Constitutive tumor necrosis factor (TNF)‐deficiency causes a reduction in spine density in mouse dentate granule cells accompanied by homeostatic adaptations of spine head size

The majority of excitatory synapses terminating on cortical neurons are found on dendritic spines. The geometry of spines, in particular the size of the spine head, tightly correlates with the strength of the excitatory synapse formed with the spine. Under conditions of synaptic plasticity, spine geometry may change, reflecting functional adaptations. Since the cytokine tumor necrosis factor (TNF) has been shown to influence synaptic transmission as well as Hebbian and homeostatic forms of synaptic plasticity, we speculated that TNF‐deficiency may cause concomitant structural changes at the level of dendritic spines. To address this question, we analyzed spine density and spine head area of Alexa568‐filled granule cells in the dentate gyrus of adult C57BL/6J and TNF‐deficient (TNF‐KO) mice. Tissue sections were double‐stained for the actin‐modulating and plasticity‐related protein synaptopodin (SP), a molecular marker for strong and stable spines. Dendritic segments of TNF‐deficient granule cells exhibited ∼20% fewer spines in the outer molecular layer of the dentate gyrus compared to controls, indicating a reduced afferent innervation. Of note, these segments also had larger spines containing larger SP‐clusters. This pattern of changes is strikingly similar to the one seen after denervation‐associated spine loss following experimental entorhinal denervation of granule cells: Denervated granule cells increase the SP‐content and strength of their remaining spines to homeostatically compensate for those that were lost. Our data suggest a similar compensatory mechanism in TNF‐deficient granule cells in response to a reduction in their afferent innervation.


INTRODUCTION
Dendritic spines are highly motile protrusions on dendrites of many neurons in the central nervous system (Dunaevsky et al., 2001) forming axo-spinous synapses with glutamatergic afferents (DeFelipe et al., 1988). They are basic functional units of signal integration detecting the coincidence of pre-and postsynaptic activity (Yuste & Denk, 1995). Of note, spines are modified by synaptic activity and the geometry of spines has been tightly linked to the functional properties of their synapses (Bonhoeffer & Yuste, 2002;Kasai et al., 2003;Matsuzaki et al., 2004). In particular, spine head size is associated with synaptic strength (Fifková & van Harreveld, 1977;Kasai et al., 2010;McKinney, 2010), postsynaptic density area (Harris et al., 1992), and AMPA-receptor density (Matsuzaki et al., 2001(Matsuzaki et al., , 2004Noguchi et al., 2011;Zito et al., 2009), thus making spine head size a structural surrogate marker -or at least an indicator -for local synaptic activity.
Tumor necrosis factor (TNF) is a pleiotropic cytokine that has been implicated in a wide range of physiological, that is, synaptic plasticity (Heir & Stellwagen, 2020;Santello & Volterra, 2012) and pathophysiological processes, that is, inflammation (Locksley et al., 2001;Santello & Volterra, 2012). In the central nervous system, TNF has also been shown to affect synaptic function with its net effect depending strongly on its concentration (Heir & Stellwagen, 2020;Maggio & Vlachos, 2018;Santello & Volterra, 2012). At low concentrations, that is, in a physiologic range, TNF increases synaptic strength (Stellwagen & Malenka, 2006;Stellwagen et al., 2005). It is also permissive for different forms of homeostatic synaptic plasticity Steinmetz & Turrigiano, 2010;Stellwagen & Malenka, 2006) and it has recently been shown to promote the ability of neurons to express LTP (long-term potentiation; Maggio & Vlachos, 2018). Since TNF also reduces GABAergic transmission by promoting GABAa-receptor endocytosis (Stellwagen et al., 2005), TNF has been suggested to play a role in the fine tuning of excitation/inhibition of neuronal networks (Santello & Volterra, 2012;Stellwagen et al., 2005). These physiological functions of TNF, which require only basal concentrations of the cytokine are considered largely distinct from TNF actions in pathophysiological contexts (Heir & Stellwagen, 2020;Santello & Volterra, 2012).
Since the number and geometry of spines are tightly linked to synaptic function, and since TNF has been linked to synaptic plasticity and SP, we wondered whether lack of TNF could result in changes of spines and/or SP within spines, which would be indicative of an altered network function. Our data show that constitutive TNF-deficiency does indeed cause structural changes of dendrites of dentate granule cells and suggest that these structural changes reflect a compensatory homeostatic response: TNF-deficient granule cells may compensate a reduced number of dendritic spines, that is, fewer afferent synapses, by homeostatically increasing the strength of some of their remaining spines, that is remaining synapses. Every effort was made to minimize distress and pain of animals.

Intracellular injections of granule cells in fixed tissue
After delivery, animals were kept in an in-house scantainer for a minimum of 24 h. Animals were killed with an overdose of intraperitoneal pentobarbital and subsequently intracardially perfused (0.1 M Phosphate buffer saline (PBS) containing 4% paraformaldehyde (PFA)). Tail biopsies were obtained after death to reconfirm the genotype. Brains were taken out immediately after perfusion, postfixed (18 h, 4% PFA in 0.1 M PBS, 4 • C), washed trice in ice-cold 0.1 M PBS, sectioned (250 μm) on a vibratome (Leica VT 1000 S), and stored at 4 • C until use. Intracellular injections of granule cells in fixed slices were performed as previously described (Arends & Jacquin, 1993;Hick et al., 2015;Yap et al., 2020), with modifications. Hippocampal slices were placed in a custom-built, transparent, and grounded recording chamber filled with ice-cold 0.1 M PBS. The chamber was attached to an epifluorescence microscope (Olympus BX51WI; 10× objective LMPlanFLN10×, NA 0.25, WD 21 mm) mounted on an x-y translation table (Science Products, VT-1 xy Microscope Translator). Sharp quartz-glass microelectrodes (Sutter Instruments, QF100-70-10, with filament) were pulled F I G U R E 1 Age distribution of TNF-WT and TNF-KO mice. Mice used in the present study were age-matched. There were no significant differences between the age distribution of the two groups. n.s. p = 0.144. Brunner-Munzel U-test. TNF-WT, n = 7; TNF-KO, n = 7 using a P-2000 laser puller (Sutter Instruments). Microelectrodes were tip-loaded with 0.75 mM Alexa568-Hydrazide (Invitrogen) in HPLCgrade water (VWR Chemicals, HiPerSolv CHROMANORM) and subsequently back-filled with 0.1 M LiCl in HPLC-grade water. Microelectrodes were attached to an electrophoretic setup via a silver wire and 500 MΩ resistance. The tip of the microelectrode was navigated into the granule cell layer using a micromanipulator (Märzhäuser Wetzlar, Manipulator DC-3K). A square-wave voltage (1 mV, 1 Hz) was applied using a voltage generator (Gwinstek SFG-2102). Granule cells were filled under visual control for at least 10 min or until no further labeling was observed. (Figures 2a and 2b). Injected sections were fixed overnight (4% PFA in PBS, 4 • C, in darkness) and washed in 0.1 M PBS.

Confocal microscopy of fixed hippocampal slices
Confocal imaging of fixed dendritic segments from identified, Alexa568-labeled dentate granule cells in the outer molecular layer (OML) of the suprapyramidal blade was done with an Olympus FV1000 microscope and a 60× oil-immersion objective (UPlanSApo, NA 1.35, Olympus) using FV10-ASW software with 5× scan zoom at a resolution of 1024 × 1024 pixels. To ensure localization of dendritic segments in the OML the hippocampal fissure was used for orientation. Dendritic segments located at a distance of 10-20 μm from the hippocampal fissure were identified and traced for a maximal length of 50 μm. Only segments within the OML were used for analysis and only one dendritic segment was used per labeled cell. Three-dimensional image stacks of such dendritic segments (0.15 μm z-axis step size) were F I G U R E 3 Quantification of dendritic spine head area and synaptopodin (SP) cluster area. (a) Maximum projection of a confocal image stack showing a dendritic segment of a granule cell. For analysis, dendritic spines of all shapes were assessed manually on z-stacks of dendritic segments. Only protrusions extending laterally in the x-y plane, not above or below the dendrite, and exceeding the dendrite for at least 5 pixels (0.2 μm) were included for analysis. For spine head area the maximum cross-sectional area of the spine head was acquired. Crossing dendritic segments or branch points were avoided to facilitate spine attribution to a given segment. For imaging of dendritic segments, imaging parameters were set to capture the dendritic segment as bright as possible without oversaturating the spines. For imaging of SP, all imaging parameters were kept the same for all images.

2.5
Image processing and data analysis Images obtained were deconvolved with Huygens Professional Version 17.10 (Scientific Volume Imaging, The Netherlands, http://svi.nl).
Image processing and data analysis were then performed using Fiji version 1.52h (Schindelin et al., 2012). Spines were identified and analyzed using established criteria (Holtmaat et al., 2009). Prior to quantification, all images were randomized. Images were renamed by one experimenter (M. Rietsche) and subsequently analyzed by a second experimenter (D. Smilovic) blind to genotypes. Dendritic spines of all shapes were assessed manually on z-stacks of dendritic segments in the OML. Only protrusions emanating laterally in the x-y directions, not above or below the dendrite, and exceeding the dendrite for at least 5 pixels (0.2 μm) were included for analysis (Holtmaat et al., 2009;Vlachos, Müller-Dahlhaus, et al., 2012;Yap et al., 2020). The length of each segment was determined ( Figure 3a). For spine head area and SP cluster area measurements, the largest maximum cross-sectional area of the spine head or SP cluster in one of the x-y planes within the z-stack was manually measured using a predefined gray-value as a cutoff for the border of the spine head or SP cluster ( Figure 3b). A spine was considered SP+ if the SP cluster overlapped with the spine head, neck, and/or base in both the x-z and y-z directions when scrolling through the z-stacks ( Figure 3c). The subcellular location of SP clusters in the spine head, spine neck, spine base, or dendritic segment was noted.
SP clusters were considered within the spine head, if most (> 80%) of the SP-cluster was located within the identified area of the spine head (see Figure 3c). SP-clusters were considered within the spine neck, if most (> 80%) of the cluster was located outside the dendritic shaft border, between the identified area of the spine head and the shaft, where a fluorescently filled, visible, spine neck was marked. SP clusters were considered associated with the spine base, if they were found within 0.2 μm of the intersection between the dendritic spine and the dendritic shaft border. SP clusters were considered inside dendritic shafts if they did not meet any of the aforementioned criteria but were still localized within the investigated dendritic segment.
identified in one of the x-y planes and measured. Scale bar = 5 μm.
(b) Only spines with heads protruding at least 0.2 μm from the parent dendrite (parallel white lines) were analyzed. Spines co-localizing with an SP cluster in the x-y, x-z, and y-z directions when scrolling through the z-stack were considered to be SP+ (arrow). Scale bar = 0.5 μm. (c) Spine head area and SP cluster area were defined as the largest x-y cross-sectional area obtained in a z-stack. Images containing the largest area of spine head (middle column, orange outline of spine head, asterisk) and SP cluster (right column, orange outline of SP-cluster, asterisk) are highlighted. Scale bar = 0.5 μm

Statistical analysis
Statistical tests and n-values used for testing are indicated in the figure captions. Statistical analysis was performed using R 4.0.4 (R Core Team, 2013) called via R-script from LabVIEW scripts (National Instruments, Austin, Texas). Robust methods were applied throughout as specified for the different conditions: (1) Comparison of magnitude between two groups: Brunner-Munzel U-test (function "brunnermunzel.permutation.test" from R-library "brunnermunzel") (Brunner & Munzel, 2000), (2)  tion between two attributes: Spearman's rho (R-function "cor" employing method "spearman"), and (7) robust linear regression between two attributes: R-function "rlm" employing method "MM" from library "MASS". Confidence intervals for Spearman's rho and for regression slope and intercept were obtained using the resampling function "bootstrap" from R-library "bootstrap." In case of multiple post hoc tests pvalues were adjusted using the function "p.adjust" employing method "hochberg". If p values were less than 0.05, the null-hypothesis was rejected (*p < 0.05, **p < 0.01, ***p < 0.001). Quantitative data are displayed either in box plots (box encloses 25-75% quartiles, dividing line in box represents median, x labels mean, whiskers represent maximum / minimum or median ± 1.5 times the distance between the 25% and 75% quartiles in case of presence of extreme values, which are shown as additional dots) or as bar plots (mean + SEM) including the individual data points as dots. Diagrams were created using Microsoft Office Excel.

Granule cell dendrites of TNF-KO mice exhibit a reduced spine density
Previous work showed that TNF is an important factor in the control of synaptic strength (Beattie et al., 2002;Santello et al., 2011;Stellwagen et al., 2005;Stück et al., 2012). Since synaptic strength and spine geometry are tightly linked (Bonhoeffer & Yuste, 2002;Fifková & van Harreveld, 1977;Kasai et al., 2003;Matsuzaki et al., 2004), we speculated that genetic knockout of TNF in vivo may have a structural correlate at the level of spines. To address this question, we first studied dendritic segments of Alexa568-filled granule cells (Figure 4a-c) in the outer molecular layer of the DG of TNF-deficient and age-matched C57BL/6J control mice. TNF-KO mice exhibited a significant reduction in spine density (Figure 4d): whereas wildtype mice had 2.03 spines/μm, TNF-KO mice had 1.62 spines/μm, that is, ∼20% fewer spines. Next, we analyzed spine head area, since spine head area correlates well with synaptic strength and the density of AMPA-Rs.
Average spine head area was not significantly different between genotypes (Figure 4e), although a trend towards higher values was seen in TNF-KO segments. Finally, we calculated total spine head area per segment (Figure 4f), which illustrates how changes in spine density and head area affect the available spine head area for neurotransmission.
This parameter takes the number of spines into account and shows that the total spine head area per segment decreases in the TNF-KO mice. After analyzing dendritic segments, we shifted our attention to the entire population of spines. The trend seen in Figure 4e was confirmed to be significant (Figure 4g). The cumulative distribution revealed a highly significant difference between the two genotypes, with differences most prominent at the beginning of the curve, that is, small spine heads, and at the end of the curve, that is, large spine heads ( Figure 4h).

TNF-KO mice show an increase in the fraction and size of large spines
To investigate this further, we distinguished three categories of spines mice had more small (∼54% compared to ∼49%) and large (∼17% compared to ∼13%) sized spines than controls, whereas TNF-WT granule cells had more medium sized spines (~38% compared to~30%) compared to TNF-deficient cells.

Large SP ± spines are selectively enlarged in TNF-KO mice
We now divided SP+ and SP-spines into the three size categories (c.f. Figures 5a and 5b). Most SP+ spines were found belonging to  Figure 7a). SP+ spines belonging to the small category did not differ significantly between TNF-deficient and TNF-WT granule cells (Figure 7b; n.s. p = 0.593). In contrast, SP-spines showed a significant reduction of ∼10% in spine head area in TNF-KO mice (Figure 7b; **p = 0.003). There was no significant difference between genotypes for spines belonging to the medium sized subgroup (n.s. p = 0.862 for both, SP+ and SP-spines; Figure 7b). We conclude from these findings, that (i) the increase in spine head area of large spines observed in TNF-KO granule cells (Figure 5b) is the result of an enlargement of large SP+ spines (Figure 7b), and, (ii) the reduction in spine head area of small spines (Figure 5b) is the result of a diminution of small SP-spines ( Figure 7b).

SP cluster size is increased in spines of TNF-deficient granule cells
The fact that SP+ spines of TNF-deficient granule cells have larger heads made us wonder whether this increase is matched by a corresponding increase in SP clusters, since these two parameters are highly correlated (Lenz et al., 2021;Yap et al., 2020). Indeed, average SP cluster areas were ∼25% bigger in TNF-deficient granule cell segments (Figure 8a). Similarly, the cumulative distribution of SP cluster areas was right-shifted in the mouse mutant compared to control (Figure 8b). SP-clusters were preferentially found in the spine head of both genotypes, without a significant shift in proportions between genotypes (Figure 8c, d). Finally, we analyzed the relationship between SP-cluster area and spine head area. Both genotypes showed a strong positive correlation between the two parameters ( Figure 8e).
Linear regression revealed a non-significant trend for TNF-mutants to have relatively smaller SP-clusters when comparing equally sized spine heads.

DISCUSSION
In the present study, we analyzed the effects of constitutive TNFdeficiency on the structure of dendritic spines. The rationale behind this question were earlier data implicating TNF in the modulation of synaptic transmission and the fine-tuning of excitation/inhibition balance of synaptic networks (Heir & Stellwagen, 2020;Santello & Volterra, 2012;Stellwagen et al., 2005). Although basal levels of TNF are low, complete absence of TNF may impair these physiological functions in synaptic transmission and may cause changes in network activity accompanied by structural alterations. Since some TNF-effects may require SP/spine apparatus (Maggio & Vlachos, 2018), we also investigated SP in spines of TNF-deficient neurons. The main findings of our study can be summarized as follows: (1) Granule cells of TNF-deficient mice have ∼20% fewer spines than wildtype controls.
(2) TNF-deficient mice have an altered distribution of spine head sizes: they show higher fractions of large and of small spines.
(3) Although the fraction of SP+ spines was comparable between genotypes, TNF-deficient mice exhibited larger SP+ spines with larger SP clusters. (4) Small SP-spines were smaller in TNF-deficient mice compared to controls. This pattern of changes suggests that granule cells of TNF-deficient mice have fewer afferent synapses and that the reduced number of synapses is homeostatically compensated for by an increase in head size and SP cluster size of the remaining spines (Figure 9).

Structural alterations of spines in TNF-deficient mice are similar to changes observed after entorhinal denervation
This pattern of changes, that is, fewer spines and larger SP+ spines, is very similar to homeostatic changes observed after entorhinal denervation of granule cells. In this experimental denervation model, granule cell spine density is significantly reduced following entorhinal deafferentation (Caceres & Steward, 1983;Vlachos, Becker, et al., 2012;Vuksic et al., 2011) and the loss of spines is homeostatically compensated for by an increase in synaptic strength and an increase in SPcluster size of the remaining spines (Vlachos, Becker, et al., 2012;Vlachos, Ikenberg, et al., 2013). Such a homeostatic response may help to keep denervated neurons within their physiological firing range (Platschek et al., 2016) and may promote information flow through a partially denervated brain area (Deller & Frotscher, 1997;Steward, 1994).
In the present study focusing on TNF-deficient granule cells we recognized a comparable situation, since the reduced density of spines of TNF-deficient granule dendrites in the outer molecular layer of the dentate gyrus is indicative of a reduced excitatory innervation from the entorhinal cortex. How this reduced innervation comes about and which developmental processes may underlie this change in this constitutive knock-out mouse model is currently unknown and a limitation of our study. Although a developmental role of TNF or even a retraction of some spines into the shaft cannot be fully excluded, we consider it more likely that alterations in the balance of network excitation/inhibition could have caused secondary changes in the density of dendritic spines.
This interpretation is in line with studies linking spinogenesis and spine density to afferent activity (Drakew et al., 1996;Engert & Bonhoeffer, 1999;Jourdain et al., 2003;Knott et al., 2006;Maletic-Savatic et al., 1999;Nägerl et al., 2004;Segal et al., 2003). Regardless of the cause, as (i) most spines of adult neurons are innervated (Knott et al., 2006), and, (ii) excitatory input on adult spiny neurons terminates almost exclusively on spines (Mates & Lund, 1983), a reduced density of spines is a bona fide structural indicator of a reduced glutamatergic innervation of granule cells in the dentate gyrus of TNF-deficient mice.
In addition to a reduced spine density, TNF-deficient neurons show an increase in the size of large spines. These spines are characterized by large PSDs (Harris et al., 1992), a high density of AMPA-R (Béïque et al., 2006;Matsuzaki et al., 2001Matsuzaki et al., , 2004Noguchi et al., 2011;Zito et al., 2009), and (this study) by large SP-clusters. Thus, they are strong spines, contributing much to the excitatory drive of a neuron. In contrast, many small spines lack AMPA-R and may represent "silent spines" (Kerchner & Nicoll, 2008). These spines are weak spines, contributing little to the excitatory drive. Since both parameters, that is, spine head size (Matsuzaki et al., 2001;Noguchi et al., 2011;Zito et al., 2009) as well as the presence of SP-clusters (Vlachos et al., 2009) F I G U R E 9 Summary diagram illustrating the structural differences between dendritic segments of TNF-WT and TNF-KO granule cells (a, b) TNF-KO dendrites exhibit ∼20% fewer spines compared to controls. Although average spine head area is not significantly different, the size distribution of spines has changed: Compared to controls, the fraction of large spines as well as the average size of large spines is increased in the KO. The fraction of small spines is also increased, but this group of spines showed a decreased average head size. Thus, the KO exhibits relatively more large spines, which are also larger and at the same time more small spines, which are also smaller. These bidirectional changes explain why average spine head size between genotypes is not significantly different. Furthermore, large spines were found to be associated with the plasticity-related protein Synaptopodin (green clusters). The larger spines of TNF-KO mice also exhibited larger SP clusters

TNF and SP/spine apparatus are linked
TNF and SP have been previously linked in the context of inflammation (Strehl et al., 2014) and Hebbian plasticity (Maggio & Vlachos, 2018).
In the first context, that is, inflammation and high levels of TNF, SP was found to be negatively regulated, whereas in the second context, that is, Hebbian plasticity, TNF required SP for its plasticity-promoting effect. This complex relationship may be explained by the divergent effects of TNF at different concentrations: high concentrations of TNF are present in the context of inflammation, whereas lower concentrations of TNF modulate synaptic plasticity under physiological conditions (Heir & Stellwagen, 2020;Santello & Volterra, 2012). Thus, the relationship between TNF and SP may depend on the specific condition of an experiment, that is, whether TNF is present in high concentrations, for example, inflammation, slightly elevated concentrations, for example, plasticity conditions, or constitutively, for example, in naive animals.
In this study, we used a constitutive TNF-deficient mouse and found that TNF-deficiency was associated with larger SP-clusters in spines compared to wildype. A trivial explanation of this result is that TNF and SP are negatively correlated. This interpretation contrasts, however, with earlier observations, in which parallel increases in glial TNF  and SP-cluster size (Vlachos, Ikenberg, et al., 2013) were reported following entorhinal denervation, suggesting that TNF and SP are positively correlated in this condition. We speculate that the compensatory increase in SP cluster size seen in constitutively TNFdeficient mice may not be the result of TNF-deficiency. Rather, other regulators of homeostatic synaptic plasticity, such as retinoic acid (Lenz et al., 2021), which enlarges both spine heads as well as SP clusters, could contribute to these changes. Further in vitro and in vivo experiments are required to resolve the question, whether the increase in SP cluster sizes seen in the present study is the result of TNF-deficiency or the result of denervation-related adaptations.

Competitive interactions between strong and weak spines may underlie the decrease in size of small spines
The increased average size of large spines was accompanied by a reduced average size of small spines. This reduction in size could be the result of competitive interactions between the large and strong SP+ spines and spines in their neighbourhood: Large SP+ spines can compete with newly formed spines via a NMDA-R/CaMKII-dependent mechanism (Vlachos, Helias et al., 2013) and may reduce them in size. This interaction takes the precise temporal correlation of calcium entering the cell into account (Helias et al., 2008). Such a mechanism would be homeostatic in nature, since it keeps the average spine head size constant. Indeed, as was shown recently (Jungenitz et al., 2018), granule cells control spine head size across their dendritic arbour. In this earlier study, high-frequency stimulation of the middle molecular layer of the dentate gyrus caused LTP and an increase in spine head size in the stimulated zone. In the nonstimulated outer molecular layer, however, dendritic segments of the same granule cell exhibited a decrease in spine head size (Jungenitz et al., 2018), indicative of hetereosynaptic long-term depression (LTD) in this layer (Abraham & Bear, 1996;Christie & Abraham, 1992;Jedlicka et al., 2015). Thus, the decrease in the average size of small spines may be secondary to the increase in the strength of large SP+ spines.

4.4
Changes of spine density and size may contribute to learning and memory deficits of TNF-deficient mice The behavior of constitutive TNF-deficient mice has been studied and an impaired learning and memory retention performance were reported (Baune et al., 2008). It is conceivable that the structural changes we report here contribute to such behavioral deficits.
A reduction of granule cell spine density by ∼20% is significant and may be associated with a reduced excitatory drive onto dentate granule cells. Furthermore, the compensatory strengthening of some synapses, while keeping the neuron within a physiologic firing range, could also limit the degrees of freedom a granule cell has for plasticity. Since synaptic information storage capacity has been linked to spine size classes (Bromer et al., 2018), it is conceivable that homostatically enlarged and highly stable SP+ spines are saturated in this respect, that is they cannot be strengthened further and do not contribute to informational changes induced by plasticity. The concomitant reduction of the fraction of medium sized spines, i.e. of those spines able to undergo both structural LTP as well as structural LTD, may further limit the ability of TNF-deficient granule cells to express plasticity and may thus diminish their ability to integrate into neuronal ensembles encoding novel contexts (Abdou et al., 2018).

CONFLICT OF INTEREST
TD received funding from Novartis for a lecture on human brain anatomy. The other authors declare that the research was conducted in the absence of any commercial or financial relationships that could be constructed as a potential conflict of interest.

AUTHORS CONTRIBUTIONS
DS, MR, MV acquired data, DS and AD analyzed data, TD and MV conceived and supervised the study, TD, DS, AD, and MV wrote the manuscript with contributions from all authors. All authors were involved in data interpretation and critically revising the manuscript.
All authors read and approved the final manuscript.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available from the corresponding author upon reasonable request.

PEER REVIEW
The peer review history for this article is available at https://publons. com/publon/10.1002/cne.25237.