Bari,  ‘ Aldo Moro ’ , PiazzaGiulio Cesare, 11, 70120 Bari, Italy. E-mail address:  alessandro.bertolino@uniba.it (A. Bertolino).European Neuropsychopharmacology (2017)  27 , 928 – 939  con 󿬁 rm the association between the Glu-PRS and both performance in the domain of Attentionand brain activity during the VAC.We found that performance in Speed of Processing and Working Memory was not associatedwith any of the Polygenic-Risk-Scores. The Glu-PRS, but not the PRS was associated withAttention and brain activity during the VAC. The speci 󿬁 c effects of Glu-PRS on Attention andbrain activity during the VAC were also con 󿬁 rmed in the replication sample.Our results suggest a pathway speci 󿬁 city in the relationship between genetic risk forschizophrenia, the associated cognitive dysfunction and related brain processing. &  2017 Elsevier B.V. and ECNP. All rights reserved. 1. Introduction Schizophrenia is a brain disease with an important geneticcomponent accounting for 65 – 80% of heritability (Lichtensteinet al., 2009; Sullivan et al., 2012, 2003). There is evidence that genetic risk for schizophrenia results by a collective burden of many low-penetrance gene mutations together with a smallnumber of highly penetrant mutations (Purcell et al., 2009).Consistently, Genome Wide Association Studies (GWAS) havecon 󿬁 rmed that a considerably high number of Single NucleotidePolymorphisms (SNPs) distributed across the entire genome isassociated with the disorder. For example, recent analyses bythe Schizophrenia Working Group of the Psychiatric GenomicsConsortium on almost 37,000 cases and 113,000 controls(PGC2) have identi 󿬁 ed 128 SNPs, occurring in 108 loci thatare genome-wide associated with the disorder (SchizophreniaWorking Group of the Psychiatric Genomics, 2014). In order totest the polygenic burden of this disease, previous studies havecaptured the en masse additive variation across nominallyassociated loci into quantitative scores (Polygenic Risk Scores-PRSs), and related the scores to disease state in independentsamples (Cross-Disorder Group of the Psychiatric Genomicset al., 2013, Ripke et al., 2013). These scores are constructed from alleles showing association with schizophrenia in GWASand thus re 󿬂 ect additive genetic risk for the disease andrepresent a powerful tool to understand the genetic under-pinnings of schizophrenia. Nevertheless, these cumulative non-speci 󿬁 c indices of genetic risk do not provide any speci 󿬁 cinsight into the biology associated with these genetic variants.One possible way to overcome such a limitation is to consideralleles associated with known speci 󿬁 c biological pathways anduse them to construct pathway-speci 󿬁 c PRSs. Moreover, it ispossible to hypothesize that PRSs recapitulating variation in aspeci 󿬁 c signaling pathway and associated with a disease aremore strongly associated with distinct speci 󿬁 c endophenotypesof the same disease as compared with the en masse PRS byvirtue of their biological speci 󿬁 city. In line with such ahypothesis, studies have demonstrated that disruption inspeci 󿬁 c neuronal signaling pathways is associated with varia-tion in endophenotypes of schizophrenia (Allen et al., 2009,Cannon, 2005), including cognitive de 󿬁 cits typical of thedisorder. Therefore, it is plausible to hypothesize that PRSsbuilt on genetic variants belonging to speci 󿬁 c pathways canpredict variation in these phenotypes more robustly than riskscores re 󿬂 ecting the overall risk for the disorder.Among cognitive de 󿬁 cits that are endophenotypes of schizophrenia (Bertolino and Blasi, 2009), impairments inattention and WM are well established. Several studies haveidenti 󿬁 ed attention as a cognitive domain that encompassesthe primary de 󿬁 ciencies of schizophrenia (Nuechterleinet al., 2004). Consistently, schizophrenia patients per-form poorer than healthy controls at the Continuous Per-formance Test (CPT) (Cornblatt and Malhotra, 2001), atask that is used to assess attention performance in hum-ans. Consistently, neuroimaging studies have reportedmodi 󿬁 cation in prefrontal cortex processing of attentionstimuli in patients with schizophrenia (Blasi et al., 2010;Delawalla et al., 2008).Similarly, robust de 󿬁 cits in WM have been identi 󿬁 ed inschizophrenia patients relative to healthy controls. In fact,as described in two comprehensive meta-analyses (Lee andPark, 2005; Forbes et al., 2009), de 󿬁 cits in working memoryare present in schizophrenia across a number of measures,speci 󿬁 cally, tasks assessing maintenance and/or manipula-tion of auditory, visual, lexical, or semantic information.Consistently, imaging studies have shown prefrontal activityduring WM to be altered in patients with the disorder(Callicott et al., 1998).A number of studies have also suggested that variation inattention and WM performance is associated with modula-tion in glutamatergic signaling both in animal models andhumans. In fact, studies in rodents have demonstrated thatacute administration of the noncompetitive N-methyl-D-aspartate (NMDA) receptor antagonist ketamine inhibitssustained attention in rats performing a visual signaldetection task (Hillhouse et al., 2015) and that the repeatedadministration of ketamine can impair the ability to sustainattention as assessed in the  󿬁 ve-choice serial reaction timetask (5-CSRTT) (Nikiforuk and Popik, 2014). Furthermore,the same ketamine administration protocol induces de 󿬁 citsin mice attention set-shifting abilities as assessed with theattention set-shifting task (ASST), an effect that is alle-viated by the subsequent administration of antipsychoticmedication (Kos et al., 2011). In humans, studies havedemonstrated that NMDA antagonists including ketamineand phencyclidine exacerbate psychosis in patients withschizophrenia (Malhotra et al., 1997). Furthermore, thesemolecules induce in healthy subjects phenomena reminis-cent of positive and negative symptoms (Krystal et al.,1994), as well as cognitive dysfunction reminiscent of thatassociated with the disorder, especially in the domain of attention (Lieberman et al., 2008, Malhotra et al., 1996). Among others, de 󿬁 cits in strategic shifting of visual atten-tion (Fuchs et al., 2015) and goal-driven biases in orienting929A Polygenic Risk Score of glutamatergic SNPs  spatial attention (Gouzoulis-Mayfrank et al., 2006) havebeen related to treatment with NMDAR antagonists. Finally,fMRI studies have reported ketamine to modulate DLPFCactivity during performance of the Continuous PerformanceTask (Honey et al., 2008).Growing evidence has also suggested a role for glutama-tergic signaling in WM both in animal models and in humans.For example, preclinical studies have indicated that NMDAR signaling is critically involved in the maintenance of informa-tion in WM (Fellous and Sejnowski, 2003, Seamans et al., 2003) while other research lines have reported that inrodents, NMDAR antagonists attenuate PFC activity asso-ciated with WM maintenance (Homayoun et al., 2005;Jackson et al., 2004) and reduce WM performance (Robertset al., 2010).Consistent results have been reported in humans (Krystalet al., 1994; Oye et al., 1992) with fMRI studies suggesting brain activity related to WM to be affected by pharmaco-logical modulation of NMDAR activity (Honey et al., 2004).Moreover, genetic studies have shown that genetic variationassociated with NMDAR expression levels in PFC is corre-lated with WM behavioral performance and related PFCactivity in healthy volunteers (Pergola et al., 2016). Finally,genetic variation in genes of relevance to glutamatergicsignaling has been implicated in modulation of attentionand WM performance in schizophrenia and related prefron-tal cortex processing (Gallinat et al., 2007).Taken together, this evidence suggests that a putativepathophysiological mechanism of the symptoms of schizo-phrenia, including attention and WM de 󿬁 cits, may involvedysregulation in glutamatergic neurotransmission.Based on this previous evidence, we hypothesized thatcumulative genetic risk for schizophrenia deriving from thePGC2 SNPs within genes involved in glutamatergic neuro-transmission is associated with measures of attention andWM processing, including behavioral performance andrelated brain activity. In particular, in two independentsamples of healthy volunteers we used SNPs associated withschizophrenia in PGC2 to construct two different PRSs: the 󿬁 rst, encompassing all SNPs included in PGC2 and thusaccounting for overall genetic variation associated withthe disorder (PRS); the second, including only SNPs relatedto genes implicated in glutamatergic signaling pathways(Glu-PRS). In particular, gene selection for Glu-PRS compu-tation was carried out referring to pathway analysis resultsreported in PGC2 and indicating  GRIN2A ,  GRM3 ,  SRR  and CLCN3  genes as belonging to glutamatergic neurotransmis-sion pathway. Two of these genes, namely  GRIN2A  and GRM3 , respectively code for an N-methyl-D-aspartate(NMDA) receptor subunit (McBain and Mayer, 1994) and amember of metabotropic glutamate receptor subfamily(Tanabe et al., 1992), thus playing a key role in neuronalglutamatergic neurotransmission.  SRR  gene codes for theenzyme Serine Racemase, which operates at the neuronallevel by converting l-serine to d-serine (Wolosker et al.,1999), an endogenous ligand of the NMDA receptor (Matsuiet al., 1995). Finally,  CLCN3  gene codes for a member of thevoltage-gated chloride channel (ClC) family (ClC-3)(Jentsch, 2008) that modulates synaptic strength at gluta-matergic synapses by regulating the amount of neurotrans-mitter stored in a single synaptic vesicle as well as itslikelihood of fusion (Guzman et al., 2014).PRS and Glu-PRS were then employed to test the hypoth-eses that 1) risk related to glutamatergic neurotransmission(re 󿬂 ected by Glu-PRS) was associated with sustained atten-tion, WM and related brain activity; 2) the associationbetween overall genetic risk for schizophrenia as recapitu-lated by the PRS with attention and WM behavioral andbrain imaging phenotypes of interest would be weakerif any.In order to test these hypotheses, in a  󿬁 rst exploratorystep, we adopted a data reduction approach to cognitiveperformance in one of the two samples through a FactorAnalysis (FA) and used Glu-PRS and PRS as predictors of cognitive performance re 󿬂 ected by the three Factors, whichwe named Attention, Working Memory and Speed of Proces-sing, based on factor loadings on cognitive test indices.Since we found Glu-PRS to have a speci 󿬁 c effect onattention and related prefrontal processing while showingno impact on the other two cognitive domains, in a secondcon 󿬁 rmatory step, we used the second independent sampleof healthy volunteers to replicate FA structure and con 󿬁 rmthe effect of Glu-PRS on Attention and prefrontal cortexprocessing during a sustained attention task. 2. Experimental procedures 2.1. Participants and Polygenic Risk Scores calculation 2.1.1. Samples 2.1.1.1. Discovery sample.  We recruited 260 healthy Caucasiansubjects from the region of Apulia (males,  N  = 123; mean age, 28years [SD = 8 years]) within the discovery sample. All subjects werescreened using the Non-Patient Structured Clinical Interview forDSM-IV to ensure they were unaffected by any psychiatric condi-tion. Additional exclusion criteria were represented by a signi 󿬁 canthistory of drug or alcohol abuse, active drug abuse in the previousyear, head trauma with a loss of consciousness, any other signi 󿬁 cant Table 1  Demographic and neuropsychological characteristics of the discovery and replication fMRI samples. Discovery sample mean(SD) Replication sample mean(SD)N  151 51 Age  26(10.7) 25(6.1) Gender Ratio (m:f)  73:78 26:25 Socio-Economic Status  44.53(16.04) 37.69(16.02) Handedness  0.76(0.39) 0.80(0.55) IQ   109.34(9.94) 112.18(12.06) A. Rampino et al.930  medical condition. All subjects underwent a battery of tasksmeasuring behavioral performance associated with different cogni-tive domains (see below). Furthermore, a sub-group of this samplecomprising 151 healthy Caucasian subjects (males,  N  = 73; meanage, 26 years [SD = 10.7]) from the region of Apulia underwent fMRIwhile performing an attention control task (see below) (Table 1). 2.1.2. Replication sample The replication sample included 73 healthy Caucasian subjects fromthe same region (males,  N  = 34; mean age, 26 years [SD = 6.4]).Screening procedures and inclusion criteria were the same as in thediscovery sample. All these individuals were tested with the samebattery of cognitive tasks used for the discovery sample. Further-more, a sub-group of the replication sample including 51 healthysubjects (males,  N  = 26; mean age, 25 years [SD = 6.1]) underwentthe same fMRI protocol as in the discovery sample (Table 1).All participants in the current study provided written informedconsent according to the guidelines of the Declaration of Helsinki.Protocols and procedures were approved by the ethics committee of the University of Bari Aldo Moro. 2.2. SNP selection and genotype determination SNPs reported in PGC2 (Schizophrenia Working Group of thePsychiatric Genomics, 2014) for exceeding genome wide signi 󿬁 cance(  p r 5  10 -8 ) were selected. These were 128 LD independent SNPs(see Schizophrenia Working Group of the Psychiatric Genomics (2014)for LD independence criteria). All individuals were genotyped usingan Illumina HumanOmni2.5-8 v1 BeadChip platform. More in thedetail, approximately 200 ng DNA was used for genotyping analysis.DNAwas concentrated at 50 ng/ml (diluted in 10 mM Tris/1 mM EDTA)with a Nanodrop Spectrophotometer (ND-1000). Each sample waswhole-genome ampli 󿬁 ed, fragmented, precipitated and re-suspended in appropriate concentrations of hybridization buffer.Denatured samples were hybridized on the prepared IlluminaHumanOmni2.5-8 v1 BeadChip. After hybridization, the BeadChipoligonucleotides were extended by a single labeled base, which wasdetected by  󿬂 uorescence imaging with an Illumina Bead ArrayReader. Normalized bead intensity data obtained for each samplewere loaded into the Illumina GenomeStudio (Illumina, v.2010.1)with cluster position  󿬁 les provided by Illumina, and  󿬂 uorescenceintensities were converted into SNP genotypes. After genotypes werecalled and the pedigree  󿬁 le was assembled, we removed SNPsshowing minor allele frequency  o 1%, genotype missing rate 4 5%,or deviation from Hardy-Weinberg equilibrium (  p o 0.0001). Indivi-dual data were also removed if their overall genotyping rate wasbelow 97%. Sample duplications and cryptic relatedness were ruledout through identity-by-state (IBS) analysis of genotype data.The genotyping platform we used included 97 out of 128 SNPsreported in PGC2.Genotype condition at missing genotype calls was imputed by thegenotype imputation and haplotype phasing program IMPUTE 2(Howie et al., 2009). 2.3. PRS calculation Genotype data were used to calculate a Polygenic Risk Score for eachindividual in both the discovery and the replication sample followingthe method reported by Purcell and coll. (Purcell et al., 2009). Inparticular,  󿬁 rst we assigned an  Allelic Load Factor (ALF)  to eachsubject, i.e. a score re 󿬂 ecting the number of risk alleles carried by thesubject at each SNP selected.  ALFs   were assigned according to thefollowing criteria: 1 to the homozygotes for the protective allele, 2 tothe heterozygotes, 3 to the homozygotes for the risk allele. Eachassigned  ALF   was then multiplied by the natural logarithm of the oddsratio (OR) related to each SNP (ORs were provided in SupplementaryTable 2 of (Schizophrenia Working Group of the Psychiatric Genomics,2014).For OR  o 1 we used the natural logarithm of the reciprocal OR (1/OR). The resulting values were summed for each individual, so thateach individual had a whole genome PRS for further analyses.The calculation procedure is summarized in the followingformula: X ½ Allelic Load Factor    log e OR    if OR  4 1 X  Allelic Load Factor    log e 1OR     if OR  o 1 2.4. Glu-PRS calculation Determination of Glu-PRS was carried out following the sameprocedures reported for PRS calculation. In particular, SNPs asso-ciated with schizophrenia in PGC2 and belonging to genes that PGC2attributed to  Glutamatergic Neurotransmission Pathway   wereselected (Schizophrenia Working Group of the PsychiatricGenomics, 2014). The SNP list included rs9922678 belonging to GRIN2A  gene (location: 16p13.2; encoded protein: NMDA ionotropicglutamate receptor subunit GluN2A), rs4523957 belonging to  SRR gene (location: 17p13.3; encoded protein: Serine Racemase),rs10520163, belonging to  CLCN3  (location: 4q33; encoded protein:H + /Cl  exchange transporter 3) and rs12704290, belonging to GRM3  gene (location: 7q21.12; encoded protein: metabotropicGlutamate Receptor 3/mGluR3). 3. Cognitive performance and FactorAnalyses The same procedures for  Factor Analysis  were adopted inboth the discovery and the replication sample. All subjectswere assessed with a battery of cognitive tests including i)the N-Back Working Memory paradigm, ii) the ContinuousPerformance Test, assessing sustained attention, iii) TrailMaking A and B for the assessment of visual memory, and iv)the Controlled Oral Word Association Test (COWAT) whichmeasures Semantic and Phonological Fluency. In order toreduce type I error in subsequent analyses and based onprevious evidence that performance at multiple cognitivetests can be correlated hiding implicit architecture in thecognitive pro 󿬁 le of a sample (Genderson et al., 2007), weperformed a data dimension reduction of cognitive informa-tion with a Factor Analysis (FA). FA was performed on aCorrelation Matrix including all pairwise Pearson's correla-tions between scores at cognitive tests used, followed byVarimax rotation and retaining factors with an eigenvalue Z 1 (Guttman, 1954; Kaiser, 1970). Factors were then named according to cognitive function probed by the testsloading on each Factor. Contribution of each cognitive testto a single Factor was considered signi 󿬁 cant (i.e., relevantfor Factor interpretation and naming) if correspondingfactor loading was  4 0.7 (default cut-off value set byStatistica 10 software).Factor scores for each Factor were assigned to eachindividual and were used as performance measures insubsequent analyses. Pearson's test was carried out in orderto assess degree of correlation between factor loadings inthe two (discovery and replication) samples (Table 2).931A Polygenic Risk Score of glutamatergic SNPs  4. Association of Glu-PRS and PRS withcognition 4.1. Discovery sample In the discovery sample, correlation between Glu-PRS withcognitive performance was assessed with a multiple regres-sion having Glu-PRS as predictor and factor scores at each Factor  as the dependent variables. Based on previousevidence supporting the association of genetic variation inglutamate signaling with I.Q. (Pergola et al., 2016) andbecause of the established relationship of age with cogni-tive performance (Tucker-Drob, 2011), age and scores at theWAIS test for I.Q. assessment were entered in the analysesas nuisance variables. Bonferroni correction was adopted toreduce type I errors (number of tests = number of dependentvariables = 3). Correlation between global PRS and cognitiveperformance was assessed with a multiple regression havingPRS as predictor and factor scores at each Factor as thedependent variables. Age and scores at the WAIS test for I.Q. assessment were entered as nuisance variables. 4.1.1. Replication sample In the replication sample, we explored the correlationbetween Glu-PRS with performance in the cognitive domainof Attention. Such a correlation was assessed with a linearregression having Glu-PRS as predictor and factor scores atthe Attention Factor as the dependent variable. Age andWAIS I.Q. were entered as nuisance variables in the analysis.Because of our strong  a priori  hypothesis on the effect of Glu-PRS on Attention, a one-tailed signi 󿬁 cance thresholdwas adopted. 4.1.2. Cross-sample prediction With the aim to support replicability and factor structurevalidity, we investigated cross-sample prediction betweenthe two samples. In particular, we projected the factorloadings of the Attention Factor detected in the discoverysample (Table 2) onto the replication data to obtainprojected Attention Factor scores.  Vice versa , we alsoprojected the factor loadings of the Attention Factordetected in the replication sample onto the discovery data.Finally, projected Attention Factor scores within eachsample were used as the dependent variable in a newregression model including Glu-PRS as predictor along withage and WAIS I.Q. as nuisance variables (same model as inthe discovery and the replication samples). Based on our  a priori  hypothesis regarding the association between Glu-PRSand Attention performance, one-tailed p-values wereapplied. 5. fMRI 5.1. Neuropsychological task Given the behavioral effect that we found on the attentiondomain, we focused on investigation of fMRI activity duringattention processing. With this aim, the Variable Atte-ntion Control (VAC) task was used to elicit different andincreasing levels of attention control processing. The task iscomposed of stimuli made of arrows with 3 different sizeseither pointing to the right or to the left. 42 small arrowsare embedded in 6 medium arrows, which are in turnembedded in a single large arrow. Stimuli are presentedwith a cue word above each stimulus (big, medium, orsmall). Subjects are instructed to focus on the cue word and Table 2  Neuropsychological variables included in our Factor Analysis and related factor loadings detected separately in thediscovery and in the replication sample. Variable Attention Speed of processing Working memory Discovery sampleReplicationsampleDiscovery sampleReplicationsampleDiscovery sampleReplicationsampleOne Back RT  a  0.068   0.100   0.128 0.173   0.849* 0.858* One Back accuracy   0.172   0.135 0.032 0.098 0.737*   0.716* Two Back RT  a  0.070   0.158   0.091 0.255   0.840* 0.700* Two Back accuracy   0.099 0.071 0.133   0.140 0.820*   0.888* Phonologic Fluency   0.269   0.049 0.141   0.590   0.105 0.125 Semantic Fluency   0.342 0.027 0.137   0.440   0.026   0.118 Trail Making Test A   0.095 0.144   0.500 0.361   0.151 0.304 Trail Making Test B   0.111 0.048   0.974* 0.876*   0.105 0.220 Trail Making Test B-A   0.091   0.006   0.917* 0.846*   0.066 0.128 FS HIT (CPT)  0.714* 0.604 0.179 0.156 0.174   0.088 HIT RATS (CPT)  0.793* 0.889* 0.131   0.010 0.317   0.112 DP S (CPT)  0.811* 0.861*   0.124   0.030 0.217 0.013 DPC S (CPT)  0.819* 0.874*   0.058   0.017 0.127 0.104 r-squared  b 0.803 0.764 0.932 Loadings marked with (*) are 4 0.7. a RT: Reaction times. b r-squared of the Pearson Correlation between factor loadings of the discovery and replication sample. A. Rampino et al.932
Search
Similar documents
View more...
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks