T1 relaxation time is prolonged in healthy aging: a whole brain study

Background/aim Measurement of tissue characteristics such as the longitudinal relaxation time (T1) provides complementary information to the volumetric and surface based structural analyses. We aimed to investigate T1 relaxation time characteristics in healthy aging via an exploratory design in the whole brain. The data processing pipeline was designed to minimize errors related to aging effects such as atrophy. Materials and methods Sixty healthy participants underwent MRI scanning (28 F, 32 M, age range: 18–78, 30 young and 30 old) in November 2017–March 2018 at the Bilkent University UMRAM Center. Four images with varying flip angles with FLASH (fast low angle shot magnetic resonance imaging) sequence and a high-resolution structural image with MP-RAGE (Magnetization Prepared - RApid Gradient Echo) were acquired. T1 relaxation times of the entire brain were mapped by using the region of interest (ROI) based method on 134 brain areas in young and old populations. Results T1 prolongation was observed in various subcortical (bilateral hippocampus, caudate and thalamus) and cortical brain structures (bilateral precentral gyrus, bilateral middle frontal gyrus, bilateral supplementary motor area (SMA), left middle occipital gyrus, bilateral postcentral gyrus and bilateral Heschl’s gyrus) as well as cerebellar regions (GM regions of cerebellum: bilateral cerebellum III, cerebellum IV V, cerebellum X, cerebellar vermis u 4 5, cerebellar vermis u 9 and WM cerebellar regions: left cerebellum IX, bilateral cerebellum X and cerebellar vermis u 4 5). Conclusion T1 mapping provides a practical quantitative MRI (qMRI) methodology for studying the tissue characteristics in healthy aging. T1 values are significantly increased in the aging group among half of the studied ROIs (57 ROIs out of 134).


Introduction
According to the 2019 Revision of the World Population Prospects, by 2050, people over the age of 65 will be 16% of the total population [1]. Such a change in the pattern of population growth in favor of elderly people emphasizes the importance of aging studies in neuroimaging. During healthy aging, functional [2] and structural [3] changes occur in the brain. Beyond healthy aging, a notable amount of the aging population is susceptible to neurodegenerative diseases. The most common of them is Alzheimer's disease, which affects approximately 30% of people aged 85 or older [4]. Understanding healthy brain aging is therefore crucial to be able to detect the early biomarkers of the age-related diseases.
Age-related structural brain changes including cortical thickness and signal intensity were reported in various areas [5]. These changes are so profound that age prediction from brain magnetic resonance (MR) images can feasibly be performed with machine learning [6] or based on the contrast at the tissue borders [7]. On the other hand, there are also age-related changes in the brain in intrinsic characteristics of the tissues. It is demonstrated that the main driver of the age-related gray matter (GM) volume loss in subcortical areas is the tissue property changes rather than atrophy [8]. In several studies, myelin, water content and iron were reported as the main factors that contribute to the structural changes in the aging brain. In MR imaging, these differences are observed through longitudinal relaxation time T 1 , the longitudinal relaxation rate R 1 (R 1 = 1/T 1 ) and transverse relaxation time T 2 .
However, due to several technical limitations, the MRI signal bears inadequate information regarding the microstructure of the underlying tissues [9] through the T 1 , R 1 and T 2 measures. First, the signal is a nonlinear mix of several signal contrasts (T 1 , T 2 or proton density (PD)). Second, choice of the data acquisition parameters, such as inversion time, echo time, flip angle, can only be determined suboptimally. Third, hardware artefacts in conventional MRI are difficult to circumvent through reverse engineering the signal. Due to these factors, use of conventional MRI techniques does not allow prediction of underlying properties of the human brain tissue. On the other hand, T 1 relaxation time is a good indicator of local tissue density (such as water content), macromolecules (e.g., myelin), paramagnetic concentration (e.g., iron), as well as the lipid and protein compositions of the underlying tissue. Another superior property of the T 1 is that it is less affected by hardware artefacts [10]. Additionally, T 1 of the human brain has also been reported as a biomarker of development and maturation [11]. The abovementioned properties emphasize that T 1 is a suitable parameter to investigate aging changes.
The studies that address quantification of T 1 in the whole brain are scarce. In healthy aging, the studies that performed T 1 mapping can be grouped based on the subject populations. Most of the studies conducted in this area are crosssectional designs. Steen et al. [12] examined age-related T 1 changes in 18-72 years old 55 healthy volunteers and reported that T 1 increased with age in the genu, frontal white matter, occipital white matter, putamen, and thalamus. Contrary to the subcortical areas, significant decrease in T 1 with age was observed in cortical gray matter [12]. In a comprehensive study of 115 healthy subjects aged 4 to 72, it was reported that T 1 exhibits a quadratic pattern in such a way that it begins decreasing in adolescence and reaches to minimum in 40-60 years but then starts to increasing [13]. Another study with 56 normal volunteers with ages 11-71 years reported a positive linear relationship between T 1 and age in the genu of the corpus callosum (CC) and frontal white matter (WM) regions and a negative linear relationship in the left substantia nigra (SN) [14]. In line with the outcomes of Cho et al. (1997), the quadratic patterns were also shown in 3 frontoparietal WM regions and the right SN in this study [15]. In a recent study on 70 healthy subjects aged 20-76, a significant increase T 1 in thalamus and WM but a decrease in amygdala, nucleus accumbens and the ventralinferior putamen were reported [16]. Another recent study of 211 healthy participants aged 20-89 years found that there was negative linear relationship between T 1 and age in putamen, thalamus and head of the caudate nucleus, but positive linear correlation in frontal lobe WM, globus pallidus and genu of the CC [17]. In all these crosssectional studies, the age range of the participants in the cohort seem to be a major confounding factor prohibiting comparison of the findings.
Despite the abundance of the crosssectional studies, there are only two longitudinal designs investigating agerelated changes of T 1 . In a part of the LBC1936 longitudinal study (wave 2: N = 653, mean age = 73; wave 3: N = 442, mean age = 76) hippocampal T 1 was demonstrated to significantly decrease across waves [18]. In the other longitudinal study (7-year period,17 healthy subjects, 51-77 years), an age related decrease of mean T 1 in GM cortex and unchanged mean T 1 in WM were reported [19]. As seen from the literature, there is no consistency in the reported T 1 changes during aging and it seems that the limited number of ROIs targeted in these studies prohibit a general comparison across the findings.
In this study, we aimed to conduct an exploratory approach to understand T 1 relaxation time differences in aging by studying the entire brain. The data processing pipeline was designed to minimize the errors related to aging effects such as atrophy. It is well known that agerelated GM-WM loss and also brain atrophy become prominent after 70-80 years of life [20]. The age range of the old group participated in our study was chosen specifically to circumvent this problem.

Participants
A total of 63 participants volunteered to participate in the study; however, 1 young and 2 old subjects were excluded from the study because of their claustrophobia. Hence, total of 60; 30 young (mean = 26.36 years, SD = 2.69, 12 F, 18 M) and 30 old (mean = 67.46 years, SD = 4.89, 16 F, 14 M) subjects participated in the study. Demographic information of the remaining 30 participants is given in Table. Classification of the participants into age groups was conducted according to the aging criterion of the United Nations [21].
The recruiting of the participants was conducted via distributed fliers, social media and with the help of our circle of acquaintances. All the participants signed informed consent according to the principles of the Declaration of Helsinki and the study was approved by Ankara University Clinical Research Ethical Committee. Exclusion criteria were history of a neuropsychological/ psychiatric disorder, alcoholism, use of medication affecting the central nervous system (CNS) and cognitive decline in the aging group. None of the participants reported having a neuropsychological/psychiatric disorder or suffering from alcoholism and all of them stated that they were not using any medication affecting CNS. In addition, the participants included in the study did not have any physical disability preventing the application of the cognitive tests (visual, hearing, etc.), nor a metal prosthesis or pacemaker, and claustrophobia which would endanger their MR scanning. To determine whether the older participants were cognitively healthy, standardized mini mental state examination (SMMSE) [22] and geriatric depression scale (GDS) [23] were administered. The cutoff scores were 25 and 11, respectively. All participants fulfilled these criteria, as indicated in Table. 2.2 Procedure Structural MRI data were collected at the National Magnetic Resonance Research Center (UMRAM) in Bilkent University, using a 3T MRI system (Siemens Magnetom Trio, Germany) in November 2017-March 2018. The scan time was 20 min. High resolution 3D anatomical brain images were collected with of MP-RAGE (Magnetization Prepared -RApid Gradient Echo) protocol (TR = 2500 ms, TE = 3.16 ms, bandwidth = 199 Hz/pixel, matrix 256 × 256, slice thickness 1mm, 256 slices, FOV = 256 × 256 (axial), number of averages = 1). Then 4 brain images with four different flip angles (3°, 5°, 15°, 30°) that adhered to the same imaging coordinates with the MPRAGE sequence were collected with FLASH (fast low angle shot magnetic resonance imaging) sequence (TR = 20 ms, TE = 4.15 ms, bandwidth = 199 Hz/pixel, matrix 256 × 256, with slice thickness 3 mm, 44 slices, FOV = 256 × 256 (axial), number of averages = 1). MPRAGE and FLASH sequences were preferred because MPRAGE yields high resolution images with low specific absorption rate (SAR) even in high magnetic field MRIs, and FLASH sequence introduces images varied contrast images in short scanning times. Moreover, these sequences are widely available in various scanners.

Data processing 2.3.1 Preprocessing
Data collection is prone to head motion due to the long scan time. To remove head motion, images were deobliqued using AFNI's 3dWarp program [24]. In addition, skull removal and bias field correction was performed with FAST tool [25]. For standardization, the images were aligned to the standard stereotaxic space (TLRC) by auto_ tlrc program of AFNI [24].

T 1 mapping
It was previously reported that variable flip angle (VFA) provides a better alternative to conventional methods through improved precision and speed [26]. Hence, we used VFA with 4 different flip angles (FA) for mapping the spin lattice relaxation time (T 1 ) of each voxel. FLASH is an appropriate sequence providing opportunity to collect images at different contrasts [27], allowing enhancement of data through different flip angles. For this purpose, we collected four FLASH images with varying flip angles (3°, 5°, 15°, 30°).
Eq. 1 demonstrates the intensity value I (x, y, z) of the voxel at (x, y, z) coordinates acquired with FLASH sequence. The intensity is calculated by tissue characteristics such as magnetization transfer constant (M 0 ), longitudinal relaxation time (T 1 ), transverse relaxation time (T 2 ) and scanning parameters such as repetition time (TR), echo time (TE) and flip angle (α).
Iα(x,y,z) = M0(x,y,z) e -TE/T 2 * sin(α) Based on the requirement of VFA method, we collected four images with different flip angles. For really small flip angles (e.g., α = 3°) cos(α) approaches to 1; thus, the equation (1) is reduced to Eq. 2 [28]: Iα(x,y,z) = M0(x,y,z) e -TE/T 2 * sin(α) In this way, the intensity value of the FLASH image collected with 3° flip angle is described as a constant c = M 0 (x, y, z)sin (3°˚). Hence, Equation 1 can be rewritten as follows: Iα(x,y,z) = M0(x,y,z) e -TE/T 2 * sin(α) I α (x, y, z) in Equation (3) represents the intensity value observed in FLASH images acquired with 5°, 15°, and 30° flip angles, respectively, where c is acquired from the image with α = 3°. So far, all the parameters in Eq. (3) are known except for T 1 . The usage of 3 equations derived from 3 images and the only one unknown parameter (T 1 ) make this problem overdetermined. Based on previous studies, the range of the T 1 is determined as 0-4000 ms. For all these candidate values of T 1 , intensity I α (x, y, z) is calculated for all three images (α = 5°, 15°, and 30°) based on Eq. (3). Then, computed theoretical I α for each T 1 and measured real I α in image is subtracted and squared for the voxel (x, y, z). Through a least square error fit, the T 1 value

Segmentation
FSL's FAST tool [25] is used to segment the MPRAGE image, into three tissue types: WM, GM, and cerebrospinal fluid (CSF). The MPRAGE image is aligned to the T 1 map image through the FLASH images. Hence, the binary CSF mask produced from the MPRAGE image can be used to remove the CSF from the T 1 maps, to take care of the data processing errors due to the atrophy in the cortex and ventricular enlargement in the subcortical areas of the aging group. The binary masks of WM and GM produced from the MPRAGE are then used as a basis for T 1 measurements in specific ROIs depicted by the atlas (see Figure 1). The usage of WM masks is especially important to minimize the partial volume effects (PVE) in the aging brain. These procedures are illustrated in Figure 1.  positive rate is controlled using family-wise error (FWE) correction for multiple comparisons, and thresholded at p < 0.05 at the cluster level.

Subcortical area
A Kolmogorov-Smirnov test indicated that T 1 values in this region had a normal distribution (p ≥ 0.05). Independent Samples t-test demonstrated that average T 1 value in bilateral hippocampus, caudate and thalamus of the old group were significantly higher than those of younger counterparts. Figure 2a shows the average T 1 values that significantly increased in the aged group.

Cerebellum
Kolmogorov-Smirnov test indicated that the T1 values in GM regions of the cerebellum were not normally distributed (p ≤ 0.05). A Mann-Whitney U test indicated significant age-related differences in average T 1 measurements in 9 GM regions of the cerebellum. The average T 1 values of old participants measured in bilateral cerebellum III, cerebellum IV V, cerebellum X, cerebellar vermis u 4 5 and cerebellar vermis u 9 were significantly higher than those of the young group as seen in Figure 2b. Similarly, a Kolmogorov-Smirnov test indicated that T 1 values measured in WM regions in the cerebellum did not have a normal distribution (p ≤ 0.05). The results of the Mann-Whitney U test showed that the average T 1 value measured in left cerebellum IX, bilateral cerebellum X and cerebellar vermis u 4 5 of the old participants were significantly higher than those of young participants (Figure 2c).  Figure 2c. Overall, we have demonstrated that there is a trend indicating that average T 1 values measured in the brain significantly prolonged with increasing age. We did not observe a significant reduction of T 1 in any of the ROIs.

Discussion and Conclusion
Quantitative MRI techniques with different contrast mechanisms are used in the clinic to provide valuable information about both normal and pathological brain  tissues. There are several studies reporting a relationship between myelin and T 1 [30,31]. Reduced myelination of the underlying tissue tends to prolong T 1 [32]. Rooney et al. (2007) showed the relative contribution of iron content to R 1 maps [33]. Higher iron concentration in deep brain nuclei [34] and cortex [35] reduces T 1 . In addition, T 1 is linearly proportional to the water content [36] and agerelated reduced water content tends to reduce T 1 [35]. Furthermore, it is important to consider that these factors are coupled with each other. While myelin is a key factor determining the interpretation of T 1 , the water content of the myelin is a confounding factor. Previous studies reported differences of T 1 in subcortical and cortical areas of patients with Parkinson's [37] and Alzheimer's disease [38]. These outcomes indicate the potential of T 1 mapping in neurodegenerative disease research. Determination of T 1 variation in the brain is also important from a cytoarchitectural point of view. It was shown that R 1 maps of the visual cortex and retinotopic maps are associated with many visual area borders [39].
In our study, T 1 values in the whole brain of young and old participants were measured based on 134 ROIs from the CA_N27_ML atlas. This allowed us to investigate T 1 changes during healthy aging comprehensively. To the best of our knowledge, this is the first study investigating agerelated changes of T 1 relaxation time in the ROIs selected from the whole brain using an exploratory approach. The major finding of this study is that age-related T 1 prolongation is observed in various subcortical and cortical brain structures, as well as cerebellar regions. In a cohort of young (18-35 years old) and early aging (60-78 years old) groups, age-related prolongation of T 1 in several subcortical, cortical, and cerebellar areas (57 of 134 ROIs) is observed. All the significant changes observed in this study are related to prolongation of T 1 . We did not observe a significant decrease in T 1 in any of the ROIs between these two age groups.
In terms of subcortical structures: We observed significant T 1 increases in bilateral caudate, thalamus and hippocampus in the older group relative to younger participants. The findings of our study in these subcortical areas replicated the outcomes of some previous studies indicating an age-related T 1 increase in basal ganglia [40] and thalamus [12,16,17] in elderly. A possible origin of increased T 1 in subcortical regions in our experiment might be explained by the concomitant loss of myelin in these structures. Age-related demyelination in thalamus [41] and in basal ganglia [42] have been previously reported. Although increased iron content in elderly shortens T 1 , a degeneration of the myelin sheets has been observed in elderly due to iron accumulation [43]. Water content also influences T 1 and it is strongly correlated with iron in several regions including caudate and thalamus [34]. Unfortunately, the underlying region specific mechanism of the association among iron, myelin and T 1 is poorly understood [34,42]. Therefore, further studies with complementary techniques (such as MR spectroscopy) and multiparameter approaches investigating myelin water fraction (MWF) are needed to unveil the individual aging mechanisms.
Our outcomes contradict with some recent studies which reported a decreased T 1 [16] and no change [44] in deep GM structures (e.g., basal ganglia) of old participants. This conflict might stem from the difference between the ROI measurements in these studies and ours. For instance, in one of these studies, manual drawing of the ROIs on a single midslice on the population-averaged map was used [16]. In our study 134 3D ROIs are created for each participant's MR image by registering to the stereotaxic space and atlas. We developed an automated data processing pipeline free from biases stemming from subjective measurements. Therefore, the discrepancy between our results and others' may be attributed to the choices in such quantification techniques. In the other study [44] presenting conflicting outcomes with ours, T 1 was measured on an ROI which was a combination of several subcortical structures (e.g., caudate, putamen, thalamus, etc.). Such combination of structures might cause losing regionally-specific information which is important in aging [45]. Finally, the CSF T 1 values are prominently different than those of GM and WM areas, deserving special treatment. We removed the CSF areas in subcortical and cortical regions to minimize the effect of age-related atrophy and ventricular enlargements on tissue concentrations and to measure T 1 more precisely. We believe that some portion of the reduction of T 1 reported in other studies might result from confounding factors that were not accommodated during the initial data processing steps.
In terms of cortical structures: Our findings of elongated T 1 are in line with some studies on somatosensory cortex [46] and on occipital gyrus [47]. On the other hand, a decrease in T 1 with aging [12,44,48,49] or no change in cortex [50] were reported in other works. The difference between the age range of the participants in Gracien et al.'s study (2017) and ours possibly presents the main source of this contradiction (i.e. 55-71 years in the former, 18-78 years in the latter). Earlier, it has been reported that the T 1 differences may follow a quadratic pattern [13]. Gracien's study (2017) might be focusing on the decreasing range of this pattern while our study focuses on the increasing range. Global T 1 measurements based on frequency distributions (or spectra) were conducted in two studies [48,49]. In these studies, the aging effect was reported as decreased T 1 in GM and increased T 1 in WM. These studies did not distinguish the T 1 changes with respect to ROI. Hence, a comparison between these studies and ours is impossible.
In terms of cerebellum, we found T 1 prolongation in several structures including cerebellum III, X, and vermis. There are very few studies investigating T 1 changes in aging in the cerebellar area. The sheet-like structure of cerebellum is composed of complex cellular layers which are really thin (≤1 mm) [51]. Moreover, the myelination degree [52] varies in different cortical depths, also cell sizes, distribution and densities across the cerebellar lobes [53]. Perhaps the complexity of the cerebellar anatomy is a prohibiting factor in ROI based analysis of T 1 changes in cerebellum. We have replicated outcomes of Saito (2009) reporting an increase in WM anterior subsegment of cerebellum. On another front, one recent study by Badve et al. (2015) reported that age-related changes were nonexisting in cerebellum and cerebellar vermis [14]. In Badve and colleagues' study, quantitative metrics were measured with magnetic resonance fingerprinting (MRF) technique, in which slice thickness was higher than that in our study. Resolution differences might have been effective in the conflicting outcomes between their results and ours.
Although there are abundant previous studies focusing on subcortical structures, the literature is not rich in evaluating the T 1 variation in cortex and cerebellum. The complex and folded topology of cortex and cerebellum is a possible reason of the absence of qMRI and microstructural studies in these areas. There are several limitations imposed by the aging process and complexity of the automatic processing of brain images. The cortical thinning [5] is a crucial factor that would introduce PVE and worsens interpretation of the cortical data which already has a complex structure.
The present study has several limitations that should be addressed. The variable flip angle method which we have used in the present study is reported to overestimate T 1 values in vivo due to imperfect spoiling and B 1 bias [54]. Although the correction methods for flip angle inhomogeneities were proposed previously [55], many qMRI studies-including ours-lack a correction while computing the T 1 maps. Only after flip angle corrections are done on individual brains, the T 1 evaluation methods would indicate accurate changes between age groups. It is important to identify within which range flip angle inhomogeneities are tolerated to predict T 1 maps. To test the restrictions of our study, we ran simulations to compute the range of errors in predicting the T 1 values. We can safely state that our results are valid, if the variability of flip angles remain in the range of ±10% in comparison to the settings of the MR protocol (results available upon request).
This study has a crosssectional design which limits the interpretation individual aging patterns. Longitudinal studies should be conducted in the future to provide a better understanding of the age-related changes and minimize intersubject variations. To sum up, our report of increased T 1 characteristics in healthy aging is centered on young and old adults and limits the generalization to ages outside this age range.
Furthermore, the interpretation of T 1 measurements is affected by other factors such as the initial atlas registration and regional boundaries of the anatomical structures. Another limitation worthy to note is the method of registration. There are two different age groups whose anatomical properties differ a lot. We have created special masks for each participant to remove CSF for a better differentiation of WM and GM. This way, at least we were able to remove the age-dependent increase in ventricles and the effects of atrophy. Age-specific atlases may be used for future studies. In addition, when we consider biological variation introduced by the regional heterogeneity of the aging patterns, the complexity of predicting MR signal characteristics represented by T 1 should be considered carefully.

Acknowledgment/disclaimers/conflict of interest
This article is based on the doctoral study of Hayriye Aktaş Dinçer. The authors are grateful to Düzce University, Bilkent University and Middle East Technical University for partially supporting the research infrastructure, MR scans and laboratory environment. More specifically we would like to thank Prof. Ergin Atalar for data acquisition and METU BAP office for graduate student project funding. The authors are indebted to Ece Dinçer and İlayda Çalışkan for assistance during scanning and subject surveys.
The authors declare that they have no conflict of interest.

Informed consent
All procedures performed in studies involving human participants were in accordance with the ethical standards according to the principles of the Declaration of Helsinki and this study was approved by Ankara University Clinical Research Ethical Committee (Protocol Number:13-416-12). All the participants signed informed consent.