Detailed segmentation of human hippocampal and subicular subfields using a combined approach

Hippocampal and subicular subfields (dentate gyrus, CA4, CA3, CA2, CA1, prosubiculum, subiculum, presubiculum and parasubiculum) of human brains have become the major targets of extensive ex vivo and in vivo neuroimaging studies. The subfield segmentation in previous studies relied heavily on parcellation based on conventional Nissl stains. Our recent study (Ding and Van Hoesen, J Comp Neurol, 523:2233-2253, 2015), however, has revealed significantly more reliable and accurate parcellation of all the subfields in both hippocampal head and body regions using a combined approach. This approach combines traditional Nissl staining and modern neurochemical labeling on closely adjacent sections. The resulting atlases of human hippocampi in our recent study are further supported by gene expression patterns in each subfield and thus will provide a very helpful guide for hippocampal and subicular segmentation on neuroimaging scans. A comparison of our recent results with those in previous studies is also discussed.

The hippocampal formation (HF) mainly includes hippocampal and subicular cortices.The former is composed of the dentate gyrus (DG) and hippocampal subfields CA1-4 [1]   while the latter consists of four or five subicular subfields including prosubiculum (ProS), subiculum (Sub), presubiculum (PrS), postsubiculum (PoS; caudal part of PrS) and parasubiculum (PaS) [2,3] .Many studies have demonstrated that HF plays a critical role in learning, memory, motivation, emotion, stress and reward [4,5] .Abnormalities and changes in HF have been observed in many major neurological and mental diseases such as Alzheimer's disease, medial temporal lobe epilepsy, schizophrenia and major depression [6,7,8,9] .To further investigate specific functions of the HF subfields, and to detect early changes of these subfields in associated diseases, many researchers in human neuroscience have put their efforts into defining reliable, accurate and automatic segmentation of the HF subfields on MRI scans [10,11,12,13] .The prerequisite for these segmentations is accurate histology-based parcellation of the HF subfields and reliable correlation of these subfields with visible macroanatomic landmarks, since the resolution of current MRI techniques has not reached the cellular level.Current segmentations of human HF subfields on MRI scans are mainly based on several histology-based parcellations or atlases: Rosene and Van Hoesen [2] , Amaral and Insausti [14] , Duvernoy [15] and Mai et al. [16] .These parcellations are mainly based on Nissl and/or myelin stains with little or no supporting neurochemical data.As pointed out by Lorente de No [1] , it was very difficult to confidently define the borders of HF subfields based only on Nissl stains.More recent investigation of human medial temporal cortices has shown that a combined analysis of multiple cellular markers can reveal reliable, consistent and accurate subfields of complex
Inspired by the success of this combined approach, we have recently investigated the hippocampal head and body regions using the same approach as well as using dense sequential section sampling [20] (see Fig. 1 for summary).In this recent study we utilized analysis of the distribution patterns in the HF with combinations the following cellular chemicals: neuronal nuclear antigen (NeuN) or Nissl substance, parvalbumin (PV), calbindin-d28k (CB) and a non-phosphorylated neurofilament protein (NPNFP, stained with SMI-32 antibody) [20] .Our main results are the identification of clear-cut boundaries of all hippocampal and subicular subfields and detailed organization of the HF in the complex and curved hippocampal head region.In addition, these findings also suggest that the boundaries of the HF subfields determined using only Nissl stain are not very accurate and reliable.The following is a comparison of our major findings [20] with those in previous studies.Some recently available molecular evidence supporting our definition of HF subfields is also provided on the base of analysis of raw human brain data downloaded from the Allen Institute website: www.brain-map.org).

Identification of ProS and its separation from Sub and CA1
A very challenging task in human HF segmentation is to define ProS and determine its boundaries.According to Lorente de No [1] , the term ProS was first introduced by Vogt in 1919 to assign the region between Sub and the lower blade of Ammon's horn (i.e., CA1) although its extent and borders were not described or shown.Lorente de No [1] labeled ProS as a large region located between typical CA1 and the Sub, which was much smaller than the ProS, and further subdivided the ProS into three subregions (ProSa, b and c).Later, Rosene and Van Hoesen [2] redefined the boundaries of the ProS based on a combined analysis of Nissl stain and AChE histochemistry, which strongly labeled the ProS but not adjoining CA1 and Sub, making the borders of the ProS much sharper and more accurate.Thus the redefined ProS by the later authors [2] is a narrow zone (about 1.2-1.6 mm in width in the hippocampal body region) and roughly corresponds to Lorente de No's ProSc and a small part of ProSb.ProSa and a large part of ProSb belong to Sub.For this reason (may be also others), some authors chose to treat the strongly AChE-labeled narrow zone as an overlap region of CA1 and Sub [14,15,16] .Moreover, Amaral and Insausti [14] proposed that the term ProS should be dropped unless more evidence is available for the establishment of the existence of the ProS.Recently a comparative study across human, monkey and rodent [3] has provided several lines of evidence for the existence of the redefined ProS since this ProS has cyto-, chemo-and geno-architecture as well as connectivity that is distinct from its neighbors (CA1 and Sub).As revealed with neurochemical markers such as AChE and TH [2, 3]   , the ProS in human hippocampal head region (i.e., ProSu, u for uncal [20] ) is not narrow at all but rather wide [3,20] (about 3-5 mm in width; also see Figs. 1C, D; 2A-C).The ProSu corresponds to the region between CA1u and Subu; both ProSu and typical ProS display weaker PV and NPNFP labeling [20] .The ProSu/ProS is a long, sizeable zone that extends all the way from hippocampal head to tail regions.Therefore, it is not proper to treat ProS as an overlap zone of CA1 and Sub.This also implies that previous inclusion of ProS into Sub is less accurate [14,15,16] .Taken together, ProS/ProSu is an independent entity of the HF and is distinguished by strong AChE, TH and neurotensin staining and weaker PV, NPNFP, NEFH, PCP4 and GFRA1 labeling [3, 20]   .In contrast to ProS/ProSu, the Sub/Subu was strongly labeled with many other markers such as GFRA1 [3] , NEFH (or NPNFP, see Fig. 2 Inset), GABRQ (Fig. 3B), RGS4 (Fig. 4A), and PCP4 (Fig. 4C).

Definition of CA2 boundaries with CA3 and CA1
Another challenging task in human HF segmentation is to determine CA2 boundaries since this would not only affect the extent of CA2 but also that of CA3 and CA1 fields.In previous studies the borders of CA2 were variably defined by different authors, but one key feature which distinguishes CA2 from CA3 has been generally accepted.This feature is that CA2 does not receive mossy fiber innervation from DG while CA3 does [1,2,14,15,20] .Unfortunately, either positive or negative mossy fiber staining was not correlated with commonly used Nissl stain on adjacent sections in most previous studies [14,15,16,21] .For this reason, the accuracy of CA2 borders needs to be reexamined and clarified with reference to mossy fiber layer.For example, CA2 was described by some authors to have the most compact and narrow pyramidal cell layer (py) on Nissl preparations [14,15] .In contrast, other authors showed that CA2 is the region distal to the compact and narrow py region (see Fig. 5A of Braak [21] and Fig. 8B of Rosene and Van Hoesen [2] ).In fact, it is very difficult to locate the CA2/CA3 border without referring to the mossy fiber layer, as concluded by Lorente de No [1] .In our recent study of human hippocampus [20] , adjacent sections were stained for mossy fiber markers (CB and/or NPNFP) and cytoarchitectonic markers (Nissl or NeuN), respectively.We found that CA2 corresponds consistently to a wider and less compact py part of the hippocampus, which is located distal to the compact and narrow py region; the latter region roughly corresponds to CA3a [20] .We also determined that strong CB immunoreactivity (IR) is one of the reliable markers for human CA2 [20] ; this is consistent with the results from in situ hybridization (ISH) study [22] .As shown in Fig. 3A, B and Inset 1, the relatively strongly CB-ISH stained CA2 region does not correspond to the most narrow py region but is rather distal to this narrow region, which actually corresponds to CA3a (Fig. 3).Another gene marker for human CA2 is PCP4, which strongly labels CA2 but lightly  (CaT) and choroid plexus (#) in the inferior lateral ventricle (iLV).Inset: Two ex vivo MRI images parcellated according to our HF atlas as summarized in Figure 1.The left MRI image is at the level of Fig. 4C or Fig. 1H and the right one at about the level between Fig. 1E and F. The white arrowhead in the left image indicates the very shallow depression on the dorsal aspect of the py, which often marks the border between CA3 and CA2 according to our recent study [20] .Bars: 797μm in A-C; 1cm in the Inset.

Existence of CA4 and its difference from CA3
One unresolved topic in HF parcellation is related to the existence of CA4.The term CA4 was first introduced by Lorente de No [1] , who assigned it to the hilar portion of the hippocampus across rodent, monkey and human brain based on Nissl preparations and/or Golgi staining.Later studies confirmed the existence of CA4 [2,15,21] , and clearly separated CA4 from the polymorphic layer of the DG (DGp) in the hilar region of monkey and human hippocampi [2] .However, the existence of CA4 in rodent is questioned by Amaral [24] who found that the CA4 labeled by Lorente de No in the hilar region of the rodent is actually DGp; this was confirmed by others [2] .Amaral and Insausti [14] further suggested that since rodent does not have a CA4, the term CA4 used in monkey and human should be dropped to avoid confusion.Consequently, the CA4 region in monkey and human brains was subsumed into CA3 in their studies [14,25] .In other cases, a modified term, hilar portion of CA3 (CA3h) was used by some authors in order to distinguish it from typical CA3 [20,26,27] .To clarify if CA4 or CA3h is really distinct from typical CA3, two human brain ISH datasets of the Allen Institute (http://human.brain-map.org [22]and http://brainspan.org [28]) were searched for differential gene expression of CA4/CA3h versus typical CA3.As shown in Fig. 3B, strong GABRQ expression was found in typical CA3 while almost no expression was detected in CA4/CA3h and DGp.Similarly, strong and weak expression of HTR2C was seen in CA4/CA3h and DGp, respectively (data not shown).Furthermore, DGp can be distinguished from CA4/CA3h using other gene markers such as CARTPT, TRHDE, ADCYAP1, semaphorin 3C and tachykinin 1-3, all of which were expressed strongly in DGp but weaker or weak in CA4/CH3h (Figs. 3 Insets 3 and 4; 5A, C, D).Therefore, the evidence presented here has demonstrated that the hilar region of human hippocampus is composed of two distinct subregions: DGp and CA4/CH3h.Given clear differences between CA4/CA3h and typical CA3, and the fact that the term CA4 has been commonly used since Lorente de No in human literature by many groups of researchers [2,8,12,15,21] , the term CA4 rather than CA3 should be encouraged although CA3h is also acceptable.

Identification of the border between PrS and PaS
The border between PrS and PaS can be difficult to identify although the border between PrS and Sub can be easily defined in Nissl preparations.In our recent study, we found that the PrS/PaS border can be easily identified in PV-stained sections, in which the superficial principal layer (SPL) of the PrS was strongly stained while adjoining PaS (PaSb) was very weakly labeled [20] .In addition, PDYN and PCP4 are found to be reliable gene markers for this border because these two genes were strongly expressed in SPL of the PrS with very weak expression in the superficial part of the entire PaS (Fig. 4B, C).In the same sections, PaS can be easily distinguished from adjoining entorhinal cortex (EC) since the EC shows strong expression of PDYN (in layers 2 and 3a) and PCP4 (in layers 2 and 3b) (see Fig. 4B, C).Lastly, it should be pointed out that PrS and PaS gradually shifts from more lateral to more medial locations along A-P axis of the hippocampus (see Fig. 1).This finding could be helpful for HF segmentation on MRI scans across A-P levels.Previously, the PrS and PaS were often merged into Sub in MRI segmentation studies [12] .

Segmentations along anterior-posterior axis of human HF
In addition to the reliable and accurate HF subfield segmentation described above, our recent study also revealed detailed organization and parcellation of the hippocampal head region [20] .Briefly, typical HF subfields extend into the head region lateroventrally (typical subfields) and then modified slightly as the uncus forms and turns posteriorly (uncal subfields).Finally, at the bottom of the uncus, the hippocampal subfields turn vertically toward the dorsal aspect of the uncus (vertical subfields).The vertical subfields are greatly modified and display distinct cyto-and chemo-architecture from the typical subfields [20] .No detailed studies have been carried out on the head region previously [2,3,14,15,16] and thus the entire head region of human hippocampus was treated as the equivalent of rodent ventral hippocampus [29] .Our results suggest differential functions of human HF along A-P axis with vertical and uncal hippocampus probably corresponding to rodent ventral hippocampus.In fact, differential gene expression can be found in typical, uncal and vertical HF subfields.For instance, strong TRHDE expression was seen in typical CA2-4 and DGp but was much weaker in the uncal CA2-4; there was little or no expression in vertical CA2-4 (Fig. 5A,  B; u for uncal and v for vertical).Strong GABRQ expression was observed in typical and uncal CA3 but not in vertical CA3 (Fig. 3B).In contrast, uncal CA4 and vertical CA3-4 but not typical CA3-4 display strong ADCYAP1 expression (Fig. 5C) and vertical but not uncal and typical CA3 show strong GABRG1 expression (data not shown).Much stronger CARTPT expression was found in uncal and vertical subfields than typical subfields (Fig. 5D).

Accurate parcellation of HF subfields on MRI scans
One of the main findings in our recent study [20] is the generation of three versions of a detailed atlas for the human hippocampal head region.No previous studies have produced such detailed atlases [14,15,16] .To show how our HF atlases

Figure 1 .
Figure 1.Summary of HF subfield parcellation in human hippocampal head and body regions.This figure is modified and color-coded from figure 6 of our recent research article [20]).A large portion of the prosubiculum (ProSu) is seen medial to CA1u in the anterior head region (A-E).Note that DGp (p in H) was included in the DG/DGu while CA3h was replaced with CA4/CA4u (4/4u).Bar: 1.0cm.

Figure 2 .
Figure 2. Border-definition of ProSu, CA1u and Subu supported by gene expression patterns.A-C: Closely adjacent sections stained for TH (A), HTR2A (B) and Nissl (C) showing distinct gene expression in ProS/ProSu from CA1/CA1u.TH and HTR2A are strongly expressed in ProS/ProSu and CA1/CA1u, respectively.Note the strong expression of HTR2A in layer 5 of the EC and lateral nucleus of the amygdala in B. Inset: NPNFP-stained sections in the anterior HF showing distinct labeling intensity in Sub/Subu (strong) and ProS/ProSu (weak).Bars: 797μm in A; 1600μm in B and C.

Figure
Figure 4. Gene expression in subicular complex and adjoining regions.A: Strong RGS4 expression was seen in Sub, PrS, CA2, CA3a and EC.B: Strong PDYN expression was observed in PrS with much weaker expression in adjoining PaS.Strong expression was also found in layers 2 and 3a of the EC.C: Strong PCP4 expression was seen in Sub, PrS and the EC (L2 and L3b) as well as in CA2 and DG (granular cell layer).Note the strong expression in the caudal caudate nucleus(CaT) and choroid plexus (#) in the inferior lateral ventricle (iLV).Inset: Two ex vivo MRI images parcellated according to our HF atlas as summarized in Figure1.The left MRI image is at the level of Fig.4Cor Fig.1Hand the right one at about the level between Fig.1E and F. The white arrowhead in the left image indicates the very shallow depression on the dorsal aspect of the py, which often marks the border between CA3 and CA2 according to our recent study[20] .Bars: 797μm in A-C; 1cm in the Inset.

Figure 5 .
Figure 5. Differential gene expression along anterior-posterior axis of hippocampus.A and B: TRHDE expression in vertical/uncal HF (A) and typical HF (B).Note the weak, intermediate and strong expression respectively in vertical, uncal and typical hippocampal subfields.Note also the strong expression in the choroid plexus (#).C: ADCYAP1 expression in vertical, uncal and typical HF subfields.Strong and weak expression was observed in CA4v/CA4u of the vertical/uncal and typical hippocampus, respectively.Differential expression in DGp-v, DGp-u and DGp was also observed.Note the strong expression in layer 2 of the PrS and PaS and in the choroid plexus (#).D: CARTPT expression in vertical/uncal and typical HF subfields.Strong versus weaker expression was found in the vertical/uncal versus typical hippocampus.Bars: 797μm in A and B; 806μm in C and D.