Mirror trends of plasticity and stability indicators in primate prefrontal cortex

Abstract

Research on plasticity markers in the cerebral cortex has largely focused on their timing of expression and role in shaping circuits during critical and normal periods. By contrast, little attention has been focused on the spatial dimension of plasticity-stability across cortical areas. The rationale for this analysis is based on the systematic variation in cortical structure that parallels functional specialization and raises the possibility of varying levels of plasticity. Here we investigated in adult rhesus monkeys the expression of markers related to synaptic plasticity or stability in prefrontal limbic and eulaminate areas that vary in laminar structure. Our findings revealed that limbic areas are impoverished in three markers of stability: intracortical myelin, the lectin Wisteria floribunda agglutinin, which labels perineuronal nets, and parvalbumin, which is expressed in a class of strong inhibitory neurons. By contrast, prefrontal limbic areas were enriched in the enzyme calcium/calmodulin-dependent protein kinase II (CaMKII), known to enhance plasticity. Eulaminate areas have more elaborate laminar architecture than limbic areas and showed the opposite trend: they were enriched in markers of stability and had lower expression of the plasticity related marker CaMKII. The expression of glial fibrillary acidic protein (GFAP), a marker of activated astrocytes, was also higher in limbic areas, suggesting that cellular stress correlates with the rate of circuit reshaping. Elevated markers of plasticity may endow limbic areas with flexibility necessary for learning and memory within an affective context, but may also render them vulnerable to abnormal structural changes, as seen in neurologic and psychiatric diseases.

Graphical abstract

Systematic variation in cortical structure across areas suggests varying levels of plasticity. Here we show that in prefrontal cortices of adult monkeys markers related to synaptic plasticity are high in limbic areas while markers of stability are low; the opposite trend is seen in eulaminate areas, which have greater laminar elaboration than limbic areas. High plasticity markers may render limbic areas flexible for learning and memory but also vulnerable to neurologic and psychiatric diseases.

Introduction

Plasticity is the property of neural circuits to reshape their connectivity by experience to achieve novel functions (Paillard, 1976; Will et al., 2008; Berlucchi & Buchtel, 2009). In the adult cerebral cortex the best studied mechanisms for circuit reshaping require structural modification of synapses at two levels. The first level involves changes in the strength of existing excitatory synapses by modification of postsynaptic receptors that result in long term potentiation (LTP) or long term depression (LTD) (Lisman et al., 2012; Cooke & Bear, 2014; Shipton & Paulsen, 2014; Lisman, 2017). The second level involves disassembling old synapses and forming new synapses with participation of axon terminals and dendritic spines (Gogolla et al., 2007; Holtmaat & Svoboda, 2009; Holtmaat et al., 2013). Higher rate of disassembly and formation of synapses during postnatal critical periods is followed by circuit stabilization during adult normal periods (Nabel & Morishita, 2013; Takesian & Hensch, 2013).

There is a dearth of studies that compare synaptic plasticity in a spatial dimension across brain regions during normal periods. This is particularly striking for the cerebral cortex, which is a heterogeneous structure characterized by systematic variation across areas [reviewed in Barbas (2015)]. For example, cortical lamination is least differentiated in cortical limbic areas and is increasingly more elaborate along a series of six-layered (eulaminate) areas. Other architectonic and cellular features, like myelin content and spine density of pyramidal excitatory neurons, also vary systematically across areas in gradients that parallel laminar elaboration (Sanides, 1970; Barbas & Pandya, 1987; 1989; Nieuwenhuys, 2013; Elston & Fujita, 2014; Barbas & García-Cabezas, 2015; Medalla & Luebke, 2015; Medalla et al., 2017). Plasticity in the adult cortex was suspected to be higher in limbic areas based on cellular features (Barbas, 1995), which was subsequently supported by findings that in adult primates anterior cingulate (limbic) areas have higher capacity for spine and synapse formation than dorsolateral (eulaminate) prefrontal areas (Sasaki et al., 2015).

Here we systematically investigated cellular and molecular markers that either enhance or limit synaptic plasticity in limbic and eulaminate prefrontal areas that vary in laminar structure. Specifically, we studied three markers known to limit plasticity: The first was parvalbumin (PV) expressed in some inhibitory neurons, whose activation helps end critical periods and limits LTP and LTD in adult cortex (Saez & Friedlander, 2016). The second was the lectin Wisteria floribunda agglutinin (WFA), a marker of perineuronal nets (PNNs) reported to surround PV inhibitory neurons (Hartig et al., 1992; Hartig et al., 1994). Chemical removal of PNNs in the cortex restores plasticity to the level of the critical period (Pizzorusso et al., 2002; Sorg et al., 2016; Lensjo et al., 2017). The third marker was intracortical myelin, which is also associated with ending critical periods and reducing spine turnover and LTP in the cortex (Akbik et al., 2012; Schwab & Strittmatter, 2014; Boghdadi et al., 2017). The second group of markers, with presumed opposite expression, included the enzyme CaMKII, because of its key role in mediating LTP (Lisman et al., 2012; Colgan & Yasuda, 2014). We also examined two glial markers: the glial fibrillary acidic protein (GFAP), which labels activated and reactive astrocytes and is an indicator of cellular stress (Sofroniew & Vinters, 2010; Hol & Pekny, 2015), and the ionized calcium binding adapter molecule 1 (Iba1), which is specific for microglia (Ito et al., 1998; Torres-Platas et al., 2014). The Iba1 is a good marker for resting microglia in non-pathological states (Hendrickx et al., 2017).

Our findings revealed that limbic areas are enriched in markers that enable plasticity, but are impoverished in markers that underlie stability, while eulaminate areas show the opposite trend. High expression of plasticity markers in limbic areas may facilitate engagement in emotions and memory, but may also account for their increased vulnerability to psychiatric and neurologic diseases (Arnold et al., 1991; Mayberg et al., 2005; Zikopoulos & Barbas, 2010).

Methods and materials

Animal cases, perfusion and tissue processing

We obtained data from prefrontal cortex of 11 young adult rhesus monkeys (Macaca mulatta). Detailed protocols were approved by Institutional Animal Care and Use Committees (Harvard Medical School and Boston University School of Medicine) according to NIH guidelines [DHEW Publication no. (NIH) 80-22, revised 1996, Bethesda, MD, USA].

Animals were deeply anesthetized with a lethal dose of sodium pentobarbital (~50 mg/kg, intravenous, to effect) and perfused transcardially, with either 4% paraformaldehyde in cacodylate buffer or PBS (0.1M, pH 7.4; cases AL, AN, AQ, AS, AT, AV, AZ, BB, & BD), or saline followed by 6% paraformaldehyde in PB, 0.1M, pH 7.4 (cases AJ, & AK). These cases were also used for other studies (Barbas, 1993; Dombrowski & Barbas, 1996; Zikopoulos & Barbas, 2006; Ghashghaei et al., 2007; García-Cabezas & Barbas, 2016). Brains were removed from the skull, photographed, cryoprotected in ascending sucrose solutions (10–30% in PBS 0.01M at pH 7.4), frozen in −75 °C isopentane (Fisher Scientific, Pittsburg, PA, USA) for rapid and uniform freezing (Rosene et al., 1986), and cut in the coronal plane on a freezing microtome at 40 or 50 µm to produce ten matched series.

Assays and stains

To estimate neuron, astrocyte, and microglia density we stained series of sections for Nissl (n = 3 cases) which stains all neurons and glia and allows their distinction (García-Cabezas et al., 2016). Intracortical myelin was stained in series of sections using the Gallyas silver technique [n = 5 cases; Gallyas (1979); Zikopoulos et al. (2016)]. We employed immunohistochemical methods to label PV (n = 3 cases), which labels a neurochemical and functionally distinct class of inhibitory neurons in primate cerebral cortex (DeFelipe, 1997), WFA (n = 2 cases), which recognizes N-acetylgalactosamine-containing epitopes in the CS-GAG chains and labels extracellular PNNs (Hartig et al., 1992; Hartig et al., 1994), αCaMKII (n = 2 cases), GFAP (n = 2 cases), and Iba1 (n = 2 cases). Briefly, free-floating sections were rinsed in PBS (0.01M, pH 7.4), incubated in 0.01 M sodium citrate buffer, pH 8.5, at 80–85°C for 30 min for antigen retrieval (only for WFA staining), incubated for 1 h in 0.05 M glycine, and pre-blocked [10% normal goat or horse serum, 5% bovine serum albumin (BSA), and 0.2% Triton-X in PBS]. Sections were then incubated overnight in WFA (Biotinylated Wisteria floribunda lectin, cat. no. B-1355, Vector Laboratories, Burlingame, CA, USA; diluted 1/200) or in primary antibody against PV (mouse anti-PV, cat. no. 235, Swant Antibodies, Marly, Switzerland; diluted 1:3,000), αCaMKII (mouse anti-αCaMKII, cat. no. 1481703, Boehringer Manheim, Indianapolis, IN, USA; diluted 1:400), GFAP (rabbit anti-GFAP, cat. no. G9269, Sigma-Aldrich, St. Louis, MO, USA; diluted 1:500), or Iba1 (goat anti-Iba1, cat. no. ab5076, Abcam, Cambridge, MA, USA; diluted 1/1000 in PBS, 1% normal goat or horse serum, 1% BSA, and 0.1% Triton-X), rinsed in PBS and incubated for 4 h in the respective secondary biotinylated antibody (goat anti-mouse IgG, cat. no. BA-9200; goat anti-rabbit IgG, cat. no. BA-1000; horse anti-mouse IgG, cat. no. BA-2000; or horse anti-goat, BA-9500; Vector Laboratories, diluted 1:200 in PBS, 1% normal goat or horse serum, 1% BSA, and 0.1% Triton-X). For biotinylated WFA no secondary antibody was needed. We then incubated sections for 1 h in avidin-biotin horseradish peroxidase complex (AB-HRP kit; Vectastain PK-6100 ABC Elite kit, Vector Laboratories; diluted 1:100 in 0.01M PBS with 0.1% Triton X-100), rinsed in PBS and processed for 2–3 min for the peroxidase-catalyzed polymerization of diaminobenzidine (DAB; Vector or Zymed Laboratories Inc., South San Francisco, CA, USA; 0.05% DAB and 0.004% H₂O₂ in PBS). Sections were mounted on gelatin-coated slides and dried, and some were counterstained for Nissl (García-Cabezas et al., 2016), dehydrated in graded alcohols, cleared in xylenes and coverslipped with mounting media (Permount, Fisher Scientific; or Entellan, EM Sciences, Hatfield, PA, USA).

Unbiased estimate of PV and Nissl stained neurons, astrocytes, and microglia

We estimated the density of parvalbumin-positive (PV+) neurons, their proportion in the entire neuron population, and the density of astrocytes and microglia. Measures were obtained from representative columns along the depth of the gyral part of anterior cingulate areas 25 and 32, medial area 10 (10m), and caudal dorsolateral area 46d (Figure 1), based on the maps of Barbas and Pandya (1989). We used the unbiased stereological method of the optical fractionator (Gundersen, 1986; Howard & Reed, 1998) in conjunction with a commercial system (StereoInvestigator; MicroBrightField, Inc., Williston VT, USA), as described [e. g., García-Cabezas and Barbas (2014a; 2014b)]. We first drew contours of layers in each column (layers I and II–III in all areas; layers IV–VI in areas 25 and 32; layers IV and V–VI in eulaminate areas 10m and 46d) to estimate the number of neurons, astrocytes, and microglia by laminar groups or for entire columns (I–VI). We counted PV+ neurons at 400× and Nissl stained neurons, astrocytes, and microglia at 1000× from a minimum of three evenly spaced sections/case/area, using systematic random sampling. The counting frame (disector) size for PV+ neurons was 200 µm and for neuron, astrocyte, and microglia counts it was 50–60 µm, based on pilot study. We used as guard zones the top and bottom of each section (minimum 2 µm in 10–15 µm sections after tissue shrinkage). We measured section thickness at each counting site using the program software. The height of the counting frame was 5 µm and grid spacing was 100–300 µm. Neurons, astrocytes, and microglia were counted if their nuclei fell within the counting frame or touched the two acceptance lines but not the two forbidden lines (Howard & Reed, 1998). These parameters yielded a sampling fraction with a coefficient of error of <10% per contour, with the exception of PV+ neurons in layer IV where the error was <15%, due to the small laminar volume, as recommended (Gundersen, 1986; Howard & Reed, 1998). We computed cell density by dividing the estimated number of counted cells with the estimated volume of each contour.

Figure 1.

Limbic and eulaminate areas of the monkey prefrontal cortex differ in laminar structure. A, B, Maps of monkey prefrontal cortex (Barbas & Pandya, 1989). A, Medial surface; B, Lateral surface. The maps show areas with the lowest (black) and highest (lightest grey) laminar elaboration. C–F, Photomicrographs of areas 25, 32, 10m, and 46d stained with Nissl. C, D, Areas 25 and 32 have a rudimentary layer IV (dysgranular). Layer I is thick and shows poor delimitation with layer II. Deep layers V–VI are more prominent than superficial layers II–III. E, Eulaminate (I) area 10m has six layers. F, Layer IV in eulaminate (II) area 46d is better developed than in area 10m. Layer I is thinner than in areas 25 and 32 and is delineated from layer II. Superficial layers II–III are denser than in limbic areas 25 and 32. Abbreviations: MPAll, medial periallocortex; WM, white matter. Arabic numerals show cortical areas according to Barbas and Pandya (1989). Roman numerals indicate cortical layers. Calibration bar in F applies to C–F.

Optical density analysis of intracortical myelin, WFA, αCaMKII, and GFAP

We quantified intracortical myelin content, WFA, αCaMKII or GFAP expression using optical density from photomicrographs of representative columns from each area. We first captured images under bright-field (myelin, WFA, and αCaMKII) or dark-field (GFAP) with a CCD camera (Olympus DP70) mounted on an optical microscope (Olympus BX 51) connected to a personal computer using image software (DP Controller). We captured images at 100× (UPlanFl 10×/0.30 Japan) with the same light exposure and obtained optical density measurements from 5 to 15 images taken from 3 to 8 sections per marker/case/area.

We imported photographs into MATLAB (MATLAB and Statistics Toolbox Release R2015b, The MathWorks, Inc., Natick, MA, USA) to convert into gray scale. Each bright-field image was inverted by subtracting pixel values from the darkest possible pixel value, so that pixels with strong staining had higher numerical values than the background. We measured the mean grey level density for each image and obtained overall mean gray level values for each area. We then estimated the grey level density along the depth of normalized columns, encompassing cortical thickness from the pial surface to the white matter. This involved two steps for each image: First, we computed a vertical grey level density profile along the depth of the cortical region. Second, we divided each profile into 20 bins and averaged density values across images for each bin.

Statistical analyses

We employed one-way ANOVA for overall comparison of markers across areas and laminar groups. For analyses that showed significant differences (p<0.05) we performed post hoc pair comparisons (Bonferroni method). We report p-values, F-statistics, and degrees of freedom (shown as subscripts: F_{between groups, within groups}). Data were tabulated in Excel (Office 365, Microsoft) and analyses were performed using MATLAB.

To demonstrate global similarities/differences among prefrontal areas using all markers simultaneously (10 parameters) we performed nonmetric multidimensional scaling (NMDS), which allows visualization of high-dimensional data into a low two-dimensional space that approximates pairwise distances between data points. Each area was initially represented using a feature vector with ten dimensions: PV+ neuron density: (1) across all layers, (2) layers II–III, (3) layers IV–VI in areas 25 and 32 or layers V–VI in eulaminate areas 10m and 46d; PV+ to neuron ratio: (4) across all layers, (5) layers II–III, (6) layers IV–VI in areas 25 and 32 or layers V–VI in eulaminate areas 10m and 46d; mean gray level values: (7) of WFA, (8) myelin, (9) αCaMKII, and (10) GFAP. Data were z-scored to remove scale-related effects. We employed the NMDS algorithm of MATLAB using the stress criterion for goodness of fit. The resulting NMDS diagram displayed areas in a 2-dimensional space that closely fit the Euclidean distances among areas in the high-dimensional space, in which the relative proximity of areas represents their relative similarity/dissimilarity.

Photography for figures

We photographed representative columns with labeling under bright-field (Nissl, PV, myelin, WFA, and αCaMKII) or dark-field (GFAP) to assemble in figures using Adobe Illustrator CC software (Adobe Systems Incorporated, San José, CA, USA). We made minor adjustment of overall brightness and contrast, but did not retouch images.

Results

Progressive laminar elaboration from medial to dorsolateral prefrontal areas

Figure 1 shows the cytoarchitecture of prefrontal areas with the lowest (anterior cingulate limbic areas 25 and 32), intermediate (medial area 10, 10m), and most elaborate (dorsolateral area 46, 46d) laminar structure. These areas were used for comparison of plasticity-stability related markers. Areas 25 and 32 are dysgranular, with a rudimentary layer IV (Figure 1C, D). Area 10m, situated anterior to area 32, has six layers (eulaminate) and has been categorized as eulaminate I (Figure 1E) (Dombrowski et al., 2001). Area 46d has the best developed lamination with a prominent layer IV, and has been categorized with eulaminate II areas (Figure 1F). There is a notable increase in neuron density in the upper layers seen from Figure 1C to 1F.

Density of PV+ inhibitory neurons is higher in eulaminate areas

PV+ neurons were sparsely distributed in layers II–VI of area 25 and formed a thin but conspicuous band in layer V (Figure 2A). In area 32, PV+ neurons were comparable to area 25, but were more evenly distributed in layers II–VI (Figure 2B). Area 10m had more PV+ neurons than areas 25 or 32 (Figure 2C), and area 46d had the most, distributed in layers II–VI (Figure 2D).

Figure 2.

Distribution of parvalbumin positive (PV+) neurons in limbic and eulaminate areas of the monkey prefrontal cortex. A–D, Photomicrographs of areas 25, 32, 10m, and 46d. A, In area 25, PV+ neurons are sparse across layers and form a thin band in layer V. B, PV+ neurons are more evenly distributed within layers in area 32 than in area 25. C, D, Eulaminate areas 10m and 46d have more PV+ neurons than areas 25 and 32 with a dense band in layers IV and V. E, PV+ neuron density is higher in eulaminate areas 10m and 46 than in areas 25 and 32 across layers. F, PV+ neuron density is higher in eulaminate areas 10m and 46 than in areas 25 and 32 in the superficial layers. G, PV+ neuron density is higher in eulaminate areas 10m and 46 than in areas 25 and 32 in the deep layers (IV–VI for limbic, V–VI for eulaminate). H, Layer IV, which is distinct in areas 10m and 46d, has the highest PV+ neuron density. I, The proportion of PV+ neurons for the entire neuron population is higher in eulaminate areas 10m and 46d than in limbic areas 25 and 32 across layers. J, The proportion of PV+ neurons for the neuron population in superficial layers is higher in eulaminate areas 10m and 46d than in limbic areas 25 and 32. K, The proportion of PV+ neurons for the neuron population in the deep layers is higher in eulaminate areas 10m and 46d than in limbic areas 25 and 32 (layers IV–VI for limbic, V–VI for eulaminate). L, The proportion of PV+ neurons for the neuron population in layer IV is high in areas 10m and 46d. WM, white matter. Roman numerals indicate cortical layers. Asterisks in E, F, I–K indicate significant differences between pairs of areas, as determined by post hoc analysis (Bonferroni method) conducted after one-way ANOVA. Scatter plots in E–L represent individual cases denoted by different symbols; greyscale horizontal bars represent case averages; vertical lines on bars show the standard error. Calibration bar in D applies to A–D.

One-way ANOVA followed by post hoc comparisons revealed significantly higher density of PV+ neurons (neurons/mm³) in areas 10m and 46d than in limbic areas 25 and 32 across layers (area 25 = 3,736 ± 470; area 32 = 3,522 ± 382; area 10m = 7,167 ± 762; area 46d = 7,613 ± 308; ANOVA, p = 0.002, F_3,8 = 12.19; Figure 2E). These differences were also significant for the superficial layers II–III (area 25 = 4,388 ± 564; area 32 = 4,638 ± 645; area 10m = 8,522 ± 709; area 46d = 9,797 ± 346; ANOVA, p = 0.001, F_3,8 = 14.71; Figure 2F), but not for the deep layers (IV–VI in limbic areas 25 and 32, V–VI in eulaminate areas 10m and 46d; area 25 = 4,597 ± 666; area 32 = 3,561 ± 503; area 10m = 6,383 ± 1,272; area 46d = 6,255 ± 752; ANOVA, p = 0.24, F_3,8 = 1.72; Figure 2G). In layer IV of areas 10m and 46d, PV+ neuron density was higher than in other layers (area 10m = 15,906 ± 1,229; area 46d = 15,877 ± 1,243; Figure 2H).

We then estimated neuron density in Nissl stained sections as a first step to compute PV+ neurons as a proportion of the entire neuron population for each area and laminar group. Neuron density (neurons/mm³) was overall higher in eulaminate areas 10m and 46d than in limbic areas 25 and 32 across layers I–VI (area 25 = 48,652 ± 4,063; area 32 = 46, 695 ± 534; area 10m = 56,452 ± 5,692; area 46d = 53,121 ± 3,828), as well as in layers II–III (area 25 = 47,577 ± 3,059; area 32 = 49,961 ± 857; area 10m = 58,954 ± 7,977; area 46d = 60,463 ± 6,262), but these differences were not statistically significant (ANOVA, p = 0.52, F_3,8 = 0.81; p = 0.45, F_3,8 = 0.97; and p = 0.13, F_3,8 = 2.53 respectively). The most striking difference was in layer IV of eulaminate areas 10m and 46d, which was denser than other layers (area 10m = 98,303 ± 6,991; area 46d = 111,400 ± 2,132) consistent with previous findings (Dombrowski et al., 2001).

We then computed the density of PV+ neurons as a proportion of the entire neuron population, and found that it was higher in eulaminate areas 10m and 46d than in limbic areas 25 and 32 across layers (area 25 = 0.08 ± 0.01; area 32 = 0.09 ± 0.03; area 10m = 0.13 ± 0.002; area 46d = 0.15 ± 0.02; ANOVA, p = 0.003, F_3,8 = 11.17; Figure 2I). The highest proportion of PV+ neurons was also seen in the group of superficial layers II–III (area 25 = 0.09 ± 0.01; area 32 = 0.09 ± 0.02; area 10m = 0.15 ± 0.01; area 46d = 0.17 ± 0.03; ANOVA, p = 0.014, F_3,8 = 6.81; Figure 2J), and the deep layers (layers IV–VI in limbic areas 25 and 32 or layers V–VI in eulaminate areas 10m and 46d; area 25 = 0.07 ± 0.01; area 32 = 0.07 ± 0.02; area 10m = 0.09 ± 0.01; area 46d = 0.13 ± 0.02; ANOVA, p = 0.011, F_3,8 = 7.4; Figure 2K). Post hoc comparisons revealed significant differences between each limbic area and eulaminate area 46d across layers, as well as for superficial layers II–III and deep layers. In layer IV of areas 10m and 46d the proportion of PV+ neurons was comparable to layers II–III (area 10m = 0.15 ± 0.01; area 46d = 0.14 ± 0.01; Figure 2L).

PNNs density is higher in eulaminate areas

In area 25 there was scant PNN labeling with WFA in layers II–III but there was a band of stronger staining in the neuropil in layer V (Figure 3A). Area 32 had slightly more label of PNNs in layers II–III than area 25 (Figure 3B). In area 10m, WFA staining was comparable to area 32 (Figure 3C). In area 46d there was more labeling of PNNs across layers II–VI than in the other areas (Figure 3D). Layer I did not show WFA staining in any area. There was evidence of PNNs label around some pyramidal neurons across areas as well (Figure 3A, black arrow).

Figure 3.

Perineuronal net (PNN) label by the lectin Wisteria floribunda agglutinin (WFA) in limbic and eulaminate areas of the monkey prefrontal cortex. A–D, Photomicrographs of areas 25, 32, 10m, and 46d stained for WFA. A, area 25 shows scant label in layers II–III and a band of WFA staining in the neuropil of layer V; black arrow points at PNN in a pyramidal neuron. B, C, Areas 32 and 10m show more label for WFA in layers II–III and IV–V than area 25. D, Area 46d shows the highest label for WFA across layers II–VI compared to other areas. E, The mean gray level index through the depth of the cortex shows higher levels of WFA staining in area 46d. F, WFA content increases towards the middle layers in the four areas, shown along the course from the surface of the cortex (left) to the edge of the white matter (right); the highest density is found consistently in area 46d and the lowest in area 25. WM, white matter. Roman numerals indicate cortical layers. Scatter plots in E represent individual cases denoted by different symbols; greyscale horizontal bars represent case averages; vertical lines on bars show the standard error. Calibration bar in D applies to A–D.

The above findings were corroborated by measuring the mean gray level index of WFA staining through the depth of the cortex (Figure 3E). The variation in grey level density of WFA along the depth of normalized columns showed an increase towards the middle layers in the four areas (Figure 3F). The middle bins showed higher content of WFA in area 46d than in the other areas, as shown by separation of the function for this area (Figure 3F).

Intracortical myelin content is higher in eulaminate areas

Area 25 had the sparsest myelinated axons, arranged in vertical arrays in layers IV–VI (Figure 4A), a pattern that was more elaborate in area 32 (Figure 4B). In eulaminate area 10m vertical arrays of myelinated axons were thicker and interwoven with abundant horizontal myelinated axons in layers IV, V, and VI, and there were more myelinated axons in superficial layers II–III compared with areas 25 and 32 (Figure 4C). Eulaminate area 46d had vertical myelinated axons forming thick bundles that extended from the white matter to layer III and above, which were denser than in the other areas (Figure 4D).

Figure 4.

Myelin content in limbic and eulaminate areas of the monkey prefrontal cortex. A–D, Photomicrographs of areas 25, 32, 10m, and 46d stained with the Gallyas technique for myelin. A, B, In areas 25 and 32 the content of intracortical myelin is lower than in eulaminate areas. C, D, Area 10m and area 46d show progressive increase of intracortical myelin. E, The mean gray level index of myelin through the depth of the cortex also shows this trend. F, Myelin content increases towards the white matter in the four areas, and is higher in the middle-deep bins of areas 46d and 10m than in areas 25 and 32. WM, white matter. Roman numerals indicate cortical layers. Scatter plots in E represent individual cases denoted by different symbols; greyscale horizontal bars represent case averages; vertical lines on bars show the standard error. Calibration bar in D applies to A–D.

Measurement of the mean gray level index of myelin through the depth of the cortex corroborated the qualitative patterns described above (Figure 4E). The variation in grey level density of myelin along the depth of normalized columns showed an increase towards the white matter in the four areas. The middle-deep bins showed higher content of myelin in areas 46d and 10m than in areas 25 and 32, revealed by the separation of the four functions (Figure 4F).

αCamKII expression is higher in limbic areas

The density of αCaMKII showed the opposite trend in the four areas than PV+ neurons or myelin. In areas 25 and 32, αCaMKII expression was dense in the neuropil of layers I, II, and superficial part of layer III, and was moderate in the deep part of layer III and layers IV–VI (brown label; Figure 5A, B). In area 10m expression of αCaMKII was dense in layers I and II, moderate in layers III, V, and VI and light in layer IV (Figure 5C). In area 46d αCaMKII expression was dense in layer I, moderate in layer II, superficial layer III and layer VI, and very light in the deep part of layer III, layer IV, and upper layer V (Figure 5D).

Figure 5.

Expression of the alpha subunit of the calcium/calmodulin-dependent protein kinase II (αCaMKII) in limbic and eulaminate areas of the rhesus monkey prefrontal cortex. A–D, Photomicrographs of areas 25, 32, 10m, and 46d stained for αCaMKII. A, Area 25 shows high neuropil expression of αCaMKII across layers (dark brown). B, C, In area 32 and in area 10m, αCaMKII expression is lower than in area 25. D, Area 46d shows lower expression of αCaMKII than areas 25, 32, and 10m across layers except in layer I. E, The mean gray level index of αCaMKII also shows the trend of A–D. F, αCaMKII expression decreases towards the white matter in the four areas. WM, white matter. Roman numerals indicate cortical layers. Asterisks in E indicate significant differences between pairs of areas, as determined by post hoc analysis (Bonferroni method) conducted after one-way ANOVA. Scatter plots in E represent individual cases denoted by different symbols; greyscale horizontal bars represent case averages; vertical lines on bars show the standard error. Calibration bar in D applies to A–D.

One-way ANOVA followed by post hoc comparisons revealed significantly lower mean gray level index of αCaMKII in area 46d than in areas 25 and 32 (ANOVA, p = 0.001, F_3,4 = 57.71; Figure 5E). The variation in grey level density along the depth of normalized columns showed that αCaMKII expression decreased towards the white matter in the four areas (Figure 5F).

GFAP expression is higher in limbic areas but astrocyte density is comparable across areas

Expression of GFAP, a marker of activated astrocytes, revealed a distinct and pronounced trend across the four areas. Area 25 showed dense and uniform expression of GFAP across layers (Figure 6A, yellow label). A similar pattern was evident in area 32, with the exception of moderate expression in the middle cortical layers (Figure 6B). In eulaminate areas 10m and 46d, dense GFAP expression was restricted to layers I and VI with moderate expression in layer II and superficial layer III, due to GFAP labeling of the processes of interlaminar astrocytes located in layer I. The deep part of layer III, layer IV and layer V had light expression of GFAP (dark region in Figure 6C, D).

Figure 6.

Expression of glial fibrillary acidic protein (GFAP) in limbic and eulaminate prefrontal areas in rhesus monkeys. A–D, Photomicrographs of areas 25, 32, 10m, and 46d stained using immunohistochemistry for GFAP (gold label). A, Area 25 shows dense GFAP labeling across layers. B, GFAP labeling is also dense in area 32 but the middle layers show moderate labeling (less gold labeling). C, D, The middle layers in eulaminate areas 10m and 46d show light labeling of GFAP with dense expression in layers I, II, and VI. E, The mean gray level index of GFAP through the depth of the cortex also shows this trend. F, GFAP expression decreases from layer I and the white matter towards the middle part of the cortex in the four areas, a pattern that is more pronounced in eulaminate areas 10m and 46d. WM, white matter. Roman numerals indicate cortical layers. Asterisks in E indicate significant differences between pairs of areas, as determined by post hoc analysis (Bonferroni method) conducted after one-way ANOVA. Scatter plots in E represent individual cases denoted by different symbols; greyscale horizontal bars represent case averages; vertical lines on bars show the standard error. Calibration bar in D applies to A–D.

One-way ANOVA followed by post hoc analysis of the mean gray level index showed that areas 25 and 32 had significantly higher GFAP expression than eulaminate area 46d (ANOVA, p = 0.002, F_3,4 = 16.56; Figure 6E). The variation in grey level density along the depth of normalized columns showed that GFAP expression was higher in layer I and layer VI while the middle part of the cortex around layer IV had the lowest level (Figure 6F).

We then estimated astrocyte density in Nissl stained sections to investigate whether it correlated with GFAP expression or not. This analysis revealed that the densities of astrocytes in layers I–VI were not significantly different among areas (area 25 = 22,448 ± 2,058; area 32 = 23,357 ± 968; area 10m = 23,515 ± 1688; area 46d = 21,515 ± 1,484; ANOVA, p = 0.89, F_3,8 = 0.22).

Iba-1 expression and microglia density are comparable across areas

Expression of the calcium binding protein Iba1, which labels resting microglia (Hendrickx et al., 2017), was comparable across the areas studied. Labeled macrophages were homogeneously distributed across areas and layers forming a network that looked like a starry sky. Iba1-labeled cells had a small cell body and ramified cytoplasm, consistent with resting microglia (Torres-Platas et al., 2014). We also estimated microglia density in Nissl stained sections and found comparable numbers across areas in layers I–VI (area 25 = 5,908 ± 318; area 32 = 6,426 ± 495; area 10m = 6,099 ± 194; area 46d = 5,688 ± 706; ANOVA, p = 0.83, F_3,8 = 0.3) consistent with the above findings.

Overall segregation of prefrontal areas by plasticity-stability markers

Non-metric multidimensional scaling (NMDS) made it possible to condense features along a 10-dimensionsal space into a 2-dimensional space, to facilitate visualization while preserving differences among the four areas. NMDS revealed a clear separation between limbic (left) and eulaminate areas (right), and also showed separation between each pair of cortices (Figure 7). Stress was negligible (~6×10⁻¹⁷) indicating that the 2-dimensional space accurately reproduced the differences among areas.

Figure 7.

Nonmetric multidimensional scaling (NMDS) diagram shows separation between limbic areas 25 and 32 (left) and eulaminate areas 10m and 46d (right) with a negligible level of stress (~6×10⁻¹⁷), indicating that the 2-dimensional space accurately reproduced the differences among areas.

Discussion

Our findings revealed that limbic prefrontal areas showed significantly lower expression of markers associated with stability and higher expression of the marker αCaMKII associated with plasticity and GFAP, which is related to cellular stress (Sofroniew & Vinters, 2010; Hol & Pekny, 2015). In contrast, eulaminate areas showed the opposite trend. These findings were predicted from the graded increase in laminar elaboration from limbic to eulaminate areas (Barbas, 2015). This evidence suggests that the potential for synaptic plasticity is higher in limbic areas than in eulaminate areas (Barbas, 1995). These features are consistent with a key role of limbic areas in emotions, learning and memory and eulaminate prefrontal areas in cognition (Fuster, 2008; Pessoa, 2008; John et al., 2013; Anderson et al., 2015; Chanes & Barrett, 2016; Barbas & García-Cabezas, 2017).

Variation of plasticity related markers suggests higher synaptic plasticity in limbic areas

The reshaping of neural circuits by experience in adult cortex is constrained by factors that regulate synaptic plasticity, including postsynaptic receptors and synapse turnover (Holtmaat & Svoboda, 2009; Lisman et al., 2012; Takesian & Hensch, 2013; Lisman, 2017). Inhibitory circuitry limits synaptic plasticity during normal periods that follow the highly plastic critical periods, as seen by blockade of glutamic acid decarboxylase or GABA_A receptors (Harauzov et al., 2010; Saez & Friedlander, 2016). Specifically, digestion of PNNs around PV inhibitory neurons reverses synaptic plasticity to the critical period (Pizzorusso et al., 2002; Romberg et al., 2013; Lensjo et al., 2017). A striking finding here is the significantly lower proportion of PV+ neurons from the entire neuron population in prefrontal limbic areas than in the best delineated eulaminate areas. The significance of this finding is based on functional evidence that PV+ neurons exert strong perisomatic inhibition of nearby pyramidal neurons (DeFelipe, 2002).

A significant proportion of PV neurons in the cerebral cortex are coated by extracellular PNNs (Hartig et al., 1992; Hartig et al., 1994). The development of PNNs coincides with the end of critical periods and PNN removal in adult life reverts plasticity levels to critical periods (Pizzorusso et al., 2002; Sorg et al., 2016; Lensjo et al., 2017). We found that PNNs density, as labeled by WFA, is higher in eulaminate area 46d than in areas 25, 32, and 10m. Comparable patterns of labeling of WFA across prefrontal areas were found in the cebus monkey (Cruz-Rizzolo et al., 2011).

The pattern of myelin content followed a similar trend as PV neuron density. Myelin associated proteins limit cortical synaptic plasticity, spine turnover and LTP (Akbik et al., 2013; Schwab & Strittmatter, 2014; Boghdadi et al., 2017). This pattern is consistent with more elaborate dendritic trees and more spines in limbic than in eulaminate areas in the prefrontal cortex of primates (Elston et al., 2005a; Elston et al., 2005b; Elston et al., 2005c; Elston et al., 2006; Elston et al., 2011; Sasaki et al., 2015; Medalla et al., 2017) .

In contrast, αCaMKII, which is a key modulator of synaptic plasticity and crucial for LTP (Lisman et al., 2012), showed high expression in the neuropil of limbic area 25, was intermediate in areas 32 and 10m, and low in area 46d. In area 46d only layer I had high expression of αCaMKII. These findings may reflect differences in spine labeling, and suggest that in limbic areas LTP is facilitated across layers. In contrast, in area 46d LTP may be more easily attained in layer I, which also lacks PV neurons and PNNs and is the major recipient of feedback projections from areas with less elaborate laminar structure, including limbic areas (Barbas, 2015).

As summarized in Figure 8, the opposite trends in the expression of plasticity and stability markers suggest that prefrontal limbic areas are more plastic than eulaminate areas. In contrast, eulaminate cortices with the best delineated laminar structure may be more stable.

Figure 8.

Markers of plasticity in the cortex parallel laminar differentiation. A, Cartoon depicts expression of factors that limit synaptic plasticity, which are higher in eulaminate than in limbic areas. B, Conversely, expression of αCaMKII, known to enhance synaptic plasticity, is higher in limbic areas; GFAP expression, a marker of cellular stress, is also higher in limbic areas than in eulaminate areas. C, The distribution of these markers suggests that cortical plasticity and stability change systematically with laminar differentiation as shown in the cartoon of cellular density across areas.

GFAP expression suggests more activation of astrocytes in limbic areas

We found comparable densities of astrocytes and microglia across limbic and eulaminate areas. The distribution of cells labeled for Iba1, a marker that labels resting microglia, was also comparable across areas. Iba1-labeled cells across areas in our material showed features of resting microglia and comparable levels across areas. In contrast, labeling for GFAP, a marker of activated astrocytes (Sofroniew & Vinters, 2010; Hol & Pekny, 2015), was higher in limbic areas 25 and 32 than in eulaminate area 10m and especially in area 46d, which has the best delineated laminar structure. This evidence suggests that astrocytes in limbic cortices normally have higher levels of cellular stress. Astrocytes can impact synaptic plasticity because they participate in cellular and molecular processes related to synapse and neurotransmitter regulation (Singh & Abraham, 2017). The higher expression of GFAP in limbic areas may reflect higher demands of neurons engaged in synaptic plasticity functions.

Plasticity-stability trends along the cerebral cortex

Our analysis focused on four representative prefrontal areas, but data in the literature suggest comparable distribution of markers associated with plasticity and stability across the primate cerebral cortex. For instance, PV expression and PNNs density varies in the human temporal areas (Ding et al., 2009) with a notable increase progressively from limbic to eulaminate areas. Similarly, classical studies have shown that intracortical myelin increases from limbic to eulaminate areas in all lobes and systems of the primate brain (Sanides, 1970). At the other extreme of the plasticity-stability spectrum, expression in adult cortex of the growth-associated phosphoprotein 43 (GAP-43), a marker related to synaptic plasticity and LTP (Benowitz et al., 1989; Benowitz & Routtenberg, 1997), has a complementary expression to myelin, with much higher expression in limbic areas than in eulaminate cortices; the latter include the highly differentiated primary areas (Benowitz et al., 1989). Other cellular features also vary in parallel with laminar elaboration in the cortex and the expression of plasticity-stability related markers. For instance, in the monkey temporal and frontal lobes spine density and the length of dendritic arborization varies consistently across cortical areas (Elston, 2003; Elston et al., 2005a; Elston et al., 2005b; Elston et al., 2005c; Elston et al., 2006; Sasaki et al., 2015; Medalla et al., 2017). Laminar elaboration in inferior temporal and occipital areas of the macaque monkey increases in the anterior to posterior direction as shown in Figure 9 of Hilgetag et al. (2016). In the same areas, spine density, dendritic size, and dendritic complexity decrease in the anterior to posterior direction as shown in Figure 6 of Elston (2003). These structural features show steady changes across areas in parallel with changes in elaboration of laminar structure from limbic to eulaminate cortices.

Our findings, along with observations from the literature, suggest that limbic areas, which are found at the foot of every cortical system (Barbas, 1986; 1988), are more plastic, and their neurons have larger dendritic arbors with more spines, consistent with their multimodal, associative functions. In contrast, eulaminate areas across systems are more stable along the axis of increasing laminar elaboration. Pyramidal neurons in eulaminate areas have smaller dendritic trees with fewer spines, consistent with more specialized functions, a pattern accentuated in the highly laminated primary sensory cortices (Barbas, 2015).

Plasticity-stability features and implications for pathology

Cortical areas are not equally susceptible to neurologic and psychiatric diseases. For example, aggregation of tau protein in Alzheimer’s and α-synuclein in Parkinson’s disease starts in limbic cortices and spreads gradually towards eulaminate areas (Arnold et al., 1991; Duyckaerts et al., 1998; Braak et al., 2006; Brettschneider et al., 2015). Similarly, the lower density of the powerful PV neurons may render limbic cortices vulnerable to epileptiform activity (Barbas, 1995; Zikopoulos & Barbas, 2013). Anterior cingulate limbic areas, including areas 25 and 32, are also preferentially implicated in psychiatric diseases like depression (Mayberg et al., 2005) and autism (Zikopoulos & Barbas, 2010; 2013).

The observation of differential involvement of areas in neurologic and psychiatric diseases led Oskar and Cécile Vogt to propose the idea of pathoclisis, or selective vulnerability. The Vogts speculated that specific differences in the physico-chemical composition of cells provided the basis for differences in susceptibility to insult and disease in different brain regions, though there was little evidence to substantiate the idea at the cellular and molecular levels (Vogt & Vogt, 1922; Klatzo, 2003). Based on novel findings here, we suggest that the complement of cellular and molecular features at once endows limbic areas with plasticity but also vulnerability to disruption in disease. High plasticity entails high metabolic activity and cellular stress, as revealed by higher expression of GFAP in anterior cingulate limbic areas compared with eulaminate cortices. These findings provide the basis to investigate in future studies factors that facilitate the highly plastic processes of learning, emotions and memory, as well as the trigger that leads to disruption and pathology in neurologic and psychiatric diseases.

Abbreviations

αCaMKII

alpha subunit of calcium/calmodulin-dependent protein kinase II

GFAP

glial fibrillary acidic protein

LTD

Long term depression

LTP

Long term potentiation

Iba1

Ionized calcium binding adapter molecule 1

NMDS

nonmetric multidimensional scaling

PNNs

perineuronal nets

PV

parvalbumin

WFA

lectin Wisteria floribunda agglutinin

WM

white matter