Ultrastructural heterogeneity of human layer 4 excitatory synaptic boutons in the adult temporal lobe neocortex

Synapses are fundamental building blocks that control and modulate the ‘behavior’ of brain networks. How their structural composition, most notably their quantitative morphology underlies their computational properties remains rather unclear, particularly in humans. Here, excitatory synaptic boutons (SBs) in layer 4 (L4) of the temporal lobe neocortex (TLN) were quantitatively investigated. Biopsies from epilepsy surgery were used for fine-scale and tomographic electron microscopy to generate 3D-reconstructions of SBs. Particularly, the size of active zones (AZs) and of the three functionally defined pools of synaptic vesicles (SVs) were quantified. SBs were comparably small (∼2.50 μm2), with a single AZ (∼0.13 µm2) and preferentially established on spines. SBs had a total pool of ∼1800SVs with strikingly large readily releasable (∼ 20), recycling (∼ 80) and resting pools (∼850). Thus, human L4 SBs may act as ‘amplifiers’ of signals from the sensory periphery and integrate, synchronize and modulate intra- and extra-cortical synaptic activity.


126
The main goal of this study was to quantify several morphological parameters representing structural 143 correlates of synaptic transmission and plasticity in L4 excitatory SBs in the human TLN. For this purpose, 144 150 SBs and 155 AZs were completely reconstructed out of five series of 70-100 ultrathin sections/series 145 using biopsy material from TLE surgery (see Material and Methods). 146 EM investigation of L4 in the human TLN revealed a dense neuropil containing neuronal cell bodies, 147 astrocytes and their fine processes, dendrites and SBs of different shape and size (Fig. 2, 3) and traversing 148 apical dendrites of L5 pyramidal neurons (Fig. 1B) which are much thicker in diameter and were thus not 149

sampled. 150
Synaptic complexes in L4 were formed by either presynaptic en passant or endterminal boutons (Fig. 2), 151 with their prospective postsynaptic target structures, either a cell body of a neuron, or dendritic shafts of 152 different calibers (17.7 %) or spines of distinct sizes and types (thin 58.20%; filopodial 5.70%; mushroom 153 7.80%; stubby 3.50% and 7.10% were not classifiable). In our samples SBs were predominantly (~82% of 154 the total) found on dendritic spines, ~3% of which are SBs establishing either two or three synaptic contacts 155 on the same spine and numerous contacts with either the same or different dendrites (Fig. 2B, C). 156 Interestingly, only ~40 % of the spines in our sample contained a spine apparatus (Figs. 2B, 3A, B), a 157 specialized form of the endoplasmic reticulum. 158 L4 SBs were on average small, with a mean surface area of 2.50 ± 1.78 µm 2 , and a mean volume of 0.16 159 ± 0.16 µm 3 , respectively. SBs were oval to round with a form factor, ranging from 0.27 to 0.77 and a mean 160 of 0.56 ± 0.09. Beside relatively large SBs (11.54 µm 2 ; 1.09 µm 3 ) also very small ones (0.42 µm 2 ; 0.01 µm 3 ) 161 were sampled and quantified. Hence, a huge variability was observed with respect to the shape and size of 162 SBs as indicated by the large SD, CV and variance (Table 2) regardless of their target structures. 163 Interestingly, a high correlation between the surface area and volume of SBs was observed as indicated by 164 the coefficient of correlation (R 2 ; Fig. 5A). 165 In larger SBs, several mitochondria (range 1 to 4) of different shape and size (0.02 ± 0.03 µm 3 ) were 166 present, occupying ~13% of the total bouton volume ( Figs    In our study, the average total pool of SVs was 1820.64 ± 980.34 SVs (ranging from 368 to 5053) 256 occupying ~7% (0.01 µm 3 ) of the total volume of SBs, although the range and the SD indicated a huge 257 variability in total pool size (Table 2) at individual SBs. Strikingly, the total pool in the human TLN was 258 already ~3-fold larger when compared with L4 and L5 SBs in rats (Rollenhagen et al. 2015, although 259 only a low correlation was found between the total pool of SVs and PreAZ surface area (Fig. 5E) as well as 260 volume of SBs (Fig. 5G) implying that the total pool of SVs were independent from the size of the SBs. 261 The distribution pattern of SVs made it impossible to morphologically distinguish the three functionally 262 we assumed that the RRP was located at a distance (perimeter p) of ≤10 nm and ≤20 nm from the PreAZ 266 representing docked and primed SVs fused to the PreAZ. The second pool, the RP, is constituted by SVs 267 within 60-200 nm, which maintained release on moderate (physiological) stimulation. The resting pool, 268 consisted in all SVs further than ≥200 nm, preventing depletion upon strong or repetitive stimulations, but 269 which under normal physiological conditions remains unused. 270 Using the same perimeter criteria as at human L5 SBs (Yakoubi et al. 2018), the RRP/AZ was extremely 271 large with an average of 20.20 ± 18.57 at p10 nm and increased by nearly 2.5-fold (48.59 ± 39.02) at p20 272 nm. However, both pools were characterized by a large variability as indicated by the SD, CV and variance 273 (Table 2) suggesting differences in Pr, synaptic efficacy, strength and paired-pulse behavior at individual 274 SBs. Interestingly, no correlation was found for the p10 nm and p20 nm RRP with the surface area of PreAZs 275 However, no correlation was found between the RRP at p10 nm (Fig. 6A) and p20 nm (Fig. 6B) and the 280 total pool of SVs minus the RRP at p10 nm and p20 nm, respectively. A weak correlation was observed for 281 the RP (p60-p200 nm) and the total pool of SVs ( Fig. 6C-E). Finally, no correlation existed for the resting 282 (p500 nm) and the total pool of SVs (Fig. 6F). 283 Taken together, although small in surface area and volume, SBs in L4 of the TLN have strikingly large 284 RRPs, RPs and resting pools when compared with CNS synapses of comparable size or even much larger 285 terminals (see Discussion), but all pools were characterized by a huge variability and were not correlated 286 and hence independent from the SB and PreAZ size. 287

EM tomography of L4 excitatory SBs in the human TLN 289
High-resolution EM tomography was carried out on a sample of small to large SBs with different AZ sizes 290 to look for the organization of SVs, in particular those of the RRP and to test the hypothesis that larger 291 PreAZs display more primed or fused SVs than smaller ones. vs. p10 nm (blue dots) and p20 nm (red dots) RRPs, respectively. *Data points were fitted by linear 301 regression and the R 2 is given for each correlation. total pool of SVs vs. the RP at p60 nm. D, The total pool of SVs vs. the RP at p100 nm. E, The total pool 306 of SVs vs. the RP at p200 nm. F, The total pool of SVs vs. the resting pool at p500 nm.

308
The results of the tomography were three-fold: First, all SBs analyzed, regardless of their target structures, (20.20 ± 18.58 SVs; see also Table 1). Secondly, there was a tendency that larger PreAZs contained more 314 'docked' vesicles ( Fig. 7A, B); providing a larger 'docking' area allowing the recruitment of more SVs. 315 However, in a few cases also SBs with a smaller PreAZ were found that had the same number of 'docked' 316 vesicles (Fig. 7C, E). Finally, so-called MDVs and clathrin-coated pits were clearly identified at several 317 group (red cluster) had large AZs with 0.14 ± 0.08 µm 2 for both PreAZs and PSDs surface area; whereas 337 the SBs in the larger group (blue cluster) had smaller AZs (0.12 ± 0.06 µm 2 ) surface areas. 338 The clustering according to SV pools led also to two groups of SBs, namely SBs (blue cluster) with 339 25.38 ± 21.09 SVs at p10 nm; 427.29 ± 288.94 SVs in the RP and 995.88 ± 696.49 SVs in the resting pool, 340 respectively. SBs belonging to the red cluster with a smaller pool size compared to the first group: 18.69 ± 341 17.58 SVs at p10 nm; 368.85 ± 234.72 SVs at RP and 814.45 ± 484.73 SVs at the resting pool respectively. 342 Although two different clusters existed that further helped to identify subclasses of SBs according the 343 structural parameters, excitatory L4 SBs in the TLN were relatively similar (Fig. 8C). 344 Thus, the CA revealed that the AZs and pools of SVs were the structural parameters that best 345 characterized the SBs in L4 of the human TLN, clustering them into two major groups accordingly. 346

Movie 1: EM tomography of L4 SBs in the human TLN
Two SBs terminating on dendritic spines, both with a large nonperforated AZ occupying half of the pre-and postsynaptic apposition zone with a relatively large pool of SVs, one of which is clustered around the AZ (lower right corner). Note, also the SB establishing a synaptic contact with a dendritic shaft (upper left corner). All three SBs contain a single large mitochondrion. Scale bar 0.25 µm. The present study is the first comprehensive and coherent structural study of L4 excitatory SBs in the 377 temporo-basal human TLN using high-end fine-scale EM and tomography. Although SBs in any given 378 region of the brain are composed of nearly the same structural subelements, it is their individual and specific 379 composition that makes them unique entities, perfectly adapted to their function in the microcircuit in which

Relevance and implications of the density of synaptic contacts measurements 391
Synaptic density measurement can be a useful tool to not only describe the synaptic organization of a 392 particular area, nuclei and even layers in different brain regions, but also the degree of connectivity 393 underlying the computational properties of a given brain area. 394 Meanwhile several studies in various animal species and brain regions included such an analysis (for . Strikingly, a huge difference in the mean density of synaptic contacts was found between our 401 study (2.37*10 6 ± 2.19*10 6 ) and the existing data ranging from 9.13 ± 0.63*10 8 (Alonso-Nanclares et al. this study. It has to be noted that a large proportion of the AZs on spines nearly occupied two-third or even 438 the entire pre-and postsynaptic apposition zone, suggesting that excitatory synaptic transmission is highly 439 efficient at these structures by increasing the docking area for primed and 'docked' SVs which is further 440 supported by our EM tomography experiments (see Fig.7). Interestingly, only a weak correlation between 441 the PreAZ surface area with that of the bouton was found in human and cortical SBs in rat (Rollenhagen et  Only a weak correlation was found between the surface area of the PreAZ and the total pool of SVs, with 446 no correlation for the p10 nm and p20 nm RRP, respectively. A slight tendency was found that SBs with 447 larger PreAZs contained more 'docked' vesicles which is in contrast to our quantitative analysis concerning The already large size of the total pool of SVs in L4 SBs in the human TLN also predict comparably 472 large RRPs and RPs. Indeed, the putative RRP was on average 20.20 ± 18.57 (p10 nm) and doubled to 48.59 473 ± 39.02 (p20 nm) SVs/AZ, ~5-fold larger than those in rats (p10 nm 3.9 ± 3.4, p20 nm 11.56 ± 4.2; 474 Rollenhagen et al. 2018), and 8-10-fold larger than that in L4 SBs in rats (p10 nm 2.0 ± 2.6, p20 nm 6.3 ± 475 6.4; Rollenhagen et al. 2015), respectively. Comparison with even larger CNS synaptic terminals revealed 476 a more than 12-fold and 8-fold difference (hippocampal MFBs p10 nm 1.6 ± 1.5, p20 nm 6.2 ± 4.1; This notion is even more supported by the size of the putative RP/AZ which was ~380 SVs at human L4 485 SBs, ~130 and~200 SVs in rat L4 and L5 SBs, respectively, ~3700 SVs for adult rat MFBs, but nearly 4-486 fold larger than that reported for the rat Calyx of Held (~60 vesicles). Finally, the resting pool of SVs is 487 large enough to rapidly replenish the RRP and RP and thus guarantee only a partial depletion even at 488 repetitive high-frequency stimulation. Finally, rat L4 spiny neurons are highly interconnected with ~200 other excitatory spiny neurons within 562 a 'barrel' column: in turn ~300-400 L4 spiny neurons converge onto a single L2/3 and L5 pyramidal neuron 563 (Lübke et al. 2003). This intracolumnar connectivity is not only a major determinant for reliable signal 564 transduction in L4, acting as 'feed-back amplifiers' even for weak signals from the sensory periphery, but 565 also for the safe and reliable distribution of signals to other neurons located in different layers within the 566 cortical column with which L4 neurons are interconnected. 567 Assuming such a scenario described above exists in L4 of the human TLN, several structural parameters 568 contribute to its function as an important associational area involved in the induction, maintenance and 569 regulation of various computations underlying perception, executive control, learning and memory in which 570 the TLN plays an important role. Hence several structural subelements may contribute to reliable signal 571 transduction: The shape and size of AZs and the large number of SVs in the RRP and RP implying a high 572 P r underlying high synaptic efficacy and strength that contribute to feed-back amplification of even weak 573 sensory signals, and in addition may also enhance TL intracortical information processing. 574 The astrocytic coverage preventing glutamate spillover further guarantees a direct control and sharpening 575 of the transmitted signals. On the other hand, the large variability in AZ size and the three pools of SVs may 576 be involved in the sorting, modulation, and further discrimination of intrinsic and extrinsic signals by 577 neurons in the TLN. Together, all these characteristics ensure the proper wiring and firing of neurons in L4 578 of the human TLN, to accomplish its function as an input-recipient layer and help to explain information 579 processing from incoming signals of the sensory periphery, within the TLN and from brain regions with 580 which the TLN is interconnected. 581 582

Material and Methods 583
Human neocortical tissue processing for EM 584 Biopsy material was obtained from three male and three female patients (25-63 years in age, see Table 3 During surgery, blocks of neocortical access tissue from the temporo-basal regions of the inferior temporal 600 gyrus (Fig. 10) were taken far from the epileptic focus and may thus be regarded as non-affected (non-601 epileptic) tissue as routinely monitored by preoperative electrophysiology and magnetic resonance imaging 602 (MRI). Other evidence that confirms the 'normality' of biopsies and rules out the effect of disease and 603 treatment is the homogeneity of synaptic parameters analyzed among patients as shown by the boxplots 604 (Fig. 11). This has also been demonstrated by other recent structural and functional studies using the same 605  Hu_110204_VII  Female  36  4  GGL  LTG, LEV  53  Hu_110520_V  Female  25  12  AHS  LTG  25  Hu_160217_III  Female  25  23  GGL  Zebinix, LEV  25  Hu_161118_IV  Male  33  5  Gliosis  LEV, CBZ  25  Hu_170801_I  Male  63  24  AHS  LEV, LTG, CBZ  22  Hu_181120  Male  49  36 AHS Vimpat, ZNS / buffered sucrose (300 mOsm, pH 7.4) at room temperature in the dark. After visual inspection and thorough 613 washing in PB they were dehydrated in a series of ethanol starting at 20% to absolute ethanol followed by 614 a brief incubation in propylene oxide (twice 2 min; Fluka, Neu-Ulm, Germany). Sections were then 615 transferred into a mixture of propylene oxide and Durcupan TM resin (2:1, 1:1 for 1hr each; Fluka, Neu-Ulm, 616 Germany) and stored overnight in pure resin. The next day, sections were flat-embedded on coated glass 617 slides in fresh Durcupan TM , coverslipped and polymerized at 60°C for 2 days. The density of synaptic contacts in a given volume is a valuable parameter to assess the structural and 672 functional changes in the brain, which are linked to the age, pathological or experimental conditions (Rakic 673 et al. 1994;DeFelipe et al. 1999). The density of synaptic contacts was unbiasedly estimated in six patients 674 (see Table 3) using the physical dissector technique (Mayhew 1996 (dendritic spines or shafts). Finally the density of synaptic contacts (Nv) per 1 mm 3 (see Table 1) was 683 calculated using the formula below: 684 where Qdis the number of synaptic contacts per dissector and Vd is the volume of the dissector given by: 686 Number of dissectors x frame area x section thickness. 687 688

3D-volume reconstructions and quantitative analysis of L4 SBs 689
All electron micrographs composing each series were imported, stacked, and aligned in the reconstruction 690 software OpenCAR (Contour Alignment Reconstruction; for details see Sätzler et al. 2002). Synaptic 691 structures of interest were outlined on the outer edge of their membranes throughout the series. 3D-volume 692 reconstructions were then generated and the following structural parameters were analyzed: 1) surface area 693 diameter of clear synaptic and DCVs, and 5) total pool of SVs and the RRP, RP and resting pool. 695 The PreAZ surface area was measured by extraction from that of the presynaptic terminal membrane. 696 The size of the PSD was the perimeter ratio between the outlines of the PSD to that of the synaptic contact. 697 Synaptic cleft width measurement was performed only on synaptic contacts cut perpendicular to the AZ. 698 The distance between the outer edge of the pre-and postsynaptic membranes at the center of the synaptic 699 contact and at the two lateral edges was measured and averaged for each synaptic contact. All SVs were 700 marked throughout each SB and their diameters were individually measured. To determine the distribution 701 profile of the SVs, the minimal distance between each SV membrane to the contour lines of the PreAZ was 702 measured throughout the SB in every single image of the series. Large DCVs were only counted in the 703 image where they appeared largest (for details see Yakoubi et al. 2018). 704 No correction for tissue shrinkage was performed. Recently it has been shown by high pressure freezing, 705 cryo-substitution and subsequent EM that no significant differences in quantitative parameters of synaptic 706 structures as measured here were found when compared with conventionally embedded electron 707 microscopic material (Korogod et al. 2015). 708 709

Focused ion beam scanning electron microscopy 710
In this study FIB-SEM was used on L4 of the human TLN to investigate the dynamic changes of the neuropil 711 through a large z-dimension (Movie 2). 712 Immediately after explantation, one additional neocortical access tissue sample from a female patient 713 (63 years in age) of the Gyrus temporalis inferior was immersion-fixed in an ice-cold mixture of phosphate-714 buffered 4% PFA and 2.5% GA for 4hr. Subsequently, the samples were post-fixed overnight in 0.15M 715 cacodylate buffer (CB) + 2% PFA, 2.5% GA and 2mM CaCl 2 before they were embedded in 4% Agar-Agar 716 dissolved in water. After removing access Agar-Agar, vibratome sections of 150 µm thickness were cut 717 (VT1000S, Leica Microsystems GmbH, Wetzlar, Germany) in the frontal (coronal) plane through the human 718 TLN. Sections were collected in multi-well plates in 0.3M CB + 4 mM CaCl2 and thoroughly washed (5x 3 719 min) with 0.15M CB + 2mM CaCl2. Thereafter, sections were incubated in 0.15M CB + 1.5% potassium 720 hexocyanoferrate (II), 2% osmium tetroxide and 2mM CaCl2 for 1hr on ice, in the dark. After washing (5x 721 3 min) with deionized water ("MilliQ", Merck Millipore, Burlington, Massachusetts, USA), sections were 722 placed in an aqueous 1% thiocarbohydrazide solution for 20 min followed by another washing step with 723 deionized water (5x 3 min). This was followed by another treatment with an aqueous 2% osmium tetroxide 724 solution for additional 30 min at room temperature, in the dark and washing with deionized water (5x 3 725 min). Block contrasting was conducted with a filtered, aqueous 1% uranyl acetate solution, overnight at 726 4°C, in the dark. On the next day, samples were washed with deionized water (5x 3 min) and stained with 727 lead aspartate (20mmol lead nitrate in a 30mmol L-aspartic acid solution, pH 5.5) for 30 min at 60°C. After 728 thorough washing with deionized water (3x 5 min), sections were dehydrated through an ascending series 729 of ice-cold, aqueous ethanol dilutions (30%, 50%, 70%, 90%, 100%, each 5 min, 2x 100%, anhydrous, each 730 10 min) before they were transferred into propylene oxide (2x 10 min). Finally, the samples were infiltrated 731 with an ascending series of Durcupan ACM (Sigma-Aldrich) in propylene oxide (2:1; 3:1, each for 1hr and 732 pure Durcupan ACM, overnight) before the sections were flat-embedded between 2 overhead projector foils, 733 which in turn were placed between 2 microscopic glass slides and polymerized at 60°C for two days. 734 For the quantitative analysis of L4 synaptic boutons 3D-volume reconstructions were made based on z-735 stacks obtained using focused ion beam (FIB) scanning electron microscopy (SEM). Based on the overall 736 appearance of the sample, an area of interest was trimmed out of a flat-embedded section, using a 4 mm 737 biopsy puncher, which was then glued onto a pre-polymerized resin block. Excess resin was removed around 738 the tissue using a histology diamond knife on an ultramicrotome (UC7, Leica Microsystems GmbH). The 739 tissue sample was removed from the resin block with a razor blade and was then glued onto a SEM 740 aluminum specimen stub using colloidal silver paste. The sample was dried in a vacuum chamber overnight, 741 then sputter-coated with platinum/palladium for 15 s and finally placed into the FIB-SEM (Crossbeam© 742 540, Carl Zeiss, Oberkochen, Germany) for 3D analysis. 743 A trench was milled with the FIB at 30 kV/30 nA, polishing of the surface was performed at 30kV/3nA 744 and fine milling for data acquisition was performed at 30kV/7nA. The cross-section surface was imaged 745 with an electron energy of 2keV and an electron beam current of 500pA using an in-column energy-selective 746 backscatter electron detector. The dwell time was 10 µs with line average 1. The pixel size in the XY-plane 747 was 10 nm and the slice thickness (Z-direction) was 50 nm yielding a voxel size of 10 nm x 10 nm x 50 nm. 748 The image acquisition software Atlas 3D (Ver. 5.2.0.125, ZEISS, Oberkochen, Germany) allowed the 749 automated collection of 3D SEM datasets using automated correction algorithms for drift, focus and 750 astigmatism (Movie 2). The advantage in using the FIB-SEM technique is three-fold: 1) a much higher 751 throughput of different tissue samples at once; 2) definition of a much a larger region of interest per sample 752 and 3) increase of the z-dimensions of the individual samples. However, the disadvantage of this method is 753 still the weaker resolution of single SVs compared to TEM. This approach, together with TEM, will be used 754 in future studies for further image processing, 3D-volume reconstructions and subsequent data analysis. 755 756 757

EM tomography of L4 SBs in the TLN 758
EM tomography was carried out on 200-300 nm thick sections cut from blocks prepared for ultrathin 759 sectioning as described above. Sections were mounted on either pioloform-coated line or slot copper grids 760 (Plano, Wetzlar, Germany) and were counterstained with uranyl acetate and lead citrate following a slightly 761 modified staining protocol as described by Reynolds (1963). Subsequently, sections were examined with a 762 JEOL JEM 1400Plus, operating at 120 kV and equipped with a 4096x4096 pixels CMOS camera (TemCam-763 F416, TVIPS, Gauting, Germany). Tilt series were acquired automatically over an angular range of -60° to 764 +60° at 1° degree increments using Serial EM (Ver. 3.58; Mastronarde 2005). Stack alignment and 765 reconstruction by filtered backprojection were carried out using the software package iMOD (Ver. 4.9.7; 766 Kremer et al. 1996). Final reconstructions were ultimately filtered using a median filter with a window size 767 of 3 pixels. CA was performed based on the structural parameters investigated (see Table 2), to further identify different 773 groups i.e. types of SBs by running a CA using MATLAB and Statistics Toolbox Release 2016b (The 774 MathWorks, Inc., Natick, MA, USA; for details see Yakoubi et al. 2018). Then a zero-mean normalization 775 was performed as the parameters had different units. This was followed by a PCA to reduce our large dataset 776 to a smaller set of uncorrelated variables called PCs, but still containing most of the information in the 777 original dataset. Subsequently, we performed a HCA on the simplified dataset composed of the PCs, as the 778 original data were not labeled (Fig. 8, see also Yakoubi et al. 2018). 779 780

Statistical analysis 781
The mean value ± SD, the median with the 1 st and 3 rd quartile, the R 2 , the coefficient of variation (CV), 782 skewness and variance were given for each parameter. The p-value was considered significant only if 783 p<0.05. Boxplots were generated to investigate inter-individual differences for each structural parameter 784 (Fig. 11). The non-parametric Kruskal-Wallis H-test analysis was computed, using InStat (GraphPad 785 Software Inc., San Diego, CA, USA), as some of the analyzed parameters were not normally distributed as 786 indicated by the skewness. Correlation graphs between several structural parameters were generated.