OUP user menu

★ Editor's Choice ★

Neural bases of orthographic long-term memory and working memory in dysgraphia

Brenda Rapp, Jeremy Purcell, Argye E. Hillis, Rita Capasso, Gabriele Miceli
DOI: http://dx.doi.org/10.1093/brain/awv348 588-604 First published online: 17 December 2015


Spelling a word involves the retrieval of information about the word’s letters and their order from long-term memory as well as the maintenance and processing of this information by working memory in preparation for serial production by the motor system. While it is known that brain lesions may selectively affect orthographic long-term memory and working memory processes, relatively little is known about the neurotopographic distribution of the substrates that support these cognitive processes, or the lesions that give rise to the distinct forms of dysgraphia that affect these cognitive processes. To examine these issues, this study uses a voxel-based mapping approach to analyse the lesion distribution of 27 individuals with dysgraphia subsequent to stroke, who were identified on the basis of their behavioural profiles alone, as suffering from deficits only affecting either orthographic long-term or working memory, as well as six other individuals with deficits affecting both sets of processes. The findings provide, for the first time, clear evidence of substrates that selectively support orthographic long-term and working memory processes, with orthographic long-term memory deficits centred in either the left posterior inferior frontal region or left ventral temporal cortex, and orthographic working memory deficits primarily arising from lesions of the left parietal cortex centred on the intraparietal sulcus. These findings also contribute to our understanding of the relationship between the neural instantiation of written language processes and spoken language, working memory and other cognitive skills.

  • dysgraphia
  • spelling
  • working memory
  • orthography


Acquired agraphia was a topic of intense interest in the early days of modern neurology (Gordinier, 1903). The major figures of the period—Wernicke, Lichtheim, Hughlings-Jackson, Déjérine, Exner, Charcot, and others—were profoundly interested in determining whether agraphia could dissociate from either spoken language or motor deficits because they understood that the answer to this question would shed light on the brain’s capacity to instantiate evolutionarily recent cognitive skills—such as writing—independently of evolutionarily older skills. The debate has continued along similar lines with some researchers recently proposing that neural substrates are repurposed or ‘recycled’ to specifically support evolutionarily recent skills such as written language (Dehaene and Cohen, 2007), while others have argued that these skills are entirely parasitic on ‘primary’ skills (Price and Devlin, 2003). Although interest in these issues has continued, dysgraphia (we will use the terms ‘agraphia’ and ‘dysgraphia’ interchangeably) and written language production have received relatively little research attention from neurology, neuroscience, cognitive science, psychology, or even language pathology. The current study fills an important gap in the existing literature, reporting on a voxel-based lesion mapping analysis of 33 individuals with acquired dysgraphia affecting either only orthographic long-term memory (LTM) (n = 17) or orthographic working memory (n = 10), or both (n = 6). The work is novel in the application of this analytic approach to a substantial number of individuals with well-defined deficits and because of its focus on distinguishing LTM from working memory substrates. The findings provide a robust foundation for understanding the neural bases of the central processes of written language production and the associated dysgraphias, as well as contributing to broader debates regarding the neural and cognitive relationships between written language and other more basic cognitive systems.

The cognitive processes of spelling

Producing written language draws on certain cognitive processes that are not specific to written language. Thus, writing in response to heard speech (e.g. taking a phone message) requires the cognitive and neural machinery of auditory analysis and speech processing (Fig. 1). Likewise, writing either in response to speech input or to communicate ideas or concepts (e.g. writing a letter) recruits the cognitive and neural machinery for the representation of concepts and word meanings (semantics). Further, all formats for expressing the spelling of words—writing, typing, saying letter names, etc.—recruit motor processes shared by other tasks that use the same muscles. Importantly, however, producing written words additionally involves processes assumed to be specific to written language production, although the question of the ‘selectivity’ of orthographic processes has been debated (for opposing views see Cohen et al., 2002 and Price and Devlin, 2003). These core spelling processes can be grouped into central processes [orthographic LTM, orthographic working memory, and phonology-orthography conversion (POC)] and peripheral processes (allographic/letter shape selection and graphic-motor planning). In the literature, orthographic LTM is sometimes referred to as the orthographic lexicon and orthographic working memory as the graphemic buffer.

Figure 1

The cognitive architecture of spelling. Schematic depicting the cognitive architecture of spelling, highlighting central (core) and peripheral processes. P = phonological/phonology; O = orthographic/orthography; WM = working memory.

Spelling a familiar word, regardless of the task or the modality/format of output, involves retrieving orthographic information from orthographic LTM, the repository of memory representations of familiar word spellings. Orthographic LTM spelling representations consist, at a minimum, of letter identities (graphemes) and their order/position (e.g. L1 + I2 + O3 + N4). It is generally assumed that orthographic LTM representations consist of abstract letter identities, rather than letter names or letter shapes (for a review, see Fischer-Baum and Rapp, 2014). In addition to spelling familiar words, we are also able to generate reasonable or plausible spellings for unfamiliar or pseudowords (e.g. ‘fleem’ → FLEEM, FLEAM, PHLEEM, etc.) and this ability relies on POC processes that use information about the relationships between sounds and letters learned from experiences in spelling words.

Orthographic representations, whether retrieved from orthographic LTM or assembled from POC, are processed by a limited-capacity orthographic working memory system that is not only responsible for the active maintenance of letter identities and their order, but also interacts with the peripheral processes to ensure the correct serial production of the letters. In turn, the peripheral processes produce letters in a specific modality and format and, to that end, select specific letter shapes (case and font) for written spelling, letter names for oral spelling, or hand shape configurations for typing. This information is then used to drive the motor planning/execution processes in the respective effector systems (right or left hand, mouth, etc.).

The dysgraphias

Research on acquired dysgraphias has shown that disruption of different components of the spelling system leads to characteristic behavioural profiles (Buchwald and Rapp, 2009). Damage/disruption to orthographic LTM results in frequency-sensitive spelling: higher frequency words are spelled more accurately than lower frequency words, presumably because the former have more robust neural representations than the latter due to their higher frequency of use prior to the brain injury. In contrast, for these deficits, the length of words (number of letters) does not affect the likelihood of error. If damage to orthographic LTM occurs in the context of spared POC processes, then the errors produced reflect the operation of the POC system, which processes the stimulus as it would an unfamiliar word. If the target word has only common/regular sound-letter mappings (e.g. ‘camp’) the POC system will likely generate a correct spelling, whereas if it contains uncommon/irregular mappings (e.g. ‘sauce’) then a phonologically plausible error (e.g. SOSS) will likely be produced. If orthographic LTM damage occurs in the context of severe damage to the POC system, various error types may be observed, including: semantic errors (‘lion’ → TIGER), other words (‘lion’ → LINT), pseudowords (‘lion’ → LONP) and failures to respond.

The profile for orthographic working memory deficits is markedly different from that of orthographic LTM deficits. Because of the limited capacity nature of working memory systems, damage/disruption to orthographic working memory has properties similar to those observed for working memory deficits in other domains (phonological, visual or spatial). Namely, orthographic working memory deficits are characterized by sensitivity to the number of items held in orthographic working memory, such that the probability of incorrectly producing a letter increases with the number of letters in the word. These deficits are relatively insensitive to word frequency, although mild frequency effects have been reported (Sage and Ellis, 2004; Buchwald and Rapp, 2009). The error profiles are also distinctive, with responses reflecting disruption to letter identity and order, resulting in letter substitutions, deletions, transpositions and additions (‘lion’ → LIOT, LIN, LINO or LIONT) that comparably affect both words and pseudowords. Letter substitutions in orthographic working memory do not share visual or motoric features with the target letters (Rapp and Caramazza, 1997) although they do share abstract properties such as consonant/vowel status (Caramazza and Miceli, 1990), consistent with the fact that the information processed in this system corresponds to abstract graphemes.

These highly characteristic profiles permit the identification of individuals with damage to specific components of the spelling process on the basis of behavioural testing alone and without consideration of lesion locus, allowing for an unbiased selection of cases for the examination of the neural substrates supporting the affected processes.

The neural substrates of core spelling processes

Orthographic long-term memory

Deficits affecting orthographic LTM (corresponding to the clinical category of surface dysgraphia) are characterized by greater difficulty spelling irregular versus regular words and producing phonologically plausible errors. In a seminal study, Rapcsak and Beeson (2004) examined eight right-handed individuals with left inferior temporal occipital damage and found significant effects of regularity (regular words spelled more accurately than irregular ones) and lexicality (words spelled more accurately than pseudowords). They further found that the most commonly lesioned areas were the left lingual, fusiform and parahippocampal gyri with lateral extension into the inferior temporal gyrus (Tsapkini and Rapp, 2010; Purcell et al., 2014, and with Kanji-selective agraphia in Japanese (Kawahata et al., 1988; Soma et al., 1989; Sakurai et al., 1994). Orthographic LTM has also been associated with damage to other neural regions. For example, since the early work of Déjérine (1892), the spelling of familiar words has been thought to rely on the angular gyrus (Beauvois and Dérouesné, 1981; Roeltgen and Heilman, 1985). In addition, orthographic LTM deficits have been associated with hypoperfusion or infarct to posterior inferior frontal gyrus [IFG; Brodmann area (BA) 44/45] (Hillis et al., 2002, 2004).

Existing functional neuroimaging studies of spelling provide converging evidence for the role of both the ventral temporal and posterior IFG sites in spelling, with some findings supporting their specific association with orthographic LTM. Both the Purcell et al. (2011b) and Planton et al. (2013) functional neuroimaging meta-analyses found a consistent association of activation in left ventral cortex and IFG with central spelling processes. Rapp and Lipka (2011) and Rapp and Dufor (2011) provided evidence specifically linking these two regions with orthographic LTM, finding effects of lexical frequency in these two regions, with Rapp and Dufor (2011) additionally finding these areas to be insensitive to word length. Additionally, Matsuo et al. (2001) found greater blood oxygen level-dependent response in the left fusiform region for Japanese Kanji compared to Kana forms (Matsuo et al., 2001), with the former, but not the latter, requiring orthographic LTM processing.

It is noteworthy that the ventral temporal region is also central to the reading process, and has been referred to as the visual word form area (Cohen et al., 2000). In fact, several functional neuroimaging studies (Purcell et al., 2011a; Rapp and Dufor, 2011; Rapp and Lipka, 2011) have found that the same ventral temporal and posterior IFG sites are active in both reading and spelling, indicating that orthographic LTM is shared by reading and spelling (for converging evidence from lesion studies, see Rapcsak and Beeson, 2004; Philipose et al., 2007; Tsapkini and Rapp, 2010).

Orthographic working memory

Very little research, either in the lesion or neuroimaging literatures, has examined the neural substrates of orthographic working memory. Cases of orthographic working memory impairment have implicated the left frontal and parietal lobes, temporal and occipital cortex and the basal ganglia (for a review, see Cloutman et al., 2009). However, these cases often involved large strokes with multiple cognitive deficits, complicating interpretation. Cloutman et al. (2009) used diffusion and perfusion-weighted imaging (DWI/PWI) within 48 h of stroke onset to identify regions that distinguished individuals with orthographic working memory deficits from those without. They found that precentral and premotor areas, post-central gyrus (BA 4, 6, 2 and 3) and also subcortical white matter underlying prefrontal BA 48 and the caudate were most strongly associated with orthographic working memory impairment. Applying a similar approach, Hillis et al. (2002) found that hypoperfusion of BA 18/19 distinguished between individuals with and without orthographic working memory impairments.

In the only functional neuroimaging study to specifically examine orthographic working memory in spelling, Rapp and Dufor (2011) found that activity in the intraparietal sulcus/superior parietal lobule and posterior superior frontal sulcus/superior frontal gyrus was sensitive to word length but not frequency. Interestingly, it has also been claimed that the left posterior superior parietal cortex is engaged in reading tasks that are either highly attentionally demanding (Cohen et al., 2008) or require serial allocation of attention (Carreiras et al., 2015).

Phonology-orthography conversion processes

Although POC processes are not a focus of this investigation, we include a brief review as context for the discussion of orthographic LTM and orthographic working memory. POC disruption produces disproportionate difficulties in spelling unfamiliar words or pseudowords compared to words, despite accurate perception of the auditory stimuli. In terms of neural substrates, Henry et al. (2007) found that a group of individuals with perisylvian lesions exhibited greater difficulty with pseudowords than words (clinically referred to as ‘phonological agraphia’) and that the greatest lesion overlap was in the inferior frontal gyrus/frontal operculum, precentral gyrus and insula and, to a lesser degree, the superior temporal gyrus and the supramarginal gyrus. Other reports have also implicated posterior perisylvian regions, such as the anterior-inferior supramarginal gyrus (Bub and Kertesz, 1982; Roeltgen et al., 1983; Alexander et al., 1992; Philipose et al., 2007) and the angular gyrus (Hillis et al., 2002; Sheldon et al., 2008). In terms of the scant functional neuroimaging research on POC processes, Ludersdorfer et al. (2015) reported increased activity in left superior temporal gyrus for pseudoword compared to word spelling. Also, Omura et al. (2004) found that Japanese Kana writing, which relies heavily on POC processes, recruited left pre-motor cortex.

Current study

In sum, there is fairly strong convergence from lesion and functional neuroimaging studies for a role of the left mid-fusiform/ventral temporal cortex (BA 37) in orthographic LTM processes. There is also some neuroimaging evidence for the involvement of posterior IFG in orthographic LTM, although for this region there is no clear convergence from the lesion literature. With regard to orthographic working memory, the lesion-based evidence is mixed, with most individuals having large lesions; however, the (scant) neuroimaging evidence implicates regions within the left superior parietal lobule and the superior frontal sulcus. The current study addresses these important gaps in our understanding of the neural bases of spelling and dysgraphia by bringing into common registration a substantial number of cases of individuals with well-defined deficits affecting orthographic LTM or orthographic working memory and applying voxel-based lesion mapping methods to identify the neural substrates specifically associated with these processes.

Materials and methods


The 33 research participants were studied in the laboratories of B. Rapp (n = 25), G. Miceli (n = 5) and A. Hillis (n = 3), over a period of ∼15 years; the set includes 16 previously unpublished cases (Table 1). The inclusion criteria were: (i) behaviourally well-documented orthographic LTM or orthographic working memory deficits, or both; and (ii) structural scanning of sufficient quality for the voxel-based analyses. Importantly, deficit diagnosis was based strictly on behavioural findings and lesion locus was not an inclusion consideration. Consent for research participation was obtained using procedures consistent with the Declaration of Helsinki and approved by the ethical committees of the institutions where the research was performed.

View this table:
Table 1

Demographic and lesion characteristics

Age (years)Deficit typeScan typeSlice distance (mm)Lesion locationVolume (mm3)Post-onset scan time
PatientSexHandEducation (years)AetiologyLanguage
1-DSNFR16Tumour resectionEnglish68.8O-LTMT1-weighted1Left O/T Lobe21 5442.2 y
2-DPTMR19Tumour resectionEnglish36O-LTMT1-weighted1Left O/T Lobe24 7043.97 y
3-LHDFR18StrokeEnglish71.9O-LTMT1-weighted1Left O/T Lobe86 1205.8 y
4-DBYFR14StrokeEnglish54.3O-LTMT1-weighted1Left O/T Lobe50733.7 y
5-JGLFR16StrokeEnglish71.6O-LTMT1-weighted1Left O/T Lobe37 3682.8 y
6-DBAFR12StrokeItalian46O-LTMT1-weighted6Left O/T Lobe29 4487.97 y
7-PALMR8StrokeItalian57O-LTMT2-weighted6Left O/T Lobe76 36417 days
8-DHYMR16StrokeEnglish37.3O-LTMT1-weighted1Left F/P Lobe114 5592.9 y
9-VBRFR12StrokeEnglish56.7O-LTMT1-weighted1Left F/P Lobe107 3755.2 y
10-AESFL/R16StrokeEnglish58.1O-LTMT1-weighted1Left F/P Lobe224 91915.9 y
11-LSSML18StrokeEnglish54.1O-LTMT1-weighted1Left Posterior F Lobe51 3393.3 mo
12-DSKMR16StrokeEnglish64.9O-LTMT1-weighted1Left F/P Lobe145 7585.3 y
13-RHNFL19StrokeEnglish75.7O-LTMT1-weighted1Left Posterior F Lobe13 7452.5 y
14-SJNFR10StrokeEnglish35O-LTMPWI/T2-weighted2Left Posterior F Lobe44 957Same day
15-BBLFR12StrokeEnglish66O-LTMPWI/T2-weighted2Left Posterior F Lobe18 185Same day
16-CCTFR16StrokeEnglish59O-LTMPWI/T2-weighted2Left Posterior F Lobe39 891Same day
17-CIEFL14StrokeEnglish56.4O-LTMCT scan5Left Posterior F Lobe33 9227.0 mo
18-LPOFR18StrokeEnglish42.3O-WMT1-weighted1Left P Lobe80 21129 days
19-JREFR18StrokeEnglish76O-WMT1-weighted1Left F/P Lobe90 92013.04 y
20-DTEFR16StrokeEnglish80.7O-WMT1-weighted1Left F/P Lobe43 6801.6 y
21-PQSMR16StrokeEnglish54.4O-WMT1-weighted1Left F/P Lobe116 7406 y
22-BWNMR18StrokeEnglish87.4O-WMT1-weighted5Left P Lobe63 37914.9 y
BWN Secondary lesion:Right P Lobe16 00312.9 y
23-RSBMR18StrokeEnglish66.1O-WMCT scan2Left P Lobe58 27312.2 y
24-STEMR8StrokeItalian64.3O-WMCT scan0.5Left Posterior F Lobe14 6285.4 mo
25-FGIMR8StrokeItalian69O-WMT1-weighted5Left Posterior F Lobe72 2121.03 y
26-CRIMR12StrokeItalian69.9O-WMCT scan5Left Posterior F Lobe85 4706.3 mo
27-CSSMR15StrokeEnglish63.5O-WMT1-weighted5Left F/P Lobe81 4734.2 y
28-WCRMR18StrokeEnglish66.25mixedT1-weighted1Left F/P/O/T193 0767.5 y
29-MSOMR18StrokeEnglish46.1mixedT1-weighted1Left F/T Lobe176 6559.4y
30-KSTMR14StrokeEnglish62.5mixedT1-weighted1Left F/P Lobe30 6734.9 y
31-AEFFR16StrokeEnglish57.3mixedT1-weighted1Left F/P Lobe209 4808.6 y
32-DLHML23StrokeEnglish59.3mixedT1-weighted1Left F/P/O/T349 51210 y
33-KMNMR14StrokeEnglish54.9mixedT1-weighted1Left F/P60 2562.9 y
  • Orthographic LTM deficits: Cases 1–17, orthographic working memory deficits: Cases 18–27, and orthographic mixed deficits: Cases 28–33. Results of this investigation (Fig. 4) reveal that the orthographic LTM group fractionates into two groups, one with left ventral temporal lesions (Cases 1–7) and one with posterior/inferior frontal lesions (Cases 8–17). F = Frontal; P = Parietal; T = Temporal; O = Occipital.

Table 1 reports basic participant information: 17 were male, 29 right-handed (one was ambidextrous), education ranged from 8–19 years (junior high school to PhD), and five were native Italian speakers while the rest were native English speakers. Thirty suffered a single left-hemisphere stroke, one had bilateral strokes, and two had tumour resections. Twenty-five were studied in the chronic stage (more than 12 months post-stroke), five within 2–30 weeks of stroke and three within 48 h of stroke. For acute stage cases, the imaging modality was PWI; for all others, either CT, T1- or T2-weighted MRI scans were used.

Spelling and cognitive profiles

Seventeen participants had documented orthographic LTM impairment (five males), 10 had orthographic working memory impairment (seven males), and six had both orthographic LTM and working memory impairments (five males); these latter cases will be referred to as having orthographic mixed impairment. Table 2 reports data key to documenting the impairments and Table 3 reports performance on other relevant language and cognitive tasks (orthographic LTM: Cases 1–17; orthographic working memory: Cases 18–27; orthographic mixed: Cases 28–33).

View this table:
Table 2

Characteristics of spelling performance relevant for classification of individuals with orthographic LTM deficits (Cases 1–17) and orthographic working memory deficits (Cases 18–27) and with both orthographic LTM and working memory deficits (Cases 28–33)

Frequency: high versus low OR Regularity: regular versus irregular*Length (short versus long) OR Length of PPEs versus correct responses for irregular words only*Error distribution in word spelling PPE/PIN/OTHWD/DKPseudoword spellingPublication/s with additional information
1-DSN96% versus 74% (110)71% versus 71% (56)75%25%0%0%97% (34)Purcell et al. (2014)
2-DPT98% versus 80% (231)94% versus 88% (34)100%0%0%0%97% (34)Purcell et al. (2014)
Tsapkini et al. (2010, 2011)
3-LHD92% versus 84% (212)96% versus 86% (56)100%0%0%0%91% (34)Purcell et al. (2014)
Schubert et al. (2013)
4-DBY95% versus 82% (110)89% versus75% (56)92%8%0%0%97% (34)Purcell et al. (2014)
5-JGL77% versus 49% (70)45% versus 46% (77)58%42%0%0%85% (20)
6-DBA99% versus 83% (360)*PPEs: 6.14 versus Correct: 6.70*88%12%0%0%98 (40)
7-PAL99% versus 68% (261)*PPEs: 7.1 versus Correct: 7.0*94%6%0%0%95% (41)
8-DHY86% versus 54% (100)70% versus 70% (100)13%52%34%0%3% (34)Buchwald and Rapp (2009)
9-VBR90% versus 67% (98)73% versus 62% (89)16%53%30%0%9% (34)Buchwald and Rapp (2009)
10-AES96% versus 67% (51)83% versus 80% (49)12%43%46%0%29% (34)Fischer-Baum and Rapp (2012)
11-LSS23% versus 6% (70)11% versus 18% (56)0%82%15%0%0% (34)Fischer-Baum and Rapp (2014)
12-DSK54% versus 17% (70)29% versus 29% (56)6%78%17%0%3% (34)
13-RHN79% versus 54% (56)58% versus 48% (42)32%38%11%0%65% (20)
14-SJNNA61% versus 32%62%NANANA88%Hillis et al. (2004; Case 3)
15-BBL78% versus 41%70% versus 61% (128)70%NANANA81%Hillis et al. (2004; Case 1)
16-CCT64% versus 37% (71)25% versus 21%% (56)90%NANANA96%Hillis et al. (2004; Case 2)
17-CIE60% versus 9% (70)29% versus 39% (56)0%56%70%37%6% (34)
18-LPO75% versus 61% (56)89% versus 46% (56)50%44%6%0%36% (14)
19-JRE29% versus 26% (70)25% versus 0% (56)22%78%0%0%15% (20)Rapp (2005)
20-DTE36% versus 29% (56)64% versus 0% (56)4%94%2%0%17% (18)
21-PQS43% versus 32% (56)64% versus 11% (56)39%55%6%0%41%(34)
22-BWN64% versus 50% (56)86% versus 53% (487)14%79%5%0%24% (34)Buchwald and Rapp (2009)
23-RSB49% versus 46% (70)89% versus 68% (56)7%87%6%0%68% (31)Buchwald and Rapp (2009)
Rapp and Kane (2002)
24-STE58% versus 60% (80)58% versus 31% (389)4%83%13%0%28% (80)
25-FGI8% versus 8% (80)8% (183) versus 0% (155)0%92%8%0%1% (80)Zazio et al. (2013)
26-CRI55% versus 65% (80)74% versus 51% (1660)1%81%18%0%23% (80)Costa et al. (2011)
27-CSS91% versus 85% (200)98% versus 63% (130)63%33%4%0%57% (33)Goldrick and Rapp (2007)
28-WCR36% versus 18% (70)68% versus 50% (56)2%90%6%2%0% (20)
29-MSO34% versus 17% (70)39% versus 11% (56)0%12%16%72%68% (34)
30-KST71% versus 39% (70)75% versus 36% (56)16%36%8%8%0% (20)
31-AEF79% versus 57% (56)93% versus 43% (56)10%45%0%0%25% (5)
32-DLH14% versus 0% (26)8% versus 7% (26)4%88%8%0%NA
33-KMN37% versus 6% (70)39% versus 3.6% (56)4%83%9%4%0% (20)
  • Bolded cells = statistically significant P < 0.05. Accuracy = words correct. In parentheses = total number of stimuli. NA = not available; PPE = phonologically plausible errors; PIN = phonologically implausible non-word errors; OTHWD = other word errors (formal, semantic, morphological); DK = ‘don’t know’ responses.

View this table:
Table 3

Data on word comprehension, reading and naming and other working memory tasks for individuals with orthographic LTM deficits (Cases 1–17) and orthographic working memory deficits (Cases 18–27) and with both orthographic LTM and working memory deficits (Cases 28–33)

Word comprehensionDigit spanSpatial spanReadingSpoken picture naming
1-DSN91st percentile (PPVT)676 (WMS)93% (60)52% (130)
2-DPT92nd percentile (PPVT)NANANA100% (160)98% (260)
3-LHD55th percentile (PPVT)875 (WMS)61% (160)25% (65)
4-DBY99th percentile (PPVT)545 (WMS)56% (160)92% (65)
5-JGL60th percentile (PPVT)535 (Corsi)43% (60)53% (40)
6-DBA100% (WPM)53NA100% (30) WPM20% (15)
7-PAL94% (PPT)545 (Corsi)100% (30) WPM100% (29)
8-DHY87th percentile (PPVT)525 (WMS)97% (104)94% (35)
9-VBR58th percentile (PPVT)NANANA100% (112) LD93% (61)
10-AES39th percentile (PPVT)53NA53% (60)40% (134)
11-LSS7th percentile (PPVT)6NANA72% (104)83% (79)
12-DSK55th percentile (PPVT)524 (Corsi)55% (60)55% (106)
13-RHN98% (PPT)655 (Corsi)100% (60)98% (66)
14-SJN (Case 3)100% (WPM)NANANA93% (34)100% (30)
15-BBL (Case 1)100% (WPM)NANANA92% (34)93% (30)
16-CCT (Case 2)100% (WPM)NANANA92% (34)100% (30)
17-CIE1st percentile (PPVT)NANANA73% (60)78% (40)
18-LPO73th percentile (PPVT)565 (WMS)98% (160)90% (60)
19-JRE94th percentile (PPVT)534 (WMS)97% (60)92% (66)
20-DTE91th percentile (PPVT)545 (Corsi)90% (60)98% (40)
21-PQS98% (PPT)445 (Corsi)68%(60)95% (40)
22-BWN94th percentile (PPVT)3NA4 (WMS)77% (254)75% (254)
23-RSB97th percentile (PPVT)425 (WMS)88% (104)77% (192)
24-STE100% (WPM)4NA5 (Corsi)95%(60) WPM72% (30)
25-SDI98% (WPM)5NANA97% (60) WPM93% (30)
26-CRI97% (WPM)5NA5 (Corsi)95% (60) WPM73% (30)
27-CSS42nd percentile (PPVT)4NANA79% (216)82% (516)
28-WCR84th percentile (PPVT)424 (Corsi)95% (60)90% (66)
29-MSO50th percentile (PPVT)325 (Corsi)47% (85)83% (66)
30-KST10th percentile (PPVT)325 (Corsi)42% (85)56% (66)
31-AEF1st percentile (PPVT)325 (Corsi)88% (60)80% (66)
32-DLH58th percentile (PPVT)326 (WMS)86% (72)62% (60)
33-KMN27th percentile (PPVT)<2<23 (Corsi)33% (60)85% (66)
  • Comprehension was assessed with the PPVT or auditory word-picture matching (WPM) task. Reading was assessed with an oral reading task unless written word picture matching (WPM) or written lexical decision (LD) are indicated. Spoken picture naming corresponds to naming of object pictures. Parentheses = total number of stimuli. NA = not available; PPVT = Peabody Picture Vocabulary Test (Dunn and Dunn, 1981), PPT = Pyramids and Palm Trees (picture version) (Howard and Patterson, 1992), WMS = Wechsler Memory Scale (Wechsler, 1997); Corsi Blocks (Mueller and Piper, 2014).

As discussed in the ‘Introduction’ section, disruptions to orthographic LTM and working memory are distinguished by their differential sensitivity to the frequency and length (in letters) of words in spelling. Additionally, these deficits are differentiated by error types, such that for individuals with orthographic LTM deficits (with intact POC processes) errors are primarily phonologically plausible (PPEs) while, in contrast, orthographic working memory deficits result in phonologically implausible errors (e.g. letter substitutions, deletion, insertions and transpositions). Differences across labs, languages and the long time-span of data collection resulted in differences in the behavioural tests and criteria used to identify orthographic LTM and working memory deficits.

With regard to documenting orthographic LTM deficits, Cases 1–5, 8–13 and 17 were identified because they exhibited significant effects of word frequency but not length. Cases 14–16 (corresponding to Cases 1–3 in Hillis et al., 2004) were first determined to have pure agraphia based on the finding of higher than normal error rates in spelling in the context of normal error rates in oral naming, repetition and word comprehension. Orthographic LTM deficits were then identified if, in spelling, individuals exhibited a significant frequency effect, produced PPEs, and spelled pseudowords more accurately than words. Cases 6 and 7 were Italian speakers with orthographic LTM deficits. Because Italian has a highly transparent orthography, most words can be spelled correctly via either the orthographic LTM or the POC systems and, therefore, orthographic LTM impairments in Italian were diagnosed based on their higher than normal rates of PPEs on words with unpredictable sound-letter correspondences (e.g. CUOCO/QUOTA) (Rates of PPEs for Italian neurotypical individuals = 2%). Furthermore it is noteworthy that, as seen in Table 2, the orthographic LTM group apparently contained two subgroups: those with high accuracy in pseudoword spelling (65–98%) and those with low accuracy (3–29%). This indicates that the latter subgroup likely suffered from an additional severe impairment in POC processes, a hypothesis confirmed by the expected differences between the two subgroups in their rates of PPEs in word spelling (32–100% versus 0–16%, respectively).

Orthographic working memory deficits (Cases 24–26) in Italian were identified by exhibiting a length effect and a predominance of letter errors (substitutions, transpositions, omissions, additions), in the absence of frequency effects. The English-speaking orthographic working memory cases exhibited a significant effect of length but no (or mild) effects of frequency (Cases 18–23 and 27). Overall, the error profile of the orthographic working memory cases was as expected, with the errors consisting primarily of phonologically implausible errors (44–94% of errors).

Orthographic mixed deficits (Cases 28–33) were all identified in the Rapp lab and were included in the study because they exhibited significant effects of both length and frequency.

Establishing good comprehension (semantic representations) of word stimuli strengthens the diagnosis of orthographic LTM and working memory deficits. Word comprehension was generally evaluated with tasks requiring matching a heard word with a correct picture amongst distractor pictures, including: Peabody Picture Vocabulary Test (PPVT; Dunn and Dunn, 1981), the Pyramids and Palm Trees picture version (Howard and Patterson, 1992), word/picture matching subtests of the Batteria per l’Analisi dei Deficit Afasici (BADA) for the Italian participants (Miceli et al., 2006), and also non-commercial word/picture verification tasks. The results (Table 3) indicate word comprehension was within the normal range for all participants in the orthographic working memory group (Cases 18–27) and in the orthographic LTM group (Cases 1–17) except Cases 11 and 17, who did, however, show normal comprehension for concrete/imageable words.

The additional cognitive measures reported in Table 3 include digit and spatial span, as well as word reading and picture naming. Digit span was evaluated by verbally presenting increasingly longer strings of digits for spoken repetition. Visual/spatial span was evaluated either with the Spatial Span subtest of the Wechsler Memory Scale (Wechsler, 1997) or the Corsi blocks task (Mueller and Piper, 2014). Both tasks require participants to observe a subset of a total of 9 or 10 blocks/squares in a display as they are tapped (or highlighted) in a particular order. Set size increases across trials and participants are asked to reproduce each sequence. Reading was evaluated with oral word reading, written word comprehension or written lexical decision tasks. Spoken picture naming was evaluated with a variety of different picture sets.

The results of these additional measures indicate that for the orthographic LTM and working memory groups: spatial span scores were within the normal range (cut-off = 4), digit span was above the 25th percentile (except for Case 22) and both groups displayed a comparable range of accuracies in reading (orthographic LTM: 43–100%; orthographic working memory: 68–98%). In terms of spoken naming the orthographic working memory showed a somewhat higher accuracy range (72–98%) than the orthographic LTM group (25–100%). It should be noted, however, that it is not appropriate to strictly compare performance across groups in reading and spoken picture naming given the wide range of tasks/stimuli administered. Overall, it is important that low performance on these tasks was not selectively associated with either of the two groups. This reduces the possibility that any areas of lesion overlap identified with a specific group in subsequent analyses reflects differences between the groups in terms of the cognitive functions reported in Table 3 and, therefore, increases confidence that lesion overlap differences are related to the clear differences between the groups in terms of their orthographic LTM and working memory abilities.

Neuroimaging measures and data preprocessing

As indicated in Table 1, the majority of the structural scans were T1-weighted images acquired on a 3 T scanner with a slice distance of 1 mm (20/33 scans); a further 5/33 were obtained from lower resolution T1 scans with a slice distance of 5–6 mm. When T1-weighted images were not available, the following were used: 4/33 participants had CT scans (0.5 to 5 mm slice distance); 1/33 participants had T2-weighted images (6 mm slice distance); 3/33 participants had combined PWI/T2-weighted scans with PWI slice distance of 7 mm and T2 slice distance of 2 mm (the latter were used for normalization). For analysis, the slice sets for each scan were stacked into a 3D volume and then re-sliced to a voxel size of 1 mm3. These volumes were then reoriented to align with a plane through the anterior and posterior commissures (AC-PC) in order to match the standard template orientation. Although the AC-PC plane was not identifiable in the PWI images, they were transformed in the same manner as their concomitantly acquired T2-weighted images.

Lesion identification and normalization

Lesions were drawn in MRIcron by identifying the regions with signal abnormalities (Rorden and Brett, 2000) from the CT, T1- or T2-weighted image. For lesions identified via PWI, the time-to-peak arrival of gadolinium was identified in each voxel in the infarcted hemisphere; abnormalities corresponded to voxels with at least a 2.5 s delay in the gadolinium arrival time for each voxel compared to the arrival time in an intact region in the contralateral hemisphere (Hillis et al., 2004); normalization was performed on the corresponding T2-weighted image.

Each brain volume was normalized to a standard T1-weighted, CT, or FLAIR-T2 template (Rorden et al., 2012). To account for lesion-based signal abnormalities during normalization, a cost function masking procedure was used (Andersen et al., 2010). For this, the lesion volume was first smoothed by an 8 mm full-width at half-maximum (FWHM) Gaussian and then set as an exclusionary mask during normalization procedures such that the transformation parameters were derived from the intact brain tissue only (Brett et al., 2001). The normalization transformation parameters were then applied to the lesion volumes themselves to obtain lesion volumes normalized to MNI space. The normalized lesion maps were overlaid on a template brain; all voxel coordinates are reported in MNI space.

Statistical analysis of neuroimaging data

First, to provide an overview of the extent and overlap of the lesions for the three deficit groups, overlays of the individual normalized lesion maps for each of the three groups (orthographic LTM, orthographic working memory and orthographic mixed) were generated; these maps were thresholded to only include voxels with a minimum of three lesions. These results are presented in Fig. 2.

Figure 2

Lesion overlap and voxel-based lesion mapping analysis for all groups. Lesions are presented on left hemisphere sagittal slices with the x-value in MNI coordinates reported below the slices. The colour scale reflects the number of overlapping lesions; only voxels with a minimum of three lesions are depicted. (A) Overlapping lesions for the set of the 17 individuals with orthographic LTM (O-LTM) deficits. (B) Overlapping lesions for the set of the 10 individuals with othographic working memory (O-WM) deficits. (C) Overlapping lesions for the set of the six individuals with both deficit types (O-Mixed) deficits. (D) The results of Analysis 1, testing (at each voxel) for differences in presence/absence of lesion for individuals with deficits affecting orthographic LTM or orthographic working memory. This analysis included orthographic LTM, orthographic working memory, and orthographic mixed deficit lesions. Clusters of significant difference are presented on left hemisphere sagittal slices, with the x-value in MNI coordinates reported below each slice. All clusters are FDR (false discovery rate) corrected for multiple comparisons at a P < 0.05. The colour scale reflects the z-values of the significant clusters. Positive z-values indicate clusters specifically associated with the orthographic LTM deficit; negative z-values indicate clusters specifically associated with the orthographic working memory deficit.

Next, we performed two voxel-based lesion mapping analyses. In Analysis 1, in order to identify the neural substrates selectively associated with orthographic LTM or orthographic working memory deficits we considered the distributions of lesions from all participants, including individuals with mixed deficits. In Analysis 2, for reasons discussed below concerning lesion volume differences, we restricted the analysis to only the lesions from the orthographic LTM or working memory groups.

The goal of the analyses was to determine whether or not each voxel was specifically and significantly associated with one deficit type or the other. For both analyses, the voxel measure was binary (either intact or damaged) and the behavioural measures were also binary (presence/absence of orthographic working memory deficit or working LTM deficit). Rorden et al. (2007) discusses using both continuous and binary behavioural measures for lesion-symptom mapping and the MRIcron software includes a method for analysis of binary datasets amongst the voxel-based lesion–symptom mapping (VLSM) methods. We note, however, that the term VLSM is more typically used to refer to analyses that involve continuous behavioural measures. Therefore, when referring to our analyses, we have not used the term VLSM, but preferred to refer to them simply as voxel-based lesion mapping analyses. Use of this terminology highlights that the statistical testing takes place at the voxel level and that the goal is to understand the relationship between lesions and deficits. For both analyses, we used the non-parametric Liebermeister test, applying the MRIcron ‘Nonparametric Mapping (NPM)’ tool (Rorden et al., 2007; http://www.sph.sc.edu/comd/rorden/mricron/). For recent publications that apply this approach see Ptak and Schnider (2010) (n = 20 participants) and Papageorgiou et al. (2012) (n = 30), among others. Additionally, it is worth noting that one major concern regarding lesion-deficit mapping approaches is that the areas identified with a deficit may simply correspond to areas most likely to be lesioned for reasons related to the nature of vascular distribution. However, this concern is reduced because we are performing contrast analyses rather than association analyses. In a contrast analysis it is precisely the voxels that are lesioned in both deficit types that are ‘discounted’ as they are not specifically linked to only one deficit type. Finally, to control for multiple comparisons, a false discovery rate (FDR) of 0.05 alpha level was adopted and we limited our interpretation to voxels that were damaged in at least 10% of participants and to clusters with an extent larger than 100 mm3.


Lesion overlap

For the 17 individuals in the orthographic LTM group (Fig. 2A) there were two left hemisphere areas with extensive lesion overlap: one in the posterior frontal lobe and another in the ventral temporal lobe. Each of these areas included foci with lesions from at least seven individuals from the orthographic LTM group (depicted in dark orange). Furthermore, there was no spatial overlap between the frontal and temporal lobe lesions, such that no cases with orthographic LTM had lesions affecting both areas. These findings reveal that the orthographic LTM group consisted of two subgroups with distinct lesion topography. For the orthographic working memory group (Fig. 2B), there was considerable lesion overlap in both the parietal and the superior frontal lobe, with the most densely populated lesions occurring in the parietal lobe, centred on the intraparietal sulcus region (depicted in pink). For the orthographic mixed group, Fig. 2C indicates there are multiple areas of high lesion overlap in both frontal and parietal lobes, including portions of the left inferior frontal gyrus (pars opercularis), middle frontal gyrus, insular cortex, central operculum, supramarginal gyrus and angular gryus.

As seen in Fig. 2, in addition to the apparently spatially distinct sites of lesion concentration for the orthographic LTM and working memory groups, there was also considerable overlap across the orthographic LTM and working memory groups, especially in frontal and anterior parietal areas. This overlap makes it especially important to determine if there are areas within these overlap regions that are selectively associated with each of the deficit types; this question was the focus of Analyses 1 and 2.

Voxel-based lesion mapping analyses

Analysis 1

This analysis identified regions associated with either orthographic LTM or working memory deficits based on the lesions of all the participants, including the mixed cases. The results of the analysis, as indicated in Fig. 2D, confirmed the basic distinctions already apparent in Fig. 2A–C indicating different left hemisphere regions associated with orthographic LTM or working memory deficits, with a largely parietal locus associated with orthographic working memory and fronto-temporal sites associated with orthographic LTM. However, before carrying out a more detailed evaluation of these findings, we needed to address a concern common to this type of analysis approach, namely that results can be affected by differences in lesion volumes. To evaluate this possibility, we carried out a comparison of lesion volumes across the three participant groups. The comparison shows that the mean and standard error of the lesion volumes for the individuals with only orthographic LTM and working memory deficits were 63.3 cm3 [standard error (SE) = 8.8] and 72.3 cm3 (SE = 13.9), respectively and that these values were not significantly different from one another [t(27) = 0.9, P = 0.5]. In contrast, the individuals with mixed deficits, had significant larger lesion volumes (mean = 169.9 cm3; SE = 46.8) than either the orthographic LTM group [t(21) = 5.57, P = 0.007] or the orthographic working memory group [t(14) = 5.25, P = 0.02]. This analysis raises the concern that some of the effects observed in Analysis 1 could be driven by the inclusion of a set of lesions that were, on average, over twice as large as the lesions in either the orthographic LTM or working memory group. Larger lesion sizes may produce a larger number of different deficits, increasing the possibility that areas of significant overlap in the larger lesion group may reflect additional cognitive deficits outside the cognitive functions of interest. In fact, this possibility is consistent with the observation (Table 3) that the mixed deficit group had generally lower word comprehension and digit span scores than the other two groups. To address these concerns regarding lesion volume differences, we carried out Analysis 2 excluding the orthographic mixed deficit group, focusing instead on the two groups with single deficits and comparable lesion volumes.

Analysis 2

As listed in Table 4 and depicted in Fig. 3, this analysis identified three clusters selectively associated with orthographic LTM impairment and one cluster specifically associated with orthographic working memory impairment. A comparison of Figs 2D and 3 reveals remarkably similar results, but a number of fairly small clusters observed in Analysis 1 were no longer significant in Analysis 2. While one cannot be sure what these small clusters represent, the fact that they are associated with the set of large lesions indicates that one cannot rule out the possibility that they are unrelated to the spelling deficits.

Figure 3

Voxel-based lesion mapping comparison of orthographic LTM and working memory deficit groups. The results of Analysis 2, testing (at each voxel) for differences in presence/absence of lesion for individuals with deficits affecting orthographic LTM or working memory. This analysis was restricted to individuals with only orthographic LTM or working memory deficits. Clusters of significant difference are presented on medial-to-lateral left hemisphere sagittal slices (on left), and superior-to-inferior axial slices (on right); x (for sagittal slices) and z (for axial slices) MNI coordinates are reported to the right of each slice. All clusters are FDR (false discovery rate) corrected for multiple comparisons at a P < 0.05. The colour scale reflects the z-values of the significant clusters. Positive z-values reflect the orthographic LTM deficit clusters; negative z-values reflect the orthographic working memory deficit clusters.

View this table:
Table 4

Results of lesion comparison analysis

Group comparisonCentre of mass (MNI)Peak ZCluster size (1 mm3)Anatomical structures (left)% of cluster
O-LTM > O-WMCluster 1−49−39−192.33749Inferior temporal gyrus48%
Fusiform gyrus23%
White matter29%
Cluster 2−52−191821492IFG pars opercularis21%
IFG pars triangularus10%
Frontal operculum10%
Middle frontal gyrus3%
White matter56%
Cluster 3−3818162.3688Central operculum56%
Postcentral gyrus13%
White matter31%
O-LTM < O-WM−39−50364.423 708Superior parietal lobe24%
Angular gyrus18%
Supramarginal gyrus14%
Superior LOC13%
Parietal operculum3%
White matter27%
  • x, y, z correspond to coordinates in MNI space. O = orthographic; WM = working memory; LOC = lateral occipital cortex.

For each cluster identified, we used the xjview (http://www.alivelearn.net/xjview8/) matlab toolbox to determine the percentage distribution of each cluster’s volume across neuroanatomical regions defined using the Harvard-Oxford Cortical and Subcortical Structural Atlases (http://www.cma.mgh.harvard.edu/). With regard to the three orthographic LTM lesion sites (Fig. 3 and Table 4), the largest cluster is centred in the left inferior temporal gyrus extending into the left fusiform gyrus. It is associated with 7/17 of the individuals with orthographic LTM deficits and all of these seven lesions are contained within the left temporal and occipital lobes; we refer to these cases as the orthographic LTM ventral temporal subgroup. In contrast, there is an orthographic LTM frontal subgroup, associated with two clusters: one is centred in the left pars opercularis, extending into the left pars triangularis and the frontal operculum, and another is situated primarily in the central operculum, with a centre of mass that is 40 mm from the pars opercularis cluster. Combined, these two posterior frontal sites are associated with 10/17 of the individuals with orthographic LTM deficits and with all of individuals in the orthographic LTM frontal subgroup. Of the 10 cases with frontal orthographic LTM, four involve both frontal sites, three involve only the IFG and three only the central operculum site. These findings provide clear evidence of the fractionation of the orthographic LTM deficits into (at least) two subgroups based on lesion neurotopography, with lesions centred either in the left inferior temporal gyrus/fusiform gyrus or in the posterior IFG. Whether the two posterior IFG lesion sites should be considered to be further fractionations corresponding to functionally independent deficit loci is unclear given their proximity.

With regard to the neural substrates selectively associated with orthographic working memory deficits, a single large cluster of voxels was identified, affecting left posterior parietal/occipital regions including the angular gyrus, superior lateral occipital cortex and extending inferiorly to include a small portion of the supramarginal gyrus (Fig. 3 and Table 4). Of the 10 individuals with orthographic working memory deficits, all have lesions within the region of significant overlap, and 8/10 have lesions that include the area of peak overlap (−32, −54, 34). Furthermore, none of these individuals have lesions that extend down into the ventral temporal lesion area associated with orthographic LTM ventral temporal subgroup.

A strong prediction from these results is that the orthographic mixed deficit group, because they suffer from both orthographic LTM and working memory deficits, should have damaged voxels within the lesion overlap areas associated with both deficit types. This prediction is upheld: all six of the individuals with orthographic mixed deficits have lesions within the orthographic working memory parietal/occipital cluster and all six have lesions within the orthographic LTM IFG cluster; in addition, 4/6 in the orthographic LTM central cluster and 4/6 in the superior portion of the inferior temporal gyrus region associated with orthographic LTM.


Voxel-based lesion mapping analyses were carried out to characterize the lesion topography of 33 individuals with acquired dysgraphia subsequent to stroke who had been classified, solely on the basis of behavioural criteria, as suffering from disruption to the orthographic long-term and/or working memory components of the spelling process. The analysis revealed: (i) a distinct neurotopography of lesions producing orthographic LTM and working memory deficits; (ii) fractionation of orthographic LTM into sites in the left ventral temporal cortex and the posterior IFG; and (iii) lesions giving rise to orthographic working memory impairments were centred in the area of the left intraparietal sulcus. These findings serve to clarify and extend the previous literature regarding the neural bases of spelling and dysgraphia and also to address and raise several additional issues.

Relationship of the current findings with previous neuroimaging and lesion-based reports

Figure 4 depicts the findings of the current study along with those of a meta-analysis of functional neuroimaging studies of spelling in neurotypical adults (Purcell et al., 2011b; Planton et al., 2013). As seen in Fig. 4, there is striking overall correspondence between the results of the neuroimaging meta-analysis and the current voxel-based symptom lesion mapping findings. In fact, each of the four lesion overlap clusters reported in Table 3 has a close correspondence with an activation likelihood peak, with Euclidean distances between them ranging only from 9–28 mm. In addition to the spatial correspondences, there is close correspondence between the functions assigned to these sites from the lesion-based and neuroimaging methods. The left posterior frontal and ventral temporal sites linked to orthographic LTM deficits were associated in functional MRI with the orthographic LTM profile of blood oxygen level-dependent sensitivity to word frequency but not length. Furthermore, the parietal site linked to orthographic working memory lesions was associated in functional neuroimaging with sensitivity to word length but not frequency (Rapp and Dufor, 2011).

Figure 4

Voxel-based lesion mapping results depicted alongside spelling neuroimaging findings. Comparison of the lesion mapping results from the current study (left) with meta-analysis of functional neuroimaging studies of spelling (Purcell et al., 2011b; used with permission). All clusters are projected onto the surface of a canonical brain with a search depth of 16 mm. Left = left hemisphere.

In terms of the lesion literature regarding orthographic LTM deficits, the current study provides strong confirmation of previous reports of the relationship between lesions of the left ventral temporal region and orthographic LTM deficits (Rapcsak and Beeson, 2004; Tsapkini and Rapp, 2010; Purcell et al., 2014). Importantly, the findings also strengthen claims regarding the role of the left posterior IFG in orthographic LTM based on hypoperfusion studies in acute stroke (Hillis et al., 2002, 2004). Note that while the current study includes three of the five individuals reported in Hillis et al. (2004), it does not include any from the Hillis et al. (2002) report. In addition to these three, it also includes seven chronic lesion cases of orthographic LTM impairment with frontal lesions. Finally, the findings are consistent with the Rapcsak and Beeson (2004) argument that the angular gyrus may not play a critical role in orthographic LTM, contrary to the long-standing claim based on Déjérine (1892). Although the absence of findings in the angular gyrus does not rule out that the area may be involved in orthographic LTM, this study actually indicates that lesions in this area are more likely to be associated with orthographic working memory deficits. To account for the longstanding view linking the angular gyrus with orthographic LTM deficits, Rapcsak and Beeson (2004) argued that many of the angular gyrus lesions supporting its association with orthographic LTM extended inferiorly to include regions of the ventral temporal cortex, an interpretation that strengthens the association of ventral temporal areas with orthographic LTM deficits.

With regard to lesion-based evidence of orthographic working memory substrates, as indicated earlier, the evidence is scant and mixed although the parietal site identified in this study has been associated with orthographic working memory deficits in at least some previous reports (Miceli et al., 1985). In addition, Fig. 2B also indicates considerable frontal involvement in orthographic working memory lesions, even though the statistical analysis did not reveal a selective association within the frontal lobe with orthographic working memory deficits. Nonetheless, the observed lesion overlap is consistent with claims of the role of precentral and premotor areas in orthographic working memory in acute stroke (Cloutman et al., 2009), with the role of Exner’s area in pure dysgraphia and graphic-motor planning more generally (Roux et al., 2009) and with the finding of blood oxygen level-dependent sensitivity to word length but not frequency in superior prefrontal cortex (Rapp and Dufor, 2011). However, given the considerable lesion overlap in this region between the orthographic LTM and working memory groups, more cases may be needed to identify if there are indeed frontal areas selectively associated with orthographic working memory.

Finally, it is worth mentioning that the findings regarding pseudoword spelling have some relevance to understanding the substrates subserving POC processes. In this regard, while the pseudoword spelling accuracy of the frontal orthographic LTM subgroup included 6/10 individuals with very poor pseudoword spelling (0–29% accuracy), the individuals with the ventral temporal lesions all had good to excellent pseudoword spelling (85–98% accuracy). This suggests that key pseudoword spelling processes are unlikely to be situated within the ventral-temporal lesion region and points, instead, to some role for frontal/perisylvian areas, a finding generally consistent with Henry et al. (2007).

Questions of domain and modality specificity in working memory

Phonological and visual working memory processes have been generally linked to the left hemisphere parietal sites associated in this study with orthographic working memory. Specifically, the supramarginal gyrus has been previously associated with phonological working memory (Paulesu et al., 1993), while the superior and posterior parietal/occipital areas identified in this study have been associated in the functional neuroimaging literature with visual (Todd and Marois, 2004) and spatial working memory (Wager and Smith, 2003). One account of this association across domains and modalities is that although there are distinct working memory processes, they occupy adjacent parietal subregions, with orthographic working memory constituting a distinct working memory process. Another possibility is that different domains and modalities of working memory are undifferentiated and depend upon a shared function that is distributed across this general area.

While it is beyond the scope of this paper to fully review this issue, there are some behavioural findings from the current study that generally favour the hypothesis of domain-specific working memory processes and substrates. While several of the orthographic working memory cases (Table 3) do have somewhat reduced phonological (digit) spans, several do not, indicating that orthographic and phonological working memory deficits do not necessarily co-occur. Similarly, all but two of the cases with orthographic working memory have spatial span scores within the normal range and the average forward spatial span for the orthographic working memory group (4.75) is not different from that of the orthographic LTM group (5.0) (t = 1.05, P < 0.05). This dissociation of orthographic working memory from spatial working memory is also consistent with claims in the functional neuroimaging literature of reading-specific letter processing in bilateral occipitoparietal cortex (Cohen et al., 2008) and with claims of letter-position working memory coding for visually presented letters in posterior parietal cortex (Carreiras et al., 2015). Both areas reported in these neuroimaging studies are near (within 1 cm of) the region highly associated with orthographic working memory in the current study. The findings suggest orthography-specific working memory/attentional processes in left hemisphere parietal areas that are generally homologous to right hemisphere areas commonly associated with visual working memory/attention networks (Wager and Smith, 2003). For reading and spelling, these parietal orthographic working memory areas may interact with the ventral temporal cortex orthographic LTM region providing the ‘top-down’ amplification of orthographic information (Cohen et al., 2008) and serially directed attention required for letter selection in production.

The multiple components of orthographic LTM

One interpretation of the fractionation of the lesion loci for the orthographic LTM deficit group into the IFG and ventral temporal sites is that these regions together instantiate a single, distributed orthographic LTM function. Another hypothesis is that the two regions instantiate distinct cognitive functions that both contribute to orthographic LTM with the frontal site involved in cognitive control for lexical orthographic selection and the ventral temporal region responsible for the storage of word spellings. According to the second hypothesis, the two regions would work interactively for the retrieval and selection of word spellings. Generally consistent with this view are claims that the ventral temporal region supports the representation of orthographic word forms in reading (visual word form area) (Cohen et al., 2002; Glezer et al., 2009) and serves as the interface between lexical orthographic and semantic information in spelling (Purcell et al., 2014). The IFG region, on the other hand, has been frequently associated with cognitive control generally, and with word selection specifically (Thompson-Schill et al., 1997; Vuong and Martin, 2011; Harvey et al., 2013). One specific interpretation of the posterior IFG’s contribution to orthographic lexical selection derives from claims of the IFG’s role in the perisylvian phonological network (Fiez et al., 2006). Under that view, the IFG contributes to orthographic lexical selection by providing phonological constraints (Henry et al., 2007) on the selection of orthographic representations in ventral temporal cortex. This view predicts that individuals with orthographic LTM deficits and frontal lesions would necessarily have damage affecting the phonological (POC) processes of spelling. However, as indicated by performance on pseudoword spelling reported in Table 2, while most of the frontal orthographic LTM group did indeed have POC impairments, at least 3 of the 10 individuals had normal pseudoword spelling performance (Cases 14–16), making it less likely that the IFG’s role in orthographic lexical selection is based entirely or necessarily on providing phonological constraints on lexical selection. It does not, however, diminish the possibility that the IFG carries out some type of control in orthographic lexical selection.

While nothing in the current dataset clearly favours one of these specific interpretations of the orthographic LTM lesion fractionation over another, the finding of a fractionation is nonetheless important as it provides a foundation for asking more detailed questions regarding the functional roles of the frontal and ventral temporal areas in spelling.


The neural distribution of lesions selectively affecting orthographic working memory and long-term memory was mapped using voxel-based lesion mapping methods. The results constitute substantial progress in our understanding of the neural bases of the core processes of spelling and how these may be affected in acquired dysgraphia. The findings also have broader implications for our understanding of the relationship between written language processes on the one hand and spoken language, visual and spatial processes on the other.


We gratefully acknowledge the support of the multi-site NIH-supported grant DC012283 examining the neurobiology of language recovery in aphasia to B.R., as well as NIH grant DC05375 to A.H. and grants from Provincia Autonoma de Trento and Fondazione CaRiTRo to G.M.


We thank Jennifer Shea and Chloe Haviland for their many valuable contributions to this project.

inferior frontal gyrus
long-term memory
phonology-orthography conversion
phonologically plausible errors
perfusion-weighted imaging


View Abstract