Abstract
Maternal brain adaptations occur in response to pregnancy, but little is known about how parity impacts white matter (WM) microstructure and WM ageing trajectories later in life. Utilising global and regional brain-age prediction based on multi-shell diffusion MRI data, we investigated the association between previous childbirths and WM brain age in 8,895 women in the UK Biobank cohort (age range = 54 - 81 years). The results showed that a higher number of previous childbirths was associated with lower WM brain age, in line with previous studies showing less evident grey matter (GM) brain ageing in parous relative to nulliparous women. Both global WM and GM brain age estimates showed unique contributions to the association with previous childbirths, suggesting partly independent processes. Corpus callosum contributed uniquely to the global WM association with previous childbirths, and showed a stronger relationship relative to several other tracts. While our findings demonstrate a link between reproductive history and brain WM characteristics later in life, longitudinal studies are required to understand how parity influences women’s WM trajectories across the lifespan.
1. Introduction
Maternal brain adaptations have been shown during pregnancy and postpartum, with dynamic alterations across brain regions depending on time since delivery [1, 2, 3, 4, 5]. While some structural brain changes revert post parturition [6], recent studies indicate that some effects of pregnancy may be long-lasting [2, 5], potentially influencing brain trajectories later in life [7, 8, 9, 10]. However, neuroimaging studies of the maternal brain have largely focused on grey matter (GM) volume [1, 2, 10, 11, 12, 13, 4, 14] and cortical thickness [3, 15], and less is known about the effects of pregnancy on brain white matter (WM) microstructure.
Emerging evidence from animal models suggests that pregnancy may induce WM plasticity [16, 17, 18]. Specifically, pregnant mice exhibit increases in oligodendrocyte progenitor cell proliferation, oligodendrocyte generation, and in the number of myelinated axons, indicating an enhanced capacity for myelination in the maternal brain [16]. Pregnancy-induced remyelination may partly explain why pregnancy seem to cause remission of multiple sclerosis (MS), an auto-immune disease that attacks the myelin sheath [19]. In line with this, slower disability progression has been found in parous MS patients after 18 years, compared with nulliparous patients [20]. This effect was strongest in patients that gave birth after disease onset, indicating favourable effects of pregnancy-related adaptations on disease mechanisms in MS.
While the influence of childbirth on WM trajectories in healthy women is largely unknown, one study reported larger regional WM volumes in mothers compared to non-mothers, as well as maternal WM increases that were linked to changes in empathetic abilities during the postpartum period [14]. In line with these findings, a diffusion tensor imaging (DTI) [21] study in rats found that fractional anisotropy (FA), which quantifies the degree of diffusion directionality, in the dentate gyrus increased significantly during pregnancy. However, whole-brain diffusivity also increased in pregnant rats compared to nulliparous rats [17], indicating global changes in the characteristics of molecular water movement - potentially linked to increased extracellular water in the brain during pregnancy [6].
In a recent study comparing longitudinal changes in human brain morphology during pregnancy and two years of pubertal development, no WM changes were observed in mothers, nor in female adolescents [22]. However, as adolescence is known to involve substantial WM remodelling [23, 24, 25, 26, 27], the lack of effects could possibly reflect insensitivity of the method used to assess WM changes (T1-weighted estimation of total WM volume) [22]. In development and ageing studies, WM microstructure is commonly investigated using DTI [21], which yields metrics that are highly sensitive to age [28]. However, the accuracy of the DTI approach is limited by factors such as, for example, crossing fibres. These obstacles have motivated the development of advanced biophysical diffusion models such as white matter tract integrity (WMTI) [29], which is derived from diffusion kurtosis imaging (DKI) [30], and spherical mean technique (SMT) [31, 32]. Based on assumptions about the underlying microstructure, these models enable estimation of the separable contribution of diffusion in the intra- and extra-axonal space [33], which provides higher biological specificity, 1. e. additional information about the microstructural environment [34].
In the current study, we utilised four diffusion models (DTI, DKI, WMTI, SMT) to build predictive models of WM brain ageing, and investigated associations between brain-age estimates and previous childbirths in a sample of 8,895 UK Biobank women (mean age ± standard deviation = 62.45 ± 7.26). In line with studies suggesting that distinct and regional brain-age prediction models may provide additional detail [11, 35, 36, 37], we used separate models to estimate i) global WM ageing, ii) global GM ageing to test for modality-specific contributions, and iii) WM ageing in 12 major WM tracts in order to identify regions of particular importance for maternal brain ageing.
2. Methods and Materials
2.1. Sample characteristics
The initial sample was drawn from the UK Biobank (www.ukbiobank.ac.uk), and included 9,899 women. 899 participants with known brain disorders were excluded based on ICD10 diagnoses (chapter V and VI, field F; mental and behavioral disorders, including F00 - F03 for Alzheimer’s disease and dementia, and F06.7 ‘Mild cognitive disorder’, and field G; diseases of the nervous system, including inflammatory and neurodegenerative diseases (except G55-59; “Nerve, nerve root and plexus disorders”). An overview of the diagnoses is provided in the UK Biobank online resources (http://biobank.ndph.ox.ac.uk/showcase/field.cgi?id=41270), and the diagnostic criteria are listed in the ICD10 diagnostic manual (https://www.who.int/classifications/icd/icdonlineversions). In addition, 99 participants were excluded based on magnetic resonance imaging (MRI) outliers (see section 2.2) and 11 participants were excluded based on missing data on the number of previous childbirths, yielding a total of 8,895 participants that were included in the study. Sample demographics are provided in Table 1.
2.2. MRI data acquisition and processing
A detailed overview of the UK Biobank data acquisition and protocols is available in [38] and [39]. For the diffusion-weighted MRI data, a conventional Stejskal-Tanner monopolar spin-echo echo-planar imaging sequence was used with multiband factor 3. Diffusion weightings were 1000 and 2000 s/mm2 and 50 non-coplanar diffusion directions per each diffusion shell. The spatial resolution was 2 mm3 isotropic, and 5 anterior-posterior vs 3 anterior-posterior images with b = 0 s/mm2 were acquired. All diffusion data were processed using an optimised diffusion pipeline [40] consisting of 6 steps: noise correction [41, 42], Gibbs-ringing correction [43], estimation of echo-planar imaging distortions, motion, eddy-current and susceptibility-induced distortion corrections [44, 45], spatial smoothing using fslmaths from FSL (version 6.0.1) [46] with a Gaussian kernel of 1mm3, and diffusion metrics estimation. DTI and DKI derived metrics were estimated using Matlab R2017a (MathWorks, Natick, Massachusetts, USA) as proposed by Veraart and colleagues [47]. The DTI metrics included mean diffusivity, FA, axial diffusivity, and radial diffusivity [21]. The DKI metrics included mean kurtosis, axial kurtosis, and radial kurtosis [30]. WMTI metrics included axonal water fraction, extra-axonal axial diffusivity, and extra-axonal radial diffusivity [29]. SMT metrics included intra-neurite volume fraction, extra-neurite mean diffusivity, and extra-neurite radial diffusivity [31]. See [40] for details on the processing pipeline.
Tract-based spatial statistics (TBSS) was used to extract diffusion metrics in WM [48]. Initially, all maps were aligned to the FMRIB58_FA template supplied by FSL, using non-linear transformation in FNIRT [49]. Next, a mean FA image of 18,600 UK Biobank subjects was obtained and thinned to create a mean FA skeleton. The number N = 18,600 was obtained from the processing of the two first UKB data releases. The maximal FA values for each subject were then projected onto the skeleton to minimise confounding effects due to partial volumes and any residual misalignments. Finally, all diffusion metrics were projected onto the subject-specific skeletons. WM features were extracted based on John Hopkins University (JHU) atlases for white matter tracts and labels (with 0 thresholding) [50], yielding a total of 910 WM features including mean values and regional measures for each of the diffusion model metrics. For the region-specific brain age models, 12 tracts of interest used in previous ageing and development studies were extracted [51, 52]; anterior thalamic radiation (ATR), corticospinal tract (CST) cingulate gyrus (CG), cingulum hippocampus (CING), forceps major (FMAJ), forceps minor (FMIN), inferior fronto-occipital fasciculus (IFOF), inferior longitudinal fasciculus (ILF) superior longitudinal fasciculus (SLF), uncinate fasciculus (UF), superior longitudinal fasciculus temporal (SLFT), and corpus callosum (CC). The diffusion MRI data passed TBSS post-processing quality control using the YTTRIUM algorithm [53], and were residualised with respect to scanning site using linear models.
For the GM data, raw T1-weighted MRI data for all participants were processed using a harmonised analysis pipeline, including automated surface-based morphometry and subcortical segmentation. In line with recent brain-age studies [10, 36, 54, 55], we utilised a fine-grained cortical parcellation scheme [56] to extract cortical thickness, area, and volume for 180 regions of interest per hemisphere, in addition to the classic set of subcortical and cortical summary statistics from FreeSurfer (version 5.3) [57]. This yielded a total set of 1,118 structural brain imaging features (360/360/360/38 for cortical thickness/area/volume, as well as cerebellar/subcortical and cortical summary statistics, respectively). Linear models were used to residualise the T1-weighted MRI data with respect to scanning site, intracranial volume [58], and data quality using Euler numbers [59] extracted from FreeSurfer. To remove poor-quality data likely due to motion, participants with Euler numbers of standard deviation (SD) ± 4 were identified and excluded (n = 80). In addition, participants with SD ± 4 on the global MRI measures mean FA, mean cortical GM volume, and/or subcortical GM volume were excluded (n = 10, n = 5 and n = 4, respectively), yielding a total of 8,895 participants with both WM (diffusion-weighted) and GM (T1-weighted) MRI data.
2.3. Brain-age prediction
Brain-age prediction is a method in which a machine learning algorithm estimates an individual’s age based on their brain characteristics [60]. This estimation is then compared to the individual’s chronological age to estimate each individual’s brain-age gap (BAG), which is used to identify degrees of deviation from normative ageing trajectories. Such deviations have been associated with a range of clinical risk factors [37, 55, 61] as well as neurological and neuropsychiatric diseases [62, 63, 64, 36]. They have also been assessed in previous studies of parity and brain ageing [9, 10, 11, 54].
Separate brain-age prediction models were run for global WM and GM, and for each of the WM tracts using the XG Boost regressor model, which is based on a decision-tree ensemble algorithm (https://xgboost.readthedocs.io/en/latest/python). XGboost includes advanced regularisation to reduce overfitting [65], and uses a gradient boosting framework where the final model is based on a collection of individual models (https://github.com/dmlc/xgboost). For the global WM and GM models, principal component analyses (PCA) were run on the features to reduce computational time. The top 200 PCA components, explaining 97.84% of the total variance, were used as input for the WM model, and the top 700 components, explaining 98.07% of the variance, were used as input for the GM model. The model parameters were set to maximum depth = 4, number of estimators = 140, and learning rate = 0.1 for the the global and tract-specific WM models, and maximum depth = 5, number of estimators = 140, and learning rate = 0.1 for the global GM model, based on randomised searches with 10 folds and 10 iterations for hyper-parameter optimisation.
The models were run using 10-fold cross-validation, which splits the sample into subsets (folds) and trains the model on all subsets but one, which is used for evaluation. The process is repeated ten times with a different subset reserved for evaluation each time. Predicted age estimates for each participant were derived using the Scikit-learn library (https://scikit-learn.org), and BAG values were calculated using (predicted — chronological age). To validate the models, the 10-fold cross validations were repeated ten times, and average R2, root mean square error (RMSE), and mean absolute error (MAE) were calculated across folds and repetitions.
2.4. Statistical analyses
The statistical analyses were conducted using Python 3.7.6. All variables were standardised (subtracting the mean and dividing by the SD) before entered into the analyses, and p-values were corrected for multiple comparisons using false discovery rate (FDR) correction [66]. Chronological age was included as a covariate in all analyses, adjusting for age-bias in the brain age predictions as well as age-dependence in number of childbirths [67, 68].
2.4.1. Previous childbirths and global WM ageing
To investigate associations between number of previous childbirths and global WM brain ageing, a linear regression analysis was run using global WM BAG as the dependent variable, and number of childbirths as the independent variable. To control for potential confounding factors, the analysis was rerun including assessment location, education, ethnic background, body mass index (BMI), diabetic status, hypertension, smoking and alcohol intake, menopausal status (‘yes’, ‘no’, ‘not sure, had hysterectomy’, and ‘not sure, other reason’), oral contraceptive (OC) and hormonal replacement therapy (HRT) status (previous or current user vs never used), and experience with stillbirth, miscarriage, or pregnancy termination (‘yes’, ‘no’) as covariates. In total, 7,725 women had data on all variables and were included in these analyses. In addition, we tested for mean differences in global WM BAG between parous and nulliparous women using an independent samples t-test, and estimated Cohen’s d as effect size [69].
2.4.2. Previous childbirths and WM ageing versus GM ageing
To compare the contributions of global WM and GM brain ageing to the association with previous childbirths, a multiple regression analysis was run with both WM and GM based BAG estimates as independent variables and number of childbirths as the dependent variable, before eliminating one modality at the time to compare the log-likelihood of the full and reduced models. The significance of model differences was calculated using Wilk’s theorem [70] as , where ΔLL = LL1 – LL2; the difference in log-likelihood between the reduced model (LL1) and the full model (LL2).
Next, we tested for differences between the GM and WM BAG associations with number of previous childbirths using a Z test for correlated samples [71] with where m1 = model 1 (WM); m2 = model 2 (GM); β = beta coefficients from the linear regressions between number of childbirths and each model; σ = standard errors of the beta coefficients; ρ = age-adjusted correlation between the modality-specific brain age gap estimates.
2.4.3. Previous childbirths and regional WM tracts
To test for unique contributions by each tract to the global WM association with previous childbirths, a multiple regression analysis was run with all tract-specific BAG estimates as independent variables and number of childbirths as the dependent variable, before eliminating the tracts one at a time to compare the log-likelihood of the full and reduced models. The significance of model differences was calculated using Wilk’s theorem as described in Section 2.4.2. In addition, the reduced χ2 values for each of the models were calculated to account for the difference in number of input variables to the full and reduced models (13 for the full model including 12 tracts + age, versus 11 for the reduced models where each of the tracts were eliminated one by one). Next, we performed separate regression analyses for each tract-specific BAG estimate versus number of childbirths, before testing for differences between the associations using pairwise Z tests for correlated samples (Eq. 1; Section 2.4.2).
3. Results
The age prediction accuracies for the global WM and GM models, as well as each of the the tract-specific WM models are shown in Table 2. The associations between number of previous childbirths and BAG estimates based on each of the predictions are shown in Table 3.
3.1. Previous childbirths and global WM ageing
A higher number of previous childbirths was associated with lower global WM BAG, as shown in Table 3. The model including potential confounding factors showed a corresponding association of β = –0.04, SE = 0.007, t = – 5.10, p = 3.46 × 10−7, indicating that assessment location, education, ethnic background, BMI, diabetic status, hypertension, smoking and alcohol intake, menopausal status, and OC and HRT use could not fully explain the association between number of childbirths and global WM BAG. The mean difference in global WM ageing between parous and nulliparous women was t = 3.53,p = 4.15 × 10−4; effect size d = 0.09 ± 0.03 (d error).
3.2. Previous childbirths and WM ageing versus GM ageing
The age prediction based on the WM model showed higher accuracy compared to the GM prediction (R2 of 0.51 versus 0.32), as shown in Table 3. To directly compare the model predictions, a post-hoc Z test for correlated samples (Eq. 1; Section 2.4.2) was run on the model-specific fits of predicted versus chronological age (Pearson’s r values). The result showed a significant difference in model performance in favour of the WM model; Z = −11.90, p = 1.06 × 10−32.
When comparing regression models including both WM and GM-based BAG estimates to models including only one of the modalities, both the WM-based and the GM-based estimates were found to contribute uniquely to the association with number of previous childbirths, as shown in Table 4. The Z test for differences in associations (Eq. 1; Section 2.4.2) revealed similar associations between number of childbirths and WM-based versus GM-based BAG estimates, as shown in Table 5.
3.3. Previous childbirths and regional WM tracts
Significant (p < 0.05) associations between a higher number of previous childbirths and lower WM BAG estimates were found for ATR, CG, CING, FMIN, IFOF, SLFT, and CC, as shown in Table 3. The correlations between the tract-specific BAG estimates are shown in Figure 1. CC contributed uniquely to the global WM association with number of previous childbirths, as shown in Table 6. Pairwise Z tests for differences in associations revealed that ATR and FMIN had significantly stronger associations with previous childbirths compared to SLF, while CC was more strongly associated with previous childbirths than CST, CG, FMAJ, IFOF, ILF, SLF, UF, and SLFT, as shown in Figure 2.
4. Discussion
The current study investigated the association between previous childbirths and WM brain age by utilising global and region-specific brain-age prediction. The results showed that a higher number of previous childbirths was associated with lower brain age in global WM, as well as in WM tracts including ATR, CG, CING, FMAJ, FMIN, IFOF, SLFT, and CC. CC contributed uniquely to the global WM association with previous childbirths, and showed a stronger relationship with previous childbirths relative to several other tracts. When assessing global WM compared to GM brain age estimates, both modalities showed unique contributions to the association with previous childbirths. Taken together, these results indicate an association between previous childbirths and global WM ageing later in life, with regional effects that may be particularly prominent in CC.
4.1. Previous childbirths and global WM ageing
During pregnancy, several adaptations in the female body and brain take place in order to meet the needs and demands of the offspring, and to secure adequate expression of maternal caregiving [72]. Maternal adaptation in WM may thus be induced to meet these new demands, such as promoting myelination to ensure increased efficiency of neural transmission in relevant WM tracts. While speculative, our results may reflect a long-term benefit of pregnancy-induced WM plasticity, potentially promoting favourable WM trajectories later in life [73]. In support for long-term positive effects of childbirth on WM health, parity is associated with protective effects on age-related decline in learning, memory, and brain health in rats [74]. Further evidence for beneficial effects of parity on brain ageing stems from a study showing that telomeres are significantly elongated in parous relative to nulliparous women [7], suggesting that parity may slow the pace of cellular ageing.
The current results are also in line with previous studies in MS patients showing beneficial effects of pregnancy on WM health [14, 16, 17, 18, 20] and long-term disability progression following childbirth [20]. Oestradiol, a type of oestrogen that increases 300-fold during pregnancy [75], has been linked to pregnancy-induced MS remission [76], likely due to its anti-inflammatory and neuroplastic properties [77]. Postnatally and during the transition to menopause, oestradiol levels drop rapidly and may promote a pro-inflammatory immune environment [78], which has been linked to a high risk of relapse or worsening of symptoms in women suffering from MS [79, 80]. These findings suggest that high oestradiol levels may have protective effects on WM. Further evidence for this stems from hormonal replacement studies in postmenopausal women: long-term oestrogen use has been associated with greater WM volumes [81], indicating a protective effect on WM loss in ageing. However, oestrogen exposure has also been associated with GM atrophy [82] and higher rates of ventricular expansion in menopausal women [83], and some evidence suggests that genetic factors may influence how oestrogen exposure affects brain health [54, 84, 85]. Beside oestrogen, other hormones such as progesterone, prolactin, oxytocin, and cortisol also fluctuate during pregnancy and may regulate WM plasticity [86, 87, 88]. While the influence of hormone exposure on brain ageing trajectories is currently unclear, other pregnancy-induced adaptations such as the proliferation of regulatory T cells or fetal microchimerism may also represent mechanisms underlying potential long-term benefits of pregnancy on brain ageing (for a review see [86]). Future studies may target the links between hormone- and immune-related neuroplasticity in pregnancy, and the potential effect of these processes on women’s brain ageing trajectories.
4.2. Modality-specific and regional effects
In line with recent studies demonstrating high age prediction accuracy based on diffusion imaging data [37, 63, 89, 90, 91], the WM prediction showed higher accuracy compared to the GM model, of which the accuracy corresponded to our previous UK Biobank studies [10, 11, 54]. Importantly, we found unique contributions by both models, suggesting that the diffusion-based WM model may pick up variance not explained by the T1-based GM model. These findings highlight the relevance of assessing brain characteristics using different MRI modalities to increase our understanding of possible long-term effects of pregnancy on the brain.
The most prominent regional WM effect of childbirth was seen in the CC, showing both a unique contribution and a stronger association relative to several other tracts, potentially indicating regional variations. While the volume of most WM tracts increase from childhood to young adulthood, peaks around the fifties, and subsequently declines [28, 51, 92, 93, 94, 52], CC volume has been shown to peak already in the beginning of the thirties, exhibiting an earlier onset of age-related decline relative to other WM tracts [52]. Sex differences have also been found in CC ageing, with steeper volumetric declines in men relative to women [95]. While little is known about pregnancy-induced alterations in specific WM regions, an increased number of myelinated axons in the CC have been found in healthy pregnant rats [16], and increased CC remyelination has been observed in pregnant rat models of demyelination [16, 18]. Although speculative, our findings could potentially reflect a mitigating effect of parity on age-related CC volumetric decline. However, CC is also the most accessible WM structure to investigate, given its size and location in the brain, and the relative simple and coherent microstructural milieu may be easier to resolve using diffusion MRI than other pathways with more complex tissue structure. The tract extraction procedure could thus result in higher signal-to-noise ratio for the CC than for the remaining tracts, rendering it more sensitive to tests of WM associations with childbirth.
4.3. Study limitations
The cross-sectional design of the current study represents a major limitation, and longitudinal studies following women through pregnancy, postpartum, and into midlife and older age are required to infer causality between the observed associations. Furthermore, a complex interplay of numerous underlying processes likely influence the link between parity and WM trajectories. While the current study controls for a range of confounding factors including neurological disease, mental disorders, education, lifestyle behaviours, and cardiovascular risk, the number of children a women gives birth to - as well as their brain health across the lifespan - may also depend on genetic predispositions, life circumstances, and additional aspects of general health.
Our results could potentially reflect long-term effects of pregnancy-related processes such as myelination. However, the exact neurobiological underpinnings of diffusion metrics cannot be directly inferred, and although we utilised advanced diffusion modelling which is sensitive to biophysical tissue properties [34], the biological substrates underlying these metrics remain to be elucidated by future studies. In addition, controlling for the effect of extracellular water or indices of hydration [96] as well as including measures of WM hyperintensities [97, 98] could potentially provide more accurate models of WM ageing.
While the UK Biobank dataset enables detection of subtle effects due to its large sample size, the cohort is homogeneous with regard to ethnic background (97% white participants in the current study), preventing any conclusion about associations between reproductive history and WM ageing across ethnic groups. The cohort is also characterised by a “healthy volunteer effect” [99], suggesting that it is not representative of the general population [100]. Hence, the presented results may not apply to populations beyond those represented in this cohort. However, in context of the historical lack of research on women’s brain health [101], the current results represent a contribution that may prompt further study into how female biology influences neural processes involved in normal ageing - as well as autoimmune conditions and Alzheimer’s disease, of which the risks are higher for women relative to men [102, 103].
4.4. Conclusion
In summary, the current study found an association between a higher number of previous childbirths and lower WM brain age, in line with previous studies showing relationships between parity and brain characteristics in midlife and older age [9, 11, 15]. As outlined above, a complex interplay of numerous underlying processes likely influence the link between previous childbirths and brain health in older age. Thus, while our results may suggest that reproductive history influences women’s WM ageing trajectories, prospective longitudinal studies assessing this multi-factorial relationship are greatly needed to increase the knowledge about women’s brain health across the lifespan.
Data availability statement
The data that support the findings of this study are available through the UK Biobank data access procedures (https://www.ukbiobank.ac.uk/researchers).
Author contributions
I.V., L.T.W., and A-M.G.dL. designed the study; I.I.M., T.K., C.B., and A-M.G.dL. processed the data; I.V., D.B., and A-M.G.dL. performed the data analyses; I.V., C.B., I.I.M., T.K., D.B., G.R., T.M., L.T.W., and A-M.G.dL. interpreted the data; I.V. and A-M.G.dL. drafted and finalised the manuscript; C.B., I.I.M., T.K., D.B., G.R., T.M., and L.T.W. critically revised the first draft and approved the final manuscript.
Acknowledgements
This research has been conducted using the UK Biobank under Application 27412. The work was performed on the Service for Sensitive Data (TSD) platform, owned by the University of Oslo, operated and developed by the TSD service group at the University of Oslo IT-Department (USIT). Computations were also performed using resources provided by UNINETT Sigma2 – the National Infrastructure for High Performance Computing and Data Storage in Norway. While working on this study, the authors received funding from the Research Council of Norway (I.V; Student scholarship; AM.G.dL.; 286838, T.K.; 276082; L.T.W.; 273345, 249795, 298646, 300768, 223273; C.B.; 250358), the South-East Norway Regional Health Authority (L.T.W.; 2018076, 2019101), and the European Research Council under the European Union’s Horizon 2020 research and innovation programme (L.T.W.; 802998).
References
- [1].↵
- [2].↵
- [3].↵
- [4].↵
- [5].↵
- [6].↵
- [7].↵
- [8].↵
- [9].↵
- [10].↵
- [11].↵
- [12].↵
- [13].↵
- [14].↵
- [15].↵
- [16].↵
- [17].↵
- [18].↵
- [19].↵
- [20].↵
- [21].↵
- [22].↵
- [23].↵
- [24].↵
- [25].↵
- [26].↵
- [27].↵
- [28].↵
- [29].↵
- [30].↵
- [31].↵
- [32].↵
- [33].↵
- [34].↵
- [35].↵
- [36].↵
- [37].↵
- [38].↵
- [39].↵
- [40].↵
- [41].↵
- [42].↵
- [43].↵
- [44].↵
- [45].↵
- [46].↵
- [47].↵
- [48].↵
- [49].↵
- [50].↵
- [51].↵
- [52].↵
- [53].↵
- [54].↵
- [55].↵
- [56].↵
- [57].↵
- [58].↵
- [59].↵
- [60].↵
- [61].↵
- [62].↵
- [63].↵
- [64].↵
- [65].↵
- [66].↵
- [67].↵
- [68].↵
- [69].↵
- [70].↵
- [71].↵
- [72].↵
- [73].↵
- [74].↵
- [75].↵
- [76].↵
- [77].↵
- [78].↵
- [79].↵
- [80].↵
- [81].↵
- [82].↵
- [83].↵
- [84].↵
- [85].↵
- [86].↵
- [87].↵
- [88].↵
- [89].↵
- [90].↵
- [91].↵
- [92].↵
- [93].↵
- [94].↵
- [95].↵
- [96].↵
- [97].↵
- [98].↵
- [99].↵
- [100].↵
- [101].↵
- [102].↵
- [103].↵