Literacy, but not memory, is associated with hippocampal connectivity in illiterate adults

Background The influence of hippocampal connectivity on memory performance is well established in individuals with high educational attainment. However, the role of hippocampal connectivity in illiterate populations remains poorly understood. Methods Thirty-five illiterate adults were administered a literacy assessment (Test of Functional Health Literacy in Adults - TOFHLA), structural and resting state functional MRI and an episodic memory test (Free and Cued Selective Reminding Test). Illiteracy was defined as a TOFHLA score below 53. We evaluated the correlation between hippocampal connectivity at rest and both free recall and literacy scores. Results Participants were mostly female (57.1%) and Black (84.8%), with a median age of 50 years. The median TOFHLA literacy score was 28.0 [21.0;42.5] out of 100 points and the median free recall score was 30.0 [26.2;35] out of 48 points. The median gray matter volume of both the left and right hippocampi was 2.3 [2.1; 2.4] cm3. We observed a significant connectivity between both hippocampi and the precuneus and the ventral medial prefrontal cortex. Interestingly, the right hippocampal connectivity positively correlated with the literacy scores (β = 0.58, p = 0.008). There was no significant association between episodic memory and hippocampal connectivity. Neither memory nor literacy scores correlated with hippocampal gray matter volume. Conclusions Low literacy levels correlate with hippocampal connectivity in illiterate adults. The lack of association with memory scores might be associated with low brain reserve in illiterate adults.


Introduction
Life expectancy is increasing in low-and middle-income countries (LMIC), and consequently, the prevalence of dementia is rapidly rising. Between 2015 and 2050, the prevalence of dementia is expected to increase by 138% in countries like Brazil, compared to an increase of 56% in high-income countries (HIC) (Prince et al., 2015). Whereas most causes of dementia have no curative treatments so far, diseasemodifying drugs for Alzheimer's disease have a high cost and a controversial e cacy (Cummings et al., preventing dementia is a powerful strategy to mitigate the high burden of the disease on patients, caregivers, and society. Research shows that in LMIC 48% of dementia cases could be prevented if 12 modi able dementia-risk factors were controlled . Low educational level ranks highest among these factors. The prevalence and incidence of dementia in illiterate older adults is two and ve times higher than in literate adults (César-Freitas et al., 2022;Nitrini et al., 2009;Ribeiro et al., 2022), respectively. Up to 19% of dementia cases can be attributed to low educational attainment in HIC and up to 30% in LMIC, where low education is more prevalent (Mukadam et al., 2019;Norton et al., 2014). An increase in educational attainment in HIC is believed to have contributed to the recently observed decline in dementia incidence (Wu et al., 2017). In the Framingham study, the incidence of dementia is declining only among persons who have at least a high-school degree (Satizabal et al., 2016). Although other socioeconomic determinants of health associated with high education may play a role in the apparent protective trends, several studies support education as an independent factor leading to lower risk of dementia (Lu et  Previous research in low literate older adults (mean of four years of formal education) showed that episodic memory correlated with the integrity of white matter bundles that connect the hippocampus with the precuneus and with hippocampal volume (Resende et al., 2017). However, this correlation was signi cant only amongst the group with more than four years of formal education (Resende, Rosen, et al., 2018). The role of hippocampal connectivity on episodic memory performance in adults is still a matter of debate (Aggleton & Brown, 1999;Bhattacharyya, 2017;Eichenbaum, 2004;Sidhu et al., 2013;van Kesteren et al., 2010) and there are no studies in illiterate adults. The default mode network is important for memory processing (Staffaroni et al., 2018) and it is affected in patients with dementia of the amnestic type (Malotaux et al., 2022;Seeley et al., 2009;Zhou et al., 2010).
In the present study, we used structural and functional MRI to de ne whether there was an association between episodic memory and hippocampal volumes and functional connectivity in illiterates.
Understanding the brain mechanisms involved in episodic memory processing in illiterates can help unveil possible markers of successful interventions to improve memory in these populations, to mitigate the memory problems caused by neurodegenerative process that comes with aging.

Population
We used a community-based participatory research approach to collaborate with a basic-literacy training program for adults that is sponsored by the government. This program targets illiterate adults that did not have the opportunity to go to school when they were at the school age and want to learn how to read and write later in life. Adults aged 40 to 80 years-old that spontaneously enrolled in those late-life educational programs in the city of Belo Horizonte, Brazil, from February to July 2019 were invited to participate in the present research. Forty-three persons signed the informed consent and agreed to participate in the research. Sociodemographic and smoking habits were collected through a structured questionnaire. The level of physical activity was assessed with the Baecke scale (Baecke et al., 1982;Rocha et al., 1992).
Depression, anxiety and alcohol abuse were investigated by the Mini International Neuropsychiatric Interview (Sheehan et al., 1998). All evaluations were conducted upon entry in the late-life literacy program before any literacy training. The socioeconomic levels were determined using the ABEPE (Brazilian Association of Research Companies) framework that categorize households into different socioeconomic levels. This classi cation considers various factors such as income, education, and ownership of goods to determine the living standards of households. The level A category represents the highest socioeconomic level with high income levels, advanced education, and ownership multiple properties and luxury goods. The level B category includes households with a relatively high socioeconomic status, although slightly lower than those in level A. These households generally have good incomes, tertiary education, and own properties and durable goods. The level C category encompasses households with a middle socioeconomic status. They usually have moderate incomes, secondary education, and may own a house or apartment. The levels D and E represent households with a lower socioeconomic status that often have low incomes, limited education, and may live in rented accommodations or informal settlements. They may face signi cant economic challenges and lack of access to basic services. They often live in poverty, struggling to meet their basic needs and relying on government assistance programs.

Literacy and cognitive assessment
Participants that enroll in those late-life government sponsored programs have various degrees of reading and writing skills. Some never attended formal school while others attended for few years. Their reading abilities vary from inability to recognize letters to some reading capacity, without comprehending the meaning of the text. Therefore, we used the Test of Functional Health Literacy in Adults (TOFHLA) (Parker et al., 1995), validated for Brazilian Portuguese (Maragno et al., 2019), to evaluate the participant's literacy skills across different levels. Previous studies determined that a score equal or lower than 53 de nes illiteracy (Apolinario et al., 2015). Before any prepossessing of the images, all T1-weighted images were visually inspected for quality control. One image was excluded because of a large artifact. T1-weighted images undergone bias eld correction using N3 algorithm, the segmentation was performed using SPM12 uni ed segmentation (Ashburner & Friston, 2005). A customized group template was generated from the segmented gray and white matter tissues and cerebrospinal uid (CSF) by non-linear registration template generation using Large Deformation Diffeomorphic Metric Mapping framework . Native subjects' space gray and white matter were geometrically normalized to the group template, modulated, and then smoothed in the group template. The applied smoothing used a Gaussian kernel with 8 ~ mm full width half maximum. Every step of the transformation was carefully inspected from the native space to the group template. First, functional and anatomical data were preprocessed using a exible preprocessing pipeline (Nieto-Castanon, 2020) including realignment with correction of susceptibility distortion interactions, slice timing correction, outlier detection, direct segmentation and MNI-space normalization, smoothing, and band-pass ltering. Functional data were realigned using SPM realign & unwarp procedure (Andersson et al., 2001), where all scans were coregistered to a reference image ( rst scan of the rst session) using a least squares approach and a 6 parameter (rigid body) transformation (Friston et al., 1995), and resampled using b-spline interpolation to correct for motion and magnetic susceptibility interactions.
Temporal misalignment between different slices of the functional data (acquired in interleaved Siemens order) was corrected following SPM slice-timing correction procedure (Henson et al., 1999;Sladky et al., 2011), using sinc temporal interpolation to resample each slice BOLD timeseries to a common midacquisition time. Potential outlier scans were identi ed using ART (Whit eld-Gabrieli, 2009) as acquisitions with framewise displacement above 0.9 mm or global BOLD signal changes above 5 standard deviations (Power et al., 2014). A reference BOLD image was computed for each subject by averaging all scans excluding outliers. Functional and anatomical data were normalized into standard MNI space, segmented into grey matter, white matter, and CSF tissue classes, and resampled to 2 mm isotropic voxels following a direct normalization procedure (Calhoun et al., 2017) using SPM uni ed segmentation and normalization algorithm (Ashburner & Friston, 2005) with the default IXI-549 tissue probability map template. Functional data were smoothed using spatial convolution with a Gaussian kernel of 8 mm full width half maximum. Last, BOLD signal timeseries were bandpass ltered between 0.01 Hz and 0.1 Hz.
In addition, functional data were denoised using a standard denoising pipeline  including the regression of potential confounding effects characterized by white matter timeseries (5 CompCor noise components), CSF timeseries (5 CompCor noise components), motion parameters and their rst order derivatives (12 factors) , outlier scans (below 13 factors) (Power et al., 2014), session effects and their rst order derivatives (2 factors), and linear trends (2 factors) within each functional run, followed by bandpass frequency ltering of the BOLD timeseries (Hallquist et al., 2013) between 0.008 Hz and 0.09 Hz. CompCor stands for Component-based noise correction method (Behzadi et al., 2007) that computes the average BOLD signal as well as the largest principal components orthogonal to the BOLD average, motion parameters, and outlier scans within each subject's eroded segmentation masks. Those CompCor noise were estimated within the white matter and CSF.
Seed-based connectivity maps and ROI-to-ROI connectivity matrices were estimated characterizing the patterns of functional connectivity with 164 HPC-ICA networks (Whit eld-Gabrieli & Nieto-Castanon, 2012) and Harvard-Oxford atlas ROIs (Desikan et al., 2006). Functional connectivity strength was represented by Fisher-transformed bivariate correlation coe cients from a weighted general linear model (weighted-GLM (Nieto-Castanon, 2020)), de ned separately for each pair of seed and target areas, modeling the association between their BOLD signal timeseries. To compensate for possible transient magnetization effects at the beginning of each run, individual scans were weighted by a step function convolved with an SPM canonical hemodynamic response function and recti ed. The seed-based connectivity analyses were done placing a seed in each hippocampi using the Harvard-Oxford automated atlas (Desikan et al., 2006). The ROI-to-ROI connectivity matrices analyzed were the ones between each hippocampus and the ventral medial pre-frontal (VMPFC), each hippocampus (HC) and the Precuneus (PCC) and between the VMPFC and PCC. Finally, the group-level analyses were performed using a GLM. For each individual voxel a separate GLM was estimated, with rst-level connectivity measures at this voxel as dependent variables (one independent sample per subject), and groups as independent variables. Voxel-level hypotheses were evaluated using multivariate parametric statistics with random-effects across subjects and sample covariance estimation across multiple measurements. Inferences were performed at the level of individual clusters (groups of contiguous voxels). Cluster-level inferences were based on parametric statistics from Gaussian Random Field theory (Worsley et al., 1996). Results were thresholded using a combination of a cluster-forming p < 0.001 voxel-level threshold, and a familywise corrected p-FDR < 0.05 cluster-size threshold (Chumbley et al., 2010) Demeaned age was used as a covariate in all neuroimaging analyses.

Statistical analyses
Continuous variables were depicted in median and interquartile intervals; categorical variables were depicted in frequencies. GLM considering age, sex and total intracranial volume as covariates were used to calculate the correlation between episodic memory, literacy levels, brain connectivity and hippocampal volumes. In the rst model, the FCSRT free-recall sum of attempts was the dependent variable, and the predictors were the functional connectivity between each HC separately and the VMPFC, between each HC and precuneus, and between the VMPFC and PCC, as well as with each hippocampal volume. In the second model, the literacy level measured by the TOFHLA total score was the dependent variable and the predictors were the same depicted above.

Results
The nal sample had 35 participants. We excluded three participants that had claustrophobia and did not tolerate the brain MRI, one participant whose scan had artifacts that precluded the analysis, three that were left-handed and one that scored 98 in the TOFLHA and was, therefore considered literate. The median age was 50 years, 57.1% (n = 20) of participants were women and 84.8% (n = 28) were Blacks ( Table 1). The median TOFHLA score was 28 with an interquartile interval of 21.0 to 42.5. See the text for more details about the socioeconomic levels.
The seed-based connectivity analysis at rest showed a signi cant connectivity between both HC and the VMPFC and PCC, and other brain regions (Fig. 1). However, we failed to nd a signi cant association between the HC-VMPFC connectivity and episodic memory measured by the FCSRT free recall sum of attempts (Table 2). On the other hand, we found signi cant associations between the low TOFHLA scores and the HC-VMPFC connectivity (Table 3). Interestingly, the association was in opposite directions in each hippocampus. On the right side, the stronger the HC-VMPFC connectivity, the better the literacy scores (β = 0.58, p = 0.004), whilst on the left side, the stronger the connectivity, the worse the literacy scores (β=-0.39, p = 0.041). Age and sex did not signi cantly correlate with the association between HC connectivity and memory or literacy scores.

Discussion
In a group of middle-aged adults, the performance on a literacy test, even low enough to be considered illiterate per the literature (Apolinario et al., 2015), correlated with the HC-VMPFC connectivity. The association between low literacy levels and HC-VMPFC may suggest the role of even some literacy on cognitive reserve mechanisms. In contrast, we can speculate that the lack of association between episodic memory performance and hippocampal connectivity might re ect that this reserve is not enough to strengthen the role of hippocampal connectivity in memory abilities.
Cognitive reserve refers to distinct cognitive mechanisms, developed across the lifespan, that make a person more resilient or resistant to cognitive decline caused by brain damage (Stern et al., 2023). A higher level of cognitive reserve equips the brain to compensate through more e cient brain activation patterns that are more exible and resilient to neurodegeneration or other forms of brain injury (Stern et al., 2023). Because higher cognitive reserve is associated with more tolerance to hippocampal atrophy HC-VMPFC connectivity and eventually prevent cognitive impairment in this population. Our nding may substantiate the hypothesis that improved hippocampal e ciency, re ected in stronger connections between the hippocampus and critical areas for memory processing such as the prefrontal cortex, may impact cognitive reserve even with some schooling. However, because our study was cross-sectional, we cannot demonstrate causality.
The TOFHLA test has been widely used to measure literacy level (Fan et  neural correlates of literacy measured by literacy tests, and not years of education, is less studied. A previous study showed that higher literacy skills measured by the REALM-SF test correlated with brain structural connectivity, but not with hippocampal volumes (Resende et al., 2022). Interestingly, we found that the very low literacy levels measured by the TOFHLA in our sample was signi cantly associated with the HC-VMPFC connectivity. We speculated that this nding may re ect how even low levels of literacy can relate to brain functioning, shedding light on a possible mechanism of cognitive reserve in this illiterate population.
In terms of episodic memory and brain connectivity, there is still a debate in the literature. The FCSRT is a traditional episodic memory test that has two versions (verbal and visual). The neural correlates of the verbal version have been more explored, while the visual version was less studied. Because the participants were illiterate, the visual version of the FCSRT was more appropriate. The few studies that explored the neural basis of the visual FCSRT test were conducted in persons with high educational level.
One study with 14 participants compared the brain activation by the visual FCSRT between novel and repeated stimuli and showed that activations in left superior temporal and left prefrontal cortices were signi cantly associated with episodic memory (Diamond et al., 2007). Other brain areas activated through the FCSRT stimuli were the inferior parietal lobule, precuneus, hippocampus and In our study, the lack of association between episodic memory measured by the visual version of the FCSRT and the HC-VMPFC connectivity might be explained by the fact that we did not use task-based functional MRI as the previous studies used, but resting state functional MRI, which might be less sensitive to cognitive-brain correlations (Rasero et al., 2018). Another possibility is that illiterates use less their HC-VMPFC connectivity for memory processing, which might suggest a low cognitive reserve in this group. The fact that we found a signi cant relationship between literacy levels and the HC-VMPFC connectivity may support this theory, because, as the literacy levels increase, the association becomes Our next goal is to explore the effects of adult-literacy training in brain structural and functional connectivity as well as in cognitive abilities, to determine whether adult-literacy acquisition might have a bene cial effect on dementia prevention. Eventually, we will be able to inform public policies to increase educational attainment in adulthood with a substantial impact on lowering dementia burden worldwide. representatives. The purpose, procedures, potential risks, and bene ts of the study were clearly explained, ensuring that participants understood their rights and had the opportunity to ask questions. All personal information and data collected from participants were treated with con dentiality. Identifying information was anonymized and stored securely, limiting access to authorized researchers only. Any data presented in the manuscript has been de-identi ed to ensure the privacy and con dentiality of participants. The study protocol was reviewed and approved by the relevant institutional or independent ethics committee, ensuring that it complied with ethical guidelines and safeguarded the welfare and rights of participants.

Competing interests
The authors declare that there are no con icts of interest that could have in uenced the design, conduct, or reporting of the study. Financial or personal relationships that could potentially bias the research ndings were disclosed and managed appropriately.
Author's Contributions EPFR: Conceived the study, contributed to the study design, supervised data collection and analysis, interpreted the results, and wrote the manuscript, VLP, ALCS: Participated in data collection and contributed to data analysis, CVF: Contributed to the study design, and assisted in data collection, HJR: Contributed to the study design, and critically revised the manuscript, JAB: Assisted in neuroimaging data analysis, contributed to the interpretation of results, and provided critical feedback on the manuscript, YC: Assisted in neuroimaging data preprocessing, analysis and interpretation. LLG: Assisted with neuroimaging acquisition and preprocessing the data, LCS: Assisted in data interpretation and contributed to the critical discussion of results, and reviewed the manuscript, LR: Assisted in data collection and interpretation, LTG: contributed to the study design, supervised data collection and analysis, interpreted the results, and critically revised the manuscript, FIPM: Assisted in data collection, contributed to the literature review, and provided critical revisions to the manuscript, PC: Conceived and designed the study, provided overall supervision, contributed to data interpretation, and critically revised the manuscript.
All authors have read and approved the nal version of the manuscript.

Availability of data and materials
Page 16/24 The data that supports the ndings are available upon reasonable request. Aggregated and anonymized data, as well as additional information related to the study methodology, can be made available to interested researchers. Requests for data access should be addressed to the corresponding author, Dr. Elisa de Paula França Resende (email: elisaresende@gbhi.org), who will assess each request on a caseby-case basis in consultation with the research team and in compliance with applicable data protection regulations and institutional policies.

Figure 1
Correlation between hippocampal connectivity and low literacy levels.
The statistical map is displayed on an in ated brain image. The heat maps represent the T statistical value for the connectivity between the right and left hippocampal seed and the other clusters. Blue means anticorrelation and red means positive correlation. The graph depicts the correlation between literacy levels measured by the Test of Functional Health Literacy Assessment (TOFHLA) and the right HC-VMPFC connectivity.