Article

ISSN 0006-2979, Biochemistry (Moscow), 2026, Vol. 91, No. 6, pp. 979-991 © Pleiades Publishing, Ltd., 2026.

979

Metabolic Status and Morphological Changes

of Astrocytes under Reductive Stress invitro:

Neural Network Analysis

Svetlana V. Novikova

, Nataliya A. Kolotyeva

1,a

, Arseniy K. Berdnikov

Taisia E. Shcheveleva

, Ivan V. Simkin

, Yulia K. Komleva

, Egor V. Yakovlev

Nikita P. Kryuchkov

, Stanislav O. Yurchenko

, and Alla B. Salmina

Russian Center of Neurology and Neurosciences, 125367 Moscow, Russia

Bauman Moscow State Technical University, 105005 Moscow, Russia

e-mail: kolotyeva.n.a@neurology.ru

Received January 30, 2026

Revised June 4, 2026

Accepted June 14, 2026

Abstract— Reductive stress caused by excessive accumulation of reducing equivalents (NADH, NADPH, gluta-

thione), is increasingly recognized as a pathogenetic factor at the early stages of neurodegenerative diseases.

However, its impact on the morphofunctional properties of astrocytes remains poorly understood. We per-

formed the first comprehensive quantitative assessment of biochemical and morphological changes in cul-

tured primary rat astrocytes under chronic reductive stress modeled using dithiothreitol (DTT, 500μM,  24  h).

To enable morphometric analysis of live, unstained cells, we developed and trained a semantic segmentation

model based on the YOLOv11 architecture, which enables objective assessment of complex branched cell

morphology in phase-contrast images while avoiding fixation-related artifacts. DTT-induced reductive stress

caused a significant increase in the total pool of nicotinamide coenzymes (p  <  0.01) and elevated mito-

chondrial superoxide production without compromising mitochondrial membrane potential. Morphometric

analysis revealed a sustained enlargement of astrocyte soma area and increase in the number of astrocyte

processes, along with a reduced branching complexity. These findings provide new insights into the role of

redox imbalance in regulating glial function and may expand our understanding of the early mechanisms

of neurodegenerative diseases.

DOI: 10.1134/S0006297926600274

Keywords: metabolic plasticity, astrocytes, dithiothreitol (DTT), reductive stress, deep machine learning

* To whom correspondence should be addressed.

# These authors contributed equally to this study.

INTRODUCTION

In recent years, metabolic plasticity – the ability

of neural, glial, and endothelial cells in the neuro-

vascular unit to flexibly reorganize their energy me-

tabolism in response to changing functional demands

and external conditions – has gained recognition as

a key concept in neuroscience  [1,  2]. The metabolic

phenotype of central nervous system (CNS) cells is

highly dynamic rather than static, undergoing con-

tinuous modifications throughout ontogenesis and in

response to variations in synaptic activity, as well

as during the progression of neurodegenerative dis-

orders, neuroinflammatory processes, and ischemic

damage  [3,  4]. The development of relevant cellular

models of CNS disorders is critical for advancing new

technologies for disease prevention and treatment  [5].

Moreover, such models are used in regenerative med-

icine and provide valuable platforms for the discov-

ery, development, and screening of new therapeutic

agents, including compounds capable of crossing the

blood–brain barrier  [6].

Astrocytes, the principal components of the glial

network in the CNS, play a key role in maintaining

brain homeostasis by providing metabolic support

NOVIKOVA et al.980

BIOCHEMISTRY (MOSCOW) Vol. 91 No. 6 2026

to neurons, regulating extracellular ion composition,

and participating in detoxification processes and for-

mation of functional neurovascular units. For several

decades, oxidative stress has been considered a major

contributor to aging and neurodegenerative disorders.

However, a growing body of evidence indicates that

excessive reduction of the intracellular environment,

known as reductive stress, also represents an import-

ant pathogenetic factor in disease development  [7].

Reductive stress is characterized by the excessive ac-

cumulation of reducing equivalents, including NADH,

NADPH, and reduced glutathione (GSH) [8, 9]. Under

physiological conditions, redox pairs NAD

/NADH,

NADPH/NADP

, and GSH/GSSG regulate numerous

metabolic and signaling pathways, thereby maintain-

ing cellular redox homeostasis  [10]. Importantly, re-

ductive stress does not occur independently of oxida-

tive stress. Rather, these two states may precede each

other, coexist, or alternate, generating a complex cy-

cle of mutually reinforcing redox dysregulation that

contributes to the early stages of neurodegeneration

and cell aging in the brain  [11,  12].

Several studies have demonstrated a relationship

between the morphofunctional state of cells and dy-

namics of cellular redox homeostasis. The balance be-

tween oxidative and reductive processes ensures the

maintenance of basic cellular functions and directly

influences cell proliferation, differentiation, migra-

tion, and the expression of structural and signaling

proteins  [13,  14]. Currently, the impact of reductive

stress on the morphofunctional properties of astro-

cytes remains largely unexplored. Given the high met-

abolic plasticity of these cells, their response to redox

imbalance may reflect both adaptive and pathological

processes.

In  vitro, reductive stress is commonly modeled

using exogenous reducing agents. One of the most

widely used compounds is dithiothreitol (DTT). The

molecular mechanism of DTT action is based on its

ability to reduce disulfide bonds with the generation

of free sulfhydryl groups, which leads to the disrup-

tion of proper protein folding in the endoplasmic re-

ticulum, accumulation of misfolded proteins, and acti-

vation of unfolded protein response(UPR). In addition

to inducing ER stress-related signaling pathways, DTT

shifts the cellular redox equilibrium toward a more

reduced state [13, 15-17].

Conventional approaches to astrocyte morpholog-

ical analysis typically rely on cell fixation and im-

munocytochemical staining, i.e., procedures that can

alter cellular morphology. Phase-contrast microsco-

py offers a non-invasive alternative and enables re-

al-time visualization of live cells. However, automated

analysis of phase-contrast images is challenging due

to the complex morphology of astrocytes and inher-

ently low contrast of such images. Recent advances

in deep learning, particularly convolutional neural

networks (CNNs), have significantly improved the

accuracy of biological image segmentation  [18,  19].

Here, we developed an enhanced approach for the

automated segmentation of astrocytes in phase-con-

trast images using the YOLOv11 (You Only Look Once

version  11) architecture, which enabled objective,

high-throughput morphometric analysis and facilitat-

ed the detection of subtle phenotypic changes that

may not be discernible with traditional microscopy

techniques.

The aim of this study was to investigate biochem-

ical changes and provide a quantitative characteriza-

tion of morphological rearrangements in cultured rat

astrocytes under DTT-induced reductive stress.

MATERIALS AND METHODS

Isolation and culturing of primary astrocytes.

Primary astrocyte cultures were prepared from the

brains of neonatal Wistar rats (postnatal days  0-2)

using an adapted protocol for enzymatic/mechanical

cell dissociation as described in [20,  21]. Briefly, cor-

tical tissue was isolated, rinsed in calcium- and mag-

nesium-free phosphate-buffered saline (PBS; pH  7.4;

Gibco, USA), minced into small fragments with a

scalpel, and incubated in 0.05% trypsin/0.02% EDTA

solution (Gibco) for 15  min at 37°C. Following enzy-

matic digestion, the tissue was washed twice with

PBS and once with complete culture medium to inac-

tivate trypsin. Cells were then mechanically dissociat-

ed by gentle pipetting in culture medium. Complete

culture medium consisted of 90% Minimum Essential

Medium Eagle (MEM; Gibco), 10% fetal bovine serum

(FBS; HyClone, USA), 2  mM GlutaMAX (Gibco), and

10  mM HEPES (Sigma-Aldrich, USA), pH  7.2-7.4.

The resulting cell suspension was centrifuged

at 100g for 3  min. The pellet was resuspended in

complete medium, and cells were seeded into two

T-25 culture flasks. Cell number and viability were

determined using a Countess automated cell count-

er (Invitrogen, USA). Cultures were maintained in a

humidified CO

incubator (RWD Life Science, China)

at37°C in 5%  CO

at98% relative humidity. To enrich

the astrocyte population and reduce contamination

by microglia and oligodendrocytes, cells were washed

with prewarmed PBS, detached using trypsin–EDTA,

resuspended in culture medium, and replated into

poly-L-lysine-coated 96-well plates (2-5×10

  cells/well)

and 6-well plates (2-4×10

  cells/well) (Servicebio,

China). All experiments were performed using

first-passage cultures after they reached 90-95% con-

fluence. Cell morphology and growth were moni-

tored with an Olympus CKX41 inverted microscope

(Olympus, Japan).

METABOLIC STATUS AND MORPHOLOGICAL CHANGES OF ASTROCYTES UNDER REDUCTIVE STRESS 981

BIOCHEMISTRY (MOSCOW) Vol. 91 No. 6 2026

Modeling reductive stress in astrocytes. Re-

ductive stress was induced on days 4-5 after pas-

saging, when astrocyte cultures had reached 90-95%

confluence and formed a uniform monolayer. One

hour prior to treatment, cells were transferred to

serum-free medium consisting of 90% MEM, 2  mM

GlutaMAX, and 10  mM HEPES, pH  7.2-7.4. Reductive

stress was induced with the low-molecular-weight re-

ducing agent dithiothreitol (DTT; Sigma-Aldrich) at a

final concentration of 500  μM for 24  h at 37°C  [22].

The control group was subjected to the same proce-

dure except DTT solution was replaced with an equal

volume of distilled water (vehicle). The selected DTT

concentration (500  μM) and exposure duration (24  h)

were based on published studies and preliminary

observations demonstrating effective induction of

reductive stress without causing immediate cytotox-

icity. Previous studies have shown that acute (≤1  h)

and prolonged (≥3-24  h) reductive stress trigger fun-

damentally different transcriptional and translational

responses in  vitro  [15]. The use of serum-free medi-

um was a critical methodological requirement be-

cause serum proteins, particularly albumin and oth-

er thiol-containing molecules, can rapidly bind and

neutralize DTT [22,  23]. Maintaining both control and

DTT-treated cultures under identical serum-free con-

ditions enabled evaluation of the specific contribution

of DTT-induced reductive stress.

Assessment of total NAD

/NADH content. Total

NAD

/NADH content was determined using a commer-

cial NAD

/NADH Colorimetric Assay Kit (E-BC-K804-M;

Elabscience, China) with an Inno-S plate spectropho-

tometer (LTek, South Korea) at 450  nm. Prior to analy-

sis, cells were washed three times with prewarmed

PBS. Protein concentration in cell lysates was deter-

mined using the bicinchoninic acid (BCA) assay with

a commercial kit (E-BC-K318-M; Elabscience, China).

Total NAD

/NADH content was calculated from a stan-

dard calibration curve and normalized to the protein

content in each sample.

Assessment of mitochondrial membrane po-

tential. Mitochondrial membrane potential (ΔΨm)

in astrocyte cultures exposed to reductive stress and

in control cells was assessed using the tetramethyl-

rhodamine ethyl ester (TMRE) assay. TMRE (Abcam,

UK) was added to the culture medium to a final con-

centration of 500  nM, and cells were incubated for

15  min at 37°C in the dark according to the manu-

facturer’s protocol. To verify signal specificity and

complete mitochondrial membrane depolarization,

additional subgroup of control cells was treated with

20  μM FCCP for 10  min at 37°C in a CO

incubator.

Following incubation, cells were washed three times

with Locke’s solution to remove excess dye. Fluores-

cence intensity was measured using a SpectraMax

microplate reader (Molecular Devices, USA) con-

trolled by SoftMax Pro software, with excitation and

emission at 549 and 575  nm, respectively. Intracellu-

lar localization of TMRE was confirmed by fluores-

cence imaging using the EVOS M7000 Imaging System

(Thermo Fisher Scientific, USA). For statistical analy-

sis, mean values from several biological replicates

(n  =  5 independent experiments) were used. Each

experiment included 20 technical replicates for the

DTT-treated group and 20 technical replicates for the

control group. TMRE fluorescence intensity was ex-

pressed relative to the control group, which was set

to 100%. Statistical significance was evaluated using

the one-sample Wilcoxon signed-rank test.

Mitochondrial superoxide production in

live cells was assessed using the fluorescent probe

MitoSOX Green (Invitrogen, USA) according to the

manufacturer’s instructions. Cells were incubated

with MitoSOX Green at a final concentration of 1  μM

for 30  min at 37°C in a humidified CO

incubator.

Next, cells were washed three times with Hank’s Bal-

anced Salt Solution (HBSS) containing calcium and

magnesium. Fluorescence intensity was measured

using a SpectraMax microplate reader with exci-

tation and emission at 488 and 510  nm, respectively.

Fluorescence values were normalized to the mean

fluorescence intensity of the corresponding control

group in each plate and expressed as a percentage

of control.

Immunofluorescence staining of GFAP. Astro-

cytes were fixed with 4% paraformaldehyde (Sigma-

Aldrich, Germany) for 15  min at room temperature

and permeabilized with 0.1%  Triton X-100 (Calbio-

chem, USA) for 10  min at 4°C. Non-specific binding

sites were blocked by incubation with 5%  goat serum

(Sigma-Aldrich) for 60 min. Next, cells were incu-

bated with primary rabbit anti-glial fibrillary acidic

protein (GFAP) antibodies (dilution, 1  :  750; #DF6040,

Affinity Biosciences, USA) for 6  h at 4°C in a humidi-

fied chamber. Both primary and secondary antibodies

were diluted in IHC Diluent buffer (Leica Biosystems).

After five washes with PBS, cells were incubated with

secondary FITC-conjugated goat anti-rabbit IgG (dilu-

tion, 1  :  350; #SAA544Rb18, Cloud-Clone Corp., USA)

for 2  h at room temperature in the dark. Cells were

subsequently washed twice with PBS and mounted

under a coverslip using Fluoroshield mounting me-

dium containing 4’,6-diamidino-2-phenylindole dihy-

drochloride (DAPI, Sigma-Aldrich) for 15  min. Nega-

tive-control samples were processed identically except

for omission of the primary antibody. Fluorescence

images were acquired using an EVOS  M7000 imaging

system (Thermo Fisher Scientific). Quantitative image

analysis was performed with ImageJ software (ver-

sion  1.47; National Institutes of Health, USA). GFAP

expression was quantified as the integrated fluores-

cence intensity in the green channel and normalized

NOVIKOVA et al.982

BIOCHEMISTRY (MOSCOW) Vol. 91 No. 6 2026

to the number of DAPI-positive nuclei within the cor-

responding field of view.

Phase-contrast microscopy. Images were ob-

tained using an automated EVOS  M7000 Imaging Sys-

tem equipped with a built-in environmental chamber

to maintain physiological culture conditions (37°C,

5.0%  CO

, and >80% humidity) (Fig.  1b). For each

well, a randomly selected 5×5 area was captured at

40× magnification. Each frame had a resolution of

2048×1536 pixels, corresponding to a field of view of

320×240  μm. To generate a neural network training

dataset, 439 phase-contrast images of live astrocytes

were used, including 221 images from previously

published sources  [24] and 218 newly acquired imag-

es. Primary astrocytes were seeded into 96-well plates

and imaged daily over a 10-day period. This approach

enabled the capture of different stages of cell growth

and proliferation while ensuring biological diversity

within the training dataset. Images acquired for the

analysis of reductive stress-induced morphological

changes were excluded from model training. Spe-

cifically, 98 images from the control group and 200

images from the DTT-treated group were collected

on day  5 of culturing. At this time point, cells had

formed a stable monolayer with 90-95% confluency,

providing optimal conditions for modeling reductive

stress.

Data preparation and annotation were carried

out using the LabelMe software  [25]. Annotation was

performed for a dataset of phase-contrast images of

live astrocytes. Only astrocytes that were fully con-

tained within the image frame and did not overlap

with other objects were annotated. The annotated

dataset was randomly divided into training (75%),

validation (15%), and test (10%) subsets. Images from

the DTT-treated and control groups were not annotat-

ed and were analyzed exclusively using the trained

model.

Extraction of astrocytes morphological charac-

teristics. Morphological feature extraction was per-

formed using a multi-step image analysis pipeline.

Following cell segmentation, the binary mask of each

astrocyte was used to identify the soma and cellu-

lar processes. The soma region was isolated using an

anisotropic Gaussian filter, while the remaining mask

area was classified as processes. To characterize the

process architecture, binary masks were skeletonized

and converted into branching graphs using a depth-

first search algorithm. Process length was calculated

as the number of pixels comprising each skeleton

branch. Branching nodes were identified through

analysis of a 3×3 pixel neighborhood; a pixel was

classified as a branching node when more than three

neighboring pixels belonged to the skeleton. For each

segmented cell, the following morphological param-

eters were automatically quantified: total cell area,

soma area, total process length, number of processes,

and number of branching nodes.

Statistical analysis. Biochemical data were

analyzed and visualized using GraphPad Prism  10

(GraphPad Software, USA) and the programming R

language. Data distribution normality and homogene-

ity of variances were assessed using the Shapiro–Wilk

and Brown–Forsythe tests, respectively. For normal-

ly distributed data with equal variances, differences

among groups were evaluated using one-way analysis

of variance (ANOVA), followed by the Tukey’s post hoc

test for multiple comparisons. When the assumptions

of normality and/or homoscedasticity were violated,

the non-parametric Kruskal–Wallis test was applied,

followed by the Dunn’s post  hoc test. Data are pre-

sented as mean ± standard error of the mean (SEM).

Differences were considered statistically significant

at p  <  0.05.

Comparison between morphometric parame-

ters of astrocytes in control and DTT groups. To

evaluate the effects of DTT treatment on astrocyte

morphology, a Bayesian estimation approach (Bayes-

ian Estimation Supersedes the t-test, BEST) was em-

ployed  [26]. This method was selected because of its

robustness to outliers and violations of normality

assumptions, which are common in morphometric

datasets characterized by substantial cellular het-

erogeneity. In addition, Bayesian inference provides

direct estimates of effect size and enables probabi-

listic interpretation of results, offering advantages

over conventional frequentist approaches. For each

morphometric parameter, data were modeled using

a Student’s t-distribution, which provides increased

robustness to outliers and non-normal data. Posteri-

or distributions of the mean and standard deviation

were estimated for each group. The main estimated

quantities were the posterior distributions of (i)  the

difference in group means (DTT – control), (ii)  the

difference in standard deviations, and (iii)  the stan-

dardized effect size. Practical significance was evalu-

ated using the region of practical equivalence (ROPE),

defined as ±5% of the mean value of a correspond-

ing parameter calculated from the pooled dataset. For

each parameter, the position of the 94% highest den-

sity interval (HDI) relative to the ROPE was assessed

according to the following criteria: (i)  if the 94% HDI

lay entirely outside the ROPE, the difference was

considered meaningful and statistically significant;

(ii)  if the 94% HDI lay entirely within the ROPE, the

difference was considered negligible and statistically

absent; (iii)  if the 94% HDI partially overlapped the

ROPE, the result was regarded as inconclusive. The

ROPE boundaries (±5% of the pooled mean value)

were selected based on estimates of typical segmen-

tation error and the inherent biological variability of

astrocyte morphometric parameters.

METABOLIC STATUS AND MORPHOLOGICAL CHANGES OF ASTROCYTES UNDER REDUCTIVE STRESS 983

BIOCHEMISTRY (MOSCOW) Vol. 91 No. 6 2026

Neural network training for segmentation.

To address the semantic segmentation of native as-

trocytes in phase-contrast microscopy images, two

convolutional neural network architectures were

evaluated: the two-stage detector Mask R-CNN (Re-

gion-Based Convolutional Neural Network) with the

Inception-ResNet-v2 backbone [24, 27], and the one-

stage segmentation detector YOLOv11s-seg (You Only

Look Once version  11)  [28]. Mask R-CNN is a two-stage

object detection and segmentation framework known

for its high segmentation accuracy, particularly when

dealing with objects exhibiting complex and irregular

morphologies. However, the two-stage design results

in increased computational complexity and slower

inference compared with single-stage approaches.

In contrast, YOLOv11s-seg performs object detection,

classification, and instance segmentation simultane-

ously in a single forward pass through the network.

Key advantages of YOLOv11 include significantly

higher processing speed and lower computational re-

source requirements compared to Mask R-CNN.

Model training and optimization were conduct-

ed in Python using the PyTorch framework and the

Ultralytics library [28]. Hyperparameter optimization

for YOLOv11s-seg was performed using the automat-

ed hyperparameter search framework Optuna. Seg-

mentation performance was assessed at both pixel

and object levels. The Dice coefficient was calculat-

ed at the pixel level as the Intersection over Union

(IoU) between predicted and ground-truth segmenta-

tion masks, background pixels were excluded from

the calculation to minimize the effects of class im-

balance inherent to phase-contrast microscopy im-

ages. Precision and recall were calculated at the

object level (individual cells) by matching predicted

and actual cell borders using the IoU) threshold ≥0.5.

The combination of object-level evaluation and Dice-

based overlap analysis reduced the influence of dom-

inant background regions and provided a more reli-

able assessment of each model’s ability to accurately

detect and segment such morphologically complex

cells as astrocytes.

RESULTS

Training of the astrocyte segmentation model.

The trained Mask R-CNN  [24] and YOLOv11s-seg mod-

els were evaluated using the same set of phase-con-

trast images of native astrocyte cultures. Hyperpa-

rameter optimization for the YOLOv11s-seg model

was performed using the Optuna library. During op-

timization, the following parameters were varied:

(i)  initial learning rate (lr0), from 1×10

−5

to 1×10

−4

;

(ii)  input image size (640, 900, and 1200 pixels); and

(iii)  batch size (4, 8, and 16 images).

For each hyperparameter configuration, the mod-

el was trained for 100 epochs using the AdamW op-

timizer. The average precision at the IoU threshold

of 0.5 (mAP50) on the validation set served as the

objective function. A total of 10 optimization itera-

tions were conducted. The optimal configuration con-

sisted of the initial learning rate of 4.15×10

−5

, image

size of 1200×1200  pixels, and batch size of 16 imag-

es. Following hyperparameter selection, the model

was retrained for 150 epochs using the optimized

settings (one epoch corresponds to a complete pass

of all training samples through the neural network).

To monitor overfitting, model performance was eval-

uated on the validation set after each epoch. Final

segmentation performance was assessed on an inde-

pendent test set comprising 10% of the annotated im-

ages, which was excluded from both model training

and hyperparameter optimization. Evaluation metrics

included the Dice coefficient, accuracy, precision, and

recall.

The Mask R-CNN model achieved a Dice coef-

ficient of 0.72, whereas the YOLOv11s-seg model

reached a Dice coefficient of 0.73. The Dice coeffi-

cient measures the overlap between predicted and

ground-truth segmentation masks, ranging from 0

(no overlap) to 1 (perfect agreement). Although both

models demonstrated comparable segmentation ac-

curacy, YOLOv11s-seg showed substantially better

classification performance. In particular, accuracy in-

creased from 0.29 to 0.60, precision increased from

0.71 to 0.76, and recall increased from 0.33 to 0.73.

The more than twofold increase in recall indicates a

markedly improved ability to identify astrocyte pixels

and a significant reduction in missed cellular struc-

tures compared with the Mask R-CNN model  [24].

These results demonstrate that the YOLOv11s-seg ar-

chitecture provides more reliable automated astro-

cyte detection and enables more comprehensive cell

extraction for subsequent morphometric analysis.

A comparison of segmentation and classification met-

rics for both models is presented in Table 1.

Table 1. Comparison of segmentation (Dice coefficient)

and classification (accuracy, precision, and recall) met-

rics for the Mask R-CNN and YOLOv11 models

Metric Mask R-CNN [24] YOLOv11

(current model)

Dice coefficient 0.72 0.73

Accuracy 0.29 0.60

Precision 0.71 0.76

Recall 0.33 0.73

NOVIKOVA et al.984

BIOCHEMISTRY (MOSCOW) Vol. 91 No. 6 2026

Fig. 1. Comparison of segmentation results: (a)  original phase-contrast image of astrocyte culture; (b)  segmentation result

using Mask R-CNN; (c)  segmentation result using YOLOv11.

For a visual comparison of the algorithms, Fig.  1

presents segmentation results obtained on identical

regions of astrocyte culture using Mask R-CNN and

YOLOv11. The proposed method based on the YOLOv11

architecture demonstrates a clear advantage in astro-

cyte detection while maintaining comparable segmen-

tation quality. Specifically, despite a negligible change

in the Dice coefficient, recall increased by more than

twofold. This improvement is particularly import-

ant for quantitative analysis of astrocyte cultures, as

missed detections can introduce systematic bias in the

estimation of their morphological parameters.

Extraction of morphological characteristics

of astrocytes. Segmented images of astrocytes were

used for extraction of morphometric characteristics

for subsequent quantitative analysis, statistical com-

parison of experimental groups and interpretation

of changes in astrocyte morphology under different

culture conditions. For each cell, structural indicators

reflecting its morphological state and branching com-

plexity were determined including (1)  cell area  (μm

(2)  soma area  (%), (3)  average process length  (μm),

(4)  number of branching nodes, and (5)  number of

processes. The results of analysis of all astrocytes in

the image were saved in a CSV file, where for each

cell, a unique identifier and a full set of morphomet-

ric parameters were indicated. For quality control of

segmentation and feature extraction, annotated imag-

es were generated with overlays of cell boundaries

and morphological landmarks (Fig.  2).

Statistical analysis of differences between con-

trol and reductive stress groups. To assess the ef-

fect of DTT on astrocyte morphology, a Bayesian com-

parison of the control and experimental groups was

performed using the BEST approach  [26]. For each

morphometric parameter, posterior distributions of

differences in means (DTT − control), standard devi-

ations, and standardized effect size were analyzed.

Results were interpreted by assessing the 94% HDI

position relative to the ROPE.

Fig. 2. Representative annotated output (image) of segmentation and morphometric analysis of an astrocyte. Cell bound-

aries are shown in yellow; soma boundaries indicated in blue; branching nodes are marked in red. Magnification, 40×;

0.156  μm per pixel.

METABOLIC STATUS AND MORPHOLOGICAL CHANGES OF ASTROCYTES UNDER REDUCTIVE STRESS 985

BIOCHEMISTRY (MOSCOW) Vol. 91 No. 6 2026

Fig. 3. Total cell area of astrocytes. Bayesian comparison of morphometric parameters between the control and DTT-treated

groups: a) posterior distribution of the difference in mean total cell area (DTT − control, μm

, b) posterior distribution of

the difference in standard deviations (μm

), c)posterior distribution of standardized effect size(μm

). Black horizontal line

shows 94% HDI; vertical line indicates zero difference.

Fig. 4. Soma area of astrocytes. Bayesian comparison of soma area between the control and DTT-treated groups: a) pos-

terior distribution of the difference in means (DTT − control, μm

), b) posterior distribution of the difference in standard

deviations(μm

); c)posterior distribution of standardized effect size (μm

). 94%HDI and zero difference value are shown.

Total cell area. The posterior distribution of the

difference in mean total cell area showed a slight

positive shift (by 9.0  μm

); however, the 94% HDI

spanned both negative and positive values (−20.8 to

37.4  μm

; Fig.  3a). The posterior probability of the in-

crease in the total cell area in the experimental group

was 71.9%. In contrast, the posterior distribution of

the difference in standard deviations was entirely be-

low zero (mean difference, −47.2  μm

; 94% HDI  <  0;

Fig.  3b), indicating a pronounced reduction in the

intercellular variability of total cell area under DTT

treatment. The standardized effect size was close to

zero (mean, 0.041; 94% HDI, −0.096 to 0.17; Fig.  3c).

Therefore, DTT exposure did not produce a credible

change in mean total cell area, but was associated

with a clear reduction in the variability of this pa-

rameter across cells.

Soma area. For the soma area, the posterior dis-

tribution of the difference in means was shifted mark-

edly toward positive values, indicating an increase

of 35.0  μm

in the DTT-treated group. The 94% HDI

was completely above zero (13.5 to 55.9μm

; Fig.  4a),

and the posterior probability of the increase in the

soma area exceeded 99.9%. The standardized effect

size was positive and small-to-moderate (mean 0.19;

94% HDI, 0.074 to 0.31; Fig.  4c), indicating a consis-

tent but modest shift in the soma area. In contrast,

the posterior distribution for the difference in stan-

dard deviations overlapped zero (mean difference,

−3.9  μm

; Fig.  4b), indicating no pronounced changes

in the intercellular variability in soma size between

the groups.

Total process length. The posterior distribution

of the difference in mean total process length showed

a slight positive shift (0.19  μm), although the 94% HDI

included zero (−0.19 to 0.56  μm; Fig.  5a). The posteri-

or probability of the increase in this parameter was

82.9%, indicating a trend but not sufficient evidence

for a reliable effect. The standardized effect size was

small (mean, 0.081; 94% HDI, −0.079 to 0.24; Fig.  5c),

and the posterior distribution of the difference in

standard deviations overlapped zero (mean differ-

ence, 0.12 μm; Fig.  5b). Therefore, exposure to DTT

did not produce a statistically significant increase in

the total length of astrocyte processes.

Number of processes. The posterior distribu-

tion of the difference in mean number of processes

was completely shifted into the positive region (mean

difference, 0.47 processes; 94% HDI: 0.25 to 0.67;

Fig.  6a). The posterior probability of the increase in

this parameter in the experimental group was effec-

tively100%. The standardized effect size was positive

NOVIKOVA et al.986

BIOCHEMISTRY (MOSCOW) Vol. 91 No. 6 2026

Fig.  5. Total process length in astrocytes. Bayesian comparison of the total astrocyte process length between the control

and DTT-treated groups: a)  posterior distribution of the difference in means (DTT – control, μm); b)  posterior distribution

of the difference in standard deviations (μm); c)  posterior distribution of the standardized effect size  (μm). Black line in-

dicates 94%  HDI.

Fig.  6. Number of astrocyte processes. Bayesian comparison of the number of astrocyte processes between the control and

DTT-treated groups: a)  posterior distribution of the difference in means (DTT − control), b)  posterior distribution of the

difference in standard deviations, c)  posterior distribution of the standardized effect size. The 94%  HDI is shown.

Fig.  7. Number of branching nodes of astrocytes. Bayesian comparison of the number of branching nodes of astrocyte

processes between the control and DTT-treated groups: a)  posterior distribution of the difference in means (DTT – control);

b)  posterior distribution of the difference in standard deviations; c)  posterior distribution of the standardized effect size.

Black line indicates 94%  HDI.

and moderate (mean, 0.24; 94% HDI, 0.13 to 0.34;

Fig.  6c), indicating a stable and reproducible effect.

The posterior distribution of the difference in stan-

dard deviations overlapped zero (mean difference,

0.084; Fig.  6b). Overall, DTT treatment produced a ro-

bust and statistically significant increase in the num-

ber of astrocyte processes.

Number of branching nodes. The posterior dis-

tribution of the difference in the mean number of

branching nodes was shifted toward negative values

(mean difference, −0.19 nodes), with the 94% HDI

largely below zero (−0.40 to −0.002; Fig.  7a). The

posterior probability of the decrease in the number

of branching nodes was 96.5%. The posterior distri-

bution of the difference in standard deviations was

entirely below zero (mean difference, −0.33; Fig.  7b),

indicating a reduction in the intercellular variability

of branching. The standardized effect size was small

METABOLIC STATUS AND MORPHOLOGICAL CHANGES OF ASTROCYTES UNDER REDUCTIVE STRESS 987

BIOCHEMISTRY (MOSCOW) Vol. 91 No. 6 2026

Table 2. Results of Bayesian comparisons of astrocyte morphometric parameters in the control and DTT-treated

groups

Morphometric parameter

Mean difference

(DTT − control)

94% HDI Effect size Conclusion

Total cell area, μm

+9.0 [−20.8; 37.4] 0.04 no statistically significant changes

Soma area, μm

+35.0 [13.5; 55.9] 0.19 statistically significant increase

Total process length, μm +0.19 [−0.19; 0.56] 0.08 inconclusive

Number of processes +0.47 [0.25; 0.67] 0.24 statistically significant increase

Number of branching nodes −0.19 [−0.40; −0.002] −0.13 weak directed effect

(mean, −0.13; 94%  HDI, −0.27 to near zero; Fig.  7c).

Therefore, exposure to DTT was accompanied by a

moderate decrease in the astrocyte process branching

complexity, along with a reduction in the variability

of this parameter between cells.

A summary of Bayesian comparisons of astrocyte

morphometric parameters is presented in Table  2.

Overall, DTT-induced reductive stress was associated

with an increase in the soma area and the number of

astrocyte processes, alongside a decrease in the num-

ber of branching nodes.

Metabolic status of astrocytes. Incubation of

primary astrocyte cultures with DTT (500  μM, 24  h)

resulted in a statistically significant increase in the

total intracellular content of NAD

/NADH compared

to the control (p  <  0.01; Fig.  8a). Analysis of mito-

chondrial membrane potential revealed no signifi-

cant differences between the experimental and con-

trol groups (Fig.  8b). However, a pronounced increase

in the mitochondrial superoxide production was ob-

served in astrocytes following 24-h exposure to DTT

(p  <  0.01) (Fig.  8c), indicating the development of ox-

idative stress. These data suggest that astrocytes pos-

sess compensatory mechanisms that help preserve

mitochondrial homeostasis even under shifts in the

redox state.

Fig.  8. a)  Total NAD

/NADH content; b)  mitochondrial membrane potential; c)  mitochondrial superoxide production in

astrocytes after incubation with DTT (500  μM, 24  h). Data are presented as mean  ±  SEM; **  p  <  0.01; ns,  not significant.

d)  Fluorescent staining of astrocytes with TMRE (500  nM); magnification, ×40.

NOVIKOVA et al.988

BIOCHEMISTRY (MOSCOW) Vol. 91 No. 6 2026

Fig.  9. a)  GFAP expression in astrocytes after 24-h incubation with DTT. Data are presented as mean  ±  SEM; ns,  not signif-

icant. b)  Immunocytochemical staining of cultured astrocytes: green, anti-GFAP immunofluorescence; blue, DAPI-stained

nuclei; magnification, ×40.

We also evaluated GFAP expression in primary

astrocyte cultures after 24  h of incubation with DTT

(Fig.  9, a  and  b). A trend toward reduced GFAP ex-

pression was observed in the experimental group,

but the differences did not reach statistical signifi-

cance.

DISCUSSION

In this study, we present the first quantitative

assessment of specific morphological alterations in

astrocytes in  vitro under conditions of DTT-induced

reductive stress. To achieve this, we developed and

trained a semantic segmentation model based on the

YOLOv11 architecture, which enabled high-precision

morphometric analysis of more than 2700 cells exhib-

iting complex branched morphologies. The resulting

software pipeline for post-processing of phase-con-

trast microscopy images of live, unstained astrocytes

leverages machine learning to avoid artifacts associ-

ated with fixation and staining and enables quantita-

tive tracking of key morphological parameters during

cell culturing, including total process length, number

of processes, degree of branching, and cell and soma

areas.

For groupwise comparison of morphometric pa-

rameters, we applied a Bayesian estimation approach

(BEST), selected for its robustness to outliers and

deviations from normality, which is particularly im-

portant for morphometric datasets characterized by

high cellular heterogeneity. Modeling based on the

Student’s t-distribution provided robust estimates of

effect sizes and a probabilistic interpretation of in-

tergroup differences, thereby supporting a more reli-

able biological interpretation of changes induced by

reductive stress.

Our results demonstrate that incubation of as-

trocytes with DTT led to an increase in the soma

area and number of processes, without statistically

significant changes in the total cell area or total pro-

cess length, while showing a trend toward reduced

branching complexity. Together, these findings indi-

cate a partial morphological remodeling of astrocytes

in response to DTT-induced reductive stress. The ob-

served enlargement of the soma area, in the absence

of changes in the overall cell area and process length,

along with the increased process number and a ten-

dency toward decreased branching complexity, may

reflect a redistribution of cellular volume and cyto-

skeletal reorganization in response to altered redox

homeostasis.

DTT-induced reductive stress primarily affects re-

dox-dependent processes in the endoplasmic reticu-

lum, as extensively described in the literature [13, 14,

29]. It has been shown that DTT induces pronounced

morphological changes in dermal fibroblasts and tu-

mor cells, disrupts actin cytoskeleton organization,

alters cell shape and spreading, reduces cell–cell

contacts, and impairs cell motility  [13]. By reducing

both intra- and intermolecular disulfide bonds, DTT

interferes with the native folding of proteins, includ-

ing cytoskeletal and extracellular matrix components,

thereby directly affecting adhesion, migration, and

morphological integrity of glial cells.

From a biochemical perspective, DTT-induced

reductive stress in primary astrocyte cultures was

accompanied by a significant increase in the total

pool of nicotinamide coenzymes. Notably, this re-

ductive shift coincided with elevated mitochondrial

production of superoxide in DTT-treated astrocytes.

Importantly, no significant changes in the mitochon-

drial membrane potential were observed, suggesting

preservation of the overall mitochondrial function

METABOLIC STATUS AND MORPHOLOGICAL CHANGES OF ASTROCYTES UNDER REDUCTIVE STRESS 989

BIOCHEMISTRY (MOSCOW) Vol. 91 No. 6 2026

and a compensatory adaptation to the imposed redox

perturbation.

Existing literature suggests that an excess of

reducing equivalents can drive reverse electron

transport in the respiratory chain, thereby promot-

ing electron leakage to oxygen and ROS generation

[11, 15, 30]. Chronic reductive stress has also been

shown to elicit secondary oxidative stress responses

via hyperactivation of the ERO1 pathway, resulting in

paradoxical ROS accumulation. In this context, genes

involved in oxidative quality control (OQC) serve as

critical regulators of redox homeostasis  [15]. Further-

more, recent evidence indicates that DTT may de-

plete S-adenosylmethionine pools through activation

of methyltransferase-dependent pathways, potentially

contributing to broader metabolic remodeling  [31].

Collectively, these findings support the concept of a

dualistic nature of redox imbalance, which may play

a role in early pathogenic mechanisms of neurode-

generative diseases, where reductive stress is increas-

ingly recognized as a potential initiating or permis-

sive factor for subsequent oxidative damage  [8].

Several methodological limitations should be

considered when interpreting these results. Accurate

assessment of cellular redox state ideally requires

quantification of both oxidized and reduced forms

of key cofactors, as well as their ratios (NAD

/NADH,

NADP

/NADPH, and GSH/GSSG), which allows for an

unambiguous elucidation of directions of shifts in

the redox balance  [30]. In our study, only the total

NAD(H) pool was measured, which primarily reflects

alterations in astrocyte metabolism and in the bal-

ance between the synthesis and degradation of nic-

otinamide coenzymes. While not a direct redox state

index, this parameter nonetheless indicates a meta-

bolic response of astrocytes to the DTT-induced re-

ductive stress.

The proposed approach to morphometric analy-

sis based on the use of neural networks also has

inherent limitations related to the training dataset

constraints. Because phase-contrast microscopy does

not reliably delineate individual astrocytes in dense-

ly packed or overlapping cultures, model training

was performed exclusively on isolated, non-overlap-

ping cells. Consequently, in experimental conditions,

where partial cell overlap occurs, the model may

generate segmentation artifacts, including merging

of adjacent cells into a single object. This limitation

is common to many automated image analysis ap-

proaches for glial cultures and necessitates cautious

interpretation of morphometric outputs under high

cell density conditions. Nevertheless, consistent im-

age selection criteria and uniform processing across

all experimental groups support the conclusion that

observed differences reflect a genuine astrocytic re-

sponse to reductive stress.

CONCLUSION

Here, we performed for the first time a compre-

hensive quantitative assessment of morphofunctional

and metabolic changes in astrocytes under DTT-in-

duced reductive stress in  vitro. For this purpose, we

developed a novel approach for automated recogni-

tion of live, unstained astrocytes in phase-contrast

images, followed by morphological quantification

using machine learning based on the YOLOv11 deep

neural network architecture. This method enabled

high-precision analysis of more than 2700  cells, pro-

viding an objective assessment of subtle phenotypic

changes that are not readily accessible using conven-

tional microscopy techniques involving fixation and

staining.

Our results demonstrate that reductive stress,

even in the absence of pronounced mitochondrial

depolarization, induces measurable morphological

and metabolic remodeling in astrocytes. Specifically,

we observed an increase in the soma area and in

the number of cellular processes, while total cell area

and total process length remained unaffected, accom-

panied by a modest but directed reduction in branch-

ing complexity. Concurrently, we detected elevated

ROS production and an increase in the total pool of

nicotinamide coenzymes, indicating a shift in the in-

tracellular redox balance. Collectively, these changes

may reflect adaptive astrocytic response aimed at

preserving cellular and mitochondrial homeostasis

under redox pressure, as well as early events in the

development of pathological cascades. These findings

highlight the importance of further studies on the

role of redox modulation in the regulation of astrog-

lial function, particularly at the early stages of neu-

rodegenerative processes, where reductive stress may

act as a previously underappreciated yet critically im-

portant pathogenic factor.

In addition, the proposed machine learning-based

pipeline substantially accelerates image processing

and offers a broadly applicable framework for auto-

mated quantitative analysis in cellular neurobiology.

This approach facilitates faster data acquisition and

may contribute to the development of more refined

in  vitro disease models and personalized therapeutic

strategies for CNS disorders.

Abbreviations

CNN convolutional neural network

DAPI 4′,6-diamidino-2-phenylindole

DTT dithiothreitol

GFAP glial fibrillary acidic protein

HDI highest density interval

IoU Intersection over Union

ROPE region of practical equivalence

TMRE tetramethylrhodamine ethyl ester

NOVIKOVA et al.990

BIOCHEMISTRY (MOSCOW) Vol. 91 No. 6 2026

Acknowledgments

The authors express their gratitude to R.  S.  Mudariso-

va and A.  I.  Olkhovskaya for their assistance in con-

ducting experiments with primary astrocyte cultures

and spectrophotometric measurement of metabolites.

Contributions

N.A.K., A.B.S., E.V. Ya., and S.O.Yu. developed the con-

cept, supervised the study, and edited the manuscript;

S.V.N., A.K.B., T.E.Shch., and I.V.S. conducted experi-

ments and processed the data; E.V. Ya., Yu.K.K., and

N.P.K. discussed the results; N.A.K., S.V.N., A.K.B., I.V.S.,

and T.E.Shch. wrote the text of the article.

Funding

This study was supported by the State Assignment

“Controlled Angiogenesis: Collective Cell Dynamics and

New Physical Methods of Exposure”, no.  0705-2025-

0010, to the Bauman Moscow State Technical Universi-

ty (morphological analysis) and the State Assignment

“Aberrant Metabolic Plasticity of Cells of the Neuro-

vascular Unit in Brain Pathology”, no.1023101100004-

9-3.1.8;3.1.4 (biochemical studies).

Ethics approval and consent to participate

All experimental procedures using laboratory animals

were performed in accordance with the current leg-

islation of the Russian Federation, international prin-

ciples of the Basel Declaration, and in compliance

with the requirements for the humane treatment of

laboratory animals. The study protocol was approved

by the Ethics Committee of the Russian Center of

Neurology and Neurosciences (protocol No.  10-9/23,

December 20, 2023).

Conflict of interest

The authors of this work declare that they have no

conflicts of interest.

REFERENCES

1. Formolo, D. A., Cheng, T., Yu, J., Kranz, G. S., and Yau, S. Y. (2022) Central adiponectin signaling – a metabolic

regulator in support of brain plasticity, Brain Plast., 8, 79-96, https://doi.org/10.3233/BPL-220138.

2. Salmina, A. B. (2023) Metabolic plasticity in developing and aging brain, Neurochem. J., 17, 325-337, https://

doi.org/10.1134/s1819712423030157.

3. Camandola, S., and Mattson, M. P. (2017) Brain metabolism in health, aging, and neurodegeneration, EMBO J.,

36, 1474-1492, https://doi.org/10.15252/embj.201695810.

4. Di Russo, F., and Lucia, S. (2021) Special issue: neural bases of cognitive processing, Brain Sci., 11, 1286, https://

doi.org/10.3390/brainsci11101286.

5. Salmina, A. B., Malinovskaya, N. A., Morgun, A. V., Khilazheva, E. D., Uspenskaya, Y. A., and Illarioshkin, S. N.

(2022) Reproducibility of developmental neuroplasticity in in vitro brain tissue models, Rev. Neurosci., 33, 531-

554, https://doi.org/10.1515/revneuro-2021-0137.

6. Salmina, A. B., Komleva, Y. K., Malinovskaya, N.A., Morgun, A. V., Teplyashina, E. A., Lopatina, O.L., Gorina, Y. V.,

Kharitonova, E. V., Khilazheva, E. D., and Shuvaev, A. N. (2021) Blood-brain barrier breakdown in stress and

neurodegeneration: biochemical mechanisms and new models for translational research, Biochemistry (Moscow),

86, 746-760, https://doi.org/10.1134/S0006297921060122.

7. Mercola, J. (2025) Reductive stress and mitochondrial dysfunction: The hidden link in chronic disease, Free

Radic. Biol. Med., 233, 118-131, https://doi.org/10.1016/j.freeradbiomed.2025.03.029.

8. Manford, A. G., Rodriguez-Perez, F., Shih, K. Y., Shi, Z., Berdan, C. A., Choe, M., Titov, D. V., Nomura, D. K.,

and Rape, M. (2020) A cellular mechanism to detect and alleviate reductive stress, Cell, 183, 46-61.e21, https://

doi.org/10.1016/j.cell.2020.08.034.

9. Krakowiak, A., and Pietrasik, S. (2023) New insights into oxidative and reductive stress responses and their

relation to the anticancer activity of selenium-containing compounds as hydrogen selenide donors, Biology

(Basel), 12, 875, https://doi.org/10.3390/biology12060875.

10. Berridge, M. V., Herst, P. M., and Prata, C. (2023) Cellular reductive stress: is plasma membrane elec-

tron transport an evolutionarily-conserved safety valve? Redox Biochem. Chem., 5-6, 100016, https://doi.org/

10.1016/j.rbc.2023.100016.

11. Yan, L. J. (2014) Pathogenesis of chronic hyperglycemia: from reductive stress to oxidative stress, J. Diab. Res.,

2014, 137919, https://doi.org/10.1155/2014/137919.

12. Wu, Y. H., and Hsieh, H. L. (2023) Effects of redox homeostasis and mitochondrial damage on Alzheimer’s dis-

ease, Antioxidants (Basel), 12, 1816, https://doi.org/10.3390/antiox12101816.

13. Turishcheva, E. P., Ashniev, G. A., Vildanova, M. S., and Smirnova, E. A. (2023) Endoplasmic reticu-

lum stress inducer dithiothreitol affects the morphology and motility of cultured human dermal fi-

broblasts and fibrosarcoma HT1080 cell line, Russ. J. Dev. Biol., 54, 309-323, https://doi.org/10.1134/

S1062360423050065.

METABOLIC STATUS AND MORPHOLOGICAL CHANGES OF ASTROCYTES UNDER REDUCTIVE STRESS 991

BIOCHEMISTRY (MOSCOW) Vol. 91 No. 6 2026

14. Zhang, L., Zhang, J., Ye, Z. W., Muhammad, A., Li, L., Culpepper, J. W., Townsend, D. M., and Tew, K. D. (2024)

Adaptive changes in tumor cells in response to reductive stress, Biochem. Pharmacol., 219, 115929, https://

doi.org/10.1016/j.bcp.2023.115929.

15. Maity, S., Rajkumar, A., Matai, L., Bhat, A., Ghosh, A., Agam, G., Kaur, S., Bhatt, N. R., Mukhopadhyay, A.,

Sengupta, S., and Chakraborty, K. (2016) Oxidative homeostasis regulates the response to reductive endoplasmic

reticulum stress through translation control, Cell Rep., 16, 851-865, https://doi.org/10.1016/j.celrep.2016.06.025.

16. Jia, C., Shi, Y., Xie, K., Zhang, J., Hu, X., Xu, K., Li, M., and Chu, M. (2019) Vph2 is required for protection

against a reductive stress in Candida albicans, Biochem. Biophys. Res. Commun., 512, 758-762, https://doi.org/

10.1016/j.bbrc.2019.03.146.

17. Ma, W. X., Li, C. Y., Tao, R., Wang, X. P., and Yan, L. J. (2020) Reductive stress-induced mitochondrial dysfunction

and cardiomyopathy, Oxid. Med. Cell Longev., 2020, 5136957, https://doi.org/10.1155/2020/5136957.

18. Suleymanova, I., Balassa, T., Tripathi, S., Molnar, C., Saarma, M., Sidorova, Y., and Horvath, P. (2018) A deep

convolutional neural network approach for astrocyte detection, Sci. Rep., 8, 12878, https://doi.org/10.1038/

s41598-018-31284-x.

19. Jiang, Y., Yang, M., Wang, S., Li, X., and Sun, Y. (2020) Emerging role of deep learning-based artificial intelligence

in tumor pathology, Cancer Commun. (Lond), 40, 154-166, https://doi.org/10.1002/cac2.12012.

20. Weikert, S., Freyer, D., Weih, M., Isaev, N., Busch, C., Schultze, J., Megow, D., and Dirnagl, U. (1997) Rapid

-dependent NO-production from central nervous system cells in culture measured by NO-nitrite/ozone

chemoluminescence, Brain Res., 748, 1-11, https://doi.org/10.1016/s0006-8993(96)01241-3.

21. Stelmashuk, E. V., Kapkaeva, M. R., Rozanova, N. A., Aleksandrova, O. P., Genrikhs, E. E., Oblomov, V. V.,

Novikova, S. V., and Isaev, N. K. (2022) The effect of a neuroinflammation inducer on components of the neu-

rovascular unit of the brain in vitro [in Russian], Sechenov Physiol. J., 108, 686-696, https://doi.org/10.31857/

S0869813922050107.

22. Al-Ani, B., Hewett, P. W., Ahmed, S., Cudmore, M., Fujisawa, T., Ahmad, S., and Ahmed, A. (2006) The release

of nitric oxide from S-nitrosothiols promotes angiogenesis, PLoS One, 1, e25, https://doi.org/10.1371/journal.

pone.0000025.

23. Cleland, W. W. (1964) Dithiothreitol, a new protective reagent for Sh groups, Biochemistry, 3, 480-482, https://

doi.org/10.1021/bi00892a002.

24. Yakovlev, E. V., Simkin, I. V., Shirokova, A. A., Kolotieva, N. A., Novikova, S. V., Nasyrov, A. D., Denisenko, I. R.,

Gursky, K. D., Shishkov, I. N., Narzaeva, D. E., Salmina, A. B., Yurchenko, S. O., and Kryuchkov, N. P. (2024)

Machine learning approach for recognition and morphological analysis of isolated astrocytes in phase contrast

microscopy, Sci. Rep., 14, 9846, https://doi.org/10.1038/s41598-024-59773-2.

25. Russell, B. C., Torralba, A., Murphy, K. P., and Freeman, W. T. (2007) LabelMe: a database and web-based tool

for image annotation, Int. J. Comp. Vis., 77, 157-173, https://doi.org/10.1007/s11263-007-0090-8.

26. Kruschke, J. K., and Liddell, T. M. (2018) The Bayesian new statistics: hypothesis testing, estimation, me-

ta-analysis, and power analysis from a Bayesian perspective, Psychon. Bull. Rev., 25, 178-206, https://doi.org/

10.3758/s13423-016-1221-4.

27. He, K., Gkioxari, G., Dollar, P., and Girshick, R. (2017) Mask R-CNN. International Conference on Computer Vision

(ICCV), Venice, URL: https://www.computer.org/csdl/proceedings-article/iccv/2017/1032c980/12OmNrIJqCA.

28. Ultralytics. Ultralytics YOLO11 Documentation, URL: https://docs.ultralytics.com/models/yolo11.

29. Raj, G., Kumar, M., Shreya, S., Bhardwaj, S., Priya, R., Mangalhara, K. C., and Jain, B. P. (2025) Dithiothreitol

induced endoplasmic reticulum stress and its role in neurodegeneration in Caenorhabditis elegans, World J.

Biol. Chem., 16, 111110, https://doi.org/10.4331/wjbc.v16.i4.111110.

30. Xiao, W., and Loscalzo, J. (2020) Metabolic responses to reductive stress, Antioxid. Redox Signal., 32, 1330-1347,

https://doi.org/10.1089/ars.2019.7803.

31. Gokul, G., and Singh, J. (2022) Dithiothreitol causes toxicity in C. elegans by modulating the methionine-homo-

cysteine cycle, Elife, 11, e76021, https://doi.org/10.7554/eLife.76021.

Publisher’s Note. Pleiades Publishing remains neutral with regard to jurisdictional claims in published

maps and institutional affiliations. AI tools may have been used in the translation or editing of this article.