Presurgical Microstructural Coherence Predicts Cognitive Change for Bariatric Surgery Patients

Abstract

Objective

This observational study examined the relationship between presurgical white matter microstructural coherence and cognitive change after weight loss. We hypothesized that higher baseline fractional anisotropy (FA) would predict greater baseline and change cognition.

Methods

A sample of 24 adults (BMI ≥ 35 kg/m²) underwent neuropsychological assessment at baseline and twelve weeks post bariatric surgery. A brain MRI was administered at baseline and processed through Tract-Based Spatial Statistics to compute FA in white matter tracts of interest (TOIs). Composite scores for attention, learning, processing speed, executive function, verbal fluency, working memory, and overall cognition were calculated.

Results

As expected, FA in some TOIs was significantly (p < .05) positively associated with change in cognition. Inverse relationships were observed between baseline FA and presurgical cognition, which may be explained by increased medial and radial diffusivity and preserved axonal diffusivity. Cognition generally improved after surgery; however, relative yet clinically non-significant deterioration was observed on learning measures. Poorer baseline cognitive performance was associated with greater post-surgical cognitive improvement.

Conclusions

Presurgical microstructural coherence is associated with magnitude of cognitive change after weight loss. An observed reduction in learning suggests that bariatric surgery may lead to negative outcomes in some cognitive domains, at least temporarily.

Introduction

Obesity impacts over 650 million people and the rate of obesity has grown nearly three-fold since 1975 (1). Associations between obesity and cardiac and metabolic disorders, including diabetes (2), hyperlipidemia, and hypertension (3), are well-documented (4). Recent research has focused on the relationship between weight and cognition in conjunction with these comorbidities, revealing a link to impaired performance on tasks of episodic memory (5), verbal learning (6, 7), concept formation and set shifting (8, 9), psychomotor skills (10) and selective attention (11). Further research has chronicled the relationship between age, obesity, and cognition (12). Among young adults with obesity, studies have highlighted compromised working memory (13). Additionally, midlife obesity may be more cognitively detrimental than late-life obesity due to accelerated brain aging (14).

Obesity-related cognitive decline may be partly explained by low-grade, chronic inflammation, which can cause structural damage within the body and brain and engender metabolic disorders, exacerbating existing issues (12, 14, 15). A circular relationship has been proposed, in which genetic predisposition for obesity and lower executive function (EF) predict weight gain, which further impairs cognition via brain inflammation, hyperlipidemia, and insulin resistance (16). Additional research implicates the role of comorbidities. Among a sample of Swedes aged 35 - 90, Nilsson and Nilsson (2009) observed an interaction between overweight and age for episodic memory, but this effect was eliminated when controlling for hypertension, stroke, and diabetes (17).

Recent evidence suggests baseline cognition is associated with weight loss success. In participants with obesity, initial cognition predicts twelve-month weight loss (18). Individuals with superior cognition, particularly EF, demonstrate better decision-making and impulse control, and may experience greater success with diet and exercise. Another line of research has examined effects of weight loss on cognition. Weight loss may improve cognition among elderly individuals with obesity (19). In a metanalysis on intentional weight loss, improvements were noted in memory, attention, and EF among adults with obesity, but not overweight (20). Among bariatric surgery candidates, enhanced memory was observed twelve weeks post-surgery, while non-surgical controls exhibited memory decline (21). Initial cognitive ability may predict subsequent weight loss, leading to a decrease in cognitively detrimental comorbidities and/or neuroinflammation, further improving cognition.

In this study, we utilize diffusion tensor imaging (DTI) to examine microstructural coherence in participants with obesity. In line with previous research, we report fractional anisotropy (FA), which measures uniformity of diffusion along the axon. Generally, higher FA indicates healthier white matter and superior cognition. Obesity is independently associated with lower FA in several tracts throughout the brain (22) and this relationship is mediated by white matter damage (22). Reduced FA appears to be specifically associated with obesity, rather than overweight. Stanek et al. found no difference between participants with and without overweight, but lower FA was observed in the those with obesity (23). This relationship is further complicated by reductions in FA due to normal brain aging. Several studies have reported inverse relationships between BMI and FA in the corpus callosum (CC), fornix, and corona radiata across ages, with significant interactions between BMI and age (24). Conversely, some researchers have observed positive associations between obesity and FA. Birdsill et al. reported abdominal obesity was associated with higher FA, independent of comorbidities (25). Preliminary evidence suggests weight loss may lead to improvements in microstructural coherence – Hu et al. found that structural connectivity of the dorsolateral prefrontal cortex to anterior cingulate gyrus increased in gastric sleeve patients one month and six months post-surgery (26). Further, Wang et al. found increased structural connectivity in sleeve gastrectomy patients between the hypothalamus, bilateral superior frontal gyri, left amygdala, and orbitofrontal cortex at 12 months post-surgery (27). In a meta-analysis of microstructural coherence differences in obesity, Daoust et al. reported increases in FA from physical activity-induced weight loss (28).

The current study seeks to define the relationship between baseline microstructural coherence and cognitive change after weight loss. We recruited bariatric surgery candidates and collected brain MRI and neuropsychological assessments at baseline and follow-up. Several tracts of interest (TOIs*) were selected a priori based on initial literature search and prior experience indicating the tracts most vulnerable to adverse effects of metabolic, vascular, and other physiological consequences of obesity and diabetes, as well as involvement in connecting functional anatomic regions associated with cognition. Greater microstructural coherence reflects greater structural integrity of the brain and should provide a better foundation for subsequent cognitive improvement following neurophysiological alterations from weight loss. In effect, intact brain structure is necessary for optimal cognitive performance, such that reduced baseline microstructural coherence may limit baseline cognitive capacity and potential for cognitive improvement.

Primary Hypothesis:

Higher baseline FA predicts improvement in cognition after weight loss.

Secondary Hypothesis:

Higher baseline FA predicts greater baseline cognitive performance.

Further analyses characterized the relationship between baseline and change cognition after weight loss. Exploratory analyses assessed associations between related DTI metrics and cognition.

Methods

Study Design

Bariatric surgery candidates were recruited at UF Health Bariatric Surgery Center during nutrition classes for an NIDDK-funded study that aims to characterize longitudinal effects of weight loss on cognition and brain health. Recruitment was ongoing from 2014 to 2021. G*Power was used to compute a desired sample size of 54 for regression analyses to detect a medium effect size of .35, representative of research in this area, with 15 predictors at α < 0.05. Due to difficulties associated with the COVID-19 pandemic, participant drop-off, and data quality concerns, the achieved sample size was 55 for baseline and 24 for follow-up. Although the current study appears underpowered based on these calculations, we argue that these results still contribute to the literature and are adequately powered when considering the longitudinal nature of the study. Ethics approval was obtained from the University of Florida Gainesville Health Science Center Institutional Review Board (IRB-01) and study procedures were conducted in accordance with IRB-01 ethical guidelines. Participants were monitored to ensure safety. The current observational study did not seek to confirm efficacy or efficiency of bariatric surgery. Participants completed a medical history, cognitive battery, and MRI brain scan at pre-surgical baseline and 12-months post-surgery. Participants were pre-screened prior to enrollment to confirm eligibility. Inclusion criteria included age between 20-75, BMI ≥ 35 kg/m², MRI compatibility, and no history of neurological disease. Participants were excluded at baseline for demonstration of mild cognitive impairment (MoCA total score < 18). Participants were compensated for participation.

Participants

Data were collected and analyzed for 55 participants (47 female) by trained research staff. Our sample included 36 White, 13 Black, 3 Latino, and 3 who identified as other. Five participants identified as Hispanic/Latino and 49 as not Hispanic/Latino (ethnicity data was missing for one participant). Average age was 42.76 years and average years of education was 13.89. Average BMI was 47.01 kg/m². Demographic and clinical characteristics are expanded upon in Table 1.

Table 1.

Demographic and clinical characteristics of the study sample

Characteristic
Mean Age (yrs.) (+/− SD)	42.76 ± 12.26
Mean Education (yrs.) (+/− SD)	13.89 ± 2.69
Sex (% women)	85.5
Race (% Caucasian)	65.5
Clinical Indices
Mean BMI (kg/m2) (+/− SD)	47.01 ± 8.26
Class Obesity (%)
Class II	20
Class III	80
Type-2 diabetes (%)	41.8
Mean HbA1C level (+/− SD)	6.11 ± 1.57
HbA1C (%)
Normal	53.8
Pre-diabetic	23.1
Diabetic	23.1
CPAP use (%)	26.4
Hypertension (%)	54.5
Hyperlipidemia (%)	29.1
CAD (%)	1.8
Hepatitis A (%)	1.8
Kidney Problems (%)	3.6
Liver Problems (%)	10.9
Hypothyroidism (%)	13
Surgery Type
Gastric Bypass/Roux en Y	13
Gastric Sleeve	12

Note: BMI = body mass index, HbA1C = hemoglobin A1c, CPAP = continuous positive airway pressure, CAD = coronary artery disease.

Measures

Demographic and Clinical

Participants self-reported weight, race, ethnicity, age, and years of education. Medical history assessed metabolic, cardiovascular, neurological, psychological, and physical conditions. Neuropsychological tasks included the Montreal Cognitive Assessment (MoCA) (N = 25), Paced Auditory Serial Addition Test (PASAT) (N = 23), California Verbal Learning Test Second Edition (CVLT-II) (N = 23), Adaptive Rate Continuous Performance Test (ARCPT) (N = 19), Stroop Color and Word Test (N = 25), Trail Making Test (N = 25), FAS Verbal Fluency (N = 25), and Animals Verbal Fluency (N = 25)*.

*N indicates number of participants with baseline and follow-up data. Neuropsychological tasks are described in Table 2.

Table 2.

Description of neuropsychological tests and measures

Test Measure	Test Description
Attention
ARCPT sensitivity D', ARCPT inconsistency	Adaptive test of vigilance and sustained attention. Participants press a key in response to a pre-specified sequence of letters. Indices examined included sensitivity D' (measure of overall accuracy) and inconsistency.
Learning
CVLT-II total free recall, long-delay free recall	Assesses ability to learn and recall a list of words over several trials and after a delay. Indices examined included total free recall (total number of words correctly recalled over five trials) and long-delay free recall (number of words correctly recalled after delay). Note that an alternate form was used at follow-up to limit practice effects.
Processing Speed
TMT part A	Tasks of visual attention that assesses speed at which participant can connect 25 numbered circles correctly and in order. Examined time to complete task.
Stroop color naming	In separate trials, participants read from a list of color names and state the color of X's on a page. Number of colors correctly named over 45s was examined.
ARCPT final ISI	Smallest interval between stimuli in which participants can maintain 80% accuracy on ARCPT.
Executive Function
Stroop color-word test, Stroop color-word interference	Participants are required to inhibit an automatic response to read a word by identifying its color.
TMT part B	Participants connect an alternating sequence of numbers and letters, switching between numbers to letters with time for task completion used as the dependent measure.
Verbal Fluency
Letter fluency	Participants rapidly generate words beginning with a specified letter over three trials of one minute each. Total number of words correct over the three trials are examined.
Semantic Fluency	Participants rapidly generate words belonging to a specified semantic category in a single trial of one minute. Total number of correct responses is examined.
Working Memory
PASAT	Test of information processing speed, flexibility, and calculation skills.
Overall Cognition	Composite score generated by averaging Z scores for the above domain composite scores.

Note: MoCA = Montreal Cognitive Assessment, PASAT = Paced Auditory Serial Addition Test, CVLT-II = California Verbal Learning Test Second Edition, ARCPT = Adaptive Rate Continuous Performance Test, TMT = Trail Making Test; change calculated as follow-up minus baseline.

Neuroimaging Acquisition and Processing

Neuroimaging data was collected on a 3T Siemens Verio scanner at UF Health Shands in Gainesville, Florida. Whole brain T1 images were collected in 176 sagittal, 1.0 mm interleaved slices (TR = 2530 ms, TE = 3.5 ms, FOV = 256 mm²). DTI images were acquired in 73 sagittal, 2.0 mm interleaved slices (TR = 17300 ms, TE = 81 ms, FOV = 256 mm²). T1 images were segmented using Freesurfer Version 6.0.0 (29). Manual edits were performed to correct segmentation errors (30) and scans were visually assessed for artifacts. Eddy correction (Version 5.0.11) was performed to correct for EPI-related distortions and slice-to-volume motion. A diffusion tensor model was fitted to each voxel using dtifit from FMRIB’s Diffusion Toolbox. DTI images were then pre-processed using Tract-Based Spatial Statistics (TBSS) (31). FA skeletons were extracted from the TBSS output. Bilateral white matter TOIs were indexed from the JHU white matter tractography atlas. FLIRT was used to align the standard MNI152 1mm brain to the standard 1mm FMRIB58 brain and the saved transformation was used to FLIRT white matter TOIs to FMRIB58 space using nearest neighbor interpolation with 12 DOF. Transformations were individually inspected for accuracy. CC TOIs were generated using Freesurfer. Trained researchers analyzed segmentation for accuracy; control points were added to correct for white matter underestimation. Freesurfer brains were transformed to standard MNI152 space and then FLIRTed to FMRIB58 space. The resulting image was used to FLIRT the Freesurfer automatic segmentation to fMRIB58 space; transformations were individually inspected for accuracy. TOIs were extracted using fslmaths and binarized with FSL (32) to create masks. fslmaths -mas was used with combined bilateral TOI masks and FA skeletons to generate TOIs with FA data inside, preserved only where the skeleton exists, for all participants. FA values were extracted and imported into SPSS for further analysis described below.

Statistical Analysis

Statistical analyses were conducted using IBM SPSS Statistics v25 and R 4.2.2. Descriptive analyses identified demographic and clinical characteristics. Cognitive change scores were calculated by subtracting baseline from follow-up. Histograms and Q-Q plots were generated to assess normality and non-normal data were Winsorized where normality was an assumption by replacing values three standard deviations above or below the mean with the mean. Cognitive change data was Winsorized after calculation. Paired sample t-tests were computed to compare cognitive performance between timepoints. To account for participant drop-off, independent samples t-tests were performed to compare participants who underwent surgery and were seen for follow-up and participants who only received surgery. Independent samples t-tests were also computed to assess sex differences for comorbidities, cognition, and DTI metrics. Bivariate correlations were performed to assess association between baseline FA and comorbidities – those significantly associated with FA (e.g., age, BMI, hypertension, hyperlipidemia) were included as covariates in regression analyses. Results are reported in Supporting Information Table S2. Education was also included as a covariate.

Composite scores were developed for cognitive domains using a priori knowledge of each test (see Table 2). Winsorized scores were recoded as needed to ensure positive values indicate greater performance for baseline or improvement for change scores. Z-scores for measures associated with each domain were calculated and then averaged to produce composites for attention, processing speed, memory, EF, and verbal fluency. Z-scores were then calculated for each domain and averaged to produce overall cognition composites.

Adaptive LASSO was conducted in R to limit predictors regressed on cognition. This method is used to select a subset of appropriate predictors from a larger pool of variables. Deviating from the standard LASSO, coefficients are penalized by their associated tuning parameters. We fit the adaptive LASSO with 5-fold cross-validation using the glmnet package in R. Selected predictors were then regressed on baseline and change scores for each cognitive domain using multivariable linear regression. To account for multiple comparisons, we applied a Bonferroni correction and divided the standard α < .05 by seven – the number of cognitive domains tested – to produce an adjusted α < .00714.

To further characterize the relationship between cognition pre- and post-surgery, baseline and change composites were correlated for each domain.

Results

Paired-samples t-tests comparing Winsorized baseline and post-surgical neuropsychological scores indicate a trend toward improved cognition after weight loss (see Table 3). Significant improvements were found for Stroop color test (mean = 2.8 ± 4.13, t = 3.387, p = .002) and TMT part B (mean = −9.69 ± 15.92, t = −3.042, p = .006). Decreased performance was observed for CVLT-II total free recall, CVLT-II long-delay free recall, and TMT part A. Attenuated performance was significant for CVLT-II total free recall (mean = −3.52 ± 6.90*, t = −2.447, range = 26, p = .023), with participants exhibiting minor changes in TMT part A (mean = 0.01 ± 7.80, t = −0.150, p = .882) and CVLT-II long-delay free recall (mean = −0.36 ± 2.56, t = −0.668, p = .512) scores. Figure 1 depicts the distribution of composite scores for overall cognition at baseline and twelve weeks post-surgery.

Table 3.

Means and standard deviations for baseline and follow-up cognitive performance measures. Results of paired samples t-test comparing Winsorized baseline and follow-up cognitive performance.

	Baseline		Follow-up		Follow-up - Baseline
Measure	Mean ± SD	N	Mean ± SD	N	Mean ± SD	T- Score	N	P- value
MoCA	25.11 ± 3.10	55	26.6 ± 3.25	25	0.87 ± 2.24	1.946	25	.063
PASAT Trial 1	33.17 ± 10.23	52	38.61 ± 8.54	23	2.96 ± 9.75	1.454	23	.160
PASAT Trial 2	28.24 ± 7.78	50	31.26 ± 7.65	23	1.83 ± 6.08	1.440	23	.164
CVLT-II Total Free Recall	50.42 ± 8.81	55	49.35 ± 7.29	23	−3.52 ± 6.90*	−2.447	23	.023
CVLT-II Long-Delay Free Recall	11.09 ± 3.17	55	11.27 ± 2.99	22	−0.36 ± 2.56	−0.668	22	.512
ARCPT Sensitivity D	2.64 ± 0.56	51	2.98 ± 0.53	23	0.07 ± 0.50	0.610	19	.550
ARCPT Final ISI	72.66 ± 64.26	51	50.63 ± 27.96	23	−3.69 ± 24.91	−0.645	19	.527
ARCPT Inconsistency	7.1 ± 4.16	51	6.52 ± 3.72	23	1.30 ± 3.55	1.591	19	.129
Stroop Word	89.85 ± 17.25	55	90.4 ± 15.74	25	2.36 ± 6.02	1.960	25	.062
Stroop Color	70.15 ± 12.63	54	73.32 ± 12.41	25	2.80 ± 4.13**	3.387	25	.002
Stroop Color Word	37.5 ± 10.41	54	41.72 ± 10.29	25	2.20 ± 6.75	1.629	25	.116
Stroop Color Word Interference	−0.47 ± 8.46	54	3.07 ± 8.21	25	1.20 ± 7.14	0.841	25	.409
TMT Part A	25.46 ± 9.82	54	25.07 ± 8.66	25	−0.20 ± 6.71	−0.150	25	.882
TMT Part B	80.86 ± 49.59	55	59.16 ± 20.56	25	−9.69 ± 15.92**	−3.042	25	.006
FAS Verbal Fluency Total	37.91 ± 11.73	55	40.92 ± 9.96	25	2.28 ± 7.16	1.593	25	.124
Animals Verbal Fluency Total	21.13 ± 4.74	55	21.76 ± 5.13	25	1.40 ± 3.99	1.755	25	.092

Note: MoCA = Montreal Cognitive Assessment, PASAT = Paced Auditory Serial Addition Test, CVLT-II = California Verbal Learning Test Second Edition, ARCPT = Adaptive Rate Continuous Performance Test, TMT = Trail Making Test.

Figure 1.

Violin plots depicting the distribution of composite scores for overall cognition at baseline (blue) and twelve weeks post-surgery (green). Shaded dots represent individual data points.

Results of independent samples t-tests suggest that participants who underwent surgery and returned for follow-up and participants who did not return after surgery were generally equivalent at baseline but the follow-up group possessed higher level of education (mean difference = 2.033 ± .679, t = 2.996, p = .004), lower BMI (mean difference = −4.694 ± 2.067, t = −2.271, p = .028), and higher baseline cingulum (cingulate gyrus) FA (mean difference = .024 ± .012, t = 2.042, p = .046) and the splenium (mean difference = .023 ± .011, t = 2.044, p = .046).

Means and standard deviations for pre-surgical FA were calculated for TOIs (see Table 4). FA was generally higher for medial tracts, such as the splenium (mean = 0.89 ± .04) and lower for longer, more lateral tracts, such as the SLF (mean = 0.55 ± .04). Tracts of highest FA included anterior and posterior CC and parts of the cingulum. Locations of TOIs are depicted in Figure 2.

Table 4.

Means and standard deviations for FA in TOIs. Values may range between 0 and 1.

WM TOIs	Baseline FA
Anterior Thalamic Radiation	0.55 ± .04
Corona Radiata	0.53 ± .04
Cingulum (Cingulate Gyrus)	0.74 ± .05
Cingulum Hippocampus	0.67 ± .10
Forceps Minor	0.65 ± .04
Superior Longitudinal Fasciculus	0.55 ± .04
Uncinate Fasciculus	0.48 ± .05
Genu of Corpus Callosum	0.67 ± .14
Body Corpus Callosum	0.64 ± .10
Splenium Corpus Callosum	0.89 ± .04

Figure 2.

White matter TOI masks (green) overlaid on top of standard 1mm FMRIB58 brain.

Baseline FA was correlated with baseline cognitive composites. Paradoxically, higher FA in some tracts was associated with poorer cognition. At baseline, FA in the cingulum (cingulate gyrus) was significantly positively correlated with attention (r = .308, p = .028), while cingulum hippocampus FA was significantly anticorrelated with learning (r = −.274, p = .043).

Although mostly non-significant, correlations between baseline FA and baseline BMI trended positive. Forceps minor FA was significantly positively associated with baseline BMI (r = .287, p = .034). We observed weak-to-negligible, non-significant inverse relationships between baseline BMI and FA in the cingulum (cingulate gyrus) (r = −.227, p = .096), genu (r = −.009, p = .946) and splenium (r = −.128, p = .350).

Forceps minor FA was significantly anticorrelated with baseline age (r = −.437, p = .001), hypertension (r = −.377, p = .005), hyperlipidemia (r = −.295, p = .029), and A1C (r = −.274, p = .049. CC FA was significantly positively correlated with baseline age (r = .411, p = .002). Cingulum (cingulate gyrus) FA was significantly anticorrelated with baseline A1C (r = −.388, p = .004).

Independent samples t-tests revealed a significant difference between sexes in the forceps minor. FA was slightly greater (~.03) for women than men. These comparisons are not ideal due to the ratio of women to men, and it is unclear whether these findings are driven by actual differences or the disproportionate sample.

Greater FA was predictive of poorer baseline cognition. Higher FA in the cingulum (cingulate gyrus), cingulum hippocampus, forceps minor, genu, and splenium was predictive of poorer baseline learning. Cingulum hippocampus FA was negatively associated with baseline overall cognition. In contrast, higher FA generally predicted cognitive improvement at follow-up. Higher FA in the cingulum hippocampus, genu, and body of the CC predicted improvement in processing speed. Cingulum hippocampus FA was also positively associated with overall cognition.

Baseline composites for each domain were anticorrelated with their associated change score. These correlations were moderate and significant for learning (r = −.475, p = .022) and EF (r = −.498, p = .011). Results of multivariable linear regression analyses are reported in Table 5.

Table 5.

Results of multivariable linear regression using BL FA and covariates to predict baseline and change cognition. Predictors were determined by adaptive LASSO; blanks indicate the variable was not selected as an appropriate predictor in the LASSO.

Measures	R-squared	p-val	β ATR	β CR	β CCG	β CH	β FM	β SLF	β UF	β GCC	β BCC	β SCC
Baseline
Learning	0.493	0.002***			−3.214	−0.193	−2.635			−0.352		−6.056
Overall Cognition	0.357	< .001***				−2.312
Change
Processing Speed	0.468	0.118				0.408				0.514	1.479
Overall Cognition	0.198	0.030*				2.166

Note: ATR = Anterior Thalamic Radiation, CR = Corona Radiata, CCG = Cingulum (Cingulate Gyrus), CH = Cingulum Hippocampus, FM = Forceps Minor, SLF = Superior Longitudinal Fasciculus, UF = Uncinate Fasciculus, GCC = Genu of Corpus Callosum, BCC = Body of Corpus Callosum, SCC = Splenium of Corpus Callosum.

^{* =}sig. at α = .05.

^{** =}sig. at α = .01.

^{*** =}sig. at .00714.

Adaptive LASSO and multivariable linear regression were repeated with AD, MD, and RD as predictors. Results are reported in Supporting Information Table S1. Higher baseline AD, MD, and RD were generally predictive of greater baseline cognition. Greater genu and forceps minor AD were predictive of superior baseline performance in nearly all domains. Higher forceps minor MD also consistently predicted greater cognition. Higher AD, MD, and RD were generally associated with post-surgical cognitive decline. Greater CC RD predicted slower post-surgical processing speed. Change analyses did not reach significance after multiple comparison correction.

Discussion

To our knowledge, this study is the first to analyze the relationship between baseline microstructural coherence and cognitive change after weight loss. Results indicate a positive relationship between presurgical FA and postsurgical cognitive change. Higher FA in the cingulum hippocampus, genu, and body of the CC predicted improvement in processing speed. The genu and body of the CC transmit information across hemispheres and have been previously associated with performance in several domains (33, 34). Cingulum hippocampus FA was also positively associated with overall cognition. This TOI disperses memory information throughout the brain and is integral for learning and memory recall. These analyses did not survive multiple comparison correction, but our findings suggest that greater presurgical FA predicts cognitive improvement after weight loss.

Paradoxically, higher FA was associated with poorer baseline cognition. Higher FA in the cingulum (cingulate gyrus), cingulum hippocampus, forceps minor, genu, and splenium was predictive of poorer learning. Cingulum hippocampus FA was also negatively predictive of overall cognition. These TOIs form a series of interlocking connections that may be particularly affected in obesity. The forceps minor links the frontal lobes and crosses the genu, which also has connections to the cingulum (cingulate gyrus). The cingulum hippocampus is implicated in learning and memory and has connections to the splenium. Post hoc analyses assessed if these paradoxical relationships are driven by abnormalities in other measures of diffusivity. We noted several unexpected trends between AD, MD, and RD and cognition. While higher AD and lower RD and MD are generally associated with greater FA and thus greater cognition, we found that higher AD, MD, and RD were predictive of superior cognition among our participants with morbid obesity. Higher AD, MD, and RD were also generally associated with greater post-surgical cognitive decline. Higher RD indicates lower myelin concentration, while higher MD signals less restricted diffusion. Therefore, FA within our sample may be inflated by decreased myelin, resulting in greater RD, driving greater MD, and ultimately greater FA. These findings indicate that, within our sample of participants with morbid obesity, higher FA represents greater myelin damage, explaining both unexpected negative associations between FA and baseline cognition and greater cognitive recovery observed among participants with lower baseline FA. Given the study limitations (discussed below), future research assessing myelin health in morbid obesity is needed to support these findings.

Our results imply that participants with greater cognitive impairment at baseline exhibit greater cognitive recovery. These individuals display greater cognitive benefits after weight loss, suggesting a threshold of improvement for patients with milder baseline cognitive deficits. It is possible that change in cognition attenuates, and individuals with initially high and low cognition reach equilibrium. Future analyses utilizing 18-month follow-up data will investigate this hypothesis. Regardless, our findings that baseline microstructural coherence may predict lasting change indicate a need for increased emphasis on obesity prevention and early intervention. Educating patients on healthy habits may prevent obesity and comorbidities that increase inflammation and induce brain aging.

The observed negative association between FA and obesity is not entirely novel. In 2017, Birdsill et al. described a positive relationship between waist circumference and FA (25). Intriguingly, the authors noted their results appeared to be driven by lower AD, MD, and RD (25). We attribute differences between results to the cohorts each study utilized; while we focused on bariatric surgery patients with severe obesity, Birdsill et al. included participants with a wide range of BMIs (25). Nonetheless, both studies provide support for a complex relationship between FA and obesity.

We may also look to the theory of cognitive reserve to explain these confounding results. Proponents of this concept postulate that biological or cognitive mechanisms may account for clinically normal or better than expected cognition in cases of brain pathology (35). In the context of healthy aging, there is evidence of increased brain efficiency within the Ventral Attention Network, which may compensate for age-related changes in attention (36). Effects of obesity-induced inflammation are functionally alike normal brain aging, so it is reasonable that bodily responses are similar or even equivalent. Perhaps within our sample of adults with morbid obesity, negative relationships between FA and cognition may be explained by increased network efficiency or other compensatory mechanisms. Further research is needed to explore this possibility.

Previous work within our lab has identified improvements in memory at twelve-months post-surgery, however, current results support cognitive recovery in other domains just twelve weeks post-surgery. We observed significant improvement on TMT Part B, a commonly used measure of EF. There is evidence of an inverse relationship between performance in this domain and obesity, however, our results suggest weight loss may improve obesity-related EF deficits. These short-term effects may be due to metabolic changes associated with bariatric surgery. Unexpectedly, we observed attenuated post-surgical learning at twelve weeks post-operative. These results contradict findings by Gunstad et al. in 2010 that memory improves twelve weeks post-surgery (37). To prevent practice effects, participants were given the CVLT-II Standard Form at baseline and Alternate Form at follow-up; it is possible that participants struggled more to learn and recall words on the Alternate Form. This is unlikely given that previous research has established acceptable test-retest reliability between forms (38). Rather, we propose that although weight loss and accompanying reduction in comorbidities may lead to improvements in some cognitive dimensions, postoperative complications of bariatric surgery may also produce attenuation in other areas of cognition, at least temporarily. It is possible that participants may experience minor cognitive deficits due to decreased nutritional intake. Nutritionists at UF Health estimate that recipients of bariatric surgery consume roughly 1000 calories daily and our participants reported symptoms such as hair loss and fatigue. Additionally, anesthesia may affect both cognition and brain health; changes in brain health may occur even after relatively short surgeries, such as knee replacement, which typically lasts about two hours (39, 40). LAP-BAND and gastric sleeve surgery last 1-2 hours, while gastric bypass may range from 2-3 hours, thus it is plausible that observed deficits in learning may be related to anesthesia effects. Future research is needed to explore potential effects of undernutrition and anesthesia on bariatric surgery patients.

There were several limitations to the current study. Due to logistical issues, few control participants received diffusion imaging, so comparisons with a stable-weight group could not be drawn. Furthermore, we cannot exclude that our results may be driven by test-retest or ceiling effects. The small sample size allowed for limited power; we expect that with a larger sample, more statistically significant changes would have been observed and potentially provided further support for our theory of domain-specific decreases in cognition. It is worth noting, however, that these trends were merely statistically significant and not clinically significant. In other words, the observed decline in learning would not be sufficient to warrant diagnosis of memory decline. Due to small sample size, we also chose not to conduct multiple comparison correction for t-tests comparing cognition between timepoints, which would increase risk of type II errors. Instead, we note an increased possibility of type I errors, which further necessitates replication. Data analyzed included only baseline and short-term follow-up data. We intend to replicate these analyses using 18-month follow-up data to determine if observed relationships between FA and cognition hold for participants who have lost more weight. We anticipate assessments further from surgery are also less likely to be affected by anesthesia exposure and post-surgical nutritional deficit. We also note our sample skews heavily female. Aly et al. found that women are more likely than men to undergo bariatric surgery (41). Future studies exploring the relationship between microstructural coherence and weight loss should aim to recruit a more heterogeneous sample, potentially with a diet or exercise-based weight loss intervention.

Conclusion

Results suggest pre-surgical brain health, specifically microstructural coherence, is associated with cognitive change after bariatric surgery. Our data indicates that weight loss may lead to improvements in some cognitive domains, but obesity-related white matter deterioration and health factors may continue to impact other areas, such as learning. We also identified novel relationships between presurgical FA and cognition at baseline and after weight loss. Post-hoc analyses revealed a complex relationship between white matter metrics and cognition, highlighting the importance of assessing additional measures of diffusivity. Our findings provide support for the notion that additional effort should be allocated toward promoting healthier lifestyles and increasing obesity prevention measures.

Study Importance Questions.

What is already known about this subject?

• Preliminary evidence suggests weight loss may improve cognitive ability

• Obesity is associated with decreased microstructural coherence, which may, in part, be the result of chronic low-grade inflammation in the brain

What are the new findings in your manuscript?

• Weight loss in patients with obesity generally leads to improved cognition, particularly executive function.

• Baseline white matter microstructural coherence positively predicts cognitive recovery after weight loss.

How might your results change the direction of research or the focus of clinical practice?

• We find that cognitive recovery is influenced by baseline microstructural coherence; thus further emphasis on weight management may be beneficial in preventing obesity-related cognitive decline

• Future research should investigate the long-term relationship between baseline microstructural coherence and post-surgical cognitive change. It is possible that cognitive improvement may continue with weight loss and/or patients may reach a threshold of cognitive recovery.