Genomic Anatomy Of The Hippocampus

Neuron

Article Genomic Anatomy of the Hippocampus Carol L. Thompson,1 Sayan D. Pathak,1 Andreas Jeromin,1 Lydia L. Ng,1 Cameron R. MacPherson,2 Marty T. Mortrud,1 Allison Cusick,1 Zackery L. Riley,1 Susan M. Sunkin,1 Amy Bernard,1 Ralph B. Puchalski,1 Fred H. Gage,3 Allan R. Jones,1 Vladimir B. Bajic,2 Michael J. Hawrylycz,1 and Ed S. Lein1,* 1Allen

Institute for Brain Science, Seattle, WA 98103, USA African National Bioinformatics Institute, University of the Western Cape, Bellville 7535, Cape Town, South Africa 3Salk Institute for Biological Studies, La Jolla, CA 92037, USA *Correspondence: [email protected] DOI 10.1016/j.neuron.2008.12.008 2South

SUMMARY

Availability of genome-scale in situ hybridization data allows systematic analysis of genetic neuroanatomical architecture. Within the hippocampus, electrophysiology and lesion and imaging studies demonstrate functional heterogeneity along the septotemporal axis, although precise underlying circuitry and molecular substrates remain uncharacterized. Application of unbiased statistical component analyses to genome-scale hippocampal gene expression data revealed robust septotemporal molecular heterogeneity, leading to the identification of a large cohort of genes with robust regionalized hippocampal expression. Manual mapping of heterogeneous CA3 pyramidal neuron expression patterns demonstrates an unexpectedly complex molecular parcellation into a relatively coherent set of nine expression domains in the septal/temporal and proximal/distal axes with reciprocal, nonoverlapping boundaries. Unique combinatorial profiles of adhesion molecules within these domains suggest corresponding differential connectivity, which is demonstrated for CA3 projections to the lateral septum using retrograde labeling. This complex, discrete molecular architecture provides a novel paradigm for predicting functional differentiation across the full septotemporal extent of the hippocampus. INTRODUCTION The role of the hippocampus in learning and memory is well established, with increasing evidence for a role in anxiety-related behaviors as well (Bannerman et al., 2004). These functions are differentially distributed along the septotemporal axis of the hippocampus. Septal (dorsal) hippocampal lesions result in selective impairment in spatial memory tasks but not anxietyrelated measures, whereas temporal (ventral) lesions result in the converse impairments (Bannerman et al., 2002; Moser and Moser, 1998b). Functional differentiation along the long axis of the hippocampus is evolutionarily conserved and has been described in rats (Bannerman et al., 2002; Jung et al., 1994; Moser and Moser, 1998a; Vann et al., 2000), monkeys (Colombo

1010 Neuron 60, 1010–1021, December 26, 2008 ª2008 Elsevier Inc.

et al., 1998), and humans (Small et al., 2001) using various approaches, including selective lesions and behavioral analysis, c-fos activation, electrophysiology, and functional MRI. Copious anatomical evidence exists for differential afferent, efferent, and intrahippocampal connectivity along the septotemporal axis of the hippocampus that could underlie functional differences. The entorhinal cortex, providing the major input to the dentate gyrus (DG) via the perforant path, can be divided into bands projecting to different septotemporal levels of the DG. Lateral and intermediate bands through the medial and lateral entorhinal cortex, which convey most of the information from sensory cortical areas via the perirhinal and postrhinal cortices, preferentially target the septal half and third quarter of the DG, respectively. A medial band through the entorhinal cortex, in contrast, targets the most temporal quarter of the DG (Burwell and Amaral, 1998; Dolorfo and Amaral, 1998; Ruth et al., 1988; van Groen et al., 2003). Amygdalar hippocampal afferents selectively target temporal portions of CA3, CA1, and subiculum (Petrovich et al., 2001). Differential afferent projections are also reflected by neurochemical localization, with greater cholinergic innervation of septal hippocampus (Amaral and Kurz, 1985; Milner et al., 1983) and greater temporal innervation from fibers containing dopamine (Verney et al., 1985), noradrenaline, and serotonin (Gage and Thompson, 1980; Haring and Davis, 1985; Kohler et al., 1981) and various neuropeptides (Caffe et al., 1987; Gall et al., 1981; Kohler et al., 1987; Mantyh et al., 1984; Pazos et al., 1985; van Leeuwen et al., 1985). In contrast to the simple lamellar model of intrahippocampal connectivity (Anderson et al., 1971), projections from the DG to CA3 to CA1 exhibit complex proximal/distal and septotemporal topography (Amaral and Witter, 1989; Ishizuka et al., 1990). Furthermore, recurrent associational projections in the hilus and CA3 project extensively along the septotemporal axis (Ishizuka et al., 1990). Hippocampal efferents project topographically onto the lateral septum, with septotemporal levels of CA3 and CA1 targeting distinct portions of the septum that in turn reciprocally innervate functionally discrete hypothalamic regions (Risold and Swanson, 1997). In addition, there is reciprocal connectivity between the temporal hippocampus and the amygdala (Petrovich et al., 2001), and temporal CA1 and subiculum selectively project to prefrontal cortex (Verwer et al., 1997). Septal and temporal hippocampus also differ in cellular and circuit properties. Septal hippocampus has more place cells than temporal hippocampus, and temporal place cells have lower spatial selectivity (Jung et al., 1994). Temporal CA1 shows greatly reduced long-term potentiation (LTP) induction relative to

Neuron Genomic Anatomy of the Hippocampus

septal CA1 (Papatheodoropoulos and Kostopoulos, 2000), whereas temporal CA1 selectively exhibits long-term depression (LTD) with low-frequency burst stimulation (LFBS) (Izaki et al., 2000). In models of epilepsy, temporal hippocampus exhibits more rapid kindling (Racine et al., 1977), and slices through temporal hippocampus exhibit greater epileptiform bursting in high potassium (Bragdon et al., 1986). In contrast, septal CA1 shows selective vulnerability to ischemic insults (Ashton et al., 1989). Differential functional connectivity and intrinsic cellular function are likely to be reflected by specific patterns of gene expression. There is extensive evidence for gene expression specifically delineating classically defined hippocampal subfields CA1, CA2, CA3, and the DG (Datson et al., 2004; Lein et al., 2004, Lein et al., 2005). Recently, several genes have been identified with differential septotemporal expression in CA1 (Leonardo et al., 2006), CA3, and DG (Lein et al., 2006). The recent availability of genome-scale in situ hybridization (ISH) data (Lein et al., 2006) allows a ‘‘genomic anatomy’’ approach to understanding the functional architecture of the hippocampus based on correlated expression patterns across many genes. The present study uses a combination of statistical component analysis techniques and more traditional manual anatomical boundary mapping to assess the molecular and cellular architecture of the hippocampus, identify boundaries and domains within hippocampal subfields, and generate a high-resolution, molecularly defined anatomical map that can be related to differential functional properties along the septotemporal axis of the hippocampus. RESULTS Computational Analysis of Hippocampal Gene Expression The Allen Brain Atlas (ABA) provides a powerful data set for analysis of statistical relationships between different spatially separated neuronal populations. These data consist of cellular resolution ISH data for >20,000 unique transcripts sampled across the entire adult (P56) C57BL/6J mouse brain, algorithmically quantified and mapped to a common anatomical coordinate framework (Lein et al., 2006). This framework allows the application of clustering and component-based methods to understand molecular anatomical subdivisions of brain architecture. In the present context, dimensions of expression level and 3D spatial coordinate can be analyzed to identify principal spatial domains displaying similar gene expression profiles. Applied to the hippocampus, the resulting relational or ‘‘genomic anatomy’’ should correlate with conventional anatomical delineations (e.g., DG, CA1, CA3) based on previous work (Lein et al., 2004, 2006), but may also suggest more complex anatomical parcellation reflecting differential ontogeny and function of specific portions of the hippocampus. Various matrix decomposition methods exist for identifying principal features of multidimensional data sets, including principal components analysis (PCA) (Hastie et al., 2001), independent component analysis (ICA) (Bell and Sejnowski, 1997), and nonnegative matrix factorization (NMF) (Lee and Seung, 1999). NMF in particular has been successfully applied to identify complex spatial and topological features and is particularly appropriate for image-based gene expression data

since the dimensions of space and expression level are correlated and nonnegative. NMF-based methods were applied to data from the ABA (www.brain-map.org). As described (Lein et al., 2006; Ng et al., 2007), images of uniformly (100–200 mm) spaced sections for each gene were algorithmically quantified and aligned in 3D to a common anatomical framework derived from a Nissl-based reference atlas (Dong, 2008). This framework consists both of gross anatomical delineations (e.g., hippocampus) and an underlying coordinate grid of (200 mm)3 3D voxels spanning the entire brain. These data were restricted to a matrix of voxels (n = 1290) spanning the hippocampus over a set of genes, where the matrix elements represent expression level for each gene at every voxel location. The original NMF method (Lee and Seung, 1999) does not explicitly account for local spatial relationships, and variants that do so have recently been described to enhance detection of local topological structure in facial feature detection (Zhang et al., 2008). Since neighboring (200 mm)3 voxels tend to exhibit highly correlated gene expression profiles in the ABA data set, we took a similar approach by modifying the NMF method (mNMF) to incorporate neighborhood constraints via Markov Random Fields (Besag, 1986) (see Experimental Procedures). mNMF decomposition was performed using 2686 genes from the ABA showing hippocampal expression and for which coronal plane data was available, as these data register more accurately to the coronal reference model used (Table S1 available online). Since the ABA was generated iteratively (Lein et al., 2006), with screening in the sagittal plane followed by coronal replicates for genes displaying regionalized expression, this coronal gene set is highly selective for heterogeneous, nonubiquitous expression patterns. The number of decomposition modes must be set a priori and was varied from 2 to 20 (Document S1). The data reduction afforded by mNMF allows readily interpretable results by plotting the resulting modes, each consisting of a set of (mostly) spatially contiguous voxels partitioning the hippocampal volume, back on to sections of the anatomical reference model (Figure 1). The initial modes correspond to canonical neuroanatomical subdivisions consisting of discrete neuronal subtypes, providing an intuitive validation of the approach to obtain meaningful partitioning of hippocampal anatomy based on gene expression. For example, the initial two modes correspond to small granule cells of the DG and pyramidal neurons of Ammon’s horn (Figure 1, second row). Decomposition into three modes further divided the pyramidal cell layer into canonical CA1 and CA3 fields (Figure 1, third row). Surprisingly, four modes differentiated a discrete temporal domain spanning all three subfields at the temporal pole (Figure 1, fourth row), indicating that temporal hippocampus displays a molecular profile distinct from the septal hippocampus. Increasing numbers of modes further divides the hippocampus, although these data are less interpretable and the statistical fit of the data as measured by intermode contrast declines steeply beyond four modes (Document S1). A similar result partitioning the hippocampus into the septal DG, septal CA1, septal CA3, and the temporal hippocampus was also obtained using a recursive correlation-based clustering method (Document S2). Another approach that empirically matched individual gene expression patterns well was to apply mNMF to the voxel sets

Neuron 60, 1010–1021, December 26, 2008 ª2008 Elsevier Inc. 1011

Neuron Genomic Anatomy of the Hippocampus

Figure 1. NMF Decomposition of Hippocampal Gene Expression (A) (Row 1) Series of Nissl-stained sections spanning the rostrocaudal extent of the hippocampus, with major fields annotated. (Rows 2–4) Results of NMF classification of hippocampal voxels into two through four modes, respectively, color coded and plotted onto the same Nissl sections in row 1. Two modes differentiate DG (blue) from the CA fields (red). Three modes correlate closely to CA1 (red), CA3 (green), and the DG (light blue). Addition of a fourth mode delineates a zone at the temporal pole of the hippocampus (yellow, arrows) spanning all subfields. (Rows 4–6) NMF decomposition of CA1, CA3, and DG derived from threemode decomposition of the entire hippocampus in row 3. The results of a two-mode decomposition are shown, dividing each subfield into rostral/septal (red) and caudal/temporal (green) division (arrows). (B and C) Heterogeneous septotemporal expression of individual gene patterns match NMF boundaries. Kctd4 is expressed in caudal ([B], right panels) but not rostral (left panel) CA1 and avoids the temporal portion of CA3 (arrowhead). Trhr is expressed in temporal DG (arrows) and temporal CA3 (arrowheads). DG, CA1, CA2, CA3: major hippocampal subfields; vSub: ventral subiculum.

contained in modes identified from the whole hippocampus decomposition (iterative mNMF). Decomposition of CA1, CA3, and dentate gyrus voxel sets from the initial three-mode partition further subdivided each region along the septotemporal axis of the hippocampus (Document S1). The first two modes for each

1012 Neuron 60, 1010–1021, December 26, 2008 ª2008 Elsevier Inc.

region are shown in Figure 1A, along with individual genes matching these modes (Figures 1B and 1C). CA1 is divided into rostral and caudal domains more or less in the coronal plane (Figure 1A), a pattern seen for the gene Kctd4 (Figure 1B). The DG is divided into dorsal/septal and ventral/temporal domains

Neuron Genomic Anatomy of the Hippocampus

(Figure 1A), mirrored by expression of thyrotropin-releasing hormone receptor (Trhr, Figure 1C). CA3 is divided into septal and temporal domains (Figure 1A), matching septal expression of Kctd4 (Figure 1B) and temporal expression of Trhr (Figure 1C). This boundary is very robust and recapitulated by many individual gene expression patterns, as described further below. These individual gene examples demonstrate that regardless of higher-order cross-subfield correlations, heterogeneous expression within each hippocampal subfield frequently occurs independently (e.g., regionalized expression of Kctd4 in CA1 and CA3 but uniform expression in the DG), and at least for the hippocampus an iterative NMF strategy is an effective means of predicting novel genetic partitions based on large-scale gene expression data. Decomposition of individual subfields into additional modes produces further septotemporal divisions, potentially indicating a significantly more complex partitioning along this axis (Document S1). Septotemporal Heterogeneity in CA3 Manual Mapping of Gene Expression Boundaries To gain a more detailed understanding of molecular partitioning along the septotemporal axis of the hippocampus, cellular resolution expression patterns of 6000 genes including the set analyzed by mNMF were manually analyzed. CA3 demonstrated a particularly high degree of heterogeneity, and over 300 genes were identified with robust regionalized expression patterns (Table S2). Many different individual gene expression patterns are observed in CA3, with differential expression along the septotemporal and proximal/distal (between the DG and CA2) axes, and less frequently between the inner and outer pyramidal cell populations in the radial dimension (Lein et al., 2006). Genes with particularly discrete boundaries allowed a detailed mapping of expression domains across the full extent of CA3 on to a reference Nissl series, facilitated by the uniform, densely sampled plane of section-matched expression data available for each gene in the ABA. Approached in this way, gene expression boundaries delineating nine discrete subdivisions within the CA3 pyramidal cell layer were identified, with numerous genes recapitulating each of these boundaries. The majority of the differentially expressed genes in CA3 obey these boundaries (Table S2), and any individual gene expression pattern could therefore be described as comprising some subset of these expression ‘‘domains.’’ It should be noted that great variability exists from gene to gene with respect to enrichment versus binary restriction to a particular region of CA3. Furthermore, many genes display some gradient-like characteristics, but tend either to taper off at one of these boundaries or appear to some degree as step gradients, showing signal drop-off at these boundaries. Figure 2 illustrates the boundaries and domains identified in CA3. To describe these complex anatomical data, individual genes demonstrating each observed boundary are shown, with approximate locations plotted on a plane-matched series of Nissl-stained sections spanning the entire rostrocaudal extent of the hippocampus. CA3 in this Nissl series is defined here as the portion of the pyramidal cell layer between the DG and the CA2 subfield (defined on the basis of robust and consistent boundaries delineated by many genes: Figure S1) (Lein et al., 2005). The regions or domains of CA3 between boundaries are

numbered 1–9 according to their proximal/distal and septotemporal locations in CA3. Regions 1–3 cover roughly the septal third of CA3 (from proximal to distal), regions 4–6 span approximately the mid-septotemporal half (from proximal to distal), and regions 7–9 cover the temporal pole of CA3. Genes were selected to demonstrate these boundaries where possible as series of extensions from the septal pole of CA3, with each gene adding one additional domain in the septal-to-temporal axis to the extent of the previous gene. For example, Ttn1 is expressed only in the septal, proximal-most (to the DG) portion of CA3 (region 1; Figure 2). Fmo1 is expressed in regions 1 and 2, extending slightly more temporally and distally. Prkcd is expressed in regions 1–3, and so on up to Dkk3, which is expressed in regions 1–8, excluding only the temporal-most tip of CA3 comprising region 9. Differentiation between small temporal regions 7–9 is shown in Figure S2. Certain boundaries are observed more frequently than others. For example, the boundaries between regions 4/7 and 6/7 are particularly robust and frequently observed, and these correspond well to the temporal CA3 division identified by NMF in Figure 1A. To demonstrate the reproducibility of these boundaries across multiple genes and specimens, 24 additional genes displaying these boundaries are shown in Figure S3. The complexity of the subdivisions delineated by gene expression is best appreciated by mapping subdomains onto a 3D model of CA3 (Figure 3). Viewed in 3D, the septal two-thirds of CA3 is clearly divided into a series of diagonal bands oriented septal-distal (toward CA2) to temporal-proximal (toward the DG; Figure 3B). This banding is highly reminiscent of the organization of recurrent associational projections (Ishizuka et al., 1990), as discussed below. The temporal pole, on the other hand, consists of a series of domains wrapped around the proximal and distal edges. Combinatorial Molecular Profiles of CA3 Domains Many more complex patterns are observed as well, consisting of subsets of the domains described above. Each subdivision of CA3 defined above displays a distinct molecular profile defined by combinatorial patterns of gene expression, exemplified by genes involved in differential cell adhesion, ionic conductance, and transcriptional regulation. Genes displaying robust regional expression in each of these functional ontological categories are plotted as a series of heat maps representing densitometric quantification of ISH signal, with subdivisions of CA3 organized linearly from septal to temporal as effectively a ‘‘flat map’’ of CA3 (Figure 4). For example, many cell adhesion genes exhibit complex, partially overlapping expression patterns, predominantly starting at either the septal or temporal pole of CA3 and ending at different boundaries in between. Individual domains are not necessarily defined by single genes; rather, each domain is defined by the combinatorial expression of sets of adhesion molecules, such that each CA3 subdivision expresses a unique complement of adhesion molecules. Similar patterns for ion channels suggest heterogeneity in septotemporal physiological properties as well. Many specific expression patterns are recapitulated across multiple genes (Table S2). A series of genes selectively expressed in regions 7–9 are shown in Figure S4. Interestingly, the CA3 NMF decomposition described above gives a similar pattern of septotemporal partitioning when the CA3 volume is decomposed into more modes

Neuron 60, 1010–1021, December 26, 2008 ª2008 Elsevier Inc. 1013

Neuron Genomic Anatomy of the Hippocampus

Figure 2. Mapping Gene Expression Boundaries in CA3 Nine divisions of the CA3 pyramidal cell layer can be identified on the basis of gene expression boundaries, displayed on a panel of eight coronal planes of section through the entire hippocampus moving from rostral (top) to caudal (bottom). Exemplar genes for each boundary are presented mainly as a series of extensions from the septal pole, with the first gene (Ttn) expressed solely in the most septal region of CA3 proximal to the DG and each gene moving to the right expressed in

1014 Neuron 60, 1010–1021, December 26, 2008 ª2008 Elsevier Inc.

Neuron Genomic Anatomy of the Hippocampus

Figure 3. Three-Dimensional Modeling of CA3 Molecular Anatomy Color-coded 3D models of major hippocampal subfields CA1 (red), CA2 (dark blue), CA3 (green), and DG (yellow) (A) and gene expression-based subdivisions of CA3 (B). Four different orientations demonstrate the organization of CA3 subdivisions, which can be seen to divide septal CA3 into a series of diagonal bands oriented septal-distal (toward CA2) to temporal-proximal (toward DG). CA2 (dark blue) is included in both models as a fiducial reference, and color coding of CA3 divisions matches that in Figure 2. 3D orientation bars: lateral, red; ventral, green; rostral, blue. Scale bar, 1 mm.

than described in Figure 1. A direct comparison of the CA3 model from Figure 2 to the NMF decomposition is shown in Figure S5. While these partitions were not easily matched to individual gene patterns, the additive parts-based nature of the NMF approach appears to predict spatially coherent domains that are reflected by the combinatorial overlap of expression patterns. Reciprocal, Nonoverlapping CA3 Gene Expression Domains CA3 gene expression boundaries are reciprocal, such that individual genes delineate each boundary from both sides. For example, Fmo1 and Mas1 reciprocally delineate the boundary between regions 2 and 3 (Figures 5A and 5B), Itga7 and Plagl1 the boundary between regions 3 and 5 (Figures 3C and 3D), Ptgs2 and Coch the boundary between regions 6 and 7 (Figures 5E and 5F), and Loxl1 and Coch the boundary between regions 4 and 7 (Figures 5G and 5H). Interestingly, the expression boundaries for these genes are stable across late postnatal development, indicating that this partitioning is established fairly early in development (Figure S6). To determine how discrete these gene expression boundaries are at a cellular level, double fluorescent ISH was used to label mRNAs for these same pairs of highly restricted genes reciprocally abutting each border on the same tissue sections. In each case expression proved to be mutually exclusive, with no detectable coexpression in cells at the boundaries. In some cases the boundaries were extremely sharp, as between regions 6 and 7 (Figures 5K and 5O) and 4 and 7 (Figures 5G and 5H). In other cases, cells labeled for each gene commingle at the boundary (e.g., Plagl1 and Itga7; Figures 5C and 5D), but no cells were observed that expressed both genes. An additional level of complexity is presented by a single-cellthick band of neurons along the border of the pyramidal cell layer and stratum oriens that strongly expresses a small set of genes including the procollagen gene Col6a1 (Figures 6A and 6I) and

suppressor of tumorigenicity 18 (St18; Figures 6E and 6I). These appear to be pyramidal neurons, since most genes are coordinately expressed throughout the pyramidal cell layer including this outer layer, and because GABAergic cell markers (and markers for nonneuronal cell populations) do not delineate a similar population along this boundary (data not shown). Some genes also appeared to selectively lack expression along the inner portion of the pyramidal cell layer, such as the potassium channel subunit Kcnq5. Colabeling for Kcnq5 with these two outer layer markers demonstrates, as above, that these genes label nonoverlapping populations of pyramidal cells (Figures 6A–6H). Further heterogeneity exists within this outer population, as St18 colabels with Col6a1 entirely in septal CA3, but in a decreasing subset moving temporally along the septotemporal axis (Figures 6I–6O). In general, the areal extent of labeled cells in this band adjacent to stratum oriens mirrors areal boundaries described above for the full radial thickness of CA3, and this expression is also annotated in Table S2 when possible. Hippocampal Connectivity and Gene Expression Boundaries It is well established that there is heterogeneous afferent, efferent, and intrinsic hippocampal connectivity that varies along the transverse and septotemporal axes, and tracing studies have indicated that relatively discrete borders are associated with many of these projections (Dolorfo and Amaral, 1998; Ishizuka et al., 1990; Risold and Swanson, 1997; van Groen et al., 2003; Verwer et al., 1997). Many of the molecular patterns we observe correlate with trends reported by tracing studies (Figure 7). For example, anterograde and retrograde labeling studies in rat subdivide the DG into three rough domains comprising the septal half and subsequent two more temporal quarters based on connections from discrete bands in the entorhinal cortex

the previous regions as well as one additional region, generally slightly more temporal and distal to the DG. The regions or expression domains circumscribed by these boundaries are numbered 1–9 based on their relative septal to temporal location. For each gene, the boundary of its expression is shown by a black bar across CA3, and color-coded arrowheads and numbers delineate the boundaries and domains between boundaries as they are added moving from left to right. These boundaries are compiled onto a plane-matched Nissl series in the left-most column. The most temporal tip (region 9) is not shown by gene expression but is delineated with black arrows in the Nissl series and for Dkk (regions 1–8). CA2 and the boundary between CA1 and ventral subiculum are delineated by blue bars in all sections. Scale bars: rows 1 and 2, 421 mm; rows 3–8, 842 mm.

Neuron 60, 1010–1021, December 26, 2008 ª2008 Elsevier Inc. 1015

Neuron Genomic Anatomy of the Hippocampus

Figure 4. Combinatorial Profiles of Cell Adhesion/Axon Guidance Molecules, Ion Channels, and Transcription Factors in CA3 Subdivisions Heat map representation of densitometric quantification of gene expression in each functional category across nine subdivisions of CA3 from septal to temporal, demonstrating clustering of enriched genes along septal to temporal subdivisions. Data are normalized to [0,1] corresponding to minimum and maximum measurements for that gene, with the corresponding continuous color scale ranging from black (0) to bright copper (1). Assignment of genes to functional categories based on AmiGO (Ashburner et al., 2000), DAVID (Dennis et al., 2003), MGI, or PANTHER (Mi et al., 2005) annotation.

(Burwell and Amaral, 1998). Likewise, we identified genes that differentially label septal and temporal DG, dividing the DG into three domains through their overlap (Figures 7 and S7). The temporal portion of CA1 specifically projects to frontal cortex (Verwer et al., 1997), a pattern observed in regionalized CA1 gene expression (Figures 7 and S7). Within CA3, gene expression boundaries divide the septal 2/3 into a series of diagonal bands oriented distal/septal to proximal/temporal, similar to the orientation of autoassociational collaterals of pyramidal cells within CA3

(Ishizuka et al., 1990) (Figure 7). Amygdalar inputs target very specific small regions in temporal CA3 (Petrovich et al., 2001), similar to the domain structure of gene expression in this region. A topographically organized projection of CA3 to the lateral septum has also been described, with different septotemporal levels of CA3 projecting to distinct portions of the septum (Risold and Swanson, 1997), and previous studies have argued that this topography is established via opposing gradients of axon guidance molecules and their receptors in the hippocampus and Figure 5. Gene Expression Defines Reciprocal Boundaries in CA3 (A–H) ISH data for a set of genes exemplifying the reciprocal, nonoverlapping nature of CA3 gene expression boundaries. (A) Fmo1, (B) Mas1, (C) Itga7, (D) Plagl1, (E) Ptgs2, (F and G) Coch, (H) Loxl1. White arrows in (A)–(H) indicate the location of CA2, and black arrows and corresponding numbers indicate the specific boundary being illustrated in each panel. (I–P) Double fluorescent ISH for pairs of genes defining reciprocal boundaries in CA3 at low magnification (I–L) or high magnification (M–P). Sections are counterstained with DAPI (blue). (I and M) Mas1 (green) and Fmo1 (red), (J and NJ) Plagl1 (green) and Itga7 (red), (K and O) Coch (green) and Ptgs2 (red), (L and P) Coch (green) and Loxl1 (red). Scale bars: (A and B) 500 mm, (C–H) 500 mm, (I–L) 1 mm, (M–P) 250 mm.

1016 Neuron 60, 1010–1021, December 26, 2008 ª2008 Elsevier Inc.

Neuron Genomic Anatomy of the Hippocampus

Figure 6. Inner and Outer CA3 Pyramidal Cells Display Nonoverlapping Gene Expression Profiles Double fluorescent ISH for pairs of genes differentiating inner (adjacent to stratum radiatum) from outer pyramidal cells in CA3 at low (A, E, and I) and high (B–D, F–H, and J–O) magnification. Kcnq5 (green) is nonoverlapping with Col6a1 (red; [A–D]) or with St18 (red; [E–H]), shown as individual gene labeling (B and C and F and G) or colabeling (A and D and E and H). (I–O) St18 labels a subpopulation of Col6a-expressing cells with differential colabeling in septal (J and M), midseptotemporal (K and N), or temporal CA3 (L and O) shown with (J–L) and without DAPI (M–O). Scale bars: (A and E) 1 mm, (B–D and F–H) 50 mm, (I) 500 mm, (J–O) 50 mm.

septum (Zhang et al., 1996). To test directly whether molecular boundaries observed in CA3 correlate with efferent connectivity, we combined retrograde labeling from CA3 axon terminals in the septum with ISH for a marker gene delineating the boundaries described above. Injections of Alexa Fluor 555-conjugated cholera toxin into the lateral septum back-labeled CA3 pyramidal cells. Three cases are shown in Figure 8, demonstrating that cells retrogradely labeled from the caudal lateral septum consistently avoid the portion of CA3 proximal to the dentate gyrus (Figures 8C, 8H, and 8M), corresponding closely to the domain 2/3 boundary described above. Colabeling these sections for a marker of this boundary, Fmo1 (see Figures 2 and 5) showed a close apposition of this gene expression boundary with the extent of backlabelled neurons (Figures 8E, 8J, and 8O). DISCUSSION The availability of genome-scale ISH data provides great opportunities to understand the detailed cellular and functional archi-

tecture of the nervous system on the premise that molecular phenotype reflects and/or underlies function. These data allow a data-driven, unbiased ‘‘genomic anatomy’’ approach that takes advantage of large-scale cellular resolution data to reveal organizational principles that are difficult to glean based on single-gene or targeted hypothesis-driven experimentation. Application of statistical component analysis and detailed mapping of gene expression to the hippocampus reveals a complex yet discrete molecular architecture defined by highly combinatorial expression patterns of many genes. These findings provide a molecular substrate for previously described anatomical data demonstrating complex in vivo hippocampal architecture, with a level of complexity in stark contrast to a lamellar model of hippocampal circuitry reflecting that portion of intrinsic circuitry preserved in an in vitro transverse hippocampal slice. In particular, the current findings provide strong evidence for molecular regionality that could underlie functional differentiation along the septotemporal axis of the hippocampus. This approach should be generally informative applied to other brain regions or other large-scale ISH data sets as well. Figure 7. Similarities between Molecular Domains and Extra- and Intrahippocampal Connectivity (Top panels) Schematic views of the topography of projections from the entorhinal cortex to the DG, forward and recurrent projections from CA3, and from the amygdala onto the entire hippocampal formation. (Bottom panels) 3D modeling of molecularly defined subdivisions of DG, CA3, and CA1 delineated by gene expression. Temporal CA1 is delimited by the gene Dio3 (red), while the DG is divided into three regions based on the expression of the genes Cyp7b1 (septal, red), Trhr (temporal, green), and their overlap (middle, yellow). Entorhinal-dentate (Burwell and Amaral, 1998 [Reprinted with permission of Wiley-Liss, Inc., a subsidiary of John Wiley & Sons, Inc]), CA3 intrahippocampal (Deadwyler and Hampson, 1999 [Reprinted with permission of Wiley-Liss, Inc., a subsidiary of John Wiley & Sons, Inc]), originally modified from (Ishizuka et al., 1990 [Reprinted with permission of Wiley-Liss, Inc., a subsidiary of John Wiley & Sons, Inc]), and amygdalohippocampal (Petrovich et al., 2001 [Reprinted with permission of Elsevier]) schematics reprinted with permission.

Neuron 60, 1010–1021, December 26, 2008 ª2008 Elsevier Inc. 1017

Neuron Genomic Anatomy of the Hippocampus

Figure 8. CA3 Efferent Projections to the Lateral Septum Mirror Molecular Boundaries Combined fluorescent ISH with retrograde labeling of CA3 pyramidal cells following injection of Alexa Fluor 488 conjugated-CTB into dorsal caudal lateral septum. Injection sites for three separate cases (16, 17, 18) are shown in panels (B), (G), and (L), with corresponding atlas plates in panels (A), (F), and (K) (Dong, 2008). CTB (red) labeling in CA3 (C, H, and M) closely apposes the expression domain of Fmo1 mRNA (green; [D, I, and N]). The merged image is shown at higher magnification in (E), (J), and (O). Arrows denote boundary between Fmo1 and CTB labeling. Scale bars: 500 mm (B–D, G–I, and L–N) and 200 mm (E, J, and O).

Hippocampal Boundaries, Domains, and Cellular Identities A number of organizational principles have emerged from this broad analysis of hippocampal gene expression. Perhaps the most striking finding is the degree of molecular diversity within excitatory neuronal populations, generally considered fairly homogeneous. Gene expression patterns define contiguous groups of neurons in each hippocampal subfield, generally dividing the septal or dorsal hippocampus from the temporal or ventral hippocampus. In all subfields, the sharpest gene expression boundaries involve the most temporal quarter of the hippocampus, roughly the region defined as a principle mode by the NMF analysis. Gene expression in each hippocampal subfield appears to be regulated independently. CA3 molecular architecture is characterized by a series of reciprocal, nonoverlapping boundaries extending along the septal/temporal and proximal/ distal axes, and graded expression in CA3 frequently appears as a step gradient, in contrast to more continuous gradients observed in DG and CA1. Although these regions were not analyzed in the same detail as CA3, gene expression ‘‘boundaries’’ identified in these regions appear less discrete, and for CA1 in particular, are not reciprocal in the sense that we could not identify genes that abut each boundary from both sides. Several functional categories are highly represented among differentially expressed genes in CA3, suggestive of the functional relevance for this restricted expression. By far the most abundant broad category, including more than half of all differentially expressed genes, involves differential cell adhesion, axon guidance, and cell-cell communication (including neuropeptides and neurotransmitter receptors). This high representation suggests that a great deal of septotemporal heterogeneity relates to the establishment and maintenance of differential

1018 Neuron 60, 1010–1021, December 26, 2008 ª2008 Elsevier Inc.

functional connectivity. Previous studies have postulated links between differential cell adhesion molecule expression and axon guidance in the process of establishing topographic specificity in hippocampal projections. For example, the ephrin receptor Epha5 is expressed in a medial to lateral gradient in the developing hippocampus, with its corresponding ligands expressed in a reciprocal fashion in hippocampal target tissues such as the lateral septum (Zhang et al., 1996). While axon guidance and establishment of topographic specificity are generally considered developmental phenomena, there is increasing evidence that cell adhesion molecules are involved in synaptic plasticity and maintenance of neural networks in adult animals (de Wit and Verhaagen, 2003; Pinkstaff et al., 1999). Furthermore, it is worth noting that neurogenesis persists in the adult DG, and it may be necessary to maintain cues for axonal pathfinding for newly generated granule cells to establish appropriate connectivity within CA3 (Zhao et al., 2006). For example, several homophilic cell adhesion molecules, including Pcdh21, Robo2, Cdh9, and Cdh8, are expressed both in septal DG and in septal CA3. The majority of the remaining differentially expressed genes relate in some way to intrinsic cellular functions, most notably ion channels and transcription factors. The systematic compilation of differentially expressed genes provides a window into the molecular specification of cellular identity. Clearly the unique molecular identity of a particular cell is defined by the combinatorial overlap of many expressed genes. In CA3, we have shown nine discrete regions along the septotemporal axis that can be defined by gene expression. Many of these regions are not represented uniquely by expression of any single gene, but rather can be defined by the overlap of many gene expression domains, providing each CA3 subdivision with a unique complement of expressed genes. Notably,

Neuron Genomic Anatomy of the Hippocampus

members of many different cell adhesion (cadherins, collagens, IgG superfamily) and axon guidance (ephrins, slits, robos, semaphorins) families exhibit combinatorial expression in a cascade of partially overlapping domains. While it is common to discuss the transcription factor code specifying the fate of a cell during development, it would appear that unique adhesion molecule and ion channel profiles may serve to determine functional connectivity and intrinsic electrophysiological profiles of discrete hippocampal neuronal populations as well.

CA1- and CA3-specific NMDA receptor knockouts, demonstrating differential hippocampal subregion function in spatial learning tasks (Nakazawa et al., 2002; Tsien et al., 1996). Similar transgenic approaches to recapitulate discrete septotemporal hippocampal subdivisions, coupled with an array of developing technologies for selective cell ablation (Luquet et al., 2005), neuronal silencing (Lerchner et al., 2007; Tan et al., 2006), and transneuronal labeling (Wickersham et al., 2007), hold great promise for elucidating the fine functional architecture of the hippocampus.

Functional Correlates of Molecular Boundaries To make sense of the complicated molecular anatomy of the hippocampus, it is necessary to synthesize knowledge of functional connectivity, molecular boundaries, and molecular function (e.g., Figures 4 and 7). A variety of tracing studies have indicated that there are discontinuous features associated with (topographically organized) intrinsic, afferent and efferent hippocampal projections, frequently with relatively discrete borders (Dolorfo and Amaral, 1998; Ishizuka et al., 1990; Risold and Swanson, 1997; van Groen et al., 2003; Verwer et al., 1997). In the current study, we similarly found a consistent, relatively discrete boundary of efferent CA3 projections to the caudal lateral septum by retrograde labeling and established that this boundary correlates with one of the molecular boundaries (between regions 2 and 3). Many of the other molecular patterns observed are consistent with trends observed by tracing studies (Figure 7), which, along with the overrepresentation of cell adhesion genes, suggests that a great deal of regionalized gene expression reflects underlying hippocampal circuitry. Differential physiological properties have also been reported across septal and temporal hippocampus, likely the result of differential ion channel expression (Figure 4). For example, septal versus temporal CA1 pyramidal neurons exhibit differential induction of LTP and LTD (Izaki et al., 2000; Papatheodoropoulos and Kostopoulos, 2000), while epileptiform bursting can be readily induced in CA3 of temporal but not septal hippocampal slices (Bragdon et al., 1986). Interestingly, this latter work indicated that the greatest bursting could be induced in the second-most temporal slice, suggesting a regionalized expression of ion channels underlying these responses. The hippocampus also receives differential innervation across its septotemporal extent from various neurotransmitter systems, with stronger temporal innervation by fibers containing dopamine (Verney et al., 1985), serotonin (Gage and Thompson, 1980), NPY (Kohler et al., 1987), and TRH (Pazos et al., 1985). There is a corresponding localized expression of cognate receptors in temporal CA3 of the dopamine receptor Drd2, serotonin receptor Htr4, NPY receptor Npy1r, and TRH receptor Trhr (see Table S2). The comprehensive molecular description of ion channels and receptors should be a powerful predictor for differential intrinsic membrane properties and ligand-induced excitation among discrete hippocampal neuronal populations and guide future experimentation to understand their functional relevance. Comprehensive molecular mapping based on analysis of genome-wide ISH data provides a complex molecular framework for considering the functional architecture of the hippocampus, and also the means for functional analysis through targeted genetic manipulation. This approach has been used to generate

EXPERIMENTAL PROCEDURES High-Throughput Nonisotopic In Situ Hybridization All procedures were approved by the Allen Institute Institutional Animal Care and Use Committee. P56 male C57BL/6J mice (Jackson Labs West, Sacramento, CA) were used for most experiments, except for a small developmental series using P14 and P28 animals. High-throughput data generation was performed using a semiautomated nonisotopic digoxigenin-based colorimetric ISH platform as described previously (Lein et al., 2006). Briefly, gene-specific riboprobes were hybridized to 25 mm thick postfixed sections sampled uniformly either across the entire brain or restricted to the hippocampus for high sampling density series used for 3D reconstruction of gene expression patterns. Nonnegative Matrix Factorization ISH data were quantified and registered in 3D to a common anatomical reference framework provided by a Nissl-stained reference atlas as previously described (Lein et al., 2006; Ng et al., 2007). Expression levels were calculated for each gene for the set of (200 mm)3 voxels spanning the hippocampus. These expression data were used to generate an N 3 V matrix A, representing N genes in V 3D voxels. 2868 genes displaying expression above a minimum threshold (Ng et al., 2007) and for which coronal ISH data were available were used for this analysis across 1290 voxels representing the hippocampus (see Table S1). To identify distinct subdivisions of the hippocampus based on spatial gene expression patterns, the data were decomposed into defined numbers of modes, each mode defined as some subset of the V voxels. We extended the method of Non-Negative Factorization (NMF) (Lee and Seung, 1999) using Markov Random Fields (MRF) (Besag, 1986) to account for spatial influence in the data. NMF is an iterative matrix factorization technique that differs from principal components analysis in that the matrix factors are constrained to have positive and interpretable component entries. In the present implementation (mNMF) the matrix factors are modified during each iteration of the decomposition via a MRF model to incorporate gene expression from a voxel’s spatial neighborhood. This additional use of MRF acts as a spatial smoothing and yields more robust clusters. Details and the full decomposition results for 2 to 20 modes are provided in Document S1. Expression Boundary Mapping and Densitometric Quantification of Colorimetric In Situ Hybridization Data Gene expression boundaries were initially identified by mapping approximate boundaries of 300 genes exhibiting regionalized CA3 expression onto a highdensity reference Nissl series. The vast majority of observed expression patterns could be accounted for a set of expression boundaries dividing CA3 into nine discrete subdomains. The boundaries exhibited by particularly robust exemplars, such as those shown in Figure 2, were used for final boundary delineations used to model these regions in 3D and to bound regions for quantitative analysis of gene expression across subdomains. For the 103 genes in Figure 4, bitmap images of ISH data were quantified using Scion Image (Scion Corporation, Frederick, MD) on 12 hippocampus-containing coronal ISH sections per gene. For each gene a rectangular region of interest was used to sample the mean density (average gray value) from each of the nine subregions of CA3, with five measurements collected per CA3 subdomain per gene where possible (several regions were only present on one to three available brain sections). Transition areas between subdomains were avoided when possible to improve the accuracy of the quantitation. It should be noted that this method does not discriminate between expression in pyramidal cells of the hippocampus and other cells such as interneurons, nonneuronal cells, and

Neuron 60, 1010–1021, December 26, 2008 ª2008 Elsevier Inc. 1019

Neuron Genomic Anatomy of the Hippocampus

the outer layer of CA3, and does not correct for changes in cell density across different subdomains, tending to attenuate quantitative differences between pyramidal cell expression between different subdomains and leading to generally lower values for relatively low density regions (e.g., region 1 in Figure 4). Due to large variance in background levels between sections, a method for outlier detection appropriate for small sample sizes, the median absolute deviation statistic (MAD), was used prior to plotting results in Figure 4 (Wilcox, 2001).

REFERENCES

3D Modeling of Hippocampal Gene Expression Hippocampal expression patterns were visualized in 3D by either reconstructing the raw ISH data from aligned sections (DG) or by transposing expression patterns onto a model of the hippocampus (CA1 and CA3) as described previously (Lein et al., 2005). Briefly, a reference model was created by aligning a set of high-density Nissl images spanning the hippocampus in Adobe PhotoShop (Adobe Systems, San Jose, CA), and digitally ‘‘dissecting’’ each structure of interest (e.g., CA1, CA2, CA3, DG) from surrounding brain regions for individual visualization. Observed gene expression patterns were transposed onto these reference sets to generate independent aligned image series for each division (except for DG patterns, for which raw ISH data were used on independent templates). 3D modeling and surface rendering of each region and gene expression-defined subdivision was performed using the 3D Constructor plug-in of ImagePro Plus (v. 5.1; Media Cybernetics, Silver Spring, MD).

Anderson, P., Bliss, T.V., and Skrede, K.K. (1971). Lamellar organization of hippocampal pathways. Exp. Brain Res. 13, 222–238.

Fluorescent In Situ Hybridization Double-label fluorescent ISH was performed using a variation of the colorimetric protocol. Briefly, riboprobes were labeled with either digoxigenin-UTP or dinitrophenyl-11-UTP (DNP; Perkin Elmer). A DNP-labeled probe and a DIG-labeled probe were hybridized simultaneously. Tyramide signal amplification was performed for each probe individually, using either anti-DIG-HRP with tyramide-biotin, or anti-DNP-HRP with tyramide-DNP for amplification. Visualization of signal was achieved using either streptavidin-Alexa Fluor 488 (Invitrogen/Molecular Probes) or anti-DNP-Alexa Fluor 555 (Invitrogen/ Molecular Probes), and automated fluorescence microscopy. See Document S3 for complete methodological details. Retrograde Labeling Combined with ISH Stereotaxic injections of Alexa Fluor 555-conjugated cholera toxin B subunit (Invitrogen) were made into the dorsal portion of the caudal lateral septum of P56 mice using coordinates from Paxinos and Franklin (2004) (0.4 mm M/L; 0.15–0.3 mm A/P; 2.8 mm D/V) using a Picrospritzer and pulled glass pipette. 0.2–2.0 ml of 0.5% w/v labeled toxin was delivered under isoflurane anesthesia using five pulses of 50–80 ms duration at 30 psi. Five to seven days postinjection to allow transport of the tracer, animals were sacrificed and brains processed for fluorescent ISH using Alexa Fluor 488 detection to allow visualization of retrogradely labeled cells and ISH signal on the same tissue sections. SUPPLEMENTAL DATA

Amaral, D.G., and Kurz, J. (1985). An analysis of the origins of the cholinergic and noncholinergic septal projections to the hippocampal formation of the rat. J. Comp. Neurol. 240, 37–59. Amaral, D.G., and Witter, M.P. (1989). The three-dimensional organization of the hippocampal formation: a review of anatomical data. Neuroscience 31, 571–591.

Ashburner, M., Ball, C.A., Blake, J.A., Botstein, D., Butler, H., Cherry, J.M., Davis, A.P., Dolinski, K., Dwight, S.S., Eppig, J.T., et al. (2000). Gene ontology: tool for the unification of biology. The Gene Ontology Consortium. Nat. Genet. 25, 25–29. Ashton, D., Van Reempts, J., Haseldonckx, M., and Willems, R. (1989). Dorsalventral gradient in vulnerability of CA1 hippocampus to ischemia: a combined histological and electrophysiological study. Brain Res. 487, 368–372. Bannerman, D.M., Deacon, R.M., Offen, S., Friswell, J., Grubb, M., and Rawlins, J.N. (2002). Double dissociation of function within the hippocampus: spatial memory and hyponeophagia. Behav. Neurosci. 116, 884–901. Bannerman, D.M., Rawlins, J.N., McHugh, S.B., Deacon, R.M., Yee, B.K., Bast, T., Zhang, W.N., Pothuizen, H.H., and Feldon, J. (2004). Regional dissociations within the hippocampus—memory and anxiety. Neurosci. Biobehav. Rev. 28, 273–283. Bell, A.J., and Sejnowski, T.J. (1997). The ‘‘independent components’’ of natural scenes are edge filters. Vision Res. 37, 3327–3338. Besag, J. (1986). On the statistics of dirty images. J. Roy. Statist. Soc. Ser. B. Methodological 48, 259–302. Bragdon, A.C., Taylor, D.M., and Wilson, W.A. (1986). Potassium-induced epileptiform activity in area CA3 varies markedly along the septotemporal axis of the rat hippocampus. Brain Res. 378, 169–173. Burwell, R.D., and Amaral, D.G. (1998). Perirhinal and postrhinal cortices of the rat: interconnectivity and connections with the entorhinal cortex. J. Comp. Neurol. 391, 293–321. Caffe, A.R., van Leeuwen, F.W., and Luiten, P.G. (1987). Vasopressin cells in the medial amygdala of the rat project to the lateral septum and ventral hippocampus. J. Comp. Neurol. 261, 237–252. Colombo, M., Fernandez, T., Nakamura, K., and Gross, C.G. (1998). Functional differentiation along the anterior-posterior axis of the hippocampus in monkeys. J. Neurophysiol. 80, 1002–1005. Datson, N.A., Meijer, L., Steenbergen, P.J., Morsink, M.C., van der Laan, S., Meijer, O.C., and de Kloet, E.R. (2004). Expression profiling in laser-microdissected hippocampal subregions in rat brain reveals large subregion-specific differences in expression. Eur. J. Neurosci. 20, 2541–2554.

The Supplemental Data can be found with this article online at http://www. neuron.org/supplemental/S0896-6273(08)01056-8.

de Wit, J., and Verhaagen, J. (2003). Role of semaphorins in the adult nervous system. Prog. Neurobiol. 71, 249–267.

ACKNOWLEDGMENTS

Deadwyler, S.A., and Hampson, R.E. (1999). Anatomic model of hippocampal encoding of spatial information. Hippocampus 9, 397–412.

This work was sponsored by the Allen Institute for Brain Science. The authors wish to thank the Allen Institute founders, P.G. Allen and J. Patton, for their vision, encouragement, and support. The authors also wish to acknowledge Theresa Zwingman and Maureen Howell for assistance with mining efforts to identify regional hippocampal expression patterns; Simon Smith for assistance with histological preparations; Jolene Kidney, Maureen Howell, Andrew Boe, and Tracy Lemon for assistance with tract tracing and sectioning; John Hohmann and John Morris for assistance with the developmental analysis of hippocampal gene expression; Amanda Ebbert, Lon Luong, Kimberly Smith, Melissa Reding, and Cathy Copeland for supporting production and scanning of fluorescent ISH images; Chinh Dang for supporting database needs; and Paul Wohnoutka for supporting data production. Accepted: December 8, 2008 Published: December 24, 2008

1020 Neuron 60, 1010–1021, December 26, 2008 ª2008 Elsevier Inc.

Dennis, G., Jr., Sherman, B.T., Hosack, D.A., Yang, J., Gao, W., Lane, H.C., and Lempicki, R.A. (2003). DAVID: Database for Annotation, Visualization, and Integrated Discovery. Genome Biol. 4, 3. Dolorfo, C.L., and Amaral, D.G. (1998). Entorhinal cortex of the rat: topographic organization of the cells of origin of the perforant path projection to the dentate gyrus. J. Comp. Neurol. 398, 25–48. Dong, H.W. (2008). The Allen Atlas: A Digital Brain Atlas of C57BL/6J Male Mouse (Hoboken, NJ: John Wiley & Sons). Gage, F.H., and Thompson, R.G. (1980). Differential distribution of norepinephrine and serotonin along the dorsal-ventral axis of the hippocampal formation. Brain Res. Bull. 5, 771–773. Gall, C., Brecha, N., Karten, H.J., and Chang, K.J. (1981). Localization of enkephalin-like immunoreactivity to identified axonal and neuronal populations of the rat hippocampus. J. Comp. Neurol. 198, 335–350.

Neuron Genomic Anatomy of the Hippocampus

Haring, J.H., and Davis, J.N. (1985). Differential distribution of locus coeruleus projections to the hippocampal formation: anatomical and biochemical evidence. Brain Res. 325, 366–369.

Papatheodoropoulos, C., and Kostopoulos, G. (2000). Decreased ability of rat temporal hippocampal CA1 region to produce long-term potentiation. Neurosci. Lett. 279, 177–180.

Hastie, T., Tibshirani, R., and Friedman, J.H. (2001). The Elements of Statistical Learning: Data Mining, Inference, and Prediction (New York: Springer).

Paxinos, G., and Franklin, K.B.J. (2004). The Mouse Brain in Stereotaxic Coordinates, Compact, Second Edition (Amsterdam, Boston: Elsevier Academic Press).

Ishizuka, N., Weber, J., and Amaral, D.G. (1990). Organization of intrahippocampal projections originating from CA3 pyramidal cells in the rat. J. Comp. Neurol. 295, 580–623.

Pazos, A., Cortes, R., and Palacios, J.M. (1985). Thyrotropin-releasing hormone receptor binding sites: autoradiographic distribution in the rat and guinea pig brain. J. Neurochem. 45, 1448–1463.

Izaki, Y., Takita, M., and Nomura, M. (2000). Comparative induction of longterm depression between dorsal and ventral hippocampal CA1 in the anesthetized rat. Neurosci. Lett. 294, 171–174.

Petrovich, G.D., Canteras, N.S., and Swanson, L.W. (2001). Combinatorial amygdalar inputs to hippocampal domains and hypothalamic behavior systems. Brain Res. Brain Res. Rev. 38, 247–289.

Jung, M.W., Wiener, S.I., and McNaughton, B.L. (1994). Comparison of spatial firing characteristics of units in dorsal and ventral hippocampus of the rat. J. Neurosci. 14, 7347–7356.

Pinkstaff, J.K., Detterich, J., Lynch, G., and Gall, C. (1999). Integrin subunit gene expression is regionally differentiated in adult brain. J. Neurosci. 19, 1541–1556.

Kohler, C., Chan-Palay, V., and Steinbusch, H. (1981). The distribution and orientation of serotonin fibers in the entorhinal and other retrohippocampal areas. An immunohistochemical study with anti-serotonin antibodies in the rats brain. Anat. Embryol. (Berl.) 161, 237–264.

Racine, R., Rose, P.A., and Burnham, W.M. (1977). Afterdischarge thresholds and kindling rates in dorsal and ventral hippocampus and dentate gyrus. Can. J. Neurol. Sci. 4, 273–278. Risold, P.Y., and Swanson, L.W. (1997). Connections of the rat lateral septal complex. Brain Res. Brain Res. Rev. 24, 115–195.

Lee, D.D., and Seung, H.S. (1999). Learning the parts of objects by non-negative matrix factorization. Nature 401, 788–791.

Ruth, R.E., Collier, T.J., and Routtenberg, A. (1988). Topographical relationship between the entorhinal cortex and the septotemporal axis of the dentate gyrus in rats: II. Cells projecting from lateral entorhinal subdivisions. J. Comp. Neurol. 270, 506–516.

Lein, E.S., Zhao, X., and Gage, F.H. (2004). Defining a molecular atlas of the hippocampus using DNA microarrays and high-throughput in situ hybridization. J. Neurosci. 24, 3879–3889.

Small, S.A., Nava, A.S., Perera, G.M., DeLaPaz, R., Mayeux, R., and Stern, Y. (2001). Circuit mechanisms underlying memory encoding and retrieval in the long axis of the hippocampal formation. Nat. Neurosci. 4, 442–449.

Kohler, C., Schultzberg, M., and Radesater, A.C. (1987). Distribution of neuropeptide Y receptors in the rat hippocampal region. Neurosci. Lett. 75, 141–146.

Lein, E.S., Callaway, E.M., Albright, T.D., and Gage, F.H. (2005). Redefining the boundaries of the hippocampal CA2 subfield in the mouse using gene expression and 3-dimensional reconstruction. J. Comp. Neurol. 485, 1–10. Lein, E.S., Hawrylycz, M.J., Ao, N., Ayres, M., Bensinger, A., Bernard, A., Boe, A.F., Boguski, M.S., Brockway, K.S., Byrnes, E.J., et al. (2006). Genome-wide atlas of gene expression in the adult mouse brain. Nature 445, 168–176. Leonardo, E.D., Richardson-Jones, J.W., Sibille, E., Kottman, A., and Hen, R. (2006). Molecular heterogeneity along the dorsal-ventral axis of the murine hippocampal CA1 field: a microarray analysis of gene expression. Neuroscience 137, 177–186. Lerchner, W., Xiao, C., Nashmi, R., Slimko, E.M., van Trigt, L., Lester, H.A., and Anderson, D.J. (2007). Reversible silencing of neuronal excitability in behaving mice by a genetically targeted, ivermectin-gated Cl- channel. Neuron 54, 35–49.

Tan, E.M., Yamaguchi, Y., Horwitz, G.D., Gosgnach, S., Lein, E.S., Goulding, M., Albright, T.D., and Callaway, E.M. (2006). Selective and quickly reversible inactivation of mammalian neurons in vivo using the Drosophila allatostatin receptor. Neuron 51, 157–170. Tsien, J.Z., Huerta, P.T., and Tonegawa, S. (1996). The essential role of hippocampal CA1 NMDA receptor-dependent synaptic plasticity in spatial memory. Cell 87, 1327–1338. van Groen, T., Miettinen, P., and Kadish, I. (2003). The entorhinal cortex of the mouse: organization of the projection to the hippocampal formation. Hippocampus 13, 133–149. van Leeuwen, F.W., van Heerikhuize, J., van der Meulen, G., and Wolters, P. (1985). Light microscopic autoradiographic localization of [3H]oxytocin binding sites in the rat brain, pituitary and mammary gland. Brain Res. 359, 320–325.

Luquet, S., Perez, F.A., Hnasko, T.S., and Palmiter, R.D. (2005). NPY/AgRP neurons are essential for feeding in adult mice but can be ablated in neonates. Science 310, 683–685.

Vann, S.D., Brown, M.W., Erichsen, J.T., and Aggleton, J.P. (2000). Fos imaging reveals differential patterns of hippocampal and parahippocampal subfield activation in rats in response to different spatial memory tests. J. Neurosci. 20, 2711–2718.

Mantyh, P.W., Hunt, S.P., and Maggio, J.E. (1984). Substance P receptors: localization by light microscopic autoradiography in rat brain using [3H]SP as the radioligand. Brain Res. 307, 147–165.

Verney, C., Baulac, M., Berger, B., Alvarez, C., Vigny, A., and Helle, K.B. (1985). Morphological evidence for a dopaminergic terminal field in the hippocampal formation of young and adult rat. Neuroscience 14, 1039–1052.

Mi, H., Lazareva-Ulitsky, B., Loo, R., Kejariwal, A., Vandergriff, J., Rabkin, S., Guo, N., Muruganujan, A., Doremieux, O., Campbell, M.J., et al. (2005). The PANTHER database of protein families, subfamilies, functions and pathways. Nucleic Acids Res. 33, D284–D288.

Verwer, R.W., Meijer, R.J., Van Uum, H.F., and Witter, M.P. (1997). Collateral projections from the rat hippocampal formation to the lateral and medial prefrontal cortex. Hippocampus 7, 397–402.

Milner, T.A., Loy, R., and Amaral, D.G. (1983). An anatomical study of the development of the septo-hippocampal projection in the rat. Brain Res. 284, 343–371.

Wickersham, I.R., Finke, S., Conzelmann, K.K., and Callaway, E.M. (2007). Retrograde neuronal tracing with a deletion-mutant rabies virus. Nat. Methods 4, 47–49.

Moser, M.B., and Moser, E.I. (1998a). Distributed encoding and retrieval of spatial memory in the hippocampus. J. Neurosci. 18, 7535–7542.

Wilcox, R. (2001). Fundamentals of Modern Statistical Methods (New York: Springer)., pp. 35–36.

Moser, M.B., and Moser, E.I. (1998b). Functional differentiation in the hippocampus. Hippocampus 8, 608–619.

Zhang, J.H., Cerretti, D.P., Yu, T., Flanagan, J.G., and Zhou, R. (1996). Detection of ligands in regions anatomically connected to neurons expressing the Eph receptor Bsk: potential roles in neuron-target interaction. J. Neurosci. 16, 7182–7192.

Nakazawa, K., Quirk, M.C., Chitwood, R.A., Watanabe, M., Yeckel, M.F., Sun, L.D., Kato, A., Carr, C.A., Johnston, D., Wilson, M.A., and Tonegawa, S. (2002). Requirement for hippocampal CA3 NMDA receptors in associative memory recall. Science 297, 211–218. Ng, L., Pathak, S., Kuan, L., Lau, C., Dong, H.-W., Sodt, A., Dang, C., Avants, B., Yushkevich, P., Gee, J., et al. (2007). Neuroinformatics for Genome-wide 3-D Gene Expression Mapping in the Mouse Brain. IEEE/ACM Trans. Comput. Biol. Bioinform. 4, 382–393.

Zhang, T., Fang, B., Tang, Y.Y., He, G., and Wen, J. (2008). Topology preserving non-negative matrix factorization for face recognition. IEEE Trans. Image Process. 17, 574–584. Zhao, C., Teng, E.M., Summers, R.G., Jr., Ming, G.L., and Gage, F.H. (2006). Distinct morphological stages of dentate granule neuron maturation in the adult mouse hippocampus. J. Neurosci. 26, 3–11.

Neuron 60, 1010–1021, December 26, 2008 ª2008 Elsevier Inc. 1021