back to africa: tracing dyslexia genes in east...

23
Back to Africa: Tracing dyslexia genes in east Africa ELENA L. GRIGORENKO 1,2,3,10 , ADAM NAPLES 2 , JOSEPH CHANG 4 , CHRISTINA ROMANO 1 , DAMARIS NGOROSHO 5 , SELEMANI KUNGULILO 6 , MATTHEW JUKES 7,8 , DONALD BUNDY 9 1 Child Study Center, Yale University Medical School, New Haven, CT, USA; 2 Yale University PACE Center, New Haven, CT, USA; 3 Department of Psychology, Moscow State University, Moscow, Russia; 4 Department of Statistics, Yale University, New Haven, CT, USA; 5 The Agency for the Development of Educational Management, Bagamoyo, Tanzania; 6 Muhimbili University College of Health Sciences, Dar-es-Salaam, Tanzania; 7 Department of Infectious Disease Epidemiology, Imperial College, Partnership for Child Development, London, UK; 8 Institute of Education, School of Lifelong Education and International Development, University of London, London, UK; 9 School Health and Nutrition, World Bank, Washington, DC, USA; 10 Child Study Center, Yale University Medical School, 203 South Frontage Road, New Haven, CT, 06510, USA Abstract. A sample of Swahili-speaking probands with reading difficulties was identified from a large representative sample of 1,500 school children in the rural areas of Tanzania. Families of these probands (n = 88) were invited to participate in the study. The proband and his/her siblings received a battery of reading-related tasks and performance on these tasks was recorded and treated as phenotypic data. Molecular- genetic analyses were carried out with 47 highly polymorphic markers spanning three previously identified regions of interest harboring susceptibility loci for reading difficul- ties: 2p, 6p, and 15q (DYX1–DYX3). The analyses revealed the involvement of these regions in the development of reading difficulties in Swahili. The linkage signals are especially pronounced for time (compared with error) indicators of reading difficulties. These findings are easily interpretable because in transparent languages such as Swahili deficits in reading are more related to the rate/speed of reading and reading-related pro- cesses than to the number of errors made. In short, the study incrementally advances the field by adding an understudied language and an understudied population to the variety of languages and populations in the field of molecular-genetic studies of reading difficulties. Key words: Candidate genes, Dyslexia, Linkage analyses, Regional mapping, Swahili Introduction The realization that complex human disorders arise from a background of distinct yet related processes interacting with each other has driven the last decade of genetic research into the etiology of such disorders. A number of authors have discussed the need to ‘‘dissect’’ complex Reading and Writing (2007) 20:27–49 Ó Springer 2006 DOI 10.1007/s11145-006-9017-y

Upload: others

Post on 09-Aug-2020

0 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Back to Africa: Tracing dyslexia genes in east Africajtc5/papers/AfricaDyslexia_Grigorenko_07.pdf · For example, Grigorenko and colleagues (Grigorenko et al., 1997) used continuous

Back to Africa: Tracing dyslexia genes in east Africa

ELENA L. GRIGORENKO1,2,3,10, ADAM NAPLES2, JOSEPHCHANG4, CHRISTINA ROMANO1, DAMARIS NGOROSHO5,SELEMANI KUNGULILO6, MATTHEW JUKES7,8, DONALDBUNDY9

1Child Study Center, Yale University Medical School, New Haven, CT, USA; 2YaleUniversity PACE Center, New Haven, CT, USA; 3Department of Psychology, Moscow

State University, Moscow, Russia; 4Department of Statistics, Yale University, NewHaven, CT, USA; 5The Agency for the Development of Educational Management,Bagamoyo, Tanzania; 6Muhimbili University College of Health Sciences, Dar-es-Salaam,Tanzania; 7Department of Infectious Disease Epidemiology, Imperial College, Partnership

for Child Development, London, UK; 8Institute of Education, School of LifelongEducation and International Development, University of London, London, UK; 9SchoolHealth and Nutrition, World Bank, Washington, DC, USA; 10Child Study Center, Yale

University Medical School, 203 South Frontage Road, New Haven, CT, 06510, USA

Abstract. A sample of Swahili-speaking probands with reading difficulties was identifiedfrom a large representative sample of �1,500 school children in the rural areas of

Tanzania. Families of these probands (n = 88) were invited to participate in the study.The proband and his/her siblings received a battery of reading-related tasks andperformance on these tasks was recorded and treated as phenotypic data. Molecular-genetic analyses were carried out with 47 highly polymorphic markers spanning three

previously identified regions of interest harboring susceptibility loci for reading difficul-ties: 2p, 6p, and 15q (DYX1–DYX3). The analyses revealed the involvement of theseregions in the development of reading difficulties in Swahili. The linkage signals are

especially pronounced for time (compared with error) indicators of reading difficulties.These findings are easily interpretable because in transparent languages such as Swahilideficits in reading are more related to the rate/speed of reading and reading-related pro-

cesses than to the number of errors made. In short, the study incrementally advances thefield by adding an understudied language and an understudied population to the variety oflanguages and populations in the field of molecular-genetic studies of reading difficulties.

Key words: Candidate genes, Dyslexia, Linkage analyses, Regional mapping, Swahili

Introduction

The realization that complex human disorders arise from a background ofdistinct yet related processes interacting with each other has driven thelast decade of genetic research into the etiology of such disorders. Anumber of authors have discussed the need to ‘‘dissect’’ complex

Reading and Writing (2007) 20:27–49 � Springer 2006DOI 10.1007/s11145-006-9017-y

Page 2: Back to Africa: Tracing dyslexia genes in east Africajtc5/papers/AfricaDyslexia_Grigorenko_07.pdf · For example, Grigorenko and colleagues (Grigorenko et al., 1997) used continuous

phenotypes into their correlating components, both for what are tradi-tionally viewed as ‘‘disorders of the body’’ (e.g., diabetes, Chiu et al.,2005; obesity, Chen et al., 2005; hypertension, Kammerer et al., 2004) and‘‘disorders of the brain’’ (e.g., autism, Alarcon, Yonan, Gilliam, Cantor,& Geschwind, 2005; depression, Nash et al., 2004). However, althoughthe necessity and usefulness of such dissection is largely accepted at thelevel of phenotypic analyses when the disorder is characterized, there is noconsensus on what to do about these correlated components in the con-text of linkage and association analyses.

For example, in the field of developmental dyslexia (DD),1 almost allresearchers base their definition of ‘‘affected’’ status on comprehensiveevaluation batteries that typically include indices of cognitive processessuch as (1) phonological processing; (2) orthographical processing; (3)speed of naming; (4) single-word recognition; (5) working memory; and(6) general cognitive functioning. The specifics of what is assessed andhow might vary from group to group, but, in general, there is consensusthat these processes must be assessed to generate an accurate represen-tation of the componential deficit characteristic(s) of DD.

A noticeable diversity of approaches is present at the stage of geneticanalysis of these componential data. Our review of the literature hasidentified the following approaches:

(1) The componential data are processed and summarized in such away that a set of complex, threshold-based categorical phenotypes isgenerated at the level of behavior; this set is then mapped onto a categoryat the level of the disorder itself, resulting in the affected–unaffecteddiagnosis (Fagerheim, Raeymaekers, Tønnessen, Pedersen, Tranebjærg,& Lubs, et al., 1999; Kaminen et al., 2003). Thus, these researchers as-sume that componential processes contribute to a complex but ultimatelyunivariate phenotype. Subsequently, this categorical phenotype is used inlinkage and association studies. A continuous analogy of this approach isexemplified in the Colorado group�s use of the discriminant function(Cardon et al., 1994). These researchers recruited continuous indices ofmultiple processes in establishing the discriminant function, but then usedonly a single discriminant score to establish the affected–unaffectedphenotype, both in continuous and categorical representations.

(2) The componential data are processed with the assumption thatcertain clusters of indices describe specific higher-order constructs. Thereare diverse approaches that fit this category. For example, Grigorenkoand colleagues (Grigorenko et al., 1997) used continuous indices toderive specific deficit-based categorical scores; in this interpretation, DDis a multifaceted disorder and each facet can be represented throughspecific deficits individually or collectively. A variant of this approach,

28 ELENA L. GRIGORENKO ET AL.

Page 3: Back to Africa: Tracing dyslexia genes in east Africajtc5/papers/AfricaDyslexia_Grigorenko_07.pdf · For example, Grigorenko and colleagues (Grigorenko et al., 1997) used continuous

conceptually, is to measure multiple continuous phenotypes and thenattempt to reduce the dimensionality of the DD phenotype by applyingfactor-analytic techniques and deriving higher-order variables.

(3) The componential data are preserved so that all processes measuredare represented in phenotypic analyses. Correspondingly, the researchershere include as many phenotypes in their linkage and association analysesas there were measured phenotypes (Deffenbacher et al., 2004; Fisheret al., 2002; Gayan et al., 1999). Although most of the analyses to datehave been univariate, there has been an attempt to use a multivariateapproach (Marlow et al., 2003).

There is no consensus at this point on what particular phenotypingapproach will produce the most precise and parsimonious findings. Col-lectively in the field of DD, all these phenotyping approaches have gener-atedanumber of interesting and important results, convincingly supportingthe hypotheses of the role of genes in the manifestation of DD, althoughspecifics of this involvementmight vary for different phenotyping strategies.The origin of these findings can be traced to the 1983 paper by Smith andcolleagues that triggered the establishment of the molecular-genetic field ofDD (Smith,Kimberling, Pennington,&Lubs, 1983), whereDDwas treatedas a single categorical phenotype. In that paper, the field obtained its firstcandidate region for DD, a region somewhere around the centromere onchromosome 15. The precision of genetic mapping was so low at that timethat the boundaries of the region were huge, by genetic standards, andsubsequent attempts to work with chromosome 15 resulted in both repli-cations and nonreplications simply because researchers looked at varioussubregions of this initially flagged piece of the chromosome.

Since that cornerstone paper, the field has expanded immensely andthe current state of affairs is quite remarkable: The field has nine candi-date regions to entertain (Grigorenko, 2005a) and four candidate genes toconsider. These regions are recognized as DD candidate regions; they areabbreviated as DYX1–9 and refer to the regions on chromosomes 15q,6p, 2p, 6q, 3cen, 18p, 11p, 1p, and Xq, respectively. However, this isclearly not the end of the story – new regions of interest are reported on aregular basis, the latest being 2q (Raskind et al., 2005).

A number of different research groups work on these loci in an attemptto identify plausible candidate genes. Four successful attempts have beenannounced in the literature: one for the 15q region – the candidate geneknown as DYX1C1 (Taipale et al., 2003); two for the 6p region – thecandidate genes known as KIAA0319 (Cope et al., 2005; Deffenbacheret al., 2004; Francks et al., 2004) and DCDC2 (Meng et al., 2005;Schumacher et al., 2006), and one for the 3cen region, ROBO1 (Hannula-Jouppi et al., 2005). Yet, after the first presentation of the DYX1C1 gene,

29TRACING DYSLEXIA GENES IN EAST AFRICA

Page 4: Back to Africa: Tracing dyslexia genes in east Africajtc5/papers/AfricaDyslexia_Grigorenko_07.pdf · For example, Grigorenko and colleagues (Grigorenko et al., 1997) used continuous

somewhat controversial evidence followed that challenged the associationbetween DYX1C1 and dyslexia (Bellini et al., 2005; Cope et al., 2005;Grigorenko, Ngorosho, Romano, Turechek, & Yrigollen, 2004; Menget al., 2005; Scerri et al., 2004). The association between KIAA0319 alsoawaits further confirmation, as there is at least one nonreplication (Barr,2005). DCDC2 also proves to be controversial (Paracchini, 2005). Finally,to our knowledge, no replications of ROBO1 have yet been attempted.Although the field has not yet converged on ‘‘firm’’ candidates, it isremarkable and of great scientific interest that all four current candidategenes for DD are involved with biological functions of neuronal migra-tion, axonal crossing, and dendrite development.

In summary, the field of DD remains an attractive, colorful mantle,stitching together different phenotypic definitions, a variety of reading-related processes, a number of regions of genetic susceptibility, andcandidate genes. It is possible that this colorfulness is attributable, in part,to the fact that groups contributing to the molecular-genetic field of DDwork in a variety of different languages, including English, Finnish,Norwegian, German, Swedish, Russian, and Italian. Here we add evenmore color to this exciting mantle by summarizing regional linkagestudies we carried out in a set of Swahili-speaking sib pairs.

As of today, work has been confined largely to languages and genotypesof European origin. Some of the language-specific differences in findingsmight be put into perspective by the analysis of genotypes and languagesfrom a different continent. We conducted a study in Tanzania amongSwahili speakers and hypothesized that similar candidate regions would beidentified as linked to the phenotype of reading difficulties. We alsohypothesized that, because of the nature of Swahili (an orthographicallytransparent language), the strongest linkage indices will be obtained withtime indicators rather than error indicators. Driven by this assumption, wewanted to explore a set of time- and error-based phenotypes on a variety ofindicators in Swahili. Althoughwe are conscious of experiment-wideType Ierror rates with a number of phenotypes, we think it is important, especiallywhen a new language is introduced for studies, to consider as full andexplicit a picture for manifestation of reading difficulties as possible.

Method

Participants

The selection of participants was carried out as follows. Initially, tests ofreading and spelling were developed in Swahili (Alcock & Ngorosho,

30 ELENA L. GRIGORENKO ET AL.

Page 5: Back to Africa: Tracing dyslexia genes in east Africajtc5/papers/AfricaDyslexia_Grigorenko_07.pdf · For example, Grigorenko and colleagues (Grigorenko et al., 1997) used continuous

2003, 2004; Alcock et al., 2000). These tests were then administered to1,476 children aged 8–14 (mean age = 12.24, SD = 1.25) studying atprimary schools (Grigorenko, Ngorosho, Jukes, & Bundy, 2006). TheReading test included a composite of tests of letter, word, and sentenceidentification; the Spelling test was a test of spelling words read aloud (fordetails, see Partnership for Child Development, 2002). The tests wereadministered twice – at the baseline and then 16 months after the base-line. Correspondingly, the selection of the probands was based on theavailability of data from 1,476 children on the four indicators: spellingand reading at the baseline and then 16 months after the baseline.

Swahili is one of a number of widely used languages of Africa (http://www.ethnologue.com/show_language.asp?code=swh), spoken through-out Tanzania, Kenya, and Uganda and in parts of Mozambique, Somalia,South Africa, and the Democratic Republic of Congo. Its linguistic lineageplaces it in the Niger–Congo group of languages, which includes more than1,500 different African languages. The most proximal grouping of Swahiliand other related languages is into the group of Narrow Bantu languages.Because Swahili�s transcription into the Roman alphabet was fairly sys-tematic (Fabian, 1986; Hombert & Hyman, 1999) rather than naturallyevolved, it is one of the most regularly spelled languages in existence today(Amberber & Collins, 2002; Heine & Nurse, 2000). This characteristic ofSwahili means that once readers master all possible phoneme–graphemecorrespondences, they can use the phonological route in both decoding andspelling any word. Thus, it is theoretically possible that, once the code isdeciphered, a Swahili reader can read and spell successfully all familiar andunfamiliar words. Yet, that is not the case, and, just as in all regular orirregular languages studied so far (Goulandris, 2003), there is a substantialgroup of children who have difficulty mastering reading and spelling (Al-cock & Ngorosho, 2003, 2004). It is that group of children – who havedifficulty learning the phoneme–grapheme code even in one of the world�smost regular languages in the presence of adequate education (i.e., at least3 years of schooling) – that was of interest in this study. This group is alsolikely to maintain the deficit even after mastering the code, but the primarymanifestation of the deficit might be in fluency rather than accuracy ofreading.

After considering the distribution of both reading and spelling mea-sures and taking into account characteristics of the distribution, we usedthe following criteria to define the sample of probands (for more details,see Grigorenko et al., 2006). First, all four measures (Reading andSpelling at times 1 and 2) were regressed on the grade the child attendedat times 1 and 2, respectively; the standardized residuals from theseregression analyses were saved. Second, the distributions of the residuals

31TRACING DYSLEXIA GENES IN EAST AFRICA

Page 6: Back to Africa: Tracing dyslexia genes in east Africajtc5/papers/AfricaDyslexia_Grigorenko_07.pdf · For example, Grigorenko and colleagues (Grigorenko et al., 1997) used continuous

were investigated; all four distributions were bimodal, with the uppermode being below or at the 75th percentile. Correspondingly, all childrenwho scored at or below the upper mode on at least one of the fourindicators formed the pool of probands. Of these children, 98 childrenhad siblings attending primary school grades and, as reported by theirmothers, sharing common parents. As a group, the probands scoredsignificantly lower than the rest of the sample at both time points (Wilks�k = .992, P<.01 and Wilks� k = .992, P<.05, for Time 1 and Time 2,respectively) on all four selection indicators: Reading at Time 1(meansample = 26.0, meanprobands = 22.3, F1,1143 = 8.9, P<.005) andTime 2 (meansample = 37.6, meanprobands = 33.9, F1,1071 = 8.0, P< .005)and Spelling at Times 1 (meansample = 35.7, meanprobands = 31.1,F1,1143 = 8.0, P<.005) and Time 2 (meansample = 41.9, meanprobands= 38.6, F1,1071 = 4.8, P<.05). There were no or only sporadic differ-ences between the group of probands and the rest of the sample on the twoIQ proxies used in the study –Digit Span (forward and backward; a subtestof theWISC), andan indicator of verbal fluency.Thus, our selection criteriaresulted in the identification of a group of probands who, on average, didnot differ substantively from the population-based sample in their abilitylevels, but that did differ in their levels of reading and spelling achievement.2

The siblings of the children identified as probands were invited to partici-pate in the study.However, for a variety of reasons,mostly absenteeismanddecision not to participate, we were not able to assess siblings of 14 of theprobands. The assessment and specimen collection resulted in the avail-ability of 84 sib pairs and 4 sib trios. The mean age of the children in theproband sample was 13.35 (SD = 1.30) and in the sibpair sample – 12.13(SD = 1.30). The sibpair sample gender structure was approximatelybalanced: 45.9% of the children were boys and 54.1% were girls.

Signed informed assent and consent forms were obtained from allchildren and their parents, respectively, before assessments began. Ap-proval for the study was obtained from the Tanzania Ministry of Healthand Ministry of Education and Culture at national, regional, district, andward levels and also by the schools and teachers participating in thestudy. Ethical clearance was obtained from the Institute of Child Health,London, UK; Yale University, New Haven, USA; and the TanzaniaFood and Nutrition Centre, Dar es Salaam, Tanzania.

Behavioral assessment

The probands and their siblings were administered the following test batteryPhonemic awareness. Phonemic awareness was assessed with a Swahilimodification of Rosner�s Test of Auditory Analysis Skills, RAAS (Rosner,

32 ELENA L. GRIGORENKO ET AL.

Page 7: Back to Africa: Tracing dyslexia genes in east Africajtc5/papers/AfricaDyslexia_Grigorenko_07.pdf · For example, Grigorenko and colleagues (Grigorenko et al., 1997) used continuous

1999). Participants are required to make a new word by deleting a specifiedelement (e.g., Say meat. Now say it again, but don�t say [m]; a parallelexample in Swahili is Sema dua [prayer]. Sasa sema tena, lakini usitamke[d]; the resulting word is ua [flower]). The trial items and the first few testitems involve deletion of a morpheme in a compound word or a syllable ina two-syllable word. The 40 remaining items involve omitting a phoneme,first in the initial position, then at the end of the word, then from conso-nant blends at the beginning of the word, and, finally, phoneme blends inthe middle of the word. Two indicators of performance were measured:number of errors (mean = 22.86, SD = 9.92, where an error counted as 1and no error counted as 0) and time of performance (mean = 10.20 s,sd = 1.13, where the total time on task was divided by the number ofitems).

Phonemic memory (short-term verbal memory, STVM). Short-termverbal memory was assessed with a nonword repetition task. Specifically,the children were presented with 40 two-to-five-syllable pseudowords andasked to repeat these words. All words used combination of letterspermissible in Swahili (e.g., sabe, peneta, pamineri, rambatajika). Theassessment was modeled after a number of assessments available in theliterature (Brady, 1997; Gathercole, Willis, Baddeley, & Emslie, 1994).The tester counted the number of errors in each word (mean = 2.17,SD = 2.1, where an error counted as 1) and time of performance(mean = 1.00 s, SD = .18, where the total time on task was divided bythe number of items).

Rapid naming. Naming fluency was assessed with the Test of RapidNaming for Colors, Objects, Numbers, and Letters (RAN). The RANconstituted a conventional presentation of Denckla and Rudel�s (1976)four tests, with the instructions translated into the appropriate Swahili.Each of the four tests consists of a chart containing five different items(i.e., five colors, five letters) presented in horizontal rows of 10 items each,repeated in a random order. The time needed to name the 50 items fromleft-to-right, top-to-bottom was registered for each card and then anaverage for each stimulus was calculated (mean = .77, SD = .43;mean = .32, SD = .12; mean = .66, SD = .30; mean = .46,SD = .26, for Colors, Objects, Numbers, and Letters, respectively). As isoften seen, the naming of Colors and Objects took longer than that ofNumbers and Letters (Stanovich, 1981).

Digit–word naming task (D/W NT). This task was modeled after thetask developed by Stanovich (1981). Similar to other naming tasks, thistask was to name the stimuli as fast as possible. Four sets of 20 stimulieach were presented – numbers from 1 to 10 printed as digits and as wordson two separate cards, and numbers from 11 to 20 printed as digits and as

33TRACING DYSLEXIA GENES IN EAST AFRICA

Page 8: Back to Africa: Tracing dyslexia genes in east Africajtc5/papers/AfricaDyslexia_Grigorenko_07.pdf · For example, Grigorenko and colleagues (Grigorenko et al., 1997) used continuous

words on two separate cards. Errors in naming were calculated(mean = 6.34, SD = 10.64 and mean = 1.08, SD = 3.34, for Wordsand Numbers, respectively). The time needed to name the numbersprinted as digits and the numbers printed as words was registered and anaverage per stimulus was calculated (mean = .53, SD = .48 andmean = .31, sd = .17, for Words and Numbers, respectively).

Reading aloud words and nonwords (RA W/NW). A task of reading 16randomly intermixed legal three- and four-syllable pseudowords and 17real words of medium to low frequency was administered to the partici-pants. Each child was asked to read the words and nonwords one by one,as quickly as possible. Because the decoding of nonwords and unfamiliarreal words in phonemically simple languages appears to be driven by thesame psychological processing (Seymour, Aro, & Erskine, 2003), we de-rived a combined coefficient for decoding (by summing the number oferrors/time for words and nonwords). Correspondingly, two indicators ofperformance were obtained: the level of accuracy in decoding both wordsor nonwords (mean = 39.92, SD = 15.70, where an error counted as 0, aself-corrected error counted as 1, and a correctly decoded word countedas 2) and performance time (mean = 2.14 s, SD = 1.73, where the totaltime on task was divided by the number of items).

Sentence repetition (SR). This task included a repetition of 6 ‘‘tongue-twister’’ sentences. Specifically, the child was asked to repeat a sentence ofthe following nature: ‘‘How can a clam cram in a clean cream can?’’(http://www.uebersetzung.at/twister/en.htm) as quickly and accurately aspossible. An example of an analogous sentence in Swahili is ‘‘Kale kak-uku kakwekwe ka kaka kako kwako kaka?’’ The examiner registered thenumber of errors during the task (mean = 20.52, SD = 8.15) and thetime per sentence (mean = 45 sec, SD = 30 s).

Genetic data

DNA was extracted from �1 ml of whole blood using the Qiagen Flex-igene� kit. Only siblings were genotyped. Three regions of interest wereincluded in these analyses: 2p, 6p, and 15q. The regions were of differentlengths because we investigated previously reported regions and at-tempted to include predominantly the markers that have previouslydemonstrated the presence of the signal or bracketed the signal. Theregion on 2p spanned 29.7 cM and included the markers D2S177,D2S2306, D2S2174, D2S391, D2S1352, D2S378, D2S2183, D2S2315,D2S337, and D2S296 (in map order). The region on 6p covered 8.8 cMand included the following STRs in map order: JA01, JA02, GAAT3A06,

34 ELENA L. GRIGORENKO ET AL.

Page 9: Back to Africa: Tracing dyslexia genes in east Africajtc5/papers/AfricaDyslexia_Grigorenko_07.pdf · For example, Grigorenko and colleagues (Grigorenko et al., 1997) used continuous

AFM342xe5, D6S506, D6S1686, D6S1050, D6S1663, JA03, D6S1660,D6S461, D6S1691, D6S276, JA04, D6S1571, D6S1281, D6S2233,D6S2238, JA05, D6S2252, D6S1260, D6S105, D6S1001, D6S2227,D6S2217, D6S1624, D6S258, JA08, and JA06. Finally, the region on 15qextended over 1.8 cM and covered the markers GATA143C02, D15S97,GABRB3, D15S822, D15S975, D15S219, D15S156, and D15S217. Thegenotyping technology used multiplex PCRs, fluorlabeled primers,pooling in each lane, and data collecting on an ABI sequencer usingGenescan and Genotyper software.

Statistical analyses

The quality of the genotyping was assessed by investigating Mendeliansegregation errors. The raw data were analyzed with GAS 2.3 (http://users.ox.ac.uk/�ayoung/GAS23/gmang23.htm) and all inconsistencieswere attempted to be resolved. When resolution by reevaluation of thegenotyping data was not possible, missing genotypes were assigned to theindividual in question. All genotypic relationships were verified and, whenappropriate, reclassified with PREST (McPeek & Sun, 2000). In allanalyses, we used the marker map positions obtained from the UCSCgenome browser (http://genome.ucsc.edu). The distances were convertedto Haldane cM because one of the software pieces used in our analyses(LOKI, see below) requires the use of this mapping function.

The distribution of allele frequencies was assessed as follows. First, weestimated allele frequencies from the sample, using the counts of thenumber of times each allele appeared. Second, we created a controlsample of 50 DNAs from the larger sample of the children in the generalstudy (see above); this sample was drawn at random from �1,000 indi-viduals who were screened for their reading and spelling scores and didnot qualify as probands. We then obtained allele estimates from thiscontrol sample based on the counts of the number of times each allele wasobserved. We compared the distributions and did not observe any distinctdifferences. To verify this conclusion, all analyses with SOLAR wereconducted with two different sets of allele frequencies – first from thesibpair sample and second from the control set of DNAs. The results werevirtually identical. Therefore, here we present only the results obtainedwith the set of allele frequencies estimated from the sibpair sample.

Multiple analytical paradigms were implemented in these analyses.First, we carried out variance component-based analyses to partition theobserved variation of the trait into a sum of variances including variancesattributable to (a) a major gene at a particular location and (b) a

35TRACING DYSLEXIA GENES IN EAST AFRICA

Page 10: Back to Africa: Tracing dyslexia genes in east Africajtc5/papers/AfricaDyslexia_Grigorenko_07.pdf · For example, Grigorenko and colleagues (Grigorenko et al., 1997) used continuous

polygenic component. These analyses were conducted using SOLAR 1.7.3(http://www.sfbr.org/sfbr/public/software/solar/) and Merlin (http://www.sph.umich.edu/csg/abecasis/Merlin/). For SOLAR, two modelswere fitted at each marker locus – M0, the null polygenic model, and M1,an additive model, where the additive variance attributable to a QTL atthat location is estimated, the dominance variance attributable to thatQTL is assumed to be zero, and a residual additive polygenic componentis specified. Logarithm of odds (LOD) scores were obtained by comparingthe log-likelihoods of M1 to M0, in other words, by comparing themaximum likelihoods between a model that assumes the presence of atrait-related QTL and an unlinked polygenic background to a modelwithout the susceptibility QTL. For Merlin, we used the ‘‘variance-components’’ option for the analyses. The tests for both softwareprograms were multipoint, and we intentionally used two different vari-ance-components approaches to investigate the convergence of results.

Second, in case the presence of the selected samples and nonnormalityof some of our phenotypes introduced biases to the results, we carried outmultipoint analyses allowing for multiple putative QTLs without a priorispecification of a number of potential loci. Specifically, we estimated theposterior distribution of a number of parameters (the number of QTLscontributing to the trait, their location, and the genotype effects for eachQTL) of interest for linkage analyses, conditioning these parameters onthe observed data and their assumed prior distribution. For this purpose,we used LOKI 2.4 (http://www.stat.washington.edu/thompson/Genepi/Loki.shtml), which implements a Bayesian reversible-jump Markov chainMonte Carlo (Heath, 1997). Because of the specifics of this approach, theoutputted values do not include conventional P-values and LOD scores.Here, as an indicator of the strength of evidence for linkage of a trait to aparticular interval of the genome, we present a statistic known as theBayes Factor (Kass & Raftery, 1995) or L-score. L-scores were computedas follows (Neuman et al., 2002). For each sampling iteration (200,000),the position of each putative QTL, marker frequencies, covariate effects,and residual variance parameters were all updated. For each subsequentrun, the previous QTL run was deleted and a new QTL was created. Theprior probability of finding a QTL linked to a 1 cM bin is 1/t, where t isthe total map length of the genome (approximately 33 Morgans).3 If, for aparticular iteration, there are n QTLs in the model, the prior probability,P, of at least one QTL located in the bin is 1)(1)1/t)n. The posteriorprobability, q, is 1 or 0 depending on whether at least one QTL is locatedin the 1 cM bin. The Bayes Factor (L-score) for each bin is estimated byaveraging q/P over all iterations. Regions that have a high probability ofcontaining a QTL will have considerably elevated L-scores compared

36 ELENA L. GRIGORENKO ET AL.

Page 11: Back to Africa: Tracing dyslexia genes in east Africajtc5/papers/AfricaDyslexia_Grigorenko_07.pdf · For example, Grigorenko and colleagues (Grigorenko et al., 1997) used continuous

with surrounding regions. The Bayes Factors cannot be quantified interms of a LOD score, but guidelines for estimating the importance of theBayes Factor indicate that L-scores in the range of 3–20 are consideredpositive signals, 20–150 are considered strong, and over 150 very strong(Kass & Raftery, 1995).

The results for SOLAR and LOKI were obtained with the variableGender introduced into the model as a covariate.

Results

As per the description of the phenotypes, all indicators used for pheno-typing in this study were quantitative. These indicators are divisible intotwo large groups, one including all counts of errors committed duringperformance of a given task and the second including measures of timespent on a given task. Specifically, the group of phenotypes indicative ofnumber of errors while performing a given task included (1) RAAS er-rors, (2) short-term verbal memory (STVM) errors, (3) digit naming task(D/W NT) errors, (4) word naming task (W NT) errors, (5) reading aloudwords and nonwords (RA W/NW) errors, and (6) sentence repetition(SR) errors. The phenotypes indicating time spent on a particular taskincluded (1) RAAS time, (2) short-term verbal memory (STVM) time, (3)RAN Colors time, (4) RAN Objects time, (5) RAN Numbers time, (6)RAN Letters time, (7) digit naming task (D NT) time, (8) word namingtask (W NT) time, (9) reading aloud words and nonwords (RA W/NW)time, and (10) sentence repetition (SR) time.

To simplify the presentation, the results of variance-component anal-yses carried out with SOLAR and Merlin are presented in graphs, com-bining the results for all error-based and time-based indicators (seeFigures 1 and 2). The same strategy is applied to the LOKI results,4 withthe exception that they are presented in separate figures (see Figures 3and 4), because the L-scores are not directly comparable to the LODscores. The most impressive results are presented in the text as well.

Chromosome 2

The investigated region on chromosome 2 (2p22.2-2p14) spanned about30 cM. This region has been previously investigated by a number ofresearchers (Chapman et al., 2004; Fagerheim et al., 1999; Fisher et al.,2002; Kaminen et al., 2003; Petryshen, Kaplan, Hughes, Tzenova, &Field, 2002) and is known to contain a susceptibility region for DD

37TRACING DYSLEXIA GENES IN EAST AFRICA

Page 12: Back to Africa: Tracing dyslexia genes in east Africajtc5/papers/AfricaDyslexia_Grigorenko_07.pdf · For example, Grigorenko and colleagues (Grigorenko et al., 1997) used continuous

Figure 1. Variance-components linkage analyses for markers on chromosomes 2, 6,and 15 conducted with SOLAR (left panel) and Merlin (right panel) for indicators oferrors. Note. All Phenotypes were analyzed in each software package. However, for

clarity, the figure includes only those phenotypes for which the LOD score exceeded 1.

38 ELENA L. GRIGORENKO ET AL.

Page 13: Back to Africa: Tracing dyslexia genes in east Africajtc5/papers/AfricaDyslexia_Grigorenko_07.pdf · For example, Grigorenko and colleagues (Grigorenko et al., 1997) used continuous

(DYX3, 2p16-p15). In this sample we confirmed the importance of thisregion. In the variance-components analytical frameworks, multiplephenotypes demonstrated signals with LOD-scores greater than 1 (seeFigures 1 and 2), and one phenotype (RA W/NW time) generated LOD-score greater than 3 (LOD-score of 3.28 with P = .0127 by SOLAR andLOD-score of 3.09 with P = .00008 by Merlin). The results obtainedfrom LOKI supported these findings; there were a number of phenotypeswhose L-scores ranged between 3 and 20 (see Figures 3 and 4). Collec-tively, these results place the strongest signal at about 5 cM into theinvestigated region (at �42–44 Mb into chromosome 2, 2p21) and, pos-sibly, a secondary signal at about 20cM into the studied region (at �57–58 Mb into chromosome 2, 2p16.1).

Chromosome 6

The region on chromosome 6 (6p22.3-p22.1) covered about 8 cM. Thisregion is known as DYX2; it has been extensively studied in DD (for areview, see Grigorenko, 2005a) and includes two candidate genes describedearlier, KIAA0319 and DCDC2 (see above). A number of phenotypesdemonstrated confirmatory signals with LOD-scores greater than 1 (seeFigures 1 and2). The strongest signalswere generatedbyWNTerrors (4.83withP<.10)5 byMerlin) and SentenceRepetitionTime (LOD-score of 3.3withP = .00005 byMerlin). Similarly, a number of L-scores greater than 3were obtained (see Figures 3 and 4), and the phenotypes of RAN Colorstime, short-term verbal memory (STVM) errors, and digit naming task (DNT) errors generated the L-scores of 140, 250, and 120, indicating thepresence of very strong signals (see Figure 5). Summarizing these findings,it can be stated that these results point to three regions of interest, one at0–2 cM into the studied interval (at�22 Mb, 6p22.3), the second at�6 cMinto the studied interval (at �28 Mb, 6p22.1), and the third at �8 cM intothe studied interval (at �29 Mb, 6p22.1).

Chromosome 15

The region on chromosome 15 included approximately 1.8 cM (15q12).This region has previously been studied in the context of research on amuch larger region on chromosome 15q and is known as DYX1 region,extending through 15q11.2–15q26.3 (Chapman et al., 2004; Grigorenko etal., 1997; Marino et al., 2004; Schulte-Korne et al., 1998). This regionharbors yet another candidate gene for dyslexia, DYX1C1 (15q21.3, see

39TRACING DYSLEXIA GENES IN EAST AFRICA

Page 14: Back to Africa: Tracing dyslexia genes in east Africajtc5/papers/AfricaDyslexia_Grigorenko_07.pdf · For example, Grigorenko and colleagues (Grigorenko et al., 1997) used continuous

Figure 2. Variance-components linkage analyses for markers on chromosomes 2, 6,

and 15 conducted with SOLAR (left panel) and Merlin (right panel) for indicators oftime. Note. All Phenotypes were analyzed in each software package. However, forclarity, the figure includes only those phenotypes for which the LOD score exceeded 1.

40 ELENA L. GRIGORENKO ET AL.

Page 15: Back to Africa: Tracing dyslexia genes in east Africajtc5/papers/AfricaDyslexia_Grigorenko_07.pdf · For example, Grigorenko and colleagues (Grigorenko et al., 1997) used continuous

above). Because we have investigated this region as a candidate region inthis sample and did not find any evidence of the gene�s involvement(Grigorenko et al., 2004), in this work we examined a more centromericregion. As evident from Figures 1 and 2, a number of phenotypes showedpositive suggestive signals with LOD-scores around 2. The highest LOD-scores were observed for W NT errors (LOD-score of 2.90 withP = .00013 by Merlin) and RA W/NW time (LOD-score of 3.77P = .0052 with SOLAR). Both phenotypes showed supportive evidencefor the region with nearby markers as well (see Figures 1 and 2). How-ever, LOKI was unable to place any QTLs in this region, possibly due toits small size.

Discussion

The field of molecular–genetic studies of reading today represents avariety of approaches using categorical and continuous, multivariate,univariate, and composite phenotypes. The field encompasses a variety ofapproaches: whole-genome scans, regional studies, and associationstudies. In this work, we decided to concentrate on a set of individualmeasures obtained by means of a number of tasks eliciting a variety ofreading-related processes.

A number of issues should be commented on in interpreting the resultsof our investigation. First, likely the most interesting aspect of our workwas that we investigated molecular bases of deficient reading-relatedprocesses in a sample of probands and their sibling in Swahili. To ourknowledge, this is the first molecular-genetic study of reading-relatedprocesses outside of the developed world in a language that has notpreviously been studied for this purpose.

Of note is that all three genomic regions considered in this studyshowed indicators of contributing to the molecular bases of readingdifficulties. The consistency of findings provided by three different ana-lytical platforms (SOLAR, Merlin, and LOKI) are reassuring.

Second, when considered holistically and speaking imprecisely, theresults appear to indicate that ‘‘more’’ and ‘‘stronger’’ signals are asso-ciated with time- rather than error-based phenotypes. This finding comesalmost as no surprise because Swahili is a language with regularorthography where errors are infrequent and most of the variance inreading ability is reflected in reading speed rather than accuracy; otherlanguages with similar transparent orthographies include Finnish(Leinonen et al., 2001), German (Landerl, 2001; Landerl & Wimmer,

41TRACING DYSLEXIA GENES IN EAST AFRICA

Page 16: Back to Africa: Tracing dyslexia genes in east Africajtc5/papers/AfricaDyslexia_Grigorenko_07.pdf · For example, Grigorenko and colleagues (Grigorenko et al., 1997) used continuous

2002; Wimmer, 1996), Italian (Zoccolotti et al., 1999), and Russian(Grigorenko, 2005b). This finding underscores again the importance ofdeveloping phenotypes that are linguistically sensitive and informed.

Third, regions on chromosomes 2 and 6 produced multiple peaks(unlike the pattern of results on chromosome 15q). One of the peaks onchromosome 2, although somewhat smaller than the other peak, landsright by the marker (D2S378), where a strong linkage signal was previ-ously identified (Fagerheim et al., 1999). The two more telomeric peakson chromosome 6 surround the area in which the two candidate genes onchromosome 6 are harbored. Moreover, the middle peak is right wherethe first signal (around the marker D6S105) was identified on chromo-some 6 (Cardon et al., 1994, 1995). These multiple peaks might representmultiple QTLs. They might also reflect some hidden misspecification inthe data.

Finally, it is very important to recognize the limitations of this study.Specifically, it is a small sample with a variety of phenotypes. Bothconsiderations instill an element of uncertainty/skepticism about results:

Figure 3. MCMC linkage analyses for markers on chromosomes 2 and 6 conductedwith LOKI for indicators of errors.

42 ELENA L. GRIGORENKO ET AL.

Page 17: Back to Africa: Tracing dyslexia genes in east Africajtc5/papers/AfricaDyslexia_Grigorenko_07.pdf · For example, Grigorenko and colleagues (Grigorenko et al., 1997) used continuous

The sample is low on power and the presence of many phenotypes mightincrease chances of false positives. However, one cannot forget that allthese regions have been implicated in studies of reading before; becausestatistical requirements for replication studies are considerably more re-laxed, we believe that we provide a sufficient amount of evidence of theimportance of these three regions.

We began this article by providing a brief summary of strategies forphenotyping used in the field of dyslexia. Here we selected one that al-lowed us to represent the fullest picture of the results we obtained for ourphenotypes. It is possible that various attempts to summarize theseindicators cohesively through a factor or discriminate score or through‘‘if–then’’ decision-making trees will lead us to a stronger, more localizedsignal. It is also possible that multivariate techniques will help us to graspthe underlying picture more clearly. For now, however, we limit ourselvesto the cautious but rather convincingly supported statement that sus-ceptibility regions DYX1–3 appear to be of importance in a sample ofsiblings speaking Swahili.

Figure 4. MCMC linkage analyses for markers on chromosomes 2 and 6 conductedwith LOKI for indicators of time.

43TRACING DYSLEXIA GENES IN EAST AFRICA

Page 18: Back to Africa: Tracing dyslexia genes in east Africajtc5/papers/AfricaDyslexia_Grigorenko_07.pdf · For example, Grigorenko and colleagues (Grigorenko et al., 1997) used continuous

Figure 5. Selected L-score graphs for MCMC linkage analyses for chromosome 6:

RAN Colors time, short-term verbal memory (STVM) errors, and digit naming task(D NT) errors.

44 ELENA L. GRIGORENKO ET AL.

Page 19: Back to Africa: Tracing dyslexia genes in east Africajtc5/papers/AfricaDyslexia_Grigorenko_07.pdf · For example, Grigorenko and colleagues (Grigorenko et al., 1997) used continuous

Acknowledgement

This research was supported primarily by the Partnership for ChildDevelopment, with headquarters at Imperial College, London, UnitedKingdom. The PCD in turn received major support from the James S.McDonnell Foundation. This work also received partial support from agrant under the Javits Act Program (Grant No. R206R00001), adminis-tered by the Institute for Educational Sciences, U.S. Department ofEducation, and from a grant P01 HD 21887, administered by the U.S.National Institutes of Health.We express our gratitude to our manyTanzanian colleagues who assisted us in data collection and processing.Moreover, the project would never have been completed without thesupport of the Tanzanian Ministries of Education and Health and localauthorities in Bagamoyo. We also express our gratitude to Dr. LindaJarvin for her assistance with transporting the samples and to Ms.Robyn Rissman for her editorial aid. Finally, our special thanks aredue to the children and their families who participated in this research.

Notes

1. Here we use the terms developmental dyslexia and specific reading disability inter-

changeably, referring to a developmental condition of difficulty mastering mentalprocessing and representing of written print.

2. None of the selection measures were used in the linkage analyses so the linkage

results were not to be biased.3. This is probably an overly conservative solution. Bayes factors should not particularly

change much when the prior is spread over the candidate regions, since the prior andposterior probabilities in each bin would increase proportionately. We have consid-

ered this possibility in a number of tests and obtained results similar to those pre-sented here. We did not, however, undertake a comprehensive exploration of thisissue.

4. The analyses are presented for s = 2, although multiple s(s) were considered.

References

Alarcon, M., Yonan, A. L., Gilliam, T. C., Cantor, R. M., & Geschwind, D. H. (2005).Quantitative genome scan and Ordered-Subsets Analysis of autism endophenotypessupport language QTLs. Molecular Psychiatry, 10, 747–757.

Alcock, K. J., & Ngorosho, D. (2003). Learning to spell a regularly spelled language isnot a trivial task – Patterns of errors in Kiswahili. Reading & Writing, 16, 635–666.

45TRACING DYSLEXIA GENES IN EAST AFRICA

Page 20: Back to Africa: Tracing dyslexia genes in east Africajtc5/papers/AfricaDyslexia_Grigorenko_07.pdf · For example, Grigorenko and colleagues (Grigorenko et al., 1997) used continuous

Alcock, K. J., & Ngorosho, D. (2004). Interaction between phonological andgrammatical processing in single word production in Kiswahili. Language & Speech,47, 1–30.

Alcock, K. J., Nokes, K., Ngowi, F., Musabi, C., Mbise, A., & Mandali, R. et al. (2000).

The development of reading tests for use in a regularly spelled language. AppliedPsycholinguistics, 21, 525–555.

Amberber, M., Collins, P. (Eds.). (2002). Language universals and variation. Praeger,

Westport, CT.Barr, C. (2005). Linkage studies of reading disabilities and ADHD in the chromosome 6p

and 15q regions. Toronto, CA: SSSR Annual Meeting: Pre-conference.

Bellini, G., Bravaccio, C., Calamoneri, F., Cocuzza, M. D., Fiorillo, P., & Gagliano, A.et al. (2005). No evidence for association between dyslexia and DYX1C1 functionalvariatns in a group of children and adolescents from Southern Italy. Journal of

Molecular Neuroscience, 27, 311–314.Brady, S. A. (1997). Ability to encode phonological representations: An underlying

difficulty of poor readers. In B. A. Blachman (Ed.), Foundations of readingacquisition and dyslexia: Implications for early intervention (pp. 21–47). Mahwah,

NJ: Lawrence Erlbaum Associates, Publishers.Cardon, L. R., Smith, S. D., Fulker, D. W., Kimberling, W. J., Pennington, B. F., &

DeFries, J. C. (1994). Quantitative trait locus for reading disability on chromosome

6. Science, 226, 276–279.Cardon, L. R., Smith, S. D., Fulker, D. W., Kimberling, W. J., Pennington, B. F., &

DeFries, J. C. (1995). Quantitative trait locus for reading disability: Correction.

Science, 268, 1553.Chapman, N. H., Igo, R. P., Thomson, J. B., Matsushita, M., Brkanac, Z., & Holzman,

T. et al. (2004). Linkage analyses of four regions previously implicated in dyslexia:Confirmation of a locus on chromosome 15q. American Journal of Medical Genetics

(Neuropsychiatric Genetics), 131B, 67–75.Chen, G., Adeyemo, A. A., Johnson, T., Zhou, J., Amoah, A., & Owusu, S. et al. (2005).

A genome-wide scan for quantitative trait loci linked to obesity phenotypes among

West Africans. International Journal of Obesity, 29, 255–259.Chiu, Y. F., Chuang, L. M., Hsiao, C. F., Hung, Y. J., Lin, M. W., & Chen, Y. T. et al.

(2005). An autosomal genome-wide scan for loci linked to pre-diabetic phenotypes

in nondiabetic Chinese subjects from the Stanford Asia-Pacific Program ofHypertension and Insulin Resistance Family Study. Diabetes, 54, 1200–1206.

Cope, N., Harold, D., Hill, G., Moskvina, V., Holmans, P., & Owen, M. J. et al. (2005).

Strong evidence that KIAA0319 on chromosome 6p is a susceptibility gene fordevelopmental dyslexia. American Journal of Human Genetics, 76, 581–591.

Denckla, M. A., & Rudel, R. G. (1976). Naming of object drawing by dyslexia andother learning disabled children. Brain and Language, 3, 1–16.

Deffenbacher, K. E., Kenyon, J. B., Hoover, D. M., Olson, R. K., Pennington, B. F., &DeFries, J. C. et al. (2004). Refinement of the 6p21.3 quantitative trait locusinfluencing dyslexia: linkage and association analyses. Human Genetics, 115, 128–

138.Fabian, J. (1986). Language and colonial power. Berkley: University of California Press.Fagerheim, T., Raeymaekers, P., Tønnessen, F. E., Pedersen, M., Tranebjærg, L., &

Lubs, H. A. (1999). A new gene (DYX3) for dyslexia is located on chromosome 2.Journal of Medical Genetics, 36, 664–669.

46 ELENA L. GRIGORENKO ET AL.

Page 21: Back to Africa: Tracing dyslexia genes in east Africajtc5/papers/AfricaDyslexia_Grigorenko_07.pdf · For example, Grigorenko and colleagues (Grigorenko et al., 1997) used continuous

Fisher, S. E., Francks, C., Marlow, A. J., MacPhie, I. L., Newburry, D. F., & Cardon,L. R. et al. (2002). Independent genome-wide scans identify a chromosome 18quantitative-trait locus influencing dyslexia. Nature Genetics, 30, 86–91.

Francks, C., Paracchini, S., Smith, S. D., Richardson, A. J., Scerri, T. S., & Cardon, L.

R. et al. (2004). A 77-kilobase region on chromosome 6p22.2 is associated withdyslexia in families from the United Kingdom and from the United States. AmericanJournal of Human Genetics, 75, 1046–1058.

Gathercole, S. E., Willis, G. S., Baddeley, A. D., & Emslie, H. (1994). The children�s testof non-word repetition: A test of phonological memory. Memory, 2, 103–127.

Gayan, J., Smith, S. D., Cherny, S. S., Cardon, L. R., Fulker, D. W., & Brower, A. M.

et al. (1999). Quantitative-trait locus for specific language and reading deficits onchromosome 6p. American Journal of Human Genetics, 64, 157–164.

Goulandris, N. (Ed.). (2003). Dyslexia in different languages: A cross-linguistic

comparison. Whurr Publishers, London.Grigorenko, E. L. (2005a). A conservative meta-analysis of linkage and linkage-

association studies of developmental dyslexia. Scientific Studies of Reading, 9, 285–316.

Grigorenko, E. L. (2005b). If John were Ivan: Would he fail in reading? In R. M. Joshi& P. G. Aaron (Eds.), Handbook of orthography and literacy (pp. 303–320).Mahwah, NJ: Lawrence Erlbaum Associates.

Grigorenko, E. L., Ngorosho, D., Jukes, M., & Bundy, D. (2006). Reading in able anddisabled readers from around the world: Same or different? An illustration from astudy of reading-related processes in a Swahili sample of siblings. Journal of Reading

Research, 29, 104–123.Grigorenko, E. L., Ngorosho, D., Romano, C., Turechek, L., & Yrigollen, C. (2004).

Two failed attempts to replicate the association between DD and DYX1C1/EKN1.Behavior Genetics, 34, 642–643.

Grigorenko, E. L., Wood, F. B., Meyer, M. S., Hart, L. A., Speed, W. C., & Shuster, A.et al. (1997). Susceptibility loci for distinct components of developmental dyslexia onchromosomes 6 and 15. American Journal of Human Genetics, 60, 27–39.

Hannula-Jouppi, K., Kaminen-Ahola, N., Taipale, M., Eklund, R., Nopola-Hemmi, J.,& Kaariainen, H. et al. (2005). The axon guidance receptor gene ROBO1 is acandidate dene for developmental dyslexia. PLoS, 1, e50.

Heath, S. C. (1997). Markov Chain Monte Carlo segregation and linkage analysis foroligogenic models. American Journal of Human Genetics, 61, 748–760.

Heine, B., Nurse, D. (Eds.). (2000). African languages: An introduction. Cambridge

University Press, New York.Hombert, J.-M., & Hyman, L. M. (1999). Bantu historical linguistics. Washington, DC:

CSLI Publications.Kaminen, N., Hannula-Jouppi, K., Kestila, M., Lahermo, P., Muller, K., & Kaaranen,

M. et al. (2003). A genome scane for developmental dyslexia confirms linkage tochromosome 2p11 and suggests a new locus on 7q32. Journal of Medical Genetics,40, 340–345.

Kammerer, C. M., Gouin, N., Samollow, P. B., VandeBerg, J. F., Hixson, J. E., & Cole,S. A. et al. (2004). Two quantitative trait loci affect ACE activities in Mexican–Americans. Hypertension, 43, 466–470.

Kass, R. E., & Raftery, A. E. (1995). Bayes factors. Journal of American StatisticalAssociation, 90, 773–795.

47TRACING DYSLEXIA GENES IN EAST AFRICA

Page 22: Back to Africa: Tracing dyslexia genes in east Africajtc5/papers/AfricaDyslexia_Grigorenko_07.pdf · For example, Grigorenko and colleagues (Grigorenko et al., 1997) used continuous

Landerl, K. (2001). Word recognition deficits in German: More evidence from arepresentative sample. Dyslexia: An International Journal of Research & Practice, 7,183–196.

Landerl, K., & Wimmer, H. (2002). Deficits in phoneme segmentation are not the core

problem of dyslexia: Evidence from German and English children. AppliedPsycholinguistics, 21, 243–262.

Leinonen, S., Muller, K., Leppanen, P. H. T., Aro, M., Ahonen, T., & Lyytinen, H.

(2001). Heterogeneity in adult dyslexic readers: Relating processing skills to thespeed and accuracy of oral text reading. Reading and Writing, 14, 265–296.

Marino, C., Giorda, R., Vanzin, L., Nobile, M., Lorusso, M. L., & Baschirotto, C. et al.

(2004). A locus on 15q15–15qter influences dyslexia: Further support from atransmission/disequilibrium study in an Italian speaking population. Journal ofMedical Genetics, 41, 42–48.

Marlow, A. J., Fisher, S. E., Francks, C., MacPhie, I. L., Cherny, S. S., Richardson, A.J., Talcott, J. B., Stein, J. F., Monaco, A. P., & Cardon, L. R. (2003). Use ofmultivariate linkage analysis for dissection of a complex cognitive trait. AmericanJournal of Human Genetics, 72, 561–570.

McPeek, M. S., & Sun, L. (2000). Statistical tests for detaction of misspecifiedrelationships by use of genome-screen data. American Journal of Human Genetics,66, 1076–1094.

Meng, H., Hager, K., Held, M., Page, G. P., Olson, R. K., & Pennington, B. F. et al.(2005). TDT-association analysis of EKN1 and dyslexia in a Colorado twin cohort.Human Genetics, 118, 87–90.

Meng, H., Smith, S. D., Hager, K., Held, M., Liu, J., & Olson, R. K. et al. (2005).DCDC2 is associated with reading disability and modulates neuronal developmentin the brain. Proceedings of the National Academy of Sciences of the United States ofAmerica, 102, 17053–17058.

Nash, M. W., Huezo-Diaz, P., Williamson, R. J., Sterne, A., Purcell, S., & Hoda, F.et al. (2004). Genome-wide linkage analysis of a composite index of neuroticism andmood-related scales in extreme selected sibships. Human Molecular Genetics, 13,

2173–2182.Neuman, R. J., Yuan, B., Gerhard, D. S., Liu, K.-Y., Yue, P., & Duan, S. et al. (2002).

Replication of linkage of familial hypobetalipoproteinemia to chromosome 3p in six

kindreds. Journal of Lipid Research, 43, 407–415.Paracchini, S. (2005). Functional analysis of the risk haplotype for dyslexia on

chromosome 6p22. Toronto, CA: SSSR Annual Meeting: Pre-conference.

Partnership for Child Development (2002). Heavy schistosomiasis associated with poorshort-term memory and slower reaction times in Tanzanian school children.Tropical Medicine & International Health, 7, 104–117.

Petryshen, T. L., Kaplan, B. J., Hughes, M. L., Tzenova, J., & Field, L. L. (2002).

Supportive evidence for the DYX3 dyslexia susceptibility gene in Canadian families.Journal of Medical Genetics, 39, 125–126.

Raskind, W. H., Igo, R. P. J., Chapman, N. H., Berninger, V. W., Thomson, J. B., &

Matsushita, M. et al. (2005). A genome scan in multigenerational families withdyslexia: Identification of a novel locus on chromosome 2q that contributes tophonological decoding efficiency. Molecular Psychiatry, 10, 699–711.

Rosner, J. (1999). Test of auditory analysis skills. Novato, CA: Academic TherapyPublications.

48 ELENA L. GRIGORENKO ET AL.

Page 23: Back to Africa: Tracing dyslexia genes in east Africajtc5/papers/AfricaDyslexia_Grigorenko_07.pdf · For example, Grigorenko and colleagues (Grigorenko et al., 1997) used continuous

Scerri, T. S., Fisher, S. E., Francks, C., MacPhie, I. L., Paracchini, S., & Richardson, A.J. et al. (2004). Putative functional alleles of DYX1C1 are not associated withdyslexia susceptibility in a large sample of sibling pairs from the UK. Journal ofMedical Genetics, 41, 853–857.

Schulte-Korne, G., Grimm, T., Nothen, M. M., Muller-Myhsok, B., Cichon, S., &Vogt, I. R. et al. (1998). Evidence for linkage of spelling disability to chromosome15. American Journal of Human Genetics, 63, 279–282.

Schumacher, J., Anthoni, H., Dahdouh, F., Konig, I. R., Hillmer, H. M., & Kluck, N.et al. (2006). Strong genetic evidence of DCDC2 as a susceptibility gene for dyslexia.American Journal of Human Genetics, 78, 52–62.

Seymour, P. H. K., Aro, M., & Erskine, J. M. (2003). Foundation literacy acquisition inEuropean orthographies. British Journal of Psychology, 94, 143–174.

Smith, S. D., Kimberling, W. J., Pennington, B. F., & Lubs, H. A. (1983). Specific

reading disability: Identification of an inherited form through linkage analyses.Science, 219, 1345–1347.

Stanovich, K. E. (1981). Relationships between word decoding speed, general name-retrieval ability, and reading progress in first-grade children. Journal of Edcuational

Psychology, 73, 809–815.Taipale, M., Kaminen, N., Nopola-Hemmi, J., Haltia, T., Myllyluoma, B., & Lyytinen,

H. et al. (2003). A candidate gene for developmental dyslexia encodes a nuclear

tetratricopeptide repeat domain protein dynamically regulated in brain. Proceedingsof the National Academy of Sciences of the United States of America, 100, 11553–11558.

Wimmer, H. (1996). The nonword reading deficit in developmental dyslexia: Evidencefrom children learning to read German. Journal of Experimental Child Psychology,61, 80–90.

Zoccolotti, P., Luca, M.de, di Pace, E., Judica, A., Orlandi, M., & Spinelli, D. (1999).

Markers of developmental surface dyslexia in a language (Italian) with highgrapheme–phoneme correspondence. Applied Psycholinguistics, 20, 191–216.

Address for correspondence: Elena L. Grigorenko, Child Study Center, Yale University, 203 SouthFrontage Road, CT 06510, New Haven, CT, USAPhone: +1-203-432-4660; Fax: +1-203-432-8317; E-mail: [email protected]

49TRACING DYSLEXIA GENES IN EAST AFRICA