id
stringlengths 11
17
| article_id
stringlengths 8
11
| path
stringlengths 11
60
| section_title
stringlengths 1
1.33k
| educational_score
float64 0
5.16
| domain
stringclasses 4
values | document_type
stringclasses 5
values | domain_scores
listlengths 0
3
| document_type_scores
listlengths 0
4
| text
stringlengths 1
110k
| authors
listlengths 0
8.02k
| article_url
stringlengths 3
63
| license_type
stringclasses 1
value | license_url
stringclasses 15
values | language
stringclasses 45
values | language_score
float64 0
1
|
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
PMC11278230_p25
|
PMC11278230
|
sec[4]/p[1]
|
5. Clinical Implications
| 3.923828 |
biomedical
|
Other
|
[
0.9931640625,
0.00516510009765625,
0.0019054412841796875
] |
[
0.0535888671875,
0.7080078125,
0.2364501953125,
0.001728057861328125
] |
Recognizing the signs of CF early is crucial for maintaining well-being and the ability to provide effective care. To achieve this, healthcare organizations could screen HCWs and nurses using questionnaires and then conduct interviews with specialized psychologists for those identified as being at greater risk, such as individuals with long-term COVID-19, for developing CF. This approach would enable the implementation of effective prevention measures for helping those who help. Practical interventions could include self-care, seeking support, maintaining a healthy lifestyle, practicing mindfulness and stress reduction, and prioritizing worker well-being through comprehensive care initiatives. Other significant interventions could include regular psychological support, wellness programs, and personalized coping strategies, thus helping to improve both the quality of patient care and the mental health of healthcare workers and nurses. Each of these aspects significantly impacts the well-being of neurological healthcare teams and healthcare organizations, ultimately enhancing the efficiency of care for patients with neurological disorders.
|
[
"Rosaria De Luca",
"Mirjam Bonanno",
"Maria Grazia Maggio",
"Antonino Todaro",
"Carmela Rifici",
"Carmela Mento",
"Maria Rosaria Anna Muscatello",
"Milva Veronica Castorina",
"Paolo Tonin",
"Angelo Quartarone",
"Maria Elena Pugliese",
"Rocco Salvatore Calabrò"
] |
https://doi.org/10.3390/jcm13144200
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
PMC11278230_p26
|
PMC11278230
|
sec[5]/p[0]
|
6. Limitation and Future Research
| 4.058594 |
biomedical
|
Study
|
[
0.9990234375,
0.0004260540008544922,
0.0005230903625488281
] |
[
0.9990234375,
0.00021851062774658203,
0.0004782676696777344,
0.00005692243576049805
] |
This study has some limitations that need to be acknowledged. First, the two samples analyzed were not large enough to be representative of the entire nurse and HCW population. Future larger studies with randomized samples could be beneficial to further investigate the extent of CF in healthcare workers in Italy. Second, there was a lack of follow-up, and there were no comparisons with pre-COVID-19 conditions. Third, we did not focus on the possible types of bias in the administration of questionnaires (such as recall bias, selection bias, etc.) that can be used as a checklist for identifying potential problem/errors (questions with problematic wording; questionnaires that are too long; the administration of questionnaires to populations with cultural differences) when designing and administering the questionnaire . Finally, we did not consider job resource variables, such as support from colleagues and supervisors. In the future, new studies with a larger sample and follow-up would be useful to support the results obtained and provide adequate interventions to prevent situations that put nurses and HCWs at risk, considering the role of long-term COVID-19. Indeed, future research could focus on interventions to manage risk factors for CF among nurses and HCWs.
|
[
"Rosaria De Luca",
"Mirjam Bonanno",
"Maria Grazia Maggio",
"Antonino Todaro",
"Carmela Rifici",
"Carmela Mento",
"Maria Rosaria Anna Muscatello",
"Milva Veronica Castorina",
"Paolo Tonin",
"Angelo Quartarone",
"Maria Elena Pugliese",
"Rocco Salvatore Calabrò"
] |
https://doi.org/10.3390/jcm13144200
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
PMC11278230_p27
|
PMC11278230
|
sec[6]/p[0]
|
7. Conclusions
| 3.957031 |
biomedical
|
Study
|
[
0.99755859375,
0.0016412734985351562,
0.0007181167602539062
] |
[
0.99462890625,
0.003673553466796875,
0.0012969970703125,
0.0002338886260986328
] |
According to our study, CF is a very common symptom affecting nurses and HCWs and, therefore, a potential public health problem in different clinical settings, especially during pandemics. Nurses and HCWs suffering from Long COVID seem to be more affected by CF, and this issue deserves further investigation. The complexity of a patient’s care pathway, mainly in chronic neurological conditions, requires an enormous commitment that can lead to burnout and CF, which should be considered to initiate preventive interventions aimed at helping those who help, for the well-being of patients, healthcare teams, and healthcare organizations.
|
[
"Rosaria De Luca",
"Mirjam Bonanno",
"Maria Grazia Maggio",
"Antonino Todaro",
"Carmela Rifici",
"Carmela Mento",
"Maria Rosaria Anna Muscatello",
"Milva Veronica Castorina",
"Paolo Tonin",
"Angelo Quartarone",
"Maria Elena Pugliese",
"Rocco Salvatore Calabrò"
] |
https://doi.org/10.3390/jcm13144200
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
PMC11278242_p0
|
PMC11278242
|
sec[0]/p[0]
|
1. Introduction
| 4.164063 |
biomedical
|
Study
|
[
0.99853515625,
0.0002543926239013672,
0.0012178421020507812
] |
[
0.9873046875,
0.0024776458740234375,
0.01019287109375,
0.00010752677917480469
] |
In photoelectrocatalysis, the interface between the semiconducting photoelectrode and the electrolyte plays a central role; it is the place where excited charges and reactant molecules meet, and is where the actual chemical reaction takes place. As such, a full understanding of the structure of the interface and its dynamics under reaction conditions is desirable. Unfortunately, this is quite difficult because of the complexity of the system and reactions. The semiconductor is often doped and nanostructured, and the electrolyte often consists of water with soluted ions. An electrostatic potential is applied through the interface, and illumination induces a further redistribution of charges. This poses challenges to experimental characterisation techniques. Moreover, different aspects of the system are related to very different length and time scales; alignment of water molecules a few Ångstroms from the surface depends on the electric fields present there; however, these can often only be understood by considering the electrostatics of the whole interface, spanning up to hundreds of micrometers, or even of the whole device.
|
[
"Nicola Seriani",
"Paola Delcompare-Rodriguez",
"Dhanshree Pandey",
"Abhishek Kumar Adak",
"Vikram Mahamiya",
"Carlos Pinilla",
"Hala J. El-Khozondar"
] |
https://doi.org/10.3390/ma17143460
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
PMC11278242_p1
|
PMC11278242
|
sec[0]/p[1]
|
1. Introduction
| 4.324219 |
biomedical
|
Study
|
[
0.9970703125,
0.0005688667297363281,
0.0023822784423828125
] |
[
0.99853515625,
0.00022530555725097656,
0.0011320114135742188,
0.00007927417755126953
] |
Much of today’s understanding remains based on continuum models that were developed decades ago , and provide a simplified picture of the system; analysis of the experimental data often relies on models that assume full depletion in the space charge layer, a planar interface, constant dielectric behaviour, and independence from atomic details. Although many of the shortcomings of these models are recognised , this insight often remains at the qualitative level. Therefore, it is highly desirable to devise simple analytic models that overcome some of the common assumptions in order to provide easier insight into the role of the different variables and can be used for systematic comparisons of experimental situations. For the case of relevance here, one important issue is that the capacitance of the Helmholtz layer becomes more important when the semiconductor is heavily doped. This is already known to the point that its values are sometimes extracted from electrochemical impedance spectroscopy data . Nonetheless, the consequences are still often only discussed qualitatively. In addition, while nanostructuring is recognised as an important factor that can modify interface behaviour , it is rarely included in quantitative analysis, meaning that a systematic understanding of its effects is still missing. In the case of hematite, experiments on nanorods have been analysed while taking into account modified geometries . It should be noted that analytic expressions are highly desirable, as they are easy to handle and provide conceptual insights into the role of the different properties. For this reason, we focus on a simple analytic treatment of the spherical and cylindrical cases. In this work, we study the interface in a continuum model, including the space charge layer, the Helmholtz layer, and the diffuse Gouy–Chapman layer. The model considers the presence of doping for both a spherical nanoparticle and a nanorod, and we compare the results to those for the planar interface . The results are focused on two systems of particular interest, namely, that of a photoelectrode constituted by doped hematite ( α − Fe 2 O 3 ), and that in which the photoelectrode is constituted by titania (TiO 2 ). Hematite is considered a promising photoanode for water splitting , and has been extensively studied both experimentally and theoretically in recent years, while titania was historically the first oxide to display photocatalytic water splitting and remains the reference material in the field .
|
[
"Nicola Seriani",
"Paola Delcompare-Rodriguez",
"Dhanshree Pandey",
"Abhishek Kumar Adak",
"Vikram Mahamiya",
"Carlos Pinilla",
"Hala J. El-Khozondar"
] |
https://doi.org/10.3390/ma17143460
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
PMC11278242_p2
|
PMC11278242
|
sec[0]/p[2]
|
1. Introduction
| 4.132813 |
biomedical
|
Study
|
[
0.99658203125,
0.00037670135498046875,
0.002887725830078125
] |
[
0.99951171875,
0.00025343894958496094,
0.000331878662109375,
0.00004297494888305664
] |
These two cases are highly interesting because the electrostatic fields that regulate charge dynamics both towards the interface and at the interface itself play a crucial role in ensuring that holes reach the interface in high concentrations and that the reaction kinetics are favourable . We quantitatively analyse the interplay of doping and nanostructuring. Low doping ensures that the space charge layer is quite extended, favouring charge collection over recombination, while the potential that a nanostructure can sustain before being completely depleted is quite small. On the other hand, high doping reduces the size of the space charge layer, while nanostructuring is decisive in ensuring that the space charge layer represents a large volume fraction of the semiconductor. Finally, we show that at high doping electric fields in the Helmholtz layer can be as high as those that are known to cause freezing in pure water . This raises questions about the dynamics of water at the interface in the presence of high doping and high applied potential.
|
[
"Nicola Seriani",
"Paola Delcompare-Rodriguez",
"Dhanshree Pandey",
"Abhishek Kumar Adak",
"Vikram Mahamiya",
"Carlos Pinilla",
"Hala J. El-Khozondar"
] |
https://doi.org/10.3390/ma17143460
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
PMC11278242_p3
|
PMC11278242
|
sec[0]/p[3]
|
1. Introduction
| 2.078125 |
biomedical
|
Study
|
[
0.63623046875,
0.0019311904907226562,
0.362060546875
] |
[
0.50244140625,
0.482177734375,
0.0138702392578125,
0.0015773773193359375
] |
In the following sections, we first discuss the analytic model, then present the results for the electrochemical interface between the oxide and a water electrolyte. Finally, we close the article with our conclusions.
|
[
"Nicola Seriani",
"Paola Delcompare-Rodriguez",
"Dhanshree Pandey",
"Abhishek Kumar Adak",
"Vikram Mahamiya",
"Carlos Pinilla",
"Hala J. El-Khozondar"
] |
https://doi.org/10.3390/ma17143460
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999995 |
PMC11278242_p4
|
PMC11278242
|
sec[1]/sec[0]/p[0]
|
2.1. Poisson–Boltzmann Equations in the Spherical Case
| 4.367188 |
biomedical
|
Study
|
[
0.98974609375,
0.0006785392761230469,
0.0095367431640625
] |
[
0.9970703125,
0.0015783309936523438,
0.0014190673828125,
0.00009578466415405273
] |
The model describes the space charge layer in the depletion approximation, the Helmholtz layer, and the diffuse Gouy–Chapman layer. The assumptions are analogous to those employed in for the planar interface. The Poisson–Boltzmann model assumes that the interface consists of three regions: the space charge layer in the semiconductor, the Helmholtz layer in the liquid in direct contact with the solid, and then the diffuse or Gouy–Chapman layer, also in the liquid. The space charge layer has a thickness that depends on the applied potential, and contains a fixed and uniform density of donors; the corresponding electrons are completely removed from the space charge layer (depletion approximation), meaning that the charge of the space charge layer is uniquely determined by its width and by the donor density. The Helmholtz layer is the thin region between the surface of the semiconductor and the closest point at which dissolved ions can come to the surface; it behaves like a capacitor with a given dielectric constant. Finally, the diffuse or Gouy–Chapman layer extends into the liquid, in principle to infinity, and the concentration of ions is determined self-consistently by Boltzmann statistics and the electrostatic potential. In the planar case, the equations have an exact analytic solution . For the spherical case, an approximate analytic solution is available, with an exact expression for the space charge and Helmholtz layers and a robust analytic approximation for the diffuse Gouy–Chapman layer . Before introducing a similar robust approximation below for the cylindrical case, we first discuss the spherical case. In this case, the semiconductor is a sphere; as shown in Figure 1 a, the semiconductor has a radius R; the space charge layer is located in the region between a radius R 1 and R (0 ≤ R 1 ≤ R), the Helmholtz layer is between R and R H , and beyond R H is the diffuse Gouy–Chapman layer. The boundary conditions are ϕ S C ( R 1 ) = 0 , ∂ ϕ S C ∂ r ( R 1 ) = 0 , l i m r − > ∞ ϕ e l ( r ) = ϕ e l ( + ∞ ) . The first condition is a choice of the zero of the potential, the second is a continuity condition for the electric field, which is strictly zero for r ≤ R 1 , and the third amounts to fixing the total potential drop across the whole interface.
|
[
"Nicola Seriani",
"Paola Delcompare-Rodriguez",
"Dhanshree Pandey",
"Abhishek Kumar Adak",
"Vikram Mahamiya",
"Carlos Pinilla",
"Hala J. El-Khozondar"
] |
https://doi.org/10.3390/ma17143460
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
PMC11278242_p5
|
PMC11278242
|
sec[1]/sec[0]/p[1]
|
2.1. Poisson–Boltzmann Equations in the Spherical Case
| 4.1875 |
biomedical
|
Study
|
[
0.99609375,
0.00028634071350097656,
0.003574371337890625
] |
[
0.99853515625,
0.0007805824279785156,
0.00046062469482421875,
0.00004750490188598633
] |
Here, we summarize the results; a detailed derivation is in the Supplementary Materials . The potential drop across the electrochemical interface is (1) V a p p − V f b = Δ ϕ t o t a l = Δ ϕ s c + Δ ϕ H + Δ ϕ e l , where V a p p is the applied potential, V f b is the flatband potential, and Δ ϕ s c , Δ ϕ H , and Δ ϕ e l are the potential drops across the space charge layer, Helmholtz layer, and Gouy–Chapman layer, respectively. As in , we express all quantities as a function of Δ ϕ S C . We also use the quantities (2) A 0 = 6 ϵ 0 ϵ S C e N D and (3) B = e 2 c 0 k T ϵ 0 ϵ e l . Given a certain radius R 1 for the inner border of the space charge layer, the potential drop is (4) Δ ϕ S C = e N D 6 ϵ 0 ϵ S C 1 R R 3 − 3 R 1 2 R + 2 R 1 3 . Conversely, for a given potential drop in the space charge layer, the inner border of the space charge layer is located in (5) R 1 = R 1 2 + c o s 1 3 a r c c o s − 1 + 2 A 0 R 2 Δ ϕ S C − 2 π 3 .
|
[
"Nicola Seriani",
"Paola Delcompare-Rodriguez",
"Dhanshree Pandey",
"Abhishek Kumar Adak",
"Vikram Mahamiya",
"Carlos Pinilla",
"Hala J. El-Khozondar"
] |
https://doi.org/10.3390/ma17143460
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
PMC11278242_p6
|
PMC11278242
|
sec[1]/sec[0]/p[2]
|
2.1. Poisson–Boltzmann Equations in the Spherical Case
| 4.226563 |
biomedical
|
Study
|
[
0.97705078125,
0.0003921985626220703,
0.0227508544921875
] |
[
0.998046875,
0.0014400482177734375,
0.0002837181091308594,
0.00004750490188598633
] |
The other two contributions to the potential drop are (6) Δ ϕ H = e N D 3 ϵ 0 ϵ H R 3 − R 1 3 1 R − 1 R H , (7) ϕ e l ( R H ) − ϕ e l ( + ∞ ) = 2 k T e l n { e 2 k T ϵ S C ϵ e l d ϕ S C d r ( R ) R 2 R H 1 1 − B R H + 1 + e 2 k T ϵ S C ϵ e l d ϕ S C d r ( R ) R 2 R H 1 1 − B R H 2 } , where (8) ∂ ϕ S C ∂ r ( R ) = − e N D 3 ϵ 0 ϵ S C R − R 1 3 R 2 . These are the equations that define the potential drop across the electrochemical interface. As discussed in the Supplementary Materials . they reduce to the planar case in the limit of infinite radius. These expressions are valid in the depletion approximation for 0 ≤ Δ ϕ S C ≤ Δ ϕ S C l i m i t − s p h , with (9) Δ ϕ S C l i m i t − s p h = e N D R 2 6 ϵ 0 ϵ S C = R 2 A . At this limiting value of the potential drop, the space charge layer includes the whole semiconductor sphere, and cannot be extended further. This last formula can be used to check whether the limit has been reached during an experiment.
|
[
"Nicola Seriani",
"Paola Delcompare-Rodriguez",
"Dhanshree Pandey",
"Abhishek Kumar Adak",
"Vikram Mahamiya",
"Carlos Pinilla",
"Hala J. El-Khozondar"
] |
https://doi.org/10.3390/ma17143460
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
PMC11278242_p7
|
PMC11278242
|
sec[1]/sec[1]/p[0]
|
2.2. Poisson–Boltzmann Equations in the Cylindrical Case
| 4.332031 |
biomedical
|
Study
|
[
0.9931640625,
0.000461578369140625,
0.006549835205078125
] |
[
0.9990234375,
0.0006241798400878906,
0.00043964385986328125,
0.000057756900787353516
] |
The cylindrical case is very interesting in view of the existence of nanostructures with cylindrical symmetry, such as nanotubes and nanorods. We use a similar notation to the previous section: the semiconductor is a cylinder of radius R, and the space charge layer is located in the region between a radius R 1 and R (0 ≤ R 1 ≤ R); R is a geometric feature of the nanostructure, while R 1 depends on the applied bias. The general boundary conditions are ϕ S C ( R 1 ) = 0 , ∂ ϕ S C ∂ r ( R 1 ) = 0 , l i m r − > ∞ ϕ e l ( r ) = ϕ e l ( + ∞ ) ; the first condition fixes the zero of the potential, the second condition forces the electric field to go to zero continuously at R 1 , and the third condition fixes the potential difference between the center of the cylinder and the region far from the cylinder. The solutions are as follows, (10) Δ ϕ S C = ϕ S C ( R 1 ) − ϕ S C ( R ) = e N D 2 ϵ 0 ϵ S C R 2 2 − R 1 2 2 − R 1 2 l n R R 1 The above equation is transcendental in R 1 , meaning that it is not possible to find an explicit expression for R 1 as a function of Δ ϕ S C . (11) Δ ϕ H = e N D 2 ϵ 0 ϵ H R 2 − R 1 2 l n ( R H R ) In analogy to the approximation used in for the spherical case, an approximate analytic expression is employed for the potential drop in the Gouy–Chapman layer (see Supplementary Materials for a derivation): (12) Δ ϕ e l = 4 k T e a r c t a n h C + 1 + C 2 where (13) C = ϵ e l ϵ S C k T e 1 + 2 B R H R 1 ∂ ϕ S C ∂ r ( R ) and (14) B = e 2 c 0 k T ϵ 0 ϵ e l . For convenience, we again report here the explicit expressions for ∂ ϕ S C ∂ r ( R ) : (15) ∂ ϕ S C ∂ r ( R ) = − e N D 2 ϵ 0 ϵ S C R − R 1 2 R . These equations are valid up the situation where the space charge layer fills the whole semiconductor. In this limiting case, the maximal potential drop in the semiconductor is reached: (16) Δ ϕ S C l i m i t − c y l = ϕ S C ( R 1 ) − ϕ S C ( R ) = e N D R 2 4 ϵ 0 ϵ S C or (17) Δ ϕ S C l i m i t − c y l = ϕ S C ( R 1 ) − ϕ S C ( R ) = 3 R 2 2 A 0 .
|
[
"Nicola Seriani",
"Paola Delcompare-Rodriguez",
"Dhanshree Pandey",
"Abhishek Kumar Adak",
"Vikram Mahamiya",
"Carlos Pinilla",
"Hala J. El-Khozondar"
] |
https://doi.org/10.3390/ma17143460
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
PMC11278242_p8
|
PMC11278242
|
sec[1]/sec[2]/p[0]
|
2.3. Calculation Parameters
| 4.203125 |
biomedical
|
Study
|
[
0.9970703125,
0.0003161430358886719,
0.0025730133056640625
] |
[
0.99951171875,
0.00024199485778808594,
0.00024008750915527344,
0.000040650367736816406
] |
The parameters needed for the calculation are the dielectric constants ϵ S C , ϵ H , and ϵ e l for space charge layer, Helmholtz layer, and Gouy–Chapman layer, respectively, along with the doping density in the semiconductor N D , the ion concentration in the electrolyte c 0 , the width of the Helmholtz layer L H , and the flatband potential V f b . Most of these are taken directly from the experimental literature. For hematite, the value ϵ S C = 57 from has been used; moreover, as discussed in , ϵ e l = 64.42 has been used . To illustrate the effect of doping in hematite, we highlight two extreme situations, one with low doping and one with high doping, for which the data have been taken from the experiments by Iandolo et al. and Le Formal et al. , respectively. Silicon is normally present as a substitutional dopant in the 4+ oxidation state, which should be compared with the 3+ oxidation state of iron in hematite. This results in one extra electron per silicon atom. Because oxygen in the oxide lattice is in the 2− oxidation state, the oxygen vacancy usually results in two extra electrons in the material. Both cases result in n-doping being present in the oxide. The experimental data are summarised in Table 1 . On the contrary, L H and ϵ H are not available from the experimental literature; the values of L H = 4.4 Å and ϵ H = 25.3 have been taken from theoretical works . A thorough discussion on these theoretical values can be found in . For titania, the value ϵ S C = 50 has been used, as in . For titania, we chose experiments with the same NaOH-based electrolyte as in the case of hematite to ensure that we could use the same dielectric constants. NaOH-based electrolytes are very common in this kind of experiment . For the Helmholtz layer of titania, we used the value L H = 10 Å, as in .
|
[
"Nicola Seriani",
"Paola Delcompare-Rodriguez",
"Dhanshree Pandey",
"Abhishek Kumar Adak",
"Vikram Mahamiya",
"Carlos Pinilla",
"Hala J. El-Khozondar"
] |
https://doi.org/10.3390/ma17143460
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
PMC11278242_p9
|
PMC11278242
|
sec[2]/p[0]
|
3. Results and Discussion
| 4.207031 |
biomedical
|
Study
|
[
0.99658203125,
0.0005712509155273438,
0.002651214599609375
] |
[
0.99951171875,
0.000148773193359375,
0.00029468536376953125,
0.00004982948303222656
] |
For the cases of hematite with low and high doping concentrations, the potential drops in the three regions of the interface (the space charge layer, Helmholtz layer, and Gouy–Chapman layer) are shown in Figure 2 for the spherical and cylindrical geometries, with the results for the planar interface from shown for comparison. In all cases, the low-doping regime displays the typical behaviour of a semiconductor interface, with by far the largest contribution to the potential drop provided by the space charge layer. On the other side, the highly doped semiconductor displays intermediate behaviour between a pure semiconductor and a metal, with a substantial portion of the potential drop being located in the Helmholtz layer. This behaviour is more pronounced in the planar interface. Counterintuitively, it is damped in the nanostructures; the reason for this is that the small radius of curvature enhances more the electric field in the space charge layer, resulting in a smaller electric field in the Helmholtz layer. In the case of high doping, the electric field in the Helmholtz layer reaches values higher than 100 mV/Å for the planar interface, while in the nanoparticle with a radius of 5 nanometers it reaches values of at most ∼80 mV/Å. Notably, 5 nm was the radius of the typical nanostructures in recent experiments with hematite and titania . We note that such high electric fields have been shown to lead to freezing of pure water . It would be interesting to understand how the dynamics of interfacial water are affected under these conditions; however, this is beyond the scope of the present paper. On the contrary, the potential drop across the Gouy–Chapman layer, though more pronounced in the highly doped case, remains smaller than 100 mV under all considered conditions.
|
[
"Nicola Seriani",
"Paola Delcompare-Rodriguez",
"Dhanshree Pandey",
"Abhishek Kumar Adak",
"Vikram Mahamiya",
"Carlos Pinilla",
"Hala J. El-Khozondar"
] |
https://doi.org/10.3390/ma17143460
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
PMC11278242_p10
|
PMC11278242
|
sec[2]/p[1]
|
3. Results and Discussion
| 4.128906 |
biomedical
|
Study
|
[
0.9873046875,
0.0002865791320800781,
0.01220703125
] |
[
0.9990234375,
0.0005660057067871094,
0.0004203319549560547,
0.000043272972106933594
] |
The trends observed in hematite are even more pronounced in titania due to the even higher doping concentration reported in . In this case, N D is so high that most of the potential drop takes place in the Helmholtz layer. Even at the highest applied potentials, Δ ϕ S C remains smaller than 400 meV. This behaviour is more similar to that of a metal than that of a typical semiconductor. We note that such high doping concentrations have been reported for several nanostructured oxides; for example, Cristino et al. reported N D = 7.5 × 10 20 cm − 3 in nanostructured WO 3 films , and Shaddad et al. reported N D = 9.25 × 10 19 cm − 3 in dendritic nanostructured Bi 2 O 3 films .
|
[
"Nicola Seriani",
"Paola Delcompare-Rodriguez",
"Dhanshree Pandey",
"Abhishek Kumar Adak",
"Vikram Mahamiya",
"Carlos Pinilla",
"Hala J. El-Khozondar"
] |
https://doi.org/10.3390/ma17143460
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
PMC11278242_p11
|
PMC11278242
|
sec[2]/p[2]
|
3. Results and Discussion
| 3.496094 |
biomedical
|
Study
|
[
0.83447265625,
0.0007638931274414062,
0.164794921875
] |
[
0.9931640625,
0.006099700927734375,
0.0005431175231933594,
0.00010001659393310547
] |
It could be questioned whether the employed model is still valid under such high electric fields. In fact, the high electric fields predicted here for the Helmholtz layer should have a non-negligible effect on the water dynamics in the layer, probably contributing to polarisation of any water there. This in turn should affect the dielectric constant of the Helmholtz layer, decreasing its value and leading to modified behaviour on the part of the interface. While the effects of a field-dependent dielectric constant are not included in the model, it can be speculated that a reduction of the dielectric constant would further increase the electric field. However, a full treatment is beyond the scope of this paper.
|
[
"Nicola Seriani",
"Paola Delcompare-Rodriguez",
"Dhanshree Pandey",
"Abhishek Kumar Adak",
"Vikram Mahamiya",
"Carlos Pinilla",
"Hala J. El-Khozondar"
] |
https://doi.org/10.3390/ma17143460
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
PMC11278242_p12
|
PMC11278242
|
sec[2]/p[3]
|
3. Results and Discussion
| 4.359375 |
biomedical
|
Study
|
[
0.998046875,
0.00047397613525390625,
0.0013666152954101562
] |
[
0.9990234375,
0.0002510547637939453,
0.0007829666137695312,
0.0000712275505065918
] |
At this point, it might be useful to underline the conceptual difference between the quantities considered here and the quantities obtained in typical atomistic simulations. The potentials and fields reported here are only those due to the charging of the interface away from the conditions of flatband potential. At the flatband potential, all fields and potential differences considered in the present work are zero. This is a striking difference from the typical case of atomistic simulations, where the atomic nuclei carry a point charge. There, electric fields, which we might call intrinsic atomic electric fields, are always present, and can be sizable or even larger than those reported here. However, they are always present, even at the conditions of flatband potential, where excited charges do not produce macroscopic drift currents. Therefore, the photocurrents that are of relevance for photocatalysis are related to the “macroscopic” electric fields considered in the present work, which are superimposed upon the intrinsic atomic electric fields. This explains why, for example, Sang et al. found huge electric fields (up to 3 V/Ångstrom) even in the neutral interface by means of classical molecular dynamics . The fields calculated here are superimposed upon these intrinsic contributions. Thus, the situation described here is more akin to atomistic simulations in which an external electric field is imposed. Indeed, Futera et al. found that water dissociates at the interface with hematite in external electric fields of 75 mV/Å and above, with strong reorganisation. However, Futera et al. also noticed that external fields of 100 mV/Å created numerical problems for their molecular dynamics algorithm. Given that in their simulations the external field was naturally screened by water polarisation, the total electric field is bound to be smaller than our value of 100 mV/Å. Thus, the high-field regime depicted here is in fact largely unexplored.
|
[
"Nicola Seriani",
"Paola Delcompare-Rodriguez",
"Dhanshree Pandey",
"Abhishek Kumar Adak",
"Vikram Mahamiya",
"Carlos Pinilla",
"Hala J. El-Khozondar"
] |
https://doi.org/10.3390/ma17143460
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
PMC11278242_p13
|
PMC11278242
|
sec[2]/p[4]
|
3. Results and Discussion
| 4.230469 |
biomedical
|
Study
|
[
0.99755859375,
0.00045561790466308594,
0.0019369125366210938
] |
[
0.99951171875,
0.0002281665802001953,
0.00025153160095214844,
0.00004553794860839844
] |
A second result of this work is the calculation of the limiting potential drop that can be sustained inside the nanostructure. This can be readily obtained from Equation ( 9 ) for a spherical nanoparticle and from Equation ( 17 ) for the case of a nanorod. Full depletion of the semiconductor corresponds to setting R 1 to zero. For the spherical nanoparticle, we obtain (18) Δ ϕ S C l i m i t − s p h = e N D R 2 6 ϵ 0 ϵ S C = R 2 A 0 , while for the nanorod we obtain (19) Δ ϕ S C l i m i t − c y l = e N D R 2 4 ϵ 0 ϵ S C = 3 R 2 2 A 0 , where (20) A 0 = 6 ϵ 0 ϵ S C e N D . Figure 4 shows the limiting potential drop Δ ϕ t o t l i m i t = V a p p − V f l a t b a n d across the interface, for which the space charge layer extends over the whole nanoparticle in dependence on the particle’s radius, as well as the corresponding potential drop Δ ϕ S C l i m i t across the space charge layer. In the case of low doping these two quantities almost coincide, whereas the Helmholtz contribution is substantial in the case of high doping. At low doping , the space charge layer is very extended, to easily over hundreds of nanometers, leading to charge depletion even in fairly large nanostructures. On the contrary, high applied potentials are necessary at high doping in order to deplete nanoparticles of a few nanometers. In either case, the solid electrode has no possibility to accommodate further potential increases above the limiting potential, which are taken care of either by an increase in the electric fields in the liquid electrolyte or by charging of the substrate. This makes more precise the observation in that both experiments eventually reached a regime in which the semiconductor (a thin film in and a cauliflower-like structure in ) was completely depleted, in other words, where the space charge layer extended over the whole semiconductor.
|
[
"Nicola Seriani",
"Paola Delcompare-Rodriguez",
"Dhanshree Pandey",
"Abhishek Kumar Adak",
"Vikram Mahamiya",
"Carlos Pinilla",
"Hala J. El-Khozondar"
] |
https://doi.org/10.3390/ma17143460
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
PMC11278242_p14
|
PMC11278242
|
sec[2]/p[5]
|
3. Results and Discussion
| 4.164063 |
biomedical
|
Study
|
[
0.9931640625,
0.0005578994750976562,
0.006317138671875
] |
[
0.99951171875,
0.00021600723266601562,
0.00026988983154296875,
0.000043272972106933594
] |
The results for the nanostructures illustrate that despite the very different doping in the two experiments, both result in situations where a large part of the semiconductor belongs to the space charge layer thanks to the fine nanostructure of the highly doped sample. To quantify this, we calculated the width and volume fraction of the space charge layer over the total volume of the semiconductor, with the results shown in Figure 5 and Figure 6 , respectively. Figure 5 a shows the width of the space charge layer in the spherical case as a function of the total potential drop across the interface. As expected, the width is much smaller in the case with high doping than in the case with low doping. However, as shown in Figure 6 , nanostructuring of a highly doped electrode can lead to a situation where the fraction of the total volume that is taken up by the space charge layer is comparable to that in the case with low doping. The figure shows that at a doping concentration of N D = 7.00 × 10 20 cm − 3 , a particle of 5 nm radius has a similar volume fraction of space charge layer to a particle of 100 nm at a lower doping concentration of N D = 2.28 × 10 18 cm − 3 . Considering that it is often assumed that the holes that can reach the surface and participate in the photocatalysis are mostly produced in the space charge layer, and as such are partly protected from recombination, these results quantitatively shown how high doping concentrations can work when accompanied by nanostructuring. The width and volume fraction of the space charge layer in the cylindrical case are reported in Figure 5 c,d and Figure 6 c,d, respectively. Not unexpectedly, the cylindrical case is intermediate between the spherical and planar cases. Nonetheless, in this case nanostructuring again brings the volume fraction of small-radius nanorods in the highly doped case into line with that of the low-doping case for nanorods with a larger radius. Therefore, the effect of decreased space layer width can again be obviated through nanostructuring.
|
[
"Nicola Seriani",
"Paola Delcompare-Rodriguez",
"Dhanshree Pandey",
"Abhishek Kumar Adak",
"Vikram Mahamiya",
"Carlos Pinilla",
"Hala J. El-Khozondar"
] |
https://doi.org/10.3390/ma17143460
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
PMC11278242_p15
|
PMC11278242
|
sec[2]/p[6]
|
3. Results and Discussion
| 4.148438 |
biomedical
|
Study
|
[
0.99267578125,
0.0006504058837890625,
0.006847381591796875
] |
[
0.9990234375,
0.00039649009704589844,
0.0004856586456298828,
0.000053882598876953125
] |
The electric field in the space charge layer is crucial to ensuring electron–hole separation and avoiding recombination. Therefore, it is usually assumed that an extended space charge layer leads to higher performance, as it leads to decreased recombination over a large portion of the semiconductor. This would speak in favour of low doping in the semiconductor. On the other hand, higher doping fosters conductivity inside the semiconductor, though at the price of higher recombination. Here, we show that high doping with nanostructuring makes the best of both situations, as is it possible to have a high volume fraction of the semiconductor in the space charge layer (thereby avoiding recombination) while also harvesting the advantages of high doping. Therefore, we propose that photocatalytic performance can best benefit from high doping when carried out together with fine nanostructuring of the material. In addition, we show that this could contribute to better charge transport, avoiding the pitfalls of high recombination, and provide a quantitative framework for evaluating this connection. Finally, we propose that the optimal doping concentration should increase with decreasing size of the nanostructure, a prediction that could be tested experimentally in future research.
|
[
"Nicola Seriani",
"Paola Delcompare-Rodriguez",
"Dhanshree Pandey",
"Abhishek Kumar Adak",
"Vikram Mahamiya",
"Carlos Pinilla",
"Hala J. El-Khozondar"
] |
https://doi.org/10.3390/ma17143460
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
PMC11278242_p16
|
PMC11278242
|
sec[3]/p[0]
|
4. Conclusions
| 4.148438 |
biomedical
|
Study
|
[
0.9970703125,
0.00041222572326660156,
0.002552032470703125
] |
[
0.99951171875,
0.00015926361083984375,
0.00043892860412597656,
0.00004571676254272461
] |
In this paper, the effect of doping and nanostructuring on the electrochemical interface between an oxide (hematite or titania) and a water electrolyte has been investigated using a Poisson–Boltzmann model, including the space charge layer, Helmholtz layer, and Gouy–Chapman layer. Thanks to robust yet simple approximations for the diffuse layer, compact analytic expressions are reported for the potential drop across the interface in the cases of spherical and cylindrical nanostructures. For spherical nanoparticles and nanorods, simple formulae for the limiting potentials at which the space charge layer includes the whole semiconductor are provided. These can be used in future experiments to estimate whether full depletion of the semiconductor has been reached. At low doping, nanostructures can already be fully depleted at potentials of only a few mV.
|
[
"Nicola Seriani",
"Paola Delcompare-Rodriguez",
"Dhanshree Pandey",
"Abhishek Kumar Adak",
"Vikram Mahamiya",
"Carlos Pinilla",
"Hala J. El-Khozondar"
] |
https://doi.org/10.3390/ma17143460
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999999 |
PMC11278242_p17
|
PMC11278242
|
sec[3]/p[1]
|
4. Conclusions
| 4.121094 |
biomedical
|
Study
|
[
0.98388671875,
0.0003705024719238281,
0.0155181884765625
] |
[
0.99853515625,
0.0010576248168945312,
0.00032019615173339844,
0.000045180320739746094
] |
At high doping densities, a substantial part of the potential drop is located in the Helmholtz layer, contrary to what happens in a conventional semiconductor. This results in huge electric fields in the liquid, which can be as high as 100 mV/Å, comparable to electric fields that induce freezing in pure water. This effect is slightly reduced by nanostructuring; for nanoparticles of 5 nm, the electric field in the Helmholtz layer reaches values of at most ∼80 mV/Å. The potential drop in the Gouy–Chapman layer is negligible at low doping, while at high doping it is smaller than 100 mV under all considered conditions. It is highly necessary that local electric fields and water dynamics at the interface be tested experimentally in future research.
|
[
"Nicola Seriani",
"Paola Delcompare-Rodriguez",
"Dhanshree Pandey",
"Abhishek Kumar Adak",
"Vikram Mahamiya",
"Carlos Pinilla",
"Hala J. El-Khozondar"
] |
https://doi.org/10.3390/ma17143460
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999999 |
PMC11278242_p18
|
PMC11278242
|
sec[3]/p[2]
|
4. Conclusions
| 4.070313 |
biomedical
|
Study
|
[
0.99658203125,
0.00020194053649902344,
0.003406524658203125
] |
[
0.9990234375,
0.0005488395690917969,
0.0002510547637939453,
0.00003743171691894531
] |
It is usually assumed that high doping improves the charge dynamics in hematite but reduces the space charge layer width, i.e., it reduces the portion of the semiconductor from which charges can be collected. However, nanostructuring corrects the latter negative effect. We show quantitatively that in highly doped nanostructures the space charge layer can occupy a similar volume fraction as in low-doped microparticles. These results imply that the optimal doping concentration should become larger as the size of the nanostructures decreases, a behaviour that would be interesting to test experimentally.
|
[
"Nicola Seriani",
"Paola Delcompare-Rodriguez",
"Dhanshree Pandey",
"Abhishek Kumar Adak",
"Vikram Mahamiya",
"Carlos Pinilla",
"Hala J. El-Khozondar"
] |
https://doi.org/10.3390/ma17143460
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
PMC11278242_p19
|
PMC11278242
|
sec[3]/p[3]
|
4. Conclusions
| 4.101563 |
biomedical
|
Study
|
[
0.998046875,
0.00027871131896972656,
0.001506805419921875
] |
[
0.99853515625,
0.00032401084899902344,
0.0011081695556640625,
0.000047326087951660156
] |
This work provides a systematic quantitative framework for understanding the effect of doping and nanostructuring on electrochemical interfaces by providing a complete set of equations for the description of these interfaces at arbitrary doping concentrations. These results suggest that it remains necessary to better characterize the dynamics of the interface at the atomistic level, especially in the presence of high electric fields. Moreover, in future experiments it will be necessary to characterise the properties of the electrochemical interface in detail in terms of local electric fields in order to allow for direct comparison with our model.
|
[
"Nicola Seriani",
"Paola Delcompare-Rodriguez",
"Dhanshree Pandey",
"Abhishek Kumar Adak",
"Vikram Mahamiya",
"Carlos Pinilla",
"Hala J. El-Khozondar"
] |
https://doi.org/10.3390/ma17143460
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057743_p0
|
39057743
|
sec[0]/p[0]
|
1. Introduction
| 3.21875 |
biomedical
|
Other
|
[
0.5947265625,
0.0011844635009765625,
0.404052734375
] |
[
0.08392333984375,
0.90087890625,
0.014892578125,
0.0003752708435058594
] |
Deep-learning algorithms have been providing effective solutions for many tasks, contributing to the advancement of domains such as speech recognition and computer vision . One important computer vision task that is being tackled with such algorithms is image segmentation , a process that labels each pixel into categories, e.g., background and various cell types. However, deep-learning models require large quantities of annotated data for training. In addition, the provided annotations should also be of high quality. Specifically, the annotations should be accurate by providing information that reflects the reality within the input, and be complete, meaning that they provide all the information required for the given task, e.g., all pixels from an image have an associated label in a segmentation task.
|
[
"Serban Vădineanu",
"Daniël M. Pelt",
"Oleh Dzyubachyk",
"Kees Joost Batenburg"
] |
https://doi.org/10.3390/jimaging10070172
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057743_p1
|
39057743
|
sec[0]/p[1]
|
1. Introduction
| 4.09375 |
biomedical
|
Study
|
[
0.99853515625,
0.0002963542938232422,
0.0009431838989257812
] |
[
0.88525390625,
0.006195068359375,
0.10821533203125,
0.00027298927307128906
] |
For many biomedical imaging tasks, including cell imaging , the annotations are created manually by domain experts. Due to the limited availability of experts, the annotation process is often tedious , limiting the capacity for annotating the large quantities of data required by deep-learning algorithms. As a result, the general adoption of deep learning for such specialized domains may be considerably hindered. An annotation process with fewer quality constraints could significantly reduce the burden on expert annotators, enabling them to produce annotated images within a shorter time frame. For instance, when creating segmentation masks, the boundary of every cell in the image has to be carefully delineated. By providing coarser delineations, only annotating a subset of all cells, or relying on automatic but inaccurate segmentation tools based on classical image processing, a much faster annotation process can be achieved. However, training directly on low-quality annotations harms the performance of cell segmentation deep-learning algorithms . Thus, it becomes apparent that a solution that leverages inaccurate annotations to expand costly training data sets can greatly benefit the adoption of deep learning for cell segmentation.
|
[
"Serban Vădineanu",
"Daniël M. Pelt",
"Oleh Dzyubachyk",
"Kees Joost Batenburg"
] |
https://doi.org/10.3390/jimaging10070172
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999999 |
39057743_p2
|
39057743
|
sec[0]/p[2]
|
1. Introduction
| 3.855469 |
biomedical
|
Study
|
[
0.869140625,
0.000675201416015625,
0.13037109375
] |
[
0.78125,
0.056427001953125,
0.1619873046875,
0.0003204345703125
] |
Learning from imperfect or missing labels due to annotation constraints is a long-standing issue associated with machine-learning tasks. In the case of cell segmentation, obtaining large amounts of labeled data requires time-consuming efforts by experts with specialized knowledge of the task. One field concerned with this problem is weakly-supervised learning, where the aim is to train deep-learning algorithms to produce complete segmentation masks by only providing the models with partial annotations. Such techniques usually vary in the amount of information that is present in the annotations, which can include bounding boxes , rough sketches of shape contours , geometrical shape descriptors in the form of center points and lines , or partially-annotated segmentation areas . Despite their promising results, these techniques are generally tailored towards a single type of inconsistency, which can limit their applicability.
|
[
"Serban Vădineanu",
"Daniël M. Pelt",
"Oleh Dzyubachyk",
"Kees Joost Batenburg"
] |
https://doi.org/10.3390/jimaging10070172
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999999 |
39057743_p3
|
39057743
|
sec[0]/p[3]
|
1. Introduction
| 4.070313 |
biomedical
|
Study
|
[
0.99755859375,
0.0002987384796142578,
0.0022335052490234375
] |
[
0.98583984375,
0.0007395744323730469,
0.01325225830078125,
0.00008493661880493164
] |
Directly accounting for labelling errors, implicit consistency correction methods compensate for inaccuracies in the annotated input during the training process by, for instance, reducing the influence of gradients coming from segmentation areas of lower confidence , by using a teacher–student architecture to change the label of less confident areas in the annotation mask , or by using adversarial training to only annotate high-confidence areas of unlabeled data . On the other hand, explicit consistency correction solutions provide fine adjustments to the output of trained deep-learning models . Similarly to weakly supervised techniques, these methods lack a broad applicability and their utilization depends rigidly on custom architectures. When it comes to improving the provided labels, Yang et al. developed a solution for iteratively adjusting the manual annotations of retinal vessels by employing generative adversarial networks. Their framework, however, only produces small adjustments, relies on a relatively large amount of high-quality annotations, and may suffer from the challenges associated with generative models, e.g., mode collapse and convergence failure .
|
[
"Serban Vădineanu",
"Daniël M. Pelt",
"Oleh Dzyubachyk",
"Kees Joost Batenburg"
] |
https://doi.org/10.3390/jimaging10070172
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999995 |
39057743_p4
|
39057743
|
sec[0]/p[4]
|
1. Introduction
| 3.878906 |
biomedical
|
Study
|
[
0.96728515625,
0.00029540061950683594,
0.0325927734375
] |
[
0.90625,
0.0760498046875,
0.017730712890625,
0.00020325183868408203
] |
Also concerned with annotation scarcity, few-shot segmentation aims to segment new query images by leveraging information from relatively few support images with a limited amount of annotations. However, these approaches generally require additional training tasks with a large set of semantic classes whose annotations can be costly to obtain. The need for manual annotations can also be avoided by employing general foundation models such as the Segment Anything Model (SAM) , or cell-specific models such as Cellpose . However, although the applicability of such models is not confined to a single image modality or cell type, they do not generalize well to images outside their vast training pool. For instance, the SAM is not accurate when the targets have weak boundaries , which can be the case with cell images , whereas Cellpose is sensitive to variations in the texture of the objects . This may make these general solutions less suitable than techniques trained for a specific cell type.
|
[
"Serban Vădineanu",
"Daniël M. Pelt",
"Oleh Dzyubachyk",
"Kees Joost Batenburg"
] |
https://doi.org/10.3390/jimaging10070172
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057743_p5
|
39057743
|
sec[0]/p[5]
|
1. Introduction
| 4.253906 |
biomedical
|
Study
|
[
0.99853515625,
0.0004146099090576172,
0.0008826255798339844
] |
[
0.99853515625,
0.0006103515625,
0.0007696151733398438,
0.00008338689804077148
] |
In summary, although there are many methods designed for improving deep-learning segmentation with incorrect or incomplete labels, these solutions generally tackle specific types of inconsistencies, e.g., boundary uncertainty, require custom architectures or training schemes, or have considerable annotation requirements for additional training tasks. In this paper, we present a method designed to be applied to a wide set of inconsistencies, with low data requirements and a flexible training scheme allowing for a straightforward integration in other pipelines. We propose a framework for effectively obtaining large amounts of high-quality training data with limited required human annotation time for the task of cell segmentation. Our approach is based on manually annotating a small training set of high quality, which we then enlarge with a much larger set with low-quality annotations (possibly produced with considerably less human effort). In order to leverage the low-quality annotations, we train a convolutional neural network to learn the mapping for upgrading a low-quality annotation to a high-quality one by presenting it with both high-quality annotations as well as low-quality versions of these annotations. We create multiple types of erroneous annotations by perturbing the high-quality annotations with a function that approximates potential errors resulting from a low-quality annotation process. Moreover, we show that this perturbation function does not need to exactly replicate the annotation errors present in the low-quality annotations in order for a good mapping to be trained. The training process requires pairs of the perturbed annotations with their corresponding images as input for the upgrade network with the unperturbed, high-quality annotations as targets. We apply the learned mapping to the large low-quality set to enhance its annotations. Finally, we combine the initial small set of well-annotated data together with the larger set with upgraded annotations and use them for training accurate deep-learning models for the task of cell segmentation. By separating the inconsistency correction step, i.e., the upgrading of annotations, from the segmentation step, we enable our framework to tackle a wide array of inconsistencies and we facilitate its integration into other segmentation pipelines.
|
[
"Serban Vădineanu",
"Daniël M. Pelt",
"Oleh Dzyubachyk",
"Kees Joost Batenburg"
] |
https://doi.org/10.3390/jimaging10070172
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999995 |
39057743_p6
|
39057743
|
sec[0]/p[6]
|
1. Introduction
| 3.884766 |
biomedical
|
Study
|
[
0.99462890625,
0.00019884109497070312,
0.00507354736328125
] |
[
0.9951171875,
0.002979278564453125,
0.001575469970703125,
0.00009948015213012695
] |
This paper is an extended version of our paper published in . We extend our previous version by: Providing a formal introduction of the perturbations used throughout the experiments together with illustrative examples; Adding extensive experiments to further evaluate (i) the training requirements of the upgrade network, (ii) the effect of inconsistencies in the annotations of the high-quality data set, (iii) the trade-off between annotation cost and segmentation performance when upgrading the low-quality annotations, (iv) the performance of segmentation networks based on the annotation quality; Adding comparisons with related solutions; Adding experiments on complex RGB cell segmentation data sets; Expanding the discussion by elaborating on the potential consequences of our observations and scenarios under which our framework can contribute the most.
|
[
"Serban Vădineanu",
"Daniël M. Pelt",
"Oleh Dzyubachyk",
"Kees Joost Batenburg"
] |
https://doi.org/10.3390/jimaging10070172
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057743_p7
|
39057743
|
sec[1]/sec[0]/sec[0]/p[0]
|
2.1.1. Synthetic Data
| 4.121094 |
biomedical
|
Study
|
[
0.99853515625,
0.00026702880859375,
0.0010814666748046875
] |
[
0.99951171875,
0.0002238750457763672,
0.0001285076141357422,
0.00003606081008911133
] |
We opted to use synthetic data to study most aspects of our method since their ground truth annotations do not suffer from the inconsistencies a human annotator may induce. Thus, we can be confident that such external factors do not influence the outcomes of our experiments. Also, to isolate the effect of a particular type of inconsistency in the low-quality set, we apply perturbation functions (see Section 2.2.3 ) throughout the experimentation with synthetic data. We employ three data sets , which consist of microscopy images of HL60 nuclei cells, granulocytes, and both cell types, respectively, produced by a virtual microscope extracted from the Masaryk University Cell Image Collection ( https://cbia.fi.muni.cz/datasets ) . Each data set consists of 30 volumes of 129 slices, each containing 565 × 807 16-bit pixels. We filtered the volumes by eliminating the slices with empty labels, which resulted in differently sized volumes, averaging 84 slices per volume. In addition, 25 volumes were used for training, while 5 volumes were kept only for testing. In Figure 1 , we show sample slices and their corresponding high-quality annotations of the synthetic data sets. Since they are organized in volumes, we want to avoid selecting high-quality annotations of adjacent slices since such samples show little variation in their input and may be less informative when training the upgrade network than more distant slices. Consequently, we sample by subdividing a volume into a number of sections equal to the number of slices we want to select. We then select the middle slice of each section, thus ensuring an equidistant separation between slices. Additionally, when we sample from multiple volumes, we similarly partition each volume, but we select every next slice from a section belonging to a different volume in a circular manner. For instance, when taking a total of 5 slices from 5 volumes, the first slice will be selected from the center of section 1 from volume 1, the second from section 2 of volume 2 and so on.
|
[
"Serban Vădineanu",
"Daniël M. Pelt",
"Oleh Dzyubachyk",
"Kees Joost Batenburg"
] |
https://doi.org/10.3390/jimaging10070172
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39057743_p8
|
39057743
|
sec[1]/sec[0]/sec[1]/p[0]
|
2.1.2. Real Data
| 4.132813 |
biomedical
|
Study
|
[
0.99951171875,
0.00020694732666015625,
0.0004012584686279297
] |
[
0.99951171875,
0.0001659393310546875,
0.00023090839385986328,
0.00004208087921142578
] |
We also employ two manually-annotated data sets: the EPFL Hippocampus data set and a large-scale data set for colonic nuclear segmentation called Lizard . The EPFL data set is comprised of a training and a testing volume, each containing 165 slices of 768 × 1024 8-bit grayscale pixels. This set of images, obtained using focused ion beam scanning electron microscopy, is commonly used for benchmarking mitochondria segmentation algorithms, whose monitoring can provide, for instance, insights into the development of neurodegenerative diseases . The Lizard data set contains histology RGB images of colon tissue of varying sizes with instance labels for each cell. Among the six cell types annotated in this data set, we selected the most prevalent category, i.e., epithelial cells, as our target and the remaining cells as background objects. This choice allows us to test our method on the largest number of samples, which ensures that we obtain the most statistically significant results. We split the images into 500 × 500 patches, with 100 pixels overlapping between patches and removed patches that did not contain epithelial cells. We partitioned the resulting set into 1209 training and 288 testing patches. In this case, we assume the corresponding provided ground-truth annotations to be of high quality. For each data set, we select a small subset of samples for which we keep the high-quality annotations while perturbing the annotations of the remaining samples to generate the low-quality set. This perturbation step is performed only once per annotated image.
|
[
"Serban Vădineanu",
"Daniël M. Pelt",
"Oleh Dzyubachyk",
"Kees Joost Batenburg"
] |
https://doi.org/10.3390/jimaging10070172
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
39057743_p9
|
39057743
|
sec[1]/sec[1]/p[0]
|
2.2. Method
| 3.095703 |
biomedical
|
Study
|
[
0.8505859375,
0.0009264945983886719,
0.1485595703125
] |
[
0.822265625,
0.1759033203125,
0.001384735107421875,
0.00047469139099121094
] |
In Figure 2 , we illustrate an overview of our method. We consider a high-quality annotation process that produces labels in a slow and costly manner and a low-quality annotation process, yielding labels faster and cheaper. Within a given time frame, the processes would generate a small data set with high-quality labels and a larger lower-quality set. We apply perturbations to the well-annotated labels and we use the perturbed labels together with their corresponding images as input to train an upgrade model. We employ the upgrade model to enhance the labels of the larger data set, which we use in conjunction with the well-annotated samples to train the final segmentation model.
|
[
"Serban Vădineanu",
"Daniël M. Pelt",
"Oleh Dzyubachyk",
"Kees Joost Batenburg"
] |
https://doi.org/10.3390/jimaging10070172
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999995 |
39057743_p10
|
39057743
|
sec[1]/sec[1]/sec[0]/p[0]
|
2.2.1. Background
| 4.257813 |
biomedical
|
Study
|
[
0.9951171875,
0.0004203319549560547,
0.00429534912109375
] |
[
0.9990234375,
0.0006823539733886719,
0.0003581047058105469,
0.00005036592483520508
] |
We apply our framework to the segmentation task of 2-dimensional vector-valued (e.g., RGB, grayscale) images. In this paper, we define an image as a matrix of pixels x ∈ R N × M × C , where N , M , and C represent the number of rows, columns, and channels, respectively. The goal of segmentation is to create a mapping from a given input x to the target y ∈ Z N × M in order to provide a separation between the different entities within that image. Essentially, a label is attributed to each pixel according to the entity that it belongs to. When using deep learning for image segmentation, this mapping is approximated using convolutional neural networks (CNNs), f δ : R N × M × C → R N × M , which require a set of image-target pairs X = { ( x 1 , y 1 ) , ( x 2 , y 2 ) , … , ( x N t , y N t ) } , to train their parameters, δ . The process of training neural networks usually involves successive predictions based on the input x and adjusting the parameters such that the loss between the predictions and the labels is minimized. In order to achieve the desired results, the network requires well-annotated training samples. We describe the annotation process that produced high-quality labels as the output of the high-quality annotator, (1) A H Q : R N × M × C → A H Q , that receives an input image x and produces a label that belongs to the set of high-quality annotations, A H Q , i.e., it is both complete and correct. Such an annotation can be the result of a consensus between multiple experts or can require a slow and careful delineation of the shape of each element in x by a single expert. Additionally, we define the set of well-annotated images, X H Q = { ( x , A H Q ( x ) ) } , needed to train the network parameters, (2) δ ^ = argmin δ ∑ ( x , y ) ∈ X H Q L ( f δ ( x ) , y ) , where L is a loss function. Due to their large parameter count, these models are generally prone to overfitting and therefore require large quantities of well-annotated samples.
|
[
"Serban Vădineanu",
"Daniël M. Pelt",
"Oleh Dzyubachyk",
"Kees Joost Batenburg"
] |
https://doi.org/10.3390/jimaging10070172
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057743_p11
|
39057743
|
sec[1]/sec[1]/sec[1]/p[0]
|
2.2.2. Perturbation-Removal Framework
| 4.207031 |
biomedical
|
Study
|
[
0.99755859375,
0.00023365020751953125,
0.002002716064453125
] |
[
0.99462890625,
0.004352569580078125,
0.0009541511535644531,
0.00007218122482299805
] |
Since producing a sufficient number of high-quality annotations may prove unfeasible for cell segmentation, the required annotations may be supplied via a less rigorous annotation process. A low-quality annotation process would, for instance, result from an individual expert who quickly produces the annotation, without spending additional time on finer shape details or on removing ambiguities. Also, for setups that require consensus, the label can come from a single expert, a person in training, or a non-expert, thus reducing the annotation costs. Alternatively, the low-quality annotations can even be the product of traditional segmentation techniques (e.g., thresholding, graph cut , Otsu ) or machine-learning-based algorithms, removing the need for a human annotator in this stage of the process. For instance, one easily-applicable strategy to produce low-quality annotations is to simply train a segmentation network on the few available high-quality samples and then use its predictions on the remaining unannotated samples as low-quality annotations. We define the low-quality annotator (3) A L Q : R N × M × C → A L Q , as a function that produces labels that are either incorrect or incomplete or both, thus, being included in the set of low-quality annotations, A L Q .
|
[
"Serban Vădineanu",
"Daniël M. Pelt",
"Oleh Dzyubachyk",
"Kees Joost Batenburg"
] |
https://doi.org/10.3390/jimaging10070172
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057743_p12
|
39057743
|
sec[1]/sec[1]/sec[1]/p[1]
|
2.2.2. Perturbation-Removal Framework
| 4.171875 |
biomedical
|
Study
|
[
0.95849609375,
0.00063323974609375,
0.0408935546875
] |
[
0.9970703125,
0.0024433135986328125,
0.0003910064697265625,
0.0000623464584350586
] |
Training solely with low-quality annotations generally leads to inaccurate results . Thus, we propose a solution to enhance the quality of a larger set of low-quality annotations, which we utilize to enlarge an initially small set of high-quality annotations. Our framework requires a small number of well-annotated images, X H Q , together with a substantially larger set of images and their low-quality annotations, X L Q = { ( x , A L Q ( x ) ) } , with | X H Q | < | X L Q | . We aim to enhance A L Q ( x ) to A H Q by finding the upgrade function (4) U : ( R N × M × C , A L Q ) → A H Q , which translates an annotation of the input image created by the low-quality annotator to an annotation belonging to the space of high-quality annotations. In order to create both high- and low-quality versions of annotations, we utilize a perturbation function that aims to approximate the unknown mapping from a high-quality annotation to a low-quality one. We handcraft function (5) P : A H Q → A L Q , which applies perturbations to a high-quality annotation to create an annotated image that approximates a faster, but lower-quality, annotation process. The choice for such a function can vary by task and data set, with implementations that can include heuristics or even learning the perturbations from the data. In our work, we assume that we can approximate the perturbation function, P , by implementing a custom stochastic version of it. Additionally, we assume that the function U that maps the low-quality label to a high-quality one is a learnable function. We employ the high-quality set to generate many ( x , P ( A H Q ( x ) ) ) pairs. Given the stochastic nature of our chosen perturbation function, we can generate multiple perturbed versions of the same high-quality annotation; thus, we only require a small number of ( x , A H Q ( x ) ) pairs. We utilize the generated pairs to train an upgrade network, u θ , parameterized by θ , which approximates U by finding (6) θ ^ = argmin θ ∑ ( x , A H Q ( x ) ) ∈ X H Q L ( u θ ( x , P ( A H Q ( x ) ) ) , A H Q ( x ) ) , where L is a loss function. After training u θ , we apply it to our lower-quality set. In this way, we enhance the low-quality annotations, which results in the pairs ( x , u θ ( x , A L Q ( x ) ) ) of input images and upgraded annotations. Finally, we use both the enhanced ( x , u θ ( x , A L Q ( x ) ) ) pairs and the initial high-quality ( x , A H Q ( x ) ) pairs as training samples for our final segmentation task. Therefore, our segmentation CNN f δ will be obtained as (7) δ ^ = argmin δ ( ∑ ( x , y ) ∈ X H Q L ( f δ ( x ) , y ) + ∑ ( x , y ) ∈ X L Q L ( f δ ( x ) , u θ ^ ( x , y ) ) ) .
|
[
"Serban Vădineanu",
"Daniël M. Pelt",
"Oleh Dzyubachyk",
"Kees Joost Batenburg"
] |
https://doi.org/10.3390/jimaging10070172
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057743_p13
|
39057743
|
sec[1]/sec[1]/sec[1]/p[2]
|
2.2.2. Perturbation-Removal Framework
| 4.117188 |
biomedical
|
Study
|
[
0.98974609375,
0.0005931854248046875,
0.0095062255859375
] |
[
0.9765625,
0.02288818359375,
0.0005154609680175781,
0.0001659393310546875
] |
Algorithm 1 shows the pseudocode of a segmentation pipeline that makes use of our upgrade network. The requirements of our framework are (1) a small set with high-quality annotations, (2) a larger set with low-quality annotations, and (3) a perturbation function. The objective of this pipeline is to obtain the parameters δ of a well-trained segmentation network. We initially train the upgrade network u θ only on the high-quality data X H Q , whose labels we perturb with the previously selected perturbation function, P . We aim here to obtain predictions from input images and perturbed labels that match the high-quality annotations as closely as possible. After estimating the parameters of u θ , we apply it to X L Q , whose images and resulting upgraded annotations we employ, in conjunction with X H Q , to estimate the parameters δ of a segmentation network. Algorithm 1 Upgrade Framework Require: X H Q , X L Q , P return δ (1) Train the upgrade network u θ : for ( x , y ) ∈ X H Q do Perturb y : P ( y ) Predict upgraded label: u θ ( x , P ( y ) ) Compute loss: L ( u θ ( x , P ( y ) ) , y ) end for Estimate θ ^ according to Equation ( 6 ) (2) Upgrade low-quality set and expand segmentation training data: for ( x , y ) ∈ X L Q do Upgrade low-quality label: u θ ( x , y ) end for Estimate δ ^ according to Equation ( 7 )
|
[
"Serban Vădineanu",
"Daniël M. Pelt",
"Oleh Dzyubachyk",
"Kees Joost Batenburg"
] |
https://doi.org/10.3390/jimaging10070172
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39057743_p14
|
39057743
|
sec[1]/sec[1]/sec[2]/p[0]
|
2.2.3. Producing Low-Quality Annotations
| 4.066406 |
biomedical
|
Study
|
[
0.99853515625,
0.00017249584197998047,
0.0011806488037109375
] |
[
0.990234375,
0.00923919677734375,
0.0003018379211425781,
0.0001061558723449707
] |
We designed our method for the task of binary cell segmentation, where the object of interest is a single type of cell. In order to apply our perturbation function, we require the instance label of every cell in the image. Therefore, considering E cells in image x , we define L ⊂ Z as the set of all cell instance labels, with | L | = E . Our label then becomes (8) y n m = i , if x n m belongs to cell i ∈ L , 0 , otherwise 1 ≤ n ≤ N , 1 ≤ m ≤ M .
|
[
"Serban Vădineanu",
"Daniël M. Pelt",
"Oleh Dzyubachyk",
"Kees Joost Batenburg"
] |
https://doi.org/10.3390/jimaging10070172
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
39057743_p15
|
39057743
|
sec[1]/sec[1]/sec[2]/p[1]
|
2.2.3. Producing Low-Quality Annotations
| 4.003906 |
biomedical
|
Study
|
[
0.998046875,
0.00019299983978271484,
0.00174713134765625
] |
[
0.98388671875,
0.0153656005859375,
0.0007367134094238281,
0.00009495019912719727
] |
We apply three types of perturbations (omission, inclusion, and bias), introduced in , which are designed to reflect the incompleteness and inaccuracy of the cell segmentation masks resulting from an annotation process with fewer resources. For instance, a much shorter annotation time can be spent by using segmentation masks that only contain a proportion of the total cells present in the image. Moreover, allowing for inconsistencies in cell recognition in the form of inclusions can also reduce the time an annotator spends choosing which cells to include in the segmentation mask. Finally, by eliminating the need to provide correct cell border delineations, we can expect a boost in the annotation speed.
|
[
"Serban Vădineanu",
"Daniël M. Pelt",
"Oleh Dzyubachyk",
"Kees Joost Batenburg"
] |
https://doi.org/10.3390/jimaging10070172
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
39057743_p16
|
39057743
|
sec[1]/sec[1]/sec[2]/p[2]
|
2.2.3. Producing Low-Quality Annotations
| 3.666016 |
biomedical
|
Study
|
[
0.94140625,
0.00038695335388183594,
0.058197021484375
] |
[
0.994140625,
0.00528717041015625,
0.00039696693420410156,
0.00008171796798706055
] |
Given the relatively ill-defined distinction between low-quality and high-quality annotations, we will further consider as low-quality annotations only the ones affected by large degrees of perturbations, i.e., 70% omission, 70% inclusion, a bias of 6, or a combination of perturbations. Thus, we only consider as low-quality the annotations that significantly diverge from the gold standard. In Figure 3 , we illustrate an example of an annotation where all three perturbation types are present and highlighted. Alternatively, we investigate the case where the low-quality annotator A L Q would not imply any human effort. This can happen when A L Q are produced by a segmentation network trained on the small number of samples in the high-quality set A H Q . In this case, the generation of low-quality annotations is disentangled from the perturbations that we apply when training the upgrade network.
|
[
"Serban Vădineanu",
"Daniël M. Pelt",
"Oleh Dzyubachyk",
"Kees Joost Batenburg"
] |
https://doi.org/10.3390/jimaging10070172
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057743_p17
|
39057743
|
sec[1]/sec[2]/p[0]
|
2.3. Experimental Setup
| 4.109375 |
biomedical
|
Study
|
[
0.9931640625,
0.0005655288696289062,
0.006328582763671875
] |
[
0.99951171875,
0.00034546852111816406,
0.00017845630645751953,
0.00004297494888305664
] |
We designed our experimental setup around a PyTorch implementation of UNet . UNet features an encoder–decoder architecture with skip connections between the encoding layers and the decoding layers of the same spatial resolution. We employed 4 convolutional blocks in the encoder and 4 in the decoder, with a block containing 2 convolutional and 2 batch normalisation layers. We treat both the segmentation and upgrade tasks as binary pixel-wise classification tasks. Thus, the output of the network in both cases is a two-channel image with the first channel’s pixels being 0 if they belong to the foreground and 1 if they belong to the background, with the opposite holding true for the second channel. All activations between layers are ReLU functions, with the exception of the last layer, where the output is processed by a soft-max function. We train the network until there is no improvement in the validation score for 10 consecutive epochs, at which point we only keep the model with the highest score. Our loss function is the Dice loss, and we update the network’s parameters according to ADAM optimization algorithm , with a learning rate of 10 − 5 and a batch size of 4. We partition our data into training and testing with an additional 80/20 split of the training data into training and validation. Finally, we present our results by reporting the Sørensen–Dice coefficient computed over the entire test set and averaged over 5 runs. We validated our comparisons by using the Wilcoxon non-parametric test .
|
[
"Serban Vădineanu",
"Daniël M. Pelt",
"Oleh Dzyubachyk",
"Kees Joost Batenburg"
] |
https://doi.org/10.3390/jimaging10070172
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999999 |
39057743_p18
|
39057743
|
sec[2]/p[0]
|
3. Results
| 3.546875 |
biomedical
|
Study
|
[
0.9541015625,
0.00067901611328125,
0.045074462890625
] |
[
0.99560546875,
0.0034275054931640625,
0.0007529258728027344,
0.00012922286987304688
] |
We performed a series of experiments to analyze various aspects of our proposed framework. In Section 3.1 , we use the synthetic data sets with objective ground truth to measure the quality gain of upgraded annotations under various sets of assumptions. On the same data sets, we also evaluate the benefits of expanding the segmentation training data with upgraded annotations in terms of segmentation performance and annotation cost ( Section 3.2 ). Furthermore, in Section 3.3 , we validate our previous observations on real manually-annotated data. Lastly, we show, in Section 3.4 , a case study of an application where our solution can be integrated to improve the prediction quality of a segmentation network trained with insufficient samples.
|
[
"Serban Vădineanu",
"Daniël M. Pelt",
"Oleh Dzyubachyk",
"Kees Joost Batenburg"
] |
https://doi.org/10.3390/jimaging10070172
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39057743_p19
|
39057743
|
sec[2]/sec[0]/p[0]
|
3.1. Analysis of the Upgrade Network
| 4.050781 |
biomedical
|
Study
|
[
0.9951171875,
0.00038504600524902344,
0.004573822021484375
] |
[
0.99951171875,
0.00018227100372314453,
0.00013625621795654297,
0.00003254413604736328
] |
To assess the optimal training set size for the upgrade network u θ , we created various training sets by varying both the total number of annotated slices and the number of volumes from which the annotated slices were selected. The models were trained to upgrade annotations affected by 70% omission, 70% inclusion, and a bias of 6, respectively. The results presented in Figure 4 show that the upgrade network requires just 5 well-annotated slices to improve the quality of the annotations, regardless of the applied perturbation. We also notice that the resulting quality of the upgraded annotations plateaus quickly to Dice values > 0.9. We report the optimal number of training slices for different perturbations together with the corresponding Dice score of the upgraded annotations in Table 1 .
|
[
"Serban Vădineanu",
"Daniël M. Pelt",
"Oleh Dzyubachyk",
"Kees Joost Batenburg"
] |
https://doi.org/10.3390/jimaging10070172
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057743_p20
|
39057743
|
sec[2]/sec[0]/p[1]
|
3.1. Analysis of the Upgrade Network
| 4.035156 |
biomedical
|
Study
|
[
0.99658203125,
0.0003368854522705078,
0.00299835205078125
] |
[
0.99951171875,
0.0002491474151611328,
0.00013387203216552734,
0.00003606081008911133
] |
So far, we assumed that we can perfectly model the errors affecting the low-quality annotations with our perturbation functions. However, in practice, it might be difficult to exactly match the type and severity of the perturbations present in the data. To account for that, we relax this assumption by allowing a mismatch between the error generated by the perturbation functions and the errors in X L Q . In Table 2 , we report the effect of such mismatch on the performance of the upgrade network when the annotations of X L Q contain 30 % omission, 30 % inclusion, and a bias severity of 4, respectively. We observe that, even when not reaching the highest Dice scores, the upgraded annotations show high Dice scores when u θ is trained on the highest perturbation level. This implies that varying the presence of a large proportion of the cell masks can be more beneficial for training u θ than aiming to exactly match the amount of error present in the X L Q .
|
[
"Serban Vădineanu",
"Daniël M. Pelt",
"Oleh Dzyubachyk",
"Kees Joost Batenburg"
] |
https://doi.org/10.3390/jimaging10070172
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39057743_p21
|
39057743
|
sec[2]/sec[0]/p[2]
|
3.1. Analysis of the Upgrade Network
| 4.128906 |
biomedical
|
Study
|
[
0.9990234375,
0.00024044513702392578,
0.0006928443908691406
] |
[
0.99951171875,
0.00018405914306640625,
0.0001888275146484375,
0.000035881996154785156
] |
In addition to the perturbation function, another essential requirement of our solution is the presence of a high-quality set of annotations for training u θ . Since we use synthetic data, the quality of this set is ideal, which, however, is not expected from manual annotations for many reasons, including inter-observer variability or limited available resources. We model these inaccuracies by introducing moderate amounts of perturbations into the high-quality set. Figure 5 illustrates that the upgrade networks trained on the larger HL60 cells are robust to imperfect HQ annotations, whereas the ones trained on granulocytes are more sensitive due to the comparatively smaller footprint of the cells. Thus, the same amount of perturbation affects the quality of the granulocytes annotations more drastically than that of HL60 cells. Despite allowing for a moderate amount of omission and inclusion perturbations, the networks trained on granulocytes show a sharp drop in performance for bias since this type of perturbation introduces the greatest variation in shape relative to cell size among the two data sets.
|
[
"Serban Vădineanu",
"Daniël M. Pelt",
"Oleh Dzyubachyk",
"Kees Joost Batenburg"
] |
https://doi.org/10.3390/jimaging10070172
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057743_p22
|
39057743
|
sec[2]/sec[0]/p[3]
|
3.1. Analysis of the Upgrade Network
| 4.050781 |
biomedical
|
Study
|
[
0.9990234375,
0.00034236907958984375,
0.0006198883056640625
] |
[
0.99853515625,
0.00017344951629638672,
0.0011806488037109375,
0.00005340576171875
] |
We compare our solutions with works tackling the issue of training biomedical image segmentation models with imperfect or incomplete annotations. We selected techniques that employ full-size segmentation masks for training and that apply corrections to these masks to either fill incomplete areas or remove incorrect ones. Also, although we compare the selected methods for all our perturbation types, it is important to note that Partial Labeling was designed for setups closer to omission than the other perturbations, whereas Confident Learning tackles uncertain areas at the border of the masked areas resembling more our bias perturbation. In Table 3 , we observe that our method generates comparable results with Partial Labeling for omission and Confident Learning for bias perturbation. However, among all three perturbation types, our framework performs consistently better than the other solutions, showing wider applicability to different types of inconsistency.
|
[
"Serban Vădineanu",
"Daniël M. Pelt",
"Oleh Dzyubachyk",
"Kees Joost Batenburg"
] |
https://doi.org/10.3390/jimaging10070172
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
39057743_p23
|
39057743
|
sec[2]/sec[1]/p[0]
|
3.2. Segmentation Improvements
| 4.042969 |
biomedical
|
Study
|
[
0.99169921875,
0.0004520416259765625,
0.0078582763671875
] |
[
0.99951171875,
0.0003402233123779297,
0.00019180774688720703,
0.00003731250762939453
] |
In Section 3.1 , we investigated the capability of the upgrade network to improve the quality of annotations affected by errors. In this section, we are analyzing whether adding the upgraded annotations to the training set results in improved segmentation performance and reduced overall annotation costs. In Table 1 , we report different scenarios under which X H Q and X L Q can be used to train networks for segmentation. Given an initial data set with low-quality annotations, we can use it directly as a training set for segmentation (LQ only column in Table 1 ). We can also spend additional resources on improving the quality of a small number of annotations and utilize them in conjunction with the low-quality set (column HQ + LQ) or we can employ the high-quality set alone for training (column HQ). Finally, we can use our framework for upgrading the low-quality annotations and, together with X H Q , forming a larger training set of improved quality for the segmentation network (column HQ + upgraded). In order to ensure that the synthetic data cannot be easily segmented based on the pixel intensity levels, we use as baseline a simple thresholding solution in which the input images are segmented by selecting a threshold via grid search with a step of 1% of the maximum pixel intensity. For each data set, we select a single threshold that yields the highest Dice score on the training set. The low baseline results in Table 1 reflect the complexity of the simulated data sets. Our results show that, for most cases where u θ improved the quality of annotations, the addition of samples with upgraded annotations translated into a higher segmentation performance of the final segmentation network on the test data.
|
[
"Serban Vădineanu",
"Daniël M. Pelt",
"Oleh Dzyubachyk",
"Kees Joost Batenburg"
] |
https://doi.org/10.3390/jimaging10070172
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
39057743_p24
|
39057743
|
sec[2]/sec[1]/p[1]
|
3.2. Segmentation Improvements
| 4.121094 |
biomedical
|
Study
|
[
0.998046875,
0.00042819976806640625,
0.0014629364013671875
] |
[
0.99951171875,
0.00015175342559814453,
0.0002186298370361328,
0.00003701448440551758
] |
From Table 1 , we observed that adding the upgraded annotations to the training set results in better segmentation. However, this performance gain resulted from upgrading a large number of low-quality annotations, which may also prove difficult to produce in practice. To account for this, we perform an experiment analyzing the trade-off between annotation cost and performance. For a fixed number of slices, we select 10% of them to have high-quality annotations, while the rest have low-quality annotations. We apply our framework to this set of slices and compare against segmentation networks trained with low-quality annotations, i.e., 0% high-quality slices, and against segmentation networks trained on high-quality slices only, i.e., 100% high-quality slices. We define the annotation cost as the equivalent number of low-quality annotations that would be produced with the same effort as a given annotation. For instance, for a low-quality annotation, the equivalent number of low-quality annotations is 1, while for a high-quality annotation, this number will differ depending on the particularities of the task, such as the data sets or the experience of the annotators, as is the case with works comparing annotation costs in the literature . For illustration purposes, we consider the equivalent number of low-quality annotations for a high-quality annotation to be 5. We observe in Figure 6 that, except for bias perturbation, the segmentation networks trained with our framework are the most cost-effective option for reaching the highest Dice scores. When it comes to bias, the variation in cell size induced in the training set with low-quality annotations forces the network to learn an “average” cell size that matches more closely the ground truth in the test set. However, in cases where the bias is more systematic, we expect a considerable drop in performance for the networks trained only with low-quality labels.
|
[
"Serban Vădineanu",
"Daniël M. Pelt",
"Oleh Dzyubachyk",
"Kees Joost Batenburg"
] |
https://doi.org/10.3390/jimaging10070172
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
39057743_p25
|
39057743
|
sec[2]/sec[2]/p[0]
|
3.3. Enhancing Manual Annotations
| 4.148438 |
biomedical
|
Study
|
[
0.9990234375,
0.00023949146270751953,
0.000827789306640625
] |
[
0.99951171875,
0.00021564960479736328,
0.0002605915069580078,
0.00003820657730102539
] |
In Section 3.1 , we showed that the upgrade network is able to improve low-quality annotations of synthetic images under various circumstances. Here, we expand our analysis by validating our observations on real cell images. We integrate the two described real data sets in a scenario emulating the process experts may undertake to enhance the quality of their annotations. Our goal is to assess whether the quality gains reported in Table 1 can be similarly reproduced on real manually-annotated data. We consider a setup where the constraints on the annotation process are accurately captured by the perturbation functions used during the training of the upgrade network. With omission, we model an expert that deliberately ignores most cells in an image, focusing only on 30% of them. Inclusion allows for the presence of other structures that, for instance, can result from using networks trained on other cell data sets, or from foundation models. Bias would allow the annotator to either focus on the “core” of the cell, as shown in Figure 7 m,o, or on the wider cell area without rigorously delineating the boundaries. Figure 7 shows the results of the upgrade network trained on 24% of the training samples of EPFL, and on 20% of Lizard’s, respectively. We notice, both qualitatively and quantitatively, that our solution can successfully upgrade annotations affected by high perturbation levels, requiring a relatively low number of high-quality annotations for real, more complex data sets. Also, the large quality increase for omission and bias highlights the potential of our framework to expand the size of cell data sets with relatively low effort for producing the low-quality annotations.
|
[
"Serban Vădineanu",
"Daniël M. Pelt",
"Oleh Dzyubachyk",
"Kees Joost Batenburg"
] |
https://doi.org/10.3390/jimaging10070172
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
39057743_p26
|
39057743
|
sec[2]/sec[3]/p[0]
|
3.4. Case Study: Upgrading Low-Quality Predictions
| 4.136719 |
biomedical
|
Study
|
[
0.99658203125,
0.00037670135498046875,
0.0031280517578125
] |
[
0.99951171875,
0.0003113746643066406,
0.0001990795135498047,
0.0000426173210144043
] |
We showcase here an example where the upgrade network can be applied in a scenario requiring no manual annotation cost for producing the low-quality annotations. In this case, X H Q can be employed to train a segmentation network whose predictions can then be further used as the cheap annotations of X L Q . We consider the predictions of a segmentation network trained with 10 well-annotated samples of Lizard data set in a setup similar to . We use the same X H Q for training our upgrade network. We opted for a set of perturbations that would guide u θ to compensate for prediction inaccuracies that we visually assessed. At each training iteration, we perform a 50% omission, followed by the inclusion of 10% of the segments extracted by Felzenszwalb’s algorithm to emulate missing or mispredicted structures. We add salt and pepper noise with a 10% probability to mimic the observed gaps in the segmented area as well as the small clusters of false positive pixels that can be noticed in Figure 8 d. Finally, the resulting label is subjected to bias perturbation with a bias of 6 to guide the upgrade network towards better delineation of cell boundaries. The results, shown in Figure 8 , demonstrate the potential of our method to refine the predictions of an undertrained segmentation network. We achieve a 22% improvement in the quality of the predictions without requiring additional supervision. Moreover, by visually inspecting the results, we notice that u θ achieves good separation between the individual shapes, a property not captured by the Dice score metric. These delineated shapes can then be used, for instance, to facilitate a further instance segmentation step.
|
[
"Serban Vădineanu",
"Daniël M. Pelt",
"Oleh Dzyubachyk",
"Kees Joost Batenburg"
] |
https://doi.org/10.3390/jimaging10070172
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057743_p27
|
39057743
|
sec[3]/p[0]
|
4. Discussion
| 4.066406 |
biomedical
|
Study
|
[
0.9970703125,
0.0003745555877685547,
0.0024051666259765625
] |
[
0.99951171875,
0.000339508056640625,
0.00019371509552001953,
0.0000393986701965332
] |
Our results reported in Table 1 indicate that, with as few as 10 well-annotated images, we can improve low-quality annotations to a level comparable with the gold standard. In addition, as can be seen in Figure 4 , the performance of the upgrade network relative to the size of the high-quality data set follows a logarithmic trend. Therefore, continuously increasing the size of X H Q will not generate meaningful improvements. By knowing the logarithmic trend of the performance of u θ , the end user of our framework would benefit from being able to decide more easily when enough high-quality data has been gathered and annotated, since, once u θ performs well for a certain size of X H Q , little improvement can be expected when the size of the training is increased. Furthermore, we showed that the upgrade network produced positive results for all considered cell data sets. The only requirements are a small high-quality set of annotations, a separate larger set of low-quality annotations, and a perturbation function that can map a high-quality annotation to multiple lower-quality versions of it, resembling the quality within the low-quality set. Since our requirements are independent of the data set, we expect our method to also work on other image modalities where our assumptions are met. This also applies to data collected in the three-dimensional regime, such as tomography. In this case, our framework can be applied on each individual slice separately.
|
[
"Serban Vădineanu",
"Daniël M. Pelt",
"Oleh Dzyubachyk",
"Kees Joost Batenburg"
] |
https://doi.org/10.3390/jimaging10070172
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057743_p28
|
39057743
|
sec[3]/p[1]
|
4. Discussion
| 4.097656 |
biomedical
|
Study
|
[
0.99560546875,
0.0005173683166503906,
0.0037555694580078125
] |
[
0.99951171875,
0.00027370452880859375,
0.00035834312438964844,
0.00004112720489501953
] |
We observed that using both the upgraded annotations of the low-quality set together with the small well-annotated set generally results in higher segmentation scores. Moreover, we noticed that the highest Dice scores are obtained when the upgrade model is both trained with and applied to annotations perturbed with 70% omission, 70% inclusion, or a bias of 6. We also saw in Figure 6 that our framework can be a cost-effective solution to increase the performance of segmentation networks when the annotation time is a constraint. Moreover, by comparing with other works targeting the enhancement of imperfect annotations, we showed that our upgrade network can handle a wider variety of perturbations than existing techniques. Thus, our solution is well-suited for being embedded into an annotation process with limited resources, rather than for fine-tuning, where there is a wide gap between the cost of producing a low-quality annotation and the cost of producing a high-quality one. For instance, for automatically-produced annotations by a non-learning algorithm, the only costly requirement would be to manually enhance a small proportion of them, on which the upgrade network can be trained. Moreover, as shown in Figure 8 , our solution is flexible enough to be used for upgrading predictions of a network trained with insufficient data. These upgraded predictions can then be used to enlarge the existing data set or be further adjusted by experts, reducing the overall annotation time.
|
[
"Serban Vădineanu",
"Daniël M. Pelt",
"Oleh Dzyubachyk",
"Kees Joost Batenburg"
] |
https://doi.org/10.3390/jimaging10070172
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
39057743_p29
|
39057743
|
sec[3]/p[2]
|
4. Discussion
| 4.011719 |
biomedical
|
Study
|
[
0.982421875,
0.0004813671112060547,
0.01715087890625
] |
[
0.99951171875,
0.000324249267578125,
0.00019109249114990234,
0.000037610530853271484
] |
We also noticed the benefit of training for high perturbation levels, i.e., 70% omission, 70% inclusion, and a bias of 6, when we tested the robustness of our solution with respect to discrepancies between the perturbation levels used to train the upgrade network and the perturbation levels in the low-quality set. In Table 2 , we saw that, generally, when we train for the highest perturbation level we reach comparable, or higher, performance than when training on the same perturbation applied to generate the low-quality set. Since, in practice, the annotation inaccuracies can have a systematic, i.e., annotator-specific, component and a random component, it may prove impossible to exactly model these inaccuracies through perturbations. Thus, the robustness to discrepancies in perturbation levels shown by our framework can indicate its potential applicability in practical scenarios. We additionally showed that our framework is robust to reductions in the quality of X H Q . Figure 5 shows that we can expect a relatively small drop in performance when we moderately reduce the quality of the well-annotated set. This observation may imply that the annotation process of X H Q can become less costly, e.g., requiring fewer experts per high-quality annotation, while still being able to produce annotations to train a well-performing upgrade network. However, the less information is present in an annotation, e.g., small cell areas, the more sensitive the framework becomes to inconsistencies.
|
[
"Serban Vădineanu",
"Daniël M. Pelt",
"Oleh Dzyubachyk",
"Kees Joost Batenburg"
] |
https://doi.org/10.3390/jimaging10070172
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39057743_p30
|
39057743
|
sec[3]/p[3]
|
4. Discussion
| 4.023438 |
biomedical
|
Study
|
[
0.99658203125,
0.0002903938293457031,
0.00298309326171875
] |
[
0.99951171875,
0.0003211498260498047,
0.0002415180206298828,
0.00003463029861450195
] |
Given that we focused solely on cell segmentation, we are unable to conclude with certainty whether or not our framework is applicable to other image segmentation applications where the goal would diverge from the cell segmentation setup, for instance by requiring the segmentation of a single contiguous target object. However, considering that our framework does not demand a specific type of annotation, as long as sufficient realistic low-quality versions of the high-quality annotations can be created with enough variety between them, we expect the upgrade network to still be applicable. Despite this, further experimentation is required to ensure that our requirements are met by other segmentation applications. Another limitation presented by our work is the lack of integration of the third dimension for volumetric data sets. This can be tackled in the future by, for instance, employing an architecture with 3D convolutions as the upgrade network. Finally, throughout our experimentation, we upgraded only annotations suffering from high levels of inconsistencies, while ignoring the fine-tuning of less severe cases. We expect our upgrade network to not perform similarly well on such cases, given that the small errors would not allow for much variation in the generation of the low-quality versions of the annotations. This would then impede the network from learning a generalizable mapping from a low-quality annotation to a high-quality one.
|
[
"Serban Vădineanu",
"Daniël M. Pelt",
"Oleh Dzyubachyk",
"Kees Joost Batenburg"
] |
https://doi.org/10.3390/jimaging10070172
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
39057743_p31
|
39057743
|
sec[4]/p[0]
|
5. Conclusions
| 3.386719 |
biomedical
|
Study
|
[
0.97314453125,
0.00048613548278808594,
0.026458740234375
] |
[
0.8720703125,
0.1265869140625,
0.0008769035339355469,
0.00035500526428222656
] |
We presented our framework for enlarging training data sets with limited human annotation costs by only requiring a small set of data with high-quality annotations and a larger set with low-quality annotations that would require little or no human annotation effort. We utilize a small high-quality data set whose annotation quality is reduced for providing it as input to an upgrade network that learns the mapping from a low-quality annotation to a high-quality one. We then use the upgrade network to enhance the annotation quality of the larger low-quality set.
|
[
"Serban Vădineanu",
"Daniël M. Pelt",
"Oleh Dzyubachyk",
"Kees Joost Batenburg"
] |
https://doi.org/10.3390/jimaging10070172
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39057743_p32
|
39057743
|
sec[4]/p[1]
|
5. Conclusions
| 3.910156 |
biomedical
|
Study
|
[
0.990234375,
0.0004215240478515625,
0.00934600830078125
] |
[
0.9990234375,
0.0007600784301757812,
0.000331878662109375,
0.00006860494613647461
] |
We observed that our solution is applicable to at least three types of annotation inconsistencies (omission, inclusion, and bias), that it is robust to changes in the annotation quality of the training set, and that it can have wider applicability than existing works. We showed that our work can be applied to enhance the low-quality predictions of a network trained on an insufficient number of samples. Finally, we showed that the networks trained on data sets enlarged by our method present higher segmentation scores than only training on high-quality data.
|
[
"Serban Vădineanu",
"Daniël M. Pelt",
"Oleh Dzyubachyk",
"Kees Joost Batenburg"
] |
https://doi.org/10.3390/jimaging10070172
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39057403_p0
|
39057403
|
sec[0]/p[0]
|
1. Introduction
| 3.986328 |
biomedical
|
Study
|
[
0.99951171875,
0.00014710426330566406,
0.00032806396484375
] |
[
0.83984375,
0.01065826416015625,
0.14892578125,
0.00041961669921875
] |
Sea stars, animals belonging to the phylum Echinodermata, produce varied low molecular weight metabolites, the chemical structures that differ significantly from the metabolites of various representatives of terrestrial flora and fauna. The most abundant natural products in starfish were established to be sterols, polyhydroxysteroids, steroid glycosides, ceramides, cerebrosides, and gangliosides . These secondary metabolites have been reported to exhibit diverse biological activities such as cytotoxic , antitumor , anti-inflammatory , neuritogenic , etc. .
|
[
"Alla A. Kicha",
"Dmitriy K. Tolkanov",
"Timofey V. Malyarenko",
"Olesya S. Malyarenko",
"Alexandra S. Kuzmich",
"Anatoly I. Kalinovsky",
"Roman S. Popov",
"Valentin A. Stonik",
"Natalia V. Ivanchina",
"Pavel S. Dmitrenok"
] |
https://doi.org/10.3390/md22070294
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39057403_p1
|
39057403
|
sec[0]/p[1]
|
1. Introduction
| 4.324219 |
biomedical
|
Study
|
[
0.9990234375,
0.0002849102020263672,
0.0008006095886230469
] |
[
0.998046875,
0.0002791881561279297,
0.0018482208251953125,
0.0000616908073425293
] |
Starfish of the genus Henricia Gray, 1840 (order Spinulosida, family Echinasteridae) inhabit mainly temperate and arctic waters. About 50 species of starfish belong to this genus. Henricia spp. are widely distributed in the North Pacific Ocean, especially in the Bering and Okhotsk Seas. Species of the genus Henricia are highly variable. Many of them are very close to each other, making their identification difficult in some cases . It is known that sea stars contain a variety of polar steroid metabolites. So far, efforts have been made to study the steroid composition of seven species of the genus, namely, H. leviuscula (earlier erroneously named as H. laeviuscola ) , H. downeyae , H. sanguinolenta , H. leviuscula leviuscula , H. aspera , H tumida , and H. derjugini . Most of the isolated polar steroids are either sulfated or non-sulfated polyhydroxysteroids or structurally related to them monoglycosides with a monosaccharide residue at C-3 of the aglycon or so-called “two-chains” glycosides with two monosaccharide units attached at different positions of the aglycon, namely, in the steroid nucleus at C-3 and in the steroid side chain at C-24. Summarily, these studies demonstrate a high diversity of polar steroids in sea stars of the genus Henricia , which is consistent with the wide biological variability of species in this genus. However, “classical” asterosaponins, which are monosulfated steroid oligoglycosides with five to six monosaccharide residues, were not found in the species studied, with the exception of henricioside A from H. leviuscula .
|
[
"Alla A. Kicha",
"Dmitriy K. Tolkanov",
"Timofey V. Malyarenko",
"Olesya S. Malyarenko",
"Alexandra S. Kuzmich",
"Anatoly I. Kalinovsky",
"Roman S. Popov",
"Valentin A. Stonik",
"Natalia V. Ivanchina",
"Pavel S. Dmitrenok"
] |
https://doi.org/10.3390/md22070294
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39057403_p2
|
39057403
|
sec[0]/p[2]
|
1. Introduction
| 4.183594 |
biomedical
|
Study
|
[
0.99951171875,
0.00019562244415283203,
0.00021147727966308594
] |
[
0.982421875,
0.000545501708984375,
0.0167999267578125,
0.00014066696166992188
] |
Steroid metabolites isolated from sea stars of the genus Henricia exhibited diverse biological effects. Thus, some compounds have shown antifungal activity , the ability to inhibit cell division of fertilized sea urchin eggs , cytotoxic activity against non-small-cell lung human carcinoma , and hemolytic effect against mouse erythrocytes . In addition, leviusculoside G from H. leviuscula was shown to induce apoptosis in cancer cells and decrease the pro-carcinogenic transformation of normal cells. A possible molecular mechanism was proposed through the induction of p53-dependent apoptosis and inhibition of AP-1, NF-κB, and ERKs activities. Thereby, steroid metabolites isolated from Henricia spp. are of interest for further study of their structures and biological activity, especially as anti-cancer and cancer-preventive compounds .
|
[
"Alla A. Kicha",
"Dmitriy K. Tolkanov",
"Timofey V. Malyarenko",
"Olesya S. Malyarenko",
"Alexandra S. Kuzmich",
"Anatoly I. Kalinovsky",
"Roman S. Popov",
"Valentin A. Stonik",
"Natalia V. Ivanchina",
"Pavel S. Dmitrenok"
] |
https://doi.org/10.3390/md22070294
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39057403_p3
|
39057403
|
sec[0]/p[3]
|
1. Introduction
| 4.117188 |
biomedical
|
Study
|
[
0.99951171875,
0.0002703666687011719,
0.00024390220642089844
] |
[
0.99951171875,
0.00018990039825439453,
0.00025463104248046875,
0.00006562471389770508
] |
In continuation of the study on the chemical constituents of the sea stars , herein we report the results of our investigation of polar steroid metabolites from an ethanolic extract of the Far Eastern starfish Henricia leviuscula spiculifera H.L. Clark, 1901 (order Spinulosida, family Echinasteridae), collected near Urup Island (Kuril Islands) in the Sea of Okhotsk. We have isolated and structurally elucidated four new polyhydroxysteroid glycosides 1 – 4 . The anti-cancer activity of 1 – 3 against several types of human cancer cells has been investigated. A bioassay of compound 4 was not carried out since it was isolated in insufficient amounts. In addition, the influence of 3 , the most active of the tested compounds, on the cell cycle, regulation of expression of cell cycle proteins, and inhibition of phosphorylation of protein kinases has been studied.
|
[
"Alla A. Kicha",
"Dmitriy K. Tolkanov",
"Timofey V. Malyarenko",
"Olesya S. Malyarenko",
"Alexandra S. Kuzmich",
"Anatoly I. Kalinovsky",
"Roman S. Popov",
"Valentin A. Stonik",
"Natalia V. Ivanchina",
"Pavel S. Dmitrenok"
] |
https://doi.org/10.3390/md22070294
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39057403_p4
|
39057403
|
sec[1]/sec[0]/p[0]
|
2.1. Structure Determination of Compounds 1 – 4
| 4.007813 |
biomedical
|
Study
|
[
0.99951171875,
0.00021517276763916016,
0.00041961669921875
] |
[
0.9990234375,
0.00087738037109375,
0.00023114681243896484,
0.00010377168655395508
] |
Three new monosulfated polyhydroxysteroid glycosides and one new unsulfated related monoside were isolated from an ethanolic extract of the sea star Henricia spiculifera by means of chromatographic techniques (column chromatography on Polychrom 1, Si gel, and Florisil followed by reverse-phase high-pressure liquid chromatography on Diasfer-110-C18 and YMC-Pack Pro C18 columns). These substances were designated as spiculiferosides A ( 1 ), B ( 2 ), C ( 3 ), and D ( 4 ) .
|
[
"Alla A. Kicha",
"Dmitriy K. Tolkanov",
"Timofey V. Malyarenko",
"Olesya S. Malyarenko",
"Alexandra S. Kuzmich",
"Anatoly I. Kalinovsky",
"Roman S. Popov",
"Valentin A. Stonik",
"Natalia V. Ivanchina",
"Pavel S. Dmitrenok"
] |
https://doi.org/10.3390/md22070294
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39057403_p5
|
39057403
|
sec[1]/sec[0]/p[1]
|
2.1. Structure Determination of Compounds 1 – 4
| 4.648438 |
biomedical
|
Study
|
[
0.9990234375,
0.0006928443908691406,
0.0003178119659423828
] |
[
0.998046875,
0.001079559326171875,
0.0006694793701171875,
0.0003757476806640625
] |
Spiculiferoside A ( 1 ) has the molecular formula C 45 H 77 O 22 SNa determined from the peak of [M − Na] − ion at m / z 1001.4622 in the (–)HRESIMS and from the peak of the cationized molecule [M + Na] + at m / z 1047.4419 in the (+)HRESIMS. The fragment ion peak at m / z 97 [HSO 4 ] − in the (−)ESIMS/MS spectrum of the ion with m / z 1001 [M − Na] − and the fragment ion peaks at m / z 927 [(M + Na) − NaHSO 4 ] + and 143 [Na 2 HSO 4 ] + in the (+)ESIMS/MS spectrum of the ion with m / z 1047 [M + Na] + showed the presence of a sulfate group in 1 . The IR spectrum of 1 revealed absorption bands due to hydroxy and sulfate groups. The 13 C-NMR and DEPT spectra of 1 exhibited the presence of 45 carbon atoms in the molecule, including 5 methyls, 11 methylenes, 24 methines, two quaternary carbon atoms, one oxygenated tertiary carbon, and two methoxyl groups . The 1 H- and 13 C-NMR spectra of 1 contained signals of protons and carbon atoms of two angular methyl groups (δ H 0.95, 1.43, both s; δ C 15.3, 18.7 ppm, H 3 C-18, H 3 C-19, respectively), five oxygenated methine groups (δ H 3.64, m; δ C 80.6, HC-3), (δ H 4.25, m; δ C 74.7, HC-4), (δ H 4.25, m; δ C 76.2, HC-6), (δ H 4.27, td, J = 9.5, 3.1 Hz; δ C 70.1, HC-15), (δ H 3.31, m; δ C 84.0, HC-24), and one oxygenated tertiary carbon atom (δ C 76.8, C-8), characteristic of 3β,4β,6β,8,15α,24-hexahydroxysteroid aglycon, glycosylated at the positions C-3 and C-24, which was previously found in forbeside J from the starfish Asterias forbesi .
|
[
"Alla A. Kicha",
"Dmitriy K. Tolkanov",
"Timofey V. Malyarenko",
"Olesya S. Malyarenko",
"Alexandra S. Kuzmich",
"Anatoly I. Kalinovsky",
"Roman S. Popov",
"Valentin A. Stonik",
"Natalia V. Ivanchina",
"Pavel S. Dmitrenok"
] |
https://doi.org/10.3390/md22070294
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
39057403_p6
|
39057403
|
sec[1]/sec[0]/p[2]
|
2.1. Structure Determination of Compounds 1 – 4
| 4.511719 |
biomedical
|
Study
|
[
0.9990234375,
0.0006117820739746094,
0.00025343894958496094
] |
[
0.99853515625,
0.0005755424499511719,
0.00066375732421875,
0.0002446174621582031
] |
Analysis of the 1 H- 1 H COSY and HSQC correlations made it possible to establish the spin systems of protons and the corresponding sequences of carbon atoms from C-1 to C-7, from C-9 to C-12 through C-11, from C-14 to C-17, from C-20 to C-21, and from C-22 to C-27 . The relative configurations 3β, 4β, 6β, and 15α of hydroxyl substituents in the steroid core and the 5α-cholestane skeleton in 1 were determined based on proton correlations in the ROESY spectrum from H-3 to Hα-1 and H-5; from H-5 to Hα-7 and H-9; from H-14 to H-9 and H-17; from H 3 -18 to Hβ-11, Hβ-12, and H-15; and from H 3 -19 to Hβ-1, and Hβ-2 . The main cross peaks of protons and carbon atoms in the HMBC spectrum confirmed the general structure of the steroid aglycon in 1 . The resonance value of the methyl group H 3 -21 at δ H 0.90, as well as the presence in the ROESY spectrum of correlations of protons from Hβ-12 to H 3 -21, from H-17 to H 3 -21, and from H 3 -18 to H-20, indicated a 20 R configuration of the asymmetric center .
|
[
"Alla A. Kicha",
"Dmitriy K. Tolkanov",
"Timofey V. Malyarenko",
"Olesya S. Malyarenko",
"Alexandra S. Kuzmich",
"Anatoly I. Kalinovsky",
"Roman S. Popov",
"Valentin A. Stonik",
"Natalia V. Ivanchina",
"Pavel S. Dmitrenok"
] |
https://doi.org/10.3390/md22070294
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
39057403_p7
|
39057403
|
sec[1]/sec[0]/p[3]
|
2.1. Structure Determination of Compounds 1 – 4
| 4.625 |
biomedical
|
Study
|
[
0.9990234375,
0.0007867813110351562,
0.0002646446228027344
] |
[
0.998046875,
0.0008268356323242188,
0.0008001327514648438,
0.0004215240478515625
] |
In the 1 H-NMR spectrum of 1 , three chemical shifts of anomeric protons were observed at δ H 4.44, 5.00, and 4.34, associated with signals of carbon atoms at δ C 102.6, 109.2, and 104.9 in the HSQC spectrum, respectively. These data indicated the existence of three monosaccharide residues in glycoside 1 . The coupling constants 7.5 and 7.7 Hz of two anomeric protons exhibited the β-glycosidic bonds of the corresponding monosaccharide residues, and a wide singlet of the third anomeric proton showed the presence of an α-glycosidic bond in this monosaccharide residue. The ESIMS/MS spectrum of the [M − Na] − ion with m / z 1001 revealed fragment ion peaks corresponding to the loss in a hexose at m / z 839 [(M − Na) – C 6 H 10 O 5 ] − and the simultaneous loss in a hexose and a di- O -methyl-pentose at m / z 679 [(M − Na) − C 6 H 10 O 5 − C 7 H 12 O 4 ] − . Respectively, the ESIMS/MS spectrum of the [M + Na] + ion with m / z 1047 exhibited fragment ion peaks arising due to the loss in a hexose at m / z 885 [(M + Na) − C 6 H 10 O 5 ] + , the simultaneous loss in a hexose and a di- O -methyl-pentose at m / z 725 [(M + Na) − C 6 H 10 O 5 − C 7 H 12 O 4 ] + , the simultaneous loss in a hexose and a sulfoxypentose at m / z 651 [(M + Na) − C 6 H 10 O 5 − C 5 H 7 O 7 SNa] + , and the simultaneous loss in a hexose, a sulfoxypentose, and a di- O -methyl-pentose at m / z 491 [651 − C 7 H 12 O 4 ] + . Therefore, according to the ESIMS/MS and NMR spectra, molecule 1 contains hexose, di- O -methyl-pentose, and sulfoxypentose units. Acid hydrolysis of glycoside 1 with 2 M CF 3 COOH yielded three monosaccharides, which, after obtaining 2-octylglycoside derivatives by treatment with ( R )-(–)-2-octanol and subsequent acetylation according to the procedure of Leontein et al. , were identified by GC as 2,4-di- O -methyl-D-xylose, L-arabinose, and D-glucose.
|
[
"Alla A. Kicha",
"Dmitriy K. Tolkanov",
"Timofey V. Malyarenko",
"Olesya S. Malyarenko",
"Alexandra S. Kuzmich",
"Anatoly I. Kalinovsky",
"Roman S. Popov",
"Valentin A. Stonik",
"Natalia V. Ivanchina",
"Pavel S. Dmitrenok"
] |
https://doi.org/10.3390/md22070294
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057403_p8
|
39057403
|
sec[1]/sec[0]/p[4]
|
2.1. Structure Determination of Compounds 1 – 4
| 4.71875 |
biomedical
|
Study
|
[
0.99853515625,
0.0012941360473632812,
0.00033092498779296875
] |
[
0.9970703125,
0.0009164810180664062,
0.0013227462768554688,
0.0006999969482421875
] |
The sequences of protons and the carbon atoms associated with corresponding protons, as well as the relative proton configurations of monosaccharide residues, were assigned using 1 H- 1 H COSY, HSQC, HMBC, and ROESY experiments . Irradiation of anomeric protons in 1D TOCSY experiments allowed us to refine the chemical shifts and coupling constants of the carbohydrate moiety protons. The spectral data of the two monosaccharide units were in good agreement with those for the terminal residues of 2,4-di- O -methyl-β- d -xylopyranose and β- d -glucopyranose . The position of the terminal 2,4-di- O -methyl-β- d -xylopyranose unit at C-3 of aglycon was confirmed by the cross-peaks between H-1′ and H-3, C-3, the linkage of the terminal residue of β- d -glucopyranose to C-5″ of the internal residue of α-L-arabinofuranose was indicated by the cross-peaks between H-1‴ and H 2 -5″, C-5″, and the linkage of the α-L-arabinofuranose residue to C-24 of aglycon was fixed by the cross-peaks between H-1″ and H-24, C-24 in the ROESY and HMBC spectra, respectively. A comparison of the proton and carbon signals of the internal monosaccharide residue of glycoside 1 with those of the five-substituted α- l -arabinofuranose residue of kurilensoside B from the starfish Hippasteria kurilensis showed that the chemical shifts of H-3″ and C-3″ were deshielded from δ H 3.94 to 4.67 and from δ C 79.2 to 84.7, respectively, and the signal of C-2″ was shielded from δ C 83.8 to 82.0. These facts clearly revealed the location of the sulfate group at C-3″of 5″-substituted residue of α- l -arabinofuranose in 1 . In addition, the signal of C-1″ at δ C 109.2 unambiguously indicated the α-configuration of the anomeric center of the arabinofuranose residue . The 24 S configuration was proposed based on the similarity of the 13 C-NMR spectroscopic data for the side chain of glycoside 1 with those for other related (24 S )-24- O -α- l -arabinofuranosides previously isolated from starfish . Consequently, the structure of spiculiferoside A ( 1 ) was elucidated as the (24 S )-3- O -(2,4-di- O -methyl-β- d -xylopyranosyl)-24- O -[β-D-glucopyranosyl-(1→5)-3- O -sulfate-α- l -arabinofuranosyl]-5α-cholestane-3β,4β,6β,8,15α,24-hexaol, sodium salt. Glycoside 1 is a triglycoside and contains two carbohydrate moieties, one of which is attached to C-3 of the steroid core, and the other is located at C-24 of the aglycon side chain. Only five such “two-chain” triglycosides from see stars were previously known . In addition, the five-substituted 3-OSO 3 -α- l -Araf residue was found for the first time in steroid glycosides from starfish.
|
[
"Alla A. Kicha",
"Dmitriy K. Tolkanov",
"Timofey V. Malyarenko",
"Olesya S. Malyarenko",
"Alexandra S. Kuzmich",
"Anatoly I. Kalinovsky",
"Roman S. Popov",
"Valentin A. Stonik",
"Natalia V. Ivanchina",
"Pavel S. Dmitrenok"
] |
https://doi.org/10.3390/md22070294
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999995 |
39057403_p9
|
39057403
|
sec[1]/sec[0]/p[5]
|
2.1. Structure Determination of Compounds 1 – 4
| 4.398438 |
biomedical
|
Study
|
[
0.99951171875,
0.0003769397735595703,
0.00027632713317871094
] |
[
0.998046875,
0.0012044906616210938,
0.0003361701965332031,
0.0001919269561767578
] |
Spiculiferoside B ( 2 ) has the molecular formula C 39 H 67 O 17 SNa, determined from the peak of [M − Na] − ion at m / z 839.4108 in the (–)HRESIMS and from the peak of cationized molecule [M + Na] + at m / z 885.3896 in the (+)HRESIMS. The fragment ion peak at m / z 97 [HSO 4 ] − in the (−)ESIMS/MS spectrum of the ion with m / z 839 [M − Na] − and the fragment ion peaks at m / z 765 [(M + Na) − NaHSO 4 ] + and 143 [Na 2 HSO 4 ] + in the (+)ESIMS/MS spectrum of the ion with m / z 885 [M + Na] + showed the presence of a sulfate group in 2 . The IR spectrum of 2 revealed absorption bands due to hydroxy and sulfate groups.
|
[
"Alla A. Kicha",
"Dmitriy K. Tolkanov",
"Timofey V. Malyarenko",
"Olesya S. Malyarenko",
"Alexandra S. Kuzmich",
"Anatoly I. Kalinovsky",
"Roman S. Popov",
"Valentin A. Stonik",
"Natalia V. Ivanchina",
"Pavel S. Dmitrenok"
] |
https://doi.org/10.3390/md22070294
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057403_p10
|
39057403
|
sec[1]/sec[0]/p[6]
|
2.1. Structure Determination of Compounds 1 – 4
| 4.644531 |
biomedical
|
Study
|
[
0.9990234375,
0.0008473396301269531,
0.00025534629821777344
] |
[
0.998046875,
0.0008783340454101562,
0.0008463859558105469,
0.0004565715789794922
] |
The 1 H-NMR spectrum of 2 included two resonances in the deshielded region due to anomeric protons at δ H 4.44 and 4.98, which correlated in the HSQC spectrum with corresponding carbon resonances at δ C 102.7 and 109.5, respectively ( Table 2 ). The ESIMS/MS spectrum of the [M − Na] − ion with m / z 839 indicated fragment ion peaks corresponding to the loss in a di- O -methyl-pentose at m / z 679 [(M − Na) − C 7 H 12 O 4 ] − and a sulfoxypentose at m / z 211 [C 5 H 7 O 7 S] − . Accordingly, the ESIMS/MS spectrum of the [M + Na] + ion with m / z 885 revealed fragment ion peaks arising due to the loss in a di- O -methyl-pentose at m / z 725 [(M + Na) − C 7 H 12 O 4 ] + , the loss in a sulfoxypentose at m / z 651 [(M + Na) − C 5 H 7 O 7 SNa] + , and the simultaneous loss in a di- O -methyl-pentose and a sulfoxypentose at m / z 491 [651 − C 7 H 12 O 4 ] + . A detailed comparison of the 1 H-, 13 C-NMR, DEPT, COSY, HSQC, HMBC, and ROESY spectroscopic data of glycoside 2 with the corresponding data of glycoside 1 showed that 2 had the same 3β,4β,6β,8,15α,24-hexahydroxy-5α-cholestane aglycon, glycosylated at C-3 with a 2,4-di- O -methyl-β- d -xylopyranose residue, and at C-24 with a sulfated α- l -arabinofuranose residue, and differed from 1 only in the absence of a terminal β- d -glucopyranose residue. A comparison of the signals of the protons and carbon atoms of the monosaccharide unit at C-24 in glycoside 2 with the corresponding signals of the terminal α-L-arabinofuranose residue of forbeside J showed that the resonance of C-3″ was deshielded from δ C 78.7 to 84.5; the resonances of C-2″ and C-4″ were shielded from δ C 84.0 to 82.1 and from δ C 85.0 to 84.4, respectively, and the signal of H-3″ was deshielded from δ H 3.86 to 4.46 in accordance with α- and β-effects of sulfation. In this way, the position of a sulfate group in the α-L-arabinofuranose in 2 was defined as C-3″. Compound 2 was subjected to mild solvolysis with a mixture of dioxane and pyridine to give desulfated derivative 2a , which was identified by comparison of the HRESIMS and 1 H-, 13 C-NMR, and HSQC data with those of forbeside J . As a result, the absolute configuration at C-24 in 2 was proposed as S by analogy with forbeside J ( 2a ). On the basis of the above-mentioned data, the structure of spiculiferoside B ( 2 ) was defined as the (24 S )-3- O -(2,4-di- O -methyl-β- d -xylopyranosyl)-24- O -(3- O -sulfate-α- l -arabinofuranosyl)-5α-cholestane-3β,4β,6β,8,15α,24-hexaol sodium salt.
|
[
"Alla A. Kicha",
"Dmitriy K. Tolkanov",
"Timofey V. Malyarenko",
"Olesya S. Malyarenko",
"Alexandra S. Kuzmich",
"Anatoly I. Kalinovsky",
"Roman S. Popov",
"Valentin A. Stonik",
"Natalia V. Ivanchina",
"Pavel S. Dmitrenok"
] |
https://doi.org/10.3390/md22070294
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
39057403_p11
|
39057403
|
sec[1]/sec[0]/p[7]
|
2.1. Structure Determination of Compounds 1 – 4
| 4.382813 |
biomedical
|
Study
|
[
0.99951171875,
0.000354766845703125,
0.0002741813659667969
] |
[
0.99853515625,
0.0010900497436523438,
0.00031828880310058594,
0.0001785755157470703
] |
Spiculiferoside C ( 3 ) has the molecular formula C 39 H 67 O 16 SNa determined from the peak of [M − Na] − ion at m / z 823.4162 in the (–)HRESIMS and from the peak of cationized molecule [M + Na] + at m / z 869.3936 in the (+)HRESIMS. The fragment ion peak at m / z 97 [HSO 4 ] − in the (−)ESIMS/MS spectrum of the ion with m / z 823 [M − Na] − and the fragment ion peaks at m / z 749 [(M + Na) − NaHSO 4 ] + and 143 [Na 2 HSO 4 ] + in the (+)ESIMS/MS spectrum of the ion with m / z 869 [M + Na] + showed the presence of a sulfate group in 3 . The IR spectrum of 3 revealed absorption bands due to hydroxy and sulfate groups.
|
[
"Alla A. Kicha",
"Dmitriy K. Tolkanov",
"Timofey V. Malyarenko",
"Olesya S. Malyarenko",
"Alexandra S. Kuzmich",
"Anatoly I. Kalinovsky",
"Roman S. Popov",
"Valentin A. Stonik",
"Natalia V. Ivanchina",
"Pavel S. Dmitrenok"
] |
https://doi.org/10.3390/md22070294
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39057403_p12
|
39057403
|
sec[1]/sec[0]/p[8]
|
2.1. Structure Determination of Compounds 1 – 4
| 4.363281 |
biomedical
|
Study
|
[
0.99951171875,
0.0004398822784423828,
0.00016796588897705078
] |
[
0.9990234375,
0.0004892349243164062,
0.0004429817199707031,
0.00019240379333496094
] |
The ESIMS/MS spectrum of the [M − Na] − ion with m / z 823 indicated fragment ion peaks corresponding to the loss in a di- O -methyl-pentose at m / z 663 [(M − Na) − C 7 H 12 O 4 ] − and a sulfoxypentose at m / z 211 [C 5 H 7 O 7 S] − . Accordingly, the ESIMS/MS spectrum of the [M + Na] + ion with m / z 869 revealed fragment ion peaks arising due to the loss in a di- O -methyl-pentose at m / z 709 [(M + Na) − C 7 H 12 O 4 ] + and the loss in a sulfoxypentose at m / z 635 [(M + Na) − C 5 H 7 O 7 SNa] + . Examination of the HRESIMS, ESIMS/MS, 1D, and 2D NMR spectra of glycoside 3 and the corresponding data of glycoside 2 clearly showed the presence of identical monosaccharide residues and steroid side chain in both compounds: 2,4-di- O -methyl-β- d -xylopyranose unit attached to C-3 of steroid nucleus and 3-sulfoxy-α- l -arabinofuranose unit attached to C-24 of the steroid side chain ( Table 1 and Table 2 ).
|
[
"Alla A. Kicha",
"Dmitriy K. Tolkanov",
"Timofey V. Malyarenko",
"Olesya S. Malyarenko",
"Alexandra S. Kuzmich",
"Anatoly I. Kalinovsky",
"Roman S. Popov",
"Valentin A. Stonik",
"Natalia V. Ivanchina",
"Pavel S. Dmitrenok"
] |
https://doi.org/10.3390/md22070294
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
39057403_p13
|
39057403
|
sec[1]/sec[0]/p[9]
|
2.1. Structure Determination of Compounds 1 – 4
| 4.558594 |
biomedical
|
Study
|
[
0.9990234375,
0.0005397796630859375,
0.0002009868621826172
] |
[
0.998046875,
0.0007257461547851562,
0.0008144378662109375,
0.00033092498779296875
] |
Most of the resonances in the 1 H- and 13 C-NMR spectra of 3 , related to the steroid moiety, were close to the corresponding values for 2 . However, in the 13 C-NMR spectrum of 3, no signal was observed for carbon atom C-8 at δ C 76.8 because of the absence of a hydroxyl group at this position. 1 H- 1 H COSY, HSQC, and HMBC cross-peaks supported the presence of spin proton sequences in the steroid core of 3 at C-1 to C-9, at C-9 to C-12 through C-11, at C-8 to C-14, and at C-14 to C-17 . The ROESY correlations from H-4 to H-6, from H-5 to H-3 and H-9, from Hα-7 to H-14, from H-8 to H 3 -18 and H 3 -19, from H 3 -18 to Hβ-12 and H-15, and from H 3 -19 to Hβ-1 and Hβ-2 confirmed that the 5α/8β/9α/10β/13β/14α steroid nucleus in 3 had a 3β,4β,6β,15α-tetrahydroxy substitution . Thereby, the structure of spiculiferoside C ( 3 ) was established as the (24 S )-3- O -(2,4-di- O -methyl-β- d -xylopyranosyl)-24- O -(3- O -sulfate-α- l -arabinofuranosyl)-5α-cholestane-3β,4β,6β,15α,24-pentaol sodium salt. The 3-OSO 3 -α- l -Araf residue was included in spiculiferosides B ( 2 ) and C ( 3 ) and was previously found in only one steroid glycoside from starfish Oreaster reticulatus .
|
[
"Alla A. Kicha",
"Dmitriy K. Tolkanov",
"Timofey V. Malyarenko",
"Olesya S. Malyarenko",
"Alexandra S. Kuzmich",
"Anatoly I. Kalinovsky",
"Roman S. Popov",
"Valentin A. Stonik",
"Natalia V. Ivanchina",
"Pavel S. Dmitrenok"
] |
https://doi.org/10.3390/md22070294
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057403_p14
|
39057403
|
sec[1]/sec[0]/p[10]
|
2.1. Structure Determination of Compounds 1 – 4
| 4.621094 |
biomedical
|
Study
|
[
0.9990234375,
0.0007882118225097656,
0.00026345252990722656
] |
[
0.998046875,
0.0008993148803710938,
0.0007648468017578125,
0.00047588348388671875
] |
The molecular formula C 32 H 56 O 10 of spiculiferoside D ( 4 ) was elucidated from the peaks of [M − H] − ion at m / z 599.3794, [M + Cl] − ion at m / z 635.3561, and [M + CHO 2 ] − ion at m / z 645.3847 in the (–)HRESIMS and from the peak of the cationized molecule [M + Na] + at m / z 623.3770 in the (+)HRESIMS. The fragment ion peaks at m / z 467 [(M − H) − C 5 H 8 O 4 ] − , 449 [(M − H) − C 5 H 10 O 5 ] − , and 131 [C 5 H 7 O 4 ] − in the (−)ESIMS/MS spectrum of the ion with m / z 599 [M − H] − and the fragment ion peaks at m / z 491 [(M + Na) − C 5 H 8 O 4 ] + and 473 [(M + Na) − C 5 H 10 O 5 ] + in the (+)ESIMS/MS spectrum of the ion with m / z 623 [M + Na] + indicated the presence of a pentose unit in 4 . A thorough comparison of NMR spectra of compound 4 and desulfated derivative 2a (Experimental section) exhibited that both compounds contained the same 3β,4β,6β,8,15α,24-hexahydroxy-5α-cholestane aglycon and α-L-arabinofuranose residue at C-24 of the side chain and differed from each other only in the absence of a 2,4-di- O -methyl-β- d -xylopyranose residue at C-3 of the steroid moiety of 4 . In accordance with this, the chemical shifts of H-3 and C-3 of 4 compared to 2a were shielded from δ H 3.64 to 3.50 and from δ C 80.5 to 73.1, respectively, while the signals of C-2 and C-4 were deshielded from δ C 25.2 to 26.6 and from δ C 74.6 to 77.5, respectively, according to α- and β-effects of deglycosylation . The structure of glycoside 4 was confirmed by DEPT, 1 H- 1 H COSY, HSQC, HMBC, and ROESY experiments . Thus, it was established that spiculiferoside D ( 4 ) was the (24 S )-24- O -(α-L-arabinofuranosyl)-5α-cholestane-3β,4β,6β,8,5α,24-hexaol.
|
[
"Alla A. Kicha",
"Dmitriy K. Tolkanov",
"Timofey V. Malyarenko",
"Olesya S. Malyarenko",
"Alexandra S. Kuzmich",
"Anatoly I. Kalinovsky",
"Roman S. Popov",
"Valentin A. Stonik",
"Natalia V. Ivanchina",
"Pavel S. Dmitrenok"
] |
https://doi.org/10.3390/md22070294
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999999 |
39057403_p15
|
39057403
|
sec[1]/sec[1]/p[0]
|
2.2. The Effect of Compounds 1 – 3 on Cell Viability and Proliferation of Human Normal and Cancer Cells
| 4.0625 |
biomedical
|
Study
|
[
0.99951171875,
0.00022339820861816406,
0.0002186298370361328
] |
[
0.99951171875,
0.0002359151840209961,
0.0002663135528564453,
0.000059545040130615234
] |
In the present work, the effect of compounds 1 , 2 , and 3 on cell viability of human embryonic kidney HEK293, melanoma SK-MEL-28, breast cancer cells MDA-MB-231, and colorectal carcinoma HCT 116 cells was determined by MTS assay in a 24 h cell treatment .
|
[
"Alla A. Kicha",
"Dmitriy K. Tolkanov",
"Timofey V. Malyarenko",
"Olesya S. Malyarenko",
"Alexandra S. Kuzmich",
"Anatoly I. Kalinovsky",
"Roman S. Popov",
"Valentin A. Stonik",
"Natalia V. Ivanchina",
"Pavel S. Dmitrenok"
] |
https://doi.org/10.3390/md22070294
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
39057403_p16
|
39057403
|
sec[1]/sec[1]/p[1]
|
2.2. The Effect of Compounds 1 – 3 on Cell Viability and Proliferation of Human Normal and Cancer Cells
| 4.140625 |
biomedical
|
Study
|
[
0.99951171875,
0.00028514862060546875,
0.0002143383026123047
] |
[
0.99951171875,
0.00020611286163330078,
0.0004062652587890625,
0.00006812810897827148
] |
The tested compounds were less cytotoxic against normal cells, HEK293, and two types of cancer cells, SK-MEL-28 and MDA-MB-231 . On the other hand, these compounds were found to suppress cell viability of colorectal carcinoma cells HCT 116 more effectively, with great impact of compound 3 . Compound 1 at concentrations of 1, 10, 50, and 100 µM inhibited cell viability of HCT 116 cells by 0%, 2%, 6%, and 35%, respectively; compound 2 inhibited cell viability at 1, 10, 50, and 100 µM—0%, 0%, 6%, and 30%, respectively, while 3 at the same experimental conditions suppressed the cell viability by 0%, 0%, 37%, and 54%, respectively . IC 50 was reached only for compound 3 , which was 87.6 µM with a selective index (SI) of 1.5 after 24 h of HCT 116 cells’ treatment .
|
[
"Alla A. Kicha",
"Dmitriy K. Tolkanov",
"Timofey V. Malyarenko",
"Olesya S. Malyarenko",
"Alexandra S. Kuzmich",
"Anatoly I. Kalinovsky",
"Roman S. Popov",
"Valentin A. Stonik",
"Natalia V. Ivanchina",
"Pavel S. Dmitrenok"
] |
https://doi.org/10.3390/md22070294
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
39057403_p17
|
39057403
|
sec[1]/sec[1]/p[2]
|
2.2. The Effect of Compounds 1 – 3 on Cell Viability and Proliferation of Human Normal and Cancer Cells
| 4.101563 |
biomedical
|
Study
|
[
0.99951171875,
0.00028014183044433594,
0.00023317337036132812
] |
[
0.99951171875,
0.0001844167709350586,
0.0003211498260498047,
0.000059545040130615234
] |
Since compounds 1 , 2 , and 3 possessed more significant cytotoxic activity against HCT 116 cells, we determined their effect on the proliferation of HCT 116 cells. All the tested compounds slightly inhibited cell proliferation at concentrations ranging from 1 to 50 µM within 72 h of treatment . Compound 1 at 100 µM decreased cell growth by 35%, 31%, and 40% after 24, 48, and 72 h of cells’ treatment, respectively . Compound 2 (100 µM) was shown to inhibit cell proliferation by 30%, 25%, and 27% after 24, 48, and 72 h of cell incubation, respectively . Compound 3 (100 µM) possessed the highest anti-proliferative activity and suppressed proliferation of HCT 116 cells by 55%, 57%, and 60% after 24, 48, and 72 h of treatment, respectively .
|
[
"Alla A. Kicha",
"Dmitriy K. Tolkanov",
"Timofey V. Malyarenko",
"Olesya S. Malyarenko",
"Alexandra S. Kuzmich",
"Anatoly I. Kalinovsky",
"Roman S. Popov",
"Valentin A. Stonik",
"Natalia V. Ivanchina",
"Pavel S. Dmitrenok"
] |
https://doi.org/10.3390/md22070294
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057403_p18
|
39057403
|
sec[1]/sec[2]/p[0]
|
2.3. The Effect of Compounds 1 – 3 on the Colony Formation of Human Colorectal Carcinoma Cells
| 4.082031 |
biomedical
|
Study
|
[
0.99951171875,
0.00031375885009765625,
0.00019562244415283203
] |
[
0.99951171875,
0.00014674663543701172,
0.0003590583801269531,
0.00006848573684692383
] |
More promising data were obtained in the results of the studies on the effects of 1 – 3 on microcolony formation by tumor cells. In the present study, the colony-inhibiting activity was investigated in HCT 116 cells using the soft agar assay. Non-toxic concentrations of 10, 20, and 40 µM of the investigated compounds were chosen for further experiments. All the tested compounds were found to significantly decrease colonies’ numbers of colorectal carcinoma cells dose-dependently . Compound 1 at concentrations of 10, 20, and 40 µM inhibited colony formation in HCT 116 cells by 18%, 39%, and 65%, respectively ; 2 —by 25%, 48%, and 81%, respectively , and 3 —by 19%, 56%, 87%, respectively . Compound 3 was found to have the most significant colony-inhibiting activity against HCT 116 cells among all the compounds studied and was, therefore, selected for further investigation of the molecular mechanism of its anti-cancer action.
|
[
"Alla A. Kicha",
"Dmitriy K. Tolkanov",
"Timofey V. Malyarenko",
"Olesya S. Malyarenko",
"Alexandra S. Kuzmich",
"Anatoly I. Kalinovsky",
"Roman S. Popov",
"Valentin A. Stonik",
"Natalia V. Ivanchina",
"Pavel S. Dmitrenok"
] |
https://doi.org/10.3390/md22070294
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057403_p19
|
39057403
|
sec[1]/sec[3]/p[0]
|
2.4. The Effect of Compound 3 on Cell Cycle Progression and Molecular Mechanism of Anti-cancer Action in Human Colorectal Carcinoma Cells
| 4.078125 |
biomedical
|
Study
|
[
0.99951171875,
0.00019598007202148438,
0.00018668174743652344
] |
[
0.99951171875,
0.0002758502960205078,
0.0002536773681640625,
0.00005990266799926758
] |
The fundamental abnormality that leads to the development of cancer is the continuous, unregulated proliferation of cancer cells. Instead of responding appropriately to signals that control normal cell behavior, cancer cells grow and divide uncontrollably, invading normal tissues and organs and eventually spreading throughout the body . Since compound 3 inhibited proliferation and colony formation of colorectal cancer cells HCT 116, we checked whether compound 3 could regulate cell cycle distribution by flow cytometric analysis. Cell cycle progression was examined after treatment of HCT 116 cells with 10, 20, and 40 μM of 3 for 72 h.
|
[
"Alla A. Kicha",
"Dmitriy K. Tolkanov",
"Timofey V. Malyarenko",
"Olesya S. Malyarenko",
"Alexandra S. Kuzmich",
"Anatoly I. Kalinovsky",
"Roman S. Popov",
"Valentin A. Stonik",
"Natalia V. Ivanchina",
"Pavel S. Dmitrenok"
] |
https://doi.org/10.3390/md22070294
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39057403_p20
|
39057403
|
sec[1]/sec[3]/p[1]
|
2.4. The Effect of Compound 3 on Cell Cycle Progression and Molecular Mechanism of Anti-cancer Action in Human Colorectal Carcinoma Cells
| 4.136719 |
biomedical
|
Study
|
[
0.99951171875,
0.00027680397033691406,
0.00017130374908447266
] |
[
0.99951171875,
0.0002453327178955078,
0.0003485679626464844,
0.00006836652755737305
] |
It was found that the treatment of HCT 116 cells with 3 resulted in a dose-dependent increase in cells in the G2/M phase compared to the control group. Compound 3 at 10, 20, and 40 µM was shown to increase the amount of HCT 116 cells in G2/M phase by 16%, 42%, and 71%, respectively, with a corresponding reduction in the percentage of cells in the G0/G1 phase by 0%, 15%, and 25%, respectively, and S phase by 9%, 11%, and 20%, respectively, compared to the control group . These data suggest that the inhibition of cell proliferation of HCT 116 cells is mainly associated with the induction of G2/M cell cycle arrest.
|
[
"Alla A. Kicha",
"Dmitriy K. Tolkanov",
"Timofey V. Malyarenko",
"Olesya S. Malyarenko",
"Alexandra S. Kuzmich",
"Anatoly I. Kalinovsky",
"Roman S. Popov",
"Valentin A. Stonik",
"Natalia V. Ivanchina",
"Pavel S. Dmitrenok"
] |
https://doi.org/10.3390/md22070294
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057403_p21
|
39057403
|
sec[1]/sec[3]/p[2]
|
2.4. The Effect of Compound 3 on Cell Cycle Progression and Molecular Mechanism of Anti-cancer Action in Human Colorectal Carcinoma Cells
| 4.46875 |
biomedical
|
Study
|
[
0.9990234375,
0.0005626678466796875,
0.00019288063049316406
] |
[
0.99853515625,
0.0003809928894042969,
0.0008463859558105469,
0.00020396709442138672
] |
Next, we turned our attention to the molecular mechanism of anti-cancer action of compound 3 associated with the inhibition of cell proliferation of HCT 116 cells via the regulation of a series of important cell cycle proteins and the activation of mitogen-activated protein kinases (MAPK) by Western Blot assay. Extracellular-signal-related kinase p44/42 MAPK (Erk1/2) is known to be an important participant in the MAPK signaling pathway . ERK1/2 plays a well-established role in regulating cell cycle progression by activation of multiple transcription factors such as Elk1, c-Jun, c-Myc, and c-Fos, which control the expression of proteins important for cell-cycle progression, including Cyclin D1 and p21WAF1/CIP1 . Cyclin-dependent kinases (CDK) are major players in cell proliferation that regulate cell cycle checkpoints and transcription events in response to extracellular and intracellular signals. CDK dysregulation is certain to be a hallmark of cancer and an attractive target in cancer therapy. CDK activity is primarily regulated by the binding of CDK catalytic subunits to Cyclin partners and CDK inhibitors. The complex formed by CDK4 and Cyclin D1 has been strongly implicated in the control of cell proliferation and prognoses in human malignancies . In this regard, we examined the influence of 3 on the expression of CDK2, CDK4, Cyclin D1, and p21. The investigated compound was found to dose-dependently down-regulate the expression of CDK2 and Cyclin D1 but not CDK4. The expression of the inhibitor of CDK/Cyclin complex—p21 was significantly increased by 3 compared to non-treated HCT 116 cells . The treatment of HCT 116 cells by 3 was demonstrated to cause the inhibition of phosphorylation of c-Raf, MEK1/2, and ERK1/2 kinases .
|
[
"Alla A. Kicha",
"Dmitriy K. Tolkanov",
"Timofey V. Malyarenko",
"Olesya S. Malyarenko",
"Alexandra S. Kuzmich",
"Anatoly I. Kalinovsky",
"Roman S. Popov",
"Valentin A. Stonik",
"Natalia V. Ivanchina",
"Pavel S. Dmitrenok"
] |
https://doi.org/10.3390/md22070294
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057403_p22
|
39057403
|
sec[1]/sec[3]/p[3]
|
2.4. The Effect of Compound 3 on Cell Cycle Progression and Molecular Mechanism of Anti-cancer Action in Human Colorectal Carcinoma Cells
| 4.105469 |
biomedical
|
Study
|
[
0.99951171875,
0.00022518634796142578,
0.00013899803161621094
] |
[
0.9990234375,
0.0003485679626464844,
0.00040435791015625,
0.00008934736251831055
] |
Our results provided evidence that the coordinated alteration of the expression of cell cycle proteins and inhibition of the phosphorylation of the ERK1/2 MAPK signaling cascade were likely the basis of the anti-cancer effect of compound 3 on the proliferation of colorectal carcinoma cells HCT 116.
|
[
"Alla A. Kicha",
"Dmitriy K. Tolkanov",
"Timofey V. Malyarenko",
"Olesya S. Malyarenko",
"Alexandra S. Kuzmich",
"Anatoly I. Kalinovsky",
"Roman S. Popov",
"Valentin A. Stonik",
"Natalia V. Ivanchina",
"Pavel S. Dmitrenok"
] |
https://doi.org/10.3390/md22070294
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39057403_p23
|
39057403
|
sec[2]/sec[0]/p[0]
|
3.1. General Procedures
| 3.136719 |
biomedical
|
Other
|
[
0.99658203125,
0.00030803680419921875,
0.0032405853271484375
] |
[
0.44921875,
0.54833984375,
0.0016040802001953125,
0.0009341239929199219
] |
Optical rotations, Perkin-Elmer 343 polarimeter (PerkinElmer, Waltham, MA, USA). NMR spectra, Bruker Avance III 500 HD (Bruker, Göttingen, Germany) at 500.13 MHz ( 1 H)/125.76 MHz ( 13 C), Bruker Avance III 700 spectrometer (Bruker, Bremen, Germany) at 700.13 ( 1 H)/176.04 MHz ( 13 C), internal standard CD 3 OD at δ H 3.30/ δ C 49.0. HRESIMS spectra, Bruker Impact II Q-TOF mass spectrometer (Bruker, Bremen, Germany); sample concentration in MeOH 0.001 mg/mL. HPLC, Agilent 1100 Series chromatograph (Agilent Technologies, Santa Clara, CA, USA) with a differential refractometer; columns Discovery C18 (5 µm, 10.0 × 250 mm, Supelco, Bellefonte, PA, USA), YMC-Pack Pro C18 (5 µm, 10.0 × 250 mm, YMC Co., Ltd., Kyoto, Japan), and Diasfer-110-C18 (5 µm, 4.0 × 250 mm, BioChemMack, Moscow, Russia). GC, Agilent 6580 Series chromatograph (Agilent Technologies, Santa Clara, CA, USA), HP-1 MS capillary column (0.32 mm × 30 m) over the temperature range 100−270 °C at 5 °C/min, carrier gas He (1.7 mL/min), injector temperature 250 °C, detector temperature 270 °C. LPLC, column sorbents Polychrom 1 (powdered Teflon, 0.25–0.50 mm, Biolar, Olaine, Latvia), Si gel (63–200 µm, Sigma-Aldrich, Switzerland), and Florisil (60–100 µm, Sigma-Aldrich, Co., St. Louis, MI, USA).
|
[
"Alla A. Kicha",
"Dmitriy K. Tolkanov",
"Timofey V. Malyarenko",
"Olesya S. Malyarenko",
"Alexandra S. Kuzmich",
"Anatoly I. Kalinovsky",
"Roman S. Popov",
"Valentin A. Stonik",
"Natalia V. Ivanchina",
"Pavel S. Dmitrenok"
] |
https://doi.org/10.3390/md22070294
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999994 |
39057403_p24
|
39057403
|
sec[2]/sec[1]/p[0]
|
3.2. Animal Material
| 1.798828 |
biomedical
|
Other
|
[
0.703125,
0.0017642974853515625,
0.294921875
] |
[
0.3515625,
0.64501953125,
0.0013523101806640625,
0.0019407272338867188
] |
Specimens of Henricia leviuscula spiculifera Clark, 1901 (order Spinulosida, family Echinasteridae) were collected near Urup Island (Kuril Islands, Sea of Okhotsk) at a depth of 85–89 m using a small trawl . Taxonomical identification of species was determined by Mr. Boris B. Grebnev (G.B. Elyakov PIBOC FEB RAS, Vladivostok, Russia). A voucher specimen (no. 051-039a) has been deposited in the collection of G.B. Elyakov PIBOC FEB RAS, Vladivostok, Russia.
|
[
"Alla A. Kicha",
"Dmitriy K. Tolkanov",
"Timofey V. Malyarenko",
"Olesya S. Malyarenko",
"Alexandra S. Kuzmich",
"Anatoly I. Kalinovsky",
"Roman S. Popov",
"Valentin A. Stonik",
"Natalia V. Ivanchina",
"Pavel S. Dmitrenok"
] |
https://doi.org/10.3390/md22070294
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
39057403_p25
|
39057403
|
sec[2]/sec[2]/p[0]
|
3.3. Extraction and Isolation
| 4.234375 |
biomedical
|
Study
|
[
0.9990234375,
0.0004553794860839844,
0.00034928321838378906
] |
[
0.9990234375,
0.0006856918334960938,
0.0002168416976928711,
0.00015079975128173828
] |
Freshly collected specimens of starfish H. leviuscula spiculifera were immediately frozen after fishing. The sliced specimens (1.1 kg) were extracted twice with EtOH (2.0 L/kg) at room temperature. The extract was evaporated under reduced pressure, and the residue (83.2 g) was dissolved in H 2 O (0.5 L). The H 2 O-soluble fraction was passed through a Polychrom 1 column (7.5 × 75 cm) and eluted with H 2 O and then with EtOH. The combined EtOH eluate was concentrated under reduced pressure, and the resulting total fraction (9.7 g) was chromatographed over a Si gel column (6.5 × 15 cm) using CHCl 3 /EtOH (stepwise gradient, 5:1−1:3, v / v ). The obtained fractions were further purified on Florisil columns (7 × 15 cm) using CHCl 3 /EtOH (stepwise gradient, 3:1 to 1:3, v / v ) to yield four main fractions (1–4). Fr. 1 (242 mg) was subjected to HPLC on a Discovery C18 column (65% aq. EtOH, flow rate: 2.6 mL/min) and further separated on a YMC-Pack Pro C18 column (80% aq. MeOH, flow rate: 1.1 mL/min) to afford pure 4 (2.3 mg, t R 16.9 min). Fr. 2 (641 mg) was separated by HPLC on a Discovery C18 column (MeOH/H 2 O/1M NH 4 OAc, 55:44:1, v / v / v , flow rate: 1.7 mL/min) to give pure 2 (44 mg, t R 24.3 min) and 3 (17 mg, t R 36.7 min). Fr. 3 (235 mg) was subjected to HPLC on a YMC-Pack Pro C18 column (54% aq. EtOH, flow rate: 2.0 mL/min) and purified repeatedly on the same column (50% aq. EtOH, flow rate: 2.0 mL/min) to afford pure 1 (14 mg, t R 13.9 min).
|
[
"Alla A. Kicha",
"Dmitriy K. Tolkanov",
"Timofey V. Malyarenko",
"Olesya S. Malyarenko",
"Alexandra S. Kuzmich",
"Anatoly I. Kalinovsky",
"Roman S. Popov",
"Valentin A. Stonik",
"Natalia V. Ivanchina",
"Pavel S. Dmitrenok"
] |
https://doi.org/10.3390/md22070294
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39057403_p26
|
39057403
|
sec[2]/sec[3]/p[0]
|
3.4. Compound Characterization Data
| 4.25 |
biomedical
|
Study
|
[
0.9990234375,
0.0006651878356933594,
0.0005450248718261719
] |
[
0.9951171875,
0.004261016845703125,
0.0002357959747314453,
0.0003216266632080078
] |
Spiculiferoside A ( 1 ): Colorless powder; [α] D 25 : −27.6 ( c 0.43, MeOH); IR (KBr) ν max 3441, 2931, 1641, 1448, 1422, 1265, 1081, 1046 cm −1 ; (−)HRESIMS m / z 1001.4622 [M − Na] − ; (+)HRESIMS m / z 1047.4419 [M + Na] + ; (−)ESIMS/MS of the [M − Na] − ion with m / z 1001: 839 [(M − Na) − C 6 H 10 O 5 ] − , 679 [(M − Na) − C 6 H 10 O 5 − C 7 H 12 O 4 ] − , 661 [(M − Na) − C 6 H 10 O 5 − C 7 H 12 O 4 −H 2 O] − , 97 [HSO 4 ] − ; (+)ESIMS/MS of the [M + Na] + ion with m / z 1047: 927 [(M + Na) − NaHSO 4 ] + , 885 [(M + Na) − C 6 H 10 O 5 ] + , 725 [(M + Na) − C 6 H 10 O 5 − C 7 H 12 O 4 ] + , 651 [(M + Na) − C 6 H 10 O 5 − C 5 H 7 O 7 SNa] + , 633 [651 − H 2 O] + , 491 [651 − C 7 H 12 O 4 ] + , 143 [Na 2 HSO 4 ] + ; 1 H- and 13 C-NMR data of aglycon moiety, see Table 1 ; 1 H- and 13 C-NMR data of carbohydrate moiety, see Table 2 .
|
[
"Alla A. Kicha",
"Dmitriy K. Tolkanov",
"Timofey V. Malyarenko",
"Olesya S. Malyarenko",
"Alexandra S. Kuzmich",
"Anatoly I. Kalinovsky",
"Roman S. Popov",
"Valentin A. Stonik",
"Natalia V. Ivanchina",
"Pavel S. Dmitrenok"
] |
https://doi.org/10.3390/md22070294
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999995 |
39057403_p27
|
39057403
|
sec[2]/sec[3]/p[1]
|
3.4. Compound Characterization Data
| 4.238281 |
biomedical
|
Study
|
[
0.9990234375,
0.0005459785461425781,
0.0005450248718261719
] |
[
0.99658203125,
0.0029087066650390625,
0.00020515918731689453,
0.00022363662719726562
] |
Spiculiferoside B ( 2 ): Colorless powder; [α] D 25 : –27.8 ( c 0.71, MeOH); IR (KBr) ν max 3440, 2930, 1632, 1446, 1423, 1263, 1088, 1046, 1024, 979 cm −1 ; (−)HRESIMS m / z 839.4108 [M − Na] − ; (+)HRESIMS m / z 885.3896 [M + Na] + ; (−)ESIMS/MS of the [M − Na] − ion with m / z 839: 679 [(M − Na) − C 7 H 12 O 4 ] − , 661 [(M − Na) − C 7 H 12 O 4 − H 2 O] − , 211 [C 5 H 7 O 7 S] − , 152 [C 3 H 5 O 5 S] − , 97 [HSO 4 ] − ; (+)ESIMS/MS of the [M + Na] + ion with m / z 885: 765 [(M + Na) − NaHSO 4 ] + , 725 [(M + Na) − C 7 H 12 O 4 ] + , 651 [(M + Na) − C 5 H 7 O 7 SNa] + , 633 [651 − H 2 O] + , 491 [651 − C 7 H 12 O 4 ] + , 143 [N 2 HSO 4 ] + ; 1 H- and 13 C-NMR data of aglycon moiety, see Table 1 ; 1 H- and 13 C-NMR data of carbohydrate moiety, see Table 2 .
|
[
"Alla A. Kicha",
"Dmitriy K. Tolkanov",
"Timofey V. Malyarenko",
"Olesya S. Malyarenko",
"Alexandra S. Kuzmich",
"Anatoly I. Kalinovsky",
"Roman S. Popov",
"Valentin A. Stonik",
"Natalia V. Ivanchina",
"Pavel S. Dmitrenok"
] |
https://doi.org/10.3390/md22070294
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999995 |
39057403_p28
|
39057403
|
sec[2]/sec[3]/p[2]
|
3.4. Compound Characterization Data
| 4.210938 |
biomedical
|
Study
|
[
0.9990234375,
0.0005273818969726562,
0.0005273818969726562
] |
[
0.9970703125,
0.0027065277099609375,
0.0002027750015258789,
0.00021004676818847656
] |
Spiculiferoside C ( 3 ): Colorless powder; [α] D 25 : –17.6 ( c 0.6, MeOH); IR (KBr) ν max 3441, 2938, 1641, 1458, 1423, 1270, 1087, 1045, 984 cm −1 ; (−)HRESIMS m / z 823.4162 [M − Na] − ; (+)HRESIMS m / z 869 [M + Na] + ; (−)ESIMS/MS of the [M − Na] − ion with m / z 823: 663 [(M − Na) − C 7 H 12 O 4 ] − , 645 [(M − Na) − C 7 H 12 O 4 − H 2 O] − , 211 [C 5 H 7 O 7 S] − , 152 [C 3 H 5 O 5 S] − , 97 [HSO 4 ] − ; (+)ESIMS/MS of the [M + Na] + ion with m / z 869: 749 [(M + Na) − NaHSO 4 ] + , 709 [(M + Na) − C 7 H 12 O 4 ] + , 635 [(M + Na) − C 5 H 7 O 7 SNa] + , 617 [635 − H 2 O] + , 143 [Na 2 HSO 4 ] + ; 1 H- and 13 C-NMR data of aglycon moiety, see Table 1 ; 1 H- and 13 C-NMR data of carbohydrate moiety, see Table 2 .
|
[
"Alla A. Kicha",
"Dmitriy K. Tolkanov",
"Timofey V. Malyarenko",
"Olesya S. Malyarenko",
"Alexandra S. Kuzmich",
"Anatoly I. Kalinovsky",
"Roman S. Popov",
"Valentin A. Stonik",
"Natalia V. Ivanchina",
"Pavel S. Dmitrenok"
] |
https://doi.org/10.3390/md22070294
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39057403_p29
|
39057403
|
sec[2]/sec[3]/p[3]
|
3.4. Compound Characterization Data
| 4.402344 |
biomedical
|
Study
|
[
0.99853515625,
0.00046324729919433594,
0.0007777214050292969
] |
[
0.9970703125,
0.0026092529296875,
0.0002944469451904297,
0.00016748905181884766
] |
Spiculiferoside D ( 4 ): Colorless powder; [α] D 25 : –11.0 ( c 0.21, MeOH); (−)HRESIMS m / z 599.3794 [M − H] − , 635.3561 [M + Cl] − , 645.3847 [M + CHO 2 ] − ; (+)HRESIMS m / z 623.3770 [M + Na] + ; (−)ESIMS/MS of the [M − H] − ion with m / z 599: 467 [(M − H) − C 5 H 8 O 4 ] − , 449 [(M − H) − C 5 H 10 O 5 ] − , 131 [C 5 H 7 O 4 ] − ; (+)ESIMS/MS of the [M + Na] + ion with m / z 623: 605 [(M + Na) –H 2 O] + , 491 [(M + Na) − C 5 H 8 O 4 ] + , 473 [(M + Na) − C 5 H 10 O 5 ] + ; 1 H-NMR (CD 3 OD, 700.13 MHz): δ H 0.90 (d, J = 6.9 Hz, H 3 -26), 0.90 (d, J = 6.9 Hz, H 3 -27), 0.91 (d, J = 6.0 Hz, H 3 -21), 0.95 (s, H 3 -18), 0.96 (dd, J = 12.4, 3.2 Hz, H-9), 0.99 (m, H′-22), 1.00 (m, H′-1), 1.17 (d, J = 9.5 Hz, H-14), 1.22 (m, H′-12), 1.23 (m, H-5), 1.31 (m, H′-23), 1.33 (m, H-17), 1.33 (m, H-20), 1.42 (s, H 3 -19), 1.47 (m, H′-11), 1.58 (m, H-22), 1.58 (m, H-23), 1.59 (dd, J = 15.0, 3.2 Hz, H′-7), 1.61 (m, H′-2), 1.72 (m, H-1), 1.72 (m, H′-16), 1.80 (m, H-11), 1.83 (m, H-25), 1.88 (m, H-2), 1.90 (m, H-16), 1.96 (m, H-12), 2.40 (dd, J = 15.0, 3.0 Hz, H-7), 3.30 (m, H-24), 3.50 (m, H-3), 4.05 (m, H-4), 4.24 (m, H-6), 4.27 (td, J = 9.5, 3.2 Hz, H-15), 4.91 (d, J = 1.7 Hz, H-1′), 3.96 (m, H-2′), 3.83 (dd, J = 6.6, 4.0 Hz, H-3′), 3.97 (m, H-4′), 3.74 (dd, J = 12.0, 3.0 Hz, H-5′), 3.63 (dd, J = 12.0, 5.2 Hz, H-5′); 13 C-NMR (CD 3 OD, 176.04 MHz): 41.1 (C-1), 26.6 (C-2), 73.1 (C-3), 77.5 (C-4), 50.7 (C-5), 76.2 (C-6), 45.3 (C-7), 76.8 (C-8), 57.7 (C-9), 36.8 (C-10), 19.3 (C-11), 42.7 (C-12), 45.5 (C-13), 66.6 (C-14), 70.1 (C-15), 41.7 (C-16), 55.9 (C-17), 15.3 (C-18), 18.7 (C-19), 36.3 (C-20), 19.0 (C-21), 32.8 (C-22), 28.8 (C-23), 84.8 (C-24), 31.8 (C-25), 18.4 (C-26), 18.3 (C-27), 109.5 (C-1′), 83.9 (C-2′), 78.7 (C-3′), 85.1 (C-4′), 62.9 (C-5′).
|
[
"Alla A. Kicha",
"Dmitriy K. Tolkanov",
"Timofey V. Malyarenko",
"Olesya S. Malyarenko",
"Alexandra S. Kuzmich",
"Anatoly I. Kalinovsky",
"Roman S. Popov",
"Valentin A. Stonik",
"Natalia V. Ivanchina",
"Pavel S. Dmitrenok"
] |
https://doi.org/10.3390/md22070294
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999995 |
39057403_p30
|
39057403
|
sec[2]/sec[4]/p[0]
|
3.5. Acid Hydrolysis of Compound 1 and Determination of Absolute Configurations of the Sugars by GC
| 4.222656 |
biomedical
|
Study
|
[
0.9990234375,
0.00047087669372558594,
0.00031495094299316406
] |
[
0.998046875,
0.0014181137084960938,
0.000301361083984375,
0.00013554096221923828
] |
Compound 1 (1.2 mg) in a solution of 2 M TFA (1.0 mL) was heated in a H 2 O bath at 100 °C for 2 h. The reaction mixture was diluted with H 2 O (0.5 mL), washed with CHCl 3 (3 × 0.5 mL), and then evaporated under reduced pressure. ( R )-(−)-2-octanol (Aldrich) (0.4 mL) and one drop of conc. TFA was added to the dried residue, and the reaction mixture was heated in a glycerol bath at 130 °C for 6 h. The solution was concentrated under reduced pressure and treated with a mixture of Py/Ac 2 O (1:1, 0.5 mL) for 24 h at room temperature. The reaction mixture was evaporated under reduced pressure, and the resulting acetylated 2-octylglycosides of monosaccharides were analyzed by GC using the corresponding standard samples prepared in the same manner. The retention times of four tautomeric forms (two pyranoses and two furanoses) for each monosaccharide derivative from 1 were as follows: 2,4-di- O -methyl- d -xylose (19.96, 20.04 min); L-arabinose (22.44, 22.79, 23.04, and 23.33 min); and D-glucose (26.31, 26.95, 27.18, and 27.46 min). The retention times of the standard samples were as follows: 2,4-di- O -methyl- d -xylose (19.95, 20.03 min); L-arabinose (22.48, 22.83, 23.08, and 23.37 min); and D-glucose (26.32, 26.97, 27.19, and 27.48 min).
|
[
"Alla A. Kicha",
"Dmitriy K. Tolkanov",
"Timofey V. Malyarenko",
"Olesya S. Malyarenko",
"Alexandra S. Kuzmich",
"Anatoly I. Kalinovsky",
"Roman S. Popov",
"Valentin A. Stonik",
"Natalia V. Ivanchina",
"Pavel S. Dmitrenok"
] |
https://doi.org/10.3390/md22070294
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
39057403_p31
|
39057403
|
sec[2]/sec[5]/p[0]
|
3.6. Solvolysis of Compound 2
| 4.394531 |
biomedical
|
Study
|
[
0.9990234375,
0.0005354881286621094,
0.0003046989440917969
] |
[
0.99853515625,
0.0009703636169433594,
0.0003757476806640625,
0.00017380714416503906
] |
A solution of 2 (2.8 mg) in a mixture of dioxane/Py (1:1, 0.8 mL) was heated at 100 °C for 4 h. The reaction mixture was evaporated under reduced pressure and purified by HPLC on an analytical Diasfer-110-C18 column with 80% aq. MeOH (0.5 mL/min) as an eluent system to give pure desulfated derivative 2a (1.2 mg): (−)HRESIMS m / z 759.4547 [M − H] − , 819.4754 [M + C 2 H 4 O 2 ] − ; (+)HRESIMS m / z 783.4535 [M + Na] + ; (−)ESIMS/MS of the [M − H] − ion with m / z 759: 627 [(M − H) − C 5 H 8 O 4 ] − , 609 [(M − H) − C 5 H 10 O 5 ], 467 [(M − H) − C 5 H 8 O 4 − C 6 H 10 O 4 ] − , 131 [C 5 H 7 O 4 ] − ; (+)ESIMS/MS of the [M + Na] + ion with m / z 783: 765 [(M + Na) − H 2 O] + , 623 [(M + Na) − C 6 H 10 O 4 ] + ; 1 H-NMR and 13 C-NMR data were identical with those of forbeside J : 1 H-NMR (CD 3 OD, 700.13 MHz): δ H 0.90 (d, J = 6.8 Hz, H 3 -26), 0.90 (d, J = 6.8 Hz, H 3 -27), 0.91 (d, J = 6.8 Hz, H 3 -21), 0.95 (s, H 3 -18), 0.97 (dd, J = 12.5, 3.0 Hz, H-9), 0.99 (m, H′-22), 1.01 (m, H′-1), 1.17 (m, H-14), 1.22 (m, H′-12), 1.23 (m, H-5), 1.31 (m, H′-23), 1.33 (m, H-17), 1.34 (m, H-20), 1.43 (s, H 3 -19), 1.47 (m, H′-11), 1.59 (m, H-22), 1.59 (m, H-23), 1.59 (m, H′-7), 1.70 (m, H′-2), 1.72 (m, H′-16), 1.74 (m, H-1), 1.81 (m, H-11), 1.83 (m, H-25), 1.90 (m, H-16), 1.96 (m, H-2), 1.97 (m, H-12), 2.41 (dd, J = 15.0, 2.9 Hz, H-7), 3.31 (m, H-24), 3.64 (m, H-3), 4.25 (m, H-4), 4.26 (m, H-6), 4.27 (td, J = 10.2, 3.6 Hz, H-15), 4.92 (d, J = 1.8 Hz, H-1′), 3.96 (m, H-2′), 3.83 (dd, J = 6.6, 4.0 Hz, H-3′), 3.97 (m, H-4′), 3.74 (dd, J = 11.9, 3.3 Hz, H-5′), 3.63 (m, H-5′), 4.44 (d, J = 7.5 Hz, H-1″), 2.92 (dd, J = 9.1, 7.6 Hz, H-2″), 3.43 (t, J = 8.9 Hz, H-3″), 3.17 (m, H-4″), 3.14 (dd, J = 11.0, 10.0 Hz, H-5″), 4.00 (dd, J = 11.0, 4.8 Hz, H-5″), 3.46 (s, 2″-OMe), 3.61 (s, 4″-OMe); 13 C-NMR (CD 3 OD, 176.04 MHz): 41.1 (C-1), 25.2 (C-2), 80.5 (C-3), 74.6 (C-4), 50.5 (C-5), 76.2 (C-6), 45.2 (C-7), 76.7 (C-8), 57.7 (C-9), 36.9 (C-10), 19.3 (C-11), 42.7 (C-12), 45.5 (C-13), 66.5 (C-14), 70.1 (C-15), 41.7 (C-16), 55.9 (C-17), 15.3 (C-18), 18.6 (C-19), 36.3 (C-20), 19.0 (C-21), 32.8 (C-22), 28.8 (C-23), 84.8 (C-24), 31.8 (C-25), 18.4 (C-26), 18.3 (C-27), 109.5 (C-1′), 83.9 (C-2′), 78.7 (C-3′), 85.1 (C-4′), 62.9 (C-5′), 102.5 (C″-1), 84.7 (C″-2), 76.8 (C″-3), 81.0 (C″-4), 64.2 (C″-5), 59.0 (2″-OMe), 61.0 (4″-OMe).
|
[
"Alla A. Kicha",
"Dmitriy K. Tolkanov",
"Timofey V. Malyarenko",
"Olesya S. Malyarenko",
"Alexandra S. Kuzmich",
"Anatoly I. Kalinovsky",
"Roman S. Popov",
"Valentin A. Stonik",
"Natalia V. Ivanchina",
"Pavel S. Dmitrenok"
] |
https://doi.org/10.3390/md22070294
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057403_p32
|
39057403
|
sec[2]/sec[6]/p[0]
|
3.7. Reagents
| 1.473633 |
biomedical
|
Other
|
[
0.9931640625,
0.0012598037719726562,
0.005619049072265625
] |
[
0.168212890625,
0.82763671875,
0.0015697479248046875,
0.0027751922607421875
] |
The phosphate-buffered saline (PBS), L-glutamine, penicillin/streptomycin solution (10,000 U/mL, 10 µg/mL), Minimum Essential Medium Eagle (MEM), Dulbecco’s Modified Eagle Medium (DMEM), McCoy’s 5A modified medium (McCoy’s 5A), and Basal Medium Eagle (BME) were purchased from the Sigma-Aldrich company (St. Louis, MO, USA). The MTS reagent 3-[4,5-dimethylthiazol-2-yl]-2,5-diphenyltetrazolium bromide was purchased from Promega (Madison, WI, USA). The trypsin, fetal bovine serum (FBS), and the protein marker PageRulerTM Plus Prestained Protein Ladder were purchased from Thermo Fisher Scientific (Waltham, MA, USA).
|
[
"Alla A. Kicha",
"Dmitriy K. Tolkanov",
"Timofey V. Malyarenko",
"Olesya S. Malyarenko",
"Alexandra S. Kuzmich",
"Anatoly I. Kalinovsky",
"Roman S. Popov",
"Valentin A. Stonik",
"Natalia V. Ivanchina",
"Pavel S. Dmitrenok"
] |
https://doi.org/10.3390/md22070294
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057403_p33
|
39057403
|
sec[2]/sec[6]/p[1]
|
3.7. Reagents
| 2.015625 |
biomedical
|
Other
|
[
0.99609375,
0.0005383491516113281,
0.0032558441162109375
] |
[
0.3623046875,
0.634765625,
0.001461029052734375,
0.0014781951904296875
] |
The cell lysis buffer (10×), CDK2 , CDK4 , Cyclin D1 , p21 Waf1/Cip1 , phospho-c-Raf (Ser338) , phospho-p44/42 MAPK (phospho-Erk1/2) (Thr202/Tyr204) , p44/42 MAPK (Erk1/2) , phospho-MEK1/2 (Ser217/221) , and MEK1/2 antibodies were obtained from Cell Signaling Technology (Danvers, MA, USA); β-actin and horseradish peroxidase (HRP) conjugated secondary antibody from rabbit and mouse were purchased from the Sigma-Aldrich company (St. Louis, MO, USA).
|
[
"Alla A. Kicha",
"Dmitriy K. Tolkanov",
"Timofey V. Malyarenko",
"Olesya S. Malyarenko",
"Alexandra S. Kuzmich",
"Anatoly I. Kalinovsky",
"Roman S. Popov",
"Valentin A. Stonik",
"Natalia V. Ivanchina",
"Pavel S. Dmitrenok"
] |
https://doi.org/10.3390/md22070294
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999999 |
39057403_p34
|
39057403
|
sec[2]/sec[7]/p[0]
|
3.8. Cell Lines and Cell Culture Conditions
| 3.871094 |
biomedical
|
Study
|
[
0.99951171875,
0.00017952919006347656,
0.00024437904357910156
] |
[
0.97802734375,
0.0202178955078125,
0.0015468597412109375,
0.0002467632293701172
] |
Human cell lines, including embryonic kidney HEK293 , melanoma SK-MEL-28 (ATCC ® HTB-72™), breast cancer MDA-MB-231 (ATCC ® HTB-26™), and colorectal carcinoma cells HCT 116 (ATCC ® CCL-247™) were obtained from American Type Culture Collection (ATCC, Manassas, VA, USA). All cells were maintained according to ATCC protocols and routinely checked for contamination with mycoplasma.
|
[
"Alla A. Kicha",
"Dmitriy K. Tolkanov",
"Timofey V. Malyarenko",
"Olesya S. Malyarenko",
"Alexandra S. Kuzmich",
"Anatoly I. Kalinovsky",
"Roman S. Popov",
"Valentin A. Stonik",
"Natalia V. Ivanchina",
"Pavel S. Dmitrenok"
] |
https://doi.org/10.3390/md22070294
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999995 |
39057403_p35
|
39057403
|
sec[2]/sec[7]/p[1]
|
3.8. Cell Lines and Cell Culture Conditions
| 4.058594 |
biomedical
|
Study
|
[
0.9990234375,
0.0005354881286621094,
0.000362396240234375
] |
[
0.8291015625,
0.1676025390625,
0.0025615692138671875,
0.0008630752563476562
] |
HEK293 cells were grown in MEM medium; SK-MEL-28 and MDA-MB-231 cells were cultured in DMEM medium, while HCT 116 cells were maintained in McCoys’ 5A medium according to the manufacturer’s instructions. Culture media were supplemented with 10% heat-inactivated fetal bovine serum (FBS) and 1% penicillin–streptomycin solution. The cells were cultured at 37 °C in a humidified atmosphere containing 5% CO 2 .
|
[
"Alla A. Kicha",
"Dmitriy K. Tolkanov",
"Timofey V. Malyarenko",
"Olesya S. Malyarenko",
"Alexandra S. Kuzmich",
"Anatoly I. Kalinovsky",
"Roman S. Popov",
"Valentin A. Stonik",
"Natalia V. Ivanchina",
"Pavel S. Dmitrenok"
] |
https://doi.org/10.3390/md22070294
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39057403_p36
|
39057403
|
sec[2]/sec[8]/p[0]
|
3.9. Cell Viability Assay
| 4.117188 |
biomedical
|
Study
|
[
0.99951171875,
0.00032067298889160156,
0.00015020370483398438
] |
[
0.99755859375,
0.002063751220703125,
0.0004801750183105469,
0.00011926889419555664
] |
The CellTiter 96 ® Aqueous One Solution Cell Proliferation Assay kit (MTS) was used for cell viability analysis and performed according to the standard protocol. Briefly, cells (1 × 10 4 /200 µL) were seeded in 96-well plates and incubated for 24 h in a humidified atmosphere containing 5% CO 2 . Then, they were treated with DMSO (control) and compounds T1, B1, and B2 at 1, 10, 50, and 100 µM for an additional 24 h. The MTS reagent (20 μL/well) was added to the cell culture medium and incubated at a 37 °C incubator for 2 h. Cell viability was examined at 490/630 nm using a Power Wave XS microplate reader (BioTek, Winooski, VT, USA).
|
[
"Alla A. Kicha",
"Dmitriy K. Tolkanov",
"Timofey V. Malyarenko",
"Olesya S. Malyarenko",
"Alexandra S. Kuzmich",
"Anatoly I. Kalinovsky",
"Roman S. Popov",
"Valentin A. Stonik",
"Natalia V. Ivanchina",
"Pavel S. Dmitrenok"
] |
https://doi.org/10.3390/md22070294
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
39057403_p37
|
39057403
|
sec[2]/sec[8]/p[1]
|
3.9. Cell Viability Assay
| 4.074219 |
biomedical
|
Study
|
[
0.99951171875,
0.00018227100372314453,
0.0001493692398071289
] |
[
0.99853515625,
0.0011262893676757812,
0.00043511390686035156,
0.00007301568984985352
] |
The concentration at which the compounds exert half of their maximal inhibitory effect on cell viability (IC 50 ) was calculated by the AAT-Bioquest ® online calculator . The selectivity index (SI) was calculated as described previously using the following formula: SI = IC 50 of the compounds in normal cell (HEK293)/IC 50 of the same compounds in human colorectal adenocarcinoma cell line (HCT 116).
|
[
"Alla A. Kicha",
"Dmitriy K. Tolkanov",
"Timofey V. Malyarenko",
"Olesya S. Malyarenko",
"Alexandra S. Kuzmich",
"Anatoly I. Kalinovsky",
"Roman S. Popov",
"Valentin A. Stonik",
"Natalia V. Ivanchina",
"Pavel S. Dmitrenok"
] |
https://doi.org/10.3390/md22070294
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999995 |
39057403_p38
|
39057403
|
sec[2]/sec[9]/p[0]
|
3.10. Cell Proliferation Assay
| 4.121094 |
biomedical
|
Study
|
[
0.99951171875,
0.0003733634948730469,
0.00017702579498291016
] |
[
0.99853515625,
0.0011167526245117188,
0.00035309791564941406,
0.00009840726852416992
] |
HCT 116 cells (8 × 10 3 /200 µL) were seeded in 96-well plates and incubated for 24 h in a CO 2 incubator. The cells’ monolayers were washed with phosphate-buffered saline (PBS) to remove unattached cells. The attached cells were incubated with fresh medium containing DMSO (control) and B2 (0–100 µM) for 24, 48, and 72 h. Subsequently, the cells were incubated with 15 µL MTS reagent for 2 h, and the absorbance of each well was measured at 490/630 nm using a microplate reader (Power Wave XS, USA).
|
[
"Alla A. Kicha",
"Dmitriy K. Tolkanov",
"Timofey V. Malyarenko",
"Olesya S. Malyarenko",
"Alexandra S. Kuzmich",
"Anatoly I. Kalinovsky",
"Roman S. Popov",
"Valentin A. Stonik",
"Natalia V. Ivanchina",
"Pavel S. Dmitrenok"
] |
https://doi.org/10.3390/md22070294
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057403_p39
|
39057403
|
sec[2]/sec[10]/p[0]
|
3.11. Anchorage-Independent Cell Growth Assay
| 4.082031 |
biomedical
|
Study
|
[
0.99951171875,
0.0001856088638305664,
0.00017631053924560547
] |
[
0.9990234375,
0.000583648681640625,
0.0003337860107421875,
0.00006449222564697266
] |
Soft agar assay was performed as described previously . Briefly, the cells were counted and seeded into 6-well plates at a density of 8×10 3 /per well with 0.3% BME agar containing 10% FBS and DMSO (control) or various concentrations of compounds 1 , 2 , and 3 (10, 20, and 40 µM). The number of the colonies was determined using a Motic microscope AE 20 and ImageJ software bundled with 64-bit Java 1.8.0_112 (NIH, Bethesda, MD, USA) 14 days later.
|
[
"Alla A. Kicha",
"Dmitriy K. Tolkanov",
"Timofey V. Malyarenko",
"Olesya S. Malyarenko",
"Alexandra S. Kuzmich",
"Anatoly I. Kalinovsky",
"Roman S. Popov",
"Valentin A. Stonik",
"Natalia V. Ivanchina",
"Pavel S. Dmitrenok"
] |
https://doi.org/10.3390/md22070294
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
39057403_p40
|
39057403
|
sec[2]/sec[11]/p[0]
|
3.12. Cell Cycle Assay
| 4.132813 |
biomedical
|
Study
|
[
0.99951171875,
0.00032067298889160156,
0.00016570091247558594
] |
[
0.99853515625,
0.0008687973022460938,
0.00036644935607910156,
0.00009268522262573242
] |
HCT 116 cells (3 × 10 5 ) were seeded in 60 mm dishes and incubated for 24 h in a CO 2 incubator. The attached cells were treated by DMSO (control) or 3 (10, 20, and 40 µM) for 72 h. Then, cells were harvested, washed with ice-cold 1× PBS, and fixed with 70% ethanol. HCT 116 cells were incubated overnight at –20 °C, and then fixed cells were collected by centrifugation at 4000 rpm for 10 min and rinsed with 1× PBS. The cell pellet was resuspended in Muse™ Cell Cycle Reagent , and the cells were incubated for 30 min at RT in the dark. The DNA content was assessed by measuring the fluorescence intensity by flow cytometry (Muse™ Cell Analyzer). Results were expressed as a percentage of cells in the G0/1, S, and G2/M phases of the cell cycle, associated with the DNA content profile histograms
|
[
"Alla A. Kicha",
"Dmitriy K. Tolkanov",
"Timofey V. Malyarenko",
"Olesya S. Malyarenko",
"Alexandra S. Kuzmich",
"Anatoly I. Kalinovsky",
"Roman S. Popov",
"Valentin A. Stonik",
"Natalia V. Ivanchina",
"Pavel S. Dmitrenok"
] |
https://doi.org/10.3390/md22070294
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39057403_p41
|
39057403
|
sec[2]/sec[12]/p[0]
|
3.13. Western Blot Assay
| 4.113281 |
biomedical
|
Study
|
[
0.99951171875,
0.0002472400665283203,
0.0001811981201171875
] |
[
0.99853515625,
0.0008406639099121094,
0.0003693103790283203,
0.00007712841033935547
] |
HCT 116 cells (1.0 × 10 5 /mL) were seeded in 100 mm dishes and incubated for 24 h at 37 °C in a CO 2 incubator. The cells were treated by DMSO (control) or 3 (10, 20, and 40 µM) for 72 h. Then, cells were harvested and lysed by 1× cell lysis buffer (“Cell Signaling Technology”, Danvers, MA, USA) according to the manufacturer’s protocol. Cells’ protein content was determined by the DC protein assay (Bio-Rad, Hercules, CA, USA). Lysates of protein (20–40 µg) were exposed to 10% or 12% SDS-PAGE and electrophoretically transferred to polyvinylidene difluoride membranes (PVDF) (Millipore, Burlington, MA, USA). The membranes were blocked with 5% non-fat milk (Bio-Rad) for 1 h and then incubated with the respective specific primary antibody at 4 °C overnight. Protein bands were visualized using an enhanced chemiluminescence reagent (ECL) (Bio-Rad, Hercules, CA, USA) after hybridization with an HRP-conjugated secondary antibody.
|
[
"Alla A. Kicha",
"Dmitriy K. Tolkanov",
"Timofey V. Malyarenko",
"Olesya S. Malyarenko",
"Alexandra S. Kuzmich",
"Anatoly I. Kalinovsky",
"Roman S. Popov",
"Valentin A. Stonik",
"Natalia V. Ivanchina",
"Pavel S. Dmitrenok"
] |
https://doi.org/10.3390/md22070294
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057403_p42
|
39057403
|
sec[2]/sec[13]/p[0]
|
3.14. Statistical Analysis
| 3.167969 |
biomedical
|
Study
|
[
0.9990234375,
0.00018870830535888672,
0.0005884170532226562
] |
[
0.9775390625,
0.0208282470703125,
0.0016164779663085938,
0.00024890899658203125
] |
All of the assays were performed in at least three independent experiments. Results are expressed as the mean ±standard deviation (SD). Statistical procedures were performed using one-way ANOVA and Tukey’s HSD tests with * p < 0.05, ** p < 0.01, and *** p < 0.001.
|
[
"Alla A. Kicha",
"Dmitriy K. Tolkanov",
"Timofey V. Malyarenko",
"Olesya S. Malyarenko",
"Alexandra S. Kuzmich",
"Anatoly I. Kalinovsky",
"Roman S. Popov",
"Valentin A. Stonik",
"Natalia V. Ivanchina",
"Pavel S. Dmitrenok"
] |
https://doi.org/10.3390/md22070294
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057403_p43
|
39057403
|
sec[3]/p[0]
|
4. Conclusions
| 4.378906 |
biomedical
|
Study
|
[
0.99951171875,
0.00038886070251464844,
0.0003235340118408203
] |
[
0.9990234375,
0.00022995471954345703,
0.0004782676696777344,
0.00010383129119873047
] |
Polar steroid compounds from the ethanolic extract of the Sea of Okhotsk starfish Henricia leviuscula spiculifera were investigated. New monosulfated steroid glycosides, spiculiferosides A, B, and C, and a new unsulfated related monoglycoside, spiculiferoside D, were isolated, and their chemical structures were characterized. Three of them contain two carbohydrate chains, which are located at positions C-3 and C-24 of the polyhydroxylated cholestane aglycone. Spiculiferosides B and C are biosides, and spiculiferoside A, in contrast, has three monosaccharide residues. Previously, only five such “two-chains” triglycosides were known from sea stars, kurilensosides A, B, C, and I, found in the Far Eastern starfish Hippasteria kurilensis , and planciside D isolated from the tropical starfish Acanthaster planci , which also contain two carbohydrate patterns attached to the steroid core and aglycon side chain. The 5-substituted 3-OSO 3 -α-L-Araf residue of spiculiferoside A was discovered and described for the first time in steroid glycosides of starfish. Moreover, the 3-OSO 3 -α-L-Araf residue that comprised spiculiferosides B and C, was previously found only in one steroid glycoside from the sea star Oreaster reticulatus . Interestingly, we did not find “classical” oligoglycosides (asterosaponins) in the starfish H. leviuscula spiculifera as in most previously studied species of the genus Henricia .
|
[
"Alla A. Kicha",
"Dmitriy K. Tolkanov",
"Timofey V. Malyarenko",
"Olesya S. Malyarenko",
"Alexandra S. Kuzmich",
"Anatoly I. Kalinovsky",
"Roman S. Popov",
"Valentin A. Stonik",
"Natalia V. Ivanchina",
"Pavel S. Dmitrenok"
] |
https://doi.org/10.3390/md22070294
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |