ISSN: 2638-9231

Research Article

Annals of Behavioral Neuroscience

Computational RSM Modeling of
Neuromorphofunctional Relations of Dentate
Nuclear Neurons and Dentatostriate Inter-Cluster
Mapping with the Dentatostriate Neural Network
Reconstruction: RLSR/PCR Regression and Canonical
Correlation Analysis
Grbatinić I1*, Krstonošić B2, Marić D2, Purić N3 and Milošević N1
Laboratory for digital image processing and analysis, Institute of Biophysics, University of Belgrade, Serbia
1

2
Institute of Anatomy, University of Novi Sad, Serbia
3
Klinical Center of Serbia, Belgrade, Serbia
*
Correspondence: Ivan Grbatinić, Laboratory for Digital Image Processing and Analysis, Faculty of Medicine,
University of Belgrade, Dr Subotic 6, Belgrade, Serbia, E-mail: [email protected]

Received date: March 12, 2019; Accepted date: April 24, 2019; Published date: April 29, 2019

Abstract
Aim: The aim of this study is to find relational connections (interdependence) between the two most general categorical
aspects of a neuron, i.e., between the form (morphology) and its function, using as a model for this task dentate
nucleus neurons. Furthermore, the configuration of the dentatostriate nucleotopic inter-cluster mapping of the
dentatostriate neural network is investigated in order to determine mutual, inter-neuronal, neuromorphofunctional
remote influence, i.e. the neuromorphofunctional relations at the level of a neural network.
Materials and methods: (Semi) virtual dentate and neostriate adult human neuronal samples were used.
Neuromorphological parameters of each neuron have been directly measured, i.e. experimentally determined,
whereas the corresponding neurofunctional parameters have been theoretically obtained. The neuromorphological
parameters determine the following properties of a neuron: neuron shape, compartmental length and size/
surface, dendritic branching, complexity and organization of neuronal morphology. The group of neurofunctional
parameters determines functional aspects of action potential (AV/AP), as well as neurofunctional properties of the
perikaryodendritic compartment of a neuron. Data analysis is performed using response surface (RSM) modeling,
along with partial least-squares (PLSR) and principal component regression analysis (PCR), accompanied by canonical
and Pearson correlation analysis. A stepwise algorithm formulates the complete data analysis.
Results: Obtained RSM models represent response-predictor relations, where a neuromorphological/functional
response parameter is expressed as a function in terms of parameters of other category (morphology/function).
Additionally, RSM modeling is also used to decipher the symmetry of the dentatostriate inter-cluster neural network
by the corresponding inter-cluster inter-nuclear mapping, using so-called integral parameters/variables, obtained on
a computational, theoretical manner. The obtained network is a fully connected, symmetric, Hopfield neural network.
Conclusion: Neuronal morphology and function are definitely interrelated and depend on each other. By
intensity, however, this interconnectedness can be treated as mild to moderate. It is determined by elementary
neuromorphofunctional relations, observed at the macroscopic, phenomenological level, i.e. only through measured

Annal Behav Neurosci, 2(1): 168-196 (2019) 168

parameters as their observable and explicit manifestation without considering the microscopic, molecular causality
of them. These relations are the strongest when acting upon a single neuron and their mutual remote influence on
each other weakens in neural circuits and networks up to 10% of deterministic relational interconnection strength
observed at the level of single neuron relations.

Keywords: Neuromorphofunctional relations, RSM modeling, Dentatostriate inter-cluster mapping

Introduction a typical representative neuron of the entire caudate

nucleus, i.e., the average type, to say (Figure 2e and f), in
The general question concerning relation(s) between order to show its sharing of common neuromorphological
morphology and function of cells is a very intriguing features with the others, specific and characteristic types
question, however, still not hitherto completely (I and II), where nicely demonstrate itself as a transitional
deciphered in purely experimental systems. This is (intermediary form) type.
particularly the case for neurons, on a first place due
to an extremely wide variation in their morphology Parallel to the question of neuromorphofunctional
and consequently the function [1]. Studies dealing with interrelations at a single-neuron level is also a question
neuromorphofunctional interrelations, aiming to decode of influence of systemic neural network activity upon an
types of parametric mutual interdependence between individual neuron in terms of reshaping its morphology
morphology of a neuron and its function, described at a and function. The DN and neostriatum are interconnected
precise mathematical manner and performed at a single- by numerous bidirectional neural projections of which
neuron level are based predominantly upon data of indirect pathways represent the greatest majority. The
experimental, ionic current measurements, where specific dentato-rubro-thalamic pathway represents the most
neuromorphofunctional interrelations are theoretically important of them. The other pathways go via the inferior
determined or implicitly implied [2-6]. Different types olivary complex, paramedian and pontine nuclei [6,13-
of neurons, classically identified and described by their 15]. Figure 3a and b show some of these interconnections.
unique morphologies, are also only distinguished by their
repertoire of ion channels and their characteristic, output
firing patterns [3]. Experimental and computational
studies have shown that various ion channel distributions
lead to very different, simple to complex firing behaviors
of neurons, thus revealing the exact, mathematical
form of neuromorphofunctional interdependence [2-6].
Parallel to these are computational studies based upon
the cable theory which explain spatial extensions of
cable-like structures and corresponding dependences of
Figure 1: The 3D neuroanatomical reconstruction of
the space-time voltage dynamics [7-9].
dentate nucleus gross morphology.

In our previous study [10], we determined two

According to our and hypothesis by other authors
neuromorphotopologically different clusters (types)
[4-6,16] part of the key unlocking and deciphering the
of dentate nucleus (DN) neurons (Figure 1), namely,
connection between categorical aspects of a neuron
the external (EBC) and internal border neurons (IBC),
(morphology/function) lies in the other one. Using a
differing in terms of both their morphology and
computational approach, we are going to investigate
topology. The topological cluster of central neurons,
these neuromorphofunctional relations for the
neuromorphologically, can be divided into the EBC and
neuronal types of the human DN, separately analyzed.
IBC histotypes [11]. In our other, analogous study, dealing
By starting from the fact of already proven existence of
with the histological organization of the neostriatum, we
connection between neuronal morphology and function,
discovered three histomorphologically different clusters:
we further want to investigate this by answering on
the two caudate clusters (I and II) and putaminal cluster
the following questions: 1) how they are connected,
[12]. However, in this study, we did not determine the
meaning what type of computational connection exists
topology of the caudate clusters. Classical representative
between morphology and function, i.e., what is their
neurons of each of the identified clusters from both of
mutual interdependence, parametrically and separately
the nuclei, are shown in figure 2. We have also included
analyzed, and 2) what is the strength of this connection;
Annal Behav Neurosci, 2(1): 168-196 (2019) 169
giving us the quantification of their dependence on one the way, it should also be noted that, in this study, our aim
side, and of their autonomy on the other. In addition, is not to give a detail description and causal explanation
based on the obtained results, inter-cluster mapping of every obtained model. This would be cumbersome and
and reconstruction of the dentatostriate neural network redundant due to a large number of models and it would
will be performed in order to investigate mutual, inter- be possible only in future studies. Here, to summarize, we
neuronal, neuromorphofunctional remote influence of just list them as enough, self-explanatory for the moment,
the network upon a single neuron. This would also serve in purposes of concretizing the generally known fact of
as a general example of such influence and behavior of structure/function interrelations through given models
cells making connection with the neighborhood. Because as their computational, theoretical representatives.
of highly complex models, due to a large number of
parameters and consequent semi-experimental/semi-
computational approach we intend to use here, rough
polynomial RSM methodology will be applied upon
experimentally collected neuromorphological data in
order to simplify desired relations and their further
analysis and interpretation.

Figure 3: Dentatostriate interconnections.

Materials and Methods

Impregnation procedure and acquisition of a 2D
binary neuron image
Brain-tissue samples used in this study have been
acquired and collected during the period of 2013-2014 at
the Department of Forensic Medicine, Faculty of Medicine
of the University of Novi Sad, Serbia with the approval
of the Ethics Committee of the University of Novi Sad.
The tissue samples of the neostriate and dentate nuclei
were taken from the brains of 30 adult human cadavers,
independently of gender and brain side, with the age
range 32-88 years old, 24 male and 6 female. Criteria for
tissue sampling were the absence of both brain damage
Figure 2: The 2D binary and planar 3D reconstruction and diagnosed cardio/cerebro-vascular diseases.
images of neuron representatives of the dentate
nucleus and neostriate neuromorphological neuron The dentate and neostriate nuclei were cut at 2.5
clusters. mm thick tissue blocks, impregnated after the Kopsch-
Bubenaite method [17], dehydrated in the increasing
In contrast to exact modeling methods, only concentration of alcohol and embedded in paraffin.
numerical, approximate methods, such as RSM, can Then, serial horizontal sections were cut at 90 mm thick
give us the adequate quantification of the strength of slices and mounted on glass slides. In this way, the
neuromorphofunctional relations, given as a model fit. By preparations were prepared for the following selection

Annal Behav Neurosci, 2(1): 168-196 (2019) 170

process. neuronal image. The last and less significant criterion,
used for parameter classification, is defined according
Each preparation sample is observed under a light to whether a parameter value is directly obtained by
microscope at the magnification of 400x. The only measurement analysis, performed upon a neuronal
criterion for sample selection was a clear visualization of image or the parameter is derived using the other ones.
impregnated, innumerable neurons. Such tissue section As it is expected, the standardized parameters are all
was treated as successful and chosen for the further derived but some of the unstandardized ones as well.
analysis. Neurons of such sections were then transformed The quantitative neuromorphological parameters are all
into digital images by digital camera Leica DC 100 explained in detail in our previous studies [10,18]. Here,
(Leica Microsystem Wetzlar GmbH, Wetlzar, Germany), we just list them for completeness.
accompanied by the adequate software package (Leica
Microsystem Wetzlar Ltd., Heerbrurg, Switzerland). Each Compartmental parameters of the size of a
neuron was photographed and recorded from six image neuron
slices by increasing the depth of optical focus by 15 mm.
Although suitable surface measurements, this group
Obtained serial images of neurons were then imported
of parameters determine the sizes of various aspects and
into a software specialized for digital image analysis
compartments of a BNI2D. Naturally, they are expressed
and reconstruction, the Image J. Using the average Z–
in microns squared (µm2). They include the following
projection method, all selected images were projected
parameters: a) the surface of an entire neuronal area
onto an image stack along the axis perpendicular to
(AN), b) the surface of a neurosoma (perikaryon) (AS) [19],
the image plane, while spines, axons and background
c) the surface of a dendritic tree (ADT), d) the surface
artifacts were digitally removed. Images of neurons
of an entire neuronal field area (ANF), e) the surface
processed on such manner were finally transformed to
of a dendritic field area (ADF) and f) the surface of a
binary images. The obtained image represents a digital
perineuronal space area (APNS).
2D binary image of a neuron (BNI2D), shown as black on
a white background (Figure 2 a, c, e, g, I, k).
The AN parameter measures the actual surface
Quantitative neuromorphological parameters size of the entire area of a BNI2D (Figure 4a). The AS
parameter represents the real, actual surface of a BNI2D
In this study, quantitative, computational parameters
perikaryon image, after the dendrites were previously
describe various aspects of neuronal morphology
digitally removed (Figure 4a). The ADT parameter is a
and function of the DN and neostriatum, in general.
measure of the actual, real surface of a dendritic tree,
The DN parameters served for the purposes of
after the perikaryon is removed. The ANF parameter
neuromorphofunctional, computational RSM
measures the surface of a minimal radial field occupied
modeling, while the neostriate parameters were used
by a BNI2D (Figure 4b). The ADF parameter is the surface
in the dentatostriate inter-cluster mapping. They are
of a minimal radial field occupied by a dendritic tree,
summarized in tables 1-3.
after the neurosoma was digitally removed. The APNS
parameter is an inter-dendritic surface, i.e. the surface
As it can be seen from the Tables, the vast majority
between dendrites. It is obtained using the ANF polygon,
of parameters are the same i.e. determined for both of
after the entire neuron was digitally removed. Its value
the nuclei, with some exceptions such as the RB and RT
is calculated as the difference between the ANF and AN
parameter. They can be devided into the following classes:
parameter,
1) compartmental neuron size (surface) parameters, 2)
shape, 3) compartmental length, 4) dendritic arborization
APNS = ANF – AN. (1)
(branching), 5) organizational parameters of neuronal

morphology and 6) computational neuromorphological
Since a perineuronal space is filled up with neuroglial
complexity parameters. This is the main classification of
cells, as well as with dendrites, small interneurons
parameters. The most of them are unstandardized. This
and axon bundles [20], the APNS parameter might be
means that they are not given as the ratios between other
thought of as a partial neuromorphological measure of
parameters, i.e., they are not standardized with respect
neuro-neuroglial (neuroglia-to-neuron) supportiveness
to some other parameter(s). Some of the parameters, so
and interconnectedness. Alternatively, from
called PCA parameters, are calculated using the principal
neurophysiological view, it can be comprehended as a
component analysis (PCA) of pixel distributions of a binary
partial neurotrophic strength of a particular neuron.

Annal Behav Neurosci, 2(1): 168-196 (2019) 171

 average dendritic width.

Standardized parameters of compartmental

neuronal length
All of these parameters are standardized with respect
to the AS parameter. In this way the dependence and
Figure 4: The graphical representation of the three variation of compartmental length upon the neuron
surface parameters, namely AN, AS and AN.
body size is excluded. The AS parameter is chosen as
Neuronal shape parameters a standardizing factor since it is the main categorizing
parameter of neurons according to their size-every
They define and quantify neuron shape and the
particular transmitter type of a neuron is characterized
shape of its compartments on a BNI2D. They include
by a specific range of body size. Therefore, the modeling
the following parameters: a) the fractal dimension of a
is more accurate on account of a single factor. They
native, unmodified BNI2D (DN), b) the fractal dimension
are: a) standardized total dendritic length (Lst), b) the
of a BNI2D outline (DO) c) the index of asymmetry of a
standardized radius of the NMAX circle (RCst) and c)
neuron (MN), d) the index of neuroperikaryon asymmetry
standardized average dendritic width (DWDTHst). They are
(MS) and e) neuronal axialization index (DAX).
obtained on the following ways

The DN parameter is a quantitative measure of real

neuron shape (Figure 5a, c). The DO parameter measures
2
the fractal dimension of a neuron outline, i.e. the fractal
dimension of the border of the neuron (Figure 6). The
MN parameter is the inverse circularity of actual, real 3
neuron shape as a measure of the neuron roundness
and shape compactness. Analogously, the MS parameter
is the inverse circularity of real, actual neurosomatic 4
shape as a measure of perikaryon roundness and shape
compactness. Unstandardized branching parameters of den-
dritic arborization
The DAX quantifies how much a neuron is elongated These parameters give a number of dendrites of
in a 2D space, far mostly on account of the elongation of various branching orders. They include the following: a)
the dendritic tree. It was introduced in morphometry by the number of primary dendrites (NPD), b) the number
Yelnik et al. [21]. of secondary dendrites (NSD), c) the number of tertiary
dendrites (NTD), d) the number of higher order dendrites
Unstandardized compartmental length parame- (NHOD), e) the total number of (all) dendrites (NDALL), f)
ters of a neuron the total number of dendrites above the first order (NR),
Expressed in microns (µm), they measure and g) the total number of tertiary and higher order dendrites,
determine lengths of various compartments and aspects counted together (NTHOD) and h) dendritic centrifugal
of a BNI2D. They include the following parameters: a) branching order (DCBO) (Figure 8). The DCBO parameter
total dendritic length (L), b) the radius of the NMAX circle represents an average dendritic branching order. Though
(RC), c) the radius of the circle with the maximal number it represents a ratio, it’s not a standardized parameter,
of dendritic branching points (RB), d) the radius of the as it describes the elementary property of a neuron. The
circle with the maximal number of terminal dendrites (RT) other parameters in this group are self-explanatory.
and e) average dendritic width (DWDTH).
Standardized dendritic arborization branching
The L parameter is a sum of all individual dendritic parameters
lengths [22,23]. The RC parameter determines the Only one parameter belongs to this category, namely,
position of maximal dendritic arborization density [19] the standardized dendritic centrifugal branching order
(Figure 7). In a similar manner, the parameters RB and (DCBOst). It is calculated as the following ratio
RT are also obtained and partly represent the branching
parameters as well. The DWDTH parameter measures
Annal Behav Neurosci, 2(1): 168-196 (2019) 172
 (MDCBO), b) the dendritic surface of influence (DSI), c)
5 partial dendritic surface (DSP), d) dendritic clustering
index (degree) (DCLD) and e) dendritic orientation degree
Where NMIN is a complexity parameter (explained (DPOL).
below). All of the branching parameters could also be
comprehended as partial complexity parameters as well, The MDCBO quantifies the extent of radial symmetry
due to their high quantitative and topological variations, of a dendritic tree in terms of its branching degree. The
largely contributing that way to dendritic branching DSI PCA parameter represents the covariance of dendritic
pattern complexity. space coordinates. It quantifies the dendritic dispersion
degree of a neuron in a 2D space. The other PCA
parameter, DSP represents the mere, uniform density of
an entire dendritic arborization, without the complexity
component. The DCLD parameter gives a measure
of dendritic clustering intensity, i.e. it is the dendritic
clusterization degree (index). The DPOL measures the
orientation of an entire neuron in a 2D space. It is also
introduced by Yelnik et al. [21].

Standardized neuromorphological organization

parameters
Figure 5: The calculation of the fractal DN and DS In this group is only standardized DSP parameter (DSPst).
parameters. It is standardized with respect to the ADF parameter,

or equivalently

From this standardization it can be seen that a neuron

with the large ratio between the actual dendritic area
(ADT) and its occupied field (ADF) would have the higher
partial dendritic surface (DSPst), as it is naturally expected
due to its higher dendritic space density.

Unstandardized neuromorphological
complexity parameters
It can be said that this is the most important group
Figure 6: The Richardson circle/radius counting of the neuromorphological parameters from the view
method. of describing morphological features of a neuron.
This does not mandatorily imply significant response-
Unstandardized organization parameters of factor dependence in terms of these parameters.
neuronal morphology Neuromorphological complexity as BNI2D detail and
shape complexity [24] can be reduced to branching
Parameters in this group express numerically a degree
pattern complexity, highly dependent on branching
of neuromorphological organization of the compartments
characteristics of dendritic tree and its order, expressed
and various parts of a neuron in a 2D space and in
through the branching parameters and its topological
relation to each other. They include the following set of
orientation of in a 2D space. They are: a) the fractal
parameters: a) dendritic (branching) polarization index
dimension of a skeletonized neuron image (DS) [25], b)

Annal Behav Neurosci, 2(1): 168-196 (2019) 173

the maximal complexity index of dendritic arborization Neurofunctional parameters
density (NMAX) [19,23], c) the minimal complexity index of
There is no yet a computational neuromorphofunctional
dendritic tree density (NMIN) and d) dendritic branching
model that would predict behavior of a bunch of neural
pattern complexity (CDF). The DS selectively reflects only
parameters in terms of values of other parameters. To this
dendritic branching pattern complexity or the degree
end, we wanted to make it possible using an approximate,
of dendritic tree curvature (Figure 5b and c). The NMAX/
computational, algorithmic approach by combining
NMIN parameters measure dendritic arborization
experimentally and theoretically obtained neural data. In
complex density (Figure 7). The CDF parameter is the
this study, the neurofunctional parameters are not directly,
most integrative of all of the four ones, describing entire
experimentally measured for each individual neuron, but
neuron shape complexity.
rather theoretically determined on the following way.
According to the results of Uusisaari and Knöpfel [6], DN
neurons are classifed into five neurotransmitter types:
1) small GABAergic (GABA-decarboxylase positive/GAD+)
inferior olive projecting neurons (GABA/GAD-SIOP)
with neurosoma size 35 ± 15 µm2, 2) large GABAergic
neurons (GABA/GAD-L), body size 175 ± 125 µm2, 3) small
GLUergic neurons (Glu-S), perikaryon size 200 ± 100
µm2, 4) large GLUergic neurons (Glu-L), perikaryon size
450 ± 150 µm2, and 5) large intrinsically inactive Gly-ergic
neurons (Gly-LI), soma size 250 ± 80 µm2. Additionally,
Figure 7: The calculation of neuromorphological com- they further featurize each neuron neurofunctionaly
plexity parameters. according to its neurotransmitter type membership.
Using their classification, our neurofunctional mapping
Standardized neuromorphological complexity of our each individual DN neuron is achieved by
parameters performing the identification of it in terms of its most
probable neurotransmitter type. Based on theoretically
As for the most important one, the most of
generated Gaussian distributions and by using the
standardizations used in this study is introduced for
Uusisaari’s results, this is done on a stochastic manner
the CDF parameter. As significant response-predictor
by assigning the most probable neurotransmitter type to
dependences are expected in the RSM models for this
each DN neuron according to its body size. Then, using
parameter due to the highest degree of it integrative
again the same results, the neurofunctional parameters
nature, compared to other complexity parameters,
are then determined in a similar manner using Gaussian
neurons must be matched with respect to many
distribution pattern. As a final result, for each DN
parameters in order to obtain valid results. The CDF/ADFst
neuron is obtained a predicted value of a particular
parameter is obtained as the following ratio,
neurofunctional parameter according to the value of
its categorical cell size. Similar principle is applied to
neostriate neurons for the purposes of dentatostriate
8
intercluster mapping using the results of Koos and
Analogously to this, the other parameters in this group Tepper [26-28]. In this way we were able to observe
are CDF(NDALL)st, CDF(NPD)st, CDF(NSD)st, CDF(NTD)st, relations of neuromorphological parameters in terms
CDF(NHOD)st, CDF(NR)st, CDF(NTHOD)st and CDF(DCBO)st. It of neurofunctional ones and vice versa. Thus, using the
should be noted that the CDF is predominantly standardized one known and predefined relation, neuronal body size/
in terms of the branching parameters as its major function, [6], we are trying to decipher other analogous
determinants. The NMAXst is obtained as the following ratio relations for other parameters. Since data distributions
of the neurofunctional parameters, according to the
neurotransmitter neuronal types they belong to, are
9 redundant due to a large number of these parameters
and simulaneously represent just intermediar products
of computational simulation, they are not explicitly
Tables 1-3 show the measured neuromorphological shown. The obtaned final data used for the generation of
parameters for the neostriatum and DN. the models and their analysis are presented in tables 1-4.

Annal Behav Neurosci, 2(1): 168-196 (2019) 174

 The DN neurofunctional parameters can be categorized Samples of neurons
as action potential (AP/AV) parameters, determined
In this study, we resorted to the use of a virtual DN
for an action potential induced by 1) spontaneous
sample, as the natural sample was not of satisfying size
firing of a neuron, 2) hypopolarizing/depolarizing and
and neuron classifying neural network showed stabile
3) hyperpolarizing injection current stimulus. The
behavior on the virtual sample [12]. Our DN virtual sample
perikaryodendritic compartmental parameters represent
consisted of 861 EBN and 476 IBN with the total of 1337
the fourth category (Table 4, Figure 9). The spontaneous
DN neurons. Caudate cluster I with 457 interneurons,
AV parameters are: the frequency of spontaneous AP
489 cluster II and 391 putaminal interneurons constitute
(FS), its width, i.e. duration (WS), volatage (VS), the voltage
the structure of the natural sample of the neostriatum.
of its slow phase of afterhyperpolarization (VSHS), the
As naturally expected, these two samples are separately
voltage of the fast phase of afterhyperpolarization (VFHS),
analyzed for different purposes.
and the total voltage of afterhyperpolarization (VHS).
The depolarization-induced AP parameters are: bursting Statistical data analysis
volatage (VBD), bursting frequency (FBD), the volatage of
bursting threshold (VTBD), the voltage of depolarizing Data are analyzed using response surface (RSM)
phase (VDD), the total voltage of depolarizing AP (VD) modeling with partial least-squares (PLSR) and principal
and the voltage of afterhyperpolarization (VHD), with component regression analysis (PCR), along with Pearson
the varieties minimal (MINVHD), maximal (MAXVHD) and canonical correlation analysis. Artificial neural
and average (AVRGVHD). Similar is aplicable to the networks were used for the functional clustering of
hyperpolarizing AP (Appendix). The FIPSP, RTAUIPSP and neurons as a well-proved classification tool [10-12,29]. It
DTAUIPSP parameters are the frequency and constant should be noted that PCA analysis is used in two different
times of rize and decay of inhibitory postsynaptic potential, contexts in this study. Namely, in one context, when
respectively, and represent constant characteristics applied to an entire neuron sample, it is used as a neuron
for each individual neuron. The neostriatal functional classification tool, whereas in the second, when applied
parameters are bursting voltage (VB) and frequency (FB) to a particular neuron image, it represented a basis for
(Table 2, Appendix). calculations of some neuromorphological parameters
(DSI, DSP and DAX).

RSM modeling

RSM is a collection of mathematical and statistical

techniques made for empirical model building based
upon experimentally collected data. In contrast to exact
techniques, it represents one of approximate stochastic
modeling methods. By careful design of experiments, the
objective is to optimize a response, output variable, which
is influenced by several independent, input variables.
An experiment is a series of tests, called runs, in which
Figure 8: Dendritic branching ordering.
changes are made in the input variables in order to
identify the changes in the output response. It represents
the method of polynomial, response-factor dependence
modeling that enables mathematical, computational,
theoretical relationships between predictors and
response parameter through adjusted coefficients,
formally enclosing and defining a law of response-factor
dependence. The general formula for RSM model is
10

Where RSM is a response surface, according to

Figure 9: AV neuroelectrophysiological parameters. the suitable graphical presentation. Very often, valid
models are multi-argument functions. Thus,
is an RSM function representing response-
factor dependence model and E is a model fitting error
Annal Behav Neurosci, 2(1): 168-196 (2019) 175
 Table 1: The mean ± SD, t- and p-values for the neuromorphological parameters of DN neurons.

Parameters EBN IBN t–value p–value

Unstandardized parameters
Surface parameters
AS 128.52 ± 36.31 111.25 ± 45.54 2.31 p<0.05
AN 7701.04 ± 2444.44 6613.31 ± 7.56 p<0.001
2648.45
ADT 6577.57 ± 1843.05 6615.9 ± 2682.69 -0.31 p>0.05
ANF 38144.39 ± 34130.73 ± 4.96 p<0.001
13269.58 15692.67
ADF 30880.53 ± 29104.61 ± 2.86 p<0.01
10917.93 10737.34
APNS 22854.2 ± 6334.52 20367.66 ± 6.39 p<0.001
7610.38
Length parameters
L 1707.66 ± 458.73 1478.13 ± 524.61 8.32 p<0.001
RC 42.12 ± 11.62 39.62 ± 12.17 3.7 p<0.001
RB 44.1 ± 9.3 43.29 ± 11.05 1.42 p>0.05
RT 90.32 ± 15.71 87.55 ± 18.54 2.89 p<0.01
DWDTH 4.56 ± 3.53 5.71 ± 4.97 -4.89 p<0.001
Shape parameters
DO 1.45 ± 0.04 1.43 ± 0.05 11.75 p<0.001
MS 1.2 ± 0.05 1.2 ± 0.08 0.39 p>0.05
MN 4.66 ± 8 4 ± 2.88 1.73 p>0.05
Branching parameters
NDALL 31.32 ± 5.87 32.53 ± 9.39 -2.88 p<0.01
NPD 7.61 ± 1.36 7 ± 1.56 7.52 p<0.001
NSD 9.34 ± 3.39 11.96 ± 5.49 -10.78 p<0.001
NTHOD 16.35 ± 4.18 15.07 ± 6.59 4.34 p<0.001
Complexity parameters
NMAX 15.53 ± 2.82 14.12 ± 3.68 7.84 p<0.001
DS 1.29 ± 0.04 1.27 ± 0.05 10.14 p<0.001
CDF 4.76 ± 2.52 5.08 ± 3.72 -1.85 p>0.05
Standardized parameters
Length parameters
Lst 2688.1 ± 1290.89 2293.3 ± 1651.03 4.84 p<0.001
(DWDTH)st 0.01 ± 0.003 0.01 ± 0.01 -5.92 p<0.001
RCst 1.83 ± 0.95 1.77 ± 1.2 1.1 p>0.05
Complexity parameters
CDF(ADF)st 0.003 ± 0.006 0.002 ± 0.003 3.35 p<0.01
CDF(NDALL)st 0.17 ± 0.16 0.18 ± 0.17 -0.46 p>0.05
CDF(NPD)st 0.68 ± 0.5 0.79 ± 0.82 -3.13 p<0.01
CDF(NSD)st 1.3 ± 2.92 0.67 ± 0.67 4.64 p<0.001
CDF(NTHOD)st 1.71 ± 3.05 0.6 ± 0.6 7.82 p<0.001

Annal Behav Neurosci, 2(1): 168-196 (2019) 176

 Table 2: The data mean ± SD for the unstandardized parameters of neostriate interneuron morphology and
function.

Parameters Caudate cluster I Caudate cluster II Putaminal cluster

Surface parameters
AS 446.21 ± 261.36 297.95 ± 155.625 4520.53 ± 2753.47
AN 33370.37 ± 15770.96 ± 5995.24 19340.28 ± 11168.11
10896.47
ADT 28908.24 ± 9984.96 12791.45 ± 5442.62 14819.75 ± 9772.81
ANF 170627.09 ± 57470.96 ± 25640.58 78820.44 ± 56069.08
53394.63
ADF 166164.96 ± 54491.45 ± 25168.78 74299.92 ± 55136.37
52757.44
APNS 137256.72 ± 41700 ± 20840.53 59480.17 ± 47145.93
47410.49
Shape parameters
DO 1.17 ± 0.02 1.15 ± 0.02 1.16 ± 0.02
DN 1.65 ± 0.48 1.12 ± 0.33 1.46 ± 0.07
MS 1.29 ± 0.2 1.26 ± 0.18 1.31 ± 0.28
MN 61.02 ± 15.12 31.2 ± 13.49 41.00 ± 23.08
DAX 0.47 ± 0.26 0.49 ± 0.26 0.46 ± 0.25
Branching parameters
NPD 6.37 ± 2.35 5.87 ± 2.17 6.25 ± 2.42
NSD 8.78 ± 3.33 7.35 ± 3.02 7.91 ± 4.04
NTD 8.3 ± 3.96 5.57 ± 3.47 6.17 ± 4.33
NHOD 10.49 ± 8.32 5.3 ± 5.21 5.96 ± 7.03
NTHOD 18.78 ± 11.05 10.87 ± 7.42 12.13 ± 10.27
NR 27.56 ± 13.06 18.22 ± 8.67 20.04 ± 12.96
NDALL 33.93 ± 13.94 24.09 ± 9.32 26.30 ± 14.12
DCBO 2.59 ± 0.38 2.33 ± 0.44 2.29 ± 0.48
Complexity parameters
DS 1 ± 0.03 1±0 1.16 ± 0.08
CDF 0.79 ± 1.11 0.78 ± 0.93 0.78 ± 1.12
NMAX 23.34 ± 12.24 28.26 ± 11.95 30.26 ± 12.52
NMIN 5.62 ± 3.64 5.42 ± 5.06 5.31 ± 4.17
Organization parameter
DSI 0.83 ± 0.19 0.81 ± 0.2 0.83 ± 0.19
DSP 37136.80 ± 17176.05 ± 10138.57 18959.12 ± 14901.21
15633.22
DCLD 0.2 ± 0.32 0.22 ± 0.35 0.21 ± 0.46
MDCBO 3137.92 ± 939.44 1870.35 ± 1323.32 1561.68 ± 950.07
DPOL 71.31 ± 40.01 64.22 ± 39.14 60.79 ± 36.91
Neurofunctional parameters
VB 51.7 ± 19.4 48.11 ± 10.79 49.33 ± 17.95
FB 163.17 ± 89.52 209.17 ± 79.4 162.75 ± 88.61

Annal Behav Neurosci, 2(1): 168-196 (2019) 177

 Table 3: The data mean ± SD for the standardized parameters of neostriate interneuron morphology.

Parameters Caudate cluster I Caudate cluster II Putaminal cluster

Lst 2822.78 ± 2188.84 864.22 ± 729.88 871.45 ± 1098.93
DWDTHst 0.03 ± 0.01 0.03 ± 0.02 0.027 ± 0.02
RCst 0.36 ± 0.74 0.21 ± 0.45 0.22 ± 0.71
DSPst 0.23 ± 0.11 0.34 ± 0.15 0.30 ± 0.21
CDF(ADF)st 0.5 ± 0.81 1.8 ± 2.14 0.20 ± 0.33
CDF(NDALL)st 0.03 ± 0.04 0.04 ± 0.05 0.04 ± 0.07
CDF(NPD)st 0.15 ± 0.24 0.17 ± 0.3 0.15 ± 0.21
CDF(NSD)st 0.11 ± 0.17 0.13 ± 0.19 0.13 ± 0.22
CDF(NTD)st 0.11 ± 0.16 0.19 ± 0.28 0.15 ± 0.30
CDF(NHOD)st 0.13 ± 0.3 0.14 ± 0.24 0.10 ± 0.22
CDF(NR)st 0.04 ± 0.05 0.06 ± 0.07 0.07 ± 0.14
CDF(NTHOD)st 0.05 ± 0.08 0.11 ± 0.18 0.10 ± 0.27
CDF(DCBO)st 0.3 ± 0.42 0.34 ± 0.41 0.36 ± 0.57
DCBOst 0.83 ± 0.98 1.35 ± 2.27 1.24 ± 3.24
NMAXst 8.12 ± 11.28 19.79 ± 41.73 18.74 ± 52.72

[30]. of interdependence), while the model error quantifies

the autonomy of each of these categorical instances-
Obtained RSM models, represented by final equations, morphology/function, i.e. their independence of the
are generated in a stepwise manner. A starting point in other one. Additionally, very common nonlinearity of RSM
model generation is one of chosen factors, according models, especially when a large number of parameters is
to the significance evaluation of a particular parameter. involved, ensures their ability to catch complex patterns of
Then, the ANOVA based contribution weighting is such relations, giving us therefore the maximal precision
performed upon the selected factor (parameter) and in quantification of their strength.
its Fisher quotient value and alpha error (p) level are
determined. The model adjusted-R2 value is then also Results
determined as a measure of fitting strength. If the chosen
In this study, the idea was to model
factor markedly suites the model, it is preserved in further
neuromorphofunctional relations of DN neurons in
analysis, when additional terms are added. Next ANOVA
terms of the experimentally measured morphological
checking evaluates the significance of the next factor but
and theoretically determined neurofunctional
in couple with the previous one(s), forming a partial sum.
parameters. The study design is bidirectional, i.e., the
As significant additional terms are getting included in the
neuromorphological parameters are expressed in terms
model, the adjusted-R2 value increases. The analysis stops
of neurofunctional ones and vice versa (Tables 5-8).
when there are no significant terms left to be included in
the model. As it is expected, there are two kinds of terms,
If a response parameter is one of the determined
namely, factor (a single parameter) and interaction terms
parameters, whether neuromorphological or functional,
(a product of parameters) [31] and the general formula of
the model is defined as a non-integral type model and
the obtained RSM model has the form,
if it is of integral (combined) nature (explained below)
the model is an integral type model, representative for
a whole parameter category (morphological/functional).
Every model is analyzed separately for the EBN and IBN
11 cluster. They are developed by a specific factor-extraction
algorithm (Figure 10). Two main hierarchical steps in
Where C is a fitting coefficient, while xn represent
model development are PLSR/PCR factor analysis, as the
parameters of the model. In this study, the fitting
first step, and RSM modeling as the second, i.e. the most
adjusted-R2 value of a model is taken as a measure of
significant factors are first extracted using PLSR/PCR
neuromorphofunctional relational strength (the strength
analysis and then an RSM model is built upon them.
Annal Behav Neurosci, 2(1): 168-196 (2019) 178
 Table 4: The mean ± SD, t– and p–values for the neurofunctional parameters of DN neurons.

Parameters EBN IBN t–value p–value

Spontaneous AP parameters
FS 5.65 ± 3.21 6.35 ± 4.06 -3.47 p<0.01
WS 2.14 ± 0.39 2.14 ± 0.5 -0.21 p>0.05
VS 64.91 ± 3.29 64.79 ± 3.94 0.61 p>0.05
VFHS 6.88 ± 2.42 7.23 ± 2.78 -2.44 p<0.05
VSHS 9.33 ± 4.26 8.76 ± 4.89 2.22 p<0.05
VHS 16.18 ± 3.64 15.97 ± 3.76 0.98 p>0.05
Depolarization–induced AP parameters
VBD 51.45 ± 10.33 51.45 ± 11.76 0.002 p>0.05
FBD 57.86 ± 8.08 59.78 ± 8.75 -4.03 p<0.001
VTBD 30.44 ± 7.17 29.83 ± 6.78 1.52 p>0.05
VDD 81.84 ± 8.16 81.19 ± 8.85 1.35 p>0.05
VD 80.92 ± 5.66 80.64 ± 6.68 0.81 p>0.05
MINVHD 29.28 ± 13.12 30.23 ± 15.02 -1.2 p>0.05
MAXVHD 43.86 ± 21.49 45.88 ± 23.42 -1.6 p>0.05
AVRGVHD 36.24 ± 17.41 37.92 ± 19.22 -1.63 p>0.05
Hyperpolarization–induced AP parameters
MINVTBH 15.06 ± 7.8 14.3 ± 9.46 1.59 p>0.05
MAXVTBH 16.31 ± 10.22 17.51 ± 10.97 -2 p<0.05
AVRGVTBH 16.05 ± 8.32 16.12 ± 9.5 -0.14 p>0.05
MINVBH 11.72 ± 6.67 11.2 ± 7.57 1.3 p>0.05
MAXVBH 62.21 ± 24.92 61.06 ± 28.87 0.76 p>0.05
AVRGVBH 37.58 ± 16.88 37.92 ± 19.95 -0.33 p>0.05
WBH 1263.86 ± 764.83 1232.57 ± 868.36 0.68 p>0.05
AVRGVHH 26.41 ± 11.36 26.22 ± 13.39 0.28 p>0.05
MAXVHH 38.48 ± 17.15 38.91 ± 18.78 -0.43 p>0.05
MINVHH 13.43 ± 6.45 13.39 ± 6.88 0.12 p>0.05
AVRGVH 27.98 ± 11.46 28.2 ± 13.51 -0.32 p>0.05
MAXVH 40.41 ± 20.9 42.04 ± 21.45 -1.36 p>0.05
MINVH 24.51 ± 12.27 25.66 ± 12.98 -1.61 p>0.05
FBH 18.46 ± 7 18.09 ± 8.27 0.87 p>0.05
Perikaryodendritic compartmental parameters
IPSC 69.58 ± 28.93 69.69 ± 36.25 -0.06 p>0.05
FIPSP 13.05 ± 11.45 13.31 ± 14.04 -0.37 p>0.05
RTAUIPSP 2.73 ± 2.36 2.62 ± 2.47 0.86 p>0.05
DTAUIPSP 19.08 ± 16.81 14.83 ± 13.72 4.71 p<0.001

The third or preliminary and simultaneously a form of nonlinear dependence, linear dependence (co-
intermediate, optional step is adequate correlation linearity), and independence. All except independence
analysis (Pearson/canonical) between response and are taken into account, i.e. tolerated for being included
predictor parameters, included at least in one of the cluster in the model, meaning that no bias toward relation type
models (EBN/IBN). It is not used for factor extraction but is preferred.
rather as a rough modeling tool, where preliminary, inter-
parameter relations are presented, other way unobvious, The only criterion is the significance of predictor
hidden by adequate RSM models. The relations can take contribution to the model or Pearson/canonical

Annal Behav Neurosci, 2(1): 168-196 (2019) 179

correlation significance (p<0.05). The presence of It just means that the obtained models are defined
statistically significant (p<0.05) correlation is taken as in terms of their validity, a type of inter-parametric
a criterion for inter-parametric dependence and its dependence and the error level of unexplained noise, else
absence (p>0.05) for independence. An adequate inter- greater than zero but tolerable (<10%) for valid models.
parametric nonlinear relationship is just determined by a If the error level was not satisfying (>10%), the RSM
specific RSM model. The main purpose of RSM modeling model was proclaimed invalid and the type of parameter
was the factor analysis of response variable and not data interdependence unknown (N/A).
extrapolation, nor prediction.
It has been mentioned that non-integral RSM models
represent the models of one measured or determined
parameter, belonging to one category, as a function in
terms of PLSR/PCR selected parameters of the other
category. They are presented in tables 5-8 for the entire
parameter set. Here are listed just those with the greatest
adjusted-R2 as the representatives of each parameter
category.

12

13
Figure 10: The flux diagram of neuromorphofunctional
modeling used in this study.

RSM models and inter-parametric

14
neuromorphofunctional correlations
As aforementioned, in this study, RSM models are
divided to non-integral and integral type according to the
fact whether a response variable is a measured parameter
(experimentally or theoretically) or combined theoretical,
computationally obtained variable, representative for
an entire parameter category. Taking into account the
extreme complexity of the models due to a large number
15
of parameters involved, the summation convention
is used, thus impeding us to show model coefficients
explicitly. Instead, as it was already mentioned, the
emphasis is put to predictor extraction and the type of
dependence (linear vs. nonlinear) in order to formalize
mathematically the low of dependence. Of course, this
From the AS RSM model it can be seen that
does not justify expectation for the obtained RSM models
neurofunctional parameters which partly determine the
to be exact, i.e. to have zero-error level, since they
size of neurosoma are the VS, VBD, VDD and naturally,
represent approximate numerical modeling solutions.
the transmitter type of a neuron, as already shown by
Only exact modeling methods fulfill this condition but
Uusisaari and Knöpfel [6] (Figure 11-13).
cannot give quantification of the relational strength of the
connection between neuronal morphology and function.
In order to extend the description of parameter

Annal Behav Neurosci, 2(1): 168-196 (2019) 180

interdependence, adequate response-factor correlations impossible to obtain by measurement, i.e., beyond the
are determined in order to obtain the relation direction. boundary values of the real parameters. This is due to
Thus, for example, the AS/VBD correlation is very highly the error cumulative effect at the boundaries of the
significant (p<0.001), positive for the EBN cluster, meaning region. They are in fact valid only for very narrow region,
that with larger AV burst voltage, the neuron would have approximately in the middle of a model RSM surface but
its perikaryon increasing (Figure 13) and from the model we chose to take them as they are since they nicely present
itself, it can be seen that this relation is nonlinear, i.e. far the mathematical type of dependence. Other models can
more extensive for AS values greater than 100 µm2. For be interpreted on a similar manner as well. Although
the IBN neuron cluster the situation is different, since VBD coefficients are not shown due to redundancy of data,
correlation with neurosoma size is not significant (p>0.05). thus not allowing the analysis of dependence directions,
By taking a look at the model itself reveals that for the IBN tables 5-8 demonstrate clear bidirectional connections
cluster, the VBD is just appearing in the interaction terms, between parameterized features like neuron shape
implying its significance only through the conglomerate complexity and neurosoma symmetry, for example,
of factors involved, i.e. through their combined effect with firing patterns etc. The adjusted-R2, error level and
whereas its single influence is not enough to reach a level precision of the model would then signify the strength
of significance. On the other hand, the analysis of the AS of dynamical inter-parametric connection, responsible
EBN model shows that the VBD does stand alone in one for the determination and intensity with which the law is
of the model terms, thus making it one of the significant taking place.
carriers of the model, in terms of single factor influence,
which correlation naturally supports. By the way, the
ordinal character of transmitter type (TT) as a multiplier
mathematically does not change the behavior of non-
interacting terms to interactive ones. Similar relations
can be extracted by analyzing the other factors of this
and other models (Tables 5-8, Figures 14-16). 16

Figure 11: PLSR/PCR analysis of predictors determining

the AS parameter for the EBN and IBN clusters.

Integral RSM models represent the models where the

response parameter is combined, representative, integral
variable (parameter) of an entire parameter category
(morphological/functional). They are obtained using the
combination of matrix manipulation techniques such
as concatenation, multiplication and squared deviation
summation applied upon the all parameters of one
deterministic category (morphology/function). They are
obtained for each DN neuronal cluster (EBN and IBN)
using the following formula,

It should be noticed that the 3D graphs of RSM models

(Figures 12 and 15) entail outliers as well as theoretical
Figure 12: The graphical presentations of the adequate
values of response parameter, which are practically

Annal Behav Neurosci, 2(1): 168-196 (2019) 181

 Table 5: Autoneurofunctional determination of the EBN neuromorphological parameters.

Response Predictors/Factors Model Adj R–sq(%) Adeq Prec

AS VS VDD VBD TT QI 80.26 72.97
AN MAXVH AVRGVBH WBH VDD IPSC TT LNI 53.58 52.97
L VD WBH VDD VBD MAXVBH IPSC TT LNI 37.05 42.46
NPD WBH DTAUIPSP TT LNI 21.86 27.56
NMAX AVRGVHH MAXVHD TT LNI 12.84 32.59
DO VD DTAUIPSP VDD AT TT LNI 13.84 28.62
DS VDD AVRGVBH AT TT LNI 29.07 41.51
RC DTAUIPSP IPSC AT TT LNI 13.27 25.63
RT VD VTBD AT TT LNI 33.87 31.31
RB VD IPSC TT LNI 41.97 43.41
ANF VS WBH IPSC VDD VBD TT LNI 33.09 36.63
NTHOD DTAUIPSP LNI 3.28 22.21
MN N/A N/A N/A N/A
CDF IPSC FIPSP TT LI 5.87 30.09
NSD MINVTBH VFHS DTAUIPSP VDD VBD TT LNI 19.22 21.69
ADF AVRGVBH MINVHH IPSC DTAUIPSP TT LNI 14.62 46.64
MS TT LNI 2.43 6.66
ADT IPSC FBD MAXVBH VTBD WBH TT LNI 23.88 38.21
APNS VDD TT LNI 12.24 32.86
Lst MINVHD VDD VBD AT TT LNI 6.34 20.74
DWDTH MINVHD WBH DTAUIPSP VDD VBD AT TT LNI 28.42 55.25
DWDTHst FIPSP AVRGVTBH FBH WBH IPSC TT LNI 44.08 44.62
RCst MINVHD MAXVBH TT LNI 7.33 19.86
CDF(ADF)st DTAUIPSP LNI 1.1 13.12
CDF(NDALL)st FIPSP MAXVHD FBH AVRGVTBH MAXVBH LNI 25.33 50.03
VTBD WBH IPSC TT
CDF(NPD)st FIPSP FBH IPSC TT LNI 15.49 50.34
CDF(NTHOD)st DTAUIPSP FIPSP WBH IPSC TT LNI 11.57 28.03
CDF(NSD)st DTAUIPSP FIPSP MINVBH FBD IPSC TT LNI 9.27 28.66
NDALL DTAUIPSP TT LNI 8.12 18.42

L–linear, I–interaction, NI–noninteractive, Q–quadratic, C–cubic (example, LNI–linear model without interaction).

AS RSM model response surfaces for the EBN and IBN correlations are shown in figures 18 and 20, respectively.
clusters. From the integrative models, it can be seen that all of the
Where X̅ represents the mean of the each parameter parameters are contributing to the model, predominantly
belonging to one category (morphology/function) and through interactions, thus implying very complex
to a DN cluster and k=100 is an arbitrary constant. In implicate relations (Figures 21-23).
fact, they are inverse Fourier transforms of the above-
formulated integral variables (Figure 17) but since In order to analyze a general character of
the suitable RSM models had lower performance on neuromorphofunctional relations, the inter-integrative
periodic functions, their Fourier transforms are kept as correlations are performed. They are obtained using the
representative integral variables. The morphological canonical correlation analysis between the two categories
EBN integral model is shown to be valid (Figures 18- of parameters and Pearson correlation between the
20). However, the IBN model could not be obtained as integrative variables. The both canonical correlations are
significant one (not shown). Its PLSR/PCR analysis and significant (p<0.05) (Figure 24a), but the Pearson ones are

Annal Behav Neurosci, 2(1): 168-196 (2019) 182

 Table 6: Autoneurofunctional determination of the IBN neuromorphological parameters.

Response Predictors/Factors Model Adj R–sq(%) Adeq Prec

AS MAXVTBH RTAUIPSP VBD TT QI 81.71 41.36
AN TT LNI 35.43 20.79
L IPSC MAXVTBH TT LNI 25.11 20.48
NPD MAXVBH TT LNI 16.14 13.57
NMAX FIPSP IPSC TT LNI 6.16 15.92
DO TT LNI 3.24 5.67
DS FIPSP TT LNI 18.77 15.46
RC FIPSP TT LNI 12.52 19.02
RT FIPSP AVRGVTBH MINVTBH TT LNI 36.61 23.14
RB TT LNI 33.72 21.58
ANF TT LNI 31.74 19.4
NTHOD N/A N/A N/A N/A
MN WBH TT CI 1.25 7.84
CDF TT LNI 7.98 7.86
NSD VBD VFHS MINVTBH VD VS IPSC FIPSP LNI 28.22 21.55
MAXVTBH AVRGVTBH TT
ADF N/A N/A N/A N/A
MS TT LNI 5.46 6.74
ADT WBH AT TT LNI 11.4 16.32
APNS N/A N/A N/A N/A
Lst N/A N/A N/A N/A
DWDTH VD VSHS MINVHD AVRGVBH MAXVBH LNI 37.82 32.05
TT
DWDTHst VSHS TT LNI 26.16 20.37
RCst WBH TT LNI 6.69 10.71
CDF(ADF)st MAXVBH VSHS VHS WBH MAXVTBH LNI 16.15 18.49
AVRGVTBH MINVTBH TT
CDF(NDALL)st TT LNI 12.63 19.79
CDF(NPD)st TT LNI 10.03 8.58
CDF(NTHOD)st IPSC FBH TT LNI 9.09 9.27
CDF(NSD)st MINVBH TT LNI 13.74 13.92
NDALL TT LNI 7.26 10.44

not (p>0.05) (Figure 24b). parameters: NTHOD, MN and CDF (ADF)st. For the IBN
cluster the insignificant were also the NTHOD and MN. In all
The relations between a type of activity of neurons other cases ANOVA was very highly significant (p<0.001).
(intrinsic/extrinsic) and neurotransmitter type in terms The scheme of the integrative neuromorphofunctional
of all neuromorphological factors are also obtained network shown in figure 27 is symmetric fully connected,
(Figures 25 and 26). The significant for the activity of partly due to direct model connections and partly on
the EBN neurons are the AN (p<0.05), L (p<0.05), NPD account of indirect ones.
(p<0.01), APNS (p<0.05), Lst (p<0.05), while for the IBN
Neurofunctional classification of DN neurons
cluster are significant CDF(NPD)st (p<0.05), DWDTH (p<0.05),
DWDTHst (p<0.01), Lst (p<0.05), ADT (p<0.05), ANF (p<0.05) Neurofunctional classification based just upon
and RC (p<0.05). Considering the neuromorphological the functional parameter set was selectively applied
TT determination, ANOVA was insignificant for the EBN to neurons in order to investigate whether the
cluster (p>0.05), concerning the following morphological neurofunctional classification follows the morphological
Annal Behav Neurosci, 2(1): 168-196 (2019) 183
one.

Of IBN neuron morphology the FBH and AVRGVTBH

parameters, the adequate RSM model is not significant.

This would tell us more about neuromorphofunctional

relations if such classification would be proven as
existing and valid. PCA analysis as unsupervised method,
followed by artificial neural network application as
supervised one, is used for this task. Unfortunately,
the results were devastating showing that such kind of
correlation is excluded, despite the fact that the ROC
curve doesn’t seem so bad (Figure 28). Table 9 proves the
result of neurofunctional classification failure, where it
can be seen that the majority of neurons is, according to
morphology, wrongly assigned to the EBN cluster (leaded
by the beta error concept, sensitivity threshold taken is at
least 80%). This result is also supported by the result of
correlation-comparison analysis where just 6.47% of the
neuromorphofunctional correlations are mismatched
correlations (Figure 29) [10,12].

Figure 15: The graphical presentations of the adequate

VTBD RSM model response surfaces for the EBN and
IBN clusters.

Figure 13: Pearson and canonical inter-parameter

causal predictor-response correlations have driven by
the adequate response AS neuromorphofunctional RSM
models for the EBN and IBN dentate nucleus neuron
clusters.

Figure 16: Pearson and canonical inter–parameter

Figure 14: PLSR/PCR analysis of predictors determining causal predictor-response correlations driven by the
the VTBD parameter for the EBN and IBN dentate adequate response VTBD neuromorphofunctional RSM
clusters. models for the EBN and IBN dentate
Annal Behav Neurosci, 2(1): 168-196 (2019) 184
 Table 7: Autoneuromorphological determination of the EBN neurofunctional parameters.

Response Predictors/Factors Model Adj R–sq(%) Adeq Prec

FS AS L NPD DO RB ADT APNS Lst DWDTHst RCst CDF(NTHOD)st LNI 25.55 36.39
WS entire set LNI 44.16 40.42
VS AS L NPD NMAX DO RC NSD ADF ADT Lst RCst DWDTH NDALL LNI 22.33 48.32
VBD AS AN L NPD NMAX DS RB NSD MS APNS Lst DWDTH RCst LNI 24.12 33.5
CDF(NTHOD)st
FBD NPD DO DS RC RB CDF NSD ADF MS ADT Lst DWDTHst RCst LNI 22.75 33.16
CDF(NDALL)st CDF(NTHOD)st CDF(NPD)st
VTBD AS NPD DS RB ANF MN CDF NSD ADF ADT APNS Lst DWDTH CI 94.56 24.27
RCst CDF(NDALL)st
VFHS AS AN L DO RC ANF MN CDF ADF APNS LST DWDTHst CD- LNI 28.61 31.96
F(NTHOD)st CDF(NPD)st NDALL
VSHS AN L NPD NMAX DO RC RB ANF NTHOD MN NSD APNS DWDTH LNI 25.55 35.89
CDF(NTHOD)st CDF(NPD)st CDF(NSD)st
VHS AS AN L NPD NMAX RT RB ANF NTHOD MN CDF ADF ADT Lst LNI 25.17 36.62
DWDTH CDF(NPD)st NDALL
VDD AS AN L NMAX DS ANF NM CDF ADT APNS DWDTH RCst CDF(N- LNI 32.19 45.13
DALL)st
VD AS NPD NMAX RB NTHOD MN NSD ADT ADF DWDTH DWDTHst LNI 20.01 34.13
PCst CDF(NTHOD)st CDF(NSD)st
MAXVHD AS L NPD RB ANF MN CDF APNS Lst RCst CDF(NDALL)st LNI 31.88 46.84
MINVHD AS L RB ANF MN APNS Lst DWDTH DWDTHst RCst CDF(NPD)st LNI 32.87 48.95
AVRGVHD AS L NPD RB ANF MN APNS Lst DWDTH DWDTHst RCst LNI 31.82 49.42
MINVTBH AN DO DS RB ANF MN ADF Lst DWDTH DWDTHst RCst CDF(N- LNI 34.03 47.44
DALL)st
MAXVTBH AS AN L NPD NMAX RB NSD ADF MS RCst CDF(NDALL)st LNI 28.03 50.16
AVRGVTBH AS AN NMAX DS RB ANF NSD Lst RCst CDF(NDALL)st CDF(NSD) LNI 35.27 47.53
st
AVRGVBH AS NPD RB NTHOD MN NSD ADT DWDTH DWDTHst RCst CD- LNI 21.22 41.09
F(NTHOD)st CDF(NSD)st NDALL
MAXVBH AS L NPD RB ANF NTHOD ADF APNS Lst DWDTHst RCst NDALL LNI 23.24 36.57
MINVBH AS L RT RB ANF ADF APNS Lst DWDTHst RCst CDF(NPD)st CD- LNI 21.29 29.25
F(NSD)st
WBH AS L RT RB ANF ADF APNS Lst DWDTH DWDTHst RCst CDF(N- LNI 20.9 38.53
DALL)st
AVRGVHH AS NPD DS RB ANF NTHOD MN ADT Lst DWDTHst RCst CD- LNI 22.95 36.27
F(NSD)st CDF(NDALL)st
MAXVHH AS L NPD RB ANF NTHOD NSD ADF MS APNS Lst DWDTHst LNI 27.82 41.96
RCst CDF(NPD)st NDALL
MINVHH AS AN L RB ANF MN NSD Lst DWDTHst RCst CDF(NPD)st LNI 33.28 48.98
AVRGVH AS L DS RB ANF MN NSD ADT Lst DWDTH DWDTHst RCst CD- LNI 27.55 39.72
F(NPD)st
MAXVH AS L NPD RB ANF MN CDF NSD ADF APNS Lst DWDTHst RCst LNI 29.82 42.62
MINVH AS L NPD RB ANF NTHOD MN APNS Lst DWDTHst RCst CDF(N- LNI 33.91 47.21
PD)st NDALL
FBH AS AN NPD DS RB NTHOD ADF Lst RCst CDF(NSD)st NDALL LNI 26.43 40.77
IPSC AS AN L DS RB NSD ADF Lst RCst CDF(NSD)st CDF(NDALL)st LNI 23.23 34.36
FIPSP AS AN L DS RB CDF(NDALL)st CDF(NPD)st CDF(NTHOD)st LNI 15.38 36.19
NDALL
RTAUIPSP AS L DS NSD RCst CDF(NPD)st CDF(NTHOD)st NDALL LNI 18.07 43.42
DTAUIPSP AS AN L NPD RC RB NTHOD MN ADT Lst RCst DWDTH CDF(N- LNI 23.73 39.89
PD)st CDF(NTHOD)st

Annal Behav Neurosci, 2(1): 168-196 (2019) 185

 Table 8: Autoneuromorphological determination of the IBN neurofunctional parameters.

Response Predictors/Factors Model Adj R–sq(%) Adeq

Prec
FS AS DS RB ANF MN CDF ADT Lst RCst NDALL LNI 17.09 19.82
WS L DO DS NTHOD CDF DWDTH CDF(ADF)st CDF(NDALL)st LNI 23.92 33.06
VS Entire set of neuromorphological parameters LI 50.92 19.41
VBD AS AN L NMAX RC ADT DWDTH RCst CDF(ADF)st LNI 16.87 31.59
FBD AS RC ADF RCst CDF(ADF)st CDF(NDALL)st LNI 8.66 27.07
VTBD AS ANF Lst DWDTH RCst CDF(ADF)st CDF(NDALL)st 4thI 27.38 29.5
VFHS L NPD NMAX RC RB CDF Lst DWDTH DWDTHst CDF(ADF)st LNI 31.4 35.43
CDF(NPD)st
VSHS DO RC RB ANF CDF MS Lst DWDTH DWDTHst CDF(ADF)st LNI 39.26 23.79
CDF(NPD)st
VHS ANF ADT DWDTH LNI 18.73 41.81
VDD AS AN L NPD NMAX RC ADT Lst DWDTH DWDTHst RCst LNI 31.49 40.42
VD AS ANF DWDTH RCst LNI 14.24 36.43
MAXVHD AS L NMAX DO RB ANF MN Lst RCst CDF(NSD)st LNI 21.72 25.84
MINVHD AS DO RB ANF MN CDF ADT Lst RCst CDF(ADF)st CDF(NPD)st LNI 23.47 24.61
AVRGVHD AS L NMAX DO RC RB ANF MN CDF ADT Lst RCst CDF(ADF)st LNI 23.36 25.11
MINVTBH AS AN NPD NTHOD Lst DWDTH CDF(ADF)st CDF(NDALL)st LNI 22.26 21.63
CDF(NPD)st
MAXVTBH RC RT NTHOD MS RCst CDF(ADF)st NDALL LNI 14.61 24.19
AVRGVTBH NPD NMAX RC RT NSD MS DWDTH RCst CDF(ADF)st LNI 18.91 21.7
AVRGVBH AS RC RB CDF ADT DWDTH RCst CDF(ADF)st LNI 14.81 24.84
MAXVBH AS DO RB ANF NTHOD CDF NSD ADT Lst RCst CDF(ADF)st LNI 21.06 21.79
NDALL
MINVBH L NTHOD CDF NSD Lst DWDTHst CDF(ADF)st CDF(NPD)st LNI 24.05 25.95
CDF(NSD)st NDALL
WBH AS DO RB ANF CDF Lst DWDTH DWDTHst RCst LNI 18.19 25.88
AVRGVHH ADT RC RCst CDF(ADF)st LNI 10.73 24.16
MAXVHH AS NMAX DO RC ANF MN CDF ADT Lst RCst CDF(ADF)st LNI 21.42 21.3
MINVHH NPD NMAX ANF CDF RCst CDF(ADF)st CDF(NPD)st LNI 25.39 27.74
AVRGVH AS DO RB ANF MN CDF ADT Lst RCst CDF(ADF)st LNI 19.28 22.99
MAXVH AS L NMAX DO RC RB ANF MN CDF Lst RCst CDF(ADF)st LNI 23.43 26.2
MINVH AS L NMAX DO RC RB ANF MN CDF Lst RCst CDF(ADF)st LNI 24.71 25.82
FBH AS NMAX DS ANF CDF ADT Lst RCst CDF(ADF)st LNI 18.84 23.01
IPSC AS RT CDF ADT RCst LNI 9.62 26.72
FIPSP AS AN L ADT RCst LNI 10.68 22.48
RTAUIPSP AS AN L NTHOD CDF NDALL NSD Lst CDF(ADF)st CDF(NPD)st LNI 11.93 14.67
DTAUIPSP entire set LI 66.29 15.09

Table 9: ANN parameters in the case of the neurofunctional classification of DN neurons.

Neuron % of correctly classified % of misclassified Classification Modified odds

type neurons neurons ratio ratio
EBN 85.4 14.6 5.85 6.04
IBN 50.8 49.2 1.03

Annal Behav Neurosci, 2(1): 168-196 (2019) 186

Dentatostriate inter-cluster mapping that approximate network contribution to modulation
of individual neuromorphofunctional relations would be
Having in mind the proved neuromorphofunctional
about 10%, which would further mean that neuron is the
feature (quantified by suitable parameters) relations
strongest automodulator of its morphology and function
characterizing a single neuron, we hypothesized that
through their mutual interplay, while the modulating
neural inter-cluster connections might behave as some
effect of the network is negligible compared to it.
kind of neuromodulators, acting upon these relations.
This in turn would enable decoding and mapping of
Discussion
specific, to say nucleo/clusterotopic inter-nuclear neural
projections that connect neuron clusters of two nuclei, Hitherto experimental and theoretical computational
in our case the DN and neostriatum, thus revealing and work proves undoubtedly that the structure and function
reconstructing the configuration of an entire inter-nuclear of a cell, in general, are interconnected on a very complex
inter-cluster neural network. As glanced, we tested this manner and especially if this issue is about neurons [5].
hypothesis on the dentatostriate inter-cluster network. This fact is true definitely for neuron types that can be
In our previous study [10] we showed the existence of differentiated at the level of qualitative studies, i.e., for
two DN morphologically different neuron clusters that neurons of obviously different morphology, observable
we named, according to their topology, EBN and IBN even by a naked eye [2,4]. However, the situation gets more
neurons, whereas in the neostriatum the existence of complicated in the case of theoretical, computational
three clusters is proved: the caudate cluster I and II of studies when much finer and subtle neuromorphological
unknown topological distribution and putaminal neuron differences are involved, as our results show.
cluster.

In order to neuromorphologically decode and analyze

the configuration of this neural network we again
resorted to RSM methodology. First, a representative
neuromorphofunctional integrative variable is created,
on the abovementioned manner, for each of the five
nuclear clusters of neurons, from both of the nuclei,
using the both of the categorical sets of parameters,
neuromorphological and functional one. Then, RSM
methodology is used where the projecting clusters of one
nucleus are treated as factors, determining a response
variable of a targeted, single cluster of the other nucleus.
These relations are bidirectionally analyzed for each of
the resulting dentatostriate inter-cluster combinations.

The general idea was that potential detection and

extraction of some projecting clusters as significant for
the network RSM model would be interpreted as specific
unidirectional nucleotopic inter-cluster mapping. This
combined with non-significant, or other words, bidirectional
connections would make the network nucleotopically
inhomogeneous, i.e. asymmetric. However, if no such
clusters exist, the configuration of the network would
be treated as symmetric, as in our case (model fit 10 ±
2.32%) (Figure 3c). It seems that indirect inter-nuclear
connections are unifying the network without specific
inter-nuclear inter-cluster projections (Figure 3a and b).
The observed neural network interconnection strength is
about 10% of the one observed in the case of an average
neuromorphofunctional interconnectedness observed
at the level of a single neuron. This would further imply Figure 17: Integrative parameter/variable plots.
Annal Behav Neurosci, 2(1): 168-196 (2019) 187
 In this study, essentially, we had three
neuromorphofunctional objectives: the first one is to
computationally relate the structure and function of DN
neurons, the second is to see whether these relations are
different between the two neuromorphotopological DN
neuron types, namely, EBN and IBN, and the third one
is to observe any potential neurofunctional differences
between two morphological types of neurons, as maybe
Figure 18: PLSR/PCR analysis of predictors determining the strongest indicator of neuromorphofunctional
the integral neuromorphological parameters relations. According to the results of our study, it is
(MRFPVINTGR) for the EBN and IBN dentate clusters. obvious that the structure and function of DN neurons are
interrelated, the result already proven by other studies
[2-6]. Additionally, we show that these relations are not
the same for both of the clusters. Thus, for example, the
neurofunctional parameters that partially determine the
size of neurosoma of EBN neurons are the VS, VBD, VDD
and the neurotransmitter type of a neuron, whereas the
other set of parameters is responsible for IBN soma size
determination. Just in the sense of neuroadaptation,
this would generally mean that a neuron would have
its soma size influenced by AV voltage and the type of
activity (spontaneous/intrinsic vs. stimulated/extrinsic).
This is probably the expression of underlying interplay of
a myriad of synaptic inputs bombarding the neurosoma,
behaving like neurotrophic factors and stimulating it
simultaneously to become larger. The figure 12b shows
that an EBN neuroperikaryion approximately linearly
increases as the voltage of depolarizing burst (VBD) also
increases. This can be interpreted from the aspect of
neural support of suitable output or from the aspect of
neural response to a similar neurosynaptic input. In the
case of output neurosignal analysis, this result implies
that larger and more mature neurosoma is required in
order to support higher AV bursting voltages in response
to depolarizing neural stimulus. Analogously, in the case
of input analysis, the input bursting neurosignal of higher
voltage would induce development of a larger neurosoma.
This law tends to a limit shown as a plateau on figure 12d
for IBN neurons, where VBD is discretized parameter.
This tells us that although stimulative to some extent,
in both cases, high bursting voltage eventually leads to
the exhaustion of neuron electrical abilities, expressed
as shrinkage and eventually death, in extreme cases,
which, although not explicitly shown by extrapolation
of our results, is a well known experimental fact [32-
34]. Naturally, this cannot be taken as the only factor
conglomerate determining neuron body size. Intrinsic,
genetically determined factors are playing the main role,
Figure 19: The graphical presentations of the thus preparing the perikaryon for adequate neurosignal
adequate integrative neuromorphological parameter integration [35]. For every model, presumptions may
(MRFPVINTGR) RSM model response surfaces for the be posed in terms of morphological neuroadaptation
EBN cluster (the adequate IBN model is insignificant).
Annal Behav Neurosci, 2(1): 168-196 (2019) 188
and functional neurostimulation in order to support
and produce an adequate output, respectively. Hence,
the particular shape and complexity of a neuronal
perikaryodendritic compartment is required for
adequate ion channel compartmentalization in order to
produce an adequate output. Alternatively, the neuron is
dynamically shaped in response to neurosynaptic stimuli,
leading to current neuron shape and level of complexity,
with expected further evolution, of course, but under the
government of found law, presented through the given
model.

Figure 22: The graphical presentations of the adequate

integrative neurofunctional parameter/variable
(FNPVINTGR) RSM model response surfaces for the EBN
and IBN clusters.

In general, the strength of neuromorphofunctional

connection may be treated as mild to moderate,
with a few exceptions, characterized by a strong
neuromorphofunctional connection (Tables 5-8).
However, the proved neuromorphological, subtle
Figure 20: Pearson and canonical inter–parameter but significant, quantitative differences between DN
causal predictor–response correlations driven by the clusters are not strong enough to cluster accordingly
adequate response integrative neuromorphological their neurophysiological properties. The functionality of
(MRFPVINTGR) RSM models for the EBN and IBN dentate DN remains undivided, homogenous, thus making the
nucleus neuron clusters. nucleus integrated as a whole at the neurofunctional
level. It appears that neuromorphological differentiation
is necessary condition for such unified function to take
place. This cannot be generalized since an analogous
situation, showed for stellate interneuron subtypes
of the rat cerebral neocortex, revealed different
neurophysiology assigned to different histological
subtypes [16].

Generally speaking, the problem of structure (form)

Figure 21: PLSR/PCR analysis of predictors of a neuron and its function interconnection is still a
determining the integral neurofunctional parameter/ theoretical problem, not yet proved in real experimental
variable (FNPVINTGR) for the EBN and IBN dentate systems. Nevertheless, some glances of elementary
clusters. building blocks that would relate these aspects of neural

Annal Behav Neurosci, 2(1): 168-196 (2019) 189

existence are found on a computational manner. They can compartmental size ratio with or without distribution
be classified into basic structural factors that determine change of ion channels. This is due to the fact that
firing patterns and other physiological properties of various morphologies influence various firing patterns
neurons and neurofunctional ones, modulating neural for the same ion channel distribution [4]. Subtle change
morphology. The first group would include: structure of in diameter causes drastic changes in firing complexities
neuronal compartmentalization, trans compartmental [36]. Large relative dendritic area would induce firing
distribution of ion channels, single compartment ion in bursts, as well as low dendrosomatic conductance
channel essay, absolute value of compartment diameter/ following it, as Uusisaari and Knöpfel showed in their
size, presence of compartmental branching, inter- study [6] as well as our RSM models. Neurocomplexity and
compartmental conductance, inter-compartmental size dendritic branching characteristics would probably affect
ratio (relative compartmental size) and in particular, the physiological properties of a neuron by themselves
dendrite/perikaryon surface/membrane area ratio [2]. directly, while the length and shape parameters would
Synaptic inputs and the tendency of neurons for them, do the same by inducing specific ion channel distribution
with modulating effects upon their neuromorphology and by change in the inter-compartmental conductance.
constitute the second group of factors [2,10]. The specific combinations of these elementary relations,
acting through a single neuromorphological parameter
as well as through their particular combination in an
adequate RSM model, determine the model fitting
properties and the strength of neuromorphofunctional
interconnectedness for a given neuromorphological
response.

Figure 23: Pearson and canonical inter–parameter causal

predictor–response correlations driven by the adequate
response integrative neurofunctional (FNPVINTGR) RSM
models for the EBN and IBN dentate nucleus neuron
clusters.

Now, after considering all of this, the only thing

remained is to decode every of our neuromorphological
parameters in terms of these elementary interrelations
and to view, interpret and explain our models in terms Figure 24: Integrative inter–parameter correlations for
of them. According to this, neuronal size parameters the two DN clusters.
would influence neurofunctional properties, by
affecting the compartment diameter property, i.e. inter- Using the same template, the interpretation of the

Annal Behav Neurosci, 2(1): 168-196 (2019) 190

response neurofunctional parameters in terms of the
neuromorphological ones is a little bit different, due to
the fact of simultaneous existence of indirect influence.
Namely, the neuromorphological modulation influenced
by neuron function can be partly comprehended as direct
product of synaptic input stimulation but also as a need of
a neuron for suitable morphology, in order to obtain the
desired functionality. The second is probably genetically
coded and takes place predominantly during the period of
neural embryonic development [35]. Naturally, in neural
networks there are chains of this interplay: morphology
determines the function and then through synaptic
coupling this function influences morphology, which
completes the circle of interplay [2,4,37,38]. According
to these studies, it appears that a neural network, when
exposed to learning of some new information, relies
only on its function and thus compensates the lack of
adequate morphology. However, if still exposed to the
same stimuli, it modulates its morphology accordingly
and thus completes the learning process. It should
keep in mind that these are just partial, not completely
deterministic, stochastic relationships, which justify the
obtained adjusted-R2 values, never reaching the value of
100%.

Figure 26: Differences between neurotransmitter DN

neuron types in terms of complexity parameters.

The semi-theoretical experimentation with the

dentatostriate inter-cluster networks and mapping of
these connections revealed that these interactions are
weakening between remote neurons with more neurons
inserted in the neural chain. The influence of one neuron
morphology and/or function to a distanced neuron is thus
negligible as our RSM mapping reveals. Hence, this kind
of network can be treated as symmetrically coupled on a
Hopfield manner, without real insight in its configuration
but the functionality of such network resembles and is
reminiscent to a symmetric one.

In conclusion, it can be said that neuronal morphology

and function are definitely interrelated and depend on
each other. Although this is an irrefutable, very well
proved and known fact, now proved once again by our
results, by intensity, however, this interconnectedness
can be treated only as mild to moderate. Morphology
and function of a neuron, have a high level of self-
Figure 25: Neurodynamics of activity type of DN sustained autonomy (independence). It is determined by
neurons in terms of the neuromorphological predictors. elementary neuromorphofunctional relations, observed
at the macroscopic, phenomenological level, i.e. only
through measured parameters as their observable

Annal Behav Neurosci, 2(1): 168-196 (2019) 191

and explicit manifestation without considering the Serbia, project number III41031.
microscopic, molecular (ion channel) causality of them.
They show that almost every aspect of morphology and/
or function plays some role in determination of the other
one. These relations are the strongest when acting upon
a single neuron and their mutual remote influence on
each other weakens in neural circuits and networks up to
10% of deterministic relational interconnection strength
observed in relations at the single neuron level.

Figure29: Neuromorphofunctional correlation–

comparison analysis plots.

Appendix
List of abbreviations and explanations

• ADF: surface area of a neuron dendritic arborization

field, measures the size of a minimal region occupied
by the neuron dendritic tree in a 2D space after
removing the neurosoma
• ADT: surface area of a dendritic tree, measures the
Figure 27: The general interplay network of dendritic arborization size of the neuron
neuromorphofunctional relations obtained by the RSM • AN: surface area of a neuron i.e. the image area
modeling. occupied by the entire neuron 2D binary image,
measures the size of the neuron
• ANF: surface area of a neuron field, measures the
size of a minimal region occupied by the neuron in
a 2D space
• ANN: artificial neural network
• AP/AV: action potential
• APNS: surface area of a neuron perineuronal space,
measures the size of a total minimal region between
dendrites
• AS: surface area of a neurosoma, measures the soma
Figure 28: Unsupervised (PCA) and supervised
size of the neuron
(ANN) classification of DN neurons according to
• AVRGVBH: average bursting volatage of
neurofunctional parameters.
hyperpolarization-induced AP
• AVRGVH: total average voltage of hyperpolarization-
Conflict of Interests Statement
induced AP
The authors declare that there is no conflict of interests • AVRGVHD: average voltage of afterhyperpolarization
regarding the publication of this paper. of depolarization-induced AP
• AVRGVHH: average afterhyperpolarization volatage
Acknowledgments of hyperpolarization-induced AP
This study was supported by the Ministry of Education, • AVRGVTBH: average bursting volatage treshold of
Science and Technological Development, Republic of hyperpolarization-induced AP

Annal Behav Neurosci, 2(1): 168-196 (2019) 192

• BNI2D: 2D binary neuron image space i.e. the orientation of its dominant elongation
• CDF: complexity of dendritic field, as a complexity axis
parameter, it measures the total curvature and detail • DS: fractal dimension of skeletonized neuron image,
image complexity of the dendritic arborization, the represents one of neuron complexity parameters
most integrative of all complexity parameters • DSI: dendritic surface of influence, dendritic
• CDF/ADFst: standardized CDF parameter with respect dispersion index/degree of a neuron i.e.
to the ADF parameter multidirectional dendritic radiation index/degree
• CDF/DCBOst: standardized CDF parameter with or dendritic radiation degree/index in all directions,
respect to the DCBO parameter represents the measure of dendritic dispersion in a
• CDF/NDALLst: standardized CDF parameter with 2D space obtained as a covariance of dendritic space
respect to the NDALL parameter coordinates
• CDF/NHODst: standardized CDF parameter with • DSP: partial dendritic surface, represents a measure
respect to the NHOD parameter of dendritic non-complex density
• CDF/NPDst: standardized CDF parameter with • DSPst: standardized DSP parameter with respect to
respect to the NPD parameter the ADF parameter
• CDF/NRst: standardized CDF parameter with respect • DTAUIPSP: constant time of decay of inhibitory
to the NR parameter postsynaptic potential
• CDF/NSDst: standardized CDF parameter with • DWDTH: average dendritic width of a neuron
respect to the NSD parameter • DWDTHst: standardized DWDTH parameter with
• CDF/NTDst: standardized CDF parameter with respect to the AS parameter
respect to the NTD parameter • EBN: external border neurons of the DN
• CDF/NTHODst: standardized CDF parameter with • FB: bursting frequency
respect to the NTHOD parameter • FBD: bursting frequency of depolarization-induced
• CRSKEL: circularity of neuron shape measured on AP
a skeletonized 2D binary neuron image, represents • FBH: burst frequency of hyperpolarization-induced
an intensity of shape roundness and geometrical AP
compactness, closeness to circle shape, for • FIPSP: frequency of inhibitory postsynaptic potential
skeletonized image far away from circle/low value • FS: frequency of spontaneous AP
(<1), the auxiliary parameter used for MDCBO • GABA/GAD-L: large GABA-ergic neurons
calculation • GABA/GAD-SIOP: small inferior olive projecting
• DAX: neuron axialization index, PCA parameter, GABA-ergic neurons
quantifies how much a neuron is elongated in a 2D • Glu-L: large Glu-ergic neurons
space, far mostly on account of the dendritic tree • Glu-S: small Glu-ergic neurons
elongation • Gly-LI: large intrinsically inactive Gly-ergic neurons
• DCBO: dendritic centrifugal branching order, • IBN: internal border neurons of the DN
represents average dendritic branching order • INTGRFNPV: integrative representative
• DCBOst: standardized DCBO parameter with respect neurofunctional parameter/variable
to the NMIN parameter • INTGRMRFPV: integrative representative
• DCLD: dendritic clustering degree/index, measures neuromorphological parameter/variable
the tendency of neuron dendritic clustering • IPSC: intensity of inhibitory postsynaptic current
• DN: fractal dimension of a neuron, one of neuron • L: total dendritic length, represents the sum of all
shape parameters, quantifies the shape of the individual dendritic lengths
neuron from one aspect of its observation (the entire • Lst: standardized L parameter with respect to the AS
image of it) or cerebellar dentate nucleus (context parameter
specific) • MAXVBH: maximal bursting volatage of
• DO: fractal dimension of a neuron outline, one of hyperpolarization-induced AP
neuron shape parameters, quantifies the shape • MAXVH: total maximal voltage of hyperpolarization-
of the neuron from one aspect of its observation induced AP
(outline) • MAXVHD: maximal voltage of afterhyperpolarization
• DPOL: dendritic orientation degree, expressed in of depolarization-induced AP
angle degrees, measures dendritic orientation in a 2D • MAXVHH: maximal afterhyperpolarization volatage

Annal Behav Neurosci, 2(1): 168-196 (2019) 193

 of hyperpolarization-induced AP dendrites, which a theoretical circle makes centered
• MAXVTBH: maximal bursting volatage treshold of in geometrical center of the neurosoma, represents
hyperpolarization-induced AP one of neuron complexity parameters, it can be
• MDCBO: dendritic/branching polarization index, thought of as a minimal complexity of the particular
measures how much dendritic tree is asymmetric on neuron
account of its average branching degree • NPD: number of the first order/primary dendrites i.e.
• MINVBH: minimal bursting volatage of dendrites originating directly from the neurosoma
hyperpolarization-induced AP • NR: total number of dendrites above the first order
• MINVH: total minimal voltage of hyperpolarization- • NSD: number of the second order dendrites i.e.
induced AP dendrites originating directly from the first order
• MINVHD: minimal voltage of afterhyperpolarization dendrites
of depolarization-induced AP • NTD: number of the third order dendrites i.e.
• MINVHH: minimal afterhyperpolarization volatage of dendrites originating directly from the second order
hyperpolarization-induced AP dendrites
• MINVTBH: minimal bursting volatage treshold of • NTHOD: total number of dendrites above the second
hyperpolarization-induced APMN - index of neuron order
asymmetry, as a reciprocal value of neuron circularity, • PCA: principal component analysis
represents an intensity of shape non-roundness and • PCd1: first principal component i.e. one of coordinate
geometrical irregularity, difference from circle in axes along the maximal variation of data
geometrical sense and from its symmetry • PCd1x: x component of PCd1 (with respect to
• MS: index of neurosoma asymmetry, as a reciprocal Cartesian coordinate system using vector approach)
value of neurosoma circularity, represents an • PCR - principal component regression
intensity of shape non-roundness and geometrical • Perikaryodendritic compartment of a neuron -
irregularity, difference from circle in geometrical neurosoma/perikaryon with the dendrites
sense and from its symmetry • PLSR: partial least-squares regression
• MSKEL - index of asymmetry of neuron shape • PRN: perimeter of a neuron perikaryodendritic
measured on a skeletonized 2D binary neuron compartment, the auxiliary parameter used for MN
image, as a reciprocal value of CRSKEL, represents an calculation
intensity of shape non-roundness and geometrical • PRS: perimeter of a neurosoma, the auxiliary
irregularity, difference from circle in geometrical parameter used for MS calculation
sense and from its symmetry, for skeletonized image • RB: radius of a circle with the maximal number of
far away from circle/large value (>1), the auxiliary dendritic branching points
parameter used for MDCBO calculation • RC: radius of a circle with maximal number of
• NDALL: total number of all dendrites of a neuron intersections with dendrites (NMAX circle), measures
• Neurosoma: neuron/cell body, perikaryon or just a distance of maximal dendritic complexity from
soma, here in the text these terms are used on an equal neurosoma representing that way the position of
basis but they are not equal to perikaryodendritic maximal dendritic arborization density
compartment of a neuron since it also includes the • RCst: standardized RC parameter with respect to the
dendrites AS parameter
• NHOD: number of the higher order dendrites i.e. • RSM: response surface methodology
dendrites originating directly from the third and • RT: radius of a circle with the maximal number of
higher order dendrites terminal dendrites
• NMAX: maximal number of intersections with • RTAUIPSP: constant time of rise of inhibitory
dendrites, which a theoretical circle makes centered postsynaptic potential
in geometrical center of the neurosoma, represents • VB: bursting voltage
one of neuron complexity parameters, it can be • VBD: bursting volatage of depolarization-induced AP
thought of as a maximal complexity of the particular • VD: total voltage of depolarization-induced AP
neuron • VDD: voltage of depolarizing phase of depolarization-
• NMAXst: standardized NMAX parameter with respect induced AP
to the NMIN parameter • VFHS: voltage of fast phase of afterhyperpolarization
• NMIN: minimal number of intersections with of spontaneous AP

Annal Behav Neurosci, 2(1): 168-196 (2019) 194

• VHS: total voltage of afterhyperpolarization of 10. Grbatinić I, Milošević N. Classification of adult human
spontaneous AP dentate nucleus border neurons: Artificial neural
• VS: volatage of spontaneous AP networks and multidimensional approach. J Theor
• VSHS: voltage of slow phase of afterhyperpolarization Biol. 2016;404:273-284. Doi: https://doi.org/10.1016/j.
of spontaneous AP jtbi.2016.06.011
• VTBD: bursting volatage treshold of depolarization- 11. Grbatinić I, Marić DL, Milošević NT. Neurons from the
induced AP adult human dentate nucleus: Neural networks in the
• WBH: bursting width/duration of hyperpolarization- neuron classification. J Theor Biol. 2015;370:11-20.
induced AP Doi: https://doi.org/10.1016/j.jtbi.2015.01.024
• WS: width/duration of spontaneous AP
12. Grbatinić I, Milošević N, Krstonošić B. The
neuromorphological caudate-putaminal clustering
References
of neostriate interneurons: Kohonen self-organizing
1. Ramón Y, Cajal R. The Croonian lecture: La fine maps and supervised artificial neural networks with
structure des centres nerveux. Proc Roy Soc Lond multivariate analysis. J Theor Biol. 2018;438:96-115.
[Biol]. 1894;55(331-335):444-468. Doi: https://doi. Doi: https://doi.org/10.1016/j.jtbi.2017.11.013
org/10.1098/rspl.1894.0063 13. Bostan AC, Strick PL. The cerebellum and basal ganglia
2. Connors BW, Regehr WG. Neuronal firing: Does are interconnected. Neuropsychol Rev. 2010;20(3):261-
function follow form? Curr Biol. 1996;6(12):1560-1562. 270. Doi: https://doi.org/10.1007/s11065-010-9143-9
Doi: https://doi.org/10.1016/S0960-9822(02)70771-9 14. Habas C. Functional imaging of the deep cerebellar
3. Llinas R. The intrinsic electrophysiological properties nuclei: A review. Cerebellum. 2010;9(1):22-28. Doi:
of mammalian neurons: Insights into central nervous https://doi.org/10.1007/s12311-009-0119-3
system function. Science. 1988;242(4886):1654-1664. 15. Middleton F. Basal ganglia and cerebellar loops:
Doi: https://doi.org/10.1126/science.3059497 Motor and cognitive circuits. Brain Res Brain Res Rev.
4. Mainen ZF, Sejnowski TJ. Influence of dendritic 2000;31(2-3):236-250. Doi: https://doi.org/10.1016/
structure on firing pattern in model neocortical S0165-0173(99)00040-5
neurons. Nature. 1996;382(6589):363-366. Doi: 16. Kawaguchi Y, Kubota Y. Correlation of physiological
https://doi.org/10.1038/382363a0 subgroupings of nonpyramidal cells with parvalbumin-
5. Pinsky PF, Rinzel J. Intrinsic and network and calbindinD28k-immunoreactive neurons in layer
rhythmogenesis in a reduced traub model for CA3 V of rat frontal cortex. J Neurophysiol. 1993;70(1):387-
neurons. J Comput Neurosci. 1994;1(1-2):39-60. Doi: 396. Doi: https://doi.org/10.1152/jn.1993.70.1.387
https://doi.org/10.1007/BF00962717 17. Schierhorn VH, Doedenes K, Nagel I. Quantitative
6. Uusisaari M, Knöpfel T. Functional classification studies on the comparability of neurohistological
of neurons in the mouse lateral cerebellar nuclei. results in rat cortical pyramids produced by different
Cerebellum. 2011;10(4):637-646. Doi: https://doi. Golgi methods (author’s transl). J Hirnforsch.
org/10.1007/s12311-010-0240-3 1977;18(5):423-429.
7. Deitcher Y, Eyal G, Kanari L, et al. Comprehensive 18. Grbatinić I, Milošević N, Krstonošić B. The
morpho-electrotonic analysis shows 2 distinct classes neuromorphological caudate-putaminal clustering
of L2 and L3 pyramidal neurons in human temporal of neostriate interneurons: Kohonen self-organizing
cortex. Cereb Cortex. 2017;27(11):5398-5414. Doi: maps and supervised artificial neural networks with
https://doi.org/10.1093/cercor/bhx226 multivariate analysis. J Theor Biol. 2018;438:96-115.
8. Elbasiouny SM. Development of modified cable Doi: https://doi.org/10.1016/j.jtbi.2017.11.013
models to simulate accurate neuronal active 19. Milošević NT, Ristanović D, Marić DL, et al. Morphology
behaviors. J Appl Physiol. 2014;117(11):1243-1261. and cell classification of large neurons in the adult
Doi: https://doi.org/10.1152/japplphysiol.00496.2014 human dentate nucleus: A quantitative study. Neurosci
9. Iannella N, Launey T, Abbott D, et al. A nonlinear Lett. 2010;468(1):59-63. Doi: https://doi.org/10.1016/j.
cable framework for bidirectional synaptic plasticity. neulet.2009.10.063
Phillips W, ed. PLoS ONE. 2014;9(8):e102601. Doi: 20. Chan-Palay V. Cerebellar dentate nucleus:
https://doi.org/10.1371/journal.pone.0102601 Organization, cytology and transmitters. 1977.
https://www.springer.com/in/book/9783642665004.
Annal Behav Neurosci, 2(1): 168-196 (2019) 195
21. Yelnik J, Percheron G, François C, et al. Principal and ridge analyses. 2nd ed. Hoboken, N.J: John Wiley;
component analysis: A suitable method for the 2007.
3-dimensional study of the shape, dimensions and 31. Atkinson AC, Donev AN, Tobias RD. Optimum
orientation of dendritic arborizations. J Neurosci experimental designs, with SAS. Oxford: Oxford
Method. 1983;9(2):115-125. Doi: https://doi. University Press; 2007. https://global.oup.com/
org/10.1016/0165-0270(83)90125-5 academic/product/optimum-experimental-designs-
22. Ristanović D, Krstonošić B, Milošević NT, et al. with-sas-9780199296606?cc=in&lang=en&
Mathematical modelling of transformations of 32. Albayrak S, Zhao Q, Siesjö BK, et al. Effect of transient
asymmetrically distributed biological data: An focal ischemia on blood-brain barrier permeability in
application to a quantitative classification of spiny the rat: correlation to cell injury. Acta Neuropathologica.
neurons of the human putamen. J Theor Biol. 1997;94(2):158-163. Doi: https://doi.org/10.1007/
2012;302:81-88. Doi: https://doi.org/10.1016/j. s004010050688
jtbi.2012.02.027
33. Bartus RT, Dean RL, Mennerick S, et al. Temporal
23. Ristanović D, Milosević NT, Stefanović BD, et al. ordering of pathogenic events following transient
Morphology and classification of large neurons global ischemia. Brain Res. 1998;790(1-2):1-13. Doi:
in the adult human dentate nucleus: A qualitative https://doi.org/10.1016/S0006-8993(97)01414-5
and quantitative analysis of 2D images. Neurosci
34. Bidmon HJ, Oermann E, Schleicher A, et al. Copper-
Res. 2010;67(1):1-7. Doi: https://doi.org/10.1016/j.
zinc superoxide dismutase and isolectin B4 binding
neures.2010.01.002
are markers for associative and transhemispheric
24. Pantic I, Harhaji-Trajkovic L, Pantovic A, et al. Changes diaschisis induced by focal ischemia in rat cortex.
in fractal dimension and lacunarity as early markers Neurosci Lett. 1997;228(3):163-166. https://www.ncbi.
of UV-induced apoptosis. J Theor Biol. 2012;303:87-92. nlm.nih.gov/pubmed/9218633
Doi: https://doi.org/10.1016/j.jtbi.2012.03.013
35. Kasper EM, Lübke J, Larkman AU, et al. Pyramidal
25. Ristanović D, Milošević NT, Stefanović IB, et al. Cell neurons in layer 5 of the rat visual cortex. III.
image area as a tool for neuronal classification. J Differential maturation of axon targeting, dendritic
Neurosci Methods. 2009;182(2):272-278. Doi: https:// morphology, and electrophysiological properties. J
doi.org/10.1016/j.jneumeth.2009.06.004 Comp Neurol. 1994;339(4):495-518. Doi: https://doi.
26. Koós T, Tepper JM. Inhibitory control of neostriatal org/10.1002/cne.903390404
projection neurons by GABAergic interneurons. 36. Moore JW, Westerfield M. Action potential propagation
Nat Neurosci. 1999;2(5):467-472. Doi: https://doi. and threshold parameters in inhomogeneous regions
org/10.1038/8138 of squid axons. J Physiol. 1983;336(1):285-300. Doi:
27. Tepper JM, Bolam JP. Functional diversity and https://doi.org/10.1113/jphysiol.1983.sp014581
specificity of neostriatal interneurons. Curr Opin 37. Chklovskii DB. Synaptic connectivity and neuronal
Neurobiol. 2004;14(6):685-692. Doi: https://doi. morphology: Two sides of the same coin. Neuron.
org/10.1016/j.conb.2004.10.003 2004;43(5):609-617. Doi: https://doi.org/10.1016/j.
28. Tepper JM, Koós T, Wilson CJ. GABAergic microcircuits neuron.2004.08.012
in the neostriatum. Trends Neurosci. 2004;27(11):662- 38. Stiefel KM, Sejnowski TJ. Mapping function onto
669. Doi: https://doi.org/10.1016/j.tins.2004.08.007 neuronal morphology. J Neurophysiol. 2007;98(1):513-
29. Krishna SA, Orr JA, Westenskow DR. Classification of 526. Doi: https://doi.org/10.1152/jn.00865.2006
red blood cell images using a neural network. Proc
Annu Symp Comput Appl Med Care. 1994:975. https://
www.ncbi.nlm.nih.gov/pubmed/7950079
30. Box GEP, Draper NR. Response surfaces, mixtures,

Copyright: © Grbatinić et al. This is an Open Access article distributed under the terms of
the Creative Commons Attribution License, which permits unrestricted use, distribution, and
reproduction in any medium, provided the original work is properly cited.

Annal Behav Neurosci, 2(1): 168-196 (2019) 196