Alzheimer disease (AD) is the most common dementing disorder in Western societies, accounting for about two-thirds of all dementia cases. There are two different aetiological forms of AD. The more common sporadic form is a multifactorial disease with both genetic and environmental factors contributing to the disease process. There are also early-onset, familial forms of AD with autosomal dominant inheritance, which account for 5% of AD cases.1 The most clearly established risk gene for both sporadic and familial AD is apolipoprotein epsilon.2

Monozygotic (MZ) twins have all genes in common, whereas dizygotic (DZ) twins share on average half of their segregating genes. These genetic differences can be used in large samples to estimate the relative contribution of genetic and environmental factors on multifactorial disorders. A higher concordance of AD has been found in MZ than DZ twins in studies based on large twin cohorts,3^–6 implicating the role of heritability. However, the discordance in MZ pairs demonstrate that environmental factors also contribute; these factors can potentially include prenatal and epigenetic effects.7 8

Besides the pathological changes including plaques and tangles, metabolic changes are seen in AD as well. These include changes in neurotransmitter systems9 and glucose metabolism. 2-Deoxy-2[18F]-fluoro-d-glucose (FDG) is a tracer used to measure cerebral glucose metabolic rate (GMR) in positron emission tomography (PET).10 In AD patients, cerebral GMR has been found to be extensively reduced with relative sparing of the primary sensory and motor cortices.11 Patients with mild cognitive impairment (MCI)12 have lowered GMR in the hippocampus and the posterior cingulate, and GMR reductions in inferior parietal cortex and posterior cingulate gyrus predict MCI patient’s rapid progression to AD.13^–16

Besides MCI patients, FDG-PET can be applied to studying twin pairs discordant for AD. The metabolic and structural changes detected by neuroimaging in AD and MCI patients can be part of the disease process, or they can reflect characteristics of persons with a genetic liability to the disease. With twin pairs discordant for AD, one can attempt to distinguish disease characteristics from indicators of genetic liability to AD. If the neuroimaging changes seen in AD patients are absent in the unaffected MZ and DZ co-twins of AD patients, these changes can be attributed to the disease process itself. On the other hand, if the findings are present in these co-twins, it would suggest that they reflect a familial liability to disease. Further, if the changes in the MZ co-twins resemble the changes in AD patients more than those of the DZ co-twins, this would suggest that the familial liability is genetic. On the other hand, if MZ and DZ co-twins of AD patients do not differ (but have more changes than unrelated healthy controls), the liability would be due to an acquired familial effect, such as an exposure in childhood affecting both siblings.

Previous FDG-PET studies including one to three MZ twin pairs discordant for AD have shown controversial results.17^–19 A more recent study including seven female MZ twin pairs discordant for AD showed reduced GMR in the non-demented co-twins temporoparietally, but to a lesser degree than in the demented co-twins.20 The finding can be an indication of early, preclinical AD pathogenesis or an increased familial risk of AD. In a similar study including nine DZ twin pairs discordant for AD, no differences were found between non-demented co-twins and controls21 based on a region of interest (ROI) analysis.

In this study, we examined cerebral GMR in seven MZ and nine DZ twin pairs discordant for AD using voxel-based image analysis methodology. Because of the differences in the genetic resemblance between MZ and DZ twins, the inclusion of both MZ and DZ twins in the same study allows a more powerful means to examine cerebral GMR as an indicator of genetic liability versus disease process indicator AD. The voxel-based analysis was used as it gives more explorative and objective information compared with the manual ROI-based analysis. As statistical testing is done at the voxel level covering the whole brain space, not just the data in the ROIs, no a priori hypothesis about the locations of the possible differences between subject groups is required. To obtain quantification of regional glucose metabolism, an automated ROI analysis conducted in a standard brain space was performed. As this method is based on spatial normalisation of brain images, it eliminates the possibility of observer induced bias caused by manually drawing the ROIs separately for each subject.

METHODS

Subjects and controls

All same-sexed DZ twins born 1937 or earlier and all MZ twins born 1934 or earlier from the older Finnish Twin Cohort22 23 were invited to participate in a telephone interview evaluating their cognitive functions, the TELE,24 which has been shown to be able to differentiate between AD patients and healthy controls.25 The twins’ zygosity was assessed by a validated questionnaire,26 and monozygosity was confirmed by genotyping of multiple genetic markers.20 Twin pairs discordant for dementia according to TELE were asked to participate in a detailed neuropsychological examination and an FDG-PET study at the Turku PET Centre; seven MZ twin pairs and nine DZ twin pairs were included in the present study. Co-twins’ discordance in cognitive performance was confirmed with a neuropsychological test battery measuring episodic and semantic memory, attention, language and visuospatial abilities. Two of the cognitively more impaired MZ co-twins and one of the cognitively less impaired MZ co-twins were found to fulfil the criteria for MCI. MCI was diagnosed according to the criteria suggested by Petersen et al.12 All patients had memory impairment and possibly mild decline in other cognitive domains, the clinical dementia rating (CDR) was 0.5, global cognition was normal, mild changes in activities of daily living were allowed, but no subject had dementia. For simplicity, the cognitively less declined co-twins are referred as non-demented and the cognitively more declined co-twins as demented, respectively. MRI scans were also performed with a 1.5 T scanner on all subjects to screen for possible brain lesions.

The control group consisted of 13 healthy, unrelated volunteers who were matched to the twins at group level for age and sex. The controls had no history of neurological or psychiatric diseases or abnormal findings in clinical neurological examination. The ages, sex distributions and Mini Mental State Examination (MMSE) scores of the twins and controls are shown in table 1 and a scatter plot of the MMSE scores in fig 1. The differences in MMSE scores between non-demented MZ and DZ co-twins or between demented MZ and DZ twins were not statistically significant. The study protocol was approved by the Joint Ethical Committee of the University of Turku and the Turku University City Hospital.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1 Scatter plot of the Mini Mental State Examination (MMSE) scores of the gropus. The cloud is jittered. DZ, dizygotic; MZ, monozygotic.

Positron emission tomography (PET) imaging

FDG-PET imaging was performed in all subjects with a GE Advance PET scanner (General Electric Medical Systems, Milwaukee, Wisconsin). Axial FOV was 15.4 cm consisting of 35 planes. After reconstruction, the final image resolution was 6–7 mm. Attenuation was performed with two rotating rod sources (nominal activity 400 MBq each), and iterative segmented attenuation correction was used. Cerebral GMR was determined using a Gjedde–Patlak plot27 according to a protocol reported previously.21 Written informed consent was obtained from all subjects before PET imaging. The mean injected activity was 257.4 (45.9) MBq. The PET data of the MZ20 and DZ21 twins have been previously published using manually drawn ROIs.

Statistical parametric mapping (SPM) analysis

Preprocessing of imaging data and voxel-based statistical analysis were performed using SPM2 (Wellcome Department of Imaging Neuroscience, University College London, London) and Matlab 6.5 (MathWorks, Natick, Massachusetts). Spatial normalisation of GMR images was performed with a ligand specific template being prepared with a MRI-aided procedure. The ligand-specific template approach seems to be an optimal and widely used procedure for spatial normalisation, and it is described in detail elsewhere.28 In brief, the FDG template was generated from 14 [¹⁸F]FDG scans from healthy subjects that were not included the present study. First, individual MRI images were coregistered to summated [¹⁸F]FDG images, and then MRI images were normalised using T1 weighted MRI template delivered with SPM2. Normalisation parameters were finally used to create normalised [¹⁸F]FDG summated images written onto the template bounding box. The [¹⁸F]FDG template image was calculated as a mean of these images. To ensure the symmetry of the template, it was averaged with its mirror copy (left–right flipped), and finally the template was smoothed using an 8 mm Gaussian kernel as 8 mm smoothing is used in SPM2 standard normalisation estimation.

The twins’ and controls’ parametric GMR images were spatially normalised to MNI (Montreal Neurological Institute) space with SPM2 using summated images and [¹⁸F]FDG template described above. The normalised images were written using bilinear interpolation and smoothed using 12 mm Gaussian kernel.

The voxel-based comparison between the groups was performed for the whole brain without any a priori hypothesis about possible location of the possible differences between groups. Groups were compared using SPM2 with two different one-way ANOVAs: one including the controls and the non-demented MZ and DZ co-twins, and the other including the controls and all the demented co-twins pooled into together. The pooling was done as it can be assumed that both the MZ and DZ demented co-twins manifest the same disease process. It allowed us to include more subjects in the same analysis, which increases its statistical power. In all SPM analyses, cluster level interpretation was utilised, and p<0.05 (corrected at cluster level) was considered significant. Also, a one-way ANOVA was performed comparing the demented MZ and DZ co-twins separately with the controls.

Automated ROI analysis

An automated ROI analysis was used to quantitatively determine the regional glucose metabolic rate (rGMR) in different brain regions; the methodology of the analysis is described in detail elsewhere.29 The mean image of the 11 spatially normalised MRIs was used as anatomic reference in drawing the ROIs to ensure a common stereotaxic space. ROIs were drawn bilaterally and symmetrically to frontal cortex (Brodmann areas (BA) 10 and 46), lateral temporal cortex (BA 20–22), parietal cortex (BA 19 and 39), occipital cortex (BA 17–18), anterior and posterior cingulate gyrus, medial temporal lobe (including hippocampus and amygdala), caudate, putamen, thalamus and cerebellar cortex using Imadeus Academic 1.20 (Forima, Turku, Finland). As the same set of ROIs was used for all subjects, this method eliminates the operator-induced bias resulting from drawing the ROIs separately for each subject. The rGMR values were calculated from the spatially normalised parametric GMR images using these ROIs. The rGMR values are presented in µmol/ml/min.

The rGMR valus were statistically analysed with SPSS 13.0 for Windows (SPSS, Chicago). The controls, demented co-twins, non-demented MZ co-twins and non-demented DZ co-twins were compared with linear mixed models. Again, the demented co-twins were pooled together. Repeated-measures analysis was used with twin pair as a random effects variable to account for the possible correlation between co-twins’ rGMR values. Differences with p<0.05 were considered significant. rGMR values were analysed separately for both hemispheres in ROIs where asymmetry was significant in a paired samples t test.

RESULTS

In the voxel-based SPM one-way ANOVA including the controls as well as MZ and DZ non-demented co-twins, a statistically significant reduction in GMR was seen bilaterally in the inferior frontal, the lateral temporal, the parietal and the medial temporal cortices in the non-demented MZ co-twins compared with controls. Also, in the thalamus, putamen and right amygdala significant differences were found (fig 2, tables 2, 3). In the lateral temporal cortex the statistical significance was greater in the left hemisphere. No statistically significant differences were found between the controls and non-demented DZ co-twins.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2 Presentations of the voxel-based glucose metabolism analyses between groups using statistical parametric mapping (SPM). The palette from red to yellow indicates regions where statistically significant differences between groups were found; yellow indicates a more significant difference. The comparison between controls and non-demented dizygotic (DZ) co-twins is not shown, as no significant differences were found. (A) Controls versus non-demented monozygotic (MZ) co-twins, p<0.001 corrected at cluster level. (B) Controls versus demented co-twins pooled together, p<0.001 corrected at cluster level. (C) Controls versus demented MZ co-twins, p<0.001 corrected at cluster level. (D) Controls versus demented DZ co-twins, p<0.001 corrected at cluster level.

In the one-way ANOVA including the controls as well as the MZ and DZ co-twins pooled together, significant reductions in GMR were found extensively in the frontal, parietal, lateral temporal and occipital cortices, and in the medial temporal lobe in all demented co-twins compared with the controls. In addition, significant differences were seen in limbic structures and subcortical nuclei. In the lateral temporal cortex, the significance was statistically greater in the left hemisphere than in the right hemisphere. In the primary sensory–motor or visual cortices no significant differences were found. Visualisations of the differences between controls and twin groups are shown in fig 2, and a summary of the SPM results of the non-demented and demented co-twins is provided in tables 2, 3, respectively.

An automated ROI analysis was used to quantitatively determine the rGMR. In the non-demented MZ co-twins, rGMR was most markedly reduced as compared with controls in the parietal cortex, where the rGMR in the non-demented MZ co-twins was 14% lower than in the controls. rGMR was also significantly lower in the lateral temporal cortex, the putamen and the thalamus. No statistically significant differences were found between the non-demented DZ co-twins and the controls. In the demented co-twins, rGMR values were significantly lower than in controls in every ROI except occipital cortex and cerebellar cortex. The difference was greatest in the caudate (32% lower than controls), putamen (20% lower), thalamus (20% lower), posterior cingulate (19% lower) and parietal cortex (19% lower). The mean rGMR values of the twins and the controls are shown in table 4. In addition to the original ROIs, we created additional bilateral ROIs to the orbitofrontal cortex and posterolateral parieto-frontal cortex (PFC) guided by the results of the SPM analysis between the controls and non-demented MZ co-twins. The rGMR of the non-demented MZ co-twins was reduced in these areas, but no statistically significant differences compared with the controls were found, possibly because of differences between the statistical methods in the SPM and automated ROI analysis. The results of the additional ROIs are shown in table 5.

One of the non-demented MZ co-twins had MCI, so we performed a t test between the non-demented MZ co-twins and the controls for the quantitative GMR values excluding this subject. The results were consistent with the analysis including all subjects, as a significant difference was found in lateral temporal and right parietal cortices as well as in the caudate, putamen and thalamus. Because the MZ twin group consisted only of women, we did a subgroup automated ROI analysis including only female subjects from all groups: nine controls, all seven MZ twin pairs and six DZ twin pairs. Their age or MMSE-score distributions did not differ from the entire sample. The auto-ROI analysis did not differ either except for the putamen, where no significant difference between controls and non-demented MZ co-twins was found in the subgroup analysis.

We also performed one-way ANOVAs for the automated ROI data comparing the controls with both demented MZ and DZ co-twins. In the analysis comparing the controls with both demented co-twin groups separately, the results were similar to those from the repeated-measures analysis. The comparison between the demented DZ co-twins and controls yielded significant differences in every ROI, whereas the analysis including the controls and demented MZ co-twins resulted in similar findings to those from the repeated-measures analysis. In the SPM analysis the demented DZ co-twins have slightly more extensive areas of significant reductions than the demented MZ co-twins (fig 2). When the results were reanalysed using the cerebellar vermis as a reference, no significant differences were seen in the non-demented MZ or DZ groups, but the findings remained significant in the demented co-twins.

To estimate the influence of the degree of genetic similarity to the AD patients on cerebral glucose metabolism, we did a regression analysis between genetic load and rGMR values in ROIs affected by AD. In line with an additive model of gene action, the genetic load was determined as 0 for controls, 0.5 for DZ non-demented co-twins and 1 for MZ non-demented co-twins. The regression was significant in the lateral temporal cortex, where the multiple correlation coefficient was −0.41 (p = 0.03), and in the thalamus, where the coefficient was −0.40 (p = 0.03). All correlations are shown in table 6.

DISCUSSION

We found the cerebral GMR to be extensively reduced in the demented co-twins (demented MZ and DZ co-twins pooled together) as compared with controls. The reduction was greatest in the subcortical nuclei and the posterior cingulate, and in the medial temporal lobe and the frontal, lateral temporal and parietal cortices. In contrast, no significant reductions were found in the primary sensory–motor or primary visual cortex. The findings in the cerebral cortex and medial temporal lobe are consistent with previous FDG-PET studies on AD patients.11 30^–32 As the subcortical GM regions are anatomically small structures, the atrophy associated with AD combined with the partial volume effect could account for the significant reductions found in these structures. Also, in the voxel-based analysis the displacement of subcortical structures due to ventricular enlargement can result in artefact hypometabolism in these structures. In the demented DZ co-twins, the reductions as compared with controls were somewhat more extensive than in the MZ co-twins, which could be a reflection of the slightly lower MMSE scores found in the demented DZ co-twins.

In the non-demented MZ co-twins, the cerebral GMR was reduced as compared with controls in the inferior frontal cortex, the lateral temporal cortex, the parietal cortex, the right amygdala, and the thalamus and putamen. However, as significant differences between the controls and non-demented MZ-twins were found in the SPM but not in the ROI analysis in orbital frontal cortex and cerebellum, it is possible that the SPM findings in these regions might be artefactual. In contrast to the non-demented MZ co-twins, no reduction was found in the non-demented DZ co-twins in any cerebral region. As the MMSE scores of the non-demented MZ and DZ co-twins were practically equal, it is unlikely that the different findings reflect a difference in the general level of cognitive functioning between the non-demented co-twin groups. However, one of the non-demented MZ co-twins met the criteria for MCI, but excluding this subject from analyses yielded similar results. The MZ twin pairs were all female, but as the results of the subgroup analysis including only women did not differ from the results including all subjects, the different findings cannot be explained by differences in sex distributions between the MZ and DZ twins. The functional changes found in non-demented MZ co-twins seem to precede structural changes, since in the non-demented MZ twins no hippocampal atrophy was seen,33 although glucose consumption was impaired.

As stated above, the results based on manual drawing of ROIs of the twins have been previously published (MZ twins20 and DZ twins21). However, this is the first FDG-PET study on twin pairs discordant for AD in which both MZ and DZ twin pairs were studied. Also, a voxel-based analysis was performed, which is an objective method without any possible operator-induced bias. Furthermore, an automated ROI analysis was also performed, enabling a quantitative comparison of the MZ and DZ data, since the same ROIs applied to spatially normalised images are used for both of these twin groups. Our results are consistent with the previous studies using manual ROI drawing. In these studies, GMR was found to be reduced temporo-parietally in the non-demented MZ co-twins, but no reduction was found in the non-demented DZ co-twins. In addition to the differences found previously by manual ROI drawing, we found the glucose metabolism to be more extensively reduced in the non-demented MZ co-twins, which may reflect the advantages of voxel-based analysis, as the voxel-based analysis included brain regions not covered by the manual ROI drawing. The elimination of observer-induced bias can result in detection of more real differences between subject groups, but unfortunately it is also possible that some of the differences found in the SPM analysis may be artefactual. However, it may also result from different ROI drawing or statistical analyses. Also, the SPM analysis allowed us to create additional ROIs guided by the SPM results. This allowed us to quantitatively analyse the areas outside the original ROIs in which the GMR seemed to be reduced. As the whole brain volume was analysed, our results confirm that the non-demented DZ co-twins have no reductions in GMR. Using traditional ROI-based methodology, some areas of reduced GMR could have been missed, if they were located outside the ROIs.

The identical analysis for both twin groups enables evaluation of the possible genetic versus environmental contribution to FDG uptake and susceptibility to develop AD. A reduction in cerebral GMR was found in the non-demented MZ co-twins but not in the DZ co-twins. It can be assumed that the finding is caused by different characteristics of MZ and DZ twin pairs. It is presumed that both MZ and DZ twins have equally shared environmental factors but differ in genetic relatedness. Thus, the different findings in the non-demented MZ and DZ co-twins GMR could be attributed to genetic factors. These factors seem strong enough to affect MZ twins only. The relatively small sample size is unlikely to explain why no reductions were found in the non-demented DZ co-twins, since the GMR values were virtually identical to those of controls. The small number of subjects reflects the infrequency of twin pairs discordant for AD, as hundreds of twins were interviewed for this study.

As the distribution of the lowered GMR in the non-demented MZ co-twins is not entirely typical for early AD, it can be stated that the non-demented MZ co-twins possess a genetically determined background hypometabolism that might be related to vulnerability for AD. The finding that no significant differences were found in cerebellar-corrected GMR between non-demented MZ co-twins and controls further supports this. Hence, it is possible that the genetic predisposition for AD could be a lower resting GMR. However, the confirmation of this requires testing larger samples in the future.

The heritability of cerebral GMR has been studied in only seven female MZ twin pairs.34 In that study, significant intraclass correlations were found between co-twins’ GMR in prefrontal and orbital frontal cortex, as well as in the basal ganglia. These correlations probably reflected the same metabolic phenomena, as the correlations were not independent but depended on the correlation between GMR in other brain regions. We found the genetic load for AD to affect the GMR relatively strongly in the putamen, caudate and thalamus, even though the correlation was statistically significant only in the thalamus. We also found a significant correlation between genetic dose and GMR in the lateral temporal cortex. The GMR of all these areas was also markedly reduced in the demented co-twins, and hence the correlation can be seen as an indication of a specific vulnerability for AD.

In conclusion, we found the cerebral glucose metabolism to be lowered in the demented co-twins in cerebral regions affected by AD. Also, the non-demented MZ co-twins showed reductions in GMR in several cortical regions and cingulate gyrus, suggesting that these changes are early indicators of the disease process before onset of clinical dementia or are genetic markers of susceptibility to AD. In contrast, no reductions in cerebral GMR were found in the non-demented DZ twins, which may be due to genetic differences. To establish whether the changes in the non-demented twins represent an early pathological process of AD, a follow-up study on the non-demented co-twins would be required.