Atrophy of the hippocampus (HC) is common in many diseases affecting the brain, such as Alzheimer’s disease (AD; Braak and Braak, 1991), frontotemporal lobar degeneration (FTLD; Lindberg et al., 2012), schizophrenia (Bogerts et al., 1993; Altshuler et al., 1998; Heckers, 2001), depression (Bremner et al., 2000; Vakili et al., 2000; Campbell et al., 2004) and post-traumatic stress disorder (Van Rooij et al., 2015). The HC can from a gross anatomical perspective be divided into three sub-parts moving from anterior to posterior: the hippocampal head, the hippocampal body and the hippocampal tail; further described in Section “A Gross Anatomical Division of the Hippocampus”) and in Malykhin et al. (2007).

Further, on the basis of the characteristics of the cellular cytoarchitecture, the HC can be divided into six different subfields: The cornu ammonis (CA) 1–4, the dentate gyrus and the subiculum (Duvernoy and Bourgouin, 1998). Various subfields may be implicated in different forms of brain dysfunction. For example, AD is in early stages associated with atrophy of primarily CA1 and subiculum (Carlesimo et al., 2015; for review see de Flores et al., 2015). CA1 is also vulnerable to ischemia and mesial lobe epilepsy, CA2 to schizophrenia and subiculum to aging and several other forms of dementia such as FTLD (Van Hoesen and Hyman, 1990; Tabuchi et al., 1992; West et al., 1994; Bobinski et al., 1998; Adriano et al., 2012; Lindberg et al., 2012). Since different risk factors and different diseases have partly unique patterns of cellular degeneration, several parcellation protocols have been developed to obtain the volume of each subfield from structural magnetic resonance images (MRI; reviewed in Yushkevich et al., 2015). What is in common for most of these protocols is that they involve segmentation at very high resolution perpendicular to the long axis of HC. One single coronal slice of HC may contain all six subfields. In contrast, some subfields such as the subiculum are present on all slices along the long axis of the HC. The boundary between the subiculum and other subfields in the coronal view is often defined at sub-voxel level but spans several centimeters on the long axis of the HC.

While this is a valid division from a neuropathological perspective (Duvernoy and Bourgouin, 1998), it is less relevant from a functional perspective. The anterior CA1 and subiculum project to the amygdala, the prefrontal cortex, the temporal pole and the nucleus accumbens (Fanselow and Dong, 2010). The posterior parts of these regions project to the mammillary bodies, the retrosplenial cortex and the anterior cingulate (Aggleton, 2012). The anterior and posterior CA1 and subiculum are in fact involved in such different functions that some authors suggests that the dorsal (posterior in primates) and ventral (anterior in primates) parts of the HC could potentially be regarded as two functionally distinct structures (Fanselow and Dong, 2010). The posterior HC supports functions of locomotion, orientation, movement, navigation and exploration (Fanselow and Dong, 2010) and in detailed spatial and autobiographical (episodic) memory (Strange et al., 2014), while the anterior HC is associated with motivational behavior, neuroendocrine and autonomic (hypothalamic) functions (Fanselow and Dong, 2010). In view of the clinical symptoms of AD with deficits of episodic memory and spatial orientation, it could be hypothesized that AD-pathology may selectively affect the posterior HC.

One approach to investigate differences along the axis of the HC is to divide the HC or an HC subfield on the basis of gross anatomical landmarks of the hippocampal head, body and tail. Using this approach, one study found that both AD and semantic dementia were associated with atrophy of the CA1 and subiculum. However, these subfields were more atrophic in the hippocampal head in the semantic dementia group. The same study also investigated the effect of Aβ pathology in patients with mild cognitive impairment (MCI). MCI patients with Aβ pathology displayed atrophy in the CA1 and the subiculum, while MCI patients without Aβ pathology only had atrophy in the CA1 (La Joie et al., 2013).

Another way to investigate how atrophy is distributed along the anterior-posterior axis of HC is to perform shape analysis of the whole HC. While this approach often shows involvement of the hippocampal body in AD (Gerardin et al., 2009; Lindberg et al., 2012; Tang et al., 2016), the method is limited in the sense that it does not produce volumetric data on specific subfields.

The purpose of this study is to further develop the hippocampal subfields segmentation of the presubiculum (defined as the subiculum in this study) from Freesurfer (FS) 5.3. We aim to study volumetric differences along the anterior-posterior axis of this region in relation to Aβ₄₂ and P-tau levels in cerebrospinal fluid (CSF; as a proxy for tau and Aβ pathology) and memory performance in cognitively normal participants (CN), subjects with subjective cognitive decline (SCD) and subjects with amnestic MCI.

The purpose of this study was to investigate whether a consideration of differences between anterior and posterior parts of a hippocampal subfield could improve our knowledge about how different disease pathologies affect the region. A suitable model to test this hypothesize is early AD. There is evidence that early clinical symptoms of AD, such as deficits in episodic memory (Ranganath and Ritchey, 2012) and scene processing (Lee et al., 2008; Ranganath and Ritchey, 2012; Strange et al., 2014), are associated with the functionality of the posterior hippocampal system. Further, it has been shown that the subiculum is commonly affected in AD.

In accordance with our hypothesis we found that Aβ pathology (measured with CSF Aβ₄₂) and tau pathology (measured with CSF P-tau) were more associated with the volumes of the posterior parts than with the anterior parts of the subiculum. Further, we found that in SCD patients, low CSF Aβ₄₂ levels were associated with reduced volume of left posterior but not anterior subiculum in comparison with SCD subjects with normal levels of Aβ₄₂. The same effect was found with regard to tau pathology on the contralateral side. Interestingly we found a significant interaction between diagnosis and location (anterior vs. posterior atrophy) in both SCD and MCI with low Aβ₄₂ or elevated P-tau levels when comparing these participants with CN with normal levels of the CSF biomarkers. Exploring this interaction further revealed that the patient group had more pronounced atrophy in posterior than the anterior subiculum. Such interaction was not present in SCD with normal levels of Aβ₄₂ or P-tau, and modest in MCI with normal levels of Aβ₄₂ (Supplementary Figures S2, S3). The results found in the SCD and MCI subjects could potentially be interpreted in terms of progression. The largest difference between anterior and posterior segments is seen in SCD subjects with abnormal CSF biomarkers (Figure 7). In MCI with abnormal CSF biomarkers the interaction effect is still significant. When inspecting analysis in Figure 6 we can however see that the whole left subiculum is significantly atrophied in MCI with low CSF Aβ₄₂-levels, which could indicate that these patients have progressed so far in the disease process that atrophy has spread also to the anterior part of the subiculum.

In the total sample, we found that P-tau levels >52 ng/l were associated with decreased volume of the left posterior but not anterior subiculum in participants with normal Aβ₄₂ levels. Both the anterior and the posterior left subiculum were reduced in participants with P-tau levels >70 ng/l, regardless of Aβ₄₂-status. This could also be interpreted as support for the hypothesis that in early stages the posterior subiculum is more sensitive to AD-pathology than the anterior part. A cutoff for P-tau of 52 ng/l is low. But also CN with P-tau >52 ng/l displayed a tendency to reduction in the volume of the posterior but not anterior left subiculum compared to CN with levels below that of cutoff (data not shown).

Together these data clearly show the advantage of considering the differences between the anterior vs. posterior part of single subfields when studying how pathological processes affect the HC. To our knowledge, this has not been considered in a systematic way in previous studies. One reason why this has received so little attention in MRI studies is probably that current subfield terminology does not subdivide single subfields in the anterior-posterior direction (Duvernoy and Bourgouin, 1998). However, knowledge about the functional specialization of anterior vs. posterior HC has improved during recent decades (reviewed in Fanselow and Dong, 2010; Strange et al., 2014). This is also paralleled by increasing knowledge about the cellular characteristics within specific HC subfields. Two studies on the Allen Brain Atlas³ of C57BI/6 mice found that the pyramidal neurons of CA1 and CA3 displayed a clear laminar specificity in spatially distinct molecular domains distributed along the anterior-posterior axis (Thompson et al., 2008; Dong et al., 2009). Genes that are specifically expressed in the CA1, CA2 or CA3 are also clearly segregated among the dorsal, intermediate and ventral segments of these subfields (Dong et al., 2009; Fanselow and Dong, 2010). In recent parcellation models for CA1 and CA3 these subfield are divided into CA1 and CA3 d (dorsal), CA1 and CA3 i (intermediate), and CA1 and CA3 v (ventral; Fanselow and Dong, 2010). While less studied, there is evidence from neuropathological studies suggesting that the structural and functional subdomains found in CA1 are also present in the subiculum (Risold and Swanson, 1996; Swanson, 2004). Altogether, many studies support a subdivision of the HC subfields along the anterior-posterior axis. Another reason why this has not been considered in currently used subfield segmentation approaches is probably that the boundaries between the dorsal, intermediate and posterior parts of HC are not as clearly defined as those between the different subfields (Dong et al., 2009). A distinct division based on connectivity or on function will probably never be possible as changes occur gradually along the axis of HC (Aggleton, 2012; Strange et al., 2014). A possible way of handling this is the approach taken in this study in which one HC subfield was divided into 10 equally long sections along the long axis of the HC.

While our plots indicate that a posterior > anterior atrophic gradient is present in subiculum bilaterally in participants with abnormal levels of CSF Aβ₄₂ and P-tau, does this effect seem to be particularly pronounced on the left side (see Figures 2, 3, 7). We also found that the left posterior subiculum was affected earlier by elevated P-tau levels than the left anterior subiculum in participants with normal CSF Aβ₄₂. This interaction was not significant on the right side. A left > right rate of atrophy has been found by Morra et al. (2009) in a large longitudinal study including 490 subjects. The authors propose that the right hemisphere may potentially lag behind the left in the very early stages of dementia. A left > right hippocampal volume loss has also been found in cross-sectional studies. Qiu et al. (2008) found that cognitively normal subjects with a Clinical Dementia Rating scale (CDR) of 0.5 (questionable AD) displayed atrophy of the left lateral hippocampal tail. We found atrophy in the left but not the right hippocampal body and tail in a group of patients with mild AD compared to controls subjects (Lindberg et al., 2012). Apostolova et al. (2010) investigated 169 amnestic MCI patients and found left > right atrophy of the hippocampal body and tail in participants who later converted to AD. They propose that this may be related to an ascertainment bias as standardized cognitive testing for patients suspected of having a dementia diagnosis is biased towards verbally mediated tasks which are more sensitive to left-side pathology. It is possible that our finding is partly driven by a selection bias as suggested in the study by Apostolova et al. (2010). However, CN subjects with low Aβ₄₂-levels also have decreased volume of the total left (F_(1,283) = 4.52, p = 0.03) but not right (F_(1,277) = 1.19, p = 0.28) subiculum (Supplementary Figures S4A,B), which indicates that there may be an underlying biological mechanism that contributes to this asymmetry.

Another finding that may be related to a selection bias is that SCD patients without pathology also seem to have more atrophy of the posterior subiculum. Furthermore, in MCI patients with normal levels of Aβ₄₂ we see an interaction between diagnosis (CN vs. MCI) and location (Supplementary Figure S2). We believe that this can be explained by a selection bias in which people with atrophy of the posterior hippocampal system are more likely to be diagnosed as preclinical AD than people with atrophy of the anterior system.

The most important limitation of this study is that only one subfield was analyzed. It would of course strengthen our argument if we could show a more significant anterior-posterior atrophic gradient in other subfields that have previously been shown to be affected in AD, including CA1. However, we do not consider the segmentations of subfields other than presubiculum reliable when using FS 5.3. Another limitation is that the hippocampal tail is excluded in FS 5.3 (and 6). It is possible that we would have fond an even stronger association between AD pathology and the posterior subiculum if we could have included measurements from the hippocampal tail.

It has been reported that FS 6 provides a better parcellation scheme than FS 5.3 which is more consistent with known histopathological anatomical borders of hippocampal subfields (Iglesias et al., 2015). However in addition to an improved parcellation schema FS6 is also parcellating the molecular cellular layer (Iglesias et al., 2015). But while the molecular layer is present in the different subparts of the subiculum, the parcellation scheme only provides the total volume of the molecular layer. This creates a difficult problem in terms of quality control. To visually inspect if the molecular layer is correctly segmented we need a MRI image with very high resolution. On regular structural MRI images we are basically forced to accept the segmentation provided by FS6 without any possibility to do quality control of the results. If the molecular layer had been subdivided so that we would know how much volume of this layer that belongs to each subdivision of the subiculum, we could ourselves determine whether to present the total volume of each subicular subdivision or the molecular and other cellular layer separately based on the resolution of the MRI-image in the study. In FS 5.3 we are still able to do a visual quality control of the subicular subfield which is the reason why this older version was used in this study.

One further limitation is that two results have a relatively weak p-value (≥0.045). We have emphasized this by denoting them as borderline significant. The interaction significant between high vs. low levels of Aβ₄₂ and location for the left subiculum in SCD, as well as the interaction between diagnosis CN normal vs. SCD high levels of P-tau and location for the right subiculum thus needs to be validated in a replication study.

Finally it should be acknowledged that the hypothetical model that relates episodic memory particularly to the functionality of the posterior HC has been challenged in some recent articles (see Zeidman and Maguire, 2016). As presented, we found that the posterior part of the subiculum was generally more associated with performance on delayed recollection than the anterior part.

The main goal of the study was to investigate if we could confirm whether specific regions of a single subfield along the anterior-posterior axis of HC could be selectively vulnerable to pathology. We found strong support for this hypothesis. If the whole volume of the subiculum had been considered, no effect of CSF Aβ₄₂ or P-tau would have been found in the SCD group. Low CSF Aβ₄₂ was however associated with significant volume loss of the left posterior subiculum, while elevated P-tau was associated with volume loss in the right posterior subiculum.

In conclusion, we have shown that emerging early Aβ and tau pathology, predominantly affects the posterior subiculum which suggests that future segmentations approaches of HC subfields would benefit from considering volumetric differences along the anterior-posterior axis.

Study concept and design was formulated by OL. GM created the application for dividing the subicular subfield from Freesurfer into volumetric subsection. Acquisition, analysis or interpretation of data were done by OL, GM, ES, SP, L-OW, EW, OH.

OH has acquired research support (for the institution) from Roche, GE Healthcare, Biogen, AVID Radiopharmaceuticals, Fujirebio, and Euroimmun. In the past 2 years, he has received consultancy/speaker fees (paid to the institution) from Lilly, Roche, and Fujirebio. The other 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.