To Go or Not to Go: Exploring brain activation during response inhibition reading tasks

Objective: Response inhibition is an understudied component of reading that aids in the selection of appropriate responses amidst complicated tasks. Our objective was to explore the contribution of brain regions associated with response inhibition processing in reading tasks that vary in difficulty of response inhibition. 
Method: Participants (N = 15) completed two go/no-go reading tasks while in a functional magnetic resonance imaging (fMRI) scanner, with the instructions to “name aloud the letter strings that spell a real word.” For the minimal response inhibition condition, the foils, which are stimuli that should not be repsonded to, were nonwords with unfamiliar spelling and sound (e.g., “bink”). For the maximal response inhibition condition, the foils were pseudohomophones with unfamiliar spelling but familiar sound (e.g., “pynt”). The following brain regions associated with decision-making processes were analyzed: the anterior cingulate cortex (ACC), the dorsomedial prefrontal cortex (DMPFC), the inferior frontal gyrus (IFG), the middle temporal gyrus (MTG), the middle occipital gyrus (MOG), and the posterior insula (PI). 
Results: Significant differences in activation within the nonword task were found for the DMPFC and the PI (the ACC approached significance). Significant differences in activation within the pseudohomophone task were found for the ACC, the MTG, and the PI. The IFG was found to be greatly activated for all words that had familiar phonemes (sounds). The MOG was found to be activated across all tasks. 
Conclusion: We provide evidence for differential response inhibition processing in the decision-making network during reading tasks. This work is a necessary step in better understanding response inhibition ability for individuals with and without reading impairments.


Introduction
Decision-making can be divided into two categories: autonomic processes and executive processes (Kahneman, 2003). Autonomic processes are those that are innate and reflexive, whereas executive processes are those that are effortful and conscious. A subdivision of executive processes is response inhibition (Stevens et al., 2015), which plays an important role in various reading tasks (e.g., go/no-go; Cummine, Aalto, Ostevik, Cheema, & Hodgetts, 2018). For example, the pronunciation of exception words like "yacht" requires the inhibition of usual language processes that would cause you to pronounce the word /jaet∫t/ (to rhyme with "patched"), and instead requires retrieval from stored internal vocabulary for the correct pronunciation, /jαt/ (to rhyme with "caught"). The objective of this project was to explore the extent to which brain regions associated with decision-making processes are activated during reading tasks varying in response inhibition difficulty. It is hypothesized that brain regions involved in decision-making will be differentially activated during reading tasks varying in response inhibition difficulty.

Response Inhibition
Response inhibition is the ability to suppress a preponent (i.e., natural or habitual) behaviour or action (Li, Huang, Constable, Sinha, 2006). Response inhibition allows for the selection of appropriate responses amidst complicated situations and/or foils. One of the more commonly-used approaches to assess response inhibition is through the use of go/no-go or stop-signal tasks (Li et al., 2006).
A participant may be presented with multiple go signals that are then followed by intermittent nogo (stop) signals which measure a participant's ability to quickly inhibit a task. Interest in investigating response inhibition stems from its association with many neurological conditions.
For example, impaired response inhibition has been found in individuals with attention deficit hyperactivity disorder (ADHD; Aron & Poldrack, 2005) and has also been used extensively in the reading literature (e.g., regularity effects-see section 1.3;Cummine et al., 2018;. In the imaging space, several brain regions have been implicated as playing a marked role in response inhibition. For example, the anterior cingulate cortex has been reported to be involved in response selection (Botvinick, 2007) and the dorsomedial prefrontal cortex involved in error detection (Holroyd, Yeung, Coles, & Cohen, 2005;Modirrousta & Fellows, 2008) (see Figure 1). The extent to which specific inhibition effects are evident at the neural level during reading is not well understood.

Reading
The dual route model of reading proposes that there are two pathways that work together to aid in reading: the sublexical and lexical pathways (Coltheart, Rastle, Perry, Langdon, & Ziegler, 2001) 1 .
The lexical pathway is primarily used for the reading of familiar words; it requires retrieval from an individual's stored internal vocabulary, that is, their lexicon. For example, exception words (e.g., "yacht") and regular words (e.g., "hat") rely on the lexical pathway to be read correctly. June 2019 Figure 1. Visual representation of brain regions involved in decision-making processes and reading.

Response Inhibition and Reading
The role of response inhibition has been studied in different capacities. One example is the regularity effect, which is the finding that words with typical spelling-to-sound correspondence (i.e., regular words) are read aloud more quickly than words with an atypical spelling-to-sound correspondence (i.e., exception words) (Hino & Lupker 1998, 2000Cummine, Amyotte, Pancheshen, & Chouinard, 2011;Cummine et al., 2013). This finding has been taken as evidence for the additional response inhibition that is needed for exception words.
Exception words are stimuli that produce two competing responses (e.g., /jaet∫t/ and /jαt/ for the word /yacht/). An additional example of the ways in which response inhibition has been studied is the modulation of task difficulty. Researchers have explored how response selection and inhibition change as a task goes from relatively easy (i.e., reliance on the lexical pathway-press "1" if the word spells a real word) to relatively hard (i.e., reliance on both lexical and sublexical pathways-press "1" if the word sounds like a real word). However, the extent to which such response inhibition processes are evident from brain region activation, when an individual is reading, has not been well established.

Summary
Response inhibition is an executive process that is commonly studied through go/no-go stop signal tasks (refer to 1.1). Response inhibition has been shown to be important for reading, yet it is still not well understood at the level of the brain.
Furthermore, the exact brain regions responsible for response inhibition reading tasks are not clear (refer to 1.3). This investigation will compare brain activation across low complexity to high complexity (refer to 1.2) response inhibition reading tasks.
This will contribute towards establishing a more comprehensive understanding of the brain regions required for response inhibition reading tasks.

Materials
The stimuli consisted of 200 high and low frequency real words (e.g., regular and exception words; see Appendix 1). Stimuli that should not be responded to, also called foils, included 50 nonwords (e.g., "norve") and 50 pseudohomophones (e.g., "whyle"). Nonword foils were created by changing one or two letters of the real words and the pseudohomophones were compiled from current literature (e.g., Cummine et al., 2011). All stimuli were matched for onset phoneme (initial word sound), length, bigram sum (frequency of two adjacent letters), frequency (in the case of the real words), phonological neighborhood (sets of words that differ by a single sound), and orthographic neighborhood (words of the same length that differ by only one letter) (see Balota et al., 2007). The words were presented in two different lists: mixed with non-words (words mixed in with nonword foils; total=150), and mixed with pseudohomophones (words mixed in with pseudohomophone foils; total=150). Participants

Methodology
were instructed to read aloud only those words that spelled a real word, and to remain silent when they saw a nonword or pseudohomophone.
In the minimal response inhibition condition, the foils were nonwords (unfamiliar spelling and sound, e.g., "bink"). These foils represented a less complex response inhibition task because accuracy only required recognition of familiar words (i.e., real words; refer to section 1.2) and no response for unfamiliar words (i.e., nonwords). In the maximal response inhibition condition, the foils were pseudohomophones (unfamiliar spelling with familiar sound, e.g., "pynt"). This foil represents a more complex response inhibition task, because it requires differentiating words with familiar spelling and sound from pseudohomophones, which only have familiar sound. In order to be accurate for this foil, participants must decode the PHP (refer to 1.2) and then inhibit a response because PHPs have incorrect spelling.

Procedure
Participants came to the neuroimaging centre, which is located on the University of Alberta campus, where the Magnetic Resonance Imaging (MRI) technician ensured they could safely take part in the study (i.e., no contraindications to go in the MR scanner).
Prior to going into the MRI scanner, participants were provided with information about the nature of the tasks they would be completing. Stimuli were presented using a data projector connected to the computer running E-Prime software (Psychology Software Tools, Inc., http://www.pstnet.com). For each condition an event-related design was used.

Behavioural Responses
Verbal responses of the participants were analyzed using Audacity (http://audacity.sourceforge. net/), a free software used to manipulate audio files. Noise removal algorithms were implemented to reduce the ambient noise from the magnet.
Response time (in milliseconds) was measured as the difference between a visual word onset and the vocal response onset. Correct responses were averaged across trials for each condition and participant. Data were then entered into SPSS for a repeated measures analysis of variance.

Brain Activation
Step 1. Preprocessing (individual subject level): The first five image volumes were used to achieve a steady state of image contrast and were discarded prior to analysis. The remaining volumes were classified as activity during the task or activity during rest and were subject to standard pre- In-scanner (fMRI) response inhibition task procedure. Participants were presented with fixation crosses in between stimulus presentation. There were two different tasks, the nonword task and the pseudohomophone task. The nonword task consisted of presenting real word letter strings (e.g., "plant") and nonword letter strings (e.g., "bink"). The pseudohomophone task consisted of presenting real word letter strings (e.g., "plant") and pseudohomophone letter strings (e.g., "pynt"). Stimulus presentation was randomized in each task; refer to section 2.2 for more details on task procedure.
Spectrum | InterdIScIplInary undergraduate reSearch Step 3. Second-level analysis (group level): Second level analysis included averaging data of all participants to create a mean activation map for each condition. Using a one-sample t-test, mean activation maps were significant at t (14) = 3.179, p < 0.05 at the group level.   Figure 4).

Dorsomedial prefrontal cortex (DMPFC):
A significant difference in mean percent activation for NWs compared to words (in the NW foil) (p=0.034) was found (refer to Table 2; Figure 4).

Middle Temporal Gyrus (MTG):
A significant difference in mean percent activation for PHPs compared to words (in the PHP foil) (p=0.013) was found (refer to Table 3; Figure 4).

Inferior frontal gyrus (IFG):
No significant differences were found between the stimuli. High mean percent activation was found for the words (in the NW foil), the PHPs, and the words (in the PHP foil). In contrast, low mean Spectrum | InterdIScIplInary undergraduate reSearch  The bars represent mean percent activation for different types of words: nonwords (dots); pseudohomophones (solid); words within the nonword task (checkered); words within the pseudohomophone task (striped). The black boxes indicate significant differences in activation from performed t-tests (refer to Table 3 & 4). The brackets indicate the function of that brain region in reading tasks (e.g., the ACC is important for common decision-making processes). Outliers >2.5 std. dev. were assessed and removed on a region by region basis (from each of the IFG, MOFC, PI, and ACC).
The purpose of the present work was to explore the extent to which brain regions associated with decision-making processes are differentially active during reading tasks that vary in difficulty of response inhibition. Here, we found evidence for percent activation was found in the NWs (Figure 4).

Middle Occipital Gyrus (MOG):
No significant differences were found between the stimuli. High mean percent activation was found across all tasks; NWs, the words (in the NW foil), the PHPs, and the words (in the PHP foil; Figure 4).

Posterior insula (PI):
A significant difference in mean percent activation for NWs compared to words (in the NW foil) (p < 0.001) and PHPs compared to words (in the PHP foil; p=0.011) was found (refer to Table 2 & 3; Figure 4). differential activation in brain regions associated with decision-making processes (ACC), error detection (DMPFC) and semantic processing (MTG).
Equally valuable, we provide additional information on brain regions that were not sensitive to complex decision-making per se, but instead reflect differences in more general language function.
These findings are discussed in further detail below.

Task Complexity
As previously outlined, assessing the spelling of real words relies on the retrieval of information that is already known (lexical processes; refer to 1.2) and relatively quick to access. In the NW task, rejection of NWs is also relatively straightforward as the decoding process does not lead to any familiar print (i.e. written words) or sound information. In contrast, assessing the spelling of PHPs produces unfamiliar print but familiar sound information, making this task more complex than the NW task. Our behavioural Spectrum | InterdIScIplInary undergraduate reSearch 9 doi: PUBLISHED: Published:

10.29173/spectrum52
June 2019 data supports the notion that the PHP task is indeed more complex, as the PHP task led to significantly longer reaction times in participants than the NW task. Ultimately, individuals needed to resolve the conflicting no-go (print information) and go (sound information) information to complete the task successfully, and this required additional time.

Common decision-making processes: Anterior
Cingulate Cortex (ACC) The ACC regulates decision-making, specifically in situations that require response override (Botvinick, 2007). An example of response override is incongruent trials of the stroop task-a common task in psychology where one must override saying the colour a letter string spells and instead say the colour of the letter string. The difference in activation of the ACC for nonwords compared to words in the nonword foil approached significance and was significant for PHPs compared to words in the PHP foil. Specifically, the NWs and PHPs showed a smaller mean percent activation than the words. Recall that the task required responses for NWs and PHPs to be withheld and responses for words to be said aloud. Therefore, this finding indicates that activation of the ACC was greater when overt response was required for the task.
This supports the idea that the ACC was involved in deciding which stimuli to inhibit a response for and which stimuli to allow a response for, affirming the ACC's role in response inhibition decisions.
Greater negative activation (i.e., greater inhibition) of the DMPFC for NWs compared to words may indicate that NWs are detected as errors in this condition (i.e., they were unfamiliar; when an intermittent nonword was presented amidst the real words, the unfamiliar word was detected as an error). In contrast, the lack of difference in activation between PHPs and real words may indicate that the familiar sounds associated with PHPs (i.e., they sound like a real word) were not detected as an error. Thus, it is likely that two different strategies were employed for the NW task versus the PHP task. The less complex NW task used automatic evaluation of words encoded in memory to determine if a word was familiar or not. If the letter string was familiar, a response was permitted, whereas, if the letter string was unfamiliar, the response was inhibited. In contrast, the more complex PHP task required evaluation of the stimulus via grapheme to phoneme decoding, in order to recognize that the PHP was not a real word. As such, it was not automatically coded as an error. In line with this notion, the PHP task had less negative activation (a smaller inhibitory response), indicating it was not automatically coded as an error.

Semantic: Middle Temporal Gyrus (MTG)
Researchers have shown the MTG to be implicated in semantic control (Davey et al., 2016). Spelling processing (Burton, 2001) and storing familiar speech sounds (Guenther, 2006 (Vorobyev et al., 2004). Visual word form processing includes the identification of shapes, letters, and words prior to, or in parallel to, identification of sound and/or meaning (Dehaene & Cohen, 2011). Visual word form processing is critical for reading as evidenced by the high mean percentage activation of the MOG across each task and for each stimulus type.
Posterior Insula (PI): While previous work has implicated the anterior insula in speech and articulatory production (Oh, Duerden, & Pang, 2014), The PHP task took participants significantly longer to complete than the NW task, supporting the notion that the PHP task was more complex.