Comparison of histological delineations of medial temporal lobe cortices by four independent neuroanatomy laboratories

The medial temporal lobe (MTL) cortex, located adjacent to the hippocampus, is crucial for memory and prone to the accumulation of certain neuropathologies such as Alzheimer's disease neurofibrillary tau tangles. The MTL cortex is composed of several subregions which differ in their functional and cytoarchitectonic features. As neuroanatomical schools rely on different cytoarchitectonic definitions of these subregions, it is unclear to what extent their delineations of MTL cortex subregions overlap. Here, we provide an overview of cytoarchitectonic definitions of the entorhinal and parahippocampal cortices as well as Brodmann areas (BA) 35 and 36, as provided by four neuroanatomists from different laboratories, aiming to identify the rationale for overlapping and diverging delineations. Nissl-stained series were acquired from the temporal lobes of three human specimens (two right and one left hemisphere). Slices (50 μ m thick) were prepared perpendicular to the long axis of the hippocampus spanning the entire longitudinal extent of the MTL cortex. Four neuroanatomists annotated

MTL cortex subregions on digitized slices spaced 5 mm apart (pixel size 0.4 μm at 20Â magnification).Parcellations, terminology, and border placement were compared among neuroanatomists.Cytoarchitectonic features of each subregion are described in detail.
Qualitative analysis of the annotations showed higher agreement in the definitions of the entorhinal cortex and BA35, while the definitions of BA36 and the parahippocampal cortex exhibited less overlap among neuroanatomists.The degree of overlap of cytoarchitectonic definitions was partially reflected in the neuroanatomists' agreement on the respective delineations.Lower agreement in annotations was observed in transitional zones between structures where seminal cytoarchitectonic features are expressed less saliently.The results highlight that definitions and parcellations of the MTL cortex differ among neuroanatomical schools and thereby increase understanding of why these differences may arise.This work sets a crucial foundation to further advance anatomically-informed neuroimaging research on the human MTL cortex.
It is important to mention that the MTL cortex (incorporating the ambient, parahippocampal, and fusiform gyri) is not a single homogeneous entity, but rather exhibits high variance in cytoarchitecture.Its anteromedial portion comprises three-layered, allocortical structures of the hippocampus, whereas its outmost lateral subregions exhibit the key characteristics of six-layered isocortex (Braak & Braak, 1995;Ding & Van Hoesen, 2010;Filimonoff, 1947;Insausti et al., 1995Insausti et al., , 2017;;Van Hoesen et al., 2000).This heterogeneous composition has led neuroanatomists to distinguish various subregions of the human MTL cortex with partially overlapping terminology.
In their seminal cytoarchitectonic brain maps, Brodmann (1909) as well as von Economo and Koskinas (1925) established two distinct atlases of the MTL cortex (Figure 1b).On Brodmann's map, it spans Brodmann Area (BA) 28 (area entorhinalis), BA34 (area entorhinalis dorsalis), BA35 (area perirhinalis), and BA36 (area ectorhinalis).The term entorhinal cortex (ERC) has been used as an umbrella term for BA28 and BA34, corresponding to its portions located on the anterior parahippocampal and ambient gyri, respectively (Insausti et al., 1995(Insausti et al., , 2019)).The perirhinal cortex is anteriorly, laterally, and posteriorly adjacent to the ERC and corresponds to BA35.BA36 has previously been referred to as ectorhinal cortex (Augustinack, Huber, et al., 2013;Brodmann, 1909;Van Hoesen et al., 2000) but has also been grouped with BA35 under the term perirhinal cortex (Ding & Van Hoesen, 2010;Insausti et al., 2017;Kivisaari et al., 2012).Importantly, Brodmann's map is clearer in the anterior MTL cortex but less so in its posterior portion, also referred to as parahippocampal cortex (PHC; not to be confused with parahippocampal gyrus, which is the gross anatomical term for the gyrus adjacent to the hippocampus).In fact, Brodmann (1909) proposed a region termed BA48 (area retrosubicularis) to border BA35 posteriorly.However, Brodmann was unable to reliably identify this region in humans, and thus it was not included in Brodmann's final neuroanatomical atlas, leaving some ambiguity in the posterior MTL cortex.
If corresponding to any BA, the PHC may include some posterior portions of BA36.
The anatomical map established by von Economo and Koskinas (1925) depicts the MTL cortex as areas HA, HB, HC of the regio hippocampi (associated with the ERC) and part of TG (TGa) anteriorly (regio polaris), as well as regions TH (THa) and TF posteriorly (regio fusiformis; Figure 1b).Areas THa and TGa are possible correlates of BA35.THa lies posteromedial to TGa on the crown of the parahippocampal gyrus and adjacent to the parasubiculum, presubiculum, or transsubiculum of the hippocampal formation (Palomero-Gallagher et al., 2020).
Region TH extends from the lateral crown of the parahippocampal gyrus to the lateral bank of the collateral sulcus and is laterally bordered by the fusiform region TF (von Economo & Koskinas, 1925).
Part of region PH (von Economo & Koskinas, 1929)-a large transitional area spanning the posterior parietal and temporal lobes-has been considered part of the PHC as well (Stenger et al., 2022).BA36 may be related to parts of TG and posterior TH (regio temporo-hippocampica).Further descriptions of these regions in humans, macaques, and chimpanzees can be found elsewhere (Bailey et al., 1950;Bailey & von Bonin, 1951;von Bonin & Bailey, 1947).
Since the establishment of these first exhaustive neuroanatomical atlases, histological studies have benefitted from methodological advancements (e.g., digitalization and immunohistochemistry) and neuroanatomists have developed and specified these existing brain maps continuously.Such advancements have led to the distinction of subregions at remarkably detailed levels.For instance, BA35 has been subdivided into BA35a and BA35b based on their distinct cytoarchitecture (Augustinack, Huber, et al., 2013;Ding & Van Hoesen, 2010;Van Hoesen et al., 2000) and several subfields of the ERC have been identified (Behuet et al., 2021;Braak & Braak, 1992;Filimonoff, 1947;Insausti et al., 1995Insausti et al., , 2017;;Krimer et al., 1997).A multitude of additional parcellations can be found in the extensive literature on nonhuman primates (e.g., Blatt & Rosene, 1998;Insausti et al., 1987;Suzuki & Amaral, 2003).In addition to the improvement of histological methods, histological datasets have grown.In combination with the rise of in vivo neuroimaging techniques, this development has provided crucial insights into interindividual differences in neuroanatomy.
For example, it has become apparent that the MTL cortex exhibits great interindividual variation in micro-as well as macrostructure (Augustinack, van der Kouwe, et al., 2013;Ding & Van Hoesen, 2010;Insausti et al., 1998;Ono et al., 1990).In particular, the rhinal and collateral sulci vary in depth and ramifications.Variance in collateral sulcus depth and continuity have been linked to differing gross anatomical locations of MTL cortex subregions (Ding & Van Hoesen, 2010;Insausti et al., 1998;Taylor & Probst, 2008).Considering the different histories and subsequent development of anatomical studies, it is unsurprising that neuroanatomical schools rely on different definitions of the MTL cortex and its subregions.
The described micro-and macroanatomical variance within the MTL cortex poses a challenge to researchers who aim to distinguish MTL cortex subregions on magnetic resonance imaging (MRI) scans for a wide range of research applications.However, accurate delineations are crucial as MTL cortex subregions show distinct involvement in processes of basic and clinical neuroscientific interest.For instance, it has been proposed that the perirhinal cortex (here referring to BA35 and BA36) and the PHC belong to distinct networks of cognition (Maass et al., 2015;Navarro Schröder et al., 2015;Ritchey et al., 2015;Squire & Zola-Morgan, 1991; but also note Doan et al., 2019), making their accurate distinction critical.With regard to neurodegeneration, subregions of the MTL cortex are differently targeted by pathology and healthy aging (Llamas-Rodríguez et al., 2022;Stark & Stark, 2016).The importance of accurate MTL cortex parcellation becomes evident in the case of Alzheimer's disease, in which early neurofibrillary tau tangles are located in the transentorhinal region, likely corresponding to BA35 (Braak & Braak, 1985, 1991;Ding & Van Hoesen, 2010;Insausti et al., 1998).
While the neuroanatomical investigation of the MTL cortex is progressing, proposed parcellations and definitions of subregions used for neuroimaging studies continue to diversify (Yushkevich, Amaral, et al., 2015).This diversity may stem either from inter-individual neuroanatomical variability or, as segmentation protocols are informed by different neuroanatomical laboratories, from diverging definitions of brain structures.To clarify the influence of the latter, it is necessary to examine the overlap of different neuroanatomical laboratories when delineating brain structures on histological samples.Here, we provide an overview of the cytoarchitectonic definitions of MTL cortex subregions as provided by four neuroanatomists (J.C.A., R.I., O.K., and S.L.D.), based at four different laboratories.We focus on subregions ERC, BA35, BA36, and PHC (Figure 1c-e), as this parcellation is commonly found in the neuroanatomical literature and has been translated into MTL cortex segmentation protocols for MRI (Augustinack, Huber, et al., 2013;Berron et al., 2017;Fischl et al., 2009;Insausti et al., 1998;Xie et al., 2019;Yushkevich, Pluta, et al., 2015).We additionally compare annotations of these subregions on histological sections of the same three specimens for all neuroanatomists.The insights from this detailed analysis of MTL cortex subregional delineations will provide a basis for the ongoing Hippocampal Subfields Group's effort to develop a harmonized segmentation protocol for MTL cortex subregions to be used for in vivo MRI (Olsen et al., 2019;Wisse et al., 2017;Yushkevich, Amaral, et al., 2015).Moreover, the present study will benefit human neuroimaging studies of the MTL cortex in a broader sense by allowing more precise localization of MTL cortex structures, more valid interpretation of results, and higher comparability across studies.

| Specimens
Three human MTL specimens, two from the Human Neuroanatomy Laboratory (HNL) at University of Castilla-La Mancha (UCLM, Albacete, Spain) and one from the Center for Neurodegenerative Disease Research at the University of Pennsylvania (UPenn, Philadelphia, USA), were used.Human brain specimens were obtained in accordance with the UCLM Ethical Clinical Committee and the University of Pennsylvania Institutional Review Board guidelines.Written consent was obtained in all HNL cases.Pre-consent during life and, in all cases, next-of-kin consent at death was given at UPenn.The three specimens were selected to represent both sexes, both hemispheres, a wide age range, cases with and without reported neurodegenerative disease diagnosis, and a broad variability in depth and length of the collateral sulcus to evaluate the location of MTL cortical fields in relationship to macrostructural landmarks.Table 1 provides demographic and diagnostic details of the selected specimens.Note that for ease of interpretation, the left hemisphere specimen (specimen 2) was mirrored in all figures to match the orientation of the other specimens.

| Hemisphere preparation and diagnostic pathology
At UCLM, fixation was performed by intracarotid perfusion with the brain in the skull (in situ; Insausti et al., 2023) before brain removal at autopsy with 4% paraformaldehyde in 0.1 M phosphate buffer.At UPenn, brain specimens were fixed in 10% neutral buffered formalin after autopsy.At both centers, hemispheres were fixed for at least 4 weeks.The opposite hemisphere was sampled for diagnostic pathology according to the National Institute of Aging and Alzheimer's Association (NIA-AA) protocol (Hyman et al., 2012).After hemisphere fixation, an intact tissue block containing the full temporal lobe was dissected and placed in the fixative for 7 Tesla (T) MRI scanning.
Beginning at the temporal pole, specimens were cut coronally following the plane of the 7 T MRI into four parallel 20 mm thick blocks using a custom three-dimensional mold that was generated from a postmortem 7 T MRI.Using dry ice, cryoprotected blocks were frozen and serially sectioned into 50 μm sections using a sliding microtome coupled to a freezing unit after digital block-face images were taken.
Every resulting section was collected in a freezing protectant solution.Every 10th section was immediately mounted and Nisslstained with thionin for cytoarchitectonic analysis, resulting in a 0.5 mm gap between adjacent stained sections.The remaining tissue was saved for further use.Every 20th section (1 mm gap between stained sections) was stained using AT8, a phosphorylated tau antibody immunohistochemistry stain.All sections were counterstained with lower intensity thionin.Sections were mounted on glass slides, digitally scanned, and uploaded to a cloud-based digital histology archive which supports anatomical labeling (see Appendix S1).

| Histological annotations
For the purposes of this study, Nissl-stained sections 5 mm apart were selected to be annotated, resulting in a total of 15-16 digitized slices (pixel size 0.4 μm at 20Â magnification) annotated per specimen.This selection of sections aimed to balance workload for the neuroanatomists while ensuring thorough coverage of the MTL cortex subregions.First, after receiving instruction on the digital annotation tool (see Appendix S1), annotations were completed independently by each neuroanatomist (J.C.A., O.K., R.I., S.L.D.).Second, J.C.A., O.K., and R.I. attended a Hippocampal Subfields Group working group meeting in June 2022 during which their annotations and definitions of MTL cortex subregions were discussed.At this meeting each region was discussed individually, comparing the histological annotations of the neuroanatomists-focusing on the main discrepancies and similarities-and clarifications on the reasoning of specific delineations and definitions were provided.Third, J.C.A., O.K., and S.L.D. received access to the full range of serial histology slices to further inform their annotations and potentially adjust them, which two of them completed.R.I. had access to the full range of serial histology slices from the beginning, as part of another project.Fourth, all four neuroanatomists were consulted for follow-up clarifications and asked to provide the seminal cytoarchitectonic features per region that they used to inform their annotations.Assessments of agreement among neuroanatomists were performed per subregion in a qualitative manner.

| RESULTS
All neuroanatomists determined the borders of ERC, BA35, BA36, and PHC, with finer subdivisions of these subregions made by some neuroanatomists following common practices in their respective laboratories.The following sections first provide a short introduction to the key cytoarchitectonic concepts that informed the neuroanatomists' annotations.Next, the cytoarchitectonic features, definitions, and parcellations of each MTL cortex subregion, as provided by the neuroanatomists, are described and compared.

| Key cytoarchitectural concepts used for MTL cortex parcellation
Cytoarchitectonic features such as neuronal size, type, organization, and packing density in the different cellular layers were the main source of information used for annotations.At times, some neuroanatomists additionally referred to gross anatomical landmarks in their justifications of border placement.Table 2 provides definitions of key T A B L E 1 Demographic, diagnostic, and neuropathological information of the three specimens.cytoarchitectonic concepts that were relevant for the delineation of MTL cortex subregions.
An important concept was granularity, which here focuses on the number and organization of granule cells in layer IV (or the equivalent in ERC; see Table S1) of the cortex (Figure 2).The degree of granularity can range from the complete lack of granule cell bodies (agranular cortex) to the presence of few, loosely arranged granule cells (dysgranular cortex), to the dense packing of highly organized granule cells (granular cortex ;Brodmann, 1909;Insausti et al., 2017;von Economo & Koskinas, 1925).Changes in granularity are one of the main distinguishing features of MTL cortex subregions which reflect the transition from periallocortex (agranular), proisocortex (dysgranular), to isocortex (granular; Brodmann, 1909;Ding & Van Hoesen, 2010;Insausti et al., 2017;von Economo & Koskinas, 1925).A cytoarchitectonic feature that can coincide with an absent granule cell layer IV is the lamina dissecans.It is a thoroughly cell-sparse, agranular layer and, in the MTL cortex, specific to the ERC.It divides the cortex into two halves or, if two laminae dissecans are present, encloses a small layer of large pyramidal cells between two cell-free layers (layers Va and Vb in Insausti et al., 1995; see also Insausti et al., 2017;Stephan, 1975).Another concept used to parcellate MTL cortex subregions is the presence or absence of layer II cell islands (Figure 2a).These prominent clusters of large stellate cells, also termed pre-α cell islands (Braak & Braak, 1985), are most prominent in the ERC (Ding & Van Hoesen, 2010;Insausti et al., 1995;Schön et al., 2022).Lastly, columnarity is a feature that is differently expressed throughout layer II and/or III in different subregions of the MTL cortex (Figure 2b).It can be defined as the irregular arrangement of cells, forming cell accumulations that are oriented perpendicular to the pial surface and are separated by cell-sparse parts.

| Cytoarchitectonic and gross anatomy features of MTL cortex subregions
Annotations of all neuroanatomists primarily relied on cytoarchitectonics, not on the gross anatomical appearance of the MTL cortex.
Only one neuroanatomist referred to gross anatomical hallmarks as a relevant source of information for border definition, as described previously (Ding & Van Hoesen, 2010).Table 3 provides a short overview of each structure's core cytoarchitectonic features as stated by the neuroanatomists.A more detailed overview can be found in Table 4, which also highlights points of agreement and disagreement among the neuroanatomists.Although the neuroanatomists agreed upon the presence of these features, they were weighed differently by the neuroanatomists when making boundary placement decisions.
T A B L E 2 Key cytoarchitectonic concepts used for the delineation of the entorhinal cortex, Brodmann areas 35 and 36, and the parahippocampal cortex.and external (or superficial, between pial surface and lamina dissecans) principal layer (Braak & Braak, 1985;Insausti et al., 2017;Rose, 1927).
In some sections of the ERC, two cell-free layers are formed: an external (sometimes referred to as proper lamina dissecans) and an internal cell-free layer (sometimes also referred to as lamina dissecans, while not by all neuroanatomists; see Table S1).Note that the terminology used to distinguish cellular layers in the ERC differs from that used in most of the cortex due to ontogenetic differences (Behuet et al., 2021;Braak, 1972).This special case of the ERC has been resolved differently by neuroanatomical laboratories with some staying closer to conventional cortical nomenclature than others (see Table S1 for an overview; Behuet et al., 2021;Braak & Braak, 1991;Insausti et al., 1995).
Between the cell-free layers lies a distinct layer of large, darkly stained pyramidal neurons (Braak & Braak, 1985;Insausti et al., 2017; and sulcus semiannularis as well as superiorly by the periamygdaloid cortex, reflecting the substantial cytoarchitectonic differences between the six-layered ERC and the adjacent three-layered periamygdaloid cortex (Insausti et al., 2017).The medial border of the ERC was consistently placed adjacent to the hippocampal fissure, as well as the pre-or parasubiculum (Figures 3b,c and 4a; see also Ding & Van Hoesen, 2010;Insausti et al., 1995Insausti et al., , 1998Insausti et al., , 2017;;Williams et al., 2023).As already depicted by Brodmann (1909), the ERC was encapsulated by perirhinal BA35 at its anterior, posterior, and lateral ends in all three specimens.Compared with the highly consistent superior and medial borders, there was some, albeit minor, variability in the placement of the ERC's lateral border with BA35, potentially due to the gradual transition of cytoarchitectonic features between the ERC and BA35 in this region.Figure 6b provides an example of slight variability in the placement of the border between ERC and BA35 in an anterior section, including a clarification of the different cytoarchitectonic features leading to this difference in border placement between neuroanatomists.

| Brodmann area 35
All neuroanatomists agreed that BA35, also called perirhinal cortex, is a heterogeneous structure located at the transition of periallocortex (i.e., ERC) to proisocortex.However, there was disagreement about the specific cortex type of BA35 as it was labeled dysgranular proisocortex by some, while others described both agranular periallocortical (e.g., BA35a) and dysgranular, proisocortical (e.g., BA35b) portions within BA35 (see Table 3, Table 4).Nevertheless, neuroanatomists agreed on several cytoarchitectonic features being characteristic to BA35.The pyramidal cells found in layers II and V are relatively large, though not as large as in the ERC.Layers II and III are organized in a distinct columnar fashion (Figure 2b).BA35 shows a unique layer III (by some neuroanatomists referred to as layer IIIu; Ding et al., 2009) which is formed by exceptionally large cells and lies internally to a typical layer III (see Figure 5b2; Augustinack, Huber, et al., 2013;Ding et al., 2009;Ding & Van Hoesen, 2010).At the medial border with the ERC, these features are expressed to form an oblique border, first appearing superficially, and extending into the midst of the cortex in a For the perirhinal and ectorhinal cortices there was disagreement on the type of cortex (indicated by "or").
b The seminal features below do not refer to the isocortex of the posterior PHC (Stenger et al., 2022). c The gross anatomical location of subregions varies and is, among other things, dependent on the depth and ramification of the collateral sulcus (see Ding & Van Hoesen, 2010).d Note that some neuroanatomists only refer to the external cell-free layer as lamina dissecans.
e See Table 4 for details.
T A B L E 4 Agreement and disagreement among neuroanatomists about cytoarchitectonic features of the MTL cortex subregions.lateral direction.Layer IIIu is difficult to distinguish from typical layer III using Nissl-staining, especially for non-experts (Figure 5b; Ding & Van Hoesen, 2010).However, as the pyramidal cells in layer IIIu are affected early by Alzheimer's disease-related neurofibrillary tau tangles (Braak & Braak, 1985;Ding & Van Hoesen, 2010), tau staining impressively revealed the unique, wedge-like shape of BA35's medial border in specimen 2 (Figure 5b1).BA35 closely corresponds to Braak's transentorhinal region (Braak & Braak, 1985).While for some neuroanatomists these terms are synonymous and refer to the same region, others stated that only the medial portions of their BA35 annotations (e.g., BA35a) can be referred to as transentorhinal cortex.
Across annotations, BA35 borders BA36 anterolaterally and PHC posterolaterally.Like the ERC, annotations in midsections of BA35 exhibited higher agreement among neuroanatomists (Figures 3e,f and 4, Figure S2), compared with anterior (Figure 3d) or posterior transitional sections.Figures 3 and 4 also show that, compared with annotations of the ERC, there was more variation among annotations in the medial-lateral dimension (see also Figure S2).Specifically, the lateral border of BA35 with BA36 varied among neuroanatomists.This was likely driven by the neuroanatomists' differing cytoarchitectonic definitions of BA35 with regards to layer IV granularity and cortex type (Table 4) and more importantly, by the gradual expression of layer IV granule cells when moving laterally (Figure 5b). Figure 6a F I G U R E 3 Annotations performed by the neuroanatomists for the anterior, mid-section, and posterior parts of four subregions of the MTL cortex.Millimeters indicate distance from the temporal pole.Sections were chosen to reflect the varying overlap of annotations across specimens and gross anatomical locations within the MTL cortex.a, anterior; BA, Brodmann area; ERC, entorhinal cortex; i, inferior; l, lateral; m, medial; p, posterior; PHC, parahippocampal cortex; s, superior.
provides an exemplary slice visualizing the differing annotations that resulted from these diverging definitions.Meanwhile, the medial border with the ERC had high agreement with minor variability (Figures 3d and 4), representing the large consensus on the cytoarchitectonic features of BA35 and the ERC in this region (i.e., cell islands in layer II of the ERC vs. large layer II/III cells arranged in columns in BA35).

| Brodmann area 36
BA36 was referred to as ectorhinal cortex by the neuroanatomists.
Like BA35, neuroanatomists disagreed on the cortex type and thereby the extent of the layer IV granularity for BA36.While two neuroanatomists denoted BA36 as dysgranular proisocortex, others considered it a fully granular isocortical structure.The neuroanatomists agreed that layers II and III have smaller cells relative to the ERC and BA35, layer V is homogenous with middle to large pyramidal cells, and layer VI fluently transitions into white matter (Figure 2c).
BA36 borders other regions of the temporopolar cortex anteriorly and PHC as well as the peristriate cortex/BA19 posteriorly.Higher disagreement among neuroanatomists could be observed in these anterior and posterior transitional regions as opposed to the midsection of BA36 (Figures 3g-i and 4, Figure S2).Among all described structures, the annotations of BA36 exhibited the highest degree of disagreement in the medial-lateral dimension.The lateral border with BA20 varied profoundly.While three neuroanatomists included the lateral bank of the collateral sulcus (i.e., corresponding to Bordmann's original mapping of BA36), the annotations of one neuroanatomist additionally spanned the entire fusiform gyrus and medial bank of the occipitotemporal sulcus (i.e., also including some of BA20 on Brodmann's map or area TF as defined by von Economo and Koskinas; Figures 3h   and 4).Similar to the border of BA35 with BA36, disagreements on the lateral border of BA36 with BA20 were at least partially caused by the gradual changes in granularity, which is susceptible to individual interpretation, along the medial-lateral axis (Figure 5b) and the neuroanatomists' different definitions of BA36 as either a proisocortical or isocortical structure (Table 4, see Figure 6a for how diverging definitions can lead F I G U R E 6 Examples of differences in delineations of the neuroanatomists in border regions of the MTL cortex.Panel (a) shows differing border placement due to disagreement in cytoarchitectonic definitions of BA35 and BA36.In (a1), the neuroanatomist annotated the lateral border of BA36 with region TE due to a change in granularity in layer IV (dashed lines).In the medial portion of this section (BA36), layer IV is present but not as prominent as in the lateral (region TE).In (a2), the neuroanatomist annotated a similarly positioned border but for regions BA35 and BA36.They also described a change in organization of layer IV granular cells (dashed lines).Moreover, smaller layer II cells in BA35 than in BA36 (circled) were indicated.Panel (b) shows a slightly different border placement in a transitional section between ERC and BA35.In (b1), the neuroananatomist identified a wedge-like transition between the two cortices (dotted line), with gradually emerging layer IIIu-cells (marked by arrows).Layer II cells (circled) are larger in ERC than in BA35 and the lamina dissecans is present only in ERC (dashed lines).In (b2), the neuroanatomist also identified cell clusters in layer II of the ERC.In BA35 (circled), cells in layers II and III are organized in elongated columns (dashed lines).Panel (c) shows differing border placement along the longitudinal axis of BA36 and PHC in an anterior to posterior fashion.In (c1), the neuroanatomist distinguishes BA36 and region TE on the most anterior slice.Here, BA36 is characterized by a gradual transition of gray into white matter (arrow) and a thinner layer IV (dashed lines) than found in region TE.Density of layer II cells is higher in region TE than in BA36 (circled).This portion posteriorly borders with PHC, with characteristic small cells and high cell packing density (circled) as well as a betterdeveloped layer IV compared to BA36 (dashed lines).One section further posterior, only PHC is annotated due to a dysgranular layer IV (dashed lines), which is less developed than in region TE, and has a higher cell density than BA36 (circled).In (c2), another neuroanatomist annotated the whole anterior-most section as BA36, due to an absent layer IIIu and smaller layer IV cells (dashed lines) compared to BA35.The next section, moving one slice posterior, was annotated PHC as layer IV is well-developed (dashed lines) and smaller layer V and VI cells than found in BA36 (circled).In the adjacent posterior slice, PHC was annotated based on the same criteria.BA, Brodmann area; ERC, entorhinal cortex; PHC, parahippocampal cortex.
to differing annotations).Further, one neuroanatomist stated that their placement of the lateral BA36 border was guided by the position of the fundus of the occipitotemporal sulcus.

| Parahippocampal cortex
The PHC is another highly heterogeneous MTL subregion that spans periallocortical and proisocortical portions.In fact, one neuroanatomist stated that PHC even includes granular isocortex at its lateral end.Importantly, the neuroanatomists agreed that parcellation based on granularity is more difficult in the PHC than in more anterior subregions of the MTL cortex.Analogous to more anterior parts of the MTL cortex comprising the ERC and BA35, the medial bank of the collateral sulcus exhibits higher cytoarchitectonic heterogeneity at the level of the PHC.Neuroanatomists used various cytoarchitectonic definitions of the PHC which led to substantial differences in annotations, especially in their posterolateral extent (Table 4).In line with many of the previously proposed parcellations of the PHC in nonhuman primates (Blatt & Rosene, 1998;Ding & Van Hoesen, 2010;Suzuki & Amaral, 2003;Tranel et al., 1988) and in humans (Bailey & von Bonin, 1951;von Economo & Koskinas, 1925), its medial portion was referred to as region TH by two of the neuroanatomists.Here, layer II and III cell bodies are rather small, layer IV is absent (agranular), and layers V and VI are homogeneously transitioning into one another (Figure 2d; note that here an incipient granule cell layer may be visible).Relative to BA35, the medial PHC has a higher cell density and wider internal layers (Figure 2d).Neuroanatomists who distinguished the PHC subregions divided the lateral PHC into either as TFm (medial) and TFl (lateral; as introduced in the Macaca fascicularis monkey by Suzuki & Amaral, 2003) or as TL (as described also in primates by Blatt & Rosene, 1998).Here, layer II is thicker than in the medial PHC, layer III exhibits high columnar organization, weakly organized granule cells form an incipient dysgranular layer IV, and the transition of layers V and VI is gradual and almost imperceptible (Figure 2e).One neuroanatomist additionally included more posterior regions (referred to as Ph1-3) in their annotations of the PHC (see Figure 4 and Figure S2).These isocortical regions were characterized by a fully organized granular layer IV, though its prominence varied among the single subregions (see Appendix S1 for more details on regions Ph1-3).
All neuroanatomists placed the PHC posterior to BA35, lateral to the parasubiculum or transsubiculum, and medial to the fundus of the posterior collateral sulcus extending as far as the posterior tip of the corpus callosum.Similar to the medial ERC border, the medial PHC border was annotated with high consistency in midsections (i.e., at the level of the hippocampal body; Figure 3).Figures 3, 4, and 6c additionally show the high disagreement in border placement in the posterior and lateral extent of the PHC.Note that if the isocortical Ph subregion is not considered, which has been proposed to be part of the human PHC only recently (Stenger et al., 2022) and was annotated only by one neuroanatomist, the agreement among neuroanatomists on the PHC would be higher.

| DISCUSSION
In this study, we give an overview of the cytoarchitectonic definitions of MTL cortex subregions (ERC, BA35, BA36, and PHC), aiming to examine and understand the underlying reasons for overlapping and diverging annotations made by four independent neuroanatomists.
We find highest agreement of border placement for the ERC and lowest agreement for the PHC, while annotations of BA35 and BA36 show intermediate overlap.Importantly, the degree of agreement on definitions of cytoarchitectonic features does not necessarily correspond to agreement in border placement, since even structures with substantial agreement in described seminal features could show highly diverging annotations and vice versa.This partial mismatch of annotations and formal cytoarchitectonic definitions is due to additional factors influencing inter-and intra-rater reliability of neuroanatomical annotations as discussed below.Despite these differences, an extensive overlap of annotations among neuroanatomists is observed.
Overall, the presented dataset of histological annotations uniquely informs in vivo neuroimaging research.

| Various factors influence the differences among neuroanatomists' annotations
Based on the present results, we identify several factors which appear to influence the (dis-)agreement in annotations among neuroanatomists, underlining the complexity of annotating MTL cortex subregions.
First, disagreements are partially caused by varying definitions of features of MTL cortex subregions, as discussed above.These differences may partially stem from different traditions, backgrounds, and research foci of the neuroanatomists.For example, neuroanatomical schools may put particular emphasis on the subregions' structural or functional connectivity, receptor architecture, or chemoarchitecture.
Subsequently, this may lead to different perceptions and interpretations of histological information.For instance, we observed that while neuroanatomists agreed on the formal definition of different types of cortex, they did not allocate these to the same MTL cortex subregions.
An example is proisocortex and the disagreement on whether this includes only BA35 or both BA35 and BA36.In several cases, the proisocortical BA35b of one and the proisocortical BA36 of another neuroanatomist were delineated almost identically.Additional variability may be introduced by the fact that neuroanatomical schools differ in their use of gross anatomical information for their annotations.This ranges from explicitly considering gross anatomy for border placement, using this information for orientation purposes, to dismissing gross anatomical landmarks and solely relying on histological information.
Second, we observed that annotations diverged in transitional areas between structures.Transition sharpness clearly differs within the MTL cortex, influencing the overlap of annotations.For example, the transition between the ERC and hippocampal structures is relatively salient, leaving little room for uncertainty and leading to higher agreement among neuroanatomists.Meanwhile, there are other transitional areas, such as the area between BA35 and BA36, where the features of multiple subregions are expressed.As the neuroanatomists may weigh these features differently, they may come to different conclusions and place the respective border at different locations.This ambiguity does not only impact inter-rater reliability; neuroanatomists also highlighted that intra-rater reliability (i.e., one neuroanatomist rating the same specimen twice) would likely be lower in these transitional areas.
Third, the neuroanatomists' certainty and, in turn, the agreement of annotations, is influenced by aspects related to the sectioning procedure chosen in this study.The difficulty of annotating transitional zones may increase when borders between structures do not run strictly parallel or perpendicular to the provided sections.Even when using a consistent plane of sectioning (i.zones in all directions, transitions between sections (i.e., along the longitudinal axis) appear to cause more uncertainty than those within sections (i.e., on the medial to lateral axis).This may be because continuous information is only available within histological slices, making it more difficult to perceive differences between MTL cortical regions along the longitudinal axis.It is likely that this effect was exacerbated since three neuroanatomists were initially provided with histological slices spaced 5 mm apart, making subtle changes in cytoarchitectonic features more difficult to detect.
A fourth influencing factor pointed out by the neuroanatomists is the historical focus (or lack of it) on some subregions over the others.The PHC, for example, is historically understudied in its seminal cytoarchitectonic features, connectivity, and borders with other structures.Most research on the PHC has been done in non-human primates (e.g., Blaizot et al., 2004;Suzuki & Amaral, 1994;von Bonin & Bailey, 1947) and translational studies have only recently emerged (e.g., Stenger et al., 2022).
Fifth, neuroanatomists pointed out that annotating at sulcal fundi is generally challenging as most cellular layers are compressed (see e.g., von Economo & Koskinas, 1929).This factor is relevant to the annotation of the border between BA35 and BA36 as well as the lateral border of the PHC as both are often located in proximity to the fundus of the collateral sulcus.
Finally, neuroanatomists indicated that their uncertainties were amplified in the third specimen due to severe neurodegeneration of the MTL cortex.Cell loss that particularly affects certain neuronal populations and cortical layers distorts the overall cytoarchitectonic appearance of the cortex, potentially posing a significant challenge for neuroanatomical annotations.The uncertainty regarding the exact border locations may have contributed to higher disagreement across all annotated subregions in this specific specimen.
Performing annotations as presented in this study requires neuroanatomists to integrate various features of the provided specimen.The need for harmonization in neuroimaging is demonstrated in the variety of MRI segmentation protocols dedicated to the MTL (Berron et al., 2017;Feczko et al., 2009;Insausti et al., 1998;Kivisaari et al., 2020;Olsen et al., 2013;Pruessner et al., 2002;Wisse et al., 2017;Xie et al., 2019;Yushkevich, Pluta, et al., 2015).Our study shows that while there is notable agreement among neuroanatomists in the cytoarchitectonic features referenced, there is currently no single neuroanatomical ground truth for the delineation of MTL cortex subregions on histological slices.As segmentation protocols for in vivo MRI commonly rely on a single neuroanatomical school, profound differences are inevitable.Consequently, in vivo neuroimaging studies of the MTL often lack comparability, negatively impacting their generalizability and usefulness for meta-analyses.To resolve this issue, harmonized definitions of MTL cortex subregions are needed for neuroimaging.Our present study sets an important foundation for this effort by providing an overview of how MTL cortex subregions are defined and delineated by different neuroanatomical schools.Our observations of highly overlapping annotations in certain regions are especially encouraging for neuroimaging research that operates at resolutions that make many of the observed differences negligible.The Hippocampal Subfields Group is currently working towards establishing a harmonized segmentation protocol of the MTL cortex based on the presented findings.

| Strengths and limitations
A major strength of the present study is the integration of input from four leading expert neuroanatomists in the field, forming a first-of-its-kind annotated dataset of MTL cortex subregions in three histological specimens representing varying demographic backgrounds and anatomical variability.Yet, the limitations of the study must also be considered.
First, no quantitative investigation of the differences in annotations was performed.Due to the numerous factors that influence the annotations of the neuroanatomists, we believe that it is not warranted to assess the differences in annotations quantitatively.Furthermore, focusing on quantitative measures of inter-rater agreement may lead to false conclusions and obscure the fine-grained picture that we are describing in this study.Instead, we show that considering the complexity of this work is of utmost importance when investigating the fine-grained neuroanatomy of the MTL cortex.
Second, the generalizability of our results is limited as the presented information is based on a small sample of three specimens.It has been reported that the location of MTL cortex subregions varies with collateral sulcus depth and ramification (Ding & Van Hoesen, 2010;Insausti et al., 1998).Although we included samples with different collateral sulcus depths and ramifications to account for this to some extent, a sample size of three specimens is clearly insufficient to fully characterize the role of anatomical variants in MTL subregion delineations.Unfortunately, because of the time and effort required ($9 h per specimen) to generate a dataset comprising the entirety of macroscopic MTL cortex phenotypes it was not possible to perform the current comparison between neuroanatomical schools in a larger sample.Nevertheless, we believe the characterization of the three specimens in this study still provide valuable insights.
Third, generalizability may also be impaired as we included "only" four neuroanatomists.There may be experts in the field who rely on neuroanatomical procedures that were not detailed here (e.g., intracellular filling).However, once again, we note that those techniques would be in principle limited to a small amount of the volume of the whole MTL cortex.Still, to our knowledge, the shown synthesis of neuroanatomical expertise goes beyond the efforts of other studies in the field, substantially improving upon the diverging opinions.
Finally, Nissl-stained sections provide optimal cytoarchitectural boundaries (Swanson, 2003), making them reliable for performing mass parcellations (Williams et al., 2023).Meanwhile, immunochemical markers may be used in combination with Nissl-stained sections to annotate complex and heterogeneous regions like the MTL cortex (e.g., Ding et al., 2009;Ding & Van Hoesen, 2010).The fact that we utilized Nissl-stained sections may have negatively affected the certainty of some neuroanatomists.However, the additional use of complementary techniques would not disambiguate the borders for all neuroanatomists.The time and effort required to implement such techniques, thus, do not justify their use.

| CONCLUSION
In summary, our detailed histological investigation shows that, aside from some disagreements, there is high overlap of definitions and delineations of subregions in the MTL cortex.These findings not only advance our neuroanatomical understanding of this important brain region but also provide a rich foundation for all neuroimaging research investigating the human MTL.Moreover, the in-depth understanding of disagreements among neuroanatomists likely translates to other brain regions, where similar degrees of disagreement are potentially present.This is evident for the hippocampus, where differences in annotations were observed among neuroanatomists (Olsen et al., 2019) and for neocortical regions as evidenced by different automated segmentations for in vivo MRI (Huizinga et al., 2021;Gomez-Ramirez et al., 2022), potentially reflecting underlying differences in cytoarchitectonic definitions as well.

F
I G U R E 1 Overview of the cortical MTL structures investigated in the present study.Panel (a) shows an inferior-medial view of the brain cut at the midline with labels for relevant gyri and sulci.Panel (b) approximates two historical human brain maps (bI: Brodmann, 1909; bII: von Economo & Koskinas, 1925).Panel (c) is a three-dimensional visualization of the MTL cortex (in green) embedded in the left hemisphere indicating important terminology and identifying the approximate position of panels (d) and (e).Panels (d) and (e) visualize the rough position of the MTL cortex subregions investigated in this study.Panel (d) shows an anterior coronal section of the MTL cortex including the ERC and BA 35 and 36 labels.The color gradient indicates certainty of borders (blue: higher agreement, gray: lower agreement).Panel (e) shows a more posterior section of the MTL cortex showing the PHC.Panel (f) explains the abbreviations used in panels (a-e).
Figure 4a (see FigureS2for the other specimens).Generally, there was high agreement on border placement particularly in the midsection of the ERC with slight variability shown at its anterior and posterior ends (Figures3a,c and 4a).Indeed, neuroanatomists stated that the seminal cytoarchitectonic features of the ERC emerge only gradually at these anterior and posterior transitional sections, making annotations more difficult.An example of gradually developing layer II cell islands and lamina dissecans in the anterior ERC is shown in Figure5a.In line with the previous neuroanatomical literature, the ERC was consistently bordered anteriorly by the rhinal/collateral sulci Abbreviations: BA, Brodmann area; ERC, entorhinal cortex; L, layer; PHC, parahippocampal cortex.a Note that some but not all neuroanatomists refer to this layer as IIIu.b Medial PHC was used by O.K. and TH by the other neuroanatomists.c TF can be divided into a medial and lateral part (TFm and TFl).TL is synonymous with TFm.TF was annotated by J.C.A. and R.I. TL/TFm was annotated by S.L. D. O.K. referred to this region as lateral PHC.

F
I G U R E 4 Overview of annotations of all four MTL cortex subregions in specimen 1 visualizing the overlap among neuroanatomists.Note the lower agreement (i.e., lighter colors) on transitional slices compared with higher agreement (i.e., darker colors) in midsections.The analogous visualization of annotations in specimens 2 and 3 can be found in FigureS2.Millimeters indicate distance from the temporal pole.BA, Brodmann area; ERC, entorhinal cortex; PHC, parahippocampal cortex.F I G U R E 5 Gradual transition of cytoarchitectonic features in the MTL cortex.Panel (a) shows the gradual transition of the seminal cytoarchitectonic features of the ERC in an anterior-to-posterior direction in the anterior MTL cortex in specimen 2 (a-I to a-IV).Layer II cell islands are increasingly expressed moving posteriorly (a in a-I to a-IV).At the anterior level (b in a-I), layer IV comprises granule cells.In a-II and a-III cell-free sublaminae (laminae dissecans) can be identified.Two laminae dissecans are visible in the very posterior slice (b in a-IV).Note that some neuroanatomists only refer to the external cell-free sublamina as lamina dissecans.The annotations of the respective section are indicated below each slice.Panel (b) shows the gradual change from lamina dissecans (periallocortex) to a fully expressed granular layer IV in the medial-to-lateral direction along the collateral sulcus (cs) in specimen 1 (left slice).AT8 staining in panel (b) (right slice) exemplifies the wedge-like appearance of layer III (medial dotted line), made up of uniquely large pyramidal cells (referred to as layer IIIu by some neuroanatomists), which is characteristic for BA35.Panel (b1) shows a magnified image of the unique layer III in medial BA35 using AT8 staining.Panels (b2-4) indicate the increasing thickness, cell density, and organization of layer IV moving laterally along the collateral sulcus.Neuroanatomists annotated neighboring sections 0.5 mm anterior to the slices displayed in (b) as slices with both AT8 staining and Nissl staining-based annotations were unavailable according to the study protocol.Neuroanatomists annotated the section neighboring (b2) BA35 (2/4) and BA36 (2/4), the section neighboring (b3) BA36 (3/4) and transition from BA35 to BA36 (1/4), and the section neighboring (b4) BA36 (3/4) and non-MTL cortex (1/4).Millimeters indicate distance from the temporal pole.BA, Brodmann area; ERC, entorhinal cortex.
e., orthogonal to anterior commissure-posterior commissure (AC-PC) line) the irregular surface of the MTL-alongside the obliquity of sulci and other gross anatomical landmarks-can add complexity and ambiguity to the interpretation of boundaries.In such cases, it is possible that sections of one subregion are interrupted by portions of other types of cortices (e.g., appearance of olfactory cortex surrounded by the ERC).In the present dataset, slices were cut perpendicular to the long axis of the hippocampus to aid the translation to thick-sliced in vivo MRI.For most neuroanatomists this is different from the common coronal sections cut perpendicular to the AC-PC line direction, which may have decreased certainty in border placement as well.While we observed disagreement of annotations in transitional Braak and Braak (1991)uropathological staging were derived from contralateral sampling.Abbreviations: A, amyloid score based on Thal phases for amyloid deposits; ADNC, Alzheimer's disease neuropathologic change; B, neurofibrillary tangle stage, modified fromBraak and Braak (1991); C, CERAD neuritic plaque score; NFT, neurofibrillary tangles (AT8 staining); PSP, progressive supranuclear palsy.
Hyman et al., 2012)considered protected health information in the US, the exact age of this specimen is not disclosed.bPathologyseverity rating (A:B:C score; seeHyman et al., 2012).
T A B L E 3 Key features of the MTL cortex subregions.