The black box of global aphasia: Neuroanatomical underpinnings of remission from acute global aphasia with preserved inner language function

OBJECTIVE
We studied an unusual case of global aphasia (GA) occurring after brain tumor removal and remitting one-month after surgery. After recovering, the patient reported on her experience during the episode, which suggested a partial preservation of language abilities (such as semantic processing) and the presence of inner speech (IS) despite a failure in overt speech production. Thus, we explored the role of IS and preserved language functions in the acute phase and investigated the neuroanatomical underpinnings of this severe breakdown in language processing.


METHOD
A neuropsychological and language assessment tapping into language production, comprehension, attention and working memory was carried out both before and three months after surgery. In the acute stage a simplified protocol was tailored to assess the limited language abilities and further explore patient's performance on different semantic tasks. The neuroanatomical dimension of these abrupt changes was provided by perioperative structural neuroimaging.


RESULTS
Language and neuropsychological performance were normal/close to normal both before and three months after surgery. In the acute stage, the patient presented severe difficulties with comprehension, production and repetition, whereas she was able to correctly perform tasks that requested conceptual analysis and non-verbal operations. After recovering, the patient reported that she had been able to internally formulate her thoughts despite her overt phonological errors during the episode. Structural neuroimaging revealed that an extra-axial blood collection affected the middle frontal areas during the acute stage and that the white matter circuitry was left-lateralized before surgery.


CONCLUSIONS
We deemed that the global aphasia episode was produced by a combination of the post-operative extra-axial blood collection directly impacting left middle frontal areas and a left-lateralization of the arcuate and/or uncinated fasciculi before surgery. Additionally, we advocate for a comprehensive evaluation of linguistic function that includes the assessment of IS and non-expressive language functions in similar cases.


Introduction
I have thoughts that I can't manage to translate into sounds.

Martino Sclavi in his book
The Finch in My Brain about postneurosurgical recovery (Sclavi, 2017).
Global aphasia (GA) is the most severe form of aphasia, affecting virtually all facets of language production and comprehension. The prognosis of language recovery is usually poor in the affected patients. In some cases, this form of language impairment may evolve into a less severe (although still chronic) aphasic syndrome. Interestingly, it has been shown that, despite failures in overt speech production, individuals with milder forms of aphasia may have preserved inner speech (IS), defined as the subjective experience of speaking in one's head (IS, Fama, Hayward, Snider, Friedman, & Turkeltaub, 2017;. For instance, aphasic patients with anomia often frustratingly experience that they can "say or hear the correct words in their mind" despite failing to produce them out loud (Fama et al., 2017). Indeed, there is some evidence that IS and overt speech can dissociate in aphasia (Geva et al., 2011), indicating that the two are at least partially independent functions. Further, recent theoretical models assume that IS can have different levels of information specification, from full articulatory and phonological specification (concrete or expanded IS) to only semantics without specific formulation and articulation (abstract or condensed) (see Fernyhough, 2004;Geva et al., 2011;Geva and Fernyough, 2019;Grandchamp et al., 2019). In this respect, we use the term to refer to the subjective experience of verbalizing in one's head, independent from overt speech that can vary from the experience of a voice percept with precise articulatory specification (expanded inner speech) to a conceptual message without articulatory specification (condensed inner speech, see Grandchamp et al., 2019;Loevenbruck et al., 2018).
The concept of IS has generated an intriguing and unresolved debate in linguistics. It is argued whether language and thought processes are functionally inseparable or not (Gleitman & Papafragou, 2005;Sokolov, 2012;Vygotsky, 1987). The "lingualism" theory suggests that there is no thought without language (Sapir, 1941;Whorf, 1956;Wittgenstein, 1922, p. 196, see also Vicente & Martinez Manrique, 2011), whereas the "language of thought" stream proposes that language is inessential to private thought (Fodor, 1975;for a review;Hermer-Vazquez, Spelke, & Katsnelson, 1999, see also ;Edwards, Ellams, & Thompson, 1976). According to the first theory, mental operations such as mental calculation per se are supported by IS (Hermer-Vazquez et al., 1999;Sokolov, 2012), whereas the other standpoint proposes that the brain may hold these processes separately (Basso, De Renzi, Faglioni, Scotti, & Spinnler, 1973;Hamsher et al., 1991). Recently, this controversy has been revived in the light of recent anatomo-functional observations. For example, results from voxel-lesion symptom mapping studies in patients and functional imaging studies (using fMRI) using linguistic and non-linguistic tasks in healthy participants have provided partial evidences in support of both views (Baldo et al., 2005;Fedorenko & Varley, 2016;Geva et al., 2011).
Evidence of whether individuals with GA preserve their IS ability is rather scarce (Bek, Blades, Siegal, & Varley, 2010). In order to address this gap of knowledge, we present an unusual case of an individual who presented GA in the first few days after brain tumor removal but maintained a good level of performance in tasks requiring conceptual analysis and nonverbal operations. Captivatingly, she fully recovered within the first month after surgery and could report on her subjective experience while dealing with GA, which suggested a preserved IS. We explored the neuroanatomical underpinnings that sustained her recovery from the GA episode and put them together with her objectively measured language performance and subjective reports, during and after the episode.

Case presentation
TD, a 35 year old, right-handed woman, was admitted to the Neurosurgery Department of the Bellvitge Academic Hospital as a candidate for brain tumor resection (diffuse astrocytoma, WHO, II). TD was in a stable partnership in a city of moderate size and was a mother of two. She accomplished secondary school (10 years of formal education), being a Spanish speaker with a high level of Catalan (qualifying as a SpanisheCatalan bilingual) and some knowledge of English. The study protocol was accepted by the Hospital Universitari of Bellvitge Ethical Committee in accordance with the principles of the Declaration of Helsinki and the participant signed the informed consent for the participation in the study. As this report is a case study, no part of the study procedures or analyses were pre-registered in a time-stamped, institutional registry prior to the research being conducted. We report all data exclusions, all exclusion criteria, whether exclusion criteria were established prior to data analysis, and all measures in the study. All data reported formed part of standard medical care. In this sense, MRI, neuropsychological, and language perioperative testing was performed before surgery, intrasurgically and from 3 to 7 months after the surgical intervention. In terms of behavioral testing e given that this report concerns the acute post-surgical global aphasia episode, we included bedside assessment results as a centric point of our focus. Additionally, we added baseline and early recovery scores for comparison. With regard to the neuroimaging data, a standard medical care protocol included a set of anatomical and functional images, but, for this report, we included only the pre-surgical Diffusion and T1 weighted images, the T1-weighted and Fluid-Attenuated Inversion Recovery Sequence (FLAIR) during the acute post-surgical stage and an anatomical, T1 weighted image at 7 months post-surgery. We excluded the intrasurgical results from the present report. No exclusion criteria were established prior to data analysis, given that all the results were previously used for clinical purposes.
The conditions of our ethics approval do not permit public archiving of the data supporting this study, according to the corresponding ethics committee of our hospital (Comite Etic d'Investigaci o Clínica, Bellvitge Hospital). Readers seeking access to this data should contact the lead author (JS). Access will be granted to named individuals in accordance with ethical procedures governing the reuse of sensitive data. Specifically, the person responsible of the research team should be willing to sign a document in which they commit to maintain confidentiality with regard to the shared dataset.

Neuropsychological and language assessment
The presurgical and post-surgical (3 months after surgery) neuropsychological and language assessment involved a preset collection of standardized tests used at the Neurology Ward of Hospital Universitari de Bellvitge for all brain surgery candidates. Following this, we measured: handedness (Edinburgh Inventory, Oldfield, 1971), verbal comprehension (Token Test, De Renzi andFaglioni, 1978), semantic (animals) and phonological (letter p) verbal fluency, attention and working memory (Digit Span from the WAIS e Wechsler Adult Intelligence Scale) (Wechsler, 1997) and naming (Boston Naming Task, Goodglass and Kaplan, 2001) and a home-made simplified version of the picture-naming task, composed by 60 black and white drawings selected from a standard stimuli database (Snodgrass & Vanderwart, 1980, see also : Havas et al., 2015). After establishing that TD was SpanisheCatalan bilingual, a handful of additional tasks was added allowing to test both languages intraoperatively, as well as her language switching profile (Rodriguez-Fornells et al., 2012) and abilities (for more details on this task, please consult our previous work: Sierpowska et al., 2013Sierpowska et al., , 2018. Spanish normative data for these tasks was obtained from the Neuronorma database by Peña-Casanova and collaborators (Peña-Casanova, 2005; Peña-Casanova, Quinones-Ubeda, Gramunt-Fombuena, Aguilar, et al., 2009;Aranciva et al., 2012;Casals-Coll et al., 2013) and from the Spanish adaptation of the Wechsler Memory scale III edition (Wechsler, 2004). In the acute postsurgical stage (4 days after surgery), the neuropsychologist team intended to evaluate TD using the same protocol. However, given the patient's unstable condition, the assessment was readapted to a more approachable form and also repeated three days later (7th day post-surgery, consult Table 1 for a more comprehensive overview). In this sense, an attempt to perform the Token test was undertaken showing a very poor performance and, thus, a sentence comprehension task was proposed as well (TBR e Test de Barcelona Revisado) (Peña-Casanova, 2005). In the task of repetition of words and nonwords, non-words were substituted by syllables and syllable pairs to maximally simplify the stimuli set and in this way, make it more approachable to the patient. Automated series naming was measured by weekdays naming (spoken) and counting to ten (both spoken and written) (all tests included in the TBR) to test for automated speech output and to scan for possible language switching errors. Additionally, two semantic matching tasks were carried out: Pyramids and Palm Trees (PPT, Howard and Patterson, 1992) and Environmental Sound Recognition Task (Bozeat, Lambon Ralph, Patterson, Garrard, & Hodges, 2000). These latter tasks were useful to ensure if the hidden (semantic) facets of language processing are still preserved despite the breakdown in overt language production. Attention, short term and working memory were measured by a Symbol Search task (included in the WAIS) and Corsi Cubes. The Symbol Search (cancellation) task also allowed to rule out the possible appearance of hemineglect. With regard to motor function evaluation, measures of unilateral and bilateral posture sequences and imitation were implemented (see Table 1 for more details).
Legal copyright restrictions prevent public archiving of the test materials used in the study, which can obtained from the copyright holders in the cited references.

Clinical neuroimaging data acquisition and analyses
Anatomical (T1-weighted and diffusion tensor) images were collected at the Hospital Universitari de Bellvitge before surgery allowing the presurgical damage assessment and support intraoperative neuronavigation. Given the delicate condition of TD in the acute post-surgical stage it was only possible to acquire a short protocol including T1 and Fluid Level Attenuation Imaging Resonance (FLAIR). Seven months after surgery, structural neuroimaging was acquired again for the post-surgical followup. In this report we show only the T1 weighted data. The scanner used was the Philips Intera 1.5-T system with a maximum field gradient strength of 76 mT/m. T1-weighted images were acquired using: slice thickness 1.1 mm, number of slices 150, TR 25 msec, TE 4.6 msec, flip angle 30 , matrix 320 Â 320, FOV 240 mm, and voxel size .75 Â .75 Â 1.1 mm.
Table 1 e Language, cognitive and motor function assessment results at baseline (before surgery), in the acute post-surgical stage (4 days (D4) or 7 days (D7) after surgery) and during early recovery (3 months after surgery).
T1-weighted and FLAIR images allowed us to visualize the brain tumor and post-surgical edema/cavity location. DWI data allowed us to visualize 7 major language-related white matter tracts: the three segments of the arcuate fasciculus (AF; fronto-temporal, parieto-temporal and fronto-parietal connections), the inferior longitudinal fasciculus (ILF), the inferior fronto-occipital fasciculus (IFOF), the uncinate fasciculus (UF) and the frontal aslant tract (FAT). The tracts were visualized for both hemispheres and carried out according to the anatomical landmarks comprised in WM atlas (Catani and Thiebaut de Schotten, 2008), for more specific details please consult the following previous reports (Franç ois et al., 2016;Sierpowska et al., 2015Sierpowska et al., , 2017Sierpowska et al., , 2019.

Results
3.1. Brain tumor location and major language-related white matter organization The discrete brain lesion was located within the middle frontal gyrus (MFG) and subjacent white matter (see Fig. 1, left  panel). Preoperatively, diffusion tensor imaging (DTI) analyses revealed that the major white tracts belonging to the ventral (IFOF, ILF and UF) and dorsal (AF) language processing systems were well-preserved within the left hemisphere. However, the same tracts organization in the right hemisphere showed that the fronto-temporal connection of the AF was missing (see Fig. 1), which can be classified as the most extreme type of left c o r t e x 1 3 0 ( 2 0 2 0 ) 3 4 0 e3 5 0 lateralization of the AF (according to Catani, Jones, & ffytche, 2005). Also in the right hemisphere, the UF was only scarcely represented. The frontal aslant tract (FAT) was not found within the left hemisphere and was underrepresented on the right side.

3.2.
Neuropsychological and language assessment before surgery According to a fine-grained evaluation before surgery, TD scored within the normal range in naming, comprehension, short term memory, working memory and verbal fluency (see Table 1). She did not report any difficulties in sensory and motor functioning. Surgery itself was performed using brain mapping (sensory, motor, language naming and language switching) with precise intraoperative anatomical neuronavigation. During the awake surgical phase, TD could successfully communicate with the team and showed a good performance in the intraoperative tasks.

Bedside neuropsychological and language assessments
Unexpectedly, in the bedside assessment 2 days after surgery, TD presented severe difficulties at virtually all levels of spoken communication: production, repetition, and comprehension, meeting the diagnostic criteria of GA (Table 1). Her speech difficulties at motor level were representative of both apraxia of speech and dysarthria and TD's attempts to communicate required a huge effort and eventually lead to frustration, which was both observed by the neuropsychologist as well as expressed by the patient after recovery (see the transcript of her testimony at the end of this section). However, the patient did not present signs of hemiparesis (see Table 1 for motor function assessment scores and Fig. 2 for an original image of TD's handwriting). During both the hospital stay and the first days at home, TD tried to communicate with her spouse through drawings and gap-fill exercises designed by him (see Fig. 2). Both of these home-made tasks revealed severe errors in phonological processing, such as phonological (phonemic) and neologistic paraphasias in writing and languageswitching errors (see Fig. 2A). Structural neuroimaging revealed that in this acute, post-surgical phase, a portion of the left MFG was lesioned during tumor resection (see Fig. 1, middle panel). Additionally, a subdural extra-axial blood collection temporarily affected a considerable part of the MFG e adjacent pericavitarian areas (including white matter) and provoked a slight midline shift, which aggravated the patient's condition even after excision which was of a relatively small size. During the post-surgical recovery, TD was treated with corticosteroids (in line with our institution practice).
In the follow-up neuroimaging assessment (7 months after surgery), structural data confirmed the disappearance of the extra-axial blood collection and the midline shift (Fig. 1, lower  section, right panel). At the same time, TD successfully returned to a normal neuropsychological and language status (Table 1). Consequently, she reported on her subjective experience during the acute stage of GA. TD stated being aware that her family and medical team members were constantly trying to communicate with her but that she could not understand them, corroborating a severe disruption in language comprehension (see also Table 1, for her performance in the Token Task). She also reported that she was able to formulate thoughts and even to internally elaborate the messages she wanted to convey, but that she could not utter them. Importantly, TD was aware of her errors when she attempted to communicate, in her own words: When I was writing and it was incorrect, I knew it was incorrect and that was when I was getting angry and I was saying to myself: I don't want to write anymore or to do anything else (original transcript in Spanish: Cuando escribía y lo escribía mal, yo s e que lo escribía mal y entonces es cuando me daba rabia y decía: no quiero escribir m as ni nada ya). A score of 49 out of 52 in the Pyramids and Palm Trees Test 7 days post-surgery confirmed her ability to correctly match appropriate pictorial representations, and thus access the semantic representations. The use of these tasks was particularly useful, since, in this way, we could assess semantic processing bypassing the difficulties in auditory comprehension and overt speech production. During the same acute GA episode, she was also able to successfully complete non-verbal exercises, such as completing simple sequences of symbols that follow a certain rule (triangleecircleesquare, triangleecircleesquare Fig. 2B).

Discussion and clinical relevance
This uncommon case of acute global aphasia shows a great potential of language functional recovery after surgery compromising the integrity of the left MFG vicinity. Moreover, it clearly supports that verbal and non-verbal tasks may be selectively affected by focal brain lesions, explores the limits of covert versus overt language production and opens new, alluring questions about the role of IS in GA. Visual inspection of the presurgical structural neuroimaging data showed that the left lateralization of the AF was among the most prominent features of the patient's brain structural organization. Given that it is highly likely that the acute post-operative pericavitarian blood collection hindered the correct transfer of information with the left dorsal stream for language, we could assume that the right hemispheric analogs of the dorsal system were less aptly equipped to compensate for language impairments. In addition to the leftward AF lateralization, the frontal aslant tract (FAT) was scarcely present in both hemispheres. This could explain the motor component of TD speech difficulties, given the growing evidence on the involvement of the FAT in lexical selection, language planning, production and speech motor control (Chernoff et al., 2018;Dick, Garic, Graziano, & Tremblay, 2019;Halai, Woollams, & Lambon Ralph, 2017;Kemerdere et al., 2016;Kinoshita at al., 2015;Sierpowska et al., 2015). On the other hand, the pathological switching instances, as appearing in the gap-fill exercises during the acute post-surgical phase, may be explained by the impact of the pericavitarian blood collection on the left middle and inferior frontal areas. Pathological language switching was observed as an aftermath of a similar tumor resection involving this region, the left MFG (see Sierpowska et al., 2013). The present report nicely builds on the very recent study by Finkl et al. (2019) showing that core language-related white matter tracts did not differ between deaf signers and healthy participants, but differences were observed in speech-related processing and production white-matter pathways, as the left FAT and frontal premotor-parietal segment of the superior longitudinal fasciculus. Thus, these authors suggest that language processing may be to a certain extent independent of speech processing. However, one may find that our report challenges a recent proposal by Geva and Fernyhough (2019), who suggest that the dorsal language stream has a role in supporting the psychological phenomenon of IS. Their premise is based on a temporal overlap of the psychological phenomenon and neural development e IS emerges in early childhood, in parallel with the maturation of the dorsally located white matter fascicles. This discrepancy is especially important given that the authors claim that IS development relies mostly on the fronto-temporal and frontoparietal white matter connectivity e both of which were the most likely affected by TD's postsurgical blood collection.
Interestingly, however, the authors also state that it would be a mistake to assume that the emergence of IS is developmentally constrained by dorsal pathway maturation only, since the biological maturation and sociocultural experience may obey bidirectional causation. They also add that further research on the interplay between the ventral and the dorsal language streams should be undertaken. Importantly, in our patient, the impact of the lesion on comprehension difficulties is not easily explained neither by the extent of the lesion, nor by its location, challenging the prototypical neural underpinnings of global aphasia. However, frontal lobe damage could lead to WM memory problems affecting the auditory STM loop and consequently impacting normal comprehension (see Katz & Goodglass, 1990). Actually, when taking a closer look at the Token Test scores in the acute post-surgical state, it can be seen that TD was able to follow the commands from the initial section of the test, indicating the relationship between her difficulties and the length of the utterances (also confirmed by the fact that TD was able to repeat Corsi cubes sequences and follow up simple instructions up to the point when these surpassed a critical level of complexity). This observation suggests that in similar cases, it would be beneficial to carry out a more thorough exploration of working memory and include a naming test, where patients can point to the image of a named object (e.g., from the Cambridge Semantic Battery, Adlam, Patterson, Bozeat, & Hodges, 2010) to assess comprehension at a single word level. Another useful addition could be a task for comprehension of simple sentences controlled for syntactic complexity.
Additionally, with regards to the dorsal processing system, a recent report by Fama et al. (2017) conveyed that dorsal stream lesions (as is most probably the case in the present study) correlate with a high proportion of subjectivelyreported successful IS in patients with aphasia (see also Geva et al., 2011, for similar, voxel-lesion symptom mapping results). Interestingly, several evidences suggest that TD might have had preserved IS during the acute stage of GA. Her good semantic matching results (PPT and Environmental sounds test) and attempts of response in informal gap-fill exercises and formal testing support at least partial preservation of lexical access and/or utterances that she intended to produce despite her phonological errors in formal testing (i.e., in writing and in repetition). The preservation of word retrieval despite articulatory output impairments has been thought to underlie the preservation of IS in other aphasic patients (Fama et al., 2017). Additionally, according to TD's reported subjective experience, she was able to internally elaborate conceptual messages.
An interesting point in support of potentially preserved IS in TD may also be provided by her awareness of the phonological errors, which we find particularly puzzling. Observing this, one might be prone to consider the preservation of error monitoring in language production in TD as an indirect proof of the existence of accurate internal representations (e.g., "the expected word or lexical output") that are needed to detect the mismatch between the internal covert production (or writing output in this case) and the expected correct item. Most models in language production indeed assume the existence of mechanisms allowing for the self-monitoring of one's own speech production (Dell, 1985;Levelt, 1983Levelt, , 1989Motley, Baars, & Camden, 1983;Postma, 2000). However, it is unclear to what extent feedforward models of error monitoring are required for successful IS (Gallagher, 2004;Jones & Fernyhough, 2007;M€ oller, Jansma, Rodriguez-Fornells, & Mü nte, 2007; but see; Whitford et al., 2017). As an example, error awareness is possible without proper word retrieval, as it has been reported in tip-of-tongue language production states. Also relevant in this respect, Fernyhough (2004) suggested that IS can take different forms, varying from containing fully specified information (semantic, syntactic and phonological) to only semantics, which would imply that having an accurate phonological and/or articulatory representation is not a necessary condition for IS. Similarly, Grandchamp et al. (2019) recently proposed a neurocognitive model in which IS can vary along a condensation dimension, from condensed IS (without articulatory specification) to expanded IS (involving articulatory specification and the perception of a voice percept), and argue that expanded forms of IS entail verbal monitoring through internal feedforward models. Altogether, it remains unclear to which extent the presence of error monitoring in TD could be indirectly speaking in favor of preserved IS.
With regards to the motor aspects of the postoperative neurological profile of TD's e it is worth noting that the global aphasia did not occur together with hemiparesis, which adds to a handful of similar reported cases (see : Pai, Krishnan, Prashanth, & Rao, 2011, Shindo et al., 2013Paik & Kim, 2014). As an additional comment, it is worth considering that the errors in TD's handwriting (e.g., letter substitutions) might be attributed to the damage of the Exner's area (BA6). This region located in the "foot" of the MFG is considered to be "a graphic image motor center", where the abstract, orthographic representations are converted into motor representations (Exner, 1881). Along this line and in brain tumor population specifically, Roux et al. (2009) reported that direct electrical stimulation of BA6 resulted either in impaired handwriting or an incapacity of combining handwriting with other language tasks. However, we are rather skeptical with this interpretation e indeed, we reiterate that a variety of written language production errors including also language switching and neologisms suggest rather a general breakdown in phonological processing and/or control mechanisms. Finally, it is important to note that none of the errors in written production were of a visual nature (i.e., named incorrectly due to an ill-defined image), but it was difficult to discern if any of TD's errors was of orthographic nature.
Summing up, we consider that the evidence points to a partial preservation of language function during the GA episode and infer that some form of IS was involved. At the same time, we admit that inference based on subjective reports is not enough to unequivocally conclude in favor of preserved IS. We nevertheless consider that the present case study hints at the conception that overt language production tasks may not be predictive of the actual functional status of individuals with severe aphasia. There may still be a silent, covert reality that is usually left unassessed. We advocate for a fine-grained behavioral testing (for example, by using the Varieties of inner speech questionnaire, McCarthy-Jones & Fernyhough, 2011 and/or selected subtest of the Cambridge Semantic battery, Adlan et al., 2010) in order to understand the extent to which IS and other non-expressive language abilities are preserved in patients with GA.

Declaration of Competing Interest
The authors declare no competing interests.