We selected participants from different pre-existing longitudinal cohorts at the Memory Clinic of the Division of Psychiatry Research and Psychogeriatric Medicine, University of Zurich. Briefly, participants were recruited from the outpatient population of the Memory Clinic or through advertisements in the local media. HC were retrospectively and additionally recruited through inquiries of caregivers or relatives of the patients. MCI was diagnosed according to Winblad et al. (2004), based on performance in multiple tests covering the following cognitive domains: episodic memory, executive function, attention/psychomotor processing speed, language and visual-constructive abilities. Impairment was defined if at least one score per domain was 1.5 SD below group means provided by test-specific normative data. Conversion to dementia was diagnosed when clinical work up indicated progression and significant impairment in activities of daily living. This was assessed by a multidisciplinary team under the supervision of an experienced psychiatrist.

For the present study, inclusion criteria for MCI subjects were: amnestic MCI (single or multiple domain) diagnosis and availability of MRI data at baseline and follow-up. Exclusion criteria were: left-handedness, significant medication or drug abuse as well as clinically significant neurological and psychiatric or internal disease that may affect cognition, MRI findings of infarction or other focal lesions, multiple lacunes or lacunes in critical memory structures. A total of 33 baselines from subjects with amnestic MCI were considered for the present study. The population was stratified into subjects with stable cognitive abilities during an approximately 2-year follow-up (MCI-S, n = 23), and subjects with future cognitive worsening and conversion to probable AD (MCI-CB, n = 10) within a 2-year time frame during follow-up. Additionally, data from the MCI-CB group at time of conversion was obtained (MCI-CC, n = 10). Inclusion criteria for HC were: stable cognitive health ascertained by clinical work up and neuropsychological testing during an approximately 2-year follow-up. Exclusion criteria were: MRI exclusion criteria, left-handedness, evidence for abuse of alcohol and drugs, psychiatric, neurological or significant other system diseases. Three groups of HC were identified for group wise age and gender matching with MCI-S, MCI-CB, and MCI-CC, and MRI data was acquired following the description in section Magnetic Resonance Image Acquisition. The final demographic details are presented in Table 1. This study was approved by the cantonal ethics committee of canton Zurich, Switzerland, in accordance with the Helsinki Declaration. All participants provided written informed consent prior to study inclusion.

All (MRI) were performed on the same 1.5 Tesla Phillips Achieva scanner using an 8-element head coil. Whole-brain high-resolution 3D T₁-weighted structural data was obtained by using the following scanning parameters: 166 slices, repetition time: 6.9 ms, echo time: 3.2 ms, flip angle: 8°, field of view: 240 × 240 × 166 mm (anterior-posterior, foot-head, right-left), slice thickness: 1 mm, total scan time: 15 min.

Segmentation of the striatum, thalamus and thalamic nuclei was performed using a recently developed label-fusion-based segmentation method that had previously proven its high accuracy (Chakravarty et al., 2013). Briefly, the MAGeT-Brain algorithm applies multiple automatically generated templates from a single atlas derived from manually segmented serial histological data comprising 108 basal ganglia and thalamic structures as defined using three different references (Schaltenbrand and Wahren, 1977; Hirai and Jones, 1989; Gloor, 1997). We used two of the segmentations produced from the MAGeT-Brain pipeline, the first are the whole striatum (caudate and putamen) and thalamus, and the second are the thalamic subnuclei as per the Hirai and Jones definitions (1989). The thalamus was segmented into pulvinar-, anterior-, and central nuclei and lateral dorsal-, lateral posterior-, medial dorsal nuclei, ventral anterior nuclei (VA), ventral lateral nuclei (VL), ventral posterior nuclei (VP) and lateral geniculate nucleus (LGN) and medial geniculate nucleus (MGN) as per the Hirai and Jones (1989) nomenclature. Segmentation of the hippocampus and its subfields was performed using five high-resolution atlases developed and validated for use with MAGeT-Brain (Winterburn et al., 2013; Pipitone et al., 2014). The hippocampus was segmented into cornu ammonis (CA) 1, CA2-CA3, CA4/Dentate gyrus, strata radiatum/lacunosum/moleculare, and subiculum.

Striatal and thalamic shape analysis was performed by using an adapted surface-based methodology (Magon et al., 2014; Raznahan et al., 2014; Shaw et al., 2015). Briefly, surface-based representations of the striatum and thalamus were defined on the input atlas. The nonlinear portions of the transformations that map each subject to the input template were concatenated and then averaged to limit the effects of noise and error and to increase precision and accuracy. Next, the dot product between the nonlinear deformation vector (of the inverse of the averaged atlas-to-subject transformation) and the surface normal at each vertex (a unit vector describing the direction perpendicular to the surface) was estimated. This measure provides an estimate of the local measure of inward or outward displacement along the normal. Then, surface-area values were blurred using a 5 mm surface-based diffusion smoothing kernel (Raznahan et al., 2014; Chakravarty et al., 2015). Resulting inward and outward displacements (measured in millimeters) were estimated relative to a detailed subcortical atlas previously described (Chakravarty et al., 2015). An inward displacement (contraction) represents a surface that is deformed inwards relative to the model that we were using and vice-versa for the outward displacement (expansion).

Cortical thickness was estimated by using the automated CIVET pipeline (version 1.1.10; Montreal Neurological Institute at McGill University, Montreal, Quebec, Canada). Briefly, the native images were linearly registered to the symmetric ICBM 152 template (Collins et al., 1994; Mazziotta et al., 2001). Intensity nonuniformities were corrected using the N3 algorithm (Sled et al., 1998). The skull was removed (Smith, 2002), and brain tissue was segmented into white matter, gray matter, cerebrospinal fluid (CSF) using the Intensity—Normalized Stereotaxic Environment for Classification of Tissues (INSECT) algorithm (Zijdenbos et al., 1998; Tohka et al., 2004). Deformable models were used to construct the inner white matter surface and gray matter-CSF interface in both hemispheres (Kim et al., 2005) revealing 40,962 vertex points at each surface. Cortical thickness was then measured as the distance, in millimeters, between each vertex point at the inner and the corresponding point at the outer surface using the method proposed by Lerch and Evans (2005). The cortical thickness maps were blurred using a 20 mm diffusion smoothing kernel to increase signal-to-noise ratio and statistical power.

The comparison of gray matter volumes across groups requires taking into account interindividual variability in brain morphology. Values of intracranial volume (ICV) indicate premorbid brain volume, and thus are often used to adjust volumes for subsequent volume analyses. When examining volume reductions in neurodegenerative disease, considering ICV allows for estimation of atrophy caused by neurodegenerative mechanisms rather than by interindividual differences in head size and brain morphology.

In the present work, the FreeSurfer pipeline (version 5.1.0) was used to calculate total intracranial volume (eTIV) representing an estimate for ICV as described in Buckner et al. (2004). Briefly, each individual is registered to an atlas template. The Atlas Scaling Factor obtained by this transformation represents the whole-brain volume adjustment that is required to match each individual to the atlas template and is thus used to automatically generate eTIV. This automated method has been shown to be equivalent to manual correction (Buckner et al., 2004) and has previously been used for normalization in several AD studies (Westman et al., 2011, 2012, 2013).

Group comparisons of demographic and conitive data were applied using analysis of variance (ANOVA) or Mann-Whitney U-Test. Pearson's chi-square test was used for categorical variables. Tests were performed with a significance level of p < 0.05. Between-group differences in volumetric raw data (MCI-S, MCI-CB and MCI-CC vs. matched HC) were examined by including age and gender as covariates in the multiple linear regression models. These analyses were repeated by using volumes relative to eTIV (volume/eTIV^*100) in order to adjust volumes for differences in head size. P-values resulting from volume analyses were adjusted for multiple testing by using Bonferroni-Holm correction (level of significance for hippocampus, striatum and thalamus starting with p < 0.05/2; level of significance for hippocampal subfields starting with p < 0.05/10). The same models were performed for investigating between-group shape and cortical thickness differences. Vertex-wise analyses results are reported on a q-value corrected for multiple testing, using a false discovery rate (FDR) of 10% as in previous publications in our group (Wheeler et al., 2013; Janes et al., 2014). Receiver operating characteristic (ROC) curves were computed to evaluate and compare the accuracy of striatal and thalamic shape alterations, and of hippocampus total volumes as established AD imaging marker for discriminating HC from MCI-CB. ROC curves are produced by plotting the true positive rate (sensitivity) against the false positive rate (1-specificity) for different thresholds. The area under the curve (AUC) is then calculated and provides information about the ability of the morphometric data to discriminate between patients and HC. AUC values of 1.0 indicate perfect discriminative abilities; values of < 0.6 indicate poor discriminative abilities. For these analyses, between group t statistics from comparison between HC and MCI-CB were used. More precisely, for all vertices whose t-values constituted a local minimum (or maximum), the average of the displacement values in mm was computed for each subject. This was done separately for each structure and hemisphere. Statistical analyses were performed with IBM SPSS statistics 21 and RMINC package (R for Medical Imaging NetCDF; https://github.com/Mouse-Imaging-Centre/RMINC), an image analysis software library developed for the R statistical environment (http://www.r-project.org).

Apart from reduced bilateral striatal volumes, and when analyzing volumes relative to eTIV, there were no volume differences in any of the investigated structures in MCI-S when compared with HC (see Table 2). Significance and p-values were similar when using raw volumes instead of volumes relative to eTIV (see Table 3). A segmentation map of the thalamus is shown in Figure 1, and of the hippocampus in Figure 2.

In contrast, pronounced reductions in volumes relative to eTIV were found in MCI-CB (see Table 2). In particular, and apart from CA2-CA3 volumes, all bilateral hippocampal subfield volumes were smaller in MCI-CB compared to HC (right CA1 t = 3.23, p = 0.005; subiculum t = 3.29, p = 0.005; CA4/Dentate gyrus t = 3.80, p = 0.002; strata t = 4.55, p < 0.001/left CA1 t = 5.18, p < 0.001; subiculum t = 4.96, p < 0.001; CA4/Dentate gyrus t = 4.66, p < 0.001; strata t = 5.91, p < 0.001; df = 3,16) after correction for multiple testing. With regard to thalamic subnuclei, and although statistically significant, effects of volume reductions in bilateral VP in MCI-CB (right p = 0.047, left p = 0.015) did not survive the correction for multiple testing. Significance and p-values were similar when using raw volumes instead of volumes relative to eTIV. However, some of the right hemispheric differences in hippocampal subfield volumes did not quite achieve the level of significance (CA1 p = 0.057). Additionally, and although statistically significant, two right hemispheric effects did not survive the correction for multiple comparisons (CA4/Dentate gyrus p = 0.017; subiculum p = 0.027) (see Table 3).

Further extended reductions in volumes relative to eTIV revealed in MCI-CC (see Table 2). More precisely, and after correction for multiple testing, all bilateral hippocampal subfield volumes were smaller in MCI-CC compared to HC (left CA1 t = 5.09, p < 0.001; subiculum t = 5.00, p < 0.001; CA2-CA3 t = 4.01, p = 0.001; CA4/Dentate gyrus t = 6.66, p < 0.001; strata t = 9.41, p < 0.001/right CA1 t = 3.24, p = 0.005; subiculum t = 3.91, p = 0.001; CA2-CA3 t = 2.18, p = 0.044; CA4/Dentate gyrus t = 4.47, p < 0.001; strata t = 5.53, p = 0.000; df = 3,16). Similar as in MCI-CB, effects of VP volume reductions in MCI-CC (right p = 0.054, left p = 0.009) did not survive the correction for multiple testing. Significance and p-values were similar when using raw volumes instead of volumes relative to eTIV. However, two right hemispheric effects in hippocampal subfield volumes did not quite achieve the level of significance (CA1 p = 0.074; CA2-CA3 p = 0.057) (see Table 3).

Analyses on data corrected for multiple testing by using FDR at q < 0.10 revealed no cortical thinning in MCI-S compared to HC. Reduced cortical thickness, however, was found in MCI-CB and MCI-CC compared to HC (Figure 3, Table 4). Significant effects were limited to medial areas such as the left parahippocampal cortex, left subgenual cingulate, and left region of the uncus in MCI-CB. Similar regions revealed cortical thinning in the right hemisphere, with significance only at q < 0.15 though (Figure 4). Importantly, the pattern of cortical thinning extended to the right hemisphere in MCI-CC, where cortical thinning in bilateral parahippocampal cortices and bilateral regions of the uncus now achieved an appropriate level of significance.

Analyses on data corrected for multiple testing by using FDR at q < 0.10 revealed no striatal or thalamic shape alterations in MCI-S, but pronounced striatal and thalamic displacements in MCI-CB and MCI-CC compared to HC.

Thalamic contractions and expansions are presented in Figure 4A. In contrast to MCI-S, MCI-CB, and MCI-CC revealed contractions which were limited to dorsal and medial parts, and were more pronounced in the left than in the right hemisphere. Specifically, MCI-CB exhibited contractions in bilateral dorsal aspects of the pulvinar, bilateral dorsal aspects of VP, and in left medial aspects of VP and medial dorsal nuclei. Again, the pattern of alterations had further continued in MCI-CC, exhibiting more pronounced contractions extending from dorsal aspects of the pulvinar and VP to dorsal aspects of VL, lateral posterior nuclei, and VA in the right hemisphere. In contrast, contractions in the left hemisphere were now limited to dorsal aspects of VL, VA, and medial dorsal nuclei. However, there was a tendency toward significant contractions (q = 0.15) in dorsal aspects of the pulvinar, VP and lateral posterior nuclei as well. Thalamic expansions in turn were limited to ventral and medial parts. MCI-CB revealed expansions in bilateral ventral aspects of the central nuclei, VA, VL, and VP. MCI-CC revealed the same, though more pronounced pattern of expansions in both hemispheres, with additional expansion in the medial aspect of the medial dorsal nuclei.

Striatal displacements are presented in Figure 4B. Again in contrast to MCI-S, the other groups displayed contractions, predominantly in the left hemisphere and most pronounced in ventral (inferior) aspects. More precisely, MCI-CB revealed contractions in medial parts of the putamen and anterior parts of the striatum (caudate head). The same pattern was found in MCI-CC, with more pronounced alterations in left ventral (inferior) medial parts of the putamen and with continued spreading to the left dorsal medial striatum (caudate body) and to right ventral (inferior) aspects of the anterior striatum (caudate head). Similar to the contractions, striatal expansions were more pronounced in the left hemisphere: MCI-CB showed pronounced expansions in ventral aspects of the anterior striatum (caudate head) and lateral putamen. MCI-CC showed a similar pattern, but with further continued expansions to ventral (inferior) aspects of the left striatum (caudate tail), and to ventral aspects of the anterior striatum (caudate head) of the right hemisphere.

For these analyses, between group t statistics from comparison between HC and MCI-CB were used. More precisely, for all vertices whose t-values constituted a local minimum (or maximum), the average of the displacement values in mm was computed for each subject. This was done separately for each structure and hemisphere. The analyses revealed contractions of the right thalamus and left striatum having an AUC of 0.98 (95% confidence interval [CI] 0.929-1.000, p < 0.001) and of 0.96 (95%CI 0.883-1.000, p < 0.001) for discriminating HC from MCI-CB (Figure 5), both AUC confirming excellent accuracy and high statistical significance similar to left hippocampus volume (raw volume AUC = 0.95, 95%CI 0.861-1.000, p < 0.001; volume relative to eTIV AUC = 0.95, 95%CI 0.848-1.000, p < 0.001) and increased accuracy and statistical significance when compared to the right hippocampus volume (raw volume AUC = 0.80, 95%CI 0.572-1.000, p < 0.05; volume relative to eTIV AUC = 0.84, 95%CI 0.633-1.000, p < 0.05). The same AUC analyses carried out for thalamic expansions (left AUC = 0.92 95%CI 0.767-1.000, p = 0.001; right AUC = 0.93 95%CI 0.816-1.000, p = 0.001), thalamic contractions left (AUC = 0.86, 95%CI 0.685-1.000, p = 0.007) and striatal expansions left (AUC = 0.88, 95%CI 0.730-1.000, p = 0.004) revealed lower but still statistically significant accuracy for discriminating HC from MCI-CB.

There was no significant difference in cortical thickness in MCI-S compared to HC, but in MCI-CB in left parahippocampal regions including the uncus with further propagation to bilateral regions in MCI-CC. This pattern is consistent with the literature (Braak and Braak, 1991b; Mitchell et al., 2002) and with morphometric changes previously associated with early stages of AD (Mitchell et al., 2002; Drago et al., 2011). Like others (Lebedev et al., 2013), we also identified cortical thinning in the subgenual cingulate region in MCI-CB. Functionally, this region has been related to normal sadness reactions (Phan et al., 2002), and reduced volumes have been observed in patients with major depressive disorder (Drevets et al., 2008). Although clinically significant depression was one of the exclusion criteria in our study, very subtle and subclinical depressive symptoms that are difficult to quantify may be related to this finding in MCI-CB, leaving the exact mechanism unclear. This region has furthermore been associated with the uptake of serotonin (Lanzenberger et al., 2013). Serotonin is involved in the regulation of sleep (Portas et al., 2000) which is disturbed in AD (Westerberg et al., 2012).

MCI-S revealed no global or local hippocampal volume reductions compared to HC. However, in accordance with the AD literature (Stepan-Buksakowska et al., 2014) and our expectations, we found reduced volumes of bilateral hippocampi in MCI-CB and MCI-CC as well as reduced local volumes of all but one (CA2-CA3) bilateral subfields already in MCI-CB, and of all bilateral subfields in MCI-CC.

Hippocampal atrophy represents the key imaging marker in AD research. The successful identification of hippocampal subfields by using high-field MRI, however, has offered a more refined approach (Mueller and Weiner, 2009; Yushkevich et al., 2009; Mueller et al., 2010; Antharam et al., 2012; Wisse et al., 2014a). Advances in segmentation and analysis techniques have now enabled the field to identify hippocampal subfield alterations on images obtained from standard clinical systems. Corresponding with the pattern of neurofibrillary tangle formation, predominant CA1 atrophy has been found in MCI (Apostolova et al., 2006a; Atienza et al., 2011; Pluta et al., 2012; La Joie et al., 2013; Yushkevich et al., 2015a), with some of these studies having reported additional subicular alterations (Atienza et al., 2011; La Joie et al., 2013) and alterations in additional subfields (Amaral et al., 2016). A more extended pattern also including CA2 and CA3 or even CA4/Dentate gyrus has been identified in AD (Apostolova et al., 2006a; Frisoni et al., 2008; Frankó et al., 2013; Li et al., 2013). Importantly, alterations in these subfields have been related to impaired memory functions in amnestic MCI (Atienza et al., 2011; Ferrarini et al., 2014). Local analysis of the structure, therefore, has been suggested advantageous for the early detection of dementia (Tang et al., 2014).

Nevertheless, only a few studies have examined hippocampal subfields in future converters. Even though they have applied surface-based subfield imaging without providing volumetric information, results of alterations in CA1 and/or subiculum in future converters to MCI (Apostolova et al., 2010) or to AD (Apostolova et al., 2006b; Chételat et al., 2008; Frankó et al., 2013) are in agreement with our own findings. In line with further progression of tangle accumulation (Braak and Braak, 1991b), we found similar bilateral CA2-CA3 volumes in MCI-CB but reduced bilateral CA2-CA3 volumes in MCI-CC compared to HC. Additionally, we found bilateral CA4/Dentate gyrus volumes were lowered not only in MCI-CC but also in MCI-CB. Although CA4/Dentate gyrus volume reduction has mainly been identified in dementia stages (Frankó et al., 2013; Li et al., 2013), earlier alterations in this subfield have also been reported by others (Pluta et al., 2012). We also identified reduced volumes of bilateral strata lacunosum/radiatum/moleculare in MCI-CB and MCI-CC, and our results are supported by studies reporting stratum radiatum and stratum lacunosum/moleculare of CA1 being affected by early tangle accumulation (Braak and Braak, 1997; Thal et al., 2000; Braak et al., 2006) and atrophy in mild AD (Kerchner et al., 2010). It shoud be further noted that only a few protocols exist allowing for the reliable, automated segmentation of hippocampal subfields (see Wisse et al., 2014b for a critical discussion; Yushkevic et al., 2015b). Here, we used a multi-atlas based segmentation approach (Chakravarty et al., 2013; Voineskos et al., 2015) that performs as well as other methods used in the field (Pipitone et al., 2014).

MCI-S did not reveal any deviation from HC, neither in thalamic total or subfield volumes nor in shape alterations. Contrary to our expectation, the same was true for both total and local thalamus volumes in MCI-CB and MCI-CC. Indeed reduced thalamic total volumes have been reported in amnestic MCI (Hahn et al., 2016) and MCI-CB (Bossa et al., 2011). However, thalamic atrophy in AD is debated, particularly in early disease stages (Xuereb et al., 1991; Braak and Braak, 1991a; Paskavitz et al., 1995), where only a few regions are affected by a small number of isolated neurofibrillary tangles (Braak and Braak, 1991b). Moreover, our results are in agreement with other studies having observed similar volumes in amnestic MCI, MCI and AD patients (Cho et al., 2014).

But first and foremost, and despite the absence of volumetric differences, we identified a widespread pattern of shape alterations in MCI-CB and MCI-CC with displacements showing similar if not improved discrimination between HC and MCI-CB compared to hippocampus volume. These findings may represent a critical MRI-based marker for AD. Confirming our expectation, the pronounced thalamic shape alterations in VA found in MCI-CB and MCI-CC cover regions that are affected from early neurofibrillary tangles and later occurring amyloid deposition (Braak and Braak, 1991b). Importantly, the thalamus plays a crucial role in the Papez circuit with the anterior thalamus and the pulvinar, both having shown shape alterations in MCI-CB and MCI-CC, being directly connected to the hippocampus (Zarei et al., 2010). Furthermore, frontostriatal circuits link dorsolateral prefrontal, anterior cingulate, and orbitofrontal cortex regions (Alexander et al., 1986) via the striatum / globus pallidus (Haber, 2003) to VA and medial dorsal aspects of the thalamus (Tekin and Cummings, 2002; Bonelli and Cummings, 2007), aspects that again have shown shape alterations in MCI-CB and MCI-CC. Further significant shape alterations were found in the VP, VL and lateral posterior nuclei connecting the structure with the somatosensory, motor, premotor and prefrontal and temporal and parietal cortices (Behrens et al., 2003; De bourbon-Teles et al., 2014). Consistent with these extensive connections, thalamic regions affected from shape alterations in MCI-CB and MCI-CC have been linked with memory and frontal executive, attention, visuospatial perception, and emotion processing (Wilke et al., 2013; Saalmann, 2014; De bourbon-Teles et al., 2014; Arend et al., 2015), with all of these functions being impaired early in AD (Klekociuk et al., 2014).

To our knowledge, there has been only one other study investigating subcortical shape alterations in MCI-S and MCI-CB, without a HC group (Tang et al., 2014). The contractions found in the pulvinar and dorsal aspects of the VP in MCI-CB are consistent with contractions that have been identified in amnestic MCI by an earlier study of our group (Leh et al., 2015). Additionally, the authors have documented contractions in regions that remained unaffected in the present MCI-CB sample such as the VL and lateral posterior nuclei. These, however, revealed contractions in the MCI-CC group. Our finding of contractions in dorsal, medial dorsal and anterior regions with more pronounced results in the left than in the right hemisphere in MCI-CB and MCI-CC, contractions in the pulvinar and dorsal as well as medial dorsal regions in MCI-CC in turn are consistent with other study results (Qiu et al., 2009; Cho et al., 2014; Stepan-Buksakowska et al., 2014; Hahn et al., 2016). The few results about thalamic expansions in MCI and AD are inconsistent. In contrast to expansions in ventral regions in both MCI-CB and MCI-CC, the earlier study from our group has identified expansions in more lateral aspects in amnestic MCI, whereas others have reported expansions in medial or even dorsal aspects in MCI and AD (Qiu et al., 2009). Further studies are warranted to find out whether there is a typical AD-related pattern of thalamic expansion.

Against our expectation, we observed reduced bilateral striatal volumes in MCI-S but similar volumes in MCI-CB and MCI-CC when compared to HC. The most plausible explanation for this seem to be age- and gender-related shrinking, or age-related changes in dopamine and frontostriatal networks (Klostermann et al., 2012) resulting in reduced striatal volumes. Due to age- and gender-matched MCI-S and HC groups, however, this is rather unlikely in the present study. Rather, the result may indicate non-AD pathology in our MCI-S sample. This assumption is further supported by cognitive profiles differing between MCI-S and MCI-CB. In comparison with their respective HC group, MCI-S showed significantly reduced cognitive scores not only in memory tests but also in tests assessing executive functions, attention and language abilities (see Table 1). In contrast, and apart from impaired memory functions, MCI-CB performed worse than their respective HC in the category fluency task and in the boston naming test but not in attentional tasks. Both, the category fluency task and the boston naming test have closely been linked with semantic knowledge (Delis and Kaplan, 2001), which in turn is thought to rely on temporal brain regions (Levy et al., 2004) known to be affected early in AD (Tondelli et al., 2012). Genereally speaking, MCI-S in our study might indeed represent non-AD cognitive impairment mainly based on attentional deficits influencing other cognitive processes and leading to a more widespread pattern of cognitive impairment whereas our MCI-CB group showed an AD-typical pattern of cognitive impairment.

Normal striatal volumes in MCI-CB and MCI-CC in turn are most likely attributable to the early disease stages of each group: striatal amyloid deposition and tangle formation have evidenced at late histopathological disease stages, and mainly after dementia onset (Braak and Braak, 1990, 1991a,b; Thal et al., 2002; Beach et al., 2012). Hence, the striatum in MCI-CB and MCI-CC may still be free of typical AD pathology. Correspondingly, similar volumes of the striatum, putamen and caudate in MCI and future converters compared to HC have been reported by others (Bossa et al., 2011; Roh et al., 2011).

As expected, however, shape analyses showed no alterations in the striatum in MCI-S but pronounced displacements in MCI-CB and MCI-CC. The observed shape alterations were limited to the left striatum in MCI-CB, but displacements had propagated within the structure and to the right hemisphere in MCI-CC. Furthermore, and as with the thalamic shape alterations, these striatal shape alterations revealed similar abilities for discriminating HC from MCI-CB compared to hippocampus volume. Aspects of the striatum showing most pronounced contractions and expansions such as the caudate head, body and tail as well as medial and lateral putamen have been linked to a wide range of cognitive functions. These regions are involved in attention, planning and memory (Cummings, 1995), all of which are impaired early in AD (Klekociuk et al., 2014).

Again, the study from Tang et al. (2014) represents the only study we are aware of having documented shape alterations in MCI-S and MCI-CB; however no HCs were compared in this study. In agreement with other studies, however, they demonstrated patterns of striatal contractions in MCI and AD patients (Qiu et al., 2009; Tang et al., 2014; De Jong et al., 2014) that were comparable to the patterns found in the current study. Interestingly, only a few studies have documented striatal expansions in general (Tang et al., 2014; De Jong et al., 2014), though these expansions were less pronounced than in the present study (visual inspection).

It is beyond the aim of our study to draw direct inferences about the neuronal correlate of shape alterations. Given the pronounced connections with other disease-relevant structures (Leh et al., 2008), however, striatal and thalamic shape alterations may represent secondary downstream effects. A similar effect has been proposed by Stepan-Buksakowska et al. (2014). Specifically, volume reductions in the hippocampus and other early affected cortical regions may have caused subsequent morphometric changes in connected regions such as the thalamus and the striatum, without generating volume reductions yet. Although we cannot rule out the possibility of contractions representing atrophy-related alterations, the applied surface-based shape analyses provide local, but not fully comprehensive information about the entire structure. Accordingly, and as it has been shown in the present study, structural shape alterations are not necessarily associated with corresponding volume changes. Hence, our results of shape differences in the absence of volume differences highlight the importance of considering shape changes along with established volume measures.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. RN and CH are founder/co-founder and board members of Neurimmune Holding AG, Schlieren.