“Subpial Fan Cell” — A Class of Calretinin Neuron in Layer 1 of Adult Monkey Prefrontal Cortex

Layer 1 of the cortex contains populations of neurochemically distinct neurons and afferent fibers which markedly affect neural activity in the apical dendritic tufts of pyramidal cells. Understanding the causal mechanisms requires knowledge of the cellular architecture and synaptic organization of layer 1. This study has identified eight morphological classes of calretinin immunopositive (CRet+) neurons (including Cajal-Retzius cells) in layer 1 of the prefrontal cortex (PFC) in adult monkey (Macaca fasicularis), with a distinct class — termed “subpial fan (SPF) cell” — described in detail. SPF cells were rare horizontal unipolar CRet+ cells located directly beneath the pia with a single thick primary dendrite that branched into a characteristic fan-like dendritic tree tangential to the pial surface. Dendrites had spines, filamentous processes and thorny branchlets. SPF cells lay millimeters apart with intralaminar axons that ramified widely in upper layer 1. Such cells were GABA immunonegative (-) and occurred in areas beyond PFC. Interspersed amidst SPF cells displaying normal structural integrity were degenerating CRet+ neurons (including SPF cells) and clumps of lipofuscin-rich cellular debris. The number of degenerating SPF cells increased during adulthood. Ultrastructural analyses indicated SPF cell somata received asymmetric (A — presumed excitatory) and symmetric (S — presumed inhibitory) synaptic contacts. Proximal dendritic shafts received mainly S-type and distal shafts mostly A-type input. All dendritic thorns and most dendritic spines received both synapse types. The tangential areal density of SPF cell axonal varicosities varied radially from parent somata — with dense clusters in more distal zones. All boutons formed A-type contacts with CRet- structures. The main post-synaptic targets were dendritic shafts (67%; mostly spine-bearing) and dendritic spines (24%). SPF-SPF cell innervation was not observed. Morphometry of SPF cells indicated a unique class of CRet+/GABA- neuron in adult monkey PFC — possibly a subtype of persisting Cajal-Retzius cell. The distribution and connectivity of SPF cells suggest they act as integrative hubs in upper layer 1 during postnatal maturation. The main synaptic output of SPF cells likely provides a transminicolumnar excitatory influence across swathes of apical dendritic tufts — thus affecting information processing in discrete patches of layer 1 in adult monkey PFC.

Layer 1 contains a numerically small population of excitatory and inhibitory cells -most, possibly all, being local circuit neurons (LCNs). In the medial prefrontal cortex (mPFC) of adult macaque monkeys, layer 1 has approximately 560 neurons under 1 mm 2 of pial surface -less than 0.5% of the total number of neurons in a column of cortex spanning layers 1-6 (Figures 1A-C; Gabbott and Bacon, 1996a,b). LCNs containing the inhibitory neurotransmitter gamma-aminobutyric acid (GABA) represent about 90% of neurons in layer 1 and are predominantly situated mid-lamina, whereas the GABA immunonegative (-) LCNs (presumed excitatory cells), which account for ∼10% of layer 1 neurons, are mainly located directly beneath the pia and toward the boundary with layer 2 (Figures 1D-F).
Compared with other cortical LCNs, the morphology and synaptic connectivity of CRet+ layer 1 neurons during development and persisting into adulthood remain understudied -especially in primates (Del Rio and DeFelipe, 1997;Meskenaite, 1997;Džaja et al., 2014). The detailed neuroanatomical research described here uses archived material to classify the morphological varieties of CRet+ neurons (including persisting Cajal-Retzius cells) in layer 1 of adult macaque monkey PFC. With this comparative foundation, the study then concentrates on a unique class of CRet+/GABA-layer 1 neuron, present in adulthood. The characteristic features of this exceptional neuronal phenotype are its location directly beneath the pial surface and a smooth unipolar soma with a thick primary dendrite that gives rise to a horizontal fan-like dividing arbor. As a result of the latter structural hallmark these distinctive CRet+ neurons have been named "subpial fan (SPF) cells." By using correlated light and electron microscopy, the synaptic input to identified SPF cell dendrites is defined. Furthermore, data are presented detailing the spatial distribution of SPF varicose axonal arbors and the ultrastructural identity of local and distant post-synaptic target structures. In addition to SPF cells with normal structural integrity, degenerating CRet+ neurons (including SPF cells) were present in layer 1 -for completeness, an account of SPF cells undergoing degeneration is also given.
The observations are related to the structure and function of other LCN neuron types and afferent systems ramifying in layer 1 -expressly their relation with the Cajal-Retzius cell family. Although previous studies highlight the structural complexity of layer 1 neurons in primates (Meyer et al., 1999;Rakic and Zečević, 2003) consideration is given, where appropriate, to studies in lower mammals. Also considered is the possible structural and functional relationship of SPF cells with the apical dendritic tufts in layer 1 derived from pyramidal cell minicolumns (Marín-Padilla, 1984, 2015Gabbott, 2003).
While the study has exclusively used CRet to identify layer 1 LCN subtypes, it provides new and important information about the morphology and synaptic connectivity of SPF cellswhich contribute to the function of this lamina in the PFC of a primate species. Of significance, primate dorsolateral (dl) PFC is involved in the executive control of goal directed behaviors, whereas mPFC mediates emotional and autonomic functions (Vogt, 2009;Passingham and Wise, 2012). Moreover, evidence suggests that abberations in the maturation of PFC underlie specific neurodevelopmental disorders and psychiatric conditions in humans (Iafrati et al., 2014;Schubert et al., 2015).

Subjects and Experimental Methods
Post-mortem brain tissue for this study was obtained from Professor Kevan Martin and Dr John Anderson (Oxford/Zurich), Professor Alan Cowey (Oxford) and Professor Wolfram Schultz (Cambridge). The tissue came from eight normal young adult/adult monkeys (Macaca fasicularis; 5 male and 3 female; 4-12 years of age). Several of the cases had been used in experiments studying the monkey visual system and in experiments exploring reward mechanisms in the brain. All surgical and related procedures were conducted in accordance with the Society for Neuroscience "Policy on the use of animals in neuroscience research" and were licensed separately at the Universities of Oxford and Cambridge under the U.K. Animals (Scientific Procedures) Act, 1986. At the end of these experiments monkeys had been given lethal overdoses of anesthetic. When sufficiently anesthetized animals were transcardially perfused initially with 0.9% physiological saline and then with either 4% paraformaldehyde alone or with 3% paraformaldehyde and 0.5-2.0% glutaraldehyde (Taab, Reading, UK) in 0.1M phosphate buffer (pH 7.4) at room temperature.
Selected sets of serial coronal sections, together with tangential sections through the uppermost layers of the cortex, were rinsed in TRIS buffer containing 0.5% Triton X-100 for 1-2 h or were freeze-thawed. Some sections were treated in a microwave at 600 W for 60-90 s in TRIS buffer pH 6.0. Sections were subsequently washed in 20% normal goat serum [30-60 min diluted in TRIS buffer pH 7.4 at room temperature (RT)] and incubated in a primary polyclonal antiserum against calretinin [Code 7696; SWant raised in rabbit (Schwaller et al., 1993); dilution 1000-2500 with 0.01% NaN 3 ] for 2-3 days at 4 • C or overnight at RT.
Immunolabeling was visualized with standard immunoperoxidase procedures using a species matched Vectastain ABC kit (Vector laboratories, Peterborough) and developed with either: (i) 3,3 ′ -diaminobenzidine tetrahydrochloride (DAB) as chromogen [incubating sections in TRIS (pH 7.4) with 0.05% DAB (Sigma) and 0.01% H 2 O 2 for 3-10 min at RT], or (ii) using the SG (slate gray) Vectastain peroxidase substrate kit (SK-4700; Vector laboratories, Peterborough) by incubating sections for 3-8 min at RT. Specific immunolabeling was absent from sections incubated without primary anti-serum or peroxidase linked antibody. (Importantly, the dark yellow lipofuscin-rich profiles referred to below were seen in the untreated tissue sections prior to immunocytochemistry and exposure to osmium tetroxide -see Figures 4I-K, 13E,G,I, 14B,D). Sections for light microscopy alone were mounted in series on glass slides, air-dried, dehydrated in alcohol, passed through xylene, and embedded in DePeX mountant then coverslipped.
Archived Nissl stained sections were used to define the areal and laminar cytoarchitecture of frontal cortices (Figures 1A-C; Carmichael and Price, 1994;Saleem et al., 2013).

Light-Microscopy
Sections were initially examined in a light microscope and structures of interest recorded with through-focus photomicrographs, digital images, and drawings.

Dendritic Morphometry
Eight distinct classes of CRet+ neuron in layer 1 were identified qualatively on the basis of dendritic architecture (Figures 2-4). Quantitative analyses were then used to provide morphometric parameters indexing each class and to test whether SPF cells represented a distinct type of CRet+ cell. Accordingly, the dendritic arbors of 12 well-labeled neurons per class were reconstructed in three dimensions (3D) using a Neurolucida/Neuroexplorer (MicroBrightField, USA) computer system attached to a Leitz light microscope. Arbors were assessed quantitatively using the following procedures: (i) construction of dendrograms (Uylings and Van Pelt, 2002); (ii) Sholl analyses; (iii) distributions of dendritic branch points with respect to the maximum (100%) arbor length; and (iv) a "Weighted-Segment/Tip" (WeST) analysis. This index reflects, to a first approximation, the overall structural topology of a LCN dendritic arbor -composed of one or more binary dendritic trees. Each segment in a dendritic tree was given the value of its centrifugal order (i.e., primary dendrite (root segment) = 1, secondary dendrite = 2, tertiary dendrite = 3, etc). The total weighted values for all segments in a single dendritic tree were calculated then divided by the number of terminal dendritic tips in the tree. A mean WeST value was calculated for all the dendritic trees from one neuron and averaged across cells (n = 10) from the same class. Significant interclass differences were assessed using ANOVAs followed by multiple post-hoc Bonferroni t-tests (significance p < 0.05).

3-D Reconstruction/Rotation of SPF Cells
The dendritic and axonal arbors of 15 SPF cells were reconstructed in 3D. The locations of varicosities (putative synaptic boutons) along SPF cell axons were recorded in the 3D data sets -likewise the positions of CRet+ punctae in close association with identified SPF cell dendrites (Figures 5E,G ′ ). Neurons were subsequently rotated and viewed either in the perpendicular plane (90 • ) or in surface view (0 • ) with respect to the pia (see lower left inset in Figure 7).
The tangential boundary distributions of the SPF cell dendritic arbors were assessed by calculating their major caliper axis. Arbors were then superimposed along the major axis with somatic centroids as root points. Arbor boundaries were transformed so that their maximal extents along the major axis were the same (100%) -providing an overall comparison of SPF cell dendritic arbor polarization ( Figure 3J).

Spine Densities of SPF Cell Dendrites
Linear spine densities (number of spines per micron) along spine bearing primary dendritic segments (lengths >50 µm) selected from at least five CRet+ cells per class were calculated as described previously (Gabbott et al., 1995).
Areal Density of Axonal Varicosities -"Local" vs. "Distant" Zones Preliminary analysis of SPF axons (n = 8; from areas 24b, 32,46,9) in the tangential plane indicated a variation in the clustering of identified axonal swellings with respect to distance from parent somata (Figure 8). To assess this quantitatively, individual axonal arbors were divided into two separate concentric regions -"local" (within the vicinity of the soma) and "distant" zones ( Figure 8). A general guideline was established that the division between the two zones occurred between 150 and 200 µm radius from the parent soma (see Figures 8D,E). This guideline was adjusted to accommodate individual axonal arbors (Figures 8A,C).
Axonal swelling density per unit area of cortical surface was then calculated (c.f. linear density per unit axon length) for each zone within an individual arbor (Figure 8). Thirty counting quadrats (optimized at 50 × 50 µm) were positioned randomly by computer within each local and distant zone of the arbor. A "test" quadrat was defined as lying entirely within the respective FIGURE 2 | Calretinin immunopositive (CRet+) neurons in layer 1 of monkey PFC -Classes 1-5. (A) Class 1. Area 32. Surface view of a superficial horizontal fusiform CRet+ neuron (n) lying immediately below the pial surface. Note the widespread distribution of labeled processes which ramified throughout the depth of layer 1 (asterisk). A deposit of cellular debris (non-immunolabeled) is indicated (small arrow). Montaged photograph. Scale bar: 25 µm. (B) Reconstruction of the CRet+ neuron seen in (A). Thick dendrites emerge from opposite ends of the highly fusiform soma. Numerous thin varicose processes arise from dendrites and ramify radially to form a swirling varicose plexus in the vicinity of the soma and its main dendrites. Two labeled processes that course distally before being truncated (small arrows). This neuron is immunomorphologically identified as a Cajal-Retzius cell. Scale bar: 50 µm.  zone and containing an axonal segment from the identified SPF cell with at least one varicosity. The total number of identified swellings in each test quadrat were then counted (with upper/left inclusion borders and lower/right exclusion borders; Gabbott and Somogyi, 1986). Mean "local" and "distant" axonal varicosity areal densities were calculated per arbor and then for the total sample of the eight arbors. Overall mean "local" and "distant" data were compared statistically using a two-tailed Student t-test (significance p < 0.05).
Finally, an optimized lattice (divided into 20 × 20 µm squares) was used to visualize the tangential density distribution of axonal swellings from individual SPF cells. The relative frequency of varicosities in each square was derived and contour plots created using Surfer 7 mapping software (Golden Software, Colorado. USA).

Ultrastructural Analysis
Selected well-labeled SPF neurons (n = 12) were examined ultrastructurally using correlated light and electron microscopy (Gabbott et al., 2012). In brief, long series of serial ultrathin sections were cut through identified neural structures seen in the light microscope, collected on Formvar-coated single slot grids, stained with Reynold's lead citrate and subsequently examined in an electron microscope (Gabbott et al., 2012).

Synaptic Input to Identified SPF Cells
The somata, dendrites, thorns, filamentous processes, and spines of SPF cells were traced through serial sections to identify asymmetric (A-type) and symmetric (S-type) synaptic inputs (Peters et al., 1991). The CRet+ and CRet-nature of presynaptic boutons was noted. Specific structures were reformed from ultrathin section profiles using Reconstruct (Fiala, 2005).

Synaptic Output of Identified SPF Cells
Post-synaptic target structures were categorized as either dendritic spine (sp) or shaft (sh), cell body (cb), axon initial segment (ais), or unclassifiable due to lack of structural criteria in the post-synaptic element (?). A selected sample of local (n = 74) and distant (n = 79) boutons (with clearly identified synaptic junctions) were analyzed from the axonal arbors of five representative SPF cells (areas 9, 24b, 25, 32, and 46d) in four animals. Possible differences in the distribution of the cellular compartments innervated by the "local" and "distant" axonal arbors were assessed. In addition, each compartment category was subdivided into CRet+ or CRet-post-synaptic targets.
Electron micrographs of identified structures were produced photographically, scanned and digital image files created.

Combined Pre-Embedding CRet and Post-Embedding GABA Immunocytochemistry
Coronal and parasagittal tissue sections (from two animals fixed with high glutaraldehyde content) were reacted immunocytochemically for CRet using the pre-embedding immunocytochemical method described above -with Vector SG as peroxidase substrate. Selected sections were then embedded in resin. Serial semithin 0.5-2 µm thick resin sections were cut through the processes and somata of identified CRet+ SPF cells (from areas 32, 24, and 46v) and subsequently reacted for GABA using post-embedding immunocytochemistry -with DAB as peroxidase substrate (Gabbott and Bacon, 1996a). The GABA immunoreactive nature of identified CRet+ SPF cells (n = 7) was then assessed by the detection of intense darkbrown GABA immunolabeling within their SG immunolabeled somatic/nuclear profiles. [Note: SG reaction product in CRet+ immunolabeled neurons does not prevent subsequent detection of GABA immunolabeling in the same material visualized with DAB]. The material was also used to assess the presence of GABA immunoreactivity in 100 CRet+ neurons throughout layer 1 in area 32 -thereby allowing the ratio of CRet+/GABAto CRet+/GABA+ neurons to be derived. The depth distributions of GABA immunonegative (GABA-) and GABA immunopositive (GABA+) nuclear profiles through layer 1 and into the upper part of layer 2 ( Figure 1F) were also   One pole of the ellipsoid soma (arrow) gives rise to a thick branching dendrite that recurves and courses tangentially directly beneath the pial surface. Thorn-like processes arising proximally from the main dendrite are indicated (small white arrows). Region outlined is enlarged in inset which shows a CRet+ puncta abutting a dendritic thorn (d, dendrite). A dark vacuolated cellular profile is shown enlarged in (E ′ ) Capillary, c. Scale bar: 100 µm. Inset shows (F) Reconstruction of the neuron in (E) (arrow). Fine thorn-like processes (small arrows) emanate from the thick main process (see inset). A small-caliber terminal process gives rise to several varicosities (one of which is encircled). Scale bar: 100 µm (G) Photomicrograph of the SPF soma -note lack of filaments arising from soma (c.f. Figures 2C,F). A thin axon-like processes (ax) emerges from one pole of the soma. Swelling along an immunopositive process (black circle). CRet+ punctae ( determined. The number of GABA+ and GABA-nuclear profiles in a sampling strip (250 µm wide), divided vertically into twenty 10 µm deep sampling tiers, were counted. (The left and bottom borders of each tier represented exclusion boundaries - Gabbott and Somogyi, 1986). A sufficient number of randomly positioned sampling strips (n = 10 in each animal; right hemisphere) were used to compile depth distribution histograms for equal numbers (n = 200) of GABA+ and GABA-nuclear profiles (Figures 1D-F).

Illustrations
All image files were finally imported into Adobe Photoshop © CS6, adjustments made for contrast and brightness, photographic montages then assembled and composite illustrations prepared.

CRet+ Immunoreactivity Throughout Prefrontal Cortex
Specific CRet+ immunolabeling was present throughout layers 1-6 of the PFC areas studied and in the underlying white matter. The morphologies and quantitative laminar distributions of calretinin immunopositive (CRet+) neurons in macaque PFC have been previously described (Condé et al., 1994;Gabbott and Bacon, 1996a,b;Gabbott et al., 1997). The peak distributions of CRet+ neurons in primate PFC was in layer 2 with a small fraction (<1%) of the total CRet+ population present in layer 1 ( Figure 1G -area 32; Condé et al., 1994;Gabbott and Bacon, 1996b).

Classification of Layer 1 CRet+ Cells
Several classes of CRet+ neurons were immunolabeled throughout the depth of layer 1 and were present across all PFC areas (Figures 2, 3A-F, 4D,G,H). The classification starts with bi-and multipolar cells and proceeds to unipolar cells. The term "superficial" refers to neurons in upper layer 1 while "deep" applies to cells in mid/lower layer 1. Eight main classes were identified. These consisted of: (1) Superficial horizontal fusiform cells in upper layer 1 with large-sized highly ovoid somata (diameters of area equivalent circles: d.circ = 20-35 µm; max/min caliper diameter ratio = 4.6 ± 0.4; mean ± SEM, n = 18). Somata commonly gave rise to two (infrequently three) thick primary dendrites from opposite poles. These dendrites coursed parallel with the pia and issued long horizontal axon-like processes with numerous swellings (Figures 2A-F). Axon-like fibers also emerged from somata ( Figures 2C,F). Classical axons arose from the soma and/or branch points along primary dendrites -they ramified extensively in the vicinity of the parent somata. Distal axonal fibers were truncated ( Figure 2B); (2) Superficial horizontal multipolar cells with medium/largesized somata (d.circ = 15-30 µm) located in upper layer 1 (Figures 2H,I). The primary dendritic processes (3-6 in number) lay parallel to the pia. They formed interwoven trees that were highly elongated (c.500+ µm length; c.200+ µm width; Figures 2H,I). The primary segments had frequent expansions along their lengths ( Figure 2I). Numerous fine-caliber processes arose from dendritic stems and extended toward the pia giving rise to terminal bulbs abutting the pial surface ( Figure 2J). One or more axon-like processes commonly emerged from cell bodies and/or from primary dendritic shafts -ramifying extensively both locally and distally throughout layer 1. (3) Standard multipolar neurons situated predominantly in the lower two-thirds of layer 1 (c. 50-150 µm below pia; Figure 1C) with fine axons that ramified locally in lower layer 1/upper layer 2 (Figures 2K-N). The dendrites had simple bifurcating patterns and could have low numbers of dendritic spines (0.47 ± 0.12 spines/micron; mean ± SEM, n = 5); (4) Midget neurogliaform cells located in upper layer 1 ( Figure 2O). A small-sized multipolar neuron (d.circ = 5-10 µm) with fine-caliber radiate dendrites which branched frequently in the vicinity of the soma -numerous swellings were present along their lengths. The axons of these cells were not seen; (5) Tufted cells. Neurons with highly elongated somata (max/min caliper diameter ratio = 5.4 ± 0.6; mean ± SEM, n = 15: Figures 2P,Q) with horizontal tufts of long dendrites from opposite somatic poles. Some somata issued fine hair-like filaments, as shown in Figure 2P;  FIGURE 7 | Continued surface as: P, perpendicular; Ob, oblique; S, surface]. All cells had a single fine axon-like processes (ax) emerging either from the opposite somatic pole (cells a, c, d, f, g, h, i), from the side of the soma (cell b), base of the main dendrite (cell e) or proximally along the primary process (cell j). Axonal varicosities are indicated (small arrows). Note the long thorns of cells a, g and j and the terminal interwoven dendritic processes of cell g (c.f. cells in Figure 8). Some cells had lipofuscin-rich regions associated with their dendrites (shown encircled). Areal locations of cells: a, area 46d; b, area 25; c, area 32; d, area 24a; e, area 46d; f, area 32; g, area 24c; h, area 32; i, area 46v; j, area 9; k, area 24c. Scale bar: 50 µm.
(6) Deep horizontal bipolar neurons in mid/lower depths of layer 1 (Figures 3A-C). These cells had linear aspiny dendrites several hundreds of microns in length oriented horizontally -the axons arose from the cell body or proximal dendrite (Figures 3A-C, 4H); (7) Simple unipolar neurons aligned horizontally in layer 1.
Most common in the lower two-thirds of layer 1 (c.50-150 µm below pia; Figure 1C). These cells possessed a single small/medium-caliber dendrite and a fine axonal process ( Figures 3D, 4G, 5I). The somata of these neurons were smooth and dendrites could sometimes issue appendages ( Figure 5I). (The unidendritic nature of these cells was not due to sectioning artifacts); and, (8) Subpial fan (SPF) cells are horizontal neurons located immediately beneath the pia ( Figure 1C). They possessed unipolar somata from which arose a single stout primary dendrite that branched repeatedly into an arbor that fanned out across a plane parallel with the pial surface (Figures 3E,F). The morphology and synaptology of SPF cells are described in detail below.

Comparative Dendritic Morphometry Identifies SPF Cells as a Unique Class
The quantitative results indicated significant differences in the dendritic architecture of class 1-8 CRet+ neurons in layer 1 (Figures 3G-I). A salient feature of cells in classes 2-5 were short lengths of lower order (< 4 • ) dendritic segments (Figure 3G), with the vast majority of the corresponding branch points being located proximal to parent somata ( Figure 3H). Also evident were the long total lengths of the complex dendritic arbors of class 2 neurons and the relative simplicity of class 7 unipolar cells ( Figure 3G).
With regard to class 8 SPF cells, the dendrogram plots illustrate the frequent presence of higher dendritic orders (>7 • ) -a characteristic not found in other classes. Indeed, the mean total length of SPF cell higher order dendrites (>150 µm from soma) was significantly greater than all other cell types, notably class 1/2 neurons [ Figure 3G; class 8 > class 2: +24.3% ± 1.82 (mean ± SEM, n = 8); t-test p < 0.01].
In sum, these results highlight SPF cells as a specific morphometric class of CRet+ neuron in layer 1 of adult monkey mPFC. Further, as indicated in the orientation plot shown in Figure 3J, the dendritic arbors of SPF cells were markedly polarized about their somata.

Class 8: "Subpial Fan (SPF) Cells"
Light Microscopical Features of Identified SPF Cells Areal/sublaminar location and frequency SPF cells occurred across all PFC areas being more common in the depths of the cingulate and principal sulci ( Figure 1B). They were also present in other areas beyond PFCfor example, temporopolar (area 38), anterior insular, and orbitofrontal cortices ( Figure 1B). Their number was highly variable across individuals, being more numerous in younger cases (see below). Cells lay immediately beneath the pial surface or in upper layer 1 (Figures 3E,F, 4, 5E, 6A-E, 7,  9A,A ′ , 13A,H, 14F).
The somata of SPF cells were situated millimeters apart (c.1-10 mm) and could be found clustered with other CRet+ neurons in upper layer 1 (Figure 5I). They occurred with a linear density of approximately 2-5 per 10 mm of pia (estimate derived using 200 µm thick sections).

Shape and size of somata
SPF cells possessed smooth somata that were ovoid, pyriform or circular in outline (max/min caliper diameter ratio = 1.2 ± 0.4; mean ± SEM, n = 32: see Figures 3-9). Somatic sizes ranged from 15.3 to 30.8 µm (d.circ 21.4 ± 2.2 µm; mean ± SEM, n = 22). The other pole gives rise to a proximally bifurcating thick primary dendrite with numerous dendritic "thorns" (shaded region and inset) and terminates in a fan-like plexus of fine processes (arrow). A proximal dendritic branch bifurcates into several fine varicose axonal fibers (ax dendritic) that ramify extensively -both locally and at a distance from the soma. Boundary

Dendritic architecture
The characteristic feature of these unipolar neurons was a single horizontal primary dendrite that usually gave rise to numerous repeatedly dividing daughter branches with increasingly shorter segment lengths -giving the whole dendritic arbor the appearance of a "corymb-like" fan ( Figures 3G, 5A, 7). The main dendrite of some cells extended for over 200 µm before branching into a radiate spray of higher order segments (Figures 4F, 6E, 7 -cell g, 8D).
In addition, lower order dendrites could give rise to numerous fine caliber filamentous strands of varying lengths (Figures 6B,F,G, 7 -cells f,i, 8A). These willowy strands were much thinner and usually longer than the thorns described above (Figures 6F,G). Proximal dendrites also gave rise to low numbers of thin-and mushroom-type dendritic spines (0.24 ± 0.03 spines per µm; mean ± S.D.; Figure 6H). Distal dendrites could narrow gradually into varicose "axon-like" processes ( Figures 5E,F, 7 -cell J, 8D).
3D reconstructions/rotations confirmed that the dendritic (and axonal) processes of SPF cells were oriented parallel with the pial surface and lay flat directly beneath the pia (Figures 4M,  5C,H,I, 7, 8, 9A).
The immunolabeled dendritic and axonal arbors of neighboring SPF cells did not overlap.

SPF cells in close association with capillaries
A conspicuous feature of SPF somata and dendrites was their frequent close association with blood vessels penetrating layer 1 from the pial surface (Figures 5D,E, 6A-C, 9A, 11F, 13F,  14F-H).

CRet+ punctae associated with SPF cells
CRet+ punctae (presumed pre-synaptic boutons) of unknown origin were found in close contact with the dendritic arbors of SPF cells (Figures 5E,H,G,G ′ ,I). Three SPF cells had a single CRet+ axonal swelling closely opposed to their somata (see below - Figures 11F,G).

Ultrastructure of Identified SPF Cells Somata
Cell bodies possessed a single nucleolus within a heavily infolded large nucleus that occupied a large proportion of the soma ( Figure 11F). The cytoplasm contained a typical array of subcellular organelles (Figures 9D,E; see Peters et al., 1991). A high number of axosomatic inputs were present comprised of both asymmetric (A-type; 21% occurence) and symmetric (Stype; 79%) junctions (Figures 9D,E). S-type input came from CRet-presynaptic boutons whereas A-type input was derived from mainly CRet-and occasional CRet+ boutons (Figures 9D,E,  11F,G).

Dendritic shafts
Labeled dendritic shafts contained microtubules and mitochondrial profiles (Figure 11K). Unlabeled A-type and S-type axodendritic synapses were present -however, their distribution was uneven: A-type junctions were more frequent (1.3x) distally whereas S-type junctions were more common (1.5x) proximally, especially over primary dendritic shafts. S-type junctions were formed by both CRet+ and CRet-boutons formed S-type junctions.

Spines
The profiles of dendritic spines varied in shape and size (stem lengths/head diameters) -ranging from stubby protrusions, to "lollipop" and long thin drumstick profiles up to 3 µm in length. All spines investigated (n = 21) received CRet-A-type synaptic input with 71% (n = 16) found to receive additional CRet-S-type input (Figures 9F,G).

Filamentous processes
These processes were typically thin (<3 µm) and extended for considerable distances into the surrounding neuropil ( Figure 9H). The profiles of long mitochondria were central components of these structures ( Figure 9I).

Thorns
Dendritic thorns originated as prominences from the profiles of dendritic shafts (Figures 10A,B). The base of thorns contained mitochondria and clusters of ribosomes (Figures 10B,E). These organelles continued into the stems of thorns assuming positions alongside streams of longitudinally running filaments. Mitochondria were present throughout their length ( Figure 10D). Thorns received both A-type (65% occurrence) and S-type (35%) synaptic contacts (Figures 10A,C,D). A-type input was from CRet-boutons; and S-type input from both CRet+ and CRet-boutons (Figures 10B,C). CRet-A-type junctions were commonly found either adjacent or opposite the entrance to thorns (Figures 10B,E).

Axonal connectivity
SPF cell axonal arbors were unmyelinated and divided into local and distant segments -with the areal density of axonal varicosities differing significantly between zones (see above). Following correlated light and electron microscopical analyses, identified axonal varicosities along these axons were confirmed as CRet+ synaptic boutons establishing exclusively A-type junctions with CRet-post-synaptic targets (Figures 9B,C, 11A-H). These boutons contained predominantly round synaptic vesicles (Figures 9G, 10C, 11A-H). CRet+ boutons of unknown origin were found throughout layer 1 neuropil ( Figure 5G) establishing S-type synaptic contacts with unidentified targets (not illustrated). Table 1 provides a quantitative analysis of the cellular compartments post-synaptic to CRet+ boutons from identified SPF cells -data are presented separately for "local" and "distant" segments. The post-synaptic targets were mainly dendritic shafts (sh) and spines (sp) -with the sh/sp ratio 2.1 locally and 3.7 distally. Analyses of serial sections indicated that a maximum of 3 axonal boutons were found to innervate a single post-synaptic target and that most (>85%) of the axodendritic synapses innervated sparsely-spiny thin-caliber processes ( Table 1). Of note, the three axosomatic contacts were onto different somata which received both A-and S-type input.
A summary map of the synaptic connectivity of SPF cells is presented in Figure 12.

Cellular Debris and Degenerating SPF Cells in Layer 1
In the light microscope, irregular profiles (d.circ. c.5-17 µm) with a dark yellow/orange color (indicative of lipofuscin) were frequently found in the upper tier of layer 1, particularly beneath the pia (Figures 2-8, 13). Such profiles were present in all PFC areas studied and occurred, either singly or in clusters, with a linear density of 1-6 per mm length of pia (Figures 4A,B,D,J,K). They were commonly associated   Table 1). Data for axonal boutons lying "local" and "distant" to parent somata are given separately. Post-synaptic compartments: dendritic spine, sp; dendritic shaft, sh; cell body, cb; axon initial segment, ais; and unidentified elements?
with the processes and somata of SPF cells -as well class 1/2 neurons (Figures 2A,C,C ′ , 2H-J, 4A,B,I-K, 6A-F, 7, 8B, 13A,C,E, especially G,I,J ′′ ). Some of the profiles were clearly vacuolated with closely apposed microglial cells (Figures 5E ′ , 14D). The coloration of these lipofuscinrich profiles darkened following treatment with osmium tetroxide. Figures 13G,I, 14B show the distinction between gray immunolabeling for CRet alone (using Vector SG kit) and the lipofuscin containing profiles. Accordingly, it was possible to identify a large number of CRet+ SG labeled cellular profiles in upper layer 1 with lipofuscin in their cytoplasmata ( Figure 4I).
Interspersed among healthy CRet+ neurons in upper layer 1 were CRet+ cells with somatodendritic compartments at various stages of cytoplasmic vacuolation. Such neurons were most frequently located directly underneath the pia and represented less than 1% of the healthy CRet+ cell population in layer 1. These degenerating neurons were spaced apart with a linear density of 2-10 per 10 mm of pia and were composed of putative Cajal-Retzius cells ( Figure 2G) and identified SPF  Figure 6, cell g in Figure 7, and the cells in Figures 8A,C). Positions of dark lipofuscin profiles (thick arrows) indicated (thick arrows -see also photographic insert) Cell nuclei (n). cells (Figures 13A-E,H,H ′ ,J). Noteworthy were subpial CRet+ somatic profiles with unlabeled regions of cytoplasm as illustrated in Figures 13F,F ′ . Identified SPF cells undergoing vacuolation were present in all areas examined (Figure 13). The degree of cellular vacuolation varied considerably between SPF cells within an area. Some cells had vacuoles present only in higher order dendrites, others throughout their dendritic arbors, whilst additional cells were almost entirely vacuolated (Figure 13). Where present, dendritic thorns were much swollen ( Figure 13E). Heavily vacuolated somata either lacked an axon or possessed the remnants of a dilated initial segment ( Figures 13E -cells b,d, G,J). This variation represented a continuum -with approximately five times the number of SPF cells undergoing degeneration in adult cases than those with clear structural integrity (see below).
Ultrastructural analysis of identified vacuolated SPF neurons confirmed that their somata and dendrites were undergoing various stages of structural fragmentation characteristic of cellular degeneration (Figures 14A,E). These somata would contain increased numbers of much enlarged lysosomes, bulky lipofuscin granules associated with regions of vacuolated cytoplasm, dense membranous inclusions and pyknotic nuclei (Figures 14A,E). In the electron microscope, the dark yellow irregular lipofuscin-rich profiles were found to be regions of neuropil with large amounts of condensed cellular debris undergoing advanced stages of degeneration -sometimes containing highly condensed pyknotic nuclei ( Figure 14C). Degenerating SPF cells and the clumps of cellular debris in upper layer 1 were surrounded by neuropil with normal structural integrity -for example, containing undisrupted synaptic junctions ( Figure 14E).

Comparison between Young Adult and Adult Cases
A preliminary analysis indicated that SPF cells (with normal structural integrity) were more common (approximately x1.5) in younger animals (4-8 years) than older animals (9-12 years). Older adults had more pyknotic neurons (including degenerating SPF cells: Figure 13) and cellular debris in upper layer 1 across all areas.

Combined Pre-Embedding CRet and Post-Embedding GABA Immunocytochemistry
Using this procedure the nuclear profiles of all seven SPF cells tested were found to be devoid of specific GABA immunoreactivity -similar to pyramidal shaped profiles in layers 2-6 ( Figures 14F-I). After GABA immunolabeling, some neighboring CRet-somatic/nuclear profiles were found to be strongly GABA+; for example, ovoid CRet-/GABA+ cell profiles oriented tangentially in upper layer 1 (Figure 14H).

DISCUSSION
This neurocytological study illustrates the morphological variety of calretinin immunopositive (CRet+) cells in layer 1 of the young adult and adult macaque monkey PFC. The diverse morphologies of these neurons have not been fully classified previouslyalthough the studies of Condé et al. (1994) and Gabbott and Bacon (1996a) document several of the CRet+ subtypes present in the current classification. The survey provided here served as a necessary comparative background for a comprehensive account of SPF cells -the main focus of the study.

Technical Considerations
The investigation used perfusion-fixed adult monkey material with thick sections (upto 250 µm) being prepared in the major sectioning orientations -especially in the plane tangential to the pial surface. Thick sections gave the possibility of unequivocally tracing the intricate dendritic arbors of SPF cells from their unipolar parent somata. These factors coupled with excellent penetration of the CRet antiserum (and other immunoreagents) into tissue immediately subjacent to the pia gave detail-rich immunomorphology in both the light and electron microscopes.
Whilst the dendrites of CRet+ SPF cells displayed strong immunoreactivity their axonal arbors showed marked variation in labeling -ranging from short initial segments ( Figure 5G) to darkly labeled expansive arbors with numerous varicosities (Figures 8A,B). The gradation may reflect variations in CRet concentration -sometimes below the level of immunocytochemical detection. Therefore, it is unlikely that the full extents of the immunolabeled SPF cell axonal arbors were consistently revealed.

CRet+ Neuron Classes in Layer 1
On the basis of light microscopical and morphometric features eight classes of CRet+ neurons in layer 1 of adult monkey PFC were identified. As discussed below, classes 1 and 2 are probably "classical" Cajal-Retzius cells -with the possibility that some superficial unipolar class 7 neurons also belong to  this group. Class 8 SPF cells may share a common lineage with these neurons. Moreover, evidence (from across mammalian species) indicates that neurons in classes 1, 2, 7, and 8 are probably excitatory in function (Del Río et al., 1995; this study), whereas most neurons in classes 3-6 are likely GABA-containing inhibitory LCNs (Gabbott and Bacon, 1996a,b and unpublished observations;Condé et al., 1994;Varga et al., 2015). Furthermore, class 4 cells are similar to the Golgi-impregnated neurons in human layer 1 described as neurogliaform cells by Retzius (1893; see cell 8 in Figure 14J of this study) and as dwarf stellate cells by Ramón y Cajal (1899b).

Class 8 SPF Cells in Adult Monkey PFC
SPF cells represent a morphologically distinct class of CRet+ neuron in layer 1 of adult monkey PFC -with similar neurons present in other areas outside PFC. As far as the author is aware, such cells have not been previously recognized. Golgi-impregnation investigations and other immunocytological studies, using antisera against calretinin, as well as the calcium binding proteins calbindin and parvalbumin, charting the preand postnatal cellular development of monkey cortex do not specifically identify neurons with the morphology of SPF cells (Huntley and Jones, 1990;Yan et al., 1995a,b;Gabbott and Bacon, 1996a;DeFelipe, 1997;Del Rio and DeFelipe, 1997;Meskenaite, 1997;Ding et al., 2000). An exception is the study of Condé et al. which illustrates a CRet+ cell in layer 1 of adult monkey PFC with the structural rudiments of a SPF cell and remark on the "long spine-like processes" arsing from dendrites -no additional comments are provided (Condé et al., 1994 - Figure 5D). The reasons for the previous lack of identification are unclear. They may relate to the extreme outer position of SPF cells, their comparative rarity, histochemical staining/immunolabeling procedures, species differences, section thickness and orientation effects. The last two factors would limit the extent of defining morphological features present in a single section -e.g., Condé et al. (1994) used 40 µm cryostat sections. This could lead to SPF cells being considered as simple unipolar neurons (see Figure 5I). Alternatively, if only segments of their dendritic fans are present in thin sections, such fragments could be attributed to the elaborate dendritic/axonal arbors possessed by some Cajal-Retzius cells. This highlights the continuing debate about which layer 1 cells qualify for membership of the special Cajal-Retzius family across development and into adulthood (Meyer et al., 1999;Zečević and Rakic, 2001;Martinez-Cerdeno and Noctor, 2014;Marín-Padilla, 2015: see especially DeFelipe et al., 2013. Interestingly, Anstötz et al. (2014) and Ma et al. (2014) have studied the electrophysiological properties, neurochemical content and connectivity of Cajal-Retzius cells in the developing cortex of mice. They illustrate the dendritic and axonal arbors of typical and atypical unipolar Cajal-Retzius cells which bear some resemblance to the SPF cells described in adult monkey. However, the somatodendritic compartments of the cells in postnatal mouse reside mostly in deep layer 1 with their extensive axonal arbors ramifying throughout the lamina -unlike SPF cells in monkey.
Are SPF cells present in human PFC? The study of Gabbott et al. (1997) provided evidence of CRet+ Cajal-Retzius cells in layer 1 of aged human mPFC (>60 years). However, CRet+ neurons with the morphology of SPF cells in adult monkey were not observed in the human material examined. The comparatively much older ages of the human cases studied (Amlien et al., 2016), together with post-mortem interval (<18 h) and immersion fixation effects may underlie the absence of SPF cells in human mPFC.
In Golgi-impregnation studies, Ramón y Cajal (1890) describes "very singular cells" in cortical layer 1 of neonatal rodents, rabbits and cats. Retzius (1893Retzius ( , 1894 described more elaborate counterparts (sometimes exclusive variants) of the "Cajal"sche Zellen' in human primate fetal cortex ( Figure 14J) -with Ramón y Cajal (1909) subsequently confirming their presence at postnatal ages. The cells of Retzius and of Ramón y Cajal fall into an expansive polymorphic family; both at different developmental stages and across species. Ramón y Cajal writes: " [They] show many variations in the shape, number, and direction of the processes" and "... there does not exist any rigorous frontier" (DeFelipe and Jones, 1988). Despite this, two of the main varieties are firmly represented in the present CRet immunocytochemical study of adult monkey PFC (Marín-Padilla, 1984;DeFelipe and Jones, 1988, pp 201-210).
Ramón y Cajal (1890) defines fusiform cells in the upper part of the molecular layer in rabbit cortex: "The dominant form is that of a perfect spindle, with two long dendrites directed parallel to the fibrillary layer ... the rectilinear dendrites give off, almost at a right angle, small ascending branches... The axon is at least double ... commonly [emerging] at the point at which the polar dendrites turn to ascend. The axons run in opposite directions ... giving off at intervals fine, ascending, branched filaments that come to complicate notably, the plexus of the superficial [sublayer]" (DeFelipe and Jones, 1988, pp11-12). An example of this cell type in humans drawn by Retzius (1893) is presented here in Figure 14J -cell 7 (c.f. Figures 2A-F this study).
Horizontal cells in human fetal brain are described by Ramón y Cajal as appearing "... throughout the thickness of the molecular layer in the form of fusiform, triangular, or stellate cells with very long horizontal processes from which, at right angles, emerge an infinity of small ascending branches; [these] end under the pia in a thick varicosity. The more superficially placed cells (marginal cells) are conical, have an external base adherent to the pia and a central pedicle from which emerge very numerous and enormously long processes ... it is impossible to distinguish among the various elongated processes any with the characteristics of axons or the [typical] properties of dendrites" (DeFelipe and Jones, 1988, p153 special cells of the first layer" and "Cells of the plexiform layer with short axon" in new born human motor cortex are provided by DeFelipe and Jones (1988, pp201-210).
The CRet+ cells in classes 1 and 2 conform to the above descriptions of fusiform and horizontal cells, respectively -with the versions in adult monkey being significantly more complex than those in lower mammals.
In addition, vertically aligned marginal/pyriform Cajal-Retzius cells are also present during corticogenesis in humans (Ramón y Cajal, 1909 -Figure 341, cell A;Retzius, 1893 -Figures 1/4 in Plate III;Meyer and González-Hernández, 1993 -Figures 5, 6) -however, similar cells were not seen in the nonhuman primate material investigated here. Further, the studies of Retzius and Ramón y Cajal identify unipolar cells in upper layer 1 of human cortex [Retzius (1893), reproduced in Figure 14J cells 1, 4 and 5; Ramón y Cajal (1909) Chap. XXIV Figure 338, cell E]. Ramón y Cajal (1909) also describes "unipolar fusiform cells" in lower layer 1 of prenatal and newborn mammals (rat, cat, dog). These unipolar neurons have dendrites which branch infrequently and taper gradually along their course -some processes giving out numerous side-shoots with bulbous endings (termed "thick nodules" by Retzius) abutting the pia -similar to Figure 2J in this study. Of note, Ramón y Cajal (1899b) identifies a "rudimentary cell with a short axon" in layer 1 of the precentral gyrus of a 1 month old child -not dissimilar to the initial portion of cell j in Figure 7 (this study). However, Ramón y Cajal's illustration shows a unipolar neuron with a simple horizontal dendrite from which emerges a short axon (Ramón y Cajal, 1899b - Figure 93, cell E).
Consequently, the superficial horizontal SPF cells identified immunomorphologically in this study of adult monkey, with axons and characteristically branching fan-like dendritic arbors disposed tangentially in upper layer 1, are qualitatively and quantitatively distinct from the Golgi-impregnated fusiform, horizontal, and vertical marginal cells described by Ramón y Cajal and Retzius (and others) in a variety of species including primates (Marín-Padilla, 1984, 2015Meyer and González-Hernández, 1993). However, the possibility exists that some of the superficial CRet+ unipolar (Class 7) cells seen here (Figures 2N, 5I) and the unipolar cells illustrated by Retzius (1893; see Figure 14J -cell 5 in this study) and Ramón y Cajal are highly truncated versions of SPF cells.
The likely excitatory function of SPF cells (demonstrated by the lack of GABA immunoreactivity together with specific ultrastructural features such as spines and thorny dendritic branchlets, as well as significant S-type input proximal to the soma and A-type axonal synaptic junctions) aligns them with the special cells of Retzius and Ramón y Cajal and segregates them from GABAergic LCNs in layer 1. Accordingly, SPF cells may be "atypical" members of the persisting CRet+ Cajal-Retzius family -defining a way-point across a broad structural (and probable functional) spectrum in adult monkeys (Džaja et al., 2014;Gil et al., 2014).

Demise of SPF Cells
A limitation of this study is the lack of a full pre-/postnatal timeline for the observations reported here. Nonetheless, clear evidence is presented across all PFC areas studied of degenerating and highly pyknotic SPF cells (Figures 13, 14A,C,E) and neurons of the classical Cajal-Retzius type in young adult and adult monkey ( Figure 2G). Indeed degenerating SPF cells were more numerous in the older animals. Interestingly, Retzius (1893) illustrates vacuolated Cajal'sche Zellen in human fetal cortex likely undergoing the early stages of cellular degeneration similar to monkey SPF cells (Figure 14J vs. Figure 13E). Peters and colleagues (Peters et al., 2001;Peters and Sethares, 2002) studied age-related structural changes in neurons and glia throughout layer 1 in areas 46 and 17 of adult (5-12 years) and aged (25-35 years) monkeys. These authors describe the ultrastructural characteristics of rare pyknotic neurons in layer 1. Degenerating SPF cells have similar features (Figure 14) and may be represented in this population (see also Derer and Derer, 1990).
One striking structural component of upper layer 1 in adult monkey PFC are deposits of lipofuscin-rich cellular debris (Figures 4I-K) -some being closely opposed to the somata of SPF cells or enveloping their dendrites (Figures 5E, 6E,F, 7, 13). The stereological study of O'Kusky and Colonnier (1982) reports a highly significant loss of layer 1 neurons between birth and young adulthood (0-8 years) in area 17 of monkey. Given the late postnatal maturation of PFC, similar alterations would result in large numbers of degenerating neurons contributing to the widespread deposition of cellular debris in upper layer 1.

A Non-Synaptic Function of SPF Cells?
A non-synaptic function of SPF cells may relate to the close structural proximity of their somata and dendrites with blood vessels. If SPF cells contain reelin (similar to CRet+ Cajal-Retzius cells) they are strategically placed to secrete this protein across the blood brain barrier via endothelial cells into penetrating blood vessels (Perez-Costas et al., 2015). This would impact the postnatal maturation of neural circuitry within the perfusion domain of neighboring blood flow (Blinder et al., 2013;Chai et al., 2015). Figure 12A presents a summary map of synapses over SPF cell dendrites. Although the identities of these inputs are currently undefined, they can be predicted by considering the axonal systems known to ramify in layer 1. A-type inputs to SPF cells are likely derived from (i) other excitatory neurons in layer 1, (ii) the recurrent axonal collaterals of pyramidal cells situated in deeper layers, and (iii) the terminal axonal plexi of long-distance projections from pyramidal neurons in ipsilateral and contralateral cortical areas (Figure 15A; Martin, 1984;Vogt, 2009). In addition, layer 1 in PFC receives direct excitatory drive from specific and non-specific thalamic nuclei which also provide input to mid-layers of the cortex and layer 6 (Jones, 2007). Since no CRet+ boutons with A-type synapses were found to innervate SPF cell dendrites, intralaminar excitatory inputs from other SPF cells and CRet+ Cajal-Retzius cells is unlikely, at least in the adult.

Synaptic Connectivity of SPF Cells in Adult Monkeys
Inhibitory input to SPF cells may come from CRet+ and CRet-GABAergic neurons situated throughout the depth of layer 1 (see Figures 1D,F; also Muralidhar et al., 2014) and from inhibitory LCNs in deeper layers ( Figure 15A). For example, evidence in rat and cat when translated to monkey, suggests FIGURE 15 | (A) Schematic diagram showing possible sources of synaptic input (?) to SPF cells. (B) Diagram illustrating the organization of pyramidal cell bodies in agranular area 32 into radial minicolumns (spaced 30+µm apart in adult monkey PFC). The apical dendrites of pyramids in layers 3-6 form bundles as they ascend through the cortex. Layer 2 pyramids contribute to a separate bundling system. Each bundle gives rise to a spray of apical dendritic tufts -the main post-synaptic targets of SPF cells. (C) Internuncial role of SPF cells. Extrinsic and intrinsic axonal systems in layer 1 provide common input (a) to the apical dendritic tufts of pyramidal cells and SPF cells. This input may be differentially distributed across cell types. However, SPF cells and dendritic tufts may (?) receive additional input from specific sources (b/c respectively). SPF cells are strategically placed to integrate afferent input and provide a feed-forward excitation influencing neural activity in target apical dendritic tufts. that S-type axodendritic input to SPF cell dendrites might come from GABAergic double bouquet, bipolar and Martinotti cells in layers 2-6 -these neuron types have ascending axons innervating upper layer 1 (Marín-Padilla, 1971;Ramos-Moreno and Clascá, 2014). Of significance here, Martinotti cells may be either CRet+ or CRet- (Wang et al., 2004). Both CRet+ and CRet-S-type inputs were present over the dendrites of SPF cells with CRet+ S-type boutons located more proximally ( Figure 5H). These data suggest the strong proximal influence of GABAergic input on dendritic information processing in SPF cells.
The CRet-S-type synaptic input to SPF cells in layer 1 might also come from cholinergic basal forebrain neurons, and dopaminergic, serotoninergic, histaminergic, and noradrenergic brainstem nuclei (Figure 15A: Mrzliak et al., 1995;Latsari et al., 2002;Janusonis et al., 2004;Jin and Panula, 2005;Huo et al., 2009;Arrovo et al., 2014). Axons from these systems have varicosities that form conventional S-type synaptic contacts as well as numerous swellings without defined specialized membraneous junctions. Of interest are recent studies (Ramos-Moreno and Clascá, 2014;Chen and Kriegstein, 2015) demonstrating long range subcortical inputs to layer 1 originating from GABAergic projection neurons in the zona incerta of rodents having a role in the pre-and post-natal functioning of neurons in the molecular layer.
The dendritic arbors of SPF cells are embedded in the above fiber paths, terminal arborizations and local neural circuits (Figures 15A,C). As mentioned above the pial surface density of SPF cells is likely to be greatly reduced in the adult compared with earlier stages of cortical development -when the processes of functionally active SPF cells (and cells of the Cajal-Retzius family) could intermingle and overlap extensively.
The clustered pattern of boutons along the distal axonal segments of SPF cells (and their prominent innervation of dendritic shafts) may reflect a selective innervation of two apical dendritic systems derived from the known minicolumnar radial organization of the cortex (Figure 15B; Gabbott, 2003). Namely, the distribution mosaic of pyramidal cells in layer 2 (Gabbott et al., unpublished observations;Ichinohe et al., 2003) and the dendritic bundles formed by pyramidal cells in layers 3-6 ( Figure 15B, this study; Arikuni et al., 1983;Marín-Padilla, 1990;Schlaug et al., 1995;Gabbott, 2003).
Morphometric calculations indicate that about 5000 pyramidal cells distributed in layers 2-6 with apical dendritic tufts that ramify within the small patch of layer 1 innervated by an adult SPF cell axon (∼100,000 µm 2 ) have a probability (>0) of receiving a synaptic contact (Arikuni et al., 1983;Schlaug et al., 1995;Peters and Sethares, 1996). In the adult this innervation would be sparse. However, the immunocytochemical evidence presented is likely to have greatly underestimated both the spatial extent and bouton density of SPF cell axons ( Figure 8A). In the earlier stages of PFC maturation such connectivity is probably more abundant and widespread with a possible convergence of SPF cell output into specific (mini) columns -perhaps onto select populations of pyramidal cells (see below).
Furthermore, the small proportion of non-spiny (smooth) dendritic shafts innervated by SPF cells (Table 1) likely belong to CRet-/GABA+ layer 1 LCNs (Figures 1D,F, 14H). In rat, these GABAergic neurons have been shown to furnish not only extensive lateral axonal arbors in layer 1 but also vertical axons innervating superficial pyramidal cells and LCNs, as well as pyramids deeper in the cortex (Muralidhar et al., 2014;Lee et al., 2015).

Persisting SPF Cells as "Integrative Hubs"
The presence of SPF cells with normal structural integrity in young adult and adult monkey PFC suggests that these neurons remain functional (Figure 11). SPF cells are immersed in the same extrinsic and intrinsic afferent pathways as their main synaptic targets -the feltwork of apical dendritic tufts in layer 1 (Figure 15B). However, SPF cells, with their polarized fan-like dendritic arbors, may be privy to specific combinations and strengths of inputs (Figures 15A,C). Of functional significance is the fan-like architecture of SPF cell dendrites and their short dendritic thorns. The latter structures receive numerous A-type and S-type inputs ( Figure 12A) and have the electromorphometry to significantly influence processing mechanisms in parent dendritic shafts (Han and Heinemann, 2013;Chabrol et al., 2015). Therefore, SPF cells (together with neurons of the Cajal-Retzius family and other excitatory neurons in layer 1) could act as local internuncial hubs responsible for feed-forward excitation affecting the integration of neural information within the apical dendritic tufts of target pyramidal cells (Figures 15B,C). These apical tufts would be derived from many rows of minicolumns ( Figure 15B). This strategic role would influence intra-/interareal and subcortical pyramid projection pathways from discrete patches of cortex ( Figure 15B). It is proposed that such regions could be finalizing the developmental refinement of synaptic circuitry subserving emotional and cognitive processes during the protracted postnatal maturation of primate PFC (Petanjek et al., 2011;Bianchi et al., 2013;Amlien et al., 2016).
In conclusion, this study identified eight classes of CRet+ neurons (including Cajal-Retzius cells) in layer 1 of adult macaque monkey PFC, with a morphologically distinct type of CRet+/GABA-neuron, named "subpial fan (SPF) cell, " described comprehensively. SPF cells may be related to the persisting cells identified by Ramón y Cajal and by Retzius in humans. A full characterization of this rare neuron class, and its potential role in the developmental trajectory of primate cortex is open to future combined genetic, molecular, anatomical and electrophysiological research.

AUTHOR CONTRIBUTIONS
The author designed and undertook the research, acquired and analysed the data, wrote the manuscript and approved it for publication.