The hippocampus plays an integral role in theories of cognitive aging (O’Shea et al., 2016) and supports a wide range of neuropsychological functions as a result of its expansion through human evolution (Pine et al., 2021). Hippocampus structural alterations are a transdiagnostic feature of psychiatric disorders (Brosch et al., 2022) and play a central role in neurologic disorders such as Alzheimer’s Disease (Toda et al., 2019) and its functional sequelae (Rao et al., 2023). The advent of web-based interactive visualization tools using hippocampus volumes obtained in very large cohorts (Nobis et al., 2019, Pomponio et al., 2020) now provide rapid and objective estimates of age trends and corresponding percentiles, and allow users to input individual hippocampus volume data for normative comparison. Better understanding deviations in the normal trajectory of age-related associations with hippocampus volume could help clarify differential contributions of age- and disorder specific genetic mechanisms to hippocampus morphology in different brain disorders (Bahrami et al., 2022) and neurodevelopmental models of psychiatric disorders (Knight et al., 2022, Sun et al., 2023).

The majority of magnetic resonance (MR) imaging studies investigating the relationship between age and hippocampus volume have focused on its total structure. The hippocampus is structurally heterogeneous, however, with multiple subregions that have different afferent and efferent connections (Knierim, 2015, Shinohara and Kohara, 2023). Hippocampus segmentation is a critical goal of neuroscience given a distinction among subregions in their underlying neuroanatomical connections and functional correlates that have relevance for neuropsychiatric and neurological disorders (Genon et al., 2021). Although several groups have investigated the relationship between subfield/subregion volumes and aging (e.g., Malykhin et al., 2017; Pereira et al., 2014; Wisse et al., 2014a; Mueller and Weiner, 2009; Raz et al., 2015; Mueller et al., 2007; Shing et al., 2011; de Flores et al., 2015; La Joie et al., 2010; Wolf et al., 2015), findings across these studies have been inconsistent.

Manual mensuration of the hippocampus and its subregions from MR images remains the gold-standard (Malykhin et al., 2010, Mueller et al., 2007, Rhindress et al., 2015; Wisse et al., 2014a), but is impractical to implement in very large samples that can now number in the tens of thousands (Dima et al., 2022, Pomponio et al., 2020). There are several computerized algorithms available that assist in hippocampus segmentation. HippUnfold (DeKraker et al., 2022) is a surface-based approach that incorporates hippocampus morphology to facilitate anatomical registration among individuals. In contrast to other approaches that rely solely on volumetry, HippUnfold maps the hippocampus to a flat rectangle thus preserving its topology and can produce indices related to CA subfields 1–4 and the dentate gyrus from a T1-weighted contrast. Another approach, Automatic Segmentation of Hippocampal Subfields (ASHS), provides an automated segmentation of mesiotemporal lobe structures that permits identification of subfields from T2-weighted images or along the anterior/posterior hippocampal axis from T1-weighted scans (Yushkevich et al., 2015).

Another widely used automated approach for hippocampal subregion delineation, which uses both ex-vivo post-mortem and in-vivo MR imaging data (Iglesias et al., 2015), has been implemented in newer versions of the FreeSurfer image analysis suite (Desikan et al., 2006). This approach resolves some limitations of earlier versions to provide improved accuracy and better alignment with histological studies. In addition, there is also improved translation to the hippocampal head/tail and the ability to resolve additional subregions (e.g., molecular layer), thus avoiding the need for geometric criteria to trace boundaries. Empirical studies demonstrate that FreeSurfer provides reliable hippocampus subregion volumes assessed using intra-class correlation coefficients, percentage volume differences and percentage volume overlap (i.e., Dice indices) (Brown et al., 2020, Worker et al., 2018), but with the reliable identification and purported validity of some subregions (e.g., hippocampal fissure) being potentially problematic (Iglesias et al., 2015, Kahhale et al., 2023). In addition, it should be noted that hippocampus subfield volumes generated automatically from FreeSurfer using standard T1-weighted 1 mm3 acquisitions may be insufficient for identifying the stratum radiatum lacunosum moleculare, which is critical for distinguishing among some hippocampus subfields, and could be a limitation of this approach (Wisse et al., 2021, Wisse et al., 2014b).

Prior studies investigating the relationship between age and hippocampus volume may include apriori assumptions regarding the distribution of the data (e.g., quadratic) and/or the use of a restricted age range that can meaningfully affect the maximum age at which a brain volume peaks (Fjell et al., 2010). To address these issues several studies investigating the relationship between age and total hippocampus volume in large (>10,000) datasets incorporated a wide age range of participants and used fractional polynomial regression (Dima et al., 2022) or generalized additive models (Pomponio et al., 2020) to address nonlinear effects. Dima et al., (2022) reported that total hippocampus volume was largest during the first 2–3 decades of life whereas in the study by Pomponio et al. (2020) peak volume occurred later in adulthood. Few studies, however, have investigated the trajectory of individual hippocampus subregions across the age span using flexible statistical modeling approaches that can model a distribution of data without any apriori assumptions regarding its shape as would be evident from linear, quadratic or cubic modeling. In one study that used spline modeling Bussy et al. (2021) reported that some subfields (e.g., CA4DG) demonstrated a pattern of volume loss across the age range of 18–81 years, but that the majority of them showed minimal age-associated relations until approximately the sixth decade of life at which point there was accelerated volume loss.

Many research studies and clinical investigations have demonstrated the importance of the hippocampus in a wide variety of memory functions including episodic encoding and retrieval (Spaniol et al., 2009), spatial memory (Li and King, 2019), facial memory (Tsukiura, 2012) and word learning (Davis and Gaskell, 2009). A large number of studies have investigated different memory correlates of whole hippocampal volume assessed using structural MR imaging, although findings have been inconsistent. For example, an early meta-analysis identified a positive relationship between hippocampus volume and memory functioning among older healthy individuals (Van Petten, 2004); in contrast, less hippocampus volume was paradoxically associated with better memory performance in healthy young adults (Chantôme et al., 1999, Foster et al., 1999). A recent meta-analysis, however, that included 25 studies involving 1357 typically developing children and adolescents, reported a significant positive association between total hippocampus volume and memory functioning (Botdorf et al., 2022).

A few studies investigated memory functions in relationship to individual hippocampus subregions and were largely restricted to ages ≥ 18 analyzed using cross-sectional designs. Prior work investigating the memory correlates of CA subfields reported that greater right CA2/CA3 volume was associated with better letter number sequence score performance (Voineskos et al., 2015), greater left CA4/DG volume was associated with better figure recall (Voineskos et al., 2015) and that larger volumes of left CA hippocampal subfields was associated with better verbal episodic memory (Aslaksen et al., 2018). The investigation of other hippocampus subregions revealed that preservation of the fornix was associated with stability of associative memory (Foster et al., 2019) and right hippocampal tail volume was associated with spatial memory, but that left hippocampal body volume was associated with delayed verbal memory (Chen et al., 2010). In one of the few longitudinal studies conducted to date greater entorhinal cortex volume loss was reportedly correlated with worse memory performance (Rodrigue and Raz, 2004).

In the current study we modeled the age trajectory of 4 hippocampus subregions derived from automated segmentation of the hippocampus using FreeSurfer in a large cohort of 674 healthy individuals ranging in age from 6 to 85 years. We used natural splines with different degrees of freedom (and corresponding “knots”) to provide flexibility in identifying the best fitting model without apriori assumptions regarding the distribution of data. We further investigated whether hippocampus subregion volume mediated the relationship between age and memory performance.