## Participants

Thirty right-handed undergraduate students (15 males and 15 females; mean age = 21.92 years, range: 18–25 years) were recruited to participate in this study from Beijing Normal University, China. They majored in a wide range of disciplines except for mathematics. These participants reported having no history of neurological disorders or head injuries. The experiment was fully explained to the participants before informed consent was obtained. This study was approved by the Institutional Review Board of the Institute of Cognitive Neuroscience and Learning at Beijing Normal University.

## Materials

The algebraic task included the operational principles for addition, subtraction, multiplication, and division acquired by students in primary school. The arithmetic task included the numerical calculations for addition, subtraction, multiplication, and division. Each problem included two expressions, and the average length of the problems was controlled (Fig. 1). All expressions were presented in white against a black background (with a red, green, and blue value of 0, 0, 0). The participants were given 16 algebra and 16 arithmetic problems to complete.

The stimulus presentation and behavioral data recordings were programmed using E-prime software (Version 1.1, Psychology Software Tools, Inc., www.pstnet.com) on a Pentium 4 laptop. Stimuli were projected onto a translucent screen placed at the back of the magnet bore. Participants viewed the screen through a mirror mounted on the head coil at a distance of 30 cm from the eyes.

## Procedure

Prior to scanning, the participants underwent a training session with the same type of materials as a formal experiment to ensure that they understood the instructions of this experiment. After that, participants were required to complete the experimental tasks in the scanner.

The scanning session was organized into two runs, each lasting 192 s. Each run consisted of four experimental blocks (two experimental blocks for each task) and four baseline blocks (Fig. 1). The balanced Latin square design [42] was used to counterbalance the order effect of the two experimental tasks. Each block with four trials lasted for 24 s, and the presentation order of trials was random. There was a 1-min rest period after each run.

For each trial in the experimental blocks, two expressions were presented synchronously on the screen, one at the top and one at the bottom. Participants were asked to judge whether the two expressions were equal. For each trial in the baseline blocks, two arrows were presented synchronously on the screen, one at the top and one at the bottom. Participants were asked to judge whether the two arrows pointed in the same direction. Each trial lasted 6 s, and a fixation cross was displayed to fill the remaining time if the participant did not use all 6 s. Half of the participants responded to the trials by pressing a key on a response box on their left with their left index finger when the two mathematical expressions or arrows were equal. The remaining participants responded to the trials by pressing a key on a response box on their right with their right index finger. Both accuracy and speed were emphasized.

## fMRI data acquisition

Images were obtained using a Siemens (Munich, Germany) 3T Trio MRI scanner using a standard eight-channel head coil. After automatic shimming of the magnetic field, three-dimensional high-resolution T1 anatomical images were acquired for co-registration with the functional images. Next, functional volumes were acquired using a multiple slice T2*-weighted echo planar imaging (EPI) sequence with the following parameters: in-plane resolution = 3.125 × 3.125 mm2; repetition time = 2000 ms; echo time = 30 ms; flip angle = 90°; field of view = 240 × 240 mm2; matrix dimensions = 64 × 64; field of view = 200 mm; and slice thickness = 4 mm. The entire brain was imaged in 32 slices.

## Statistical analysis of the fMRI data

Individual MRI datasets were analyzed using SPM12 software (Wellcome Department of Imaging Neurosciences, University College London, UK; http://www.fil.ion.ucl.ac.uk/spm). All volumes were realigned to the first volume and spatially normalized to a common value to correct for whole brain differences over time. Images were then smoothed using an isotropic Gaussian kernel of 4 mm and high-pass filtered at a cut-off of 128 s.

*Brain activation analysis*. After preprocessing, parameter-estimated images for individual participants across the whole brain were calculated. Group analyses with random effects were conducted using the one-way analysis of variance (ANOVA) on the brain activation maps of all participants with material type as the independent variable. Then, the brain activation for each type of material relative to fixation was calculated. Brain activations for the two types of materials were compared. We used the thresholds from the lenient *p* < .05 to *p* < .005, to the stringent *p* < .001, uncorrected, with a minimum cluster size of 20 voxels.

*Brain-behavior correlation analysis.* A single-trial (item-wise) interindividual correlation analysis that has been reported in previous fMRI and event-related potential (ERP) studies was used to examine the brain-behavior correlations (Li et al., 2020; Zhou et al., 2018). First, the correlation between the brain activation maps and the reaction times (RTs) was determined for each trial. Then, a one-sample *t*-test on the correlation coefficients obtained for all trials of one type of processing against zero was performed. The analysis was performed separately for algebra and arithmetic tasks. The single-trial correlation has been reported as more effective than the traditional mean-trial correlation as it can filter out much of the noise that exists after the first step [41].

The traditional mean-trial interindividual correlation was also used, and the results were compared with those of the single-trial correlation. The mean brain activation map and the mean RT for each type of processing for each participant were determined. Then, a correlation analysis between the mean brain activation map and the RT for all participants was performed. We used the thresholds for two brain-behavior correlation analyses from the lenient* p* < .05 to *p* < .005, to the stringent *p *< .001, uncorrected, with a minimum cluster size of 20 voxels.

Further, we defined seven functional regions of interest (ROIs) based on results from previous studies for the semantic, phonological, and visuospatial networks. Based on the meta-analysis of 120 functional neuroimaging studies of semantic processing (Binder et al., 2009), we defined four ROIs of the semantic network in the left hemisphere that included the middle temporal gyrus (Montreal Neurological Institute (MNI) coordinates [−45, −21, −15]), inferior frontal gyrus (MNI coordinates [−51, 24, 0]), dorsomedial prefrontal cortex (MNI coordinates [−6, −48, 30]), and angular gyrus (MNI coordinates [−57, −45, 30]). Based on the fMRI studies of phonological processing for arithmetic (Zhou et al., 2007), we defined two ROIs of the phonological network in the left hemisphere that included the precentral gyrus (MNI coordinates [−45, 4, 30]) and supplementary motor area (MNI coordinates [−3, 3, 58]). Based on the fMRI studies of visuospatial processing for arithmetic [43], we defined one ROI of the visuospatial network in the left superior parietal lobule (MNI coordinates [−32, −68, 56]). Each ROI was a sphere with a radius of 9 mm. These ROIs were used to compare the levels of brain-behavior correlations between algebra and arithmetic.