Introduction
Late-life depression (LLD) and frailty are two common and often co-occurring conditions in the elderly. LLD may include both syndromal major depression and subsyndromal presentations of depression. Frailty is a geriatric syndrome associated with declines in health and function. Clinical and epidemiological studies show a strong association between these conditions, with 38.6% of frail elders having clinically significant depressive symptoms and 40.4% of those with LLD meeting the criteria for prefrailty and frailty (Soysal et al., Reference Soysal2017). This evidence suggests a bidirectional relationship in which LLD may accelerate frailty development and frailty may exacerbate LLD (Mayerl et al., Reference Mayerl, Stolz and Freidl2020; Oude Voshaar et al., Reference Oude Voshaar2021; Brown et al., Reference Brown2020). However, the mechanisms underlying this interplay remain unclear.
Although much is known regarding the structural brain changes of LLD or frailty, little is known regarding their joint effects on the brain. Previous studies including individuals with LLD or frailty have shown significant subcortical and cortical grey matter atrophy (Benjamin and Steffens, Reference Benjamin and Steffens2011; Lopez-Sanz et al., Reference López-Sanz2018), with the atrophy being more pronounced in the hippocampus, amygdala, anterior cingulate, striatum, and the frontal gyri (Benjamin and Steffens, Reference Benjamin and Steffens2011; Wen et al., Reference Wen, Steffens, Chen and Zainal2014; Chen et al., Reference Chen2015; Kant et al., Reference Kant2018). Studies using diffusion-weighted imaging (DWI) to quantify microstructural damages in the white matter (WM) (Lopez-Sanz et al., Reference López-Sanz2018; Wen et al., Reference Wen, Steffens, Chen and Zainal2014), in particular, fractional anisotropy (FA), and mean diffusivity (MD) measures, also showed that LLD or frailty have been associated with significantly lower FA values (indicating reduced axonal integrity) and higher MD values (indicating more inflammatory changes in the WM tracts) (Lopez-Sanz et al., Reference López-Sanz2018; Charlton et al., Reference Charlton, Lamar, Zhang, Yang, Ajilore and Kumar2014; Chen et al., Reference Chen, McQuoid, Payne and Steffens2006). The most commonly affected tracts in both LLD and frailty were the corpus callosum, uncinate fasciculus, superior fronto-occipital fasciculus, and anterior limb of the internal capsule (Lopez-Sanz et al., Reference López-Sanz2018; Tian et al., Reference Tian, Williams, Landman, Resnick and Ferrucci2020; Wan et al., Reference Wan2020; Reppermund et al., Reference Reppermund2014).
Frailty and LLD share several imaging abnormalities suggesting the presence of a depression–frailty phenotype. However, no previous study has evaluated the effect of frailty and LLD on structural and microstructural brain abnormalities. Therefore, this study aimed to investigate the effect of the association between LLD and frailty on brain structure. Based on prior studies, we hypothesized that the presence of LLD with frailty would be associated with WM changes in the frontal and temporal regions.
Methods
Sample
We included 31 participants (14 with LLD+Frailty; 17 Never-Depressed+Robust “comparison group”) in this analysis. They were recruited from an ongoing cohort that aims to evaluate the impact of LLD on hallmarks of biological aging conducted in the Centre for Addiction and Mental Health, Toronto, Canada.
DEFINITION OF MAJOR DEPRESSION
All study participants were evaluated by a trained geriatric psychiatrist, and the diagnosis of a major depressive episode was based on the Diagnostic and Statistical Manual of Mental Disorders 5th ed (DSM–5; American Psychiatric Association, 2013) diagnostic criteria for single episode or recurrent major depressive disorder without psychotic features. The severity of the depressive episode was assessed using the Patient Health Questionnaire (PHQ-9) and Montogomery-Asberg Depression Rating Scale (MADRS) (Levis et al., Reference Levis, Benedetti and Thombs2019; Montgomery and Asberg, Reference Montgomery and Asberg1979).
DEFINITION OF FRAILTY
Frailty was assessed using the FRAIL scale, a validated questionnaire that evaluates the five components of the frailty phenotype: fatigue, resistance, ambulation, medical comorbidity, and weight loss (Kojima, Reference Kojima2018). Each item corresponds to a single question, with each scored as 0 (absent) or 1 (present). Possible FRAIL Scale scores range from 0 to 5, classifying subjects as robust (score = 0), prefrail (score = 1–2), or frail (score = 3–5). For this analysis, we combined individuals identified as prefrail (n = 13) and frail (n = 1) in the same group, since prefrailty shares many of the health risks associated with frailty and is a significant risk factor for developing frailty (Kojima, Reference Kojima2018).
COMPARISON GROUP (CG)
Individuals without a past or current history of a major depressive episode and physically robust were included as a CG.
For both groups (LLD+Frailty and Never-Depressed+Robust), the exclusion criteria included a history/current diagnosis of bipolar disorder, schizophrenia, or other major psychiatric disorder; history of dementia or intellectual disability; history of severe head trauma; history of alcohol or substance abuse; history of a major central nervous system disorder or past stroke; current unstable medical condition; a diagnosis of HIV, and chronic use of anti-inflammatory medications. Exclusion criteria for control participants included the use of antidepressants or mood stabilizers for other medical conditions. The CAMH ethics board approved this study, and all participants provided written informed consent after a detailed description of the study.
ADDITIONAL CLINICAL MEASURES
Global cognitive performance was assessed using the Montreal Cognitive Assessment (MoCA) and the Repeatable Battery for the Assessment of Neuropsychological Status (RBANS) (Nasreddine et al., Reference Nasreddine2005; Randolph et al., Reference Randolph, Tierney, Mohr and Chase1998).
MRI acquisition and processing
Subjects were imaged at the Toronto Neuroimaging (ToNI) Facility at the University of Toronto (Toronto, ON, Canada), using a 3T Siemens Prisma MRI Scanner (Siemens Medical Solutions, Malvern, PA, USA). A 3D T1-weighted neuroanatomical scan was acquired using a magnetization-prepared rapid acquisition gradient echo (MPRAGE) sequence, for cortical and subcortical segmentation (1mm isotropic, TR = 2500.0 ms, TE = 2.9 ms, TI = 1070.0 ms, 178 slices, FOV = 256 × 256 mm2, 8° flip angle). Structural MRI data were processed using the Freesurfer Software Suite (http://surfer.nmr.mgh.harvard.edu, version 6.0).
Subcortical regional volumes (thalamus, caudate, putamen, pallidum, hippocampus, amygdala, accumbens area) were defined according to the Desikan-Killiany atlas (Desikan et al., Reference Desikan2006). These subcortical regional volumes were segmented and processed through the pipelines ‘recon-all’ command. Specifically, the ‘recon-all’ image processing involved (1) nonbrain tissue removal; (2) spatial normalization; (3) intensity inhomogeneity correction; (4) tissue-type classification; and (5) parcellation of subcortical regions (Fischl et al., Reference Fischl2004).
Vertex-wise cortical thickness values were obtained through the Freesurfer surface-based processing stream. Specifically, the surface-based processing stream involved (1) processing images through the ‘recon-all’ command, which has been described above; (2) defining boundaries between white and grey matter and boundaries between grey matter and cerebrospinal fluid; (3) registering individual-level data to a Freesurfer average template, and (4) spatially smoothing images with a Gaussian kernel (FWHM = 10 mm) (Schaer et al., Reference Schaer, Cuadra, Tamarit, Lazeyras, Eliez and Thiran2008). All volumes and reconstructed surfaces were assessed with FreeSurfer quality assurance tools and by visual inspection to ensure accuracy.
DWI acquisition and processing
DWI data were acquired via a multishell high angular resolution diffusion image (b-values = [500, 1000, 2000, and 3000 s/mm2], b0s = 3, 1.7mm isotropic, TR = 4200.0 ms, TE = 89 ms, 103 slices, FOV = 1260 × 1260 mm2, 90° flip angle). Quality control compared several metrics (namely absolute motion, relative motion, percent outliers, average signal-to-noise ratio, and mean framewise displacement) against a threshold of plus or minus two standard deviations beyond the mean for each metric. A single participant from the depressed–frail group was removed from the analysis after surpassing this threshold in three of the five metrics.
DWI data were preprocessed to remove eddy current-induced distortions and fit a diffusion tensor via the FMRIB Software Library (https://fsl.fmrib.ox.ac.,uk version 5.0.10). Tract-based spatial statistics (TBSS) was then used to perform a voxel-wise statistical analysis comparing the two participant groups on measures of FA and MD. Image processing for TBSS included the erosion of individual subject FA images; aligning all preprocessed FA images to a 1×1×1 standard space (FMRIB58_FA) using nonlinear regression; merging all standard-space FA images into an averaged FA skeleton; thresholding and binarizing of the aforementioned skeleton; and creating a mask of voxels to which all subsequent processing was restricted (Smith et al., Reference Smith2004). These steps resulted in two 4D images for each subject: one image representing subjects’ normalized FA and MD data, respectively.
Statistical analysis
Between-groups differences of subcortical volume were tested with separate univariate analyses of covariance including sex, age, and estimated total intracranial volume (eTIV) as covariates, and corresponding effect sizes were computed using Cohen’s f. Significance was determined using p-values that were corrected for testing across multiple regions (n = 14) using the false discovery rate < 0.05. All analyses were performed in R.
The FreeSurfer command tool, mri_glmfit-sim, was used to analyze between-group differences in cortical thickness at each vertex on the cortical surface. Specifically, mri_glmfit-sim was used to perform cluster-wise correction for multiple comparisons by running a permutation simulation with 10,000 iterations. The initial vertex-wise threshold and the subsequent cluster-forming threshold were set to p < 0.05. P-values were adjusted for testing in two hemispheres. The corresponding general linear model included sex and age as covariates.
The FSL software suit command tool, randomize, was used to test between-group differences in FA and MD at each voxel along the sample-specific mean FA skeleton, representing the centers of the major WM tracts common to all subjects in the sample. Separate permutation simulations, run with 10,000 iterations and threshold-free cluster enhancement, were used to identify regions of significant between-groups differences in FA and MD. Significance was determined using family-wise error-corrected (FWER) p-values that were further adjusted for testing across multiple phenotypes: pFWER = 0.025. All analyses were done controlling for sex and age. When a significant between-groups difference was detected, cluster means and labels were extracted using fslmeants and autoaq, respectively, the latter with the JHU ICBM-DTI-81 White-Matter Labels atlas. Post-hoc analyses, including effect size computation (Cohen’s f), were performed in R.
Results
Participant demographics and clinical characteristics are shown in Table 1. The scores for the FRAIL scale, PHQ-9, MADRS, sex frequency, and years of education were significantly different between groups. The two groups did not differ in age, MoCA, and RBANS score. We found no significant between-groups differences in regional subcortical volumes (Table 2) and vertex-wise cortical thickness when comparing the LLD+Frail and Never-depressed+Robust groups.
Table 1. Sample characteristics
Abbreviations: t = Independent Sampled t-test; X 2 = Chi squared test; MoCA = Montreal Cognitive Assessment; PHQ-9 = Patient Health Questionnaire-9; MADRS = Montgomery-Asberg Depression Rating Scale; RBANS = Repeatable Battery for the Assessment of Neuropsychological Status.
Table 2. Difference in Subcortical Volumes Among LLD-Frail and CG
There were no significant differences observed when comparing subcortical volumes between LLD-Frail and CG individuals. Abbreviations: CV = covariates; eTIV = estimated total intracranial volume; FDR = false discovery rate; CG = Never-depressed+Robust; LLD-Frail = depressed, frail.
In the analysis of DWI data, one participant (Never-depressed+Robust group) had head motion that interfered with the DWI analysis. Therefore, the total sample for this analysis comprised 30 participants (LLD+Frail (n = 14) and Never-depressed+Robust (n = 16)). We detected a significant between-group difference (LLD+Frail > Never-depressed+Robust) in MD measures, adjusted for sex and age, in one cluster (48,225 voxels; peak voxel: pFWER = 0.005, MNI coord. (X, Y, Z) = −26, −11, 27) (Figure 1). The corresponding effect size was large (Cohen’s f = 0.808). The spatial extent of this cluster was large, covering nearly 38% of the mean FA skeleton, representing the centers of the major WM tracts, the corpus callosum, anterior and superior corona radiata, internal and external capsule, and superior longitudinal fasciculus (Figure 1). On the other hand, we identified no significant between-group differences in voxel-wise FA when comparing LLD+Frail and Never-depressed+Robust groups. Including years of education as an additional covariate reduced the overall spatial extent of the effect, but preserved its MD specificity and direction, as well as its localization over the same major WM tracts (Supplementary Table 1 and Supplementary Figure 1).
Figure 1. Regions of significant between-groups difference in MD (red: LLD-Frail > CG), overlaid onto the mean FA skeleton (green). A single cluster representing several WM tracts, the corpus callosum, anterior and superior corona radiata, internal and external capsule, and superior longitudinal fasciculus had a significant between-group difference (LLD+Frail > Never-depressed+Robust) in MD measures. Abbreviations: FA = fractional anisotropy; FWER = family-wise error rate; CG = Never-depressed+Robust; LLD-Frail = depressed, frail; MD = mean diffusivity.
Discussion
In this study, we showed a significantly higher MD values in LLD+Frail compared to Never-depressed+Robust individuals. The difference in MD values between LLD+Frail and Never-depressed+Robust individuals had a large effect size and extended to many WM tracts that support inter-hemispheric communication, emotional, and cognitive processing. MD is a DTI metric that measures the average water diffusion in a voxel, representing WM microstructure and is viewed as an indirect measure of inflammatory activity, with higher values indicating lower regional cellular density and higher local inflammatory activities (O’Donnell and Westin, Reference O’Donnell and Westin2011). Our findings suggest that individuals with LLD and frailty have a significant loss of WM integrity compared to Never-depressed+Robust subjects and are consistent with previous studies showing LLD and frailty to be independently associated with abnormal WM integrity and higher neuroinflammation (Lopez-Sanz et al., Reference López-Sanz2018; Wen et al., Reference Wen, Steffens, Chen and Zainal2014; Chen et al., Reference Chen, McQuoid, Payne and Steffens2006).
The mechanisms underlying the abnormal WM microstructure in LLD and frailty remain unclear. However, since MD is sensitive to cytotoxic edema and WM primary contains glial cells, we hypothesize that inflammation and low-grade cytotoxic edema within glial cells may potentiate LLD and frailty (Ampo et al., Reference Ampo, Mendes-Silva, Goncalves, Bartley, Kuchel and Diniz2022; Diniz et al., Reference Diniz2022; Aizenstein et al., Reference Aizenstein2016). Our results indicate that LLD+Frailty is associated with a higher neuroinflammatory burden in tracts related to interhemispheric communication, emotional, and cognitive processing. This is in line with a recent study showing that subjects with LLD+Frailty have more systemic inflammatory and mitochondrial dysfunction burden when compared with Never-depressed+Robust individuals (Ampo et al., Reference Ampo, Mendes-Silva, Goncalves, Bartley, Kuchel and Diniz2022). Therefore, we hypothesize that a higher inflammatory burden in the central nervous system and the periphery can be a shared mechanism of depression and frailty in older adults and its associated loss of WM tract integrity.
The present study did not show a significant difference in subcortical volume or cortical thickness between frail LLD participants and controls. Grey matter atrophy and ventricular enlargement have consistently been observed in LLD and frail individuals (Chen et al., Reference Chen, McQuoid, Payne and Steffens2006; Benjamin and Steffens, Reference Benjamin and Steffens2011; Lopez-Sanz et al., Reference López-Sanz2018). Our findings may suggest a lack of statistical power due to a small sample size analyzed. On the other hand, these results may also be because the majority of the participants included in this analysis had a milder presentation of the frailty syndrome and might not have yet developed significant gray matter and subcortical structural changes.
The results of this study should be viewed in light of several additional limitations. First, the sample size is relatively small, raising the risk of type II error. Second, our analysis did not include additional CGs (LLD+Robust and Never-depressed+Frail), limiting our ability to infer the extent LLD or frailty contributed to the results. Third, this is a cross-sectional study that limits one’s ability to infer casualty. Fourth, the observational nature of imaging will make this study prone to residual cofounding in which a common, unknown factor is responsible for the significant statistical associations. The small sample size may also make it hard to control for specific cofounding variables, introducing bias into our analysis. Finally, the study has limited external validity as the participants were recruited from two nearby study centers, and the depressed subjects mainly consisted of prefrail females. Future studies utilizing larger samples, additional CGs, and a longitudinal design are necessary to confirm our hypothesis and evaluate the prognostic significance of elevated inflammatory burden in these individuals.
In conclusion, we showed that the LLD+Frailty is associated with significant microstructural abnormalities within WM tracts compared to Never-depressed+Robust individuals. Our findings indicate that heightened neuroinflammatory burden can be a shared mechanism underlying both conditions and suggests the possible existence of a depression–frailty phenotype in older adults.
Conflict of interest
The authors do not have any conflict of interest to report.
Source of funding
This work was funded by NIMH grant R01MH115953.
Description of authors’ roles
E. Shuster: Study design, statistical analyses, and drafting of the manuscript.
A. Miles: Neuroimaging analyses, statistical analyses, and drafting of the manuscript.
L. Heyland: Neuroimaging analyses, statistical analyses,
N. Calarco: Neuroimaging analyses, statistical analyses,
J. Jeyachandra: Neuroimaging analyses, statistical analyses,
S. Mansour: Neuroimaging analyses, statistical analyses
A.Voineskos: Study design, intellectual input, and drafting of the manuscript.
D.Steffens: Intellectual input, and drafting of the manuscript.
Y.Nikolova: Study design, statistical analyses, intellectual input, and drafting of the manuscript.
B. Diniz: Study design, statistical analyses, intellectual input, and drafting of the manuscript.
Supplementary material
To view supplementary material for this article, please visit https://doi.org/10.1017/S1041610223000066
