text
stringlengths 12
2.52M
| meta
dict |
|---|---|
Background
==========
Gonorrhoea is a major public health concern globally \[[@B1]\]. In 2011, 39 179 gonorrhoea cases were reported from 28 European Union (EU)/European Economic Area (EEA) Member States, with an overall incidence of 12.6 cases per 100,000 population \[[@B2]\]. In most of the non-EU/EEA countries of the World Health Organization (WHO) European Region (mostly in the former Soviet Union and Yugoslav Republic), during the last two decades the incidence has rapidly declined. This decline has also been observed in the largest non-EU/EEA country, that is, Russia (\>142 million inhabitants). However, in 2011 Russia still reported an incidence of 38 cases per 100,000 population, which was the highest country incidence in the WHO European Region \[2,3,<http://data.euro.who.int/cisid>\]. Furthermore, the reported gonorrhoea incidences in Russia are underestimated, which is due to the large heterogeneity in health care settings nationally, differences in access to testing, suboptimal diagnostics, case reporting, e.g. lack of reporting of cases diagnosed in the private sector, and surveillance \[[@B3]-[@B5]\].
Unfortunately, the etiological agent of gonorrhoea, *Neisseria gonorrhoeae,* has developed antimicrobial resistance (AMR) to essentially all antimicrobials introduced as first-line treatment. Currently, ceftriaxone is the only recommended first-line option for antimicrobial monotherapy in many countries globally \[[@B6]-[@B12]\]. Most worryingly, rare verified treatment failure of pharyngeal gonorrhoea with ceftriaxone have been reported \[[@B13]-[@B18]\] and the first few extensively-drug resistant (XDR) gonococcal strains with high-level resistance to ceftriaxone were described recently \[[@B16],[@B19],[@B20]\]. In this era of hard-to-treat and possibly emergence of untreatable gonorrhoea, the WHO \[[@B21]\], European Centre for Disease Prevention and Control (ECDC) \[[@B22]\] and Centers for Disease Control and Prevention (CDC), USA \[[@B23]\] have published action/response plans to mitigate and control the spread of multidrug-resistant gonorrhoea. One key action emphasized for public health purposes in all these action/response plans is to substantially enhance the quality assured surveillance of gonococcal AMR worldwide \[[@B21]-[@B23]\].
In the WHO European Region, the European Gonococcal Antimicrobial Surveillance Programme (Euro-GASP) is operating in the EU/EEA since 2004. In Euro-GASP, 68% (21/31) of the EU/EEA countries are included in the gonococcal AMR surveillance \[[@B3],[@B12]\]. However, in the non-EU/EEA countries of the WHO European Region, quality assured gonococcal AMR surveillance only exist in 13% (3/23) of the countries \[[@B3]\] and, in general, the awareness and knowledge regarding gonococcal AMR, which is crucial for informing the empirical treatment guidelines, is limited \[[@B3]-[@B5]\]. Nevertheless, in 2004 the national Russian GASP (RU-GASP) was initiated. The RU-GASP has been quality assured in accordance with WHO standards and, for international comparability of AMR data, the 2008 WHO *N. gonorrhoeae* reference strains are used as quality controls \[[@B24]-[@B27]\]. For molecular epidemiological typing of gonococci, the *N. gonorrhoeae* multiantigen sequence typing (NG-MAST) has been used in many countries worldwide \[[@B28]\]. However, very limited genetic characteristics of gonococcal strains circulating in Russia have been published and only two NG-MAST studies have been performed, examining isolates from 2004--2005 \[[@B29],[@B30]\].
The aims of the present study were to examine the prevalence and trends of *N. gonorrhoeae* resistance, to previous and current antimicrobial treatment options, from 2009 to 2012 in Russia and the genotypic distribution of *N. gonorrhoeae*, by means of NG-MAST, isolated in 2011 and 2012 in Russia.
Methods
=======
Study population
----------------
As previously described \[[@B24],[@B25]\], dermatovenereological dispensaries situated all over Russia are surveyed in RU-GASP. In the present study, mainly consecutive culture positive gonorrhoea patients attending 12--46 dispensaries from January 2009 to December 2012 were included. Urethral and cervical specimens from females and urethral specimens from males were collected. All specimens were cultured on selective gonococcal agar media, and the *N. gonorrhoeae* isolates were preserved in cryomedium at -70°C and subsequently transported to the SRCDV for complete species verification and centralized AMR testing, as previously described \[[@B24],[@B25]\]. At the SRDCDV, all isolates were confirmed as *N. gonorrhoeae* by identification of typical colonies on selective culture agar media, Gram negative diplococci in microscopy, rapid oxidase reaction, and a sugar utilization test \[[@B31]\]. All examined gonococcal isolates were cultured and stored as part of the routine diagnostics (standard care) and no patient identification information is used in RU-GASP.
Antimicrobial susceptibility testing
------------------------------------
At the SRCDV, the minimum inhibitory concentration (MIC, mg/L) of ceftriaxone (0.002-4 mg/L), spectinomycin (0.125-512 mg/L), azithromycin (0.002-4 mg/L), penicillin G (0.016-16 mg/L), and ciprofloxacin (0.002-128 mg/L) was determined using agar dilution method, according to the recommendations from the US Clinical and Laboratory Standards Institute \[CLSI; 32\]. All antimicrobial powder was purchased from Fluka Analytical (Steinheim, Germany). For azithromycin, for which the CLSI does not state any breakpoints, the MIC breakpoints from the European Committee on Antimicrobial Susceptibility Testing (EUCAST; <http://www.eucast.org/clinical_breakpoints>) were used. For quality control, as recommended by the CLSI \[[@B32]\] the *N. gonorrhoeae* reference strain АТСС 49226 was included in each testing. The 2008 WHO *N. gonorrhoeae* reference strains \[[@B26]\] were also included in the quality control on a regular basis. β-lactamase production was identified using nitrocefin discs, according to the manufacturer's instructions (Cefinase discs; Becton Dickinson, Cockeysville, Md, USA).
Isolation of genomic DNA
------------------------
DNA was isolated from bacterial suspensions using the DNA express kit (Lytech Ltd, Moscow, Russia), according to the instructions from the manufacturer.
Molecular epidemiological typing
--------------------------------
For molecular epidemiological typing, NG-MAST was performed on gonococcal isolates from 2011 (n = 421) and 2012 (n = 100), as previously described \[[@B33]\]. NG-MAST allele numbers of the more variable segments of *porB* and *tbpB*, and sequence types (STs) were assigned using the NG-MAST website (<http://www.ng-mast.net>).
Statistical analysis
--------------------
Statistical analysis was performed using the Statistica software version 9.0 PL (StatSoft Corporation, Cracow, Poland). Z-test for comparison of proportions was used. The level of significance was set at *P* \< 0.05.
Results
=======
Patient characteristics
-----------------------
*N. gonorrhoeae* isolates (one isolate per patient) from 1200 patients (959 males and 241 females), 407 patients (324 males and 83 females), 423 patients (295 males and 128 females), and 106 (65 males and 41 females) in 2009, 2010, 2011 and 2012, respectively, were examined. The mean ages of the males (n = 1643) were 26.8 years (median age: 25 years; range: 15 to 64 years) and the mean ages of the females (n = 493) were 25.3 years (median age: 24 years; range: 16 to 76 years). The age distribution was relatively similar during the four years investigated.
Antimicrobial susceptibility of *N. gonorrhoeae* isolates in 2009--2012 (n = 2136) in Russia
--------------------------------------------------------------------------------------------
The results of the antimicrobial susceptibility testing of all isolates are summarized in Table [1](#T1){ref-type="table"}.
######
**Antimicrobial resistance and β-lactamase production in*Neisseria gonorrhoeae*isolates (n = 2136) from Russia in 2009--2012**
**Number (%) of resistant isolates**
---------------------------- -------------------------------------- ------------ ------------ -----------
**Ciprofloxacin** 533 (44.4) 217 (53.2) 138 (32.6) 27 (25.5)
(R \> 0.5 mg/L)^*a*^
**Penicillin G** 115 (9.6) 51 (12.5) 56 (13.2) 12 (11.3)
(R \> 1 mg/L)^*a*^
**Azithromycin** 28 (2.3) 20 (4.9) 70 (16.5) 18 (17.0)
(R \> 0.5 mg/L)^*a*^
**Spectinomycin** 16 (1.3) 18 (4.4) 49 (11.6) 1 (0.9)
(R \> 64 mg/L)^*a*^
**Ceftriaxone** 0 0 0 0
(R \> 0.25 mg/L)^*a*^
**β-lactamase production** 4 (0.3) 0 2 (0.5) 0
^*a*^Breakpoints for resistance according to the US Clinical and Laboratory Standards Institute \[CLSI; 32\].
For azithromycin, for which CLSI does not state any breakpoints, the resistance breakpoint from the European Committee on Antimicrobial Susceptibility Testing \[EUCAST; <http://www.eucast.org/clinical_breakpoints>\] was used.
Briefly, in 2012 the proportion of isolates with *in vitro* resistance was 25.5%, 17.0%, 11.3%, 0.9%, and 0% for ciprofloxacin, azithromycin, penicillin G, spectinomycin, and ceftriaxone, respectively. During 2009--2012, the proportions of *N. gonorrhoeae* isolates resistant to ciprofloxacin, penicillin G, azithromycin and spectinomycin ranged from 25.5% to 44.4%, 9.6% to 13.2%, 2.3% to 17.0% and 0.9% to 11.6%, respectively. The overall number of β-lactamase producing *N. gonorrhoeae* isolates was 6 (0.3%). In general, the resistance level to penicillin G was stable, the resistance level to ciprofloxacin was declining, however, the level of resistance to azithromycin was increasing significantly (*P* \< 0.05) (Table [1](#T1){ref-type="table"}). However, the highest MIC of azithromycin detected was 8 mg/L and no isolates with high-level resistance to azithromycin (MIC ≥ 256 mg/L) have yet been found in Russia. Worryingly, gonococcal isolates with low-level resistance to spectinomycin were identified in all the surveyed years. Nevertheless, no isolates with high-level resistance to spectinomycin (MIC ≥ 1024 mg/L) have yet been identified in Russia and the spectinomycin MICs of the identified isolates only ranged from 128 to 256 mg/L. All isolates from 2009 to 2012 were susceptible to ceftriaxone (Table [1](#T1){ref-type="table"}). Nevertheless, using the European EUCAST breakpoint (<http://www.eucast.org>; R \> 0.125 mg/L), in total 58 (2.7%) of the isolates during 2009--2012 were resistant to ceftriaxone. Interestingly, the prevalence of the isolates resistant to ceftriaxone according to the EUCAST breakpoint decreased significantly (*P* \< 0.05), i.e. 48 (4.0%), 8 (2.0%), 2 (0.5%) and 0 (0%) isolates were found in 2009, 2010, 2011 and 2012, respectively. Furthermore, in general the annual MIC distribution for ceftriaxone appeared to shift to lower MICs during the study period 2009--2012 (Figure [1](#F1){ref-type="fig"}).
{#F1}
*Neisseria gonorrhoeae* multiantigen sequence typing (NG-MAST)
--------------------------------------------------------------
The examined gonococcal isolates from 2011 (n = 421) and 2012 (n = 100) were assigned to 183 NG-MAST STs. Hundred twenty-two (66.7%) of these STs were not previously described. The most prevalent STs were ST807 (n = 41, 7.9% of isolates), ST5714 (n = 32, 6.1%), ST228 (n = 14, 2.7%), ST5042 (n = 11, 2.1%), ST1152 (n = 10, 1.9%), ST5825 (n = 10, 1.9%), ST5937 (n = 10, 1.9%), and ST5718 (n = 9, 1.9%). Five STs were represented by eight isolates, two STs by seven isolates, eight STs by six isolates, seven STs by five isolates, eight STs by four isolates, 18 STs by three isolates, 33 STs two isolates and remaining 98 STs were represented by single isolates.
In general, the most prevalent STs such as ST807, ST5714, ST228, ST5042, ST1152, ST5825, ST5937, and ST5718 had relatively low MICs of ceftriaxone, ranging from ≤0.016 mg/L to 0.064 mg/L. Notable, the two gonococcal isolates obtained in 2011 with a ceftriaxone MIC of 0.25 mg/L (resistant according to the European EUCAST breakpoint) were assigned as ST2861 and ST5929. One isolate assigned as ST1407, which is an internationally spread multidrug resistant gonococcal clone \[[@B8],[@B12],[@B15],[@B19],[@B20],[@B34],[@B35]\], was also identified in 2012 (in Ryazan, Central Federal District). This isolate was resistant to ciprofloxacin and, with exception of ceftriaxone, had similar antimicrobial resistance profile as ST1407 isolates described internationally. Surprisingly, the MIC of ceftriaxone was only 0.008 mg/L. Furthermore, the spectinomycin resistant isolates in 2011 and 2012 (n = 50) belonged to 32 different STs and were isolated in four of the seven Federal Districts of Russia. The most prevalent STs among the spectinomycin resistant isolates were ST5714 (n = 5), ST807 (n = 4), ST21 (n = 3), and ST5825 (n = 3).
Discussion
==========
The present study describes the antimicrobial resistance in *N. gonorrhoeae* isolates cultured from 2009 to 2012, and molecular epidemiological characteristics (NG-MAST) of *N. gonorrhoeae* isolates, obtained in 2011--2012, in Russia. Previously, only two minor NG-MAST studies examining Russian gonococcal isolates have been performed. Both these studies examined isolates cultured in 2004--2005 \[[@B29],[@B30]\] and, accordingly, no knowledge of the NG-MAST STs of gonococcal strains currently circulating in Russia is available.
High prevalences of resistance to ciprofloxacin and penicillin G were observed. Similar high levels of resistance to these antimicrobials have been described from basically worldwide \[[@B3],[@B6]-[@B12],[@B21]\]. Accordingly, ciprofloxacin and penicillin G should not be recommended for empiric first-line antimicrobial monotherapy of gonorrhoea in Russia or in most other countries worldwide. Nevertheless, interestingly β-lactamase producing *N. gonorrhoeae* strains have remained rare in Russia \[[@B24],[@B25],[@B29]\] as well as in other independent countries of the former Soviet Union, e.g. Belarus \[[@B36]\]. This may indicate that penicillins have not been widely used for treatment of gonorrhoea in many years and/or that no imported β-lactamase producing gonococcal strains have been established and resulted in an endemic spread in Russia during several years. The prevalence of resistance to azithromycin was also high, particularly during the latest years, that is, 16-17% in 2011--2012. However, no isolates with high-level resistance to azithromycin (MIC ≥ 256 mg/L), which have been described from several other countries \[[@B37]-[@B42]\], have yet been identified in Russia. Worryingly, gonococcal isolates with resistance to spectinomycin, which are exceedingly rare internationally \[[@B3],[@B6]-[@B9],[@B12]\], were identified in all the surveyed years and in four of the seven Federal Districts of Russia. In earlier Russian studies \[[@B24],[@B25]\], spectinomycin resistant gonococcal isolates have also been found in all the seven Federal Districts of Russia. In the present study, the spectinomycin resistant isolates (n = 50) belonged to 32 different STs. Accordingly, they did not represent any clonal spread and spectinomycin resistance has been selected in many different gonococcal strains. Spectinomycin remains also available and used for treatment of gonorrhoea in Russia, that is, despite that the level of use has decreased substantially during the last two decades. Fortunately, the resistant isolates had a spectinomycin MIC of maximum 128--256 mg/L and no isolates with high-level resistance to spectinomycin (MIC ≥ 1024 mg/L) \[[@B43]-[@B45]\] have yet been found in Russia. The molecular mechanisms for this low-level resistance to spectinomycin are commonly specific amino acid alterations in the ribosomal protein S5 \[[@B45],[@B46]\], which has been selected by frequent use of spectinomycin. Using the CLSI breakpoints \[[@B32]\], all isolates from 2009 to 2012 were susceptible to ceftriaxone (MIC ≤ 0.25 mg/L). However, using the European EUCAST breakpoint (<http://www.eucast.org>), in total 58 (2.7%) of the isolates during 2009--2012 were resistant to ceftriaxone (MIC \> 0.125 mg/L). Interestingly, the prevalence of these ceftriaxone resistant isolates decreased significantly (*P* \< 0.05), i.e. from 4.0% in 2009 to 0% in 2012. Still no treatment failure of gonorrhoea with ceftriaxone has been verified in Russia, however, isolates with a ceftriaxone MIC of ≤0.25 mg/L have resulted in failures to treat pharyngeal gonorrhoea with ceftriaxone in other countries \[[@B13]-[@B15],[@B17],[@B18]\]. Similar increases in the susceptibility to extended-spectrum cephalosporins such as cefixime and ceftriaxone have recently been reported from the United Kingdom \[[@B47]\], Slovenia \[[@B48]\] and India \[[@B49]\]. The reasons for this remain unknown, however, this might indicate that mostly appropriate treatment with ceftriaxone (in adequately high dose and quality) with or without additional azithromycin (in dual therapy regimens) are used for treatment of gonorrhoea internationally. Accordingly, the use of less potent antimicrobials for treatment might have decreased. It is essential to continuously monitor, using MIC determination, the spread of gonococcal strains with multidrug resistance and resistance to particularly ceftriaxone, spectinomycin and azithromycin and, ideally, also the antimicrobial use/misuse in Russia as well as internationally. Most worryingly, the number of isolates examined in the RU-GASP has substantially decreased the latest years, which is due to both the increased use of genetic detection of *N. gonorrhoeae* for diagnosis of gonorrhoea as well as lack of sufficient financial and political commitments. A national surveillance, including representative gonococcal isolates from all the seven Russian Federal Districts, of gonococcal AMR (ideally also treatment failures) is imperative in Russia. Essential actions aiming to implement the recently published international action/response plans \[[@B21],[@B22]\], and strengthen the culture capacity and surveillance of AMR and test-of-cure in Russia have been initiated. It is also crucial to establish and quality assure regional and national GASPs in the additional independent countries of the former Soviet Union and, for this purpose, national and international support, including political and financial commitment, is essential \[[@B3]-[@B5]\].
In Russia, for first-line empiric treatment of uncomplicated urogenital or extragenital gonorrhoea ceftriaxone (250 mg, intramuscularly), cefixime (400 mg, orally) or spectinomycin (2 g, intramuscularly) is recommended \[[@B27]\]. In practice, also fluoroquinolones, azithromycin, and other cephalosporins can be used in the treatment, and antimicrobials are easily available "over-the-counter", which needs to be abandoned. Based on the present RU-GASP data and resistance emergence and spread worldwide \[[@B3],[@B6]-[@B20],[@B24],[@B25],[@B47],[@B48]\], ceftriaxone should be the only option for first-line empiric antimicrobial monotherapy of gonorrhoea and it should be considered to increase the dose to 500 mg and/or add azithromycin (1--2 g) in the recommended first-line treatment, which is in line with the US CDC \[[@B50]\] and European \[[@B51]\] treatment guidelines. Furthermore, spectinomycin should be the alternative treatment option and only used when ceftriaxone is not available or the patient suffers from a severe β-lactam allergy. However, if pharyngeal gonorrhoea has not been excluded azithromycin is recommended to be added to the spectinomycin regimen. Cefixime, which is less potent compared to ceftriaxone and for which no data exist in Russia, should be excluded from the recommended first-line empiric treatment. This antimicrobial should only be used when injection therapy is refused by patient and should then ideally be used together with azithromycin, which is in line with the recently revised US CDC \[[@B50]\] and European \[[@B51]\] treatment guidelines. The dual antimicrobial regimens will also effectively eradicate any concomitant *Chlamydia trachomatis* infection.
NG-MAST analysis showed a diversified population of *N. gonorrhoeae* in Russia during 2011--2012, with 183 different NG-MAST STs identified among the examined 521 isolates. The high number of unique STs (n = 98) and STs that have not been previously described (n = 122) may be associated with the low number of cultured gonococcal isolates from each surveillance site, suboptimal diagnostics (only random gonorrhoea patients and/or isolates are identified), contact tracing (sexual contacts are not traced) and epidemiological surveillance (sexual transmission chains spreading an identical ST are not identified or followed-up), STs evolved locally in Russia (STs are not previously described because only two minor NG-MAST studies examining isolates from 2004--2005 \[[@B29],[@B30]\] have been previously performed in Russia) or imported from abroad. Nevertheless, some main ST clusters of, e.g., ST807 (n = 41, 7.9% of isolates), ST5714 (n = 30), ST228 (n = 14), ST5042 (n = 11), ST1152 (n = 10), ST5825 (n = 10), and ST5937 (n = 10) were identified, which indicate some larger sexual transmission chains.
Conclusions
===========
In Russia, during 2009--2012 the diversified gonococcal population showed a high resistance to ciprofloxacin, penicillin G and azithromycin. Isolates with low-level resistance to spectinomycin were also identified each year. In general, the MICs of ceftriaxone were relatively high, however, they were decreasing significantly (*P* \< 0.05) from 2009 to 2012. Ceftriaxone should be the only recommended first-line antimicrobial for empiric monotherapy of gonorrhoea in Russia. It should also be considered to increase the dose of ceftriaxone to 500 mg and/or add azithromycin (1--2 g) in the recommended first-line treatment, that is, use a dual antimicrobial therapy regimen \[[@B50],[@B51]\]. Spectinomycin should be the second-line and only used when ceftriaxone is not available or the patient suffers from a severe β-lactam allergy. Regular, quality assured national and international surveillance of AMR (ideally also treatment failures) in *N. gonorrhoeae* is crucial and it is essential to further strengthen the RU-GASP in Russia.
Competing interests
===================
The authors declare that they have no competing interests.
Authors' contributions
======================
AK, AK, NF, and MU designed and initiated the study. NF, RK, DV, VSe, and VSo coordinated and performed all the laboratory analyses. AK, NF and MU analysed and interpreted all the data, and wrote a first draft of the paper. All authors read, commented on and approved the final manuscript.
Pre-publication history
=======================
The pre-publication history for this paper can be accessed here:
<http://www.biomedcentral.com/1471-2334/14/342/prepub>
Acknowledgments
===============
The present work was supported by the Federal target program of Ministry of Health of the Russian Federation \"Prevention and Control of Social Diseases (2007--2012 years)\", subprogram \"Sexually transmitted infections\". We are grateful for the collaboration of the heads and staff of all the involved surveillance sites. Special thanks to M Tarasova, A Belikov, L Kiseleva, T Smirnova, A Amozov, K Baryschkov, N Kirpicheva, M Zemzov, M Glusmin, V Dumchenko, V Temnikov, S Ribalkin, I Shakurov, I Minullin, V Merzlyakov, N Krasnova, V Karyanov, O Sharisheva, N Dolgenitsina, Y Novikov, L Berdizkaya, V Onipchenko, M Arshynsky, G Katzina, E Krug, I Letunova, S Rumjanzev, D Schnaider, I Klemenova, G Jakovenko.
| {
"pile_set_name": "PubMed Central"
} |
1. Introduction {#sec1-sensors-17-02262}
===============
Industrial inspection has to date been an exclusively human task. However, maintaining a group of people detecting surface defects accelerates visual fatigue and misinterpretations. Automatic visual inspection allows for detection of defects by image analysis. This process improves product quality, increases the production rate, avoids errors caused by subjectivity, integrates other systems into the production line, and reduces costs \[[@B1-sensors-17-02262],[@B2-sensors-17-02262],[@B3-sensors-17-02262]\]. Industrial machine vision applications are classified into four quality categories according to the characteristics of the inspected product. These categories refer to dimensional, structural or proper assembly, superficial, and operational features \[[@B4-sensors-17-02262]\]. An estimated 10% of failures in the manufactured parts are caused by superficial defects \[[@B5-sensors-17-02262]\]. The surface quality inspection searches for holes, scratches, cracks, wear, finish, roughness, texture, joints, folds, discontinuities, etc. \[[@B6-sensors-17-02262]\].
Approaches to surface defect detection are summarized in two main research directions. The first approach uses complex lighting systems to generate contrast changes in the surface defect \[[@B7-sensors-17-02262]\]. These methods detect and recognize the defect from the measurement and characterization of pattern deformation on the surface of the object. Zhi and Johansson \[[@B8-sensors-17-02262]\] presented a method for interpreting the deformation of the fringes based on 14 shape features without measuring the depth. Caulier et al. \[[@B9-sensors-17-02262]\] proposed a set of eight features suited to the problem of surface inspection, combined with six characteristics that were developed for the classification of defects in interferometry images. Osten et al. \[[@B10-sensors-17-02262]\] carried out a study of the behavior of the fringes in different defects with the aim of developing inspection systems based on knowledge. Li et al. \[[@B11-sensors-17-02262]\] proposed an automatic inspection scheme for fabric defect detection using smart visual sensors. The system consisted of multiple smart visual sensors working independently. Martínez et al. \[[@B12-sensors-17-02262]\] presented a machine vision system, with an easily configurable hardware-software structure, for surface quality inspection of transparent parts. Neogi et al. \[[@B6-sensors-17-02262]\] presented a detailed review of vision-based steel surface inspection systems. Also, a comparison was made by typology of the defect, extracted features, and detection accuracy between different defect detection systems. In \[[@B13-sensors-17-02262]\] a machine vision system able to achieve fused image acquirement and defect inspection for the textured surface with a suitable efficiency and accuracy was presented. The second direction of investigation focuses on the 3D surface reconstruction of the object, and from the 3D point cloud features are extracted to detect and classify the defect. Pernkof \[[@B14-sensors-17-02262]\] proposed an approach for the 3D reconstruction of raw steel blocks in industrial environments using light sectioning. Ogun et al. \[[@B15-sensors-17-02262]\] proposed a method for identifying conical defects on simple surfaces and measuring its volume automatically from the 3D reconstruction of the part. Chu and Wang \[[@B16-sensors-17-02262]\] presented a automated vision-based system for measure the weld bead size and detect defects. Song et al. \[[@B17-sensors-17-02262]\] proposed a method for fabric defect identification in the textile industry using three-dimensional color phase shift profilometry. A procedure to extract fundamental quality parameters to assess the quality of welds was proposed in \[[@B18-sensors-17-02262]\]. They used a structured light system to obtain 3D reconstruction.
The works that interpret the deformation of the illumination on the defects in the 2D image present some limitations. The 2D image does not provide enough information to enable recognition between different typologies of defects. When there are variations in the scale and geometry of defects, determining the appropriate lighting system is a challenge. The geometry and texture of the object generate brightness and shadows that may be confused with a defect. From the 2D image, it is not possible to perform precise metrology of the defect. In the literature some approaches have been reported to detect and recognize defects from the 3D reconstruction of the piece. However, this is in specific domains of an application where the geometries of surfaces have few variations and only one type of defect is recognized. Because of the above, the methods of describing the defect are simple, which limits them to recognition of a set of larger defects.
The problem of recognizing defects in a 3D surface is similar to the task of recognizing objects in 3D images from partial views. This is not a simple problem, taking into account different variations that the object/surface may suffer, such as translation, rotation, scaling, noise addition, lack of information, and in some cases, non-rigid deformations. For this reason it is necessary to create robust descriptors for different variations suffered by the defect and the surface. In \[[@B19-sensors-17-02262]\] a method for the extraction of characteristics is proposed, called Point Signature, which creates a signature of each point using the intersection with the surface of a sphere centered at the point. A disadvantage of this method lies in the calculation of what the author called signature point, as the intersection of the sphere with the surface is not easy to calculate and in some cases it is necessary to interpolate points, which leads to a reduced accuracy of the signature point and increases the computational cost. In addition, the calculation of the reference vector is sensitive to noise.
Johnson and Hebert \[[@B20-sensors-17-02262]\] proposed Spin Image, which is a point descriptor based on the projection of the adjacent 3D points on a tangential 2D plane. It obtains a 2D image for that point, which becomes the characteristic. A disadvantage of Spin Image is its high dependence on the resolution of the method. Some modifications have been proposed for interpolation, increasing the computational cost. In Reference \[[@B21-sensors-17-02262]\] a method called the Point Feature Histogram (PFH) is proposed to describe the surface of the neighborhood of a point by the difference between the normals. Although an advantage of the method is the simplicity of its calculation, a greater disadvantage lies in the construction of the histogram for each point from the calculation of the differences between normals of an entire region. Full connectivity is created, which generates a smoothing effect for small changes in the region. This also affects the processing times. In Reference \[[@B22-sensors-17-02262]\] a modification to the PFH method improving the processing time was presented, called the Fast Point Feature Histogram (FPFH). It does not create full connectivity for the construction of the histogram, at the cost of reducing the discriminant ability of the method. In Reference \[[@B23-sensors-17-02262]\] the Viewpoint Feature Histogram (VFH) is presented, which is a modification to the FPFH method to describe an object globally by coding its geometry and point of view.
In the recognition of surface defects of objects on a submillimeter scale, the acquisition conditions strongly affect the recognition process. Moreover, it is considered that defects are composed of abrupt changes in the surface, which generate glare and shadows in the 3D reconstruction process and cause low point density in these defective regions. Therefore, the descriptors used to represent the underlying surface must be sufficiently stable to conditions of low point density and noise in the data. The PFH, FPFH and Spin Image, by including the normals in the calculation of the descriptor, are representation methods that highly depend on the correct estimation of the normals and their support radius, which is difficult to ensure in the presence of a defect. The PFH uses the difference between all the normals in the vicinity of a keypoint. As such, in regions with small changes the normals of these points are averaged with the normals of regions with few changes, hence causing a loss of information when describing the defect region.
This paper proposes a descriptor that allows discrimination of different regions within a 3D surface. Our descriptor is based on the estimation of local models on the surface and the difference between the model's normals in a local region.
The contributions to this paper can be summarized as follows:The new local 3D surface descriptor the Model Point Feature Histogram (MPFH) is presented, improving robustness and discriminant capabilities of the PFH method.The MPFH descriptor is constructed from the estimation of models and their weighting to the formation of the underlying region.Furthermore, a methodology for the automatic surface defects inspection is presented. For the detection of the defect, the local 3D descriptor MPFH is used in 3D reconstructed objects. From the projection 2D of the detected 3D primitives are extracted 2D features that allow to classification of the defect.
This paper is organized as follows: [Section 2](#sec2-sensors-17-02262){ref-type="sec"} provides an overview of the methodology. It emphasizes the calculation of structures, estimation of the normals, and the proposed descriptor. [Section 3](#sec3-sensors-17-02262){ref-type="sec"} describes the experimental results and, finally, the conclusions are drawn in [Section 4](#sec4-sensors-17-02262){ref-type="sec"}.
2. System Overview {#sec2-sensors-17-02262}
==================
[Figure 1](#sensors-17-02262-f001){ref-type="fig"} illustrates the scheme of the methodology proposed to recognize surface defects. Our methodology begins with surface 3D reconstruction by projecting structured light patterns ([Figure 1](#sensors-17-02262-f001){ref-type="fig"}a). As a result, a 3D point cloud is obtained which represents the sampled 3D surface ([Figure 1](#sensors-17-02262-f001){ref-type="fig"}b), followed by the calculation of the models or planes that fit each region best using the multiple structures estimation method J-linkage \[[@B24-sensors-17-02262]\] ([Figure 1](#sensors-17-02262-f001){ref-type="fig"}c), and then the normals of the surface models are estimated ([Figure 1](#sensors-17-02262-f001){ref-type="fig"}d). For each surface model, the model contribution weight to the formation of the surface region ([Figure 1](#sensors-17-02262-f001){ref-type="fig"}e) is calculated and from the relative difference between two models of the same region a histogram is generated representing the underlying surface changes ([Figure 1](#sensors-17-02262-f001){ref-type="fig"}f). Each point on the 3D surface is classified into one of five types of primitives, points belonging to hollow, hollow edge, crest, base crest, or flat surfaces ([Figure 1](#sensors-17-02262-f001){ref-type="fig"}g). In this work, it is considered that with this set of primitives it is possible to describe any typology of the defect. For instance, a bump defect will be composed of the base crest, crest, and possibly flat primitives; therefore regions with local geometric changes are detected on the surface. Finally, these regions are extracted 2D features to recognize the defect in a classification stage ([Figure 1](#sensors-17-02262-f001){ref-type="fig"}h).
2.1. Image Acquisition {#sec2dot1-sensors-17-02262}
----------------------
The 3D point cloud is acquired from a 3D reconstruction system. The technique of projecting structured light patterns was used. Specifically, sinusoidal phase shift patterns were projected. A complementary metal-oxide semiconductor (CMOS) camera (Point Grey FL3-U3-88S2C-C, Richmond, BC, Canada) was used to capture the images and a digital light projector (DLP) was used to synchronously project the sinusoidal patterns on the object \[[@B25-sensors-17-02262]\]. The acquisition scheme is shown in [Figure 2](#sensors-17-02262-f002){ref-type="fig"}. The specifications of the camera and the projector are shown in [Table 1](#sensors-17-02262-t001){ref-type="table"}. Our 3D reconstruction system had an RMS measurement error of 0.053 mm in *x*, 0.039 mm in *y* and 0.10 mm in *z*, for a measurement area of 60 mm × 60 mm.
2.2. Calculation of Structures {#sec2dot2-sensors-17-02262}
------------------------------
Superficial defects generate deformations on the object's surface; this causes the normals and curvature in a defective region to change with respect to the neighborhood region. Therefore, if the representation method includes the normal or curvature in the construction of the descriptor, its adequate estimation influences the description of the region. Given a point cloud of the sample 3D surface called *S*, let $P_{i} \in S$ be point on the surface; the *k* closest adjacent points around $P_{i}$ are denoted by $R_{k}$. The problem of choosing *k* is known as the correct scale factor and it affects the estimation of the normal vectors. The parameter *k* is also related to the minimum defect size that our system is capable of detecting. To estimate the normal to a point $P_{i}$ on the surface, a least squares local plane is fitted to $R_{k}$ and its normal vector becomes normal to the point $P_{i}$ \[[@B26-sensors-17-02262]\]. However, the normals of the planes that are adjusted to points $R_{k}$, with very big *k* or belonging to edges or corners, undergo a smoothing effect leading to misdirection, which influences the description of that region. The normals shown in red in [Figure 3](#sensors-17-02262-f003){ref-type="fig"}a present these effects.
A structure is defined as a set of points that adjusts to a plane. We call these structures "models" as they represent the surface changes in a simple manner. The region $R_{k}$ around the corner is composed of three structures or models with different normal vectors ([Figure 3](#sensors-17-02262-f003){ref-type="fig"}a). Thus, our methodology proposes using a method to estimate the multiple structures $M_{j}$ in a region $R_{k}$ and then calculate their normal vectors. A model is denoted by $M_{j} = \left\{ p_{i} \right\}_{i = 1:\tau,\tau \leq k}$. Moreover, it associates the normal vector of the model $M_{j}$ to each point $P_{i} \in M_{j}$, which makes it not strictly necessary to calculate the normals for each point of the surface, but rather ensures that the joining of the points that belong to the generated models is equal to the surface *S*. This is shown in Equation ([1](#FD1-sensors-17-02262){ref-type="disp-formula"}), where *n* represents the total number of models found on the surface. [Figure 3](#sensors-17-02262-f003){ref-type="fig"}b shows the normal vectors on the surface using the proposed methodology, where is possible observe the correct directing of the normals for points belonging to edges and corners. The number of $M_{j}$ models in a region $R_{k}$ depends on its topology, so that if the region has few variations, the number of found models is low compared to a region presenting many topological changes. [Figure 3](#sensors-17-02262-f003){ref-type="fig"}c represents this concept, where the red dots indicate the center of each estimated model on a defective surface. $$S = \bigcup\limits_{j = 1}^{n}M_{j}$$
The nature of the problem described above is comprised of a set of data that belongs to multiple structures and outliers. The challenge is to estimate the different structures in the presence of noise and geometric changes in the point cloud.
We use the approach proposed in \[[@B24-sensors-17-02262]\], the J-linkage method, to estimate the multiple structures on a local region of points $R_{k}$, which is composed of the *k*-nearest neighbors of a point $P_{i}$. As in Random Sample Consensus (RANSAC) \[[@B27-sensors-17-02262]\], J-linkage generates a random set of *M* hypothesis models called a minimum sampling set. Then the consensus set (CS) of each model is determined as the set of points that is at a smaller distance than a threshold $\epsilon$ for this particular model. From the CS of the *M* hypothesis models, a $1 \times M$ vector is built, representing the set of models to which $P_{i}$ belongs, where 1 in the column *j* of the vector represents the belonging of that point to the $M_{j}$ model and 0 represents that the point does not belonging to the model. This vector is called the preference set (PS) of a point $P_{i}$, which becomes a conceptual representation of this point, so that points within the same structure will have similar representations. They will be grouped together as in \[[@B24-sensors-17-02262]\], in a conceptual space $\left\{ {0,1} \right\}^{M}$. To measure this similarity, the Jaccard distance is used (Equation ([2](#FD2-sensors-17-02262){ref-type="disp-formula"})). *A* and *B* are two PSs different to ⊘ and $d_{J}$ takes on values between 0 and 1, for identical and different sets, consecutively. $$d_{J}\left( {A,B} \right) = \frac{{A \cup B} - {A \cap B}}{A \cup B}$$
J-linkage has been used to determine the multiple flat structures in a scene for segmentation purposes from point clouds \[[@B28-sensors-17-02262],[@B29-sensors-17-02262]\], where the difference between the functions of the flat structures is high and the amount of structures is low compared to the density of points. This difference does not arise with point clouds belonging to 3D reconstructions of surface defects, where noise can easily be an inlier of structures and generate a very large number of structures. Therefore, we first generated local regions $R_{k}$ throughout the point cloud, and then the models of each $R_{k}$ region were determined using J-linkage. As a result, the estimated models represent local changes in the surface.
A model of a plane can be represented from up to three points. However, we found experimentally that accepting models with fewer than five points increases false models due the noisy data. Increasing the number of points results in that in defective regions, where the point density is low, the models are rejected.
2.3. Normal Estimation {#sec2dot3-sensors-17-02262}
----------------------
After obtaining the *M* models belonging to a neighborhood $R_{k}$ of $P_{i}$, the normal vector to the model is determined from the analysis of the eigenvalues $\lambda$ and eigenvectors $\overset{\rightarrow}{v}$ of the covariance matrix *C*, as described in Equation ([3](#FD3-sensors-17-02262){ref-type="disp-formula"}). $$C = \frac{1}{\tau}\sum_{i = 1}^{\tau}\left( P_{i} - \overline{P} \right) \cdot \left( P_{i} - \overline{P} \right)^{T}$$
The eigenvector $\overset{\rightarrow}{v_{0}}$ of the smallest eigenvalue $\lambda_{0}$ corresponds to the approximation of the normal $\overset{\rightarrow}{N}$, if the condition $\lambda_{2} \geq \lambda_{1} \geq \lambda_{0} \geq 0$ is fulfilled \[[@B30-sensors-17-02262]\].
Using models to determine the normal vectors to a surface of a region solves to a large extent the problem of the right scale factor when calculating the normals (see [Section 3.2](#sec3dot2-sensors-17-02262){ref-type="sec"}). The problem of the scale factor is associated with the selected support radii to estimate the normals. If the scale factor (*k* or *r* search) is big, the region will undergo a smoothing effect and the normal vector will not capture its details. Algorithm 1 shows the pseudo-code for determining the models and their associated normal vectors and [Figure 4](#sensors-17-02262-f004){ref-type="fig"} shows the result of the estimation of the normal vectors on a surface sampled containing a hole defect.
Algorithm 1:
Normals estimation
2.4. Model Points Feature Histogram (MPFH) {#sec2dot4-sensors-17-02262}
------------------------------------------
After obtaining the models and its normal vectors, it is necessary to calculate a representation for each point $P_{i}$ of the surface sampled *S*. For this, we propose calculating a histogram of characteristics similar to that proposed in \[[@B31-sensors-17-02262]\].
The calculated feature is the relative difference between the normal vectors of each two models belonging to the region $R_{k}$ in a Darboux coordinate system. This coordinate system generates a 4D feature that is invariant to the translation and rotation, which has the advantage of reducing the number of parameters from 12 to 4 parameters. This reduction is composed of the coordinates $\mathbf{P}\left( {x,y,z} \right)$ and normal vectors $\mathbf{N}\left( {x,y,z} \right)$ of the two points.
The pair $\left( {\mathbf{P},\mathbf{N}} \right)$ is known as surflet. The origin will be $\mathbf{P}_{1}$, if Equation ([4](#FD4-sensors-17-02262){ref-type="disp-formula"}) is met, otherwise it will be $\mathbf{P}_{2}$ and the indexes 1 y 2 in the equations must be exchanged. This is to achieve homogeneity in the choice of the origin of the coordinate system.
Let · denote the scalar product between two vectors and × the cross product of the two vectors. $$\left| {\mathbf{N}_{1} \cdot \left( {\mathbf{P}_{2} - \mathbf{P}_{1}} \right)} \right| \leq \left| {\mathbf{N}_{2} \cdot \left( {\mathbf{P}_{2} - \mathbf{P}_{1}} \right)} \right|$$
The axes of the coordinate system are expressed in Equations ([5](#FD5-sensors-17-02262){ref-type="disp-formula"})--([7](#FD7-sensors-17-02262){ref-type="disp-formula"}), and the features indicating the differences between two surflets are described in Equations ([8](#FD8-sensors-17-02262){ref-type="disp-formula"})--([11](#FD11-sensors-17-02262){ref-type="disp-formula"}), \[[@B31-sensors-17-02262]\]. [Figure 5](#sensors-17-02262-f005){ref-type="fig"} shows the relationship between the coordinate systems and the features obtained.
u
=
N
1
v
=
P
2
−
P
1
×
u
P
2
−
P
1
×
u
w
=
u
×
v
θ
=
arctan
w
·
N
1
,
u
·
N
2
α
=
v
·
N
2
ϕ
=
u
·
P
2
−
P
1
P
2
−
P
1
d
=
P
2
−
P
1
To calculate the descriptor of a point $P_{i}$, the closest points $P_{k}$ are searched in a neighborhood $k_{2}$ which we call region $R_{k2}$. Then, the models $M_{j}$ and their centroids to which belong to the region $R_{k2}$ are determined. Hence, a vector with the set of models and another with the level of participation $\rho$ of the model in the representation of the region $R_{k2}$ are computed. $\rho$ represents the ratio between the number of points that form a model $M_{j} \in R_{k}$ and the totality of the points $M_{j}$, described in Equation ([12](#FD12-sensors-17-02262){ref-type="disp-formula"}). $\rho$ ranges from $\left( 0,1 \right\rbrack$; $\rho = 0$ is not present, because $R_{k2} \cap M_{j} \neq 0$. [Figure 5](#sensors-17-02262-f005){ref-type="fig"} illustrates the $M_{j}$ models belonging to the $R_{k2}$ region and the difference between them. $$\rho_{i} = \frac{R_{k2} \cap M_{i}}{M_{i}}$$
The PFH method \[[@B21-sensors-17-02262]\] uses the histogram of characteristics proposed in \[[@B31-sensors-17-02262]\], where the features $\left( {\theta,\alpha,\phi,d} \right)$ are found between each pair of surflets belonging to the region $R_{k2}$. It generates a set of $k_{2}\left( \frac{k_{2} - 1}{2} \right)$ quadruplets of features with complete connectivity without weighting. As a result, it averages the differences between surflets and smoothes local details of the region. On the other hand, our method is based on the relationship between the normal vectors of the models belonging to the region $R_{k}$. In addition, our proposal adds the parameter $\rho$, which represents the participation of each model to the formation of the region. This speeds up the calculation of the representation and leads to a better capture of the geometric information of the underlying 3D surface. This also reduces the computational complexity.
To calculate the MPFH representation of each point $P_{i}$, a histogram is generated from the quadruplets of the region $R_{k2}$ using the first three Darboux features and adding the weighting of the model to the formation of the region $\left( {\theta,\alpha,\phi,\rho} \right)$. Therefore, a histogram *H* with 4 × *b* bins is generated, where the first *b* bins corresponds to the quantization of the feature $\phi$, the following to $\alpha$, then to $\theta$, and finally the last corresponds to $\rho$. [Figure 6](#sensors-17-02262-f006){ref-type="fig"}a shows the MPFH representation for a point belonging to a flat region and [Figure 6](#sensors-17-02262-f006){ref-type="fig"}b for a point belonging to a hole-like defect, given *b* equals 11 as in the FPFH method.
2.5. Primitives 2D Projection {#sec2dot5-sensors-17-02262}
-----------------------------
The identification of primitives allows for detection of defects on surface but not their classification. Our proposal is to project each connected component of primitives on a plane forming a 2D image. Then, 2D geometrical features are extracted for recognition of the defect. [Figure 7](#sensors-17-02262-f007){ref-type="fig"} shows the methodological scheme proposed.
First, the primitives are clustered in connected components, using Euclidean distance. The threshold distance $d_{t}$ is selected according to resolution of the point cloud. Then, for each connected component $C_{i}$, the centroid $O_{ci}$ is calculated, and with the flat primitive $P_{i}^{f}$ nearest to this, the projection plane $\mathsf{\Pi}_{i}$ is set. All primitives $P_{c_{i}}$ are projected orthogonally to the plane $\mathsf{\Pi}_{i}$ as described in Equations ([13](#FD13-sensors-17-02262){ref-type="disp-formula"})--([15](#FD15-sensors-17-02262){ref-type="disp-formula"}). [Figure 8](#sensors-17-02262-f008){ref-type="fig"}a,b illustrates the procedure. $$\mathbf{q} = \left| {\mathbf{O}_{\mathbf{ci}} - \mathbf{P}_{\mathbf{i}}^{\mathbf{f}}} \right|$$ $$\mathbf{D} = \left| {\mathbf{q} \cdot N_{i}^{f}} \right|$$ $$\mathbf{P}_{\mathbf{proj}} = \mathbf{P}_{c_{i}} - \mathbf{D} \ast \mathbf{N}_{i}^{f}$$
$\mathbf{N}_{i}^{f}$ is the normal vector associated to $P_{i}^{f}$.
A affine transformation is applied to the plane $\mathsf{\Pi}_{i}$, so that it is parallel to the $z = 0$ plane with normal vector $\mathbf{n}_{z = 0}$($0,0,1$). Equations ([16](#FD16-sensors-17-02262){ref-type="disp-formula"})--([18](#FD18-sensors-17-02262){ref-type="disp-formula"}) show the calculate of axis and angle rotation, and Equation ([19](#FD19-sensors-17-02262){ref-type="disp-formula"}) obtains the matrix Euler-Rodriguez formula. $$\mathbf{Rot}_{\mathbf{Axis}} = \mathbf{N}_{i}^{f} \times \mathbf{n}_{z = 0}$$ $$\mathbf{u}_{ax} = \frac{\mathbf{Rot}_{\mathbf{Axis}}}{\parallel \mathbf{Rot}_{\mathbf{Axis}} \parallel}$$ $$\mathbf{Rot}_{\mathbf{Angle}} = \mathbf{\theta}_{g} = \mathbf{N}_{i}^{f} \cdot \mathbf{n}_{z = 0}$$ $$\mathbf{R} = \cos\theta_{g}\mathbf{I} + \sin\theta_{g}\left\lbrack \mathbf{u}_{ax} \right\rbrack_{\times} + \left( {1 - \cos\theta_{g}} \right)\mathbf{u}_{ax} \otimes \mathbf{u}_{ax}$$
$\left\lbrack \mathbf{u}_{ax} \right\rbrack_{\times}$ denotes the cross-product matrix of $\mathbf{u}_{ax}$, ⊗ is the tensor product, and $\mathbf{I}$ is the identity matrix. [Figure 6](#sensors-17-02262-f006){ref-type="fig"}b illustrates the result of this procedure.
The 2D image is formed by the conversion of the plane $\mathsf{\Pi}_{i}$ in mm to pixels. For that, we calculate the factor $s_{x} = \frac{1}{d_{x}}$$\frac{pix}{mm}$, where $d_{x}$ is the closest distance between $O_{ci}$ projected at $\mathsf{\Pi}_{i_{z = 0}}$ and other point. Then, the bounding box of $\mathsf{\Pi}_{i_{z = 0}}$ is calculated with the goal of translating it the coordinate origin. [Figure 8](#sensors-17-02262-f008){ref-type="fig"}c shows the 2D image result and [Figure 8](#sensors-17-02262-f008){ref-type="fig"}d,e shows 2D projections of different typologies of defects on the objects of experimentation.
According to \[[@B32-sensors-17-02262]\], the characteristics of a surface imperfection are represented in the length, width, depth, height, and area of imperfection. Hence, it is possible to combine 3D and 2D descriptors to recognize an imperfection. As explained in [Section 2.4](#sec2dot4-sensors-17-02262){ref-type="sec"}, the 3D information of the curvature of the surface is embedded in the local 3D descriptor MPFH. However, it is necessary to obtain 2D geometric information that allows recognition of the defect. We proposed to group each set of primitives from its 2D projection and then extract a set of characteristics from each region of primitives belonging to the defect. Finally, in a classification stage the defect region is recognized.
The set of characteristics was chosen from a process of selection of characteristics using a Fisher discriminant \[[@B33-sensors-17-02262]\] through the Balu toolbox Matlab \[[@B34-sensors-17-02262]\]. The most relevant features for each type of primitives were the Hu \[H1, H2, H7\] moments, the Fourier descriptors \[F3, F4, F5, F7 and F11\], the eccentricity \[Exc\], and the number Euler \[En\].
3. Results {#sec3-sensors-17-02262}
==========
In this section, we developed two types of evaluation. First, we proved the discriminative power of the proposed 3D local descriptor compared with five methods of description commonly used. Also, invariance of descriptors to noise and support radius was addressed. Second, we tested the performance of the proposed defects recognition methodology in a database with different objects and defect typology.
3.1. Database {#sec3dot1-sensors-17-02262}
-------------
According to our review, to date there have not been any public databases of 3D images with surface defects. Therefore, the database used in these experiments was constructed from 3D reconstructions of defects in welding, artificial teeth; indentations in materials, ceramics, models of artificial defects, and simulations. [Figure 9](#sensors-17-02262-f009){ref-type="fig"} shows an example of the objects used in the database. Our database contains 480 three-dimensional images with more than 2000 regions with surface defects. There are four types labeled in the database: holes, bumps, cracks and without defect.
In order to understand how difficult it is to detect and recognize a micro-metric defect on an object, [Figure 10](#sensors-17-02262-f010){ref-type="fig"}a--c shows an indentation with a test of Vickers hardness on a sample of aluminum. In [Figure 10](#sensors-17-02262-f010){ref-type="fig"}d--f a material detachment defect in an artificial tooth can be observed.
3.2. Evaluation of the Estimation of the Normals from Multiple Structures {#sec3dot2-sensors-17-02262}
-------------------------------------------------------------------------
In this section, we evaluate how the computation of the normal vector on a surface is affected when simple or multiple structure estimation methodologies are used. In [Figure 11](#sensors-17-02262-f011){ref-type="fig"}, the result of evaluating the change of a normal at a critical point, where abrupt geometrical changes take place, is shown. The number of k-neighbors, which supports the parameter to estimate the models, was varied between 15 and 250 points. As [Figure 11](#sensors-17-02262-f011){ref-type="fig"} shows, the scaling factor does not significantly influence the normal vectors calculated by estimating multiple structures. In contrast to using a unique structure, the normal vector in that region changes with the variation of support radius. This means that through the proposed methodology, better-oriented normal vectors at critical points such as edges and corners are obtained. Consequently, this will increase the discriminating capabilities of our descriptor.
3.3. Robustness to Noise in the Estimation of the Normal Vectors {#sec3dot3-sensors-17-02262}
----------------------------------------------------------------
In order to evaluate the robustness to noise in the estimation of models for calculating the normal vectors, in [Table 2](#sensors-17-02262-t002){ref-type="table"} we show a comparison of the results of calculating the normal vectors using the adjustment of multiple- structure J-linkage versus the adjustment of a single structure on a cube. In a simple structure like a cube that has three types of regions (planes, edges, and corners), normal estimation methods based on a single structure fail at edges and corners. The defects, being more complex structures, require greater precision in the estimation of the normal, which can be achieved if the estimation of normal multiples is used. The regions chosen for the evaluation were: points in planes, near edges, and on corners. Gaussian noise with a standard deviation of 5% is also added. The calculation of the normal vectors is compared with the theoretical normal vector. The spaces in the "One Structure" column are due to the fact that, regardless of the topological complexity of a region, the methods based on the estimation of unique structures will always find only one normal vector. However, for example for a cube-type region, there are three theoretical normal vectors which match the results using multiple structures. The mean square error between the calculation of the normal vectors using the estimation of multiple structures with respect to the normal theoretical vector was 0.00854, whereas with a single structure this value was 0.01117. In summary, these results show that is more stable to calculate the normal vectors using the estimation of multiple structures versus single structure, even with the presence of noise.
3.4. Evaluation of the Discriminating Properties {#sec3dot4-sensors-17-02262}
------------------------------------------------
In order to evaluate the discriminating capabilities of the MPFH method, it is compared with the PFH and FPFH descriptor with different *k* neighborhood support. The measure of similarity used for the histogram was the intersection between the descriptors for seven different regions: the point in a plane, cylinder, sphere, a point on the vertices of a triangular pyramid, square pyramid, pentagonal pyramid, and hexagonal pyramid. When the number of sides of the polygon of the base of the pyramid increases, the curvature of the vertex is close to the curvature of a sphere. These regions were chosen because they present a small change between them in the curvature.
[Figure 12](#sensors-17-02262-f012){ref-type="fig"} presents a similarity matrix. The intensity level represents the quantification of similarity between objects row and column. The rows correspond to the similarity measure for the PFH, FPFH, and MPFH descriptors consecutively, and the columns represent the similarity with different $k_{1}$ and $k_{2}$ neighborhood support. An ideal similarity matrix for a classification process would have a diagonal with black and the rest white.
It can be seen in the gray levels from [Figure 12](#sensors-17-02262-f012){ref-type="fig"}g--i that there is a lower probability of confusion of the MPFH method between different classes with respect to the PFH ([Figure 12](#sensors-17-02262-f012){ref-type="fig"}a--c) and FPFH ([Figure 12](#sensors-17-02262-f012){ref-type="fig"}d--f). These results suggest that the description made by the MPFH method facilitates a correct classification of the underlying region.
3.5. Comparison with Some 3D Local Descriptors {#sec3dot5-sensors-17-02262}
----------------------------------------------
In this section, the descriptor MPFH proposed is compared with some of the descriptors more often used in the literature like PFH, FPFH, spin image, radius-based surface (RSD), and signature of histograms of orientations (SHOT). From the database we carefully selected 2800 points belonging to different 3D surfaces regions. These points were categorized into five classes: hollow, crest, edge hollow, base crest and planar surfaces. Descriptors were evaluated in a classification task varying Gaussian noise, adding 0%, 5% and 10%.
[Figure 13](#sensors-17-02262-f013){ref-type="fig"}a compares the accuracy obtained between the descriptors. It can be seen that the accuracy of the descriptor MPFH is higher, even with the addition of noise, and although accuracy decreases with increasing the percentage of noise, it provides less of a slope than the PFH, FPFH, spin image and the radius-based surface descriptor (RSD).
In [Figure 13](#sensors-17-02262-f013){ref-type="fig"}b, the classification results varying the $r_{2}$ support radius for the calculation of the descriptors are displayed. We can also observe that while decreasing the classification accuracy with the variation in radius $r_{2}$ support, our method still has the greatest accuracy between descriptors. In [Figure 13](#sensors-17-02262-f013){ref-type="fig"}b, the qualitative results taking the 3D sampled surface of the artificial tooth, welding defect, and simulated defects models with added noise as test data are shown.
3.6. Performance Evaluation for Defects Recognition {#sec3dot6-sensors-17-02262}
---------------------------------------------------
This section evaluates the full methodology of surface defect recognition proposed in this paper. The set of training and test images consisted of 2160 regions labeled as holes, bumps, cracks and without defect.
In the detection stage, the points of the surface are classified into five primitives, the points belonging to a hollow (red), hollow edge (yellow), crest (magenta), base crest (blue), and flat (green) surfaces. The capacity of the local descriptor to represent and classify the regions of the surface can be seen in [Figure 14](#sensors-17-02262-f014){ref-type="fig"}, . However, it is also observed that misclassified primitives appear, which will affect the classification process.
The recognition stage was evaluated with the classification techniques: k-nearest neighbors (knn), linear discriminant analysis (LDA), quadratic discriminant analysis (QDA), multilayer perceptron, support vector machine (SVM) with radial basis function kernel (rbf), SVM with polynomial kernel function (poly), and SVM with a quadratic kernel (quad). In the evaluation process, we used 10-fold cross-validation. [Table 3](#sensors-17-02262-t003){ref-type="table"} shows the performance of each classifier. The performance was measured using the accuracy factor (Equation ([20](#FD20-sensors-17-02262){ref-type="disp-formula"})). $$AverageAccuracy = \frac{\sum_{i = 1}^{l}\frac{tp_{i} + tn_{i}}{tp_{i} + fn_{i} + fp_{i} + tn_{i}}}{l}$$
The parameters ($tp_{i}$) are the true positive for class $C_{i}$, the elements ($fp_{i}$) are false positives, ($fn_{i}$) are false negatives, and finally, the the parameters ($tn_{i}$) are the true negatives. The factor *l* is the class number.
[Figure 15](#sensors-17-02262-f015){ref-type="fig"} shows the result of the classification for different images in the database. It can be seen from images that the recognition stage is able to correctly classify the defect. However, can be seen that in defective regions close to the edges of the object, our method erroneously classifies the region as without defect ([Figure 15](#sensors-17-02262-f015){ref-type="fig"}d). The misclassification could be attributed to the very low density of points in this regions, which causes a misclassification in primitives, affecting the recognition.
The average processing time of the automated inspection system is composed of the acquisition step, which is delayed 0.3 s, the processing and description step with 1.1 s for a region with 1817 points and three defects, and the classification step with 0.2 s. According to the above, the automated inspection system could detect defects on a surface in 1.6 s.
4. Conclusions {#sec4-sensors-17-02262}
==============
In this paper, we have presented a new local 3D surface descriptor that improves the robustness and discriminating capabilities of the PFH method. We have demonstrated its applicability for surface quality inspection in the detection of defects on different objects. The proposed MPFH method was evaluated and compared to some of the most used state-of-the-art descriptors under a classification task. The obtained results show that our method has higher discriminating capabilities, because we tried to capture the geometric information of the underlying 3D surface more precisely by estimating the normals from the models that are adjusted to the surface and including these in the construction of the descriptor. From the identification of primitives, we propose a method of description that projects each connected component of primitives on a plane forming the 2D image. Then, 2D geometric features are extracted to recognize the defect. With this method, three types of defects (holes, bumps, and cracks), with an accuracy of 94.17%, are recognized using an SVM. In the MPFH descriptor calculation, the models that best fit the surface are estimated through an estimation technique of multiple structures or models. In this paper the models are flat surfaces, however, we believe that if models can be fitted to polynomial surfaces or splines, this would help to estimate the normal vectors more correctly and the local region would be more accurately represented.
We give particular thanks to the research groups Automatic, Electronic, and Computer Sciences at ITM, GIDIA at the Universidad Nacional de Colombia at Medellin, and GRIMA at the Pontificia Universidad Católica de Chile.
Carlos A. Madrigal contributed extensively to the entire work, in particular to the concept and implementation of the MPFH descriptor and the writing of the paper. John W. Branch and Alejandro Restrepo contributed to the work as scientific directors, designing the methodology of recognition, analyzing the results, and preparing the paper. The contribution of Domingo Mery focused on the concept of the descriptor MPFH, the analysis of results and feedback on the manuscript.
The authors declare no conflict of interest.
{#sensors-17-02262-f001}
{#sensors-17-02262-f002}
{#sensors-17-02262-f003}
{#sensors-17-02262-f004}
{#sensors-17-02262-f005}
{#sensors-17-02262-f006}
{#sensors-17-02262-f007}
{#sensors-17-02262-f008}
{#sensors-17-02262-f009}
{#sensors-17-02262-f010}
{#sensors-17-02262-f011}
######
Measurement of distances between feature vectors of points in different regions and with different $k_{1}$ and $k_{2}$ neighborhood support. (**a**) PFH con $k_{1} = 15$, $k_{2} = 45$; (**b**) PFH $k_{1} = 30$, $k_{2} = 60$; (**c**) PFH $k_{1} = 45$, $k_{2} = 75$; (**d**) FPFH $k_{1} = 15$, $k_{2} = 45$; (**e**) FPFH $k_{1} = 30$, $k_{2} = 60$; (**f**) FPFH $k_{1} = 45$, $k_{2} = 75$; (**g**) MPFH $k_{1} = 15$, $k_{2} = 45$; (**g**) MPFH $k_{1} = 30$, $k_{2} = 60$ and (**i**) MPFH $k_{1} = 45$, $k_{2} = 75$.
{#sensors-17-02262-f013}
{#sensors-17-02262-f014}
{#sensors-17-02262-f015}
sensors-17-02262-t001_Table 1
######
Camera and digital light projector specifications.
Device Specifications
------------------------------- ---------------------------------------------
Camera Point Grey FL3-U3-88S2C-C
Sensor size 1.55 $\mathsf{\mu}$m × 1.55 $\mathsf{\mu}$m
Image resolution $4096 \times 2160$ pixels
Lenses Edmund Optics *f*8.5 mm
Digital light projector (DLP) DLP LightCrafter 4500
DLP resolution $912 \times 1140$ pixels
Synchronization circuit freescale FRDM-K20D50M
Software Visual C++ + OpenCV3.0 + PCL1.8
sensors-17-02262-t002_Table 2
######
Normal vectors in corners, edges, and planes of a cube, using single (standard methods) and multiple structures (method used).
Region Theoretical Normal Vector One-Structure Theoretical Multiple-Structure J-linkage
----------------- --------------------------- --------------------------- ------------------------------
Corner 0.0/−1.0/0.0 −0.5350/−0.5350/0.6538 0.0/−1.0/0.0
−1.0/0.0/0.0 --- −1.0/0.0/0.0
0.0/0.0/1.0 --- 0.0/0.0/1.0
Corner 5% noise 0.0/−1.0/0.0 −0.5438/−0.5888/0.5979 −0.0010/−0.9997/0.0215
−1.0/0.0/0.0 --- −0.9993/0.0234/0.02866
0.0/0.0/1.0 --- 0.0711/0.0478/0.9963
Edge 0.0/0.0/1.0 0.0/−0.7071/0.7071 0.0/0.0/1.0
0.0/−1.0/0.0 --- 0.0/−1.0/0.0
Edge 5% noise 0.0/0.0/1.0 0.0027/−0.5286/0.8488 0.0765/−0.0416/0.9962
0.0/−1.0/0.0 --- 0.0159/−0.9953/−0.0952
Plane $0.0/0.0/1.0$ $0.0/0.0/1.0$ $0.0/0.0/1.0$
Plane 5% noise 0.0/0.0/1.0 0.0191/−0.0116/0.9998 0.01271/−0.0114/0.9998
sensors-17-02262-t003_Table 3
######
Results of classification of surface defects. LDA: linear discriminant analysis; knn: k-nearest neighbors; QDA: quadratic discriminant analysis; SVM (rbf): SVM with radial basis function kernel; SVM (poly): SVM with polynomial kernel function; SVM (quad): SVM with a quadratic kernel.
Classifier knn (k = 10) LDA QDA Multilayer Perceptron SVM (rbf) SVM (poly) SVM (quad)
------------ -------------- -------- -------- ----------------------- ----------- ------------ ------------
Accuracy 90.09% 86.22% 66.89% 93.95% 94.17% 87.72% 93.05%
| {
"pile_set_name": "PubMed Central"
} |
(J Am Heart Assoc. 2015;4:e002626 doi: [10.1161/JAHA.115.002626](10.1161/JAHA.115.002626))
Introduction {#jah31237-sec-0004}
============
Cardiac resynchronization therapy (CRT) is an effective treatment for patients with systolic heart failure and electrical dyssynchrony, resulting in improvements in both symptoms and mortality.[1](#jah31237-bib-0001){ref-type="ref"} Depending on the end point assessed, between 30% and 40% of patients fail to improve with CRT.[2](#jah31237-bib-0002){ref-type="ref"} Metrics created to improve patient selection and to predict response have often appeared encouraging in small single‐center studies but have lacked reproducibility when extrapolated to multicenter trials.[3](#jah31237-bib-0003){ref-type="ref"} Consequently, interest has been increasing in both the pathophysiology of dyssynchronous heart failure, in an attempt to understand the mechanical sequelae of electrical dyssynchrony, and new methods of ventricular stimulation such as endocardial and multipoint pacing to improve CRT response.[4](#jah31237-bib-0004){ref-type="ref"}, [5](#jah31237-bib-0005){ref-type="ref"}, [6](#jah31237-bib-0006){ref-type="ref"}, [7](#jah31237-bib-0007){ref-type="ref"}, [8](#jah31237-bib-0008){ref-type="ref"} An area of research is the effect of impaired electrical activation and CRT on coronary hemodynamics and physiology; recent data from animal models have indicated the importance of blood flow in CRT response.[9](#jah31237-bib-0009){ref-type="ref"}
The coronary vasculature is unique in the human body in that the majority of flow occurs during diastole. Using advanced invasive techniques, it has been possible to demonstrate that flow is mediated principally by the forward propulsion of blood through the coronary tree in systole (a dominant forward compression wave \[FCW\]) and by a relatively larger backward expansion wave (BEW; suction wave) generated by relaxation of the ventricle in diastole.[10](#jah31237-bib-0010){ref-type="ref"} Current evidence shows that the amplitude and wavelengths of these waves are affected by CRT when measured in the left main coronary artery[11](#jah31237-bib-0011){ref-type="ref"}; however, relative workload and myocardial stress are not homogenous in the dyssynchronous left ventricle. Ventricular activation with left bundle branch block results initially in septal contraction with delayed activation and contraction of the lateral wall. This early contraction occurs prior to development of tension in the lateral wall, resulting in reduced septal work compared with normal contraction. Conversely, the lateral wall contracts against a pressure‐loaded ventricle, causing an increase in lateral wall work compared with synchronous activation.[12](#jah31237-bib-0012){ref-type="ref"}, [13](#jah31237-bib-0013){ref-type="ref"}
Conventional biventricular pacing (BIVCS) delivered epicardially from the coronary sinus can increase left anterior descending artery (LAD) coronary flow, and acute changes in LAD flow may predict response to CRT, but little is known about the effects of BIVCS on the wave energy that determines coronary flow in the left‐sided coronary system.[14](#jah31237-bib-0014){ref-type="ref"}, [15](#jah31237-bib-0015){ref-type="ref"} Similarly, although data are increasing regarding the beneficial acute effects of left ventricular (LV) endocardial pacing on cardiac work and acute contractility, no data describe the effects of LV endocardial pacing on coronary blood flow.[16](#jah31237-bib-0016){ref-type="ref"}, [17](#jah31237-bib-0017){ref-type="ref"} Given the heterogeneity of regional work in the dyssynchronous ventricle, a more detailed examination of the individual epicardial arteries may give insights into the regional effects of myocardial contraction on the different constituents of coronary blood flow during both epicardial and endocardial LV pacing.
We sought to describe the effects of both standard CRT via LV epicardial pacing from the coronary sinus and biventricular endocardial pacing (BIVEN) on coronary flow. By applying wave intensity analysis to simultaneously obtained coronary pressure--Doppler flow data, we sought to describe the effect of pacing from different sites (epicardial and endocardial) on coronary physiology in both left‐sided epicardial arteries.
Methods {#jah31237-sec-0005}
=======
The study received approval from the local research ethics committee and was conducted in accordance with the Declaration of Helsinki. All participants provided written informed consent. We obtained simultaneous electrophysiological and hemodynamic measurements from 11 patients with a previously implanted standard CRT device with an epicardial LV lead in the lateral/posterolateral wall. Arterial access was gained in both the femoral and radial arteries. A 0.014‐cm Doppler wire (ComboWire model 9500; Volcano Corp) was advanced to the proximal LAD and then to a central location within the proximal LAD at which a high signal‐to‐noise ratio was obtained. A PrimeWire pressure guide wire (Volcano Corp) was placed in the LV cavity to allow measurement of acute contractility by the maximal rate of rise of LV pressure (dP/dt~max~).[18](#jah31237-bib-0018){ref-type="ref"}
To study the effects of LV endocardial pacing on coronary flow, 5 of the 11 patients agreed to undergo a more comprehensive study in which a roving endocardial pacing catheter was placed in the left ventricle to perform LV endocardial pacing. Several endocardial positions (mean of 4 \[2 for each artery\]) were tested in the 5 patients, and the site with the highest average peak velocity (APV) in each artery was selected for analysis. Patients underwent an acute pacing protocol with assessment of different pacing configurations. The study was then repeated with the ComboWire moved from the LAD to the circumflex artery (Cx). In 5 cases, the Cx was assessed before the LAD. Intracoronary adenosine was given as a bolus dose of 36 microgrammes to induce hyperemia for further investigation of the effect of pacing from different sites on coronary hemodynamics.
Baseline measurements were taken from patients in sinus rhythm in AAI mode. For patients with atrial fibrillation or complete heart block, the baseline comparator rhythm was right ventricular pacing with atrial synchronous pacing in the latter group. When comparisons were made between baseline and right ventricular pacing, those whose baseline required right ventricular pacing were omitted from analysis. All studies were performed with pacing at 10 beats above the intrinsic rate or at 70 beats per minute in the case of patients with underlying atrioventricular block to control for the Bowditch effect.[19](#jah31237-bib-0019){ref-type="ref"} Atrio‐ and ventriculoventricular delays were set, as per previous standard clinical echo optimization using the Doppler mitral inflow and LV outflow tract velocity time integral metrics, respectively. Baseline measures were reassessed for different pacing configurations to control for possible drift.
Patients received dual antiplatelet therapy prior to the procedure in case of the need for emergency percutaneous intervention and received boluses of heparin to keep an activated clotting time \>300 seconds.
The first 3 to 5 beats recorded after a change in pacing protocol were selected for analysis according to the following criteria, which are similar to those used by other groups[11](#jah31237-bib-0011){ref-type="ref"}: Lapse of an initial 10 seconds to stabilize the new pacing parameterExclusion of the beat preceding and the 3 beats following an ectopic ventricular beatDetermination that the signals were of sufficient quality for analysis, made by a physician who was blinded to the study hypotheses
Data analysis involved 2 stages. The raw data were analyzed using the StudyManager program (Volcano Corp). Wave intensity analysis was applied to the coronary data using a custom‐made program, Cardiac Waves (King\'s College London). Data were sampled at 250 Hz. Details of the methodology used to perform wave intensity analysis for data analysis have been described previously.[20](#jah31237-bib-0020){ref-type="ref"}, [21](#jah31237-bib-0021){ref-type="ref"} Briefly, a Savitzky--Golay convolution method was adopted using a polynomial filter to refine the derivatives of the aortic pressure and velocity signals.[21](#jah31237-bib-0021){ref-type="ref"}, [22](#jah31237-bib-0022){ref-type="ref"} The selected 3 to 5 consecutive cardiac cycles were gated to the ECG R wave peak, with ensemble averaging of aortic pressure, coronary pressure (P~d~), APV, and heart rate. Net coronary wave intensity (dI in the equation) was calculated from the time derivatives (dt in the equation) of ensemble‐averaged coronary pressure and flow velocity (dU in the equation), as follows: dI=dP~d~/dt×dU/dt.[10](#jah31237-bib-0010){ref-type="ref"}, [21](#jah31237-bib-0021){ref-type="ref"} Coincident backward (microcirculation‐derived) and forward (aorta‐derived) propagating waves were separated, assuming blood density to be 1050 kg/m^3^ and estimating coronary wave speed using the sum of squares method.[21](#jah31237-bib-0021){ref-type="ref"}, [23](#jah31237-bib-0023){ref-type="ref"} The areas beneath the 2 most prominent wave energies identified were analyzed and included in this report. These were the positive (aorta‐derived) FCW, occurring at the onset of systole, and the BEW, the first negative wave occurring at the onset of ventricular relaxation, identified by the onset of diastole. An example of the data following analysis are demonstrated in graphic form in Figure [1](#jah31237-fig-0001){ref-type="fig"}.
{#jah31237-fig-0001}
Statistical Analysis {#jah31237-sec-0006}
--------------------
Statistical analysis was performed using IBM SPSS version 22.0 (IBM Corp). Rather than assume normality, coronary flow data were analyzed as nonparametric variables using the Friedman test (the nonparametric equivalent of ANOVA for repeated measures).[24](#jah31237-bib-0024){ref-type="ref"} If the Friedman test was significant, comparisons between pairs of states were made using the Wilcoxon signed rank test, with probability values calculated comparisonwise. For normally distributed data, paired 2‐sided Student *t* tests were used to compare means. Demographic data are presented as means with standard deviations. Data regarding the atrio‐ and ventriculoventricular delays are presented as medians and ranges to provide clinically interpretable data. To allow for correlation of repeated measures in the same patient, linear mixed‐effect models were used to explore the relationship between coronary flow and acute changes in LV contractility.
Results {#jah31237-sec-0007}
=======
Twelve patients consented to take part in the study. The protocol could not be completed in 1 patient because of problems getting a stable signal from the ComboWire, and this patient was excluded from further analysis. Patient demographics for the remaining 11 patients are shown in [Table](#jah31237-tbl-0001){ref-type="table-wrap"}. There were no complications as a result of the acute procedure. The median atrioventricular delay was 125 ms (range 100 to 140 ms), and the median ventriculoventricular delay was left ventricle ahead by 30 ms (range 0 to 40 ms).
######
Demographic Data
Characteristics Mean (SD)
--------------------------- --------------
Age, y 59.9 (9.38)
Sex
Male 9
Female 2
Etiology
NICM 9
ICM 2
QRS morphology
LBBB 8
NSIVCD 2
RBBB 1
QRS duration, ms 156.6 (20.1)
EF before implant (%) 24 (10.5)
Sinus rhythm 8/11
Time since implant (days) 1027 (967)
ACEIs 11/11
Beta blockers 11/11
Aldosterone antagonists 9/11
ACEIs indicates angiotensin‐converting enzyme inhibitors; EF, ejection fraction; ICM, ischemic cardiomyopathy (previous infarct or flow‐limiting coronary disease at angiography); LBBB, left bundle branch block; NICM, nonischemic cardiomyopathy; NSIVCD, nonspecific intraventricular conduction delay; RBBB, right bundle branch block.
Effect on Coronary Flow With Different Pacing States {#jah31237-sec-0008}
----------------------------------------------------
There was no change in LAD flow when intrinsic conduction was compared with right ventricular pacing (−4.5%, *P*=0.123). BIVCS from the chronically implanted epicardial LV lead increased LAD flow significantly from baseline (+8.7%; *P*=0.033) and was further increased with the optimal endocardial pacing site (BIVEN; +27.0%; *P*=0.021) (Figure [2](#jah31237-fig-0002){ref-type="fig"}). BIVEN pacing resulted in an increase in LAD flow from BIVCS, but this was not significant (+6.6%; mean increase from baseline in 5 patients: BIVCS +20.2% versus BIVEN +27.0%; *P*=0.748). APV from the Cx showed no significant variation among baseline, BIVCS, right ventricle, and BIVEN (*P*=0.615) (Figure [2](#jah31237-fig-0002){ref-type="fig"}).
{#jah31237-fig-0002}
Analysis of the 8 patients with left bundle branch block demonstrated a significant increase in LAD flow with BIVCS from baseline (+10.2%, *P*=0.034) with no significant change in Cx flow (*P*=0.917). Analysis of the 8 patients in sinus rhythm demonstrated a significant increase in LAD flow with BIVCS from baseline (+10.1%, *P*=0.005) with no significant change in Cx flow (*P*=0.6).
Relationship Between Coronary Flow and Acute Myocardial Contractility (dP/dt~max~) {#jah31237-sec-0009}
----------------------------------------------------------------------------------
The change in LAD (APV) and acute hemodynamic response, as measured by dP/dt~max~, was significantly correlated with a Pearson correlation coefficient of 0.449 when all pacing states in all patients were assessed (*P*=0.015). The relationship between the 2 variables remained significant in the mixed model, which allowed for correlation within repeated measures from the same participant (β=1.006 \[95% CI 0.09 to 1.92\]; *P*=0.033) (Figure [3](#jah31237-fig-0003){ref-type="fig"}). The change in acute hemodynamic response was poorly correlated with the change in Cx flow, and there was no evidence of association using the mixed model when all pacing states were assessed (β=−0.215 \[95% CI −0.797 to 0.357\]; *P*=0.44).
{#jah31237-fig-0003}
Effect of Pacing on Wave Intensity Analysis Energy Profile {#jah31237-sec-0010}
----------------------------------------------------------
There was a significant increase in the magnitude of the BEW in the LAD with BIVCS compared with baseline (mean area 6422.8 W/m^2^ per second at baseline increased to 7281.9 W/m^2^ per second with BIVCS \[13% mean increase\]; *P*=0.004). BIVEN facilitated a significant increase in the area above the wave to a mean of 8200.3 W/m^2^ per second versus 5744.1 W/m^2^ per second (*P*=0.036). The difference in BEW between BIVEN and BIVCS for the artery was 1594.0 W/m^2^ per second, but this was not significant (*P*=0.139) (Figures [4](#jah31237-fig-0004){ref-type="fig"} and [5](#jah31237-fig-0005){ref-type="fig"}).
{#jah31237-fig-0004}
{ref-type="fig"}. BEW indicates backward expansion wave; BIVCS, conventional biventricular pacing; BIVEN, biventricular endocardial pacing; WI, wave intensity.](JAH3-4-e002626-g005){#jah31237-fig-0005}
There was no difference in the area under the FCW in the LAD with a mean increase from 2668.5 W/m^2^ per second at baseline to 3027.0 W/m^2^ per second with BIVCS (*P* value not significant). The optimal BIVEN position, however, also resulted in a significant increase in the energy of the FCW (mean area 1984.18 W/m^2^ per second to 4220.4 W/m^2^ per second \[112% mean increase\]; *P*=0.048) (Figures [4](#jah31237-fig-0004){ref-type="fig"} and [5](#jah31237-fig-0005){ref-type="fig"}).
There was no difference in the magnitude of the BEW in the Cx with BIVCS compared with baseline (mean 11 047.2 W/m^2^ per second at baseline reduced to 8666.6 W/m^2^ per second; *P*=0.237). There was a nonsignificant reduction in the size of the BEW for the patients who underwent the endocardial procedure (mean area under the BEW at baseline was 13 704.68 W/m^2^ per second versus largest BIVEN 5825.5 W/m^2^ per second; *P*=0.053). With regard to the area under the dominant FCW, there was no change in the energy from baseline with BIVCS (baseline 4501.648 W/m^2^ per second versus BIVCS 3229.2 W/m^2^ per second; *P*=0.123) and the best BIVEN (baseline 4548.48 W/m^2^ per second versus BIVEN 1910.3 W/m^2^ per second; *P*=0.176) (Figure [6](#jah31237-fig-0006){ref-type="fig"}).
{#jah31237-fig-0006}
Change in Timing of Coronary Waves With Application of Biventricular Pacing: Coronary Resynchronization {#jah31237-sec-0011}
-------------------------------------------------------------------------------------------------------
The time to the peak of the dominant BEW was significantly delayed between the LAD and the Cx at baseline in patients with a nonischemic etiology (284 ms in the LAD versus 331 ms in the Cx; *P*=0.01) (Figure [7](#jah31237-fig-0007){ref-type="fig"}). This was corrected by BIVCS (mean LAD 289 ms versus mean Cx 297 ms; *P*=0.566). The reduction of the difference between the time to peak of the BEW in the LAD versus the Cx by BIVCS was significant (mean 47 ms at baseline versus 8 ms; *P*=0.004) (Figure [7](#jah31237-fig-0007){ref-type="fig"}).
{#jah31237-fig-0007}
In assessing the FCW in a similar manner, the time to peak of the FCW at baseline was significantly different between the LAD and the Cx (30 versus 69 ms; *P*=0.03), with a reduction in the difference of this timing with BIVCS (56 ms \[LAD\] versus 53 ms \[Cx\]; *P*=0.715). The reduction of the difference to peak of the FCW was also significant (39 versus 3 ms; *P*=0.008) (Figure [7](#jah31237-fig-0007){ref-type="fig"}).
Coronary Flow Velocity Reserve in the LAD and the Cx {#jah31237-sec-0012}
----------------------------------------------------
Hyperemia was induced at baseline and with BIVCS. There was a significant difference between the baseline LAD coronary flow velocity reserve (CFVR; mean 2.35) and CFVR with BIVCS (mean 2.05; *P*=0.02). Conversely, there was a nonsignificant increase in the CFVR in the Cx from 2.1 to 2.46 with BIVCS pacing (*P*=0.349) (Figure [8](#jah31237-fig-0008){ref-type="fig"}).
{#jah31237-fig-0008}
Discussion {#jah31237-sec-0013}
==========
To our knowledge, this study was the first to comprehensively analyze the effect of epicardial and endocardial pacing on the constituent waveforms of coronary flow in the LAD and the Cx.
The principal findings were as follows: Increased flow in the LAD with BIVCS and BIVEN was the result of an increase in the BEW.In addition to an increase in the BEW, BIVEN increased coronary flow with a significant increase in the FCW in the LAD.The time to the peak of the FCW and the BEW is homogenized between the LAD and the Cx following CRT.Cx flow and the constituent determinants of flow did not alter with BIVCS or BIVEN.
The effect of biventricular pacing on coronary flow and physiology has been an area of inquiry for many years, with discussion as to whether the changes noted are merely the result of restoration of synchronous mechanical activation or related to changes in underlying myocardial oxygen supply and demand.[25](#jah31237-bib-0025){ref-type="ref"}, [26](#jah31237-bib-0026){ref-type="ref"} In recent years, there has been gradual refinement of our understanding of how coronary physiology is affected by the dyssynchronous ventricle[11](#jah31237-bib-0011){ref-type="ref"}, [27](#jah31237-bib-0027){ref-type="ref"}; however, significant disagreement remains within the literature as to the effect of pacing on coronary flow.[15](#jah31237-bib-0015){ref-type="ref"}, [18](#jah31237-bib-0018){ref-type="ref"}, [28](#jah31237-bib-0028){ref-type="ref"} Our findings provide new insight into the effect of epicardial and LV endocardial pacing on coronary physiology.
Coronary Flow: Comparison With Previous Studies {#jah31237-sec-0014}
-----------------------------------------------
The significant increase in flow in the LAD with BIVCS accords with recent studies.[14](#jah31237-bib-0014){ref-type="ref"}, [27](#jah31237-bib-0027){ref-type="ref"} The neutral effect on coronary flow in the Cx, however, differs with the findings of Itoh et al, who found that biventricular pacing increased both LAD and Cx flow, but this was in the acute setting soon after device implant. Our data may offer some explanation with regard to the differing results of studies looking at regional and global myocardial blood flow noninvasively: These results demonstrate either no change in regional myocardial blood flow or correction of a septal perfusion defect with CRT.[29](#jah31237-bib-0029){ref-type="ref"}, [30](#jah31237-bib-0030){ref-type="ref"}, [31](#jah31237-bib-0031){ref-type="ref"}, [32](#jah31237-bib-0032){ref-type="ref"} Our findings of minimal changes in the Cx combined with an increase in the LAD with BIVCS and BIVEN may explain these apparently divergent findings. If we accept that CRT increases global (left main coronary artery) flow velocity, as per Kyriacou et al, our findings suggest that this increase in flow preferentially affects the LAD, namely, there is an increase in flow to the LAD with unchanged Cx flow rather than redistribution of an unchanged global blood flow toward the LAD.[24](#jah31237-bib-0024){ref-type="ref"} This may explain why imaging modalities do not report redistribution of flow (ie, there is none) even though CRT has been noted to reverse septal perfusion abnormalities (through increased LAD flow secondary to increased global flow).
Equally, the noted significant reduction in CFVR in the LAD with BIVCS and the nonsignificant increase in CFVR in the Cx contrast with other studies showing increased CFVR in the LAD with CRT.[27](#jah31237-bib-0027){ref-type="ref"}, [28](#jah31237-bib-0028){ref-type="ref"} Our findings may appear counterintuitive; one might expect more physiological electrical activation to have a positive effect on parameters such as CFVR. Nevertheless, our reported reduction in CFVR with BIVCS may reflect the multifaceted etiology of the impaired septal flow at baseline and during hyperemia in this cohort of patients with chronically implanted biventricular devices.
CFVR levels in the LAD and the Cx were below those seen in health, suggesting a generalized blunted hyperemic response reflecting microvascular dysfunction that cannot be overcome by CRT. Components of microvascular dysfunction that CRT is likely unable to overcome include the anatomical disruption of the microvasculature by fibrosis and extracellular matrix, decreased myocardial capillary density, and impaired capillary vasodilation.[33](#jah31237-bib-0033){ref-type="ref"}, [34](#jah31237-bib-0034){ref-type="ref"}, [35](#jah31237-bib-0035){ref-type="ref"}, [36](#jah31237-bib-0036){ref-type="ref"} The only component of microvascular resistance that seems likely to be altered by CRT is the effect of compressive forces related to high end‐diastolic pressures.[25](#jah31237-bib-0025){ref-type="ref"}
It is important to interpret our findings within the context of our studied population, namely, patients who had chronically received CRT and thus potentially undergone ventricular remodeling. It is suggested that the sudden withdrawal of BIVCS results in new‐onset mechanical dyssynchrony and thus acts as potential and sudden mechanical obstruction to LAD flow with attendant increases in diastolic pressures.[37](#jah31237-bib-0037){ref-type="ref"} When BIVCS is reintroduced, this obstruction is relieved, resulting in an increase in flow. Conversely, this mechanical obstruction can be overcome during hyperemia both at baseline and with BIVCS as a result of the chronic remodeling process, resulting in similar hyperemic flows. The net effect is a reduction in CFVR when BIVCS is compared with baseline. Conversely, the flow in the Cx is relatively unaffected by the cessation of BIVCS (with a trend toward reduction in flow). In the presence of a fixed hyperemic flow, this will manifest as the nonsignificant trend toward an increase in the CFVR with BIVCS.
The potential explanation for the increase in LAD flow with BIVCS may relate to the noted correlation between LAD APV and dP/dt~max~. The finding is biologically attractive and accords with the work of Kyriacou et al on the effect of CRT on myocardial oxygen demand.[24](#jah31237-bib-0024){ref-type="ref"} The increase in flow in the LAD from baseline to BIVCS could represent a physiological increase in blood flow to help meet the increased myocardial oxygen consumption required as a result of the increased contractility and septal work. Our findings, however, do not preclude a second possibility that the increase in LAD APV with BIVCS is the result of changes in microvascular resistance secondary to correction of mechanical dyssynchrony, and the wave intensity data are consistent with the hypothesis that these local factors may play an important role.
Wave Intensity Analysis of Left‐Sided Coronary Circulation {#jah31237-sec-0015}
----------------------------------------------------------
The increase in LAD flow with BIVCS was significantly related to an increase in flow mediated by the BEW, suggesting increased suction by improved regional relaxation of the microvasculature. The increase in LAD flow during BIVEN was related to an increase in both the FCW and the BEW. Although the reason for the increase in the BEW with BIVEN is likely similar to that proposed for BIVCS, it is suggested that the increase in the FCW in the LAD is likely related to global effects of forward propulsion and, importantly, to changes in the septal microvasculature; no increase was noted in the FCW in the Cx. This change in microvasculature might be the result of the more physiological activation pattern that LV endocardial pacing is thought to be able to achieve by early activation of the specialized Purkinje network.[38](#jah31237-bib-0038){ref-type="ref"}
With regard to the Cx, we found a nonsignificant trend toward reduction in both the FCW and the BEW. With regard to the BEW, this would conform to the hypothesis that BIVCS reduces work in the lateral wall and thus that the local intramyocardial pressure generated to create the suction for the BEW is reduced. This combined assessment of the changes in the FCW and the BEW with different pacing regimens demonstrates that both regional and global mechanisms are involved in determining coronary flow to different parts of the myocardium. This duality is further developed considering the synchronization of timing of the FCW and the BEW in the LAD and the Cx following CRT.
Coronary Resynchronization {#jah31237-sec-0016}
--------------------------
By measuring the time to the peak of the FCW and the BEW in both the LAD and the Cx at baseline and with BIVCS, we are able to demonstrate for the first time in humans that there is a difference between the timing of flow in the LAD and the Cx in dyssynchronous heart failure. Furthermore, we have shown that this is corrected significantly with regard to the time to peak of both the FCW and the BEW. The implications are 2‐fold. First, such homogenization is suggestive of generalized cardiac resynchronization. Second, it has been recognized previously that in patients with normal coronary arteries, the waves measured in coronary flow are generated by the global effects of cardiac contraction and relaxation causing propulsion of blood forward (FCW) and suction of blood backward (BEW), respectively. Changes in the sizes of waves have been correlated to wall thickness in concentric hypertrophy, suggesting that waves are dependent on the myocardium and microvasculature per se and are not just generated by the intracavity pressure gradients.[10](#jah31237-bib-0010){ref-type="ref"}
Our findings relating to dyssynchronus heart failure and the effect of BIVCS are the first to demonstrate the effect of both regional and global forces on the coronary waves. It is proposed that the ability of CRT to homogenize this timing suggests that dyssynchrony causes regional changes in the microvascular energetics that can be corrected with biventricular pacing. With regard to the FCW, timing is homogenized to a time point between the time to peak of the LAD and the Cx at baseline. This accords with the concept that the FCW is generated by the forward propulsion of blood down the coronary arteries as a result of LV systolic contraction. Conversely, the homogenization of the larger BEW is achieved largely by a reduction in the time to peak of the BEW in the Cx rather than a significant change in LAD timing. This is biologically attractive because any negative pressure generated in the LV cavity in diastole relies on the microvasculature as a conduit to create the suction for the BEW. Consequently, delayed lateral wall relaxation before resynchronization of this conduit will result in a delayed time to peak in the BEW in the Cx and reduction of the time to peak of the BEW that we observed following CRT.
From the clinical perspective, our study offers potentially new explanations of how conventional CRT may exert its benefit. It is suggested that changes in coronary blood flow and wave energy need to be investigated in prospective studies at the time of implant to determine whether they may offer a role in patient selection. As noted, neither electrical resynchronization nor correction of mechanical dyssynchrony have been found to be accurate predictors of response to CRT.[3](#jah31237-bib-0003){ref-type="ref"} Although changes in coronary hemodynamics are also unlikely to be single predictors of response, it may be that they have roles in a more integrated patient‐selection pathway in difficult cases or can be used as a target for optimization in nonresponders using emerging noninvasive techniques.[3](#jah31237-bib-0003){ref-type="ref"}, [39](#jah31237-bib-0039){ref-type="ref"} Further dedicated studies are required to assess these potential clinical applications.
There is evidence that BIVEN has an acute effect on LV function beyond conventional CRT, but further research is necessary to determine the mechanisms by which this occurs. It is again suggested that this be performed at the time of device implant, with attention paid to coronary wave energetics as a component of such studies.[17](#jah31237-bib-0017){ref-type="ref"} Such research is increasingly required as device manufacturers begin to market BIVEN systems.[40](#jah31237-bib-0040){ref-type="ref"} It is highly desirable that we have an understanding of the mechanistic effects of such systems before their use in the clinical environment becomes routine.
Study Limitations {#jah31237-sec-0017}
-----------------
The patients studied were not selected for any particular clinical characteristics and reflect a real‐world CRT population; however, our sample size was small, and it is important that we remain guarded about the generalizability of our results to the whole CRT population. Nevertheless, the number of patients studied is equivalent to other similar invasive studies.[11](#jah31237-bib-0011){ref-type="ref"}, [41](#jah31237-bib-0041){ref-type="ref"} The invasive procedural burden for patients is very high and is not suitable for routine clinical practice; however, emerging noninvasive techniques to measure wave intensity are increasingly being described and may allow larger studies to be performed.[39](#jah31237-bib-0039){ref-type="ref"} All patients studied were chronically implanted with CRT devices. This makes extrapolation of our findings to preimplant patients difficult, although many of the findings could be amplified in the preimplant population. The prolonged time between implantation of the device and study participation means that any association of blood flow changes, wave energetics, and response to CRT is unclear. As noted, further studies at the time of implant are required. By fixing the heart rate, we controlled for the effect that changes in chronotropy can cause to inotropy (the Bowditch effect); however, this does prevent reflex heart rate regulation to changes in inotropy.
Conclusions {#jah31237-sec-0018}
===========
The data presented develop our understanding of coronary physiology in dyssynchronous heart failure with biventricular pacing beyond the left main coronary artery and give new insight into the effect of regional and global mechanics as well as coronary wave energies on coronary flow. BIVCS and BIVEN increased the microcirculation‐derived BEW in the LAD with no change in the Cx. The time to peak of the BEW was homogenized in the LAD and the Cx following CRT. These findings suggest that the coronary BEW is susceptible to changes in local forces rather than dependent solely on increased global LV systolic and diastolic function, as was reported previously.[11](#jah31237-bib-0011){ref-type="ref"} The demonstrated increase in LAD coronary flow with an increase in the LAD FCW with endocardial rather than epicardial pacing offers an exciting mechanism to further increase coronary flow and may be mediated by a more favorable and physiological activation pattern of the left ventricle.
Sources of Funding {#jah31237-sec-0020}
==================
Equipment was provided by a grant from Volcano Corp.
Disclosures {#jah31237-sec-0021}
===========
None.
Dr Siobhan Crichton, Division of Health and Social Care Research, King\'s College London for assistance in building the statistical models.
| {
"pile_set_name": "PubMed Central"
} |
1. Introduction {#sec1-materials-10-01138}
===============
Scandium is classified along with yttrium and other lanthanides as a rare-earth (RE) element, has an atomic number of 21, an atomic weight of 44.96, and is the 31st most abundant element in the earth's crust, exhibiting a Clarke number of 22 ppm \[[@B1-materials-10-01138]\]. Thus, scandium is more abundant than some well-known metals, such as tin and lead. Scandium is increasingly being used as an alloying element for aluminum alloys in the field of high-performance lightweight structural alloys \[[@B2-materials-10-01138],[@B3-materials-10-01138]\]. Several advantages are realized when adding trace amounts of scandium to aluminum alloys. For instance, it has been reported that the alloy Al~20~Li~20~Mg~10~Sc~10~Ti~30~ is as light as aluminum, as strong as titanium, and as hard as a ceramic \[[@B4-materials-10-01138]\]. Because of these advantages, Al--Sc alloys are being produced commercially for use in aviation and aerospace components, bicycle frames, and baseball bats, among other applications. Sc is also used in metal halide lamps, known as mercury vapor lamps. Sc is added to these lamps, which are used in automobiles as a white light source \[[@B5-materials-10-01138]\]. A few years ago, it was also found that Sc can be used in electrical devices. High-Sc-content ScAlN thin films are attracting attention because of their strong piezoelectricity. Akiyama et al. found that the piezoelectricity of Sc~*r*~Al~1‒*r*~N thin films monotonically increases with the increase in the Sc concentration, *r*. The material's piezoelectricity reaches its maximum at *r* = 43%, at which point the piezoelectric coefficient, *d*~33~, is five times that of pure AlN \[[@B6-materials-10-01138],[@B7-materials-10-01138]\]. Hashimoto et al. reported that a surface acoustic wave (SAW) resonator based on the structure ScAlN/6H-SiC exhibited resonance *Q*, antiresonance *Q*, and *K*^2^ values of 340%, 240%, and 4.5%, respectively, at 3.8 GHz \[[@B8-materials-10-01138]\]. These values suggest that Sc*~r~*Al~1-*r*~N thin films formed on a hard substrate should be suitable as SAW wideband filters for next-generation wireless communication systems. However, it is often difficult to use ScAlN thin films in MEMS devices---including acoustic ones---because of the extremely high price of Sc metal, given the difficulty associated with smelting it.
Scandium is currently produced mainly through a thermal reduction method. Since scandium oxide as a raw material is extremely stable thermodynamically, scandium is produced by converting it into scandium fluoride, which is easily reducible, and reducing it at 1873 K using metallic calcium as the reducing agent \[[@B9-materials-10-01138]\]:
Since this process involves fluoridation, it is expensive and environmentally unfriendly. Furthermore, some of the calcium remains as an impurity. Several researchers have reported other processes for smelting Sc \[[@B10-materials-10-01138],[@B11-materials-10-01138]\]. Harata et al. demonstrated that Al--Sc alloys can be produced directly via a calciothermic reduction reaction using CaCl~2~ as the flux and Al as the collector metal \[[@B11-materials-10-01138]\]. However, it is difficult to execute this process using a molten salt at high temperatures (approximately 1100 K). On the other hand, we were previously able to successfully obtain small amounts of magnesium metal by microwave irradiation. This process had a yield as high as 71% and consumed half of the energy used in the conventional process, which is known as the Pidgeon process \[[@B12-materials-10-01138]\]. In this report, we describe the application of the abovementioned process for the smelting of Sc metal by microwave irradiation and report the obtained results.
2. Experimental {#sec2-materials-10-01138}
===============
The following two chemical reactions were utilized while using microwave radiation as the heat source:
The starting material, Sc~2~O~3~, is converted into an Al--Sc alloy. For safety, we employed CaH~2~ instead of Ca metal.
3. Results and Discussion {#sec3-materials-10-01138}
=========================
[Figure 1](#materials-10-01138-f001){ref-type="fig"}a shows the temperature of the crucible measured using an infrared sensor and the microwave power as functions of the time of the chemical reaction in Equation (3). During the process, the temperature exceeded 700 °C within 10 min, and the crucible was maintained at a temperature higher than 750 °C for 30 min. It was assumed that the sample temperature was more than 1000 °C when the crucible temperature was close to 800 °C. [Figure 1](#materials-10-01138-f001){ref-type="fig"}b shows the temperature of the crucible measured using an infrared sensor and the microwave power as functions of the time for the chemical reaction in Equation (4). The temperature of the crucible reached 880 °C within 15 min, and was maintained at 880 °C for 15 min. In this case, in order to determine the sample temperature, a multiphysics simulation was also performed, and the sample temperature was determined to be the same as the crucible temperature.
It was difficult to separate the products from the residues after the chemical reaction in Equation (3). [Figure 2](#materials-10-01138-f002){ref-type="fig"}a shows the results of X-ray diffraction (XRD) measurements (Rigaku, Tokyo, Japan) and a photograph of the mixture of the residues and products of this reaction. A few strong peaks related to calcium and a few weak peaks related to aluminum can be observed in the XRD pattern. However, it was difficult to determine whether there was a shift in the aluminum peak, which is indicative of the production of the Al--Sc alloy.
After the completion of the chemical reaction in Equation (4), a thin film with a metallic luster formed on the glass tube. [Figure 2](#materials-10-01138-f002){ref-type="fig"}b shows the results of the XRD measurements and a photograph of the thin film formed on the glass tube. A few peaks can be observed in the pattern, even though their intensities are low. On the basis of JPCD references, the peak at 31° could be attributed to Sc (101). Thus, the thin film was confirmed to be Sc using XRD measurements. This is the first example of Sc metal obtained by smelting through microwave irradiation with carbon as a reduction material from ScF~3~. However, these results strongly suggest that the microwave irradiation process has immense potential for use in the smelting of scandium metal.
4. Discussion {#sec4-materials-10-01138}
=============
Usually, the temperature of chemical reactions involving microwaves is lower than that in the case of conventional heating. However, in this case, the chemical reaction described by Equation (3) did not occur. The reason that the reaction in Equation (2) was able to proceed was that the internal energy of Ca is high, because Ca is in the gas phase, and the temperature of the reaction is high. On the other hand, the chemical reaction in Equation (4) occurs at 880 °C, which is approximately half of that for the conventional method. [Figure 3](#materials-10-01138-f003){ref-type="fig"} shows the results of a multiphysics simulation. The curves show the temperatures of the wall of the crucible and the inside of the sample as functions of microwave irradiation time. It can be seen that the two temperatures are the same, which was because the reactions were performed in a vacuum. Even though the reduction agents used in Equations (2) and (4) are different, the contact points of the particles probably undergo localized heating in the case of Equation (4). Haneishi et al. observed the specific local heating at the contact points during the microwave heating of silicon carbide spheres by in situ emission spectroscopy \[[@B13-materials-10-01138]\]. The same phenomena occurred in mixed particles of ScF~3~ and C. In this case, owing to the electric current at the contact points, the specific high temperature at the contact points was assumed to be generated up to the reaction temperature. However, the average temperature of the sample consisting of the mixed particles of ScF~3~ and C was low. Specific local nonequilibrium heating---so-called "hot spots"---has been suggested for this system.
5. Materials and Methods {#sec5-materials-10-01138}
========================
[Figure 4](#materials-10-01138-f004){ref-type="fig"} shows the experimental setup consisting of a microwave generator with a phase-locked loop (PLL) oscillator, a power amplifier module (FSU-201VP-02, Fujidenpa Corp, Saitama, Japan), a plunger for impedance matching, a waveguide cavity with the TE103 mode, a rotary pump, a quartz glass tube, an infrared temperature measurement system (FTK9-R, Japan Sensor Corp, Tokyo, Japan), an alumina crucible (located in the TE103 cavity), and a sample (placed in the crucible). The input microwave power was the difference between the input power and reflected power, which were measured by the power meter incorporated in the amplifier module. Before the smelting process, the reflected power was adjusted to zero by the plunger in a short time. The temperature of the crucible was measured through a hole in the waveguide cavity by the infrared temperature measurement system. For the chemical reaction in Equation (3), a scandium oxide (Sc~2~O~3~) powder (average particle size: less than 10 μm), calcium hydride (average particle size: 100--500 μm), and an aluminum powder (average particle size: 100--300 μm) were mixed and packed in the alumina crucible to form a sample. The total weight of the mixture was 0.46 g, and the Sc~2~O~3~/CaH~2~/Al molar ratio was 0.15:0.96:0.27. The TE103 waveguide cavity has a standing electromagnetic wave in the cavity, and the maximum magnetic field in the cavity is obtained at two points. Usually, these maxima in the magnetic field are used to anneal conducting materials with microwave irradiation \[[@B14-materials-10-01138]\]. For the process described by Equation (3), a microwave power of 177 W was applied at the maximum magnetic field of the TE103 applicator, where the maximum field closer to the input port was employed, under low vacuum (10 Pa) and a crucible temperature of 780 °C. For the chemical reaction described by Equation (4), scandium fluoride powder (average particle size: less than 10 μm) and carbon particles (average particle size: 50 μm) were mixed and packed in the alumina crucible to form a sample. The total weight of the mixture was 0.68 g, and the ScF~3~/C molar ratio was 1:3. [Figure 5](#materials-10-01138-f005){ref-type="fig"} shows scanning electron microscopy (SEM) images of the various powders. For the process described by Equation (4), a microwave power of 100 W was applied at the maximum magnetic field of the TE103 applicator under low vacuum (10 Pa) and a crucible temperature of 880 °C. After the completion of the reactions, XRD measurements of the samples were performed to identify the Al--Sc metal alloy in the sample obtained by the reaction in Equation (3) and the Sc metal in the sample obtained by the reaction in Equation (4). Multiphysics simulations were performed using a finite element method called COMSOL in order to determine the reaction temperature at the center of the sample. [Table 1](#materials-10-01138-t001){ref-type="table"} summarizes the parameters of each material used in the multiphysics simulation.
6. Conclusions {#sec6-materials-10-01138}
==============
For the first time, we demonstrated that microwave irradiation can be used to smelt Sc metal. Using microwave radiation, we attempted to perform two chemical reactions. Instead of Ca, we used carbon as the reducing agent for fluoride reduction under microwave irradiation, the corresponding chemical reaction occurred at 880 °C, which is approximately half of the temperature for the conventional process. Generally, the price of carbon is cheaper than that of Ca, and carbon is also easier to use than calcium metal. Moreover, equipment with a heat capability below 1000 °C can be easily built in comparison with that with a heat capability over 1000 °C. Therefore, the reaction of carbon particles with scandium fluoride under microwave irradiation can potentially be used to smelt Sc metal.
This study was supported in part by a Grant-in-Aid for Scientific Research (S) 25249113.
Satoshi Fujii proposed the idea of scandium smelting by microwave irradiation and was primarily responsible for theoretically explaining the mechanism underlying the smelting process using FEM simulations. Eiichi Suzuki, Shuntaro Tsubaki and Naomi Inazu performed the experiments and analyzed the data. Yuji Wada provided constant supervision, and we also participated in all the scientific discussions on the topic. All the authors discussed the results and commented on the manuscript.
The authors declare no conflict of interest.
{#materials-10-01138-f001}
{#materials-10-01138-f002}
{#materials-10-01138-f003}
{#materials-10-01138-f004}
{#materials-10-01138-f005}
materials-10-01138-t001_Table 1
######
Material parameters used in COMSOL.
Material Conductivity (S/m) Relative Permittivity Relative Permeability Thermal Conductivity (W/(m × K)) Density (kg/m^3^) Heat Capacity (J/(kg × K))
------------------ -------------------- ----------------------- ----------------------- ---------------------------------- ------------------- ----------------------------
ScF~3~ + C 1000 3.53--0.058j ^1^ 1 118 2260 700
Glass (quartz) 0 4.2 1 10 2210 730
Alumina Crucible 0 1.8 1 5 3900 900
^1^ This value was determined via the perturbation cavity method. The other parameters are assumed from the COMSOL database.
| {
"pile_set_name": "PubMed Central"
} |
Preradiosurgical embolization of cerebral arteriovenous malformations (AVMs) has been warranted as a signif icant treatment to eliminate the risk of hemorrhage during the latency period after radiosurgical treatment, and to achieve AVM volume reduction to a size amenable to radiosurgical treatment, resulting in earlier obstruction \[[@B1][@B2]\]. The treatment of inappropriate factors, such as flow-related aneurysms, is warranted because proximal and pedicular aneurysms and intranidal aneurysms have a 7% to 10% annual risk of hemorrhage \[[@B3][@B4]\].
However, embolization has recently been criticized as a useless option for radiosurgery because embolization decreases the obliteration rates after stereotactic radiosurgery (SRS) compared with radiosurgical treatment alone \[[@B5]\]. Recanalization of embolized feeder after RS is seen in 5--7% of patients, and some reports question the efficacy of embolization \[[@B6][@B7][@B8][@B9][@B10][@B11][@B12][@B13][@B14]\]. The recanalization or vascular remodeling of remaining feeders tends to occur in cases with incomplete nidus embolization and far proximal feeder occlusion \[[@B15][@B16]\]. For this reason, radiosurgeons may struggle to plan the interested area because the part with pretended obstruction after embolization with a risk of late recanalization is outside the target. By contrast, proper nidus embolization will prevent recanalization \[[@B2][@B7]\] and can contribute to achieving a preferable result from SRS.
In Japan, where there are a large number of Gamma Knife centers, the treatment option with SRS following embolization has a comparatively higher rate among the combined therapeutic modalities \[[@B17]\], and this difference affects the rate of favorable results. This multicenter study, J-REAL (Japanese registry of Radiosurgery following Embolization for Arteriovenous maLformations), is a retrospective analysis of the selected AVM cases treated with SRS following embolization that was planned to clarify the efficacy of embolization and address the discrepancy in the results between Japan and western countries.
MATERIALS AND METHODS
=====================
Outline of the study
--------------------
A retrospective review of AVMs treated between 2003 and 2012 with embolization followed by SRS was performed. The data were harvested from institutes, enrolling experienced neurointerventionists who had proper and secure strategies for the endovascular treatment of AVM. The clinical materials were AVMs with a maximum diameter of more than 1 cm in patients older than 6 years of age. Patients who underwent SRS or direct surgery before embolization, except for emergency ventricular drainage, were excluded. All AVMs were embolized with NBCA mixture in one or staged sessions, and no cases with Onyx embolization were included. The SRS tool was a Gamma Knife in all cases. Patients treated with other methods, such as LINAC and X knife, were excluded. The timing of SRS was specified as within 6 months after the final embolization, and the decision of the marginal dose and targeting area was based on the guidelines of each Gamma Knife center. The operators of endovascular treatment were senior board experts trained by the Japanese society of neuroendovascular therapy with experience in more than 20 cases of AVM embolization. Similarly, the operators of the Gamma Knife were required to have experience with more than 50 radiosurgery cases for AVM.
The performance of endovascular treatment was impartially judged by multiple neurointerventionists, and the final image judgment was performed by the experienced multiple radiosurgeons 3 years after the final radiosurgery. This study was approved of the local institutional review board.
Clinical data
-------------
We investigated patent profile including gender, age, clinical manifestation, undergone initial treatment and general complications, and AVM profile including location, Spetzler-Martin grade\[[@B18]\], maximum diameter; volume, modified radiosurgery-based arteriovenous malformation score (mRBAS)\[[@B19]\] and angioarchitecture including nidus type\$, daughter nidus, draining pattern, varix, associated aneurysm, and meningeal feeders. AVM volumes were calculated from a 3-dimensional angiogram after determination of the radius (r) on three orthogonal planes using the formula for an ellipsoid (4πr1×r2×r3/3). mRBAS were determined by the following equation: AVM score = (0.1×volume in cm^3^) + (0.02×age in years) + (0.5×location). The location values are as follows: frontal/temporal/parietal/occipital/intraventricular/corp us callosum/cerebellar = 0, and basal ganglia/thalamus/brainstem = 1. Nidus type was categorized into compact type and diffuse type. Compact type was defined as AVM with clearly demarcated nidus without daughter ones. While, diffuse type was defined as AVM with abnormal vessel dilatation or a non-shunting abnormal vascular network surrounding the nidus, particularly expressed at the watershed area between different perfusion territories.
Performance of the treatment
----------------------------
### A. Performance of embolization
Data were collected for the total number of accessible / inaccessible feeders, successful / incomplete occlusion of f istulous feeders, and successful / unsuccessful nidus penetration as proximal feeder occlusion alone. The achievement level of embolization was defined in the newly provided categorization, embolization performance grade (E-grade), to exclude the operator\'s subjective preference ([Table 1](#T1){ref-type="table"}). E grade A was defined as sufficient nidus embolization with more than 50% of total number of feeders achieving nidus penetration. E grade B was defined as less than 50% achievement of nidus embolization, and E grade C was defines as failure to perform nidus embolization due to a lack of access, technical error or fistulous AVM with no nidus components. Grades A and B were subdivided according to the components and treatment of fistulous feeders. The final embolization rate was calculated as the result of the three-dimensional volume reduction of the nidus. The size of the remaining nidus after embolization was expressed as the maximum diameter of the residual nidus. Procedure-related or perioperative complications associated with embolization and the neurological deterioration due to the complications were registered.
### B. Performance of radiosurgery
The effect of SRS was evaluated at 2 years after the first operation. The post-SRS radiological evaluation was performed using the original grading system (SRS-grade). Patients with complete occlusion of AVM in the f inal angiogram were classif ied as SRS grade A. Patients who had a fine abnormal vascular network without AV shunt remaining in the site of the nidus were classif ied as SRS grade B, and those with a remaining nidus with obvious AV shunt were classified as SRS grade C. Other information concerning the radiosurgery, including the duration between embolization and SRS; major adverse events after the radiosurgery, such as rupture or delayed complications; and the final clinical outcome were registered.
Statistical analysis
--------------------
Patients with SRS grade A were categorized as the "successful occlusion group" (S group), and the patients in SRS grades B and C were categorized as the "non-successful occlusion group" (N group). These two groups were compared in terms of patient background, AVM profile and embolization performance. As a sub-analysis, SRS grade-B patients were independently studied to determine differences from SRS grade-C patients. The statistical analysis of categorical variables was performed using the χ^2^ and Fisher exact tests. The comparison of means was performed using Student\'s t-test, and an analysis of variance (ANOVA) followed by Bonferroni post hoc testing was performed, as appropriate. Predictive factors in the univariate analysis concerning size and embolization specif ication were entered into a multivariate logistic regression analysis using a stepwise method. The percentage of incidence and 95% confidence intervals (CI) were calculated for all considered variables and results. P-values \<0.05 were considered to indicate significance.
RESULTS
=======
Patient profile
---------------
A total of 73 patients met the inclusion criteria and underwent embolization followed by SRS during the study period ([Table 2](#T2){ref-type="table"}). There were 40 males and 33 females ranging from 6--78 years old with a mean age of 35.8 years. Of the 30 unruptured AVMs, the clinical manifestations were asymptomatic in 10, convulsion in 10, headache in five, and focal neurological deficits including cognitive and psychological function deficits in five. Of 43 patients with ruptured AVMs, 20 patients had intracerebral hemorrhage, 16 had intraventricular hemorrhage, and seven had subarachnoid hemorrhage. The initial treatments before embolization were medication for seizure, intracranial pressure control, and vital stabilization in 23 patients, as well as ventricular drainage in four patients. Precedent general complications were recognized in 4 patients and neurological deficits before treatment were observed in 18 patients, including mild deficits (mRS 1) in 9 patients, middle (mRS 2) in 3 patients and severe (mRS 4) in 6 patients.
AVM profile
-----------
The AVM profile is summarized in [Table 3](#T3){ref-type="table"}. Of the AVMs, the location was frontal in 20, temporal in 8, parietal in 10, occipital in 11, cerebellum in 16, thalamostriate in three and brain stem in four. Large AVMs extending over multiple areas were sorted according to the area containing of the largest part. For the Spetzler-Martin grade, there were nine AVMs in grade 1, 17 in grade 2, 32 in grade 4 and one in grade 5. The preoperative mean maximum diameter was 31.6 mm (range 11--72 mm) and the mean volume was 13.8 ml (range 1--46 ml). The distribution of the volume was less than 5 ml in 24 AVMs, 6--10 ml in 15 AVMs, 11--20 ml in 13 AVMs, 21--30 ml in 10 AVMs, 31--40 ml in 9 AVMs and more than 40 ml in 2 AVMs. The median mRBAS before embolization was 2.13, and the distribution of AVMs in mRBASs was as follows: ≤ 1.00 in 14, 1.01--1.50 in 11, 1.51--2.00 in 16 and \>2.00 in 31.
Regarding the angioarchitecture of the AVMs, 52 were classified as compacted types and 21 were classified as diffuse types; a daughter nidus was observed in five AVMs. The draining pattern of the AVMs was as follows: single superficial drainage in 18, single deep drainage in 11, multiple superficial drainage in 32 and multiple deep drainage in 12. The cases with both superficial and deep drainers were classified to the category of the main drainage side. Varices on the drainers were observed in six AVMs. Associated aneurysms were found in 22 AVMs, including 13 proximal feeder aneurysms, 3 flow-related aneurysms and 6 intranidal aneurysms. Of these, 12 aneurysms were treated with embolization in addition to the embolization for AVM. Seventeen AVMs were also supplied by meningeal feeders, and 9 of these were embolized in a manner similar to that use for AVMs supplied by pial feeders.
Performance of embolization
---------------------------
According to our embolization grading system, the AVMs were distributed in grades as follows: 12 in E-Grade A1, 20 in E-Grade A2, 19 in E-Grade B1, 18 in E-Grade B2 and two in E-Grade C ([Table 4](#T4){ref-type="table"}). The mean final volume reduction rate was 61.2%. The maximum diameter of the remaining nidus was almost zero in 7 AVMs, \<1 cm in 22 AVMs, 1--2 cm in 24 AVMs, 2--3 cm in 13 AVMs, and ≥ 3 cm in 7 AVMs. Perioperative complications occurred in 14 patients, including perioperative or delayed hemorrhage in eight patients, postoperative convulsion in two patients, and other minor complications in four patients. Of these, two patients had neurological deterioration with score changes of ≥ 2 on the modified Rankin Scale.
Performance of radiosurgery
---------------------------
The AVMs classified by SRS grade as follows: 44 in SRS-grade A, 18 in SRS-Grade B and 11 in SRS-Grade C ([Table 5](#T5){ref-type="table"}). The median duration between final embolization and SRS was 1.9 months. Delayed adverse events occurred in 4 patients, including hemorrhage in three patients and cyst formation in one patient. Of these, two patients presented with neurological deterioration and a score change of ≥ 2 on the modified Rankin Scale.
Comparison of profiles by SRS grade
-----------------------------------
The data of patients and AVM profiles were compared between the S and N groups ([Tables 2](#T2){ref-type="table"} and [3](#T3){ref-type="table"}). For patient prof iles, there was no difference between the two groups in terms of age distribution, clinical manifestation, and initial treatment methods. While the dominancy of gender is different between S group (female: 16/44 (36%)) and N group (17/29 (59%)) with female dominancy, in particular significantly different between S group and SRS grade C (9/11(82%)) with female dominancy (P=0.015).
For the clinical manifestation, initial treatments and preceding neurological deficits, there were no significant differences between the two groups. The AVM location and Spetzler-Martin Grade showed no deviation and had a similar distribution in both groups. The sizes of the AVMs also had a similar distribution. However, the mean maximum diameter (SRS-grade A: 28.7 mm, B: 32.9 mm, C: 40.7 mm) showed no significant difference between the SRS-grade A and B groups, but they were significantly larger in the SRS-grade C group than group A (P=0.013). The mean initial volume (SRS-grade A: 12.0 ml, B:12.4 ml, C: 23.5 ml) showed similar tendency and signif icant difference between Grade A and C groups (P=0.004). The mean mRBAS (SRS-grade A: 2.01, B: 1.94, C: 2.97) was also significantly higher in SRS-Grade C AVMs (P\<0.001)
Regarding angioarchitecture, a diffuse type of nidus was significantly more frequent in the N group (14/29 (48%), P=0.004), particularly in Grade C AVMs (8/11 (73%), P\<0.001), than the S group (7/44 (16%)). Multiple drainers were significantly more frequent in the N group (22/29 (76%)) than the S group (22/44 (50%) (P=0.031). The presence of a daughter nidus, associated aneurysm and varix were not different among the groups. The supply from the meningeal feeder (S group: 6/44 (14%), N group:11/29 (38%) ) was significantly frequent in the N group (P=0.023).
Comparison of embolization performance by SRS grade
---------------------------------------------------
[Table 4](#T4){ref-type="table"} shows the comparison of embolization performance by SRS grade. The E grade groups were divided into two groups by the rate of successful nidus embolization; one was the successful embolization group, including E grades A1 and A2 (n=32); and the other was the unsuccessful group, including E grades B1, B2 and C (n=41). In the former group, SRS grade A was achieved in 30 patients (94%), while in the latter, SRS grade A was achieved in only 14 patients (34%). There was a significant difference between the two groups (P\<0.001). Because a small AVM has more potential for complete occlusion after SRS without the aid of embolization, subanalysis was performed to target the relatively larger AVMs (n=43; S group: 21 (49%), N group: 22 (51%)), excluding AVMs less than 3 cm in diameter. These results also showed a significant difference between small and larger AVMs (P\<0.001). The size of the remaining nidus, one of the main factors defining the marginal dose of radiosurgery, was compared between the S and N groups. Remaining nidus was observed following embolization in 67 AVMs (S group: 38/44 (86%), N group: 28/29 (97%)). The comparison was performed between two groups that had a maximum remaining nidus diameter of more and less than 2 cm. A smaller remaining nidus (less than 2 cm in maximum residual nidus diameter) was seen in 22 cases (50%) in the S group but was seen in only 7 cases (24%) in the N group. There were signif icant differences in this analysis (P=0.03), and the subanalysis in the 43 cases of larger AVMs excluding AVMs less than 3 cm in maximum diameter also demonstrated a difference (P=0.05).
When we performed another subanalysis with another categorization, SRS-group A +B vs. group C, there was a significant difference in the rate of successful embolization (P\<0.002). There was no significant difference in the rate of procedure related complications (S group: 7/44 (16%), N group 7/29 (24%)) and the rate of morbidity between the two groups. Post-radiosurgery complications were not encountered in the S group and signif icantly higher in the N group (4/29(14%) (P=0.022)
Multivariate analysis between the size and embolization specification
---------------------------------------------------------------------
We performed a selective multivariate study focusing on AVM size and embolization specif ication. As important factors corresponding to either the S or N group, the mRBAS, nidus type, embolization grade and residual nidus size were selected and analyzed. Of these four factors, a significant difference was obtained in the nidus type (odds ratio 0.261, 95% CI: 0.073--0.938, P=0.04) and embolization grade (odds ratio = 4.397, 95% CI 2.113--9.149, P\<0.001).
DISCUSSION
==========
There have been many reports supporting the usefulness of embolization preceding surgical removal and SRS, which can make the subsequent treatment easier, safer and more successful \[[@B1][@B2][@B3][@B4][@B5][@B16]\]. By contrast, recent reports addressed the negative effect of preradiosurgical embolization. Embolization prior to SRS was associated with a lower rate of total obliteration than radiosurgery alone \[[@B12][@B14]\] and had no sufficient role in reducing the recurrence correlated with deep regions \[[@B10]\]. The main reasons for such a result might include the followings: 1) the embolization blurs the nidus margin and causes a targeting error, 2) a part of the nidus that disappears just after the proximal feeder occlusion due to the temporary flow regression is outside of the radiosurgery target and may later recanalize due to hemodynamic remodeling \[[@B11][@B12][@B13][@B14][@B20][@B21]\].
For the adverse events secondary to the radiosurgery, the rate of post-radiosurgery hemorrhage is not affected by preceding embolization \[[@B21][@B22]\]. There are conflicting reports on radiation-induced change; one shows no correlation and larger complications with preceding embolization \[[@B4][@B14][@B23]\]. Additionally, it is obvious that the larger AVM may result in a lower obliteration rate for SRS. A higher tendency of the larger nidus of the virgin state was observed in the N group in our study. Therefore, we performed a multivariate study to clarify the signif icance of specif ication or quality of the embolization independent of the original AVM size. The result suggested that a successful SRS is expected to increase by more than four times for each increase in embolization grade. Therefore, the embolization grade was a strongly independent factor that influenced the success of SRS, which may suggest that a proper and meticulous embolization strategy, highly skillful microcatheterization, and best performance of injecting liquid embolic materials are essential for achieving good results. The diffuse type of AVM makes it a challenge to perform the ideal embolization. Therefore, we should preoperatively evaluate the type of nidus, and cases without the possibility of suff icient embolization should be omitted from the plan of preradiosurgical embolization. By contrast, cases with high odds of embolization that affect the positive use of embolization will contribute to and complement the desirable result of SRS.
Although we did not compare the results between two groups with and without embolization, we showed that a high quality embolization is essential to achieve ideal radiosurgery results. The embolization of nidus is particularly important for avoiding recanalization and regional vascular remodeling, as mentioned in many previous reports from the past century \[[@B2][@B3][@B17]\]. However, when treating cases with a diffuse, large AVM in which effective embolization is difficult, we should consider novel combined treatment options. Yashar et al. \[[@B24]\] recommends that the compartments of an embolized AVM should be contained within the radiosurgery plan. According to this concept, volume-staged radiosurgery \[[@B25][@B26][@B27]\] and preembolization radiosurgery \[[@B11]\] may be useful.
This study is a retrospective case cohort study of the limited institutes employing operators with a reliable and proper technique and strategy for AVM embolization. Therefore, larger prospective studies will be needed to confirm the findings.
CONCLUSION
==========
Embolization before radiosurgery remains a controversial neoadjuvant therapy. It is clear that reducing the size and shunt flow of AVMs improves the effect of SRS. Although sufficient volume reduction should be the most important goal of embolization, false occlusion due to the temporary depression of hemodynamics may disappoint radiosurgeons. This study showed that proper embolization with a high rate of nidus penetration to avoid recanalization is important for complete, cooperative combined treatments. A proper strategy and technique is essential for promoting occlusion following SRS.
This study was supported from the grant of the 2016 Japanese Society of Neuroendovascular Therapy.
We thank Dr. Kenta Murotani (Clinical Research Center, Aichi Medical University) for statistical studies. We also thank Drs. Noriaki Matsubara and Ryo Hiramatsu (Department of Neurosurgery and Neuroendovascular Therapy, Osaka Medical College), Drs. Osamu Masuo and Kenji Kubo (Department of Neurosurgery, Wakayama Medical College), Dr. Rie Aoki (Department of Neurosurgery, Tokai University Hospital), Dr. Shuichi Tanoue (Department of Radiology, Kurume University of Medicine), Dr. Masayuki Sato (Department of Neurosurgery, Mito Medical Center), Dr. Eika Hamano (Department of Neurosurgery, National Cerebral and Cardiovascular Center) and Dr. Yosuke Tamari (Department of Neurosurgery and Endovascular Neurosurgery, Nagoya University Graduate School of Medicine) for assistance of data management.
###### Classification of the Grade of Embolization Performance (E Grade)
E Grade A1: ≥50% of successful nidus embolization for compact (plexiform) type AVM
----------------------------------------------------- ------------------------------------------------------------------------------------------------------------------------------------------------------------
E Grade A2: ≥50% of successful nidus embolization and proximal occlusion of fistulous feeders for mixed (plexiform + fistulous) type AVM
E Grade B1: \<50% of successful nidus embolization for compact (plexiform) type AVM
E Grade B2: \<50% of successful nidus embolization and proximal occlusion of fistulous feeders for mixed (plexiform + fistulous) type
AVM or remaining daughter nidus on diffuse type^\*^
E Grade C: No successful embolization (all proximal occlusion) for all feeders in any types of AVM or failed obvious size reduction (only change of vascular density)
^\*^: Including remnant daughter nidus on diffuse type
###### Patient Profile
Grade of radiosurgical effect (SRS grade) Total A B C B + C
------------------------------------------- ----------------------------- -------------- -------------- -------------- --------------- ---- ----
Patient profile
Gender (M:F) 40 vs. 33 28 vs. 16 10 vs. 8 2 vs. 9 12 vs. 17 NS
Mean age (range) 38.3 (6\~78) 38.5 (6\~78) 30.9 (10-58) 33.2 (13-62) 31.8 (10\~62) NS
Clinical Unruptured NS
Manifestation No symptom 10 6 2 2 4
Convulsion 10 4 4 2 6
Headache 5 4 1 0 1
Focal neurological symptoms 5 2 1 2 3
Ruptured NS
Intracerebral hemorrhage 20 14 4 2 6
Intraventricualr hemorrhage 16 9 4 3 7
Subarachnoid hemorrhage 7 5 2 0 2
Initial treatment NS
No treatment 42 25 9 8 17
Medical treatment 23 15 7 1 8
Ventricular drainage 4 2 2 0 2
Others 4 2 0 2 2
Precedent neurological deficits NS
None (mRS 0) 55 32 13 10 23
Mild (mRS 1) 9 5 3 1 4
Medium (mRS 2, 3) 3 2 1 0 1
Severe (mRS ≥4) 6 5 1 0 1
###### AVM Profile
Grade of radiosurgical effect (SRS grade) Total A B C B + C
------------------------------------------- ----------------------------- -------------- -------------- -------------- -------------- -------------- ------- ---------
AVM profile
location NS NS
Frontal 20 15 4 1 5
Temporal 8 4 2 2 4
Parietal 10 4 3 3 6
Occipital 11 6 2 3 5
Cerebellum 16 11 5 0 5
Thalamostriate 3 2 1 0 1
Brainstem 4 1 2 1 3
Spetzler-Martin grade NS NS
Grade 1 9 8 1 0 1
Grade 2 17 12 3 2 5
Grade 3 32 18 10 4 14
Grade 4 14 6 3 5 8
Grade 5 1 0 1 0 1
size maximum diameter (mm) 31.6 (11-72) 28.7 (12-55) 32.9 (11-45) 40.7 (20-72) 35.9 (11-72) NS 0.013
initial volume (ml) 13.8 (1-46) 12.0 (1-45) 12.4 (1-35) 23.5 (2-46) 16.6 (1-46) NS 0.004
≤5 ml 24 18 6 0 6
6-10 ml 15 11 3 1 4
11-20 ml 13 6 3 4 7
21-30 ml 10 4 4 2 6
31-40 ml 9 4 2 3 5
\>40 ml 2 1 0 1 1
mBRAS mean value 2.13 2.01 1.94 2.97 2.33 NS \<0.001
≤1.00 15 11 3 0 4
1.01-1.50 11 6 4 1 5
1.51-2.00 16 11 4 1 5
≥2.01 31 16 7 9 15
Angioarchitecture
Nidus type compact 52 37 12 3 15 0.004 \<0.001
diffuse 21 7 6 8 14
Daughter nidus 5 2 1 2 3 NS NS
Draining pattern single (superficial) 18 14 3 1 4 0.031 0.0013
single (deep) 11 8 3 0 3
multiple (superficial main) 32 17 8 7 15
multiple (deep main) 12 5 4 3 7
Associated varix 6 5 1 0 1 NS NS
Associated aneurysm no aneurysms 52 33 13 6 19 NS NS
proximal (treated) 13(5) 6(2) 3(0) 4(3) 7(3)
flow-related (treated) 3(3) 2(2) 1(1) 0 1(1)
intranidal (treated) 6(4) 3(3) 2(1) 1(0) 3(1)
Meningeal feeders no 56 38 16 4 18 0.023 0.002
yes (treated) 17(9) 6(4) 4(1) 7(4) 11(5)
mBRAS: modified radiosurgery-based arteriovenous malformation (AVM) score
###### Performance of Embolization
Grade of radiosurgical effect (SRS grade) Total A B C B + C
------------------------------------------- ------------------------------------ ------- ------- ------- ------- ------- ---------
Performance of embolization
Embolization grade E grade A1 12 12 0 0 0 \<0.001
E grade A2 20 18 2 0 2
E grade B1 19 8 7 4 13
E grade B2 18 6 5 7 12
E grade C 2 0 2 0 2
Final embolization rate 61.2% 63.6% 60.0% 53.6% 57.6% NS
Final results of embolization complete occlusion 7 6 1 0 1 NS
Size of remained nidus (diameter) \<1 cm 22 16 5 1 6
1-2 cm 24 14 8 2 10
2-3 cm 13 3 4 6 10
≥3 cm 7 5 0 2 2
Procedure-related complications total 14 7 5 2 7 NS
hemorrhage (inta- & perioperative) 8 6 2 0 2
convulsion 2 0 1 1 2
others 4 1 2 1 3
Influence of complication on outcome none 11 7 2 2 4 NS
mRS 1 rank down 1 0 1 0 1
mRS 2 rank down 1 0 1 0 1
mRS 3 rank down 1 0 0 1 1
###### Performance of Radiosurgery
Grade of radiosurgical effect (SRS grade) Total A B C B + C
------------------------------------------- -------------------------- --- --- --- ------- --- -------
Performance of radiosurgery
Post-radiosurgery complications total 4 0 1 3 4 0.022
hemorrhage, rebleeding 3 0 1 2 3
delayed cystic formation 1 0 0 1 1
Influence of complication on outcome none 0 0 0 0 0 0.022
mRS 1 rank down 2 0 0 2 2
mRS 2 rank down 1 0 1 0 1
mRS 3 rank down 1 0 0 1 1
| {
"pile_set_name": "PubMed Central"
} |
1. Introduction {#sec1}
===============
Peroneal tendon (PT) disorders are common causes of lateral and retromalleolar ankle pain \[[@B1]--[@B3]\]. PT injuries include tenosynovitis, chronic tendinopathy, subluxation and dislocation, longitudinal splits, partial or complete tears, and painful os peroneum syndrome \[[@B4]--[@B6]\]. Chronic conditions and anatomic factors have been implicated as causes of abnormalities such as chronic lateral ankle instability, cavovarus foot, low-lying peroneus brevis (PB) muscle belly, and peroneus quartus tendon \[[@B7]--[@B10]\].
The current options of treatment for PT injuries include the following: (1) nonoperative treatment, (2) peroneal tendoscopy, (3) opened debridement and tubularization of the remaining tendon, (4) tenodesis, (5) tendon transfer of the flexor hallucis longus or flexor digitorum longus, and (6) reconstruction with allograft or autograft \[[@B6], [@B11]--[@B13]\]. When the tears are irreparable, a salvage procedure is indicated with segmental resection followed by reconstruction with tenodesis, tendon transfer, or bridging the defect using allograft or autograft. However, it is unclear which of these treatment options provides the best outcomes for PB tendon tears \[[@B12]--[@B15]\].
Reconstruction with semitendinosus allograft has produced satisfactory clinical results and is effective for pain relief and restoration of tendon function \[[@B1], [@B8]\]. However, there are concerns associated with the use of the allograft, which include its cost and availability, disease transmission, delayed incorporation, and stretching of the graft. The use of hamstring autograft, on the other hand, represents a viable and accessible option that may be biologically superior \[[@B15]\].
The purpose of this study is to describe our surgical technique for the reconstruction of irreparable PB tendon tears using semitendinosus tendon autograft (STA) as an alternative to the allograft and report the short-term results of three cases.
2. Materials and Methods {#sec2}
========================
This study reports 3 patients submitted to the reconstruction of the PB tendon tears using STA, from December 2016 to May 2017. Ethical approval was granted by our hospital\'s HSPM/Ethics Committee, and the study was registered at Clinical Trials National Registry under number 2.880.187. The indication for reconstruction using STA was irreparable tears of the PB tendon. Irreparable tears are defined as the presence of a degenerative tissue associated to longitudinal tears that involves more than 50% of the cross-sectional area of the tendon \[[@B9], [@B10]\].
2.1. Preoperative Planning {#sec2.1}
--------------------------
A precise clinical evaluation and radiological exams were performed. Clinically, we searched for symptoms and signs of chronic conditions associated to PT injuries such as lateral ankle instability and cavovarus foot. Radiological exams included radiography imaging (RI) and magnetic resonance imaging (MRI). RI included all three views of the ankle as well as Saltzman view to assess hindfoot alignment. Stress X-ray views of the ankle in varus and anterior drawer test were also performed to exclude lateral ankle instability. In the MRI, we evaluated the extent of the PB tendon pathology and possible associated anatomical factors such as a low-lying muscle belly of the PB and peroneus quartus tendon (Figures [1(a)](#fig1){ref-type="fig"} and [1(b)](#fig1){ref-type="fig"}).
2.2. Postoperative Follow-up {#sec2.2}
----------------------------
At 6 months, patients were submitted to an isokinetic evaluation of the strength of both feet in eversion and inversion using an isokinetic dynamometer (CSMI HUMAC Norm, Stoughton, Massachusetts, USA).
3. Patients Presentations {#sec3}
=========================
In our three patients---(I) a 31-year-old man, (II) a 67-year-old woman, and (III) a 40-year-old man---the mechanism of injury consisted of an ankle sprain. In patient I, the sprain occurred during a soccer match and in patients II and III it occurred while walking in the sidewalk. They went to the outpatient clinic complaining of lateral pain at the hindfoot for 24, 18, and 16 months, respectively. Previous treatments elsewhere were based only on anti-inflammatory drugs, ice, and rest. On a physical exam, there were pain and swelling over the course of the PT. In patients I and II, there were no clinical signs of ankle instability or varus of the hindfoot. In patient III, a bilateral cavovarus deformity was observed. No restriction of the range of motion of the subtalar joint was observed in any of them. Radiographic images were normal. In all MRI images, there were irreparable tears of the PB tendon and anatomical conditions were noted in patients I and II, such as low-lying muscle belly of the PB tendon and a peroneus quartus tendon, respectively. Initially, we conducted a conservative treatment for six months with physiotherapy, rest, analgesics, and ankle stabilizer to restrict inversion-eversion movements, but it was proved unsuccessful.
4. Surgical Technique {#sec4}
=====================
The illustrative case (patient II) was presented for the demonstration of the surgical technique (Figures [2](#fig2){ref-type="fig"}[](#fig3){ref-type="fig"}[](#fig4){ref-type="fig"}[](#fig5){ref-type="fig"}[](#fig6){ref-type="fig"}[](#fig7){ref-type="fig"}[](#fig8){ref-type="fig"}--[9](#fig9){ref-type="fig"}). The surgery was performed with the patient placed in an oblique lateral decubitus under regional anesthesia with a nonsterile thigh tourniquet on a radiolucent operating table. The STA was harvested through a medial longitudinal incision of 3 cm at the region of the proximal lower leg with the hip externally rotated to provide a frontal view of the knee. The graft was prepared by resection of the muscle belly; then its two stumps were tubularized with a 1-0 Vicryl whip stitch. We kept the STA in its full length to ensure that the entire defect was filled after the resection of the unhealthy PB tendon ([Figure 2](#fig2){ref-type="fig"}).
A lateral curved incision over the course of the PT was performed along the posterior border of the lateral malleolus, from 3 to 4 cm proximal to the tip of the fibula extending to the fifth metatarsal base ([Figure 3](#fig3){ref-type="fig"}). During dissection, care was taken to avoid damage to the sural nerve branches inferior to the lateral malleolus. The peroneal tendon sheath and the superior peroneal retinaculum (SPR) were opened, and PB and peroneal longus (PL) tendons were exposed. Dissection was performed proximally and extended distally to isolate the compromised portion of the tendon ([Figure 4](#fig4){ref-type="fig"}). The PB was assessed, and the nonviable portion was resected. The distal stump was debrided and totally removed to prevent local pain due to a bulky suture of the STA to the remaining distal stump, under a thin skin. The STA was sutured to the proximal stump of the native PB tendon using a Pulvertaft suture with a 1-0 Vicryl ([Figure 5](#fig5){ref-type="fig"}). The suture was performed 3 cm above the tip of the lateral malleolus to prevent the volume effect of increased pressure within the retromalleolar groove. This distance was based on previous studies which recommend to place peroneal brevis tendon tenodesis to the peroneus longus to avoid pain due to the entrapment of the suture \[[@B14]\]. The distal fixation of the STA was carried out through a bone tunnel at the fifth metatarsal base to provide a bone-to-tendon fixation. Since we kept the full length of the STA, there was an adequate distal stump length remaining to fix it distally ([Figure 6](#fig6){ref-type="fig"}). The bone tunnel was drilled with a 3.2 mm drill ([Figure 7](#fig7){ref-type="fig"}) perpendicular to the long axis of the fifth metatarsal, from plantar to dorsal. The distal stump was pulled from plantar to dorsal through the tunnel and sutured back to itself with a 1-0 Vicryl ([Figure 8](#fig8){ref-type="fig"}). Alternatively, it can be fixed with a biotenodesis screw or an anchor. During the suture, the foot is positioned in neutral inversion/eversion and dorsiflexion/plantarflexion, and the tendon graft was tensioned at 50% of the maximum excursion of the PB muscle belly. The length of the reconstruction was determined at this point. The deep tissues, SPR, and skin were closed in layers. The closure of the SPR was made carefully to prevent PT subluxation. Finally, a sterile dressing and a short leg cast were applied.
4.1. Postoperative Rehabilitation {#sec4.1}
---------------------------------
Patients remain non-weight-bearing for two weeks with the cast. At 2 weeks, sutures are removed and they are placed in a walking boot (WB) with full weight-bearing as tolerated. At this period, physical therapy is initiated focusing in the dorsiflexion/plantarflexion range of motion to prevent adhesions in the tendon graft. Inversion-eversion movements are prohibited to prevent stretching of the healing tendon graft and the subsequent development of an elongated tendon with the loss of strength. Patients are instructed to always maintain the WB except for hygiene purposes and dorsiflexion/plantarflexion exercises. The patient who underwent the calcaneal osteotomy followed the same postoperative protocol as the others. At 8 weeks postoperatively, the WB is removed and the patient is transitioned into an ankle-stabilizing orthosis. A physical therapy program is oriented to start inversion-eversion movements and to progressively restore proprioception and strengthening. The ankle-stabilizing orthosis is used progressively less accordingly to the patient rehabilitation.
5. Results {#sec5}
==========
In the postoperative period, we noted no skin necrosis, wound dehiscence, autograft rupture, or any associated complications. The three anatomical conditions of the patients associated to the lesions were addressed at the same time, which included a PB low-lying muscle belly resection, a peroneus quartus tendon resection, and a lateral sliding calcaneal osteotomy for a cavovarus deformity.
After 3 months, all patients were pain free, both in the foot and at the donor site, and were able to resume labor activities. After 6 months, all patients could perform a single-heel rise and there was no restriction of the range of motion for inversion (Figures [9(a)](#fig9){ref-type="fig"} and [9(b)](#fig9){ref-type="fig"}). At this point, the isokinetic strength of both feet in eversion and inversion was assessed. The operated feet showed no strength deficit compared to the contralateral side. Eversion strength in patient I was 4%, patient II 4%, and patient III 2% stronger than the contralateral side. At this time, they were allowed to return to sports activities. At a mean follow-up of eighteen months, they were still asymptomatic and fully active.
6. Discussion {#sec6}
=============
In 1998, Krause and Brodsky were the first authors to propose a classification system to guide the treatment of irreparable tears of the PT. If less than 50% of the cross-sectional area of the tendon was viable, segmental resection and tenodesis were performed \[[@B7], [@B9]\]. Although tenodesis is a simple procedure, its clinical outcomes may be unpredictable. Almost two-thirds of the patients report pain on activities and almost 50% of the patients cannot resume full activities. Furthermore, tenodesis sacrifices the functional integrity of the muscle-tendon unit \[[@B16]\].
In 2010, Nunley and Ousema reported for the first time a technique for the management of irreparable tears of the PB tendon using a tendon allograft instead of tenodesis, tendon transfer, or a two-stage procedure. They performed this procedure in 4 patients. A PT allograft was used for the reconstruction of defects greater than 4 cm. The authors presented it as an effective treatment option with satisfactory outcomes. There were no complications associated with this procedure, and all their patients returned to a fully active life \[[@B17]\]. In a retrospective series of 14 patients who underwent a PT (PB and PL tendons) reconstruction with a peroneal or a semitendinosus allograft, Mook et al. showed that all patients returned to their preinjury activity level without pain and yielded satisfactory patient-reported outcomes \[[@B8]\].
Although there is no evidence in which surgical procedure provides the best outcomes, reconstructing the PB tendon with an allograft or autograft seems to be biomechanically superior comparing to tenodesis. Using cadaveric models, Pellegrini et al. compared the effectiveness of allograft reconstruction and tenodesis. They concluded that reconstruction of the PB tendon with allograft substantially restores distal tension when the PT were loaded to 50% and 100% of physiologic load. Tenodesis substantially decreases PB tension under both loads \[[@B1]\].
The literature regarding PB reconstruction with tendon replacement using autograft is scarce. Ellis and Rosenbaum were the first authors to describe a surgical technique for the reconstruction of the diseased PB tendon with STA, but without any clinical results \[[@B15]\]. Although tendon autograft harvest may cause morbidity at the donor site and require a second incision, there are advantages such as tissue compatibility, faster reincorporation, and remodeling rate comparing with the allografts. Recently, Cody et al. published a cohort study of 37 patients with respect to hamstring applications in foot and ankle surgery. Their objective was to evaluate muscle balance and strength of the knee after hamstring autograft harvest with an isokinetic testing. The result was that flexion strength deficit was noted only at the higher degree of flexion of the knee, but without clinical or functional impairment \[[@B18]\]. None of our three patients had residual pain at the donor site, and they were all able to return to their preinjury level of activities after 6 months. Allografts are widely used, but there are concerns including potential disease transmission, risk of immune response, timely incorporation of the graft to the host site, stretching of the graft, and higher costs \[[@B19]\].
To our knowledge, this is the first study to present the surgical technique of reconstruction of the PB tendon using STA with its clinical results. Furthermore, this is the first report to present an objective muscle force measurement during follow-up. Nevertheless, this study has limitations, mainly the small population and the short-term follow-up.
The present study suggests that reconstruction of the PB tendon with STA may be an effective alternative technique to allograft tissues for PB tendon tears. This procedure can decrease pain and restore PB strength without altering foot function. Further studies with larger populations, longer follow-up, and comparisons of PB reconstructions using autograft and allograft are needed to establish the best treatment for these injuries.
PT:
: Peroneal tendon
PB:
: Peroneus brevis
STA:
: Semitendinosus tendon autograft
RI:
: Radiography imaging
MRI:
: Magnetic resonance imaging
WB:
: Walking boot.
Conflicts of Interest
=====================
The authors have no conflict of interest in the preparation of this work. The authors did not receive directly or indirectly any financial benefit in the preparation of this work.
Authors\' Contributions
=======================
All authors have contributed directly to this study and the resulting paper, including the conception of the idea, planning, data collection with clinical and functional analysis, and editing.
{#fig1}
{#fig2}
{#fig3}
{#fig4}
{#fig5}
{#fig6}
{#fig7}
{#fig8}
{#fig9}
[^1]: Academic Editor: Johannes Mayr
| {
"pile_set_name": "PubMed Central"
} |
Introduction
============
Eye drops account for \~90% of ophthalmic drug delivery systems available ([@b1-etm-0-0-5586]). Traditional eye drops are usually lost from the ocular surface by tear washing or nasolacrimal drainage immediately after administration, with poor ocular bioavailability and numerous adverse reactions ([@b2-etm-0-0-5586]). In addition, due to the inconvenience of night administration, the pharmacological peak-valley phenomenon is prominent, greatly affecting the treatment effect of ocular drugs. With the development of novel technologies, novel ophthalmic drug delivery systems have been designed to effectively elevate ocular bioavailability by prolonging the retention time of drugs in the eye or improving the penetration of drugs into the cornea and conjunctiva ([@b3-etm-0-0-5586],[@b4-etm-0-0-5586]).
Chitosan is a natural cationic polysaccharide that is prepared from chitin after partial or complete deacetylation. Due to its excellent biological properties, chitosan and its derivatives have been widely used in ophthalmic drug delivery systems ([@b5-etm-0-0-5586]). As a novel form of administration, *in situ* gel undergoes phase transition to a non-chemically crosslinked semi-solid gel preparation at the medication site immediately after administration in solution or semi-solid format. *In situ* gel prepared from chitosan or its derivatives forms a transparent gel that has a strong affinity to the ocular surface under ocular physiological pH conditions. Certain chitosan derivatives are temperature-sensitive and may be prepared to form a temperature-sensitive gel *in situ* ([@b6-etm-0-0-5586],[@b7-etm-0-0-5586]). The temperature-sensitive gel formation is due to changes in the physical state of polymers induced by changes in hydrogen bonds or hydrophobic interactions ([@b8-etm-0-0-5586]--[@b11-etm-0-0-5586]). Cao *et al* ([@b12-etm-0-0-5586]) synthesized a novel type of isopropyl acrylamide chitosan polymer and used it to prepare a timolol maleate temperature-sensitive *in situ* gel, which provides a higher bioavailability than its respective solution.
At present, certain challenges remain regarding the development of *in situ* gels for pharmaceutical application. First, higher concentrations of polymers and non-aqueous solvent in the gel may cause eye irritation and safety issues. Furthermore, there may be a drug burst between the time-points of dripping the formulation into the eye and the formation of the gel, leading to a significant peak-valley phenomenon ([@b13-etm-0-0-5586],[@b14-etm-0-0-5586]). In the present study, a temperature-sensitive gel was prepared using chitosan, carboxymethyl chitosan and glycerophosphate, and levofloxacin was encapsulated into chitosan microspheres, which were embedded in the chitosan temperature-sensitive gel. In addition, the safety of chitosan temperature-sensitive gel carrying drug microspheres and the bioavailability they provided were evaluated.
Materials and methods
=====================
### Preparation of chitosan microspheres
Chitosan acetic acid solution (2.0 g/l; Biotemed, Qingdao, China) was prepared and adjusted to pH 5.0. Subsequently, 20 mg levofloxacin (Wuhan DKY Technology, Wuhan, China) was dissolved in 100 ml chitosan acetic acid solution with agitation at 10.06--13.42 × g. Subsequently, sodium tripolyphosphate (1.0 g/l; Tianjin Bodi Chemical, Co., Ltd., Tianjin, China) was added to obtain a suspension after agitation for 15 min. Levofloxacin chitosan microspheres (2.4 g/l) were then obtained by filtration.
### Determination of entrapment rate of levofloxacin chitosan microspheres
A 200 mg/l-stock solution of levofloxacin (20 mg) in 0.1 M acetic acid solution was prepared. A series of solutions with mass concentrations of levofloxacin of 4, 6, 8, 10, 12, 14, 16 and 18 mg/l were prepared and the absorbance at each concentration was measured at 293 nm (UV2 9200; Nanjing Xinhang Scientific Instrument Co., Ltd., Nanjing, China). Levofloxacin chitosan microspheres (2.0 ml) were centrifuged at 671 × g at 4°C for 30 min. Supernatants were obtained and diluted with 0.1 M HCl. The concentration of free levofloxacin (C~free~) in the suspension was determined. In the meantime, levofloxacin chitosan microspheres (2.0 ml; 2.4 g/l) which was the reaction solution from above were suspended in 8.0 ml 0.1 mol/l HCl solution and incubated at 60°C for 1 h for hydrolysis. After filtration using a microfiltration membrane (pore size, 0.02 µm; EMD Millipore, Billerica, MA, USA), the total concentration of levofloxacin in the filtrate (C~total~) was determined. The entrapment rate of levofloxacin chitosan microspheres was calculated using the following equation: Entrapment rate = (1 - C~free~/C~total~) × 100%. Each test was repeated for 5 times.
### Characterization of the physical and chemical properties of levofloxacin chitosan microspheres
A suspension of levofloxacin chitosan microspheres was used to determine the pH, ζ potential, particle diameter (Zetasizer 3000 HSa; Malvern Instruments, Malvern, UK) and particle size distribution. After diluting the suspension of levofloxacin chitosan microspheres, a scanning electron microscope (JEOL7500F; Jeol, Tokyo, Japan) was used to observe the morphology of the particles. In addition, an infrared spectrum was used to analyze pure chitosan powder, blank chitosan microspheres and levofloxacin chitosan microspheres. The scanning range was 400--4,000 cm^−1^ and the resolution was 4 cm^−1^.
### Preparation of chitosan-carboxymethyl chitosan temperature-sensitive gel loaded with chitosan nanoparticles
Sodium glycerophosphate (0.78 g) and carboxymethyl chitosan (0.2 g) were dissolved in 2 and 3 ml double-distilled (dd)H~2~O, respectively. Furthermore, chitosan (0.2 g) was dissolved in 5 ml 0.1 M acetic acid. With mixing, sodium the glycerophosphate solution and carboxymethyl chitosan solution were consecutively added to the chitosan solution to thereby obtain the chitosan-carboxymethyl chitosan temperature-sensitive gel system. Subsequently, 0.1 g levofloxacin chitosan microspheres were added to 10 ml chitosan-carboxymethyl chitosan temperature-sensitive gel, followed by thorough mixing.
### Evaluation of in vitro release of levofloxacin by chitosan temperature-sensitive gel loaded with drug microspheres
In the different experimental groups, the drug delivery systems were prepared via three different methods as follows: Group 1, levofloxacin was added into chitosan-carboxymethyl chitosan solution; group 2, levofloxacin was added into chitosan-carboxymethyl chitosan-sodium glycerophosphate solution; group 3, temperature-sensitive gel containing levofloxacin nanoparticles was prepared as described above. The liquid samples were put into dialysis bags (molecular weight cut-off, 7,000 Da), which were soaked in 200 ml phosphate-buffered saline (pH 7.4) at 37°C. After standing still for 20 sec, 5 ml phosphate buffer solution was transferred to a sample tube. After the formation of a solid gel in dialysis bag at 37°C, 5 ml phosphate buffer solution was also transferred to individual sample tubes at 5, 10, 15, 20, 30, 60, 120, 180, 240, 300, 360, 420, 480, 540, 600, 660 and 720 min. Each time after withdrawal of phosphate buffer solution, another 5 ml of fresh solution was replenished. For group 1, in which no gel was formed, phosphate buffer solution was also aspirated at the abovementioned time-points. The absorbance of levofloxacin in each sample tube was measured at 293 nm (UV2 9200; Nanjing Xinhang Scientific Instrument Co., Ltd.). Each sample was tested in triplicate. Using the levofloxacin standard curve, the concentrations of levofloxacin from each group at the different time-points were calculated and levofloxacin *in vitro* release curves were plotted.
### Eye retention test of chitosan temperature-sensitive gel
The experimental animal protocols of the present study were approved by the Ethics Committee of Qingdao Agricultural University (Qingdao, China). On ice, fluorescein sodium (mass fraction, 0.05%) was added into a levofloxacin drug loading system in groups 1--3. A total of 18 healthy New Zealand white rabbits (Qingdao Medicine Inspecting Institute, Qingdao, China; age, 8 months; weight, 1.5--2.0 kg) were divided into 3 groups of 6 rabbits. Rabbits had free access to food and water in a temperature-controlled environment of 25°C (45--65% humidity) and a 12-h light/dark cycle. Temperature-sensitive gel in pure form was dripped into the conjunctival sac on the left side, and the eye on the right side was used as a blank control. After drug administration, the eyes were closed under local anesthesia in accordance with experimental animal welfare guidelines. The continuous fluorescent layer of the corneal surface was observed under a slit lamp at regular intervals. The time of disappearance of the continuous fluorescent layer on the corneal surface following application of the gel was considered to be the ocular retention time of levofloxacin.
### Cell compatibility test
The chitosan-based temperature-sensitive hydrogel was prepared as a preparation of chitosan-carboxymethyl chitosan temperature-sensitive gel loaded with chitosan nanoparticles, and the thickness of the hydrogel was controlled at 1--2 mm, which could be regarded as a membrane. The membrane was cut into circular sheets with a diameter of 1.1 cm. The circular sheets were soaked in 25, 50 and 75% ethanol for 2--3 h each time and then with 75% ethanol for 1--2 days, followed by transfer to sterile D-Hank\'s buffer for soaking over 10 h with the buffer being replaced 4--5 times. The membranes were then soaked in D-Hank\'s buffer prior to use. The membranes were placed into 48-well cell culture plates containing 100 µl Dulbecco\'s modified Eagle\'s medium (DMEM; Tianjin Bodi Chemicals, Tianjin, China) supplemented with 10% fetal bovine serum. Each time-point had 5 repeats. Subsequently, rabbit corneal endothelial cells ([@b15-etm-0-0-5586]) were trypsinized and adjusted to a concentration of 5×10^5^/ml. Cell suspension (200 µl) was added into predefined wells. The negative control contained DMEM only. The cells were cultured at 37°C and 5% CO~2~. On days 3 and 5, the cells were observed and images were captured. The cells were then trypsinized and transferred to 96-well plates, followed by incubation over 12 h. Subsequently 5 mg/ml MTT (20 µl) was added to each well, followed by further incubation for 4 h. After discarding the medium, dimethyl sulfoxide (DMSO; 150 µl) was added and the wells were incubated at 37°C for 10 min. After agitation for 40 sec, the absorbance of each well was measured at 490 nm on an UV2 9200 plate reader (Nanjing Xinhang Scientific Instrument Co., Ltd.).
### Cytotoxicity test
The L929 mouse fibroblast cell line at the logarithmic growth phase was trypsinized, the suspension was adjusted to a concentration of 1×10^4^/ml, and the cells were seeded onto 96-well plates (200 µl/well). After incubation at 37°C and 5% CO~2~, the medium was replaced with DMEM that had been used to soak the chitosan-based temperature-sensitive hydrogel membrane for 24 h in the experimental group. For the control group, normal medium was added. Each group had 12 repeats. On days 3 and 5, the cells were observed under a microscope. Subsequently, 20 µl MTT solution (5 mg/ml) was added into each well and the cells were incubated for 4 h under normal culturing conditions. After discarding the medium, DMSO (150 µl) was added, followed by incubation at 37°C for 10 min. After shaking for 40 sec, the absorbance of each well was measured at 490 nm (UV2 9200; Nanjing Xinhang Scientific Instrument Co., Ltd.). The relative growth rate (RGH) was calculated as follows: RGH = mean absorbance in experimental group/mean absorbance in control group × 100%. Cytotoxicity grading was as follows: 0, RGH≥100%; 1, 75%\<RGH\<95%; 2, 50%\<RGH\<74%; 3, 25%\<RGH\<49%; 4, 1%\<RGH\<24%; and 5, RGH=0%. For grade 1, the cytotoxicity was considered acceptable; for grade 2, cytotoxicity was evaluated in combination with cell morphological analysis; for grades 3--5, cytotoxicity was considered unacceptable.
### Skin irritation test
Sterile and pyrogen-free chitosan-based temperature-sensitive hydrogel membrane (5 g) was soaked in 5 ml saline for 72 h. At 4--24 h prior to the experiments, the hair of three healthy rabbits was removed on both sides of the dorsal spine. Six points were selected on each side of the dorsal spine. The six points on the left side were included in the control group and those on the right side were included in the experimental group. Saline (0.1 ml) was injected subcutaneously into the six points of the control group, while chitosan-based temperature-sensitive hydrogel membrane extract liquid (0.1 ml) that was DMEM soaked with the gel for 24 h, was injected subcutaneously into the six points of the experimental group. At 12, 24 and 48 h after injection, each injection point of each rabbit was evaluated according to [Table I](#tI-etm-0-0-5586){ref-type="table"} to determine the artificial vascular stimulation fraction. On days 3, 5 and 7 after surgery, subcutaneous tissues were obtained from all animals under anesthesia with 1 mg/kg sodium pentobarbital and subjected to hematoxylin and eosin (H&E) staining. The samples were paraffin-embedded, sliced, and washed with xylene for 15 min twice, 100% ethanol for 5 min twice, 80% ethanol for 5 min once, and ddH~2~O for 5 min once, followed by staining with hematoxylin at 25°C for 5 min. After washing with water for 1--3 sec, the samples were washed with 1% hydrochloric acid (1 ml) and 75% ethanol solution (99 ml) for 1--3 sec prior to washing with water for 10--30 sec. After washing with ddH~2~O for 1--2 sec, the samples were stained using 0.5% eosin at 25°C for 1--3 min. Subsequently, the samples were rehydrated/dehydrated in a graded ethanol series. The samples were then washed with xylene for three times of 2 min and mounted with neutral balsam.
### Statistical analysis
The results were analyzed using SPSS v17.0 software (SPSS, Inc., Chicago, IL, USA). Values are expressed as the mean ± standard deviation. Differences were compared using Student\'s t-test. P\<0.05 was considered to indicate a statistically significant difference.
Results
=======
### Levofloxacin chitosan microspheres have an entrapment rate of 26.5±1.31%
To determine the entrapment rate of levofloxacin chitosan microspheres, the measurement data were subjected to regression analysis, resulting in the following standard equation: A=0.08752+59.951C (r=0.9998 when C\<100 mg/l) with A being the absorbance and C the mass concentration (mg/l). The measurement data were the absorbance values of the levofloxacin standard solutions at different concentrations. The linear range of the standard curve was 6--16 mg/l. Using this equation, the concentration of each diluted sample was calculated and then converted to the concentrations prior to dilution to obtain C~free~ and C~total~. The mean entrapment rate of the levofloxacin chitosan microspheres was calculated from the measurement data of five experiments ([Table II](#tII-etm-0-0-5586){ref-type="table"}). The result suggested that the entrapment rate of levofloxacin chitosan microspheres was 26.5±1.31%.
### Levofloxacin chitosan microspheres formed by chitosan and sodium tripolyphosphate have physical and chemical properties rendering them a suitable ophthalmic particle dispersion system
To characterize the levofloxacin chitosan microspheres regarding their physical and chemical properties, the pH and ζ potential, as well as the particle diameter and distribution were determined, and a scanning electron microscope was used to observe the morphology of the particles. In addition, an infrared spectrum was recorded to analyze pure chitosan powder, blank chitosan microspheres and levofloxacin chitosan microspheres. The results indicated that the pH of levofloxacin chitosan microsphere suspension was 5.87±0.04, the average particle diameter was 2,452±342 nm, the polydispersity index was 0.168±0.028 and the ζ potential was 28.62±1.7 mV. Scanning electron microscopy revealed that the shape of the levofloxacin chitosan microspheres was almost spherical ([Fig. 1](#f1-etm-0-0-5586){ref-type="fig"}). The infrared spectrum demonstrated the existence of chitosan by intense absorbance peaks at 1,063 cm^−1^ (C-O in hydroxyl or cyclic ether) and 3,363 cm^−1^ (O-H and N-H of polysaccharides) ([Fig. 2A](#f2-etm-0-0-5586){ref-type="fig"}). In addition, the formation of microspheres by chitosan and sodium tripolyphosphate was indicated by peaks at 1,399 cm^−1^ (amino acid cross-linked by ion) and 592.9 cm^−1^ (interaction between phosphate and chitosan amino site) ([Fig. 2B](#f2-etm-0-0-5586){ref-type="fig"}). These results indicate that levofloxacin chitosan microspheres that are formed by chitosan and sodium tripolyphosphate are suitable for use as an ophthalmic particle dispersion system due to their physical and chemical properties.
### Chitosan temperature-sensitive gel containing microspheres loaded with drug prevent drug burst release at the initial stage and facilitates the slow release of the drug later on
To evaluate the *in vitro* release of levofloxacin by the chitosan temperature-sensitive gel, three different groups were used with different sample preparations. In group 1, a great amount of drug was released at the initial 20 sec and the drug burst release was dramatic within the first 10 min. For group 2, marked drug release was also observed within the first 20 sec, but later on, the drug was slowly released with time. For group 3, the slow release of the drug was similar to that in group 2 after 15 min, but the burst release of the drug was effectively prohibited within the first 15 min ([Fig. 3](#f3-etm-0-0-5586){ref-type="fig"}). These results suggest that chitosan temperature-sensitive gel containing microspheres loaded with drug prevents drug burst release at the initial stage and facilitates the slow release of the drug later on.
### Chitosan temperature-sensitive gel prolongs the contact duration of levofloxacin with the eye
To examine the eye retention of chitosan temperature-sensitive gel, fluorescein sodium was added into the levofloxacin drug loading system in groups 1--3. The results indicated that the retention time of chitosan temperature-sensitive gel in groups 1--3 was 30±2, 480±25 and 540±36 min, respectively, which was consistent with the results of the *in vitro* release assay (data not shown). This result indicated that the chitosan temperature-sensitive gel markedly prolongs the contact duration of levofloxacin with the eye.
### Chitosan temperature-sensitive gel is safe and provides good biocompatibility
To evaluate the safety of chitosan temperature-sensitive gel and the biocompatibility it provides, a cell compatibility test, a cytotoxicity test and an ocular skin irritation test were performed. The corneal endothelial cells grown on top of the hydrogel in the control and experimental groups had pebble-like shapes and had formed dense monolayers by day 5 ([Fig. 4A](#f4-etm-0-0-5586){ref-type="fig"}). An MTT assay demonstrated that the number of corneal endothelial cells that were treated with DMEM soaked with the hydrogel in the experimental group was insignificantly higher than that in the control group on days 3 and 5 (P\>0.05), suggesting that chitosan-based hydrogel membrane had good cytocompatibility ([Fig. 4B](#f4-etm-0-0-5586){ref-type="fig"}). Microscopic observation indicated that the growth of L929 cells in the control and experimental groups was not significantly different ([Fig. 5A](#f5-etm-0-0-5586){ref-type="fig"}). Consistently with this, the MTT cytotoxicity assay indicated that the absorbance in the experimental group was insignificantly higher than that in control group (RGH≥100%), suggesting that the chitosan temperature-sensitive gel had a cytotoxicity of grade 0 ([Fig. 5B](#f5-etm-0-0-5586){ref-type="fig"}). The skin irritation test indicated that the control and experimental groups had nearly no congestion phenomenon and the experimental group scored 0. After stimulation for 48, 96 or 144 h, the rabbits in the two groups did not display any behavioral abnormalities. No erythematous response was observed during the entire course of the experiment. H&E staining of the injection sites on days 3, 5 and 7 was not indicative of any inflammatory response ([Fig. 6](#f6-etm-0-0-5586){ref-type="fig"}), suggesting that chitosan temperature-sensitive gel caused nearly no skin irritation. These results suggested that chitosan temperature-sensitive gel is safe and provides good bioavailability.
Discussion
==========
Chitosan has been widely assessed for use in drug delivery systems due to its biocompatibility, safety and other unique properties. In addition, it is included in the usage standards for food additives as a thickening or coating agent ([@b16-etm-0-0-5586]). At present, research on chitosan and its derivatives in ophthalmic drug delivery systems still mainly focus on improving the solubility of chitosan, increasing the drug load and enhancing the stability of preparations ([@b17-etm-0-0-5586]). However, only few studies have focused on the absorption, transportation, distribution and metabolization of drugs after administration ([@b13-etm-0-0-5586]). In the present study, the optimal proportions of the major components and manufacturing process of chitosan temperature-sensitive gel were determined, and the physical and chemical properties, the slow-release effect and the safety of chitosan temperature-sensitive gel loaded with drug-containing nanoparticles were investigated.
The pH value of artificial tears is \~8 and their addition is equivalent to that of crosslinking agent. As a result, the temperature-sensitive system easily solidifies into a gel ([@b17-etm-0-0-5586],[@b18-etm-0-0-5586]). In the present study, a temperature-sensitive *in situ* gel loaded with levofloxacin-containing microspheres was prepared. Chitosan microspheres easily attach to cornea and have a slow-release effect. Therefore, the retention time of drugs on the cornea is prolonged, the local bioavailability of drugs on the eye surface is enhanced and adverse effects of the drugs are reduced ([@b19-etm-0-0-5586],[@b20-etm-0-0-5586]).
During the preparation of levofloxacin chitosan microspheres via the ionic gelation method in the present study, levofloxacin molecules become surrounded by a network structure formed by chitosan and sodium tripolyphosphate ([@b6-etm-0-0-5586],[@b21-etm-0-0-5586],[@b22-etm-0-0-5586]). The particle size of the microspheres is in a range that meets the requirements of an ophthalmic particle dispersion system. The measured particle size determined by a particle sizing meter in the present study is the size of hydrated particles, while that determined by scanning electron microscopy is the size of dried particles. As a result, the particle size determined by scanning electron microscopy was smaller. In addition, the zeta potential suggests that the system is stable, and guaranteed the quality of the prepared temperature-sensitive *in situ* gel.
Chitosan nanoparticles have biological adhesion properties and may attach to conjunctival cells ([@b9-etm-0-0-5586],[@b23-etm-0-0-5586],[@b24-etm-0-0-5586]). As a result, the tight junction between epithelial cells of the cornea and conjunctiva may be transiently opened to increase the permeability of the drug in the cornea and conjunctiva. The *in vitro* slow release and ocular retention experiments demonstrated that chitosan temperature sensitive gel loaded with levofloxacin microspheres prolonged the contact duration between the microspheres and eyes, and ensured the steady and sustained release of the drug. Chitosan temperature-sensitive hydrogel must not have any adverse effects on the organizational structure of organisms when being in contact with tissues *in vivo* ([@b25-etm-0-0-5586],[@b26-etm-0-0-5586]). The cell compatibility test, the cytotoxicity test and the skin irritation test in the present study have demonstrated that chitosan-based temperature-sensitive hydrogel has good biocompatibility and safety. No adverse effects are observed. However, a limitation of the present study was that only an extract of the gel was applied to the cells. To conclude, the present study provided a basis for the future development of the use of chitosan-based temperature-sensitive hydrogel in ophthalmic drug delivery.
This study was supported by the Shandong Technology Development Project (grant no. 2014GHY115025), the Qingdao Science and Technology Bureau (grant no. 103315-NSH) and the National Natural Science Foundation of China (grant no. 31071014).
{#f1-etm-0-0-5586}
{#f2-etm-0-0-5586}
{#f3-etm-0-0-5586}
{#f4-etm-0-0-5586}
{#f5-etm-0-0-5586}
{#f6-etm-0-0-5586}
######
ISO 10993.10-2002 scoring standard.
ER Grade Edema response Grade
------------------------- ------- ---------------- -------
None 0 None 0
Mild 1 Mild 1
Clear 2 Clear 2
Moderate 3 Moderate 3
Severe ER or escharosis 4 Severe 4
ED, edema; ER, erythematous response.
######
Entrapment rate of levofloxacin chitosan microspheres (n=5).
C~free~ (g/l) C~total~ (g/l) Entrapment rate (%) Mean ± RSD
--------------- ---------------- --------------------- ------------
0.312 0.425 26.6
0.305 0.418 27.0
0.319 0.432 26.2 26.5±1.31%
0.314 0.429 26.8
0.322 0.436 26.1
C~free~, concentration of free chitosan; C~total~, maximum concentration of chitosan that can be released from levofloxacin chitosan microspheres; RSD, relative standard deviation.
[^1]: Contributed equally
| {
"pile_set_name": "PubMed Central"
} |
Introduction
============
BNP7787 (Tavocept, disodium 2,2′-dithio-*bis*-ethanesulfonate) is an investigational agent that is a novel, water-soluble disulfide that has been evaluated in Phase II and Phase III clinical trials in patients with advanced non-small-cell lung cancer (NSCLC).[@b1-ott-8-375] The American Cancer Society estimates that approximately 159,260 deaths from lung cancer will occur in 2014 accounting for approximately 27% of all cancer deaths in the US.[@b2-ott-8-375] Worldwide more than 1.59 million deaths per year are attributed to lung cancer.[@b3-ott-8-375] Approximately 40--65% of these deaths will be due to primary adenocarcinoma of the lung/bronchus.[@b2-ott-8-375] While there have been some advances in chemotherapeutic agents for the treatment of NSCLC, and improved treatments are available for the management of toxic side effects, there have been only modest gains in survival for patients with advanced disease. For example, modest improvements in median duration of survival have been reported with the use of bevacizumab (Avastin) and cetuximab (Erbitux) in combination with platinum-doublet chemotherapy for the treatment of patients with advanced NSCLC.[@b4-ott-8-375],[@b5-ott-8-375] Therefore, there remains an unmet need for a treatment that can offer patients with advanced NSCLC a greater clinical benefit that is measured by increases in survival and reduction in toxic side effects, with a manageable or improved safety profile.
In separate randomized multicenter Phase II and Phase III clinical trials in NSCLC patients, treatment with BNP7787 in combination with standard chemotherapy (platinum and taxane) resulted in substantial increases in the overall survival of patients with advanced adenocarcinoma of the lung in the first-line treatment setting (Hausheer et al[@b1-ott-8-375] and Hausheer et al, unpublished data, 2014). BNP7787 does not interfere with paclitaxel-induced apoptosis or with taxane-, platinum-, vinca alkaloid-, or epothilone-induced cytotoxicity in human cancer cell lines.[@b6-ott-8-375] Additionally, studies in animals demonstrated that BNP7787 did not exhibit tumor protection when given with taxane and platinum agents.[@b7-ott-8-375],[@b8-ott-8-375] Importantly, in Phase I and Phase III clinical trials, BNP7787 did not interfere with the antitumor activity of cisplatin and paclitaxel.[@b1-ott-8-375],[@b9-ott-8-375] Unlike thiol-containing drugs such as mesna and amifostine, BNP7787 can be administered to patients in large doses (eg, 18.4 g/m^2^) without notable perturbations in plasma disulfide proportions or in the strongly oxidizing extracellular environment and without any toxic side effects.[@b6-ott-8-375] We have previously reported that thiol-containing drugs may dramatically disrupt the normal homeostatic thiol and disulfide balance both in the plasma and inside the cell, and that they are often accompanied by toxic side effects;[@b6-ott-8-375] however, for disulfide-containing drugs like BNP7787, there is no obvious mechanism through which they would substantially perturb the ratios of thiols and disulfides in these compartments.[@b6-ott-8-375],[@b7-ott-8-375],[@b10-ott-8-375]
Based on computational modeling, we hypothesized that BNP7787 might interact with and modify human anaplastic lymphoma kinase (ALK). ALK belongs to the family of insulin growth factor receptor tyrosine kinases, and fusions of ALK with other genes are common in several diseases and cancers.[@b11-ott-8-375] Our focus was on ALK fusions found in NSCLC. At least seven different variants of ALK fusions with the gene encoding the echinoderm microtubule-associated protein-like 4 (EML4) are known to occur in NSCLC; EML4--ALK variants are constitutively active, exhibiting gain of function.[@b12-ott-8-375] Additionally, fusions between the tropomyosin receptor kinase fused gene (TFG) and ALK (TFG--ALK), and KIF5B and ALK (KIF5B--ALK), are also known to occur in NSCLC.[@b11-ott-8-375],[@b13-ott-8-375],[@b14-ott-8-375] EML4--ALK fusions are thought to account for approximately 3% of NSCLC cases.[@b11-ott-8-375] ALK is coupled to numerous signaling pathways that regulate cell proliferation including Ras--ERK, JAK3--STAT3, and PI3K[@b15-ott-8-375] and, therefore, represents an important target for anticancer-drug development. Herein, we characterize the structural and functional consequences of BNP7787 on the kinase domain of human ALK. Specifically, we determine the effect of BNP7787 on ALK activity in the presence and absence of the known ALK inhibitor, crizotinib (PF02341066), and report a high-resolution X-ray crystal structure of the kinase domain of ALK bearing two covalent BNP7787-derived mesna-cysteine adducts.
Materials and methods
=====================
Reagents for in vitro kinase assays
-----------------------------------
N-terminal 6His tagged recombinant human ALK expressed in baculovirus Sf21 was purchased from EMD Millipore (Billerica, MA, USA) (purity ≥60% by sodium dodecyl sulfate polyacrylamide gel electrophoresis) and aliquoted to 1 μL fractions when it was used the first time (to avoid multiple freeze/thaws for subsequent experiments). BNP7787 was prepared by a proprietary method (purity \>97%, no mesna was detected by mass spectroscopy). Kinase inhibitor, PF02341066 (crizotinib), was purchased from Selleck Chemicals, LLC. (Houston TX, USA). Polyglutamate-tyrosine (PolyGT) substrate was purchased from Sigma-Aldrich (St. Louis, MO, USA). Kinase assay buffer was prepared and consisted of 20 mM HEPES, 0.1% Brij 96, 10 mM NaF, 1 mM Na~3~VO~4~, and 10 mM MnCl~2~ adjusted to a final pH of 7.5. Half-area 96-well microplates were purchased directly from Corning Incorporated (Corning, NY, USA). ADP-Glo reagents were purchased from Promega (Madison WI, USA) and consisted of ADP, ATP, ADP-Glo, kinase detection reagent buffer, and kinase detection substrate. All other reagents were purchased from Sigma-Aldrich Co (St Louis, MO, USA). A Tecan Ultra microplate reader with XFluor software (V4.51; Tecan \[Morrisville, NC, USA\]) and RdrOle software (V4.50; Tecan) were used in this study.
ALK kinase activity assay
-------------------------
Kinase activity was evaluated using the ADP-Glo system from Promega Corporation and monitored ADP produced when ALK phosphorylated the PolyGT substrate. Assays typically contained 4 ng/μL of ALK per assay and this concentration typically gave signal to background ratios of 12 or higher (relative to control). Each assay condition was run in triplicate and experiments were repeated independently on different days to ensure reproducibility of results. PolyGT (4:1 ratio) was used as the substrate for phosphorylation and had an average polymer mass ranging from 20,000 to 50,000 g/mol. Typically, 10-μL volume assays in half-area 96-well microtiter plates contained ALK (40 ng total or 4 ng/μL), ATP (100 μM), PolyGT substrate (0.2 μg/μL), and the concentrations of BNP7787 and/or crizotinib as indicated; additionally, kinase assay buffer was added to achieve a final volume of 10 μL per assay. For most assays, a stock of ATP (1 mM) and PolyGT (2 mg/mL) was mixed 1:1 to give an ATP/PolyGT master mix of 0.5 mM ATP and 1 mg/mL PolyGT. Crizotinib was dissolved as a 1 mM stock in dimethyl sulfoxide (DMSO) and then further diluted in kinase assay buffer (DMSO-only controls were run to ensure that DMSO did not interfere with the assay).
Cloning, expression, and purification of the kinase domain of ALK for X-ray crystallographic analyses
-----------------------------------------------------------------------------------------------------
Wild type human ALK, consisting of residues 1095--1410, was cloned into a proprietary vector containing a C-terminal 6His tag. Isolated shuttle vector was transformed into DH10Bac cells. Colonies containing bacmid with transposed ALK DNA were picked and grown overnight at 37°C. Bacmid DNA was isolated by DNA isopropanol precipitation and resuspended in 100 μL of sterile water. Bacmid DNA was allowed to resuspend at room temperature for 1 hour prior to transfection. Recombinant bacmid DNA was expressed in SF9 cells at 27°C. The virus generated from a 72-hour infection was stored at 4°C.
Recombinant protein was expressed in SF9 cells at a multiplicity of infection of two in a 48-hour infection at 27°C. The cells were harvested by centrifugation and stored at −80°C. Purification of target protein was done using a two-column system (Ni-NTA and size exclusion). The cell biomass was lysed by sonification in 50 mM Tris-HCl pH 7.8, 500 mM NaCl, 10% glycerol, 20 mM imidazole (buffer A) plus Roche complete protease inhibitor tablets (Roche Diagnostics Corporation, Indianapolis, IN, USA), and 20,000 units benzonase. Target protein was extracted by binding Ni-NTA (Qiagen, Valencia, CA, USA). Protein was eluted with 250--500 mM imidazole pH 7.8. Peak fractions were pooled and aggregated protein was separated from monomeric protein via size exclusion (S200 16/60; GE Healthcare Lifesciences \[Piscataway, NJ, USA\]) in 50 mM bicine pH 8.4, 150 mM NaCl, 5 mM dithiothreitol (DTT). Monomeric protein was concentrated to ∼19 mg/mL.
Preparation of BNP7787-derived mesna adducts on ALK crystals
------------------------------------------------------------
ALK (19 mg/mL) in 50 mM bicine, pH 8.4, 150 mM NaCl, and 25 mM DTT was incubated at 4°C overnight to fully reduce the protein. DTT was removed by exchanging five times in 50 mM bicine, pH 8.4, 150 mM NaCl using ultrafiltration (10-kDa-cutoff Centricon filters). Fully reduced ALK was incubated with 5 mM BNP7787 and incubated at 4°C overnight. Protein was submitted for mass spectrometry analysis to confirm the presence of at least one BNP7787-derived mesna adduct prior to initiation of protein-crystallization experiments. The predicted mass for ALK was 36,868, and for ALK incubated with BNP7787 a mass of 37,194 was observed. This mass difference of 326 corresponds to two BNP7787-derived mesna moieties on the ALK protein (each mesna moiety has a molecular weight of 164; loss of a proton on the thiol group due to formation of a disulfide with a cysteine residue results in net mass increase of approximately 163 per mesna moiety).
Crystallization of ALK containing BNP7787-derived mesna adducts
---------------------------------------------------------------
Mass spectrometry analysis of C-terminal 6Xhis tag ALK that had been incubated with BNP7787 indicated two likely BNP7787-derived mesna adducts; however, we were unable to obtain crystals from these protein samples. As an alternative, apo-ALK crystals were soaked with BNP7787 and this yielded crystals with two BNP7787-derived mesna adducts. Briefly, the crystal of C-terminal 6Xhis tag ALK was obtained by sitting drop/vapor diffusion method by mixing 2 μL at 17 mg/mL protein (50 mM bicine pH 8.4, 150 mM NaCl, 5 mM DTT) with 2 μL of 0.1 M Tris-HCl pH 8.5, 0.2 M sodium acetate trihydrate, 30% w/v polyethylene glycol 4,000 at 20°C. Diffracting crystals appeared within 5--8 days. Before data collection, the crystals were soaked in 20 mM BNP7787 overnight and transferred into a cryoprotectant solution made up of 20% ethylene glycol v/v in crystallization buffer, after which they were flash-frozen in liquid nitrogen for data collection. Crystals diffracted to 2.1 Å. As previously mentioned, the mass spectrometry analysis of ALK after reaction with BNP7787 (data not shown) suggested two BNP7787-derived mesna adducts consistent with the X-ray structure.
Data collection and processing
------------------------------
Diffraction data were collected at the Advanced Light Source (ALS) (Berkeley, CA). BNP7787-derived mesna adducts were observed on Cys 1156 and Cys 1235. Data were processed using the program package Mosflm as part of the ccp4 program package. Image processing statistics are summarized in [Table 1](#t1-ott-8-375){ref-type="table"}.
Structure solution and refinement
---------------------------------
Data were indexed, integrated, scaled, and merged using the program Mosflm. The structure was solved by molecular replacement with Phaser using a monomer from the Protein Data Bank,[@b16-ott-8-375] and an internal structure which is similar to 2XP2.[@b17-ott-8-375] The structure was consistent with one molecule in the crystal asymmetric unit. The protein model was iteratively refit and refined using MIFit[@b18-ott-8-375] and REFMAC5.[@b19-ott-8-375] The solved structure is supported by: 1) contiguous electron density for most of the molecule; 2) landmark side chain density features matching the amino acid sequence including cysteines; 3) absence of phi-psi violations; and 4) final R/R~free~ values in the normal range. Residual density observed near Cys 1235 and Cys 1156 was modeled as BNP7787-derived mesna adducts. Final statistics are summarized in [Table 2](#t2-ott-8-375){ref-type="table"}. A number of side-chain atoms and protein fragments were not refined. The missing fragments included Gly 1123--Gly 1128 (P-loop), Ser 1136--Pro 1144 (loop connecting β2--β3), Arg 1214--Pro 1218 (loop connecting αD--αE), and Ser 1281--Arg 1284 (part of activation loop or A-loop).
Results
=======
Crystal structure of ALK bearing BNP7787-derived mesna-cysteine adducts
-----------------------------------------------------------------------
The crystal structure of ALK in complex with a BNP7787-derived mesna-cysteine adduct was completed at 2.1-Å resolution ([Figure 1A](#f1-ott-8-375){ref-type="fig"}; PDB ID code 4TT7). The protein crystallizes as a monomer in the asymmetric unit and BNP7787-derived mesna disulfide bonds were observed with Cys 1235 and Cys 1156. We refer to this process as BNP7787-mediated xenobiotic modification/modulation of protein cysteines. The mesna adduct at Cys 1156 is located in close proximity to the active site and in fact results in substantial disorder of P-loop (or phosphate binding loop) which is highly conserved in protein kinases. This disorder prevented full refinement of the P-loop. While a large fragment of the P-loop is missing from the refined structure, comparison with the P-loop of the apo-ALK suggests that the BNP7787-derived mesna adduct at Cys 1156 interferes with the positioning of Phe 1127 (one of the P-loop residues) into a small pocket now occupied by mesna, resulting in a destabilization of the loop's binding orientation ([Figure 1B](#f1-ott-8-375){ref-type="fig"}). In order to accommodate the BNP7787-derived mesna adduct in the current crystal structure, Cys 1156 flips from a totally solvent exposed orientation (in the wild type ALK) to an inside orientation, thereby modifying the P-loop.
Ligand binding site
-------------------
Close-up views of the electron density map at the sites of the BNP7787-derived xenobiotically modified ALK cysteine residues is presented in [Figure 2A and B](#f2-ott-8-375){ref-type="fig"}. For both of the BNP7787-derived mesna adducts at Cys 1235 and Cys 1156, a single conformation was observed. Both ligand binding sites are relatively solvent exposed ([Figure 2C and D](#f2-ott-8-375){ref-type="fig"}). The BNP7787-derived mesna adduct at Cys 1235 is located in the "back" of the kinase domain relative to the position of the active site. Mesna does not interact with any residues other than Cys 1235 although the sulfonate group is in close proximity to Arg 1231 ([Figure 2D](#f2-ott-8-375){ref-type="fig"}).
At the Cys 1156 site (which is located in a loop that connects β3 to αC), the mesna sulfonate group makes a water-mediated hydrogen bond with the carbonyl of Asp 1160 ([Figure 2D](#f2-ott-8-375){ref-type="fig"}). A fragment of the P-loop (Gly 1123--Gly 1128), another nearby fragment (Ser 1281--Arg 1284, which is part of activation loop or A-loop), and a number of residue side chains (such as Lys 1285) are not in the final refined structure, and interactions between a BNP7787-derived mesna adduct and these missing atoms cannot be ruled out. Additionally, interactions between a BNP7787-derived mesna adduct and Arg 1120 of another protein monomer in the crystal may also be a possibility.
This current structure of ALK with BNP7787-derived mesna adducts at Cys 1156 and Cys 1235 does not have electron density for Tyr 1282, Tyr 1283, and Lys 1285 residues which are part of the ALK activation loop (A-loop); it is not clear if the loss of density for these residues is due to the presence of a BNP7787-derived mesna adduct on Cys 1156 near this A-loop. An overlay of wild type ALK (PDB ID code 3L9P) with the ALK containing BNP7787-derived mesna adduct at Cys 1156 (PDB ID Code 4TT7) indicates that the mesna sulfonate is in potential contact distance with Lys 1285 side chain. However, as a point of reference, these residues (Tyr 1282, Tyr 1283, and Lys 1285) are also disordered in the ALK structure with crizotinib (PDB ID code 2XP2) suggesting that this is an area with inherent disorder.
BNP7787 inhibits ALK activity in vitro
--------------------------------------
ALK activity assays were run under two ATP concentrations (100 and 500 μM). ATP is often in the millimolar range in vivo,[@b20-ott-8-375] and the human body is reported to contain no more than 0.5 moles (250 g) of ATP at any time, but this supply is constantly and efficiently recycled.[@b21-ott-8-375] In vivo there are many ATP-dependent enzymes that compete for ATP binding, including kinases, synthetases, helicases, membrane transporters and pumps, chaperones, motor proteins, and large protein complexes like the proteasome;[@b22-ott-8-375] therefore, the concentrations of 100 and 500 μM ATP used herein are approximations for ATP concentrations that may be available to ALK in vivo as it competes for ATP with the various other enzymes and proteins that utilize ATP.
BNP7787 inhibited ALK with a half maximal inhibitory concentration (IC~50~) of 9.16±2.91 mM under assay conditions of 100 μM ATP ([Figure 3A](#f3-ott-8-375){ref-type="fig"}) and with an IC~50~ of 20.80±3.49 mM under assay conditions of 500 μM ATP ([Figure 3B](#f3-ott-8-375){ref-type="fig"}). Physiologically, concentrations of BNP7787 as high as 18 mM have been achieved in the clinic.[@b23-ott-8-375] BNP7787 has been administered at doses as high as 41 g/m^2^, and, with concentrations of 18.4 g/m^2^ used in Phase III trials, maximum serum concentration (C~max~) values in plasma of 10 mM are often observed;[@b23-ott-8-375] therefore, the concentrations of BNP7787 required to see an effect on ALK activity in vitro are physiologically relevant. Kinase endpoint assays like the Promega ADP-Glo assay system often classify inhibitors as competitive if their IC~50~ increases notably as the ATP concentration increases. As the ATP concentration was increased, the IC~50~ for BNP7787 also increased. From structural work, we observed that BNP7787 covalently modified ALK on Cys 1156 in a loop region of ALK that may subsequently result in partial interference with the phosphate binding site for ALK's ATP cofactor, and possibly with the A-loop residues. Therefore, while the inhibition of ALK by BNP7787 is not likely classical competitive inhibition -- where ATP and BNP7787 have nearly identical or at least significantly overlapping binding sites and only one molecule (either ATP or BNP7787) can occupy that site at a time -- it is "competitive-like" based upon the increasing IC~50~ as the ATP concentration is increased. This conclusion is supported by the X-ray crystallography studies of the ALK structure containing a BNP7787 adduct which indicate that BNP7787 modification of ALK results in a perturbation of the P-loop near where the ATP binding site is located ([Figures 1](#f1-ott-8-375){ref-type="fig"} and [2](#f2-ott-8-375){ref-type="fig"}).
Crizotinib inhibits ALK activity in vitro
-----------------------------------------
Crizotinib is a reported ATP-competitive inhibitor of ALK.[@b24-ott-8-375],[@b25-ott-8-375] In the in vitro kinase studies reported herein, we observed that crizotinib inhibited ALK with an IC~50~ of 27.2±1.83 nM (data not shown) under assay concentrations with 100 μM ATP and with an IC~50~ of 76.3±16.3 nM with 500 μM ATP (data not shown). As mentioned above, crizotinib has previously been characterized as a competitive inhibitor of ALK, with respect to ATP,[@b24-ott-8-375],[@b25-ott-8-375] and our data are consistent with this previously reported observation (note that in clinical trials where crizotinib was administered orally at doses of 250 mg twice daily, concentrations of crizotinib of 57 nM were reported[@b25-ott-8-375]).
BNP7787 potentiates the inhibitory effect of crizotinib on ALK activity in vitro (100 and 500 μM ATP conditions)
----------------------------------------------------------------------------------------------------------------
We evaluated the effect of physiologically achievable concentrations of BNP7787 near the IC~25~ and IC~50~ concentrations of crizotinib under assay conditions with either 100 or 500 μM ATP. Concentrations of crizotinib of 57 nM have been reported in clinical trials;[@b25-ott-8-375] therefore, concentrations used in these studies were within physiologically relevant ranges. BNP7787 has been administered at doses as high as 41 g/m^2^ and C~max~ values in plasma of 10 mM are typical.[@b23-ott-8-375] Under assay conditions with 100 μM ATP ([Figure 4A](#f4-ott-8-375){ref-type="fig"}), 5 mM BNP7787 in combination with 15 nM crizotinib (near the IC~25~ value for crizotinib, when ATP is 100 μM) resulted in 16% greater inhibition than 15 nM crizotinib alone, whereas 10 mM BNP7787 in combination with 15 nM crizotinib resulted in 29% greater inhibition than 25 nM crizotinib alone. Under assay conditions with 100 μM ATP ([Figure 4B](#f4-ott-8-375){ref-type="fig"}), 5 mM BNP7787 in combination with 30 nM crizotinib (near the IC~50~ value of crizotinib) resulted in 10% greater inhibition than 30 nM crizotinib alone, whereas 10 mM BNP7787 in combination with 30 nM crizotinib resulted in 19% greater inhibition than 30 nM crizotinib alone. These assays near the IC~50~ value for crizotinib (ie, 30 nM, when ATP is 100 μM) have slightly less impressive stimulation compared to 100 μM ATP and 15 nM crizotinib IC~25~ conditions. For all of the ALK assays that contained 100 μM ATP, we noted that ALK activity levels of less than 10% were not generally observed; this is a technical limitation of the assay. This means that there is only a range of 40% between the IC~50~ and the assay lower limit, so any additional inhibition attributable to BNP7787 is confined by this limit. Based on this data, BNP7787 notably potentiates the inhibitory effect of crizotinib on ALK at physiologically relevant concentrations of both BNP7787 and crizotinib. As discussed in preceding sections, BNP7787 alone or crizotinib alone were both also effective at inhibiting ALK in vitro.
Under assay conditions with 500 μM ATP ([Figure 4C](#f4-ott-8-375){ref-type="fig"}), 5 mM BNP7787 in combination with 30 nM crizotinib (near the IC~25~ value for crizotinib when ATP is 500 μM) resulted in 11% greater inhibition than 30 nM crizotinib alone; 10 mM BNP7787 in combination with 30 nM crizotinib resulted in 20% greater inhibition than 30 nM crizotinib alone; and 20 mM BNP7787 in combination with 30 nM crizotinib resulted in 30% greater inhibition than 30 nM crizotinib alone. Under assay conditions with 500 μM ATP ([Figure 4D](#f4-ott-8-375){ref-type="fig"}), both 5 mM and 10 mM BNP7787 in combination with 65 nM crizotinib (near the IC~50~ value of crizotinib when ATP is 500 μM) resulted in 6% greater inhibition than 65 nM crizotinib alone, whereas 20 mM BNP7787 in combination with 65 nM crizotinib resulted in 13% greater inhibition than 65 nM crizotinib alone.
Discussion
==========
BNP7787 is a multitargeted cysteine-specific agent that covalently modifies the catalytic domain of ALK and inhibits kinase activity
------------------------------------------------------------------------------------------------------------------------------------
BNP7787 is a novel chemo-enhancing and cytoprotective disulfide agent that has been evaluated in the clinic in patients with NSCLC.[@b1-ott-8-375] BNP7787 reacts with and forms mixed disulfides on specific cysteine residues on proteins, yielding specific mesna-cysteine adducts on the target protein. We refer to this process as BNP7787-mediated xenobiotic modulation/modification, and we have observed and characterized this in a variety of proteins important in cell growth and proliferation (data not shown). As an example, we report data herein on the crystal structure of ALK in complex with BNP7787-derived mesna adducts at 2.1-Å resolution. BNP7787-derived mesna adducts were found at Cys 1156 and Cys 1235. Both adducts are relatively solvent exposed, although the adduct at Cys 1156 clearly disrupts the orientations of the P-loop by sterically blocking the typically observed binding site for Phe 1127. Because the P-loop binds the ATP-substrate, this P-loop disruption may or may not alter the kinase activity of ALK or the inhibitory potency of its small molecule inhibitors.
The cysteine-specific, multitargeted nature of BNP7787 is important due to the fact that, except for a few types of cancer (eg, chronic myelogenous leukemia), tumor cells are known to be genomically heterogeneous, and tumors contain subpopulations of cancer cells that often express different tumor-promoting proteins or that have multiple dysregulated, distinct, but key pathways that modulate cell proliferation.[@b26-ott-8-375],[@b27-ott-8-375] The covalent modification of ALK by BNP7787 results in noteworthy disruptions of the protein's P-loop region. We hypothesize that BNP7787-mediated cysteine xenobiotic modulation/modification represents a novel mechanism of action for this type of agent, but cysteine-specific modifications, specifically posttranslational modifications of cysteine residues in proteins, are a biological mechanism that may regulate a variety of cellular processes (eg, glutathionylation,[@b28-ott-8-375] nitrosylation,[@b29-ott-8-375] prenylation,[@b30-ott-8-375] and palmitoylation[@b31-ott-8-375]).
BNP7787 is expected to remain predominantly in the disulfide form in the plasma;[@b6-ott-8-375],[@b10-ott-8-375],[@b32-ott-8-375] however, the intracellular environment and the interstitial space are likely venues for BNP7787 metabolism to mesna, mesna-disulfide heteroconjugates, and free thiols. Any of these species (BNP7787, intracellularly generated BNP7787-derived mesna, or BNP7787-derived mesna-disulfide heteroconjugates) may modify proteins in vivo. The metabolism of BNP7787 to mesna-disulfide heteroconjugates has been observed in in vitro studies and is supported by computational studies on nonenzymatic thiol transfer reactions involving physiological free thiols with BNP7787.[@b6-ott-8-375]
In summary, BNP7787 has been shown to covalently modify ALK, to inhibit ALK's kinase activity in vitro, and to potentiate the inhibitory effect of crizotinib (PF02341066) on ALK. BNP7787 is a cysteine-specific, multitargeted modulator of protein function. BNP7787 mediates the nonenzymatic xenobiotic modification of cysteine residues on proteins. BNP7787 is autocatalytic and requires no protein cofactor to xenobiotically modify cysteines, and appears to be specific for cysteine residues located within a particular structural context (ie, not all cysteines in a protein are xenobiotically modified; proprietary computational work by FHH, PYA, and PNP). BNP7787-mediated cysteine xenobiotic modification represents a novel mechanism of action for this type of agent.
**Disclosure**
Authors ARP, PNP, MZ, PYA, HK, and FHH are or were employed by BioNumerik Pharmaceuticals, Inc. Authors VLN, JB, BC-L, VS, CL are or were employed by Zenobia Therapeutics, Inc. The authors report no other conflicts of interest in this work.
{#f1-ott-8-375}
{#f2-ott-8-375}
{#f3-ott-8-375}
{#f4-ott-8-375}
######
Crystal characteristics and data collection statistics
------------------------------ --------------------------------------------
Unit cell (Å, °) 51.535 57.157 104.216 90.000 90.000 90.000
Space group P2~1~2~1~2~1~
Resolution range (Å) 46.20--2.10 (2.21--2.10)
Number of observations 102,983
Number of unique reflections 18,384
Redundancy 5.6 (5.7)
Completeness (%) 98.8 (97.5)
Mean I/sigma(I) 10.1 (2.8)
*R*~merge~ 0.135 (0.621)
------------------------------ --------------------------------------------
**Note:** Outer shell statistics are in parenthesis.
######
Crystallographic data and refinement statistics
-------------------------------- -------------------------------------------
Resolution range (Å) 46.196--2.100
Number of reflections 18,340 (17,402 working set, 938 test set)
Number of protein chains 1 (chain A)
Number of protein residues 293
Number of ligands 5
Number of waters 159
Number of atoms 2,452
Mean B-factor 21.256
*R*~work~ 0.1927
*R*~free~ 0.2448
RMSD bond lengths (Å) 0.010
RMSD bond angles (°) 1.196
Number of disallowed φψ angles 1
-------------------------------- -------------------------------------------
**Abbreviation:** RMSD, root-mean-square deviation.
| {
"pile_set_name": "PubMed Central"
} |
1. Introduction {#sec1}
===============
Brain tumors are the second most common form of childhood cancer, after acute lymphoblastic leukemia (ALL) \[[@B1]\]. Treatment for both brain tumors and ALL includes cranial RT. Given 5-year survival rates that approach 90% for children treated for ALL and 70% for those treated for brain tumors \[[@B2]\], there are currently a great many survivors of these cancers that suffer from the consequences of RT, including adverse physiological, psychological, and cognitive side effects that manifest both acutely and years later. These so called "late effects" result in lowered quality of life (QOL) \[[@B3]\] in survivors, for which there is at present no effective treatment.
RT for pediatric cancer has long been acknowledged as a primary cause of neurological complications and neurocognitive decline \[[@B4]--[@B8]\]. Childhood RT is associated with a significant decrease in IQ scores \[[@B8]--[@B14]\], thought to result from deficits in core processing functions impaired by RT, including processing speed \[[@B15]\], attention \[[@B15]--[@B18]\], working memory, and other executive functions \[[@B7], [@B19]\]. In addition to cognitive impairments, adult survivors of childhood RT also experience elevated rates of emotional distress, such as anxiety and/or depression \[[@B20], [@B21]\] and posttraumatic stress disorder \[[@B22]\]. These cognitive and emotional consequences of RT result in decreased QOL that manifests in a variety of ways. For example, adult survivors of childhood RT are less likely to obtain a college education \[[@B23], [@B24]\] or marry \[[@B5], [@B20]\] and more likely to be unemployed \[[@B24], [@B25]\].
Improving QOL for survivors necessarily involves attenuating the long-term neural consequences of RT. Ionizing radiation damages the brain directly, but in addition, it chronically suppresses cell proliferation, thereby depriving the brain of the raw materials needed for repair. Evidence indicates that it also creates a milieu that is hostile to regenerative processes. When the brain is irradiated in childhood, there is a further consequence of RT, as suppressed cell proliferation and hostile environmental conditions disrupt ongoing developmental processes. What is needed, therefore, is a treatment that can both "jump-start" cell proliferation and foster a neural environment that is conducive to plasticity. Exercise may represent one such treatment, and its restorative potential for the post-RT brain is discussed.
2. RT Disrupts Brain Development {#sec2}
================================
RT damages the brain regardless of age. However, the brains of children are still developing, and RT profoundly affects ongoing developmental processes. The potential mechanisms underlying this disruption are many, such as perturbations of vasculature \[[@B26]\] and suppression of cell proliferation \[[@B27]--[@B29]\]. Damage to the endocrine system \[[@B30], [@B31]\] has been shown to play a role, in particular, decreased expression of growth hormone (GH). GH deficiency results from the effects of a brain tumor or of therapy such as surgery, RT, or chemotherapy. Merchant et al. \[[@B32]\] report that the peak GH response within 12 months after the initiation of cranial RT depends on hypothalamic dose-volume effects and may be predicted on the basis of a linear model that sums the effects of the entire dose distribution. The rate of decline in the peak GH response may also be influenced by clinical factors indicating the severity of the disease and the type and location of tumor.
Disruption of brain development could also be due in part to cancer treatment effects on food intake. Treatment-induced nausea and vomiting, as well as gastrointestinal toxicity can lead to nutritional deficiency and changes in body composition \[[@B33], [@B34]\], which may be long-lasting. Indeed, survivors of childhood brain cancer are often underweight \[[@B35]\]. In contrast, survivors of childhood ALL are more likely to be obese, compared with age-matched controls \[[@B36]\]. Thus, treatment effects on hormone levels and nutritional intake, alone or in combination, are likely important contributors to altered neural development and, ultimately, cognitive impairments.
Animal models of pediatric RT enable controlled study of mechanisms that contribute to disrupted development and, ultimately, cognitive late effects. To model the effects of RT on the developing brain, we have treated postnatal day 28 (PND28) rats with whole brain irradiation (WBI), using one of 2 regimens: single dose (20 Gy) or fractionated, in which animals received 20 Gy over the course of 5 days (4 Gy/d). Either regimen results in a profound stunting of brain growth visible to the naked eye (see [Figure 1](#fig1){ref-type="fig"}), although the effect is clearly bigger with single dose treatment.
Using this model, we can probe the cellular, chemical, and structural effects of RT that contribute to decreased brain size and cognitive impairments in adulthood. To enhance translational value, we are using imaging techniques to discover RT-induced changes *in vivo* that predict future cognitive impairments before they manifest. For example, we are using magnetic resonance imaging (MRI) and diffusion tensor imaging (DTI) to assess RT-induced structural changes and ^1^H magnetic resonance spectroscopy (MRS) to assess chemical changes following RT (see [Figure 2](#fig2){ref-type="fig"}). DTI has the added advantage of providing information on fractional anisotropy (FA), a measure of the functional integrity of white matter tracts. Our preliminary ^1^H MRS findings showed changes in glutamate, alanine, and lactate in RT brains, compared to sham controls. In addition, FA analysis showed a significant decline in fimbria volume and mean fimbria FA value in RT brains compared to controls. These changes were observed three months prior to the detected cognitive changes shown in [Figure 3](#fig3){ref-type="fig"}, suggesting that imaging changes can be used as early markers of cognitive decline.
3. RT-Induced Suppression of Cell Proliferation Contributes to Cognitive Impairments {#sec3}
====================================================================================
Because ionizing radiation kills dividing cells, it is effective at treating cancer, yet devastating to noncancerous tissue in the brain. Although mature neurons are postmitotic and therefore not directly affected by radiation, the brain\'s actively dividing neural stem cells (NSC) are largely wiped out, even by very low doses \[[@B28]\]. This is problematic, as it decreases the availability of new neurons in neurogenic regions of the brain and of new glia (oligodendrocytes and astrocytes) in nonneurogenic areas.
The dentate gyrus (DG) of the hippocampus, along with the lining of the lateral ventricles (the subventricular zone, or SVZ), is one of the few neurogenic regions of the adult mammalian brain (see \[[@B37]\] for review). Animal studies indicate that ongoing neurogenesis in this region is important for cognition. For example, newly generated neurons are kept alive by effortful learning (for review see \[[@B38]\]) and are needed for the formation of long-term spatial memory \[[@B39]\]. Analysis of postmortem human tissue following cancer treatment shows an almost complete lack of hippocampal neurogenesis \[[@B40]\], the functional importance of which is attested to by the cognitive impairments observed in survivors \[[@B41]\].
Animal models have yielded direct links between RT-induced decrements in hippocampal neurogenesis and cognitive impairments. Many studies have focused on deficits in spatial performance (place learning or spatial memory) and trace fear conditioning, since these are hippocampus-dependent functions. Spatial impairments have been observed in conjunction with decrements in DG cytogenesis following both fractionated \[[@B42], [@B43]\] and single-dose WBI \[[@B44], [@B45]\]. Our group also has noted performance deficits in a spatial task after fractionated irradiation (see [Figure 3(c)](#fig3){ref-type="fig"}). Decreased fear conditioning has also been associated with radiation-induced suppression of DG cytogenesis \[[@B46]--[@B48]\]. In sum, an increasing body of evidence implicates suppression of hippocampal neurogenesis as a causative factor in cognitive impairments following RT.
However, suppression of hippocampal neurogenesis is likely only part of the story. NSCs in nonneurogenic brain regions, such as the cortex, differentiate into glia \[[@B49]\]. A plentiful supply of glial cells is essential for neuronal health and function \[[@B50], [@B51]\], so reduced proliferation of NSCs due to RT could contribute to cognitive impairments by reducing the availability of glia. For example, problems with executive functions are widely reported in adult survivors of childhood RT. Executive functions develop linearly during adolescence, in apparent conjunction with myelination of the frontal lobes \[[@B19]\]. Frontal lobe white matter appears particularly vulnerable to RT \[[@B52]\], and RT-induced damage to white matter tracts may, in large part, underlie the neurocognitive deficits experienced by adult survivors of childhood cancer \[[@B19], [@B53]\]. Myelination is dependent on a ready supply of healthy oligodendrocytes, which is in turn dependent on adequate proliferative activity of NSCs. RT-induced ablation of NSCs in nonneurogenic regions could therefore contribute to cognitive impairments.
To provide direct evidence that RT-induced suppression of gliogenesis contributes to frontal lobe dysfunction, animal models of frontal lobe-dependent tasks are important. The 5-choice serial reaction time task (5-CSRTT) is a reliable means by which to assess prefrontal cognitive processes in the rodent. This automated task measures several aspects of visual attention, specifically divided, sustained, and selective attention, as well as processing speed and impulsivity \[[@B54]\]. The task requires the animal to detect brief flashes of light that appear in one of five apertures (see [Figure 3(a)](#fig3){ref-type="fig"}) and then nose-poke into the aperture that the light appeared in. The animal is given 5 seconds in which to make the nose-poke response. Correct responses are rewarded by a food pellet being dispensed into a magazine at the rear of the testing chamber (see [Figure 3(a)](#fig3){ref-type="fig"}). To provide motivation, animals are food restricted. Between stimulus presentations, there is an intertrial interval (ITI), and the animal must inhibit responding during this interval, because premature responses result in a short time-out period during which there are no trials, and thus food reward cannot be obtained. In performing this task, the animal has to sustain attention to all 5 of the apertures in order to constantly monitor where the light stimulus will be presented. Incorrect responses (nose-pokes into an aperture other than that in which the light was presented) indicate impaired attention. Measures of impulsivity are collected through responses that are characterized as perseverative and/or premature. Perseverative responses are defined as continuous nose-pokes in additional apertures. Nose-pokes made before the light is presented are considered premature responses. Processing speed measures are based on various latency times that are collected throughout the task.
We have used the 5-CSRTT to probe for impairment of prefrontal cognitive processes following fractionated irradiation (4 Gy/d for 5 days). Irradiated animals and shams were trained to perform the 5-CSRTT and then tested 6 and 9 months postirradiation. Our preliminary findings indicate that the irradiated animals are significantly less accurate at nose-poking into the correct aperture, suggesting that they have attentional impairments (see [Figure 3(b)](#fig3){ref-type="fig"}). Future experiments will focus on replicating these impairments in irradiated animals and determine whether they are linked to reduced gliogenesis in frontal regions.
4. RT Creates a Brain Milieu Hostile to Plasticity {#sec4}
==================================================
The microenvironment of the brain is regulated and protected by specific barriers, which include the vascular endothelial barrier (also called the blood-brain barrier, or BBB) at the capillary-parenchyma interface and the epithelial barrier (blood-cerebrospinal fluid barrier) at the choroid plexus \[[@B55]\]. The BBB is more than a physical barrier: it plays a fundamental role in regulating the movement of substances between the blood and the CNS (see [Figure 4(a)](#fig4){ref-type="fig"}). The microvascular network is also the site of the BBB, and the endothelial cells (ECs) that make up the microvascular network barrier contain few pinocytotic vesicles and adhere to each other via tight junctions \[[@B56]\]. Tight junctions limit paracellular transport of hydrophilic compounds into the CNS as compared to non-CNS vessels \[[@B57], [@B58]\]. Also, astrocytes in close proximity to the ECs add another impediment to paracellular transport by biochemically conditioning the ECs and strengthening the tight junctions between them \[[@B59]\]. ECs coat, in a single layer, the interior of all blood vessels. Because of this intertwined fate with the circulatory system, ECs play a unique role in maintaining physiological homeostasis, controlling the movement of substances across from the blood compartment into the different tissues and organs with varying demands and function \[[@B60]\]. The ECs also play an important immune function through leukocyte surveillance and extravasation by regulating adhesion integrins and cytokine production \[[@B61]\]. In particular, they have been shown to directly secrete tumor necrosis factor (TNF) \[[@B62]\]. Thus, damage to the ECs compromises the integrity of the BBB.
When the barrier between the vascular supply of the brain and the CNS parenchyma is disrupted, excess extravasation of proteins, biologic-response molecules (e.g., growth factors, cytokines, and clotting factors), inflammatory cells, and therapeutic drugs can damage the brain \[[@B55], [@B63]--[@B65]\]. The disruption of the BBB (see [Figure 4(b)](#fig4){ref-type="fig"}) has been identified as a consequence of various diseases/injuries such as multiple sclerosis, ischemia, HIV, hypertension, brain tumors, CNS injury, and radiation exposure \[[@B65]--[@B69]\], wherein inflammatory cells are able to penetrate the BBB and destroy the myelin surrounding the axons. Demyelination and myelin thinning have been reported in the CNS following RT \[[@B70]--[@B73]\]. Felts et al. have also shown that RT-induced BBB permeability prolonged the induced demyelination of neurons \[[@B74], [@B75]\]. We and others \[[@B76]--[@B79]\] have demonstrated that there is an increase in BBB permeability following RT, which is caused in part by EC damage, as expressed by changes in tight junction integrity and by vesicle formation postirradiation. RT-induced EC damage has been investigated \[[@B80]--[@B82]\] with the aim of elucidating the effect on initiating and/or sustaining radiation side effects. Eissner et al. \[[@B83]\], as well as others \[[@B81], [@B82]\], have shown that when irradiated, ECs *in vitro* and *in vivo* undergo apoptosis at a higher percentage than other cells. Our studies using electron microscopy show that RT causes damage to the tight junctions \[[@B77]\], which is also connected to the observed increase in BBB permeability. In addition, several studies, including our own, have shown an increase in BBB permeability and an increase in the number of vesicles following fractionated cranial irradiation \[[@B77]--[@B79]\].
Such damage to the microvasculature and breach of the BBB can disturb the delicate brain microenvironment and expose the brain parenchyma and neural cells to noxious substances \[[@B69], [@B77], [@B85]\]. This microenvironment imbalance can set into motion a chain of events (such as cytokines release), magnifying the original signal and finally causing measurable late-term tissue damage in the irradiated brain that may play a role in cognitive impairment \[[@B86]\]. We and others have shown that RT induces an inflammatory response as indicated by an increase in TNF-*α* and intercellular adhesion molecule-1 (ICAM-1) signaling in the brain \[[@B87]--[@B90]\]. We have reported activated astrocytes after treatment with single and fractionated RT \[[@B77], [@B91]\]. Prolonged gliosis can create glial scar sites, which have been theorized to inhibit axonal regeneration or remyelination \[[@B92], [@B93]\]. We have demonstrated that this inflammation response is related to an increase in BBB permeability following RT and that it is abrogated when treated with antibodies to TNF-*α* or ICAM-1 \[[@B76], [@B90]\]. In a histological study on mouse brains we observed significant changes 120 days following fractionated RT: fewer neurons, a significant decrease in myelin suggesting complete destruction of the parts of the white matter at 120 days following RT, and at 90 days following RT, we observed swelling of nerve fibers and increased thickening of the myelin sheaths (see [Figure 5(b)](#fig5){ref-type="fig"}) indicative of dying axons.
5. The Restorative Potential of Exercise for the Post-RT Brain {#sec5}
==============================================================
Given its myriad beneficial effects on the brain, exercise has been suggested as a treatment for a wide variety of brain maladies, from aging \[[@B94]\] to alcoholism \[[@B95]\]. In the case of aging, exercise has been shown to have a remarkable restorative effect, encouraging the resurgence of atrophied regions such as white matter tracts \[[@B96]\] and the hippocampus \[[@B97]\], and improving cognition \[[@B98]\]. Such effects are particularly encouraging for the post-RT brain, since it shares many things in common with the aged brain, such as decreased cell proliferation, decreased growth hormone, and increased inflammation. Moreover, in both cases these conditions worsen over time, to ultimately create a neural milieu in which plasticity is suppressed.
In aged rodents, exercise can increase proliferation of NSCs \[[@B99]\], suggesting that it has neurogenic potential even in a system in which cell proliferation is drastically reduced. Encouragingly, exercise has been reported to increase hippocampal neurogenesis in the irradiated brain in rodent models \[[@B100], [@B101]\]. There are likely multiple mechanisms of this enhanced neurogenesis. Neurogenesis is tightly linked to the microenvironment \[[@B102]\] and is known to be suppressed under conditions in which there is unchecked inflammation \[[@B27]\] or a lack of trophic \[[@B103]\] or hormonal support \[[@B104]\]. Exercise has been shown to increase growth hormone \[[@B104]\] and reduce inflammation \[[@B105]\], two potential ways in which it could counteract the suppressive environment created by RT.
Recent research has begun to elucidate the important role that microglia have in maintaining the neurogenic niche \[[@B106]--[@B108]\]. Unfortunately, radiation severely disrupts microglial distribution, alters their morphology (see [Figure 6](#fig6){ref-type="fig"}), and decreases their numbers \[[@B109]\], effects that likely contribute to RT-induced neurogenesis impairment. It has been shown that, after radiation, microglia in the SVZ rebound more quickly than those in the DG. This may explain why neurogenesis recovers better in the SVZ, compared to the dentate \[[@B110]\]. Voluntary exercise has been shown to increase microglia \[[@B111]\], suggesting a further means by which exercise could help to restore a microenvironment conducive to cell proliferation.
Exercise might also have an enhancing effect on gliogenesis in the post-RT brain. As described above, glia are essential for the integrity and function of the cortex. Exercise has been shown to enhance cortical gliogenesis in the intact brain \[[@B112]\], and our future efforts will include determining whether post-RT exercise enhances cortical gliogenesis and ameliorates impairments in the 5-CSRTT.
In addition to these many direct benefits, it is important for survivors of childhood RT to exercise, in order to counteract chronic conditions that arise from cancer treatment, such as impaired pulmonary and cardiac function \[[@B113]\]. Emotional problems like depression and anxiety may also decrease with regular exercise. Unfortunately, treatment-induced fatigue, cardiorespiratory problems, and muscular deconditioning tend to promote sedentary habits during treatment that linger into adulthood, with the result that childhood cancer survivors are much less physically active than their healthy peers (see \[[@B113]\] for review). Cranial RT in particular is associated with sedentary habits in adulthood \[[@B114], [@B115]\]. However, recent studies suggest that physical fitness is an achievable goal for childhood cancer survivors \[[@B35], [@B113]\], so the beneficial effects of exercise observed in animal models can be followed up in human patients.
Fortunately, rodents show no reluctance to exercise after RT, and initial studies suggest that exercise is capable of ameliorating RT-induced deficits in both neurogenesis and cognition. Voluntary running in adulthood has been shown to restore neurogenesis in mice irradiated early in life \[[@B100]\], suggesting that exercise may be a feasible means by which to promote cell proliferation in adult survivors of childhood RT. Furthermore, it may be able to attenuate RT-induced cognitive impairments. A recent study showed that voluntary running ameliorated radiation-induced spatial memory decline 4 months after radiation as well as partially restored neurogenesis in the DG \[[@B101]\].
While these results are encouraging, continued study of the effects of exercise on the RT brain in animal models is essential. For one thing, it is important to better understand the effects of exercise in the context of the RT brain. In particular, it is necessary to determine whether exercise has an adequate neural substrate on which to work. For example, one well-established effect of exercise is its ability to induce angiogenesis \[[@B116], [@B117]\], an effect that depends upon the brain\'s capacity to produce new ECs. Given the suppressive effect of RT on cell proliferation, it is possible that the angiogenic effect of exercise would be limited in the post-RT brain. In short, it is possible that the effects of exercise would be dampened in the post-RT brain, reducing its potential as a stand-alone treatment. Further study may indicate that exercise would be most useful as an adjuvant therapy. For example, stem cell replacement shows promise in the irradiated rodent brain \[[@B118]\] and may eventually be possible in humans if the hostile environment created by RT can be made more permissive for growth and repair. Exercise represents a viable means by which to achieve this, and future studies should address the potential of exercise to neutralize the hostile environment created by RT, as a preliminary step in restorative treatments.
6. Conclusions {#sec6}
==============
Both human and animal studies indicate that suppressed cell proliferation and the hostile neural environment induced by cranial RT contribute to subsequent cognitive impairments. These effects of RT may be alleviated, at least to some extent, by exercise. It has been well established that exercise can both promote cell proliferation as well as foster a neural environment permissive for plastic change. Early evidence from animal models indicates that exercise has the capacity to do both in the post-RT brain. However, further studies are needed, in order to determine whether RT-induced perturbations of the microenvironment could limit the plasticity-enhancing effects exercise has to offer.
{#fig1}
{#fig2}
![Data from our group showing RT-induced cognitive deficits. (a) Schematic illustration of the 5-CSRTT apparatus. (b) Fractionated X-ray radiation (4 Gy/d for 5 days) restricted to the frontal cortex of young rats significantly reduced choice accuracy on a 5-CSRTT 6 and 9 months following RT (\**P* \< 0.05). (c) Fractionated WBI in young rats impaired pliancy on a hippocampus-dependent strategy-switching task in the Morris water maze \[[@B119]\] 3-4 weeks following RT (\**P* \< 0.05).](NP2013-698528.003){#fig3}
{#fig4}
{#fig5}
{#fig6}
[^1]: Academic Editor: Chitra D. Mandyam
| {
"pile_set_name": "PubMed Central"
} |
INTRODUCTION
============
Endometrial cancer is the third most common gynecologic cancer, and its incidence and mortality are increasing in Korea, where more than 1,700 new cases are diagnosed and approximately 250 deaths occur every year from the disease \[[@B1][@B2]\]. The incidence rate increases every year by 6.9% and the mortality rate, by 6.7%. As such, the burden of this disease can be expected to increase in the near future.
The optimal management of endometrial cancer remains one of the most debated issues, with many differences and discrepancies among gynecologic oncologists. There is significant variability in the treatment algorithms used at different institutions and even among physicians within the same institution. Unresolved questions remain as to the value and extent of lymphadenectomy, the optimal adjuvant therapy for intermediate or high-risk endometrial cancer.
In 2009, the Korean Gynecologic Oncology Group (KGOG) conducted a survey to assess surgical practice patterns for endometrial cancer in Korea \[[@B3]\]. A previous survey by KGOG showed substantial differences in the surgical procedures used for the treatment of endometrial cancer between Korean gynecologic oncologists. Since then, there has been level I evidence from randomized controlled trials for surgical and adjuvant treatment in the field of endometrial cancer \[[@B4][@B5][@B6][@B7]\]. However, certain discrepancies still exist between the clinical guidelines and the actual practice adopted by clinicians. To identify current practice patterns in the surgical and adjuvant management of endometrial cancer in Korea, we conducted a survey of KGOG members.
MATERIALS AND METHODS
=====================
This survey was initiated by the Uterine Corpus Committee of the KGOG. In December 2014, we mailed the questionnaire to all 218 KGOG members. The last date for receipt of responses was set for February 13, 2015. The survey was estimated to take 10 minutes to complete and was submitted electronically. All data were stored automatically by the website SurveyMonkey (<http://ko.surveymonkey.com>), and all responses were anonymous. The respondents were asked about demographic characteristics, including their current practice settings and years since completing fellowship training. The respondents were asked 20 questions regarding surgical and adjuvant procedures for endometrial cancer. Each question referenced a detailed clinical scenario. The survey questions are provided in [Supplementary Table 1](#S1){ref-type="supplementary-material"}.
The data were analyzed using frequency distributions and nonparametric tests. In the event of missing data, percentages were determined on the basis of the number of responses received.
RESULTS
=======
1. Respondents\' demographics
-----------------------------
[Table 1](#T1){ref-type="table"} shows the demographics of the survey respondents. Of the 218 KGOG members who received the survey questionnaire, 108 (49.5%) responded. Most were men (87%) and were aged 41 to 50 years (52.8%). Of the respondents, 92.6% had completed a fellowship and 7.4% were currently fellows. Almost half of the respondents (47%) completed their fellowship training more than 10 years ago.
2. Surgical management
----------------------
Scenario 1 (mode of surgery for presumed stage I endometrial cancer): the patient was diagnosed with presumed stage I/grade 1 endometrioid adenocarcinoma. [Fig. 1](#F1){ref-type="fig"} shows the respondents\' treatment preferences for the mode of surgery. The majority of respondents (81%) would recommend laparoscopy and only 8.5% stated laparotomy. Approximately 10% of respondents preferred robotic surgery for this clinical scenario.
Scenario 2 (management for incidentally found stage I endometrial cancer): the patient was incidentally diagnosed with endometrial cancer. A general gynecologist performed a total hysterectomy and bilateral salpingo-oophorectomy without lymphadenectomy.
Scenario 2A (incidentally found stage I/grade 1 endometrial cancer): the final pathology report shows stage IA/grade 1 endometrioid adenocarcinoma. [Table 2](#T2){ref-type="table"} lists the respondents\' preferences in this scenario. Only 9.5% would recommend additional therapy. All respondents who recommended additional therapy preferred laparoscopic lymph node dissection.
Scenario 2B (incidentally found stage I/grade 3 endometrial cancer): the final pathology report reveals stage IA/grade 3 disease. [Table 2](#T2){ref-type="table"} lists the respondents\' preferences. The majority of respondents (85.6%) would recommend additional therapy in this scenario. Among those who recommended additional therapy, vaginal brachytherapy (36.4%) and laparoscopic lymph node dissection (34.1%) were the most commonly preferred options.
Scenario 3 (extent of lymphadenectomy for presumed stage I disease)
Scenario 3A (presumed stage IA/grade 1 endometrial cancer): the patient was diagnosed with presumed stage IA/grade 1 endometrioid adenocarcinoma based on preoperative magnetic resonance imaging (MRI) and biopsy. Preoperative cancer antigen 125 was within the normal range. [Table 3](#T3){ref-type="table"} shows the respondents\' treatment preferences for the extent of surgery. Of all the respondents, 19.8% stated that lymphadenectomy could be omitted and 21.7% recommended selective lymphadenectomy based on sentinel biopsy or frozen results for patients with presumed stage IA disease. More than half of the respondents (58.5%) preferred lymphadenectomy, of which 31.1% preferred pelvic lymphadenectomy only, 15.1% preferred pelvic and para-aortic lymphadenectomy up to the level of the inferior mesenteric artery (IMA), and 12.3% preferred pelvic and para-aortic lymphadenectomy up to the level of the renal vein.
Scenario 3B (presumed stage IB/grade 1 endometrial cancer): the patient was diagnosed with presumed stage IB/grade 1 disease based on preoperative MRI and biopsy. Respondents\' preferences for the extent of surgery are shown in [Table 3](#T3){ref-type="table"}. The majority of respondents (93.3%) would recommend lymphadenectomy in this scenario, of which 21.4% preferred pelvic lymphadenectomy, 35.0% preferred pelvic and para-aortic lymphadenectomy up to the IMA level, and 36.9% preferred pelvic and para-aortic lymphadenectomy up to the level of the renal vein.
Scenario 3C (presumed stage IB/grade 3 endometrial cancer): the patient was diagnosed with presumed stage IB/grade 3 disease based on preoperative MRI and biopsy. Respondents\' preferences for the extent of surgery are shown in [Table 3](#T3){ref-type="table"}. The majority of respondents (97.1%) indicated that lymphadenectomy would be recommended, of which 10.7% preferred pelvic lymphadenectomy, 34.0% preferred pelvic and paraaortic lymphadenectomy up to the IMA level, and 52.4% preferred pelvic and para-aortic lymphadenectomy up to the level of the renal vein.
3. Adjuvant treatment
---------------------
[Table 4](#T4){ref-type="table"} shows respondents\' preferences for adjuvant therapy in completely staged endometrial cancer. Respondents were asked to complete a table showing the preferred adjuvant therapy based on pathologic findings. In patients with stage IA/grade 1 disease, all respondents indicated that observation was preferable. Our survey reveals that more than 70% of members administered adjuvant therapy except stage IA/grade 1 or stage IA/grade 2: more than 90% administered adjuvant therapy when patients exhibited stage IA/grade 3, stage IB/grade 2 or stage IB/grade 3 disease. In patients with stage IB/grade 3 disease, the majority of respondents (99%) would recommend adjuvant therapy, among which whole pelvic radiation therapy (WPRT) was preferred by 34.4% and WPRT and brachytherapy was preferred by 26%.
Scenario 4 (adjuvant treatment for stage II): the patient has stage II endometrioid adenocarcinoma with less than half myometrial invasion. Respondents\' preferences for adjuvant therapy are listed in [Table 5](#T5){ref-type="table"}. The majority of respondents (89.6%) would recommend adjuvant therapy for stage II disease, among which WPRT was preferred by 31.8% and vaginal brachytherapy was preferred by 29.4%.
Scenario 5 (adjuvant treatment for stage IA and positive cytology): the patient has stage IA/grade 1 endometrioid adenocarcinoma and malignant cytology. [Table 5](#T5){ref-type="table"} shows the respondents\' preferences for adjuvant therapy. Over half of respondents (55%) would recommend adjuvant therapy, with 57.7% indicating chemotherapy as their first preference.
Scenario 6 (adjuvant treatment for stage IIIC1): The patient has stage IIIC1 endometrioid adenocarcinoma with more than half of myometrial invasion. [Table 5](#T5){ref-type="table"} shows the respondents\' preferences for adjuvant therapy. All respondents indicated that adjuvant therapy should be performed in this scenario. More than half of the respondents who recommended adjuvant therapy stated that concurrent chemoradiotherapy (CCRT) would be their first choice of adjuvant therapy.
Scenario 7 (adjuvant treatment for stage IIIA): The patient has stage IIIA/grade 1 endometrioid adenocarcinoma with less than half myometrial invasion. Respondents\' preferences of adjuvant therapy are listed in [Table 5](#T5){ref-type="table"}. The majority of respondents (95.7%) suggested the use of adjuvant therapy in microscopic ovarian metastasis. The most common treatment choices were chemotherapy (42.0%) and CCRT (31.8%).
DISCUSSION
==========
When it came to the mode of surgery, there was a general consensus among KGOG members that minimally invasive surgery was preferable. However, we observed differences between KGOG members in current practice patterns for the treatment of endometrial cancer. This discrepancy was particularly prominent for the extent of surgery and adjuvant therapy options.
The role of minimally invasive surgery in endometrial cancer is expanding. Between 2009 and 2015, there have been significant increases in the proportion of KGOG survey respondents who think that minimally invasive surgery is appropriate for early-stage endometrial cancer (from 49% to 91.5% \[n=41/84 to 97/106\], p\<0.001) \[[@B3]\]. With similar outcomes for laparoscopy and laparotomy in the Gynecologic Oncology Group (GOG) LAP2 study \[[@B7]\], minimally invasive surgery is now considered standard for the treatment of endometrial cancer in Korea. Similarly, a recent survey by the Society of Gynecologic Oncology (SGO) showed that the majority of respondents (85.5%) preferred minimally invasive surgery for the staging of endometrial cancer \[[@B8]\]. However, robotic surgery is not commonly performed in Korea: while laparoscopy is the preferred option in Korea, the SGO survey showed that 97% of respondents now perform robotic gynecologic procedures, compared with 29.0% in the 2007 survey. The SGO survey demonstrated that more than half of the respondents who performed robotic surgery stated that they used it for 50% or more of all their gynecologic cases.
For the extent of surgery for presumed low-risk endometrial cancer, there was a significant difference in opinion among the KGOG respondents, and we found a clear discrepancy between the clinical guidelines and the actual practice adopted by clinicians. As two large randomized controlled trials have demonstrated that comprehensive surgical staging does not improve progression-free or overall survival \[[@B5][@B6]\], the current National Comprehensive Cancer Network (NCCN) guidelines recommend a more selective and tailored lymphadenectomy approach in early-stage endometrial cancer \[[@B9][@B10]\]. In this survey, we observed that approximately 60% of respondents recommended lymphadenectomy at least at the pelvic level. The KGOG survey performed in 2009 showed that approximately 67% of respondents preferred routine pelvic lymphadenectomy \[[@B3]\]. As such, there has not been a significant decrease in the proportion of respondents preferring routine pelvic lymphadenectomy for early-stage endometrial cancer in recent years (67% to 58.5% \[n=56/84 to 62/106\], p=0.249). Considering the results of surveys on lymphadenectomy conducted in other countries, we found significant variation between different geographic regions: While countries in Asia (72.8%) and Central Europe (55.6%) routinely perform lymphadenectomy, countries in the north and south of Europe as well as the United Kingdom, United States, and Canada perform routine lymphadenectomy in less than a third of endometrial cancer cases \[[@B11]\].
The anatomic borders of the lymphadenectomy continue to be controversial in intermediate- and high-risk endometrial cancer. About 37% members defined the upper border of para-aortic lymphadenectomy as the renal vein for intermediate-risk endometrial cancer (presumed stage IB/grade 1), while more than half defined the upper border of para-aortic lymphadenectomy as the renal vein for high-risk endometrial cancer (presumed stage IB/grade 3). Compared to just 11% of members preferring to remove nodes to the level of the renal vein in the 2010 SGO survey \[[@B12]\], the percentage of Korean gynecologic oncologists performing para-aortic lymphadenectomy up to the renal vein is relatively high.
Our survey is the first to assess practice patterns for adjuvant treatment in patients with endometrial cancer in Korea, as we did not ask about adjuvant practice in the 2009 survey. Indications for adjuvant therapy for patients with endometrial cancer are determined by various pathologic prognostic factors such as histologic subtype, grade, lymph node metastasis, myometrial invasion, and lymphovascular space invasion. Patient selection criteria for adjuvant therapy and optimal regimens for endometrial cancer have not yet been established in clinical practice. The current NCCN guidelines allow a broad range of adjuvant therapy options for endometrial cancer.
We observed different adjuvant therapy patterns for the treatment of disease confined to the uterus and for disease with adnexal or nodal involvement. Our survey showed that WPRT and/or vaginal brachytherapy are commonly performed for Stage I endometrial cancer. Four trials evaluated the role of WPRT in endometrial cancer and failed to show improved overall survival \[[@B13][@B14][@B15][@B16]\]. As Postoperative Radiation Therapy for Endometrial Carcinoma-2 (PORTEC-2) showed that vaginal brachytherapy has similar vaginal and pelvic control rates and overall survival to those for WPRT, vaginal brachytherapy is a reasonable choice for patients with stage I endometrial cancer \[[@B4]\]. As evidence that this finding has been adopted Korea, vaginal brachytherapy was the preferred option among KGOG members. Despite adjuvant therapy with WPRT or vaginal brachytherapy, a significant proportion of patients still experience distant metastases. Although the role of adjuvant chemotherapy is currently being studied (GOG 249, PORTEC-3), adjuvant chemotherapy has been adopted among some clinicians even for the treatment of disease confined to the uterus.
Among KGOG members, there is a consensus that patients with stage IIIA and stage IIIC1 disease need adjuvant therapy. However, there is a broad range of opinion on the optimal means of adjuvant therapy. Generally, chemotherapy or CCRT are the preferred options for patients with extrauterine disease. This trend may be a reflection of the current guidelines favoring chemotherapy based on the results of the GOG-122 trial \[[@B17]\]. When we examined survey results from other countries, we found that chemotherapy was the preferred option for adjuvant therapy among Japanese Gynecologic Oncology Group members (79.9%) \[[@B18]\], who performed radiotherapy in just 13% of cases and did not consider CCRT as a treatment option.
While our survey only captures the reported practice patterns of 50% of KGOG members, there is clear variation among practicing gynecologic oncologists. This study is limited by both reporting biases and the response rate. In addition, we did not include non-endometrioid histology and considered only endometrioid adenocarcinoma.
The surgical extent of lymphadenectomy is still one of the most controversial topics in the management of endometrial cancer. Furthermore, there is a broad range of options for adjuvant therapy in the treatment of this disease. In general, radiotherapy is preferred for stage I and stage II disease and chemotherapy is preferred for stage III disease. Currently, the Korean guidelines for treating endometrial cancer are in the process of being revised. For standardizing practice patterns and improving guideline adherence, further studies will be required to identify the differences between actual practice and the revised guidelines once they are published.
**CONFLICT OF INTEREST:** No potential conflict of interest relevant to this article was reported.
Supplementary Material
======================
###### Supplementary Table 1
{#F1}
###### Respondents\' demographics
Variable No. (%)
------------------------------------------------------------------- -----------
Age (yr)
30-40 21 (19.4)
41-50 57 (52.8)
51-60 21 (19.4)
61-70 9 (8.3)
\>70 0
Sex
Male 94 (87.0)
Female 14 (13.0)
No. of years since fellowship training
Currently in fellowship training 8 (7.4)
Up to 5 years since completion 22 (20.4)
6-10 years since completion 27 (25.0)
11-15 years since completion 20 (18.5)
\>15 years since completion 31 (28.7)
No. of endometrial cancer cases managed in your center (per year)
≤20 25 (23.1)
21-40 30 (27.8)
41-80 25 (23.1)
81-100 11 (10.2)
101-150 8 (7.4)
\>150 9 (8.3)
###### Surveyed Korean Gynecologic Oncology Group members\' additional therapy recommendations for incidentally found stage I endometrial cancer
Scenario 2A^\*^ 2B^†^
---------------------------------------- -------- -------
Final pathology IAG1 IAG3
Recommend adjuvant therapy (%) 9.5 85.6
Type of therapy recommended (%)
Laparoscopy for lymph node dissection 100 34.1
Laparotomy for lymph node dissection 0 3.4
Vaginal brachytherapy 0 36.4
Whole pelvic radiation therapy 0 18.2
Chemotherapy 0 8
^\*^2A, incidentally found stage I/grade 1 endometrial cancer. ^†^2B, incidentally found stage I/grade 3 endometrial cancer.
###### Extent of lymphadenectomy for presumed stage I disease
Scenario 3A^\*^ 3B^†^ 3C^‡^
------------------------------------------------------- ---------------- ---------------- ----------------
Preopative biopsy results (grade) 1 1 3
Preopative MRI results (myometrial invasion) Less than half More than half More than half
Extent of surgery (%)
SH/BSO 19.8 0 0
SH/BSO+sentinel biopsy 6.6 2.9 1.9
SH/BSO+selective lymphadenectomy with frozen results 15.1 3.9 1.0
SH/BSO+PLND 31.1 21.4 10.7
SH/BSO+PLND/PALND (up to IMA level) 15.1 35.0 34.0
SH/BSO+PLND/PALND (up to renal vein level) 12.3 36.9 52.4
BSO, bilateral salpingo-oophorectomy; IMA, inferior mesenteric artery; MRI, magnetic resonance imaging; PALND; para-aortic lymph node dissection; PLND, pelvic lymph node dissection; SH, simple hysterectomy.
^\*^3A, presumed stage IA/grade 1 endometrial cancer. ^†^3B, presumed stage IB/grade 1 endometrial cancer. ^‡^3C, presumed stage IB/grade 3 endometrial cancer.
###### Surveyed Korean Gynecologic Oncology Group members\' adjuvant therapy recommendations for stage I endometrial cancer according to stage and grade
Adjuvant options (%) IAG1 IAG2 IAG3 IBG1 IBG2 IBG3
---------------------- ------ ------ ------ ------ ------ ------
Observation 100 70.8 9.4 28.1 8.3 1
VB 0 19.8 46.9 35.4 37.6 17.7
WPRT 0 6.3 28.1 25 26 34.4
WPRT+brachytherapy 0 1 4.2 4.2 13.5 26
Chemotherapy 0 1 7.3 4.2 7.3 8.3
CCRT 0 1 4.2 3.1 7.3 12.5
CCRT, concurrent chemoradiation therapy; VB, vaginal brachytherapy; WPRT, whole pelvic radiation therapy.
###### Surveyed Korean Gynecologic Oncology Group members\' adjuvant therapy recommendations for completely staged endometrial cancer
Scenario Stage II Stage IA and positive cytology Stage IIIC1 Stage IIIA
--------------------------------- ---------------- -------------------------------- --------------------- ---------------------
Adjuvant therapy (%)
Yes 89.6 55.3 100 95.7
No 10.4 44.7 0 4.3
If yes, preferred treatment (%)
1 WPRT (31.8) Chemotherapy (57.7) CCRT (53.2) Chemotherapy (42.0)
2 VB (29.4) CCRT (17.3) Chemotherapy (28.7) CCRT (31.8)
3 WPRT+VB (21.2) WPRT (13.5) WPRT+VB (12.8) WPRT (17.0)
4 CCRT (14.1) Hormone therapy (7.7) WPRT (5.3) WPRT+VB (9.1)
CCRT, concurrent chemoradiotherapy; VB, vaginal brachytherapy; WPRT, whole pelvic radiation therapy.
| {
"pile_set_name": "PubMed Central"
} |
Background
==========
In cancer research, high-throughput profiling has been extensively conducted, searching for genomic signatures with predictive power for traits or clinical outcomes. In this article, we analyze cancer prognosis studies, where the clinical outcomes are metastasis-free, overall, or other types of survival. We focus on microarray gene expression studies but note that the proposed approach is also applicable to data generated using other profiling techniques. In recent studies, gene signatures have been constructed for the prognosis of breast cancer, lymphoma, ovarian cancer, and cancers of many other organs \[[@B1]\].
The construction of molecular signatures for cancer prognosis has been investigated in many studies. This study complements and significantly advances from existing studies along the following directions. First, it accounts for the coordination among genes using the weighted coexpression network, whereas many existing studies ignore such coordination and assume the interchangeability of genes. Second, with properly constructed representative features, the proposed approach can accommodate the second-order effects of genes, whereas in many existing studies, only the linear effects of genes are considered. More importantly, this study provides convincing evidence that the higher-order representative features, which have often been neglected, can improve the predictive power. Thus, this study provides a way to improve over existing methodologies for the construction of prognosis signatures.
In the analysis of cancer genomic data, dimension reduction or feature selection is usually needed along with model estimation. Feature selection methods target at selecting a subset of genes, whereas dimension reduction methods construct a small number of representative features (sometimes referred to as \"super genes\" or \"latent genes\" in the literature) using the linear combinations of all genes. The approach developed in this article belongs to the family of dimension reduction approaches. Published studies have shown that the performance of feature selection and dimension reduction methods is data-dependent with no one dominating another \[[@B2],[@B3]\].
Many existing analysis methods assume the interchangeability of gene effects and ignore the interplay among them. Extensive biomedical studies have shown that there is an inherent coordination among genes and, essentially, all biological functions of living cells are carried out through the coordinated effects of multiple genes \[[@B4],[@B5]\]. Gene pathways and networks are perhaps the two most effective ways to describe the coordination among genes. Compared with pathway-based analysis, network-based analysis may have the following advantages. First, network-based analysis can use all the genes, whereas pathway-based analysis uses only annotated genes. Since many genes are not or only partially annotated, network-based analysis can be more comprehensive than pathway-based analysis. Second, in network-based analysis, the \"distances\" between genes are weighted (i.e., continuous measurements). Unlike in pathway-based analysis, we can infer not only whether two genes are connected but also the strength of connectedness. In this article, we focus on network-based analysis and defer a comprehensive comparison of pathway- and network-based methods to future studies.
In network analysis, nodes represent genes. Nodes are connected if the corresponding genes have similar biological functions and/or similar expression patterns across samples. There are subsets of nodes called \"modules\" that are tightly connected to each other. In this article, we adopt the weighted coexpression network (<http://www.genetics.ucla.edu/labs/horvath/CoexpressionNetwork>), which is built on the understanding that the coordinated coexpressions of genes encode interacting proteins with closely related biological functions and cellular processes \[[@B6]\]. Extensive empirical studies have shown that modules in the weighted coexpression networks often have important biological implications. Moreover, genes within the same modules tend to have related biological functions \[[@B7],[@B8]\].
In cancer prognosis studies, the sample sizes are often small. With some modules having large sizes, dimension reduction is needed when conducting module-based network analysis. The approach developed in \[[@B9],[@B10]\] and references therein proceeds as follows. First, principal component analysis is conducted within each module separately. Second, the first principal components, referred to as \"eigengenes\" in the literature, are identified. Following \[[@B11]\] and others, we refer to the principal components as \"representative features\". Third, the representative features are used as covariates in downstream model-building. There has been a rich literature on principal component analysis in gene expression studies \[[@B1],[@B11]-[@B13]\]. Of note, many of those studies, for example \[[@B13]\], also recommend using only the first principal components.
In this article, for cancer prognosis studies with microarray gene expression measurements, we use the weighted coexpression networks and corresponding modules to describe the coordination among genes. Building on existing eigengene-based studies, we investigate the effects of higher-order representative features, which include principal components other than the first ones and second-order terms (quadratics and interactions). This study significantly differs from published ones. Specifically, unlike \[[@B1]\] and others, it is conducted in the context of network analysis, which may provide further insights into cancer biology beyond gene- and pathway-based analyses. Unlike in \[[@B9],[@B10]\], higher-order representative features are investigated and shown to have important implications for prediction. In addition, since the dimensionality of representative features considered in this study is considerably higher than that in previous studies, regularized estimation is conducted. Unlike in \[[@B11]\], gene modules (as opposed to pathways) are the functional units. More importantly, this study considers the joint effects of multiple modules, whereas \[[@B11]\] studies the marginal effects.
Methods
=======
Data and Model
--------------
### Construction of weighted coexpression network and its modules
Besides the weighted coexpression network, there are other ways of constructing gene networks. Examples include the Boolean network, the Bayesian network, use of continuous models, and others. Compared with other networks, the weighted coexpression network has the following advantages. First, it uses only gene expression measurements and does not require any additional biological experiments. Second, it is computationally simple and can be constructed using existing software. And third, quite a few published studies have shown that it has satisfactory empirical performance \[[@B7],[@B8]\]. On the other hand, it may also have limitations. The network is defined based on the correlations among gene expressions, which may not contain all information on the coordination among genes. In addition, the network construction is unsupervised. In this article, we focus on the weighted coexpression network and defer the comparison of different networks to future studies.
Construction of the weighted coexpression network consists of the following steps.
1\. For genes *k*and *j*(= 1\... *d*), compute *cor*(*k, j*), the Pearson correlation coefficient of their expressions. Compute the similarity measure *S*(*k, j*) = \|*cor*(*k, j*)\|;
2\. Compute the adjacency function *a~k,j~*= *S^b^*(*k, j*), where the adjacency parameter *b*is chosen using the scale-free topology criterion;
3\. For gene *k*(= 1 \... *d*), compute its connectivity $C_{k} = {\sum_{u = 1}^{d}a_{k,u}}$;
4\. For genes *k*and *j*(= 1 \... *d*), compute the topological overlap based dissimilarity measure *d~k,j~*= 1 *- ω~k,j~*where *ω~k,j~*= (*l~k,j~*+ *a~k,j~*)/(*min*(*C~k,~C~j~*) + 1 *- a~k,j~*) and $l_{k,j} = {\sum_{u = 1}^{d}{a_{k,u}a_{j,u}}}$. Define the dissimilarity matrix *D*, whose (*k, j*)th element is *d~k,j~*;
5\. Conduct hierarchical clustering with matrix *D*. Apply the dynamic tree cut approach to cut the clustering tree (dendrogram) and identify the resulting branches as gene modules \[[@B14]\].
In Steps 1 and 2, the adjacency measure between genes is defined using the power transformation of correlation coefficients. We adopt the weighted network, which can measure not only whether two genes are connected but also the strength of connection. The power adjacency function has the attractive factorization property. In our study, we find that for the six datasets analyzed, *b*= 6 (which has been suggested in published studies) can lead to results nearly satisfying the scale-free topology criterion. This criterion has been motivated by the observation that metabolic networks exhibit scale-free topology. In Step 4, we use the topological overlap dissimilarity measure \[[@B15]\], which has been found to result in biologically meaningful modules. In addition, this measure is relatively robust in describing the interconnectedness between genes \[[@B8]\]. The advantage of the dynamic tree cut approach in Step 5 has been discussed in \[[@B14]\].
### Construction of representative features
Assume that *M*modules have been constructed. For module *m*(= 1 \... *M*), denote *m~i~*as the number of genes within this module and $U_{1}^{m}...\, U_{m_{i}}^{m}$ as the gene expressions. With principal component analysis, any linear combination of *U*s can be written as
$$\alpha_{1}U_{1}^{m} + ... + \alpha_{m_{i}}U_{m_{i}}^{m} = \gamma_{1}X_{1}^{m} + ... + \gamma_{m_{k}}X_{m_{k}}^{m},$$
where $X_{1}^{m}...\, X_{m_{k}}^{m}$ are the *m~k~*PCs and *m~k~*is the rank of $(U_{1}^{m},....,U_{m_{i}}^{m})$. In particular, $X_{i}^{m}\text{s}$ have unit norms and are the linear combinations of $U_{i}^{m}\text{s}$, $X_{i}^{m}$ and $X_{j}^{m}$ are orthogonal to each other when *i*≠ *j*. Variation explained by $X_{i}^{m}$ decreases as *i*increases.
Several studies propose using $(X_{1}^{1},...,X_{1}^{M})$ (i.e., the first principal components from the *M*modules) as covariates in downstream analysis \[[@B9],[@B10],[@B13]\]. In this study, we are interested in not only the first principal components but also the other principal components, as well as quadratics and interactions of principal components. Specifically, we consider the following four sets of representative features:
(R1) Consider $\left\{ X_{1}^{1},...,X_{1}^{M} \right\}$. That is, the first principal components from all modules. This set of *M*representative features has been investigated in previous studies and will serve as a benchmark. To unify notations, denote *Z*~0,\ *i*~= *X*~1~*^i^*(*i*= 1 \... *M*) and *Z*= (*Z*~0,1~*, \..., Z*~0,*M*~);
(R2) Denote *Z*~0,*i*~= *X*~1~*^i^*(i = 1\... M). For 1≤ *i≤j≤M*, define *Z~i,j~*= *Z*~0,*i*~× *Z*~0,*j*~. Consider *Z*= (\..., *Z*~0,*i*~*,.., Z~i,j~*,\...). This set of representative features is composed of the first principal components from all modules, their quadratics, and their second order interactions;
(R3) Within module *m*(= 1,\...,*M*), select the top *m\**principal components such that *ξ*% of the total variation of gene expressions is explained \[[@B16]\]. Define $P = {\sum_{m = 1}^{M}{m*}}$. Consider $Z = (Z_{0,1},...,Z_{0,P}) = (X_{1}^{1},...,X_{1}^{1},...X_{1}^{M},...,X_{M}^{M}{}_{*})$. This set of representative features is composed of principal components that can sufficiently explain the variation of gene expressions. In our study, we set *ξ*% = 80%, which is slightly smaller than that adopted in \[[@B16]\]. Our data analysis suggests that, because of the extremely noisy nature of gene expressions, a huge number of principal components are needed to explain, for example, 95% of the variation;
(R4) We first construct the *P*principal components as with (R3). Denote $(Z_{0,1},...,Z_{0,P}) = (X_{1}^{1},...,X_{1}^{1},...X_{1}^{M},...,X_{M}^{M}{}_{*})$. For 1≤ *i*≤ *j≤P*, define *Z~i,j~*= *Z*~0,*i*~× *Z*~0,*j*~. Consider *Z*= (\..., *Z*~0,*i*~*,\..., Z~i,j~,..*.), the set composed of the *P*principal components and their quadratics.
Among the above sets of representative features, (R1) has been investigated elsewhere and will be used as a benchmark. With (R3), we analyze multiple principal components per module. In the detection of marginally differentially expressed pathways, Ma and Kosorok \[[@B11]\] show that higher-order principal components may have important implications. With (R2) and (R4), we are able to account for the interactions among genes within the same modules. More importantly, we are able to account for the interactions among different modules using their principal components. The relatively small number of principal components per module makes it possible to study the interactions, which are extremely difficult to study in gene-based analysis.
Following a similar spirit, there may be other ways of defining the representative features. For example, for module *m*, it is possible to accommodate the interactions among genes by conducting principal component analysis with the set $\left\{ U_{i}^{m}:1 \leq i \leq m_{i} \right\} \cup \left\{ U_{i}^{m} \times U_{j}^{m}:1 \leq j \leq m_{i} \right\}$. However, when the module sizes are large, such construction may be computationally expensive. Our exploration suggests that (R1)-(R4) are the relatively simpler and more intuitive ways of constructing the representative features.
### Statistical modeling
Denote *T*and *C*as the survival and random censoring times, respectively. We assume that the gene expressions are associated with cancer survival through the Cox proportional hazards model, where the conditional hazard function is λ(*t\|Z*) = λ~0~(*t*) exp(*β\' Z*). Here λ~0~(*t*) is the unknown baseline hazard, and *β*is the unknown regression coefficient. Under right censoring, one observation consists of (*Y*= *min*(*T, C*), Δ = *I*(*T≤C*), *Z*). Assume *n*iid observations (*Y~i~, δ~i~*, *Z~i~*)*, i*= 1 \... *n*. Denote *r~i~*= {*k*: *Y~k~≥ Y~i~*} as the at-risk set at time *Y~i~*. The log-partial likelihood function is $R(\beta) = {\sum_{i}\delta}_{i}\left\{ \beta^{\prime}Z_{i} - \log({\sum_{k \in r_{i}}{\exp(\beta^{\prime}Z_{k})}}) \right\}$. Here we describe the relationship between genes and cancer survival using the representative features. As the representative features are functions of genes and their second-order terms, we can rewrite the Cox models in terms of genes.
Regularized estimation
----------------------
Although the dimensionality of *Z*is expected to be smaller than that of the original gene expressions, it may still be comparable to or even larger than the sample size, particularly with (R2)-(R4). In addition, it is possible that only a subset of the representative features is associated with cancer survival. Thus, we consider regularized estimation, which can effectively \"stabilize\" estimation and discriminate cancer-associated representative features from noises. With (R1) and (R3), we use the TGDR (Threshold Gradient Directed Regularization) approach \[[@B17]\]. As shown in \[[@B17]\] and follow-up studies, TGDR has a lower computational cost and empirical performance comparable to or better than that of alternative methods. With (R2) and (R4), we modify the TGDR to better accommodate the second-order terms. Particularly, under the modified approach, when a second-order term is included in the model, its corresponding first-order terms are automatically included.
### The TGDR algorithm
The TGDR approach can be used for regularized estimation when representative features (R1) and (R3) are adopted. Let Δ*ν*be a small positive increment. In numerical studies, we set Δ*ν*= 10^-3^. Denote 0 ≤ *τ*≤ 1 as the threshold. The TGDR algorithm consists of the following steps.
1\. Initialize *β*= 0;
2\. With the current estimate of *β*, compute the vector of gradient *g*= ∂*R*(*β*)/∂*β*. Denote the *j*th element of *g*as *g~j~*;
3\. Compute the thresholding vector *f*, where its *j*th element is *f~j~*= *I*(\|*g~j~\| ≥τ*× *max~l~\|g~l~\|*);
4\. Update the estimate *β~j~*= *β~j~*+ Δ*ν*× *g~j~*× *f~j~*;
5\. Iterate Steps 2-4 *K*times, where *K*is determined via cross validation.
### The modified TGDR algorithm
When representative features (R2) are (R4) are used, the following algorithm can better accommodate the
second order terms.
1\. Initialize *β*= 0;
2\. With the current estimate of *β*, compute the vector of gradient *g*= ∂*R*(*β*)/∂*β*. Denote *g~i,j~*as the component of *g*that corresponds to *Z~i,j~*;
3\. Compute the thresholding vector *f*. Denote *f~i,j~*as its component corresponding to *Z~i.j~*. Define
$$f_{i,j} = \begin{cases}
{I(|g_{i,j}| \geq \tau \times max_{l,u}|g_{l,u}|)} & {\text{for}\, i > 0,j > 0} \\
{I(|g_{i,j}| \geq \tau \times max_{l,u}|g_{l,u}|)} & \\
{\text{OR}{\sum_{u}{f_{i,u} > 0\,}}\text{OR}{\sum_{u}{f_{j,u} > 0}})} & {\text{for}\, i = 0,j = 0.} \\
\end{cases}$$
4\. Denote *β~i,j~*as the component of *β*that corresponds to *Z~i,j~*. Update the estimate *β~i,j~= β~i,j~+*Δν *× g~i,j~× f~i,j~*,
5\. Iterate Steps 2-4 *K*times, where *K*is determined via cross validation.
### Remarks
The above two approaches use thresholding for regularized estimation and feature selection. Specifically, at each iteration, the gradients, which measure the relative importance of representative features, are computed. More important representative features tend to have larger gradients. Only the important representative features are selected, and their estimated regression coefficients are updated. The iteration stops when a cross validation-based criterion is reached. The second algorithm ensures that, if a second-order term is selected, its corresponding first-order terms are selected. This cannot be automatically achieved with the first algorithm. We refer to \[[@B17]\] for a more detailed discussion of thresholding regularization. Both approaches involve tuning parameters *τ*and *K*, which are selected using V-fold cross validation. In data analysis, we set *V*= 5 and search over *τ*= 1.0, 0.95, 0.9, \..., 0.05, 0. Research code written in R is available at <http://www.med.yale.edu/eph/faculty/labs/ma/> for the construction of network modules and representative features and regularized estimation.
Results
=======
Analysis of cancer prognosis studies
------------------------------------
### Data collection
We collect six cancer prognosis studies with microarray measurements. We refer to them as data D1-D6 and provide brief descriptions below and in Table [1](#T1){ref-type="table"}.
######
Description of datasets.
Data Disease Platform Gene Sample
------------------------------- --------------- ----------------- ------- --------
D1: Rosenwald et al. (2003) MCL cDNA 8810 92
D2: Dave et al. (2004) FL Affymetrix 44928 187
D3: Rosenwald et al. (2002) DLBCL cDNA 7399 240
D4: Sotiriou et al. (2003) Breast cancer cDNA 7650 98
D5: van\'t Veer et al. (2002) Breast cancer Oligonucleotide 24481 78
D6: Huang et al. (2003) Breast cancer Affymetrix 12625 71
Gene/Sample: number of genes/subjects profiled.
**D1**. A study using microarray expression analysis of mantle cell lymphoma (MCL) was reported in \[[@B18]\]. Among 101 untreated patients with no history of previous lymphoma, 92 were classified as having MCL based on established morphologic and immunophenotypic criteria. Survival times of 64 patients were available, and the other 28 patients were censored. The median survival time was 2.8 years (range 0.02 to 14.05 years). Lymphochip DNA microarrays were used to quantify mRNA expressions in the lymphoma samples from the 92 patients. Gene expression data on 8,810 cDNA elements was available.
**D2**. A study was conducted to determine whether the survival risk of patients with follicular lymphoma (FL) can be predicted by gene expression profiles of the tumors \[[@B19]\]. Fresh-frozen tumor biopsy specimens from 191 untreated patients who had received a diagnosis of follicular lymphoma between 1974 and 2001 were obtained. The median age at diagnosis was 51 years (range: 23 to 81), and the median follow-up time was 6.6 years (range: less than 1.0 to 28.2). The median follow-up time among patients alive at the last follow-up was 8.1 years. Eight records with missing survival information are excluded from analysis. Affymetrix U133A and U133B microarray genechips were used to measure gene expressions of 44,928 probes.
**D3**. Rosenwald et al. \[[@B20]\] reported a diffuse large B-cell lymphoma (DLBCL) prognosis study. This study retrospectively collected tumor biopsy specimens and clinical data for 240 patients with untreated DLBCL. The median follow-up was 2.8 years, with 138 observed deaths. A lymphochip cDNA microarray was used to measure the expressions of 7,399 genes.
**D4**. Sotiriou et al. \[[@B21]\] reported a study correlating gene expression measurements generated using cDNA with clinico-pathological characteristics and clinical outcomes in an unselected group of 99 node-negative and node-positive breast cancer patients. In the original analysis, the Cox model was used to identify genes that were marginally significantly associated with relapse-free survival. In this study, we analyze the 98 patients with complete survival information.
**D5**. Van\'t Veer et al. \[[@B22]\] reported a breast cancer prognosis study investigating the time to distant metastasis. Ninety-seven (97) lymph node-negative breast cancer patients 55 years old or younger participated in this study. Among them, 46 developed distant metastases within 5 years. Complete information was available for 78 subjects. Expression levels of 24,481 probes were measured.
**D6**. Despite major progress in breast cancer treatment, the ability to predict metastasis of the tumor remains limited. Huang et al. \[[@B23]\] reported a study investigating metastastic states and relapses in breast cancer patients. Affymetrix genechips were used for the profiling of 71 samples. Expression measurements on 12,625 probes were available.
Among the above studies, three used cDNA, one used oligonucleotide arrays, and two used Affymetrix genechips for profiling. We process each dataset separately as follows. We conduct microarray normalization using a lowess normalization approach for cDNA data and a robust normalization approach for Affymetrix data \[[@B24]\]. We impute missing measurements using the K-nearest neighbors approach. We select 2,000 genes with the largest variances for downstream analysis. Since we expect the number of genes associated with cancer prognosis to be far less than 2,000, and since we are more interested in genes with a high level of variation, we conduct this unsupervised screening to reduce computational cost. In addition, recent studies have shown that prescreening may improve feature selection accuracy \[[@B25]\]. We then rescale gene expressions to have zero median and unit variance.
### Evaluation of prediction performance
Main objectives of cancer genomic studies include marker identification and predictive model-building. Despite the fast accumulation of knowledge on the biological functions of genes, there is still a lack of commonly accepted, objective ways of determining the accuracy and implications of identified markers. Thus, as in many published studies, we focus on the prediction performance of the models and identified representative features. It is expected that the evaluation of prediction performance can also provide an indirect evaluation of the biological implications of the models and representative features. For prognosis of some cancers, for example breast cancer, there are multiple independent studies. Thus, there is a possibility of making cross-study prediction. However, cross-study prediction demands comparability between studies \[[@B26]\]. Without having access to all the details on experimental set-up, we are unable to determine whether there are prognosis studies fully comparable to the six studies we analyze. Thus, in the study, we choose not to conduct cross-study prediction evaluation.
As an alternative, we consider a cross validation-based approach. We acknowledge that cross validation-based evaluation has a small chance of generating overly optimistic results. However, the adopted V-fold cross validation-based approach is expected to be reasonably objective. In addition, different sets of representative features are evaluated using the same approach. Thus, the evaluation results are expected to be meaningful. The cross validation-based evaluation proceeds as follows. (a) Randomly partition data into V subsets with equal sizes. In our numerical study, set *V*= 5; (b) For *υ*= 1 \... *V*, remove subset υ from data; (c) With reduced data, carry out the cross validation and regularized estimation. Denote the estimated regression coefficient as ${\hat{\beta}}^{({- \upsilon})}$ (d) Compute the predictive risk scores ${\hat{\beta}}^{({- \upsilon})}{}^{\prime}$*Z*for the removed subjects; (e) Repeat Steps (b)-(d) over all subsets; (f) Compute two summary statistics. (f.1) The first is the logrank statistic \[[@B27]\]. Dichotomize the predictive scores at the median. Create two risk groups. Compute the logrank statistic, which measures the difference of survival between the two groups. Under the Null, the representative features have no predictive power, and the logrank statistic is χ^2^distributed with degree of freedom 1. We compute the logrank statistic using the R function *survdiff*. (f.2) The second is the concordance index, which is computed using the R function *rcorr.cens*. A larger concordance index indicates better predictive power, with concordance index equal to 0.5 corresponding to random guess. This evaluation approach has been extensively used in cancer genomic studies.
### Analysis results
For each dataset, we construct the weighted coexpression network and its modules. For dataset D1-D6, 12, 10, 13, 10, 11, and 13 modules are constructed, respectively. More details are available in Additional File [1](#S1){ref-type="supplementary-material"}. We then construct the four sets of representative features.
We conduct the cross validation-based prediction evaluation and present the logrank statistics and concordance indices in Table [2](#T2){ref-type="table"}. We can see that, when (R1), the first principal components from all modules, are used, five out of the six logrank statistics are significant at the 0.05 level. This observation is in line with the satisfactory results observed in \[[@B9],[@B10]\] and others. We also find that, the (R2)-(R4) logrank statistics can be larger than those of (R1), which suggests that prediction performance can be improved by incorporating higher-order representative features. The improvement is considerably large for dataset D2 and D4. Another finding is that the prediction performance of representative features (R2)-(R4) is data-dependent. Particularly, two datasets have (R2), two have (R3), and the other two have (R4) logrank statistics as the largest. Examining the concordance indices suggests reasonable predictive power of the representative features and similar conclusions as with the logrank statistics.
######
Data analysis results: prediction logrank statistics and concordance indices.
Logrank statistic Concordance index
---- ------------------- ------------------- ----------- ----------- ------ ---------- ---------- ----------
D1 15.30 19.10 **34.70** 0.18 0.74 0.70 **0.77** 0.50
D2 0.25 **4.56** 0.60 0.46 0.61 **0.67** 0.51 0.58
D3 10.40 0.32 **19.00** 2.14 0.62 0.53 **0.69** 0.55
D4 3.89 **13.80** 11.40 0.01 0.63 **0.66** 0.64 0.54
D5 7.95 7.50 7.50 **12.30** 0.72 0.70 0.70 **0.76**
D6 6.27 2.15 6.46 **7.99** 0.65 0.61 0.69 **0.73**
Larger logrank statistics and concordance indices correspond to more predictive power. A logrank statistic greater than 3.84 is significant at the 0.05 level.
To gain further insights, we also conduct the following analysis. The TGDR and modified TGDR algorithms are capable of selecting a small number of important representative features. For each dataset, we examine the selection results for the representative feature set with the largest logrank statistic. Detailed results are presented in Additional File [2](#S2){ref-type="supplementary-material"}. In addition, although the models are constructed using representative features, we can rewrite using genes and their second-order terms. Principal components are the linear combinations of all genes within specific modules. The statistical models we construct are sparse at the representative feature level but not at the gene level. We examine the top 20 genes and/or interactions of genes with the largest regression coefficients. Since all gene expressions have been normalized to have equal variances, the magnitude of regression coefficients can provide a rough measure of the relative importance of genes.
**D1**. The representative features (R3) are adopted. Six principal components (\#2, 5, 6, 7, 9, 11) from module \#2 and 1 principal component from module \#9 are identified. It is interesting to note that in module \#2, the first principal component is not identified. Among the 20 genes with the largest regression coefficients, there are established cancer markers, including for example genes PTK2, PCNA, and PRKACA. There are also new discoveries that need further investigation.
**D2**. The representative features (R2) are adopted. We identify 6 principal components, quadratics of 4 principal components, and 4 interactions among 5 principal components. We conclude that among the 10 modules, only 6 are cancer-associated. In addition, the interactions among modules have non-ignorable effects. We also examine the top 20 regression coefficients and find that 9 of them come from individual genes and 11 come from interactions of genes.
**D3**. The representative features (R3) are adopted. Five out of 13 modules are identified as associated with prognosis. More specifically, 15, 1, 4, 1, and 1 principal components are identified in module \#2, 4, 5, 12, and 13, respectively. For all of the 5 identified modules, the first principal components are identified. We examine the 20 genes with the largest regression coefficients and find that quite a few belong to the MHC (major histocompatibility complex) family. Of note, we conduct probe-level (as opposed to gene-level) analysis, with the consideration that different probes may correspond to different segments of the same genes.
**D4**. The representative features (R2) are adopted. Eight out of 10 modules are identified as associated with prognosis. More specifically, we identify 8 principal components, quadratics of 3 principal components, and 10 interactions among 8 principal components. As with data D2, we also observe the nonzero effects of interactions among modules. When examining the top 20 regression coefficients, we find that 12 come from individual genes, 1 comes from the quadratic of a gene, and the remaining 7 come from interactions of genes.
**D5**. The representative features (R4) are adopted. Six out of 11 modules are identified as associated with prognosis. Among them, the first principal components are identified in 5 modules (all except module \#5). We identify 18 principal components, quadratics of 8 principal components, and 12 interactions among 14 principal components. When examining the top 20 regression coefficients, we find that all of them come from individual genes. Among the top 20 genes, there are several known breast cancer markers, including for example genes IL8, N-myc, PRKA6, and others.
**D6**. The representative features (R4) are adopted. Three out of 13 modules are identified as associated with prognosis. The first principal component is identified in only 1 of the 3 modules. We identify 10 principal components, quadratics of 2 principal components, and 7 interactions among 9 principal components. When examining the top 20 regression coefficients, we find that all of them come from individual genes.
Simulation
----------
To better understand properties of the proposed representative features and regularized estimation approaches, we conduct simulation studies. As observed gene expressions usually do not fit specific parametric distributions \[[@B28]\], we randomly sample gene expressions of 200 subjects without replacement from D1-D3 combined (the lymphoma datasets). We use subjects as the sampling units so that the correlation structure among genes is kept. We then randomly split the 200 samples into a training set and a testing set, each with 100 subjects. We construct the weighted coexpression network, its modules, and 4 sets of representative features using the training set. With (R1)-(R4), respectively, we randomly select 10 representative features as associated with prognosis and set the rest as noises. The prognosis-associated representative features have their non-zero regression coefficients generated randomly from *Unif*\[-0.5, 0.5\]. The survival times are then generated from the Cox model with λ~0~(*t*) = 0.5 (i.e., constant baseline hazard). The censoring times are generated independent of survival. We adjust the censoring distribution so that the censoring rate is about \~ 40%. Thus, there are a total of 4 different data-generating models, with (R1)-(R4) being the \"true\" representative features.
With the training set, we use the representative features (R1)-(R4) and proposed regularized approaches for estimation. This step reflects the fact that, in practice, it is unknown which set of representative features is appropriate. We make predictions for subjects in the testing set using the training set estimates. The logrank statistic and concordance index are computed for evaluation of prediction performance. We note that, unlike in practical data analysis, in the above simulation, the training and testing sets are completely independent.
Summary statistics based on 500 replicates are shown in Table [3](#T3){ref-type="table"}. Simulation suggests the importance of properly specifying the representative features. With data S1, where (R1) is the model generating representative features, all four sets of representative features can lead to satisfactory prediction performance. This observation is reasonable considering that (R1) is a subset of (R2)-(R4). Similar observations and reasonings hold for data S2 and S3. However, when (R4) is the true data generating representative features, results under (R1)-(R3) are significantly less satisfactory.
######
Simulation study: mean prediction logrank statistics and concordance indices based on 500 replicates.
Logrank statistic Concordance index
---- ------------------- ------------------- ------- ------- ------ ------ ------ ------
S1 94.15 94.92 90.72 88.57 0.95 0.95 0.95 0.94
S2 4.92 59.62 7.32 82.03 0.60 0.88 0.62 0.93
S3 39.45 45.70 76.19 68.53 0.80 0.82 0.90 0.88
S4 2.27 29.09 4.54 80.57 0.57 0.79 0.60 0.93
Simulation seems to suggest that (R4), the most complicated set of representative features, is the proper choice under all four simulation scenarios. A drawback of (R4) is its high computational cost, particularly when there are a moderate to large number of modules, which may make it less appealing in practical data analysis. In addition, the simulation settings may still be overly simplified compared with what is observed. With real data, as can be seen from Table [2](#T2){ref-type="table"} (R4) is not necessarily dominatingly better. There are multiple reasons for the different patterns observed in Table [2](#T2){ref-type="table"} and [3](#T3){ref-type="table"}. The first is that, with practical data, (R1)-(R4) do not necessarily include the true data-generating mechanisms. The second is that, with real data, there may not be a clear cut between signals and noises. Instead of a small number of large signals, there may be a large number of small signals. In addition, with simulated data, the survival is determined by gene signatures. In contrast, in practice, the survival may also be affected by other risk factors such as cancer treatment history, which explains the smaller predictive power observed in Table [2](#T2){ref-type="table"}.
Discussion
==========
For cancer prognosis studies with gene expression measurements, we describe the interplay among genes using the weighted coexpression network and use principal component analysis techniques to reduce the dimensionality of gene expressions. This study complements and advances from existing studies by investigating the contribution of higher-order representative features to predictive power. The four sets of representative features investigated in this study share some desired properties with other principal components-based analysis. For example, the computational cost is affordable, and the majority of the variation of gene expressions can be accounted for.
As the dimensionality of representative features may be moderate to large, the TGDR and a modification of the TGDR are used for regularized estimation and feature selection. In \[[@B17]\] and several follow-up studies, it is shown that the TGDR has performance comparable to or better than that of existing alternatives. As the TGDR cannot automatically accommodate the second-order terms, a modification of the TGDR is necessary. Examination of Table [2](#T2){ref-type="table"} shows that some of the prediction logrank statistics and concordance indices are small, suggesting possible local optimums. We note that there are many available approaches that can be used for regularized estimation. For example, penalization approaches have attracted extensive attention in recent statistical and bioinformatic literature \[[@B2]\]. However, we note that most existing (including penalization) approaches may have a problem with local optimums. A satisfactory solution to this problem is highly challenging and warrants separate investigation.
In this study, the proposed research question is investigated using both real and simulated data. In data analysis, without data from independent comparable studies, we conduct cross validation-based prediction evaluation. Such an evaluation is expected to be reasonably fair. However, independent confirmation studies will be needed to fully validate the findings. With the 6 real datasets analyzed, 3 different sets of representative features have the best prediction performances. This finding is in line with \[[@B11]\] and is not surprising considering the extreme complexity and heterogeneity of cancer. Examination of individual regression coefficients suggests that different datasets may have significantly different scenarios. Particularly, for some datasets, the quadratics and interactions among genes may have important implications. Our investigation does not yield a way to suggest the \"optimal\" representative features. Our recommendation is that, in practical analysis, *researchers need to experiment with different sets ofrepresentative features*.
In some previous network-based analysis, geneset enrichment analysis has been conducted to investigate whether modules identified as associated with prognosis are enriched with certain pathways or represent certain biological processes. We note that such analysis is also possible in this study. However, consider a hypothetical module with only 2 principal components. Consider the following two different scenarios. Under scenario 1, the second principal component is identified as associated with prognosis. Under scenario 2, the first principal component and its quadratic are identified as associated with prognosis. Under both scenarios, this module is identified as associated with prognosis. However, an important goal of this study is to discriminate between those two scenarios. Considering that the enrichment analysis will lead to the same results under those two scenarios and thus can be misleading, we choose not to conduct enrichment analysis.
We have investigated second-order representative features. In a similar manner, it is possible to consider third- or even higher-order terms. Such an effort may considerably increase the dimensionality and computational cost. We construct the representative features in an unsupervised manner, which has low computational cost and can be easily implemented using existing software. In recent principal component analysis studies, it has been suggested that supervised methods may outperform unsupervised methods \[[@B13],[@B29]\]. It is possible to construct the supervised counterparts of the proposed representative features.
Conclusions
===========
In this study, we propose using principal component analysis-based representative features for dimension reduction in weighted coexpression network analysis. The proposed representative features and TGDR regularized estimation provide an effective way of reducing the dimensionality, accounting for the interactions among genes within the same modules, and, more importantly, accounting for the interactions among modules. The investigation on the interactions may provide a useful addition to the literature. Our most important finding is that incorporating higher-order representative features leads to improved prediction performance, which may help build better predictive models for cancer prognosis.
Competing interests
===================
The authors declare that they have no competing interests.
Authors\' contributions
=======================
All authors were involved in the study design and writing. SM and YD conducted numerical studies. All authors read and approved the final manuscript.
Pre-publication history
=======================
The pre-publication history for this paper can be accessed here:
<http://www.biomedcentral.com/1755-8794/4/5/prepub>
Supplementary Material
======================
###### Additional file 1
**Results on network module construction**. This additional file contains the details on the network modules constructed using WGCNA.
######
Click here for file
###### Additional file 2
**Analysis results**. This additional file contains the detailed analysis results for dataset D1-D6.
######
Click here for file
Acknowledgements
================
We would like to thank the editor and two referees for insightful comments, which have led to significant improvement of the paper. This study has been supported by NSF grants DMS-0904181 (SM) and DMS-0904184 (MK), NIH grants LM009754, CA120988, CA142774 (SM and JH), CA142538 (MK), and Research of Longitudinal Data Analysis Methodology and Its Application (2009JJD910002) from Key Research Institute of Humanities and Social Sciences Program, China (SM).
| {
"pile_set_name": "PubMed Central"
} |
Introduction
============
In athletics and rehabilitation, functional capacity can be evaluated through different methods; one of them is the vertical jump test. In addition, the vertical jump serves as a predictor of anaerobic capacity, motor development and athletic ability in sports \[[@ref1]-[@ref7]\]. Other studies report that the vertical jump test seems to be an indicator for assessing functional capacity in the elderly \[[@ref8]\] and children \[[@ref9]\]. Vertical jump capacity can be measured by different variables such as vertical take-off speed, flight time, mechanical power or the displacement of the center of body mass \[[@ref10]-[@ref13]\].
Some of the classic measurement tools to calculate these variables are the force platform, video-analysis systems, photoelectric cells and contact mats \[[@ref10]-[@ref13]\]. These methods for vertical jumping measurement have excellent validity for laboratory studies. However, these tools are expensive and difficult to transport, creating difficult transferability to other environments or professional applications. Furthermore, in recent years new technologies have begun to be used for human motion studies, such as inertial sensors, which are small and portable, and provide solutions to the drawbacks of other commonly used instruments for human motion analysis \[[@ref14]\]. In some studies, accelerometric systems have been used to estimate vertical jump capacity \[[@ref15]-[@ref19]\]; these are based on the use of acceleration peak values recorded in the performance of vertical jumps.
Currently, the latest mobile phone generation usually includes inertial sensors with subunits such as accelerometers and gyroscopes that can detect acceleration and the inclination of devices. Numerous apps that display, store and transfer inertial sensor data have been developed for the operation of different mobile phones. These apps have great potential for tracking human motion parameters for research and clinical practice. Apps are being developed for use in different situations related to human movement, such as the pedometer \[[@ref20]\], or the development of an assessment tool and quantification of kinematic variables related to the fragility of the elderly \[[@ref21]\]. The wide availability of mobile phones, due to the variety of use in the daily lives of most people in developed countries, as well as their small size and portability, make them very useful tools for field study development and subsequent use in professional practice \[[@ref14]\].
Because of the advantages offered by mobile phones as tools for the study and analysis of human movement, it is of interest to check their ability to assess and analysis vertical jump tests. The aim of this study was to describe and analyze kinematics characteristics using the inertial sensor incorporated in the iPhone 4S, lower limb strength through a manual dynamometer, and the jump variables obtained with a contact mat in the squat jump (SJ) and countermovement jump (CMJ) tests from a cohort of healthy people. The squat jump (SJ) is defined as a jump that is performed from a squatting position. A counter movement jump (CMJ), which is higher, is a jump where the jumper starts from an upright standing position, makes a preliminary downward movement by flexing at the knees and hips, then immediately jumps up from that position.
Methods
=======
Design and Participants
-----------------------
This was a cross-sectional study, involving 81 jumps from 27 participants. The participants were young Health Sciences students from the University of Málaga (Spain). Participants had to meet the inclusion criteria of being healthy adults aged 18 to 35 years, without musculoskeletal or neurological dysfunction. Subjects with any of the following criteria were excluded: a history of heart disease, surgical interventions in the last year, any disability that would make the correct achievement of the tests difficult, any pain that prevented the completion of tests or neuromuscular pathology that could be aggravated by participating in the study. A physical therapist evaluated the volunteers for the presence of exclusion criteria. [Table 1](#table1){ref-type="table"} shows the characteristics of the sample.
The study complied with the principles laid out in the Declaration of Helsinki. The ethics committee of the Faculty of Health Sciences at the University of Malaga, Spain, approved the study.
######
Characteristics of sample (N=81 jumps).
Characteristic Mean (SD)
--------------------------- ---------------
Age (years) 24.30 (3.90)
Height (cm) 173.59 (9.74)
Weight (kg) 72.58 (13.01)
Body mass index (kg/m^2^) 23.95 (2.96)
Data Collection and Procedures
------------------------------
### Overview
Study subjects performed three trials of the jump tests described by Bosco \[[@ref22]\]: SJ and CMJ (with arm swing modality). The SJ is a maximum vertical jump starting from the position of leg flexion 90°, with no rebound or counter movement and with hands on hips from the beginning to the end of the jump. The CMJ is performed starting from a standing position, then a quick movement of flexion and extension of the knees, and an immediate maximum vertical jump \[[@ref22]\]. Before the start of the vertical jump tests, participants performed a warm-up on a cycle ergometer for 10 minutes. After the warm-up period, each subject was instructed in the proper way to perform each test. Before starting the test, a trial test was performed to verify that the participant had understood the instructions. Between every jump a rest period of 1 minute was set.
### Anthropometry
Anthropometric data were obtained following the guidelines of The International Society for the Advancement of Kinanthropometry (ISAK) \[[@ref23]\]. The weight was recorded with the subject barefoot and in underwear. The height is the distance from the vertex to the soles of the feet. It is measured with the subject standing in anatomical position and the occipital region, back, gluteal region and heels in contact with the height rod. The subject takes a deep breath at the time of measurement. The body mass index (BMI) was calculated by dividing weight in kilograms (kg) by height in meters squared (m^2^).
### Kinematics Variables
Linear acceleration was measured along three orthogonal axes using the iPhone 4S inertial sensor, which incorporates a three-axis gyroscope, accelerometer and magnetometer. The mobile phone was attached to a belt and fixed at L5-S1 level. Data were obtained for analysis through SensorLog, available as an Apple iPhone app. The recording rate was set at 30 milliseconds. The recordings were stored in the internal memory of the mobile phone and were then sent via email for off-line processing. A previous study \[[@ref24]\] showed that the mobile phone (iPhone) accelerometer was accurate and precise compared to a gold standard, with an intra-class correlation coefficient (*r* ^2^\>0.98). The mobile phone accelerometer showed excellent sequential increases with increased walking velocity and energy expenditure (*r* ^2^\>0.9). An accelerometer embedded into a mobile phone was accurate and reliable in measuring and quantifying physical activity in the laboratory setting \[[@ref24]\].
### Jump Measuring
The height and flight time were evaluated through the contact mat Globus Ergojump Thesys in CMJ with arm swing and SJ tests. The Globus Ergojump contact mat was validated in a prior study \[[@ref12]\].
The CMJ test is performed with the subject starting from a standing position on the contact mat. A quick flexion and extension of the knee joint with minimal stops between eccentric and concentric phases is performed. The participant can swing his arms to propel himself. The legs should be kept in extension from take-off to landing. In a specific study to determine the reliability of different countermovement tests, an intraclass correlation coefficient (ICC) of 0.88 for the CMJ was determined \[[@ref25]\].
For the SJ test, the participant is placed from the vertical position with hands on hips and with knees in flexion position of 90 degrees. Following the indication of the examiner, the subject performs a boost without any countermovement trying to achieve the maximum height in a vertical jump keeping the lower limbs in extension after take-off to landing \[[@ref12]\].
The contact mat records flight time in seconds and the height reached in centimeters. Three repetitions of each test were conducted with more than one-minute rest between each.
### Maximum Isotonic Strength of the Knee Extensors
Isotonic muscle strength of the knee extensors was evaluated by bilateral dynamometry through the digital manual dynamometer POWERTRACK (JtechMedical). This tool incorporates a load cell affixed to the distal end of the leg of the subject. The dynamometer has a digital display that shows the force applied to the load cell in Newtons and records the peak in each attempt. The validity of this dynamometer has been demonstrated, with an ICCs ranging from 0.72 to 0.85 \[[@ref26]\].
The participant is placed in a sitting position on a stretcher, his hands resting on his legs and feet hanging off the ground. The examiner places one hand to stabilize the subject's leg and the other hand to support the load cell on the subject's distal third tibia. Starting from 90° knee flexion, the subject performs a knee extension resisted by the examiner with the load cell. A full extension must be avoided, with the knee flexion reaching 5°. The maximum peak force is recorded in the digital dynamometer. The test was performed three times for each subject, with a 2-minute break between tests; the highest value was taken.
Data Processing
---------------
An off-line analysis was guided to obtain kinematic information from the accelerometer for each subject, in each trial, in the SJ and CMJ test. In this study, the mean and standard deviation was obtained from the maximum peak and minimum peak of accelerations in the three axes of movements (*x*, *y* and *z*). Furthermore, the mean and standard deviation from maximum peak and minimum peak from the resultant vector (RV) accelerations \[RV = √ (*x* ^2^ + *y* ^2^ + *z* ^2^)\] was obtained.
Statistical Analysis
--------------------
To analyze the results, a database was created from the information gathered from the participants, the inertial sensor variables, the jump test variables and maximum isotonic strength of the knee extensors variables. The Kolmogorov-Smirnov test was used as determined by the variables normality of distribution. Descriptive statistics were performed with measures of central tendency and dispersion of the variables studied. Analysis was performed with SPSS Version 20.0 (SPSS Inc, Chicago, IL, USA).
Results
=======
The Kolmogorov-Smirnov demonstrated that the distribution of the sample by gender was non-normal. [Table 2](#table2){ref-type="table"} summarizes the acceleration-based measures, the jump test measures and maximum isotonic strength of the knee extensors measures in the SJ and CMJ jump test.
######
Acceleration-based, jump test values and maximum isotonic strength in the SJ and CMJ (N=81 jumps).
-----------------------------------------------------------------
\ \ Mean(SD)
----------------------- ----------------------- -----------------
**Accelerometer SJ** \
\ Max acc X (m/s^2^) 0.585 (.449)
\ Min acc X (m/s^2^) -0.549 (.436)
\ Max acc Y (m/s^2^) 1.851 (.629)
\ Min acc Y (m/s^2^) -0.087 (.239)
\ Max acc Z (m/s^2^) 0.844 (.493)
\ Min acc Z (m/s^2^) -1.211 (.567)
\ Max acc RV (m/s^2^) 2.202 (.679)
\ Min acc RV (m/s^2^) 0.005 (.003)
**Accelerometer CMJ** \
\ Max acc X (m/s^2^) 0.819 (.5385)
\ Min acc X (m/s^2^) -0.792 (.501)
\ Max acc Y (m/s^2^) 2.036 (.645)
\ Min acc Y (m/s^2^) -1.158 (.076)
\ Max acc Z (m/s^2^) 1.046 (.549)
\ Min acc Z (m/s^2^) -1.489 (.446)
\ Max acc RV (m/s^2^) 2.472 (.631)
\ Min acc RV (m/s^2^) 0.004 (.002)
**Jump Test Mat** \
\ Jump height SJ (m) 0.223 (.076)
\ Jump time SJ (s) 0.419 (.077)
\ Jump height CMJ (m) 0.329 (.099)
\ Jump time CMJ (s) 0.511 (.088)
**Dynamometry** \
\ Right dynamometry (N) 251.92 (53.029)
\ Left dynamometry (N) 234.96 (45.846)
-----------------------------------------------------------------
Discussion
==========
Principal Results
-----------------
In the present study the kinematic variables derived from acceleration through the inertial sensor of an iPhone 4S, dynamometry of lower limbs with a handheld dynamometer, and the height and flight time with a contact mat were described in vertical jump tests from a cohort of young healthy subjects, aged between 18 and 35 years. The development of the execution in SJ and CMJ vertical jump tests is described, examined and identified under acceleration variables obtained with the mobile phone.
Comparison With Prior Work
--------------------------
To the best of our knowledge, and according to the literature reviewed so far, the present study is the first to describe and analyze the kinematic variables in vertical jump tests with the use of a mobile phone as the main instrument.
Previous studies have been found in literature that may be relevant to the present study, in which vertical jump tests were evaluated through inertial sensors \[[@ref15]-[@ref19],[@ref27],[@ref28]\]. However, none of them has instrumentalized jump tests with a mobile phone's inertial sensor. Furthermore, most of these studies have focused on searching algorithms to identify jump height \[[@ref15]\], and developing validations of accelerometers as vertical jump evaluators \[[@ref16]-[@ref19],[@ref27]\], rather than on the kinematic description of the jump tests through accelerometry variables. Only one of these studies \[[@ref28]\] describes the accelerometic characteristics and analyses the acceleration curves during an SJ. However, acceleration features of SJ were not compared with any other type of jump test. On the other hand, the instrumentalization used was a uniaxial accelerometer and not a mobile phone's triaxial accelerometer. The mobile phone's triaxial accelerometer offers the advantage of getting variables of the three axes of motion (*x*, *y* and *z*), while the uniaxial accelerometer gives us variables of only one axis of movement. In addition, today, most mobile phones contain triaxial accelerometers, which are very accessible to most people, and the acquisition of another tool to analyze human motion through acceleration values would not be necessary using a mobile phone.
From accelerometry data obtained through the mobile phone, differences in values of minimum and maximum acceleration were identified between the two types of jump tests SJ and CMJ (see [Table 2](#table2){ref-type="table"}). Higher values of acceleration in the different axes of movement (*x*, *y* and *z*) in CMJ with respect to the SJ, may be explained by the countermovement produced in CMJ test causes greater movement synergies involving higher acceleration \[[@ref29]\]. It can also be observed in the jump variables measured by the contact mat; higher mean values of jump height and jump time in the CMJ regarding SJ are shown (see [Table 2](#table2){ref-type="table"}).
Specifically, the higher maximum acceleration values in the y-axis found in the CMJ are explained by the jump modality, as the countermovement allowed in it facilitates greater vertical impulse \[[@ref29]\]. Furthermore, the fact that greater height is reached more easily at CMJ causes greater deceleration after the transition between the maximum peak height and landing; this results in higher minimum acceleration values along the y-axis with a remarkable difference between CMJ and SJ tests---the acceleration difference is the greatest between the two jumps. These perceived differences are consistent with the higher heights and flight times recorded by the contact mat from CMJ test and the known differences between CMJ and SJ tests \[[@ref29]\]. Moreover, when analyzing the graphs of acceleration of SJ performed in this study, it has been observed that in the development of this type of jump, a negative acceleration on the y-axis (see [Figure 1](#figure1){ref-type="fig"} and [2](#figure2){ref-type="fig"}) is a common occurrence prior to the start of the upward positive acceleration curve that corresponds to the beginning of the upward thrust. Because the SJ technique must be performed without any countermovement, this could be explained by the possibility that the mobile phone\'s accelerometer detects small accelerations produced by short, quick countermovements made by the subject, and examiners who determine whether the execution of the jump has been correct, would not be able to detect them with their own eyes.
For acceleration in the z-axis, the differences found may also be explained by the influence of the transition between the eccentric and concentric phases occurring before take-off for CMJ because trunk flexo-extension movements occur, involving higher acceleration values in the z-axis \[[@ref29]\].
Regarding the differences in the x-axis acceleration, it is conceivable that lateral displacements by the subject in the execution of both tests can be greater for the CMJ test, also because movement synergies in this jumping technique are implied, and higher flight times are achieved \[[@ref29]\].
The differences between tests are also maintained for the resultant vector acceleration values, although less so. This can be justified by the integration of the acceleration signals in the different axes of motion to calculate the resulting vector.
{#figure1}
{#figure2}
Limitations and Future Work
---------------------------
As a limitation in this study, we can mention the fact that there is no separation by gender analysis that would show differences between men and women in performing the jumps \[[@ref30]\]. Therefore, future studies with a larger sample to allow a normal distribution of variables, adjusted by gender analysis, could be performed.
As a fortuitous finding, upon visual inspection of the y-axis acceleration graphic in the SJ tests, we observed a previous negative acceleration at the start of the upward curve of positive acceleration. This would be interesting research for future studies to analyze curves and globally analyze the mobile phone discriminating power between correct executions of vertical jump tests without countermovement based on trunk accelerometry.
Conclusions
-----------
According to the results obtained in this study, we can conclude that the built-in iPhone 4S inertial sensor is able to measure acceleration variables for vertical jump tests SJ and CMJ in healthy young adults. The acceleration kinematics variables derivate from the mobile phone's inertial sensor are higher in the CMJ test than the SJ test.
Conflicts of Interest: None declared.
BMI
: body mass index
CMJ
: countermovement jump
ICC
: intraclass correlation coefficient
ISAK
: The International Society for the Advancement of Kinanthropometry
RV
: resultant vector
SJ
: squat jump
| {
"pile_set_name": "PubMed Central"
} |
Background
==========
Maternal mortality remains unacceptably high across much of the developing world. An estimated 358, 000 maternal deaths occurred worldwide in 2008. Sub-Saharan Africa (SSA) and South Asia accounted for 87% of the global maternal deaths \[[@B1]\]. In sub-Saharan Africa, a woman's risk of dying from treatable or preventable complications of pregnancy and childbirth over the course of her lifetime is 1 in 22, compared to 1 in 7,300 in the developed regions \[[@B2]\]. In Ethiopia, maternal mortality and morbidity levels are among the highest in the world. The Maternal Mortality Ratio (MMR) in 2005 was 673 per 100,000 live births \[[@B3]\].
Around 80% of maternal deaths worldwide are brought about by such direct causes as hemorrhage, infection, obstructed labor, unsafe abortion and high blood pressure. Severe bleeding which usually occurs after the mother has given birth is the single most feared complication claiming the lives of most mothers \[[@B4]\].
Globally, the annual percentage decline in MMR between 1990 and 2008 was only 2.3%. Most SSA countries are not on track to meet the Millennium Development Goal (MDG) targets pertaining to MMR because recent estimates suggest that the average annual rate of reduction in MMR for these countries is 1.7% \[[@B1]\].
Three-quarters of maternal deaths occur during childbirth and the post-partum period. Most of the maternal deaths will be avoidable if women have access to vital health care before, during, and after childbirth \[[@B2]\]. One of the indicators of meeting MDG5 is the proportion of women who deliver with the assistance of skilled birth attendants. In almost all countries, where health professionals attend more than 80% of deliveries, maternal mortality ratios are below 200 per 100,000 live births \[[@B5]\]. A skilled birth attendant at delivery is critical to reducing maternal deaths. In 2006, nearly 61% of the births in the developing world were assisted by skilled birth attendants. However, coverage remains low in Southern Asia (40%) and SSA (47%) -- the two regions with the greatest number of maternal deaths \[[@B2]\].
The proportion of births attended by skilled attendants in Ethiopia is very much lower than that of countries in SSA. Even for women who have access to the services, the proportion of births occurring at health facilities is very low. Nationally, only 6% of the births took place at health institutions according to the 2005 DHS (Demography and Health Survey), and there was no significant difference in proportion of institutional delivery service utilization between DHS 2000 and 2005. In Oromia Region, institutional delivery service utilization which is about 4.8%, is lower than that of the national level \[[@B3]\]. Therefore, assessing the factors affecting institutional delivery service utilization in the study area is very important to improve maternity services and thereby reduce maternal and infant deaths.
Methods
=======
A community-based cross-sectional study was conducted in Munesa Woreda from April 1--20, 2011. The Woreda is found in Arsi Zone, Oromia Regional State. Kersa, the capital of the Woreda is located 232km from Addis Ababa, the capital of Ethiopia and 57 km from the capital of Arsi Zone, Asella. Organized into 32 rural and 3 urban kebeles, the woreda has 2 health centers, 4 medium clinics, and 32 health posts. According to the 2007 census, it has a total population of 166,414 \[[@B6]\].
The participants of the study were mothers who delivered 12 months before the study began. Women who delivered in those months, especially those who reported to have delivered after 28 weeks of gestation, were included in the study regardless of the outcomes of the births. The stratified cluster sampling technique was used to select the study units. That is, by taking kebeles as clusters, one of the three urban and eight of the 32 rural kebeles were selected by simple random sampling technique, making it possible to accommodate all eligible mothers in the clusters.
The sample size was determined by using a single population proportion formula which took the following assumptions in to consideration: magnitude of institutional delivery service utilization 7%, \[[@B7]\] (**p** = 0.07); 5% level of significance (α = 0.05); 2.5% marginal error (**d** = 0.025). The final sample size was adjusted by using the design effect of 2 and 5% non-response rate. Thus, the sample size turned out to be 855.
Data was collected through face to face interviews using a structured and a pre-tested questionnaire. Ten health science students were collected data, and, the house to house survey was carried out to get eligible mothers. Three nurses from Kersa Health Center were assigned to supervise the data collection process after they were trained, together with the data collectors.
Data entry was done by using EPI Info 2002 and exported to SPSS version 16.0 software package for analysis. The data were analyzed using logistic regression to determine the effect of various factors on the outcome variable and to control confounding effects. The results were presented in the form of tables, figures and texts using frequencies and summary statistics such as mean, standard deviation, and percentage to describe the study population in relation to relevant variables(age, residence, ethnicity, religion, marital status, educational status, occupational status, distance from health center, parity, family size, ANC, and place of delivery). Valid categories were employed for the variables used in the context from EDHS and other published literature. The strength of association between independent and dependent variables was assessed using the odds ratio with 95% confidence interval.
The expression "skilled attendant" in this study means a delivery attended by skilled health providers in the community or at a health institution. But in Ethiopia, all deliveries which occur at health institutions are attended by skilled birth attendants while such skilled care is rare in the community.
Ethical clearance was obtained from the School of Public Health, the University of Gondar. A formal letter request of cooperation was written to Arsi Zone Health Department and Munesa Woreda Health Office. Verbal consent was obtained from each study participant.
Results
=======
Socio-demographic characteristics
---------------------------------
A total of 855 mothers who gave birth in the last 12 months were interviewed; out of these 739 (86.4%) were rural dwellers. The mean age of the respondents was 27.22 years with a standard deviation (SD) of 5.87 years. The majority (96.7%) of the mothers were married. Five hundred twenty seven (61.6%) of the mothers and 324 (37.9) of their husbands were unable to read and write. Seven hundred twenty-three (84.6%) of the female respondents were housewives, while 726 (84.9%) of the husbands were farmers.
Six hundred twenty nine (73.6%) of the respondents had either radio or Tv set in their houses. To reach the nearest health center, 719 (84.1%) of the participants had to walk for 30 minutes on average (Table [1](#T1){ref-type="table"}).
######
Socio-demographic characteristics of the study participants, Munesa Woreda, Arsi Zone, southeast Ethiopia, April 2011
**Variables** **Frequency (%)**
---------------------------------------- -------------------
**Age of the mothers at interview**
15-19 50 (5.9)
20-24 224 (26.2)
25-29 268 (31.3)
30-34 178 (20.8)
35+ 135 (15.8)
**Marital status**
Married 838 (98)
Divorced/separated 17 (2.0)
**Distance from health facility**
≤ 30 minute 136 (15.9)
\>30 minute 719 (84.1)
**Religion**
Orthodox 442 (51.7)
Muslim 371 (43.4)
Protestant 42 (4.9)
**Ethnicity**
Oromo 834 (97.5)
Amhara 17 (2.0)
Gurage 4 (0.5)
**Educational status of the mother**
Unable to read and write 527 (61.6)
Primary education 306 (35.8)
Secondary and post secondary 22 (2.6)
**Occupational status of the mother**
House wife 746 (87.3)
Government employee 12 (1.4)
Farmer 97 (11.3)
**Occupational status of the husband**
Farmer 726 (84.9)
Government employee 30 (3.5)
Daily laborer/Merchants 99 (11.6)
**Educational status of the husband**
Unable to read and write 324 (37.9)
Primary education 487 (57.0)
Secondary and post secondary 44 (5.1)
**Perceived economical status**
Low 220 (25.7)
Medium 422 (49.4)
High 213 (24.9)
**Family size**
≤5 412 (48.2%)
\>5 443 (51.8)
Obstetric characteristics
-------------------------
Six hundred seventeen (72.2%) of the mothers had their first pregnancy before the age of 20 years, and the minimum and maximum ages at first pregnancy were 14 and 37 years with a mean and SD of 18.51 and 2.31 years, respectively. During the last pregnancy, 297 (34.7%) of the mothers had at least one ANC (Antenatal Care) visit. Out of the total respondents, only 105 (12.3%) gave birth at health facilities (hospitals and health centers), and the vast majority (87.7%) delivered at home. Among the mothers who delivered at home, 392 (52.2%) were assisted by their families or relatives; 23 (3.1%) delivered without any assistance. Of those who went to health facilities, 66 (62.9%) delivered at health centers, 27 (25.6%) at hospital, and the rest (11.5%) at private clinics (Table [2](#T2){ref-type="table"}). Mothers gave a variety of reasons for delivering at home. For example, 450 (52.6%) said that delivering at home was the norm or the usual practice (Figure [1](#F1){ref-type="fig"}).
######
Obstetric characteristics of respondents, Munesa woreda, Arsi Zone, southeast Ethiopia, April 2011
**Variables** **Frequency** **Percent**
------------------------------------- --------------- -------------
**Age at first pregnancy(years)**
\<20 617 72.2
≥20 238 27.8
**Parity**
One 139 16.3
Two-four 390 45.6
Five and above 326 38.1
**Ever had abortion**
Yes 62 7.3
No 793 92.7
**Ever had still birth**
Yes 77 9.0
No 778 91.0
**ANC visit during last pregnancy**
Yes 297 34.7
No 558 65.3
**Number of ANC visits**
One 12 4.0
Two to Three 123 41.4
Four and above 162 54.6
**Place of last 12 month delivery**
Home 750 87.7
Health facility 105 12.3
**Assistance during home delivery**
No one 23 3.1
Trained TBAs 47 6.3
Untrained TBAs 288 38.4
Family or relatives 392 52.2
{#F1}
Factors associated with health institution delivery service utilization
-----------------------------------------------------------------------
As shown in the bivariate models, institutional service delivery utilization was significantly associated with the age, residence, occupational and educational status of the mothers, and the occupational and educational status of the husbands as well as with distance from the nearby health centers, family size, parity, and ANC visit during the last pregnancy.
On the basis the variables found to be significant in the bivariate analysis, maternal age at interview, residence, educational status of couples, ANC visit during the last pregnancy and parity were significantly associated with institutional delivery service utilization in multiple logistic regression analysis, too.
Mothers less than 20 years of age during the interview were about 6 times (AOR = 6.06, 95%CI: 1.54, 23.78) more likely to deliver at health institutions than mothers more than 35 and above. Urban mothers were about 2.3 times more likely to deliver at health institutions than rural mothers (AOR = 2.27, 95%CI: 1.17, 4.40). Mothers with secondary education and above were 4.3 times more likely to deliver at health facilities as compared to those who were not able to read and write (AOR = 4.31, 95%CI:1.62,11.46).
Regarding the educational status of husbands, mothers whose husband attended secondary school and above were 2.8 times (AOR = 2.77, 95%CI:1.07, 7.19) more likely to deliver at health institutions as compared to mothers whose husbands were unable to read and write.
ANC visit during the last pregnancy was also found to be a strong predictor of institutional delivery service utilization. Mothers who visited health facilities for ANC during pregnancy were 4.2 times (AOR = 4.18, 95%CI: 2.54, 6.89) more likely to deliver at health institutions than those who did not visit ANC during the last pregnancy. Parity was also another important factor which was associated with the place of delivery. Mothers who were delivering their first babies were 2.4 times more likely to utilize health institutions as compared to those who had five and more deliveries (AOR = 2.41, 95%CI: 1.17, 4.97) (Table [3](#T3){ref-type="table"}).
######
Bivariate and multivariate analysis of factors associated with institutional delivery service utilization in Munesa woreda, April 2011
**Variables** **Institutional delivery** **COR (95%CI)** **AOR (95%CI)**
----------------------------------------- ---------------------------- ----------------- ------------------- -------------------
**Place of residence**
Urban 26 90 2.41 (1.47,3.96 2.27 (1.17,4.40)
Rural 79 660 1 1
**Distance from health institution** **\***
≤30minute 27 109 2.04 (1.26,3.30)
\>30minute 78 641 1
**Age of the mother**
\<20 15 35 14.04 (4.38,44.9) 6.06 (1.54,23.78)
20-34 86 584 4.80 (1.74,13.39) 4.02 (1.34,12.06)
35+ 4 131 1 1
**Possession of Radio or TV** \*
Yes 79 550 1.10 (0.69,1.77)
No 26 200 1
**Occupational status of the mother** **\***
House wife 88 658 1
Government employee 4 8 3.74 (1.10,12.67)
Farmer 13 84 1.16 (0.62,2.16)
**Family size** **\***
≤5 72 340 2.63 (1.70,4.07)
\>5 33 410 1
**Educational status of the mothers**
Unable to read and write 50 477 1 1
Primary education 47 259 1.73 (1.13,2.65) 1.44 (0.92,2.24)
Secondary and post secondary 8 14 5.45 (2.18,13.63) 4.31 (1.62,11.46)
**Occupational status of the husbands** \*
Farmer 74 652 1
Government employee 7 23 2.68 (1.11,6.64)
Daily laborer/Merchants 24 75 2.82 (1.68,4.74)
**Educational status of the husbands**
Unable to read and write 30 294 1 1
Primary education 59 428 1.35 (0.85,2.15) 0.61 (0.36,1.03)
Secondary and post secondary 16 28 5.60 (2.73,11.50) 2.77 (1.07,7.19)
**Parity**
1 35 104 4.65 (2.61,8.29) 2.41 (1.17,4.79)
2-4 48 342 1.94 (1.14,3.29) 0.99 (0.55,1.80)
5+ 22 304 1 1
**ANC visit during last pregnancy**
Yes 74 223 5.64 (3.61,8.83) 4.18 (2.54,6.89)
No 31 527 1 1
\* Not significant in back ward stepwise logistic regression.
Discussion
==========
The results of the study revealed that the proportion of women who delivered at health facilities was 12.3% in the Woreda, and that the vast majority of mothers (87.7%) gave birth at home. This finding was low when compared with the national MDG reports of 2010 which showed that the percentage of deliveries attended by skilled health personnel was 25 \[[@B8]\]. The current utilization was higher than the national and the Oromia Region EDHS result of 2005 which were 6% and 4.8%, respectively \[[@B3]\]. This might be due to the time gap i.e. since 2005 there might have been improvements in accessing and utilizing of the service. The result was in line with that of a study done in North Gondar Zone in 2002, which was 13.5% \[[@B9]\]. However, it was lower than those of studies conducted in Enugu, Nigeria and southern Tanzania where the proportion of women who gave birth at health facilities was 47.1% and 46.7%, respectively. The difference could be explained by the fact that mothers in these countries had better educational status and better ANC service utilization \[[@B10],[@B11]\].
Residence of the respondents was significantly associated with institutional delivery service utilization. Mothers who lived in urban kebeles were about 2.3 times more likely to deliver in health institutions as compared to those who lived in rural kebeles. The finding was consistent with EDHS 2000 and that of a study done in North Gondar Zone \[[@B9],[@B12]\]. The study in Nigeria also showed that urban/rural differences had significant associations with the place of delivery \[[@B10]\]. This might be because in urban areas, the proportion of educated mothers was higher; they get maternal and other health services nearby and had better access to information than rural mothers.
Maternal age was also one of the predictors of institutional delivery service utilization. Mothers with less than 20 years of age at the time of the interview were about 6 times more likely to give birth in health institutions than mothers aged 35 years and above. This finding was in line with studies done in North Gondar Zone, Kenya, and Afghanistan which showed that younger women were more likely to utilize institutional delivery service as compared to older ones \[[@B9],[@B13],[@B14]\]. The possible explanation for this could be that younger women were more likely to be more literate than older women, and that older women tended to consider giving birth at home not so risky as it has been their usual experience. Furthermore, older women might belong to a more traditional cohort and thus be less likely to use modern facilities as compared to younger women.
The results of this study showed that, institutional delivery service was significantly influenced by the level of education. Women with higher level of education (secondary and above) were about 4.3 times more likely to deliver at health facilities than those who were unable to read and write. This finding is similar with those of studies conducted in different parts of Ethiopia \[[@B9],[@B12]\], Bangladesh, Nigeria, and Afghanistan \[[@B10],[@B14]-[@B16]\]. These might be due to the fact that educated women had better awareness about the benefits of preventive health care and health services. They might also have higher receptivity to new health-related information. Familiarity with modern medical culture, egalitarian relationship and better communication with husbands, more decision-making power, increased self-worth and self-confidence were also the characteristics of urban women which might have contributions to better utilization of health facility delivery than rural women.
Husbands' educational level was also one of the factors that predicted health institution delivery. Women whose husbands had secondary and post-secondary education were about 2.8 times more likely to deliver in health facilities as compared to those whose husbands were unable to read and write. Our finding was in line with that of a study done in Enugu-northeastern Nigeria and Zaria, northern Nigeria \[[@B10],[@B16]\]. The possible explanation for this might be that educated husbands might be more open toward modern medicine and aware of the benefits of giving birth at health facilities. They might also put fewer constraints on their wives\' mobility and decision-making, thus facilitating care-seeking.
ANC services can provide opportunities for health workers to promote a specific place of delivery or give women information on the status of their pregnancy which in turn alerts them to decide where to deliver. This study showed that mothers who had ANC visits during the last pregnancy were about four times more likely to deliver at health facilities compared to those who did not have any visits. The result was consistent with other studies done in North Gondar Zone, India, and Mali, which revealed that mothers who attended ANC follow up were more likely to deliver at health facilities than those who did not \[[@B9],[@B17],[@B18]\].
As a cross-sectional study requires respondents to remember information retrospectively, recall and interviewer bias are the potential limitations of this study. However, numerous scientific procedures have been employed to minimize the possible effects. To reduce the recall bias, for instance, only women who gave birth in the last one year were selected. In addition, procedures such as supervision, pretest of data collection tool, and adequate training of data collectors and supervisors were utilized. Wide confidence intervals observed in some of the variables are due to the calculation of the intervals by the sample size calculated for the prevalence.
Conclusions
===========
This study revealed that the proportion of women who gave birth at health facilities in the woreda was low. Urban residence, age below 20 years at the time of interview, parity, ANC visit during the last pregnancy, secondary and post-secondary levels of education of mothers and husbands were factors significantly associated with institutional delivery service utilization. Increasing the awareness of mothers and their partners about the benefits of institutional delivery services are recommended.
Competing interests
===================
The authors declare that they have no competing interests.
Authors' contributions
======================
AA wrote the proposal, participated in data collection, analyzed the data and drafted the paper. AG and ZB approved the proposal with some revisions, participated in data analysis and revised subsequent drafts of the paper. All authors read and approved the final manuscript.
Pre-publication history
=======================
The pre-publication history for this paper can be accessed here:
<http://www.biomedcentral.com/1471-2393/12/105/prepub>
Acknowledgements
================
We are very grateful to the University of Gondar for the approval of the ethical clearance and their technical and financial support. We would also like to thank all mothers who participated in this study for their commitment in responding to our interviews.
| {
"pile_set_name": "PubMed Central"
} |
Introduction {#Sec1_115}
============
Heavy metal pollutions are particularly hazardous contaminants in food and the environment. In general, they are not biodegradable and have long biological half-lives. According to the World Health Organization (World Health Organization [@CR21_115]) heavy metals must be controlled in food sources in order to assure public safety. Excessive concentration of food heavy metals is associated with the etiology of a number of diseases, especially cardiovascular, renal, neurological, and bone diseases (Chailapakul et al. [@CR6_115]). A major reason to monitor levels of toxic metals in foods follows from the fact that contamination of the general environment has increased.
It is known that some shrimp and crab may provide useful means of monitoring such elemental concentration levels and their impact on the aquatic environment. In [@CR5_115]BU-Olayan and Thomas showed higher trace metal levels in benthic molluscs and annelids of Kuwait Bay in the Persian Gulf compared to other regions of the word. Al-Mohanna and Subrahmanyam ([@CR1_115]) demonstrated Zn and Cu pollution swimmer crabs and attributed this to the 1991 Gulf War oil spill into Kuwait's marine environment. Alyahya et al. ([@CR2_115]) reported that concentrations of Cd, Pb, Cu, and Zn in fresh parts of the clam (M. meretrix) in the Persian Gulf near the Saudi Arabia coast line were within the acceptable standard range. To the extent of the author\'s knowledge, few studies reported heavy metals pollution of shrimp and fish in the Persian Gulf water of Iran (Pourang et al. [@CR15_115]; Raissy et al. [@CR16_115]).
Detection methods of stable elements included colorimetry, spectrography, mass spectroscopy, atomic absorption, spectrophotometry, and neutron activation analysis (NAA). Of these methods NAA has advantages over alternative methods. Detection sensitivity are greatest (\<0.01 ppm) with neutron activation analysis (NAA) (Corliss [@CR7_115]) and NAA also allows for simultaneous detection of many other metals in a sample. There is no need to sample preparation (i.e. dry ashing or wet ashing) before the analysis and intact samples can be used for analysis. Instrumental neutron activation analysis (INAA) has been successfully used to investigate heavy metals of many biological materials such as food and freshwater fish (Cunningham and Stroube [@CR8_115]; Ndiokwere [@CR14_115]) and also used to determine the body composition of salmon (Talbot et al. [@CR20_115]).
This study is aimed in determining the concentrations of the heavy metal contamination (Mn, Fe, Zn, Co and As) in green tiger shrimp (*Penaeus semisulcatus*) and blue crab (*Portunus pelagicus*) by using instrumental neutron activation analysis (INAA).
Material and methods {#Sec2_115}
====================
Collection and preparation of samples {#Sec3_115}
-------------------------------------
Swimmer crab (Portunus pelagicus) and shrimp (Penaeus semisulcatus) were captured in depth 41--42 m - Deylam port (Figure [1](#Fig1){ref-type="fig"}) by using cast nets during spring of 2011. At this study used 10 same sample size, shrimp (10--12 g) and Crab (11--13 g). Immediately, after sampling crab and shrimp were stored in a container, preserved in crushed ice and transferred to the laboratory and frozen -20°C until analyzed.Figure 1**Location map of sampling site, Deylam port, near Bushehr province area of Iran.**
Samples (whole body) were dried (60°C for 72 h) and ground through a 2 mm screen for subsequent INAA. The 500--600 mg of dried samples and standard (Fish Tissue, IAEA-407) were accurately weighed into polyethylene vials and heat sealed for irradiation.
Irradiation and counting {#Sec4_115}
------------------------
The irradiation of the samples and standard were carried out in the 40-tube specimen rack of the Tehran Research Reactor (Pool Type Reactor) at a neutron flux of 1.0 × 10^13^ n/cm/s. Each analytical portion was irradiated twice, once to analyze for elements yielding short-lived radioisotopes and a second time to analyze for elements yielding long-lived radioisotopes. Table [1](#Tab1){ref-type="table"} lists the radioisotopes, half-lives and gamma-ray lines used.Table 1**Radioisotopes, half-lives and gamma-ray energies used**IsotopeHalf-lifeGamma-ray energies (keV)^56^Mn155 min846.76, 1810.72^59^Fe44.5 d1099.25, 1291.60^60^Co5.27 y1173.2, 1332.50^76^As1.10 d559.10^65^Zn243.9 d1115.55^69m^Zn825.6 min438.63^71m^Zn236.4 min386.28
For short-lived radioisotopes, samples were irradiated for 10 seconds and counted 4 minutes later for 300 seconds and again 2 hours later for 1000 seconds. For long-lived radioisotopes, samples were irradiated for 1 hour (3600 s) and counted 3 days later for 1000 seconds and also 10 days later for 3000 seconds and 24 days later for 5000 seconds.
The samples and the standard were counted on the high pure germanium detector (resolution 1332.5 keV at 2.0 keV and efficiency of 20%).
Statistical analysis {#Sec5_115}
--------------------
The results were subjected to emulate and analysis using Maestro II and SPAN, respectively.
Animal Dead tissues handling was performed with regard to Iranian animal ethics society and local university rules.
Results and discussion {#Sec6_115}
======================
The Mn, Fe, Zn, Co and As levels were determined in the whole body of the swimmer crab and green tiger shrimp (Table [2](#Tab2){ref-type="table"}). There are remarkable differences in the swimmer crab and green tiger shrimp heavy metal concentrations. The order of the heavy metal concentrations in swimmer crab was found: Zn\>Fe\>As\>Mn\>Co, while in whole body of green tiger shrimp was Fe\> Zn\>Mn\>As\>Co.Table 2**Mn, Fe, Zn, Co and As concentrations (means ± s) in crab and shrimp samples (ppm/dry whole body)**MetalCrabShrimpMn1.91 ± 0.3325.43 ± 2.95As21.38 ± 3.318.28 ± 2.82Co0.15 ± 0.020.40 ± 0.05Fe62.87 ± 11.07288 ± 38.88Zn66.64 ± 7.6068.73 ± 7.84
The swimmer crab heavy metal concentrations were less than shrimp samples, with the exception of As. The As concentration were 18.07-24.69 and 5.46-11.10 ppm in whole body of crab and shrimp, respectively.
Pourang et al. ([@CR15_115]) examined green tiger shrimp heavy metal distribution in the northwestern (near the Bushehr Province coastline) part of the Persian Gulf. In this study highest mean of Zn concentration (43.39 ppm/fresh weight) was found in hepatopancreas. Also, the Zn levels of the exoskeleton and muscle were 8.56 and 8.98 ppm/wet weight, respectively. In the current study, the level of the Zn/fresh weight of whole body was 22.76 ppm.
Research on swimmer crab in turkey showed Zn, Mn and Fe levels of swimmer crab were 37.2-46.8, 1.30-1.9 and 4.5-6.8 ppm of dry body meat, respectively (Gökoğlu and Yerlikaya [@CR12_115]). Also, Gökoğlu and Yerlikaya ([@CR12_115]) showed the Fe level of the swimmer crab was higher than results found in the current study.
Sadiq et al. ([@CR17_115]) reported levels of Zn and Co are 165.73 and 4.66 ppm/dry crab whole body, respectively. These values are markedly higher than levels obtained in the current study (66.64 and 0.16 ppm, respectively).
Ayas and Özoğul ([@CR4_115]) showed the concentration of the Zn and Fe were 101.40-104.13 and 21.92-23.90 ppm for male, and 106.13-108.64 and 13.15-16.53 ppm for female swimmer crab, respectively.
The mean values of the heavy metals contents in edible tissue and whole body of crab (*Portunus pelagicus*) and shrimp (*Penaeus semisulcatus*) have been summarized at Table [3](#Tab3){ref-type="table"}. Data obtained from current study indicated that heavy metal contents in crab and shrimp are comparable to the other parts of Persian Gulf and world areas. According to available literature, there is no study on Mn concentration in shrimp tissue.Table 3**Comparison of mean concentrations of trace metals reported for shrimp (Penaeus semisulcatus) and crab (Portunus pelagicus) in the worldwide**LocationTissueMnFeZnAsCoReference**Crab**Persian Gulf, IranWhole body1.9162.8766.6421.380.15Present studyPersian Gulf, KuwaitMuscle0.95-206.00.31-Al-Mohanna and Subrahmanyam 2001Mersin bay, TurkeyMuscle-18.93104.82\--Ayas and Özoğul 2011Persian Gulf, Saudi Arabia\*Whole body\--165.73-4.66Sadiq et al. 1982**Shrimp**Persian Gulf, IranWhole body25.43288.068.738.280.40Present studyIskenderun bay, Turkey ^1^Muscle-18.6927.75\--Firat et al. 2008Persian Gulf, Iran\*Muscle\--41.76\--Pourang et al. 2005Persian Gulf, Saudi Arabia\*Whole body\--148.89-4.56Sadiq et al. 1982Values are expressed in ppm per dry weight either of edible tissue or whole body.1 Mean of males and females.2 Mean of four stations.\* The values calculated based on dry matter content of these species from the metal concentration per wet weight of shrimp and crab.
Although, in the current study was measured Zn and Mn concentration 56.39-221.68 and 0.14-2.01 ppm, respectively.
Conversely, in study by Al-Mohanna and Subrahmanyam ([@CR1_115]), determined maximum values for As was about 36 times higher than the value was reported.
In most studies the heavy metal concentration are reported either in various body parts of crustaceans or in their edible tissue. However in current study whole body samples were used. Firat et al. ([@CR11_115]) reported higher concentration of heavy metals in hepatopancreas compared to the gill and muscle of shrimp, similar results were reported by Pourang et al. ([@CR15_115]). Hence the heavy metal concentrations of muscle in shrimp and crab samples in the current study might be lower than the reported values.
Sadiq et al. ([@CR17_115]) reported Zn and Co concentrations in shrimp as 148 ppm and 4.56 ppm, respectively. The average Co content was determined as 0.15 ppm and 0.40 ppm in the swimmer crab and shrimp, respectively.
Level of As in the whole body of swimmer crab in the current study was higher than previous studies that reported As concentration in lobster, bivalves, crabs and fishes of Persian gulf (Mora et al. [@CR9_115]; Al-Mohanna and Subrahmanyam [@CR1_115]; Raissy et al. [@CR16_115]). Levels of As contamination in the Persian Gulf may have affected these values. Major anthropogenic sources of arsenic include mining and smelting operations, emissions from coal burning electrical generating facilities, leaching from hazardous waste facilities and from insecticide, herbicide or algicide applications (Anonymous [@CR3_115]).
Among these sources, herbicide and algicide may have major importance. In recent years local use of algicide against algae bloom has been increased. However, the majority of the As in shrimp and crab appears to be in the form of the less toxic organic form, for example the predominant form is arsenobetaine (Shiomi et al. [@CR18_115]; Sloth et al. [@CR19_115]). For this reason, the determination of the total amount of the arsenic in a sample is not sufficient to assess the risk from eating seafood, and speciation analysis is necessary.
High levels of Mn in shrimp samples were noted when compared to crab samples (25.43 ppm vs 1.91 ppm). In a geochemistry study of sediment core of Persian gulf near Bushehr port not only Mn level among selected trace metals was highest, but also the heavy metals concentration were higher than mean crust and mean world sediments (Karbassi et al. [@CR13_115]; Fazaeli [@CR10_115]). With the exception of the As, levels of other heavy metals were higher in shrimp compared to crabs. In contrast to these results Sadiq et al. (Sadiq et al. [@CR17_115]) showed higher heavy metals accumulation in crabs compared to shrimp.
The results showed swimmer crab (*Portunus pelagicus*) and shrimp (*Penaeus semisulcatus*) have been contaminated with heavy metals. Shrimp showed a higher potential to accumulate metals in their body compared to crabs. Further studies are necessary to evaluate the effect of organs, sex, size, season and site of sampling on heavy metal concentration in crab and shrimp. More speciation analysis is also necessary to determination of Pb, Hg, total and organic forms of arsenic.
**Competing interests**
The authors report no conflicts of interest. The authors alone are responsible for the content and writing of the paper.
**Authors' contributions**
MH and MGM participated in the design of the study and manuscript preparation. MAS and FZ participated in data acquisition and statistical analysis. MB participated in manuscript revision. ZA participated in data acquisition and technical assistance. MH reviewed the manuscript and provided advice. All authors read and approved the final manuscript.
The authors are grateful for the financial support provided by Agricultural, Medical and Industrial Research School (AMIRS-NSTRI), Karaj, Iran.
| {
"pile_set_name": "PubMed Central"
} |
All relevant data are within the paper and its Supporting Information files.
Introduction {#sec001}
============
Phytopathogenic bacteria cause serious economic losses by reducing the yield of marketable quality crops \[[@pone.0181499.ref001]\]. *Ralstonia solanacearum*, which causes vascular wilt disease, is one of the most destructive pathogens \[[@pone.0181499.ref002]\]. This soil-born pathogen is found worldwide and has a large range of hosts, comprising more than 200 plant species, including tomato, potato, pepper, peanut, tobacco, and banana \[[@pone.0181499.ref002]--[@pone.0181499.ref004]\]. The direct yield losses caused by *R*. *solanacearum* differ widely according to the host, cultivar, climate, soil type, cropping pattern, and strain; the losses were up to 90% for tomato, potato, and banana \[[@pone.0181499.ref002]\].
The control of bacterial wilt has been considered difficult owing to the complex nature of the pathogen: endophytic growth patterns, survival in soil, transport in water, and a wide host range and biological diversity \[[@pone.0181499.ref002], [@pone.0181499.ref005]\]. The investigation into various methods for the control of bacterial wilt disease has spanned several decades. Yuliar \[[@pone.0181499.ref002]\] reported that studies of these methods conducted between 1984 and 2014 predominantly described the biological methods (54%), followed by cultural practices (21%), chemical methods (8%), and physical methods (6%). Plant disease control has been largely dependent on the use of chemical pesticides. Unfortunately, the use of chemical pesticides has been strongly limited owing to their association with environmental pollution, poisonous effects, and antibiotic resistance \[[@pone.0181499.ref002], [@pone.0181499.ref006], [@pone.0181499.ref007]\]. As the demand for environmentally acceptable pesticides is increasing, the development of safe and effective antimicrobial agents is essential for the treatment of plant disease.
Plants contain chemically diverse compounds that can be used directly as pesticides with reduced harmful effects \[[@pone.0181499.ref008]\]. Globally, the use of botanicals as alternatives to synthetic pesticides has increased annually. In particular, the growing demand for botanical pesticides has been much higher in developed or industrialized countries because of the increase in organic food production \[[@pone.0181499.ref009]\]. Botanicals have many advantages over synthetic chemicals, such as less or no residues on food because of rapid degradation, little or no harmful effect in humans or on the environment, and cost effectiveness \[[@pone.0181499.ref009]\]. However, botanical pesticides have some limitations, such as slow and lower efficacy compared with chemical pesticides and less efficacy when applied to fields; thus, the development of novel formulations with enhanced efficacy and longer shelf life is required \[[@pone.0181499.ref010]--[@pone.0181499.ref012]\]. Nowadays, nanotechnology has been applied in formulating, thus plant parts such as fruit, leaf, bark, seed, and stem extracts have been used for synthesis of nanoparticles, as effective formulations for the phytopathogens control \[[@pone.0181499.ref012]\]. There are many botanical pesticides such as plant extracts and their compounds possess antibacterial activities against phytopathogenic bacteria including *R*. *solanacearum* \[[@pone.0181499.ref012]\]. Most current studies that have focused on botanicals for the control of *R*. *solanacearum* utilized *in vitro* assays or potted plants \[[@pone.0181499.ref013]--[@pone.0181499.ref016]\]. To the best of our knowledge, there are no commercialized botanical products currently available for the control of bacterial wilt of tomato. Therefore, the search and development of highly active botanical pesticides for the control of tomato bacterial wilt are quite necessary.
In the search for antibacterial agents from Vietnamese plants, we found that the methanol extract of aerial parts of *Sapium baccatum* was highly active against *R*. *solanacearum*. *S*. *baccatum* is widespread across South Asia and has been used as a traditional medicine in Malaysia \[[@pone.0181499.ref017], [@pone.0181499.ref018]\]. Although several compounds, including bukittingine, lupeol, betulin, β-taraxerol, taraxerone, aleuritolic acid, 3-acetoxy-aleuritolic acid, 1-hexacosanol, β-sitosterol, stigmasterol, docosyl *trans*-isoferulate, and docosanoic acid 2\',3\'-dihydroxypropyl ester have been extracted from *S*. *baccatum* \[[@pone.0181499.ref017]--[@pone.0181499.ref019]\], limited information is available about the antimicrobial activities of these components or whether *S*. *baccatum* has additional active compounds. The aims of this study were: 1) to isolate and identify antibacterial compounds from *S*. *baccatum*; 2) to examine the antibacterial spectra of the isolated compounds against plant pathogenic bacteria; and 3) to evaluate the disease control efficacies of the methanol extract of *S*. *baccatum* on bacterial wilt of tomato under greenhouse conditions.
Materials and methods {#sec002}
=====================
Bacterial strains and culture conditions {#sec003}
----------------------------------------
The following strains of plant pathogenic bacteria were used for antibacterial activity assays: *Acidovorax avenae* subsp. *cattleyae* SL4351, the causal agent of bacterial brown spot in Phalaenopsis (*Phalaenopsis* sp.); *Agrobacterium tumefaciens* SL2434, the causal agent of crown gall in apple (*Malus domestica*); *Burkholderia glumae* SL4269, the causal agent of bacterial panicle blight in rice (*Oryza sativa*); *Clavibacter michiganensis* subsp. *michiganensis* SL4135, the causal agent of bacterial wilt and canker in tomato (*Solanum lycopersicum*); *Pectobacterium carotovorum* subsp. *carotovorum* SL290, the causal agent of bacterial soft rot in potato (*Solanum tuberosum*); *Pectobacterium chrysanthemi* SL3218, the causal agent of bacterial leaf rot in aloe (*Aloe vera*); *Pseudomonas syringae* pv. *actinidiae* CJW7, the causal agent of bacterial canker in kiwifruit (*Actinidia deliciosa*); *Pseudomonas syringae* pv. *lachrymans* SL308, the causal agent of cucumber angular leaf spot (*Cucumis sativus*); *R*. *solanacearum* SL1944, the causal agent of bacterial wilt in tomato (*S*. *lycopersicum*); and *Xanthomonas arboricola* pv. *pruni* SL4370, the causal agent of bacterial spot of peach (*Prunus persica*). All of these bacteria were isolated from the infected tissues by Dr SD Lee of the National Academy of Agricultural Sciences \[[@pone.0181499.ref004]\], except for *P*. *syringae* pv. *actinidiae* CJW7, which was isolated by Prof. YJ Koh of Sunchon National University \[[@pone.0181499.ref020]\] and *R*. *solanacearum*, which was isolated by Prof. SW Lee of Dong-A University \[[@pone.0181499.ref004], [@pone.0181499.ref021]\]. All strains were grown on tryptic soy agar (TSA; Becton, Dickinson and Co., Sparks, MD, USA) or tryptic soy broth (TSB). *P*. *syringae* pv. *actinidia*e and *X*. *arboricola* pv. *pruni* were grown at 25°C for 18--36 h and all other strains were cultured at 30°C for 18--36 h.
Plant material {#sec004}
--------------
The aerial parts of *S*. *baccatum* were collected by the Department of Phytochemistry, Vietnam Institute of Industrial Chemistry (Hanoi, Vietnam). Plant species were identified by Dr The Bach Tran from the Institute of Ecology and Biological Resources (Hanoi, Vietnam) and voucher specimens were deposited in the laboratory.
Extraction and isolation of antibacterial compounds {#sec005}
---------------------------------------------------
The dry powdered material of *S*. *baccatum* (200 g) was extracted twice with 90% methanol (2 × 3 L) for 48 h at room temperature. The extracts were filtered through Whatman No. 1 filter paper and the filtrates were concentrated by using a rotary evaporator under vacuum to yield a crude extract (24.5 g). A portion of the methanol extract (15 g) was suspended in 500 mL distilled water and then successively partitioned twice with *n*-hexane, ethyl acetate (EtOAc), and *n*-butanol (BuOH). Of the four layers, the EtOAc and BuOH layers showed strong antibacterial activity against *R*. *solanacearum* as determined by the broth microdilution method \[[@pone.0181499.ref004]\]. Therefore, further isolation of active compounds was conducted from these two layers.
The EtOAc layer (1.55 g) was successively eluted on a silica gel column (2.5 × 60.0 cm, Kiesel gel 60, 100 g, 230--400 mesh, E. Merck) with mixtures of dichloromethane (DCM)/methanol (MeOH) (90:10, v/v, 200 mL; 85:15, v/v, 200 mL; 80:20, v/v, 500 mL; 70:30, v/v, 500 mL), yielding five fractions, E1--E5. The fractions were monitored with thin-layer chromatography (TLC, Silica gel 60 F~254~, 0.25 mm layer thickness; E. Merck) with the developing solvent DCM/MeOH (85:15, v/v). Fraction E2 (233 mg) was separated on a Sephadex LH20 column (2.5 × 60.0 cm, 60 g, 70--100 μm, Sigma-Aldrich) via successive elution with mixtures of DCM/MeOH (9:1, v/v, 100 mL; 8:2, v/v, 200 mL; 7:3, v/v, 300 mL), to obtain seven fractions designated E21--E27. Fraction E25 (21 mg) containing compound **2** was successfully separated on a Sephadex LH20 column (1.0 × 30.0 cm, 10 g, 70--100 μm) by using DCM/MeOH (7:3, v/v, 100 mL). Compound **2** was further purified on a silica gel column (1.0 × 30.0 cm, Kiesel gel 60, 10 g, 230--400 mesh) and eluted with DCM/MeOH (9:1, v/v, 50 mL), yielding a pure compound (5 mg).
Fraction E3 (900 mg) was separated on a Sephadex column (3.0 × 60.0 cm, 100 g, 70--100 μm; Sigma-Aldrich) and successively eluted with DCM/MeOH (8:2, v/v, 200 mL; 7:3, v/v, 500 mL; 5:5, v/v, 500 mL; 3:7, v/v, 700 mL), which yielded 10 fractions designated E31--E310. The fractions were monitored by using reversed-phase TLC (Silica gel 60 RP-18 F~254~, 0.25-mm layer thickness; E. Merck) with the developing solvent MeOH/water (W) (4:6, v/v). To obtain compound **7** (4.3 mg), fraction E35 (40 mg) was separated with a LiChroprep RP-18 column (1.0 × 30.0 cm, 10 g, 40--63 μm; E. Merck) and eluted with MeOH/W (2:8, v/v; 3:7, v/v; 4:6, v/v, 50 mL of each).
Fraction E38 (120 mg) was first purified with a LiChroprep RP-18 column (1.5 × 30.0 cm, 15 g, 40--63 μm) eluted with a mixture of MeOH/W (2:8, v/v, 50 mL; 3:7, v/v, 100 mL; 4:6, v/v, 50 mL), further purified with the Sephadex LH20 column (1.0 × 30.0 cm, 10 g, 70--100 μm) and eluted with 100% MeOH to afford compound **1** (43 mg). Fraction E310 (250 mg) was also separated with the LiChroprep RP-18 column (1.5 × 30.0 cm, 15 g, 40--63 μm) and eluted with MeOH/W (2:8, v/v, 30 mL; 3:7, v/v, 30 mL; 4:6, v/v, 100 mL), yielding compound **5** (41 mg), compound **4** (5 mg), and compound **6** (5 mg).
The BuOH layer (7 g) was loaded on a silica gel column (6.0 × 60.0 cm, Kiesel gel 60, 480 g, 230--400 mesh) and successively eluted with DCM/MeOH/W (70:25:5, v/v/v, 1.5 L; 65:30:5, v/v/v, 1.5 L; 50:45:7, v/v/v, 1 L; 30:60:10, v/v/v, 1 L), which yielded nine fractions, B1--B9. Because the active fractions B6--B8 exhibited patterns similar to those of the main components based on TLC analysis with the developing solvent DCM/MeOH/W (70:25:5, v/v/v), they were pooled and further purified. A portion (600 mg) of the combined fraction (2.3 g) was first separated on a Sephadex LH20 column (2.5 × 60.0 cm, 100 g, 70--100 μm) and successively eluted with DCM/MeOH (3:7, v/v; 2:8, v/v; 1:9, v/v; 100% MeOH; 200 mL of each) and then further purified with a Sep-Pak C18 cartridge (Sep-Pak Vac 35cc, 10 g, Waters) and eluted with MeOH/W (1:9, 2:8, 3:7, 4:6, 5:5, v/v, 100 mL of each) to yield compound **3** (50 mg).
Structure determination of antibacterial compounds {#sec006}
--------------------------------------------------
The chemical structures of the active components were determined by using mass spectrometry, nuclear magnetic resonance (NMR) spectroscopy, and comparison with values reported in the literature. The electrospray ionization mass spectra (ESI-MS) of the isolated compounds were recorded on an MSD1100 single-quadruple mass spectrometer equipped with an electrospray ionizer (Hewlett-Packard Co., Palo Alto, CA, USA). ^1^H and ^13^C NMR spectra were measured using methanol-d~4~, dimethyl sulfoxide (DMSO)-d~6~ or acetone-d~6~ (E. Merck) with a Bruker AMX-500 spectrometer (Bruker, Analytische Messtechnik Gmbh, Rheinstetten, Germany). Chemical shifts were calculated using tetramethylsilane as the internal standard.
*In vitro* antibacterial activity {#sec007}
---------------------------------
The antibacterial activities of the isolated compounds (compounds **1**--**7**) were measured with the broth microdilution method against various plant pathogenic bacteria, including *R*. *solanacearum*. Briefly, suspensions of plant pathogenic bacteria in TSB media (100 μL) with inocula of 10^6^ CFU/mL were added to the wells of sterile 96-well plates. The stock solutions of the isolated compounds (25 mg/mL) were diluted 100-fold in the first wells and then subjected to two-fold serial dilutions in the growth media. The final concentrations of the isolated compounds were in the range 7.8--250 μg/mL. DMSO (1%), which corresponded to the highest concentration and did not affect the bacterial growth, was used as the negative control. Streptomycin sulfate was used as the positive control.
The inoculated plates were incubated at 30°C (except for *P*. *syringae* pv. *actinidiae* and *X*. *arboricola* pv. *pruni*, which were incubated at 25°C) for 18--36 h after shaking at 300 rpm for 10 min on a microplate shaker. The minimum inhibitory concentration (MIC) was defined as the lowest concentration that completely inhibited the growth of the bacteria. The assay was performed three times with three replicates for each extract at all concentrations tested.
Disease control efficacy of the methanol extract of *S*. *baccatum* against bacterial wilt of tomato {#sec008}
----------------------------------------------------------------------------------------------------
To evaluate the efficacy of the methanol extract of *S*. *baccatum* in controlling tomato bacterial wilt caused by *R*. *solanacearum* SL1944 (race 1, biovar 4) \[[@pone.0181499.ref004], [@pone.0181499.ref021]\], we used 3-week-old 'Seokwang' tomato plants at the four- to five-true-leaf stage. The plants were grown in vinyl pots with a volume of 90 mL in a greenhouse and were then transplanted into vinyl pots with a volume of 180 mL (one plant per pot). Two different amounts of starting material (100 and 200 mg) of the methanol plant extract were dissolved in 2 mL MeOH and then diluted in 98 mL distilled water containing 250 μg/mL Tween-20 to obtain test concentrations of 1000 and 2000 μg/mL. *R*. *solanacearum* was grown in TSA Petri dishes at 30°C for 48 h, harvested with distilled water, and adjusted to an optical density at 600 nm of 0.1 (approximately 1.5 × 10^8^ CFU/mL). The two methanol extract solutions were applied to the soil of each pot (20 mL per pot). After 3 h of treatment, a cell suspension (20 mL) of *R*. *solanacearum* was inoculated into the soil of each pot. Streptomycin sulfate (200 μg/mL) was used as the positive control and distilled water solutions containing Tween-20 (250 μg/mL) and MeOH (2%) were used as negative controls. The plants were maintained in a controlled climate at 30 ± 2°C and a relative humidity of 70--80%. The pots were arranged as a randomized complete block with five replicates per treatment. The experiment was repeated three times \[[@pone.0181499.ref004], [@pone.0181499.ref022]\].
The disease severity was ranked daily for 14 days and recorded on a scale of 0--4 as described by He et al. (1983): 0, no symptoms; 1, one leaf wilted; 2, two or three leaves wilted; 3, four or more leaves wilted; and 4, plant dead \[[@pone.0181499.ref004], [@pone.0181499.ref023]\]. The control value was calculated by using following formula \[[@pone.0181499.ref022]\]: $$\text{Control~value~}\left( \% \right) = 100 \times \left( {\text{disease~severity~of~control~}–\text{~disease~severity~of~treatment}} \right)/\text{disease~severity~of~control}$$
Statistical analysis {#sec009}
--------------------
The data were assessed by one-way analysis of variance (ANOVA) and the significance of the treatments was determined by Tukey's honest significant difference (HSD) for multiple comparisons (*p* = 0.05). Statistical analyses were performed by using SAS software (version 12.0, SAS Institute, Cary, NC). Differences were considered statistically significant for *p* values less than 0.05.
Results and discussion {#sec010}
======================
Structure determination of antibacterial compounds {#sec011}
--------------------------------------------------
The bioassay-guided fractionation of *S*. *baccatum* crude extract (15 g) led to the isolation of seven compounds. The chemical structures of these compounds were determined based on ^1^H NMR, ^13^C NMR, and ESI-MS data and through comparison with the previously reported literature values. The compounds were identified as gallic acid (compound **1**) \[[@pone.0181499.ref024]--[@pone.0181499.ref026]\], methyl gallate (compound **2**) \[[@pone.0181499.ref024]--[@pone.0181499.ref026]\], corilagin (compound **3**) \[[@pone.0181499.ref025]--[@pone.0181499.ref027]\], tercatain (compound **4**) \[[@pone.0181499.ref028], [@pone.0181499.ref029]\], chebulagic acid (compound **5**) \[[@pone.0181499.ref025], [@pone.0181499.ref026], [@pone.0181499.ref030], [@pone.0181499.ref031]\], chebulinic acid (compound **6**) \[[@pone.0181499.ref026], [@pone.0181499.ref030], [@pone.0181499.ref031]\], and quercetin 3-O-α-L-arabinopyranoside or guaijaverin (compound **7**) \[[@pone.0181499.ref032]--[@pone.0181499.ref034]\] ([Fig 1](#pone.0181499.g001){ref-type="fig"}). The ^1^H, ^13^C NMR, and ESI-MS data of these compounds are shown in [S1](#pone.0181499.s001){ref-type="supplementary-material"}--[S5](#pone.0181499.s005){ref-type="supplementary-material"} Tables.
{#pone.0181499.g001}
These compounds have previously been isolated from plant extracts, most of them belong to tannins group which possess various bioactivities including antibacterial activity \[[@pone.0181499.ref035]\]: gallic acid and methyl gallate from extracts of *Sedum takesimense* aerial parts \[[@pone.0181499.ref004]\]; *Euphorbia helioscopia* whole plants \[[@pone.0181499.ref024]\]; *Dimocarpus longan* seeds \[[@pone.0181499.ref025]\] and *Terminalia* spp. fruits, including *T*. *bellerica*, *T*. *chebula*, and *T*. *horrida* \[[@pone.0181499.ref026]\]; corilagin from extracts of the *D*. *longan* seeds \[[@pone.0181499.ref025]\] and *Terminalia* spp. fruits \[[@pone.0181499.ref026]\], *Punica granatum* leaves \[[@pone.0181499.ref027]\], and *E*. *fischeriana* roots \[[@pone.0181499.ref029]\]; tercatain from extracts of leaves of *T*. *catappa* \[[@pone.0181499.ref028]\] and *E*. *fischeriana* roots \[[@pone.0181499.ref029]\]; chebulagic and chebulinic acids from extracts of the *D*. *longan* seeds and the *Terminalia* spp. fruits \[[@pone.0181499.ref025], [@pone.0181499.ref026], [@pone.0181499.ref030], [@pone.0181499.ref031]\]; and quercetin 3-O-α-L-arabinopyranoside from extracts of *Woodfordia fruticosa* \[[@pone.0181499.ref032]\] and *Psidium guajava* leaves \[[@pone.0181499.ref036]\], and *Vaccinium macrocarpon* powder \[[@pone.0181499.ref034]\]. To the best of our knowledge, we have reported the first isolation of these seven compounds from *S*. *baccatum*.
*In vitro* antibacterial activity {#sec012}
---------------------------------
The antibacterial activity of the compounds isolated from *S*. *baccatum* is presented in [Table 1](#pone.0181499.t001){ref-type="table"}. Among the seven compounds, methyl gallate (compound **2**) exhibited the strongest broad-spectrum activity against most of the plant pathogenic bacteria tested, with MIC values between 26.0 and 250 μg/mL (except against *A*. *avenae* subsp. *cattleyae*). Of the glucoside gallates (compounds **3**--**6**), corilagin (compound **3**) and chebulagic acid (compound **5**) showed a slightly stronger activity than that of tercatain (compound **4**) and chebulinic acid (compound **6**), respectively. Gallic acid was less active than other hydrolysable tannins (compounds **2**--**6**) and quercetin 3-O-α-L-arabinopyranoside (compound **7**) was the least active compound ([Table 1](#pone.0181499.t001){ref-type="table"}).
10.1371/journal.pone.0181499.t001
###### Minimum inhibitory concentration (MIC) values of isolated compounds against plant pathogenic bacteria.
{#pone.0181499.t001g}
Bacterium MIC (μg/mL)
-------------------------------------------------------- ---------------- ---------------- ---------------- ----------------- ---------------- ---------------- -------
***Acidovorax avenae* subsp. *cattleyae*** 104.2 ± 31.3 b \>250 104.2 ± 31.3 b 250 a 208.3 ± 62.5 a \>250 \>250
***Agrobacterium tumefaciens*** \>250 250 \>250 \>250 \>250 \>250 \>250
***Burkholderia glumae*** \>250 62.5 c 104.2 ± 31.3 c 208.3 ± 62.5 ab 166.7 ± 62.5 b 250 a \>250
***Clavibacter michiganensis* subsp. *michiganensis*** \>250 88.3 ± 31.3 \>250 \>250 \>250 \>250 \>250
***Pectobacterium carotovorum* subsp. *carotovorum*** \>250 250 \>250 \>250 \>250 \>250 \>250
***Pectobacterium chrysanthemi*** \>250 104.2 ± 31.3 \>250 \>250 \>250 \>250 \>250
***Pseudomonas syringae* pv. *actinidiae*** \>250 166.7 ± 62.5 b 250 a 166.7 ± 62.5 b 104.2 ± 31.3 c 104.2 ± 31.3 c \>250
***Pseudomonas syringae* pv. *lachrymans*** \>250 208.3 ± 62.5 \>250 \>250 \>250 \>250 \>250
***Ralstonia solanacearum*** 41.7 ± 15.6 bc 26.0 ± 7.8 c 31.3 bc 52.1 ± 15.6 b 52.1 ± 15.6 b 52.1 ± 15.6 b 250 a
***Xanthomonas arboricola* pv. *pruni*** \>250 62.5 a 88.3 ± 31.3 a 88.3 ± 31.3 a 52.1 ± 15.6 a 52.1 ± 15.6 a \>250
Compound **1**, gallic acid; compound **2**, methyl gallate; compound **3**, corilagin; compound **4**, tercatain; compound **5**, chebulagic acid; compound **6**, chebulinic acid; compound **7**, quercetin 3-O-α-L-arabinopyranoside.
Means within the same row followed by the same letter are not significantly different (*p* = 0.05) as determined by Tukey's HSD test.
Of plant pathogenic bacteria tested, *R*. *solanacearum* was the most susceptible to all the isolated compounds, followed by *X*. *arboricola* pv. *pruni*, *P*. *syringae* pv. *actinidiae*, *B*. *glume* and *A*. *avenae* subsp. *cattleyae*. With the exception of compound **7** (MIC = 250 μg/mL), the isolated compounds showed impressive antibacterial activity against *R*. *solanacearum*, with very low MICs (26.0--52.1 μg/mL). Compounds **2**--**6** also exhibited strong antibacterial activities against *X*. *arboricola* pv. *pruni* (MIC = 52.1--88.3 μg/mL) ([Table 1](#pone.0181499.t001){ref-type="table"}).
Methyl gallate (compound **2**) was much more active than gallic acid (compound **1**). The measurements of antibacterial activity in our study agreed with those of previous reports \[[@pone.0181499.ref004], [@pone.0181499.ref037], [@pone.0181499.ref038]\]. Compounds with similar structures, such as **3** and **4**, and **5** and **6**, had similar antibacterial activity and spectra ([Table 1](#pone.0181499.t001){ref-type="table"}). Corilagin (compound **3**) has been reported against *Acinetobacter baumannii* \[[@pone.0181499.ref039]\], methicillin-resistant *Staphylococcus aureus* \[[@pone.0181499.ref040], [@pone.0181499.ref041]\], and *Escherichia coli* \[[@pone.0181499.ref041]\]. Chebulagic acid (compound **5**) and chebulinic acid (compound **6**) showed moderate antibacterial activity against *A*. *baumannii* \[[@pone.0181499.ref039]\]. The antibacterial activity of quercetin-3-O-α-L- arabinopyranoside or guaijaverin (compound **7**) against *Streptococcus mutans* has been reported \[[@pone.0181499.ref036]\]. To the best of our knowledge, information on the antibacterial activities of tercatain (compound **4**) is unavailable. In addition, this is the first report of the antibacterial activities of the isolated compounds (with the exception of compounds **1** and **2**) against plant pathogenic bacteria \[[@pone.0181499.ref004], [@pone.0181499.ref042]\].
Effect of the methanol extract on tomato bacterial wilt {#sec013}
-------------------------------------------------------
In the *in vivo* experiment, wilt symptoms were observed 5 days after inoculation. The extracts efficiently suppressed the development of tomato bacterial wilt in a dose-dependent manner. At concentrations of 1000 and 2000 μg/mL, the methanol extract showed control efficacies of 100 and 100% after 7 days of inoculation, and 63 and 83% after 14 days of inoculation, respectively. The disease control efficacies of the extract at 1000 μg/mL were higher than those of 200 μg/mL streptomycin sulfate after 7 and 14 days of inoculation ([Fig 2](#pone.0181499.g002){ref-type="fig"} and [S6 Table](#pone.0181499.s006){ref-type="supplementary-material"}). No phytotoxic symptoms appeared on the treated plants.
{#pone.0181499.g002}
Recent studies revealed that several plant-derived products also exhibited potential antibacterial activity against *R*. *solanacearum* in *in vivo* tests. *Allium fistulosum* extract at concentrations of 50 and 100% significantly reduced the incidence of bacterial wilt of tomato: only 6 and 14% of the plants were affected, respectively, whereas the disease affected 61% of the plants in the untreated control \[[@pone.0181499.ref043]\]. The leaf extract of *Eichhorina crassipes* reduced the severity index of the bacterial wilt by more than 91% \[[@pone.0181499.ref044]\]. Several essential oils, such as cinnamon and clove oils \[[@pone.0181499.ref013]\], lemongrass and palmarosa oils and their components such as thymol \[[@pone.0181499.ref045], [@pone.0181499.ref046]\], were found to effectively reduce the *R*. *solanacearum* populations and incidence of bacterial wilt of tomato grown in infested soil. Methyl gallate at a concentration of 500 μg/mL showed a control efficacy of 65.2% in greenhouse conditions \[[@pone.0181499.ref042]\]. It is difficult to compare the results of our study with those of previous studies owing to a number of factors, such as different experimental design conditions, plants species, and bacterial species. In this study, the control efficacy of *S*. *baccatum* at a concentration of 2000 μg/mL was similar to that of the wettable powder formulation of the ethyl acetate layer of *S*. *takesimense* at a 200-fold dilution, as described in our previous study \[[@pone.0181499.ref004]\].
Conclusions {#sec014}
===========
In this study, seven antibacterial compounds were isolated from the methanol extracts of *S*. *baccatum*. The compounds showed potent *in vitro* antibacterial activities against *R*. *solanacearum*, except for quercetin 3-O-α-L-arabinopyranoside. In addition, most compounds exhibited strong antibacterial activities against *X*. *arboricola* pv. *pruni* and *B*. *glumae*. Methyl gallate and corilagin showed the strongest activities. These results suggested that the extracts from *S*. *baccatum* or their isolated compounds were promising antibacterial agents for the control of bacterial wilt of tomato.
Botanical pesticides are a promising alternative to reduce the harmful effects caused by use of synthetic pesticides. They have become more attractive after the increase in demand for organic food. Further research into production, formulation, and delivery may greatly assist and promote the development of botanical pesticides. Therefore, further studies are necessary to examine the toxicity of the *S*. *baccatum* extract, evaluate the disease control efficacy of the extract in various fields, and develop optimum formulations of the crude extracts for the control of tomato bacterial wilt.
Supporting information {#sec015}
======================
###### NMR data of gallic acid and methyl gallate isolated from *Sapium baccatum* in methanol-d~4~.
(DOCX)
######
Click here for additional data file.
###### NMR data of corilagin and tercatain isolated from *Sapium baccatum* in DMSO-d~6~ and acetone-d~6~, respectively.
(DOCX)
######
Click here for additional data file.
###### NMR data of chebulagic acid and chebulinic acid isolated from *Sapium baccatum* in acetone-d~6~.
(DOCX)
######
Click here for additional data file.
###### NMR data of quercetin 3-O-α-L-arabinopyranoside isolated from *Sapium baccatum* in methanol-d~4~.
(DOCX)
######
Click here for additional data file.
###### ESI-MS data of seven active compounds isolated from *Sapium baccatum*.
(DOCX)
######
Click here for additional data file.
###### Effect of the methanol extract of *Sapium baccatum* on the control of tomato bacterial wilt under greenhouse conditions.
(DOCX)
######
Click here for additional data file.
[^1]: **Competing Interests:**The authors have declared that no competing interests exist.
| {
"pile_set_name": "PubMed Central"
} |
1. Introduction {#sec1-molecules-25-01442}
===============
Genus *Gentiana* is one of the largest groups in Gentianaceae, comprising 360 species that are widespread across Northwest of Africa, Europe, America, East of Australia, and Asia \[[@B1-molecules-25-01442]\]. Many species of this genus have significant economic value and they are widely used by the food and pharmaceutical industries in the world \[[@B2-molecules-25-01442],[@B3-molecules-25-01442],[@B4-molecules-25-01442]\]. In Europe, *G. lutea* (Yellow Gentian) are traditional materials for alcoholic bitter beverages and have a function of being appetite stimulating and improving digestion \[[@B4-molecules-25-01442],[@B5-molecules-25-01442]\]. In Asia, *Gentiana* has a long history in use for medicine \[[@B2-molecules-25-01442],[@B3-molecules-25-01442],[@B6-molecules-25-01442]\]. Places, including Iran, Mongolia, Japan and Korea have literature and details about nature and the of medicinal *Gentiana* plants found in these countries \[[@B3-molecules-25-01442],[@B7-molecules-25-01442],[@B8-molecules-25-01442]\]. In China, species of *Gentiana* are diverse (about 248 species) and some of them have been an important part of traditional Chinese medicine (TCM) for a long time \[[@B1-molecules-25-01442],[@B9-molecules-25-01442]\]. Approximattely 2000 years ago, Chinese Medicine monographs, "Shen Nong Ben Cao Jing", had described and recorded function and medicinal value of Longdan (Gentianae Radix et Rhizoma: dried root and rhizome of *G rigescens*, *G. trifloral*, *G. manshurica* and *G. scabra*) and Qinjiao (Gentianae Macrophyllae Radix: dried root of *G. macrophylla*, *G. straminea*, *G. crassicaulis*, and *G. dahurica*) \[[@B10-molecules-25-01442]\]. Presently, nine species of *Gentiana* have been recorded as the official drug of Pharmacopoeia of the People's Republic of China (Ch.P. 2015 edition) \[[@B9-molecules-25-01442]\]. But besides that, *G. cephalantha*, *G. davidii*, *G. loureirii*, *G. rubicunda G, lawrencei* var. *farreri*, and other species have been used as a popular herb in folk medicine and many other ethnomedicines for remedy digestive and respiratory illnesses \[[@B11-molecules-25-01442],[@B12-molecules-25-01442],[@B13-molecules-25-01442]\]. *Gentiana* and its related species are extensively used for various health disorders due to the cheap price of traditional herbs \[[@B13-molecules-25-01442]\]. These medicinal plants have always played an important role in the health care of local people, especially in the underdeveloped area of southwest China.
Chemical and pharmacological researches have indicated that the composition of bioactive compounds is diverse according to different *Gentiana* species \[[@B2-molecules-25-01442],[@B6-molecules-25-01442]\]. Until now, more than 500 secondary metabolites have been isolated from approximately 60 species \[[@B2-molecules-25-01442]\]. Those compounds, including iridoids, triterpenoids, flavonoids, alkaloids, and other types of secondary metabolites \[[@B2-molecules-25-01442],[@B14-molecules-25-01442],[@B15-molecules-25-01442]\]. *Gentiana* species have different therapeutic properties and medicinal functions because of the complicated chemical profiles \[[@B2-molecules-25-01442],[@B13-molecules-25-01442],[@B14-molecules-25-01442]\]. For example, *G. lute* and *G. rigescens* could be used as raw materials for the preparation of the therapeutic drug for Alzheimer's disease because of neuritogenic compounds that were isolated from the two species \[[@B4-molecules-25-01442],[@B16-molecules-25-01442],[@B17-molecules-25-01442]\]. Although *G. straminea* and *G. scabra* are rich in iridoids, chemical composition and traditional uses are different between the two species. *G. straminea* is used for treating rheumatic arthritis, while *G. scabra* is used for liver protection \[[@B6-molecules-25-01442],[@B14-molecules-25-01442]\]. *G. rhodantha* and *G. rigescens* usually are often confused in traditional medicine markets in southwest China. In fact, the former is good at treating cough and other throat illnesses that are caused by fever, and, while the latter is used for chronic liver disease, inflammatory skin diseases, and clearing away heat \[[@B9-molecules-25-01442],[@B13-molecules-25-01442]\]. Those cases showed that the identification of *Gentiana* species is crucial for keeping the clinical effect consistent and ensuring patients' medication safety.
*Gentiana* species show extremely high morphological similarity and their Chinese names of species are often used in confusion in the market (see sample information). Furthermore, the powder of medicinal materials of *Gentiana* species is difficult for achieving the identification. Although pharmacognosy morphology identification or microscopic identification based on inner structural composition features and the inclusions of medicinal materials may be used for this purpose \[[@B9-molecules-25-01442]\]; these works critically depend on personal experiences. In recent years, the researches regarding authenticity identification and discrimination of *Gentiana* and its relatives were focused on DNA barcoding, ISSR amplification, and other molecular identification technologies \[[@B18-molecules-25-01442],[@B19-molecules-25-01442],[@B20-molecules-25-01442],[@B21-molecules-25-01442]\]. In addition, chromatographic and mass-spectrometric techniques were applied for species classification \[[@B10-molecules-25-01442],[@B22-molecules-25-01442],[@B23-molecules-25-01442]\]. However, these methods need a complex process of extractions, tedious pretreatment, a great number of chemical reagents, waste time, and are expensive. A rapid, high-accurate, and green authenticity identification method needs to be established to ensure the effectiveness and safety of the clinical application of *Gentiana*. In the past few decades, ultraviolet-visible (UV-Vis), Raman, and infrared (IR) spectroscopic have gained the attention of various botany scientists and pharmacognosists \[[@B23-molecules-25-01442],[@B24-molecules-25-01442],[@B25-molecules-25-01442],[@B26-molecules-25-01442]\]. Among them, near-infrared (NIR) and mid-infrared (MIR) spectroscopy are probably the most publicized technologies \[[@B27-molecules-25-01442],[@B28-molecules-25-01442],[@B29-molecules-25-01442],[@B30-molecules-25-01442]\]. These two technologies can provide detailed structural information on sample properties and composition at the molecular level \[[@B31-molecules-25-01442],[@B32-molecules-25-01442],[@B33-molecules-25-01442],[@B34-molecules-25-01442]\]. Like human fingerprints, the infrared spectrum of any substance has to be unique \[[@B31-molecules-25-01442]\]. This is the reason for NIR and MIR spectral fingerprints can be applied to identify or classify different samples \[[@B31-molecules-25-01442],[@B32-molecules-25-01442],[@B33-molecules-25-01442],[@B34-molecules-25-01442]\]. In the case of medicinal plants, chemical constituents and their ratios of biochemicals of different species can vary substantially \[[@B35-molecules-25-01442],[@B36-molecules-25-01442]\]. The IR spectroscopy could be used for the identification of medicinal species because the corresponding spectral signals of these chemicals are highly specific \[[@B35-molecules-25-01442],[@B36-molecules-25-01442]\]. Recently, successful species discrimination of *Dendrobium*, *Paris*, *Rhodiola*, *Ganoderma*, and the other genus based on IR spectroscopy has been reported \[[@B35-molecules-25-01442],[@B36-molecules-25-01442],[@B37-molecules-25-01442],[@B38-molecules-25-01442]\].
In the process of spectral discrimination, it is necessary to establish a relationship between the chemical information and sample categories by chemometrics then to establish a classification model for the class identification of unknown samples \[[@B39-molecules-25-01442]\]. Additionally, feature variable selection and model optimization strategy that are based on chemometrics are key steps during the model building \[[@B40-molecules-25-01442]\]. From the literature, it can be found that a combination of variable selection methods and different algorithms could provide multifarious modeling strategies and most of them showed the superior ability for classification and identification \[[@B41-molecules-25-01442],[@B42-molecules-25-01442]\]. With the development of modeling methods, Wolpert developed stacked generalization in the early 1990s \[[@B43-molecules-25-01442]\]. This method combines multiple models together to produce a meta-model with equal or better classification performance than the constituent parts \[[@B43-molecules-25-01442],[@B44-molecules-25-01442]\]. In theory, this modeling strategy belongs to the ensemble model, and its classification result might be better than any of the constituent sub-models \[[@B44-molecules-25-01442],[@B45-molecules-25-01442]\]. For example, Shan's research showed that the performance of an extreme learning machine model that was based on stacked generalization was more robust than the traditional model \[[@B46-molecules-25-01442]\]. Sfakianakis's research reported a similar finding \[[@B47-molecules-25-01442]\]. Although stacked generalization might be an approach for improving model prediction accuracy and robustness, there was limited reporting of this method applied to medicinal plant research.
The aim of this research was (1) to investigate the application of NIR (near-infrared) and FT-MIR (Fourier transform mid-infrared) spectroscopies to the classification of medicinal *Gentiana* and its wild relatives; (2) to select the optimal bands that identify the differences among different species; and, (3) to examine the feasibility of using stacked generalization combined with infrared spectral data to identify *Gentiana* species. The results of the study may provide some basis for the safety and effectiveness utilization of medicinal *Gentiana* resources in China.
2. Results and Discussion {#sec2-molecules-25-01442}
=========================
2.1. Spectral Fingerprint of NIR and FT-MIR {#sec2dot1-molecules-25-01442}
-------------------------------------------
[Figure 1](#molecules-25-01442-f001){ref-type="fig"} shows the raw NIR spectra and FT-MIR that were obtained from 180 samples of *G. rigescens* and their relatives. It can be seen from the raw NIR spectra that there are seven distinct absorption bands, which are located at 6920, 5781, 5669, 5174, 4761, 4331, and 4260 cm^−1^, respectively ([Figure 1](#molecules-25-01442-f001){ref-type="fig"}A). In the whole FT-MIR spectral range ([Figure 1](#molecules-25-01442-f001){ref-type="fig"}B), 3335, 2924, 2853, 1735, 1636, 1516, 1319, 1265, 1147, 1033, and 831 cm^−1^ appeared in all species.
In the range of 7171--6514 cm^−1^, *G. rhodantha* is clearly different from the other two traditional medicinal *Gentiana* species. It is interesting that the NIR spectra of *T. chinense* and *T. cordatum* are similar to *G. rigescens* and *G. crassicaulis*. The spectral intensity of *G. davidii* at 4225 cm^−1^ was different from *G. rigescens* and *G. cephalantha* ([Figure 2](#molecules-25-01442-f002){ref-type="fig"}). In fact, the three species have similar plant morphology and *G. cephalantha* and *G. davidii* are primary alternative species of *G. rigescens* in remote rural of the southwest of China.
The FT-MIR spectra of 18 species showed very similar band distributions in the whole spectral range of 3587--2827 cm^−1^, but there were differences in the relative intensities of the spectral absorption bands of samples in the range of 1780--600 cm^−1^ ([Figure 3](#molecules-25-01442-f003){ref-type="fig"}). For example, the huge spectral differences between the bands 1709--1531, 1478--1207, 1168--1130, 1114--1015, 948--883, and 822--740 cm^−1^ were observed among *G. rigescens, G. crassicaulis, G. rhodantha, G. davidii, G. pseudosquarrosa*, and *G. stragulata*. Obviously, the fingerprints of *Tripterospermum* species and *Gentiana* species were significantly different in the 1650--1600, 1579--1494, 1458--1393, 1164--1126, 1112--1090, 950--883, and 822--740 cm^−1^, respectively ([Figure 3](#molecules-25-01442-f003){ref-type="fig"}).
2.2. Exploratory Statistical Analysis {#sec2dot2-molecules-25-01442}
-------------------------------------
Before statistical analysis, all of the spectra datasets were pretreated by the second derivative and standard normal variate for improving visualization results. The score plots that were obtained after principal component analysis (PCA) on the NIR data set are shown in [Figure 4](#molecules-25-01442-f004){ref-type="fig"}. A faint clustering of samples was observed in the figure. The score-plot for PC1 vs. PC2 displays *G. squarrosa* (11) could be clearly separated from other species ([Figure 4](#molecules-25-01442-f004){ref-type="fig"}A). In Score-plot for PC1 vs. PC3, *G stragulata* (5) and *T. cordatum* (18) were clustered and samples from the *G. crassicaulis* (6) were more easily differentiated from other samples ([Figure 4](#molecules-25-01442-f004){ref-type="fig"}B).
[Figure 5](#molecules-25-01442-f005){ref-type="fig"} shows score plots that were obtained by an application of PCA on the FT-MIR spectra data. According to the scatter plot of PC1 vs. PC2, *G. stragulata* (5) and *G. pseudosquarrosa* (12) were clustered. The samples of *G. lawrencei* var. *farreri* (4), *G. rhodantha* (15), and *G. striata* (16) were both located in the middle of the PC1 and PC2 axes. Most of the samples of *G. squarrosa* (11) were significantly different from other species and they were located on the negative side of PC1 and PC2. With the exception of the above species, all of the other species are grouped into one group ([Figure 5](#molecules-25-01442-f005){ref-type="fig"}A). From the scatter plot of PC1 vs. PC3. *G. stragulata* (5) and *G. crassicaulis* (6) were each separately clustered. Additionally, samples from the *G. squarrosa* (11) could be distinguished from those of the *G. pseudosquarrosa* (12) ([Figure 5](#molecules-25-01442-f005){ref-type="fig"}B).
The grouping results indicated a potential application value of NIR and FT-MIR fingerprint for the discrimination of medicinal *Gentiana* and its related species. Nonetheless, most of *Gentiana* species would be difficult to differentiate from one another, due to the overlap of their sample score. Hence, the application of supervised pattern recognition methods, such as random forest (RF), support vector machines (SVM), and k-nearest neighbors (KNN), for the development of classification models were required for enabling one to distinguish the samples.
2.3. Single Block Models for Sample Classification {#sec2dot3-molecules-25-01442}
--------------------------------------------------
### 2.3.1. Classification Based on Full Spectra {#sec2dot3dot1-molecules-25-01442}
In the section, all of the classification models were established by full spectra data (the total number of points in NIR and FT-MIR is 1487 and 1214, respectively) and 180 samples were separated into a calibration set (108 samples) and a validation set (72 samples) by the Kennard--Stone algorithm \[[@B48-molecules-25-01442]\]. Six performance parameters, including sensitivity (SE), specificity (SP), efficiency (EFF), accuracy (ACC), Matthews correlation coefficient (MCC), and Cohen's kappa coefficient (K), were applied to evaluate the identification ability of classification models \[[@B49-molecules-25-01442],[@B50-molecules-25-01442]\]. Those parameters values range from 0 to 1, indicating a perfect classification when the values are 1 \[[@B49-molecules-25-01442]\].
For RF models, model performance depends on the proper selection of the hyperparameters, which are *n~tree~* and *m~try~* \[[@B49-molecules-25-01442]\]. [Figures S1 and S2](#app1-molecules-25-01442){ref-type="app"} show the suitable hyperparameters and variation of model mean misclassification error (MMCE) with different hyperparameters. The lower MMCE the hyperparameter was better \[[@B50-molecules-25-01442]\]. [Table 1](#molecules-25-01442-t001){ref-type="table"} and [Table 2](#molecules-25-01442-t002){ref-type="table"} present classification accuracies rates in the calibration and validation data sets of 18 species that were obtained by NIR-RF and FT-MIR-RF models. For the two models, all of the samples in the calibration set were correctly classified. Additionally, the accuracy rates of validation sets were not less than 97.22%. Although the FT-MIR-RF model had higher total validation accuracy (94.44%), its SE, MCC, and EFF values of the validation set were lower than the NIR-RF model. Hence, the phenomenon of imbalance category recognition in the FT-MIR-RF model was worse ([Table 1](#molecules-25-01442-t001){ref-type="table"} and [Table 2](#molecules-25-01442-t002){ref-type="table"}).
For the SVM models, the optimum kernel function (sigmoid, polynomial, and radial kernel) and the cost function were important for modeling \[[@B35-molecules-25-01442],[@B51-molecules-25-01442]\]. Hyperparameter optimization results showed the linear kernel had lower MMCE value than sigmoid, polynomial, and radial kernel. Hence, the linear kernel was suitable for modeling ([Figures S3 and S4](#app1-molecules-25-01442){ref-type="app"}). Subsequently, the cost function was optimized. And the most suitable values 5 and 0.05 were selected as the best cost function for the SVM models of NIR and FT-MIR, respectively ([Figures S3 and S4](#app1-molecules-25-01442){ref-type="app"}). [Table 3](#molecules-25-01442-t003){ref-type="table"} and [Table 4](#molecules-25-01442-t004){ref-type="table"} present the major parameters of the calibration and validation sets for NIR-SVM and FT-MIR-SVM models. It could be seen that the samples of 18 species were better discriminated by using the FT-MIR data set. FT-MIR-SVM model achieved 100% total accuracy for the calibration set and validation sets.
Determining parameter *k* is critical for KNN \[[@B52-molecules-25-01442]\]. Hence, this hyperparameter was optimized before modeling and the optimum *k* value for NIR and FT-MIR data set were both one ([Figures S5 and S6](#app1-molecules-25-01442){ref-type="app"}). [Table 5](#molecules-25-01442-t005){ref-type="table"} and [Table 6](#molecules-25-01442-t006){ref-type="table"} present the classification accuracies rates in the calibration and validation data sets of 18 species obtained by NIR-KNN and FT-MIR-KNN models. Although the calibration set accuracy of the NIR-KNN model reached 100%, the total validation set accuracy was 88.89%. The performance of the FT-MIR-KNN model was better than the NIR-KNN model. Its total accuracy of the validation set was 94.44%. By comparison of validation set parameters (SE, SP, MCC, and EFF), it was clear that the performance of the KNN models was worse than RF and SVM models. Additionally, the highest classification accuracy was obtained with the use of the SVM combined with the FT-MIR data set.
### 2.3.2. Feature Selection {#sec2dot3dot2-molecules-25-01442}
It is necessary to screen out the most relevant chemical information for classification with specific variables selection methods in order to improve the classifier performance. In the study, five methods were used to feature selection ([Figure 6](#molecules-25-01442-f006){ref-type="fig"}). Firstly, VIP (variable importance in projection), Boruta, GARF (genetic algorithm combined with random forest), and GASVM (genetic algorithm combined with support vector machine) were applied to select feature variables \[[@B49-molecules-25-01442],[@B50-molecules-25-01442]\]. Secondly, the intersection of feature variables that were selected by these four algorithms was calculated and the result was the fifth approach of feature selection (Venn selection). [Figure 7](#molecules-25-01442-f007){ref-type="fig"} displays the number of feature variables of each selection method. Further analysis by Venn diagram found that 101 NIR variables and 73 FT-MIR variables were common characteristic variables of the four selection methods, respectively ([Figure 8](#molecules-25-01442-f008){ref-type="fig"}). Those variables were 6.79% and 6.01% of the full NIR spectrum and full FT-MIR spectrum, respectively. In the final, 10 feature subsets were established. They were the VIP-NIR, Bor-NIR, GARF-NIR, GASVM-NIR, Ven-NIR, VIP-MIR, Bor-MIR, GARF-MIR, GASVM, and Ven-MIR subset.
Models of the RF, SVM, and KNN were established based on the optimal data sets of NIR and FT-MIR to verify the validity of the feature selection for improving modeling performance. [Table 7](#molecules-25-01442-t007){ref-type="table"}, [Table 8](#molecules-25-01442-t008){ref-type="table"}, and [Tables S7--S36](#app1-molecules-25-01442){ref-type="app"} show the recognition effect of each model for the calibration set and the prediction set.
Obviously, the use of the VIP-NIR and Ven-NIR data sets could produce better classification performance for all of the classifiers in comparison with using full spectrum information ([Table 7](#molecules-25-01442-t007){ref-type="table"}). For the SVM classifier, its accuracy of the validation set increases to 98.61% with the use of feature variables that were selected by Boruta. However, there is a slight decrease in RF classifier performance with the use of the same feature variables. In addition, there is no improvement for classifiers' performance when using GASVM. Overall, in the case of NIR models, the performance of the classifiers for different *Gentiana* species showed the best results when using SVM that was combined with Boruta or Venn feature selection.
For MIR spectral data ([Table 8](#molecules-25-01442-t008){ref-type="table"}), the performance of the RF classifiers for the classification of samples shows acceptable results with maximum validation accuracies of 97.22% and 98.61% that were obtained using VIP-MIR and Ven-MIR data sets, respectively. Similar results have been achieved in the study of the KNN models. Although the validation accuracy of the Ven-MIR-SVM model was 98.61% and lower than the full spectra SVM model, but feature selection greatly reduced the SVM models' variables and kept a good classification performance of models.
Comprehensive comparison modeling results, the optimal spectrum that was selected by Venn was effectively increasing the performance of the NIR and FT-MIR classification models. Additionally, Ven-NIR and Ven-MIR were the optimal data sets for further modeling. The 101 NIR variables and 73 FT-MIR spectral variables were the most important variables for the species discrimination ([Figure 8](#molecules-25-01442-f008){ref-type="fig"}, [Table 7](#molecules-25-01442-t007){ref-type="table"} and [Table 8](#molecules-25-01442-t008){ref-type="table"}).
2.4. Model Stacking for Sample Classification {#sec2dot4-molecules-25-01442}
---------------------------------------------
Although most of the models that were based on data sets of Ven-NIR and Ven-MIR had high accuracy, it is possible that stacked generalization could establish a model that had a better performance when compared to the individual classifiers. Through comparisons of tge classification results of [Section 2.3.1](#sec2dot3dot1-molecules-25-01442){ref-type="sec"} and [Section 2.3.2](#sec2dot3dot2-molecules-25-01442){ref-type="sec"}, it could be found that RF and SVM appear to be the most effective of individual classifiers, realizing the highest classification rates in many cases when compared to KNN. Confusion matrices that correspond to Ven-NIR-RF, Ven-NIR-SVM, Ven-MIR-RF, and Ven-MIR-SVM shows that the predicted outputs of the two algorithms might be complementary ([Supplementary Materials Tables S1--S36](#app1-molecules-25-01442){ref-type="app"}). All of the results suggest that the two learners would be the best combination of base learners. Accordingly, RF and SNV models as level-0 base learners were employed in our stacked generalization. Additionally, RF, SNV, and KNN algorithms were used at level-1 learners, respectively. In the final, a total of six scenarios were performed with stacking experiments ([Table 9](#molecules-25-01442-t009){ref-type="table"}). Additonally, [Figure 9](#molecules-25-01442-f009){ref-type="fig"} shows the schemes for stacked generalization.
For the Ven-NIR data set (101 variables), the best performing classifier was scenario A. The next best-performing classifiers were scenario B and C, respectively. For the Ven-MIR data set (73 variables), a model of scenario E showed the highest classification rates and second were scenario D and F (94.00% and 90.00% classification rate respectively). Comparing the performance of different stacking models ([Table 9](#molecules-25-01442-t009){ref-type="table"} and [Supplementary Materials Tables S37--S42](#app1-molecules-25-01442){ref-type="app"}), SVM comes out to be the best algorithm at level-1. Additionally, the stacking model, based on the Ven-MIR data set, had the highest accuracy of calibrations and validations sets. The comprehensive analysis revealed that the SVM stacking model combined with the Ven-MIR data set had the best performance (SG-Ven-MIR-SVM).
2.5. Are Model Stacking Better than Data Fusion for Gentiana Species Discrimination? {#sec2dot5-molecules-25-01442}
------------------------------------------------------------------------------------
Presently, the application of stacked generalization for establishing classification models of different medicinal plants or herbs is rather scarce. On the contrary, another modeling approach, data fusion strategy, has been widely used for classification and geographical origin traceability of herbs and foods \[[@B48-molecules-25-01442],[@B49-molecules-25-01442],[@B53-molecules-25-01442],[@B54-molecules-25-01442]\]. Some researches stated that spectra data fusion, such as low-level and mid-level fusion strategies, could improve the discrimination capacity of the classification models and those strategies were usually more efficient than single spectroscopic techniques for modeling \[[@B48-molecules-25-01442],[@B49-molecules-25-01442]\]. We select the Ven-MIR-SG-SVM model in the last section of the research to compare with six data fusion models on prediction accuracy and validate the advantage of stacked generalization in the classification of *Gentiana* species.
In this study, the FT-MIR and NIR spectral signals were straightforwardly concatenated and they constitute a low-level fusion data set (a total of 2701 variables: the total number of the points in the both MIR and NIR spectra). The mid-level data fusion data set (174 variables) was made up of feature important variables from Ven-NIR (101 variables) and Ven-MIR (73 variables) subsets ([Figure 10](#molecules-25-01442-f010){ref-type="fig"}). Finally, the low- and mid-level data fusion matrices were used to establish the RF, SVM, and KNN models, respectively ([Table 10](#molecules-25-01442-t010){ref-type="table"} and [Tables S43--S48](#app1-molecules-25-01442){ref-type="app"}). For low-level data fusion, the order of successful classification rates of three algorithms was as follows: SVM \> RF and KNN. The SVM model resulted in a total accuracy of 100%. Additionlly, the validation set accuracy of RF and KNN were both 97.22%. In the case of mid-level fusion, the SVM model still achieved a total accuracy rate of 100%. In addition, the parameters of RF and KNN models that were based on feature fusion data set of FT-MIR and NIR spectra were higher than that of low-level data fusion.
The low and mid-level data fusion approach improved the discrimination capacity of the developed models to classify *Gentiana* samples, as shown in [Table 10](#molecules-25-01442-t010){ref-type="table"}. Among the six classification models that were based on data fusion strategy, Low-SVM, Mid-RF, Mid-SVM, and Mid-KNN were the best performing model according to accuracy, kappa coefficient, and other indicators. When compared with these models, the performance of SG-Ven-MIR-SVM was as good as them ([Table 9](#molecules-25-01442-t009){ref-type="table"} and [Table 10](#molecules-25-01442-t010){ref-type="table"}). The experimental results that were obtained from the two different modeling strategies showed that both model stacking and data fusion could result in a classification model with improved accuracy and enhanced robustness. Additionally, the strategy of stacked generalization could obtain efficient classification models that are as good as data fusion by fewer variables.
As we know, the data fusion (low-level and mid-level) approaches present a fusion of all variables or most important variables (feature variables) to create a model in order to exploit the synergy of the multispectral information to obtain an optimized model \[[@B53-molecules-25-01442],[@B54-molecules-25-01442],[@B55-molecules-25-01442],[@B56-molecules-25-01442]\]. However, the calculation time might be higher when increasing variables. In contrast, stacked generalization reduces the calculation time and keeps fewer variables by combining several different classification algorithms into one meta-model \[[@B57-molecules-25-01442],[@B58-molecules-25-01442],[@B59-molecules-25-01442]\]. In the case of discrimination of *Gentiana* and its relatives, only 73 variables used in the SG-Ven-MIR-SVM model, while low-level and mid-level data fusion models utilized 2701 and 174 variables for modeling, respectively. The variables number and modeling results indicated that the stacked generalization strategy is probably an important technique for improving species classification model predictive accuracy and avoiding overfitting.
3. Materials and Methods {#sec3-molecules-25-01442}
========================
3.1. Plant Material Collection {#sec3dot1-molecules-25-01442}
------------------------------
The 18 species used in the study belong to two genera (*Gentiana* and *Tripterospermum*) of Gentianaceae ([Figure 11](#molecules-25-01442-f011){ref-type="fig"}). All of the species were collected and identified during the flowering and fruiting time of 2018 and 2019. The voucher specimens of those plants were deposited in the College of Chemistry, Biological and Environment, Yuxi Normal University, Yu'xi, China. Their collection location is shown in [Table 11](#molecules-25-01442-t011){ref-type="table"} and medicinal use in southwest China was summarized in [Table 12](#molecules-25-01442-t012){ref-type="table"}.
In the laboratory, the fresh materials were authenticated. Subsequently, the samples were wash cleaning and dried at 50 °C as soon as possible. The dried whole plant was broken into powder with high-speed disintegrator. Finally, 180 powder samples were collected (10 powder samples per species). All sample powders were screened through a 100-mesh stainless sieve to obtain same-sized particles. The powders after sieving were stored in dry zip-lock bags for a further spectra scan of NIR and FT-MIR.
3.2. Near Infrared (FT-NIR) {#sec3dot2-molecules-25-01442}
---------------------------
The samples were scanned in the Antaris II spectrometer (Thermo Fisher Scientific, Madison, WI, USA). Each powdered sample was scanned from 10,000 to 4000 cm^−1^ with a resolution of 4 cm^−1^ until 16 scans were averaged.
3.3. Fourier Transform Mid Infrared (FT-MIR) {#sec3dot3-molecules-25-01442}
--------------------------------------------
The FT-MIR spectrum was recorded using a FT-IR spectrometer (Perkin Elmer, Norwalk, CT, USA) that was equipped with a deuterated triglycine sulfate (DTGS) detector and a ZnSe ATR (attenuated total reflection) accessory (PIKE technologies, Inc. Madison, WI, USA). The spectral fingerprint of every sample was recorded bands from 4000--600 cm^−1^ while using a resolution of 4 cm^−1^ and an accumulation of 16 scans. The ATR accessory is equipped with a unique metal O-ring for sample holding in order to control the path length and thickness of the sample ([Figure 12](#molecules-25-01442-f012){ref-type="fig"}). In the beginning, the metal O-ring was placed on the reflection diamond of accessory, and then the sample powder was put on the central of O-ring metal. At last, a pressure tower on the top of the metal O-ring was used to press the powder tightly until a constant pressure (131 ± 1 bar on the scale of the micrometric pressure device) \[[@B60-molecules-25-01442]\]. Before each measurement, a laboratory air spectrum was recorded and checked for remaining water and sample residues, as well as background deduction.
Spectrum signals from 2500 to 1800 cm^−1^ were not considered for further analysis due to strong crystal absorbance \[[@B61-molecules-25-01442]\]. Furthermore, spectral regions that 4000--3700 cm^−1^ (baseline area and did not provide relevant information) and 682--653 cm^−1^ (disturbing absorption band of CO~2~) were excluded prior to chemometric analysis \[[@B62-molecules-25-01442]\].
3.4. Statistical Analysis {#sec3dot4-molecules-25-01442}
-------------------------
The principal component analysis (PCA), unsupervised technique, has been widely applied in data dimension reduction and exploratory data analysis \[[@B37-molecules-25-01442],[@B63-molecules-25-01442]\]. From PCA-loading analysis, we can also extract the characteristic variables, which lead to differences between the samples. Additionally, in general, the more important the band corresponding to the spectral variable, the larger PCA-loading value. In this study, PCA was applied to test whether the NIR and FT-MIR spectra fingerprint can result in a clustering of 180 samples and analyze the similarity and dissimilarity in spectra data between species, which might be useful for further understanding phytochemical diversity among different species. Furthermore, the results of PCA would provide reference information for the creation of classification models based while using the supervised technique.
Random forests (RF) or decision tree forests is an ensemble learning technique \[[@B64-molecules-25-01442]\]. This algorithm is based on a combination of a large set of classification and regression trees \[[@B64-molecules-25-01442]\]. After the ensemble of trees (the forest), each tree gives a classification. Finally, the model uses a vote to combine the trees' predictions \[[@B64-molecules-25-01442]\]. RF can handle extremely large datasets and deal with the "curse of dimensionality" well. Therefore, RF is robust to over-fitting, noise, and outliers, and always performs well in problems with a low feature ratio \[[@B65-molecules-25-01442]\]. All of those indicate that RF is quite competitive relative to other ensemble learning techniques.
The support vector machine (SVM) algorithm is a non-parametric supervised classification \[[@B66-molecules-25-01442]\]. Many previous studies have reported the theory and detailed mathematical explanation of this algorithm \[[@B67-molecules-25-01442]\]. As one of the most robust and accurate data mining algorithms, SVM has been implemented in many programming languages, including R, MATLAB, and so on, which has led SVM to be adopted by a much wider audience. In recent years, SVM has successfully been applied to a number of applications, such as classification of species or geographical origin traceability of food \[[@B53-molecules-25-01442],[@B68-molecules-25-01442],[@B69-molecules-25-01442]\]. It is important to note that SVM can achieve high classification accuracy whlie using a small number of training samples \[[@B56-molecules-25-01442],[@B67-molecules-25-01442]\]. Additionally, it is also a suitable classifier for high-dimensional data \[[@B53-molecules-25-01442],[@B69-molecules-25-01442]\].
The k-nearest neighbors (KNN) algorithm is a distance-based non-parametric discriminant technique \[[@B70-molecules-25-01442]\]. As its name, this algorithm uses information regarding an example's k-nearest neighbors to classify unlabeled examples and assign one of them to the most common class among the k-nearest neighbors \[[@B70-molecules-25-01442]\]. KNN has been widely used in statistical applications and it has been one of the most successful supervised classification algorithms, especially for the task of multi-class classification \[[@B31-molecules-25-01442],[@B71-molecules-25-01442]\].
Hyperparameters of RF (*n~tree~* and *m~try~*), SVM (kernel function and cost function), and KNN (*k*) were optimized by using Bayesian optimization of mlr package combined with the MMCE model \[[@B50-molecules-25-01442]\]. The lower MMCE, the hyperparameter was better \[[@B50-molecules-25-01442]\].
Feature selection ("optimal wavenumbers" for classification modeling) is a critical step in the modeling process \[[@B72-molecules-25-01442]\]. There might be some irrelevant or noisy features in data sets because of the infrared techniques provide multivariate and non-specific signals \[[@B72-molecules-25-01442],[@B73-molecules-25-01442]\]. Feature selection of NIR and FT-MIR subsets was based on five methods. The first four were VIP (features were selected by the PLS-DA combined with VIP value) \[[@B49-molecules-25-01442]\], Boruta (features were selected by the Boruta algorithm) \[[@B49-molecules-25-01442]\], GARF (features were selected by the genetic algorithm combined with RF model), and GASVM (features were selected by the genetic algorithm combined with SVM model) \[[@B50-molecules-25-01442]\]. The last was Venn, which feature variables were the intersection of the results of the first four feature selection.
3.5. Model Stacking and Data Fusion {#sec3dot5-molecules-25-01442}
-----------------------------------
Stacked generalization (stacking) is one of the ensemble learning \[[@B43-molecules-25-01442]\]. The essence of the method is combined predictions from a number of base learners (level 0 models) to generate a more powerful meta-model (level 1 models), with the aim of reducing the generalization error \[[@B43-molecules-25-01442],[@B44-molecules-25-01442],[@B45-molecules-25-01442]\]. Hence, stacked generalization is an ensemble learning method with two or more levels models. The greatest advantage of stacked generalization is the free choice of base learners. Additionally, in general, the classification results of base learners might be complementarities and this combination might be helpful in improving the performance of the final meta-model \[[@B44-molecules-25-01442]\]. Hence, investigating the best methods for constructing the ensemble classifiers was one focus of stacking.
In our study, the first level (level-0) of stacking model is composed of several weak classifiers (base learners 1, base learners 2, base learners 3, base learners n) \[[@B45-molecules-25-01442]\]. Subsequently, the predicted probabilities of basic learners are used to train the second level model (final model) \[[@B45-molecules-25-01442]\]. [Figure 9](#molecules-25-01442-f009){ref-type="fig"} shows the schemes for stacked generalization.
Unlike stacked generalization, the data fusion strategy focus is on improving the model through best combine the subset. Most of the reported data fusion strategies include low-level data fusion and mid-level data fusion (feature-level data fusion) \[[@B48-molecules-25-01442],[@B53-molecules-25-01442]\].
Low-level data fusion, as its name suggests, subsets are straightforwardly concatenated and reconstitute an independent data matrix. Subsequently, the new dataset is used to establish the classification models \[[@B53-molecules-25-01442]\]. In the case of mid-level data fusion, classification models were established by a new data set, which were formed by concatenating the feature important variables from a subset by different feature selection algorithms \[[@B53-molecules-25-01442]\]. In the research, the low- and mid-level data fusion strategies were considered. Additionally, [Figure 10](#molecules-25-01442-f010){ref-type="fig"} shows the schemes for data fusion strategies.
3.6. Model Evaluation {#sec3dot6-molecules-25-01442}
---------------------
The values of TP (Correctly identified samples of positive class), TN (correctly identified samples of negative class), FN (incorrectly identified samples of positive class), and FP (incorrectly identified samples of negative class) were calculated according to the confusion matrices of the classification models \[[@B49-molecules-25-01442]\]. Subsequently, SE, SP, EFF, ACC, MCC, and K were calculated using Equations (1) to (6). $${ACC} = \frac{\left( {{TN} + {TP}} \right)}{\left( {{TP} + {TN} + {FP} + {FN}} \right)}$$ $${SE} = \frac{TP}{\left( {{TP} + {FN}} \right)}$$ $${SP} = \frac{TN}{\left( {{TN} + {FP}} \right)}$$ $${EFF} = \sqrt{{SE} \times {SP}}$$ $${MCC} = \frac{\left( {{TP} \times {TN} - {FP} \times {FN}} \right)}{\sqrt{\left( {{TP} + {FP}} \right)\left( {{TP} + {FN}} \right)\left( {{TN} + {FP}} \right)\left( {{TN} + {FN}} \right)}}$$ $${kappa} = \frac{\left( {{Po} - {Pe}} \right)}{\left( {1 - {Pe}} \right)}$$ Po: observed agreement value, Pe: expected agreement value.
3.7. Software {#sec3dot7-molecules-25-01442}
-------------
ATR correction of the FT-MIR spectra was completed by OMNIC 9.7.7 software (Thermo Fisher Scientific, Madison, WI, USA). The other spectral data preprocessing (SNV and 2nd derivative), PCA, and VIP analysis were performed by SIMCA-P^+^ 14.0 Software (Umetrics AB, Umea, Sweden). In the study, a strategy of two levels stacked generalization was used and the models were developed with R \[[@B50-molecules-25-01442]\]. Kennard--Stone algorithm was used to set the calibration sets and validation sets of all models (MATLAB, Version R 2017a, Mathworks, Natick, MA, USA). The RF, SVM, KNN technique, and feature selection of classification models were all implemented in R software (version 3.6.1, <https://www.r-project.org/>) base on randomForest, e1071, Boruta, mlr, and class package. The Venn diagrams were completed by the tools on BMKCloud ([www.biocloud.net](www.biocloud.net)).
4. Conclusions {#sec4-molecules-25-01442}
==============
The results of this study indicated that NIR and FT-MIR spectroscopic techniques combined with chemometrics could successfully discriminate Chinese medicinal *Gentiana* and their related species. Exploratory data analysis showed the NIR and FT-MIR spectroscopy indirect reflection interspecific phytochemistry diversity of medicinal *Gentiana* among the genera level and species level. Hence, there was a potential application value of NIR and FT-MIR fingerprint for the identification of medicinal *Gentiana* and its related species. Subsequently, supervised methods of pattern recognition were used for further analysis of spectra data. Firstly, six classification models based on RF, SVM and KNN algorithms were built on the full spectra data set that was obtained by the NIR and FT-MIR spectroscopy technique, respectively. The FT-MIR-SVM model performed more effectively than other classification models. Five approaches were applied to select optimal wavenumbers in order to improve the performance of the models and filter irrelevant or noisy features in data sets. In the end, the stacking models were built by stacked generalization combined with NIR and FT-MIR feature data sets. The modeling results suggest that RF and SVM were the best combinations of base learners (level-0). When compared the performance of six stacking models, SVM comes out to be the best algorithm at level 1 and the stacking model using the Ven-MIR data set had the highest accuracy of calibrations and validation sets. In conclusion, stacked generalization combined with feature selection is probably an important technique for improving the classification model predictive accuracy and to avoid overfitting.
**Sample Availability:** Samples of the compound are not available from the authors.
The following are available online. Figure S1. The *n~tree~* (left figure) and *m~try~* (right figure) screening of RF models based on Bayesian optimization methodology and NIR full spectra data, Figure S2. The *n~tree~* (left figure) and *m~try~* (right figure) screening of RF models based on Bayesian optimization methodology and FT-MIR full spectra data, Figure S3. The kernel (left figure) and cost (right figure) screening of SVM models based on Bayesian optimization methodology and NIR full spectra data, Figure S4. The *kernel* (left figure) and cost (right figure) screening of SVM models based on Bayesian optimization methodology and FT-MIR full spectra data, Figure S5. The *K* value screening of KNN models based on Bayesian optimization methodology, Figure S6. The *n~tree~* (left figure) and *m~try~* (right figure) screening of RF models based on Bayesian optimization methodology and NIR feature variables, Figure S7. The kernel (left figures) and cost (right figures) screening of SVM models based on Bayesian optimization methodology and NIR feature variables, Figure S8. The *K* value screening of KNN models based on Bayesian optimization methodology and NIR feature variables, Figure S9. The *n~tree~* (left figures) and *m~try~* (right figures) screening of RF models based on Bayesian optimization methodology and FT-MIR feature variables, Figure S10. The kernel (left figures) and cost (right figures) screening of SVM models based on Bayesian optimization methodology and FT-MIR feature variables, Figure S11. The *K* value screening of KNN models based on Bayesian optimization methodology and FT-MIR feature variables, Figure S12. The *n~tree~* (left figures) and *m~try~* (right figures) screening of RF models based on Bayesian optimization methodology and data fusion strategy, Figure S13. The kernel (left figures) and *cost* (right figures) screening of SVM models based on Bayesian optimization methodology and data fusion strategy, Figure S14. The *K* value screening of KNN models based on Bayesian optimization methodology and data fusion strategy, Table S1. Confusion matrixes of the calibration set and validation set of RF model based on NIR full spectra data, Table S2. Confusion matrixes of the calibration set and validation set of RF model based on FT-MIR full spectra data, Table S3. Confusion matrixes of the calibration set and validation set of SVM model based on NIR full spectra data, Table S4. Confusion matrixes of the calibration set and validation set of SVM model based on FT-MIR full spectra data, Table S5. Confusion matrixes of the calibration set and validation set of KNN model based on NIR full spectra data, Table S6. Confusion matrixes of the calibration set and validation set of KNN model based on FT-MIR full spectra data, Table S7. Confusion matrixes of the calibration set and validation set of VIP-NIR-RF, Table S8. Confusion matrixes of the calibration set and validation set of Bor-NIR-RF, Table S9. Confusion matrixes of the calibration set and validation set of GARF-NIR-RF, Table S10. Confusion matrixes of the calibration set and validation set of GASVM-NIR-RF, Table S11. Confusion matrixes of the calibration set and validation set of Ven-NIR-RF, Table S12. Confusion matrixes of the calibration set and validation set of VIP-NIR-SVM, Table S13. Confusion matrixes of the calibration set and validation set of Bor-NIR-SVM, Table S14. Confusion matrixes of the calibration set and validation set of GARF-NIR-SVM, Table S15. Confusion matrixes of the calibration set and validation set of GASVM-NIR-SVM, Table S16. Confusion matrixes of the calibration set and validation set of Ven-NIR-SVM, Table S17. Confusion matrixes of the calibration set and validation set of VIP-NIR-KNN, Table S18. Confusion matrixes of the calibration set and validation set of Bor-NIR-KNN, Table S19. Confusion matrixes of the calibration set and validation set of GARF-NIR-KNN, Table S20. Confusion matrixes of the calibration set and validation set of GASVM-NIR-KNN, Table S21. Confusion matrixes of the calibration set and validation set of Ven-NIR-KNN, Table S22. Confusion matrixes of the calibration set and validation set of VIP-MIR-RF, Table S23. Confusion matrixes of the calibration set and validation set of Bor-MIR-RF, Table S24. Confusion matrixes of the calibration set and validation set of GARF-MIR-RF, Table S25. Confusion matrixes of the calibration set and validation set of GASVM-MIR-RF, Table S26. Confusion matrixes of the calibration set and validation set of Ven-MIR-RF, Table S27. Confusion matrixes of the calibration set and validation set of VIP-MIR-SVM, Table S28. Confusion matrixes of the calibration set and validation set of Bor-MIR-SVM, Table S29. Confusion matrixes of the calibration set and validation set of GARF-MIR-SVM, Table S30. Confusion matrixes of the calibration set and validation set of GASVM-MIR-SVM, Table S31. Confusion matrixes of the calibration set and validation set of Ven-MIR-SVM, Table S32. Confusion matrixes of the calibration set and validation set of VIP-MIR-KNN, Table S33. Confusion matrixes of the calibration set and validation set of Bor-MIR-KNN, Table S34. Confusion matrixes of the calibration set and validation set of GARF-MIR-KNN, Table S35. Confusion matrixes of the calibration set and validation set of GASVM-MIR-KNN, Table S36. Confusion matrixes of the calibration set and validation set of Ven-MIR-KNN, Table S37. Confusion matrixes of the calibration set and validation set of SG-Ven-NIR-RF, Table S38. Confusion matrixes of the calibration set and validation set of SG-Ven-NIR-SVM, Table S39. Confusion matrixes of the calibration set and validation set of SG-Ven-NIR-KNN, Table S40. Confusion matrixes of the calibration set and validation set of SG-Ven-MIR-RF, Table S41. Confusion matrixes of the calibration set and validation set of SG-Ven-MIR-SVM, Table S42. Confusion matrixes of the calibration set and validation set of SG-Ven-MIR-KNN, Table S43. Confusion matrixes of the calibration set and validation set of Low-RF, Table S44. Confusion matrixes of the calibration set and validation set of Low-SVM, Table S45. Confusion matrixes of the calibration set and validation set of Low-KNN, Table S46. Confusion matrixes of the calibration set and validation set of Mid-RF, Table S47. Confusion matrixes of the calibration set and validation set of Mid-SVM, Table S48. Confusion matrixes of the calibration set and validation set of Mid-KNN.
######
Click here for additional data file.
H.Y. and Y.-Z.W. designed the project and revised the manuscript. T.S. performed the experiments, analyzed the data and wrote the manuscript. All authors have read and agree to the published version of the manuscript.
This research was supported by the Key Project of Yunnan Provincial Natural Science Foundation (2017FA049), the Projects for Applied Basic Research in Yunnan (2017FH001-028), Biodiversity Survey, Monitoring and Assessment (2019HB2096001006) and the Department of Science and Technology of Yunnan Province (2018IA075).
The authors declare no conflict of interest.
{#molecules-25-01442-f001}
{#molecules-25-01442-f002}
{#molecules-25-01442-f003}
{#molecules-25-01442-f004}
{#molecules-25-01442-f005}
{#molecules-25-01442-f006}
{#molecules-25-01442-f007}
{#molecules-25-01442-f008}
{#molecules-25-01442-f009}
{#molecules-25-01442-f010}
{#molecules-25-01442-f011}
{#molecules-25-01442-f012}
molecules-25-01442-t001_Table 1
######
The major parameters of random forests (RF) model based on NIR full spectra data.
Class Calibration Set Validation Set
------- ----------------- ---------------- ------ ------ ------ -------- ------ ------ ------ ------
1 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
2 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
3 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
4 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
5 100.00 1.00 1.00 1.00 1.00 97.22 0.75 0.99 0.74 0.86
6 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
7 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
8 100.00 1.00 1.00 1.00 1.00 98.61 0.75 1.00 0.86 0.87
9 100.00 1.00 1.00 1.00 1.00 98.61 1.00 0.99 0.89 0.99
10 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
11 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
12 100.00 1.00 1.00 1.00 1.00 97.22 0.75 0.99 0.74 0.86
13 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
14 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
15 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
16 100.00 1.00 1.00 1.00 1.00 97.22 0.75 0.99 0.74 0.86
17 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
18 100.00 1.00 1.00 1.00 1.00 97.22 0.75 0.99 0.74 0.86
molecules-25-01442-t002_Table 2
######
The major parameters of RF model based on FT-MIR full spectra data.
Class Calibration Set Validation Set
------- ----------------- ---------------- ------ ------ ------ -------- ------ ------ ------ ------
1 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
2 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
3 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
4 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
5 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
6 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
7 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
8 100.00 1.00 1.00 1.00 1.00 98.61 0.75 1.00 0.86 0.87
9 100.00 1.00 1.00 1.00 1.00 98.61 1.00 0.99 0.89 0.99
10 100.00 1.00 1.00 1.00 1.00 98.61 1.00 0.99 0.89 0.99
11 100.00 1.00 1.00 1.00 1.00 98.61 1.00 0.99 0.89 0.99
12 100.00 1.00 1.00 1.00 1.00 98.61 0.75 1.00 0.86 0.87
13 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
14 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
15 100.00 1.00 1.00 1.00 1.00 98.61 1.00 0.99 0.89 0.99
16 100.00 1.00 1.00 1.00 1.00 97.22 0.50 1.00 0.70 0.71
17 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
18 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
molecules-25-01442-t003_Table 3
######
The major parameters of SVM model based on NIR full spectra data.
Class Calibration Set Validation Set
------- ----------------- ---------------- ------ ------ ------ -------- ------ ------ ------ ------
1 100.00 1.00 1.00 1.00 1.00 98.61 0.75 1.00 0.86 0.87
2 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
3 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
4 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
5 100.00 1.00 1.00 1.00 1.00 97.22 0.75 0.99 0.74 0.86
6 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
7 100.00 1.00 1.00 1.00 1.00 98.61 1.00 0.99 0.89 0.99
8 100.00 1.00 1.00 1.00 1.00 98.61 0.75 1.00 0.86 0.87
9 100.00 1.00 1.00 1.00 1.00 98.61 1.00 0.99 0.89 0.99
10 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
11 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
12 100.00 1.00 1.00 1.00 1.00 97.22 0.75 0.99 0.74 0.86
13 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
14 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
15 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
16 100.00 1.00 1.00 1.00 1.00 97.22 0.75 0.99 0.74 0.86
17 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
18 100.00 1.00 1.00 1.00 1.00 97.22 0.75 0.99 0.74 0.86
molecules-25-01442-t004_Table 4
######
The major parameters of SVM model based on FT-MIR full spectra data.
Class Calibration Set Validation Set
------- ----------------- ---------------- ------ ------ ------ -------- ------ ------ ------ ------
1 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
2 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
3 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
4 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
5 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
6 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
7 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
8 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
9 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
10 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
11 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
12 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
13 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
14 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
15 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
16 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
17 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
18 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
molecules-25-01442-t005_Table 5
######
The major parameters of K-nearest neighbors (KNN) model based on NIR full spectra data.
Class Calibration Set Validation Set
------- ----------------- ---------------- ------ ------ ------ -------- ------ ------ ------ ------
1 100.00 1.00 1.00 1.00 1.00 98.61 1.00 0.99 0.89 0.99
2 100.00 1.00 1.00 1.00 1.00 97.22 0.50 1.00 0.70 0.71
3 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
4 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
5 100.00 1.00 1.00 1.00 1.00 95.83 0.75 0.97 0.65 0.85
6 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
7 100.00 1.00 1.00 1.00 1.00 98.61 0.75 1.00 0.86 0.87
8 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
9 100.00 1.00 1.00 1.00 1.00 97.22 0.75 0.99 0.74 0.86
10 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
11 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
12 100.00 1.00 1.00 1.00 1.00 95.83 0.75 0.97 0.65 0.85
13 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
14 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
15 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
16 100.00 1.00 1.00 1.00 1.00 97.22 0.75 0.99 0.74 0.86
17 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
18 100.00 1.00 1.00 1.00 1.00 97.22 0.75 0.99 0.74 0.86
molecules-25-01442-t006_Table 6
######
The major parameters of KNN model based on FT-MIR full spectra data.
Class Calibration Set Validation Set
------- ----------------- ---------------- ------ ------ ------ -------- ------ ------ ------ ------
1 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
2 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
3 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
4 100.00 1.00 1.00 1.00 1.00 98.61 1.00 0.99 0.89 0.99
5 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
6 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
7 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
8 100.00 1.00 1.00 1.00 1.00 97.22 1.00 0.97 0.80 0.99
9 100.00 1.00 1.00 1.00 1.00 97.22 0.50 1.00 0.70 0.71
10 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
11 100.00 1.00 1.00 1.00 1.00 98.61 1.00 0.99 0.89 0.99
12 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
13 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
14 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
15 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
16 100.00 1.00 1.00 1.00 1.00 97.22 0.50 1.00 0.70 0.71
17 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
18 100.00 1.00 1.00 1.00 1.00 100.00 1.00 1.00 1.00 1.00
molecules-25-01442-t007_Table 7
######
The major parameters (accuracy and kappa) of classification models based on different NIR feature variables.
Model Hyperparameters Calibration Set Validation Set
--------------- ---------------------------------- ----------------- ---------------- ------
VIP-NIR-RF *n~tree~* = 1774, *m~try~* = 14 100 97.22 0.97
Bor-NIR-RF *n~tree~* = 452, *m~try~* = 11 100 91.67 0.91
GARF-NIR-RF *n~tree~* = 678, *m~try~* = 22 100 91.67 0.91
GASVM-NIR-RF *n~tree~* = 1763, *m~try~* = 34 100 91.67 0.91
Ven-NIR-RF *n~tree~* = 1511, *m~try~* = 2 100 94.44 0.94
VIP-NIR-SVM *kernel* = linear, *cost* = 0.01 100 97.22 0.97
Bor-NIR-SVM *kernel* = linear, *cost* = 0.05 100 98.61 0.99
GARF-NIR-SVM *kernel* = linear, *cost* = 0.1 100 93.06 0.93
GASVM-NIR-SVM *kernel* = linear, *cost* = 0.05 100 91.67 0.91
Ven-NIR-SVM *kernel* = linear, *cost* = 0.05 100 98.61 0.99
VIP-NIR-KNN *k* = 1 100 95.83 0.96
Bor-NIR-KNN *k* = 1 100 94.44 0.94
GARF-NIR-KNN *k* = 1 100 87.50 0.87
GASVM-NIR-KNN *k* = 1 100 88.89 0.88
Ven-NIR-KNN *k* = 1 100 94.44 0.94
Note: VIP-NIR, Bor-NIR, GARF-NIR, GASVM-NIR and Ven-NIR were feature subsets of NIR extracted by VIP, Boruta, GARF, SVM and their common overlap variables.
molecules-25-01442-t008_Table 8
######
The major parameters (accuracy and kappa) of classification models based on different FT-MIR feature variables.
Model Hyperparameter Calibration Set Validation Set
--------------- ---------------------------------- ----------------- ---------------- ------
VIP-MIR-RF *n~tree~* = 1334, *m~try~* = 23 100 97.22 0.97
Bor-MIR-RF *n~tree~* = 1673, *m~try~* = 13 100 95.83 0.96
GARF-MIR-RF *n~tree~* = 958, *m~try~* = 20 100 95.83 0.96
GASVM-MIR-RF *n~tree~* = 297 *m~try~* = 31 100 94.44 0.94
Ven-MIR-RF *n~tree~* = 190, *m~try~* = 10 100 98.61 0.99
VIP-MIR-SVM *kernel* = linear, *cost* = 0.05 100 100 1.00
Bor-MIR-SVM *kernel* = linear, *cost* = 0.5 100 100 1.00
GARF-MIR-SVM *kernel* = linear, *cost* = 0.10 100 100 1.00
GASVM-MIR-SVM *kernel* = linear, *cost* = 1.00 100 100 1.00
Ven-MIR-SVM *kernel* = linear, *cost* = 1.00 100 98.61 0.99
VIP-MIR-KNN *k* = 1 100 98.61 0.99
Bor-MIR-KNN *k* = 1 100 97.22 0.97
GARF-MIR-KNN *k* = 1 100 95.83 0.96
GASVM-MIR-KNN *k* = 1 100 94.44 0.94
Ven-MIR-KNN *k* = 1 100 97.22 0.97
Note: VIP-MIR, Bor-MIR, GARF-MIR, GASVM-MIR and Ven-MIR were feature subsets of FT-MIR extracted by VIP, Boruta, GARF, SVM, and their common overlap variables.
molecules-25-01442-t009_Table 9
######
The major parameters (accuracy and kappa) of the stacking models.
Scenario Data Set Model Level 1 Calibration Set Validation Set
---------- ---------- ----------------- --------- ----------------- ---------------- ------
A Ven-NIR SG-Ven-NIR- RF RF 100.00 98.61 0.99
B Ven-NIR SG-Ven-NIR- SVM SVM 100.00 97.22 0.97
C Ven-NIR SG-Ven-NIR- KNN KNN 100.00 95.83 0.96
D Ven-MIR SG-Ven-MIR- RF RF 100.00 94.44 0.94
E Ven-MIR SG-Ven-MIR- SVM SVM 100.00 100.00 1.00
F Ven-MIR SG-Ven-MIR- KNN KNN 100.00 90.28 0.90
Note: base learners (level-0) of all stacking models were RF and SNV models
molecules-25-01442-t010_Table 10
######
The major parameters (accuracy and kappa) of the data fusion models.
Data Fusion Strategy Number of Variables Models Calibration Set Validation Set
---------------------- --------------------- --------- ----------------- ---------------- ------
Low-level fusion 2701 Low-RF 100.00 97.22 0.97
Low-level fusion 2701 Low-SVM 100.00 100.00 1.00
Low-level fusion 2701 Low-KNN 100.00 97.22 0.97
Mid-level fusion 174 Mid-RF 100.00 100.00 1.00
Mid-level fusion 174 Mid-SVM 100.00 100.00 1.00
Mid-level fusion 174 Mid-KNN 100.00 100.00 1.00
molecules-25-01442-t011_Table 11
######
Source of 180 *Gentian* and *Tripterospermum* species samples.
Class Genus Species Geographical Location
------- ------------------- ----------------------------------- --------------------------------------
1 *Gentiana* *G. rigescens* Yongde, Lincang, Yunnan, China
2 *Gentiana* *G. cephalantha* Xuyong, Luzhou, Sichuan, China
3 *Gentiana* *G. davidii* Jianghua, Yongzhou, Hunan, China
4 *Gentiana* *G. lawrencei* var. *farreri* Songpan, Aba, Sichuan, China
5 *Gentiana* *G. stragulata* Songpan, Aba, Sichuan, China
6 *Gentiana* *G. crassicaulis* Lanping, Nujiang, Yunnan, China
7 *Gentiana* *G. loureirii* Jianghua, Yongzhou, Hunan, China
8 *Gentiana* *G. napulifera* Liping, QianDong-nan, Guizhou, China
9 *Gentiana* *G. praticola* Liping, QianDong-nan, Guizhou, China
10 *Gentiana* *G. piasezkii* Ningqiang, Hanzhong, Shaanxi, China
11 *Gentiana* *G. squarrosa* Songpan, Aba, Sichuan, China
12 *Gentiana* *G. pseudosquarrosa* Songpan, Aba, Sichuan, China
13 *Gentiana* *G. rubicunda* Xianfeng, Enshi, Hubei, China
14 *Gentiana* *G. rubicunda* var. *samolifolia* Wufeng, Yichang, Hubei, China
15 *Gentiana* *G. rhodantha* Nayong, Bijie, Guizhou, China
16 *Gentiana* *G. striata* Songpan, Aba, Sichuan, China
17 *Tripterospermum* *T. chinense* Tonggu, Yichun, Jiangxi, China
18 *Tripterospermum* *T. cordatum* Tonggu, Yichun, Jiangxi, China
molecules-25-01442-t012_Table 12
######
Sample information including their application in southwest of China.
----------------------------------- ------------------------ ----------------------------------------------------------------------------------------------------------------------------------------------------------------------- ----------------------------------------------------
Species Chinese Name Disease Ch.P.
*G. rigescens* Dian Longdan heat-clearing, liver protection, icterohepatitis, Japanese encephalitis, cephalalgia, swelling and pain of eye \[[@B9-molecules-25-01442],[@B13-molecules-25-01442]\] listed (2015 edition) \[[@B9-molecules-25-01442]\]
*G. cephalantha* Tou hua Longdan heat-clearing, icterohepatitis unlisted
*G. davidii* Wu ling Longdan heat-clearing, urinary tract infection, conjunctivitis \[[@B13-molecules-25-01442]\] unlisted
*G. lawrencei* var. *farreri* Xian ye Longdan trachitis, cough, smallpox \[[@B13-molecules-25-01442]\] unlisted
*G. stragulata* Shi e Longdan none reported unlisted
*G. crassicaulis* Cu jing qin jiao heat-clearing, icterohepatitis, hematochezia, rheumatism \[[@B9-molecules-25-01442]\] listed (2015 edition) \[[@B9-molecules-25-01442]\]
*G. loureirii* Hua nan Longdan heat-clearing, icterohepatitis, diarrhea, swelling and pain of eye \[[@B13-molecules-25-01442]\] unlisted
*G. napulifera* Fu gen Longdan none reported unlisted
*G. praticola* Cao dian Longdan heat-clearing, detumescence analgesic \[[@B13-molecules-25-01442]\] unlisted
*G. piasezkii* Shan nan Longdan none reported unlisted
*G. squarrosa* Lin ye Longdan heat-clearing, acute appendicitis, swelling and pain of eye \[[@B13-molecules-25-01442]\] unlisted
*G. pseudosquarrosa* Jia lin ye Longdan none reported unlisted
*G. rubicunda* Shen hong Longdan dyspepsia, bone fracture, snakebite, diminish inflammation \[[@B13-molecules-25-01442]\] unlisted
*G. rubicunda* var. *samolifolia* Xiao fan lu ye Longdan none reported unlisted
*G. rhodantha* Hong hua Longdan heat-clearing, diminish inflammation, urinary tract infection, cold, icterohepatitis, diarrhea, scald \[[@B9-molecules-25-01442],[@B13-molecules-25-01442]\] listed (2015 edition) \[[@B9-molecules-25-01442]\]
*G. striata* Tiao wen Longdan none reported unlisted
*T. chinense* Shuang hudie heat-clearing, phthisis, pulmonary abscess, irregular menstruation \[[@B13-molecules-25-01442]\] unlisted
*T. cordatum* E mei Shuang hudie bone fracture \[[@B13-molecules-25-01442]\] unlisted
----------------------------------- ------------------------ ----------------------------------------------------------------------------------------------------------------------------------------------------------------------- ----------------------------------------------------
| {
"pile_set_name": "PubMed Central"
} |
The authors wish to make the following corrections to this paper:(1)The work described in this paper encompasses part of the European Intelligent Cooperative Sensing for Improved (ICSI) traffic efficiency project. The authors, as partners of the project, want to clarify that its main contribution, which is the central part of the paper, is the cooperative ITS architecture design and implementation. Specifically, the implementation of the Collaborative Learning Unit (CLU). Accordingly, the authors have decided to reduce and modify Section 5, that should be removed and replaced with the following:
5. Field Experimentation {#sec5-sensors-17-01301}
========================
This section presents a summary of the results and the process followed with the aim of testing the architecture functionality using real data coming from predefined ICSI scenarios. As has been mentioned in previous sections, two scenarios have been selected for this experimentation, corresponding to the two field trials scheduled in Pisa (Italy) and Lisbon (Portugal).
For the first test scenario, a location in the access point of the Pisa city centre has been selected. Real data about historical pollution levels in the city has been incorporated to the experimentation. On the other hand, in Lisbon, the system was tested along the A5 highway, and the trials were executed in order to assess the performance of the communications, namely in terms of the IT2S communication. Real data about the vehicles traffic flow in the A5 highway has been incorporated in order to test the CLU in a context as close to the real one as possible.
The main goal of the experimentation is to show that the system, by the aggregation of distributed sensors data and the implementation of collaborative intelligence, can provide relevant information related to the analysed use cases to the users, and therefore improve their decision making while using their vehicles.
5.1. Urban Scenario in Pisa, Italy {#sec5dot1-sensors-17-01301}
----------------------------------
Pisa is a city in Tuscany, central Italy, on the right bank of the mouth of the River Arno on the Tyrrhenian Sea. It is the capital city of the Province of Pisa. In this city, some areas have a controlled access through Restricted Traffic Zones (RTZs) and Low Emission Zones (LEZ). These LEZs are a way to reduce the pressure of non-residential traffic in highly touristic destinations. The objective of LEZs is to control the pollution level in highly congested and populated zones. In this context, and having the collected data about the traffic flow and the parking availability, the proposed urban test scenario includes the implementation of the following use cases:Alternative transport services;Monitoring and reduction of air pollution;Alternative paths signalling/route guidance; andCooperative parking slots monitoring.
The system constantly monitors the pollution of the roads in the RTZ and LEZs of Pisa. As has been explained before, when it predicts that the level of pollution can exceed the threshold, it suggests leaving the car in the parking area, continuing the trip using alternative transport services. In addition, the system estimates in real-time the number of free slots in the parking lot. In this way, it can recommend the most appropriate parking lot to leave the vehicle. This fact highlights the opportunity to provide intermodal transport solutions. [Table 3](#sensors-17-01301-t001){ref-type="table"} shows the tasks performed by the different components of the architecture in order to accomplish the requirements of the test scenario.
The verification of some of these use cases have been reproduced in the laboratory due to the need of producing events related to current pollution levels.
5.2. Highway Scenario in Lisbon, Portugal {#sec5dot2-sensors-17-01301}
-----------------------------------------
The A5 highway of Lisbon is a 25 km (16 miles) long motorway that connects the capital city of Portugal to Cascais. The first section of this infrastructure was opened in 1944, becoming the first motorway in Portugal and one of the firsts in the world. Nowadays, it is the most travelled motorway of the country and one of the most congestion prone ones. Six GWs were installed on the motorway road side cabinets. The RSUs were interconnected, and together with the GWs made possible the implementation of the platform on the field trial location.
Field trials were performed by partners of the ICSI project (IT, BRISA and INTECS). The following summarizes the tested use cases, not being the purpose of this work the detailed description of those trials, which will be subject of forthcoming works. In this context, the proposed highway test scenario includes the implementation of the following use cases:Monitoring of anomaly in traffic flows (congestion);Accident warning; andRoad works warning.
The proposed ITS distributed architecture provides in-route traveller information about traffic and road conditions according to both static and dynamic rules. In this way, drivers who are approaching a traffic jam can take some precaution measures, like reducing the speed in advance.
Each RSU's GW is configured to get *Abnormal Traffic* events (e.g., accident or roads work warning) from the next GW on the road. Additionally, each GW also gets *Vehicle Counter* events. These events come from the GW's attached sensors, and they are delivered to the CLU in order to detect congestion using the implemented artificial intelligence. In line with this, if congestion is detected, a *Congestion Level* event is launched. Each GW is listening for *Congestion Level* events from itself and from the next GW on the road in order to act with foresight and warn the drivers about expected traffic jumps.
The produced messages have been also successfully received at the GUI Web Platform and the HMI, included as a mobile application inside the vehicles, informing that it is recommended to take exit to avoid traffic congestion or alerting about an accident with foresight. [Figure 7](#sensors-17-01301-f001){ref-type="fig"} shows a snapshot of the demonstrator application GUI for the highway test scenario.
(2)The title, according to the changes made to the paper, must be replaced with: ""Design of a Cooperative ITS Architecture Based on Distributed RSUs.""(3)The abstract should be modified as follows:The sentence: ""Finally, functional and operational results observed through the experimentation are described."" should be removed and replaced with: ""Finally, the process followed with the aim of testing the architecture functionality is described.""(4)Author contributions should be replaced with the following:**Author Contributions:** Asier Moreno and Eneko Osaba wrote the paper. Asier Perallos, Asier Moreno and Giovanni Iovino designed and conceived the architecture. Enrique Onieva and Pablo Fernandez conceived and designed the experiments related to CLU testing. Giovanni Iovino in collaboration with the partners of the ICSI project (CNR, CNIT, IT and Brisa) performed the experiments. All authors supervised the paper and provided substantive comments.
The authors apologize for any inconvenience caused to readers.
The authors declare no conflict of interest.
{#sensors-17-01301-f001}
sensors-17-01301-t001_Table 3
######
Urban scenario related tasks.
Task Test Equipment
--------------------------------------------------------------------------------------- ---------------- --- --- ---
Traffic flow monitoring in the selected area √
Parking space vacancy monitoring in via *Pietrasantina* and *Palasport* parking areas √
Air pollution level monitoring in city centre √
Data management and event publishing √
Event processing, alert messages generation √
Evolution of congestion and pollution levels prediction for the next period of time √
Suggestion of new parking area target √
Display the availability of both parking areas √
Display the LTZ area in the map √
Suggestion of alternative transport modes √
[^1]: This work is an extension of the paper "Cooperative decision making ITS architecture based on distributed RSUs", presented at the 9th International Conference on UbiQuitous Computing & Ambient Intelligence (UCAmI 2015), Puerto Varas, Chile, 1--4 December 2015.
| {
"pile_set_name": "PubMed Central"
} |
Introduction {#S0001}
============
Persistently opened sclerotomies after 23-gauge pars plana vitrectomy (PPV) represent a small percentage of the surgical complications associated with the procedure. However, some studies have reported the need for sclerotomy sutures at the end of the surgery after functional testing to determine if the sclerotomies are open. Recent advances in small-gauge transconjunctival sutureless vitrectomy have changed the approach to patient management. Sutureless vitrectomy has contributed to the 20-gauge sclerotomy procedure in the speed of healing, decreased conjunctival scarring, improved patient comfort, reduced postoperative inflammation, shortened postoperative period, and decreased surgical time.[@CIT0001][@CIT0007]
Even so, studies have reported complications associated with 23-gauge transconjunctival sutureless sclerotomy, i.e., increased permeability through sutureless sclerotomies and ocular hypotony.[@CIT0008],[@CIT0009]
The current study evaluated the possible risk factors associated with persistently open 23-gauge sclerotomies 30 and 60 days postoperatively. Various complications have been reported in association with persistently opened sclerotomies, i.e., the risk of endophthalmitis and decreased tamponade effects of silicone oil and gas injection.[@CIT0005],[@CIT0008],[@CIT0009]
The objective of this prospective, longitudinal, observational study was to evaluate diathermy as a routine technique to minimize sclerotomy leakage after 23-gauge PPV and prevent ocular hypotony.
Methods {#S0002}
=======
Three hundred and eighty-two patients who underwent 23-gauge posterior PPV for any indication at a private clinic in Rio de Janeiro, Brazil, were evaluated. Participants were followed weekly; data for analysis were recorded 30 and 60 days postoperatively. The follow-up evaluations took place from June 2013 to January 2015. This study followed the Guidelines and Standards for Research Involving Human Beings (Resolution 196/1996 of the National Health Council) and the Declaration of Helsinki (1964). The Research Ethics Committee of the Universidade Federal Fluminense in Niteroi, Brazil, approved the study protocol. All participants provided written informed consent. A single surgeon (EFD) performed the surgeries. Trial registration number: Brazilian Ethics and Research Committee Approval: CAAE: 01111312.9.0000.5243 (Universidade Federal Fluminense).
The exclusion criteria included the presence of factors that could affect healing, e.g., glaucoma or intraocular hypertension; steroid (topical or systemic) or topical nonsteroidal anti-inflammatory drug (NSAID) use exceeding 30 days; rheumatic diseases, e.g., systemic lupus erythematosus, rheumatoid arthritis, or scleroderma; and a history of episcleritis, scleritis or uveitis, pterygium, and ocular trauma.
Fifty-five patients were excluded because of ocular trauma (n=3), history of uveitis (n=3), glaucoma (n=20), history of scleritis (n=4), rheumatic diseases (n=10), and topical NSAID use exceeding 30 days (n=15). Ultimately 327 patients (327 eyes) were included.
The following patient data were collected: personal data (gender, ethnicity, and age), previous medical history (diabetes, hypertension, and rheumatic diseases), medication use, and affected eye (right/left). Degenerative scleral hyaline plaque was evaluated as either present or absent.
All patients underwent an ophthalmologic examination that included measurement of the refraction (Bausch & Lomb Inc., Rochester, NY), best-corrected visual acuity using a Snellen chart, intraocular pressure (IOP) using Goldmann tonometry, indirect ophthalmoscopy (Welch Allyn Inc., Skaneateles Falls, NY), and slit-lamp examination (Takagi Inc., Nakano Gen, Japan).
Patients who underwent additional vitrectomies, i.e., a second or third posterior PPV, were included among the study patients. The surgical duration was recorded as short (up to 30 mins), medium (longer than 30 mins to 1 hr), or long (more than 1 hr).
The equipment and materials used in the procedure were a vitrectomy unit with a 23-gauge posterior PPV kit (Constellation Vision System, Alcon Laboratories, Inc., Fort Worth, TX), scleral buckle elements (silicone circling band and tire) (FCI Inc., Pembroke, MA), silicone oil 5000 CPS (Ophthalmos Inc., Sao Paulo, Brazil), intraoperative intravitreal drugs (triamcinolone acetonide 0.1 mL, Ophthalmos), bevacizumab (Avastin 0.1 mL, Genentech, Inc., South San Francisco, CA), indocyanine green 0.1 mL (Ophthalmos), and gas injection of perfluoropropane (Alcon Inc.).
The 23-gauge angled sclerotomy was created by inserting a trocar at a 45-degree angle parallel to the limbus. Once it passed the trocar sleeve, the angle was changed perpendicular to the surface and the cannula was inserted into the eye, making a biplanar entry. It was unnecessary to reposition the 23-gauge trocar plug during surgery. There was no conjunctival displacement at the moment of performing sclerotomies.
The sclerotomies were cauterized employing bipolar cautery (Alcon) at the end of the surgery, as the method previously published by Boscia et al.[@CIT0010] External cauterization was applied routinely to the three sclerotomy sites after clearing the sclerotomies with a cotton swab. The diathermy sclerotomy closure was focused on the site of the scleral wound and, furthermore, the edges of the conjunctival wound were joined and closed with the same technique. The bipolar cauterization (diathermy) was applied to the conjunctiva and sclera until the site had a whitish appearance.
All patients received subconjunctival injections of dexamethasone (2 mg) and tobramycin (20 mg) followed by topical prednisolone acetate and ofloxacin 4 times daily and atropine twice daily for 10 days.
The sclerotomies were described as open or closed based on a slit-lamp biomicroscopy evaluation. Each anatomic sclerotomy site (superonasal, inferotemporal, and superotemporal) was observed. Functional and morphologic tests were performed to determine the sclerotomy status. During the functional test, a cotton swab was pressed on the sclerotomy incisions to identify leakage. The morphologic test identified conjunctival elevations (conjunctival bubbles), subconjunctival silicone oil, entrapment of vitreous residue, and a visible incision with minimal separation of the edges, all of which were indicated an open sclerotomy. Ocular hypotony was defined as IOP \<10 mmHg. The IOP was measured at the 1st, 4th, 7th, 15th, 30th, and 60th postoperative days.
Variables studied {#S0002-S2001}
-----------------
The main variable was slit-lamp observation of sclerotomy healing at 30 and 60 days postoperatively. Healing was characterized as a binary variable and defined as either present or absent. The secondary variables and their subcategories possibly related to persistent sclerotomy opening were age (50 years and younger, and older than 50 years), gender, ethnicity (Caucasian, Latin, and African Descendant), and eye laterality. [Table 1](#T0001){ref-type="table"} lists the personal and epidemiologic variables. Table 1Correlations between persistent open sclerotomies 30 and 60 days postoperatively and personal and epidemiologic dataVariablen (%)Persistently open 30 days post-op*P*Persistently open 60 days post-op*P*n (%)n (%)PresentAbsentPresentAbsent12 (3.6)315 (96.4)2 (0.6)325 (99.4)GenderMale149 (46.4)5 (1.5)144 (44.0)0.98\*1 (0.3)148 (45.10.99\*Female178 (53.6)7 (2.1)171 (52.3)1 (0.3)177 (54.3)Age group (years)≤50216 (66.3)6 (1.8.)210 (64.2)0.23\*1 (0.3)215 (65.7)0.99\*\>50111 (33.9)6 (1.8.)105 (32.2)1 (0.3)110 (33.6)EthnicityCaucasian46 (14.6)2 (0,6)44 (13,4)0.94\*1 (0.3)45 (13.8)0.26\*\*Latino258 (78.9)9 (2.7)249 (76.1)1 (0.3)257 (78.5)African descendant23 (6.9)1 (0.3)22 (6.7)023 (7,1)Ocular lateralityRight167 (51.0)6 (1.8)161 (49.2)0.82\*1 (0.3)166 (50.7)0.49\*Left160 (49.0)6 (1.8)154 (47.0)1 (0.3)159 (48.6)[^1]
[Table 2](#T0002){ref-type="table"} shows the data related to ocular and systemic morbidities. The comorbidity variables were diabetes (with/without diabetic retinopathy \[DR\], proliferative/non-proliferative DR, and no diabetes); arterial hypertension (present, absent, and present associated with retinal vascular occlusion complications), and degenerative scleral hyaline plaque (present/absent). Table 2Correlations between persistently open sclerotomies 30 and 60 days postoperatively with ocular and systemic morbidity dataVariablen (%)Persistently open 30 days post-op*P*Persistently open 60 days post-op*P*n (%)n (%)PresentAbsentPresentAbsent12 (3.6)315 (96.4)2 (0.6)325 (99.4)Scleral hyaline plaquePresent15 (4.6)2 (0.6)13 (4.0)0.18\*1 (0.3)14 (4.3)0.16\*Absent312 (95.4)10 (3.0)302 (92.3)1 (0.3)311 (95.1)High myopia (HM) with or without myopic macular degenerationPresent108 (33.0)5 (1.5)103 (22.6)0.73\*1 (0.3)106 (32.4)0.81\*Absent219 (66.9)7 (2.1)212 (64.8)1 (0.3)219 (66.9)Ocular hypertensionPresent33 (10.5)3 (0.9)30 (9.1)0.20\*1 (0.3)32 (9.8)0.48\*Absent294 (89.4)9 (2.7)285 (87.3)1 (0.3)293 (89.6)Diabetes mellitus (DM)Present103 (31.4)4 (1.2)99 (30.2)0.85\*1 (0.3)102 (31.2)0.84\*Absent224 (68.5)8 (2.4)216 (66.1)1 (0.3)223 (68.1)Systemic arterial hypertensionPresent116 (29.8)4 (12)112 (34.2)0.88\*1 (0.3)115 (35.1)0.75\*Absent211 (64.1)8 (2.4)203 (62.1)1 (0.3)210 (64.2)[^2]
High myopia was classified the presence of non-degenerative high myopia (\>6 diopters), presence of degenerative myopia (presence of staphyloma or myopic retinal degeneration), or no myopia.
A postoperative ocular hypertension complication was recorded as absent or present if the IOP exceeded 20 mmHg for more than 21 days.
[Table 3](#T0003){ref-type="table"} shows the data related directly to the surgical procedure. The surgical-related procedures were intravitreal silicone oil (present/absent), gas injection of perfluoropropane (present/absent), endolaser photocoagulation (present/absent), perioperative intravitreal drugs (present or absent), and duration of vitrectomy surgery (short/medium/long). Table 3Correlations between persistently open sclerotomies 30 and 60 days postoperatively and surgical element dataVariablen (%)Persistently open 30 days post-op*P*Persistently Open 60 Days Post-op*P*n (%)n (%)PresentAbsentPresentAbsent12 (3.6)315 (96.4)2 (0.6)325 (99.4)Intraocular silicone oilPresent69 (21.1)6 (1.8)63 (17.1)0.03\*2 (2.1)68 (18.9)0.04\*\*Absent258 (78.8)6 (1.8)252 (71.8)0 (0.6)257 (78.0)Endolaser procedurePresent200 (57.5)8 (2.4)192 (54.1)0.92\*1 (0.3)199 (60.8)0.68\*Absent127 (42.4)4 (1.5)123 (34.8)1 (0.3)126 (38.5)Intravitreal drugsPresent133 (40.6)4 (1.2)129 (36.9)0.81\*1 (0.3)132 (39.7)0.65\*Absent194 (59.3)8 (2.4)186 (52.0)1 (0.3)193 (57.1)Gas infusionPresent119 (36.4)3 (0.9)116 (34.2)0.59\*1 (0.3)118 (36.1)0.73\*Absent208 (63.5)9 (2,7)199 (54.7)1 (0.3)207 (63.3)Additional vitrectomyPresent67 (20.6)6 (1.6)61 (14.7)0.02\*2 (2.2)66 (18.0)0.04\*\*Absent260 (79.4)6 (2.2)254 (74.2)0 (0.6)259 (78.8)Scleral buckle elementsPresent46 (14.2)5 (3.2)41 (10.9)0.01\*1 (1.1)45 (13.1)0.66\*Absent281 (85.7)7 (7.4)274 (78.2)1 (1.8)280 (83.7)PPV surgical timeShort172 (52.5)2 (3.0)169 (49.2)0.02†0 (1.5)171 (51.0)0.22\*\*Long155 (47.4)10 (3.5)146 (36.3)2 (1.0)154 (39.0)[^3]
The procedures associated with vitreoretinal surgery, such as the use of scleral buckle elements (silicone circling band or tire) and cryopexy, were recorded as present or absent.
The sclerotomy sites and leakage associated with the three locations are analyzed in [Table 4](#T0004){ref-type="table"}. Considering the morphologic appearance of the sclerotomies, healing was described as open or closed. Each of the sites was evaluated separately. Leakage and ocular hypotony (present/absent) also were studied. Table 4Correlations of persistently open sclerotomies 30 and 60 days postoperatively and sclerotomy site with presence or absence of leakage**Variable**n (%)Persistently open 30 days post-op*P*Persistently open 60 days post-op*P*Present withPresent withoutAbsentPresent withPresent withoutAbsentLeakageLeakageLeakageLeakage**(Sclerotomy)**9814 (0.4)8 (0.8)969 (98.8)02 (0.2)979 (99.8)Superotemporal327 (33.3)2 (0.2)4 (0.4)321 (32.4)0.05\*01 (0.1)326 (33.3)0.05\*Inferotemporal327 (33.3)01 (0.1)326 (33.2)00327 (33.3)Superonasal327 (33.3)2 (0.2)3 (0.3)322 (32.8)01 (0.1)326 (33.2)[^4]
Statistical evaluation {#S0002-S2002}
----------------------
To determine the possible variables involved in persistent sclerotomy openings, the chi-square test or Fisher Exact test was used to analyze gender, age, ethnicity, hyaline plaque, eye laterality, silicone oil, gas injection, endolaser photocoagulation, intraoperative use of intravitreal drugs, scleral buckle elements, postoperative intraocular hypertension, postoperative ocular hypotony, vitrectomy surgery reoperations, sclerotomy site, and leakage. The Kruskal--Wallis test evaluated the variables with ordinal subcategories, such as diabetes, hypertension, and surgical duration.
To avoid the confounding effects of multiple factors, the samples were subjected to multivariate statistical analysis but only if the variable obtained a significance level ≤20% in the initial univariate analysis. The odds ratio (OR), confidence interval (CI), and significance (*P*) were calculated using binary logistic regression. [Table 5](#T0005){ref-type="table"} shows the results of the analysis. *P*\<0.05 was considered significant. The statistical program SPSS 20.0 (SPSS Inc., Chicago, IL) was used. Table 5Multivariate analysis with binary logistical regression between factors associated with persistent sclerotomy opening 30 and 60 days postoperativelyVariablePersistently open 30 days post-opPersistently open 60 days post-op*P*\*ORCI*P*\*ORCIIntraocular silicone oil0.711.0890.718--1.1250.450.4040.089--1.337Additional vitrectomies0.036.7317.197--10.8790.055.0228.115--32.019Scleral buckle elements0.210.5790.095--0.645Surgical time0.057.0058.209--11.199[^5][^6]
Results {#S0003}
=======
Univariate analyses ([Tables 1](#T0001){ref-type="table"} and [2](#T0002){ref-type="table"}) indicated that personal data, ocular morbidity, scleral hyaline plaque, and systemic morbidity were not correlated significantly with persistently open sclerotomies, which were significant only during 30 days postoperatively.
[Table 3](#T0003){ref-type="table"} shows that the variables related to the use of silicone oil and additional vitrectomies were significant at 30 days (*P*=0.03 and *P*=0.02, respectively) and 60 days (*P*=0.04 and *P*=0.04, respectively) postoperatively by univariate analysis.
[Table 3](#T0003){ref-type="table"} shows no significant associations between persistent sclerotomy openings and use of an endolaser photocoagulation probe, gas injection, and intraoperative intravitreal drug use.
[Table 4](#T0004){ref-type="table"} shows that the superotemporal and superonasal sclerotomy sites were most often involved in persistent opening of sclerotomy incisions. Inferotemporal sclerotomy was less susceptible to opening of the incision.
Morphologic evaluation of the incisions combined with functional testing for dynamic leakage provided different results. The morphologic test showed the possibility of a sclerotomy having an incision with an open morphologic appearance (incomplete healing) that was functionally closed (absence of leakage). This occurred in 0.8% of the sample sclerotomies during the first 30 days postoperatively and 0.2% of the sclerotomies after 60 days.
Searches for dynamic leakage at the sclerotomy sites identified leakage in 0.4% of the sample sclerotomies up to 30 days postoperatively. The study found no evidence of spontaneous sclerotomy leakage.
Multivariate analyses were significant for additional vitrectomies at 30 days and 60 days postoperatively (*P*=0.03 and *P*=0.05, respectively) and for the surgical duration at 30 days postoperatively (*P*=0.05) ([Table 5](#T0005){ref-type="table"}). The other variables, i.e., silicone oil, gas injection, endolaser photocoagulation probe, scleral buckle elements, ocular hypertension complications, and sclerotomy closure method showed no significant correlations.
Regarding surgical complications, four patients revealed leakage evidence within the initial three to four days postoperatively, ceasing spontaneously after this time. These patients had undergone previous prophylactic sclerotomy cauterization during the 23-gauge PPV, with no further treatment required. Endophthalmitis or postoperative ocular hypotension or other cases of silicone oil extrusion or gas injection did not occur in other cases.
Eight patients had partial cicatrization of the sclerotomy (considered morphologically opened, but without leakage) at 30 days postoperatively.
No patient complained at three days post-operatory time after surgery regarding any inflammation, scratch, foreign body sensation, or major inflammation after diathermy.
Discussion {#S0004}
==========
This study is the first to use diathermy in three routine sclerotomies, with or without intraoperatively sclerotomy leakage and evaluate the presence of degenerative scleral hyaline plaque as a risk factor for persistent 23-gauge sclerotomy opening. This study evaluated persistent opening of the incisions over 30 and 60 days postoperatively, unlike other studies that assessed the incision earlier.[@CIT0011]
Although 23-gauge sclerotomy has advantages, there are potential complications, i.e., wound permeability, vitreous entrapment, hypotonia, choroidal detachment, retinal detachment, endophthalmitis, and subconjunctival migration of silicone oil and gas.[@CIT0009] In cases of sclerotomy leakage, many authors prefer sutures. Although the sutureless vitrectomy techniques have improved, some cases still need intraoperative or postoperative sutures on the sclerotomy site.[@CIT0004],[@CIT0007] Chieh et al reported that 38% of the eyes needed sutures.[@CIT0012] In the current study, no eyes required suture placement to close the sclerotomy. Woo et al used intraoperative sutures in 11.2% of the patients who underwent posterior PPV, and no sutures were needed postoperatively.[@CIT0008]
Sclerotomy sutures have been reported in all sizes of microincision vitrectomies due to intraoperative leakage and prevention of ocular hypotony as well as other complications. In a prospective study by Veritti et al, the 27-gauge transconjunctival sutureless vitrectomy was compared with the 25-gauge for the treatment of primary rhegmatogenous retinal detachment. Sclerotomies were sutured in 8% of the cases in the 27-gauge group and in 29% of the cases in the 25-gauge group (*P*=0.017). Transient ocular hypotony (IOP \<10 mmHg) was observed in 2.7% of the eyes in the 25-gauge group, and in no case in the 27-gauge group.[@CIT0013] The present study did not present any cases ocular hypotony, even in cases with persistent opening of the sclerotomy.
Woo et al operated on 170 right eyes and 152 left eyes without statistical significance of ocular laterality.[@CIT0008] In cases requiring intraoperative sclerotomy suturing, the incidence of placing sutures on the dominant-hand side (site used for the vitreous cutter, endolaser photocoagulation probe, forceps, and vitrectomy scissors) was 83.3%. For the non-dominant-hand side (endoillumination site), the incidence was 69%, and for the inferotemporal site (site of the infusion cannula), the incidence was 83%.[@CIT0008]
We also found no statistical significance regarding ocular laterality, which is important for defining the sclerotomy sites with the most stress undergoing surgical manipulation, in this study the superotemporal and superonasal sites. We believe that the lower incidence of opening of the inferotemporal sclerotomy site resulted from less surgical stress at this site of infusion cannula placement.
Woo et al also used gas injection in 26.1% of the patients and silicone oil implantation in 1.2% of the patients. In this study, in contrast, silicone oil was used in 21.1% of the patients and gas injection on 36.4%.[@CIT0008]
Woo et al also observed that age below 50 years, a history of previous vitrectomy, vitreous base dissection, male gender, and high myopia (axial length ≥25 mm) were associated significantly with intraoperative sutures. Multivariate analysis identified the following as significant risk factors for intraoperative suture: a previous vitrectomy (*P*\<0.0005), age below 50 years (*P*\<0.0005), and vitreous base dissection (*P*=0.018).[@CIT0008]
In the current study, when gender, age, and high myopia were evaluated using univariate analysis, no correlation was seen with persistent opening of the incision. Additional vitrectomies and the surgical duration, which indirectly measured the surgical extent and complexity, were correlated significantly with persistent sclerotomy opening. Only the additional vitrectomies were a risk factor in the late postoperative period.
An important consideration is that other studies used different methodologies and evaluated the presence of leakage and its complications (hypotony, choroidal detachment) in greater detail than the persistence of opening of the incisions with morphologic evaluation.[@CIT0014]--[@CIT0016] Javey et al and Küçük et al reported the effects on single-step sclerotomy and/or silicone oil endotamponade, and the effect of removing the cannula over the light pipe.[@CIT0017],[@CIT0018]
The current study used diathermy as a method to close the microincision vitrectomy sclerotomies. The authors analyzed the morphological status of the sclerotomy (open or closed), the presence of leakage, and the occurrence of ocular hypotony. In addition, the main risk factors were also evaluated.
Various techniques have attempted to reduce sclerotomy leakage, i.e., use of tissue glue, polyethylene glycol-based hydrogel bandages, and absorbable sutures.[@CIT0011] In the current study, diathermy was used routinely over the sclerotomy incision intraoperatively, regardless of whether or not a sclerotomy leakage.
Duval et al reported sclerotomy suture rates ranging from 3.9% to 62% from five surgeons at the end of surgery and observed that the most important risk factor for the use of sutures in sclerotomy was the surgical technique of each surgeon.[@CIT0019]
Duval et al also reported that of 589 eyes, 227 needed sutures at one or more sclerotomy sites postoperatively. Cases in which gas injection was not used required greater use of sutures (42.9%) compared with cases in which gas was used (34.8%).[@CIT0019] In the current study, no correlation was seen between persistent opening of the incision and gas injection.
Barak et al conducted a histologic study and reported tissue fusion over the sclerotomies that sealed the outer portion of the sclerotomies in all small-gauge sclerotomies treated with diathermy.[@CIT0007]
In conclusion, the current investigation showed that even after a prolonged postoperative period, sclerotomy opening may persist in terms of its morphologic appearance, although it is functionally closed. Thus, routine diathermy of the sclerotomies and good surgical technique can prevent this. While several factors affected this outcome, the main risk factors identified were additional vitrectomies and the long surgical duration.
The authors are grateful to Professor Ronir Haggio Luiz, Federal University of Rio de Janeiro (UFRJ), for assistance with statistics and to CNPq (Brazilian Council of Research) and CAPES.
Disclosure {#S0005}
==========
The authors have no proprietary or financial interests to report in this work.
[^1]: **Notes:** n=number of patients, \*Chi-square test, \*\*Fisher exact test.
[^2]: **Notes:** n=number of patients; \*Chi-square test.
**Abbreviation:** DR, diabetic retinopathy.
[^3]: **Notes:** n=number of patients; \*Chi-square test; \*\*Fisher exact test; †Kruskal--Wallis test.
**Abbreviation:** PPV, pars plana vitrectomy.
[^4]: **Notes:** n=number of sclerotomies, \*Chi-square test. The significance refers to total values for the presence of persistent sclerotomy opening, with and without leakage (anatomic site), due to very small values for the statistical test in this variable subcategory (leakage).
[^5]: **Note:** \*0.05 statistic significance level.
[^6]: **Abbreviations:** OR, odds ratio; CI, confidence interval.
| {
"pile_set_name": "PubMed Central"
} |
Introduction {#s1}
============
Autoimmune diseases reflect the interplay between environment and genetic factors ([@B1]--[@B3]). The diseases share a substantial degree of immunopathology, including increased secretion of inflammatory cytokines by autoreactive CD4^+^ T cells and a loss of Regulatory T cells (Tregs) function ([@B4], [@B5]). Most autoimmune diseases are classified based on which organs and tissues are targeted by the damaging immune response \[e.g., primary biliary cirrhosis ([@B6]), type 1 diabetes mellitus ([@B1]), arthritis ([@B7]), and myositis ([@B8])\]. Autoimmune diseases include many types, and there is an autoimmune disease specific to nearly every organ in the body ([@B8]). Clinically, specific diagnostic methods are used for each autoimmune disease, which is tedious and costly. Therefore, it is urgent to find a universal marker to diagnose autoimmune diseases, which will provide new possibilities for autoimmune disease detection and treatment.
Tregs, characterized by expressing CD4, CD25, and Forkhead box P3 (Foxp3) transcription factor, play pivotal roles in protecting an individual from autoimmunity. These roles have been identified in mice with Treg depletion or absence, which results in the development of autoimmune gastritis, thyroiditis, multiple sclerosis (MS), type 1 diabetes, ankylosing spondylitis (AS), inflammatory bowel disease (IBD), systemic lupus erythematosus (SLE), rheumatoid arthritis (RA), Sjogren\'s syndrome (SS), and ulcerative colitis (UC) ([@B4], [@B9]--[@B11]). Thus, Tregs depletion in mice is considered a representative animal model of autoimmune disease. Two approaches are typically used to deplete Tregs in mice. In the first, Treg cells are depleted by constitutive expressing CD25. Previous studies have demonstrated that injection of depleting antibodies directed against CD25 will lead to a mild autoimmune disease ([@B11]--[@B13]). The other method is to use Foxp3^DTR^ mice, which are created by designing a construct in which cDNA encoding the diphtheria toxin receptor (DTR) is inserted into the 3′ untranslated region (3′UTR) of Foxp3. After continuous injection of DT, Treg cells are efficiently depleted, which affects multiple organs and leads to fatal autoimmune pathology ([@B14], [@B15]).
microRNAs (miRNAs) are small regulatory RNA molecules that function to regulate gene expression and play vital roles in various physiologic and pathologic processes. Our study ([@B16]) and others\' studies ([@B17]--[@B21]) have found that serum miRNAs can serve as potential biomarkers for detecting a variety of diseases, including immune diseases. Song et al. ([@B22]) found that circulating miRNAs play a key role in diagnosing congenital heart defects (CHD) and predicting CHD risk in offspring. Sharaf-Eldin et al. ([@B23]) determined that three miRNAs (miR-326, miR-223, and miR-145) expression profiles are promising diagnostic biomarkers for SLE and MS. Anaparti et al. ([@B24]) indicated miR-103a-3p as a prognostic biomarker for preclinical RA. Guo et al. ([@B25]) found that miRNA expression patterns are different in inflamed and noninflamed terminal ileal mucosa of patients with Crohn\'s disease (CD), and dysregulated miRNAs may be responsible for CD pathogenesis. According to current knowledge, immunosuppression relies partly on Tregs and involves in autoimmune disease and cancer, but the serum miRNA profiles of these diseases are less similar ([@B26]--[@B28]). Therefore, we deem that an investigation of serum miRNA profiles in immunodeficient animal models and patients with autoimmune diseases can be an easy and insightful pathway to provide valuable diagnostic and therapeutic approaches in the future.
In this study, we established two animal models of Treg depletion by using CD25 mAb in C57 mice and DT in Foxp3^DTR^ mice. miRNA low density array and quantitative reverse-transcription PCR (qRT-PCR) confirmation were used to characterize the miRNA expression profiles in serum of Treg-depleted mice. ROC curve analysis determined that six miRNAs (miR-551b, miR-448, miR-9, miR-124, miR-148, and miR-34c) could serve as valuable biomarkers for distinguishing Treg-depleted mice from controls. Then, we identified them in the serum from healthy controls, RA, SLE, SS, and UC patients. We found that three miRNAs (miR-448, miR-124, and miR-551b) could serve as novel diagnostic indicators and thereby provide some useful information about the molecular pathogenesis of autoimmune diseases.
Materials and methods {#s2}
=====================
Animals
-------
All animal experiments were performed in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals. Male 6--8-weeks-old C57BL/6J mice were purchased from the Model Animal Research Center of Nanjing University (Nanjing, China). The Foxp3^DTR^ mice were generously provided by Prof Alexander Rudensky (Memorial Sloan-Kettering Cancer Center, New York). The mice were maintained under specific pathogen-free conditions at Nanjing University.
Reagents
--------
TRIzol LS Reagent was purchased from Invitrogen. The mouse Treg staining kit\#1 was purchased from eBioscience. DT from corynebacterium diphtheria was purchased from Sigma-Aldrich. The purified rat anti-mouse CD25 antibody was purchased from BD Pharmingen. A peripheral blood lymphocyte isolation kit was purchased from Tianjin Haoyang Biological Company.
Depletion of tregs
------------------
In C57BL/6J mice, Tregs were transiently depleted by intraperitoneally injecting 0.5 mg purified rat anti-mouse CD25 antibody as we previously described ([@B29]). For Foxp3^DTR^ mice, frozen DT stocks were thawed once and 50 μg/kg of DT was injected intraperitoneally unless otherwise noted. To maximum the efficiency of CD4^+^ CD25^+^ Foxp3^+^ Treg elimination, we conducted injections every day for 7 consecutive days.
Flow cytometric analysis
------------------------
Peripheral blood and spleen were collected and analyzed by FACScalibur for CD4, CD25, and Foxp3 T cell expression as previously described ([@B29]). The results were analyzed by BD FACScalibur device.
Measurement of cytokine levels in serum
---------------------------------------
Whole blood of mice was collected without anticoagulant and centrifugated to obtain serum. The levels of TNF-α, IL-6, and IFN-γ in serum were detected with ELISA kits (R&D) following the instructions as we previously described ([@B29], [@B30]).
miRNA microarray
----------------
A minimum of 0.1 μg of total RNA was added to the GenoExplorer microRNA Expression System (GenoSensor Corporation, Tempe, AZ) containing probes in triplicate for mature miRNAs. miRNA concentrations are presented as threshold cycle (Ct). Significant differentially expressed miRNAs between the groups were analyzed and normalized to internal controls PC-U6B, U6-337, 5S-rRNA, and PC-HU5S recommended by the manufacturer. The relative concentration was calculated by the comparative Ct method (2^−ΔΔCt^). miRNAs were considered upregulated/downregulated if their Ct-values were \<35 in the control samples and their levels in the Treg-depleted samples showed at least a 2-fold increase/decrease compared to the controls.
Patients and healthy controls
-----------------------------
The serum samples were collected according to protocols approved by the Medical Ethics Committee of Nanjing Drum Tower Hospital. All the RA patients are in the active stage of disease and received disease-modifying antirheumatic drugs (DMARDs), such as methotrexate (MTX), Leflunomide (LEF), Hydroxychloroquine (HCQ). Among them, three patients received glucocorticoids (GC) 5--15 mg/days. The Disease Activity Score with 28 joint (DAS28) of RA is 5.42 ± 1.83. In SLE patients, the Systemic Lupus Erythematosus Disease Activity Index (SLEDAI) is 14.28 ± 5.8, all of them have received hormone immunosuppression and the average dosage was 25 mg/days. Healthy controls are in normal physiological conditions and show no sign of pathologic factors after health examination. The ages of healthy individuals are matched with patients. The demographic characteristics of patients and healthy controls are listed in Supplemental Tables [1](#SM1){ref-type="supplementary-material"}--[8](#SM1){ref-type="supplementary-material"}.
Serum RNA isolation and qRT-PCR
-------------------------------
Total RNA of serum was extracted using TRIzol LS Reagent (Invitrogen) following the instructions. qRT-PCR was performed on a LightCycler 480 real time PCR System (Roche, Mannheim, Germany) using TaqMan miRNA probes (Applied Biosystems) according to the instructions as we previously described ([@B16], [@B31], [@B32]).
Statistical analysis
--------------------
All data are representative of at least three independent experiments. All assays were performed in triplicate, and each experiment was repeated several times. Statistical analysis was performed using the *t*-test, when the groups \>2, one way ANOVA followed by Bonferroni\'s multiple comparisons test were used. Data are presented as the means ± SEMs of at least three independent experiments. Differences were considered statistically significant at *P* \< 0.05.
Results {#s3}
=======
Treg depletion by CD25 mAb or DT
--------------------------------
First, C57BL/6 mice and Foxp3^DTR^ mice were injected with CD25 mAb or DT to eliminate CD4^+^ CD25^+^ Foxp3^+^ Tregs. To test the efficiency of Treg depletion, Treg levels were measured in peripheral blood and spleen on day 8 (Figure [1A](#F1){ref-type="fig"}). As shown in Figures [1B](#F1){ref-type="fig"}--[D](#F1){ref-type="fig"}, CD4^+^ T cell levels increased after the CD25 mAb and DT injections. In peripheral blood, CD4^+^ CD25^+^ Foxp3^+^ Treg cells decreased significantly from 7.49 to 0.15% after CD25 mAb injection, and DT injection decreased the Treg levels from 6.59 to 0.06%. The levels of Treg were also reduced in spleen (Supplemental Figure [1A](#SM1){ref-type="supplementary-material"}). In addition, mice depleted of Tregs weighed less than control mice (Supplemental Figures [1B,C](#SM1){ref-type="supplementary-material"}). The levels of inflammatory cytokines TNF-α, IL-6, and IFN-γ were dramatically increased in serum from mice with Treg depletion compared to the mice without Treg depletion (Figure [1E](#F1){ref-type="fig"}), indicating that elimination of CD4^+^ CD25^+^ Foxp3^+^ Tregs is sufficient to disrupt immunological balance. These results suggest that Treg depletion can be used as a representative model of autoimmune diseases.
{#F1}
Microarray analysis of serum miRNAs in treg-depleted mice and qRT-PCR confirmation of changed miRNAs
----------------------------------------------------------------------------------------------------
To identify the markedly changed serum miRNAs, we first analyzed the miRNAs differentially expressed between Treg-depleted and control mice by a TaqMan low density array. Of the 381 miRNAs scanned, 110 demonstrated \>2-fold changes in the CD25 mAb group, 40 were upregulated and 70 were downregulated (Figures [2A,C](#F2){ref-type="fig"}). In the DT group, 254 miRNAs demonstrated \>2-fold changes, 36 were upregulated and 218 were downregulated (Figures [2B,C](#F2){ref-type="fig"}). Among the scanned miRNAs, 2 miRNAs (miR-551b and miR-448) were upregulated in both Treg-depleted groups, while 45 miRNAs were downregulated in both groups (Supplemental Figures [2A](#SM1){ref-type="supplementary-material"}--[C](#SM1){ref-type="supplementary-material"}).
{#F2}
To verify the microarray results, we performed qRT-PCR assay to measure the changed miRNAs in four groups (CTL, CD25 mAb, Foxp3^DTR^-CTL, and Foxp3^DTR^-DT groups, 15 mice/group). The inclusion criteria of changed miRNAs was as follows: mean fold change \>2 and *P*-value \< 0.05 between Treg-depleted groups and control groups. Among the significantly changed miRNAs, we selected six to validate (Supplemental Figure [2](#SM1){ref-type="supplementary-material"}, Figure [2D](#F2){ref-type="fig"}--[I](#F2){ref-type="fig"}). Consequently, we identified that two miRNAs (miR-551b and miR-448) were significantly increased and four miRNAs (miR-9, miR-124, miR-148, and miR-34c) were markedly decreased in serum from Treg-depleted mice.
Diagnostic value of the selected serum miRNAs
---------------------------------------------
Next, we conducted receiver-operating characteristic (ROC) curve analyses to identify the diagnostic usefulness of the 6 miRNAs for Treg-depletion mice models. ROC curve analysis revealed that the six miRNAs (miR-551b, miR-448, miR-9, miR-124, miR-148, and miR-34c) could serve as valuable biomarkers for distinguishing CD25 mAb samples from controls, with the AUC (the area under the ROC curve) values being 0.951, 0.858, 0.916, 0.991, 1.000, and 0.902, respectively (Figures [3A](#F3){ref-type="fig"}--[F](#F3){ref-type="fig"}). Likewise, the ROC curves also indicated that the six miRNAs (miR-551b, miR-448, miR-9, miR-124, miR-148, and miR-34c) could accurately discern DT samples from controls, with the AUCs being 0.964, 0.884, 1.000, 0.991, 0.938, and 0.862, respectively (Figures [3G](#F3){ref-type="fig"}--[L](#F3){ref-type="fig"}). The results suggest that the diagnostic potential of these six miRNAs in distinguishing Treg-depleted mice from controls was high.
{#F3}
Microarray-based go and KEGG analyses revealed the role of the selected serum miRNAs
------------------------------------------------------------------------------------
In order to understand the potential functions of these miRNAs, we conducted bioinformatics analysis. First, Gene Ontology (GO) analysis was performed to identify biological processes associated with the miRNA target genes (*P* \< 0.001, FDR \< 0.05). The high-enrichment GO terms targeted by the six miRNAs included biological regulation, macromolecule biosynthetic process, biosynthetic process and metabolic process (Figures [4A,B](#F4){ref-type="fig"}). KEGG annotation showed that oncogenic pathways (pathways in cancer, chronic myeloid leukemia, and the TNF signaling pathway), immune-associated pathways (T cell and B cell receptor signaling pathways, inflammatory mediator regulation of TRP channels, TGF-beta signaling pathway, cytokine-cytokine receptor interaction, and NF-kappa B signaling pathway), and important proliferative, survival, and apoptosis signaling pathways (MAPK, AMPK, ErbB, Ras, Wnt, mTOR, and p53) were significantly enriched (Figure [4C](#F4){ref-type="fig"}). Most of the pathways have already been reported to take part in immunodeficiency. For example, RAS-MAPK signaling pathway deregulation in T lymphocytes was found to result in a previously unknown primary immunodeficiency disease ([@B33]), mTOR pathway played a crucial part in regulating lymphoproliferation and aberrant differentiation in autoimmune lymphoproliferative syndrome (ALPS) ([@B34]), and the pivotal role of Wnt signaling pathway in T cell development, activation, and differentiation has recently been discovered ([@B35]). These bioinformatics interpretations may provide more evidence that the six miRNAs may have regulatory effects on immunity by affecting signaling pathways.
{#F4}
Separation of patients with autoimmune diseases from controls by miR-448, miR-124, and miR-551b
-----------------------------------------------------------------------------------------------
To further assess the diagnostic value of miRNA signatures in distinguishing patients with autoimmune diseases from controls, we measured the six miRNAs in serum samples comprising 34 healthy controls, 15 RA patients, 27 SLE patients, 15 SS patients, 12 UC patients. Supplemental Tables summarizes the demographic characteristics for the participants. QRT-PCR results indicated that miR-551b and miR-448 were significantly increased in RA, SLE, SS, and UC patients (Figures [5A,B](#F5){ref-type="fig"}), whereas miR-124 levels were decreased in RA, SLE, SS, and UC patients compared to the controls (Figure [5C](#F5){ref-type="fig"}). To further verify the three miRNAs are specific for systemic autoimmune diseases, we measured them in serum samples of inflammatory disease comprising 15 pneumonia patients, 15 HBV hepatitis patients, and 14 sepsis patients (Figures [5A](#F5){ref-type="fig"}--[C](#F5){ref-type="fig"}). The results showed no obvious differences between them and healthy controls, suggesting that the three miRNAs may represent specific biomarkers for distinguishing patients with autoimmune diseases from healthy controls.
{#F5}
Then we performed a ROC curve analysis to evaluate the diagnostic usefulness of the three miRNA in discriminating patients with autoimmune diseases from healthy controls. The ROC curve analysis showed that miR-448, miR-124, and miR-551b could serve as valuable biomarkers for distinguishing patients with autoimmune diseases from healthy controls, with the AUC being 0.91(95% CI 0.85--0.97), 0.9 (95% CI 0.833--0.967), and 0.850 (95% CI 0.769--0.932), respectively (Figures [5D](#F5){ref-type="fig"}--[F](#F5){ref-type="fig"}). Then, we analyzed the predictive accuracy of miRNA signatures: miR-448 showed a specificity of 82.4% and a sensitivity of 91.3%, miR-124 showed a specificity of 76.5% and a sensitivity of 91.3%, and miR-551b showed a specificity of 73.5% and a sensitivity of 88.4%. These results suggest that the diagnostic value of these three miRNAs to distinguish patients with autoimmune diseases from healthy individuals was high.
Discussion {#s4}
==========
Autoimmune diseases involve a complicated immunity disorder, leading to a loss of self-tolerance and following assault on endogenous tissues and cells. So far, there is no universal biomarkers for detecting almost all autoimmune diseases.
Tregs play essential roles in maintaining immune homeostasis and preventing autoimmunity induced by excessive immune activation ([@B10]). Depletion of Tregs in mice is considered a representative animal model of autoimmune disease. In this study, we established Treg depletion mice models in two ways: through CD25 mAb injection in C57BL/6 mice and DT injection in Foxp3^DTR^ mice. Both models showed significantly reduced Treg levels and increased CD4^+^ T cell levels in peripheral blood and spleen. Inflammatory cytokines, such as TNF-α, IL-6, and IFN-γ, were markedly increased in serum from Treg-depleted mice. These phenomena are consistent with the common characteristics of autoimmune diseases, so these Treg-depleted mice can be used as representative models of autoimmune diseases.
In a previous study, we showed that miRNAs are present in serum and plasma of humans and many other animals with stable, reproducible, and consistent in the serum of individuals of the same species ([@B16]). By characterizing serum miRNA expression profiles under normal conditions and in various disease states, we found that serum miRNAs are derived not only from circulating blood cells but also from other tissues directly affected by diseases. Thus, we concluded that serum miRNAs can serve as potential biomarkers for detecting various diseases ([@B16], [@B17], [@B36]). Here, we first investigated the serum miRNA profiles in animal models with Treg depletion. Low Density Array identified miRNAs with significantly different levels in Treg-depleted mice and control mice and revealed that two miRNAs (miR-551b and miR-448) were upregulated and 45 miRNAs were downregulated (\>2-fold change) in both Treg-depleted groups. QRT-PCR further confirmed that miR-551b and miR-448 were significantly increased and four miRNAs (miR-9, miR-124, miR-148, and miR-34c) were significantly decreased in Treg-depleted groups. Then, ROC curve analysis determined that 6 miRNAs (miR-551b, miR-448, miR-9, miR-124, miR-148, and miR-34c) could serve as valuable biomarkers for distinguishing Treg-depleted mice from controls. GO term and KEGG pathway annotation showed that target genes of the six miRNAs were associated with oncogenic, immune-associated, proliferative, survival, apoptosis, and inflammatory signaling pathways, and most of the pathways have already been reported to take part in immunodeficiency. Previous studies have reported that miR-551b is deregulated in CD (coeliac disease, a common autoimmune disorder of the small bowel) patients ([@B37]). Wu et al. ([@B38]) found that miR-448 is deregulated in MS patients and further promotes MS development through induction of the Th17 response. miR-9 has been found to be a putative GA-treatment responsive miRNA biomarker in EAE (experimental autoimmune encephalomyelitis) ([@B39]) and sympathetic ophthalmia ([@B40]). Previous research has demonstrated that miR-124 plays vital roles in regulating autoimmune inflammation ([@B41]--[@B46]). miR-148 might represent prognostic markers for treating autoimmune disorders, such as chronic inflammatory diseases, multiple types of cancer and heart failure in diabetics ([@B47], [@B48]). Besides, miR-34 has been reported to be correlated with RA ([@B49]).
Then, we measured six miRNAs to verify in the serum of RA, SLE, SS, UC patients, non-autoimmune diseases patients and healthy controls. QRT-PCR confirmation and ROC curve analysis determined that miR-448, miR-124, and miR-551b could serve as valuable specific biomarkers for distinguishing patients with autoimmune diseases from healthy controls.
In conclusion, we have defined a serum miRNA profiling in an animal model with autoimmune diseases. Moreover, our findings may provide a potential biomarker for diagnosing autoimmune diseases.
Ethics statement {#s5}
================
All animal care and handling procedures were performed in accordance with the National Institutes of Health\'s Guide for the Care and Use of Laboratory Animals and were approved by the Institutional Review Board of Nanjing University.
Author contributions {#s6}
====================
XC, HZ, and YW conceived and designed the study. FJ, HH, MX, and SZ participated in the experiments and drafted the manuscript. HZ contributed to the sample collection and interpretation the data. All authors read and approved the final manuscript.
Conflict of interest statement
------------------------------
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
We thank Prof. Alexander Rudensky (Memorial Sloan-Kettering Cancer Center, New York) for DT-treated Foxp3^DTR^ mice. This work was supported by the program for the National Natural Science Foundation of China (No. 31741075, 81671608), the Natural Science Foundation of Jiangsu Province (No. BK20140601 and BE2016737), the Chinese Science and Technology Major Project of China (2015ZX09102023-003), Scientific Research Foundation of Graduate School of Nanjing University (2016CL08) and the Jiangsu Young Medical Talents Project (QNRC2016004).
Supplementary material {#s7}
======================
The Supplementary Material for this article can be found online at: <https://www.frontiersin.org/articles/10.3389/fimmu.2018.02381/full#supplementary-material>
######
Click here for additional data file.
[^1]: Edited by: Laurence Morel, University of Florida, United States
[^2]: Reviewed by: Claudio Pignata, Università degli Studi di Napoli Federico II, Italy; Peter Igaz, Semmelweis University, Hungary
[^3]: This article was submitted to Autoimmune and Autoinflammatory Disorders, a section of the journal Frontiers in Immunology
[^4]: †These authors have contributed equally to this work
| {
"pile_set_name": "PubMed Central"
} |
Abbreviations, acronyms & symbols
----------------------------------- ---------------------------------------
BP = Blood pressure
CABG = Coronary artery bypass grafting
CPB = Cardiopulmonary bypass
CTA = Computerized tomography angiography
CVP = Central venous pressure
ECG = Electrocardiogram
ECMO = Extracorporeal membrane oxygenation
ICU = Intensive Care Unit
RCA = Right coronary artery
TTE = Transthoracic echocardiogram
INTRODUCTION
============
Stanford type A aortic dissection is a life-threatening emergency, which usually presents with acute onset of sharp chest pain. The incidence of aortic dissection is 3.5/100,000^\[[@r1]\]^. The most common risk factor is hypertension. Other risk factors include preexisting aortic diseases or aortic valve disease, family history of aortic diseases, history of cardiac surgery, cigarette smoking, direct blunt chest trauma and use of intravenous drugs^\[[@r2]\]^. Rapid diagnosis and treatment of acute type A aortic dissections is critical. Dissections involving the coronary arteries result in higher risk of death and frequently present with nontraditional symptoms. We present a rare case of a 36-year-old woman with a complete occlusion of right coronary artery (RCA) due to a Type A aortic dissection.
CASE REPORT
===========
A 36-year-old woman was admitted to a township hospital following a sharp chest pain and profuse sweating. Her medical history included longstanding arterial hypertension. Twelve-lead electrocardiogram revealed an ST-segment elevation in the II, III and aVF leads. The diagnosis of acute inferior myocardial infarction was considered, and the patient was immediately transferred to a local county hospital. A cardio-pulmonary resuscitation was performed in an ambulance for the sudden onset of cardiac arrest during the transfer. The vital signs recovered and the hemodynamic was stable. On day 6, a magnetic resonance imaging was conducted for the persisting and uncontrolled chest pain and she was diagnosed having a type A aortic dissection. She was transferred to our hospital by an ambulance. On arrival, her heart rate was 110 beats per minute and blood pressure (BP) was 80/45 mmHg. The Computerized tomography angiography (CTA) confirmed a Stanford Type A aortic dissection ([Figure 1](#f1){ref-type="fig"}) and moderate pericardial effusion was also identified. The dissecting space of the aorta was thrombotic from the aortic root to the aortic arch, and the dissection extended from the arch to the level of the iliac artery. The electrocardiogram showed a pathologic Q wave in the III and aVF leads. Therefore, the emergent surgery was arranged for this patient immediately.
Fig. 1Preoperative Computerized tomography angiography showing ascending and descending aortic dissection.
The surgery was performed in median sternotomy. After pericardiotomy, about 150 ml of bloody pericardial effusion was removed. The surface of the whole heart and the dilated ascending aorta were bluish. The right ventricle was motionless. Once a cardiopulmonary bypass (CPB) was established, systemic cooling was initiated. The ascending aorta was clamped just proximal to the brachiocephalic artery and transected above the sinotubular junction. When beginning to perfuse the heart with myocardial protection solution directly through the coronary ostium, we found that the right coronary ostium was totally occluded. Therefore, we opened the right atrium and administered cold blood cardioplegic solution retrogradely through the coronary sinus and antegradely through the left coronary ostium. The heart was arrested completely. After dissecting the RCA, we found RCA was bluish from the trunk to branches. The true lumen couldn't be found when we opened the RCA ([Figure 2](#f2){ref-type="fig"}). We had to give up coronary artery bypass grafting (CABG). When rectal temperature decreased to 28℃，the circulation was arrested. The primary entry tear was found in the lesser curvature of aortic arch. The selective cerebral perfusion was performed through right axillary artery cannulation and left common carotid artery. A self-designed 24 mm self-expandable elephant stent graft was inserted into the true lumen of the descending aorta as the elephant trunk, and the free vascular graft was located inside the native arch. The portion of vascular graft, which covered the orifices of 3 vessels, was wedge-shaped resected. A continuous suture (5-0 Polypropylene) was performed inside the native arch from the lower edge of the left subclavian artery to the two sides and ran through the entire native vessel wall around the orifices of the 3 branches to attach the vascular graft to the native arch. The anastomosis of the proximal of aortic arch and the vascular graft using a 24-mm Dacron graft with an 8-mm side branch was accomplished with a continuous suture. Then the prosthesis was de-aired, we restarted a full CPB through the side branch of the graft, and the patient was gradually rewarmed. The anastomosis of the reinforced ascending aortic stump and the vascular graft was accomplished with a continuous end-end suture. Blood pressure and heartbeat of the patient dramatically dropped during the removal of CPB due to the large area of myocardial infarction in the right heart, although maximal doses of vasopressor and inotropic support (dopamine 20.0µg/kg/minute, epinephrine 0.3µg/kg/minute) were given. The patient was failed to wean from CPB. Peripheral VA-ECMO was implanted with a 22-French Femoral Venous Cannula (Edwards Lifesciences, Irvine, CA, USA) in the right femoral vein and a 24-French Arterial Perfusion Cannula (Edwards Lifesciences) grafted to the right femoral artery via an 8 mm beveled Gelweave™ graft (Terumo, Scotland, UK)\". The ECMO circuit consisted of a centrifugal pump (Revolution 5; Sorin Group, Mirandola, Italy) with a membrane oxygenator (HLS module advanced; Maquet, Hirrlingen,Germany). When the ECMO (Pump Rotation: 2367R/min;Blood Flow: 3.880 L/min) started, we stopped the routine CPB. The patient BP was 89/49 mmHg and central venous pressure (CVP) was 5cm H~2~O with inotropic support (dopamine 20.0µg/kg/minute, epinephrine 0.3µg/kg/minute, nitroglycerin 2.0µg/kg/minute). The chest was closed in the usual fashion. The time of selective cerebral perfusion was 44 minutes, the total circulatory arrest time was 45 minutes, and the CPB time was 239 minutes. When the patient was taken to the ICU, the BP was 90/70mmHg and CVP was 11cm H~2~O. Three hours later, the patient recovered consciousness. The patient was extubated in 18 hours post-operation. We regularly monitored the changes of cardiac function through bedside transthoracic echocardiogram (TTE). The movement of the right ventricular wall gradually improved. On day 13 post-operative, ECMO was removed. The final postoperative TTE revealed that the left ventricular ejection fraction was 55% and the right ventricular wall moved normally. The patient recovered uneventfully.
Fig. 2An intraoperative photograph demonstrating that a complete thrombosis of the right coronary artery turned bluish.
The patient had scheduled a follow-up three months after discharge. The CTA demonstrated that the RCA was totally occluded, except a small caliber true lumen visualized ([Figure 3](#f3){ref-type="fig"}).
Fig. 3A follow-up CTA demonstrating that the right coronary artery was undeveloped and only a short period of residual cavity remained.
DISCUSSION
==========
A coronary involvement in type A aortic dissection is a serious phenomenon in clinic, which presents in 10-15% patients with aortic dissection^\[[@r2]\]^. Clinical research has shown that the mortality rate of a type A aortic dissection involving the coronary artery is high (up to 20% \~ 33.3%) after surgery^\[[@r3],[@r4]\]^. According to Neri et al.^\[[@r5]\]^, there are three mechanisms of total occlusion of right coronary ostium: type A, where ostial dissection is defined as a disruption of the inner layer limited to the area of the coronary ostium; type B, a dissection extending into the coronary artery; and type C, a coronary disruption (intimal detachment). The ideal treatment for this condition is surgery with ascending aorta replacement and CABG or a coronary repair, if possible. In our patient, acute Type A aortic dissection was accompanied with myocardial infarction and pericardial effusion. Without prompt treatment, it will result in death due to myocardial infarction or aortic rupture. The RCA dissection was already suspected based on the presence of pathologic Q on ECG on the patient's arrival. But in the surgery, we found RCA was bluish from the trunk to branches, and the RCA was totally thrombosed. The true lumen couldn't be found when we opened the RCA. There was no place available for CABG on the RCA. The follow-up CTA also demonstrated that the RCA was totally occluded, except for a small caliber true lumen visualized. After a regular surgery of Stanford type A aortic dissection, the patient was difficult to wean from CPB due to the right heart dysfunction. In this case, we excluded consideration of an intra-aortic balloon pump because of the residual dissected thoracoabdominal aorta. The only option left was to continue to use assisted circulation which helped myocardial recovery. The right ventricular circulatory support was another option, but was not approved for use in China. Besides, we had more experience on installing ECMO on patients with right ventricular failure and got satisfied outcomes. So, we chose ECMO to assist right heart function. As expected, the hemodynamic parameters were favorable after initiation of ECMO, and the right ventricular wall motion was gradually improved during the post-operation period.
This is the first report of using ECMO to successfully treat a complete occlusion of the RCA, when CABG can't be performed due to a Stanford Type A aortic dissection. This case demonstrates the value of ECMO in assisting right heart function to ensure stable hemodynamics and myocardial recovery in the type A aortic dissection with coronary involvement.
Author\'s roles & responsibilities
------------------------------------ ------------------------------------------------------------------------------------------------------------------------
YW Collected the date and wrote the article; final approval of the version to be published
ZZ Referred to the related literature; final approval of the version to be published
RX Referred to the related literature; final approval of the version to be published
DL Referred to the related literature; final approval of the version to be published
TW Referred to the related literature; final approval of the version to be published
KL Designed the study; performed the operation; and revised the manuscript; final approval of the version to be published
This study was carried out at the Second Hospital of Jilin University, Changchun, China.
No financial support.
No conflict of interest.
We should like to thank the radiology department for providing the images shown.
| {
"pile_set_name": "PubMed Central"
} |
Background
==========
Early debut of sexual activity is associated with greater risk of HIV and other sexually transmitted diseases (STDs) as well as unwanted pregnancy among teens in the United States. While the prevalence of sexual intercourse and pregnancy among teens has declined significantly in the United States since the early 1990s \[[@B1],[@B2]\], more recent data suggest that rates of teen pregnancy may be on the rise again \[[@B2]\]. Furthermore, the social and medical costs associated with STDs and teen pregnancy in the United States are among the highest in all developed countries \[[@B3]-[@B5]\]. To date, interventions for curbing teen pregnancy and the spread of STDs among teens have focused on educational programs for youth, many of which include sexual abstinence curricula.
While there is debate about the effectiveness of youth-focused abstinence education programs, prior research overwhelmingly supports the notion that parents have the ability to influence their children\'s decisions regarding sexual behavior, use of contraceptives, and disease prevention. Yet parent-based approaches to curbing teen pregnancy and STDs have been relatively unexplored. A number of studies show that parent-child communication about sex is related to delayed onset of sexual intercourse \[[@B6]-[@B12]\]; increased contraceptive use in daughters \[[@B12]-[@B14]\]; and increased disease prevention behaviors, including condom use and fewer sexual partners \[[@B6],[@B11],[@B15]\].
Given the lack of parent-based approaches to delaying sexual intercourse among adolescents, the U.S. Department of Health and Human Services launched the Parents Speak Up National Campaign (PSUNC). PSUNC is a multimedia social marketing campaign targeted to parents and is aimed at promoting parent-child communication about sex. The campaign is predicated on the body of research noted above that describes the influence of parental communication on adolescents\' decisions regarding sexual behavior. Launched nationally in June 2007, the campaign targets parents of 10- to 14-year-old children. The campaign primarily uses televised public service announcements (PSAs) supplemented by radio, print, and outdoor advertising. PSUNC television advertisements feature age-appropriate youth letting their parents know that they want to talk to them about sex and that they should talk \"early and often.\" The campaign PSAs are also designed for multiple racial/ethnic audiences, including ads targeted to a general audience and ads for African American, Hispanic, and Native American parents. The advertisements also promote the campaign\'s Web site, <http://www.4parents.gov>, which provides age-appropriate information and guidance to parents about talking to their children about sex.
The campaign is grounded in social cognitive theory, which predicts a chain of cognitive events that lead to behavioral outcomes and choices \[[@B16],[@B17]\]. The PSUNC conceptual model, published elsewhere \[[@B18]\], specifically posits that increased parent-child communication will result from the promotion of effective communication behaviors and expectations about the impact of those behaviors. As such, the campaign\'s messages directly promote parents\' self-efficacy and outcome efficacy to communicate with their children as well as normative beliefs about the age at which children should wait to have sex. These cognitive elements are purposefully embedded in specific messages within the campaign ads and include (1) parents\' self-efficacy that they can effectively talk with their children about sex, (2) parents\' outcome-efficacy that talking with their children will translate into reduced or delayed sexual activity and long-term social and health benefits for their children, and (3) social norms regarding the age to which children should wait to have sex. The PSUNC conceptual model hypothesizes that these are important cognitive factors that the campaign will impact to drive change in downstream parent-child communication behaviors.
Understanding these cognitive precursors provides information that can be used to refine and improve messaging for PSUNC specifically and for social marketing campaigns more generally. While many studies have examined the role of social cognitive factors in behavior change, this paper joins the small but important set of studies that have examined the cognitive factors that precede parent-child communication specifically. In a study of primarily African American father-son dyads, for example, both self-efficacy and outcome expectation played a role in increasing communication related to sexual behavior \[[@B19]\]. Self-efficacy was also found to be a significant predictor of communication around sexual behavior in mother-daughter dyads \[[@B20]\] and to enhance parent-child communication related to sexual abuse \[[@B21]\].
A preliminary evaluation study of PSUNC was conducted using a randomized controlled trial (RCT) \[[@B22]\]. The RCT design was chosen based on the PSUNC implementation strategy and inherent limitations of that strategy. As noted above, the campaign was implemented with PSAs that aired nationwide. PSAs, however, are predominantly aired during slots of advertising time that are donated by the television and radio stations airing them. These advertising time slots most often occur during late hours or programs that are not heavily viewed, making them likely choices for advertising donations to public service campaigns. Because PSAs most often appear during donated advertising times with sparse audiences, overall exposure to PSAs is very low. Therefore, using telephone surveys or other field-based data collection methods to measure parents\' awareness of PSUNC was not feasible. Such surveys likely would not capture measurable levels of parent exposure to the campaign, leaving little basis on which to make comparisons of parent-child communication outcomes between those exposed and those not exposed to the campaign. As such, an RCT was chosen to explicitly control parent exposure to PSUNC and ensure sufficient statistical power to assess the impact of campaign exposure on parent-child communication outcomes.
Preliminary results from the RCT found that PSUNC was efficacious in increasing parents\' initiation of conversations about sex with their children and in promoting use of the campaign Web site \[[@B22]\]. However, this study does not address how these effects manifest themselves through the PSUNC theoretical model described earlier. For example, is there evidence that the campaign generated this impact through the targeting of specific cognitive factors identified in the theoretical model that were hypothesized to influence parent-child communication? Fuller tests of this theoretical model for PSUNC have yet to be conducted.
Understanding how PSUNC influences the cognitive precursors identified in the campaign\'s theoretical model can help inform future parent-child communication campaign development in several ways. First, it is useful to determine whether the specific cognitive factors identified in the campaign\'s theoretical model are indeed associated with parent-child communication. This association speaks to the validity of the campaign strategy as outlined in the theoretical model (i.e., does the campaign target factors that are both sensitive to exposure to PSUNC messages and in fact likely to influence parent-child communication?). Second, it is useful to consider which cognitive precursors to parent-child communication are most influenced by the campaign. This facilitates further refinement of campaign messages to target outcomes that are most sensitive to campaign exposure.
The present study builds on the prior RCT for PSUNC to more specifically examine the relationship between parents\' exposure to PSUNC messages and cognitive precursors that are targeted by the campaign and theorized to influence parent-child communication. Specifically, we use the RCT data to (1) test whether the cognitive precursors targeted by the campaign are predictive of parent-child communication, (2) examine the impact of exposure to PSUNC messages on those cognitive precursors, and (3) explore possible effects of heavier exposure to PSUNC (i.e., receiving more PSUNC messages) on each of those cognitive outcomes. We then discuss implications of these findings both for future PSUNC implementation specifically and more generally for overall development of parent-based social marketing campaigns that use theoretical models.
Methods
=======
Data and Experiment Design
--------------------------
Our data come from the PSUNC Parent Efficacy Study, a randomized controlled experiment conducted with parents of 10- to 14-year-olds from the Knowledge Networks (KN) online panel. Established in 1999, the KN panel is the only online research panel that is based on probability sampling. The KN panel is recruited using random-digit-dial telephone surveys and is weighted to match U.S. Census demographic benchmarks. Individuals who do not have a computer and access to the Internet are provided MSN television service and free monthly Internet access. This allows coverage of both online and offline households in the United States.
For this study, we first identified all adult KN panelists living with at least one child between the ages of 10 and 14 years (N = 3,217). Mothers and fathers were sampled separately, and parents who did not participate were not replaced by a parent of the opposite gender. The parent and one child in the eligible age range were paired, and all study surveys referenced that one child after the parent consented to participate. A total of 2,439 parents (75.8%) responded to the study invitation and were eligible to participate. Among these, 1,969 parents (1,125 mothers and 844 fathers) completed the baseline survey in the fall of 2007. Only one parent per household participated in the study. Study sample sizes were determined by power analyses conducted prior to the study. Additional details on study recruitment and eligibility are provided elsewhere \[[@B22]\].
Parents were randomly assigned to treatment and control conditions, where treatment consisted of exposure to PSUNC advertisements and print materials and control consisted of no exposure to PSUNC messages. Random assignment was carried out by a standard randomization algorithm in the Knowledge Networks online sampling system that gave all participants equal chances of being placed in any of the experiment conditions. Because this study uses a self-administered survey, participants were not aware of the other experiment conditions to which they were not assigned. Mothers and fathers were randomized into their respective experiment conditions separately. All participants were surveyed at five points in time: (1) baseline, prior to message exposure, (2) 4 weeks post-baseline, (3) 6 months post-baseline, (4) 12 months post-baseline, and (5) 18 months post-baseline. Mothers were further randomized into treatment and booster (additional messages) conditions at 4 weeks post-baseline to assess the effects of additional PSUNC messages. A larger baseline sample of mothers (N = 1,125) was collected to accommodate this additional randomization of treatment condition mothers. We included the booster condition for mothers only based on two factors. First, it is well-known that mothers more often engage in sexual communication with their children compared to fathers and the PSUNC ads reflect this in their mother-centric messages. Second, study resources were limited and did not permit booster conditions for both mothers and fathers. Parents who were assigned to treatment conditions received exposure to PSUNC messages via online multimedia immediately following the baseline survey and prior to each of the 4-week and 6-month follow-up surveys.
Each parent that was assigned to an exposure condition read, viewed, and listened to a collection of multimedia stimuli from PSUNC. All materials were viewed online during each survey session. This included one 60-second television PSA called \"Talk to Me\" that shows adolescent children asking their parents to talk to them about waiting to have sex. The multimedia package also included one 60-second radio PSA and two print PSAs. Preceding the 6-week follow-up survey, mothers who were assigned to the booster condition reviewed the same set of multimedia materials plus two additional print PSAs, one additional 60-second radio PSA, and one additional 60-second television PSA. The additional television PSA, called \"Muffinhead,\" showed adolescent children telling their parents that they will still remain a close family if they talk to them about waiting to have sex. In addition to viewing the ads online within the survey, all treatment condition participants were mailed a DVD containing all media materials for their specific condition and were asked to view them between survey sessions.
The study questionnaire, consent procedures, and human subjects protection protocols were reviewed and approved by the sanctioned Institutional Review Board of RTI International. This study was also reviewed and approved by the Federal Office of Management and Budget prior to recruitment of the first subject (OMB control \#0990-0311).
Measures
--------
Parents completed a 64-item survey at each study time point. The survey was self-administered online by each parent participant from the KN panel. The survey consisted of questions on sociodemographic characteristics; knowledge, attitudes, and beliefs about parent-child communication; parent-child communication social norms and expectations; and media habits. Our analysis focuses on the impact of exposure to PSUNC messages on four primary measures of theorized cognitions related to parent-child communication: (1) social norms on waiting until older to have sex, (2) parent efficacy to talk to their child about sex, (3) short-term expectations about their child\'s response to parent communication about sex, and (4) long-term expectations about the impact of parent-child communication on their child\'s future success in life. Each outcome was measured as a multi-item summative scale or two-item index (Table [1](#T1){ref-type="table"}). We also examined how each of these factors is related to parent-child communication, the primary behavioral outcome variable studied in Evans et al. \[[@B22]\]. Each of these measures is described in more detail below.
######
Efficacy Study Outcome Variables
Scale/Index Items Response Categories
--------------------------------------------------------------------------------------------------------------------------------------------------------- ------------------------------------------------------------------------
**Wait Until Older Social Norm Index**
To what age do you think boys should wait before being sexually active? 1 (until 14), 2 (until 16), 3 (until 18+), 4 (until married)
To what age do you think girls should wait before being sexually active? 1 (until 14), 2 (until 16), 3 (until 18+), 4 (until married)
**Efficacy to Practice Parent-Child Communication**
How sure are you that you can always explain to your child why s/he should wait to be sexually active? 1 (not sure at all or very unsure) to 3 (completely sure or very sure)
How sure are you that you can always explain to your child how to make a boy/girl wait until ready to be sexually active? 1 (not sure at all or very unsure) to 3 (completely sure or very sure)
How sure are you that you can always explain to your child how to tell a boy/girl no if your child does not want to be sexually active? 1 (not sure at all or very unsure) to 3 (completely sure or very sure)
How sure are you that you can always explain to your child ways to have fun with a boy/girl without being sexually active? 1 (not sure at all or very unsure) to 3 (completely sure or very sure)
**Short-term Outcome Expectation Scale**
If you talk early and often with your child about sex, your child will\...
Be less likely to be sexually active as a young teen 1 (strongly disagree) to 4 (strongly agree)
Not think you are judgmental 1 (strongly disagree) to 4 (strongly agree)
Understand the benefits of waiting to have sex 1 (strongly disagree) to 4 (strongly agree)
Not listen to what you say 1 (strongly agree) to 4 (strongly disagree)
Think you are a hypocrite 1 (strongly agree) to 4 (strongly disagree)
Rebel and want to engage in sexual activity even more 1 (strongly agree) to 4 (strongly disagree)
**Long-term Outcome Expectation Index**
By effectively talking with your child about delaying sexual activity, you will be able to positively impact your child\'s future success and happiness 1 (strongly disagree) to 4 (strongly agree)
If you can convince your child to wait to have sex, s/he will have a better chance to succeed as an adult 1 (strongly disagree) to 4 (strongly agree)
### Social norms for delay of sexual initiation
This measure included two items that asked participants their views regarding how long a child should wait before becoming sexually active; one item asked about boys, and the other asked about girls. Response options were \"until they are 12,\" \"until they are 14,\" \"until they are 16,\" \"until they are 18,\" \"until they are 21,\" and \"until they are married.\" Principal factor analyses indicated that both items loaded strongly onto a single factor (Eigen value = 1.69) with high reliability (alpha = 0.94).
### Efficacy to talk to child about sex
Four items were used to assess participants\' belief in their ability to communicate with their child about sexual activity. Each item begins with the stem, \"How sure are you that you can always explain to your child\...\" Seven category response options ranged from \"completely sure\" to \"not at all sure\" for each item. These items loaded strongly into a single factor (Eigen value = 1.81) with good reliability (alpha = 0.78).
### Short-term outcome expectations
Six items ask participants to consider the immediate impacts of parent-child communications about sex. Each item begins with the stem, \"If you talked early and often with your child about sex, your child will\...\" and concludes with statements about the child\'s behavior (e.g., be less likely to be sexually active as a teen) and the child\'s perception of the parent (e.g., think you are a hypocrite). Four category response options ranged from \"strongly agree\" to \"strongly disagree\" with no neutral option. Principal factor analyses showed these items loaded into a single factor (Eigen value = 1.60) with acceptable reliability (alpha = 0.70).
### Long-term outcome expectations
Two items ask participants to consider whether delaying sexual initiation would have a positive impact on their children\'s future. Four category response options ranged from \"strongly agree\" to \"strongly disagree\" with no neutral option. These two items loaded into a single factor (Eigen value = 0.83) with adequate reliability (alpha = 0.69).
For each scale/index described in Table [1](#T1){ref-type="table"} we constructed relative change scores for each of the baseline to follow-up time periods (4 weeks, 6 months, 12 months, and 18 months). These were calculated by subtracting the baseline value of the outcome from the follow-up value and then dividing by the baseline value. The use of relative change scores helps account for the individual\'s baseline level for each outcome and potential correlation between the baseline and follow-up values \[[@B23]\].
### Parent-child communication
To examine the relationship between each of the cognitive variables described above and parent-child communication behavior, we focus on actual parent recommendations to the child to wait to have sex, which was found to be significantly impacted by exposure to PSUNC messages in Evans et al. \[[@B22]\]. This measure is derived from the survey item that asks parents \"Have you asked (recommended) that \[child name\] wait to have sex?\" We created a dichotomous indicator for whether the parent answered \"yes\" to this question.
### Control Variables
The study survey also included measures for a number of parent characteristics and other potential correlates of parent-child communication that we controlled for in our statistical analyses. These included parent marital status; educational attainment; race/ethnicity; age; full-time employment status; family structure (one or two parents in home); metropolitan statistical area (MSA) category (urban or rural); child gender; child\'s access to television, Internet, and other media in their bedroom; and an 8-item scale for parent involvement. The parent involvement scale includes items drawn from previous research \[[@B24],[@B25]\] that measure joint parent-child activities and the frequency of those activities. The specific items in the parent involvement scale measure past month frequency (\"never,\" \"less often,\" \"at least once a month,\" or \"at least once a week\") of (1) shopping, (2) going to the movies or sporting events, (3) watching television, (4) attending religious services, (5) doing homework, (6) attending a party, (7) volunteering, and (8) playing a game or a sport. Prior research shows that these items load into a single scale that has good reliability \[[@B22]\].
We also created a measure to account for potential contamination of the control condition. Although it was expected that natural exposure to PSUNC would be very low due to the limited reach of PSAs, there is still a small chance that participants in the control condition may be exposed to the campaign as it aired in the real world. The study instrument thus included a specific question on general awareness of PSUNC messages that asked \"Have you ever seen or heard ads on television or radio with the Parents Speak Up National Campaign theme or slogan?\" Participants could answer \"yes\" or \"no\" to this question. This variable was analyzed to assess the extent of possible control condition contamination and was included in our multivariable analyses to account for potential contamination.
Statistical Analysis
--------------------
### Scale Reliability of Cognitive Precursors
We conducted principle factor analysis to assess reliability of each of the four cognitive outcomes examined in this study. The factor analysis was performed using the principal factor method in Stata statistical software. Scale and index reliability were then estimated using Cronbach\'s alpha coefficient for the scale, a measure of the internal consistency of the scale determined by the average inter-item correlation and the number of items in the scale \[[@B26],[@B27]\]. We estimated Cronbach\'s alpha overall and separately for mothers and fathers in the study. Principal factor analyses indicated that the items within each of the four primary cognitive outcome scales/indices that we examined load into a single scale or index with factor loadings ranging from 0.38 to 0.92 for each item and Cronbach\'s alphas ranging from 0.69 to 0.94 for each scale/index.
### Relationship between Cognitive Precursors and Parent-Child Communication
To test whether the cognitive factors targeted by the campaign are predictive of parent-child communication, we estimated a multivariable logistic regression that links the cognitive variables to future parent-child communication in a longitudinal framework. This analysis helps to assess the overall validity of the campaign strategy as outlined in the theoretical model (i.e., does the campaign target cognitive precursors that are in fact likely to influence parent-child communication?). Specifically, we estimated the odds of parent recommendation to wait to have sex at the 18-month follow-up as a function of each of the four cognitive variables at baseline. This model included a control variable for whether the parent had already made recommendations to wait at baseline. We also included control variables for whether the parent was in the study treatment or control conditions as well as baseline covariates for each of the individual control variables described earlier.
### Impact of PSUNC on Cognitive Precursors
We used multivariable linear regressions to estimate the relationship between each time-specific change score and parent exposure to treatment conditions. We estimated separate models for each treatment effect at 4 weeks, 6 months, 12 months, and 18 months post-baseline. All control variables described earlier were included in each model. Because so little is known about father-child communication about sexual activity, we also estimated each model separately for mothers and fathers. All models were estimated using Stata Version 9 (College Station, Texas).
Results
=======
Study sample sizes, by experiment condition and survey wave, are summarized in Table [2](#T2){ref-type="table"}. A total of 1,969 parents (1,125 mothers and 844 fathers) completed the baseline survey. Between the baseline and 18-month follow up survey, overall attrition was 49.9%. Analysis of attrition by experiment condition suggests that both mothers and fathers in the treatment conditions were slightly more likely to drop out of the study after 18 months. Among mothers, 47.3% of control condition respondents dropped out after 18 months, whereas 53.1% of treatment condition respondents dropped out. Among fathers, 18-month attrition rates were 42.1% within the control condition and 52.4% within the treatment condition. We further examined attrition by a number of demographic characteristics and found that 18-month attrition rates were not significantly different by parent race/ethnicity, education, or employment status. We further compared all study outcome variables as well as all baseline control variables that were included in our multivariable models by attrition. There were no statistically significant differences in any of the study outcome variables or baseline control variables between those who completed all 5 waves of the study and those who dropped out before completing all waves.
######
Efficacy Experiment Sample Sizes
Survey Sample Sizes
-------------------- --------------------- ------- ------- ------- -----
Mothers
Control 349 326 270 233 184
Normal Treatment 776 663 266 219 175
Booster Treatment \-- \-- 275 220 189
Fathers
Control 340 321 280 230 197
Normal Treatment 504 444 365 297 240
Total 1,969 1,754 1,456 1,199 985
*Note*: Normal treatment = Exposed to core PSUNC messages; Booster treatment = Exposed to core plus additional PSUNC messages.
Baseline distributions of each outcome variable are shown in Table [3](#T3){ref-type="table"}. With the exception of the short-term expectation scale, the sample is generally clustered in the upper range of each outcome variable. However, significant proportions of both the mother and father samples are below outcome ceilings at baseline, indicating significant potential for change over time. Our use of relative change scores in our multivariable models allows for the inclusion of those already at outcome ceilings at baseline, capturing the treatment condition\'s protective effects in preventing relapse from the outcome ceilings.
######
Baseline Outcome Variable Distributions
Outcome Variable Mothers (N = 1,125) Fathers (N = 844)
----------------------------------------------------- --------------------- -------------------
**Wait Until Older Social Norm Index** \% \%
2 (minimum) 0.00 0.26
3 0.00 0.00
4 1.22 3.17
5 1.94 2.25
6 47.09 50.33
7 3.36 5.55
8 (maximum) 46.38 38.44
**Efficacy to Practice Parent-Child Communication**
6 (minimum) 0.10 0.40
7 0.20 0.13
8 6.52 9.96
9 7.64 10.36
10 12.83 15.14
11 15.89 15.54
12 (maximum) 56.82 48.47
**Short-Term Outcome Expectation Scale**
9 (minimum) 0.10 0.00
10 0.00 0.00
11 0.10 0.00
12 0.10 0.54
13 0.31 0.54
14 0.92 0.54
15 1.74 2.95
16 3.90 4.26
17 8.51 10.98
18 21.44 25.57
19 15.28 15.80
20 12.51 12.85
21 10.15 7.90
22 8.00 6.29
23 6.46 5.09
24 (maximum) 10.46 6.69
**Long-Term Outcome Expectation Index**
2 (minimum) 0.61 0.93
3 0.10 0.26
4 1.84 2.12
5 4.39 5.83
6 21.63 26.23
7 19.49 21.19
8 (maximum) 51.94 43.44
Parent sociodemographic characteristics are summarized in Table [4](#T4){ref-type="table"}. Most parents were between the ages of 33 and 55, with a mean age of 43. The sample was predominantly white with low sample sizes of African American and Hispanic parents, relative to the U.S. population. The sample also contained a higher rate of college-educated parents compared with the United States as a whole. Child gender, child age, and parent employment status statistics resembled the U.S. population.
######
Unweighted Sample Demographics of Efficacy Study Participants Who Completed All Five Survey Waves
TOTAL (*N*= 985) Mothers (*N*= 548) Fathers (*N*= 437)
----------------------------------- ------------------------ --------------------------------- ---------------------------------- ------------------------ --------------------------
**Baseline Demographic Variable** **Control (*N*= 184)** **Normal Treatment (*N*= 175)** **Booster Treatment (*N*= 189)** **Control (*N*= 197)** **Treatment (*N*= 240)**
Average parent age 43 42.8 41.6 44.9 44.7
Average child age 12.2 12.2 12.0 12.2 12.3
Parent education
Less than high school 1.1% 1.7% 1.6% 3.1% 1.7%
High school graduate 14.7% 13.7% 10.1% 15.2% 13.3%
Some college 36.4% 40.6% 42.3% 30.0% 34.2%
Bachelors degree+ 47.8% 44.0% 46.0% 51.8% 50.8%
Race/ethnicity
White 86.4% 84.6% 85.7% 88.3% 90.0%
African American 7.1% 9.7% 7.9% 1.0% 3.8%
Hispanic 3.3% 4.0% 3.7% 3.1% 2.9%
Other 3.3% 1.7% 2.7% 7.6% 3.3%
Child gender
Male 47.3% 49.1% 50.8% 60.9% 54.2%
Female 52.7% 50.9% 49.2% 39.1% 45.8%
Employment status
Full-time 50.3% 44.8% 50.8% 85.6% 88.3%
Part-time 24.0% 24.4% 25.7% 3.1% 2.9%
Not employed 25.7% 30.8% 23.5% 11.3% 8.8%
*Note*: Five survey waves = baseline and 4 weeks, 6 months, 12 months, and 18 months post-baseline. Normal treatment = Exposed to core PSUNC messages; Booster treatment = Exposed to core plus additional PSUNC messages.
Multivariate logistic regressions of the relationship between campaign-targeted cognitive precursors at baseline and parent-child communication at 18 months post-baseline are shown in Table [5](#T5){ref-type="table"}. Baseline norms favoring waiting until older to have sex were associated with parent recommendations to wait to have sex 18 months post-baseline among both mothers (OR = 1.21, p \< 0.016) and fathers (OR = 1.15, p \< 0.033). Expectations about long-term outcomes from parent-child communication were also associated with greater odds of parent recommendations to wait to have sex among both mothers (OR = 1.16, p \< 0.017) and fathers (OR = 1.41, p \< 0.001). Self efficacy to practice parent-child communication was not associated with follow-up parent recommendation to wait to have sex for either mothers or fathers. Short-term outcome expectations were only associated with parent-child communication at follow-up among fathers (OR = 1.13, p \< 0.002).
######
Multivariable Logistic Regression Showing Odds of Parent Recommendation to Wait to Have Sex at 18 Months Post-Baseline as a Function of PSUNC Cognitive Precursors at Baseline \[95% Confidence Interval\] (p-value)
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Outcome Variable: Parent Recommendation to Wait to Have Sex, 18 Months Post-Baseline
----------------------------------------------------------- -------------------------------------------------------------------------------------- ----------------------------
**Baseline Independent Variables (Cognitive Precursors)** **Mothers** **Fathers**
**Wait Until Older Norm Index** **1.21**\ **1.15**\
**\[1.04, 1.41\] (0.016)** **\[1.01, 1.31\] (0.033)**
**Self-Efficacy Scale** 0.99\ 0.92\
\[0.88, 1.11\] (0.804) \[0.82, 1.03\] (0.134)
**Short-term Expectation Scale** 1.04\ **1.13**\
\[0.97, 1.11\] (0.323) **\[1.04, 1.22\] (0.002)**
**Long-term Expectation Index** **1.16**\ **1.41**\
**\[1.03, 1.30\] (0.017)** **\[1.23, 1.61\] (0.001)**
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
*Notes:*Model controls for the following variables measured at baseline: parent recommendation to wait to have sex (at baseline); child gender; parent marital status; highest educational attainment; race/ethnicity; parent age; employment status; family structure; whether child has computer, cable television, or Internet in his or her bedroom; treatment condition; metropolitan statistical area urban status; and parental involvement. Bold odds ratios are statistically significant at p \< 0.05.
Results from the multivariate linear regression models of the relationship between exposure to PSUNC messages and these cognitive variables are summarized in Table [6](#T6){ref-type="table"}. Among mothers, we found significant treatment effects at 4 weeks (b = 0.030, p \< 0.001), 6 months (b = 0.037, p \< 0.001), and 12 months (b = 0.028, p \< 0.021) post-baseline for increased norms favoring waiting until older to have sex. For this outcome, we also found that mothers exposed to additional PSUNC messages in the booster condition exhibited greater change 6 months post-baseline in norms favoring waiting until older to have sex compared to mothers in the control condition (b = 0.027, p \< 0.006). However, there was not a significant booster effect for this outcome relative to mothers in the normal treatment condition, suggesting that normal treatment (without additional PSUNC messages) was just as impactful as booster treatment exposure at 6 months post-baseline. No treatment effects for norms regarding the age to which parents\' children should wait to have sex were observed at 18 months post-baseline.
######
Multivariable Least Squares Regressions Showing Coefficients for Association between Changes in Parent-Child Communication Cognitions and Exposure to PSUNC \[95% Confidence Interval\] (p-value)
-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Mothers Fathers
---------------------------------- ---------------------------- ----------------------------- ---------------------------- ------------------------- ------------------------- ------------------------- ------------------------- ------------------------------
**Outcomes** **4-Week Follow-up** **6-Month Follow-up** **12-Month Follow-up** **18-Month Follow-up** **4-Week Follow-up** **6-Month Follow-up** **12-Month Follow-up** **18-Month Follow-up**
**Wait Until Older Norm Index**
Normal Treatment **0.030**\ **0.037**\ **0.028**\ 0.009\ 0.018\ 0.007\ -0.006\ 0.003\
**\[0.02, 0.04\] (0.001)** **\[0.02, 0.06\] (0.001)** **\[0.01, 0.05\] (0.021)** \[-0.02, 0.04\] (0.487) \[-0.03, 0.04\] (0.088) \[-0.02, 0.03\] (0.592) \[-0.04, 0.02\] (0.685) \[-0.03, 0.04\] (0.846)
Booster Treatment \-- **0.027**\ 0.004\ 0.021\ \-- \-- \-- \--
**\[0.01, 0.05\] (0.006)** \[-0.02, 0.03\] (0.717) \[-0.01, 0.05\] (0.122)
**Self Efficacy Scale**
Normal Treatment 0.002\ **0.020**\ 0.013\ -0.003\ 0.015\ 0.010\ -0.002\ -0.006\
\[-0.01, 0.02\] (0.811) **\[0.01, 0.04\] (0.038)** \[-0.01, 0.04\] (0.322) \[-0.03, 0.03\] (0.817) \[-.01, 0.04\] (0.121) \[-0.01, 0.03\] (0.443) \[-0.03, 0.03\] (0.895) \[-0.04, 0.02\] (0.687)
Booster Treatment \-- 0.004\ -0.001\ 0.005\ \-- \-- \-- \--
\[-0.02, 0.02\] (0.686) \[-0.03, 0.03\] (0.937) \[-0.02, 0.03\] (0.736)
**Short-term Expectation Scale**
Normal Treatment 0.006\ -0.017\ 0.008\ -0.013\ 0.006\ -0.006\ -0.024\ \-**0.033**\
\[-0.01, 0.02\] (0.394) \[-0.04, 0.01\] (0.084) \[-0.01, 0.03\] (0.441) \[-0.04, 0.01 (0.300) \[-0.01, 0.02\] (0.456) \[-0.02, 0.01\] (0.478) \[-0.05, 0.00\] (0.060) **\[-0.06, -0.01\] (0.008)**
Booster Treatment \-- **-0.021**\ -0.009\ -0.016\ \-- \-- \-- \--
**\[-0.04, 0.00\] (0.037)** \[-0.03, 0.01\] (0.396) \[-0.04, 0.01\] (0.196)
**Long-term Expectation Index**
Normal Treatment **0.034**\ **0.033**\ **0.036**\ 0.004\ -0.002\ -0.012\ 0.023\ -0.020\
**\[0.01, 0.06\] (0.004)** **\[0.01, 0.06\] (0.046)** **\[0.01, 0.07\] (0.043)** \[-0.04, 0.04\]\ \[-0.03, 0.02\] (0.861) \[-0.04, 0.02\] (0.440) \[-0.01, 0.06\] (0.204) \[-0.06, 0.02\] (0.310)
(0.848)
Booster Treatment \-- **0.032**\ **0.042**\ 0.006\ \-- \-- \-- \--
**\[0.01, 0.06\] (0.044)** **\[0.01, 0.08\] (0.024)** \[-0.04, 0.05\] (0.815)
-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
*Notes*: All models control for child gender; parent marital status; highest educational attainment; race/ethnicity; parent age; employment status; family structure; whether child has computer, cable television, or Internet in his or her bedroom; metropolitan statistical area urban status; parental involvement, and an indicator variable for prior exposure to PSUNC. Normal treatment = Exposed to core PSUNC messages; Booster treatment = Exposed to core plus additional PSUNC messages. Bold coefficients are statistically significant at p \< 0.05.
We found evidence of a short-term normal treatment effect at 6 months post-baseline among mothers for increased parent efficacy to practice parent-child communication (b = 0.020, p \< 0.038). No other treatment effects were observed for this outcome. Results further indicated a negative booster condition effect on short-term outcome expectations among mothers 6 months post-baseline. Booster treatment was associated with decreased expectations about short-term outcomes from parent-child communication at 6 months post-baseline (b = -0.021, p \< 0.037) relative to mothers in the control condition. Exposure to PSUNC messages was not associated with either of these outcomes at 12 or 18 months post-baseline.
Among mothers, we found significant treatment effects at 4 weeks (b = 0.034, p \< 0.004), 6 months (b = 0.033, p \< 0.046), and 12 months (b = 0.036, p \< 0.043) post-baseline for increased expectations about long-term outcomes from parent-child communication about sex. We also found significant booster treatment effects on this outcome among mothers. Booster treatment was associated with increased long-term outcome expectations at 6 months (b = 0.032, p \< 0.044) and 12 months (b = 0.042, p \< 0.024) post-baseline, relative to mothers in the control condition. However, the booster treatment effect was not significantly different compared to the normal treatment, suggesting normal treatment without additional messages was as effective as booster treatment with additional messages at each of the 6- and 12-month follow ups. Across all cognitive outcomes and follow-up time points, no positive treatment effects were observed for fathers.
Discussion
==========
PSUNC appears to be effective, under controlled conditions, in changing social norms regarding the age until which teens should wait to have sex and expectations about long-term outcomes from parent-child communication about sex. Specifically, mothers who were exposed to PSUNC in this experiment exhibited a larger change than control mothers in their norms toward the belief that teens should wait until they are older to have sex. Mothers who were exposed to PSUNC messages also exhibited larger increases than control mothers in their beliefs that parent-child communication about sex would have a positive impact on their child\'s future success. There is limited evidence of short-run effects among mothers on parent efficacy to communicate with their child. This effect surfaces at 6 months post-baseline and dissipates thereafter.
We also found some evidence of positive booster condition effects among mothers on social norms regarding age until which teens should wait to have sex and expectations about long-term parent-child communication outcomes. However, most of these effects were relative to mothers in the control condition. We generally did not find significant differences in effects between mothers that received the normal treatment condition and those that received the booster treatment. That is, in most instances where positive treatment effects occurred, the normal treatment without additional PSUNC messages appeared to be as effective as the booster treatment. Furthermore, all exposure effects observed in this study appear to dissipate after 12 months post-baseline. This pattern is consistent with experiment protocols for exposing treatment condition parents to PSUNC media prior to each of the 4-week and 6-month follow-up surveys. This may suggest that continual exposure is needed to sustain longer-term effects.
Results from our longitudinal model of parent-child communication confirm that the two cognitive factors most impacted by PSUNC messages (i.e., the \"wait until older norm\" and \"long-term outcome expectations\") are also strongly predictive of actual parent-child communication. We found that baseline measures of the \"wait until older norm\" and long-term outcome expectations are predictive of parent recommendations to wait at 18 months post-baseline. The other two cognitive variables targeted by the campaign and examined in this study (self-efficacy to communication and short-term expectations) were less associated with recommendations to wait. Combined, these results shed new light on how the effects of PSUNC on parent-child communication as reported in Evans et al. \[[@B22]\] manifest through the PSUNC theoretical model. That is, two of the specific cognitions that PSUNC appears to impact are also predictive of parent-child communication. The cognitions that PSUNC is not associated with are less predictive parent-child communication.
These findings have important implications for future PSUNC message development. First, our findings speak to the validity of the campaign\'s basic strategy that is outlined in its theoretical model. The campaign identified specific cognitive factors that were theorized to impact parent-child communication and then targeted those cognitive factors in its messages. Our results suggest that the \"wait until older norm\" and long-term outcome expectations were appropriate cognitions to target and the campaign was successful in impacting them. The \"wait until older norm\" is perhaps the most prominent cognition in all PSUNC messages, because it is implicit in almost every ad and campaign material. However, other cognitions identified in the theoretical model were not impacted by PSUNC messages and do not appear to be predictive of parent-child communication. This suggest that if future PSUNC messages are developed, the campaign may be well-served by focusing primarily on the \"wait until older norm\" and long-term outcome expectations in its messaging.
Our findings also have implications for the development of parent-child communication messages more generally. This study highlights the importance of developing a theoretical framework that identifies appropriate cognitive precursors that are both predictive of a campaign\'s distal behavioral outcome and sensitive to campaign messages. This type of framework provides a roadmap of specific cognitions and other precursors to behavior that should be emphasized directly in campaign messages. It is important that this framework be theory-driven to ensure that parent-child communication messages are designed to take action on factors most likely to influence campaign-targeted behavioral outcomes.
This study also provides insights into potential gender differences in message processing for parent-child communication campaigns. Specifically, no exposure effects on cognitive precursors were shown for fathers. This finding contrasts with Evans et al. \[[@B22]\], who found that among both mothers and fathers, exposure to PSUNC was associated with a higher odds of parent recommendations to their child to wait to have sex. The apparent disconnect between these findings may be partly explained by actual gender differences in message processing as well as gender foci in the PSUNC messages themselves. First, many of the PSUNC messages, particularly early advertisements such as \"Muffinhead,\" are mother-centric. Mothers may also be more attentive to PSUNC messages. A number of studies find that mothers are significantly more involved in the sexual health education of their children compared with fathers \[[@B28]-[@B30]\]. Therefore, it is possible that mothers more actively process PSUNC messages that promote parent-child communication about sex. Fathers may also follow a different process between message exposure and behavioral parent-child communication. That is, fathers may be more likely to receive the message and turn directly to behavioral action and engage less in preceding social/cognitive contemplation. Such a process would suggest the hypothesis that PSUNC should have diminished effects on cognitive outcomes among fathers specifically. These are also important considerations for future message development, suggesting that gender-specific message tailoring may be appropriate for parent-child communication campaigns.
It should also be noted that while PSUNC was airing during our study period with a limited PSA distribution, we found very little evidence of control condition contamination. As expected, most participants in the control condition (96%) indicated no awareness of PSUNC television or radio ads, suggesting that contamination of the control condition with real-world exposure was very low. However, there was still a small percentage (4%) of control condition participants that did indicate some level of awareness of the PSUNC ads. This is consistent with levels of exposure that might be expected with a campaign based on PSA distributions. In our multivariable models of message effects, this variable was not significantly associated with any of the main outcome variables and none of the primary study findings were sensitive to the inclusion or exclusion of our variable for PSUNC awareness. This would suggest that our study findings are not biased by control condition contamination.
The study has a few limitations. First, we examined the effects of parent exposure to PSUNC in a controlled setting using an online survey, which may not reflect real-world conditions. Although the online experimental design provides high internal validity, the external validity of our study is limited because it does not assess the campaign within a field-based setting \[[@B31]\]. However, experimental efficacy studies are an emerging evaluation tool, particularly when media campaign timelines and other logistical considerations preclude the use of in-field evaluation designs \[[@B32]\]. Second, although the KN panel includes both Internet and non-Internet households and uses random-digit-dialing methodologies for recruitment, it may not perfectly represent the U.S. population of parents. Our sample was predominantly white with greater educational attainment compared with the U.S. population as a whole. This limits the ability to generalize the findings to the broader population of parents in the United States. Third, survey attrition is slightly higher among treatment condition parents than among control parents. This is somewhat expected given the greater time burden incurred by treatment condition parents who were exposed to experiment media stimuli and answered additional questions about the stimuli they viewed. However, we found that 18-month attrition rates were not significantly different by any of the baseline analytic control variables included in this study.
Finally, because our study utilized a self-administered online survey that most participants complete on their home computers, there is no way to confirm with absolute certainty that all treatment condition participants attentively viewed all of the PSUNC media materials. However, in addition to viewing the materials online, all treatment participants were mailed a DVD containing all media materials for the study. The study questionnaire included an additional item that asked participants whether they actually viewed all of the media materials on the DVD they received. Data based on this item show that 94% of all treatment condition participants reported viewing the DVD in its entirety. Combined with online viewing of the materials during the survey sessions, these data suggest that exposure fidelity was likely very good.
Conclusions
===========
In summary, this study offers empirical evidence on how PSUNC\'s impact on parent-child communication may manifest through its influence on cognitive precursors that are predictive of parent-child communication. However, additional questions remain. For example, more work is needed to determine what other factors, such as parent-child relationships, individual sociodemographic characteristics, and other variables, mediate the effects of PSUNC and whether these mediating processes vary by parent gender. Formal mediation analyses are needed to better address these questions. Furthermore, little is known about how parent-based communication programs translate into changes in outcomes among children of parents who receive messages like those disseminated by PSUNC. Confirmatory data from child self-reports of parent-child communication outcomes are needed to address this. Future evaluation studies for PSUNC will investigate these issues.
Competing interests
===================
The authors declare that they have no competing interests.
Authors\' contributions
=======================
KCD participated in the design and coordination of the study, developed analysis plans, and drafted the manuscript. JLB participated in the design of the study, assisted with statistical analysis and helped to draft the manuscript. WDE participated in the design and coordination of the study and helped to draft the manuscript. KK preformed the statistical analysis and assisted in drafting the manuscript. All authors read and approved the manuscript.
Acknowledgements
================
This study was funded by the U.S. Department of Health and Human Services, Office of Population Affairs. The findings and conclusions in this study are those of the authors and do not necessarily represent the views of the U.S. Department of Health and Human Services.
| {
"pile_set_name": "PubMed Central"
} |
Introduction
============
Macrolide antibiotics were shown to have immunomodulatory and anti-inflammatory, as well as antibacterial effects (Culic et al. [@b13]; Martinez et al. [@b28]; Kanoh and Rubin [@b23]). In vitro azithromycin inhibits the release of the neutrophil chemoattractant interleukin(IL)-8 and of granulocyte-macrophage colony-stimulating factor (GM-CSF) by lipopolysaccharide (LPS)-activated, human primary bronchial epithelial cells (Murphy et al. [@b32]), and in vivo azithromycin has been shown to inhibit LPS-induced mouse pulmonary neutrophilia (Bosnar et al. [@b8]). In potential agreement with these activities, azithromycin significantly decreased the rate of exacerbation in chronic obstructive pulmonary disease (COPD) patients (Albert et al. [@b2]) as well as in non-cystic fibrosis bronchiectasis patients (Wong et al. [@b43]), but the molecular mechanism of its pulmonary anti-inflammatory activity has not yet been elucidated.
Airway infiltration by polymorphonuclear leukocytes (PMNs) in response to bacterial and viral infections represents a pivotal feature of lung inflammatory reactions and is both directly and indirectly involved in most lung pathologies (Murugan and Peck [@b33]). Mobilization of PMNs to inflamed lungs is modulated by a complex interplay between cytokines, endothelial cells, and neutrophils. During the acute-phase response to infectious agents in ruminants and humans, inflammatory cytokines such as IL-1*β*, tumor necrosis factor(TNF)-*α* and IL-8, secreted by a variety of immune and nonimmune cell types, induce a strong PMNs infiltration that sometimes fail to control the infection, and may even contribute to the development of lesions in the lung (Malazdrewich et al. [@b27]; Mukaida [@b31]; Murugan and Peck [@b33]).
Since its discovery in 1986 (Sen and Baltimore [@b39]), the role of ubiquitous nuclear factor-*κ*B (NF-*κ*B) transcription factor has been extensively studied in tumorigenesis, immunity, and inflammatory responses, and its dysregulated activation has been associated with malignancies as well as to inflammatory pathological conditions (Bours et al. [@b10]; Salminen et al. [@b38]; Tornatore et al. [@b41]).
In the lungs, several noxious/inflammatory stimuli activate NF-*κ*B, including intact bacteria, Gram-negative bacterial LPS, ozone, and silica delivered directly to the airways, as well as systemic inflammatory insults such as sepsis, hemorrhage, and direct liver injury (Blackwell et al. [@b6], [@b7]). In rodent models of lung inflammation induced by LPS, pretreatment with relatively nonspecific inhibitors of NF-*κ*B activation was found to decrease lung inflammation (Lauzurica et al. [@b25]), and mice deficient in both RelA (the transactivating subunit of NF-*κ*B) and type 1 TNF receptor (TNFR1) showed impaired neutrophil recruitment to the lungs in response to LPS (Alcamo et al. [@b3]).
To dynamically investigate in vivo, the involvement of NF-*κ*B in the inflammatory response, a transgenic mouse model was developed expressing the minimal promoter of NF-*κ*B and the luciferase gene as a reporter, allowing a direct monitoring of NF-*κ*B activation by bioluminescence (Carlsen et al. [@b11]; Everhart et al. [@b18]; Kielland and Carlsen [@b24]). BLI is based on the detection of light emission from cells or tissues (Doyle et al. [@b17]), and requires an expression cassette consisting of the bioluminescence reporter gene (in this case was luciferase) under the control of a selected promoter (in this experimental setting the NF-*κ*B gene promoter) driving the reporter. To induce light production, the substrate luciferin must be provided by intravascular or intraperitoneal injection immediately prior to bioluminescence imaging (BLI) evaluation. BLI has proven to be a very powerful technique, facilitating real-time, molecular level analysis of disease progression. Nevertheless the generation, characterization, and colony management of transgenic mice may be difficult, and typically very expensive. An alternative, relatively inexpensive approach is represented by the use of transient transgenic mice obtained through the infusion of a polyethylenimine (PEI)/DNA complex containing plasmids with specific responsive elements and luciferase as a reporter gene. This technique yields significant transfection in vivo, as was recently shown by imaging pulmonary NF-*κ*B activation in LPS-treated mice (Ansaldi et al. [@b4]).
In this work, we studied the correlation between lung bioluminescence and lung inflammatory cell infiltration in mice transiently transfected with the luciferase (*luc*) gene under the control of an NF-*κ*B responsive element, upon intratracheal challenge with LPS, and we took advantage of this unique model to provide novel, in vivo evidence that the molecular mechanism of action of the anti-inflammatory activity of the macrolide antibiotic azithromycin involves the modulation of NF-*κ*B activation in lung resident cells.
Materials and Methods
=====================
Animals
-------
Female FVB (7--8 week-old) mice were purchased from Harlan Laboratories Italy (S. Pietro al Natisone, Udine, Italy). Animals were maintained under conventional housing conditions. Prior to use, animals were acclimated for at least 5 days to the local vivarium conditions (room temperature: 20--24°C; relative humidity: 40--70%), having free access to standard rat chow and tap water. All animal experiments were carried out in agreement with the revised "Guide for the Care and Use of Laboratory Animals" ([@b19]) and were approved by the Institutional Animal Care and Use Committee at Chiesi Farmaceutici. The study adhered to the ARRIVE guidelines (McGrath et al. [@b30]).
Reagents
--------
LPS (from Escherichia coli 0111:B4, product n.L3012) was from Sigma (St. Louis, MO); azithromycin (Zitromax) was from Pfizer Inc (Latina, Italy); JetPEI was from Polyplus-transfection Inc (Euroclone, Milano, Italy); NF-*κ*B vector (pGL4.32\[luc2P/NF-*κ*B-RE/Hygro\]) was from Promega (Madison, WI); bortezomib (Velcade) was from Millennium Pharmaceuticals (Cambridge, MA).
Vector characteristic and in vivo gene delivery
-----------------------------------------------
The pGL4.32\[luc2P/NF-*κ*B-RE/Hygro\] vector (GenBank/EMBL Accession Number EU581860) contains five copies of an NF-*κ*B responsive element (NF-*κ*B-RE) that drives transcription of the luciferase reporter gene luc2P (Photinus pyralis). Luc2P is a synthetically derived luciferase sequence with humanized codon optimization that is designed for high expression and reduced anomalous transcription.
The luc2P gene contains hPEST, a protein destabilization sequence. The protein encoded by luc2P responds more quickly than the protein encoded by the luc2 gene upon induction. The vector backbone contains a resistance gene to allow selection in *E. coli* and a mammalian selectable marker for hygromycin resistance.
JetPEI (Wu et al. [@b44]; Oh et al. [@b35]) was applied in vivo as a carrier for delivering DNA to lung tissues. The DNA and JetPEI were formulated according to the product manual. Briefly, 40 *μ*g of NF-*κ*B-luc reporter and 7 *μ*L of JetPEI were each diluted into 100 *μ*L 5% glucose. The two solutions were then mixed and incubated for 15 min at room temperature.
The entire mixture (app.ly 200 *μ*L) was injected into the tail vein of mice.
In vivo bioluminescence imaging
-------------------------------
Transfection per se causes a mild lung inflammatory response and NF-*κ*B activation that is detectable by BLI up to 3--4 days after DNA injection and disappears completely after 1 week. Therefore 1 week after DNA delivery, the transient transgenic mice were injected with luciferin i.p. and BLI was recorded to check the baseline activation of the NF-*κ*B pathway. Briefly, following intraperitoneal injection of luciferin (150 mg/kg) mice were lightly anesthetized with isoflurane (2.5%) and images were obtained using an IVIS imaging system (Caliper Life Sciences, Alameda, CA) at 10 and 15 min after luciferin: an average of photons emitted from the chest of the mice was quantified using Living Image® software (Caliper Life Sciences). The following day, mice were intratracheally challenged with LPS (12.5 *μ*g/mouse) and BLI was recorded at 2, 4, 7, and 24 h after LPS instillation, 15 min after i.p. injection of luciferin (150 mg/kg).
Immunofluorescence staining of luciferase, CD31 and cytokeratin 18 (CK18)
-------------------------------------------------------------------------
To identify the type of cell expressing luciferase protein, duoplex immunofluorescence staining was performed with anti-luciferase antibody and either an anti-CD31 (an endothelial cell marker) or anti-CK18 (an epithelial cell marker). Briefly, during necropsy, 4% paraformaldehyde was first injected through the right ventricle/pulmonary artery to inflate the blood vessel within the lung tissue, and subsequently injected through the trachea to inflate the alveoli. Lung samples were then embedded in paraffin and cut into 5-*μ*m sections, deparaffinized, microwaved in Citra Plus (pH 6; BioGenex, Fremont, CA) for antigen retrieval, and blocked with Antibody Diluent (Dako, Carpintiria, CA). Sections were then incubated with anti-luciferase antibody (Novus Biologicals, Littleton, CO) at 1:100 dilution and 4°C overnight, followed by anti-goat secondary antibody conjugated with Alexa Fluor 594 (Invitrogen, Grans Island, NY) for 1 h. Thereafter, sections were incubated with anti-CD31 antibody (Abcam, Cambridge, MA) at 1:100 dilution for 1 h, followed by anti-mouse secondary antibody conjugated with Alexa Fluor 488 for 1 h, and finally by 4′,6-diamidino-2-phenylindole (DAPI) for 5 min for nuclear staining. For Luciferase/CK18 duoplex staining, after deparaffinization, microwaving and blocking, sections were incubated with an anti-CK18 antibody (Abcam) at 1:500 dilution for 1 h, followed by SuperPicture horseradish peroxidase-conjugated anti-mouse secondary antibody (Invitrogen) for 5 min, and tyramide signal amplification conjugated fluorescein (PerkinElmer, Alameda, CA) for 10 min. Thereafter, sections were microwaved again, blocked and then incubated with anti-luciferase antibody at 1:100 overnight, followed by anti-goat secondary antibody conjugated with Alexa Fluor 594 for 1 h, and DAPI for 5 min for nuclear staining. Images were captured with a Vectra-2 Imaging System (PerkinElmer).
Acute pulmonary inflammation: effect of bortezomib and azithromycin
-------------------------------------------------------------------
Intratracheal (i.tr.) challenge with LPS was carried out using 50 *μ*L of LPS solution (250 *μ*g/mL in phosphate buffered solution \[PBS\]) and a 22-gauge intubator, resulting in a final dose of LPS of 12.5 *μ*g/mouse, with the control group receiving 50 *μ*L of saline i.tr.. In a limited number of mice, 4 h after LPS challenge and luciferin administration, lungs were rapidly excised and BLI recorded at 10 and 15 min after luciferin, as described for intact animals.
Bortezomib was administered by injection into the tail vein at different doses (0.5--1 mg/kg, Psallidas et al. [@b36]) 1 h before LPS i.tr. instillation.
Azithromycin (Zitromax solution) was administered p.o. by gavage at different doses (100--600 mg/kg) (Bosnar et al. [@b8]) 4 h before LPS i.tr. instillation.
Bronchoalveolar lavage and cytokine determination
-------------------------------------------------
Twenty-four hours after LPS challenge, animals were weighted, anaesthetized with isoflurane and sacrificed by bleeding from the abdominal aorta for bronchoalveolar lavage fluid (BALF) collection, performed as previously described (Nassini et al. [@b34]). Bronchoalveolar lavage (BAL) fluid supernatants were frozen at −80°C for simultaneous quantitation of multiple cytokines/chemokines using a Bio-Plex™ Cytokine Assay Kit (Bio-Rad Laboratories, Segrate, Milano, Italy).
The cell pellet was resuspended in 0.2 mL of PBS and cell counts were obtained using a particle counter (Dasit XT 1800J, Cornaredo, Milano, Italy).
p65 nuclear translocation
-------------------------
Lungs were excised and homogenized using a trans Turrax homogenizer. Cytoplasmatic and nuclear extracts were obtained using Nuclear Extraction Kit (Active Motif, La Hulpe, Belgium), and determination of p65 was carried out using TransAM NFkB p65 transcription factor assay kit (Active Motif), according to the manufacturer instructions.
Data analysis
-------------
As tests for normality were positive, statistical analysis was performed on raw data using one-way analysis of variance (ANOVA) followed by Dunnett\'s *t* post-hoc test for comparison with control groups. Experimental values were expressed as the mean and standard error of the mean (SEM) of *n* observations. (\**P* \< 0.05, \*\**P* \< 0.01).
Results
=======
As previously reported, LPS intratracheal instillation 1 week after DNA delivery of the luciferase reporter construct caused activation of NF-*κ*B that could be easily monitored by BLI, showing a very marked increase when compared to control animals (Fig. [1](#fig01){ref-type="fig"}A). Ex-vivo imaging of isolated lungs confirmed that the bioluminescence observed in vivo was associated to the activation of NF-*κ*B in the lungs (Fig. [1](#fig01){ref-type="fig"}B). *Luc* expression driven by NF-*κ*B activation was detectable by BLI as early as 2 h after LPS challenge, peaked at 4 h after treatment, with a 10-fold induction over baseline. This signal was still significantly enhanced 7 h after LPS, but returned to baseline levels 24 h after LPS challenge (Fig. [1](#fig01){ref-type="fig"}C). Double staining immunofluorescence analysis performed to identify which cell type(s) is(are) targeted by the injected plasmid DNA showed LPS-induced expression of luciferase in both epithelial (Fig. [2](#fig02){ref-type="fig"}A--D) or endothelial (Fig. [2](#fig02){ref-type="fig"}E--H) cells. While regular staining was observed for CK18 and CD31 (Fig. [2](#fig02){ref-type="fig"}I and K, and [Fig. S1](#SD1){ref-type="supplementary-material"}A and C), no signal for luciferase was observed in LPS-challenged wild type animals (Fig. [2](#fig02){ref-type="fig"}J) or in saline-challenged NF-*κ*B-luc animals ([Fig. S1](#SD1){ref-type="supplementary-material"}B and D).
{#fig01}
{#fig02}
Bortezomib (Velcade), a known proteasome inhibitor that interferes with the degradation of IkB and therefore the activation of NF-*κ*B, dose dependently inhibited luciferase expression 4 h after LPS administration (Fig. [3](#fig03){ref-type="fig"}A and B), lending support to the quantitative correlation between NF-*κ*B activation and bioluminescence determination in the transient transgenic mice. In agreement with the fact that the expression of adhesion molecules leading to white blood cells (WBC) extravasation and infiltration are results of NF-*κ*B activation (Read et al. [@b37]; Kalogeris et al. [@b22]; Dagia and Goetz [@b14]), increased numbers of WBC and neutrophils were also recovered by BAL 24 h after LPS challenge (Fig. [4](#fig04){ref-type="fig"}A and B) and this increase was also significantly inhibited by bortezomib treatment. A strong correlation was observed between bioluminescence observed 4 h after LPS administration and the number of WBC and neutrophils recovered by BAL 24 h after LPS challenge (Fig. [4](#fig04){ref-type="fig"}C), further supporting the ability of this transient transgenic mouse model to provide an accurate, in vivo determination of NF-*κ*B activation in the lung.
{#fig03}
{#fig04}
Pre-treatment of transient transgenic mice with azithromycin per os, dose-dependently inhibited bioluminescence induced by LPS tracheal instillation (9.5 ± 1.74 fold, 6.9 ± 1.82 and 4.16 ± 0.78 fold of induction over baseline at 4 h in LPS-control, azithromycin 100 mg/Kg p.o., and azithromycin 600 mg/Kg p.o., respectively) (Fig. [5](#fig05){ref-type="fig"}A and B). As observed with bortezomib, the inhibition of BLI was followed by a significant inhibition of WBC and neutrophil infiltration at 24 h confirming the anti-inflammatory activity of this compound (Fig. [6](#fig06){ref-type="fig"}A and B).
{#fig05}
{#fig06}
The inhibitory activity of azithromycin on NF-*κ*B activation, also resulted in a statistically significant inhibition in the BAL concentrations of several pro-inflammatory cytokines, such as TNF-*α*, granulocyte colony-stimulating factor (G-CSF) and monocyte chemoattractant protein-1 (MCP-1) (Fig. [7](#fig07){ref-type="fig"}A--F), while other cytokines upregulated by LPS were not affected ([Fig. S2](#SD2){ref-type="supplementary-material"}A--E). A complete list of cytokines analyzed is reported in [Table 1S](#SD3){ref-type="supplementary-material"}.
{#fig07}
Time-course of p65 nuclear translocation paralleled *luc* signal induction, with maximal accumulation occurring at 4 h, while by 24 h p65 was again localized primarily in cytoplasm (data not shown); azithromycin treatment inhibited by 52% (*P* \< 0.05) the nuclear translocation of activated NF-*κ*B observed in lung homogenates 4 h after LPS tracheal instillation.
Discussion
==========
The results obtained provided, for the first time, in vivo evidence that azithromycin treatment results in pulmonary anti-inflammatory activity associated with the inhibition of NF-*κ*B activation in the lung, as followed by bioluminescence monitoring in mice transiently transfected with luciferase gene under the control of an NF-*κ*B responsive element (Ansaldi et al. [@b4]).
While a limited number of evidence obtained in vitro using different cell types, including airways cells, pointed to the inhibition of NF-*κ*B as part of the mechanism of the anti-inflammatory activity of azithromycin (Aghai et al. [@b1]; Cigana et al. [@b12]; Matsumura et al. [@b29]; Vrancic et al. [@b42]), the only evidence available in vivo was obtained in a model of ocular inflammation, where it was shown that azithromycin treatment decreased the amount of NF-*κ*B protein detected by western blot in conjunctival homogenates, a measurement that does not actually evaluate changes in NF-*κ*B activation. On the other side, studying the activity of azithromycin in a model of LPS-induced pulmonary neutrophilia Boznar and co. could not show any inhibitory effect on NF-*κ*B activation in alveolar macrophages (Bosnar et al. [@b9]). The ability, offered by the animal model used in this study, to evaluate and monitor NF-*κ*B activation in vivo at the whole organ level, clearly showed that a significant activation of NF-*κ*B is indeed taking place in the lung upon intratracheal LPS challenge, and that azithromycin anti-inflammatory activity significantly inhibited NF-*κ*B activation-dependent bioluminescence. Luciferase co-expression with epithelial and endothelial cell-specific markers, as assessed by duoplex immunofluorescence, suggests that these cells may therefore represent potential targets of the activity of azithromycin. It must be noted that although cytokeratins are considered a typical marker of epithelial differentiation, they can also be expressed in certain vascular smooth muscle cells (Bar et al. [@b5]).
Although no infiltrating inflammatory cells showed up significantly expressing luciferase, this does not rule out the activation of NF-*κ*B in neutrophils, as their rapid turnover likely prevents them from being detected at the time-points used for LPS challenge after transient transfection.
NF-*κ*B activation upon LPS challenge resulted in a rather striking correlation between bioluminescence induced by NF-*κ*B activation and inflammatory biomarkers such as WBC and neutrophils airways infiltration, as shown by the parallelism between changes in these parameters induced by LPS alone or in the presence of increasing doses of a potent proteasome inhibitor (bortezomib) known to inhibit NF-*κ*B activation, by blocking the degradation of I*κ*B, the inhibitory unit of the NF-*κ*B complex, carried out by the 26S proteasome.
Dysregulation of NF-*κ*B activation has been involved in lung diseases such as asthma, chronic bronchitis, and COPD (Donovan et al. [@b16]; Teramoto and Kume [@b40]), but its specific contribution to disease progression is currently unknown (Lawrence et al. [@b26]). BLI provides a noninvasive approach to monitor gene expression in vivo and represent an important tool to evaluate the potential contribution of NF-*κ*B to the evolution of acute inflammatory reactions.
BLI is a powerful technique based on the detection of visible light produced during luciferase-mediated oxidation of a molecular substrate in the presence of the enzyme resulting from its expression in vivo as a molecular reporter. Bioluminescence arising from luciferase can be imaged as deep as several centimeters within tissues, allowing at least organ-level resolution. Being simple to execute and minimally invasive, BLI enables monitoring and serial quantification of biological processes without sacrificing the experimental animal. This powerful technique can therefore reduce the number of animals required for experimentation because multiple measurements can be made in the same animal over time, also minimizing the effects of biological variation.
In this study, we confirm the activity of azithromycin against pulmonary inflammation induced by LPS (Ianaro et al. [@b20]), showing a reduction in WBC and neutrophil airways infiltration and a significant decrease in the concentrations of proinflammatory cytokines in BAL. It must be noted that in this study azithromycin has been used prophylactically to prevent an acute inflammatory response mainly because we have been trying to establish a link between its anti-inflammatory activity and NF-*κ*B activation in vivo. Indeed, taking advantage of BLI on transiently transfected animals, we provide unequivocal evidence that this activity is associated with a significant inhibition of NF-*κ*B activation in vivo. Double immunofluorescence staining with antibodies targeting luciferase protein and specific cell type antigens, showed that in transient NF-*κ*B-luc transgenic mice intratracheal challenge with LPS induces luciferase expression both in epithelial and endothelial cells, suggesting a key role for these cell types in the inflammatory response to bacterial infection. Although no infiltrating inflammatory cells resulted in expressing luciferase, this does not rule out the activation of NF-*κ*B in these cells, but it rather reflects their rapid turnover.
This innovative analytical approach to the in vivo monitoring of pulmonary inflammation through the analysis of bioluminescence in transiently transfected mice, can certainly be used to test different macrolide antibiotics for their effects on NF-*κ*B activation, or, taking advantage of different vectors, to monitor different models of pathology (fibrosis, asthma etc.).
As previously reported, azithromycin was also able to significantly reduce the concentrations of G-CSF within the airways, and this activity may result in a decrease in epithelial cells-dependent neutrophil survival within the airways (Yamasawa et al. [@b45]), and contribute to the reported effects, among others, on COPD exacerbations (Albert et al. [@b2]; Yamaya et al. [@b47]).
Not much is known on the molecular mechanism of the inhibition of NF-*κ*B by azithomycin, but the possibility that the anti-inflammatory activity of macrolide antibiotics such as azythromycin or erythromycin may relate uniquely to their antibiotic activity has been ruled out by the identification of an erythromycin analog devoid of the antibiotic activity but retaining the anti-inflammatory action (Desaki et al. [@b15]). Conversely, it has been shown that azithromycin inhibits the binding of activator protein-1, nuclear factor of activated T cells, and interferon consensus sequence binding protein to the DNA-binding site in the IL-12p40 promoter (Yamauchi et al. [@b46]), and down regulation of TLR-4 receptor by azithromycin has also been reported (Iwamoto et al. [@b21]), and these could represent important mechanisms contributing to the anti-inflammatory effects of azithromycin.
In consideration of the striking correlation of NF-*κ*B activation as assessed by BLI, and downstream airway inflammatory responses, monitoring of BLI in this relatively simple model of transient transgenic mouse may represent a suitable approach for in vivo assessment of NF-*κ*B-associated inflammatory responses, and possess significant potential for the elucidation of molecular mechanisms in the pathobiology of lung diseases.
In conclusion, taking advantage of a transient transgenic mice expressing luciferase under the control of NF-*κ*B-responsive elements, we provide evidence that azithromycin inhibits NF-*κ*B activation in vivo, further supporting the important contribution of this mechanism of action to its anti-inflammatory activity.
The Authors would like to thank Fabrizio Facchinetti and Gessica Marchini for their skillful technical assistance.
Disclosures
===========
F. F. Stellari, P. Caruso, T. M. Topini, C. Carnini, M. Civelli, and G. Villetti are employees of Chiesi Farmaceutici that does not sell any of the drugs mentioned in the article.
K. P. Francis and X. Li are employees of Perkin Elmer, that sells imaging devices.
Supporting Information
======================
Additional Supporting Information may be found in the online version of this article:
######
**Figure S1.** Double immunofluorescence staining of mice lungs. Duoplex immunofluorescence staining of luciferase/CK18 (epithelial cell marker) or luciferase/CD31(endothelial cell marker) were performed on paraformaldehyde-fixed lung sections, as described in Materials and Methods**.** Double staining of lung obtained from saline-challenged NF- *κ*B-luc mice showed normal staining of epithelial cells (A) and endothelial cells (C) for CK18 and CD31, respectively. On the contrary, no staining for luciferase was observed (B and D). Original magnification: 210×.
######
**Figure S2.** (A--E). LPS-induced cytokines in bronchoalveolar lavage fluid (BALF): effect of azithromycin. Transient NF-*κ*B-luc transgenic mice were pretreated with saline or azithromycin (AZI, 100--600 mg/kg, *per os)* 4 h prior to LPS and sacrificed 24 h after tracheal instillation of LPS. A: IL1-*α*. B: IL1-b. C: IL-12. D: IL-12(p70). E: KC. Values are shown as mean ± SEM, *n* = 6 for each group.
######
**Table S1.** Cytokines analyred in BAL of LPS-challenged mice. Cytokines modulated by LPS showed a statistically significant increase in BAL concentrations obtained after LPS challenge.
[^1]: **Funding Information** Supported by Institutional Funds from Chiesi Farmaceutici S.p.A.
| {
"pile_set_name": "PubMed Central"
} |
ANNOUNCEMENT {#s1}
============
A healthy human feces sample was preenriched in modified peptone yeast (PY) medium ([@B1]) under anaerobic conditions at 37°C for 5 days. After 3 consecutive dilutions, 0.1 ml of the sample (10**^−^**^3^ dilution) was plated on anaerobic agar ([@B2]) and incubated anaerobically at 37°C for 48 h. The isolated pure colony was cultured using reinforced medium ([@B3]); the cells were harvested for DNA isolation with the phenol-chloroform extraction method. The 16S rRNA gene was amplified with 27F and 1429R primers ([@B4]), followed by amplicon sequencing. Sequence analysis with NCBI BLAST showed 100% sequence similarity with Clostridium butyricum strain TK520 (GenBank accession number [CP016332](https://www.ncbi.nlm.nih.gov/nuccore/CP016332)).
Following species identification, the culture pellet of the isolate grown as described above was outsourced for whole-genome shotgun (WGS) sequencing to Chromous Biotech Pvt. Ltd., Bengaluru, Karnataka, India.
The genomic DNA from the culture pellet was fragmented with an ultrasonicator to generate 300 to 400-bp fragments. Then 100 ng of fragmented DNA was used to prepare a paired-end sequencing library with a NEBNext UltraTMII DNA library prep kit, and sequencing was performed on an Illumina HiSeq platform.
A total of 17,652,182 paired-end reads of 150-bp read length on average with a genome coverage of 1,171× were sequenced, out of which 12,864,290 high-quality paired-end reads were filtered with the NGS QC Toolkit ([@B5]). The high-quality paired-end reads were assembled into 81 contigs with the *de novo* genome assembler SPAdes version 3.11.1 ([@B6]) and the scaffolder SSPACE-standard version 3.0 ([@B7]). The draft genome consists of 4,572,221 bp with a GC content of 28.65%, and the largest assembled scaffold is 1,961,597 bp ([@B5]). The genome sequence was annotated with the RAST server ([@B8]) and the NCBI's Prokaryotic Genome Annotation Pipeline (PGAP) ([@B9]). The genes were predicted and translated through the Prokaryotic Dynamic Programming Gene-finding Algorithm (Prodigal) program ([@B10]), following pathway identification with the KEGG Automatic Annotation Server (KAAS) ([@B11]). A total of 4,685 genes were predicted, which include 4,177 coding sequences (CDS), 11 rRNAs, and 114 tRNAs. The strain is predicted to encode about 199 proteins involved in carbohydrate metabolism and 150 proteins involved in amino acid metabolism, and 60 genes are involved in the synthesis of proteins and enzymes required for dormancy and sporulation stages. The strain also contains genes involved in the biosynthesis of biotin, riboflavin, cobalamin, thiamine, vitamin B~6~, and folate. The genome was screened to determine the putative virulence factors (virulence factor database \[VFDB\]) ([@B12]), plasmids (PlasmidFinder 2.0) ([@B13]), and antibiotic-resistant genes (Antibiotic Resistance Genes Database \[ARDB\]) ([@B14]). The genes encoding putative virulence factors such as botulinum neurotoxin (*atx*), Clostridiumdifficile toxin (*cdtA*, *cdtB*), hemolytic enterotoxin complex HBL (*hblA*, *hblB*, *hblC*, and *hblD*), nonhemolytic enterotoxin (NHE; *nheA*, *nheB*, and *nheC*), and enterotoxin (*entA*, *entB*, *entC*, and *entD*) were not found. Analysis of the genome sequence of Clostridium butyricum UBCB 70 infers that it does not contain any virulence factors or plasmid-containing antibiotic-resistant genes.
Data availability. {#s1.1}
------------------
The raw sequence reads have been submitted to the NCBI SRA under the accession number [SRR8372150](https://www.ncbi.nlm.nih.gov/sra/SRR8372150), and the whole-genome shotgun project of Clostridium butyricum UBCB 70 has been deposited in DDBJ/EMBL/GenBank under the accession number [RSEV00000000](https://www.ncbi.nlm.nih.gov/nuccore/RSEV00000000). The version described in this paper is the first version, RSEV01000000.
We acknowledge the staff at Chromous Biotech Pvt. Ltd. for their services.
[^1]: **Citation** Sulthana A, Thorramamidi A, Lakshmi SG, Madempudi RS. 2019. Whole-genome shotgun sequencing and characterization of probiotic strain *Clostridium butyricum* UBCB 70 to assess its safety. Microbiol Resour Announc 8:e01732-18. <https://doi.org/10.1128/MRA.01732-18>.
| {
"pile_set_name": "PubMed Central"
} |
Background
==========
The concept of case management is thought to be an effective and efficient approach to manage patients with chronic illness and complex health care needs \[[@B1]\]. Case management can implement key elements of the chronic care model, such as improved continuity of care by redesigning the delivery system and enhancing patients\' self-management skills \[[@B2]\]. It can also contribute to better implementation of evidence-based recommendations for diagnostic procedures, pharmaceutical treatment, life style counselling and monitoring of patients.
Case management has been implemented in a range of clinical settings \[[@B3]\] but most chronically ill patients receive most of their medical care in primary care settings, at least in countries with a strong primary care system. Therefore this seems to be the most obvious setting for case management programs. A previous review on case management in primary care concluded that interventions supervised by generalists were not effective in reducing health resource use \[[@B4]\]. Since then the body of evidence has vastly expanded so that an update of this reviews was required.
Case management is rather a generic concept than a clearly defined intervention strategy \[[@B5]\]. But conclusive answers about the effects of complex interventions like case management programs demand precise definitions for the inclusion of evidence \[[@B6]\]. Therefore we aim to base the selection of studies on a basic principle of case management which can be found in various concepts \[[@B7]-[@B9]\] (figure [1](#F1){ref-type="fig"}). The principle case management process emphasises on highly intensive individualized care contrasting *case*management to *disease*management programs \[[@B10]\]. Case Management starts with identifying cases in need for intensified management. As a next common step needs assessment and individualized planning are implemented in different concepts. Monitoring and/or re-assessment of the various actions which could be planned to manage the cases are more or less explicit elements of various case management approaches. All components of this process can be undertaken in different settings. In the context of our interest on primary care based case management we defined the involvement of a primary care physician (either general internist, general practitioner, family physician) in planning the management of individual cases as being essential.
{#F1}
Using a typology to guide (sub)classification of interventions could further improve the quality of systematic reviews on complex interventions \[[@B6]\]. The typology proposed by the Cochrane Consumer and Communication Review Group may be useful to classify different primary care based case management interventions from the perspective of the interacting \"cases\" and \"managers\" \[[@B11]\]. This may help answering the question \"why\" some case management interventions in primary care work and others do not.
We aim to review existing evidence on chronic diseases that can be effectively and efficiently cared for by implementing case management in primary care. We will further search for key components of case management which due to success or failure of these programs.
Methods/Design
==============
This systematic review is performed according to standards derived from the PRISMA Statement \[[@B12]\].
Eligibility criteria
--------------------
### Types of studies
We include randomised and non-randomised controlled trials studying the effects and/or economic implications of case management interventions compared to routine care. No language, publication date or publication status restrictions will be made. Study protocols will also be included and we aim to contact authors to provide details about ongoing publications.
### Types of participants
Studies on adult participants (18 years and above) suffering from at least one chronic condition will be included in the review. Reports on interventions of palliative care, cancer screening, primary prevention, and treatment of drug or substance abuse are excluded from this review.
### Type of intervention
Trials comparing case management interventions in which primary care physicians (alone or in collaboration with specialists) were involved in planning of the management strategy for individual cases will be included in this review. Case management is defined as consisting of all elements of the case management process as described above (figure [1](#F1){ref-type="fig"}). Monitoring in this context is defined as a periodic assessment (at least once in six months) which can result in a change of managing the case. We aim to contact authors in the case of ambiguities about key components due to lack of reporting in publications. We will include studies comparing case management interventions with routine management (usual care, attention control). The content of \"usual care\" will be carefully described in the review.
Information sources
-------------------
We will search Medline via OVID (1950- 2009), the Cochrane Central Register of Controlled trials (2009), DARE, Embase (1980 to 2009), CINAHL (1982-2009), NHS EED, PsychInfo (1887- 2009), Science Citation Index (1987- 2009), Royal College of Nursing database, Dissertation abstracts (1861-2009) and Registers for clinical trials. The search is not limited to language or publication date. Searches will be designed and conducted by TF, AM and MW and assisted by librarians. Additionally we aim to search reviews relevant for the topic to retrieve further studies of interest. The reference lists of retrieved reports will be screened for potentially relevant studies. We will ask experts in the field for support to avoid missing relevant studies.
Search
------
We use a highly sensitive search filter for randomized controlled trials \[[@B13]\] and add \"usual care\" in the filter as this is a common term used for describing the control group of case management trials. We will identify studies of relevance using the MeSH terms: primary health care.exp and patient care planning.exp. A search in titles and abstracts will be performed using a combination of free text terms: case manage\$.tw, care manage\$.tw, disease manage\$.tw, integrated care.tw, family medicine.tw, family practice.tw, general prac\$.tw, primary care.tw.
Study selection
---------------
Eligibility assessment will be performed independently by two reviewers (TF, FK). Studies will be selected unblinded according to a standardized procedure. Abstracts with incongruent assessment results will be included in full text screening without further consensus discussion. Disagreements during the full text-based study selection process will be discussed and resolved by consensus.
Data collection
---------------
Two investigators will extract data from each study independently. We use a standardized protocol and reporting form to extract trial characteristics, patient data, outcomes and study quality. The data extraction instrument will be piloted with 5 study reports at minimum which will be included in this review. We aim to refine the instrument after the pilot test. Authors of published study protocols will be asked to provide information about first results and planned publications if results have not already been published.
We will extract data from doubled reports of studies according to the following algorithm \[1. review of study protocol, 2. review of the major publication i.e. published in a \"high-impact\" journal, report of major outcome, 3. review of all other publications with quantitative data, 3. contact to authors in the case of inconsistencies within reports\]. All data of a single study will be displayed comprehensively in the review even if reported in different publications.
Data items
----------
We will design a data extraction form containing the following information: (1) characteristics of participants (age, gender, disease, disease stage and severity, method of diagnosis, drop-outs), and the trial\'s inclusion and exclusion criteria; (2) type of intervention (case finding, assessment, planning, intervention, monitoring, re-assessment; versus usual care or versus attention control); (3) type of outcome measure (including symptom level, quality of life score \[using a validated tool\], quality of care score \[validated tool\], adherence \[validated tool\], mortality, functional status, clinical parameters, surrogate parameters \[e.g. HbA1c\], admission rate, days of hospitalisation, outpatient care visits, emergency department visits, direct or indirect costs, length of follow-up, unintended effects of treatment).
Risk of bias in individual studies
----------------------------------
To assess the validity of included studies pairs of two reviewers will rate risk of bias according to predefined standards using the Cochrane Collaboration\'s tool for assessing risk of bias \[[@B13]\]. This tool has been validated but a detailed checklist is needed to use it appropriately \[[@B14]\]. Using the risk of bias tool adapted by the EPOC Group \[[@B15]\] we created a checklist which we would like to publish *a priori*as recently suggested \[[@B16]\] (see additional file [1](#S1){ref-type="supplementary-material"}: Risk of bias assessment checklist).
Planned methods for analysis
----------------------------
As a first step, we will perform qualitative synthesis of included studies using summary of findings tables. Intervention strategies will carefully be described with emphasis on the interactions between case management teams and patients using a Cochrane typology \[[@B11]\]. The content of usual care, intensity of case management interventions and training of case managers will further be used to compare included studies. Heterogeneity is a common problem of data synthesis from evaluations of complex interventions \[[@B6]\]. We will perform subgroup analyses regarding the following domains: different chronic conditions, training of case managers and intensity of the intervention. Heterogeneity tests will be calculated within subgroups. We aim to synthesize data quantitatively if adequate to the findings. Any meta-analysis will be carried out using a random effects model. Single effect sizes (standardised mean differences) may be more appropriate to report quantitative synthesis in order to compare different outcome measures. If possible, sensitivity analyses are planned to simulate different levels of involvement of primary care physicians in planning. Sensitivity and subgroup analysis will also be undertaken for studies with largely different risk of bias.
Abbreviations
=============
EPOC: Effective Practice and Organization of Care Group; PRISMA: Preferred Reporting Items for Systematic Reviews and Meta-Analyses; CINAHL: Cumulative Index to Nursing and Allied Health; NHS EED: NHS Economic Evaluation Database; DARE: Database of Abstracts of Reviews of Effects.
Competing interests
===================
The authors declare that they have no competing interests.
Authors\' contributions
=======================
TF leads the review and developed the study protocol. TF, FK, AM and MW contributed substantially to the design of the search strategy and abstraction of data. TF wrote the manuscript. AM, JS and MW critically revised it. All authors read and approved the final manuscript.
Pre-publication history
=======================
The pre-publication history for this paper can be accessed here:
<http://www.biomedcentral.com/1472-6963/10/112/prepub>
Supplementary Material
======================
###### Additional file 1
**Risk of bias assessment checklist**. This file contains a detailed checklist for the assessment of risk which will be used in this review. It is an adapted version of the tool provided by the Cochrane EPOC Group.
######
Click here for file
Acknowledgements
================
This review is a project within the research framework of the Competence Centre of General Practice Baden- Wuerttemberg which is funded by the Ministry of Science, Art and Research Baden-Wuerttemberg, Germany.
| {
"pile_set_name": "PubMed Central"
} |
Introduction {#S0001}
============
Proton pump inhibitors (PPIs) are a class of pharmacological agents used to treat acid-related disorders, such as gastroesophageal reflux disease, erosive esophagitis and peptic ulcer.[@CIT0001] Examples of PPI include omeprazole, pantoprazole and esomeprazole, which rank among the top 100 most frequently prescribed drugs in the US.[@CIT0002] They bind covalently to H^+^/K^+^/ATPase of the parietal cells of the stomach to inhibit gastric acid secretion.[@CIT0003] PPIs are generally safe and well tolerated by patients. However, their acid suppression action could lead to malabsorptions of several nutrients, such as vitamin B12 (cobalamin), iron, calcium and magnesium.[@CIT0004],[@CIT0005] These nutrients play essential roles in skeletal health; thus, a deficiency could contribute to bone loss. For example, cobalamin deficiency increases homocysteine and methylmalonic acid levels, which stimulate osteoclast formation and bone resorption.[@CIT0006] Several meta-analyses have confirmed that PPI use increases the risk of fracture despite a lack of change in the bone mineral density (BMD).[@CIT0007]--[@CIT0010] BMD is not a perfect surrogate of bone strength,[@CIT0011] and myriad non-BMD factors contribute to fractures.[@CIT0012] Thus, the differential effects of PPI on BMD and fracture risk are expected.
Limited recommendations are available to prevent bone loss induced by long-term PPI use. Currently, the US Food and Drug Administration recommends calcium and vitamin D supplementation for individuals at risk of osteoporosis and taking PPI.[@CIT0013] It is a logical approach because PPI users may suffer from calcium malabsorption.[@CIT0014] However, a meta-analysis of 33 randomised controlled trials revealed that calcium and vitamin D supplementation might not reduce nonvertebral, vertebral or total fracture risk in community-dwelling elderly.[@CIT0015] Although this meta-analysis is not without criticisms,[@CIT0016] the effects of calcium and vitamin D supplementation on bone health are still debatable. Researchers have been trying other approaches to prevent bone loss. Annatto tocotrienol prevents bone loss in animal models of bone loss induced by sex hormone deficiency and metabolic syndrome.[@CIT0017]--[@CIT0019] Annatto tocotrienol is derived from annatto bean and contains a unique mixture of vitamin E consisting solely of tocotrienol isomers, particularly gamma- and delta-tocotrienol.[@CIT0020] A previous study showed that annatto tocotrienol upregulates expression related to bone formation in orchidectomised rats.[@CIT0019] In cellular studies, annatto tocotrienol promotes the differentiation of osteoblasts by suppressing the mevalonate pathway.[@CIT0021],[@CIT0022] A human trial concluded that 12-week annatto tocotrienol in combination with calcium and vitamin D suppresses bone resorption markers in postmenopausal women with osteopaenia.[@CIT0023] However, the effects of calcium and annatto tocotrienol in combination on physical changes in the bone, such as bone microstructure and mechanical strength, have not been attempted.
Therefore, the current study aims to compare the preventive effects of calcium, calcium plus annatto tocotrienol and a commercial formula of calcium plus vitamin D, Caltrate Plus, on bone loss induced by pantoprazole. Pantoprazole is used in the long-term treatment of pathological hypersecretory conditions, such as Zollinger--Ellison syndrome and idiopathic hypersecretion.[@CIT0024] Adult rats were supplemented with pantoprazole for 60 days, which is equivalent to 5 years in humans.[@CIT0025] We hypothesise that calcium plus annatto tocotrienol is effective in preventing bone loss induced by pantoprazole and could serve as a candidate regime for PPI users to prevent osteoporosis and its associated fracture.
Materials and Methods {#S0002}
=====================
Preparation of Treatment Agents {#S0002-S2001}
-------------------------------
Pantoprazole (Xepa-Soul Pattinson, Malacca, Malaysia) was crushed into powder and dissolved in normal saline with either calcium carbonate (Bendosen Laboratory Chemicals, Bendosen, Norway) or powdered Caltrate Plus (Pfizer, New York, USA). Annatto tocotrienol (American River Nutrition, USA) consisting of 10% gamma-tocotrienol and 90% delta-tocotrienol was diluted in olive oil (Bertolli, Crawley, UK). The dose of Caltrate Plus (31 mg) used in this study was converted from the recommended dose for humans (600 mg) using body surface conversion formula.[@CIT0026] It also contains 41 IU vitamin D~3~, 2.6 mg magnesium, 0.4 mg zinc and trace amounts of copper and manganese. The dose of calcium carbonate used (77 mg) provided the same amount of elemental calcium as Caltrate Plus. The dose of annatto tocotrienol used \[60 mg/kg body weight (b.w.)\] was derived from previous studies,[@CIT0019] which showed that this dose could prevent bone loss in rat models.
Treatment of Animals {#S0002-S2002}
--------------------
Three-month-old Sprague Dawley male rats (n=30) were purchased from Laboratory Animal Resource Unit, Universiti Kebangsaan Malaysia (Kuala Lumpur, Malaysia). They were housed at Animal Laboratory, Department of Pharmacology, Faculty of Medicine, Universiti Kebangsaan Malaysia (Cheras, Malaysia) in ventilated plastic cages under standard conditions (25±2°C, 12/12 h dark/light cycle) and given tap and standard rat chow (Goldcoin, Klang, Malaysia) throughout the study period. After acclimatisation for a week, the rats were randomised into five groups (n=6/group). Four groups were given pantoprazole (3 mg/kg b.w.) via oral gavage daily in the morning, whereas another group was given equivolume of normal saline. Allocation of treatment regime (calcium carbonate, Caltrate Plus, annatto tocotrienol and the respective vehicles) is listed in [Table 1](#T0001){ref-type="table"}. Table 1Treatment Regimes of the RatsGroupMorningEveningNormal control (NC)Normal saline; oral gavageOlive oil; oral gavageNegative control (PPI)Pantoprazole \[3 mg/kg body weight (b.w.)\]; oral gavageOlive oil; oral gavageCalcium-treated (PPI+Ca)Pantoprazole (3 mg/kg b.w.) + calcium carbonate (77 mg); oral gavageOlive oil; oral gavageCalcium and annatto tocotrienol-treated (PPI+Ca+TT)Pantoprazole (3 mg/kg b.w.) + calcium carbonate (77 mg); oral gavageAnnatto tocotrienol (60 mg/kg b.w.); oral gavageCaltrate Plus-treated (PPI+CP)Pantoprazole (3 mg/kg b.w.) + Caltrate Plus (31 mg); oral gavageOlive oil; oral gavage
The body weight of the rats was monitored weekly using a digital balance. The rats were euthanised after 60 days of treatment through ketamine/xylazine overdose. Their femurs were harvested for subsequent analysis. The femurs were cleaned of soft tissues and weighed. The length and diameter of the femurs were measured using a calliper. The protocol of this study was reviewed and approved by Universiti Kebangsaan Malaysia Animal Ethics Committee (approval code: FAR/PP/2018/KOK YONG/26-SEPT./945-JAN.-2019-DEC.-2019). Principles and guides to the ethical use of laboratory animals by Universiti Kebangsaan Malaysia were followed during animal handling.
Micro-Computed Tomography Analysis {#S0002-S2003}
----------------------------------
The skeletal microstructure of the rats was assessed using Skyscan 1076 Scanner (Skyscan, Kartuizersweg Kontich, Belgium). Three-dimensional trabecular and cortical bone microstructures were analysed using CTAn software (SkyScan, Kartuizersweg Kontich, Belgium). The volume of interest for trabecular bone was the metaphyseal region of the distal femur, starting from 1.5 mm to the lowest margin of the femoral epiphyseal growth plate. A total of 200 slices were obtained from this point onwards and extending proximally. For cortical bone, another 200 slices volume of interest was selected at the mid-diaphysis region of the femur located 7.0 mm from the distal growth plate. The scans were performed at 70 kVp and 100 µA with high resolution. The trabecular parameters derived included bone volume (BV/TV), trabecular thickness (Tb.Th), trabecular number (Tb.N), trabecular separation (Tb.Sp), structural model index (SMI) and connectivity density. The cortical parameters derived included cortical thickness (Cr.Th), cortical area (Cr.Ar), tissue area (Tt.Ar) and the ratio of cortical area to tissue area (Cr.Ar/Tt.Ar).
Bone Cellular Histomorphometry {#S0002-S2004}
------------------------------
The right femurs were cleaned of soft tissue after harvesting and decalcified in phosphate-buffered 10% formalin solution with ethylenediaminetetraacetic acid (5.5%) for 2 months. Then, the bone samples were processed into paraffin blocks and sectioned at 5 µm longitudinally with a microtome (Leica Biosystem, Leica RM2235, Nussloch, Germany). The bone slides were rinsed with xylene and rehydrated with a series of alcohol solutions and stained with haematoxylin for 10 min. Next, they were rinsed with bluing reagents for 10 s and stained with eosin for 5 min. Finally, the slides were dehydrated with a series of alcohol solutions and mounted for microscopic observation.
The slides were examined using a light microscope and analysed using a cellSens Standard software (Olympus Corp, Tokyo, Japan). The region of interest is 1.5 mm below the lowest margin of the epiphyseal growth plate of the distal femur. A Weibel grid of 21 lines and 42 points was used for quantitative measures of the histological features. The lines of the grid intersecting with the trabecular bone, bone cells or osteoid were defined as bone (BS), osteoblast (Ob.S), osteoclast (Oc.S) or osteoid surface (OS). The points of the grid intersecting with the trabecular bone or osteoid were defined as bone (BV) or osteoid volume (OV).[@CIT0027] The histomorphometric parameters derived from this method include osteoblast surface (Ob.S/BS), osteoclast surface (Oc.S/BS), eroded surface (ES/BS), osteoid surface (OS/BS), osteoid volume (OV/BV).
Bone Mechanical Strength Test {#S0002-S2005}
-----------------------------
The mechanical strength of the femur was tested using a 3-point bending test with a precision universal tester (Autograph AG-10kNG; Shimadzu, Kyoto, Japan) with Trapezium X material testing operation software. In this test, the distal and proximal ends of the femur were supported by rounded edge-free notches with 10 mm distance in between. A blunt-end aluminium roller stamp was lowered gradually (5 mm/min) until the strength of 1 N was achieved. The study was terminated automatically when a loss of force \>20 N or a linear change of 2 mm was identified. The Trapezium X software received the data and calculated the load (N), displacement (mm), stiffness (N/mm), stress (N/mm^2^), strain (%) and Young's modulus (N/mm^2^).
Statistical Analysis {#S0002-S2006}
--------------------
Normality of data was determined using the Shapiro--Wilk test. All data were normally distributed and were analysed using parametric tests. Comparison of body weight adopted a time × treatment design and thus was analysed using mixed-design ANOVA with small effect analysis. Other parameters which involved end-point measurement were tested using one-way ANOVA with post-hoc pairwise comparison (Tukey or Dunnett T3). All data were presented as mean ± standard error of the mean. Statistical significance was considered at p \< 0.05. Statistical analysis was conducted using Statistical Package for Social Sciences (IBM, Armonk, USA).
Results {#S0003}
=======
Body weight significantly changed with time (p\<0.001), but the interaction between time and treatment was not significant (p=0.166). All groups, except the PPI+CP group, experienced a significant increase in body weight with time (p\<0.05). However, no significant difference in body weight was found among the groups at baseline, Day 30 and Day 60 of the experiment (p\>0.05) ([Figure 1A](#F0001){ref-type="fig"}). In addition, no significant difference in femoral weight, length and diameter was found among the groups (p\>0.05) ([Figure 1B](#F0001){ref-type="fig"}--[D](#F0001){ref-type="fig"}).Figure 1Body weight (**A**), femoral weight (**B**), length (**C**) and diameter (**D**) of the rats in different groups. No significant inter-group difference was found (p\>0.05).**Abbreviations:** NC, normal control; PPI, pantoprazole/negative control; PPI+Ca, pantoprazole group treated with calcium; PPI+Ca+TT, pantoprazole group treated with calcium and annatto tocotrienol; PPI+CP, pantoprazole group treated with Caltrate Plus.
Three-dimensional reconstruction of the trabecular and cortical bones is depicted in [Figure 2](#F0002){ref-type="fig"}. Apparent deterioration of the trabecular network occurred in the rats treated with pantoprazole compared with the normal control. The destruction of the trabecular network was partially prevented in the pantoprazole-treated rats. However, the cortical bone showed no apparent changes among the groups ([Figure 2](#F0002){ref-type="fig"}). Quantitatively, BV/TV, Tb.N and Tb.Th decreased significantly, whereas Tb.Sp and SMI increased significantly after the pantoprazole treatment (p\<0.05). Calcium supplementation with and without tocotrienol prevented the changes in BV/TV, Tb.Th and Tb.Sp (p\<0.05). Calcium supplementation per se prevented the changes in SMI, and calcium supplementation combined with tocotrienol prevented the changes in Tb.N (p\<0.05). Caltrate Plus did not prevent the changes in trabecular bone caused by pantoprazole. Administration of pantoprazole with or without other treatments did not alter the cortical parameters (Ct.Th, Ct.Ar, Tt.Ar and Ct.Ar/Tt.Ar) of the rats ([Figure 3](#F0003){ref-type="fig"}).Figure 2Three-dimensional reconstruction of the trabecular and cortical microstructures of the femoral bones of the rats.**Abbreviations:** NC, normal control; PPI, pantoprazole/negative control; PPI+Ca, pantoprazole group treated with calcium; PPI+Ca+TT, pantoprazole group treated with calcium and annatto tocotrienol; PPI+CP, pantoprazole group treated with Caltrate Plus.Figure 3Trabecular and cortical microstructures of the rats. The trabecular indices evaluated include bone volume (**A**), connectivity density (**B**), structural model index (**C**), trabecular number (**D**), trabecular separation (**E**) and trabecular thickness (**F**). The cortical indices evaluated include cortical thickness (**G**), cortical area (**H**), tissue area (**I**) and cortical area/tissue area (**J**). Letter ^a^indicates a significant difference compared to the normal control; \*indicates a significant difference compared to the PPI group.**Abbreviations:** NC, normal control; PPI, pantoprazole/negative control; PPI+Ca, pantoprazole group treated with calcium; PPI+Ca+TT, pantoprazole group treated with calcium and annatto tocotrienol; PPI+CP, pantoprazole group treated with Caltrate Plus.
Bone cellular histomorphometry revealed no significant changes in Ob.N, Oc.N, ES/BS, Os/BS and OV/BV among the groups (p\>0.05) ([Figures 4](#F0004){ref-type="fig"} and [5](#F0005){ref-type="fig"}). A similar observation was obtained with the mechanical parameters, whereby the differences in load, displacement, stiffness, stress, strain and Young's modulus were not significant among all groups (p\>0.05) ([Figure 6](#F0006){ref-type="fig"}).Figure 4Bone cellular indices of the rats evaluated using the Weibel grid technique, which include osteoblast number (**A**), osteoclast number (**B**), osteoid surface (**C**), osteoid volume (**D**) and eroded surface (**E**). No significant inter-group difference was found (p\>0.05).**Abbreviations:** NC, normal control; PPI, pantoprazole/negative control; PPI+Ca, pantoprazole group treated with calcium; PPI+Ca+TT, pantoprazole group treated with calcium and annatto tocotrienol; PPI+CP, pantoprazole group treated with Caltrate Plus.Figure 5Micrographs of the trabecular bones.**Abbreviations:** Ob, osteoblast; Oc, osteoclast; Os, osteoid.Figure 6Bone mechanical strength indices of the rats evaluated using the three-point-bending test, which include load (**A**), stress (**B**), displacement (**C**), strain (**D**), stiffness (**E**) and Young's modulus (**F**). No significant inter-group difference was found (p\>0.05).**Abbreviations:** NC, normal control; PPI, pantoprazole/negative control; PPI+Ca, pantoprazole group treated with calcium; PPI+Ca+TT, pantoprazole group treated with calcium and annatto tocotrienol; PPI+CP, pantoprazole group treated with Caltrate Plus.
Discussion {#S0004}
==========
Pantoprazole caused limited adverse changes in the trabecular microstructure of the femur of the rats but did not alter the cortical bone microstructure, cellular indices and mechanical strength. Calcium supplementation with or without annatto tocotrienol prevented the adverse changes in trabecular bone caused by pantoprazole. However, all treatment regimes did not affect the other skeletal indices of the rats receiving pantoprazole.
In this study, pantoprazole impaired the trabecular microstructures of the rats by reducing BV/TV, Tb.N and Tb.Th while increasing Tb.Sp and SMI. However, it exerted no effects on the cortical microstructures. The structure of trabecular bone provides a large surface-to-volume ratio for bone remodelling to occur; therefore, rapid changes can be observed with external stimuli like pantoprazole. By contrast, Matuszewska et al (2016) did not detect any changes in bone microstructure of Wistar rats treated with pantoprazole (3 mg/kg for 84 days).[@CIT0028] Similarly, Veggar et al (2017) did not observe changes in femoral trabecular and cortical structures of mice with impaired mobility and treated with pantoprazole (100 mg/kg bw daily, 3 weeks). However, they reported a marginally higher lumbar Tb.Th in these mice compared with the untreated mice.[@CIT0029]
In the present study, calcium supplementation with or without annatto tocotrienol prevented the degenerative changes in the trabecular microstructure of the rats treated with pantoprazole in this study. A previous study found that dietary calcium supplementation prevents the cadmium-induced deterioration of femoral microstructure (Tb.N and Cr.Th) in rats.[@CIT0030] In addition, calcium supplementation through drinking water prevents the destruction of trabecular structure in male rats due to testosterone deficiency induced by buserelin.[@CIT0018],[@CIT0031] By contrast, calcium supplementation does not enhance the trabecular bone structure in lactating rats.[@CIT0032] Surprisingly, annatto tocotrienol did not act synergistically with calcium carbonate in this study to further enhance its effects on bone microstructure. A previous report indicated that annatto tocotrienol prevents the deterioration of bone microstructure in animal models of bone loss induced by testosterone deficiency,[@CIT0018],[@CIT0019],[@CIT0031] oestrogen deficiency[@CIT0033] and metabolic syndrome.[@CIT0017] The results of the present study suggest that the bone anabolic effects of annatto tocotrienol are limited and overshadowed by the effects of calcium supplementation.
Bone cellular histomorphometry quantifies the number of osteoblasts and osteoclasts as well as the consequence of their actions, such as osteoid synthesis and bone erosion.[@CIT0034] In the present study, pantoprazole did not affect the bone cell numbers, osteoid surface/volume and eroded surface of the rats. Similarly, a previous study found that the osteoclast surface of mice with impaired mobility does not change with pantoprazole treatment.[@CIT0029] However, pantoprazole decreases the viability of osteoclast-like cells derived from human peripheral blood mononuclear cells and increases the viability of human primary osteoblasts derived from excised femoral head.[@CIT0035],[@CIT0036] Gene expressions related to osteoclastogenesis and osteoblastogenesis are not altered with pantoprazole in these in vitro studies.[@CIT0035],[@CIT0036]
Calcium supplementation with or without annatto tocotrienol did not alter bone cellular histomorphometric indices in rats receiving pantoprazole. This observation is in agreement with a previous study showing that calcium in drinking water does not alter bone cell number and activities in male rats with testosterone deficiency.[@CIT0031] However, calcium supplementation reduces erosion surface but not osteoblast and osteoclast numbers in lactating rats.[@CIT0032] Previous studies found that annatto tocotrienol improves bone cellular indices in various animal models of bone loss.[@CIT0017],[@CIT0031],[@CIT0037],[@CIT0038] Interaction between annatto tocotrienol and pantoprazole might exist, which hinders the beneficial actions of annatto tocotrienol on bone cells.
Reduced bone strength is the ultimate consequence of osteoporosis. This study showed that pantoprazole treatment for 2 months did not affect the bone strength of the rats despite the degenerative changes in trabecular bone microstructures. The 3-point bending test was performed at the mid-diaphysis of the femur, which consists mainly of cortical bone. No significant change was observed in the cortical structure; thus, the lack of effect on bone strength is expected. Pantoprazole does not alter bone strength at the femoral diaphysis, femoral neck, lumbar vertebra and humeral diaphysis of mice with impaired mobility and treated with pantoprazole.[@CIT0029] At 200 mg/kg i.p. for 8 weeks, pantoprazole also did not change the bone strength in a model of posterolateral lumbar spinal fusion in female rats. A rat fracture-healing model showed that pantoprazole (100 mg/kg, i.p. for 5 weeks) reduces the bending stiffness at the fractured bone but not at the contralateral unfractured bone.[@CIT0039]
Calcium supplementation with or without annatto tocotrienol did not improve the biomechanical parameters of the rats taking pantoprazole in this study. Similarly, calcium supplementation of the rat dams does not improve bone strength of pups.[@CIT0032] This observation contradicted the findings of Mohamad et al (2018) that calcium in drinking water improves the femoral load and Young's modulus of rats with testosterone deficiency induced by buserelin.[@CIT0018] Annatto tocotrienol also improves the bone strength of the rats in some models of osteoporosis[@CIT0017],[@CIT0018] but is ineffective in this case. Our observation is similar to a previous finding that annatto tocotrienol (60 mg/kg oral for 2 months) could not improve bone strength of orchidectomised male rats.[@CIT0040] The discrepancy between bone strength and microstructure may stem from the different regions of femur measured.
Surprisingly, Caltrate Plus, which contains calcium, trace elements and vitamin D~3~, was not effective in improving any of the skeletal health parameters. Likewise, annatto tocotrienol did not enhance the effects of calcium supplementation. We suspect a potential drug interaction between pantoprazole and tocotrienols or vitamin D. These agents are catabolised, at least partly, by cytochrome 3A4 (CYP3A4). Pantoprazole upregulates the expression of CYP3A4,[@CIT0041] which catabolises tocotrienols and vitamin D,[@CIT0042],[@CIT0043] thus reducing the level and effectiveness of these compounds in the rats. This speculation should be validated in future studies.
Several limitations should be addressed in this study. Firstly, we did not determine the effects of pantoprazole on the gastric pH, serum and bone mineral and cobalamin levels, which may be related to its effects on skeletal health. Secondly, the serum levels of tocotrienols and vitamin D were not determined; hence, we do not know the absorption and bioavailability of these compounds. Thirdly, the parathyroid level of the rats was not quantified; therefore, we did not know whether pantoprazole could cause calcium deficiency in this model and whether it was averted with the treatments. Fourthly, the serum bone formation and resorption markers were not evaluated to highlight the effects of treatment on bone turnover. These assays can be incorporated in future studies to provide a comprehensive view of the mechanism of calcium in preventing bone loss induced by pantoprazole.
Conclusion {#S0005}
==========
Pantoprazole causes adverse effects on trabecular bone microstructure, but it does not affect bone strength or cellular histomorphometric indices. Calcium supplementation with or without annatto tocotrienol prevents the degenerative changes of pantoprazole on trabecular bone microstructure. Annatto tocotrienol does not enhance the effects of calcium significantly in this regard. Therefore, chronic users of proton pump inhibitors may consider calcium supplementation as a pharmacological strategy to prevent bone loss.
We thank Universiti Kebangsaan Malaysia for funding this study (FF-2018-404). We also thank American River Nutrition for providing the annatto tocotrienol used in this study.
Disclosure {#S0006}
==========
The authors declare no conflicts of interest in this work.
| {
"pile_set_name": "PubMed Central"
} |
Background {#Sec1}
==========
In 2013, there were over 3 million breast cancer survivors in the USA, and with a 5-year relative survival rate of 89.7%, this number will continue to increase \[[@CR1]\]. However, survivors often encounter physiological and psychological problems that may influence long-term prognosis, such as weight gain, decreased physical activity, and decreased quality of life (QOL). Research shows that 50--96% of breast cancer patients gain significant weight during and after therapy \[[@CR2]\]. There is extensive evidence suggesting that being overweight or obese may increase risk of breast cancer recurrence, comorbidities, second cancers, and mortality \[[@CR3]\]. It has been previously reported that obese breast cancer survivors exhibited anywhere from 30 to 540% increased risk of death \[[@CR4]\].
Some suggested mechanisms involved in post-diagnosis weight gain are decreased physical activity (PA) as well as chemotherapy-related changes in metabolism \[[@CR2], [@CR5]--[@CR8]\]. There is an association between PA after diagnosis and long-term prognosis, with decreases in recurrence risk of 50% in women who walked 3 to 5 h per week compared to inactive women \[[@CR9]\] and significant decreases in mortality risk in survivors who walked 2 to 3 h per week \[[@CR10]\]. PA combined with a dietary intervention designed to produce weight loss in overweight and obese survivors may reduce risk of disease recurrence, morbidity, and mortality \[[@CR11]\].
There are several biological mechanisms underlying the relationship between obesity and breast cancer recurrence risk. Studies suggest that insulin, insulin-like growth factors, inflammatory status, adipokines, oxidative stress, vitamin D status, and estrogens may have a role in breast cancer recurrence and long-term prognosis \[[@CR12]\]. Increased levels of fasting insulin and insulin-like growth factors have been associated with distant recurrence and death in breast cancer survivors \[[@CR13]\]. C-reactive protein (CRP) is a marker of systemic inflammation and studies have linked increased pre-treatment CRP levels with decreased survival \[[@CR14], [@CR15]\] though this finding has not been consistently shown in other studies \[[@CR16]\]. Leptin and adiponectin are adipokines that may influence recurrence risk \[[@CR17]\]. Women with breast cancer may have dysregulation in several circadian systems, including hormonal rhythms \[[@CR18]\] such as secretion of cortisol and melatonin. A lack of normal diurnal cortisol variation has been associated with early mortality in women with metastatic breast cancer \[[@CR19]\] and low levels of melatonin have been associated with poor breast cancer prognosis as measured by tumor aggression \[[@CR20]\], differentiation \[[@CR21]\], and progression \[[@CR22]\]. Elevated levels of F2-isoprostanes, a biomarker of whole-body oxidative stress \[[@CR23]\], have been associated with increased risk of breast cancer among overweight and obese women \[[@CR24]\]. However, it is not clear whether high levels of F2-isoprostanes are associated with increased risk of recurrence. Also of interest has been the relationship between Vitamin D status and breast cancer risk and progression. Adequate vitamin D status has been inversely associated with risk of breast cancer in prospective studies \[[@CR25], [@CR26]\], and a recent review has also reported a consistent inverse association between increased levels of vitamin D and decreased risk of progression and mortality of breast cancer \[[@CR27]\]. Although more conclusive research is needed to establish a link between vitamin D levels and breast cancer prognosis, observational studies suggest that higher levels of vitamin D are associated with improved survival \[[@CR28]\]. Most of the biomarkers mentioned above, with the exception of vitamin D, seem to be elevated in overweight and obese individuals. Thus, it would be expected that a diet and exercise intervention aimed at decreasing body weight would reduce the levels of these markers, which could decrease the risk of cancer recurrence.
Cancer diagnosis and treatment are also often associated with decreased quality of life (QOL) and may adversely affect psychosocial factors such as depression, anxiety, stress, self-esteem, psychological and emotional well-being, and fatigue that may last for long periods of time regardless of treatment outcome \[[@CR29]--[@CR32]\]. Physical activity and weight loss interventions after breast cancer diagnosis have mostly led to improved quality of life, reduced fatigue and depression, and improved sleep quality \[[@CR33]--[@CR35]\], although a recent clinical trial found an increase in depressive symptoms after weight loss \[[@CR36]\].
Given the effects of obesity and a sedentary lifestyle on recurrence, comorbidities, quality of life, and mortality of breast cancer survivors, the purpose of this pilot study was to explore the effects of two weight loss interventions, a calorie restriction feeding plus exercise based weight loss intervention (CR) and a weight management counseling intervention (WM), on body composition, fitness, cancer-related biomarkers and quality of life among overweight and obese breast cancer survivors. When it comes to weight loss interventions, it is well established that those interventions aiming at behavioral change towards different lifestyle components are the most effective \[[@CR37]\]. In breast cancer survivors, recent systematic reviews have shown similar findings \[[@CR38], [@CR39]\]. Our primary objective was to determine whether a significant weight loss, of approximately 10% over a short period of time, which was the target weight loss of CR, would lead to beneficial changes in the biological and psychosocial parameters described above. Given that most behavioral weight loss interventions do not typically achieve greater than 7% weight loss over a 6-month period, we designed a feeding intervention that would allow us to look at the effect of a greater weight loss over a short period of time.
Methods {#Sec2}
=======
Study design {#Sec3}
------------
We conducted a parallel-group randomized controlled trial where 21 overweight and obese breast cancer survivors were randomized, via a random number generator computer software, into one of two intervention groups: a calorie-restricted feeding plus exercise group (CR) or a weight management counseling group (WM). The rationale for choosing the two treatments was based on our primary objective. As we were interested in determining whether a significant weight loss of at least 10% would lead to significant changes in the biological and psychosocial parameters described above, it was not realistic to expect that a lifestyle behavioral modification would lead to that much weight loss over 12 weeks. Therefore, the CR group was designed to control the feeding of participants by providing all their meals for 12 weeks, as well as to include an exercise prescription to meet with a personal trainer twice per week during these 12 weeks. The weight management counseling intervention was designed as the comparison group because it was based on behavioral and lifestyle modification, which is considered the standard of practice.
Participants were studied for a total of 18 weeks, consisting of 12 weeks of active intervention plus a 6-week follow-up period. Body weight was measured at baseline; weeks 6, 12, and 18 and body composition was evaluated at baseline and week 12. The American College of Sports Medicine (ACSM) recommends a minimum of 15--20 weeks is needed to measure efficacy of a fitness program \[[@CR40]\], thus fitness level was measured at baseline and week 18. Due to the nature of the interventions described, participants and study staff were not blinded to treatment allocations. The only blinding that took place was that of study staff conducting laboratory analysis of biological parameters.
Study participants {#Sec4}
------------------
Overweight and obese breast cancer survivors were recruited in Minneapolis, Minnesota for this pilot study between January and August 2009 via emails sent to breast cancer survivors affiliated to the Love/Avon Army of Women ([www.armyofwomen.org](http://www.armyofwomen.org/)) and via advertisements posted at the Fairview University Medical Center and Breast Center. Interested women contacted study coordinators through provided phone or email information on recruitment materials. If women were determined eligible after an initial screening interview, a screening clinic visit for verification of eligibility status was scheduled. Inclusion criteria included: postmenopausal, defined as experiencing at least 12 months without a menstrual period, diagnosed with operable invasive breast cancer (TNM staging system T1-3, pN0-pN3, M0) and treated with mastectomy or with lumpectomy and radiation, completed all surgery, radiation, and systemic chemotherapy at least 3 months prior to enrollment, BMI ≥ 27 kg/m^2^, less than seven servings of alcohol/week, willing to be randomized into either group, not planning to move away from the area during the period of the study, and non-smoker. Exclusion criteria included serious illness requiring medical treatment, inability to participate in physical activity due to severe disability, history of schizophrenia, psychosis or untreated major depression, unwilling to commute to study site once/wk, and failure to provide written informed consent. This study was approved by the University of Minnesota Institutional Review Board. Written informed consent was obtained from all participants prior to beginning any study activities. (ClinicalTrials.gov Identifier: NCT02940470).
Intervention {#Sec5}
------------
Women were randomized via a random number generator computer software to either a weight management counseling group (WM) or a 1000-calorie deficit feeding plus exercise group (CR). Prior to randomization, a registered dietitian assessed each participant's current dietary intake and physical activity level in order to determine total energy expenditure, which was estimated as the calculated basal energy expenditure adjusted for reported physical activity and dietary intake. An energy prescription of 1000-kcal deficit from estimated total energy expenditure was given to each woman.
Weight management counseling group (WM) {#Sec6}
---------------------------------------
Women randomized into WM participated in weekly 1-h weight management classes supervised by a registered dietitian for 12 weeks. Individualized guidelines for the energy-restricted diet, menu plans, and recommendations for increased physical activity levels were prescribed during the first class. The remaining classes covered topics related to short-term and long-term weight loss, including exercise and behavior modification.
Calorie restricted diet plus exercise group (CR) {#Sec7}
------------------------------------------------
Women randomized into the diet plus exercise group were provided all meals, freshly prepared, for 12 weeks. Meals were picked up daily Monday--Friday, with the Friday pick-up including meals for the weekend. The meals included breakfast, lunch, dinner, and a snack reflecting a deficit of 600--900 kcal per day (which when combined with estimated energy expenditure via the exercise program, equaled the 1000 kcal deficit per day). Macronutrient breakdown per day was as follows 55% of total energy from carbohydrates, 15% from protein, and 30% from fat, with 125--150 mg cholesterol per 1000 kcal, 12--15 g dietary fiber per 1000 kcal and providing 100% of the reference dietary intake for all essential vitamins and minerals.
Women randomized into CR were also given a progressive exercise intervention, which combined both aerobic and strength training, and included a membership to a fitness center. The exercise program was supervised by a certified trainer, who met with each participant twice per week for the first 4 weeks, and once per week thereafter. Exercise prescription was based on ACSM guidelines for weight loss and prevention of weight gain in overweight and obese adults \[[@CR41]\]. Recommendations specific for cancer survivors have since been published \[[@CR42]\] and our exercise intervention fit within these recommendations. The energy expended during exercise varied between 100--400 kcal per day in order to obtain a total energy deficit of 1000 kcal per day when combined with the dietary caloric deficit. Initially, participants exercised 3 days per week for 15--20 min each time at 60--70% of calculated heart rate maximum (220 − age). The frequency of exercise gradually increased throughout the period of the intervention until participants were exercising five to six times per week, for a total of 150--225 min per week. Progressive weight training sessions took place twice per week. A recent study found that in survivors with lymphedema, slowly progressive weight lifting had no significant effect on limb swelling and resulted in a decreased incidence of exacerbations of lymphedema, reduced symptoms, and increased strength \[[@CR43]\]. The weight training sessions were constantly monitored to ensure compliance and to make sure participants maintained their form. Participants were asked to wear heart rate monitors during their aerobic exercise sessions and keep logs of their aerobic and strength training workouts throughout the study.
Limited contact intervention {#Sec8}
----------------------------
After the first 12 weeks of the intervention, women in both groups were followed for an additional 6 weeks. Women in the CR group were instructed to continue both aerobic and weight lifting exercise at the recommended levels of 150--225 min per week of aerobic activity, and twice weekly weight lifting sessions. They also continued to keep a log of their exercise. Women in the WM were instructed to use the knowledge and skills they learned during the previous 12 weeks towards continuing weight management or further weight loss. Also, women in both groups were given printed materials containing information on a healthy diet for cancer survivors and a sample weekly menu to provide guidance if desired.
Anthropometrics and body composition {#Sec9}
------------------------------------
Body weight was assessed at baseline, weeks 6, 12, and 18 to the nearest 0.1 kg, using an electronic scale (Scale Tronix, White Plains, NY). Women in CR also weighed themselves each day they picked up their food (up to 5 days/week) and recorded weight in provided diet compliance logs. Height was measured at baseline without shoes to the nearest 0.1 cm (Scale Tronix, White Plains, NY). Body mass index (BMI) was calculated as the weight in kilograms divided by height in square meters (kg/m^2^). Body composition was assessed at baseline and week 12 using dual energy x-ray absorptiometry (DXA) (Lunar Prodigy, GE Medi, Madison, WI).
Fitness assessment {#Sec10}
------------------
A sub-maximal treadmill test was performed at baseline and week 18. After a 5-min warm-up, participants walked on the treadmill at a steady speed (2.0--3.0 miles per hour depending on participant ability), and percent grade on the treadmill was increased by 2% every 2 min until the participants reach 80% of their age-predicted maximum heart rate (max HR), defined as 220 − age. Heart rate was measured using Polar Heart Rate monitors (Polar Electro Inc., Woodbury, NY). This workload was converted into metabolic equivalents (MET) using a standard conversion formula \[[@CR44]\]. The test was performed by a certified exercise physiologist who was blinded to group assignment. This assessment was also used as a symptom-limited exercise stress test in order to detect any potential adverse effects of the intervention.
Sample collection {#Sec11}
-----------------
Blood and urine samples were obtained at baseline, and weeks 6, 12, and 18. Participants were asked to collect all of their urine output for 24 h separated into day time (7 am--10 pm) and night time (10 pm--7 am) urine, for measurement of cortisol and 6-sulfatoxymelatonin. Insulin, glucose, insulin-like growth factor (IGF)-1, IGF binding protein (IGFBP-3), F2-isoprostanes, 25-hydroxyvitamin D (25(OH)D), and inflammatory markers were measured in plasma or serum at all four time points, with the exception of F2-isoprostanes, which were measured at baseline and weeks 0, 12, and 18 and 25(OH)D, which was measured at baseline and week 12.
Laboratory measurements {#Sec12}
-----------------------
Free F~2~-isoprostanes were measured in plasma by a gas chromatography-mass spectrometry (GC-MS)-based method \[[@CR45]\] at the Molecular Epidemiology and Biomarker Research Laboratory (MEBRL), University of Minnesota, Minneapolis, Minnesota. Fasting blood glucose and plasma insulin levels were measured at the Fairview University Diagnostic Laboratories (Minneapolis, MN). Glucose levels were assessed using colorimetric reflectance spectrophotometry. Insulin levels were assessed by chemiluminescent immunoassay (Immulite, Diagnostic Products Corporation, Los Angeles, CA). The measure of insulin resistance was determined by the homeostatic model assessment (HOMA) index. The HOMA index was calculated by multiplying fasting plasma insulin (mmols/L) by fasting glucose (mmols/L) and then dividing by 22.5. HOMA correlates well (*r* = −0.83) with insulin sensitivity as measured by the gold standard euglycemic clamp. We also calculated insulin sensitivity using the quantitative insulin sensitivity check index (QUICKI = 1/\[log~(fasting\ insulin)~ + log~(fasting\ glucose)~\].
Commercially available Enzyme-Linked Immunosorbent Assays (ELISAs) were used to measure levels of IGF-1, IGFBP-3, leptin, adiponectin, C-reactive protein (CRP), urinary cortisol (in 24-h urine), interleukin-6 (IL-6), and 6-sulfatoxymelatonin (in overnight urine), a major metabolite of melatonin in urine. All analytes were assayed using kits purchased from R&D Systems Inc (Minneapolis, MN) with the exception of 6-sulfatoxymelatonin which was assayed by kit purchased from ALPCO Diagnostics (Salem, NH). Serum 25(OH)D was assayed at Heartland Assays (Ames, IA) by competitive chemiluminescence immunoassay (CLIA) using the DiaSorin LIAISON 25-OH Vitamin D Total assay \[[@CR46], [@CR47]\]. Intra- and interassay coefficients of variation were 6.6 and 2.8% for CRP, 8.2 and 6.1% for IL-6, 4.2 and 5.1% for adiponectin, 4/1 and 4.7% for leptin, 2.4 and 3.6% for IGF-1, 3.3 and 3.7% for IGFBP-3, 7.8 and 7.4% for 6-sulfatoxymelatonin, 3.3 and 2.3% for urinary cortisol, and 8.1 and 11.2% for 25(OH)D.
Quality of life {#Sec13}
---------------
Quality of life (QOL) will be assessed at baseline and weeks 6, 12, and 18 using the World Health Organization's quality of life assessment instrument WHOQOL-BREF \[[@CR48]\], which is a 26-item version of the WHOQOL-100 \[[@CR49]\].
Sample size {#Sec14}
-----------
Due to the pilot nature of this study, a sample size calculation was not performed. The initial aim of the study was to recruit between 20 and 25 participants because having 10 to 12 participants per group seemed like a sufficient sample size to inform our group of trends in the levels of the biomarkers of interest as a consequence of weight loss.
Statistical methods {#Sec15}
-------------------
Descriptive statistics were generated by cross-tabulation for categorical variables and by means for continuous variables. In order to compare the treatment groups at the four time points, analysis of variance with repeated measures was performed. Data for all continuous variables are expressed as mean (95% confidence interval).
Results {#Sec16}
=======
We recruited a total of 21 women for the study, and 20 women successfully completed all study procedures (Fig. [1](#Fig1){ref-type="fig"}). One participant dropped out of WM because her expectations were not met by the intervention. Baseline characteristics (means, standard deviations, and frequencies) of all participants are shown in Table [1](#Tab1){ref-type="table"}.Fig. 1Flowchart of participants in the study. CR = 1000-calorie deficit feeding and exercise intervention group; WM = weight management counseling group Table 1Baseline characteristics of study participants (*n* = 10 per group)CR\
(*n* = 10)WM\
(*n* = 10)Age (years)54.7 ± 8.458.4 ± 7.6Weight (kg)^a^85.5 ± 8.398.3 ± 19BMI (kg/m^2^)^a^31.5 ± 3.336.9 ± 7.7Body fat (%)49.4 ± 4.850.6 ± 3.7Bone mineral density (g/cm^2^)1.18 ± 0.111.26 ± 0.09Fitness level (METs)4.89 ± 1.484.80 ± 0.97Glucose (mg/dL)^b^80.7 ± 9.896.5 ± 21.8Insulin (mU/L)^b^9.0 ± 4.119.4 ± 20.725(OH)vitamin D (ng/mL)35.0 ± 9.433.1 ± 12Use of vitamin D supplements Yes43 No67Stage of breast cancer Stage I24 Stage II74 Stage III12Chemotherapy/radiotherapy Yes109 No01Endocrine therapy Aromatase inhibitors45 Selective estrogen receptor modulators43 None22Mastectomy Yes56 No30 Lumpectomy24Race White1010 Other00Values presented as mean ± SD, or frequencies*Abbreviations*: *CR* Calorie restricted diet plus exercise group, *WM* weight management counseling group
Every woman randomized into CR performed at least one weight lifting session per week with 40% adhering to the twice weekly weight lifting sessions for the entire 18 weeks. There was a high adherence rate for aerobic exercise training, with 70% of these women exercising for 150 (range 78--240) minutes per week. There was 100% compliance to the weekday food pick-ups, with no missed meals. In WM, adherence was 77.5% and it was calculated by dividing the number of sessions attended by all women by the total number of sessions given.
One participant in the WM was diabetic (baseline glucose and insulin levels were 146 mg/dL and 77 mU/L, respectively) and another participant in the same treatment group had impaired glucose tolerance (glucose and insulin levels were 116 mg/dL and 18 mU/L, respectively).
Table [2](#Tab2){ref-type="table"} shows body weight (kg), total body fat mass (TBFM, kg), total body fat percent, and total body lean mass (kg) by treatment group. There was improvement in these parameters in both groups, although the average weight loss in the WM group was largely driven by one participant who lost 12.9 kg (9.5% of her body weight).Table 2Body weights and body composition of breast cancer survivors (*n* = 10 per group)Week 0Week 6Week 12Week 18Body weight (kg) CR85.5 (77, 94)80.3 (71.8, 88.8)77.5 (69, 86)76.7 (68.1, 85.2) WM98.3 (89.8, 106.8)95.7 (87.2, 104.2)94.1 (85.6, 102.6)93.2 (84.6, 101.7)Body fat (%) CR49.6 (46.3, 53)--45.2 (41.9, 48.5)-- WM50.4 (47, 53.7)--49.6 (46.3, 53)--Body fat (kg) CR42.1 (35.7, 48.5)--35.4 (29, 41.8)-- WM46.1 (39.7, 52.5)--43.6 (37.2, 49.9)--Lean mass (kg) CR42 (37.9, 46)--41.6 (37.5, 45.7)-- WM45.2 (41.1, 49.3)--43.9 (39.9, 48)--Values displayed as mean (95% confidence interval)*Abbreviations*: *CR* calorie restricted diet plus exercise group, *WM* weight management counseling group
Women in CR showed an improvement in fitness level at week 18 (6.3 mets (95% CI: 5.4, 7.2) compared to baseline 4.9 mets (95% CI: 4, 5.8). This improvement in fitness level was not as evident in the WM group whose initial and week 18 fitness levels were 4.8 mets (95% CI: 3.9, 5.7) and 5.4 mets (95% CI: 4.5, 6.3), respectively.
Table [3](#Tab3){ref-type="table"} shows levels of glucose, insulin, insulin resistance, and IGF proteins over the four time points of the study. Table [4](#Tab4){ref-type="table"} shows levels of markers of inflammation, stress, and other hormones associated with breast cancer risk over time. Leptin, CRP and F2-isoprostanes showed improvement over time in the CR group.Table 3Levels of glucose, insulin and insulin related proteins in breast cancer survivors over time (*n* = 10 per group)Week 0Week 6Week 12Week 18Glucose (mg/dL) CR80.7 (72.1, 89.3)79.8 (71.2, 88.4)75.5 (66.9, 84.1)78.7 (70.1, 87.3) WM96.5 (87.9, 105.1)91.7 (93.1, 100.3)92.4 (83.8, 101)94.5 (85.9, 103.1)Insulin (mU/L) CR9 (2.7, 15.3)7.2 (0.8, 13.5)6 (0, 12.3)6.9 (0.6, 13.2) WM19.4 (13.1, 25.7)14.7 (8.4, 21)13.6 (7.3, 19.9)13.5 (7.2, 19.8)HOMA index CR1.8 (-0.3, 3.9)1.4 (0, 3.5)1.1 (0, 3.3)1.4 (0, 3.5) WM5.4 (3.3, 7.6)3.6 (1.5, 5.7)3.2 (1.1, 5.4)3.3 (1.2, 5.4)QUICKI CR0.6 (0.6, 0.7)0.7 (0.6, 0.8)0.8 (0.7, 0.9)0.8 (0.7, 0.9) WM0.5 (0.4, 0.6)0.6 (0.5, 0.7)0.6 (0.5, 0.7)0.6 (0.5, 0.7)IGF-1 (ng/mL) CR103.8 (80, 127.6)106.8 (83, 130.6)108 (84.3, 131.8)117.4 (93.7, 141.2) WM99.3 (75.5, 123.1)85 (61.3, 108.8)98 (74.3, 121.8)105.1 (81.4, 128.9)IGFBP-3 (ng/mL) CR1855 (1476, 2234)1833 (1455, 2212)1792 (1413, 2171)1740 (1362, 2119) WM1900 (1521, 2279)1578 (1199, 1957)1812 (1434, 2191)1827 (1448, 2205)Values displayed as mean (95% confidence interval)*Abbreviations*: *CR* calorie restricted diet plus exercise group, *WM* weight management counseling group, *HOMA* homeostasis model assessment, *QUICKI* quantitative insulin sensitivity check index, *IGF-1* insulin-like growth factor-1, *IGFBP-3* insulin-like growth factor binding protein-3 Table 4Levels of markers of inflammation, stress, and other hormones associated with breast cancer risk in breast cancer survivors over time (*n* = 10 per group)Week 0Week 6Week 12Week 18Leptin (ng/mL) CR47 (34.8, 59.3)21.5 (9.3, 33.8)19 (6.7, 31.2)26.2 (14, 38.5) WM53.8 (41.5, 66)37.3 (25, 49.5)36.8 (24.5, 49)38.2 (26, 50.5)Adiponectin (μg/mL) CR12.8 (9.3, 16.2)9.6 (6.2, 13)11.1 (7.7, 14.5)13.8 (10.4, 17.2) WM7.7 (4.3, 11.1)6.9 (3.5, 10.3)7.9 (4.5, 11.3)8.3 (4.9, 11.7)CRP (mg/L) CR6 (2.1, 9.8)4.3 (0.4, 8.1)3.2 (0, 7.1)3.8 (0, 7.7) WM5.2 (1.3, 9)4.4 (0.6, 8.2)7.4 (3.6, 11.3)5.9 (2, 9.7)Interleukin-6 (pg/mL) CR2.7 (1.8, 3.6)2 (1.1, 2.9)2.2 (1.3, 3.1)2.4 (1.6, 3.3) WM2.8 (1.9, 3.7)2.4 (1.5, 3.3)3.3 (2.4, 4.2)2.7 (1.8, 3.6)F2-isoprostanes (ng/mL) CR71.5 (52.5, 90.4)--50 (31.1, 68.9)52.9 (33.9, 71.8) WM81 (62, 100)--62 (43.1, 81)64.9 (46, 83.9)Cortisol (ng/mL) CR26.4 (18.3, 34.6)21.2 (13.1, 29.4)27.1 (19, 35.3)23.1 (14.9, 31.3) WM40.3 (32.1, 48.5)28.3 (20.1, 36.5)36.1 (27.9, 44.3)37 (28.8, 45.1)6-Sulfatoxymelatonin (ng/mg cr) CR8.1 (3, 13.3)8.6 (3.5, 13.8)5.8 (0.1, 11.6)10 (4.9, 15.2) WM15.8 (10.7, 21)13.1 (7.7, 18.5)11.6 (6.4, 16.7)16.9 (11.8, 22.1)Values displayed as mean (95% confidence interval)*Abbreviations*: *CR* calorie restricted diet plus exercise group, *WM* weight management counseling group, *CRP* C-reactive protein
Levels of 25(OH)D were measured at weeks 0 and 12 only. At week 0, levels of 25(OH)D were 32.9 (95% CI: 25.5, 40.3) and 35.2 (95% CI: 28.1, 42.2) in the WM and CR groups, respectively. At week 12, levels of 25(OH)D were 33.1 (95% CI: 26.1, 40.2) and 41.4 (95% CI: 34.4, 48.4) in the WM and CR groups, respectively.
Table [5](#Tab5){ref-type="table"} shows the overall quality of life (QOL) and specific QOL parameters over the four time points of the study.Table 5Quality of life and sleep scores of breast cancer survivors over time (*n* = 10 per group)Week 0Week 6Week 12Week 18Quality of life CR71.2 (66.1, 76.4)74.2 (69, 79.3)75.8 (70.8, 81)77 (71.6, 82.4) WM59 (53.6, 64.4)62.6 (57.2, 68)60.4 (54.7, 66.2)62.3 (54.7, 66.2)Physical health CR57.7 (50.9, 64.5)61.4 (54.6, 68.2)64 (57.2, 70.8)66.1 (58.9, 73.3) WM48.1 (40.9, 55.3)55.1 (47.9, 62.3)54 (46.4, 61.6)50 (42.4, 57.6)Psychological CR60.7 (53.5, 67.8)68.9 (61.7, 76)72.5 (65.3, 79.6)71.6 (64, 79.1) WM50.9 (43.3, 58.4)54.2 (46.7, 61.8)52.5 (44.5, 60.5)53.2 (45.3, 61.2)Social relationships CR78.8 (68.8, 88.8)80.6 (70.6, 90.6)78.7 (68.7, 88.7)79.2 (68.7, 89.7) WM62.4 (51.9, 72.9)63.3 (52.8, 73.8)59.4 (48.2, 70.5)69.5 (58.3. 80.6)Environment CR87.7 (80.6, 94.8)85.8 (78.7, 92.9)88.2 (81.1, 95.3)91 (83.5, 98.4) WM74.6 (67.1, 82)77.8 (70.3, 85.2)75.9 (68, 83.8)76.6 (68.7, 84.5)Sleep quality CR7.3 (5, 9.6)6 (3.7, 8.3)5.1 (2.8, 7.4)4.4 (2, 6.9) WM7.7 (5.2, 10.1)7.3 (4.9, 9.8)7.5 (4.9, 10.1)7.7 (5.1, 10.4)Values displayed as mean (95% confidence interval)*Abbreviations*: *CR* calorie restricted diet plus exercise group, *WM* weight management counseling group
Discussion {#Sec17}
==========
The primary objective of this randomized pilot trial was to determine whether a significant weight loss of approximately 10% would favorably change biological and psychosocial measures in breast cancer survivors. Our study examined the effects of a calorie restriction plus exercise feeding intervention (CR) compared to a weight management counseling group (WM) on weight, body composition, fitness, cancer-related biomarkers, and quality of life in overweight and obese breast cancer survivors. While several randomized controlled trials examining the effects of weight loss interventions on health outcomes of breast cancer survivors have been published and a few more are currently being conducted, many of these trials focused on change in weight as the primary outcome, and few have examined the impact of weight loss on psychosocial measures \[[@CR39]\]. Overweight or obese status, along with having a sedentary lifestyle, is thought to increase breast cancer risk and recurrence. Women in both groups successfully lost weight and percent body fat (%BF) during the trial. However, women in CR lost more weight and %BF compared to WM at weeks 6, 12, and 18. CR also had a significantly greater increase in fitness compared to baseline and also compared to WM. Both groups were able to maintain or continue to reduce body weight during the limited contact follow-up (weeks 12--18), although the CR participants showed greater weight loss and greater improvements in fitness level at week 18. To our knowledge, only one other randomized controlled trial of a weight loss intervention in this population resulted in a weight loss of 10% \[[@CR50]\]. It has been suggested that this weight loss level normalizes several metabolic parameters that are adversely affected by obesity and is therefore the recommended percent to lose by the National Institutes of Health to reduce risk of metabolic and cardiovascular diseases \[[@CR51]\]. Williamson and colleagues found that weight loss of ≥9.1 kg in overweight women resulted in a 25% reduction in all cause, cardiovascular, and cancer mortality \[[@CR52]\]. We observed an average loss of 9.6 kg in CR, which accounted for 11% of initial body weight. In addition, while other studies of weight loss reported a significant loss of lean mass, our study demonstrated that it is feasible to have substantial weight loss (≥10%) without concomitant loss of lean mass. Incorporating physical activity, and specifically weight training, into a weight loss program is key to maintaining or increasing muscle mass. Along with reduced adiposity and maintained lean mass, we found a significant 29.4% increase in fitness in CR. Research has shown that breast cancer survivors who engage in one or more hours/week of moderate physical activity have a lower risk of cancer recurrence ranging from 20 to 50%\[[@CR10], [@CR53]\]. In our study, 100% of the women reached a minimum of 1 h/week, 60% of the women averaged between 2--3 h/week, and 30% of the women averaged over 3 h/week.
There have also been few published studies to date looking at the effects of weight loss on a comprehensive number of biological markers such as the ones described here. Several randomized controlled trials to date have measured insulin and glucose, but few have measured adipokines or inflammatory markers other than CRP \[[@CR54]\]. Our study indicates that levels of IGF-1 may increase with weight loss, an expected finding as BMI has been inversely associated with IGF-1 levels \[[@CR55]\]. Levels of IGFBP-3 seemed to be lower as a consequence of weight loss. Studies have shown that high levels of IGFBP-3 predicted distant recurrence of breast cancer in postmenopausal women \[[@CR56]\] and high levels of IGFBP-3 have been found in tissue of breast tumors associated with poor prognosis \[[@CR57]\].
Obesity is consistently associated with vitamin D deficiency in adults \[[@CR58], [@CR59]\]. When it comes to the relationship between 25(OH)D levels and breast cancer risk, data from case-control studies indicate a statistically significant inverse relationship, whereas data from prospective studies do not support such relationship \[[@CR27]\]. Recent meta-analyses have reported statistically significant relationship between higher levels of 25(OH)D and improved breast cancer prognosis \[[@CR60]--[@CR62]\]. Levels of 25(OH)D increased in the CR group and this finding suggests that weight loss may beneficially affect vitamin D status in breast cancer survivors.
F2-isoprostanes are considered one of the best biomarkers of systemic in vivo oxidative stress \[[@CR63]\]. Systemic oxidative stress may play a role in progression of breast cancer. Breast cancer treatment may involve the generation of free radicals which induce oxidative stress and kill cancer cells; however, high levels of free radicals in the body may also adversely impact breast cancer prognosis and a recent case-control study looking at urinary levels of F2-isoprostanes in deceased breast cancer patients versus surviving patients found a significant inverse association between urinary levels of F2-isoprostanes and mortality \[[@CR64]\]. To our knowledge, this is the first study to assess levels of F2-isoprostanes in breast cancer survivors participating in a weight loss intervention. Weight loss resulted in decreases in plasma F2-isoprostanes in both groups. Based on findings from exercise intervention studies, it seems that exercise alone also decreases levels of F2-isoprostanes even in the absence of weight loss \[[@CR65], [@CR66]\], suggesting that interventions combining calorie restriction with exercise training may be the most effective to reduce F2-isoprostanes in obese individuals. Cortisol levels are often used as a surrogate of stress and previous studies have found disrupted circadian rhythms of diurnal cortisol in breast cancer patients \[[@CR19], [@CR67]\]. There were no clear changes in 24-h cortisol levels in this pilot study, and given the findings from previous studies, diurnal, rather than 24-h cortisol levels, may be considered in a future trial.
Similar to other studies of physical activity and weight loss interventions in survivors \[[@CR38]\], there were slight improvements in overall quality of life (QOL) in the CR group. However, this group also showed a decrease in sleep quality and social relationships over the study period. The CR intervention was very intense and involved a strict low-calorie diet regime, which may have been a reason why sleep quality and social relationships (the women may have reduced social activities in order to avoid food temptations, and since all meals were provided for them, they may have decreased activities that involved dining out) decreased.
Conclusions {#Sec18}
===========
Some of the strengths of our study include the randomized controlled study design, the inclusion of a limited contact follow-up period, the comprehensive array of variables measured, particularly F2-isoprostanes, which have not been measured in any previous trials, and the significant weight loss achieved in the CR group. Regarding the comprehensive array of variables measured, the rationale for assessing all these biomarkers was the reported associations between obesity or energy balance and changes in all these variables. Limitations of this study were an imbalance between the two groups with respect to baseline body weight, a highly-tailored feeding intervention, and the lack of a dietary assessment measurement to monitor dietary adherence in the WM group. Also, since we combined physical activity with a diet intervention, we cannot separate their individual effects, and we cannot determine the separate effects of the weight training and aerobic exercise. The authors acknowledge that a highly-tailored feeding intervention as the one described here does not provide breast cancer survivors with tools and resources needed to succeed in their weight management goals. However, the primary objective of this pilot trial was to determine whether a significant weight loss led to changes in biological and psychosocial parameters over a short period of time. Indeed, we found that a 10% decrease in body weight and a 16% decrease in fat mass were accompanied by improvements in several biomarkers that may translate into lower recurrence, morbidity, and mortality risks. Our next step will be to develop an intervention that combines elements of the CR group with elements of the WM group resulting in behavioral modification and personalized nutrition counseling to maximize weight loss and help breast cancer survivors maintain body weight in the long term. Future larger randomized controlled trials are needed to determine how to sustain long-term weight loss and fitness and how these can significantly affect prognosis of breast cancer and mortality from a psychosocial perspective as well as a biological perspective, particularly considering systemic biomarkers such as plasma F2-isoprostanes, C-reactive protein, insulin, IGF-1, IGFBP-3, and 25(OH)-vitamin D. It is also important to determine whether there is an optimal level of weight loss and exercise level that should be sustained over time to reduce recurrence and improve prognosis in breast cancer survivors.
%BF
: Percent body fat
BC
: Breast cancer
CR
: Calorie-restricted diet and exercise intervention
CRP
: C-reactive protein
HOMA
: Homeostatic model assessment (measure of insulin resistance)
IGF-1
: Insulin-like growth factor 1
IGFBP-3
: Insulin-like growth factor binding protein 3
IL-6
: Interleukin 6
METs
: Metabolic equivalents
PA
: Physical activity
QOL
: Quality of life
TBFM
: Total body fat mass
TBLM
: Total body lean mass
WM
: Weight management counseling intervention
Not applicable.
Funding {#FPar1}
=======
This work was supported by the National Cancer Institute, U54 CA116849, the National Center for Research Resources, M01-RR00400, and the National Research Service Award, T32CA132670.
Availability of data and materials {#FPar2}
==================================
The datasets generated during the current study are not publicly available due to privacy issues but are available from the corresponding author on reasonable request.
AA, SR, and MK conceptualized and designed the study intervention. BK, AA, and SR conducted the study interventions and participated in acquisition of the data. AA conducted the statistical analyses. AA and BK drafted the manuscript. SR and MK revised the manuscript. All authors read and approved the final manuscript.
Competing interests {#FPar4}
===================
The authors declare that they have no competing interests.
Consent for publication {#FPar5}
=======================
Not applicable.
Ethics approval and consent to participate {#FPar6}
==========================================
This study was approved by the University of Minnesota Institutional Review Board (IRB\#0807 M39681) and the University of Minnesota Cancer Protocol Review Committee (CPRC\# 2008NTLS107). Written informed consent was obtained from all participants prior to beginning any study activities. This study was registered by ClinicalTrials.gov Identifier: NCT02940470.
Publisher's Note {#FPar7}
================
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
| {
"pile_set_name": "PubMed Central"
} |
1. Introduction {#sec1}
===============
Tobacco production and consumption have risen dramatically in the developing world \[[@B1]\]. While smoking rates have declined in high-income countries, the public health burden of tobacco is shifting towards the developing world, where by 2030 more than 80% of the world\'s tobacco-related deaths will occur \[[@B2]\]. Coinciding with this shift to developing countries, health knowledge in these countries is increasing, albeit slowly in some places. While overall awareness of the health hazards of tobacco has improved in the last 15 years in China, it is still relatively poor. A household survey in China found that 81.8% of the population knew that smoking causes serious diseases, but fewer people realized the diseases that second hand smoke could present (64.3%) \[[@B3]\]. Surveys in Ghana, however, show comparatively low smoking prevalence, high awareness of health risks, limited exposure to tobacco advertising, and frequent efforts by smokers to quit \[[@B4]\].
There is evidence that the multinational tobacco industry appears to be targeting Asia and Africa as growth regions \[[@B5]\]. The Framework Convention on Tobacco Control (FCTC), to which 174 countries are currently parties, contains a number of key demand-reducing strategies, such as tobacco taxation, education about health effects (including health warnings on packages), removal of misleading product descriptors, and support for cessation. FCTC also addresses the product itself, and the World Health Organization has received advice from its Study Group on Tobacco Product Regulation on tobacco product testing, reporting requirements, and possible emissions regulation \[[@B6], [@B7]\]. The problems presented in developing countries will be multifold: to deal with the increasing public health burden, while implementing provisions of the FTC, including educating consumers about the harmful effects of cigarettes and regulating tobacco products.
Over the last five decades, as consumers have grown increasingly aware of the health hazards of smoking, tobacco companies have responded by designing and marketing seemingly lower tar and nicotine products that were positioned as less dangerous to health \[[@B8], [@B9]\]. However, the testing methodology (e.g., International Organization for Standardization (ISO) and Federal Trade Commission (FTC)) that depicted lower tar and nicotine levels was unrepresentative of human smoking behavior, therefore, labels such as "low tar" often presented on packs or in advertising were meaningless to consumers as health indicators \[[@B10]\]. To market lower tar and nicotine cigarettes, tobacco manufacturers designed their cigarettes with characteristics such as cigarette filters on the ends of rods, which are able to reduce the machine yields of tar and nicotine by 40--50% \[[@B11]\]. Additionally, ventilation holes, which appear as a ring of holes in the cigarette paper surrounding the filter, dilute tobacco smoke coming from the mouth end when smoked by a machine and further reduce tar and nicotine emissions \[[@B11]\]. However, when smoked by consumers, vents can be blocked by fingers and lips, or their effect is reduced by greater puffing effort, such that smokers inhale more tar and nicotine than would be predicted by machine testing \[[@B12]\].
Broadly speaking, cigarette emissions are predictable to a large degree from design features \[[@B13]--[@B15]\]. In light of the shifting public health burden of tobacco use toward the developing world, Calafat et al. \[[@B16]\] showed that cigarette emissions and design varied widely across WHO regions, with cigarettes sold in the Eastern Mediterranean, South East Asia, and Western Pacific Regions having higher tar and lower ventilation than those sold in the African, American, or European regions. O\'Connor et al. \[[@B17]\] examined the differences in cigarette design characteristics in high-, middle-, and low-income countries, with the general trend being that as country income group increased, cigarettes sold became more highly engineered and the nominal emission levels decreased \[[@B17]\]. All cigarettes from high-income countries had filters, compared with 95% of brands in middle-income and 86% of brands in low-income countries, and among these, the proportion having ventilated filters was 95% in high-income countries, 87.5% in middle income countries, and 44.4% in low-income countries. This current study seeks to replicate earlier findings relating cigarette design (and by extension, emissions) to country development grouping. More evidence from studies such as this one is needed in order for countries to implement meaningful regulation of tobacco, given the important links between cigarette design and smoke emissions \[[@B18]\].
2. Methods {#sec2}
==========
Methods for this project mirror a previous study by O\'Connor et al. \[[@B17]\], comparing cigarette design features of samples obtained from multiple low-, middle-, and high-income countries. Country income classification was based on the World Bank\'s Gross National Income per capita data \[[@B19]\]. The current study analyzed cigarettes from 11 countries (*N* = 111 brands) purchased between 2008--2010 (see [Table 1](#tab1){ref-type="table"}). Collaborators in each country purchased popular brands of cigarettes based on sales and prevalence data within each country. Nepal was the only country used in the current study that was also included in the previous study, but these were two separate purchases in two separate years. Packs were then shipped to Roswell Park Cancer Institute where the cigarettes were catalogued and stored at −20°C until analysis. Before testing, cigarettes were conditioned for a minimum of 48 hours at 22 ± 2.0°C and 60 ± 2.0% relative humidity in an environmental chamber.
Product testing procedures followed those previously published by the same laboratory \[[@B14], [@B17]\]. After conditioning, five cigarettes were selected from each pack for physical analysis. Digital calipers were used to measure the length of the entire cigarette, the length and diameter of the tobacco rod, and the length and diameter of the filter. Filter and tobacco weight measurements were also taken using an analytical balance. The length of the tipping paper was then recorded and observed using a light box for the presence of vent holes. Tobacco moisture and dry weight were assessed using an HR83 Moisture Analyzer (Mettler-Toledo, Columbus, OH, USA). Filter ventilation and pressure drop were assessed using a KC-3 apparatus (Borgwaldt-KC, Richmond, VA). The level of porosity of the cigarette paper was measured using the vacuum method on a PPM1000M paper porosity device (Cerulean, Milton Keynes, UK). Tar and nicotine values were obtained from product packages where these were listed ([Table 1](#tab1){ref-type="table"}).
Data analysis was completed using Statistical Package for the Social Sciences Version 16.0 (SPSS Inc., Chicago, IL, USA). Basic descriptive statistics and analysis of variance (ANOVA) were used to compare product design features by country income grouping. Discriminant function analysis was used to examine how combinations of design features distinguished low-, middle-, and high-income countries. Stepwise linear regression was used to assess the influence of design features on labeled tar and nicotine values. In these regression models, ventilation was forced into the model given extant literature on its major influence on ISO yields \[[@B13], [@B14], [@B17]\], while other design features were entered using stepwise procedures (*P*-entry = 0.10, *P*-removal = 0.15). Since tar and nicotine yields were provided on packs for only seven countries (see [Table 1](#tab1){ref-type="table"}), the remaining countries were excluded from the regression analyses.
3. Results {#sec3}
==========
Nearly all the cigarettes tested were filtered cigarettes; 100% of cigarettes from both high- and middle-income countries had filters while 89% of cigarettes from low-income countries had filters. Among filtered cigarettes, only 16.0% in low-income countries had vent holes, compared to 65.5% in middle-income countries and 82.1% in high-income countries.
ANOVA analyses ([Table 2](#tab2){ref-type="table"}) revealed basic differences in physical cigarette parameters by income groups in terms of: cigarette length (*P* = 0.001), length of the tipping paper (*P* = 0.010), filter weight (*P* = 0.017), number of vent rows (*P* = 0.003), per-cigarette tobacco weight (*P* = 0.040), ventilation (*P* \< 0.001), and paper porosities (*P* = 0.008). The average percentage of cigarette ventilation differed significantly across income groups, with means of 7.49%, 15.34%, and 26.21% for low-, middle-, and high-income groups, respectively, (*P* \< 0.001). Rod diameter, filter diameter, tobacco length, and filter length were not shown to have significant differences by income groups.
A discriminant function analysis was used to examine how linear combinations of the panel of design features distinguished among low-, middle-, and high-income countries. Two functions were derived, accounting for 71.4% and 28.6% of variance, respectively. The first function \[*X* ^2^ (22) = 45.6, *P* \< 0.002\] maximally separated the high-income group from low and middle, while the second function \[*X* ^2^ (11) = 14.1, *P* = 0.167\] separated low- and middle-income groups but did not achieve statistical significance. Examination of the structure matrix suggested that ventilation, paper porosity, cigarette length, and rod diameter distinguished high from the remaining income group brands. Analysis of classification statistics showed that the discriminant functions correctly classified 56.3% of cases, ranging from 72.2% of the high-income brands to 43.5% of the middle-income brands and 69.9% of the low-income brands.
Stepwise linear regressions were done for all cigarettes with tar and nicotine values recorded on the pack. Per-cigarette weight, tipping paper, filter diameter, tobacco length, and paper porosity were all associated independently with tar yields, after ventilation was forced into the model (Adjusted *R* square = 0.852, see [Table 3](#tab3){ref-type="table"}). For nicotine, ventilation, tipping paper, filter weight, and filter length were the variables predicting nicotine yields (Adjusted *R* square = 0.774; see [Table 4](#tab4){ref-type="table"}). When stratified by income group, regression analyses found that a number of design features contributed independently to tar yields in high-income group countries, including ventilation (*P* \< 0.001), tipping paper (*P* = 0.015), number of vent rows (*P* = 0.009), per-cigarette weight (*P* \< 0.001), cigarette length (*P* = 0.055), and filter diameter (*P* = 0.004) ([Table 3](#tab3){ref-type="table"}). Middle-income countries had five variables accounting for differences in tar: ventilation, tipping paper length, filter length, number of vent rows in the tipping paper, per-cigarette weight, and cigarette length. In low-income countries ventilation and tobacco length primarily accounted for differences in tar. Ventilation was not statistically significant in both low- and middle-income countries (see [Table 3](#tab3){ref-type="table"}).
When examining correlates of nicotine yield stratified by income group, we found a broadly similar pattern of results ([Table 4](#tab4){ref-type="table"}). In all cases, ventilation and per-cigarette weight had the strongest independent associations with nicotine yield. Other contributors did differ across income groups: filter weight for the low-income (*P* = 0.078), tipping paper length (*P* \< 0.001) and filter length (*P* \< 0.001) for middle-income countries, and tobacco length for the high-income group countries (*P* = 0.046; see [Table 4](#tab4){ref-type="table"}).
4. Discussion {#sec4}
=============
This study largely replicates an earlier study \[[@B17]\] on the differences in cigarette characteristics between high-, middle-, and low-income countries. As expected, brands in higher income countries were engineered with filters and ventilation more commonly and at higher levels than in lower income countries. Ventilation is the main factor in the differences in tar and nicotine levels among cigarettes \[[@B13]--[@B15]\], and a majority of cigarettes in higher income countries employed ventilation to affect tar and nicotine. The main features that distinguished the high-income group brands from the lower income group brands were ventilation, paper porosity, cigarette length, and rod diameter, features which dilute the smoke and/or alter the amount of tobacco available for burning.
Patterns in variability in tar across products, by income group, were slightly different than for nicotine. While middle- and low-income countries shared ventilation and tobacco length accounting for most of the variability in tar across their cigarette products, in high-income countries a wider array of design features appeared to have independent influences on tar yields. The added length of the tipping paper is particularly interesting, as it sequesters otherwise smokeable tobacco from burning in a machine test, hence lowering yields \[[@B20]\]. In some countries, maximum tar levels, as measured by standardized smoking machines, have been set, such at the "10-1-10" upper limits for tar, nicotine, and carbon monoxide in the EU \[[@B21]\]. Consumers typically believe products with lower levels to be "healthier", even though the primary way those numbers are achieved is primarily through increased ventilation. The problem arises in that consumers can directly manipulate how much tar and nicotine they obtain from their cigarettes by blocking the vent holes in the filter or indirectly by taking larger puffs, which ventilation facilitates \[[@B11]\]. In either case, consumers receive more tar and nicotine than stated on the product while believing they have reduced their risks. Given the past history of light and mild cigarettes in developed countries, health officials in developing countries need to be cognizant of these design alterations that can contribute to seemingly "healthier" (i.e., reduced machine-measured tar and nicotine) products introduced into their markets in the coming years.
Parties to the World Health Organization Framework Convention on Tobacco Control (WHO FCTC) should see this study as further reason to consider cigarette design feature reporting when proposing measures in their countries that regulate the contents and emission of tobacco products (Article 9) and tobacco product disclosures by manufacturers (Article 10) \[[@B18]\]. As noted at COP-4, "Collecting data on product characteristics, such as cigarette design features, would help Parties improve their understanding of the impact these characteristics have on smoke emission levels, properly interpret measurements obtained and, more importantly, keep abreast of any changes to cigarette design features" \[[@B18]\]. In order to have effective product regulation, it is essential that governmental authorities have accurate information about the composition of those products to understand how manufacturers are complying with regulations \[[@B18]\].
A strength of the current study is its consistency with prior findings of statistically significant differences in cigarette design between high-, middle-, and low-income countries, even though completely different sets of cigarettes were tested from different high-, middle-, and low-income countries. The replication of the study further validates the differences in cigarette design between country income groups. At the same time, this study also shared the limitations of the first study \[[@B13]\], that is, the selected brands may not be fully representative of the market within each country. In addition to this, only brands from three low-income countries were tested in this study. Future research on this topic should incorporate more design data from lower income countries. Also, the lower income countries chosen may not be completely representative of all cigarette design from lower income markets around the world.
As expected with our hypothesis, the current study shows how different cigarette design characteristics are among high-, middle-, and low-income countries. Smokers in higher income countries have been misled with cigarettes that appear to be less hazardous and have highly engineered cigarette design; lower income countries could avert these same mistakes by immediately establishing ways to regulate product ingredients and design. Public health officials need scientific evidence to better understand cigarette design and function.
R. J. O\'Connor has served as a consultant to the World Health Organization and the US Food and Drug Administration with respect to tobacco product regulation.
This work was supported by a grant from the National Cancer Institute (P01CA138389).
######
Summary of countries, income groupings and brands studied.
Income group Number of brands Year pack was collected Primary manufacturer T & N label on pack
------------ -------------- ------------------ ------------------------- --------------------------------------- ---------------------
Bangladesh Low 5 2009 British American Tobacco No
Ghana Low 7 2008 British American Tobacco Yes
Nepal Low 16 2009 Other No
Argentina Middle 10 2008 Philip Morris Some packs
Malaysia Middle 13 2008 British American Tobacco Yes
Nigeria Middle 14 2008 British American Tobacco Yes
Thailand Middle 10 2008 Thailand tobacco Monopoly No
Uruguay Middle 8 2010 Other No
Canada High 7 2009 British American Tobacco Yes
Taiwan High 11 2008 Taiwan Tobacco and Liquor Corporation Some packs
UK High 10 2010 Imperial Tobacco Yes
######
ANOVA, basic differences in physical parameters by income group.
Income group Mean Standard error Minimum Maximum ANOVA *P*
------------------------------ -------------- -------- ---------------- --------- --------- -------------------- ---------
Cigarette length Low 79.45 1.24 66.57 84.16 *F*(2,108) = 7.010 0.001
Middle 82.77 0.34 78.61 93.87
High 83.80 1.04 71.79 99.08
Rod diameter Low 7.59 0.02 7.34 7.91 *F*(2,108) = 0.079 0.924
Middle 7.56 0.02 6.84 8.02
High 7.54 0.17 2.88 8.02
Filter diameter Low 7.58 0.02 7.20 7.77 *F*(2,108) = 0.326 0.723
Middle 7.60 0.02 6.81 7.85
High 7.50 0.18 2.55 7.82
Tobacco length Low 61.51 0.56 56.96 68.58 *F*(2,108) = 0.845 0.432
Middle 60.46 0.49 54.15 70.27
High 60.40 0.88 50.21 72.46
Length of tipping paper Low 25.70 0.76 18.39 32.65 *F*(2,105) = 4.805 0.010
Middle 27.98 0.48 15.32 36.40
High 28.93 0.87 18.94 38.30
Filter length Low 19.92 0.90 8.94 27.23 *F*(2,97) = 2.552 0.083
Middle 22.89 0.88 11.04 63.26
High 21.44 0.71 14.94 26.95
Filter weight Low 0.1029 0.0055 0.0458 0.1547 *F*(2,97) = 4.263 0.017
Middle 0.1178 0.0028 0.0600 0.1556
High 0.1172 0.0037 0.0895 0.1585
Number of vent rows Low 0.33 0.19 0.00 4.00 *F*(2,93) = 6.226 0.003
Middle 1.00 0.15 0.00 4.00
High 1.46 0.30 0.00 6.00
Per-cigarette tobacco weight Low 0.6928 0.0075 0.62 0.77 *F*(2,108) = 3.324 0.040
Middle 0.6581 0.0116 0.52 1.16
High 0.6486 0.0099 0.55 0.75
Ventilation (%) Low 7.49 2.3595 0.00 42.22 *F*(2,105) = 2.299 \<0.001
Middle 15.34 1.6746 0.00 39.54
High 26.21 3.3641 0.76 68.20
Paper porosity Low 35.01 3.16 15.74 80.05 *F*(2,106) = 5.18 0.008
Middle 44.09 2.40 15.88 81.57
High 48.47 2.21 31.42 72.41
######
Design features associated with ISO tar yields across all brands (a) and stratified by country income group (b).
######
\(a\) Overall
Final adjusted *R*-square value Model Standardized coefficients Sig.
--------------------------------- ---------------- --------------------------- ---------
0.852 Vent −0.722 \<0.001
Per-cig weight 0.475 \<0.001
Tipping −0.344 \<0.001
Filter diameter 0.233 0.004
Tobacco length −0.244 0.017
Paper porosity −0.150 0.070
######
\(b\) Stratified by income group
Final adjusted *R*-square value Model Standardized coefficients Sig.
---------------------- --------------------------------- --------- --------------------------- ---------
Low 0.561 Vent −0.037 0.909
Tobacco length −0.857 0.047
Middle 0.894 Vent 0.055 0.795
Tipping −2.139 \<0.001
Filter length 2.212 \<0.001
Number of rows −0.547 0.014
Per-cigarette weight 0.620 0.015
Cigarette length −0.391 0.072
High 0.956 Vent −0.897 \<0.001
Tipping −0.308 0.015
Number of rows 0.266 0.009
Per-cigarette weight 0.522 \<0.001
Cigarette length −0.204 0.055
Filter diameter 0.282 0.004
######
Design features associated with ISO nicotine yields across all brands (a) and stratified by country income group (b).
######
\(a\) Overall
Final adjusted *R*-square value Model Standardized coefficients Sig.
--------------------------------- -------- --------------------------- ---------
0.774 Vent −0.568 \<0.001
Tipping −0.752 \<0.001
Filter weight 0.937 \<0.001
Filter length −0.447 0.059
######
\(b\) Stratified by income group
Final adjusted *R*-square value Model Standardized coefficients Sig.
---------------------- --------------------------------- --------- --------------------------- ---------
Low 0.860 Vent −0.191 0.385
Per-cigarette weight 1.372 0.013
Filter weight −0.627 0.078
Middle 0.915 Vent −0.430 \<0.001
Per-cigarette weight 0.200 0.033
Tipping −2.310 \<0.001
Filter length 2.063 \<0.001
High 0.710 Vent −0.637 \<0.001
Per-cigarette weight 0.537 0.003
Tobacco length −0.333 0.046
[^1]: Academic Editor: Vaughan Rees
| {
"pile_set_name": "PubMed Central"
} |
GENOME ANNOUNCEMENT {#h0.0}
===================
In the marine environment, the [d]{.smallcaps}-amino acids synthesized by microbes can be released into the seawater ([@B1]). [d]{.smallcaps}-Amino acids ([d]{.smallcaps}-AAs), as the α-carbon enantiomers of [l]{.smallcaps}-amino acids ([l]{.smallcaps}-AAs), are commonly known as nonproteinogenic amino acids ([@B2]), and there are few reports about marine bacteria utilizing [d]{.smallcaps}-amino acids as carbon and nitrogen sources ([@B3]). A bacterial strain, JL2886, was isolated from deep seawater at 2,000-m depth of South China Sea collected during a cruise organized by the National Natural Science Foundation of China in August 2012. Phylogenetic analysis based on the 16S rRNA gene sequences revealed that strain JL2886 belongs to the genus *Phaeobacter*, *Roseobacter* clade ([@B4]). JL2886 can utilize many [d]{.smallcaps}-AAs as a sole source of carbon or nitrogen for growth (our unpublished data).
The complete genome sequencing of strain JL2886 was performed using the PacBio RS platform (Pacific Biosciences). A 10-kb library was sequenced using P4-C2 chemistry on two single-molecule real-time (SMRT) cells. The average read length was 6,734 bp, with a sequencing depth of 289×. The continuous long reads (CLR) were assembled *de novo* using SMRT Analysis version 2.1 and the protocol PacBio Hierarchical Genome Assembly Process (HGAP) ([@B5]). The consensus polishing process resulted in a highly accurate self-overlapping contig, as observed using Gepard dotplot ([@B6]), with a length of 4,061,725 bp, in addition to five self-overlapping 678,758-bp plasmids, and the overall G+C content of strain JL2886 was 61.52%. DNA methylation was determined using the RS Modification and Motif Analysis protocol within the SMRT Portal version 1.3.3.
The genome was annotated using Prodigal version 2.6 ([@B7]), RNAmmer version 1.2 ([@B8]), and ARAGORN version 1.2 ([@B9]), as implemented in the Prokka automatic annotation tool version 1.11 ([@B10]). The main chromosome contained 12 rRNA operons, 58 tRNAs, a predicted 3,913 protein-coding genes, and 744 protein-coding genes in plasmids.
The predicted open reading frames (ORFs) were annotated through comparisons with the NCBI-NR database and KEGG protein database ([@B11]). The functional genes were then identified by association with Clusters of Orthologous Groups (COGs) classification ([@B12]) and the KEGG pathway collection ([@B13]). A total of 3,690 proteins matched to known functions in the genome. There were 2,641 proteins classified to COG categories and 2,120 proteins classified to KEGG orthologs.
The genome sequences from strain JL2886 contained 361 predicted protein-coding sequences (CDSs) related to amino acid transport and metabolism, including five CDSs for putative [d]{.smallcaps}-AA transferases, four CDSs for putative [d]{.smallcaps}-AA racemase, and one CDS for putative [d]{.smallcaps}-AA oxidases. Strain JL2886 has robust [d]{.smallcaps}-AA catabolism ability.
Accession number(s). {#s1}
--------------------
The data from this complete genome sequence have been deposited at DDBJ/EMBL/GenBank under accession no. [CP015124](http://www.ncbi.nlm.nih.gov/nuccore/CP015124). The version described in this paper is the first version, CP015124.1.
**Citation** Fu Y, Wang R, Zhang Z, Jiao N. 2016. Complete genome sequence of the [d]{.smallcaps}-amino acid catabolism bacterium *Phaeobacter* sp. strain JL2886, isolated from deep seawater of the South China Sea. Genome Announc 4(5):00913-16. doi:10.1128/genomeA.00913-16.
This work was supported by the National Programme on Global Change and Air-Sea Interaction (grant GASI-03-01-02-05), NSFC (grant 91428308), National Key Basic Research Program of China 973 project (grant 2013CB955700), and Fundamental Research Funds for the Central Universities (grant 20720150136).
| {
"pile_set_name": "PubMed Central"
} |
1. Introduction
===============
Aqueous lithium-air batteries have been considered as a promising electrochemical energy storage system for electric vehicles. The theoretical energy density for the cell reaction of 4Li + 6H~2~O + O~2~ = 4(LiOH·H~2~O) is 1910 Wh·kg^−1^, which is significantly higher than 387 Wh·kg^−1^ for typical lithium-ion batteries with a carbon anode and LiCoO~2~ cathode. The aqueous lithium-air battery system, proposed by Imanishi and co-workers, is composed of a lithium metal electrode, a lithium ion conducting polymer electrolyte based on poly(ethylene oxide) (PEO)*~x~*-lithium bis(trifluoromethanesulfonyl)imide (LiTFSI), a water-stable lithium ion-conducting glass ceramics of Li~1+*x*+*y*~(Ti,Ge)~1−*x*~Al*~x~*P~3−*y*~Si*~y~*O~12~ (LTAP), an aqueous electrolyte with saturated LiCl, and a carbon air electrode \[[@B1-membranes-03-00298]\]. At present, the only acceptable water-stable lithium ion conducting solid electrolyte is LTAP, which exhibits a lithium ion conductivity of higher than 10^−4^ S·cm^−1^ at ambient temperature, and is stable in saturated LiOH and LiCl aqueous solution, but exhibits poor chemical stability towards lithium metal \[[@B2-membranes-03-00298]\]. To construct a water-stable lithium electrode (WSLE) based on LTAP requires a chemically stable lithium conducting electrolyte interlayer between the lithium metal and LTAP. Some candidates for this interlayer have been adopted, such as lithium nitride \[[@B3-membranes-03-00298]\], lithium phosphorous oxynitride \[[@B4-membranes-03-00298]\], and polymer electrolytes \[[@B5-membranes-03-00298]\]. The two former materials are generally prepared by vapor deposition, which involves considerable cost and makes the preparation of large sized cells difficult. In contrast, the polymer electrolyte is stable in contact with lithium metal and large sized sheets can be easily fabricated.
In our previous studies \[[@B6-membranes-03-00298],[@B7-membranes-03-00298],[@B8-membranes-03-00298]\], we have reported that the interface resistance between lithium metal and polymer electrolytes in a WSLE is the dominant part of the cell resistance and an important factor in initiating lithium dendrite formation. The addition of ionic liquids into PEO~18~LiTFSI reduced the interface resistance and suppressed lithium dendrite formation \[[@B7-membranes-03-00298]\]; however, high overpotentials for lithium deposition and stripping reaction were observed at high current densities. The WSLE should be operated at the highest current density possible with low overpotentials to prevent the formation of lithium dendrites \[[@B9-membranes-03-00298]\]. The addition of low-molecular weight plasticizers to the PEO-based electrolyte is expected to improve the interface properties between lithium metal and the polymer electrolyte, and enhance the lithium ion transport number of the polymer electrolyte. Kim *et al*. \[[@B10-membranes-03-00298]\] reported the use of low molecular weight oligomer ethers as a plasticizer to enhance the transport properties of polymer electrolytes, and Naoi *et al*. reported that the addition of poly(ethylene glycol) dimethyl ether (PEGDME) into propylene carbonate (PC) with LiClO~4~ suppressed dendrite formation \[[@B11-membranes-03-00298]\]. Moreover, Yoshida *et al*. reported that the oxidative stability of glyme molecules is enhanced by complex formation with LiTFSI in a molar ratio of 1:1 \[[@B12-membranes-03-00298]\]. Recently, Jung *et al*. \[[@B13-membranes-03-00298]\] demonstrated a Li/tetraethylene glycol dimethyl ether (TEGDME)-LiCF~3~SO~3~/carbon, O~2~ cell capable of operating over many cycles with capacity and rate values as high as 5000 mAh·g^−1^ and 0.5 A·g^−1^, respectively. We have previously reported an increase in the lithium ion transport number and decreases in the interface resistance of the lithium metal and polymer electrolyte by the addition of PEGDME to PEO~18~LiTFSI, where the mean molecular weight of PEGDME was 500 \[[@B14-membranes-03-00298]\]. Naoi *et al*. \[[@B11-membranes-03-00298]\] concluded that the Gibbs activation energy for the charge transfer reaction on the lithium metal surface in PC-PEGDME-LiClO~4~ decreased with a decrease in the molecular weight of PEGDME in the range of 90--400. In this study, the interface resistance between lithium and the polymer electrolyte, and the transport properties of the PEO~18~LiTFSI-*x*TEGDME (M = 222.28 g·mol^−1^) composite polymer electrolyte (CPE) are examined as a function of the amount of TEGDME (*x*). In addition, the electrochemical performance of a Li/PEO~18~LiTFSI-2TEGDME/LTAP/saturated LiCl aqueous solution/Pt, air cell is evaluated at 60 °C.
2. Results and Discussion
=========================
2.1. Evaluation of CPEs
-----------------------
The electrical conductivities of PEO~18~LiTFSI-*x*G4 (TEGDME abbreviated as G4 hereafter) were measured as a function of *x* (the molar ratio of G4 to ethylene oxide). [Figure 1](#membranes-03-00298-f001){ref-type="fig"} presents Arrhenius plots for the conductivity of PEO~18~LiTFSI-*x*G4. The conductivity increases significantly with increasing *x* up to 2 and there is little difference in conductivity between the samples with *x* = 2 and 3. The electrical conductivity of PEO~18~LiTFSI-2G4 at 25 °C is almost one order of magnitude higher and that at 60 °C is three times higher than that of PEO~18~LiTFSI. TEGDME loosens the coordination of lithium ions with EO units in the PEO matrix and thus enhances the mobility of ions, and could also enable lithium ions to decouple from ion pairs, as shown by Kriz *et al*. \[[@B15-membranes-03-00298]\]. The activation energies for conduction in PEO~18~LiTFSI-*x*G4 at high temperature decrease with increasing *x*. That of PEO~18~LiTFSI-2G4 was calculated to be 25.3 kJ·mol^−1^, in the range of 55 to 80 °C, which is lower than that for PEO~18~LiTFSI (38.8 kJ·mol^−1^) and that for PEO~18~LiTFSI-2.0 *N*-methyl-*N*-propylpiperdinium bis(fluorosulfonyl) imide (PP13FSI) (33.8 kJ·mol^−1^) \[[@B16-membranes-03-00298]\].
{#membranes-03-00298-f001}
The overpotentials between lithium metal and polymer electrolytes for lithium deposition and stripping generally increase with current density and undergo a sudden increase with the polarization period at a current density (limiting current density). When the ionic concentration in the vicinity of cathode drops to zero, the limiting current density is reached and dendrite growth is initiated (Sand time) \[[@B17-membranes-03-00298]\]. To operate the Li/PEO~18~LiTFSI/LTAP/Pt, air cell at high current density, it is necessary to enhance the limiting current density of lithium-ion conducting PEO-based polymer electrolytes (*I*~l~), which is proportional to the salt diffusion coefficient and lithium ion transference number, as shown by Equation (1): where *C*~o~ is the initial concentration, *e* is the elemental charge, *D* is the salt diffusion coefficient, *t*~a~ is the anion transport number, and *L* is the thickness of the electrolyte. The salt diffusion coefficient of PEO~18~LiTFSI-*x*G4 was estimated using the method proposed by Ma *et al*. \[[@B18-membranes-03-00298]\] as a function of *x*. Typical curves for the natural logarithm of potential *versus* time for the Li/PEO~18~LiTFSI-*x*G4/Li cells at 60 °C are shown in [Figure 2](#membranes-03-00298-f002){ref-type="fig"}, where the cells were polarized at 50 mV prior to the potential being interrupted. A distinct linear relation, which corresponds to the linear diffusion region as the concentration gradient of the cell relaxes, is observed for all cells after sufficient time. The salt diffusion coefficient (*D*) of PEO~18~LiTFSI-*x*G4 was calculated from the slope of the linear curves using Equation (2): The calculated *D* are summarized in [Table 1](#membranes-03-00298-t001){ref-type="table"} along with the electrical conductivity results. A maximum salt diffusion coefficient of 3.37 × 10^−7^ cm^2^·s^−1^ was determined for *x* = 2.0, which is almost one order of magnitude higher than that for PEO~18~LiTFSI (3.60 × 10^−8^ cm^2^·s^−1^) and PEO~18~LiTFSI-1.44PP13FSI (2.25 × 10^−8^ cm^2^ s^−1^) \[[@B16-membranes-03-00298]\], and four times higher than that for PEO~18~LiTFSI-18 wt % PEGDME (8.38 × 10^−8^ cm^−1^ s^−1^) \[[@B14-membranes-03-00298]\].
{#membranes-03-00298-f002}
Lithium ion transference numbers for PEO18LiTFSI-*x*G4 were measured using the method reported by Evans and Vincent \[[@B19-membranes-03-00298]\]. [Figure 3](#membranes-03-00298-f003){ref-type="fig"} shows a typical cell current decay curve upon application of a DC bias of 20 mV and impedance profiles for the Li/PEO18LiTFSI-2G4/Li cell. From these results, *t*~a~ was calculated using the Evans and Vincent equation, and the results are summarized in [Table 1](#membranes-03-00298-t001){ref-type="table"}. Significant increase of the lithium ion transport number (*i.e.*, decrease of the anion transport number) was observed by addition of G4 into PEO~18~LiTFSI. The lithium ion transport number for PEO~18~LiTFSI-2G4 is slightly higher than that for LiTFSI-2G4. Based on these results for *D* and *t*~a~, the limiting current density (*I*~l~) for PEO~18~LiTFSI-*x*G4 was calculated and the results are summarized in [Table 1](#membranes-03-00298-t001){ref-type="table"}, where the thickness (*L*) was 100 μm. PEO~18~LiTFSI-2G4 exhibits the highest *I*~l~ of 15.8 mA·cm^−2^, which is higher than that for PEO~18~LiTFSI by a factor of approximately 15. The high limiting current density of PEO~18~LiTFSI-2G4 suggests that a WSLE with PEO~18~LiTFSI-2G4 could be operated at a high current density.
membranes-03-00298-t001_Table 1
######
Ionic transport properties of PEO~18~LiTFSI-*x*TEGDME.
PEO~18~LiTFSI- *x*TEGDME 25 °C(×10^6^ S·cm^−1^) 60 °C(×10^4^ S·cm^−1^) *E*a (kJ·mol^−1^) *D* (×10^7^ cm^2^·s^−1^) *t* ~Li~ ^+^ *I*~l~ (mA cm^−2^) *L* = 100 μm
-------------------------- ------------------------ ------------------------ ------------------- -------------------------- -------------- --------------------------------- ------ ------ ------
*x* = 0 5.64 1.35 5.29 1.27 115.7 38.8 0.36 0.24 1.02
*x* = 1.0 28.0 12.0 10.7 4.60 99.8 30.7 1.05 0.43 3.97
*x* = 1.5 40.7 18.3 12.5 5.62 82.4 27.1 1.50 0.45 5.87
*x* = 2.0 60.8 32.8 16.5 8.91 78.2 25.3 3.37 0.54 15.8
*x* = 3.0 68.3 39.6 16.4 9.51 72.9 25.0 2.27 0.58 11.6
{#membranes-03-00298-f003}
Low and stable interface resistance between lithium metal and polymer electrolytes is essential for the electrochemical performance of WSLEs and to reduce lithium dendrite formation \[[@B7-membranes-03-00298]\]. [Figure 4](#membranes-03-00298-f004){ref-type="fig"} shows the impedance profile changes for the Li/PEO~18~LiTFSI-*x*G4/Li cells at 60 °C as a function of storage time. PEO~18~LiTFSI-G4 and PEO~18~LiTFSI-1.5G4 show a significant increase of interface resistance with storage time; the initial cell resistance of 69 Ω·cm^2^ increased to 141 Ω·cm^2^ for PEO~18~LiTFSI-G4 after 28 days. However, low interface resistance was obtained for PEO~18~LiTFSI-2G4 and PEO~18~LiTFSI-3G4. After 28 days, PEO~18~LiTFSI-2G4 and PEO~18~LiTFSI-3G4 showed interface resistances of 34 and 74 Ω·cm^2^, respectively, which are much lower than 253 Ω·cm^2^ for PEO~18~LiTFSI after storage for 28 days \[[@B16-membranes-03-00298]\]. The interface resistance behavior for PEO~18~LiTFSI-2G4 was unusual, where the interfacial resistance increased during the initial seven days and then decreased continuously to a value lower than the original. A similar change in the interface resistance was reported for the Li/PEGDME-LiTFSI/Li cell by Bernhard *et al*. \[[@B20-membranes-03-00298]\], which could be attributed to the solid electrolyte interphase (SEI) formation-re-dissolution process \[[@B21-membranes-03-00298]\]. These impedance profiles show a diminished semicircle, which is associated with the resistance of a passivation film (SEI) formed on the lithium electrode surface by the reaction of lithium with the polymer electrolyte and the charge transfer resistance. [Figure 5](#membranes-03-00298-f005){ref-type="fig"} shows the temperature dependence of the inverse of the passivation film resistance and the charge transfer resistance for the Li/PEO~18~LiTFSI-*x*G4/Li cells as a function of *x*. The activation energy for the inverse of the passivation film resistance (*R*~p~) decreased from 76.2 kJ·mol^−1^ for PEO~18~LiTFSI to 54.7 kJ·mol^−1^ for PEO~18~LiTFSI-2G4. A decrease was also observed for PEO~18~LiTFSI-G4 and PEO~18~LiTFSI-3G4, but it was not as significant as that for PEO~18~LiTFSI-2G4. The activation energies for the inverse of the charge transfer resistance (*R*~c~) decreased with increasing *x*. The lowest value of 63.4 kJ·mol^−1^ was calculated for PEO~18~LiTFSI-3G4, in comparison with 81.7 kJ·mol^−1^ for PEO~18~LiTFSI and 68.1 kJ·mol^−1^ for PEO~18~LiTFSI-2G4. These results suggest that the low-molecular weight oligomer ether G4 could reduce the resistance of the SEI and facilitate the charge transfer reaction at the interface. The former role is similar to other additives, such as nanofillers and ionic liquids, whereas the latter role is only performed by G4.
{#membranes-03-00298-f004}
{#membranes-03-00298-f005}
2.2. Electrochemical Performance of WSLEs
-----------------------------------------
The high lithium ionic diffusion coefficient and low interface resistance with lithium metal for PEO~18~LiTFSI-2G4 motivated us to examine its role as the protective layer in a WSLE. The impedance of the Li/PEO~18~LiTFSI-2G4/LTAP/sat. LiCl aqueous solution/Pt, air cell was measured at 60 °C, using a platinized platinum reference electrode. [Figure 6](#membranes-03-00298-f006){ref-type="fig"} presents the impedance spectra of this cell for various storage times. The open circuit voltage (OCV) was stabilized at 3.48 V after one week, which is comparable with that reported previously (3.43 V) \[[@B8-membranes-03-00298]\] and slightly lower than that of the calculated OCV (3.59 V). After 28 days, a stable cell resistance of 84 Ω·cm^2^ was obtained, which is comparable with those reported previously; the cell resistance for the Li/PEO~18~LiTFSI/LTAP/1 M LiCl/Pt, air cell was 539 Ω·cm^2^ \[[@B1-membranes-03-00298]\], 118 Ω·cm^2^ for the Li/PEO~18~LiTFSI-40 nm BaTiO~3~/LTAP/1 M LiCl aqueous solution/Pt, air cell \[[@B22-membranes-03-00298]\] and 130 Ω·cm^2^ for the Li/PEO~18~LiTFSI-1.44 *N*-methyl-*N*-propylpiperdinium-bis(trifluromethanesulfonyl)imide/LTAP/1 M LiCl aqueous solution/Pt, air cell \[[@B8-membranes-03-00298]\]. An equivalent circuit proposed in our previous study was utilized to analyze the impedance spectra, which consists of the total resistances of the polymer electrolyte and the LTAP plate (*R*~b~), the interfacial resistance between the polymer electrolyte and lithium metal electrode (*R*~f1~), the interfacial resistance between the polymer electrolyte and the LTAP plate (*R*~f2~), the charge-transfer resistance (*R*~c~), and the Warburg impedance (*W*~1~) \[[@B14-membranes-03-00298]\]. *R*~b~ was quite stable at around 38 Ω·cm^2^ during the storage period. The fitting results suggest that *R*~f2~ increased from 19.4 to 29.3 Ω·cm^2^, *R*~c~ decreased from 18.0 to 7.4 Ω·cm^2^, and *R*~f1~ increased from 3.8 to 5.8 Ω·cm^2^ over the 28-day storage period.
{#membranes-03-00298-f006}
The electrochemical performance of a WSLE protected by PEO~18~LiTFSI-2G4 and LTAP was investigated in an aqueous electrolyte using platinized platinum electrodes as the counter and reference electrodes. [Figure 7](#membranes-03-00298-f007){ref-type="fig"} shows the change in potential over time for the Li/PEO~18~LiTFSI-2G4/LTAP/1 M LiCl-4 mM LiOH aqueous solution/Pt, air cell at current densities in the range of 0.5 to 4.0 mA·cm^−2^ at 60 °C. This WSLE exhibited quite low lithium plating and stripping overpotentials at high current densities up to 4.0 mA·cm^−2^. The lithium stripping and plating overpotentials at 1.5 mA·cm^−2^ were 0.10 and 0.15 V, respectively, which are lower than the best results obtained for the Li/PEO~18~LiTFSI-18 wt % PEGDME/LTAP/1 M LiCl-4 mM LiOH aqueous solution/Pt, air cell (0.29 V for plating and 0.21 V for stripping) in our previous studies \[[@B14-membranes-03-00298]\]. The low overpotentials of the cell at high current densities could be attributed to the high lithium ion diffusion coefficient, high lithium ion transport number, and low interfacial resistance of PEO~18~LiTFSI-2G4.
{#membranes-03-00298-f007}
To study the lithium dendrite formation at the interface of Li and PEO~18~LiTFSI-2G4, a long period of lithium deposition at a constant current of 1 mA·cm^−2^ was performed using the Li/PEO~18~LiTFSI-2G4/LTAP/saturated LiCl aqueous solution/Pt, air cell, where the thickness of PEO~18~LiTFSI-2G4 was around 100 μm and the impedance of the WSLE was measured at every 4 h polarization. [Figure 8](#membranes-03-00298-f008){ref-type="fig"}a shows cell potential *versus* time curves as a function of the polarization period. The lithium electrode potential increased suddenly after a 25 h polarization. This potential increase may be due to lithium deposition on LTAP by lithium dendrite formation, which would result in the formation of a high resistance layer by the reaction of lithium and LTAP \[[@B4-membranes-03-00298]\]. [Figure 8](#membranes-03-00298-f008){ref-type="fig"}b shows the impedance profiles as a function of the polarization period. The WSLE resistance increased gradually with the polarization period up to 13 h and then decreased up to 22 h polarization. The decrease of electrode resistance may be due to lithium dendrite formation. Finally, the WSLE resistance was increased to 220 Ω·cm^2^ by the short circuit of the lithium metal electrode with LTAP. The short circuit period of 25 h is approximately 2.6 times longer than that measured for a Li/PEO~18~LiTFSI/LTAP/10 M LiCl-4 mM LiOH aqueous solution/Pt, air cell \[[@B14-membranes-03-00298]\]. The long short circuit period for the Li/PEO~18~LiTFSI-2G4/LTAP/saturated LiCl aqueous solution/Pt, air cell could be explained by the high lithium ion transport number, high salt diffusion coefficient, and low interface resistance \[[@B17-membranes-03-00298],[@B23-membranes-03-00298]\].
{#membranes-03-00298-f008}
The cyclability of the Li/PEO~18~LiTFSI-2G4/LTAP/saturated LiCl aqueous solution/Pt, air cell for lithium deposition and stripping at a constant current density of 1.0 mA·cm^−2^ was measured at 60 °C, where the current was passed for 2 h. The cell voltage *versus* time profile is shown in [Figure 9](#membranes-03-00298-f009){ref-type="fig"}. After 100 cycles, the overpotential for lithium deposition and stripping slightly increased from 0.10 to 0.15 V and from 0.08 to 0.13 V, respectively. This excellent cycling performance could be ascribed to the low and stable resistance for the SEI formed between lithium metal and PEO~18~LiTFSI-2G4.
{#membranes-03-00298-f009}
3. Experimental Section
=======================
PEO (Sigma-Aldrich, M~v~ ≈ 600,000 g·mol^−1^) and LiTFSI (Wako Chemicals, Japan) were dried overnight under vacuum at 50 and 160 °C, respectively, and then transferred to an Ar-filled glove box without exposure to air. G4 (Kishida Chemical, M = 222.28 g·mol^−1^) was immersed into activated 0.3 nm molecular sieves for 1 week and the same pretreatment by molecular sieves was performed 3 times before G4 was stored in a glove box. The composite polymer electrolytes of PEO~18~LiTFSI-*x*TEGDME in the range of *x* = 0--3 were prepared using a casting method \[[@B24-membranes-03-00298]\]. LiTFSI (Wako Chemicals, Japan) and G4 were added to an acetonitrile (AN) solution of PEO (Li/O = 1/18). The mixed solution was stirred at room temperature for 12 h and then cast into a clean Teflon dish. The AN solvent was evaporated slowly at room temperature in a Ar-filled dry glove box. After evaporation of the AN, the Teflon dish was transferred to a vacuum oven and dried at 80 °C for 24 h to remove residual AN. The resultant CPEs were *ca*. 100 μm thick.
Electrical conductivity measurements of the CPEs were performed using sandwich cells of Au/CPE/Au with gold foil electrodes. The cell impedances were measured using a frequency response analyzer (Solartron 1250) in the frequency range from 0.1 Hz to 1 MHz at temperatures in the range of 25--80 °C. Z-plot software was employed for data analysis.
The LTAP plate (*ca*. 250 μm thick) was supplied by Ohara Inc., Japan. The electrical conductivity of the LTAP plate at 60 °C was 1.4 × 10^−3^ S·cm^−1^. A Li/CPE/Li cell was assembled to investigate the interfacial resistance between lithium and the CPE. The active surface area was 1.00 cm^2^. The WSLE was prepared by sandwiching lithium metal foil (200 μm thick) with a Cu thin film lead, CPE, and the LTAP plate in a plastic envelope. The envelope was evacuated, heat-sealed apart from a 0.35 × 0.35 cm^2^ area, and then subjected to an isostatic pressure of 150 MPa for 15 min to ensure good contact. The WSLE was immersed into an aqueous solution of 1 M (or 10 M) LiCl with 4 mM LiOH. A platinized Pt electrode was used as both the counter and reference electrodes. Electrochemical measurements were conducted using a multichannel potentiostat-galvanostat (Biologic Science Instruments VMP3). The lithium ion transport number *t*~Li~^+^, was determined using the Li/CPE/Li cell with a combination of AC impedance spectroscopy and the DC polarization method, as originally proposed by Evans and Vincent, and later refined by Abraham and Jiang \[[@B19-membranes-03-00298],[@B25-membranes-03-00298]\]. The cell potentials were measured after polarization for at least 4 h to obtain steady-state current data. The diffusion coefficient was estimated using a method proposed by Ma *et al*. \[[@B18-membranes-03-00298]\]. The Li/CPE/Li cell was polarized at 50 mV under potentiostatic mode for a few hours to achieve steady-state, after which the potential was interrupted and monitored at 60 °C.
4. Conclusions
==============
The PEO~18~LiTFSI-2G4 electrolyte exhibited a high lithium ion conductivity of 8.91 × 10^−4^ S·cm^−1^, a high diffusion coefficient of 3.37 × 10^−7^ cm^2^·S^−1^ at 60 °C, and a low interfacial resistance in contact with lithium metal. The Li/PEO~18~LiTFSI-2G4/LTAP/saturated LiCl aqueous solution/Pt, air cell exhibited excellent cyclability. The overpotentials for lithium deposition and stripping at 2.0 mAh·cm^−2^ for 2 h polarization at 60 °C were increased slightly from 0.10 to 0.15 V and from 0.08 to 0.13 V after 100 cycles, respectively. Therefore, the CPE is an attractive material for the interlayer between lithium metal and LTAP for the water stable lithium electrode in lithium-air batteries.
The authors declare no conflict of interest.
| {
"pile_set_name": "PubMed Central"
} |
Introduction {#Sec1}
============
Heart failure (HF) is a complex clinical syndrome and is often caused by reduced cardiac pump function, affecting approximately 1--2% of people in the Western world^[@CR1]^. The most prevalent cause of HF is an acute myocardial infarction (MI). Despite the significant reduction of early mortality and improved treatment, the prognosis of HF patients remains poor with a 5-year survival of less than 50%^[@CR2]^, stressing the need for a better understanding and treatment of this complex syndrome.
Heart failure after MI is caused by adverse left ventricular remodelling^[@CR3],[@CR4]^. Adverse remodelling is largely depending on infarct size, but also on the quality of cardiac repair, both greatly influenced by the inflammatory response after MI^[@CR5]--[@CR8]^. Although inflammatory cells are important in the clearance of debris and necrotic tissue after MI, their pro-inflammatory activity is also responsible for a variety of detrimental effects. In the (sub)acute phase of ischemia-reperfusion injury, neutrophils and monocytes play an important role in the increase in infarct size^[@CR5],[@CR9],[@CR10]^. In the chronic phase, activated and pro-inflammatory monocytes and T-lymphocytes increase adverse remodelling, which leads to impaired cardiac function^[@CR11],[@CR12]^. In both processes, leukocyte activation is key and regulated by integration of signals from activating and inhibitory cell-receptors^[@CR13]^. The majority of immune inhibitory receptors contain intracellular domains that -- upon activation -- are able to down regulate or inhibit activation signals from stimulating receptors. Thereby, they increase the threshold for leukocytes to become activated and attenuate pro-inflammatory effects. The transmembrane leukocyte-associated immunoglobulin-like receptor 1 (LAIR-1, CD305) is an inhibitory receptor that is expressed on most cells of the immune system, including natural killer cells, lymphocytes and monocytes^[@CR14]^. LAIR-1 can inhibit activating signals from ITAM-bearing receptors. Next to this, LAIR-1 is also capable of inhibiting cytokine-mediated signals and it can prevent proliferation and induce apoptosis of human myeloid leukemia cell lines^[@CR14]^. LAIR-1 is activated upon binding of its ligands including collagens or collagen-domain containing proteins such as surfactant protein D^[@CR15],[@CR16]^. It is counteracted by shedding of its ectodomain (sLAIR-1) or by secretion of its antagonist LAIR-2^[@CR17]^, which is why plasma levels of these molecules were also studied.
Though LAIR-1 is capable of regulating immune cell function by for example inhibition of target cell lysis or attenuation of the cytotoxic activity of effector T cells^[@CR18],[@CR19]^, its role *in vivo* has been scarcely addressed and evidence for its role in inflammation following acute MI is lacking. Therefore, we compared LAIR-1 expression on leukocytes and circulating levels of sLAIR-1 and LAIR-2 in patients 3 days and 6 weeks after MI, representing the acute and chronic phase of cardiac remodelling. Moreover, we studied the effect of LAIR-1 deficiency in experimental MI in mice, measuring inflammation, infarct size, adverse left ventricular remodelling and cardiac function.
Materials and Methods {#Sec2}
=====================
Study population {#Sec3}
----------------
Healthy volunteers and patients (\>18 years old) with a first time ST-elevation myocardial infarction (STEMI) and non-STEMI from the DEFI-MI (METC: NL45241.041.13) study were included in the current study. Exclusion criteria were the presence of a chronic inflammatory disease, autoimmune disorder, pregnancy and trauma or surgery in the last six months. The Medical Ethics Committee of the UMC Utrecht approved the study and all patients provided written informed consent. The study conforms to the Declaration of Helsinki.
Patient data collection {#Sec4}
-----------------------
In patients, at the moment of inclusion (3 days after MI) and 6 weeks thereafter, venous blood was drawn and collected by the Laboratory of Clinical Chemistry and Haematology of the UMC Utrecht (Fig. [1a](#Fig1){ref-type="fig"}). Whole blood was directly subjected to flow cytometry and EDTA plasma was stored at −80° Celsius in the UMC Utrecht Biobank and used for ELISA (see below). Similarly, blood was drawn from healthy controls and subjected to flow cytometry.Figure 1Timeline and experimental set-up. Healthy controls and patients with a first-time MI were included in the current study. Venous blood collection in healthy controls and at 3 days and 6 weeks after MI was used for flow cytometric analyses and to collect plasma (**a**). Mice were either subjected to ischemia-reperfusion injury or to permanent left coronary artery ligation. In the mice subjected to 30 minutes of ischemia and 24 hours of subsequent reperfusion, IS/AAR staining was performed (**b**). In the mice subjected to permanent left coronary artery ligation and sacrificed after 3 days, we performed flow cytometric analyses on various tissues and blood (**c**). In the other mice subjected to left coronary artery ligation, we performed echocardiography at baseline, 7 days and 28 days after MI and performed similar flow cytometric analyses (**d** -- lower panel). *MI: myocardial infarction; IS*/*AAR: infarct size*/*area at risk*.
Animals {#Sec5}
-------
Healthy male C57Bl/6 LAIR-1^−/−^ ^[@CR20]^ and C57Bl/6 WT littermates (age 10--12 weeks, weight 25--30 g) were housed at room temperature under 12 hour light/dark cycle in groups of maximum 5 animals (in type III cages with filtertop) under strict DM1 regulations and received standard chow and water ad libitum. All animals were genotyped prior to the experimental procedure and animal welfare was monitored daily. A blinded researcher performed surgery on randomly assigned animals (random number generation in excel to animal number, which resulted in alternating fashion of operation of WT and LAIR-1^−/−^ mice). Blinded technicians and observers performed the respective operations, data acquisition and analyses. Sample size calculation for myocardial ischemia reperfusion model was based on infarct size as the primary endpoint. With a power of 90%, alpha of 0.05, estimated effect size of 13% difference in infarct size, standard deviation of 10.4% (based on historical data) and estimated peri-operative mortality of 5% this resulted in a group size of 15 animals per group. For the permanent ligation model sample size calculation was based on end systolic volume as the primary endpoint. With a power of 90%, alpha of 0.05, estimated effect size of 20 μl difference in volume, standard deviation of 15 μl (based on historical data) and estimated peri-operative mortality of 25% this resulted in a group size of 20 animals per group. Cellular influx and collagen content in the ischemic area were defined as secondary outcomes. All animal experiments were approved by the Ethical Committee on Animal Experimentation of the University Medical Center Utrecht (Utrecht, the Netherlands) and conform to the 'Guide for the care and use of laboratory animals'.
Induction of myocardial ischemia-reperfusion injury {#Sec6}
---------------------------------------------------
The experimental set-up and timeline of myocardial ischemia-reperfusion injury is displayed in Fig. [1b](#Fig1){ref-type="fig"}. All operations were performed in the morning before noon. In a dedicated mouse operation room, anaesthesia was induced by intraperitoneal (i.p.) injection of medetomidinehydrochloride (1.0 g/kg body weight), midazolam (10.0 mg/kg) and fentanyl (0.1 mg/kg). These anaesthetics were preferred over cardioprotective propofol or volatile anaesthetics (*e.g*. isoflurane)^[@CR21]^. Mice were intubated and connected to a respirator with a 1:1 oxygen-air ratio (times/minute). A core body temperature of 37 **°**C was maintained during surgery by continuous rectal temperature monitoring and an automatic heating blanket. The heart was accessed through a left lateral thoracotomy with incision of the pericardium. The left coronary artery was ligated for 30 minutes with an 8-0 Ethilon suture (Ethicon) with a section of polyethylene-10 tubing placed over the left coronary artery (LCA). Ischemia was confirmed by bleaching of myocardium and tachycardia. After 30 minutes of ischemia, reperfusion was initiated by releasing the ligation, resulting in tissue colour recurrence. A piece of the suture was left in place to allow for accurate ligature positioning and determination of the ischemic area and the area at risk at termination. The surgical wounds were closed and subcutaneous atipamezole hydrochloride (3.3 mg/kg), flumazenil (0.5 mg/kg) and buprenorphin (0.15 mg/kg) were used as an antagonist. The evening of the day of operation and every 12 hours thereafter, subcutaneous injection of buprenorphin (0.15 mg/kg) was administered as analgesia.
Infarct size and area at risk quantification after ischemia-reperfusion injury {#Sec7}
------------------------------------------------------------------------------
In total 30 animals (15 WT and 15 LAIR-1^−/−^) were subjected to ischemia-reperfusion injury. Twenty-four hours after ischemia-reperfusion injury, mice were euthanized using sodium pentobarbital (60.0 mg/kg) and a left re-thoracotomy was performed. The LCA was ligated at the same location as it was ligated during index ischemia. The thoracic aorta was cannulated and 2% Evans blue was injected upstream in the aorta to perfuse the coronaries, allowing for staining of the remote but not the area at risk (AAR). The heart was then explanted and rinsed with 0.9% saline to remove superfluous dye. The left ventricle (LV) was dissected and a small piece of gauze was inserted in the left ventricular cavity. After one hour at −20 °C, the LV was cut into 4 equally sized sections. Sections were placed in 1% 2,3,5-triphenyltetrazolium chloride (TTC) in saline and incubated at 37 **°**C for 20 minutes. After 10 minutes, sections were turned to allow for adequate reagent contact. Then, sections were placed in formalin and photographs of both sides of each tissue section were captured using a SZH10 Olympus Zoom Stereo Microscope and IC Capture software, version 2.4. The infarct (white), border zone (red) and remote area (blue) were quantified using ImageJ (version 1.48 v). Infarct size (IS) was expressed as a percentage of the AAR and as a percentage of the LV.
Induction of myocardial infarction by permanent ligation {#Sec8}
--------------------------------------------------------
Permanent coronary artery ligation was performed as described above for ischemia-reperfusion injury, but leaving the ligature in place, resulting in a permanent occlusion of the left coronary artery. The experimental set-up including the timeline of the mice sacrificed after either 3 days or 28 days is shown in Fig. [1c,d](#Fig1){ref-type="fig"}. Surgery was performed on 23 animals (10 WT and 13 LAIR-1^−/−^) for 3 day survival and 38 animals (18 WT and 20 LAIR-1^−/−^) were included for long term survival (28 days).
Survival {#Sec9}
--------
Mice that died after MI were thoroughly inspected for the cause of death. Deaths within 48 hours after MI were considered due to perioperative complications or direct complication of MI. Cardiac rupture was confirmed by massive intrathoracic haemorrhage \>48 hours after operation and ventricular leakage of the myocardium upon perfusion of the heart with 0.9% saline. In the ischemia reperfusion model 2 LAIR-1^−/−^ died due to perioperative conditions. Of the animals exposed to 3 days permanent ligation 1 WT animal died during the surgical procedure, of animals exposed to 28 days permanent ligation 6 WT and 9 LAIT-1^−/−^ mice died. Of the 6 WT, 5 died due to cardiac rupture and 1 to unkown causes. Of the LAIR-1^−/−^ 8 died due to cardiac ruputre and 1 to unknown causes.
Echocardiography {#Sec10}
----------------
At baseline, 7 and 28 days after permanent ligation, anaesthesia was induced by inhalation of 2.0% isoflurane in a mixture of oxygen/air (1:1). Echocardiography was used to assess cardiac geometry and function. Heart rate, respiration and rectal temperature were constantly monitored and body temperature was kept between 36.0 and 38.0 °C using heat lamps. Respiration gating, a 3-dimensional motor and trigger points were used to obtain 300 transversal images of the heart during the expiratory phase, either at the end of systole or the end of diastole. These images were then used for complete 3D reconstruction of the heart. Image acquisition and analyses were performed using the dedicated Vevo® 2100 System and Software (Fujifilm VisualSonics Inc., Toronto, Canada).
Tissue processing and histological analyses {#Sec11}
-------------------------------------------
At the end of the follow-up period, mice were euthanized using sodium pentobarbital (60.0 g/kg). Blood was collected through orbital puncture in EDTA tubes. The inferior caval vein was incised and the vascular system was flushed with 5 mL phosphate-buffered saline (PBS) through right ventricular puncture.
The spleen and the mediastinal lymph nodes posterior to the heart were excised and contained in PBS for flow cytometric analyses afterwards. Then, the heart was explanted and cut in half. One half was dissected further into infarct and remote tissue and used for flow cytometry or snap-frozen in liquid nitrogen. The other half was formalin-fixed for 24 hours, embedded in paraffin and cut into 5 μm thick sections. Neutrophils were stained using a rat monoclonal mouse LY-6G (GR-1) antibody (1:400, Biolegend, 108402, 0.5 mg/ml). Rabbit-anti-rat-biotin (1:200, DAKO E0468, 0.84 g/L) was used as a secondary antibody and Streptavidin-AP (1:500, SA-5100) as a tertiary antibody. T-cells were stained using a polyclonal rabbit-anti-human CD3 antibody (1:100, Dako, A0452) and anti-rabbit-AP Powervision (pure, DPVR-110AP, Immunologic) was used as secondary antibody. Macrophages were stained using a rat-anti-mouse MAC3 antibody (1:30, BD Pharmingen, 553322, 0.5 mg/ml). Rabbit-anti-rat-biotin (1:200, DAKO E0468, 0.84 g/L) was used as a secondary antibody and Streptavidin-AP (1:500, SA-5100) as a tertiary antibody. Liquid permanent red was used as an enzyme substrate. The different subtypes of collagen were stained using goat-anti type I Collagen (1:250, Southern Biotech, 1310-01, 0.4 mg/ml), goat-anti type III Collagen (1:50, Southern Biotech, 1330-01, 0.4 mg/ml) or goat-anti type IV Collagen (1:100, Southern Biotech, 1340-01, 0.4 mg/ml) and alexa 488 donkey-anti-goat (1:250, Invitrogen, A11055, 2 mg/ml) as a secondary antibody.
Neutrophils, T-cells, macrophages and collagens were semi-automatically quantified using digital histology. Collagen content was quantified in tissue sections stained for picrosirius red and photographed under polarized light, converted to gray scale images and expressed as a percentage of the region of interest (*i.e*. infarct, remote). Images of tissue sections were captured and analysed using CellSens (Olympus Corporation, Tokyo, Japan). Of the 9 WT and 10 LAIR-1^−/−^) mice at 3 days follow-up, 1 WT and 1 LAIR^−/−^ mice were excluded for histological analysis due to the absence of a clearly identifiable infarction. For neutrophil analysis an 2 WT and 2 LAIR-1^−/−^) animals were excluded due to technical errors. Of the 11 WT and 10 LAIR1^−/−^ mice at 28 days follow-up 1 WT and 1 LAIR-1^−/−^) mice were excluded for histological analysis due to the absence of a clearly identifiable infarction. For macrophage analysis an 3 LAIR-1^−/−^) animals were excluded due to technical errors. 2. Technical errors may occur due to poor quality upon sectioning or inferior quality of immunohistochemistry stainings, which makes reliable analysis impossible.
Flow cytometric assays {#Sec12}
----------------------
Fresh human EDTA blood (50 μL) was added to an antibody mixture containing different cell surface markers to identify neutrophils and monocytes (see Supplementary Table [S1](#MOESM1){ref-type="media"}). Cells were incubated for 30 minutes in the dark at room temperature (RT). Before measurement, cells were washed and erythrocytes were lysed using Optilyse C.
To harvest single cells from heart tissue, enzymatic degradation was performed (N = 6 WT and N = 6 LAIR-1^−/−^). Infarct and remote tissue were collected 3 days after MI and cut into small pieces of around 1 mm^2^. Dissociation solution (10 × 10^2^ U/ml DNase I (Roche 04536282001), 10 mM HEPES (Life Technology 15630-080) and 2.6 U/ml Liberase TL (Roche 05401020001)) was added to the tissue and incubated at 37° Celsius for 20 minutes. Single cells of the dissociated myocardial tissue, lymph nodes and spleen were obtained through gentle filtering over a 40 µm cell strainer and subsequently incubated with an antibody mixture containing different cell surface markers to identify neutrophils, monocytes, and T- and B-lymphocytes (see Supplementary Table [S2](#MOESM1){ref-type="media"}) for 30 minutes in the dark at RT. After washing, residual red blood cells were lysed with erythrocyte-lysis buffer. All samples were measured on a Gallios flow cytometer (10 colour configuration, Beckman Coulter, Marseille, France). Kaluza Analysis Software 1.3 was used for data analysis. The gating strategy is shown in Supplementary Fig. [S1](#MOESM1){ref-type="media"}.
ELISA {#Sec13}
-----
Plasma levels of soluble LAIR-1 (sLAIR-1) and LAIR-2 were measured in duplo using a respective sLAIR-1 and LAIR-2 sandwich ELISA according to manufacturer's instructions (LifeSpan BioSciences, Seatle, WA, USA). Colorimetric analyses were performed using a spectrophotometer (450 nm). Plasma levels were calculated based on standards.
High-sensitivity multiplex immunoassay {#Sec14}
--------------------------------------
Plasma levels of IL-6 and TNFα were measured in duplo using a high-sensitivity ProcartaPlex multiplex immunoassay according to manufacturer's instructions (EPXS010-20603-901 and EPXS010-20607-901, ThermoFisher Scientific). Plasma levels were calculated based on standards. Il-6 and TNFα values were ln-transformed.
Statistical analyses {#Sec15}
--------------------
Data distribution was evaluated for normality using the d'Agostino & Pearson normality test. Data are expressed as mean ± standard deviation (SD). Skewed ELISA and immunoassay data were ln-transformed and presented as median with interquartile range (IQR). Normally distributed data were compared using a two-tailed paired (serial measurements) or unpaired t-test (separate groups). Non-normally distributed data were compared using a Wilcoxon (serial measurements) or Mann-Whitney test (separate groups). A log-rank (Mantel-Cox) test was used for survival analysis. A level of p \< 0.05 was considered statistically significant. Statistical analyses were performed using SPSS software, version 21 and GraphPad Prism, version 6.
Results {#Sec16}
=======
In patients, LAIR-1 expression on circulating cells and sLAIR-1 and LAIR-2 plasma levels differ between the acute and chronic phase after MI {#Sec17}
--------------------------------------------------------------------------------------------------------------------------------------------
Out of 24 patients included in this study, 22 patients (92%) suffered from a STEMI and 2 patients (8%) from a non-STEMI (Table [1](#Tab1){ref-type="table"}). The mean age was 58 ± 11 years and the majority of patients were male (79%). For comparison, 20 healthy volunteers were included as controls. The transmembrane expression of LAIR-1 on monocytes was significantly higher in the acute phase (3 days after MI), compared to the chronic phase (24.8 ± 5.3 at day 3 vs. 21.2 ± 5.1 MFI at 6 weeks post MI, p = 0.008; Fig. [2a](#Fig2){ref-type="fig"}), and both were significantly increased with respect to healthy controls. Subgroup analyses showed that this difference could be attributed to higher LAIR-1 expression on pro-inflammatory CD14^++^ CD16^−^ classical (25.0 ± 5.4 at 3 days vs. 21.5 ± 5.0 MFI at 6 weeks post MI, p = 0.013; Fig. [2b](#Fig2){ref-type="fig"}) and CD14^++^ CD16^+^ intermediate monocytes (27.2 ± 7.7 at day 3 vs. 19.9 ± 4.3 MFI at 6 weeks post MI, p = 0.001; Fig. [2c](#Fig2){ref-type="fig"}), but not CD14^−^ CD16^+^ non-classical monocytes (22.1 ± 9.1 at day 3 vs. 18.9 ± 9.7 MFI at 6 week, p = 0.28; Fig. [2d](#Fig2){ref-type="fig"}). Similar to monocytes, LAIR-1 expression on neutrophils was higher 3 days after MI compared to 6 weeks after MI (12.9 ± 4.7 vs. 10.6 ± 3.1 MFI, p = 0.046; Fig. [2e](#Fig2){ref-type="fig"}). There was no difference in LAIR-1 expression on CD4^+^ T-lymphocytes (8.4 ± 1.0 at day 3 vs. 8.7 ± 1.5 MFI at 6 weeks, p = 0.95; Fig. [2f](#Fig2){ref-type="fig"}) or CD8^+^ T-lymphocytes (10.6 ± 3.0 at day 3 vs 11.0 ± 3.5 MFI at 6 weeks, p = 0.92; see Supplementary Fig. [S3](#MOESM1){ref-type="media"}).Table 1Baseline characteristics of DEFI-MI patients.DEFI-MI patients (N = 24)Male sex19 (79%)Age58 ± 11 years**Cardiovascular risk factors**BMI25.5 ± 2.4Diabetes0 (0%)Hypertension5 (21%)Hypercholesterolemia8 (33%)Smoking8 (33%)**Medication use**Aspirin20 (83%)P2Y12 inhibitor24 (100%)Statin23 (96%)Beta-blocker20 (83%)RAAS-inhibitor22 (92%)**History of cardiovascular disease**CVA/TIA0 (0%)Peripheral artery disease0 (0%)Chronic kidney failure0 (0%)**Indication**STEMI22 (92%)Non-STEMI2 (8%)Clinical characteristics of DEFI-MI patients upon presentation. BMI: body mass index; RAS: renin-angiotensin system; CVA: cerebrovascular accident; TIA: transient ischemic attack; STEMI: ST-elevation myocardial infarction. Figure 2LAIR-1 expression on leukocyte subsets and sLAIR-1 and LAIR-2 plasma levels differ between the acute and chronic phase after myocardial infarction in patients. Flow cytometry showed that LAIR-1 receptor expression on monocytes (**a**) 3 days after MI was higher than 6 weeks thereafter and compared to healthy controls, which could be mainly attributed to CD14^++^ CD16^−^ classical (**b**) and CD14^++^ CD16^+^ intermediate monocytes (**c**), but not to CD14^−^ CD16^+^ non-classical monocytes (**d**). Similar to monocytes, also granulocytes showed higher LAIR-1 receptor expression in the acute compared to the chronic phase, but no difference between the chronic phase and healthy controls was observed (**e**). No difference was observed in LAIR-1 receptor expression on CD4^+^ T-lymphocytes (**f**). Though not significant, sLAIR-1 was higher 3 days after MI compared to 6 weeks (**g**). In contrast, plasma levels of LAIR-2 were lower 3 days after MI compared to 6 weeks (**h**). N = 19--22 (**a,b**,**d,e**), 14 (**c**) and 24 (**f,g**) patients, 20 healthy controls (**a**--**e**). *MI: myocardial infarction;* \**p* \< *0.05, \**\**p* \< *0.01, \*\**\**p* \< *0.001, \*\*\**\**p* \< *0.0001*.
In general, comparison with healthy controls showed that LAIR-1 expression was increased after MI on monocytes and neutrophils, either at both 3 days and 6 weeks (monocytes, CD14^++^ CD16^−^ classical monocytes) or at 3 days only (CD14^++^ CD16^+^ intermediate monocytes, CD14^−^ CD16^+^ non-classical monocytes).
Plasma levels of sLAIR-1 were slightly higher 3 days after MI compared to 6 weeks, though this was not significant (2.71 IQR \[1.35--3.87\] vs. 1.92 IQR \[1.01--2.61\], p = 0.07; Fig. [2g](#Fig2){ref-type="fig"}). In contrast, LAIR-2 levels were significantly higher after 6 weeks (238.7 IQR \[205.2--268.0\] vs. 260.3 IQR \[223.9--297.7\], p = 0.049; Fig. [2h](#Fig2){ref-type="fig"}).
The absence of LAIR-1 does not influence infarct size in mice {#Sec18}
-------------------------------------------------------------
Evans blue and TTC staining were used to quantify infarct size (IS), area at risk (AAR) and left ventricular (LV) area in both WT and LAIR-1^−/−^ mice (Fig. [3a,b](#Fig3){ref-type="fig"}). Ischemia-reperfusion injury as assessed by IS/AAR did not differ between WT and LAIR-1^−/−^ mice (37.0 ± 14.5 vs. 39.4 ± 12.2%, p = 0.63; Fig. [3c](#Fig3){ref-type="fig"}). In addition, AAR/LV was comparable between both groups (38.3 ± 14.1 vs. 36.8 ± 10.3%, p = 0.75; Fig. [3d](#Fig3){ref-type="fig"}), as was IS/LV (14.2 ± 7.4 vs. 14.9 ± 7.0, p = 0.80; Fig. [3e](#Fig3){ref-type="fig"}).Figure 3Infarct size and area at risk quantification after ischemia-reperfusion injury in mice. Evans blue and TTC staining of WT (**a**) and LAIR-1^−/−^ hearts (**b**) were used for the quantification of infarct size (IS; white), area at risk (AAR; sum of white and red area) and the left ventricle itself (LV; entire area). Ischemia-reperfusion injury assessed through IS/AAR% did not differ between both groups (**c**). Also AAR/LV% (**d**) and IS/LV% was comparable between both groups. N = 15 WT and 13 LAIR-1^−/−^ animals per group. *WT: wild-type; LAIR-1* ^−/−^ *: LAIR-1 deficient*.
Survival, cardiac geometry, and cardiac function are comparable between wild-type and LAIR-1^−/−^ mice after permanent coronary artery ligation {#Sec19}
-----------------------------------------------------------------------------------------------------------------------------------------------
Within 28 days after permanent ligation, cardiac rupture and subsequent death occurred in 6 out of 18 WT and 9 out of 20 LAIR-1^−/−^ mice (33.3 vs. 45.0%, p = 0.38; Fig. [4a](#Fig4){ref-type="fig"}). In the surviving animals, no differences were observed with respect to end-diastolic volume (EDV) at 7 days (WT 105.6 ± 14.3 vs. LAIR-1^−/−^ 113.0 ± 26.1 μL, p = 0.40; Fig. [4b](#Fig4){ref-type="fig"}) and 28 days (133.3 ± 19.3 vs. 132.1 ± 27.9 μL, p = 0.91) after permanent ligation. In addition, end-systolic volume (ESV) was comparable between both groups after 7 days (87.5 ± 16.1 vs. 91.0 ± 29.0 μL, p = 0.73; Fig. [4c](#Fig4){ref-type="fig"}) and 28 days (112.1 ± 22.2 vs. 106.9 ± 33.5 μL, p = 0.68). Correspondingly, left ventricular ejection fraction did not differ between WT and LAIR-1^−/−^ mice at 7 days (17.6 ± 4.0 vs. 21.0 ± 9.0%, p = 0.25; Fig. [4d](#Fig4){ref-type="fig"}) and 28 days (16.5 ± 5.1 vs. 20.5 ± 8.5%; p = 0.20) after MI.Figure 4Survival and geometric dimensions by 3D echocardiography in mice. Within the follow-up period of 28 days after permanent occlusion of the left coronary artery, 6 out of 18 WT mice and 9 out of 20 LAIR-1^−/−^ mice died as a consequence of cardiac rupture (**a**). In the surviving animals, EDV, ESV and LVEF were comparable 7 and 28 days after ligation (**b**--**d**). N = 12 WT and 11 LAIR-1^−/−^ surviving animals per group. *WT: wild-type; LAIR* ^*−*/*−*^ *: LAIR-1 deficient; EDV: end-diastolic volume; ESV: end-systolic volume; LVEF: left ventricular ejection fraction*.
Wild-type and LAIR^−/−^ mice show no differences in inflammatory responses following myocardial infarction {#Sec20}
----------------------------------------------------------------------------------------------------------
To confirm LAIR-1 expression on circulating leukocytes, we performed flow cytometry on baseline blood. In WT mice, LAIR-1 was expressed on CD4^+^ T-cells and CD8^+^ T- cells, but most prominent on cells of myeloid origin, amongst which neutrophils, macrophages and Ly6C expressing monocytes (Fig. [5a](#Fig5){ref-type="fig"}). While LAIR-1 expression was similar over time in T-cells and monocytes, we did observe a decrease in LAIR-1 expression on macrophages 3 days and on neutrophils 28 days after MI (Supplementary Fig. [S4](#MOESM1){ref-type="media"}). As expected, LAIR-1 was undetectable on cells from LAIR-1^−/−^ mice.Figure 5Leukocytes levels in circulation and infiltration in the murine heart. LAIR-1 expression on different leukocyte subset in peripheral blood from baseline WT mice was determined by flow cytometry (N = 9 animals). LAIR-1 expression was most pronounced on cells of myeloid origin, but also observed on CD4 and CD8 T-lymphocytes (**a**). To assess neutrophil infiltration in the infarcted murine myocardium, cardiac tissue sections were stained with Ly6G three days after MI. Representative images of WT and LAIR-1^−/−^ mice of the infarct area and border zone (**b**) showed no difference in neutrophil influx three days after permanent occlusion of the left coronary artery (N = 6 WT and 7 LAIR-1^−/−^) (**c**). Cardiac T-lymphocyte influx was not different three or 28 days after permanent occlusion of the left coronary artery (N = 8 WT and 9 LAIR-1^−/−^) (N = 11 WT and 10 LAIR-1^−/−^) (**d**). Macrophage influx on showed a difference between WT and LAIR-1^−/−^ mice in the infarct zone 28 days after MI (**e**). Scale bar 100 µm. *WT: wild-type; LAIR-1* ^−/−^ *: LAIR-1 deficient*. \**p* \< *0.05*.
Flow cytometry was performed for the characterization and quantification of leukocytes in the blood, spleen, draining lymph node and heart after MI (see Supplementary Fig. [S2](#MOESM1){ref-type="media"}). Three days after MI, a robust leukocyte influx in the heart was observed, that mostly consisted of neutrophils and CD8^+^ T-cells (Table [2](#Tab2){ref-type="table"}). No difference in white blood cell subtype composition was observed between WT and LAIR-1^−/−^ mice 3 days and 28 days after MI in all studied organs (Table [2](#Tab2){ref-type="table"} -- blood, infarct area; see Supplementary Table [S3](#MOESM1){ref-type="media"}-[S4](#MOESM1){ref-type="media"} -- remote area, lymph nodes, spleen).Table 2Leukocyte levels in blood and infarct area after 3 days of MI.BloodInfarcted myocardiumWTLAIR-1^−/−^WTLAIR-1^−/−^Neutrophils28.1 ± 23.5%22.6 ± 8.3%56.8 ± 11.9%49.4 ± 14.3%Macrophages5.4 ± 4.4%7.1 ± 4.0%25.9 ± 8.2%30.9 ± 8.7%Ly6C High monocytes2.0 ± 1.5%3.0 ± 2.0%3.0 ± 1.9%4.5 ± 1.8%Ly6C Low monocytes4.5 ± 0.6%5.0 ± 1.2%3.9 ± 2.4%4.6 ± 2.9%CD4 T-cells9.0 ± 2.6%11.4 ± 3.2%1.9 ± 1.1%2.0 ± 1.2%CD8 T-cells8.7 ± 3.8%10.9 ± 2.5%34.3 ± 6.8%42.5 ± 10%Leukocyte levels in blood and infarct area three days after myocardial infarction in mice. Percentages are given of total leukocytes. N = 6 WT and 5 LAIR-1^−/−^.
Leukocyte activation after MI increases the release of various cytokines, of which IL-6 and TNFα are well known. Using a high-sensitive multiplex immunoassay, we compared plasma cytokine levels in WT and LAIR-1^−/−^ mice. TNFα levels were borderline detectable, did not change after MI and were comparable in both groups (Supplementary Fig. [S4](#MOESM1){ref-type="media"}). In contrast to stable IL-6 levels in WT mice, IL-6 did increase in LAIR-1^−/−^ mice 3 days after MI (25.7 IQR 25.6 -- 25.8 vs. 437 IQR 166.7--586.5, p = 0.04). A subsequent decrease in IL-6 levels 28 days after MI was observed in both WT and LAIR-1^−/−^ mice (Supplementary Fig. [S6](#MOESM1){ref-type="media"}). Overall, IL-6 levels were not significantly different between WT and LAIR1^−/−^ mice.
Since neutrophils are the first to infiltrate the heart in the early phase after MI and are notorious for the additional damage they cause to the myocardium, we performed a neutrophil staining. Three days after MI, neutrophil influx was comparable between WT and LAIR-1^−/−^ mice in both the infarct area (406 ± 167 vs. 405 ± 165 cells/mm^2^, p = 0.99; Fig. [5b,c](#Fig5){ref-type="fig"}) and border zone (483 ± 245 vs. 539 ± 344 cells/mm^2^, p = 0.63). In addition, T-cells and macrophages play an important role in remodelling after MI. CD3^+^ staining showed a comparable infiltration of T-cells in the myocardium of WT and LAIR-1^−/−^ mice both in the infarct area (792 ± 343 vs. 1212 ± 1139 cells/mm^2^, p = 0.33; Fig. [5d](#Fig5){ref-type="fig"}) and border zone (733 ± 333 vs. 1021 ± 913 cells/mm^2^, p = 0.41) 3 days after MI. Similar results were observed twenty-eight days after MI in the infarct area (728 ± 616 vs. 824 ± 335 cells/mm^2^, p = 0.67; Fig. [5d](#Fig5){ref-type="fig"}) and border zone (450 ± 344 vs. 601 ± 304 cells/mm^2^, p = 0.30). Also, macrophage infiltration did not differ between WT and LAIR-1^−/−^ mice 3 days after MI in both the infarct area (1.5 ± 1.3 vs. 3.2 ± 3.4% positive area, p = 0.20; Fig. [5e](#Fig5){ref-type="fig"}) and border zone (2.1 ± 1.8 vs. 4.1 ± 4.7% positive area, p = 0.28). After 28 days however, fewer macrophages were present in the infarct area of WT mice compared to LAIR-1^−/−^ mice (1.0 ± 0.5 vs. 2.6 ± 1.4% positive area, p = 0.01; Fig. [5e](#Fig5){ref-type="fig"}) but this was not the case in the border zone (1.4 ± 1.6 vs. 2.0 ± 2.0% positive area, p = 0.54).
Fibrosis formation is not affected by the absence of LAIR-1^−/−^ after chronic myocardial infarction {#Sec21}
----------------------------------------------------------------------------------------------------
As a marker of cardiac fibrosis we stained cardiac tissue sections with picrosirius red of both WT and LAIR-1^−/−^ mice (representative pictures in Fig. [6a,b](#Fig6){ref-type="fig"}). Magnified images of the infarct area (Fig. [6c,d](#Fig6){ref-type="fig"}) show no differences in collagen content between both groups 28 days after permanent ligation (39.1 ± 19.9 vs. 40.0 ± 10.7%, p = 0.65; Fig. [6e](#Fig6){ref-type="fig"}). In addition, different collagen types were distinguished using specific a staining for collagen I, collagen III and collagen IV, respectively. Infarct tissue consists mainly of Collagen I (WT: 28.3 ± 17.9, LAIR-1^−/−^: 28.8 ± 21.3%, p = 0.97) and Collagen IV (WT: 17.5 ± 11.4, LAIR-1^−/−^: 23.4 ± 17.8%, p = 0.60), whereas collagen III comprised about 8% of the infarct area (WT: 8.4 ± 6.6, LAIR-1^−/−^: 9.0 ± 7.3%, p = 0.97). No difference in either collagen type was observed between WT and LAIR-1^−/−^ mice (Fig. [6f--h](#Fig6){ref-type="fig"}).Figure 6Collagen content 28 days after permanent coronary artery ligation. Cardiac tissue sections were stained with picrosirius red and photographed under polarized light. Representative overview images of WT and LAIR-1^−/−^ mice (**a**,**b**) and magnified images of the infarct area (**c**,**d**) show no differences in total collagen percentage in both remote and infarct regions 28 days after permanent occlusion of the left coronary artery (**e**). There was also no difference in the percentage of collagen subtypes I (**f**), III (**g**) or IV (**h**). N = 11 WT and 10 LAIR-1^−/−^ animals per group. Scale bar 100 µm. *WT: wild-type; LAIR-1* ^−/−^ *: LAIR-1 deficient*.
Discussion {#Sec22}
==========
Leukocytes and leukocyte activation, in particular monocytes and neutrophils, have been shown to play an important role in both cardiac ischemia-reperfusion injury and remodelling^[@CR22]--[@CR25]^. LAIR-1 is present on a variety of immune cells^[@CR14]^ and important in the regulation of leukocyte activation in response to an inflammatory reaction^[@CR15],[@CR26],[@CR27]^. We observed increased LAIR-1 expression on leukocytes of patients compared to healthy controls. More specifically, LAIR-1 expression on circulating monocytes and neutrophils is increased directly after MI and declines after six weeks, suggestive of immune regulation by LAIR-1 in a response to the pro-inflammatory environment of MI. In more detail, LAIR-1 expression differed on pro-inflammatory CD14^++^ CD16^−^ classical and CD14^++^ CD16^+^ intermediate monocytes. Although both are necessary for the removal of debris following MI, their effect is generally considered disproportionate and detrimental^[@CR11]^. Therefore, higher LAIR-1 expression in the acute phase after MI may be beneficial in suppressing pro-inflammatory monocyte activation to limit cardiac damage. Although LAIR-1 expression on monocytes decreases in the chronic phase after MI, the expression levels remain increased when compared to healthy controls. This is most probably linked to ongoing low-grade inflammatory response in the chronic phase of cardiac remodelling after MI^[@CR28]^. In addition, we observed higher LAIR-1 expression on neutrophils in the acute phase after MI compared to the chronic phase. Considering the observation that pro-inflammatory stimuli lead to the higher LAIR-1 expression on neutrophils^[@CR29]^, this is in agreement with the strong inflammatory response directly after MI.
Next to increased LAIR-1 expression on monocytes and neutrophils, we also observed higher levels of sLAIR-1 in the acute phase after MI, which is in line with the observation that cell activation induces shedding of LAIR-1^[@CR30]^. Although the source of sLAIR-1 remains to be elucidated, this finding suggests that inflammation in the acute setting of MI increases LAIR-1 expression even more to result in both high expression levels and a high amount of LAIR-1 shedding. Contrarily, the levels of LAIR-2, mainly produced by stimulated CD4^+^ T-lymphocytes, are lower in the acute phase of MI. This difference might be (partially) explained by the relatively decreased number of circulating CD4^+^ T-cells in the acute stage after MI compared to the chronic phase^[@CR31]^. Though both sLAIR-1 and LAIR-2 are natural antagonists of cell-bound LAIR-1, LAIR-2 has been shown to be far more potent than sLAIR-1.
These findings in patients prompted us to study if LAIR-1 is causally involved in ischemia reperfusion injury in the heart. However, in mice, the absence of LAIR-1, did not affect infarct size or cardiac remodelling after MI. Although leukocyte activating receptors and inflammation are widely recognized as important players in ischemia-reperfusion injury and remodelling after MI^[@CR32]--[@CR34]^, and despite the regulation in LAIR expression in MI patients, we were not able to establish a causal role for LAIR-1 deficiency in this regard.
The extent of ischemia-reperfusion injury is in agreement with previously performed experiments in WT mice in our laboratory^[@CR34]--[@CR36]^, infarct size after myocardial ischemia-reperfusion did not differ between WT and LAIR-1^−/−^ mice. We anticipated on increased reperfusion injury in the LAIR-1^−/−^ mice as a consequence of enhanced cellular infiltration and inflammation. However, the inflammatory response assessed in various tissues in the acute (3 days) and more chronic (28 days) inflammatory phase after MI did not differ between WT and LAIR-1^−/−^ mice. In addition, the deposition of collagen, as well as the extent of cardiac remodelling at 28 days was comparable to those observed in previously performed experiments^[@CR37],[@CR38]^, but did not differ between both groups.
Activating leukocyte receptors^[@CR34],[@CR39],[@CR40]^ and costimulatory molecules^[@CR41]^ have been shown to play an important role in myocardial reperfusion injury through modulation of the inflammatory response, whereas studies on inhibitory receptors or co-inhibitory molecules are lacking. The inhibitory effect of LAIR-1 may not provide sufficient potency for the extent of tissue damage and severity of the inflammatory response in the present model, as was previously shown *in vitro* ^[@CR42]^. This is in agreement with the observation that LAIR-1 has been shown to be primarily involved in low-grade chronic inflammatory diseases, such as cancer^[@CR19],[@CR43]^ and chronic contact dermatitis^[@CR44]^, but not so much in acute, high grade, inflammatory responses as observed in experimental autoimmune encephalitis and LPS injection^[@CR45]^. Although the chronic phase of myocardial remodelling shows a somewhat less inflammatory response than the acute phase of experimental MI, leukocyte influx is still impressive^[@CR46]^.
In addition, cell activation starts in the bloodstream^[@CR47]^, whereas inhibition of LAIR-1 is expected to occur predominantly upon the encounter of collagen in the heart. This may either be too late to efficiently inhibit the already initiated pro-inflammatory cascade, or the sheer amount of collagen ligands is too low to induce robust activation of LAIR-1. Moreover, other inhibitory receptors and/or pathways could have compensated for the absence of LAIR-1.
In conclusion, LAIR-1 expression on monocytes and neutrophils is increased in patients 3 days after MI. Though, in mice, the absence of LAIR-1 does not influence infarct size, nor does it affect inflammation, fibrosis formation and adverse left ventricular remodelling in mice four weeks after acute MI.
Electronic supplementary material
=================================
{#Sec23}
Supplementary data
Guilielmus H.J.M. Ellenbroek and Judith J. de Haan contributed equally to this work.
**Electronic supplementary material**
**Supplementary information** accompanies this paper at 10.1038/s41598-017-13678-5.
**Publisher\'s note:** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
The authors gratefully acknowledge Ben van Middelaar for technical assistance and Sjors Fens for logistic support. Publication of this work has been supported by ZonMW 'Meer Kennis met Minder Dieren' program (MKMD grant 114024043). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
G.E., J.d.H., S.d.J.: manuscript draft. G.E., J.d.H., S.d.J., F.A., I.H., L.M., S.d.J.: conception and design. G.E., J.d.H., B.v.K., M.B., S.v.d.W., M.S., S.d.J.: acquisition of data and analyses. G.E., J.d.H., B.v.K., M.B., S.v.d.W., M.S., S.d.J., F.A., L.T., M.-J.G,. I.H., P.D., G.P., L.M., S.d.J.: critically revising the manuscript for important intellectual content.
Competing Interests {#FPar1}
===================
The authors declare that they have no competing interests.
| {
"pile_set_name": "PubMed Central"
} |
The views expressed in this Commentary do not necessarily reflect the views of this journal or of ASM.
COMMENTARY {#h0.0}
==========
The remarkable achievements made by filariasis control programs testify to the success possible with focused effort and investment in tropical health interventions. Human infections with *Onchocerca volvulus*, the cause of onchocerciasis, or river blindness, and *Wuchereria bancrofti*, the major cause of lymphatic filariasis, have been the subjects of prolonged mass drug administration (MDA) and vector control campaigns that have led to marked decreases in the pathology, transmission, and incidence of these diseases ([@B1], [@B2]). Local eradication has been achieved in some areas for both, with the notable example of the all-but-complete elimination of onchocerciasis from the Americas ([@B3]).
MDA campaigns distribute drugs that eliminate the microfilarial (MF; larval) stages of these long-lived nematode parasites for many months after a single dose. Diethylcarbamazine (DEC) is used along with albendazole for the control of lymphatic filariasis in areas where other filarial infections are not coendemic. Ivermectin (IVM) is used for onchocerciasis and for lymphatic filariasis in regions where both infections occur. Removing microfilariae blocks the transmission of both parasites and reduces the pathology of onchocerciasis. These strategies have worked remarkably well, with one important exception: both drugs may cause severe adverse events (SAEs), including death, in individuals harboring heavy MF burdens of *Loa loa*, another filarial nematode ([@B4][@B5][@B6]). Because of this, onchocerciasis control programs have been limited in regions where loiasis occurs; clinical experience has shown that individuals with burdens of \>30,000 MF/ml blood are at particular risk of IVM-related SAEs ([@B4][@B5][@B8]).
Infections with *L. loa* have generally been considered to cause little to no overt pathology ([@B7], [@B8]). The distribution of *L. loa* coincides with that of *O. volvulus* in important parts of West and Central Africa ([@B6][@B7][@B8]), a factor that looms as an impediment to the timely eradication of *O. volvulus* infection via expanded IVM-based MDA campaigns in these areas ([@B9]). As the considerable majority of people infected with *L. loa* have MF levels well below the risk threshold ([@B5][@B6][@B8]), control through chemotherapy could be ramped up if a rapid and convenient diagnostic platform capable of identifying at-risk patients and, hence, excluding them from treatment, could be devised and implemented. For safety reasons, the threshold for inclusion of patients in an IVM treatment regimen is usually set at 30,000 MF/ml ([@B5][@B6][@B8]).
It is against this background that the new work by Drame et al. ([@B10]) stands out as a highly significant advance. Diagnosis of MF loads in loiasis patients has typically been done by counting MF in a small sample of blood under a microscope, a method that is invasive, time consuming, and sometimes difficult to use for quantification. The Nutman laboratory at the United States NIAID has been at the forefront of developing innovations to improve the detection and quantification of *Loa* MF, including nucleic acid-based methods ([@B11]). For use in the field, a loop-mediated isothermal amplification (LAMP) method is most advantageous in this category, since thermocycling is not required ([@B11]). This technique can identify individuals with high MF loads (\>30,000/ml) but requires training and is still invasive, as it requires a blood sample to obtain MF. An impressive advance in convenience, cost, and time was the development of an off-the-shelf hand-held cell counter ([@B12]). Blood passing through size gates in this instrument leads to sequestration of the MF in a format that allows quantification. The technique is very rapid but, again, is invasive. Recently, this group described a rapid and sensitive technique for *Loa* MF quantification based on a cell phone microscope with software that can determine MF abundance based on pixel changes in two sequential recordings ([@B13]). It also provides a rapid method to identify individuals who should be excluded from IVM treatment, but this method also requires a blood sample.
The major advance in the work reported by Drame et al. ([@B10]) is eliminating the need for invasive procedures by identifying a biomarker of *Loa* MF present in urine. Based on the insight that some proteins can escape the renal filtering process, the authors searched for *Loa* proteins in urine obtained from a loiasis patient. Proteins of \>3 kDa were precipitated with acetone and subjected to tryptic digestion. The resulting peptides were separated by reverse-phase liquid chromatography (LC) and submitted for tandem mass spectrometry (MS) analysis. A number of candidate *Loa* proteins were identified by this approach, all of which were found in MF. Filtering through bioinformatics screens eliminated proteins with significant homology to human proteins or proteins from other filariae and allowed the authors to prioritize a single *Loa* MF protein, termed LOAG_16927, which is unique to *L. loa* and has no predicted function.
The authors then developed a novel enzyme-linked immunosorbent assay (ELISA) to quantify the levels of this antigen in plasma based on a modification of the luciferase immunoprecipitation system (LIPS) format ([@B14]); using defined amounts of recombinant LOAG_16927 for competition, the authors developed a competitive LIPS assay that provided a reasonable correlation between antigen abundance and MF density in samples obtained from loiasis patients. This is the first potentially quantitative, MF-specific, antigen-based diagnostic test described for filariases and, as such, represents a significant advance. As the authors note, however, a considerable amount of work remains to be done to bring this assay to ready-to-use status for a field setting. In particular, it will have to be shown to be advantageous in terms of sensitivity, cost, time, or convenience compared to other techniques for quantifying *L. loa* MF in patient blood samples.
In addition to these important data, this paper suggests another intriguing avenue for research. It would be highly interesting to pursue the possibility that MF loads in loiasis patients could be quantified by an antigen detection assay using a urine sample. It is encouraging to note that a point-of-care circulating cathodic antigen (POC-CCA) test is commercially available for the urine-based diagnosis of schistosomiasis (see reference [@B15]), suggesting that global health applications of such tests are feasible. That assay is not intended to be accurately quantitative, whereas a test for the *L. loa* MF burden would need to distinguish individuals with MF loads in the high-risk category. For use in loiasis control programs, it is likely that a host protein with biochemical characteristics similar to those of LOAG_16927 would have to be included to control for hydration status and urine volume. Nonetheless, avoiding the need to take a blood sample could be an important step forward for the identification of individuals who should be excluded from IVM treatment in onchocerciasis campaigns. It is also possible that a urine-based assay would find value for population-based epidemiological analyses, an application that would not require rigorous quantification of the MF burden. While a significant proportion of *L. loa* infections seem to be amicrofilaremic ([@B5], [@B8]), the convenience of a noninvasive assay is an attractive feature for further investment in this area. An additional attraction of a urine-based assay is that the levels of MF appearing in the peripheral blood vary considerably over time, typically peaking in the early afternoon ([@B16]). Since MF are not thought to leave the bloodstream during periods of sequestration from the peripheral circulation, urinary antigen loads might be less variable than MF counts in blood as determined by microscopic or molecular methods. It would be well worth the investment needed to subject this possibility to experimental testing.
Additional applications of technology to detect filarial proteins in urine are readily apparent. As eradication programs proceed, it will be important to identify infected individuals in areas of low endemicity to allow targeted treatment when MDA campaigns are no longer justified. The current surveillance techniques for infected humans involve invasive procedures, including a blood sample for lymphatic filariasis and skin snips for onchocerciasis. Although MF stages of the causative parasites are not found in the blood, unlike loiasis, it is possible that a urinary protein unique to *W. bancrofti* or *O. volvulus* could be identified and an assay developed to generate yes/no responses in a urine sample. In fact, a protein that has homologs in all filariae of humans but not in other kinds of helminth parasites could also be useful, simply to identify affected individuals for more detailed parasitological analysis. Quantitative analyses would not be needed for such population surveys.
As Drame et al. point out, the development of a cheap, reliable platform capable of quantifying LOAG_16927 in a way that is diagnostically valid and economically valuable will be challenging. This work is nonetheless an exceptionally promising advance that merits additional urgent investment to complete its translation to practice. The sooner we have a method to allow the confident extension of filariasis control programs to areas where loiasis is coendemic with other filarial infections, the faster the feasible goal of eradication can be met.
**Citation** Geary TG. 2016. A step toward eradication of human filariases in areas where *Loa* is endemic. mBio 7(2):e00456-16. doi:10.1128/mBio.00456-16.
| {
"pile_set_name": "PubMed Central"
} |
{#sp1 .94}
| {
"pile_set_name": "PubMed Central"
} |
Background
==========
Extralobar pulmonary sequestration is a congenital anomaly resulting in the formation of ectopic lung tissue outside the normal lung with its own pleural investment, but no communication with the normal tracheobronchial tree. Without associated malformations, symptomatic cases are rare, and are usually found incidentally in adulthood. In such cases, differential diagnosis from other mass lesions is needed.
We present a case of necrotic extralobar pulmonary sequestration in a patient who presented with a back pain, which we were unable to diagnose correctly. We also discuss imaging features in comparison with a non-complicated case.
Case Reports
============
Case 1
------
A healthy and asymptomatic 50-year-old woman showed suspected right pleural effusion on abdominal ultrasonography carried out for screening. Chest computed tomography (CT) showed a well-circumscribed dome-like solid mass with cysts and branching fluid collection above the right dorsal diaphragm ([Figure 1](#f1-poljradiol-79-145){ref-type="fig"}). The abutting lung was displaced and slightly compressed by the mass, and internal structures were not connected to the mass. Laboratory data including tumor markers (carcinoembryonic antigen, 1.9 ng/mL; cytokeratin fragment, 1.4 ng/mL; Pro-gastrin-releasing peptide, 18.2 pg/mL) were all within normal limits. Magnetic resonance imaging showed the solid part of the mass as a region of inhomogeneous low signal intensity on T2-weighted images and enhanced strongly on contrast-enhanced T1-weighted imaging, while cysts and fluid collection within the mass were isointense with water on both T1- and T2-weighted images ([Figures 2](#f2-poljradiol-79-145){ref-type="fig"} and [3](#f3-poljradiol-79-145){ref-type="fig"}). On angiography, the mass was supplied by the dilated right inferior phrenic artery and showed strong enhancement but no drainage vein was detected. Extralobar pulmonary sequestration was suspected because of the presence of branching structures within the mass and the blood supply from the right inferior phrenic artery. Two days after preoperative embolization as operative support, the mass was resected by video-assisted thoracoscopic surgery. On surgical findings, an ectopic lung tissue-like mass was located above the right diaphragm, with its own pleural investment. The mass was connected to the diaphragm by a pedicle, without any connection to the normal lung, and was resected at the pedicle.
Macroscopically, the mass comprised solid and cystic lesions. Microscopically, various sizes of cystic dilated alveoli and fibrotic stroma were present and extralobar pulmonary sequestration was diagnosed.
Case 2
------
A 38-year-old healthy woman presented after a sudden onset of back pain that disappeared the next day, then reappeared the day after. Laboratory tests showed slight elevations in C-reactive protein (CRP) (1.91 mg/dL), creatine phosphokinase (221 IU/L), and lactate dehydrogenase (219 IU/L), and other values including white blood cell count were within normal range. Electrocardiography, abdominal ultrasonography, gastrointestinal endoscopy and chest radiography were normal. Contrast-enhanced CT of the chest and abdomen revealed a 4-cm fusiform mass in the azygo-esophageal recess ([Figure 4](#f4-poljradiol-79-145){ref-type="fig"}). That mass demonstrated iso-density with muscles, and only the peripheral rim was enhanced. It was abutting the lung, esophagus, vertebrae, and left atrium, but those surrounding structures did not show any abnormalities. A small amount of pleural effusion was seen on the right side, i.e. at the location of the mass. No other lesions that might have caused the symptoms were evident. On magnetic resonance imaging, the mass showed inhomogeneous low signal with branching signal hyperintensity on T2-weighted images, slight hyperintensity on T1-weighted images, and contrast enhancement only at the periphery ([Figures 5](#f5-poljradiol-79-145){ref-type="fig"} and [6](#f6-poljradiol-79-145){ref-type="fig"}). Four days after onset, pleural effusion increased and CRP was elevated to 5.52 mg/dL. A mass was suspected -- an acute necrotic solid tumor or complicated (hemorrhagic or infectious) cystic lesion, such as solitary fibrous tumor, neurogenic, foregut cyst or cystic teratoma. Although we could not establish the diagnosis, the mass was considered the cause of the symptoms, and video-assisted thoracoscopic surgery was attempted. Intraoperatively, a dark red-brown mass and small amounts of hemorrhagic pleural effusion were identified in the pleural cavity. Parietal pleura near the mass appeared red, which suggested spreading inflammation. The mass was easily separated from the surrounding tissue, and was connected to the mediastinum by a twisted pedicle. The pedicle passed around the esophagus, but connection to the aorta could not be recognized. The mass was resected at the pedicle. Microscopically, the mass contained circular lining cartilage and fat, and received its blood supply from an artery in the pedicle ([Figure 7](#f7-poljradiol-79-145){ref-type="fig"}). That vessel was an elastic artery, with a large diameter compared to the size of the mass. The mass displayed total hemorrhagic necrosis of the remaining native structures without granulation tissue, indicating a pattern of necrosis arising within the past several weeks. We diagnosed necrotic extralobar pulmonary sequestration with torsion based on the clinical course and surgical and pathological findings. Symptom completely resolved following resection.
Discussion
==========
Pulmonary sequestration is defined as ectopic lung tissue lacking normal communication to the tracheobronchial tree and receiving blood supply from systemic vessels \[[@b1-poljradiol-79-145]\]. This pathology is classified into intra- and extralobar types.
Intralobar pulmonary sequestration is more frequent than extralobar, and is covered by the visceral pleura in common with the normal lung. Although some cases are congenital, the majority is a result of changes to normal lung tissue after recurrent inflammation \[[@b2-poljradiol-79-145]\].
On the other hand, extralobar pulmonary sequestration is an absolutely congenital maldevelopment with its own visceral pleura, and symptomatic cases are rare \[[@b3-poljradiol-79-145]\]. Diagnosis is often made in the first year of life because of an associated anomaly such as diaphragmatic hernia, bronchogenic cyst or pericardial defect. In the absence of such malformations, extralobar pulmonary sequestration is usually found incidentally in adulthood, and must be differentiated from other mass lesions including neoplasms. Extralobar pulmonary sequestration in an adult is occasionally symptomatic, without other concomitant anomalies, because of infection, inflammation, hemorrhage, or infarction.
Extralobar pulmonary sequestration is typically found in the thorax on the left side (65--90%), and at the posterior costodiaphragmatic sulcus between the diaphragm and lower lobe of the lung (63--77%). It is sometimes found in the mediastinum, interperitoneally, or in the pericardium. Extralobar pulmonary sequestrations are between 0.5 and 15.0 cm, usually 3 to 6 cm, pyramidal or ovoid in shape, and with a vascular pedicle. Grossly, the tissue resembles lung tissue, and is covered by glistening pleural surface. The feeding artery is often from the aorta (80%), but can be from the splenic, gastric, subclavian, intercostal or internal mammary artery.
Microscopic findings vary depending on the tissue composition and degree of associated changes, from nearly normal lung tissue to tissue resembling cystic adenomatoid malformation, composed of dilated alveoli, thickened stroma, and inflamed fibrotic stroma with mucin-filled airways \[[@b4-poljradiol-79-145],[@b5-poljradiol-79-145]\]. Because secretions are retained, with no drainage due to the lack of communication with the normal tracheobronchial tree, recurrent inflammation and cystic and fibrotic changes occur over time.
A typical radiological finding is a mass in the lower hemithorax, and it is important for the diagnosis of extralobar pulmonary sequestration to identify the systemic vascular supply using any modality -- angiography, ultrasonography, CT, or magnetic resonance imaging. Lee et al. reported that 3-dimensional CT angiography could clearly show the feeding artery and drainage vein \[[@b6-poljradiol-79-145]\]. In Case 1, these could not be identified, probably because the vessels were too thin, and arterial-phase scans were not performed. The mass in Case 1 was inflamed, comprising cystic dilated alveoli and fibrotic stroma. Imaging reflected cystic dilated alveoli as cystic lesions, and stroma without dilated alveoli as solid parts, with the latter showing an inhomogeneous low signal on T2WI and contrast enhancement.
Infarcted extralobar pulmonary sequestration was reported in 9 cases (as well as another 6 cases in Japanese literature), and 3 cases showed torsion \[[@b7-poljradiol-79-145]--[@b9-poljradiol-79-145]\]. This pathology manifests with chest or abdominal pain. Huang et al. reported a case with prodromal intermittent pain \[[@b8-poljradiol-79-145]\]. Symptoms in the present case were also intermittent over the initial few days, possibly indicating changes in the state of torsion.
Imaging features of extralobar pulmonary sequestration with necrosis reflect the hemorrhagic necrosis, and were reported as: 1) a 4--5-cm ovoid-pyramidal mass in the paravertebral region; 2) a homogeneous high-density mass enhancing only on the periphery on CT; 3) reactive pleural effusion or hemothorax; and 4) no feeding artery identified \[[@b7-poljradiol-79-145]--[@b9-poljradiol-79-145]\].
The lesion in Case 2 presented all these findings. The signal on magnetic resonance imaging reflects acute hemorrhage and necrosis, and would change with the interval from onset. Probably because of thrombosis, no feeding could be detected \[[@b7-poljradiol-79-145]\]. Infarcted extralobar pulmonary sequestration without torsion was reported, and we thought this could have taken place because the abnormal vessel easily undergoes atheromatous changes \[[@b10-poljradiol-79-145]\].
In Case 2, the problem was differentiating the lesion from cystic tumor with hemorrhage or infection, because the mass was enhanced only on the periphery. The inhomogeneous low signal intensity and branching signal hyperintensity on T2-weighted images reflected the structures contained, and were also similar to the solid and fluid components of the mass in Case 1. This suggests that T2-weighted imaging might reflect the tissue structures before necrosis, to a certain degree, while other sequences cannot. Findings on T2-weighted images might thus help to distinguish these lesions from cystic tumor with hemorrhage or infection, which show homogeneous signal intensity, fluid-fluid level, and have a webbed appearance.
Conclusions
===========
We reported on two cases of extralobar pulmonary sequestration, including one case with torsion/necrosis. Radiographic imaging was able to reflect the state of hemorrhagic necrotizing lung tissue and only T2-weighted imaging reflected native structures even after infarction. Although infarcted extralobar pulmonary sequestration is a rare acute disease, understanding the pathology and its appearance with hemorrhagic necrosis may improve preoperative diagnosis.
**Conflict of interest**
The authors declare to have no conflict of interest.
{#f1-poljradiol-79-145}
{#f2-poljradiol-79-145}
{#f3-poljradiol-79-145}
{#f4-poljradiol-79-145}
{#f5-poljradiol-79-145}
{#f6-poljradiol-79-145}
{#f7-poljradiol-79-145}
[^1]: Study Design
[^2]: Data Collection
[^3]: Statistical Analysis
[^4]: Data Interpretation
[^5]: Manuscript Preparation
[^6]: Literature Search
[^7]: Funds Collection
| {
"pile_set_name": "PubMed Central"
} |
Introduction {#s1}
============
Colorectal cancer (CRC) is the third most common cancer and the second leading cause of cancer death in the United States. In 2010, approximately 142,000 people were diagnosed with CRC, and about 40% of these patients presented with advanced disease [@pone.0038422-Mnejja1]. Treatment for advanced CRC with chemotherapy is typically intended for disease control and palliation of symptoms only, and as a result, unresectable CRC remains an incurable disease. In order to improve clinical outcomes and develop new therapeutic approaches, the development of a reliable preclinical model to study CRC biology and drug sensitivities is required.
Mouse models of CRC remain one of the most useful tools to decipher the biological mechanisms underlying the oncogenic process. To date, a variety of genetically-engineered, carcinogen-induced and xenograft mouse models have been established [@pone.0038422-Uronis1], [@pone.0038422-Richmond1] and it is generally agreed that no one model is sufficient to elucidate all aspects of CRC etiology.
Genetically engineered mouse (GEM) models have been invaluable in establishing the role of many different genetic mutations and signal transduction pathways contributing to the oncogenic process and allow investigation in the context of an active immune system [@pone.0038422-Uronis1], [@pone.0038422-Richmond1]. However, many of these GEM models, primarily those involving mutation of the *APC* tumor suppressor gene, develop tumors in the small intestine rather than the colon. This makes longitudinal disease progression studies difficult in addition to lacking the genetic complexity observed in human cancers [@pone.0038422-Uronis1], [@pone.0038422-Richmond1].
Another widely used mouse models of CRC relies on the use of carcinogens to induce colorectal tumor development. Perhaps the most widely used carcinogen-based model is the Azoxymethane (AOM) model. Here, colorectal tumor development is initiated by AOM, a potent, colon-specific carcinogen through the formation of DNA adducts [@pone.0038422-Druckrey1]. Colorectal tumors derived using this model recapitulate key human pathological features observed in humans and allow investigation of the early stages of CRC. However tumor initiation and development is a time consuming process, often taking up to 6 months with tumor multiplicity and penetrance depending heavily on the mouse strain [@pone.0038422-Uronis1], [@pone.0038422-Papanikolaou1], [@pone.0038422-Kaiser1].
While GEM and carcinogen-based models have significantly enhanced our knowledge of the genetics and etiology of CRC, these models do not allow for accurate testing of cancer therapeutics to be used in the clinical setting [@pone.0038422-Kerbel1]. The most widely utilized *in vivo* model for the testing of anti-cancer drug efficacy and combinations is the xenograft model. Historically, xenografts have been established through the subcutaneous injection of genetically-defined human-derived cell lines into immune-compromised nude mice [@pone.0038422-Kendall1]. However, to date, the majority of these cell line-based xenograft models have failed to generate drug sensitivity data that translates into clinically relevant information [@pone.0038422-Kerbel1]. In addition, recent reports suggest that tumor-stroma interactions not present in cell line-based xenografts may represent an integral component in oncogenic potential and tumor drug response [@pone.0038422-Loeffler1], [@pone.0038422-Harris1]. Therefore, more recently, whole-tissue explants derived from human cancers including breast [@pone.0038422-Marangoni1], lung [@pone.0038422-Fichtner1], prostate [@pone.0038422-Pretlow1] and colorectal cancer [@pone.0038422-Arcaroli1]--[@pone.0038422-Fichtner2] have been established in an attempt to generate more clinically accurate and reliable xenograft models. However, these studies examined mainly early passage explants (\<5 generations) from predominantly primary tumors and therefore there remains the need to further characterize these models and evaluate how well they retain important characteristics of the original human tumor especially in metastatic disease.
10.1371/journal.pone.0038422.t001
###### Sites from which PDCCEs were derived.
{#pone-0038422-t001-1}
Sample ID Anatomic Location
----------- -------------------
CRC020 Liver
CRC025 Liver
CRC028 Colon
CRC030 Colon
CRC034 Liver
CRC039 Liver
CRC040 Liver
CRC043 Liver
CRC054 Liver
CRC057 Lung
CRC066 Colon
CRC075 Liver
CRC083 Lung
CRC093 Omentum
CRC096 Colon
CRC102 Liver
CRC103 Peritoneum
CRC105 Colon
CRC108 Liver
CRC119 Liver
CRC120 Liver
CRC133 Lung
CRC149 Liver
CRC159 Lung
CRC162 Liver
CRC167 Liver
CRC170 Liver
{#pone-0038422-g001}
10.1371/journal.pone.0038422.t002
###### Histological comparison of patient tumor and PDCCEs.
{#pone-0038422-t002-2}
Sample ID Source Generation %Tumor %Necrosis %Stroma %Tumor Gland Formation Histologic Type
----------- --------- ------------ -------- ----------- --------- ------------------------ -----------------------------------------
CRC008 Patient 65 5 25 80 Adenocarcinoma, NOS
PDCCE 3rd 70 10 20 70 Adenocarcinoma, NOS
CRC028 Patient 10 80 10 40 Adenocarcinoma w/signet ring
PDCCE 1st 25 5 70 40 Adenocarcinoma w/signet ring
CRC034 Patient 65 0 35 80 Adenocarcinoma, NOS
PDCCE 1st 85 5 10 75 Adenocarcinoma, NOS
CRC039 Patient 10 0 90 50 Adenocarcinoma w/mucinous
PDCCE 1st 60 0 40 70 Adenocarcinoma w/mucinous
PDCCE 6th 25 65 10 60 Adenocarcinoma w/mucinous
PDCCE 10th 35 50 15 80 Adenocarcinoma w/mucinous
CRC057 Patient 25 45 20 80 Adenocarcinoma, NOS
PDCCE 1st 40 60 0 80 Adenocarcinoma, NOS
PDCCE 5th 50 35 15 80 Adenocarcinoma, NOS
CRC066 Patient 25 45 30 80 Adenocarcinoma, NOS
PDCCE 1st 20 5 75 50 Adenocarcinoma, NOS
CRC075 Patient 35 30 25 90 Adenocarcinoma, NOS
PDCCE 1st 50 15 35 95 Adenocarcinoma, NOS
PDCCE 5th 85 5 10 75 Adenocarcinoma, NOS
PDCCE 11th 65 10 25 70 Adenocarcinoma, NOS
CRC093 Patient 15 30 55 60 Adenocarcinoma, NOS
PDCCE 1st 65 30 5 80 Adenocarcinoma, NOS
PDCCE 5th 80 10 10 85 Adenocarcinoma, NOS
CRC096 Patient 20 65 15 65 Adenocarcinoma, NOS
PDCCE 14th 65 30 5 90 Adenocarcinoma, NOS
CRC102 Patient 10 70 20 60 Adenocarcinoma, NOS
PDCCE 1st 80 3 17 80 Adenocarcinoma, NOS
CRC103 Patient 5 40 55 100 Adenocarcinoma, NOS
PDCCE 1st 10 90 0 95 Adenocarcinoma, NOS
PDCCE 6th 10 90 0 95 Adenocarcinoma, NOS
CRC120 Patient 35 0 65 50 Adenocarcinoma w/signet ring & mucinous
PDCCE 1st 85 0 15 70 Adenocarcinoma w/signet ring
CRC133 Patient 65 10 15 90 Adenocarcinoma, NOS
PDCCE 7th 10 85 5 80 Adenocarcinoma, NOS
CRC149 Patient 25 60 15 80 Adenocarcinoma, NOS
PDCCE 1st 40 55 5 70 Adenocarcinoma, NOS
CRC170 Patient 10 40 50 90 Adenocarcinoma, NOS
PDCCE 2nd 80 5 15 80 Adenocarcinoma, NOS
{#pone-0038422-g002}
In this study, we have performed a more comprehensive molecular and histological analysis of a panel of 27 matched patient-derived colorectal cancer explants (PDCCEs) from both primary and metastatic sites as an extension of our previous work [@pone.0038422-Kim1] in which we compared the gene expression profile of 14 matched PDCCEs and their corresponding human tumors. We now demonstrate that PDCCEs retain global gene expression patterns, oncogene mutation status and histological parameters present in the original human cancers. Altogether these findings suggest that PDCCEs have the potential to serve as a reliable preclinical model that can be used to develop and characterize new therapeutic targets for patients with CRC.
Materials and Methods {#s2}
=====================
Tumor Samples/Ethics Statement {#s2a}
------------------------------
A total of 27 human samples were obtained for genomic and histological analysis. All patients provided written consent to have tissue stored and used for research. Samples used for analysis in the laboratory were de-identified and not linked with any personal health information (PHI). All parts of this study were approved by the Duke Institutional Review Board. All animal studies were performed under a Duke University Institutional Animal Care and Use Committee (IACUC) approved protocol.
{#pone-0038422-g003}
Generation of Patient-Derived Colorectal Cancer Explants (PDCCEs) {#s2b}
-----------------------------------------------------------------
Colorectal tumors (both primary and metastatic) at time of surgery were collected under a Duke IRB approved protocol (Pro00002435). The tissues were washed with phosphate buffered saline (PBS) and then minced into pieces approximately ∼2 mm in size and injected into the flanks of 4-week-old NOD.CB17-PrkdcSCID-J mice obtained from Jackson Laboratories under a Duke IACUC approved protocol. Mice were observed and tumors measured with vernier calipers until the volume of the tumor ((V = L×2W×0.52 (L = longest diameter, W = shortest diameter)) reached ∼1,000 mm^3^. Tumors were then harvested, minced and re-implanted as described above until stable PDCCEs were established. At each generation, tumors were harvested and either fixed in 10% neutral buffered formalin (NBF), snap frozen in liquid nitrogen or frozen in optimal cutting temperature (OCT) medium on dry ice for further analysis.
10.1371/journal.pone.0038422.t003
###### Patient-derived colorectal cancer explants' *KRAS* and *BRAF* mutation status.
{#pone-0038422-t003-3}
Sample ID *KRAS* Status *BRAF* status
----------- ----------------------------------- ---------------
CRC020 WT (patient); C12 GGT\>GCT(mouse) WT
CRC025 WT WT
CRC028 WT WT
CRC030 WT WT
CRC034 WT WT
CRC039 C12 GGT\>GCT WT
CRC040 WT WT
CRC043 C12 GGT\> AGT WT
CRC054 C12 GGT\>GAT WT
CRC057 C12 GGT\>GAT WT
CRC066 WT WT
CRC075 WT WT
CRC083 C12 GGT\>TGT WT
CRC093 WT WT
CRC096 C12 GGT\>GCT WT
CRC102 WT WT
CRC103 C12 GGT\>GTT WT
CRC105 C13 GGC\>GAC WT
CRC108 WT WT
CRC119 C12 GGT\>GTT WT
CRC120 WT WT
CRC133 WT WT
CRC149 WT WT
CRC159 C12 GGT\>AGT WT
CRC162 WT WT
CRC167 C12 GGT\>GCT WT
CRC170 C13 GGC\>GAC WT
Histological Preparation and Examination {#s2c}
----------------------------------------
Paraffin-embedded PDCCE tissues were sectioned in 6 µm intervals and stained with hemotoxylin and eosin (H&E). Each sample was evaluated by a trained pathologist for the following histological criteria: histologic type, CDX-2 positivity, and relative percentage of tumor, necrosis, stroma, tumor gland formation and CDX-2 positive nuclei. All tissues were examined using \>10 high-powered fields per section. Tumor nuclei were evaluated for CDX-2 staining using a standard quantitative scale of 0, 1+, 2+ and 3+. Staining of tumor nuclei at 2+ and 3+ was considered positive and all cases considered positive exhibited at least 20% of tumor nuclei with staining.
Oncogene Mutation Analysis {#s2d}
--------------------------
Genomic DNA was isolated from snap frozen PDCCE tissues using a Qiagen genomic DNA isolation kit. Samples were diluted to 10 ng/µl and PCR was performed using the following primers for *KRAS*: forward 5′ GTGTGACATGTTCTAATATAGTCA 3′; reverse 5′ GAATGGTCCTGCACCAGTAA 3′ and *BRAF*: forward 5′ TCATAATGCTTGCTCTGATAGGA 3′; reverse 5′ GGCCAAAAATTTAATCAGTGGA 3′. Amplicons were sequenced by conventional methods using the forward primers.
Microarray Analysis {#s2e}
-------------------
RNA was isolated from snap-frozen PDCCE tissues using a Qiagen RNA/DNA Allprep kit, converted to cDNA and labeled by one cycle IVT. IVT labeled cDNAs were prepared according to the manufacturer's instructions, and targets hybridized to the Human U133A 2.0 GeneChip and read on an Affymetrix array scanner. Raw data was converted to. CEL files and RMA normalized. CEL files (GSE35144) are available at the Gene Expression Omnibus (GEO) data repository (<http://www.ncbi.nlm.nih.gov/geo/>). To check for sample outliers and batch effects, 3D principal components analysis of the global gene expression was performed. Batch effects were normalized using the ComBat algorithm (<http://jlab.byu.edu/ComBat/>) [@pone.0038422-Johnson1]. Unsupervised hierarchical clustering of the human tumors and matching PDCCEs was performed on the 20% of genes with the greatest coefficient of variation. Agglomerative clusters were generated using the pearson correlation coefficient and complete linkage using the R program (The R Foundation for Statistical Computing).
Software Used for Analysis {#s2f}
--------------------------
The R statistical software package is available at [www.r-project.org](http://www.r-project.org). The Bioconductor R package is available at [www.bioconductor.org](http://www.bioconductor.org). ComBat is available as an R script at <http://jlab.byu.edu/ComBat/>. Graphpad Prism is a product of Graphpad Software (La Jolla, CA, USA) and is available at [www.graphpad.com/prism/prism.htm](http://www.graphpad.com/prism/prism.htm).
Results {#s3}
=======
Histological Evaluation of PDCCEs {#s3a}
---------------------------------
A panel of 27 patient-derived colorectal cancer explants (PDCCEs) by direct transplantation of human colorectal cancer (CRC) tissues into NOD-SCID mice was created in this study. [Table 1](#pone-0038422-t001){ref-type="table"} shows the origin of the patient tumor and a total of 5 primary PDCCEs and 22 metastatic PDCCEs were generated. To assess the extent to which *in vivo* models of patient-derived colorectal cancer explants (PDCEEs) accurately recapitulate and can therefore serve as a model of the human condition, we investigated whether PDCCEs retain key biological features inherent to individual human colorectal cancers (CRC) over time. First, to evaluate the extent to which histological parameters are retained after xeno-transplantation, two independent PDCCEs were passaged through \>10 generations and evaluated histologically. Both PDCCEs examined exhibited pathological features remarkably consistent with the original patient tumor through 11 generations ([Figure 1](#pone-0038422-g001){ref-type="fig"}). Next, a comprehensive histological evaluation performed on a sub-panel of 15 matched PDCCEs and original banked tissues revealed that 15/15 PDCCEs retained pathological features similar to those observed in the matched human tumor and were characterized as histologically identical to their matched original banked sample ([Table 2](#pone-0038422-t002){ref-type="table"}). Even after 11 generations, PDCCEs retained the ability to form glands and contained CDX-2 positive nuclei comparable to the first generation PDCCEs ([Figure 2](#pone-0038422-g002){ref-type="fig"}). These data demonstrate that the histological features present in colorectal cancer, including the formation of glands and presence of stromal components are retained even in late passage explants, suggesting that unlike CRC cell line-derived xenografts, the PDCCE model provides us with a research tool that recapitulates the human condition generally not observed in other models.
PDCCEs Retain Basic Global Gene Expression Profiles Inherent to Human Colorectal Cancers {#s3b}
----------------------------------------------------------------------------------------
Next, to further evaluate the extent to which PDCCEs represent their primary human counterparts, we analyzed 27 matched patient tumors and PDCCEs by microarray analysis. Patient tumor and PDCCE gene expression data was first normalized using ComBat to minimize batch effects. Unsupervised hierarchical clustering analysis was then performed on the normalized data set and revealed three distinct clusters ([Figure 3](#pone-0038422-g003){ref-type="fig"}). Of the 27 matched patient tumor and PDCCEs, 22 pairs (81%) fell within the same cluster based on the dendrogram and 18 PDCCEs (66%) clustered directly with the original tumor sample. Altogether, these data suggest that basic global gene-expression patterns are preserved between PDCCEs and their original human counterparts.
Oncogene Mutation Status is Retained in PDCCEs {#s3c}
----------------------------------------------
For patients with advanced colorectal cancer, the testing of mutation status of oncogenes such as *KRAS* is required for guiding therapy. Specifically, patients with *KRAS* mutations show no benefit from treatment with EGFR inhibitors such as cetuximab or panitumumab, while patients whose tumors are *KRAS* WT derive benefit from anti-EGFR based therapies [@pone.0038422-Adelstein1], [@pone.0038422-Douillard1]. To determine whether these clinically-significant genomic parameters are maintained in PDCCEs, 27 matched PDCCEs and original patient samples were analyzed for *KRAS* and *BRAF* mutation status. Of the 27 matched pairs evaluated, 13 presented with activating *KRAS* mutations (codon 12 = 11; Codon 13 = 2) ([Table 3](#pone-0038422-t003){ref-type="table"}). Of these 27 matched pairs, 26/27 PDCCEs (96%) matched their original human counterpart suggesting that human colorectal cancer tissues maintained as mouse PDCCEs are genetically stable and retain oncogenic mutation status critical to CRC pathophysiology. All samples tested negative for *BRAF* mutations. Altogether, these data suggest that PDCCEs maintain the biologically complex histological, gene expression and mutation-based characteristics observed in human CRCs.
Discussion {#s4}
==========
To date, a number of mouse xenograft models have been established to investigate CRC etiology and treatment. To a large extent, these models have been generated using late passage cell lines derived from human CRCs and while significant treatment-induced tumor responses have been observed in these models, they are rarely predictive of tumor response in human patients [@pone.0038422-Kerbel1]. This is likely due in these models, at least in part, to the inherent lack of stroma in tumor-derived epithelial cell lines. Mounting evidence indicates that paracrine signaling and extracellular matrix components supplied by neighboring stromal cells play a significant role in the oncogenic potential of colorectal carcinoma and that modulation of these stromal interactions directly impact the efficacy of chemotherapy on tumor response [@pone.0038422-Loeffler1], [@pone.0038422-Harris1].
Recent attempts have been made to generate mouse xenograft models of CRC by direct transplantation of human colon tumors into immune-compromised mice [@pone.0038422-Arcaroli1]--[@pone.0038422-Fichtner2], [@pone.0038422-Mischek1]. Poupon and coworkers reported that the passage of human colon cancer tissues through a xenograft stage significantly improves the success rate of cell line derivation from human CRC metastases [@pone.0038422-DanglesMarie1]. More recently, Hohenadl and coworkers reported that histological characteristics and oncogene expression levels are retained in early passage CRC xenografts [@pone.0038422-Mischek1] while Fichtner et al., and Messersmith et al, used panels of 15- and 10 human CRC explants respectively to evaluate drug sensitivity [@pone.0038422-Arcaroli1], [@pone.0038422-Fichtner2]. Additionally, these studies used patient-derived explants mainly from the primary site and as metastatic tumors tend to be more aggressive and are more likely to differentiate, it remains unclear if PDCCEs generated from metastatic sites would maintain similarity to the original tumor. While these studies have begun to underscore the value of explant models in CRC research, a more comprehensive histological and molecular analysis on a larger panel of human CRC explants are needed to justify their use as a preclinical model to perform accurate drug efficacy analysis and predictive biomarker identification.
In this study, we demonstrate that PDCCEs generated from human adenocarcinomas with varying histological features each retain the parameters of the tumor from which they were derived at the histological, global RNA expression and oncogene mutation levels. Despite the existence of differences in the percentage of tumor stroma present between the original human tissues and those xenografted into mice, our study focuses on the malignant epithelial cells. First, the histological architecture inherent to colorectal adenocarcinoma, primarily the ability to form dysplastic glands as well as the presence of CDX-2 positive nuclei is maintained in the PDCCEs throughout multiple passages (\>10). Next, we compared the gene expression profiles between matched PDCCEs and its corresponding patient tumor and observed that 18/27 (66%) of the samples clustered directly together and 22/27 (81%) clustered within the major cluster as defined by the dendrogram.
We speculate that the 9 PDCCE samples that did not cluster directly with their corresponding original tumor may have been due to the inherent heterogeneous nature of CRC. It is plausible that the original CRC tumor samples corresponding to these 9 PDCCEs harbored small sub-populations bearing additional oncogenic events. This would in turn confer a growth advantage to these populations after being transplanted into the mouse, causing the PDCCE to have a different genetic composition than the original tissue from which it was derived. In support of this notion, it appears that most variation between the primary tumor and its PDCCE occur in early PDCCE passages and that less variation occurs through the process of passaging. For example, PDCCE CRC105 clustered with the original patient sample at PDCCE passages 1 and 11 while PDCCE CRC149 did not cluster with its original sample at either passage 1 or 5 suggesting that genetic changes occur predominantly in early passages and are maintained through later passages. It is also possible, that in these 9 samples, there may have been a greater stromal contamination resulting in a difference in their clustering pattern. These results suggest that there are indeed intrinsic differences between the matched patient tumor and PDCCE and extrapolations drawn from these models must be done so carefully. However, overall our findings suggest that the PDCCE model has the potential to be used in the investigation of new therapeutic agents that target both the malignant epithelial tumor architecture and/or stromal component and that PDCCEs can be maintained for 10 or more generations while retaining key histological parameters.
Finally, we evaluated the *KRAS* and *BRAF* mutation status of the PDCCEs and showed that the status of all but one of the oncogene mutations was retained. We observed one case (CRC020) in which a *KRAS* activating mutation was present in the PDCCE but was not detected in the original patient samples despite the fact that these samples clustered together by unsupervised cluster analysis. It is most likely that a small, undetectable population of *KRAS* mutant cells was present in the patients tumor at the time of surgical resection and that the growth advantage conferred by *KRAS* activation allowed for subsequent expansion of *KRAS* mutant cells during early PDCCE passages.
Although any single mouse model will never fully recapitulate actual findings in patients, the use of preclinical models is necessary and practical for the development of therapeutic agents and biomarkers and a crucial first step in bringing these agents to the clinic. We do realize the limitations of our model and that any finding must undergo rigorous testing to gauge its accuracy, reliability, and reproducibility and must also be retrospectively validated in multiple patient samples. Nevertheless we feel that our preclinical mouse model has the potential to be used to identify and test novel combinations of therapeutic agents and to also develop both predictive and prognostic biomarkers, which can then be systematically brought forth into the clinical setting.
The authors wish to thank the Duke microarray core facility for collecting the microarray data.
**Competing Interests:**The authors have declared that no competing interests exist.
**Funding:**This work was supported by grants from the Mentored Research Scholars Grant (119824-MRSG-10-195-01-TBG) from the American Cancer Society ([www.cancer.org](http://www.cancer.org)). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
[^1]: Conceived and designed the experiments: TO CM MM HKL BC DH. Performed the experiments: JU SM XY. Analyzed the data: JU SM DH. Contributed reagents/materials/analysis tools: SM CM MM HKL BC DH. Wrote the paper: JU DH.
| {
"pile_set_name": "PubMed Central"
} |
1. Introduction {#sec1-sensors-16-00419}
===============
Polarization-maintaining fiber (PMF) is a crucial component of integrated optical sensors and fiber-optic interferometers \[[@B1-sensors-16-00419],[@B2-sensors-16-00419]\]. Called polarization mode coupling (PMC), the optical power coupling between two orthogonal PMF polarization modes can be generated by inner structural imperfections or external perturbations along the PMF \[[@B3-sensors-16-00419],[@B4-sensors-16-00419]\]. PMC could be utilized to evaluate the characteristics of polarization devices, such as the PER of Y-waveguides \[[@B5-sensors-16-00419]\] and the angular alignment between PMFs \[[@B6-sensors-16-00419]\]. Typical PMC measurements based on white light interferometer (WLI) focus on the 1st-order interference produced by the exciting mode and coupling mode with only one occurrence of PMC \[[@B7-sensors-16-00419]\], because there is a consistent one-to-one correspondence between the 1st-order interference and the real perturbation point in the PMF under test. In reality, the light in the fast-axis caused from the coupling at a perturbation point will couple back to the slow-axis at the subsequent coupling points along fiber. If there are multiple perturbation points in a PMF under test, it will generate interference between the exciting mode and the coupling mode with more than one occurrence of PMC in the output signals of WLI. Additionally, the interference---We call it high-order interference (HOI)---Has been detected by employing the measurement system with a dynamic range of 90 dB reported in previous works \[[@B5-sensors-16-00419],[@B8-sensors-16-00419]\].
As typical multiple coupling applications, the Lyot filter and Lyot depolarizer have been employed successfully in interferometric fiber optical gyroscopes (IFOGs) \[[@B9-sensors-16-00419],[@B10-sensors-16-00419]\]. However, the HOI produced in PMF, which indicates no realistic coupling points, will be confused with the 1st-order interference without a clear analysis on the origin of HOI. For instance, the HOI in IFOG coil inspection and PMF-based sensors will cause misjudgment of the distribution of stress or spliced points. Litton Corporation has pointed out that there are thousands of possible interferences, including HOI, for a single-axis IFOG based on PMF and polarization-maintaining (PM) components \[[@B11-sensors-16-00419]\]. In the case of distribution sensors, it has been reported that with merely several coupling points there will be many spurious interference signals. Chen *et al.* suggested that "spurious peaks" will occur inevitably due to many coupling points along PMF, and it has become one of the major problems that limits the multiplexing capacity of WLI systems \[[@B12-sensors-16-00419]\]. Wang *et al.* analyzed the influence of the 2nd-order interference, which is called as the ghost coupling \[[@B13-sensors-16-00419]\], on the distributed PMC measurements by using a rotatable half-wave plate \[[@B14-sensors-16-00419]\]. Furthermore, these reports only focused on the 2nd-order interference and did not propose a universal method to identify HOI. Therefore, it is significant to determine the characteristics of HOI brought by multiple perturbation points in PMF. The traditional analysis methods for the polarization light transmission through polarization devices based on Jones matrix \[[@B15-sensors-16-00419]\] or Mueller matrix \[[@B16-sensors-16-00419]\] are also applicable to HOI. However, the computation will become extremely complex with the number increase of coupling points by these methods.
In this paper, an optical path tracking (OPT) method is presented for simplifying the analysis of polarization light transmission along PMFs with multiple perturbation points. A brief description of the OPT method is provided as the following three steps: (1) for a given scanning optical path difference (OPD), we divide an entire PMF into stable units in which the OPD is invariable; (2) we obtain the coupling intensity made by adjacent stable units and calculate the recursion formula; (3) we derive the general formulas of interference intensity for the entire PMF under test. It is demonstrated that the different HOIs will be suppressed or amplified depending on the different angle-related conditions. We present a method to identify HOI readily by altering the spliced angle between the polarizers' pigtails and PMF under test with WLI system, which is verified by a simple case of two coupling points along a PMF experimentally. Finally, the system errors induced by the angle of polarizers and spliced points are discussed, and the variation trends of intensities are obtained for different HOIs.
2. Model and Analysis {#sec2-sensors-16-00419}
=====================
2.1. WLI System with a Large Dynamic Range {#sec2dot1-sensors-16-00419}
------------------------------------------
The PMC measurement setup for fiber sensors based on WLI is shown in [Figure 1](#sensors-16-00419-f001){ref-type="fig"}. The white light from a superluminescent light-emitting diode (SLD) is divided into two beams through a 98:2 fiber coupler. Two percent of the light is for monitoring the output power of the light source, and the remaining light is polarized by a 0°-rotated polarizer 1. Then the linearly polarized light is launched into the slow-axis of PMF under test. A part of linearly polarized light along the slow-axis will be coupled into the orthogonal axis at a perturbation point of PMF. Then it will generate two optical paths (OPs) with orthogonally eigenmodes, which will induce OPD due to the birefringence $\Delta n$ of the PMF. Afterwards, the 1st-order coupling interferograms are detected with photodiodes (PD) by the scanning Mach-Zehnder interferometer (MZI) that will compensate the OPD.
The large dynamic range of system achieved in the previous works is improved from many aspects: Firstly, a differential detection is completed by adopting two PDs \[[@B8-sensors-16-00419]\]. Secondly, the dispersion of fiber-based WLI is compensated by inserting a segment of dispersion-shifted fiber (DSF) into one arm of MZI. Thirdly, we utilize the differential scanning MZI with two lenses to suppress the optical power fluctuation \[[@B17-sensors-16-00419]\]. As shown in [Figure 1](#sensors-16-00419-f001){ref-type="fig"}, a PMF with multiple perturbation points (denoted by point $X_{1}$, $X_{2}$, ..., $X_{J}$) is tested by WLI. At each perturbation point, light is coupled not only from the polarization mode along the slow-axis to that along the fast-axis, but also from the polarization mode along the fast-axis to that along the slow-axis. As previously reported in \[[@B5-sensors-16-00419]\], the resolution of the developed system can achieve nearly −90 dB, which can be utilized to evaluate the Y-waveguide with ultra-high PER. In this case, a great number of spurious interferograms, referring to the HOIs, will appear in spatial domain with large dynamic range.
2.2. Optical Path Tracking (OPT) Method {#sec2dot2-sensors-16-00419}
---------------------------------------
It has been recognized that a pair of OPs with an OPD less than the coherence length will suffer interference at the output of MZI and lead to an interferogram. For an identical scanning OPD in the spatial domain, there will be numerous possible pairs of OPs introduced by multiple perturbation points along PMF. The interferograms corresponding to the same scanning OPD with distinct OP pairs will give rise to the superposition of interference intensity. Therefore, the direct analysis of PMC for the entire PMF with multiple perturbation points, such as Jones matrix \[[@B15-sensors-16-00419]\], will be rather complicated and cannot obtain the general formulas due to complex superposition phenomenon and the occurrence of HOIs.
Here, the OPT method based on the enumeration method and graphic method is presented to simplify the analysis of PMC. The steps of OPT method can be briefly described as follows: (1) Stable unit---we divide an entire PMF into stable units based on the corresponding OPD conditions and list all the OP pairs with graphic method; (2) Recursion formula---Then we obtain the recursion formula between arbitrary adjacent stable units; (3) General formulas---Finally we extend the recursion formulas to the entire PMF under test and derive the general formulas of interference intensity. With this method, the intensity and the order of interferograms could be identified for a given OPD.
### 2.2.1. Stable Unit and Recursion Formula {#sec2dot2dot1-sensors-16-00419}
We define the segment $(X_{j - p},X_{j}\rbrack$ $(p \geq 1)$ of PMF as a stable unit with the following three characteristics: (a) the pair of OPs merely occurs once coupling between the orthogonal axes of PMF at the right end (Point $X_{j}$) of segment $(X_{j - p},X_{j}\rbrack$; (b) The position of $X_{j - p}$ satisfies that if we move it right until to Point $X_{j}$, the OPD of segment $(X_{j - p},X_{j}\rbrack$ is always invariable; (c) Point $X_{j - p}$ is chosen as the leftmost point that satisfies condition (b) in order to guarantee that all the stable units are linked end-to-end.
Then stable unit can be classified into two categories based on the corresponding OPD introduced by the OP pairs in the segment, for simplicity, we denote stable unit by $B_{(i,0)}$ with OPD = 0 and $B_{(i, + )}$ with OPD ≠ 0, respectively. Obviously, the OPDs of arbitrary adjacent stable units are different, so that we might set the sequence of the *i*th adjacent units to $B_{(i,0)} \cup B_{(i, + )}$. As shown in [Figure 2](#sensors-16-00419-f002){ref-type="fig"}, the only four kinds of connections of adjacent units can be diagramed by enumeration method. The output intensity of the PMF Segment $(X_{j - p},X_{j + q}\rbrack$ from fast-axis and slow-axis at Point $X_{j + q}$ are denoted by $P_{X_{j + q},F}$ and $P_{X_{j + q},S}$, respectively, which can be evaluated as: $$\left\{ \begin{array}{l}
{P_{X_{j + q},S} = P_{X_{j - p},S}(\rho_{j}\sqrt{1 - \rho_{j}^{2}})( - \rho_{j + q}\sqrt{1 - \rho_{j + q}^{2}}) + P_{X_{j - p},F}( - \rho_{j}\sqrt{1 - \rho_{j}^{2}})( - \rho_{j + q}\sqrt{1 - \rho_{j + q}^{2}})} \\
{P_{X_{j + q},F} = P_{X_{j - p},F}( - \rho_{j}\sqrt{1 - \rho_{j}^{2}})(\rho_{j + q}\sqrt{1 - \rho_{j + q}^{2}}) + P_{X_{j - p},S}(\rho_{j}\sqrt{1 - \rho_{j}^{2}})(\rho_{j + q}\sqrt{1 - \rho_{j + q}^{2}})} \\
\end{array} \right.,p,q \geq 1$$ where, $\rho_{j}$ and $\rho_{j + q}$ are the coupling coefficients of the Point $X_{j}$ and $X_{j + q}$, respectively. The sign of $\rho_{j}$ changes only for coupling from the fast to the slow axis as shown in \[[@B18-sensors-16-00419]\]. In most cases, it has the relation $\rho_{j} \ll 1$ in the detection for distributed polarization couplings along PMF \[[@B19-sensors-16-00419]\]. Here, we are reasonable to neglect the slight errors introduced by the approximation $\sqrt{1 - {\rho_{j}}^{2}} \approx 1$, which can be used to simplify the analysis. For any two adjacent units, Equation (1) can be rewritten as: $$\left\{ \begin{array}{l}
{P_{i,S} = - (P_{i - 1,S} - P_{i - 1,F})\rho_{i,j}\rho_{i,j + q}} \\
{P_{i,F} = (P_{i - 1,S} - P_{i - 1,F})\rho_{i,j}\rho_{i,j + q}} \\
\end{array} \right.,i,q \geq 1$$ where, $P_{i,S}$ and $P_{i,F}$ represent light intensities from the slow-axis and fast-axis after passing through the $i$th adjacent units, respectively, $\rho_{i,j}$ and $\rho_{i,j + q}$ are the coupling coefficients of the point at the right end of Segment $(X_{j - p},X_{j}\rbrack$ and $(X_{j},X_{j + q}\rbrack$, respectively. From Equation (2), the stable units linked end-to-end can be expressed as: $$\left\{ {\begin{array}{l}
{P_{i,S} = - (P_{\text{In,S}} - P_{\text{In,F}})2^{i - 1}{\prod\limits_{i = 1}^{\max\{ i\}}{\rho_{i,j}\rho_{i,j + q}}}} \\
{P_{i,F} = (P_{\text{In,S}} - P_{\text{In,F}})2^{i - 1}{\prod\limits_{i = 1}^{\max\{ i\}}{\rho_{i,j}\rho_{i,j + q}}}} \\
\end{array},\ i,q \geq 1} \right.$$ where, $P_{\text{In,S}}$ and $P_{\text{In,F}}$ are the initial intensities that launch into the slow-axis and fast-axis of the first stable unit along PMF under test, respectively.
### 2.2.2. Classifications and General Formulas {#sec2dot2dot2-sensors-16-00419}
In this section, we consider that the pair of OPs of the first and last segments of the PMF under test. As mentioned above, we set the sequence of adjacent units as $B_{(i,0)} \cup B_{(i, + )}$ to simplify the analysis. However, the two end segments of the entire PMF under test might not be always satisfied the sequence. The OPD of the first and last segments could also conform to the sequence of $\left. \{ B_{(\text{in}, + )} \cup (B_{(1,0)} \cup B_{(1, + )}) \cup \cdot \cdot \cdot \right\}$ and $\left\{ \cdot \cdot \cdot \cup (B_{({last}, + )} \cup B_{({last}, + )}) \cup B_{({out},0)} \right\}$, respectively, where the first segment $B_{(\text{in}, + )}$ and the last segment $B_{(\text{out},0)}$ satisfy the features of $B_{(i, + )}$ and $B_{(i,0)}$, respectively. Therefore, the scanning OPDs of the entire PMF can be categorized into four classifications based on the possible end segments conditions. As shown in [Figure 3](#sensors-16-00419-f003){ref-type="fig"}, the scanning OPDs of the entire PMF, for simplicity, are denoted by (A) {$B_{(1,0)}$, $B_{({out}, + )}$}, (B) {$B_{({in}, + )}$, $B_{({out},0)}$}, (C) {$B_{(1,0)}$, $B_{({out},0)}$} and (D) {$B_{({in}, + )}$, $B_{({out}, + )}$}, respectively.
The initial intensities ($P_{\text{In,S}}$ and $P_{\text{In,F}}$) and terminal intensities ($P_{\text{Out,S}}$ and $P_{\text{Out,F}}$) for the four conditions in [Figure 3](#sensors-16-00419-f003){ref-type="fig"} are expressed as: $$\begin{cases}
{P_{\text{In,S}} = \cos^{2}\theta_{1}\quad,\quad P_{\text{In,F}} = \sin^{2}\theta_{1},} & {\text{first\ segment} \in B_{(i,0)}} \\
{P_{\text{In,S}} = \sin\theta_{1}\cos\theta_{1}( - \rho_{in}),P_{\text{In,F}} = \sin\theta_{1}\cos\theta_{1}\rho_{in},} & {\text{first\ segment} \in B_{(i, + )}} \\
\end{cases}$$ $$\begin{cases}
{P_{\text{Out,S}} = P_{i,S} \cdot \cos^{2}\theta_{2}\quad,\quad P_{\text{Out,F}} = P_{i,F} \cdot \sin^{2}\theta_{2},} & {\text{last\ segment} \in B_{(i,0)}} \\
{P_{\text{Out,S}} = P_{i,S} \cdot \rho_{\text{out}}( - \sin\theta_{2})\cos\theta_{2},P_{\text{Out,F}} = P_{i,F} \cdot ( - \rho_{\text{out}})( - \sin\theta_{2})\cos\theta_{2},} & {\text{last\ segment} \in B_{(i, + )}} \\
\end{cases}$$ where, $\rho_{\text{in}}$ and $\rho_{\text{out}}$ are the coupling coefficients of the points before the first unit $B_{(1,0)}$ and after the last unit $B_{({last}, + )}$, respectively, $P_{\text{Out,S}}$ and $P_{\text{Out,F}}$ represent the output intensity from slow-axis and fast-axis at spliced point $X_{\text{out}}$, respectively. Because the polarizer is aligned to the slow-axis of its PM pigtail, the amplitude changing of polarized light that launched into slow-axis of PMF at point $X_{\text{in}}$ is $\cos\theta_{1}$, and that coupled into fast-axis is $\sin\theta_{1}$. It is similar at the spliced point $X_{\text{out}}$. Therefore, the final interference intensity with a given OPD based on Equations (3) and (4) can be expressed as: $$\begin{array}{l}
{\left| P \right| = \left| {P_{\text{Out,S}} + P_{\text{Out,F}}} \right|} \\
{\ = \left\{ \begin{array}{ll}
{2^{i - 1}T_{i}\rho_{\text{out}} \cdot \cos 2\theta_{1}\sin 2\theta_{2},} & {\text{OPD} \in \left\{ B_{(1,0)},B_{({out}, + )} \right\}} \\
{2^{i - 1}T_{i}\rho_{\text{in}} \cdot \sin 2\theta_{1}\cos 2\theta_{2},} & {\text{OPD} \in \left\{ B_{({in}, + )},B_{({out},0)} \right\}} \\
{2^{i - 1}T_{i} \cdot \cos 2\theta_{1}\cos 2\theta_{2},} & {\text{OPD} \in \left\{ B_{(1,0)},B_{({out},0)} \right\}} \\
{2^{i - 2}T_{i}\rho_{\text{in}}\rho_{\text{out}} \cdot \sin 2\theta_{1}\sin 2\theta_{2},} & {\text{OPD} \in \left\{ B_{({in}, + )},B_{({out}, + )} \right\}} \\
\end{array},\ {T_{i} = \begin{cases}
{\prod\limits_{i = 1}^{\max\{ i\}}{(\rho_{i,j}\rho_{i,j + q}),}} & {i \geq 1} \\
{1,} & {i = 0} \\
\end{cases}} \right.} \\
\end{array}$$ where, $i = 0$ represents there is no stable unit $B_{(i)}$. In addition, the central interferogram intensity is calculated as $\left| P_{central} \right| = \cos^{2}\theta_{1}\cos^{2}\theta_{2} + \sin^{2}\theta_{1}\sin^{2}\theta_{2}$.
In reality, it might occur negative stable unit denoted by $B_{(i, - )}$ while there exist a positive term $B_{(i, + )}$. Here, the connection conditions of adjacent units can be classified to (a) $B_{(i, + )} \cup B_{(i,0)} \cup B_{(i, - )}$ and (b) $B_{(i,0)} \cup B_{(i, + )} \cup B_{(i, - )}$. Similar to the above analysis, we generalize the results as follows. In case of (a), the interference intensities with given OPD situations are unchanged. In case of (b), the interference intensities only should be multiplied by $\rho_{i}^{2}$ instead of the corresponding $\rho_{i}$, which is produced at the corresponding kink point between $B_{(i, + )}$ and $B_{(i, - )}$, and the other terms are remained the same.
Some summaries can be acquired by the above analysis, if we define the interference-order as $N = N_{1} + N_{2} + \cdot \cdot \cdot + N_{i}$ that can be found in the coupling coefficients term $\rho_{1}^{N_{1}} \cdot \rho_{2}^{N_{2}} \cdot \cdot \cdot \rho_{i}^{N_{i}}(N_{i} = 0,1,2)$ in Equation (5). Because there are obviously even-number times couplings in arbitrary adjacent two units, the interference-order *N* of the four conditions in [Figure 3](#sensors-16-00419-f003){ref-type="fig"} can be summarized as $N \in$ odd-order when $\text{OPD} \in$ case (A) or (B), and $N \in$ even-order when $\text{OPD} \in$ case (C) or (D). Note that the intensities of every interferogram are related to the inject angle $\theta_{1}$ at polarizer 1 and the output angle $\theta_{2}$ at polarizer 2 in Equation (5). Especially, 45° and 0° for $\theta_{1}$ or $\theta_{2}$ would introduce interesting results. The intensity of odd-order interferences have the maximum and even-order interferences are reduced to zero when $\theta_{1}$ − $\theta_{2}$ are 0°--45°, or 45°--0°, respectively. However, the variation trend of intensities are the exactly opposite results when $\theta_{1}$ − $\theta_{2}$ are 0°--0°, or 45°--45°, respectively. Therefore, we can identify HOI by altering the spliced angle between polarizers' pigtails and the PMF.
3. Experimental Results {#sec3-sensors-16-00419}
=======================
3.1. Theoretical Estimation {#sec3dot1-sensors-16-00419}
---------------------------
A PMF (segment $X_{I}X_{O}$) including two perturbation points $X_{A}$ and $X_{B}$, for simplicity, is demonstrated experimentally to prove the model of HOI introduced by PMC. The OPD denoted by $S_{MN}$ (MN = IA, AB, and BO) between two adjacent points M and N (segment MN) can be calculated as: $$S_{MN} = \Delta n \cdot l_{MN}$$ where, $l_{MN}$ represents the length of corresponding PMF section ($l_{\text{IA}}$ = 2.16 m, $l_{\text{AB}}$ = 5.22 m, and $l_{\text{BO}}$ = 16.56 m), and the birefringence $\Delta n$ of this PMF is about 5.6 × 10^−4^.
Then, the $S_{IO}$, refers to the OPD of the whole PMF, can be expressed as $S_{IO} = {\sum{\alpha \cdot S_{MN}}}\ (\alpha = 0, \pm 1)$. All the different kinds of $S_{IO}$ are enumerated readily utilizing emulation tool. It seems obvious that there will be $(\beta^{3} - 1)/2$ kinds of $S_{IO}$ when we only consider the positive values of $S_{IO}$, where $\beta$ is the number of segment MN along PMF. Finally, we choose the corresponding formula (see Equation (5)) based on the different OPD of the first and last segments to acquire the interference intensity. Besides, for a given $S_{IO}$, the interference-order *N* will be determined by the unique formula. In the case of two perturbation points, there will be 13 possible interferograms with different scanning OPD (the positions and interferogram coefficients are listed in [Table 1](#sensors-16-00419-t001){ref-type="table"}).
3.2. Identification of HOI and Results {#sec3dot2-sensors-16-00419}
--------------------------------------
It has been recognized that we could set the angle of input-output angles of polarizers of WLI to 45°--0° or 0°--45° for testing the PMF sensors or IFOG coils. In these cases, even-order interferences are suppressed and only the 1st-order and 3rd-order interferences are exposed. The envelopes of interferograms *versus* scanning OPD with the angle combination 45°--0° are plot in [Figure 4](#sensors-16-00419-f004){ref-type="fig"}. Three interferograms could be forecast as expressed in Equation (5). The 1st-order interference denoted by interferograms A and D correspond to points $X_{A}$ and $X_{B}$, respectively, and interferogram B represents the 3rd-order interference whose intensity is $\rho_{A}^{2}\rho_{B}$.
However, there are numerous extra interferograms without explicit meanings, which are marked by the red boxes. We only need to determine whether the interferogram intensities could be amplified by altering the spliced angle between the pigtails of polarizers and PMF according to Equation (5). Subsequently, the spliced angle combinations are set to 0°--45°, 0°--0°, and 45°--45°, respectively, and attention is paid to the intensity variation at the corresponding positions of interferograms *C*, *E*, *F*, *H*, *I*, *J*, *K*, and *M* in [Figure 4](#sensors-16-00419-f004){ref-type="fig"}. As shown in [Figure 5](#sensors-16-00419-f005){ref-type="fig"}a,c, the intensities of interferograms *C*, *E*, *H*, *J*, and *K*, which represent the 2nd-order interference, and that of interferogram *M*, which represents the 0th-order interference, are enhanced to their maximum. [Figure 5](#sensors-16-00419-f005){ref-type="fig"}b shows that the intensity of interferogram *F*, which represents the 3rd-order interference, and that of interferogram *I*, which represents the 1st-order interference, are increased to the maximum. In consequence, all the interferograms marked by red boxes in [Figure 4](#sensors-16-00419-f004){ref-type="fig"} can be identified through just four times of measurement at different angles between the pigtails of polarizers and PMF under test. We can extract realistic signals (1st-order interference) and eliminate spurious signals from the results to evaluate the polarization characteristics of PMF.
It shall be noticed that there are several unexpected interferograms around interferograms, such as interferograms *C* or *M*. The reason lies in that the short PM pigtails of the two added polarizers are not taken into consideration for the proposed model. In these experiments, the lengths of PMF polarizers' pigtails are 0.30 m and 0.25 m, respectively. The pigtails could be considered and analyzed as another two segments of PMF, which will lead to the side interferograms around the characteristic interferograms.
4. Discussions {#sec4-sensors-16-00419}
==============
The positions and intensities of the total interferograms shown in [Figure 4](#sensors-16-00419-f004){ref-type="fig"} and [Figure 5](#sensors-16-00419-f005){ref-type="fig"} are listed in [Table 1](#sensors-16-00419-t001){ref-type="table"}. The 1st-order coupling corresponding to Points $X_{A}$ and $X_{B}$ are 14.9 dB and 15.0 dB, respectively, which are measured by a PER meter (ERM-102, General Photonics, Chino, CA, USA). The 2nd to 4th-order interferences could be calculated based on the 1st-order coupling. The errors are less than 2 dB, which might be caused by the small misalignments (\<1°) induced by fiber fusion splicer and PMF dispersion.
In order to further verify the HOI variation trend obtained by OPT method, the intensity variations of some interferograms including interference signals ranging from 0th-order to 3rd-order are measured. With the input polarized angle $\theta_{1}$ fixed at 0° and 45°, the output angle $\theta_{2}$ are change by 7.5° step by step, respectively. It can be seen from [Figure 6](#sensors-16-00419-f006){ref-type="fig"}a, when $\theta_{1}$ is set at 0°, interferogram *C* decreases to −70 dB with $\theta_{2}$ = 45°, and interferograms *I* and *F* reduce to less than --50 dB and −90 dB with $\theta_{2}$ = 0°, respectively. Similarly in [Figure 6](#sensors-16-00419-f006){ref-type="fig"}b, when $\theta_{1}$ is set at 45°, interferogram A of the 1st-order interference and interferogram B of the 3rd-order interference are suppressed with $\theta_{2}$ = 45°, and interferogram *H* of the 2nd-order interference and interferogram *M* of the 0th-order interference are suppressed with $\theta_{2}$ = 0°. Because of the spliced angle errors and the manufacture errors of added polarizers, the HOI and the 1st-order interference interferograms cannot be eliminated completely at the maximum slope angles as shown in [Figure 6](#sensors-16-00419-f006){ref-type="fig"}. Therefore, these HOIs can be identified far away from the maximum slope angle combinations.
The proposed method is helpful to realize the complicated OPs behaviors transmitted along PMF with perturbation points. Based on the results and discussions, the realistic coupling introduced by the splice points of PMF could be identified readily from the interference signals. As shown in [Figure 4](#sensors-16-00419-f004){ref-type="fig"}, interferograms A and D correspond to the real perturbation points X~A~ and X~B~, respectively. Meanwhile, interferograms I and L also represent the points X~B~ and X~A~, respectively, due the opposite spliced angle combinations (see [Figure 5](#sensors-16-00419-f005){ref-type="fig"}b). Besides, we can choose the angle combination conditions to control the occurrence of HOIs to acquire appropriate presentation. For the devices based single-variety such as the IFOG coil, we could directly set the angle of input-output polarizers of PMC measurement system to 0°--45° or 45°--0°. In this case, 2nd-order interference is suppressed and 1st-order interference is shown out to evaluate devices performance. For the high-precision integrated devices such as the IFOG system which contain the connection or splice points between different components, we could adopt the angle combinations of input-output polarizers of 0°--0° or 45°--45° to suppress the weaker 1st-order interference.
5. Conclusions {#sec5-sensors-16-00419}
==============
The HOI introduced by the PMCs of multiple perturbation points in PMFs is analyzed in detail. An OPT method---Based on the enumeration method and graphic method---Is presented for simplifying the analysis of polarization light transmission along PMF with multiple perturbation points. The positions and intensities of HOI interferograms can be calculated by the derived general HOI formulas utilizing OPT method. It is demonstrated that the odd-order or even-order HOIs will be suppressed or amplified depending on the angle between the added pigtails of polarizers and the PMF under test. Furthermore, the method is verified by a case of two coupling points along a PMF by a WLI system. As a result, all the characteristic interferograms including HOIs can be distinguished through just four measurements. The identification method is useful to evaluate the polarization performance of PMF, suppress the system noise of WLI and improve its sensitivity.
This work was funded by the National Natural Science Foundation of China (No. 61227013, 61422505, 61307104 and 61405044), the Program for New Century Excellent Talents in University (No. NCET-12-0623), the National Key Scientific Instrument and Equipment Development Project (No. 2013YQ040815), and the Specialized Research Fund for the Doctoral Program of Higher Education (No. 20122304110022).
Chuang Li and Jun Yang conceived and designed the experiments; Chuang Li performed the experiments; Chuang Li and Jianzhong Zhang analyzed the data; Libo Yuan and Jun Yang contributed reagents/materials; Yonggui Yuan, Bing Wu and Feng Peng contributed analysis tools; Chuang Li wrote the paper; Jianzhong Zhang and Zhangjun Yu revised and improved the paper.
The authors declare no conflicts of interest.
{#sensors-16-00419-f001}
{#sensors-16-00419-f002}
{#sensors-16-00419-f003}
{#sensors-16-00419-f004}
{ref-type="fig"}, which can be used to identify the HOI.](sensors-16-00419-g005){#sensors-16-00419-f005}
{#sensors-16-00419-f006}
sensors-16-00419-t001_Table 1
######
Interferogram measurement results.
Interferogram Position Meaning Interferogram Meaning Position (mm) Normalized Intensity/Error (dB) Order *N*
--------------- --------------------------------------------- --------------------------------------- --------------- --------------------------------- ----------- ----------- ----------- -----
0 1 0 0 0 0 0
**M** $\left| {S_{IA} + S_{AB} + S_{BO}} \right|$ $\frac{1}{2}$ 13.33 \<−70 \<−50 \<−50 −7.8/1.8 0th
**A** $\left| S_{IA} \right|$ $\rho_{A}$ 1.22 \<−70 \<−70 −15.6/0.7 \<−50 1st
**D** $\left| {S_{IA} + S_{AB}} \right|$ $\rho_{B}$ 4.09 \<−70 \<−70 −15.7/0.7 \<−50
**I** $\left| S_{BO} \right|$ $\rho_{B}$ 9.27 \<−50 −16.5/1.5 \<−70 \<−50
**L** $\left| {S_{AB} + S_{BO}} \right|$ $\rho_{A}$ 12.12 \<−50 −16.3/1.4 \<−70 \<−50
**C** $\left| S_{AB} \right|$ $\rho_{A}\rho_{B}$ 2.87 −29.7/0.2 \<−60 \<−60 \<−70 2nd
**E** $\left| {S_{IA} + S_{AB} - S_{BO}} \right|$ $\frac{1}{2}\rho_{B}^{2}$ 5.21 \<−70 \<−70 \<−70 −36.7/0.7
**H** $\left| {S_{IA} - S_{BO}} \right|$ $\rho_{A}\rho_{B}$ 8.08 \<−70 \<−70 \<−70 −30.6/0.7
**J** $\left| {S_{IA} + S_{BO}} \right|$ $\rho_{A}\rho_{B}$ 10.47 \<−70 \<−70 \<−70 −31.0/1.1
**K** $\left| {S_{IA} - S_{AB} - S_{BO}} \right|$ $\frac{1}{2}\rho_{A}^{2}$ 10.95 \<−70 \<−70 \<−70 −36.9/1.1
**B** $\left| {S_{IA} - S_{AB}} \right|$ $\rho_{A}^{2}\rho_{B}$ 1.70 \<−70 \<−70 −45.1/0.3 \<−70 3rd
**F** $\left| {S_{AB} - S_{BO}} \right|$ $\rho_{A}\rho_{B}^{2}$ 6.38 \<−70 −45.7/0.8 \<−60 \<−70
**G** $\left| {S_{IA} - S_{AB} + S_{BO}} \right|$ $\frac{1}{2}\rho_{A}^{2}\rho_{B}^{2}$ 7.60 \<−70 \<−70 \<−70 −67.7/1.9 4th
| {
"pile_set_name": "PubMed Central"
} |
All relevant data are within the manuscript and its Supporting Information files.
Introduction {#sec001}
============
Gene expression is often regulated by transcription factors (TFs). TFs bind to specific DNA-binding sites and modulate the expression of genes. Therefore, to understand transcriptional regulations, given its complexity, it is extremely important to make accurate inferences about transcription factor binding sites (TFBSs). High-throughput ChIP-seq, which is widely used to study TF--DNA interactions, provides the sequences of binding regions \[[@pone.0220207.ref001],[@pone.0220207.ref002]\]. TFBSs can be determined as the most over-represented motif in a given set of DNA sequences.
Bipartite motifs are defined as extensions of one-block TFBSs, that is, two conserved motifs separated by variable gaps. Several different types of bipartite motifs have been proposed in both prokaryotes and eukaryotes \[[@pone.0220207.ref003],[@pone.0220207.ref004]\]. Shultzaberger et al. (2001) proposed the bipartite model of ribosome binding sites, in which they are composed of a Shine--Dalgarno sequence and an initiation region in *Escherichia coli* \[[@pone.0220207.ref003]\]. In *Bacillus subtilis*, the principal sigma factor in vegetative growth, SigA, binds to the bipartite motif separated by variable gaps, TGACA\<spacer\>TATAAT \[[@pone.0220207.ref005]--[@pone.0220207.ref007]\]. Baichoo and Helmann (2002) determined the bipartite motif, TGATAAT\<spacer\>ATTATCA, of the ferric uptake repressor Fur \[[@pone.0220207.ref008],[@pone.0220207.ref009]\]. It has been reported that the global regulator AbrB can recognize bipartite motifs \[[@pone.0220207.ref010]--[@pone.0220207.ref012]\]. As in the case of eukaryotes, the existence of bipartite motifs of yeast TFs, such as ABF1 and GAL4, has been confirmed \[[@pone.0220207.ref013],[@pone.0220207.ref014]\]. It has been reported that around 30% of the promoter sequences contain bipartite motifs with constant gaps in humans \[[@pone.0220207.ref015]\]. The level of conservation of the motif M4 (ACTAYRNNNCCCR) was reported to be much higher than those for most known motifs. Similarly, the TFs CAR and RXR bind to bipartite motifs in humans \[[@pone.0220207.ref004]\]. Thus, it is conceivable that TFs work in a cooperative manner and recognize bipartite motifs to regulate gene expression \[[@pone.0220207.ref016],[@pone.0220207.ref017]\]. Several tools such as BioProspector \[[@pone.0220207.ref018]\], BiPad \[[@pone.0220207.ref019],[@pone.0220207.ref020]\], and AMD \[[@pone.0220207.ref021]\] are available for the *ab initio* prediction of bipartite motifs for a set of DNA sequences, while many tools have been developed for the prediction of one-block TFBSs, such as Consensus \[[@pone.0220207.ref022]\], Gibbs Sampler \[[@pone.0220207.ref023]\], and MEME \[[@pone.0220207.ref024]\]. BioProspector based on Gibbs sampling \[[@pone.0220207.ref018]\] and BiPad based on the entropy minimization method \[[@pone.0220207.ref019],[@pone.0220207.ref020]\] enable the identification of bipartite motifs with variable gaps. AMD identifies bipartite motifs with constant gaps by comparing the target sequences with the background sequences regardless of whether the motifs are long or short, gapped or contiguous \[[@pone.0220207.ref021]\].
Position weight matrices (PWMs) are commonly used to find and represent TFBSs \[[@pone.0220207.ref025]\]. They are based on the assumption that each nucleotide independently participates in the TF--DNA interaction. However, it has long been known that interactions between neighboring DNA bases affect TF--DNA interactions. For example, a single amino acid interacts with multiple bases simultaneously \[[@pone.0220207.ref026]\]. Zhao et al. (2012) clearly showed the existence of dinucleotide dependency in TFs \[[@pone.0220207.ref027],[@pone.0220207.ref028]\]. Indeed, PWMs perform well in modeling TFBS properties, but are inadequate for considering position interdependencies. There are interdependencies between neighboring positions of the binding sites of CRP and LexA in *E*. *coli* \[[@pone.0220207.ref029]\]. It has been reported that the method based on dinucleotide weight matrix (DWM) outperformed that based on PWM for yeast datasets \[[@pone.0220207.ref030]\]. In fact, Weirauch et al. (2013) observed an improvement of performance of motif detection upon incorporating dinucleotide interactions \[[@pone.0220207.ref028]\]. Although BioProspector and BiPad predict bipartite motifs, they are based on the assumption of independencies among bases, namely, PWM.
Here, we present a novel bipartite motif detection tool, DIpartite (bi**partite** motif detection tool based on **di**nucleotide weight matrix). DIpartite predicts the bipartite motif by considering interdependencies of neighboring positions, namely, DWM. We compared DIpartite with other available alternatives by using test datasets from prokaryote and eukaryote, namely, of CRP in *E*. *coli*, sigma factors in *B*. *subtilis*, and promoter motifs in humans.
Materials and methods {#sec002}
=====================
A novel method for predicting bipartite motifs by incorporating base-pair dependencies {#sec003}
--------------------------------------------------------------------------------------
DIpartite identifies the bipartite motifs with variable gaps based on PWM or DWM from the input sequences ([S1 Fig](#pone.0220207.s001){ref-type="supplementary-material"}). Since it is reported that the bipartite motif represents well by Shannon's entropy \[[@pone.0220207.ref003],[@pone.0220207.ref019],[@pone.0220207.ref020]\], we set the objective function to minimize the entropy. Similar to BiPad \[[@pone.0220207.ref019],[@pone.0220207.ref020]\], the algorithm of DIpartite is based on Gibbs sampling and the minimization of information content (IC) by a greedy algorithm. DIpartite adopts the Gibbs sampling strategy which initializes the motif positions for all input sequences at random, and iteratively improves the entropy of PWM or DWM by updating the motif position.
Objective function {#sec004}
------------------
Input data have *N* sequences for prediction of the bipartite motifs separated by gaps. Similar to BiPad \[[@pone.0220207.ref019],[@pone.0220207.ref020]\], the bipartite motifs are expressed as *l*~*L*~\<*d*\>*l*~*R*~, where *l*~*L*~ and *l*~*R*~ are the widths of left and right motifs, respectively, and *d* is gap length. We set the objective function to minimize Shannon's entropy for PWM or DWM of the concatenated motif of the left and right motifs, in [Eq 1](#pone.0220207.e001){ref-type="disp-formula"}: $${\hat{M}}_{LR} = {argmin}_{M_{LR}}\left( {IC}_{M_{LR}} \right)$$ where *M*~*LR*~ is the concatenated motif, and ${IC}_{M_{LR}}$ is the entropy for the motif *M*~*LR*~. Here, ${IC}_{M_{LR}}$ is given by: $${IC}_{M_{LR}} = {\sum_{i}^{j}{\sum_{x \in X}\left. - p_{i}(x) \times \log\left\{ \frac{p_{i}(x)}{b(x)} \right. \right\}}},i = \left\{ \begin{matrix}
{1,PWM} \\
{2,DWM} \\
\end{matrix}, \right.X = \left\{ \begin{matrix}
{\left\{ {A,C,G,T} \right\},\ PWM} \\
{\left\{ {{AA},{AC},\cdots,{TT}} \right\},\ DWM} \\
\end{matrix} \right.$$ where *p*~*i*~(*x*) and *b*(*x*) are the composition of *x* in the motif sites and the background sites (not motif sites), respectively. *x* is one of the mononucleotides or dinucleotides for PWM or DWM, respectively. *j* is the sum of the lengths of the left and right motifs. *p*~*i*~(*x*) and *b*(*x*) are given by: $$p_{i}\left( x \right) = \frac{f_{i}\left( x \right) + {\beta/k}}{N + \beta},k = \left\{ \begin{array}{l}
{\mspace{9mu}{4,\ PWM}} \\
{16,\ DWM} \\
\end{array} \right.$$ $$b\left( x \right) = \frac{g\left( x \right) + {\beta/k}}{n + \beta}$$ where *N* is the total number of input sequences. *f*~*i*~(*x*) is the frequency of *x* at the position *i*, that is, the mononucleotide at position *i* for PWM, or the dinucleotide at position *i* −1 and *i* for DWM. *k* is the number of the patterns, that is, *k* = 4 for PWM or *k* = 16 for DWM. *n* is the total number of the mononucleotides for PWM or dinucleotides for DWM that are not located at the motif sites. *β* is the total pseudo-count. *g*(*x*) is the frequency of *x* in the background sites. We set *β* = 1.
Overview of the algorithm {#sec005}
-------------------------
The algorithm of DIpartite works through an iterative process of calculating entropy. DIpartite is implemented in C++ and available under the CNU v3 license. Fasta and text formats are allowed as input files. Users can specify the lengths of the left and right motifs, the gap length, and PWM for the mononucleotide or DWM for the dinucleotide. The software works for OOPS (one occurrence per sequence), ZOOPS (Zero or one bipartite occurrence per sequence), or ANR (any number of repetitions).
Performance evaluation {#sec006}
----------------------
The nucleotide-level correlation coefficient (*nCC*) was used to evaluate the performance of each tools for the same input data \[[@pone.0220207.ref031]\]. *nCC* is given by: $$nCC = \frac{nTP \times nTN - nFN \times nFP}{\sqrt{\left( nTP + nFN \right)\left( nTN + nFP \right)\left( nTP + nFP \right)\left( nTN + nFN \right)}}$$ where *nTP* is the number of nucleotide positions in both known sites and predicted sites, *nFN* is the number of nucleotide positions in known sites but not in predicted sites, *nFP* is the number of nucleotide positions not in known sites but in predicted sites, and *nTN* is the number of nucleotide positions in neither known sites nor predicted sites. We adopted the combined *nCC* by adding *nTP*, *nFN*, *nFP*, and *nTN* over the data sets.
CRP {#sec007}
---
CRP binding sites in *E*. *coli* were retrieved from Regulon DB as "TF binding sites" (Release: 9.4 Date: 05-08-2017) \[[@pone.0220207.ref032]\]. For example, the motif sequences of two ECK125158203 entries were identical although the transcription unit was different, i.e., fumA and fumAC. Out of 374 sequences of CRP binding sites, 323 unique sequences ranging from 36 bp to 42bp were filtered and used for the performance comparison. The binding site lengths consisted of 16 bp (11 binding sites), 17 bp (one binding site), 20 bp (one binding site), 22 bp (308 binding sites), and 23 bp (two binding sites).
Promoter motifs in human {#sec008}
------------------------
Xie et al. \[[@pone.0220207.ref015]\] proposed the 1,460 motifs in human. We sought the motifs with the gap lengths greater than or equal to the lengths of left and right motifs. Among of them, we selected 46 motifs with more than 4-nt gaps as the test datasets of two-block motifs. The promoter sequences around the positions of each motifs (500 bp upstream to 500 bp downstream) were retrieved as the target sets.
Sigma factor {#sec009}
------------
As the dataset of bipartite motifs with variable gap lengths, the sigma factor dataset in *B*. *subtilis* from DBTBS \[[@pone.0220207.ref007]\] was used. The nine of the bipartite sigma transcription factors in *B*. *subtilis* were used. The minimum and maximum gap lengths of sigma factors were determined based on all identified binding sites: σ^A^ (344 sequences ranging from 38 bp to 93 bp, 6\<\[11,23\]\>6), σ^B^ (64 sequences ranging from 39 bp to 64 bp, 6\<\[12,18\]\>6), σ^D^ (30 sequences ranging from 44 bp to 57 bp, 4\<\[12,18\]\>8), σ^E^ (70 sequences ranging from 41 bp to 58 bp, 7\<\[12,18\]\>8), σ^F^ (25 sequences ranging from 41 bp to 71 bp, 5\<\[13,19\]\>10), σ^G^ (55 sequences ranging from 40 bp to 76 bp, 5\<\[15,20\]\>7), σ^H^ (25 sequences ranging from 41 bp to 60 bp, 7\<\[9,18\]\>5), σ^K^ (53 sequences ranging from 38 bp to 85 bp, 4\<\[9,17\]\>9), and σ^W^ (34 sequences ranging from 38 bp to 53 bp, 10\<\[13,17\]\>6).
Other programs used for comparison {#sec010}
----------------------------------
Four popular tools, namely MEME (ver. 5.0.3), BioProspector (release 2), AMD, and BiPad (ver. 2), were compared with DIpartite.
For the CRP dataset, MEME was executed with the options "-mod oops", "-dna", "-w 22", "-minw 22", and "-maxw 22". BioProspector was executed with the options "-n 50", and "-n 3". AMD was executed with the options "-MI" and "-T 1". BiPad was executed with the options "-l 22", "-r 0", "-a 0", "-b 0", "-i", and "-y 1000". AMD was executed with the option "-T 2" for two sigma datasets, i.e., σ^E^ and σ^F^. We used the background sequences for AMD: the 200 bp upstream regions of 4,314 genes in *E*. *coli* K-12 (NC_000913.3), the promoter sequences of all human genes (hg17: upstream1000.fa.gz), and the 200 bp upstream regions of 4,448 genes in *B*. *subtilis* 168 (NC_000964.3).
Results {#sec011}
=======
Interdependencies of neighboring DNA bases in CRP {#sec012}
-------------------------------------------------
CRP is one of the seven main transcription factors that influences transcriptional networks in *E*. *coli* \[[@pone.0220207.ref033]\]. It has been shown that there are interdependencies among neighboring DNA bases in CRP binding sites \[[@pone.0220207.ref029]\]. More than 300 binding sites for CRP have been registered in Regulon DB as "TF binding sites" (Release: 9.4) \[[@pone.0220207.ref032]\]. The CRP binding sites are separated by a 6-nt gap ([Fig 1A](#pone.0220207.g001){ref-type="fig"}). We measured the interdependency of CRP using the mutual information proposed by Salama and Stekel \[[@pone.0220207.ref029]\]. Strong correlations between neighboring bases were observed, for example, among positions 1, 2, and 6--8, and among positions 16--19 ([Fig 1B](#pone.0220207.g001){ref-type="fig"}). In addition, we observed the higher mutual information between the distant positions in 7, 16 and 8, 17 among the palindromic positions, followed by the position in 6 and 19. This suggests that the palindromic features of CRP binding sites would be incomplete.
![Sequence logo and heat map of CRP.\
Out of 374 CRP motifs, 308 sequences with the 22-bp motif were used. (A) Sequence logo for CRP using 308 sequences \[[@pone.0220207.ref034]\]. (B) Heat map of CRP.](pone.0220207.g001){#pone.0220207.g001}
Performance for CRP dataset {#sec013}
---------------------------
We evaluated the performance of DIpartite by using the TF binding sites of CRP. Out of 374 sequences of CRP binding sites, 323 unique sequences were used as the test dataset. Jensen and Liu (2004) analyzed the CRP binding sites as a bipartite motif and proposed the consensus sequence, tGTcA\<6,8\>CAcattt \[[@pone.0220207.ref019],[@pone.0220207.ref035]\]. We conducted motif prediction by using MEME (ver. 5.0.2), BioProspector (release 2), AMD, BiPad (ver. 2), and DIpartite for these 323 sequences of CRP binding sites ([Fig 2A](#pone.0220207.g002){ref-type="fig"}). DIpartite with the "PWM" or "DWM" options is referred to as DIpartite PWM or DIpartite DWM, respectively. Although DIpartite PWM performed best among the tested software for the one-block model, namely, the 22-bp motif, the performance was comparable among MEME, BioProspector, BiPad, and DIpartite. AMD exhibited a combined *nCC* value of less than 0.9. We assessed the performance of DIpartite by randomly sampling 100 datasets with 100 sequences from the CRP binding sites. DIpartite DWM slightly outperformed other tested tools for 100 datasets ([S2A Fig](#pone.0220207.s002){ref-type="supplementary-material"}). In addition, we tested the running time by using the CRP dataset. Although BioProspector was the fastest software among tested software, DIpartite was comparable with BiPad ([S3 Fig](#pone.0220207.s003){ref-type="supplementary-material"}).
![The performance comparison for 323 CRP sequences.\
The combined *nCC* values were plotted. (A) Summary of the results for searching the one-block motif, i.e., the 22 bp motif, by MEME, BioProspector, AMD, BiPad, DIpartite PWM and DIpartite DWM. (B) Summary of the results for searching the bipartite motifs, i.e., 6\<\[10\]\>6, 6\<\[8\]\>8, and 8\<\[6\]\>8, by BioProspector, BiPad, DIpartite PWM and DIpartite DWM.](pone.0220207.g002){#pone.0220207.g002}
For the bipartite motif, we compared BioProspector, BiPad, DIpartite PWM, and DIpartite DWM ([Fig 2B](#pone.0220207.g002){ref-type="fig"}). The performance of searching the bipartite motifs was lower than that of searching the one-block model, i.e., 0.936 by DIpartite PWM. For all three types of the bipartite motifs, DIpartite PWM and DIpartite DWM were superior to BioProspector and BiPad. DIpartite DWM was superior to DIpartite PWM in the case of 6\<\[10\]\>6. We conducted the performance comparison by using 100 datasets with 100 sequences ([S2B Fig](#pone.0220207.s002){ref-type="supplementary-material"}). DIpartite PWM outperformed other tested tools. Although the implementation of DIpartite PWM is similar to that of BiPad, DIpartite PWM slightly outperformed BiPad. This might be because DIpartite takes into consideration the background sites (not motif sites) unlike BiPad, that is, *b*(*x*) in Eq ([2](#pone.0220207.e004){ref-type="disp-formula"}). Taking the findings together, DIpartite successfully detected the binding sites of the one-block or bipartite motifs.
Performance for human dataset {#sec014}
-----------------------------
We selected the human promoter sequences as bipartite motifs with constant gaps in eukaryotes \[[@pone.0220207.ref015]\]. Of 1,460 motifs, 46 motifs with gaps larger than 4 nt were filtered. The promoter sequences around the positions of each motif (500 bp upstream to 500 bp downstream) were retrieved as the target sets. Since AMD did not detect any motifs for six motifs, namely, RGGANNNNNAKTCC (54 sequences), RKCTGNNNNNRMTTA (21 sequences), TTGRNNNNNNTCCAR (21 sequences), YMATCNNNNNGCGM (50 sequences), YTGGANNNNNNYCAA (26 sequences), and YTTGRNNNNNNGCCNR (50 sequences), these were excluded, and 40 datasets were evaluated for the performance of DIpartite. We assessed the performance for 40 motif datasets ([Fig 3A](#pone.0220207.g003){ref-type="fig"}). DIpartite DWM exhibited the highest performance (50%), followed by DIpartite PWM (48%), BioProspector (38%), MEME (20%), BiPad (8%), and AMD (3%) ([S1 Table](#pone.0220207.s005){ref-type="supplementary-material"}), indicating that DIpartite performs equivalently to or better than the other tools for detecting dipartite motifs. In addition to the result of CRP 6\<\[10\]\>6, DIpartite DWM outperformed other tested tools, suggesting that DWM might improve the bipartite motif detection. Apparently, MEME and BiPad exhibited larger interquartile ranges ([Fig 3B](#pone.0220207.g003){ref-type="fig"}), indicating that these tools outperformed DIpartite for particular motifs, but were outperformed by it for the other motifs.
{ref-type="supplementary-material"}.](pone.0220207.g003){#pone.0220207.g003}
Performance for sigma factor dataset {#sec015}
------------------------------------
We compared the performance of DIpartite with those of BioProspector, AMD, and BiPad for bipartite motifs with variable gaps. We adopted the nine bipartite sigma transcription factors in *B*. *subtilis*, namely, σ^A^ (344 sequences), σ^B^ (64 sequences), σ^D^ (30 sequences), σ^E^ (70 sequences), σ^F^ (25 sequences), σ^G^ (55 sequences), σ^H^ (25 sequences), σ^K^ (53 sequences), and σ^W^ (34 sequences) from DBTBS \[[@pone.0220207.ref007]\] as the test datasets ([Fig 4A](#pone.0220207.g004){ref-type="fig"}). DIpartite PWM performed better than BioProspector, BiPad, AMD, and DIpartite DWM for six sigma factors, with the exceptions being σ^D^, σ^E^ and σ^H^ ([Fig 4B](#pone.0220207.g004){ref-type="fig"}). While the performance of DIpartite PWM was excellent for two sigma factors (σ^A^ and σ^F^), that of DIpartite DWM was remarkable for four sigma factors (σ^B^, σ^G^, σ^K^, and σ^W^). AMD exhibited relatively low *nCC* values for all nine datasets ([Fig 4A](#pone.0220207.g004){ref-type="fig"}), unlike the results for human promoter sequences, suggesting that the variable gap lengths could affect its performance. This is reasonable because AMD was developed for detecting bipartite motifs with constant gaps. AMD with the option "-T 1" did not detect any motifs for two sigma datasets, i.e., σ^E^ and σ^F^.
{#pone.0220207.g004}
Among four sigma factors with the highest performance coefficients for DIpartite DWM, the *nCC* value for σ^K^ was greatly improved by DIpartite DWM, namely, to 0.757, indicating the presence of base interdependencies in the motif of σ^K^. We observed that the left motif of DIpartite DWM was shifted and "AC" was more over-represented, indicating that the left motif of σ^K^ might be improved. Position 7 was "T" in all 53 sequences ([Fig 5A](#pone.0220207.g005){ref-type="fig"}), consistent with the known motif in DBTBS. Similarly, the highest frequencies of the dinucleotides "AT" and "TA" were observed at positions 6 and 7, and 7 and 8, respectively ([Fig 5B](#pone.0220207.g005){ref-type="fig"}).
{#pone.0220207.g005}
The *nCC* value of σ^A^ was greatly improved by DIpartite PWM, namely, to 0.697. While the sequence logo generated from the result of BioProspector was similar to that generated from the result of DIpartite DWM, those of BiPad and DIpartite PWM was different from them ([S4 Fig](#pone.0220207.s004){ref-type="supplementary-material"}). In particular, DIpartite PWM exhibited the conserved base "T", at position 1. This result is consistent with the motif TTGACA\<\>tgnTATAAT proposed by DBTBS \[[@pone.0220207.ref007]\]. DIpartite PWM showed the sequences with minimum entropy.
We assessed the performance of DIpartite DWM in terms of the sizes of the input datasets. By randomly sampling the sequences of σ^A^ in *B*. *subtilis*, we generated 100 datasets for each including 10, 20, 50, 100, 150, 200, and 300 sequences ([Fig 6](#pone.0220207.g006){ref-type="fig"}). Upon increasing the size of the datasets, DIpartite PWM and DWM exhibited better performance. Notably, DIpartite underperformed for the datasets with 10 and 20 sequences, suggesting that DIpartite could perform well for data including more than 50 sequences. The variances of DIpartite PWM for the datasets with 200 and 300 sequences were relatively smaller than those of DIpartite DWM. One potential reason for this is that DWM consists of the frequencies of 16 dinucleotides ([Eq 3](#pone.0220207.e005){ref-type="disp-formula"}).
{#pone.0220207.g006}
Performance for the dataset with noise sequences {#sec016}
------------------------------------------------
We assessed the performance for the datasets with noise sequences. DIpartite allows the users to search the motifs for the datasets with noise sequences, known as ZOOPS. We evaluated the accuracy of detecting noise sequences by using the datasets with noise sequences. We chose the CRP datasets and human dataset as the test datasets of the one- and two-block motifs. We compared the performance of noise detection by DIpartite with that by MEME for the CRP datasets ([Table 1](#pone.0220207.t001){ref-type="table"}). DIpartite exhibited the TPRs (true positive rate), i.e., 0.835, 0.863, and 0.876 for the datasets with 25%, 50%, and 100% noise sequences, respectively. This indicates that DIpartite ZOOPS could be well tolerated with the noise sequences. Indeed, MEME exhibited the lower FPRs, but lower TPRs, suggesting that DIpartite ZOOPS would be comparable with MEME ZOOPS.
10.1371/journal.pone.0220207.t001
###### The performance of noise detection for the one-block motif.
{#pone.0220207.t001g}
MEME DIpartite
--------- ------- ------- ------- ----------- ------- -------
**FPR** 0.061 0.030 0.024 0.172 0.172 0.167
**TPR** 0.798 0.777 0.739 0.835 0.863 0.876
Noise sequences were sampled from the genome sequence of *E*. *coli*. CRP_25 consists of 323 CRP and 81 (25%) noise sequences. CRP_50 consists of 323 CRP and 162 (50%) noise sequences. CRP_100 consists of 323 CRP and 323 (100%) noise sequences.
TPR: True positive rate, FPR: False positive rate.
Finally, we compared the performance of noise detection for the two-block dataset, i.e., RYAAAKNNNNNNTTGW consisting of 44 sequences ([S1 Table](#pone.0220207.s005){ref-type="supplementary-material"}). BioProspector (*nCC* = 1) and BiPad (*nCC* = 1) outperformed DIpartite PWM (*nCC* = 0.914). Increasing the noise sequences, BioProspector and BiPad exhibited lower *nCC* values ([Table 2](#pone.0220207.t002){ref-type="table"}). DIpartite exhibited higher *nCC* values even adding the noise sequences, suggesting that DIpartite could work well for both one- and two-block motifs with noise sequences.
10.1371/journal.pone.0220207.t002
###### The performance of noise detection for the two-block motif.
{#pone.0220207.t002g}
TF_0 TF_25 TF_50 TF_100
------------------- ------- ------- ------- --------
**BioProspector** 1 0.907 −0.16 −0.16
**BiPad** 1 1 1 −0.16
**DIpartite PWM** 0.914 1 1 1
The combined *nCC* values were indicated. Noise sequences were sampled from the genome sequence of human. TF_0 consists of 44 RYAAAKNNNNNNTTGW sequences. TF_25 consists of 44 RYAAAKNNNNNNTTGW and 11 noise sequences. TF_50 consists of 44 RYAAAKNNNNNNTTGW and 22 noise sequences. TF_100 consists of 44 RYAAAKNNNNNNTTGW and 44 noise sequences.
Conclusions {#sec017}
===========
We have developed DIpartite for the detection of TFBSs, consisting of bipartite motifs. DIpartite enables *ab initio* prediction of conserved motifs based on not only PWM, but also DWM. We evaluated the performance of DIpartite compared with freely available tools, namely, MEME, BioProspector, AMD, and BiPad. Both DIpartite PWM and DWM performed equivalently to or better than these alternatives, especially in the case of the bipartite motifs with variable gaps, like for sigma factors in *B*. *subtilis*. The prediction of σ^K^ was greatly improved by taking into consideration base interdependencies. DIpartite is available for use at <https://github.com/Mohammad-Vahed/DIpartite>.
Supporting information {#sec018}
======================
###### Flowchart of DIpartite.
The input data is the sequence file including *N* sequences. DIpartite proposes bipartite motifs based on PWM or DWM. Each iteration starts from randomly generated positions. The convergence of each iteration is judged by the differences of the entropy, that is, ε. We set *ε* = 10^−8^. E~*i*~ and E~*i*−1~ correspond to the *i*th and *i*−1th entropy, i.e., ${IC}_{M_{LR}}$ ([Eq 2](#pone.0220207.e004){ref-type="disp-formula"}), respectively.
(TIFF)
######
Click here for additional data file.
###### The performance comparison for 100 CRP datasets.
100 datasets consisting of 100 sequences were generated by randomly sampling the CRP datasets. (A) Summary of the results for searching the one-block motif, i.e., the 22 bp. (B) Summary of the results for searching the bipartite motifs, i.e., 6\<\[10\]\>6, 6\<\[8\]\>8, and 8\<\[6\]\>8.
(TIFF)
######
Click here for additional data file.
###### Running times.
The datasets consisting of 20, 50, 100, 200, 500 and 1,000 sequences were generated by randomly sampling the CRP sequences. X-axis and Y-axis correspond to the number of sequences, and the running time \[s\] on a log scale. BioProspector (designated as Bio), BiPad, DIpartite PWM (designated as PWM), and DIpartite DWM (designated as DWM) were tested.
(TIFF)
######
Click here for additional data file.
######
Sequence logos for σ^A^ from the results of (A) BioProspector, (B) BiPad, (C) DIpatrite PWM, and (D) DIpartite DWM.
(TIFF)
######
Click here for additional data file.
###### The performance comparison for 40 motifs in human.
(XLSX)
######
Click here for additional data file.
We would like to thank Rafik Salam (University of Oxford, UK) and Dov Stekel (University of Nottingham, UK) for their fruitful discussions. MV would like to thank Fujii Medical International Exchanging Foundation for financial support. We also thank Edanz ([www.edanzediting.co.jp](http://www.edanzediting.co.jp/)) for editing the English text of a draft of this manuscript.
[^1]: **Competing Interests:**The authors have declared that no competing interests exist.
| {
"pile_set_name": "PubMed Central"
} |
1. Introduction {#sec0005}
===============
Nitrobenzene (\"nitrobenzol\" or \"oil/essence of mirbane\") is a pale yellow aromatic nitro compound frequently used in synthetic rubber, dye and paint industries \[[@bib0005]\]. It is a potent oxidizer of the iron moiety of haemoglobin causing methaemoglobinemia leading to its inability to transport oxygen \[[@bib0010]\]. Clinical features of nitrobenzene poisoning include gastric irritation, nausea, vomiting, cyanosis, drowsiness, seizures, coma and finally respiratory failure culminating in death \[[@bib0015]\]. Intravenous methylene blue and vitamin C are commonly used for treatment of significant poisoning \[[@bib0020]\]. We report a case of severe nitrobenzene poisoning who was managed and discharged home with oral preparations of methylene blue and vitamin C.
2. Case report {#sec0010}
==============
A 45 years female with chronic alcoholism, had presented to a primary center with low Glasgow coma scale (GCS) of 7/15 (E2 = eye opening to pain, V2= incomprehensible sounds, M3 = abnormal flexion). She had a history of ingestion of about 50 ml of 20 % nitrobenzene solution while under the influence of alcohol with an alleged suicidal intent. She was intubated at another center and was referred to us for further management.
In the emergency department(ED) of our center, she had a GCS of 2/10 T (E1= no eye reponse, VT = endotracheal intubation, M1= no motor response) and was under manual bag and endotracheal tube ventilation. Her arterial oxygen saturation showed 80 % with clinically evident central and peripheral cyanosis. Her pulse rate was 130 beats per minute and blood pressure was 90/70 mmHg. An arterial blood gas analysis showed a pH of 6.98, PCO~2~ of 34 mm of Hg, bicarbonate of 8mEq/L, lactate of 8 mmol/L and PO2 of 170 mm of Hg.
She was transferred to the intensive care unit (ICU) where on the background of severe metabolic acidosis, sodium bicarbonate infusion was started. Noradrenaline infusion was started for hypotension which was gradually tapered off in a few hours. Haemodialysis was initiated for severe metabolic acidosis via a dialysis catheter inserted in the right femoral vein.
Meanwhile, a preparation of methylene blue was initiated enterally through the orogastric tube at a dose of 2 mg/kg (total 100 mg) and repeated after 12 h. Oral vitamin C 1.5 gm immediately and three times a day thereafter was started. Oral preparations for both drugs were used due to unavailability of intravenous preparations. Intravenous thiamine 100 mg immediate and three times daily was started in view of the patient\'s history of chronic alcoholism.
The acidosis resolved after three sessions of dialysis. Mechanical ventilation was gradually weaned to a fraction of inspired oxygen (FiO~2~) of 40 % with a positive end expiratory pressure (PEEP) of 5 cm of water (H~2~O) maintaining a targeted SpO~2~ of more than 90 %. She was extubated after 30 h of mechanical ventilation. Urine output was maintained at 1--2 ml/kg/hr.
The patient was transferred to the ward on the fourth day and discharged home on the sixth on oral thiamine and vitamin C.
3. Discussion {#sec0015}
=============
Nitrobenzene is easily absorbed from the respiratory tract, the gastrointestinal tract or the skin following intentional or accidental exposure. It is highly lipophilic because of which the highest concentrations get accumulated in the liver, brain, blood and stomach \[[@bib0025]\]. In the blood, it leads to the excessive oxidation of the iron moiety of the haemoglobin molecule forming methaemoglobin. This molecule has an oxidized iron moiety (Fe^3+^) instead of the usual reduced form (Fe^2+^). This methaemoglobin molecule is incapable of oxygen transport. During physiological states, less than 1 % of the total hemoglobin is oxidized to methameoglobin. This low level is maintained mainly because of two reductive pathways present in the red blood cells namely the hexose monophosphate (HMP) pathway and the diaphorase pathways. With increase in oxidative stress as when occurs during nitrobenzene poisoning, these pathways are overwhelmed leading to increased proportions of methaemoglobin \[[@bib0010]\]. As a result, the SpO~2~ falls despite a high PaO~2~ leading to the classical description of chocolate brown blood failing to redden even on exposure to ambient air \[[@bib0030]\]. The clinical symptoms are graded according to the methaemoglobin levels. Mild symptoms of headache, fatigue and nausea occur at 20--30 %; dyspnea, lethargy and tachycardia occur at 30--45 %; arrhythmias, coma, seizures, respiratory distress and lactate acidosis occur at 50--70 %; cardiovascular collapse and death occur at levels greater than 70 %. The lethal dose reported ranges from 1 to 10gm \[[@bib0010]\].
The resultant effect is hypoxic tissue injury despite a normal PaO~2~. This causes severe lactic acidosis from the metabolic switch to anaerobic respiration, hypoxic liver injury as evidenced by increased liver enzymes, hypoxic encephalopathic changes which may be irreversible and acute kidney injury \[[@bib0005],[@bib0035]\].
Although methaemoglobin level measurement was not possible, the classic findings of a reduced SpO~2~ with cyanosis but normal PO~2~ that did not improve with administration of supplemental oxygen were adequate evidences of clinically significant methaemoglobinemia in the background of history of nitrobenzene ingestion.
The management has two aspects: first, to restore normal physiological conditions with supportive management and second, to attempt to decrease the methaemoglobin level. The first includes administration of sodium bicarbonate and intermittent haemodialysis to attenuate metabolic acidosis, endotracheal intubation and IPPV for oxygenation and use of vasopressors for tissue perfusion \[[@bib0015],[@bib0040]\]. The second entails the usage of methylene blue and rarely exchange transfusion \[[@bib0045]\]. Although ample evidence of intravenous methylene blue is present in literature supporting its use, we had to resort to enteral administration due to its unavailability. Very limited experience is present for the use of oral methylene blue in nitrobenzene poisoning \[[@bib0050]\]. The bioavailability for oral methylene blue has been found to be around 72 % and based on its usage for other therapies, a dose of 2 mg/kg (100 mg) which has been found to be both safe and effective was chosen for our patient \[[@bib0055]\]. Vitamin C further decreases the oxidative stress acting as an oxygen scavenger which too was administered orally due to it's unavailability in intravenous form. Her improvement after oral methylene blue administration is suggestive that this route of administration is a viable option for treatment when the intravenous preparation is not available.
4. Conclusion {#sec0020}
=============
Nitrobenzene poisoning in a resource limited setting can cause therapeutic inadequacies due to unavailability of intravenous methylene blue and vitamin C preparations. However even oral preparations can be effective in the successful management as evident in our case.
Ethical approval {#sec0025}
================
Not applicable.
Consent for publication {#sec0030}
=======================
Written informed consent was taken from the patient herself for publication of this case report.
Declaration of Competing Interest
=================================
The authors report no declarations of interest.
None.
| {
"pile_set_name": "PubMed Central"
} |
Introduction
============
Nucleotide sequences diverge over time due to the combined effects of point mutation and homologous recombination. Recombination events cause changes to regions of contiguous bases in single events and were generally assumed to be rare in bacteria. However, there is growing evidence that homologous recombination has a significant impact on sequence diversification during bacterial genome evolution. A recent analysis on the MLST (Multilocus sequence typing) data of 46 bacterial and two archaeal species revealed 27 (56%) species in which homologous recombination contributed to more nucleotide changes than point mutation.[@R1] The rapid genetic change introduced by homologous recombination could facilitate ecological adaption and drive pathogenesis in bacterial pathogens.[@R2]^-^[@R5]
Currently, the MLST scheme, using DNA fragments from seven housekeeping genes,[@R6] has been routinely used to characterize bacterial isolates.[@R7] The standard MLST scheme has also been extended to construct fine-scale relationships and further subdivide identical multilocus sequence types (STs) using more loci or a large amount of shared genomic sequences.[@R8]^-^[@R12] Given the common occurrence of homologous recombination, it becomes crucial to investigate the genome-wide extent of homologous recombination, which could also benefit the construction of the strain history and tracking the spread of emerging pathogens.
Identification and Quantification of Nonvertically Acquired Genes via Recombination within Identical STs
========================================================================================================
Identifying recombinational exchanges in closely related strains is challenging as recombinational exchanges involved in a small number of nucleotides may be mistaken as point mutations. Guttman and Dykhuizen (1994) have successfully examined the clonal divergence of *E. coli* strains in the ECOR group A by considering the divergence time and mutation rate and showed that recombination has occurred at a rate 50-fold higher than the mutation rate in four loci.[@R13] Feil et al. (2000) estimated the ancestral allele for the isolates that differ only one locus out of the seven MLST loci and assigned recombination based on the number of derived nucleotides from the ancestral allele and on whether the nucleotides are novel in the population.[@R14]
We adopted a new approach (illustrated in [Fig. 1](#F1){ref-type="fig"}) to identify recombinant genes in *Neisseria meningitidis* strains with identical STs,[@R15] which does not require the estimation of divergence time and ancestral alleles and can be applied on any two strains with identical STs. In brief, nucleotide substitution was assumed to follow a binomial distribution and an upper bound of genome-wide divergence ($\mu$) by point mutation was calculated for no observed substitution in all nucleotide sites of the seven MLST loci. The estimated maximum genome-wide divergence was then used as a benchmark to compute a P-value for the observed nucleotide changes of each gene in the genome to be explained by point mutation. Genes that have more than the expected number of nucleotide changes at a significance level of 0.001 were deemed as recombinant genes. Our results revealed that up to 19% of commonly present genes in *N. meningitidis* strains with identical STs have been affected by homologous recombination.[@R15]
{#F1}
In another study on *E. coli* O104 (ST678) genomes, we visualized recombinant genes by plotting the pairwise DNA distance of orthologous genes along the genome and identified 167 genes in three gene clusters that have likely undergone homologous recombination.[@R16] A reanalysis on the orthologs between *E. coli* ON2010 and 55989 (labeled as Ec55989 thereafter to avoid unnecessary confusion) genomes using both pairwise DNA distance and the P-values as described in ref. [@R15] yielded remarkably similar results ([Fig. 2](#F2){ref-type="fig"}). In fact, the use of nucleotide divergence between two genomes for homologous recombination detection has been successful in other studies,[@R5]^,^[@R17] one of which was on two *E. coli* ST131 strains. It has been observed that a higher portion (at least 9%) of core genes in the *E. coli* ST131 genomes than in the *E. coli* ST678 genomes ([Fig. 2](#F2){ref-type="fig"}) are affected by homologous recombination.[@R5] The findings in both *N. meningitidis* and *E. coli* showed extensive genomic variation within identical STs. Since many bacterial species have a comparable or higher level of recombinogenicity than *N. meningitidis* or *E. coli*,[@R1] extensive genomic variation within identical STs should be expected in many bacterial species.
![**Figure 2.** Inferring genes involved in homologous recombination by comparing orthologs between two *E. coli* strains ON2010 and Ec55989. (**A**) DNA distance was measured using DNADIST of the PHYLIP package.[@R29] (**B**) P-values were calculated based on the maximum genome-wide divergence given the seven identical MLST loci as illustrated in [Figure 1](#F1){ref-type="fig"}. For simplicity, P -values smaller than 0.0001 were shown as 0.0001. Genes located in the prophage regions were colored in blue. Please note that more genes (4207 genes in total) were examined here than in our previous study[@R16] (3794 genes), since our previous study focused on the genes present in both the O104 strains and the IAI1 strain.](mge-3-e23463-g2){#F2}
It is important to note that the high genomic variation discovered within identical STs[@R5]^,^[@R15]^,^[@R16] should not be interpreted as artifacts of these studies. The high level of genomic variation within identical STs could, instead, be explained by that many non-vertical genes within identical STs are deleterious or transiently adaptive and undergo fast rates of evolution.[@R18] In fact, the ratio of recombination to mutation rates was higher in the comparison of clonally related strains[@R13]^,^[@R14] than of relatively broadly sampled strains from the corresponding species.[@R1] Such a discrepancy between the estimated recombination-mutation ratios highlights the need for a population genetics framework for the study of recombination and bacterial genome evolution.[@R19]
Genomic Regions Involved in Recombination
=========================================
Among the three gene clusters of recombinant genes we identified in *E. coli* O104,[@R16] one gene cluster contained 125 genes and was likely involved in direct chromosomal homologous recombination specific to the ON2010 strain. These 125 genes were found in 20 different functional categories and 70 of them were found in all the studied 57 *E. coli* and *Shigella* genomes. This is consistent with the conclusion that genes from all functional categories are subject to DNA exchange.[@R20] Furthermore, the nearest phylogenetic neighbors of these genes were not clustered in a single phylogenetic group. We hypothesized that extensive recombination with a broad spectrum of strains has taken place in one genome, and this highly mosaic genome then recombined with the precursor to the ON2010 genome.
The other two gene clusters of recombinant genes in *E. coli* O104 were located in the prophage regions, but the genes in these two gene clusters were identical between ON2010 and Ec55989 genomes.[@R16] It is noteworthy that the reanalysis with more single-copy genes (with details in [Fig. 2](#F2){ref-type="fig"}) revealed 5 prophage genes involved in recombination. These prophage genes are not present in all O104 strains and the outgroup IAI1 strain. This could be explained by frequent recombination of the prophage genes with infecting phages or different prophages from other bacterial chromosomes. Since all examined O104 genomes are of conserved genome synteny, our observations support the argument that homologous (legitimate) recombination drives module exchange between phages.[@R21] Together, these findings suggest that homologous recombination takes place frequently in both core genes and dispensable genes.
Phylogenomic Consequence
========================
As the cost of sequencing drops, the characterization of bacterial isolates has utilized more shared genes or loci and shifted toward phylogenomic analysis.[@R8]^-^[@R12]^,^[@R22] Quite often, multiple gene alignments were concatenated into a single super-alignment, from which phylogenies were reconstructed using a variety of methodologies. Such a data set, also known as a supermatrix, has been demonstrated to solve previously ambiguous or unresolved phylogenies,[@R23] even in the presence of a low amount of horizontal gene transfer in the data set.[@R24] Unfortunately, the supermatrix approach becomes very sensitive to recombination when applied to strains with identical STs due to limited genuine sequence diversity. The concatenated sequences of 3794 genes in the *E. coli* O104 strains[@R16] were overwhelmed by the phylogenetic signal of the 125 recombinant genes, as many other genes are identical among the *E. coli* O104 strains ([Fig. 2](#F2){ref-type="fig"}).
The accuracy and robustness of the constructed evolutionary relationships can be improved by the exclusion of recombinogenic and incongruent sequences.[@R8]^,^[@R25] In fact, the removal of the 125 recombinant genes from the *E. coli* O104 data set[@R16] has resulted in consistent phylogenetic relationships of O104 strains by different phylogenetic approaches. One interesting finding of our *E. coli* O104 study is that the number of identical loci implemented in BIGSdb[@R26] was less sensitive to homologous recombination than the concatenated sequences of all loci.[@R16] This could be explained by the fact that recombination has affected a relatively small number of genes but introduced a substantial amount of diversity in the ON2010 genome. It is further noteworthy that supertrees, another widely used approach for phylogenomic analysis[@R22] are not suitable for characterizing strains with identical MLST types, as many individual genes are identical or nearly identical and contain no or very limited phylogenetic information for each individual gene tree.
Homologous Recombination and Pathogenic Adaptation
==================================================
Homologous recombination can bring the beneficial mutations arising in different genomes together and have a strong impact on ecological adaptation.[@R4]^,^[@R27] One well-known example was the recombination in the *penA* genes during the emergence of penicillin resistance in *N. meningitidis*.[@R28] Variation of the *penA* gene corresponding to different levels of penicillin susceptibility has also been observed between *N. meningitidis* strains with the same MLST types.[@R15] Furthermore, genetic variation within the same MLST types has been evident in the capsule gene cluster and genes used for vaccine target in *N. meningitidis*.[@R15] These observations suggest a strong relationship between homologous recombination and pathogenic adaptation involved in antibiotic resistance, capsule biosynthesis and vaccine efficacy.
Recombination-mediated pathogenic adaptation was also evident in *E. coli*. Recombination has affected *fimH* which encodes mannose-specific type 1 fimbrial adhesin, resulting in distinct fluoroquinolone-resistance profiles in ST131 strains.[@R5] A survey of the *fimH* gene on the 57 *E. coli* and *Shigella* genomes[@R16] revealed that ON2010 was the only *E. coli* O104 genome containing a *fimH* blast hit \> 10% of length ([Fig. 3](#F3){ref-type="fig"}). Except one nucleotide, the *fimH* sequence in ON2010 was identical with E24377A and S88. On the ON2010 genome scaffold, *fimH* is upstream adjacent to a fructuronic acid transporter gene *gntP*, which is universally present in all *E. coli* and *Shigella* genomes. The *gntP* gene in ON2010 was also found to be involved in homologous recombination ([Fig. 2](#F2){ref-type="fig"}), and most importantly, the most similar sequences to the ON2010 *gntP* were also in E24377A and S88 (data not shown). The shared origin between the adjacent *fimH* and *gntP* genes in ON2010 suggested that patchily distributed genes involved in pathogenesis could be introduced by homologous recombination of the conserved flanking genes.
{#F3}
The author would like to thank Drs Gabriel Moreno-Hagelsieb, Markus Friedrich and Edward Golenberg for critically reading earlier versions of the manuscript. This work was supported by funds from Wayne State University.
Previously published online: [www.landesbioscience.com/journals/mge/article/23463](http://www.landesbioscience.com/journals/mge/article/23463/)
HaoWAllenVGJamiesonFBLowDEAlexanderDC Phylogenetic incongruence in E. coli O104: understanding the evolutionary relationships of emerging pathogens in the face of homologous recombination PLoS One 2012 e33971 10.1371/journal.pone.0033971
No potential conflicts of interest were disclosed.
| {
"pile_set_name": "PubMed Central"
} |
Introduction {#sec1-1}
============
Optic neuritis (ON) is an inflammation of the optic nerve and has various causes such as demyelinating, infective, vasculitis, sarcoidosis, toxic, metabolic, and hereditary neuropathy.
Most commonly it is a demyelinating inflammation of the optic nerve and typically, initially occurs in young adults. It may occur in isolation or association with multiple sclerosis (MS) or neuromyelitis optica.[@ref1]-[@ref4]
In cases associated with MS, ON is commonly the first manifestation of the chronic disease.[@ref3] It has been observed in long-term follow-up studies that up to 75% of female and 35% of male patients initially presenting with ON may develop MS.[@ref4]-[@ref7]
Magnetic resonance imaging (MRI) of the brain at the initial presentation can demonstrate silent demyelinating lesions. Patients with isolated ON (without MRI brain lesions) have a 16%, 22% risk of progression to MS at 5, 10 years follow-up, respectively. This risk increases to 56% in patients with MRI brain lesions at 10 years follow-up.[@ref8]-[@ref12]
The demyelinating ON affects females more than males with 3:1 ratio.[@ref13] Most of the patients are between 20 and 45 years of age.[@ref8]-[@ref10] It is unilateral in 70% of cases and is rare in children. Loss of vision, periocular pain and dyschromatopsia are considered to be the triad of inflammatory ON. For most patients with ON, visual function improves gradually over several weeks, in some cases can begin to improve after 1 week, even without any treatment. However, permanent residual deficits in color vision and contrast and brightness sensitivity can occur.
We aimed to analyze the records of ON patients, presented to King Fahad Specialist Hospital, Buraidah, KSA (KFSH), in regard to their presentation, natural history, treatment outcome and risk and association of MS. Moreover, whether ON in Saudi patients behaves differently from the Western and Asian ON patients?
Methods {#sec1-2}
=======
After approval from the Ethical committee, in this retrospective study, we retrieved data from the medical records of patients who attended the ophthalmology and neurology services in KFSH Buraidah, from 2006 to 2012.
Inclusion and exclusion criteria {#sec2-1}
--------------------------------
Patients older than 12 years presented with acute onset visual symptoms such as loss of visual acuity (VA) with or without eye pain and those with afferent pupillary defects. Patients below this age present to the children's hospital.
We excluded patients who showed any evidence of hereditary, vascular, toxic, metabolic, infiltrative, or compressive optic neuropathy.
The demographic variables and ON related variables including presenting symptoms, unilateral or bilateral involvement, relevant neurological symptoms, VA, color vision, pupil and fundoscopic findings, neuroimaging results, and any treatment offered and outcome were retrieved from the files of these patients and recorded on a structured pro forma.
Statistical analysis {#sec2-2}
--------------------
Statistical software "Statistical Package for the Social Sciences (SPSS)-20.0" was used for data analysis. Descriptive analyses were done. Ratios for gender distribution and mean standard deviation (SD) were computed for age distribution.
Frequencies and percentages were computed to present the categorical variables such as age of onset of ON, unilateral or bilateral involvement, VA, visual fields, fundoscopic, and neuroimaging findings.
Results {#sec1-3}
=======
We reported 60 patients of ON; 38 females and 22 males. Age ranged between 13 and 48 years (mean ± SD = 27.6 ± 8.8). Among these 98.3% were from Middle East origin including 91.67 Saudi patients. Only one case was from Indian origin. Demographic data of patients are shown in [Table 1](#T1){ref-type="table"}.
######
Demographic data of patients
About 68.3% of our patients presented with ON in the right eye, while in 31.7% of cases the affected eye was left. In 33.3% patients, color vision was not affected, while decreased in rest of cases.
The vision in affected eye was 20/200 or worse in 78.3% of our cases and better than 20/200 in 21.7% of cases ([Table 2](#T2){ref-type="table"}).
######
Findings of ophthalmology examination
Acute phase of ON was from 11 to 20 days in 75% of our cases. It lasted for \<05 days in 3.4% of cases and took more than 21 days in 6.6% of cases ([Table 3](#T3){ref-type="table"}).
######
Duration of disease
MRI scans of the brain were done for 52 patients, for the remaining 8 patients it was not requested, and they had not developed other neurological deficits. MRI brain was normal in 17 patients (28.33%) while, 33 patients (55%) had multiple lesions, and 2 (3.33%) patients had 2 or less lesions. 31 patients with multiple lesions developed definite MS on follow-up ([Table 4](#T4){ref-type="table"}).
######
MRI findings in patients
Only 29 patients (48.3%) had received systemic steroid, and the rest of the cases (51.7%) had nonsteroidal anti-inflammatory drugs. Vision improved in 51 patients (85%) to better than 20/200. While it was 20/200 in 4 (6.7%) and \<20/200 in 5 (8.3%).
Discussion {#sec1-4}
==========
In this article, we analyzed various aspects of ON and described the profile of ON in Saudi (Middle East) patients. At present, minimal information is available on ON within Saudi population.
In this study, we selected all patients with ON, similar to the patient population of Zhang *et al.*,[@ref14] Lim *et al*.[@ref15] studies, though ON treatment trial (ONTT)[@ref10] and other studies[@ref16]-[@ref20] had described patients with idiopathic ON only.
Young adults, aged 20-45 years, typically initially present with acute ON, although atypical cases of ON may be seen in elderly patients. Bilateral ON can occur in childhood and has less risk of progression to MS.[@ref21] Mean age of our patients was from 13 to 48 years.
ON is seen more commonly in Caucasians and quite rarely in black populations.[@ref22],[@ref23] Whites of Northern European descent develop ON 8 times more frequently than blacks and Asians. Most of our patients (98.3%) were from Middle East region including 91.67% Saudi patients and only one case was of Indian origin. An interaction is found to exist between ethnic origin and the latitude at which the patient grows up.[@ref23],[@ref24]
More recently, in a report from the United Kingdom about the incidence of ON within an ethnically diverse patient population, ethnicity bias has been observed.[@ref25]
In our study, the numbers of females affected by ON (63.3%) were more than males (36.7%); a similar trend was noted in ONTT[@ref26] and other studies[@ref27]-[@ref29] as well.
Causes of ON were not evident in most of our cases, similar to Asian studies done in China,[@ref17],[@ref20] Singapore,[@ref25] and India,[@ref16] as most of their patients had unknown etiology.
Painful eye movements were present in 85% of our cases, which is in agreement with the ONTT that reported 92% of patients had painful ocular movements and was also one of presenting feature in some other studies.[@ref12],[@ref27],[@ref29],[@ref30]
We could not retrieve much information related to visual fields defects from our records. However, any type of visual field defect is possible as suggested by the ONTT.[@ref10] Studies from Taiwan[@ref20] and Japan[@ref18] have reported a diffuse depression as most common field defect in their patients.
Color vision was decreased in 66.7% or our patients. This is in agreement with ONTT[@ref10] where patients showed mixed red-green and blue-yellow color defects, either one or the other type predominating.[@ref31]
The incidence of MS associated with ON is found to be most common in populations of Western Europe and North America, located at higher latitudes and less common closer to the equator.[@ref27] MRI showed changes consistent with demyelination of the brain in 48.7% of the patients in ONTT.[@ref10] In comparison to Western Cohort, the risk of MS was found to be low in Asians with ON.[@ref17],[@ref32] ON was associated with MS in 25.5% of patients of Lim *et al*.[@ref15] and Tan.[@ref32]
In our study, 55% patients had demyelinating lesions on MRI brain, and most of them developed MS in follow-up. This high risk and association of MS is almost similar to the Western Europe and North American populations.[@ref10],[@ref23],[@ref24]
The treatment of acute ON is a symptomatic remedy of the acute symptoms of pain and decreased vision caused by demyelinating inflammation of the nerve. Various regimens of corticosteroids have been used for this purpose. Immunomodulating drugs are recommended for patients with ON whose brain lesions on MRI indicate a high risk of developing MS.[@ref26]-[@ref30]
We noticed, only 29 patients (48.3%) had received systemic steroid, and the rest of the cases (51.7%) had nonsteroidal anti-inflammatory drugs as initial treatment in the acute stage.
Vision improved in 85% of our cases, to better than 20/200, while in 15% of cases it remained 20/200 or less.
It has been reported that recovery is not as good with poor baseline VA, but even with ≤20/200 at baseline, recovery to ≥20/40 occurs in 85% recovery of visual loss occurs spontaneously starting within 2-3 weeks in 80%, stabilizing over months and continuing to improve for up to 1 year.[@ref30]
Acute phase of ON lasted from 11 to 12 days in 75% of our cases. In the ONTT, 79% and 93% of patients started to show signs of improvement within 3 and 5 weeks of onset, respectively. The severity of initial visual loss does appear to affect final visual outcome, and in the ONTT the best predictor of visual recovery was the baseline acuity at enrolment.[@ref30]
Limitations {#sec1-5}
===========
This is a retrospective analysis of the patients who had presented with ON. In some cases, we could not exactly correlate the timing of the MRI scan and the onset or duration of the ON. Furthermore, could not retrieve detailed information related to visual fields defects from our records.
Conclusion {#sec1-6}
==========
Most of our patients had idiopathic ON, and almost 85% had good visual recovery, in this regard our results are comparable with such studies done in other Asian countries.
Interestingly, our ON patient population is different from the other Asian countries in terms of high-risk and association of MS, as suggested by multiple demyelinating lesions on brain MRI in 55% of cases. This is almost similar to the Western Europe and North American populations.
History and examination are helpful to clinically identify typical cases. Only 29 patients (48.3%) had received systemic steroid as initial treatment in the acute stage, for management there should be multi-disciplinary approach including ophthalmology and neurology services. MRI brain is useful to identify cases having a high risk of developing MS. The association between ON and MS and possible use of immune modulating therapies should be discussed with such patients. We have single center retrospective observations; there should be multicenter prospective studies with a large number of patients.
| {
"pile_set_name": "PubMed Central"
} |
1. Introduction
===============
Claudins are a family of tight junctional proteins which are highly expressed in both benign and malignant ovarian tumors \[[@b1-ijms-14-10412]\]. Normal epithelial cells are held together by tight junctions (TJs), adherens junctions (AJs), and gap junctions. TJs are the apical cell-cell adhesions that are important for epithelial cell polarity and regulate paracellular permeability by blocking the free diffusion of proteins and lipids between the apical and baso-lateral domains of the plasma membrane \[[@b2-ijms-14-10412],[@b3-ijms-14-10412]\]. TJs are comprised of multiple membrane proteins such as occludin and claudin family proteins and several other associated peripheral proteins such as zonula occludens 1-3 (ZO-1, -2 and -3) \[[@b3-ijms-14-10412],[@b4-ijms-14-10412]\]. These proteins are seen at the cell membrane interface where they contribute to the formation of the TJ and interact to form the diffusion barrier. Epithelial TJs are considered to be dynamic structures and the correlation of epithelial breakdown or dysfunction with the promotion of the neoplastic process has been suggested by previous studies \[[@b5-ijms-14-10412]\].
Claudins have been shown to be essential and sufficient to form TJ strands and account for some of the selective variability of different barriers \[[@b6-ijms-14-10412],[@b7-ijms-14-10412]\]. There is evidence that disruption of the cell to cell adhesion is a critical step in the process of cellular transformation and tumor cell metastasis \[[@b8-ijms-14-10412]\]. The role of the claudins in this process is continuously being explored with new discoveries still occurring. Apart from contributing to mechanical cell adhesion at epithelial and endothelial cell interfaces, claudins also have the capacity to recruit cell signaling proteins and as such may regulate cell proliferation, differentiation and subsequent neoplastic transformation \[[@b9-ijms-14-10412],[@b10-ijms-14-10412]\].
There are 27 different types of claudins identified to date with varying cell- and tissue-specific expression \[[@b11-ijms-14-10412]\]. The expressions of claudins may also vary in different parts of the same organ. Most tissues express multiple claudins. The different claudin members may interact within a given tissue and this combination of the claudin proteins is thought to determine the strength and selectivity of the TJs. As they are cell surface proteins, most claudin positive tumor cells will show strong cell membrane staining with weak if any cytoplasmic reactivity noted in these cells. Of interest, the deregulation of the mitogen-activated protein kinase pathway can lead to the mis-localization of TJ proteins, including the claudins \[[@b12-ijms-14-10412]\]. The delocalization of claudin proteins from cell membranes is common among transformed cells and in ovarian cancer this is associated with tumor cell migration and invasion \[[@b10-ijms-14-10412],[@b13-ijms-14-10412]\]. Normal ovarian surface epithelial cells do not express either claudin-3 or claudin-4, however these claudins are both expressed at high levels in the majority of ovarian cancers \[[@b14-ijms-14-10412]--[@b26-ijms-14-10412]\].
Claudin-3 and -4 function as receptors for *Clostridium perfringens* enterotoxin (CPE), a potent cytolytic toxin. The use of this enterotoxin may be therefore exploited for therapeutic and diagnostic benefit for claudin-3 and -4 expressing tumors. Claudin expression has been identified in other gynecologic tumors as well, including cervical (preneoplastic and neoplastic lesions) and endometrial adenocarcinomas \[[@b27-ijms-14-10412]\]. Some claudins also have been shown to have a prognostic role in particular tumor types, for example, claudin-3/-4 has a prognostic role in ovarian cancer, claudin-1 in colon cancer, claudin-10 in hepatocellular carcinoma and claudin-18 in gastric cancer \[[@b28-ijms-14-10412]--[@b31-ijms-14-10412]\].
2. Structure and Function of Claudins
=====================================
Claudins are known as tetraspan membrane proteins consisting of intracellular amino and carboxy terminals, 4 transmembrane domains and 2 extra-cellular loops mediating interactions between claudins on adjacent cells \[[@b2-ijms-14-10412],[@b32-ijms-14-10412],[@b33-ijms-14-10412]\] ([Figure 1](#f1-ijms-14-10412){ref-type="fig"}). The amino acid sequences of the first and fourth transmembrane domains are highly conserved among the different claudin isoforms, however the sequences of the second and third domains are typically more diverse \[[@b34-ijms-14-10412]\]. The second extracellular loop acts a binding site for *Clostridium perfringens* enterotoxin (CPE) in claudin-3 and -4 \[[@b35-ijms-14-10412]\]. Claudin-3 and -4 consists of 220 and 209 amino acids respectively. Claudin-3 and -4 are considered to be the low and high affinity receptors for CPE respectively. Occludin, tricellulin and marvelD3 are other tetraspan transmembrane TJ proteins \[[@b4-ijms-14-10412],[@b36-ijms-14-10412],[@b37-ijms-14-10412]\]. The scaffolding proteins like ZO-1, -2 and -3 and also signaling proteins are associated with TJs by binding of their PDZ-domains to respective binding sites at the carboxy terminus of claudins \[[@b38-ijms-14-10412]\]. These membrane associated proteins govern the assembly and disassembly of TJ \[[@b32-ijms-14-10412],[@b39-ijms-14-10412]\]. The carboxy terminus of most claudins contain potential serine and/or threonine phosphorylation sites \[[@b33-ijms-14-10412]\]. The barrier function of claudins may be modulated through phosphorylation of the serine/threonine phosphorylation sites at the carboxy tail by various kinases such as cyclic AMP-dependent protein kinase and WNK4 \[[@b40-ijms-14-10412]\]. The carboxyterminal tail is the region that shows the most sequence and size heterogeneity among the claudin proteins \[[@b34-ijms-14-10412]\].
Claudins form the backbone of the TJ and are overall a highly structurally-related family of proteins with claudin-16 and -23 being the most different and claudins-6 and -9 are the most similar followed closely by claudin -3 and -4 and claudin -1 and -7 \[[@b41-ijms-14-10412]\]. Some claudin genes have been found to be closely linked in terms of their proximity in the human genome, (for example claudin-3 and -4) \[[@b34-ijms-14-10412]\]. It is however uncertain whether this genome arrangement has a function in coordinate regulation of the TJ. Interestingly, claudin-3 and -4 have been documented to have coordinate expression in several normal and neoplastic tissues and the combination is commonly found elevated in a variety of cancers \[[@b10-ijms-14-10412],[@b41-ijms-14-10412],[@b42-ijms-14-10412]\]. In general, the 27 claudin genes that have been identified are typically small and have few introns or lack introns altogether. Claudin-3 and-4 have been localized at position 7q11, both of these genes have only one mRNA transcript form. Claudin-1 also has only one transcript form and the gene *CLDN1* is located at position 3q28. *CLDN5* can be found at position 22q11 and has two variants of mRNA that may be produced after transcription. *CLDN10* has also two variants of RNA transcript but has a gene position of 13q31. The claudin proteins show a wide range of sequence similarity and the size of these proteins is approximately in the range of 205 to 305 amino acids \[[@b34-ijms-14-10412]\].
The claudins can be functionally divided into barrier-forming claudins such as claudin-1,-3,-4,-5 and pore forming claudins such as claudin-2,-7,-10 and -16 \[[@b38-ijms-14-10412]\]. In a subtype dependent manner, the expression of barrier-forming claudins decreases paracellular permeability of ions, solutes and proteins while the expression of pore-forming claudins generally increases paracellular permeability to ions. As a result, the tissue specific expression of the different claudin isoforms will determine the permeability properties of the TJs in that tissue \[[@b43-ijms-14-10412]\]. Most cells express multiple different claudin isoforms and these isoforms have the ability to co-polymerize into heteropolymers by homophilic and heterophilic interactions. The various types of claudin co-polymers then work together to regulate junctional permeability and to impart strength and selectivity to the TJ \[[@b16-ijms-14-10412],[@b44-ijms-14-10412]\].
Based on the amino-acid sequence, the claudins may also be separated into two subgroups namely classic and non-classic claudins. Classic claudins are claudin-1 through -10, -14, -15, -17, -19 and share higher homology among each other compared to non-classic claudins (claudin-11, -12, -13, -16, -18, -20 through -24) \[[@b38-ijms-14-10412]\]. Classic claudins are also more likely to share a common helix-turn-helix structure of the extracellular loop 2 which is involved in paracellular tightening \[[@b45-ijms-14-10412],[@b46-ijms-14-10412]\].
While there still remains much to be uncovered about the claudin structure multiple studies have shown that abnormalities in claudins may result in the disruption of TJ barrier function as well as alter paracellular permeability. These structural abnormalities are known to be associated with a number of pathologic processes such as pulmonary edema, diarrhea, inflammatory bowel disease and kidney disorders \[[@b47-ijms-14-10412]--[@b50-ijms-14-10412]\]. Additionally, germline mutations in these genes can lead to familial diseases such as the autosomal recessive form of non-syndromic sensorineural deafness which results from a defect in the claudin-14 gene \[[@b51-ijms-14-10412]\]. As such, it is apparent that proper cell to cell and cell to extracellular matrix interactions are essential for continued normal tissue and organ functioning. Similarly, the proteins constituting TJs, such as the claudins, are quite likely to have a central role in tumorigenesis and also in tumor spread.
Claudins are found in cell adhesions and are thought to facilitate the communication of the extracellular environment to both intracellular signaling pathways and to the cytoskeleton. Tight junction disruption in premalignant tissues can increase the likelihood of progression to a frankly invasive tumor due to passage of large solutes across epithelial barriers allowing growth factors (usually in luminal fluids in epithelial tissues) to now bind to their growth factor receptors (usually on the baso-lateral surface facing interstitial fluid and the bloodstream) and this interaction may lead to continuous stimulation of premalignant cells \[[@b13-ijms-14-10412]\]. Claudin expression may also affect the epithelial permeability to substances such as growth factors and also modulate the response of other tight junction proteins to various types of injury \[[@b52-ijms-14-10412]\].
The pattern of expression of claudins in normal tissue, benign and malignant tumors is not only complex but also organ dependent \[[@b10-ijms-14-10412],[@b41-ijms-14-10412],[@b53-ijms-14-10412]\]. Large scale serial analysis of the genome and gene expression arrays have documented higher expression of claudin-3, -4, -7 and -10 in ovarian carcinoma compared to normal ovarian surface epithelium \[[@b15-ijms-14-10412],[@b16-ijms-14-10412],[@b18-ijms-14-10412],[@b43-ijms-14-10412],[@b54-ijms-14-10412]\]. These findings have been validated using quantitative reverse transcription-polymerase chain reaction (qRT-PCR) and immunohistochemistry (IHC). Generally, studies comparing the expression of claudins in benign or borderline ovarian tumors *versus* ovarian carcinoma have not been entirely conclusive \[[@b1-ijms-14-10412],[@b14-ijms-14-10412],[@b16-ijms-14-10412],[@b55-ijms-14-10412],[@b56-ijms-14-10412]\]. However, most of the prevailing literature provides evidence that claudin-3 and -4 are highly differentially expressed in ovarian cancer and also correlate with chemo-resistance and poorer survival albeit some differing results have been reported depending on cell lines studied.
Claudin-1, -3, -4, -5 and -7 are the claudins most commonly overexpressed in ovarian tumors ([Table 1](#t1-ijms-14-10412){ref-type="table"}). Overall, in reviewing several studies, it is evident that claudin expression is altered in a variety of tumors with the most commonly identified claudins to have an altered expression being claudin-1, -3, -4, -5, -7, -10 and -16 \[[@b10-ijms-14-10412]\]. This phenomenon is likely due to the role of claudins in tumor survival and invasion, as it is not unusual for some carcinomatous tissue to lose their TJ proteins as they grow and develop \[[@b10-ijms-14-10412],[@b57-ijms-14-10412]--[@b59-ijms-14-10412]\], for example, claudin-1 and -7 are typically downregulated in hepatocellular carcinomas \[[@b5-ijms-14-10412],[@b60-ijms-14-10412]\].
TJ-tight junction, HCV- hepatitis C virus, CPE- Clostridium perfringens enterotoxin, RPE- retinal pigment epithelium, EPCAM- epithelial cell adhesion molecule.
Claudin function is regulated at many sites including at the level of the tight junction as a result of crosstalk between tight junction components \[[@b64-ijms-14-10412]\]. In addition several claudins are known to be phosphorylated by kinases which may affect both claudin position and function. Another potential mechanism of regulation of claudin expression is endocytic recycling of claudin proteins \[[@b65-ijms-14-10412]\]. At the transcriptional level, there is evidence that transcription factors such as GATA-4 and Snail are able to bind to the promoter regions of several claudin genes and affect their expression \[[@b66-ijms-14-10412],[@b67-ijms-14-10412]\]. The claudins may also be downregulated not only at the point of transcription but also post-transcriptionally via a variety of cytokines and growth factors \[[@b64-ijms-14-10412],[@b68-ijms-14-10412]\].
Epidermal growth factor (EGFR) signaling has been demonstrated to modulate the expression of the claudins in various cell types \[[@b48-ijms-14-10412],[@b69-ijms-14-10412],[@b70-ijms-14-10412]\]. Recently this mechanism of the TJ protein regulation in ovarian cancers was explored by treating both ovarian mucinous and serous cystadenocarcinoma cell lines with EGF \[[@b71-ijms-14-10412]\]. EGF was found to downregulate claudin-3 in mucinous ovarian carcinoma cell lines and claudin-4 in ovarian serous cystadenocarcinoma by inducing the degradation of these proteins with also changes in the structure and function of TJ via the MEK/ERK or PI3K/AKT signaling pathway. The pretreatment with EGFR inhibitors, MEK/ERK inhibitors and PI3K/AKT inhibitors in the ovarian mucinous cystadenocarcinoma cell lines prevented the decrease of claudin-3 by EGF. On the other hand, for serous ovarian carcinoma cell lines, pretreatment with inhibitors of EGFR, MEK/ERK but not PI3K/AKT prevented the decrease in claudin-4 by EGF. This suggests alternative mechanisms for claudin regulation by EGF among the different ovarian carcinoma subtypes *in vitro*. These results provide evidence that EGF may affect claudin and TJ function in ovarian cancer cells during cancer development. Additionally, in ovarian serous cystadenocarcinoma cell lines, EGF was found to downregulate the cytotoxic effects of CPE via claudin-4. As a result, it is also theorized that EGF may affect effective claudin-4 targeting therapy with CPE in serous cystadenocarcinoma \[[@b71-ijms-14-10412]\].
3. Claudins in Ovarian Cancer
=============================
3.1. Claudin-1 and Claudin-2
----------------------------
Claudin-1 expression has been studied and demonstrated in ovarian serous carcinoma and ovarian endometrioid carcinoma \[[@b72-ijms-14-10412]\]. Claudin-1, like claudin-3 and -4, is an epithelial specific claudin protein. The expression of claudin-1 is elevated in many types of cancer cells and is proposed to be potentially causally involved in tumor growth and progression. Claudin-1 has been shown to have anti-apoptotic activity and is thought to play a role in the expression and localization of β-catenin and E-cadherin. As such claudin-1 plays a role in the epithelial to mesenchymal transition and the c-abl-Ras-Raf-1-ERK1/2 signaling axis is important in claudin-1 induced malignant progression \[[@b73-ijms-14-10412],[@b74-ijms-14-10412]\].
Claudin-1 has been identified as one of the genes notably upregulated in ovarian cancer-initiating cells and claudin-1 overexpression in these cells leads to a low degree of cell differentiation and a high rate of invasive growth \[[@b75-ijms-14-10412]\]. It has been discovered that microRNA-155 (miR-155) targets claudin-1 with specificity and the increased expression of endogenous mature miR-155 may have an inhibitory effect on human ovarian cancer-initiating cell proliferation and invasion *in vitro* and *in vivo* through its effect on limiting claudin-1 expression. Claudin-2 has not been noted as a tight junction protein with high expression in ovarian cancer. Data regarding the expression and function of claudin-2 mainly centers around hepatocellular, breast and gastrointestinal carcinomas as well as Paget's disease \[[@b27-ijms-14-10412],[@b76-ijms-14-10412],[@b77-ijms-14-10412]\].
3.2. Claudin-3 and Claudin-4
----------------------------
Epithelial ovarian carcinoma remains the gynecologic malignancy with the highest mortality rate \[[@b78-ijms-14-10412]\]. Two-thirds of patients have advanced disease at the time of diagnosis and unfortunately the majority of patients will recur after an initial response to the combination of maximal cytoreductive surgery and combined platinum and paclitaxel-based chemotherapy \[[@b79-ijms-14-10412],[@b80-ijms-14-10412]\]. Thus the identification of novel therapeutic approaches against chemotherapy resistant/recurrent ovarian cancer remains a high priority.
Ovarian cancers of varying subtypes including mucinous, serous, undifferentiated, clear cell, and endometrioid carcinomas have been found to highly express claudin-3 and claudin-4 but normal ovarian surface epithelium does not \[[@b14-ijms-14-10412],[@b59-ijms-14-10412],[@b81-ijms-14-10412]--[@b83-ijms-14-10412]\] ([Table 2](#t2-ijms-14-10412){ref-type="table"}). This data suggests that the low-level expression of these claudins is associated with a benign condition and that high expression is more likely to be a signal of a malignant transformation. Consistent with this view the expression of claudin-3 and -4 in ovarian epithelial cells is thought to enhance neoplastic cell invasion and has been found to be associated with increased matrix metalloproteinase-2-activity and angiogenic effects \[[@b82-ijms-14-10412]\]. Some research has also suggested that up-regulation of claudin-3 may be an early event in the development of epithelial ovarian cancer and have potential application in detection of early stage disease \[[@b56-ijms-14-10412]\].
Using gene expression profiling, the differential patterns of expression between ovarian tumors and normal ovarian cells has been explored. Several groups including our own have recently used high throughput gene array technologies to compare the expression profiles of ovarian cancer to those of normal ovaries with the aim of identifying potential diagnostic and therapeutic markers for this aggressive malignancy. Claudin-3 and -4 genes have been reported to be highly differentially expressed in biologically aggressive malignancies including ovarian serous carcinoma (OSC) and the identification of claudin protein expression has proven to be of clinical relevance in this tumor and a variety of others \[[@b10-ijms-14-10412],[@b53-ijms-14-10412]\]. The mechanism of the increased claudin-3 and-4 expression in ovarian carcinoma is thought to be the result of epigenetic modifications of the claudin promoter regions in the cancer cells resulting in increased cell survival, invasion and motility \[[@b59-ijms-14-10412],[@b93-ijms-14-10412],[@b94-ijms-14-10412]\].
Our research group also examined the genetic fingerprints of ovarian serous cancer in flash-frozen tumor biopsies as well as primary and/or established ovarian cancer cell lines and compared the gene expression signature with that of normal cells including ovarian surface epithelium exposed to short-term culture or immortalized normal ovarian cell lines (HOSE). The gene expression in flash-frozen OSC was found to have a high correlation with those of purified primary ovarian serous carcinoma in short-term *in vitro* cultures. Claudin-3 and -4 were found among the most highly overexpressed genes in OSC compared to HOSE \[[@b15-ijms-14-10412]\].
As the comprehensive study of the molecular signature of ovarian cancer has identified claudin-3 and -4 as top differentially expressed genes, next to be investigated was the gene expression profile in chemotherapy-naïve *versus* chemotherapy-resistant ovarian cancer. Chemotherapy-resistant ovarian cancer was found to express the claudin-3 and -4 genes at significantly higher levels when compared with chemotherapy-naïve ovarian tumors \[[@b95-ijms-14-10412],[@b96-ijms-14-10412]\]. These ovarian cancer cell lines continued to display considerable sensitivity to CPE *in vitro* and *in vivo* regardless of their documented resistance to multiple chemotherapeutic agents \[[@b97-ijms-14-10412]\].
There is also great interest in the mechanisms and markers of platinum-resistance secondary to the importance of this drug as first-line treatment of ovarian cancer whether in the neoadjuvant or adjuvant setting. However limited information is currently available about the exact mechanisms of cisplatin resistance in ovarian cancer including whether or not claudin-3 or -4 may play roles as influx or efflux transporters of cisplatin. It is postulated that claudin-3 or -4 overexpression may inhibit the penetration of chemotherapeutic agents into ovarian cancer tissue and as a result generate chemo-resistance \[[@b98-ijms-14-10412]\].
Quantitative proteomic technology integrated with mRNA expression levels has been recently utilized in an effort to identify protein markers capable of prospectively determining chemo-resistant ovarian tumors \[[@b96-ijms-14-10412]\]. In this study a total of 1117 proteins were identified and quantified in cisplatin-sensitive and -resistant ovarian cancer cells. The relative expression of 121 of these proteins varied between the cell lines with 58 of them found to be overexpressed in cisplatin-resistant cells. Claudin-4 was identified as one of the top proteins associated with cisplatin resistance in ovarian cells with a 7.2 fold overexpression level.
A Japanese research group reported similar results showing that claudin-4 expression was higher in ovarian cancer tissue from platinum-based chemo-resistant patients *versus* chemo-sensitive patients. In this study suppression of claudin-4 resulted in a significant increase of cisplatin sensitivity and cellular accumulation of fluorescence-labeled cisplatin. Claudin-4 expression was significantly greater in ovarian cancer tissue from chemo-resistant patients compared to chemo-sensitive patients. Thirty-three out of the 43 cases (76.7%) of patients with ovarian cancer examined had positive claudin-4 expression with a significant shorter survival noted in the claudin-4 positive *versus* claudin-4 negative group \[[@b98-ijms-14-10412]\].
In contrast, a recent study by Shang *et al.* using two established cell lines provided some support to the notion that claudins-3 and -4 may serve to constrain the growth of human ovarian cancer xenograft and limit metastatic potential \[[@b42-ijms-14-10412]\]. In this study knockdown of claudin-3 and -4 increased the *in vivo* growth rate and metastatic potential of the xenografted tumors and reduced expression of these claudin proteins enhanced cell migration and invasion in *in vitro* assays \[[@b42-ijms-14-10412]\]. In the Shang *et al.* study, the loss of either claudin-3 or -4 resulted in the down-regulation of E-cadherin mRNA and protein as well as activation of β-catenin pathway signaling and as such claudin-3 and -4 may mediate interactions with other cells *in vivo* that result in reduced growth and metastatic potential through the maintenance of E-cadherin expression and by limiting β-catenin signaling \[[@b42-ijms-14-10412]\]. E-cadherin is the major structural protein of the adherens junctions and loss of E-cadherin is declared as a hallmark of the epithelial-to-mesenchymal transition through which it is speculated that cells must pass before becoming metastatic \[[@b99-ijms-14-10412],[@b100-ijms-14-10412]\]. It is known that E-cadherin acts as a negative regulator of the β-catenin signaling pathway, which is a pathway that guides cell destiny through the regulation of cell growth, motility and survival \[[@b42-ijms-14-10412],[@b101-ijms-14-10412]\]. As such, down-regulation of E-cadherin as well as activation of β-catenin pathway signaling could account for the increased metastatic potential of the ovarian cancer cell lines studied. Of interest, low-level expression of claudin-3 and claudin-4 in other human solid tumors has also been linked to a mesenchymal pattern and, as such, correlates to an overall poor survival in breast, esophageal, colon and pancreatic carcinoma \[[@b102-ijms-14-10412]--[@b105-ijms-14-10412]\].
Importantly, the Shang *et al.* study also lends support to the body of evidence indicating that most ovarian cancers arise from the distal fallopian tube epithelium even though these cancers are largely accepted to arise from multiple locations including ovarian surface epithelium \[[@b106-ijms-14-10412]--[@b108-ijms-14-10412]\]. In this study, immunohistochemical analysis was performed for claudin-3 and -4 expression in both the distal fallopian tube and tumor in six cases of serous ovarian cancer. All six cases had high claudin-3 and -4 expression in both sites. As the majority of ovarian cancers show a high expression of these claudins, it has been postulated that ovarian cancer develops from an epithelium which at its baseline or preneoplastic state normally expresses these two proteins. This same group has recently demonstrated in a single cell line that knockdown of claudin-3 and -4 resulted in marked changes in the phenotype of ovarian cells including an increased resistance to cisplatin by regulating the expression of the copper influx transporter CTR1 \[[@b109-ijms-14-10412]\].
Taken together the results of these latter studies are consistent with the conclusion that the effect of claudin-3/-4 knockdown on cisplatin resistance may be the consequence of promoting an epithelial to mesenchymal transition after the downregulation of the claudin proteins. This interpretation is supported by previous studies in gynecologic carcinosarcoma showing that high expression of claudin-3/-4 is present in the epithelial but not in the sarcomatous component of multiple carcinosarcomas studied by immunohistochemistry \[[@b89-ijms-14-10412]\].
The differences in claudin-3/-4 expression by ovarian cancer subtype and the correlation with outcome in ovarian cancer patients has also been researched by several groups \[[@b1-ijms-14-10412],[@b14-ijms-14-10412],[@b55-ijms-14-10412],[@b110-ijms-14-10412]\]. In one study, low claudin-3 protein expression was associated with a trend towards a poor survival in 115 primary ovarian carcinomas with 68.6% being of serous histology \[[@b56-ijms-14-10412]\].
One large study found that claudin-4 was expressed in nearly 70% of the ovarian cancer tissues examined and was differentially expressed across ovarian cancer subtypes, with the lowest expression noted in clear cell ovarian carcinomas. The highest percentage of expression was detected in endometrioid and mucinous subtypes (both 77.4% positive) compared to serous (72.17%) and clear cell (57.58%) subtypes. Also no association was found between claudin-4 expression and disease-specific survival in any subtype \[[@b81-ijms-14-10412]\]. In yet another study, claudin-3 and -4 were significantly up-regulated by 5-fold or more in most subtypes of ovarian epithelial carcinoma. By immunohistochemistry (IHC) claudin-3 was expressed in 81% and claudin-4 expressed in 85.7% of 84 serous adenocarcinomas respectively. Borderline serous tumors and adenomas had significantly lower expression of these proteins than the adenocarcinomas. The survival analysis in this study revealed that serous adenocarcinoma patients with high claudin-3 expression had a substantially shorter survival and multivariate analysis showed claudin-3 overexpression to be an independent negative prognostic factor \[[@b111-ijms-14-10412]\]. Consistent with these results claudin-3 gene silencing with small interfering RNA has been shown in mouse models to suppress ovarian tumor growth and metastasis \[[@b112-ijms-14-10412]\].
In contrast to these results Litkouhi *et al.* found the highest percentage of claudin-4 expression in clear cell and endometrioid subtypes of ovarian cancer however this study had a much smaller sample size which may at least partially explain the differing results. Also in this study, there was no statistically significant difference in survival found between the claudin-4 positive and claudin-4 negative groups \[[@b110-ijms-14-10412]\].
Support for the role of claudins in promoting tumor progression has also come from studies evaluating the anatomic-site related expression and the prognostic role of claudins in ovarian cancer. In one particular study, the data of immuno-stains for claudin-1, -3, -4 and -7 on pleural effusions, corresponding primary tumors and solid metastasis of ovarian cancer were all gathered in order to identify associations between anatomic site, clinic-pathologic parameters and survival. It was found that all 4 claudins were expressed in \>85% of tumors at all anatomic sites \[[@b1-ijms-14-10412]\]. Moreover, with the exception of claudin-4, all the other claudins were upregulated in ovarian cancer effusions compared with solid tumors and that the expression of claudins-3 and -7 in pleural effusions independently predicts poor survival in ovarian cancer \[[@b28-ijms-14-10412]\].
Facchetti *et al.* investigated the usefulness of claudin-4 in the diagnosis of mesothelioma and other malignancies that may mimic mesothelioma. In this study, analysis was performed on 454 tumors, including 82 mesotheliomas, 336 carcinomas of different origins, 36 non-epithelial spindle cell neoplasms as well as 97 cytological samples from a combination of reactive effusions, mesothelioma and metastatic carcinomas. Claudin-4 was consistently negative in normal and reactive mesothelium as well as in all 82 mesotheliomas but strong reactivity (using anti-claudin-4 primary antibody) was found in the significant majority of serosal metastasis from primary carcinomas particularly lung, breast, gastrointestinal tract, pancreas, ovary and primary peritoneal carcinoma. In effusions, metastatic tumor cells stained positive in 96.8% of cases. Facchetti's study therefore suggested that claudin-4 may be a pan-carcinoma marker with high sensitivity and specificity and that this claudin protein may be considered a primary immunohistochemical marker to rule out the diagnosis of mesothelioma in patients with pleural and peritoneal biopsies and effusions \[[@b113-ijms-14-10412]\].
3.3. Claudin-5
--------------
Claudin-5 is mainly present in vascular endothelial cells but is also seen in ovarian epithelial tumors but at a much lower frequency than claudins-1, -4 and -7 \[[@b1-ijms-14-10412]\]. In a study of 60 different types of ovarian lesions, sex-cord stromal tumors and cysts were mainly negative for claudins-1, -4, -5, and -7. In immature teratomas, mostly the epithelial component was usually positive and the other components were negative. Dysgerminomas did not express any of the claudins-1, -4, -5, and -7. The authors findings in this study were that claudins-1, -4, and -7 were mainly expressed in epithelial ovarian tumors \[[@b14-ijms-14-10412]\].
The role of claudin-5 and vascular endothelial growth factor (VEGF) in the development of malignant ascites was explored recently. Claudin-5 has been shown in an *in vitro* corpus luteum model to be important for the regulation of vascular permeability \[[@b114-ijms-14-10412]\]. VEGF is produced by malignant cells including ovarian cancer cells and induces angiogenesis to promote tumor growth and survival. VEGF has also has been shown to enhance vascular permeability and influence endothelial TJs \[[@b115-ijms-14-10412]--[@b117-ijms-14-10412]\]. Up to 24-fold higher VEGF levels may be induced in malignant tumors compared to benign ovarian cysts \[[@b118-ijms-14-10412]\]. These high VEGF levels are theorized to increase local permeability and result in fluid extravasation (third spacing) and thus ascites formation. The role of VEGF-dependent production of claudin-5 as a regulator of vascular permeability in ovarian cancer patients was thus investigated by studying the amount claudin-5 in peritoneal tissue as well as VEGF in serum and ascites. The researchers also established a co-culture system of both ovarian cancer cells and endothelial cells to examine whether a functional association exists between claudin-5 and increased peritoneal permeability.
The results showed that the serum and ascites of preoperative ovarian cancer patients had increased levels of VEGF and that there was a VEGF-dependent decrease of claudin-5 in endothelial cells co-cultured with ovarian cancer cells. The ovarian cancer patients had a lower amount of claudin-5 detected in the peritoneal vessels compared to healthy controls. The results suggest that one mechanism by which VEGF may induce ascites formation in ovarian cancer patients is by increasing peritoneal permeability secondary to the downregulation of the TJ protein claudin-5 in the peritoneal endothelium \[[@b119-ijms-14-10412]\].
Interest in the expression of claudin-5 and its correlation with ovarian cancer behavior also arose. This was investigated in a Finnish study of 85 serous ovarian cancer tissue samples. There was an association between claudin-5 expression and cancer grade and stage. The highest claudin-5 expression was seen in patients with high grade and advanced staged disease. Cancer-specific and overall survival was also associated with claudin-5 expression. Only 25%--30% of claudin-5 positive patients were alive at 5 years follow-up compared to 60% of claudin-5 negative patients. This study therefore suggests that increased claudin-5 expression is associated with aggressive behavior in serous ovarian adenocarcinoma \[[@b120-ijms-14-10412]\].
3.4. Claudin-6
--------------
Unlike claudin-3 and -4, which are expressed in multiple epithelial tissues, the expression of claudin-6 is more restricted and believed to be predominately found in embryonic tissues and in undifferentiated pluripotent stem cells \[[@b41-ijms-14-10412],[@b121-ijms-14-10412]\]. In this regard, previous studies have reported that claudin-6 has an important role in the development of the mouse embryonic epithelium and endodermal tissues \[[@b122-ijms-14-10412],[@b123-ijms-14-10412]\]. However, claudin-6 expression has been reported in multiple human cancers such as rhabdoid tumors, breast cancers and gastric cancers \[[@b124-ijms-14-10412]--[@b126-ijms-14-10412]\]. Importantly, our group has recently found that claudin-6 can be expressed in ovarian cancer and may represent a novel functional receptor for CPE \[[@b127-ijms-14-10412]\]. Consistent with this view, UCI-101, an ovarian cancer cell line highly sensitive to CPE, does not express claudin-3/4 and knockdown of claudin-6 in these cells decreases CPE sensitivity. Moreover, different ovarian cell lines that are resistant to the effects of CPE can be made sensitive through claudin-6 overexpression. Finally, binding assays show that CPE can indeed bind claudin-6 in cells and that this binding is associated with CPE cytotoxicity. These results establish claudin-6 as a novel receptor for CPE and introduce the possibility of a novel therapeutic target for ovarian and other cancers that express claudin-6.
3.5. Claudin-7
--------------
Previous studies have shown that claudin-7 is up-regulated in endometriosis associated endometrioid ovarian cancer \[[@b84-ijms-14-10412]\] and also frequently upregulated in other epithelial ovarian cancers along with claudins-3 and -4 \[[@b18-ijms-14-10412],[@b85-ijms-14-10412],[@b128-ijms-14-10412]\]. High claudin-7 expression has been associated with a poor response to platinum-based chemotherapy in epithelial ovarian cancer \[[@b129-ijms-14-10412]\]. However, claudin-7 is downregulated in several other cancers including head and neck, esophageal and prostate cancer \[[@b130-ijms-14-10412]--[@b132-ijms-14-10412]\]. In breast cancer, claudin-7 expression was not only found to be decreased but also to be inversely correlated with tumor grade and metastatic disease \[[@b57-ijms-14-10412],[@b133-ijms-14-10412]\]. The exact reason for the differing pattern of expression in various cancers is largely unknown but is likely related to the specific role of this claudin in these malignancies. Dahiya *et al.* evaluated claudin-7 expression levels in 95 ovarian tissue samples and cell lines using western blotting, qRT-PCR analysis and IHC. The gene for claudin-7 was found to be upregulated in all tumor samples studied and small-interfering RNA-mediated knockdown of claudin-7 in ovarian cancer cells led to significant changes in the expression of other genes as determined by microarrays. Analyses of the genes differentially expressed revealed that the genes altered in response to claudin-7 knockdown were associated with pathways implicated in various molecular and cellular functions including cell cycle growth and proliferation, cell death and development. Claudin-7 expression was associated with a net increase in invasion but also a decrease in cellular migration. Claudin-7 was found to be universal upregulated in the most common epithelial ovarian cancer subtypes (serous, clear-cell, endometrioid and mucinous) at both the mRNA and protein levels. With the use of immunobloting and qRT-PCR, the authors demonstrated that mRNA levels and protein levels were not always correlated, suggesting post-translational regulation of claudin-7 in epithelial ovarian cancer. Overall this work shows that claudin-7 is significantly upregulated in epithelial ovarian cancer and may be functionally involved in ovarian carcinoma invasion, as such claudin-7 may also represent a potential marker for ovarian cancer detection and also a target for therapy \[[@b86-ijms-14-10412]\].
The prognostic significance of claudin-7 overexpression in epithelial ovarian cancer patients including sensitivity to platinum-based chemotherapy was investigated in another study. In this study claudin-7 was found to be expressed in 69/71 (97.1%) epithelial ovarian cancers but not in normal ovaries (*p \<* 0.001) and high claudin-7 expression in primary tumors correlated with shorter progression-free survival (PFS) of patients and poor sensitivity to platinum based chemotherapy. As such claudin-7 expression may also represent an independent prognostic factor for PFS and be important in regulating epithelial ovarian cancer response to platinum-based chemotherapy \[[@b129-ijms-14-10412]\].
4. Claudin-3 and -4 are Potential Targets for CPE-Based Theranostics
====================================================================
The identification of tumor origin, the prediction of chemotherapy response and the determination of prognosis is not without merit but of even greater importance is the potential for the use of claudin-isoform specific targeting agents in malignancies with increased claudin protein expression. *Clostridium Perfringens enterotoxin* (CPE) has already been demonstrated to induce necrosis in xenograft models of claudin-4 expressing tumors \[[@b50-ijms-14-10412],[@b52-ijms-14-10412],[@b97-ijms-14-10412]\]. There is however concern that the expression of claudin-4 on normal epithelia will limit the usefulness of this anti-tumor strategy. As ovarian carcinoma is largely a disease of the peritoneal cavity, the utilization of intra-peritoneal (i.p) treatment with full length CPE holds promise for this claudin-4 expressing tumor. Further strategies *in vivo* to limit CPE toxicity to normal tissues also expressing these proteins may include local delivery of the blocking CPE peptide fragment to gut and lung via enteral and inhalation routes respectively.
As surface proteins highly expressed in chemotherapy-resistant ovarian cancer, the claudins represent attractive therapeutic targets. Claudins-3 and -4 have been shown to represent the natural receptors for CPE and as such to be the main family members of the trans-membrane tissue-specific claudin proteins capable of mediating CPE binding and cytolysis \[[@b134-ijms-14-10412]\]. Several strategies involving the use of CPE as a novel therapeutic and possibly even diagnostic compound have been investigated with more work underway in this area. In fact, multiple research groups have reported using claudin-4 as not only a therapeutic target for toxin delivery but also as a target for fluorescent molecules to assist with the localization of ovarian and breast cancer cells \[[@b98-ijms-14-10412],[@b135-ijms-14-10412]--[@b138-ijms-14-10412]\]. The binding of the CPE toxin to cells results in the formation of membrane pore complexes and rapid cell death. The clinical role of
CPE-targeted therapy therefore holds promise in claudin-3 and -4 expressing malignancy and more so has potential for the treatment of chemotherapy resistant disease \[[@b57-ijms-14-10412],[@b134-ijms-14-10412],[@b139-ijms-14-10412]--[@b141-ijms-14-10412]\]. Supporting this view, the functional cytotoxicity of CPE in metastatic androgen-independent prostate cancer overexpressing claudin-3 has been reported previously \[[@b140-ijms-14-10412]\]. CPE is produced by the anaerobic gram-positive bacterium, *Clostridium perfringens* type A strain. This strain is known to cause food poisoning and is the second most commonly reported food-borne illness in the United States. CPE is a single polypeptide of 35 kDa composed of 319 amino acids \[[@b142-ijms-14-10412]\]. The carboxy (*C*)-terminus of CPE allows for the binding, while the *N*-terminus of CPE is associated with cytotoxicity \[[@b128-ijms-14-10412],[@b143-ijms-14-10412],[@b144-ijms-14-10412]\] ([Figure 1](#f1-ijms-14-10412){ref-type="fig"}). CPE triggers lysis of epithelial cells through interaction with the claudin-3 and claudin-4 receptors with resultant collapse of the cellular colloid-osmotic equilibrium and initiation of massive permeability changes leading to osmotic cell ballooning and lysis \[[@b134-ijms-14-10412],[@b142-ijms-14-10412]\]. Not surprisingly, mammalian cells that do not express either claudin-3 or claudin-4 fail to bind CPE and are not susceptible to CPE cytotoxicity \[[@b143-ijms-14-10412],[@b145-ijms-14-10412]\].
Although CPE is a recognized as a potential therapeutic agent, several new and promising agents cannot be utilized clinically due to undesirable pharmacokinetics and/or systemic toxicity. For a drug to be effective it must be able to cross the necessary tissue barriers in order to reach to its target without significant effect on normal tissues. As most TJ modulators previously were rendered less effective as a result of general low tissue specificity and side effects such as cell exfoliation due to epithelial cell barrier dysfunction, the *C*-terminal region of CPE emerged as a promising tool to modulate TJs in a tissue-specific and direct manner \[[@b146-ijms-14-10412],[@b147-ijms-14-10412]\]. The side effects are expected to be less due to a more specific modulation of an important component of the TJ \[[@b147-ijms-14-10412]\] and the activity is restricted to tissues that express the CPE- sensitive claudin-3 and -4. The *C*-terminal fragment of CPE (*C*-CPE peptide) has been shown to act to increase drug absorption through mucosal surfaces in a reversible and concentration-dependent manner. The *C*-CPE is also able to sensitize epithelial ovarian cancer cells to the cytotoxic effects of Taxol and Carboplatin at relatively low doses in a claudin-4 dependent manner. Also compared with single agent Taxol or Carboplatin, the addition of *C*-CPE to Taxol is able to significantly suppress large tumor burdens in animals via inhibiting tumor cell proliferation and accelerating apoptosis \[[@b148-ijms-14-10412]\]. The *C*-terminal fragment of CPE (*C*-CPE) has also been shown to effectively target TNFα to ovarian cancer cells \[[@b138-ijms-14-10412]\]. In ovarian cancer, pharmacologic studies have shown a therapeutic advantage to i.p drug therapy and the combination of C-CPE and cytotoxic chemotherapy both i.p may result in enhanced therapeutic effect with reduced systemic toxicity. The fact that ovarian cancer remains confined to the peritoneal cavity for much of its natural history suggests that i.p administration of CPE may provide improved therapeutic responses compared to similar intra-venous doses for those patients with recurrent ovarian cancer \[[@b149-ijms-14-10412]\].
As there is a continued need for innovative and effective strategies to treat recurrent/chemo-resistant ovarian cancer, our research group has provided *in vivo* models demonstrating that multiple i.p injections of sublethal doses of CPE every three days significantly inhibited tumor growth in 100% of mice harboring claudin-3 and -4 positive chemotherapy resistant ovarian tumor xenografts \[[@b97-ijms-14-10412]\]. One of our most recent research endeavors in this area was to describe the *in vitro* and *in vivo* bio-activity of the *C*-terminal fragment of CPE as a potential carrier for tumor imaging agents as well as a means of intracellular drug delivery for claudin-3 and -4 positive ovarian neoplasms after i.p injection ([Figure 2](#f2-ijms-14-10412){ref-type="fig"}) \[[@b135-ijms-14-10412]\]. In this study, claudin-3 and -4 expression was determined by qRT-PCR and flow cytometry in several primary ovarian carcinoma cell lines. Both claudin-3 and/or claudin-4 genes were found to be highly expressed in all primary ovarian carcinomas when compared to normal ovarian epithelial cells. The accuracy and specificity of the CPE peptide *in vitro* against primary chemo-resistant ovarian carcinoma cell lines was assessed with cell binding assays, while confocal microscopy and biodistribution assays were performed to evaluate the localization and uptake of FITC-conjugated CPE peptide in the established tumor tissue. Ultimately, this research demonstrated that using FITC-conjugated CPE peptide, there was specific *in vitro* and *in vivo* binding to multiple primary chemo-resistant ovarian carcinoma cell lines. The biodistribution studies in the mice revealed higher uptake of the peptide in tumor cells *versus* normal tissue. A time-dependent internalization of the FITC-conjugated CPE peptide was consistently seen by confocal microscopy in chemotherapy-resistant ovarian carcinoma cells. These findings suggest that CPE peptide is a good candidate as a lead peptide for tumor therapy or for the development of new diagnostic tracers with the possibility of demonstrating disease extent preoperatively or even intra-operatively using near-infrared fluorescent imaging \[[@b135-ijms-14-10412]\].
Further work in the area of chemo-resistant ovarian cancer has demonstrated that CD44+ ovarian cancer stem cells represent a small proportion of cancer cells capable of sustaining tumor growth and chemo-resistance and these cancer stem cells highly express genes encoding claudin-4. Casagrande *et al.* showed that small interfering RNA -mediated knockdown of claudin-3/-4 expression in CD44+ cancer stem cells significantly protected cancer stem cells from CPE-induced cytotoxicity. Here again multiple sublethal doses of i.p CPE proved to be an effective strategy for the eradication of claudin-4 expressing chemo-resistant ovarian cancer stem cells in mice harboring these xenografts with a 100% reduction in tumor burden in 50% of treated mice; *p \<* 0.0001 \[[@b95-ijms-14-10412]\]. These studies and others lend support to the efficacy of using recombinant CPE protein in a dose-dependent manner for treating claudin-3 and -4 tumor cells *in vitro* and *in vivo* \[[@b97-ijms-14-10412],[@b135-ijms-14-10412]\]. In general, the *in vivo* application of recombinant CPE did not induce toxin-associated side effects, however repeated administration regionally or loco-regionally was required in order to attain a therapeutic effect \[[@b89-ijms-14-10412],[@b150-ijms-14-10412],[@b151-ijms-14-10412]\]. As progress continues in the molecular understanding of the CPE-claudin interactions, this may potentially lead to the development of enhanced recombinant CPE proteins.
Another novel approach of targeting claudin-3 and -4 expressing ovarian tumor cells is through gene therapy. Intra-tumoral gene transfer of CPE-expressing vectors can be employed for selective suicide gene therapy of claudin-3 and -4 positive tumors and was found to effect a more rapid and effective tumor cell killing *in vitro* and *in vivo* \[[@b152-ijms-14-10412]\]. Cytotoxicity of up to 100% was observed 72 h after gene transfer and was restricted to claudin-3 and -4 expressing tumor lines. Additionally the *in vivo* data from this study revealed significant inhibition of ovarian cancer xenograft growth in SCID mice \[[@b152-ijms-14-10412]\].
5. Conclusions
==============
Claudin-3, -4 and -7 are highly expressed in ovarian cancer. While the understanding of the exact role of these proteins in ovarian as well as other human tumors remains poorly defined, substantial experimental evidence has demonstrated an important role for claudin-3 and -4 in ovarian cancer cell invasion and dissemination, resistance to chemotherapy and as target of CPE treatment. In multiple preclinical *in vitro* and *in vivo* models recombinant CPE has been shown to induce a dose-dependent eradication of claudin-3 and -4 tumor cells while the carboxy-terminal fragment of CPE (*i.e.*, CPE~290--319~ binding peptide) has demonstrated promise as a carrier for tumor imaging agents and intracellular delivery of therapeutic drugs \[[@b41-ijms-14-10412],[@b151-ijms-14-10412]\]. The future design and implementation of phase 1 clinical trials in chemo-resistant and recurrent solid tumors will ultimately determine the feasibility and validity of these novel CPE-based theranostic approaches.
The authors declare no conflict of interest.
{#f1-ijms-14-10412}
{#f2-ijms-14-10412}
######
Claudin-1, -3, -4, -5 and -7 expression and function in normal tissues.
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Claudin Function Tissue specificity Involvement in disease
----------- --------------------------------------------------------------------------------------------------------------------------------------------------------------- ---------------------------------------------------------------------------------------------------------------------- ----------------------------------------------------------------------------
Claudin-1 TJ-specific obliteration of the intercellular space through Ca^2+^-independent cell-adhesion activity. Acts as a co-receptor for HCV entry into hepatic cells Strongly expressed in liver and kidney. Also expressed in heart, brain, spleen, lung and testis Ichthyosis \[[@b61-ijms-14-10412]\]
Claudin-3 TJ-specific obliteration of the intercellular space through Ca^2+^-independent cell-adhesion activity (CPE is the natural ligand) Strongly expressed in ovary, lung, pancreas, salivary gland, kidney, adrenal, small intestine, colon and thyroid Williams-Beuren syndrome \[[@b62-ijms-14-10412]\]
Claudin-4 TJ-specific obliteration of the intercellular space (CPE is the natural ligand) Strongly expressed in ovary, lung, pancreas, salivary gland, kidney, adrenal, small intestine, colon and thyroid Williams-Beuren syndrome \[[@b62-ijms-14-10412]\]
Claudin-5 Target molecule of hypoxia Strongly expressed in vascular endothelial cells. Transiently expressed during development of RPE. Expressed in lung Velocardiofacial syndrome \[[@b63-ijms-14-10412]\]
Claudin-7 TJ-specific obliteration of the intercellular space. Co-localizes with EPCAM at the lateral cell membrane and TJ Strongly expressed in kidney, GI tract, thyroid, adrenal gland and lung. Also expressed in prostate tissue Related to ability of breast cancer cells to disseminate.\
Downregulation correlates with histological grade \[[@b57-ijms-14-10412]\]
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
######
Claudin expression in gynecologic cancer.
Tumor type Claudin gene Expression compared to normal tissues References
------------- -------------- --------------------------------------------------------------------- -----------------------------------------------------------------------------------------
Ovarian CLDN3 High \[[@b16-ijms-14-10412],[@b18-ijms-14-10412],[@b54-ijms-14-10412],[@b56-ijms-14-10412]\]
CLDN4 High \[[@b16-ijms-14-10412],[@b18-ijms-14-10412],[@b43-ijms-14-10412]\]
CLDN7 High \[[@b18-ijms-14-10412],[@b84-ijms-14-10412]--[@b86-ijms-14-10412]\]
CLDN16 High \[[@b17-ijms-14-10412]\]
Endometrial CLDN2 High \[[@b87-ijms-14-10412]\]
CLDN3 High \[[@b87-ijms-14-10412]--[@b89-ijms-14-10412]\]
CLDN4 High \[[@b53-ijms-14-10412],[@b88-ijms-14-10412]\]
Cervical CLDN1 High \[[@b90-ijms-14-10412],[@b91-ijms-14-10412]\]
CLDN2 High \[[@b91-ijms-14-10412]\]
CLDN4 High \[[@b91-ijms-14-10412],[@b92-ijms-14-10412]\]
CLDN7 High \[[@b90-ijms-14-10412],[@b92-ijms-14-10412]\]
| {
"pile_set_name": "PubMed Central"
} |
Background
==========
Apc loss causes progenitor expansion in development and disease
---------------------------------------------------------------
The Wnt/ß-catenin signaling pathway acts to maintain the undifferentiated progenitor state in multiple epithelial tissues, and overactivation of this pathway is a major contributor to cancer. The tumor suppressor APC normally functions to inhibit Wnt/ß-catenin signaling, and *APC*mutations are oncogenic in tissues such as the colorectal epithelium \[[@B1]\]. During normal embryonic development, Wnt and APC activities are balanced to allow both progenitor cell expansion and differentiation of postmitotic derivatives. Zebrafish embryos homozygous for *apc*mutations exhibit mispatterning and failure of differentiation in multiple tissues including the central nervous system (CNS) \[[@B2],[@B3]\]. Furthermore, in the CNS of other vertebrates, loss of APC function specifically leads to arrest in the neural progenitor state \[[@B4]\]. Despite a clear picture of the cellular phenotypes following loss of APC, the molecular pathways underlying CNS progenitor cell expansion are largely unknown. These pathways may represent good candidates for mediators of oncogenesis in other epithelial cells.
Transcriptional targets of Wnt signaling mediate *APC*mutant phenotypes
-----------------------------------------------------------------------
The main downstream output of Wnt/ß-catenin signaling is the transcriptional regulation of target genes, mediated by Lef/Tcf family members. Typically, these targets are repressed by Lef/Tcf factors in the absence of Wnt signaling, and following Wnt activation ß-catenin translocates to the nucleus where it binds to Lef/Tcf proteins and acts as a co-activator. The identification of Wnt/ß-catenin transcriptional targets has thus been a major focus of investigation in past studies of the pathway\'s role in development and disease. Some identified target genes have been shown to be common targets in both normal embryos and the oncogenic state. For example, *mitf*is a direct target of Lef1 during melanocyte specification, and also plays an important role in melanoma progression downstream of Wnt pathway hyperactivation \[[@B5],[@B6]\]. Similarly, Wnt targets such as *ascl2*and *lgr5*may function in both intestinal epithelium homeostasis as well as colon cancer \[[@B7],[@B8]\].
Stat3 functions synergistically with Wnt signaling in cancer
------------------------------------------------------------
Like Wnt signaling, the Jak/Stat pathway has been shown to mediate proliferation and tumor growth in cancer. In particular, constitutive Stat3 activity is associated with malignancy in colon cancer \[[@B9]\], the primary carcinoma caused by *APC*mutations. A previous study showed that Wnt signaling can stimulate Stat3 activity during early zebrafish development \[[@B10]\], but the mechanism underlying this activation was not characterized. One potential mechanism of regulation has been suggested by a study in esophageal carcinoma, where *Stat3*was shown to be a transcriptional target of ß-catenin via Tcf4 \[[@B11]\]. Intriguingly, *Stat3*has also been suggested to be a target of Wnt signaling in ES cells \[[@B12]\], suggesting that this pathway may represent a developmentally important mechanism. However, the regulatory relationship between Wnt signaling and *Stat3*activation has not been explored *in vivo*in untransformed tissue.
Here we demonstrate that *stat3*is a direct transcriptional target of Wnt/ß-catenin signaling in developing zebrafish embryos. We show that increased *stat3*expression in *apc*mutants correlates with increased proliferation and failure of neuronal differentiation in the developing hypothalamus. Conditional inhibition of Jak/Stat signaling rescues proliferation defects as well as ectopic expression of progenitor markers, but not the general activation of Wnt targets or the complete process of neurogenesis. Together, these data indicate a specific function for Jak/Stat activation in mediating neural progenitor expansion downstream of APC mutations, and suggest a conserved role for this pathway in development and disease.
Results and Discussion
======================
*stat3*is a direct target of the Wnt pathway via Lef1
-----------------------------------------------------
We have previously shown that Wnt signaling, mediated by the transcriptional effector Lef1, is required for hypothalamic neurogenesis in the zebrafish brain \[[@B13]\]. To identify transcriptional targets of the Wnt pathway, we performed ChIP-seq analysis using a Lef1 antibody. Immunoprecipitation was performed using chromatin from whole 36 hours post-fertilization (hpf) embryos, corresponding with a time of high *lef1*expression in the hypothalamus. After deep sequencing of precipitated chromatin, we observed high enrichment of the *stat3*promoter region compared to total input as well as chromatin from *lef1*deletion mutant embryos. The genomic sequence identified by ChIP-seq (Figure [1A](#F1){ref-type="fig"}) contains several putative Lef/Tcf consensus binding sites (Figure [1B](#F1){ref-type="fig"}), and we confirmed the direct interaction with Lef1 using ChIP followed by quantitative PCR (Figure [1C](#F1){ref-type="fig"}).
{#F1}
We next tested whether the endogenous expression of *stat3*in the zebrafish embryo depends on Wnt-mediated transcription. We used a transgenic inducible repressor of Lef/Tcf target genes (*hs:ΔTcf*) to globally inhibit pathway activity in vivo. 28 hpf embryos were heat shocked for one hour, allowed to recover until 36 hpf, and then processed for in situ hybridization. We observed a qualitative decrease in *stat3*expression throughout embryos expressing ΔTcf, including in the hypothalamus (Figure [1D,E](#F1){ref-type="fig"}). Together, these results suggest that *stat3*is a direct transcriptional target of the Wnt pathway.
*stat3*expression and Stat3 phosphorylation are increased in *apc*mutants
-------------------------------------------------------------------------
Previous studies have reported multiple developmental defects in the CNS of *apc*mutant zebrafish embryos, including axon pathfinding errors \[[@B14]\], loss of normal brain patterning \[[@B3]\], and expansion of the putative retinal stem cell zone \[[@B2]\]. An additional striking phenotype that we observed in mutant embryos was a dramatic increase in proliferating cells particularly in the hypothalamus, accompanied by a dramatic decrease in differentiated neurons (Figure [2A](#F2){ref-type="fig"}). An earlier study identified *stat3*as a marker that was increased in *apc*mutant embryos in the putative retinal stem cell zone and the hypothalamus \[[@B2]\]. We examined *stat3*expression throughout the *apc*mutant embryo and observed a qualitative increase in mRNA levels, with specific enrichment in known CNS progenitor zones including the hypothalamus (Figure [2B](#F2){ref-type="fig"}). Quantitative PCR analysis of *apc*mutant embryos showed an increase in the level of *stat3*mRNA of 5.34 ± .09 fold (s.d., n = 3, p \< 0.05 by student\'s t-test) compared to wild-type siblings. We also found a qualitative increase in pStat3 immunostaining in the *apc*mutant hypothalamus compared to control embryos (Figure [2B](#F2){ref-type="fig"}), suggesting that *stat3*mRNA levels may normally limit the signaling output of this pathway. Based on the known roles of Stat3 function in progenitor cell maintenance, these results raised the possibility that increased Jak/Stat signaling might underlie some of the progenitor differentiation defects present in the *apc*mutant brain.
{#F2}
Increased proliferation in *apc*mutants can be rescued by blocking Jak/Stat signaling
-------------------------------------------------------------------------------------
In other tissues, *APC*mutations and Stat3 hyperactivation can both lead to increased cell proliferation. To quantify the proliferative increase in *apc*mutant zebrafish, we performed short-pulse (1 hour) BrdU labeling in wild-type and mutant embryos. At 36 hpf, significantly more cells within the developing hypothalamus of *apc*mutant embryos incorporated BrdU than in wild-type siblings (Figure [2C,D](#F2){ref-type="fig"}). These data are consistent with an increased number of progenitor cells in the CNS of *apc*mutants compared to wild-type embryos.
We next tested whether inhibition of Jak/Stat activity could reverse the increased proliferation found in *apc*mutants. To block Jak/Stat signaling, we used the Jak2 inhibitor AG-490, which has been demonstrated to prevent Stat3 phosphorylation in many other experimental systems including zebrafish \[[@B15]\] and allowed us to bypass early developmental defects resulting from *stat3*knockdown. When wild-type embryos were incubated in 40µm AG-490 from 24-36 hpf, we did not observe a significant change in the BrdU labeling index compared to untreated controls (Figure [2C,D](#F2){ref-type="fig"}). In contrast, AG-490 incubation completely reversed the increase in proliferation observed in *apc*mutant embryos, restoring the BrdU labeling index to wild-type levels (Figure [2C,D](#F2){ref-type="fig"}). Together, these data indicate that Jak/Stat signaling is required for increased proliferation in *apc*mutant brains. Our observations of increased *stat3*mRNA expression in *apc*mutants suggest that Stat3 levels may be limiting in the developing brain, and that regulation by the Wnt pathway may control the ability of Jak/Stat signaling to drive cell proliferation.
Increased progenitor marker expression in *apc*mutants requires Jak/Stat activity
---------------------------------------------------------------------------------
Because proliferation is closely linked to the progenitor cell phenotype in the developing CNS, we wanted to determine whether other markers of neural progenitors were also increased in *apc*mutants and whether this increase depends on Jak/Stat activity. We first examined the expression of *ascl1b*, which encodes a proneural bHLH transcription factor essential for neurogenesis. Using in situ hybridization, we found that *ascl1b*mRNA levels were qualitatively increased in the *apc*mutant hypothalamus at 36 hpf (Figure [3A](#F3){ref-type="fig"}). Incubation in 40µM AG-490 from 24-36 hpf was able to eliminate this increase and restore *ascl1b*expression to wild-type levels in *apc*mutants (Figure [3A](#F3){ref-type="fig"}), suggesting that increased proneural gene expression is mediated by Jak/Stat activity.
{#F3}
In the zebrafish retina, *otx1*expression marks the putative stem cell zone of the ciliary margin, and is expanded in *apc*mutants \[[@B2]\]. Otx1 and Otx2 are also expressed in the developing vertebrate hypothalamus and label neural progenitors in the zebrafish hypothalamus. We observed increased *otx1*mRNA expression in the hypothalamus of *apc*mutants (not shown), and to provide a more quantitative measurement, we examined the number of cells labeled with an antibody that recognizes both Otx1 and Otx2. Within the hypothalamus, *apc*mutants showed a significant increase in Otx1/2-positive cells at 36 hpf (Figure [3B,C](#F3){ref-type="fig"}), and this increase was rescued to wild-type levels by AG-490 incubation (Figure [3B,C](#F3){ref-type="fig"}). These data suggest that cells may be arrested in an Otx-positive progenitor state following *apc*inactivation, and that Jak/Stat function mediates this arrest.
Inhibition of Jak/Stat activity is not sufficient to rescue neurogenesis in *apc*mutants
----------------------------------------------------------------------------------------
While Jak/Stat activity is required for the expansion of CNS progenitor characteristics downstream of *apc*inactivation and *stat3*transcription, we hypothesized that this pathway is not likely to mediate all outputs of Wnt activation. Indeed, when we examined the expression of the Wnt target gene *axin2*, we observed a strong increase in mRNA expression that was not rescued by AG-490 incubation (Figure [4A](#F4){ref-type="fig"}). This result indicates that many transcriptional targets of Wnt/ß-catenin signaling are likely to be independent of Jak/Stat activity, and that these targets may act in parallel pathways. Furthermore, while AG-490 incubation could rescue increases in proliferation and progenitor gene expression, it was insufficient to restore neurogenesis in *apc*mutants. The loss of HuC/D expression observed in the hypothalamus was still seen in embryos after incubation in AG-490 (Figure [4B](#F4){ref-type="fig"}), suggesting that neural progenitors were still unable to differentiate into neurons. Therefore, other Stat3-independent targets of APC must be important for regulating the full program of differentiation. These could possibly include Wnt-independent APC targets, as has been demonstrated previously in other studies \[[@B16]\].
{#F4}
Conclusions
===========
Here we have shown that *stat3*is a direct transcriptional target of Wnt signaling in the developing embryo, and that Jak/Stat signaling mediates the expansion and maintenance of CNS progenitor characteristics downstream of Wnt hyperactivation in *apc*mutants. Together, our data suggest that transcriptional regulation of *stat3*may represent a general mechanism linking Wnt pathway overactivation to the expansion of undifferentiated cells in the disease state.
At higher doses of AG-490, we were able to completely eliminate both proliferation and progenitor marker expression in wild-type embryos (not shown). Combined with the endogenous expression pattern of *stat3*, and the fact that ΔTcf can repress *stat3*in wild-type embryos, this suggests that a Wnt/Stat3 pathway may also play an important role in normal CNS development.
Methods
=======
Zebrafish maintenance and embryo culture
----------------------------------------
Embryos were obtained from natural spawning of wild-type (AB\*), *Tg(hsp70l:tcf3-GFP)^w26^*, *Df(LG01:lef1,msxb)^x8^*, and *apc^hu745^*mutant zebrafish and were staged according to Kimmel et al., \[[@B17]\]. *lef1*deletion and *apc*mutant embryos were identified by morphology and *hs:Δtcf*embryos were identified by expression of a GFP fusion protein. All embryos were raised at 28.5°C and fixed in 4% PFA for analysis. 28 hpf *hs:Δtcf*embryos were heat shocked for 1 hour at 37°C, then allowed to recover at 28.5°C until 36 hpf. To block Jak/Stat signaling, embryos were treated with 40 uM AG-490 (Enzo) beginning at 24 hpf. For BrdU labeling, 35 hpf embryos were incubated in 10 mM BrdU in 15% DMSO for 30 minutes on ice, washed and allowed to recover for 1 hour at 28.5°C before fixation.
ChIP and qPCR
-------------
ChIP analysis was performed as described previously \[[@B18]\] with the following modifications. One hundred embryos at 36 hpf were dechorionated and fixed in 1% PFA in PBS for 15 minutes at room temperature, and then lysed in cell lysis buffer \[10 mM Tris (pH 8.1), 10 mM NaCl, 0.5% NP- 40, and protease inhibitors\] and nuclear lysis buffer \[50 mM Tris-Cl (pH 8.1), 10 mM EDTA, 1% SDS and proteinase inhibitors\] by pipetting. For each immunoprecipitation, 5 ug of anti-Lef1 antibody \[[@B13]\] was conjugated to 30 ul Dynabeads (Invitrogen) prior to applying nuclear extract. A detailed protocol is posted at: <https://wiki.zfin.org/display/prot/ZFIN+Protocol+Wiki>. Precipitated DNA fragments were purified and submitted for Illumina sequencing at the University of Utah HSC Core Facility and sequences were mapped to zebrafish genome (assembly zv7).
For qPCR analysis of ChIP fragments, total input chromatin and Lef1 immunoprecipitated chromatin from wild-type and *Df(LG01:lef1,msxb)^x8^*mutant siblings was used. For qPCR analysis of *stat3*mRNA levels, total RNA was isolated from 42 hpf wild-type and *apc^hu745^*mutants using an RNAeasy extraction kit (Qiagen) followed by DNase treatment. cDNA was synthesized by SuperScript II reverse transcriptase (Invitrogen), and *stat3*levels were normalized to *beta actin*cDNA. Quantitative real-time PCR was performed at the University of Utah HSC Core Facility.
Primers used for *stat3*ChIP qPCR are: **5\'-TGCGTATCACAACACGGTTT-3\' 5\'-ACATGTCTCTGACGCAGTCG-3\'**Primers used for *stat3*cDNA qPCR are: **5\'-CCGACTGGAAGAGGAGACAG-3\' 5\'-GCTGGACGGTGCTGAATAAT-3\'**
In situ hybridization
---------------------
Whole mount in situ hybridization was performed as described previously \[[@B13]\]. Probes for *stat3*\[[@B19]\] and *otx1*\[[@B20]\] were obtained from T. Piotrowski. Probes for *ascl1b*and *axin2*were synthesized in our laboratory. Following staining, whole embryos were mounted in 80% glycerol and imaged on a dissecting microscope, or embedded in plastic, sectioned, and imaged on a compound microscope.
Immunohistochemistry
--------------------
For BrdU and PCNA detection, fixed embryos were incubated for 1 hour in 2N HCl. Immunostaining was performed as described previously \[[@B21]\]. Antibodies were obtained from the following sources: anti-BrdU (AbD Serotec, 1:500), anti-HuC/D (Molecular Probes, 1:500), anti-OTX1/2 (Chemicon, 1:500), anti-PCNA (Sigma, 1:1000), anti-pStat3 (Tyr708, MBL, 1:1000), and secondary antibodies conjugated to Alexa Fluor 647 (Invitrogen). Following immunohistochemistry, embryos were counterstained with TO-PRO-3 (Invitrogen), and whole brains were dissected for imaging. Embryos were mounted in Fluoromount-G (Southern Biotech), and confocal images were acquired using an Olympus FV1000 microscope.
Authors\' contributions
=======================
J.L. conducted all experiments except the PCNA analysis of *apc*mutants, qPCR for *stat3*, and pStat3 staining, which were performed by X.W. R.I.D. provided oversight for the entire study and wrote the manuscript. All authors read and approved the final manuscript.
Acknowledgements and Funding
============================
Grant Sponsor: NIH (NINDS); R21NS055138
| {
"pile_set_name": "PubMed Central"
} |
Findings
========
NPC is a distinct type of head and neck cancer which is consistently associated with Epstein-Barr virus (EBV). Detection of clonal EBV genome in both precancerous lesions and invasive cancers indicates that EBV latent infection is an early event in the tumorigenesis of NPC. Since we established the EBV-positive NPC cell line C666-1 and reported it about fifteen years ago, it has been widely used for investigating host-viral interaction, elucidating the function and transcriptional regulation of EBV-encoded latent genes and miRNAs, and developing EBV targeting therapeutic strategies \[[@B1]\]. The origin of this cell line was from an undifferentiated NPC biopsy of a Hong Kong patient \[[@B1]\]. It contains normal episomal EBV genome and shows latency II EBV gene expression pattern. A number of studies demonstrated the distinct NF-κb, STAT3, AKT and NOTCH pathways in this cell line as well as the in vivo samples including EBV-positive NPC xenografts (e.g., C15, C17, xeno-2117) and primary tumors \[[@B2]\]. Recently, two novel EBV-encoded microRNAs, miR-BART21 and miR-BART22 have been discovered from this EBV-positive epithelial cell line \[[@B3]\].
Despite C666-1 being the only in vitro native EBV-infected NPC model worldwide, the EBV genome in this cell line has not been fully characterized until now. To facilitate the EBV-related studies using this unique cell line, we constructed the EBV genome map through bioinformatic analysis and experimental validation of our recent whole-genome deep sequencing results (Additional file [1](#S1){ref-type="supplementary-material"} Supplementary methodology). By 100-base pair-end genomic sequencing on Illumina HiSeq 2000 genome sequencer, the C666-1 genome was sequenced with average \>75-fold coverage as described \[[@B4]\]. A total of 2,511,210,660 reads (251 Gb) were collected from the sample. By using an approach that combines the results of two alignment strategies, namely aligning the reads to both human and EBV reference genomes (EBV-WT; GeneBank accession number AJ507799) at the same time, and aligning them first to the human genome and then the remaining reads to the EBV reference genome, we extracted a total of 857,595 kb EBV sequences from the collected C666-1 data. A high coverage value of 504 folds to EBV genome was yielded. All uniquely mapped EBV sequences were assembled into a 143,734 bpconsensus sequence with a read depth of at least 10 reads. We validated the poorly aligned and questionable regions and filled up the gaps by PCR amplification and conventional Sanger DNA sequencing. The regions failed to be assembled (e.g. with highly repetitive sequences) are represented by tracts of Ns as described previously \[[@B5]\]. A 171,317 bp complete EBV genome of C666-1 was constructed (Figure [1](#F1){ref-type="fig"}a). This newly assembled C666-1 EBV sequence was submitted to GenBank with accession number KC617875. The study was approved by the University Animal Experimentation Ethics Committee (AEEC) (13-036-MIS) of the Chinese University of Hong Kong.
{#F1}
In this study, we have assembled the EBV genome in C666-1 using high-coverage genome sequencing data. Since no PCR amplification was involved, both homogenous and heterogeneous genome variations are accurately determined. Comparing with the EBV-WT reference genomic sequence (AJ507799), we have revealed a total of 1,268 homogenous and 87 heterogeneous sequence variations. These changes include 127 indels and 1,228 SNVs. Among the SNVs, 907 are located within the coding regions and 41.3% (386/907) of them are nonsynonymous (Figure [1](#F1){ref-type="fig"}b). The sequence variations in selected SNVs were confirmed by Sanger DNA sequencing. Phylogenetic analysis of whole EBV genomes in C666-1 and the reported strains (EBV-WT, AG876, GD1, GD2, and HKNPC1) showed that C666-1 is closely related to the GD2 and HKNPC1 strains (Figure [1](#F1){ref-type="fig"}c) \[[@B5],[@B6]\]. It has great divergence with the AG876 and reference EBV-WT genome. Similar results were observed when we compared the protein sequences of various EBV lytic (BZLF1, BLLF1) and latent (EBNA1, LMP1, LMP2) genes (Figure [2](#F2){ref-type="fig"}). A number of studies have also shown that BZLF1 and LMP1 sequences of the isolates from Hong Kong NPC patients are distinct from that of the EBV-infected lymphoid cells derived in Africa or Western countries \[[@B7]\]--\[[@B9]\]. The findings imply that C666-1 might serve as an important model for studying the roles or functions of viral proteins in NPC tumorigenesis. Among the four EBV strains from South China, the isolate from NPC patient's saliva (GD1) shows the greatest divergence with those from the tumors (C666-1, GD2, HKNPC1). This finding suggests the presence of tumor-associated EBV strain(s) in NPC patients. Nevertheless, a comprehensive sequencing of EBV isolates from saliva, peripheral blood and tumor specimens in a panel of NPC patients may prove this hypothesis. A summary of non-synonymous SNVs in the majority of EBV-encoded lytic and latent genes of C666-1 strain versus those of GD2 and HKNPC1 is shown in Additional file [2](#S2){ref-type="supplementary-material"}: Table S1. In the latent genes including EBNA1, EBNA3B/3C, LMP1 and LMP2B genes, high frequencies of C666-1 specific non-synonymous SNVs were observed. The prevalence and function of these SNVs in NPC need further elucidations. Previously, we have demonstrated that multiple EBV-encoded BART miRNAs (miR-BART1-5p, miR-BART16 and miR-BART17-5p) target the 3′UTR of the LMP1 gene \[[@B10]\]. The predicted target sequences of these 3 EBV-encoded BART miRNAs in the 3′UTR of the LMP1 gene are highly conserved in the NPC-derived EBV strains. In this study, we also found no polymorphism in the predicted target sequences of the miR-BART1-5p, 16, and 17-5p in the C666-1 EBV strain.
{#F2}
Apart from the missense mutations, a homogenous nonsense mutation in the lytic gene, BNRF1, which encodes an EBV major tegument protein was found. We confirmed the mutation in C666-1 by PCR amplification and Sanger Sequencing (Figure [1](#F1){ref-type="fig"}d). This finding indicates the deficiency of BNRF1 protein expression in this in vitro EBV-positive NPC models. Notably, it was reported that EBV with BNRF1 deletion also showed efficient lytic replication and production of mature viral particles. There are no major structural alterations in the BNRF1-deleted virus \[[@B11]\]. Further elucidation of the virus production and lytic cycle of this BNRF1-deficient C666-1 strain is needed. On the other hand, a recent study has reported that BNRF1 activates viral early gene BZLF1 transcription via disrupting cellular DAXX-ATRX in 293 cells. Thus, BNRF1 deficiency may help to maintain the latent EBV genome in NPC cells \[[@B12]\]. On the other hand, loss of BNRF1 in the C666-1 strain may impact the escape from the host immune responses in the NPC patients since BNRF1 is a defined target of the EBV-specific T-helper-cell response.
In summary, we delineated the whole EBV genome sequence in C666-1, which might serve as an important resource for NPC studies. The phylogenetic analysis indicates the C666-1 strain as a representative strain for EBV-associated NPC.
Abbreviations
=============
EBV: Epstein-Barr virus; NPC: Nasopharyngeal carcinoma; Indels: Insertions/deletions; SNVs: Single-nucleotide variations; CDS: Coding sequence.
Competing interests
===================
The authors declare that they have no competing interests.
Authors' contributions
======================
KWL and KYLY designed the study; KWL, KYLY, and KFT drafted the manuscript; KKYT, KYLY and SDL participated in the bioinformatics analysis and sequence alignment; CKYM, GTYC, STC carried out the molecular genetic studies. All authors read and approved the final manuscript.
Authors' information
====================
Ken Kai-Yuen Tso and Kevin Yuk-Lap Yip are co-first authors.
Supplementary Material
======================
###### Additional file 1
Supplementary methodology.
######
Click here for file
###### Additional file 2: Table S1
Non-synonymous mutations and amino acid changes commonly found in NPC tumor samples (C666-1, HKNPC1 and GD2).
######
Click here for file
Acknowledgements
================
The research was supported by Focused Investments Scheme-A from the Chinese University of Hong Kong, and Hong Kong Research Grant Council -- GRF (471610, 471211), CRF (CUHK8/CRF/11R), Theme-Based Research Scheme (T12-403/11 and T12-401/13-R) and AoE NPC (AoE/M-06/08).
| {
"pile_set_name": "PubMed Central"
} |
Introduction {#S1}
============
Psychosis is a complex brain disorder that typically emerges in late adolescence or early adulthood ^[@R1]^, enormously impacts functioning, and results in large costs to public health ^[@R2]^. There are significant brain abnormalities in psychosis, including progressive brain tissue loss ^[@R3]^, reduced neuropil in the cortex ^[@R4]^, altered dopaminergic function ^[@R5]^, abnormal glutamate functioning ^[@R6]^, and alterations in brain white matter organization ^[@R7],\ [@R8]^. This constellation of dysfunction may exist from birth, which allows for the study of specific biological targets associated with psychosis; many of which may abnormally progress during the course of adolescent neuromaturation ^[@R9]^. Recent advances in neuroimaging, including the use of diffusion weighted imaging (DWI), has led to a better understanding of effects of disrupted white matter organization, and has the potential to allow for both earlier identification of illness and targeted treatments.
Early intervention requires valid and reliable methods of identifying youth at highest risk for developing psychosis. Youths with sub-psychotic symptoms often exhibit subtle neurobiological abnormalities like those found in psychosis ^[@R10],\ [@R11]^. Hence, many contemporary investigations focus on individuals at clinical high risk (CHR) for developing psychosis, or at-risk mental states (ARMS), in both help-seeking ^[@R12]^, and community-based samples ^[@R13],\ [@R14]^. When individuals at risk are followed over time about 30% develop a psychotic disorder within two years ^[@R15]^, most commonly schizophrenia (SZ) ^[@R15]^. Many of the deficits in cortical microcircuitry in psychosis are found to some extent in CHR, including disrupted neurotransmission ^[@R16]^, lower gray matter volume ^[@R17]^, and disruptions of white matter microstructure organization ^[@R18]^. These anomalies may reflect aberrant brain development, and quantification of these features has the potential to enhance our ability to detect those truly at risk for developing psychosis.
Diffusion weighted imaging (DWI) has facilitated the *in vivo* study of brain white matter microstructural organization. DWI measures the diffusion of water molecules through brain tissue, a process that is affected by microstructural properties of the local surrounding tissue ^[@R19],\ [@R20]^. Disrupted white matter microstructure, as indicated by low fractional anisotropy (FA) and high diffusivity, is documented in numerous brain regions in chronic SZ ^[@R7]^, while more recent investigations of early-onset psychosis ^[@R21]^ and psychosis in adolescence ^[@R22]^ report focal WM abnormalities. Typically, these findings are limited to major WM fiber tracts including the cingulum bundle, corpus callosum, and anterior limb of the internal capsule ^[@R23]^. Given recent convergent evidence that psychosis is a product of abnormal neurodevelopment, it is not surprising that WM microstructure is also aberrant in CHR individuals^[@R24],\ [@R25]^.
Yet, longitudinal diffusion imaging studies of brain white matter in early psychosis and youths with psychosis spectrum symptoms are scant ^[@R26],\ [@R27]^. In fact, there are only two longitudinal studies in early onset psychosis ^[@R28],\ [@R29]^ and two in youth at clinical high-risk for psychosis ^[@R30],\ [@R31]^. Early onset psychosis^[@R28]^ is associated with increasing FA over 2.5 years, which was not found in healthy comparison subjects, suggesting delayed maturation. Another study^[@R29]^ found early onset psychosis patients had lower FA in comparison to healthy comparison subjects, but there was no evidence that this difference changed over time. Individuals at risk for developing psychosis^[@R30]^ show associations between lower FA and negative symptoms, but DWI data from this 1.5 Tesla MRI study was limited to only the corpus callosum. Another study in at-risk youth, found association between increased FA in the superior longitudinal fasciculus and age, again suggesting delayed maturation, although the follow-up interval was only 12 months^[@R31]^. Thus, more longitudinal diffusion imaging studies are needed to directly test current models of the development and lifetime course of brain structure associated with psychosis risk.
Here, we measure white matter microstructure, longitudinally, in a large community sample of youths with and without subthreshold psychosis risk symptoms. Clinical symptoms and DWI were measured at two time points over approximately 20 months in individuals with persistent psychosis symptoms and typically developing (TD) youths. To our knowledge, there are no previous studies of this size in community youth with persistent symptoms that have been followed over time and where developmental white matter abnormalities were measured. Evidence of white matter disruption in this sample would provide convergent support for the examination of WM microstructure as a phenotype for individuals on the psychosis spectrum.
Our hypotheses were as follows: 1) age-related increases in FA and age-related decreases in diffusivity; 2) PS individuals with persistent symptoms would show abnormalities (e.g. lower FA; higher diffusivity) in white matter microstructure in regions impacted in adults with psychosis; 3) white matter abnormalities will progressively worsen with time in PS youth; and 4) alterations in diffusion metrics will be associated with clinical symptoms and cognitive performance.
METHODS & MATERIALS {#S2}
===================
All participants included in this study were initially enrolled in the Philadelphia Neurodevelopmental Cohort ^[@R13],\ [@R14],\ [@R32],\ [@R33]^ and provided informed consent or, for minors, informed assent plus parental consent. All procedures were approved by the Institutional Review Boards of the University of Pennsylvania and Children's Hospital of Philadelphia.
Participant Recruitment {#S3}
-----------------------
Briefly, participants were recruited by the Center for Applied Genomics at CHOP through a pediatric healthcare network of over 30 clinical community sites in the tristate area of Pennsylvania, New Jersey, and Delaware. Initial recruitment occurred between 2006 and 2012, which is described in full in a previous report^[@R13]^. When undergoing blood work, patients were approached for participation in the recruitment pool. The percentage of patients undergoing blood work across recruitment sites varied from 11% to 53%, with a mean of 36%. Participants provided a blood sample for genomic studies and access to Electronic Medical Records (EMRs). The EMR of each participant was screened for preliminary eligibility for PNC participation. Potential participants were included if they were between the ages of 8--21, had provided written informed consent/assent to be re‐contacted for future studies, were proficient in English, and did not appear to have significant developmental delays or physical conditions that would interfere with their ability to complete study procedures. Of the initial recruitment pool, 9,498 completed clinical and cognitive assessment. A random subsample (n=1601), stratified by age and gender, were enrolled in neuroimaging^[@R34]^.
Socioeconomic background was calculated using an environmental factor score ^[@R35]^, which incorporates neighborhood-level features based on geocoding (e.g. crime, median family income, etc.; see [Supplement](#SD1){ref-type="supplementary-material"}). Selected individuals from the imaging subsample were re-contacted for follow-up visits. Individuals who underwent follow-up MRI: 1) *completed the MRI protocol during initial enrollment*; and 2) *had either the presence of psychosis spectrum symptoms at the initial visit or were free of any psychopathology and/or medical conditions (e.g. healthy; [Figure 1](#F1){ref-type="fig"}).*
Initial Clinical Assessment (Time 1) {#S4}
------------------------------------
The initial clinical assessment included three structured screening tools to assess broad spectra of psychosis-relevant experiences and other psychopathology. Descriptions of the assessment tools and threshold classification criteria are published ^[@R13],\ [@R14]^. Briefly, subjects were classified as "Psychosis Spectrum" (PS) if they exceeded the threshold on the psychosis spectrum screen, regardless of the presence or absence of other psychopathology. Subjects were classified as "Other Psychopathology" (OP) if they did not meet criteria for PS, but exceeded criteria for one or more other psychopathology domains. Subjects were classified as "healthy" (HC) if they did not meet criteria for either PS or OP. Subjects were further screened and excluded if they had significant comorbid medical conditions, as previously described ^[@R34]^. Additional detail is provided in the [Supplement](#SD1){ref-type="supplementary-material"}.
Follow-up Clinical Assessment (Time 2) {#S5}
--------------------------------------
Mean time to clinical follow-up occurred 20.4 months after the initial visit; the range of follow-up was between 9--40 months. Descriptions of the assessment tools and threshold classification criteria at follow-up are published ^[@R36],\ [@R37]^. All participants received comprehensive clinical assessments including a semi-structured diagnostic interview to assess a broad spectrum of psychosis-relevant experiences ^[@R36]^. Individuals were classified as clinical high risk (CHR) for psychosis if they had at least one positive OR two negative and/or disorganized symptoms rated 3, 4, or 5 on the Scale of Prodromal Symptoms (SOPS) ^[@R38]^, without meeting criteria for a DSM-IV Axis I psychotic disorder. Six individuals met criteria for a psychotic disorder (PSY) at follow up. Again, subjects were classified as "Other Psychopathology" (OP) if they did not meet criteria for PS, but met DSM-IV criteria for one or more other psychopathology domains. Healthy individuals (HC) had no DSM-IV Axis I psychotic disorder, no super-threshold prodromal symptomatology, no history of psychosis in a first-degree biological relative, and no personal Axis II Cluster A diagnosis. All individuals received functioning (Global Assessment of Functioning; GAF^[@R39]^) screening inventories.
Longitudinal Clinical Categorization: {#S6}
-------------------------------------
To evaluate the stability of white matter microstructure as a relevant phenotype in psychosis risk, individuals were classified based upon the combination of the Time 1 and Time 2 clinical labels ([Figure 1](#F1){ref-type="fig"}). Given our interest in psychosis we focused our analyses on the comparison between those with persistent psychosis risk symptoms and those that were consistently healthy. If an individual was PS at the initial visit and CHR or PSY at follow-up, he/she was considered "Persistent" for psychosis risk. The "Typically Developing" (TD) group comprised healthy individuals who were free of significant psychopathology at both visit one and two. The number of months between clinical visits was similar for persistent psychosis risk \[23.51 (6.66)\] and TD \[22.57 (6.49); t(96)=1.20, p=0.23\].
Computerized Neurocognitive Battery: {#S7}
------------------------------------
Cognitive ability was measured using the Penn Computerized Neurocognitive Battery (CNB).^[@R33]^ We examined global factor scores for cognitive accuracy, speed and efficiency, which have been examined previously^[@R40]^ in relation to white matter microstructure. In general, the CNB was collected on the same day as the clinical assessment. The number of months between CNB examinations was similar for persistent psychosis risk \[23.55 (6.70)\] and TD \[22.79 (6.38);t(96)=1.20, p=0.23\].
Neuroimaging: {#S8}
-------------
MRI scans were acquired on the same 3T Siemens Tim Trio whole-body scanner, used the same 32-channel head coil and acquisition protocol. Data were collected between 2009 and 2013. The average duration, in months, between MRI scans was similar for persistent psychosis risk \[20.06 (6.25)\] and TD \[19.18 (6.40); t(96)=0.96, p=0.33\]. The range of MRI follow-up was 7--35 months.
Diffusion Imaging Sample: {#S9}
-------------------------
At Time 1, 38 Persisters and 79 TD had usable DWI data, while at Time 2, 37 Persisters and 89 TD had usable DWI data ([Figure 1](#F1){ref-type="fig"}). Of these individuals, 29 Persisters and 70 TD had complete clinical and DWI data passing QA (see DWI quality control in [Supplemental Material](#SD1){ref-type="supplementary-material"}) from *both* Time 1 and Time 2. Thus, a mixed-model statistical approach was used to take advantage of all available data. Applying a mixed effects model allowed for the use of all observations without list-wise deletion. Individuals were excluded based on health and medical history, incidental radiologic findings, and for poor DTI image quality, including both manual and automated image quality assessment ^[@R41]^. At Time 1, 10 Persisters were taking psychoactive medications, most commonly stimulants ([Table 1](#T1){ref-type="table"}). At Time 2, those medicated at Time 1 remained medicated and four additional individuals were taking psychoactive medications. A summary table of medications is provided [Supplemental Table 4](#SD1){ref-type="supplementary-material"}.
Diffusion Weighted Imaging Acquisition, Quality Control and Processing: {#S10}
-----------------------------------------------------------------------
As previously described ^[@R32],\ [@R41]^, DWI scans were obtained using a twice-refocused spin-echo (TRSE) single-shot EPI sequence (See [Supplement for sequence details](#SD1){ref-type="supplementary-material"}). The complete sequence consisted of 64 diffusion-weighted directions with b = 1000 s/mm^2^ and 7 interspersed scans where b = 0 s/mm^2\ [@R32]^. A B0 field map was acquired and used in the pre-processing ^[@R41]^. Images were checked for data quality using manual and automated methods, which are publicly available, ^[@R41]^ (<https://davidroalf.com/qascripts>). QC metrics include temporal signal-to-noise (TSNR) and mean relative motion. Following QC, diffusion data were skull stripped by generating a brain mask for each subject by registering a binary mask of a standard image (FMRIB58_FA) to each subject's brain using FLIRT ^[@R42]^. When necessary, manual adjustments were made to this mask. Next, eddy currents and head motion were estimated and corrected using FSL's eddy tool ^[@R41],\ [@R43],\ [@R44]^. The diffusion gradient vectors were rotated to adjust for motion using eddy's 6-parameter motion output. The field map was estimated and distortion correction was applied to the DWI data using FSL's FUGUE ^[@R42]^. Finally, the diffusion tensor was modeled and metrics were estimated at each voxel using FSL's DTIFIT.
Registration from native space to a template space was completed using DTI-TK ^[@R45]^. First, the DTI outputs of DTIFIT were converted to DTI-TK format and registered to PNC-specific template; the details of this procedure are published ^[@R41]^. All DTI maps were then registered (rigid, affine, diffeomorphic) to the high-resolution study-specific template using DTI-TK. FA, MD, AD, and RD were compared along a study specific white matter skeleton. Mean diffusion metrics were extracted from 10 full regions of interest (ROI; ICBM-JHU White Matter Tracts; Harvard-Oxford Atlas; [Supplemental Table 1](#SD1){ref-type="supplementary-material"}) using FSL's 'fslmeants' command and averaged across ROI. Additional details are provided in the [Supplement](#SD1){ref-type="supplementary-material"}.
Statistical Analysis {#S11}
--------------------
Demographic, clinical and cognitive group differences were examined with chi-square, t-tests and ANCOVAs, where appropriate. Prior work has demonstrated that brain development ^[@R46]^, including white matter development ^[@R47]^, is not a linear process. Thus, group-level analyses of DTI data were flexibly modeled using penalized splines within a mixed effects generalized additive mixed model (GAMM ^[@R48],\ [@R49]^). The GAMM is a commonly used statistical approach ^[@R50],\ [@R51]^ that assesses a penalty for increasing nonlinearity to avoid over-fitting and allows for structured errors such as those observed in longitudinal studies.
The following GAMM model used was: $$\left. \left. \text{DWI}\ \text{scalar}\ \text{metric}\ \left( {\text{FA}/\text{MD}/\text{RD}/\text{AD}} \right)\ \right.\sim\ s\left( {\text{AgeAtScan},\ k = 4} \right)\ + \ \text{diagnosis}\ + \ \text{sex}\ + \ \text{race}\ + \ \text{DWI}\ \text{TSNR},\ \text{random}\ \text{factor}\ = \ 1 \middle| \text{subject}. \right.$$
Diffusion (FA, MD, AD and RD), clinical and cognitive measures were modeled as the dependent measure, while the spline of age and diagnostic group were modeled as independent factors. Interaction terms (diagnosis\*age) were tested, but not significant (See [Supplemental Material](#SD1){ref-type="supplementary-material"}), thus these were not included in the final model. All models were controlled for potentially confounding effects of sex, race, and TSNR. Age was mean-centered in all analyses so that the intercept of the model can be interpreted within the age range of the study (See [Supplement for additional detail](#SD1){ref-type="supplementary-material"}). A k-value (smoothness term) of 4 was chosen based upon previous work using GAMM analyses within the PNC data in other brain phenotypes (e.g. ASL, Brain structure)^[@R50],\ [@R51]^. This previous work has found that this k-value of 4 tends not to overfit data while allowing enough flexibility over the limited age range of the PNC. The GAMM was implemented in 'mgcv' R package ^[@R49]^.
Whole brain analysis was performed using a customized non-linear implementation (R-package 'voxel')^[@R52]^ of FSL's randomise^[@R53]^ to compute tract-based spatial statistics (TBSS) ^[@R54]^ and corrected for multiple comparisons using threshold-free cluster estimation (TFCE; p\<0.05). The GAMM approach was employed for ROI data. False discovery rate (FDR) was applied to regional data to control for Type-I error probability (q-value =0.05); all regional p-values shown are FDR corrected. Intraclass correlation coefficients (ICCs) were used to measure similarity between scalar brain images and ROIs by scan visit. The I2C2 package in R^[@R55]^ was used to estimate ICCs for whole brain images. All statistics were performed using R (3.2.2) statistical software ^[@R56]^.
RESULTS {#S12}
=======
Participant Characteristics and Clinical Scales {#S13}
-----------------------------------------------
Participant characteristics are shown in [Table 1](#T1){ref-type="table"}. The diagnostic groups differed in Positive and Negative symptom scores, GAF, and environment score. Approximately 30% of youth with persistent symptoms were medicated, most commonly with a stimulant. Positive symptom scores were significantly lower at Time 2 than Time 1 in psychosis risk youth, but were unchanged in TD. Environment scores, which were only estimated based on Time 1 data, were significantly lower in the persistent youth that returned, while this score was unchanged in those TD who were followed at Time 2.
Computerized Neurocognitive Battery {#S14}
-----------------------------------
Persisters were less accurate, slower and less efficient than TD. This pattern was consistent across time as there was no interaction with age.
DWI Data Quality {#S15}
----------------
Persisters showed lower TSNR and higher head motion than TD, but there was no interaction with age.
PS youth have disrupted white matter microstructure {#S16}
---------------------------------------------------
Non-linear whole-brain TBSS revealed group differences between Persisters and TD in clusters along the major white matter skeleton for FA and RD, but not AD or MD ([Supplemental Figure 1](#SD1){ref-type="supplementary-material"}). FA differences were found in clusters comprising the Forceps Major, IFO, CGH, CST, ILF, SLF, and ATR. Differences in RD were limited to a cluster comprising the IFO and Forceps Major. There were no interactions with time. To fully elucidate these limited skeletonized results, diffusion metrics from each JHU White Matter Tract ROI were extracted and analyzed.
Regional Analysis {#S17}
-----------------
Mean DWI scalars are displayed in [Supplemental Table 2](#SD1){ref-type="supplementary-material"} and complete results of the GAMM analyses are shown in [Table 2](#T2){ref-type="table"}. Increasing age was associated with higher FA and lower diffusivity values ([Supplemental Results](#SD1){ref-type="supplementary-material"}).
### Lower FA in youth at persistent risk for psychosis. {#S18}
Persisters had lower average white matter FA as compared to TD ([Figure 2A](#F2){ref-type="fig"}). Regional FA ([Figure 3](#F3){ref-type="fig"}) was lower within the Cerebral Spinal Tracts (CST), and the Cingulum Bundle of the Hippocampus (CGH) as compared to TD. The intercept of each of these models was significant (ps\<0.001; [Supplemental Table 6](#SD1){ref-type="supplementary-material"}), but there were no interactions with age.
### Higher RD in youth at persistent risk for psychosis {#S19}
Persisters had higher average white matter RD (*p*~*FDR*~\<0.02) as compared to TD ([Figure 2D](#F2){ref-type="fig"}). The intercept of this models was significant (p\<0.001), but there was no interaction with age. No individual regions differed between Persisters and TD. There were no global or regional group differences in MD ([Figure 2B](#F2){ref-type="fig"}) or AD ([Figure 2C](#F2){ref-type="fig"}). Results for MD, AD and age and sex comparisons are presented in the [Supplemental Results](#SD1){ref-type="supplementary-material"}.
Sensitivity analysis of potential medication effects {#S20}
----------------------------------------------------
Comparisons of demographic and DWI metrics for persistent youth receiving psychoactive medications and those who were not are presented in the [Supplemental Tables 3 and 4](#SD1){ref-type="supplementary-material"}. All differences between TD and Persisters remained significant (except the CST FA) after the exclusion of medicated individuals, but the FDR corrected significance values were smaller.
Similarity of diffusion metrics over time {#S21}
-----------------------------------------
I2C2 values for diffusion metrics using the TBSS brain mask were: FA=0.71, MD=0.57, AD=0.64, RD=0.63. ICCs are shown for each diagnostic group in [Supplemental Table 6](#SD1){ref-type="supplementary-material"} for DWI scalars. ICCs for each ROI and each whole brain measure were significant (ps \<2.08×10^−5^) and ranged from 0.44 for MD in the cingulum bundle in to 0.97 for MD in the forceps major. In general, FA values had higher ICCs than measures of diffusivity.
Associations with clinical symptoms and neurocognitive performance {#S22}
------------------------------------------------------------------
Higher average MD (p=0.004), AD (p=0.03) and RD (0.003) was associated with higher positive symptom scores across TD and Persisters. The same relationship was observed for the Inferior Longitudinal Fasciculus (MD (p=0.005), AD (p=0.03), RD (p=0.003)) and the Inferior Frontal Occipital Fasciculus (MD (p=0.005), RD (p=0.003)). When analysis included only Persisters, this relationship was restricted to mean AD (p=0.03) and MD (p=0.04); see [Supplemental Figure 2A-B](#SD1){ref-type="supplementary-material"}. There were no associations with negative symptoms or GAF.
Higher average MD (p=0.04) and mean AD (p=0.04) were associated with lower CNB efficiency factor scores across TD and Persisters ([Supplemental Figure 2C-D](#SD1){ref-type="supplementary-material"}). Regionally, higher AD within the CST and UF, and higher RD within the CGH were associated with lower CNB efficiency factor scores. There were no interactions between diagnosis and clinical or cognitive performance for any diffusion outcome measure.
DISCUSSION {#S23}
==========
Using targeted longitudinal follow-up of a large community-based sample, we identified abnormalities of white matter microstructure in youth with persistent psychosis risk symptoms. As expected, non-linear age-related changes in DTI scalar measures were present in both Persisters and TD. Lower average and regional whole brain FA and higher RD were present in individuals with risk symptoms that were followed longitudinally over approximately two years. Lower FA was maximal within the cerebrospinal tracts and the cingulum bundle of the hippocampus. Yet, there was no interaction between age and diagnosis on DTI scalar measures. Higher whole-brain MD and AD were both associated with higher positive symptoms in individuals with persistent psychosis risk and lower cognitive efficiency across all participants. Overall, these findings delineate a pattern of localized, abnormal microstructural brain development in youth with persistent subthreshold psychosis symptoms.
Longitudinal studies of white matter changes in psychosis and psychosis risk are rather limited and previous results vary. Our results are similar to one previous study in early onset schizophrenia-spectrum disorders, in that we find altered diffusion in brain white matter (e.g. lower FA), without progressive worsening of these white matter abnormalities. Yet, our findings differ from that of the only previous longitudinal study in clinical high risk youth^[@R30]^. We did not find lower FA in the corpus callosum (e.g. forceps major/forceps minor), nor an association with negative symptoms. In fact, our findings indicate an association with positive symptoms. Methodological differences between these studies are important to consider as in Saito et al. ^[@R30]^ the follow-up time was shorter (12 months), FA was only measured in the callosum, and the study was completed at 1.5T with limited data quality assurance. Yet, the overall conversion rate to clinical psychosis was similar in each study (13% in the present study and 15% in Saito et al.). Our findings in youth with persistent psychosis-like symptoms add to a growing literature of white matter disruptions in youth at risk for psychosis, but importantly, indicate that this dysfunction is present in a sample of individuals with persistent *subthreshold* clinical symptoms.
The specific reduction in FA within the lateral temporal aspect of the cingulum bundle (e.g. CGH) is intriguing given the significant connections that this WM tract has with key brain structures known to affect symptoms and cognition in psychosis ^[@R57]^. The cingulum bundle (CB) is a prominent white matter tract comprising long and short-range fibers that support interactions among the prefrontal, parietal, and temporal lobes ^[@R58]--[@R60]^. The CB has connections with regions associated with memory and executive functioning including the thalamus, amygdala, hippocampus, and dorsolateral and dorsomedial prefrontal cortex ^[@R61]^. Importantly, recent neuroanatomical evidence indicates distinct subdivisions of the cingulum - including a lateral temporal subdivision with specific connections between the parahippocampus and the frontal and parietal lobes ^[@R62]^. This aspect of the cingulum bundle is a "control" pathway supporting memory, executive function and other cognitive functions ^[@R63]^. It is within this specific subdivision of the cingulum that we find significantly lower FA in youth with persistent psychosis-like symptoms. Thus, it is possible that brain regions associated with this aspect of the cingulum are the earliest to exhibit neuronal dysfunction in this sample and may contribute to higher positive symptoms and/or lower cognitive efficiency reported in the current study. Indeed, a recent study using a cross-sectional sample of psychosis spectrum youth from the PNC found these individuals to have lower gray matter volume within the medial temporal lobe ^[@R50]^, which was associated with higher positive symptoms. Yet, we acknowledge that most associations between diffusion measures and symptoms and cognitions were limited to average whole brain measures. As such, the specificity of these associations remains to be elucidated.
It is tempting to compare the similarity of our findings to those in help-seeking samples and to patients with documented psychosis. For example, altered WM organization, as indicated by lower fractional anisotropy (FA) and high diffusivity, is well documented in numerous brain regions in chronic SZ ^[@R7],\ [@R8]^, early-onset psychosis ^[@R21]^, psychosis in adolescence ^[@R22]^ and youth at risk for developing psychosis ^[@R18]^. Typically, these findings are seen in major WM fiber tracts including the corpus callosum, anterior limb of the internal capsule and cingulum bundle ^[@R23]^. However, we must consider that young adults can present with highly variable clinical symptoms during the psychosis risk period and that many will not develop significant psychotic disorders. Moreover, these individuals may also exhibit depression, anxiety, attention deficit, and conduct disorder symptoms ^[@R13],\ [@R36]^. Yet, brain white matter measures may be useful even when the psychosis risk diagnosis is unambiguous. Ultimately, assessing the predictive utility of these measures requires studies that track brain development earlier in youth and follow individuals over the long-term to determine final clinical outcomes. Nonetheless, we note the consistent parallels between what we observe here in the pre-psychotic risk period and the findings reported in psychosis patients. Identifying these common neurobiological markers prior to the emergence of super-threshold symptomatology may enhance prediction and facilitate earlier intervention.
In order to assess whether white matter abnormalities were already present or emerged during the time frame of the current study we investigated the intercept for those measures with significant group effects. The intercept was highly significant suggesting that the diffusion values (FA and RD) at the mean age of our sample differ and are consistently different across all ages included in the current study (e.g. no interaction with age). For example, mean FA across the brain of the youngest PS individual (0.44) is approximately 0.01 (\~2.5%) lower than that estimated in TD. Moreover, as there is no age-by-diagnosis interaction, this difference in FA is consistent across ages within our sample. This is clearly demonstrated in [Figure 2A](#F2){ref-type="fig"}.
While most structural MRI work in individuals along the psychosis spectrum has focused on gray matter disruption, there is compelling evidence that dysfunction of brain white matter, specifically in oligodendrocytes and myelin formation, significantly contributes to psychosis (see^[@R64]^). Myelination begins postnatally, but completes in young adulthood. Critically, reduced expression of genes associated with components of WM---oligodendrocytes and myelin---are widely replicated in SZ ^[@R64]^ and abnormal oligodendrocyte function is speculated as a primary mechanism in SZ ^[@R65]^. Thus, it is plausible that altered expression of specific neurotransmitter receptors on oligodendrocytes, which has been shown in demyelinating disorders ^[@R66],\ [@R67]^, could be one mechanism of neuronal dysfunction in psychosis. Our results do not suggest that critical changes in brain white matter occur during this period---approximately during the same period when the incidence of SZ increases---but they may indicate specific failures of brain cytoarchitecture that predispose individuals to more serious clinical symptoms. Moreover, it is possible that deficits in brain white matter, along with other clinical and behavioral symptoms during this time period, would increase confidence in the at-risk diagnosis.
Despite the large-scale nature of the PNC, several limitations should be considered. First, the PNC was set as a cross-sectional study, but particular samples of interest have been followed-up in targeted, yet smaller, investigations. Thus, the current longitudinal study, while one of the largest to date to implement diffusion MRI in youth at risk for developing psychosis, remains relatively small. In addition, we focused our analysis on individuals with persistent psychosis symptoms over two consecutive assessments. There are, however, individuals with inconsistent clinical symptoms that may provide valuable insight. However, these individuals may have more heterogeneity in clinical symptoms and likely brain phenotypes. Unlike some previous imaging studies, we only found *lower* FA and higher diffusivity values in psychosis spectrum youth. Other studies report *elevated* FA in CHR^[@R68]^ or *no difference* between CHR and healthy individuals^[@R69],\ [@R70]^. These inconsistencies may stem from relatively small at-risk samples, heterogeneity of disease course, lack of quality assurance and differences in DTI methodology. While our study is relatively small, the sample is well characterized and followed longitudinally. In addition, we believe our comprehensive approach to data QA and use of a study specific template are strengths that reduce analytic error. Regions of significantly lower FA in the whole brain analysis did not correspond directly to the regional effects. While the skeletonized approach allows for a whole brain method, TBSS is limited in inferences that can be made at the level of the individual tract. Thus, we included complementary approaches, which incorporated voxel-wise and regional approach, to improve specificity of reported deficits. Alterations in the corticospinal tract in at-risk individuals suggests that WM deficits may be linked to motoric disturbances or psychomotor slowing noted in psychosis ^[@R71],\ [@R72]^. What role these motoric disturbances play in the associated cognitive performance deficits remains unknown and should be further investigated. Our follow-up time is restricted to 20 months and longer inter-scan intervals (or additional scans) will likely improve our ability to map meaningful individual change. Yet, importantly, we show that diffusion metrics are, on average, consistent within individuals by estimating the ICCs between diffusion time points. Finally, we find significant associations between AD and MD with cognitive performance, but find no group difference in AD or MD. This finding may suggest dimensional features of brain white matter that need to be further explored with respect to symptoms and cognition, as it is unlikely that there are specific cut-points in symptoms or performance that separate those at-risk for psychosis and typically developing youth.
In conclusion, we measured WM microstructure in a targeted subsample of the PNC, a community sample of youth with and without persistent psychosis spectrum symptoms. We found convergent results demonstrating localized disruptions of white matter microstructure in youth with persistent psychotic-like symptoms compared to typically developing youth. These abnormalities were prominent within the cortico-spinal tracts and the cingulum bundle of the hippocampus. These results suggest that youth with persistent psychotic-risk symptoms have similar patterns of white matter microstructure abnormalities to those seen in clinically ascertained at-risk samples. It does not appear that these deficits depend on duration of subthreshold symptoms or the confounding influence of psychotropic medication, as very few of these individuals were receiving medication and exclusion of medicated individuals does not change the pattern of results. Taken together we believe that WM microstructure is a meaningful brain phenotype that should be measured and monitored throughout early development. Reliable identification of a "true" psychosis prodrome that signals the onset of a psychotic illness, especially early in its course, is a significant challenge. The presence of robust abnormalities, such as alterations in brain white matter, would increase confidence in the at-risk 'diagnosis', while an entirely normal profile might help to identify false positives and thus alleviate some clinical concerns.
Supplementary Material {#SM1}
======================
Thanks to the acquisition and recruitment team: Karthik Prabhakaran, Jeff Valdez, Raphael Gerraty, Marisa Riley, Jack Keefe, Elliott Yodh, Jason Blake, Prayosha Villa, R. Sean Gallagher and Rosetta Chiavacci.
**Funding Sources:** This work was supported by the National Institute of Mental Health grants MH089983, MH089924, MH087626. Additional support was provided by K01MH102609 to DRR; K23MH098130 and R01 MH107703 to TDS; R01 MH112847 to TDS and RTS; the Dowshen Program for Neuroscience at the University of Pennsylvania; and the Life Span Brain Institute (LiBI)---a collaboration between the University of Pennsylvania School of Medicine and Children's Hospital of Philadelphia. This work was also supported by a NARSAD Young Investigator Grant from the Brain & Behavior Research Foundation. The funding sources were not directly involved in study design, collection, data analysis or interpretation, nor manuscript writing.
Financial Disclosures/Conflicts of Interest:
**David R. Roalf, PhD** reports no potential conflicts of interest or financial disclosures related to this work.
**Angel Garcia de la Garza, BS** reports no potential conflicts of interest or financial disclosures related to this work.
**Monica E Calkins, PhD** reports no potential conflicts of interest or financial disclosures related to this work.
**Adon Rosen, BS** reports no potential conflicts of interest or financial disclosures related to this work.
**Tyler M Moore** reports no potential conflicts of interest or financial disclosures related to this work.
**Megan Quarmley, BA** reports no potential conflicts of interest or financial disclosures related to this work.
**Kosha Ruparel MSE** reports no potential conflicts of interest or financial disclosures related to this work.
**Cedric Huchuan Xia** reports no potential conflicts of interest or financial disclosures related to this work.
**Petra E. Rupert, BS** reports no potential conflicts of interest or financial disclosures related to this work.
**Theodore D. Satterthwaite, MD** reports no potential conflicts of interest or financial disclosures related to this work.
**Russell T. Shinohara, PhD** reports no potential conflicts of interest or financial disclosures related to this work.
**Mark A. Elliott PhD** reports no potential conflicts of interest or financial disclosures related to this work.
**Ruben C. Gur PhD** received royalties from the Brain Resource Centre.
**Raquel E. Gur MD PhD** reports no potential conflicts of interest or financial disclosures related to this work.
Supplementary information is available at MP's website
![Flow diagram of Persistent Psychosis and Typically Developing youth as enrolled and followed as part of the Philadelphia Neurodevelopmental Cohort. Of the 1601 individuals imaged as part of the PNC, 452 individuals screened positive for psychosis spectrum (PS) symptoms and 498 were considered healthy. At follow-up, 208 PS and 210 healthy were re-enrolled. Of these, 106 PS has persistent or worsening symptoms, while 153 healthy were still considered healthy. Of those imaged at follow-up, 46 youth with persistent psychosis risk symptoms and 98 TD had high quality diffusion weighted imaging acquired. Note: Not all individuals at baseline or follow-up completed the DWI portion of the neuroimaging protocol^[@R32],\ [@R34]^ or data quality was considered too poor for analysis^[@R41]^.](nihms-1518566-f0001){#F1}
{#F2}
{#F3}
######
Participant characteristics for Persistent Psychosis Risk and Typically Developing youth from the PNC at Time 1 and Time 2.
Persistent Psychosis Risk Typically Developing Statistical Analysis^[a](#TFN11){ref-type="table-fn"}^
------------------------------------------------------------------ ------------------------------- ------------------------------- -------------------------------------------------------- ------------------------------- ------------------------------ ------------------------------- ------------------------------
n 38 37 79 89
Initial Age \[range\] 15.51 (2.52) \[11.50--21.41\] 16.82 (2.61) \[12.75--22.50\] 16.64 (2.82) \[11.33--21.75\] 17.38 (3.56) \[11.25--23.58\] **F(1,97)=741.09, p\<0.001** F(1, 168)=1.17, p=0.28 F(1, 97)=0.46, p=0.50
Sex (% female) 52.6% 40.5% 51.9% 52.8% **F(1,239)=7.22, p=0.007** F(1,239)=0.23, p=0.63 F(1,239)=0.20, p=0.65
Race (% Caucasian) 18.4% 27.0% 67.1% 66.3% F(1,239)=0.41, p=0.52 **F(1,239)=47.90, p\<0.001** F(1,239)=3.53, p=0.06
Handedness (% right) 81.6% 81.1% 83.5% 83.1% F(1,239)=0.42, p=0.52 F(1,239)=0.08, p=0.76 F(1,239)=0.60, p=0.44
Maternal education^[1](#TFN2){ref-type="table-fn"}^ 13.79 (2.12) 14.03 (3.62) 15.05 (2.62) 13.55 (4.90) F(1,185)=0.11, p=0.74 F(1,130)=0.27, p=0.61 F(1,185)=0.67, p=0.41
Paternal education^[1](#TFN2){ref-type="table-fn"}^ 13.43 (2.52) 12.08 (5.46) 14.74 (2.92) 12.67 (5.76) F(1,185)=1.25, p=0.27 F(1,122)=0.81, p=0.37 F(1,185)=0.07, p=0.79
Positive symptom score^[2](#TFN3){ref-type="table-fn"}^ 20.38 (13.47) 12.34 (11.55) 2.01 (4.09) 1.44 (2.88) **F(1,192)=16.50, p\<0.001** **F(1,137)=125.83, p\<0.001** **F(1,192)=10.78, p=0.001**
Negative symptom score^[3](#TFN4){ref-type="table-fn"}^ 0.96 (0.76) 1.14 (0.74) 0.09 (0.17) 0.09 (0.24) F(1,194)=1.49, p=0.22 **F(1,148)=180.54, p\<0.001** F(1,97.4)=741.09, p=0.15
Global function score^[4](#TFN5){ref-type="table-fn"}^ (z-score) −1.77 (1.68) −1.98 (1.03) 0.24 (0.86) 0.24 (0.97) F(1,164)=0.54, p=0.16 **F(1,118)=148.51, p\<0.001** F(1,164)=3.06, p=0.08
TSNR^[5](#TFN6){ref-type="table-fn"}^ 7.31 (0.38) 7.15 (0.26) 7.43 (0.39) 7.28 (0.38) **F(1,184)=22.98, p\<0.001** **F(1,36)=7.64, p=0.006** F(1,184)=1.34, p=0.25
Average head motion^[6](#TFN7){ref-type="table-fn"}^ 0.42 (0.27) 0.41 (0.26) 0.33 (0.16) 0.34 (0.18) F(1,185)=0.24, p=0.90 **F(1,143)=5.14, p=0.02** F(1,185)=0.07, p=0.78
Psychoactive Medications^[7](#TFN8){ref-type="table-fn"}^ 10 14 0 0 F(1,122)=0.02, p=0.90 **F(1,122)=0.02, p\<0.001** F(1,122)=0.28, p=0.60
CNB Accuarcy^[9](#TFN9){ref-type="table-fn"}^ −0.27 (1.07) −0.08 (1.17) 0.48 (0.78) 0.50 (0.75) **F(1,233)=33.95, p\<0.001** **F(1,138)=19.09, p\<0.001** F(1,233)=0.54, p=0.48
CNB Speed^[9](#TFN9){ref-type="table-fn"}^ −0.30 (1.26) 0.11 (0.99) 0.19 (0.86) 0.45 (0.82) F(1,206)=1.28, p=0.08 **F(1,140)=6.45, p=0.01** F(1,206)=0.58, p=0.45
CNB Efficiency^[9](#TFN9){ref-type="table-fn"}^ −0.40 (1.14) −0.02 (1.07) 0.42 (0.82) 0.64 (0.69) **F(1,226)=23.18, p\<0.001** **F(1,133)=19.95, p\<0.001** F(1,233)=0.32, p=0.86
Envrionment Score^[10](#TFN10){ref-type="table-fn"}^ −0.57 (1.01) −0.40 (1.07) 0.16 (0.95) 0.20 (0.93) F(1,239)=0.17, p=0.68 **F(1,239)=25.79, p\<0.001** **F(1, 239)=13.07, p=0.003**
Values are shown as mean +/− standard deviation.
Values are show in years;
SIPS positive items (Calkins et al., 2015);
SIPS negative items (Calkins et al., 2015);
Global Assessment of Function scale;
Temporal signal- to-noise ratio (Roalf et al., 2016);
Mean relative root mean squared motion (Roalf et al., 2016);
See [Supplemental Table 4](#SD1){ref-type="supplementary-material"};
CNB Factor scores (Moore et al., 2016).
Enviornment score dervied as in Moore et al., 2017.
Mixed effects model was used to compare demographic variables by age and diagnostic group where appropriate. Otherwise, ANOVAs (e.g. sex, race, handedness, environment score).Note: for Age, visit number was used as the independent measure. Age was centered for all analyses.
######
Raw FA, MD, AD, and RD values for the two diagnostic groups: Typically Developing and Persistent Psychosis Risk. Overall mean values and group means are shown for each timepoint. MD, AD and RD values are (×10^−3^ mm^2/s^).
TD Persistent TD Persistent TD Persistent TD Persistent
---------------------- ----------- ------------ ----------- ------------ ----------- ------------ ----------- ------------
**ATR** **0.372** **0.366** **2.200** **2.209** **1.038** **1.035** **0.581** **0.587**
***T1*** 0.371 0.366 2.203 2.225 1.039 1.043 0.582 0.591
***T2*** 0.373 0.367 2.197 2.193 1.037 1.028 0.580 0.582
**CST** **0.553** **0.548** **1.981** **1.978** **1.121** **1.114** **0.430** **0.432**
***T1*** 0.552 0.547 1.987 1.993 1.123 1.122 0.432 0.436
***T2*** 0.555 0.549 1.975 1.963 1.119 1.106 0.428 0.429
**CGC** **0.536** **0.531** **2.080** **2.071** **1.165** **1.154** **0.458** **0.459**
***T1*** 0.532 0.530 2.091 2.086 1.167 1.161 0.462 0.462
***T2*** 0.539 0.532 2.070 2.057 1.162 1.146 0.454 0.455
**CGH** **0.390** **0.375** **2.146** **2.151** **1.039** **1.027** **0.553** **0.562**
***T1*** 0.388 0.372 2.152 2.173 1.041 1.034 0.555 0.569
***T2*** 0.391 0.379 2.139 2.130 1.036 1.019 0.552 0.555
**IFO** **0.444** **0.437** **2.223** **2.226** **1.137** **1.130** **0.543** **0.548**
***T1*** 0.443 0.436 2.228 2.243 1.139 1.138 0.545 0.553
***T2*** 0.445 0.438 2.217 2.209 1.135 1.122 0.541 0.543
**ILF** **0.471** **0.464** **2.160** **2.171** **1.121** **1.119** **0.519** **0.526**
***T1*** 0.470 0.462 2.165 2.189 1.123 1.127 0.521 0.531
***T2*** 0.471 0.466 2.156 2.152 1.119 1.112 0.518 0.520
**SLF** **0.378** **0.373** **2.170** **2.176** **1.016** **1.014** **0.577** **0.581**
***T1*** 0.376 0.371 2.179 2.193 1.019 1.020 0.580 0.586
***T2*** 0.379 0.375 2.162 2.159 1.013 1.007 0.575 0.576
**UF** **0.424** **0.415** **2.183** **2.191** **1.156** **1.145** **0.567** **0.573**
***T1*** 0.422 0.414 2.189 2.207 1.158 1.151 0.570 0.577
***T2*** 0.426 0.416 2.178 2.175 1.154 1.140 0.565 0.569
**Forceps Major** **0.589** **0.578** **2.291** **2.292** **1.398** **1.420** **0.501** **0.533**
***T1*** 0.590 0.577 2.299 2.305 1.397 1.421 0.496 0.532
***T2*** 0.588 0.579 2.283 2.278 1.400 1.420 0.505 0.534
**Forceps Minor** **0.437** **0.434** **2.402** **2.487** **1.208** **1.208** **0.592** **0.597**
***T1*** 0.437 0.431 2.394 2.486 1.208 1.214 0.591 0.603
***T2*** 0.438 0.437 2.410 2.488 1.209 1.202 0.593 0.592
**Mean Whole Brain** **0.454** **0.446** **2.393** **2.402** **1.122** **1.117** **0.531** **0.537**
***T1*** 0.452 0.445 2.391 2.420 1.124 1.124 0.532 0.542
***T2*** 0.455 0.448 2.395 2.385 1.120 1.110 0.529 0.532
ATR=anterior thalamic radiationL CST= corticospinal tractsL CGC=cingulum bundle of the cingulate gyrusL CGH=cingulum bundle of the hippocampusL IFO=inferior frontal occipital fasciculusL ILF=inferior longitudinal fasciculusL SLF=superior longitudinal fasciculusL UF=uncinate fasciculus. TD= Typically DevelopingL T1 = Time 1 visit [(PNC)l]{.smallcaps} T2 = Time 2 visit (followFup)
######
Model parameters for GAMM models for each DTI scalar metric. Significance values are FDR corrected.
Table 2A Diagnosis Sex Age EDFs & R^2^
------------------- ----------- -------- ------------ ---------------- --------- -------- ------------- ---------------- ----------- ---------------- ------ -------
ATR 0.0047 0.0027 1.73 1.65E-01 −0.0063 0.0024 ***−2.64*** ***4.39E-02*** **30.47** ***2.46E-11*** 1.93 0.22
CST 0.0109 0.0034 ***3.21*** ***1.53E-02*** −0.0032 0.0030 −1.08 3.51E-01 **6.75** ***1.71E-03*** 1.78 0.09
CGC 0.0087 0.0068 1.28 2.02E-01 −0.0139 0.0060 −2.30 7.41E-02 **32.61** ***2.10E-10*** 1.64 0.12
CGH 0.0141 0.0052 ***2.72*** ***3.53E-02*** −0.0075 0.0046 −1.63 1.72E-01 **6.24** ***2.46E-03*** 2.27 0.12
IFO 0.0066 0.0029 2.31 7.34E-02 −0.0015 0.0025 −0.58 6.27E-01 **17.95** ***7.57E-07*** 1.74 0.12
ILF 0.0060 0.0038 1.58 1.65E-01 −0.0039 0.0034 −1.17 3.45E-01 **9.50** ***4.07E-04*** 1.69 0.07
SLF 0.0051 0.0032 1.58 1.65E-01 −0.0054 0.0028 −1.91 1.14E-01 **17.36** ***2.48E-07*** 2.08 0.15
UF 0.0057 0.0039 1.44 1.90E-01 −0.0072 0.0035 −2.07 9.89E-02 **11.94** ***9.27E-04*** 1.00 0.09
Forceps Major 0.0095 0.0054 1.76 1.65E-01 0.0022 0.0047 0.47 6.38E-01 3.40 6.63E-02 1.00 −0.01
Forceps Minor 0.0039 0.0030 1.32 2.02E-01 −0.0119 0.0026 ***−4.59*** 7.37E-05 **14.44** ***1.74E-06*** 2.13 0.12
Mean White Matter 0.0079 0.0026 ***3.01*** ***2.94E-03*** −0.0060 0.0023 ***−2.59*** ***1.02E-02*** **25.08** ***2.75E-11*** 2.32 0.18
Table 2B MD
------------------- --------- -------- ------- ---------- --------- -------- ---------- ---------------- ----------- ---------------- ------ -------
ATR 0.0262 0.0114 −0.96 4.81E-01 −0.0145 0.0122 2.30 5.52E-02 **16.92** ***3.60E-07*** 1.94 0.09
CST 0.0129 0.0089 −1.03 4.81E-01 −0.0212 0.0095 1.45 2.00E-01 **15.73** ***1.29E-06*** 1.87 0.16
CGC 0.0385 0.0119 −0.02 9.81E-01 −0.0245 0.0128 **3.25** ***6.32E-03*** **41.36** ***3.08E-09*** 1.00 0.17
CGH 0.0157 0.0117 −1.13 4.81E-01 −0.0223 0.0126 1.34 2.00E-01 **26.79** ***1.80E-11*** 2.50 0.25
IFO 0.0148 0.0100 −0.80 5.31E-01 −0.0127 0.0108 1.48 2.00E-01 **18.54** ***5.22E-08*** 2.23 0.08
ILF 0.0121 0.0123 −1.39 4.81E-01 −0.0097 0.0132 0.99 3.26E-01 **13.83** ***1.79E-06*** 2.42 0.09
SLF 0.0191 0.0113 −1.11 4.81E-01 −0.0063 0.0121 1.69 1.84E-01 **9.86** ***6.79E-05*** 2.31 0.06
UF 0.0348 0.0111 −1.14 4.81E-01 −0.0229 0.0119 **3.14** ***6.32E-03*** **20.93** ***1.08E-05*** 1.00 0.14
Forceps Major −0.0715 0.0514 −1.36 4.81E-01 −0.0677 0.0557 −1.39 2.00E-01 **6.49** ***1.15E-02*** 1.00 −0.02
Forceps Minor 0.0539 0.0147 −0.65 5.71E-01 −0.0106 0.0159 **3.66** ***3.16E-03*** **4.96** ***6.03E-03*** 1.90 0.05
Mean White Matter 0.0180 0.0091 −1.68 9.46E-02 −0.0195 0.0098 **1.97** ***4.99E-02*** **19.57** ***2.78E-09*** 2.46 0.13
Table 2C AD
------------------- --------- -------- ------- ---------- --------- -------- ------- ---------- ------------- ---------------- ------ -------
ATR 0.0068 0.0044 −0.32 9.46E-01 −0.0078 0.0048 1.54 2.50E-01 **13.04** ***7.44E-04*** 1.00 0.01
CST 0.0051 0.0048 1.13 7.92E-01 −0.0018 0.0052 1.07 4.10E-01 **18.73** ***1.03E-04*** 1.00 0.10
CGC 0.0090 0.0068 1.26 7.92E-01 0.0024 0.0073 1.33 3.07E-01 3.25 7.28E-02 1.00 0.00
CGH 0.0000 0.0054 1.24 7.92E-01 −0.0156 0.0058 −0.01 9.96E-01 **22.78** ***7.73E-09*** 2.01 0.23
IFO 0.0061 0.0038 0.85 7.92E-01 −0.0064 0.0041 1.62 2.50E-01 ***10.56*** 1.03E-04 1.99 0.05
ILF 0.0008 0.0048 −0.32 9.46E-01 −0.0085 0.0052 0.17 9.59E-01 **7.58** ***1.33E-03*** 1.75 0.04
SLF 0.0030 0.0039 −0.27 9.46E-01 −0.0057 0.0042 0.77 5.52E-01 **5.22** ***2.90E-02*** 1.00 −0.01
UF 0.0105 0.0047 −0.07 9.46E-01 −0.0171 0.0051 2.21 1.43E-01 **13.49** ***7.38E-04*** 1.00 0.13
Forceps Major −0.0311 0.0186 −0.98 7.92E-01 −0.0230 0.0201 −1.68 2.50E-01 **5.64** ***2.61E-02*** 1.00 −0.02
Forceps Minor 0.0115 0.0052 −0.08 9.46E-01 −0.0045 0.0056 2.20 1.43E-01 **4.01** ***5.15E-02*** 1.00 0.00
Mean White Matter 0.0031 0.0035 0.31 7.54E-01 −0.0078 0.0038 0.88 3.81E-01 **12.99** ***8.63E-05*** 1.33 0.07
Table 2D RD
------------------- --------- -------- ----------- ---------------- --------- -------- ---------- ---------------- ----------- ---------------- ------ ------
ATR 0.0097 0.0039 −1.22 2.82E-01 −0.0034 0.0042 **2.50** ***3.26E-02*** **21.46** ***5.04E-09*** 2.15 0.14
CST 0.0041 0.0032 −2.25 1.26E-01 −0.0101 0.0035 1.28 2.26E-01 **14.77** ***1.12E-06*** 2.06 0.14
CGC 0.0159 0.0053 −0.90 3.78E-01 −0.0150 0.0057 **2.98** ***1.22E-02*** **32.97** ***2.38E-10*** 1.67 0.19
CGH 0.0080 0.0046 −2.30 1.26E-01 −0.0043 0.0049 1.76 1.33E-01 **20.92** ***1.88E-09*** 2.59 0.20
IFO 0.0045 0.0036 −1.53 2.39E-01 −0.0033 0.0039 1.24 2.26E-01 **20.50** ***6.20E-09*** 2.25 0.10
ILF 0.0059 0.0045 −1.67 2.39E-01 −0.0005 0.0048 1.31 2.26E-01 **14.82** ***6.45E-07*** 2.48 0.11
SLF 0.0085 0.0042 −1.39 2.39E-01 −0.0007 0.0045 2.03 8.61E-02 **13.98** ***1.12E-06*** 2.45 0.11
UF 0.0125 0.0042 −1.42 2.39E-01 −0.0030 0.0046 **2.94** ***1.22E-02*** **18.59** ***2.95E-05*** 1.00 0.12
Forceps Major −0.0204 0.0168 −1.54 2.39E-01 −0.0220 0.0182 −1.22 2.26E-01 **5.27** ***2.26E-02*** 1.00 0.00
Forceps Minor 0.0212 0.0051 −0.88 3.78E-01 −0.0030 0.0055 **4.14** ***4.86E-04*** **7.46** ***7.72E-04*** 2.13 0.08
Mean White Matter 0.0076 0.0032 **−2.47** ***1.42E-02*** −0.0060 0.0035 **2.34** ***2.01E-02*** **23.73** ***2.51E-11*** 2.58 0.17
ATR=anterior thalamic radiationL CST= corticospinal tractsL CGC=cingulum bundle of the cingulate gyrusL CGH=cingulum bundle of the hippocampusL IFO=inferior frontal occipital fasciculusL ILF=inferior longitudinal fasciculusL SLF=superior longitudinal fasciculusL UF=uncinate fasciculus. TD= Typically DevelopingL edf= estimated degrees of freedom;
######
Intraclass correlation coefficients for DTI scalar metrics from ROIs for Typically Developing and those with Persistent Psychosis.
Typically Developing Persistent Psychosis
----------------------- ---------------------- ---------------------- -------- -------- -------- -------- -------- --------
**ATR** *0.65* *0.67* *0.70* *0.68* *0.71* *0.61* *0.72* *0.62*
**CST** *0.77* *0.59* *0.71* *0.66* *0.86* *0.77* *0.81* *0.81*
**CGC** *0.84* *0.44* *0.70* *0.65* *0.82* *0.62* *0.75* *0.71*
**CGH** *0.72* *0.47* *0.57* *0.52* *0.79* *0.56* *0.69* *0.65*
**IFO** *0.82* *0.73* *0.71* *0.77* *0.74* *0.63* *0.65* *0.66*
**ILF** *0.83* *0.72* *0.70* *0.78* *0.78* *0.71* *0.74* *0.74*
**SLF** *0.70* *0.50* *0.53* *0.57* *0.85* *0.62* *0.55* *0.70*
**UF** *0.81* *0.62* *0.70* *0.69* *0.76* *0.66* *0.76* *0.68*
**Forceps Major** *0.89* *0.97* *0.96* *0.96* *0.90* *0.97* *0.97* *0.97*
**Forceps Minor** *0.85* *0.81* *0.77* *0.85* *0.87* *0.79* *0.69* *0.84*
**Mean White Matter** *0.78* *0.67* *0.62* *0.71* *0.81* *0.71* *0.71* *0.74*
ATR=anterior thalamic radiation; CST= corticospinal tracts; CGC=cingulum bundle of the cingulate gyrus; CGH=cingulum bundle of the hippocampus; IFO=inferior frontal occipital fasciculus; ILF=inferior longitudinal fasciculus; SLF=superior longitudinal fasciculus; UF=uncinate fasciculus. TD= Typically Developing
| {
"pile_set_name": "PubMed Central"
} |
The authors wish to make the following correction to their paper \[[@B1-antioxidants-09-00508]\]:
The number corresponding to the affiliation of Lucian Hritcu had been changed from:
**Lucian Hritcu ^1,^\***
to
**Lucian Hritcu ^2,^\***
The authors would like to apologize for any inconvenience caused to the readers by these changes. The changes do not affect the scientific results.
[^1]: These authors contributed equally to this work.
| {
"pile_set_name": "PubMed Central"
} |
Background {#Sec1}
==========
There are two forms of sphingosine kinase (SphK), SphK1 and SphK2. SphK1 has an established role in malignant biology with overexpression being associated with poor survival in patients with solid tumors \[[@CR1]--[@CR10]\] and resistance to therapy \[[@CR11]--[@CR14]\]. Furthermore, inhibitors of SphK1 have demonstrated preclinical activity in acute myeloid leukemia (AML) \[[@CR15], [@CR16]\]. The role of SphK2 has been more controversial but it is increasingly being shown to play a role in malignant disease and has been associated with poor patient outcome \[[@CR17]\]. Knockdown of SphK2 expression increases the sensitivity of cancer cells to chemotherapy \[[@CR18]--[@CR20]\], while chemical inhibition can reduce cancer cell growth in vitro \[[@CR21]--[@CR28]\] and in pre-clinical animal models \[[@CR21], [@CR24], [@CR26]\]. SphK2 inhibitors are now in phase II clinical trials for a number of cancers including B cell malignancies, following successful completion of phase I studies \[[@CR29]\]. We have recently shown that chemical inhibition of SphK2 can reduce acute lymphoblastic leukemia (ALL) cell growth, induce cell death in vitro and extend the survival of mice bearing human ALL xenografts. Furthermore, cells lacking SphK2 had a reduced capacity to induce ALL driven by the BCR/ABL1 fusion gene in WT mice, while SphK2 inhibition synergized with imatinib treatment of BCR/ABL1+ ALL in vitro and in vivo \[[@CR30]\].
Mice deficient in the tumor suppressor gene ARF are prone to malignancies, with undifferentiated sarcomas predominating (\~ 38%), followed by lymphomas (\~ 23%), carcinomas (\~ 15%) and neurological tumors (\~ 10%), with a latency of around 266 days \[[@CR31]\]. Genetic loss of material at the 9p21 locus, which includes ARF, is common in ALL, being reported in up to 45% of B lineage disease \[[@CR32]--[@CR34]\], making this a biologically relevant model. The development of tumors in these mice appears to be dependent on the aquisition of additional genetic changes as treatment with radiation or the mutagen DMBA significantly reduces latency. Here we show that blockade of T and B cell maturation by crossing ARF deficient mice onto a Rag1^−/−^ background \[[@CR35]\] resulted in an incidence of ALL of over 60%. Further crossing of these mice onto SphK2 deficient animals \[[@CR36]\] permitted the examination of the role of SphK2 in the development of ALL, demonstrating a significant reduction in disease incidence.
Methods {#Sec2}
=======
Development of mouse model {#Sec3}
--------------------------
Mice lacking the p19ARF product of the INK4a/ARF locus (ARF^−/−^) develop malignancies at a high penetrance with 80% of animals dying within the first year of life \[[@CR31]\]. To facilitate breeding we used mice where the ARF gene had been floxed (ARF^fl/fl^) (B6.129-Cdkn2atm4Cjs/Nci, \[[@CR37]\]) obtained from Graham Walker (QIMR, Queensland Australia). In order to produce an ALL model we crossed these mice with those lacking Rag1^tm1Mom^ from The Jackson Laboratory (Bar Harbour, ME, USA) \[[@CR35]\]. The resulting Mx1.Cre.ARF^fl/fl^.Rag1^−/−^ (MAR) mice were then crossed onto animals lacking SphK2 (Richard Proia (Bethesda, USA) \[[@CR36]\]) to produce Mx1.Cre.ARF^fl/fl^.Rag1^−/−^.SphK2^−/−^ animals (MARS2 mice). The deletion of the ARF gene was undertaken at 6 weeks of age by intraperitoneal injection of 15 mg/kg of PolyI:polyC every second day for a total of 3 doses and confirmed by PCR (Additional file [1](#MOESM1){ref-type="media"}: Figure S1). All mice were obtained or were backcrossed onto on a C57Bl6 background. Experimental mice were monitored for up to 400 days. Mice were defined as having ALL when at the time of death the bone marrow and spleen primarily consisted of B220^+^CD19^+^Gr1^−^ cells. Survival was analysed using the Kaplan-Meier method and SPSS Statistics software.
Mice were genotyped by PCR on genomic DNA obtained from ear punches using DirectPCR Lysis Reagent (Ear) (Viagen Biotech, Los Angeles CA) with 0.4 mg/mL proteinase K (Promega, Alexandria, NSW, Australia) (complete lysis solution). Ear punches from mice were incubated in complete lysis solution for 2 h at 56 °C and proteinase K was inactivated for 30 min at 85 °C prior to PCR. Deletion of ARF was detected in genomic DNA obtained from spleen cells recovered from culled mice. PCR reactions were performed using MyTaq DNA polymerase (Bioline, Eveleigh NSW Australia) and specific primers as indicated in Additional file [1](#MOESM1){ref-type="media"}: Table S1. The IL-2 PCR was used as a positive DNA control for the Mx1.Cre reaction. The PCR conditions were 95 °C for 1″, then 95 °C for 15″, 58 °C for 15″, 72 °C for 20″ for 35 cycles, 72 °C for 5′. Amplified products were separated on a 2% agarose (Sigma-Aldrich) gel stained with Midori Green Nucleic Acid solution (Bulldog Bio Inc., Portsmouth NH) and visualised using ChemiDoc MP Imaging System (Bio-Rad, Hercules, CA).
Flow cytometry {#Sec4}
--------------
Flow cytometry was performed using a FACSCanto 6-colour flow cytometer (BD Biosciences, San Jose CA). The following antibodies were purchased: Sca-1-PE-Cy7, c-Kit-APC, CD43-APC, IgM-PCP.Cy5.5, IgM-Biotin (Australian Biosearch, WangarraWA, Australia); B220-APC.Cy7, B220 PE-Cy5, CD11b-PE, CD11b-FITC, CD19-PE, CD19-APC.Cy7, Gr1-FITC, Streptavidin APC and Lineage Cocktail of biotinylated CD3, Gr-1, Ter119, B220 and CD11b (BD Biosciences, San Jose CA), and Streptavidin Pacific Blue (Thermofisher Scientific, North Ryde, NSW, Australia). Cells were labelled with antibodies as previously described \[[@CR30]\].
Histology and image acquisition {#Sec5}
-------------------------------
Blood films were prepared and stained with a Romanowsky stain. Tissues were fixed in 10% formalin, embedded, sectioned and stained as previously described \[[@CR38]\]. Femurs were decalcified prior to embedding as previously described \[[@CR38]\]. Images were obtained using a NanoZoomer Slide Scanner (SDR Scientific, Sydney Australia) or an Olympus BX51 microscope with images captured using a Spot RT slider camera (Diagnostic Instruments, Sterling Heights, MI) and SPOT Advanced software. Composite figures prepared using Adobe Photoshop software.
Results {#Sec6}
=======
Deletion of ARF in Rag1 deficient mice predisposes to ALL {#Sec7}
---------------------------------------------------------
Mice lacking ARF are known to develop malignancies with an increased incidence \[[@CR31]\]. To generate an ALL model we bred Mx1.Cre.ARF^fl/fl^ mice with Rag1^−/−^ mice to generate Mx1.Cre.ARF^fl/fl^.Rag1^−/−^ mice. At 6 weeks of age mice received 3 injections of polyI:polyC to delete the ARF gene producing Mx1.Cre.ARF^−/−^.Rag1^−/−^ (MAR) mice.
Rag1^−/−^ mice with deleted ARF (MAR mice) survived for up to 304 days (median 193 days) (Fig. [1a](#Fig1){ref-type="fig"}). The most common cause of death was B lineage ALL, which occurred in 61% of mice between 119 and 243 days with a median of 192 days. The remaining animals succumbed to a number of causes including other haematological malignancies, with the most common feature of non-ALL deaths being massively enlarged pale livers that sometimes contained defined tumors (Fig. [1b](#Fig1){ref-type="fig"}). However the origin of the tumors could not be determined with certainty. Many appeared to be haematological in origin based on morphology but the bone marrows mostly appeared normal (Additional file [1](#MOESM1){ref-type="media"}: Figure S2). Flow cytometric analysis of cells recovered from the bone marrow and spleens of these animals was generally uninformative.Fig. 1MAR mice develop malignancies with B lineage ALL predominating. **a** Kalpan-Meyer analysis showing the survival of MAR mice. **b** Mouse culled due to disease other than ALL showing tumors in the liver (white arrows) and an enlarged spleen (black arrow). **c** Mouse culled due to ALL showing enlarged spleen (black arrow). **d** Blood film from a mouse with ALL showing circulating lymphoblasts. Image acquired using a slide scanner and size bar represents 100 μm. Lower imaged taken on a Spot camera, original magnification 600×. **e** Flow cytometric analysis of bone marrow and spleen cells from mice culled due to ALL. Upper panels are from the same mouse. Central panels show the lowest and highest CD11b expression detected. Lower panels show typical expression of maturation markers B220, CD19, CD43 and surface IgM. Quadrants were set based on control stained cells from the same animal. **f** Section of liver from a mouse culled due to ALL showing both perivascular (thin arrow) and diffuse (thick arrow) infiltration by ALL cells. The degree of infiltration in this animal was typical. Image acquired using slide scanner and size bar indicates 250 μm
Mice that developed ALL were easily identified, demonstrating weight loss, reduced activity and/or impaired use of hind limbs and tail. One displayed hydrocephaly, with fitting. Necropsy findings were consistent with B lineage ALL with enlarged spleens and often enlarged livers, without evidence of tumors and a normal dark red colour (Fig. [1c](#Fig1){ref-type="fig"}). Mice with ALL also had elevated WBC for immune-compromised mice (median 15.2, range 2.1-286.5 cells/mL) with significant numbers of lymphoblasts present in blood smears (Fig. [1d](#Fig1){ref-type="fig"}). Lymph nodes were rarely involved with only 2 mice having visible nodes on cull and only 1 of those having significant lymphadenopathy (Additional file [1](#MOESM1){ref-type="media"}: Figure S3). Cells in the spleen and bone marrow were mostly B220 and CD19 positive (average of 73%, range 56-87 and 86%, range 73-97 respectively), lacking staining for the myeloid marker Gr1 and the T cell marker CD3, however CD11b was detected on cells from some animals (Fig. [1e](#Fig1){ref-type="fig"}). Cells from all mice with ALL were positive for immature marker CD43 and most expressed IgM on at least a proportion of the cells (Fig. [1e](#Fig1){ref-type="fig"}). The lack of lymph node involvement in the vast majority of animals, near complete replacement of the bone marrow by lymphoblasts as well as the expression of the immature marker CD43 and low expression of IgM indicate a pro- to pre-B classification of these leukemias. Other organs, primarily the liver, were infiltrated with lymphoblasts (Fig. [1f](#Fig1){ref-type="fig"}). ALL induced death tended to be earlier compared to non-ALL deaths, with the latter occurring between 68 and 304 days with a median of 229 days, although this was not statistically significant, *p* = 0.06) (Additional file [1](#MOESM1){ref-type="media"}: Figure S4). Animals that did not develop ALL mostly presented with solid tumors at a slightly later time point.
Deletion of SphK2 reduced the incidence of B ALL {#Sec8}
------------------------------------------------
A cohort of mice lacking ARF and Rag1 was also generated using the same methodology on an SphK2^−/−^ background (MARS2 mice). ARF was similarly deleted at 6 weeks of age by 3 injections of polyI:polyC. These mice also largely succumbed to conditions consistent with malignant diseases but compared to MAR mice had significantly increased overall survival with deaths occurring between 120 and \> 400 days (one mouse was electively culled disease free at 400 days) with a median of 234 days (*p* \< 0.05) (Fig. [2a](#Fig2){ref-type="fig"}). Notably there were fewer deaths resulting from ALL in MARS2 animals with only 43% of deaths being due to ALL, resulting in a significant increase in leukemia free survival in MARS2 mice (*p* = 0.044) (Fig. [2b](#Fig2){ref-type="fig"}).Fig. 2Loss of SphK2 reduces the incidence of B lineage ALL. **a**-**c** Kaplan-Meier plots showing all (**a**) and ALL-induced (**b**) deaths. Deaths due to causes other than ALL are illustrated in (**c**). Total WBC (**d**, left panel) and ALL blast counts (**d**, right panel) at the time of sacrifice are shown. \# indicates *p* \< 0.05. **e** Mouse culled due to ALL showing enlarged spleen (black arrow). **f** Blood film from a mouse with ALL showing circulating lymphoblasts. Image acquired using a slide scanner and size bar represents 100 μm. **g** Section of liver from a mouse culled due to ALL showing both perivascular (thin arrow) and diffuse (thick arrow) infiltration by ALL cells. The degree of infiltration in this animal was typical. Image acquired using slide scanner and size bar indicates 250 μm
The absence of SphK2 did not alter the nature of the ALL that developed, with latency, phenotype and disease dissemination being similar. Death due to ALL was slightly delayed in MARS2 mice (range 169 -- 253, median 219.5 days), however this was not significantly different from that of MAR mice (Fig. [2c](#Fig2){ref-type="fig"}). Interestingly the WBC in the leukemic MARS2 mice was significantly lower than in the MAR mice, as was the number of circulating blasts (Fig. [2d](#Fig2){ref-type="fig"}), however the blast percentage was similar between the two groups. Otherwise the disease was identical in MARS2 and MAR mice, with similar enlargement of spleen and liver and infiltration of other organs (Fig. [2e--g](#Fig2){ref-type="fig"}).
Discussion {#Sec9}
==========
Inhibition of sphingosine kinases has recently become of interest for the treatment of a number of conditions including malignant disease \[[@CR39]\]. Clinical trials for the SphK2 inhibitor ABC294640, are well under way with phase I studies complete \[[@CR29]\] and phase I/II and phase II trials examining hepatocellular carcinoma, Kaposi sarcoma as well as the haematological malignancies multiple myeloma and diffuse large B cell lymphoma ongoing (NCT02229981, NCT02939807 and NCT02757326). These trials have been supported by recent preclinical data from a number of groups \[[@CR23], [@CR24], [@CR26], [@CR30], [@CR40]--[@CR44]\]. The majority of these studies have focussed on solid tumors, however there are reports in haematological malignancies including multiple myeloma \[[@CR26]\] and T-ALL \[[@CR45]\], and we have previously reported a role for SphK2 in B lineage ALL \[[@CR30]\] using a BCR/ABL1-dependent model. In this study, we examined the effects of SphK2 gene deletion on the development of ALL in a model that is not dependent on forced expression of BCR/ABL1 and demonstrated that genetic deletion of SphK2 also inhibits the development of B lineage ALL independent of BCR/ABL1 expression. The similar latency and features of the disease in MAR and MARS2 mice suggests that the principal effect of SphK2 loss was on leukemia initiation rather than rate of disease progression. However, we previously demonstrated that the SphK2 inhibitor ABC294640 impedes disease progression in a xenograft model of Ph^−^ human ALL, suggesting that SphK2 loss/inhibition has some effect on disease progression \[[@CR30]\].
The reason why loss of SphK2 decreases the incidence of ALL is not entirely clear. However SphK2 has a well-established role in promoting malignant cell survival \[[@CR46]\] making it possible that in the absence of SphK2, cells with newly acquired potentially oncogenic changes are more susceptible to cell death. While precise mechanisms are yet to be determined, one potential explanation relates to CDKN1A expression. CDKN1A is an inhibitor of apoptosis induced in response to DNA damage whose expression is increased by SphK2-mediated effects on histone acetylation \[[@CR47]\]. In the absence of SphK2, induction of CDKN1A expression following DNA damage could be reduced increasing the probability of cell death. Another possible mechanism relating loss of SphK2 to the reduced incidence of ALL concerns the localization of SphK2 to the endoplasmic reticulum (ER) membrane and its involvement in sphingolipid metabolism at this site. We have recently demonstrated that inhibition of SphK2 induces unrecoverable ER stress leading to apoptosis of multiple myeloma cells and this ER stress-inducing mechanism is most likely also applicable to a range of cell types, including those of ALL, thus impacting on its development in our model \[[@CR48]\].
The lower WBC in leukemic MARS2 was interesting and although altered trafficking of lymphoid cells in SphK2^−/−^ animals might be an explanation for this observation, previous reports have demonstrated increased plasma sphingosine-1-phosphate (S1P) and resultant increased lymphocyte mobilization in SphK2^−/−^ mice \[[@CR49]\]. All but one MARS2 mouse that did not develop ALL went on to develop solid tumors at a time closer to the previously reported latency (median of 266 days) for solid tumors in ARF deficient animals \[[@CR31]\]. Since the tumors that emerged in this study could not be definitively classified, it is not possible to comment on the effects of SphK2 loss on the development of other malignancies.
Conclusions {#Sec10}
===========
We have previously demonstrated the role of SphK2 in ALL driven by BCR/ABL1 and the potential therapeutic application of SphK2 inhibitors in this disease. In this study we demonstrate that SphK2 also plays a role in the development of BCR/ABL1 negative ALL with genetic deletion of SphK2 reducing disease incidence. These findings further support and broaden the potential application of SphK2 inhibitors in the treatment of ALL.
Additional file {#Sec11}
===============
Additional file 1:Additional Data: Table S1, Figures S1-4. (DOCX 6406 kb)
ALL
: Acute lymphoblastic leukemia
AML
: Acute myeloid leukemia
ARF^−/−^
: Mice lacking the p19ARF product of the INK4a/ARF locus
ARF^fl/fl^
: Mice where the ARF gene had been floxed
ER
: Endoplasmic reticulum
MAR
: Mx1.Cre.ARF^fl/fl^.Rag1^−/−^
MARS2
: Mx1.Cre.ARF^fl/fl^.Rag1^−/−^.SphK2^−/−^
S1P
: Sphingosine-1-phosphate
SphK
: Sphingosine kinase
**Electronic supplementary material**
The online version of this article (10.1186/s40364-018-0120-4) contains supplementary material, which is available to authorized users.
Flow cytometry was performed in the Flow Cytometry Core Facility that is supported by Westmead Institute for Medical Research, Westmead Research Hub, Cancer Institute New South Wales and National Health and Medical Research Council. Histology was performed in the Histology Platform at Westmead Institute for Medical Research with the assistance of Virginia James.
Funding {#FPar1}
=======
This work was supported by an NHMRC Senior Research Fellowship (1042305) and Cancer Institute NSW Fellowship.
Availability of data and materials {#FPar2}
==================================
All data generated or analysed during this study are included in this published article \[and its supplementary information files\]. The mice are available through Australian BioResources as cryopreserved embryos.
LB, CW-B and KB made substantial contributions to conception and design of the study. DT, LB and CW-B designed the breeding strategies required for the development of the animals used in this study. VX and LB were responsible for data acquisition, analysis and interpretation of data. LB drafted the manuscript and all authors made significant contributions to revising the final document. All authors read and approved the final manuscript.
Ethics approval and consent to participate {#FPar3}
==========================================
The experiments reported here were conducted with the approval of the Animal Ethics Committee of the Western Sydney Local Health District - approval number 5107.
Consent for publication {#FPar4}
=======================
Not Applicable.
Competing interests {#FPar5}
===================
The authors declare that they have no competing interests.
Publisher's Note {#FPar6}
================
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
| {
"pile_set_name": "PubMed Central"
} |
1. Introduction {#sec1}
===============
Spinal cord injury (SCI) usually causes unreversible motor and sensory impairments. The incidence of SCI is 40 to 80 new cases per million people per year from all causes, depending on the country. For traumatic SCI, the ratio of male-to-female is around 2 : 1 \[[@B1]\]. SCI results in weak or paralyzed muscles, atrophy, walking disability, sensory dysfunction, and autonomic disorders such as autonomic dysreflexia \[[@B2]\]. Spasticity and pain are also some consequences of SCI affecting locomotor and quality of life \[[@B3]\].
The prevalence of spasticity after SCI is 65% at discharge from hospital \[[@B4]\]. In chronic stage, the prevalence is higher. Andresen et al. reported that, in chronic SCI, 71% of patients had spasticity, from the self-reported questionnaire \[[@B5]\]. Severe spasticity is not only detrimental to patients\' walking and motor function \[[@B6]\] but is also related to the presence of pain, lower quality of life, and daily activities \[[@B7], [@B8]\]. Dipiro and colleagues reported that the self-reported frequency of medication usage on spasticity did not significantly decrease from baseline to 5 years of follow-up in chronic SCI \[[@B9]\]. Therefore, finding a treatment strategy that can decrease spasticity and the use of medication might be beneficial to people with SCI.
The prevalence of chronic pain is high in people with SCI. Previous studies reported that the prevalence was around 84% and 73% in Canada and Denmark, respectively \[[@B5], [@B10]\]. Musculoskeletal pain is the most common type of chronic pain and presents early following spinal cord injury \[[@B11], [@B12]\]. The proportion of patients feeling at-level neuropathic pain is higher than below-level neuropathic pain \[[@B12]\]. Pain is highly correlated with poor mood, self-perceived health \[[@B12]\], physical functioning \[[@B13]\], and low quality of life \[[@B5]\] in SCI.
Walking ability is one of the rehabilitation goals of people with SCI, especially in people with incomplete injury. To achieve functional walking, patients require not only appropriate muscle strength and nerve innervation but also proper endurance and less fatigue. Fatigue impacts on function in 57% of individuals with SCI \[[@B14]\]. It is also more prevalent among younger SCI and SCI with shorter duration of disability \[[@B15]\]. Clinically, there are several commonly used measurement tools, such as 6-minute walk test (6MWT), 10-meter walk test (10MWT), timed up and go (TUG), Walking Index for Spinal Cord Injury (WISCI), and Functional Independence Measure-Locomotion (FIM-L), each assesses different aspects of walking ability. For example, 10MWT and 6MWT have been shown to be valid and reliable to measure ambulatory ability for individuals with SCI \[[@B16]\], and 6MWT has been suggested to be a good assessment tool of endurance \[[@B16], [@B17]\]. TUG is a simple and quick test to assess a person\'s mobility and balance and correlates well with gait speed for frail elderly \[[@B18]\] and endurance in chronic stroke \[[@B19]\]. FIM-L and WISCI address on the need of assistance when performing functional tasks. FIM-L measures the functional status (walking/wheelchair and stairs) of a person based on the level of assistance he or she requires \[[@B20]\]; it can be considered as an evaluation of the gait ability in activities of daily living. WISCI scores the walking ability according to the need for physical assistance, braces, and walking aids \[[@B21]\]. Manual muscle testing (MMT) assesses lower extremity motor score (LEMS) according to American Spinal Injury Association (ASIA).
Rehabilitation for improving pain, spasticity, and walking ability is always a challenge for clinicians. The use of robot-assisted gait training (RAGT) in the field of rehabilitation has become more widespread since this training is not limited by the individuals\' muscle paralysis level. Current systems of RAGT include Lokomat (Hocoma AG, Switzerland), G-EO systemTM (Reha Technology AG, Switzerland), Walkbot (P&S Mechanics Co., Ltd, Korea), and ReoAmbulatorTM (Motorika, USA Inc.) \[[@B22]\]. RAGT provides repetitive and functional task training which induces greater activation of the sensorimotor cortex (S1, S2) and cerebellar regions \[[@B23]\]. A meta-analysis revealed that RAGT improved walking endurance, walking independence, and lower limb muscle strength, but did not reduce spasticity \[[@B24]\]. Other than task-specific training, RAGT provides proprioceptive inputs to lower extremities. According to gate control theory, large fiber activation might be able to block noxious small fiber afferents which cause pain and spasticity. Previous studies revealed that sensorimotor activity by treadmill training decreased pain behavior and nociceptive fiber density in the spinal dorsal horn in acute, subchronic, and chronic SCI mice model \[[@B25], [@B26]\]. Previous studies also reported that rhythmic passive movement could induce spinal circuitry reorganization, restore postactivation depression, and decrease spasticity in patients with SCI \[[@B27]\]. Therefore, it is plausible to hypothesize that RAGT can reduce pain and spasticity. In the past, much work has been done on investigating the effect of RAGT on walking performance, but reports of its effect on pain and spasticity were rare.
The purpose of this meta-analysis was to compare the effects of RAGT on spasticity and pain with those of other treatments after SCI.
2. Methods {#sec2}
==========
This review integrated the results from relevant studies by following the systematic review and meta-analysis guidelines outlined in the Preferred Reporting Items for Systematic Review and Meta-Analysis (PRISMA) statement \[[@B28]\].
2.1. Types of Participants {#sec2.1}
--------------------------
This study included only SCI subjects, regardless of traumatic or nontraumatic lesion, the time since injury, age, and sex.
2.2. Types of Interventions {#sec2.2}
---------------------------
Any kind of RAGT compared with other training modalities or no training for lower limbs was included.
2.3. Outcome Measures {#sec2.3}
---------------------
Primary outcome measures were spasticity and pain. Spasticity was assessed by modified Ashworth scale (MAS) or Ashworth scale (AS) for lower limbs. Pain was assessed by the visual analog scale (VAS). The VAS is widely used to assess self-perceived pain \[[@B29]\]. It is a 10-centimeter line in which 0 represents no pain and 10 at the right edge means intolerable pain. Participants subjectively reported their pain condition on the VAS scale.
Secondary outcome measures were LEMS and walking ability assessed by 6MWT, 10MWT, TUG, WISCI, and FIM-L. LEMS assessed motor score for lower limbs according to ASIA standard. 6MWT measured the walking distance in 6 minutes. 10MWT assessed the walking speed measuring the time necessary to walk 10 meters. 6MWT and 10MWT were reliable and responsive tools in assessing walking ability in incomplete SCI \[[@B30]\]. TUG assessed the time that a subject took to rise from a chair, walk three meters, turn around, walk back to the chair, and sit down. WISCI measured improvements in ambulation in persons with SCI by evaluating the amount of physical assistance, braces, or devices required to walk 10 meters. WISCI I scored from 1 to 19 and WISCI II from 1 to 20 \[[@B31]\]. FIM was an 18-item assessment of physical, psychological, and social function. The assessor graded the functional status of a person based on the need of assistance \[[@B32]\].
2.4. Type of Studies {#sec2.4}
--------------------
RCT, non-RCT, and crossover trials (only the RAGT period) were included in analysis.
2.5. Searching Criteria {#sec2.5}
-----------------------
The searching criteria were limited to human studies and English language.
2.6. Data Sources {#sec2.6}
-----------------
Four electronic databases (PubMed, Scopus, Medline (Proquest), and Cochrane Central Register of Controlled Trials (CENTRAL)) before November 2019 using Medical Subject Heading terms combined with keywords, such as robotics, spinal cord injury, pain, and spasticity, were processed in title, abstract, and keywords. [](#supplementary-material-1){ref-type="supplementary-material"} shows the combinations used.
2.7. Study Selection {#sec2.7}
--------------------
Two authors independently searched and screened the titles, abstracts, and literatures to identify potentially relevant studies. Then, full texts of relevant studies were obtained and assessed to determine whether the articles met the inclusion criteria. Any disagreement was discussed and solved with the third author to reach a consensus in every relevant detail.
2.8. Data Extraction and Management {#sec2.8}
-----------------------------------
Two authors extracted data independently from included studies and filled into an extraction form. The following data were extracted: (1) authors; (2) year of publication; (3) study design; (4) inclusion/exclusion criteria; (5) subject demographics (age, gender, number of subjects, level of lesion, classification of ASIA, duration of injury); (6) intervention; (7) outcome measures; and (8) summary of the results.
Data at baseline and at the end of the intervention were extracted for the analysis of the effect of training. Measurements during the intervention or at follow-up were excluded due to inconsistent measuring time points used across different studies. Studies were excluded if necessary outcome measures were missing or not measured.
2.9. Quality Assessment {#sec2.9}
-----------------------
The methodological quality of the selected RCTs was independently assessed by two authors using the Cochrane risk of bias assessment tool \[[@B33]\]. For the assessment of the methodological quality of the selected cohort studies and clinical trials, the Newcastle Ottawa Scale \[[@B34], [@B35]\] was employed and done by two authors independently. Any disagreement was resolved through discussion and consensus with the third author.
2.10. Statistical Analysis {#sec2.10}
--------------------------
RCTs and non-RCTs were grouped and analyzed separately. Statistical analysis was performed using Comprehensive Meta-Analysis (CMA) version 3 to analyze the treatment effect. Mean differences and 95% confidence interval (CI) were calculated for each primary and secondary outcome. Random effect models were used to calculate the pooled mean difference estimates if heterogeneity occurred. Fixed effects models were used to calculate the pooled mean difference estimates if no heterogeneity occurred.
3. Results {#sec3}
==========
3.1. Studies Included {#sec3.1}
---------------------
A total of 223 studies from electronic databases and two studies from the reference lists of included studies were identified. In these, 105 of the selected studies were duplicates and thus were removed from analysis. Out of the retained studies, 18 studies were retained for quality synthesis which included 7 RCTs and 11 non-RCTs. The flow of studies through the review process is shown in [Figure 1](#fig1){ref-type="fig"}. Six studies were included after review for quantitative synthesis of which the characteristics are shown in [Table 1](#tab1){ref-type="table"} and [Table 2](#tab2){ref-type="table"}.
### 3.1.1. Excluded Studies {#sec3.1.1}
After screening, a total of 66 studies were eliminated. The reasons for exclusion were as follows: texts not in English version, manuscripts in the form of education page, subjects of the studies included other diagnostic groups, study purpose, and outcome measures did not meet our inclusion criteria.
### 3.1.2. Study Location {#sec3.1.2}
From the 18 studies, 7 trials were done in the United States \[[@B36]--[@B42]\], 2 in Spain \[[@B43], [@B44]\], 2 in Switzerland \[[@B45], [@B46]\], 2 in Canada \[[@B47], [@B48]\], 2 in Italy \[[@B49], [@B50]\], 2 in Japan \[[@B51], [@B52]\], and one in Germany \[[@B53]\].
### 3.1.3. Study Participants {#sec3.1.3}
A total of 222 participants from 7 RCTs were included. Seventy-nine participants from 11 non-RCTs were included. Although age was not reported in all included studies, the participants\' age of RCTs and non-RCTs ranged from 34 \[[@B46]\] to 59 \[[@B45]\] and 19 \[[@B51]\] to 62 \[[@B52]\] years, respectively. One RCT \[[@B36]\] and one non-RCT \[[@B39]\] did not report the proportion of gender. For other included RCTs, the proportion between males and females was 101 : 79. For non-RCTs, the proportion between males and females was 62 : 15.
The ASIA level was B, C, or D in RCTs and A, B, C, or D in non-RCTs. The level of injury was cervical (C1-C8) in 80 participants, thoracic in 67, C2-T9 in 46, and above T10 in 30 participants in RCTs. The cervical level of injury was C3 to C7 in 16 participants, thoracic (T3-T12) in 48, lumbar (L1-L5) in 14, and T12-L1 in one participant in non-RCTs.
3.2. Interventions {#sec3.2}
------------------
The intervention of RAGT was 3 to 5 sessions per week, 30 min to one hour for 4 to 12 weeks in RCTs. The training protocol of non-RCTs was 2 to 5 sessions, 30 min to 90 min for one week to 90 days.
The apparatus used for RAGT in these studies included Lokomat, hybrid assistive limb (HAL), Indego Exoskeleton, ReWalk, ARKE 2.0, and Ekso GT in which all included RCTs used Lokomat for training.
3.3. Risk of Bias of the Included Studies {#sec3.3}
-----------------------------------------
[Figure 2](#fig2){ref-type="fig"} summarized the risk of bias judgements related to all RCTs. In all included RCTs, only one study \[[@B36]\] had high risk of bias level. Five studies \[[@B37], [@B38], [@B44]--[@B46]\] reported randomization. Two studies did not mention randomization \[[@B36]\] or were unclear \[[@B47]\]. Allocation concealment was fulfilled by two studies \[[@B44], [@B45]\]. Three of the included studies \[[@B44], [@B45], [@B47]\] had blinding and the other three studies \[[@B37], [@B38], [@B46]\] did not report the methodology of allocation concealment. Two studies \[[@B45], [@B47]\] did intention to treat analysis and the other three studies \[[@B37], [@B38], [@B44]\] were unclear on the information about attrition bias. Two studies \[[@B36], [@B46]\] had high risk of attrition bias for not reporting their management on the drop-out data.
[Table 3](#tab3){ref-type="table"} summarized the risk of bias judgements related to non-RCTs. All studies had general to good quality. All non-RCTs recruited representative SCI subjects and no control group. All studies had secure record on training protocol but one study \[[@B39]\] did not. All studies assessed outcomes independently. The duration of three non-RCTs \[[@B43], [@B50], [@B52]\] was from 40 min to 2 weeks, most trials with 8 weeks.
3.4. Effects of the Interventions {#sec3.4}
---------------------------------
Pain and walking ability were analyzed in RCTs. Spasticity, pain, and walking ability were analyzed in non-RCTs. Summarization on spasticity and TUG in RCTs and FIM-L in non-RCTs were done without meta-analysis due to insufficient data information.
3.5. Results of Primary Outcomes: Spasticity {#sec3.5}
--------------------------------------------
Four RCTs \[[@B36], [@B38], [@B44], [@B46]\] assessed spasticity. However, different muscle groups were assessed in these studies; therefore, the data could not be pooled together. In these studies, all participants\' spasticity was mild (MAS 0-2) and none of them changed significantly after RAGT.
Six eligible non-RCTs were included but only 4 studies\' data were retained to pool for analysis. One trial was excluded for analysis because it assessed spasticity of 36 joints together \[[@B51]\]. The other one trial was excluded for analysis because the participants in this trial had no spasticity \[[@B52]\]. Out of the four non-RCTs analyzed, 2 studies \[[@B49], [@B50]\] use MAS as their outcome measure on 28 participants. The other two studies \[[@B42], [@B43]\] used AS to assess 23 participants for spasticity ([Figure 3](#fig3){ref-type="fig"}). The robotic group showed significant decrease in MAS (95%CI = −2.886 to -1.412, *p* ≤ 0.001) and AS (95%CI = −0.202 to -0.068, *p* ≤ 0.001) measures. The pooled mean difference using MAS and AS (fixed effects model) were -2.149 and -0.135, respectively.
3.6. Results of Primary Outcomes: Pain {#sec3.6}
--------------------------------------
Two RCTs \[[@B44], [@B45]\] and 3 non-RCTs \[[@B43], [@B49], [@B50]\] were included for analysis. Eighty-four and 31 participants were involved in RCT and non-RCT studies, respectively. [Figure 3](#fig3){ref-type="fig"} showed the results on the analysis of the primary outcomes of pain after RAGT. Although the trend for pain reduction was in favor of robotic group, there was no significant difference between robotic and control group, regardless of RCTs (*p* = 0.427) or non-RCTs (*p* = 0.239). The pooled mean difference (random effects model) of RCTs and non-RCTs were -0.890 and -1.676, respectively. The level of pain ranged from painless \[[@B44]\] to moderate \[[@B49], [@B50]\] in all included studies.
3.7. Results of Secondary Outcomes: LEMS and Walking Ability {#sec3.7}
------------------------------------------------------------
### 3.7.1. LEMS {#sec3.7.1}
Three RCTs \[[@B36], [@B44], [@B45]\] included 104 participants, and three non-RCTs \[[@B42], [@B52], [@B53]\] with 30 participants were pooled for analysis. Significant positive effect in favor of robotic group in both RCTs (95%CI = 1.143 to 2.732, *p* ≤ 0.001) and non-RCTs (95%CI = 1.508 to 4.839, *p* ≤ 0.001) were shown in the results of LEMS. The pooled mean differences (fixed effects model) were 1.938 and 3.173 for RCTs and non-RCTs, respectively ([Figure 4](#fig4){ref-type="fig"}).
### 3.7.2. 6MWT {#sec3.7.2}
Four RCTs \[[@B36], [@B38], [@B44], [@B47]\] and 4 non-RCTs \[[@B42], [@B43], [@B49], [@B53]\] assessed 6MWT. A total of 140 and 38 participants were involved in analysis in the RCTs and non-RCTs, respectively. Regardless of RCTs or non-RCTs, walking distance in 6MWT increased significantly in favor of robotic group (RCTs: 95%CI = 4.394 to 106.628, *p* = 0.033; non-RCTs: 95%CI = 7.218 to 52.586, *p* = 0.010). The pooled mean difference (random effects model) of RCTs and non-RCTs were 55.511 m and 29.902 m, respectively ([Figure 4](#fig4){ref-type="fig"}).
### 3.7.3. 10MWT {#sec3.7.3}
Five RCTs were included in this analysis, but the data of only four studies were pooled. One study \[[@B38]\] was excluded because no data of control group were provided. Four RCTs \[[@B44]--[@B47]\] and 5 non-RCTs \[[@B42], [@B43], [@B49], [@B52], [@B53]\] were used for subsequent data analysis ([Figure 4](#fig4){ref-type="fig"}). In these, 117 and 40 subjects of RCTs and non-RCTs were included, respectively. 10MWT significantly improved in robotic group of non-RCTs (95%CI = 0.032 to 0.213, *p* = 0.008) but not of RCTs (*p* = 0.597). The pooled mean difference (random effects model) for non-RCTs was 0.123 m/s.
### 3.7.4. TUG {#sec3.7.4}
Though one RCT \[[@B38]\] used this outcome measure, there was no sufficient data for analysis. Data from three non-RCTs \[[@B42], [@B49], [@B53]\] included 35 participants who were pooled for analysis. The result showed significant improvement in favor of robotic group (95%CI = −33.232 to -15.659, *p* ≤ 0.001). The pooled mean difference (fixed effects model) was -24.446 s ([Figure 4](#fig4){ref-type="fig"}).
### 3.7.5. WISCI {#sec3.7.5}
Five RCTs and 3 non-RCTs were included for this analysis, but 2 RCTs and one non-RCT were excluded for insufficient data provided (one RCT \[[@B38]\] and one non-RCT \[[@B42]\]). Data variability of one RCT \[[@B46]\] was too dispersed. Data of three RCTs \[[@B36], [@B44], [@B45]\] and 2 non-RCT ones \[[@B52], [@B53]\] with 104 and 10 participants for RCT and non-RCT, respectively, were finally pooled into analysis, and the results showed no significant difference (*p* = 0.265 for RCTs; *p* = 0.228 for non-RCTs) ([Figure 4](#fig4){ref-type="fig"}).
### 3.7.6. FIM-L {#sec3.7.6}
Only 2 RCT studies \[[@B36], [@B44]\] assessed FIM-L scale. The analysis included 95 subjects. The pooled result showed no significant difference between two groups (*p* = 0.122). The pooled mean difference (random effects model) was 1.853 ([Figure 4](#fig4){ref-type="fig"}). For the included non-RCTs, none of them reported FIM-L.
3.8. Publication Bias {#sec3.8}
---------------------
[](#supplementary-material-1){ref-type="supplementary-material"} demonstrated the funnel plots of VAS, 6MWT, 10MWT, TUG, WISCI, and LEMS. There were no funnel plots of MAS and AS due to only two studies of each measurement. It seemed a symmetrical funnel plot of VAS, but small study bias was identified in Egger\'s test (*p* = 0.03289). No small study bias was found in other measurements.
4. Discussion {#sec4}
=============
This meta-analysis showed RAGT decreased spasticity and improved walking ability in individuals with SCI. Furthermore, the level of pain showed no change after RAGT.
4.1. Spasticity {#sec4.1}
---------------
The current meta-analysis revealed that spasticity decreased after RAGT in non-RCTs. Several possible mechanisms could explain the reduction of spasticity after RAGT. Spasticity is defined as a velocity-dependent increase in tonic stretch reflexes with exaggerated tendon jerks \[[@B54]\]. However, spasticity also involves nonreflex component such as intrinsic muscular properties \[[@B54]\]. Mirbagheri et al. reported that RAGT reduced reflex and intrinsic stiffness of ankle in individuals with SCI \[[@B37]\].
RAGT produces rhythmic movements of lower limbs and provides sensory inputs. Previous studies reported that rhythmic passive exercise could induce spinal circuitry reorganization and decrease spasticity in patients with SCI \[[@B27], [@B55]\]. Improving spasticity and locomotor function by the activation of spinal locomotor centers might also be influenced by the repetitive elements of the therapeutic program \[[@B56]\]. RAGT is a type of repetitive functional task training. These above mechanisms might possibly explain the finding that RAGT reduces spasticity.
The reasons that the RCTs did not show significant reduction in spasticity in the RAGT group might be due to the floor effect \[[@B44], [@B46]\] (MAS 0 to 1) and the measurements done on different joints \[[@B38], [@B44], [@B46]\]. In the included studies, the subjects\' initial spasticity level was not high enough to show change after RAGT. It is suggested that subjects with more severe spasticity could be recruited for further investigation of RAGT.
4.2. Pain {#sec4.2}
---------
Pain and spasticity are intricate consequences of spinal cord injury \[[@B57]\]. Researchers suggested that pain and spasticity are closely linked \[[@B57]\]. In addition, pain and spasticity might share similar pathophysiological mechanisms \[[@B5]\]. Hence, it is reasonable to expect a reduction of pain with spasticity reduction after RAGT. However, the result of this meta-analysis did not show significant decrease of pain accompanying reduced spasticity following RAGT. This might be that the pain suffered by the participants in these included trials was not mainly from spasticity. Some other potential sources, such as muscle soreness due to excess exercise, joint pain due to malposture, or poor biomechanics, might be the cause of pain. One should also note that the neuropathic pain, more than 50% prevalence in spinal cord injured persons \[[@B58]\], was not reported in the included studies. Therefore, they lacked source of data for meta-analysis. It is suggested to be investigated in future studies.
Although, the included RCTs and non-RCTs did not show significant change in pain, the trend favored RAGT group. Past studies revealed that physical activities could relieve musculoskeletal and neuropathic pain \[[@B59], [@B60]\]. However, participants in the current meta-analysis did not subjectively feel significant alteration in pain level with VAS assessment. The baseline floor effect of mild \[[@B44]\] to moderate \[[@B50]\] intensity of pain felt by the participants might account for the nonsignificant result. Future studies that include participants with higher level of pain at baseline are suggested.
4.3. LEMS {#sec4.3}
---------
This study showed that LEMS significantly improved after RAGT. As discussed previously, rhythmic muscle activations could be detected during RAGT. In addition, weight bearing may be an important factor. RAGT provides support which allows users to load their weight on lower limbs during training. Lower limbs weight bearing and the enhancement of muscle activation may contribute to the improvement of LEMS.
Decreasing guiding force as RAGT progresses might increase the muscle strength of lower extremities. Subjects needed greater engagement to activate muscles and participate in the training program. Since the guiding force has not been quantified in the included studies, investigation of the relationship between guiding force and the improvement of LEMS is suggested in future studies.
4.4. 6MWT {#sec4.4}
---------
This meta-analysis showed that 6MWT increased significantly in favor of the RAGT group. 6MWT is an indicator of endurance. Clinically meaningful change (CMC) of 6MWT was 19-22 m in healthy older adults \[[@B61]\]. The 95% CI of the current meta-analysis includes the range 4.394-106.628 m for included RCTs and 7.218-52.586 m for included non-RCTs. Hence, RAGT can be clinically practical for endurance training.
In physiological point of view, endurance could be improved by multiple sessions of submaximal voluntary exercises \[[@B62]\]. RAGT, due to its lack of active participation from the users during training, was doubted to increase cardiopulmonary fitness in subjects \[[@B63]\]. However, increased 6MWT in this meta-analysis indicates it could improve endurance without emphasizing voluntary muscle contraction. According to Mazzoleni et al. \[[@B64]\], bilateral muscular activity increased after RAGT in people with SCI. Thus, the activation of muscles might increase the challenge to cardiopulmonary systems and, thus, increase the endurance of participants with SCI.
4.5. Walking Speed (10MWT and TUG) {#sec4.5}
----------------------------------
TUG is commonly used to assess functional mobility. It was correlated with muscle strength of the lower extremities and gait speed \[[@B18], [@B65]\]. The current meta-analysis showed that both parameters improved after RAGT in SCI. The results of 10MWT also supported that RAGT increased walking speed in SCI. The CMC of 10MWT was 0.04-0.06 m/s in healthy older adults \[[@B61]\]. The pooled mean difference (0.123 m/s) was above CMC in included non-RCTs in this meta-analysis.
Kim et al. \[[@B66]\] reported that muscle strength of the lower extremities was correlated with walking speed in chronic incomplete SCI. As shown with the result of LEMS\'s improvement, lower extremity strength might be the cause of improved walking speed.
Another explanation for improved walking speed was the strengthening of central pattern generator (CPG). Previous studies supported that RAGT, which involved rhythmic activations of lower extremities, could strengthen CPG \[[@B67]\]. CPG is an essential neural mechanism of walking. Enhanced CPG would lead to increase walking ability. The other explanation was the reduction of spasticity. Spasticity could increase the resistance of movement and interfere with gait. As shown in the above result, spasticity reduced after RAGT and, thus, resulted in less resistance during walking.
4.6. WISCI and FIM-L {#sec4.6}
--------------------
WISCI and FIM-L showed no significant difference after RAGT in this meta-analysis. The reason might be that studies investigated that WISCI and FIM-L were few in the current meta-analysis. However, the CMC for the WISCI was 1 point \[[@B68]\]. The pooled mean differences of included studies all exceed CMC (3.383 for RCTs and 5.945 for non-RCTs).
FIM was graded according to the assistance required by a subject. People with SCI might experience fear of falling that impeded transferring the improved walking abilities to functional tasks after RAGT.
4.7. Study Limitations {#sec4.7}
----------------------
The first limitation of this current review is no classification of subgroups according to the level and severity of SCI. More trials and subjects are needed for subgroup analysis. The second limitation is the risk of bias exists in all studies. Compared with the non-RCTs, the number of RCTs studies is relatively few. The third limitation is that the training protocol used in each study is not identical. Further reviews are suggested to compare effects of different protocols with increased trials.
5. Conclusions {#sec5}
==============
This meta-analysis concluded that RAGT had positive effects in improvements of spasticity and walking ability. In RCTs, walking distance and muscle strength of lower limbs improved after RAGT. RAGT can be applied in individuals with SCI without increasing pain.
This work was supported by the Ministry of Science and Technology, Taiwan (grant number MOST 107-2221-E-182-009-MY3) and the Healthy Aging Research Center (EMRPD1K0431), Chang Gung University, Taiwan. Special thanks to the Neuroscience Research Center, Chang Gung Memorial Hospital, Linkou, Taiwan.
SCI:
: Spinal cord injury
RAGT:
: Robot-assisted gait training.
Conflicts of Interest
=====================
We have no conflict of interest.
Supplementary Materials {#supplementary-material-1}
=======================
######
Appendix 1: searching keywords. Appendix 2: funnel plots and Egger\'s test of (a) VAS for non RCTs, (b) 6MWT for 3 RCTs, (c) 6MWT for non RCTs, (d) 10MWT for RCTs, (e) 10MWT for non RCTs, 4 (f) TUG for non RCTs, (g) WISCI for RCTs, (h) LEMS for RCTs, and (i) LEMS for non 5 RCTs.
######
Click here for additional data file.
{#fig1}
{#fig2}
{#fig3}
{#fig4}
######
Characteristics of included RCTs.
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Study Research design Participants Intervention Outcome measures
--------------------------------------- ----------------- --------------------------------------------------------------------------------------- ------------------------------------------------------------------------------------------------------ -------------------------------------------
Alcobendas-Maestro et al., \[[@B44]\] RCT *n* = 75 (37 in the Lokomat training group, 38 in the conventional overground group)\ 40 sessions over 8 weeks, 1 hour\ AS\
ASIA C or D\ (1) Lokomat group: 30 min with the Lokomat in each session+30 min standard physical treatment\ VAS\
Level of injury: C2 to T12 (2) Overground group: one hour standard physical treatment 6MWT, 10MWT\
WISCI II\
FIM-L, LEMS
Hornby et al., \[[@B36]\] RCT *n* = 30\ \(1\) Robotic-assisted BWSTT\ AS\
ASIA B, C, D\ (2) Therapist-assisted BWSTT\ SCATS\
Level of injury: above T10 (3) Overground ambulation with a mobile suspension system three 30-minute sessions per week, 8 weeks 6MWT, 10MWT\
WISCI\
FIM-L\
LEMS, TUG
Labruyère and van Hedel, \[[@B45]\] RCT cross over *n* = 9\ \(1\) Group 1: 16 sessions of RAGT (Lokomat) followed by 16 sessions of strength training\ VAS, 10MWT, WISCI\
ASIA D\ (2) Group 2: the same intervention in reversed order LEMS, UEMS, FET, PCI, gait symmetry, BBS\
Level of injury: C4 to T11 Body sway\
FES-I, SCIM
Lam et al., \[[@B47]\] RCT *n* = 15\ \(1\) Lokomat-resisted BWSTT (Loko-R)\ Reports of pain\
ASIA C or D\ (2) Lokomat-assisted BWSTT (control) 45 min, 3 times/week for 3 mo. 10MWT\
Lesion level below thoracic 11 or lower motoneuron injury was excluded 6MWT\
SCI-FAP
Mirbagheri et al., \[[@B37]\] RCT *n* = 46\ \(1\) RAGT group: 3 times a week over four weeks, one hour/session\ MAS\
ASIA C or D\ (2) Control group: no intervention Intrinsic stiffness K\
Level of injury: C2 to T9 Reflex stiffness G
Varoqui et al., \[[@B38]\] RCT *n* = 30\ \(1\) Lokomat group: 3 times a week over four weeks, one hour/session\ MAS\
ASIA C or D\ (2) Control group 10MWT\
Level of injury: above T10 6MWT\
TUG\
Ankle kinematic and kinetic assessments
Wirz et al., \[[@B46]\] RCT *n* = 21\ \(1\) Intervention group: 50 min/training\ MAS\
ASIA B or C\ (2) Control group: 25 min/training\ SCIM III\
Level of injury: C4 to T12 Lokomat, 3-5 days/week, 8 weeks WISCI II\
Penn, GICS
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
BWSTT: body-weight supported treadmill training; AS: Ashworth scale; MAS: modified Ashworth scale; 10MWT: 10-meter walking test; 6MWT: 6-minute walk test; WISCI: Walking Index for Spinal Cord Injury; FIM-L: Functional Independence Measure-Locomotor section; LEMS: lower extremity motor score; SCIM: Spinal Cord Independence Measure; Penn: Modified Penn Spasm Frequency Scale; TUG: timed up and go test; FET: Figure Eight Test; PCI: Physiological Cost Index; BBS: Berg balance scale; FES-I: falls efficacy scale-international version; UEMS: upper extremity motor score; SCI-FAP: Spinal Cord Injury-Functional Ambulation Profile; SCATS: Spinal Cord Assessment Tools for Spasticity; GICS: Global Impression of Change Scale.
######
Characteristics of included non-RCTs.
---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Study Research design Participants Intervention Outcome measures
--------------------------------- ------------------------------------------------ ------------------------------------ -------------------------------------------------------------------------------------------------------------------------------------------- ------------------------------------------------
Aach et al., \[[@B53]\] Single case experimental A-B (pre-post) design *n* = 8\ HAL\ 6MWT, 10MWT\
ASIA A\ 5 times per week, 90 days, mean number of sessions of 51.75 ± 5.6 TUG\
Level of injury: T8 to L2 WISCI II\
LEMS\
Muscle volume
Del-Ama et al., \[[@B43]\] Pilot study *n* = 3\ Kinesis system\ VAS\
ASIA A and D\ The first week intervention, the second week no intervention AS\
Level of injury: L1, L4, L5 10MWT\
6MWT\
Penn\
MMT
Ekelem and Goldfarb, \[[@B39]\] Case report *n* = 2\ Indego exoskeleton practice \< 4 hr per day MAS
ASIA B\
Level of injury: T4, T11
Esquenazi et al., \[[@B40]\] Prospective, single-intervention *n* = 12\ ReWalk\ Pain, fatigue (VAS)\
ASIA B\ Up to 24 sessions of 60 to 90 min duration over approximately 8 weeks (target was three times per week) AS\
Level of injury: T3-T12 HR, BP\
Skin integrity
Ikumi et al., \[[@B51]\] Case report *n* = 1\ HAL\ MAS\
ASIA A\ 60 min, 2 times per week for 5 weeks in addition to standard physical and occupational therapy Walking time and distance
Level of injury: C4
Lemaire et al., \[[@B48]\] Case report *n* = 2\ ARKE 2.0 LEPE\ Ten-point scale: pain, fatigue\
ASIA B\ 12 half-hour training sessions, four or more weeks Number of steps\
Level of injury: T6, T12 Distance travelled\
Standing duration\
Walking duration\
Number of partial steps
Manella et al., \[[@B41]\] Case report *n* = 1\ Lokomat\ AS\
ASIA A\ 40 mins, 3 times per week, 12 weeks Pendulum test of quadriceps spasticity\
Level of injury: T7 Spasm frequency and severity\
ASIA sensory scores\
LEMS\
FIM
Mazzoleni et al., \[[@B49]\] Single group *n* = 7\ 20 sessions, 3 sessions/week, FES-cycling system (Pegaso) followed by 20 sessions, 3 sessions/wk, overground robotic exoskeleton (Ekso GT) MAS\
ASIA A\ PSFS\
Level of injury: T4-T12 SCIM\
Spasticity and pain through a 0-10 points NRS\
ISCI\
6MWT, 10MWT\
TUG
Stampacchia et al., \[[@B50]\] Single group *n* = 21\ Ekso GT\ Pain and spasticity (0-10 points scale)\
ASIA A, B, D\ 40 min MAS\
Level of injury: C7, L1-L2, dorsal PSFS\
PGIC
Watanabe et al., \[[@B52]\] Case report *n* = 2\ HAL\ MAS\
ASIA C, D\ 3-4 times per week, for a total of 8 sessions, in addition to conventional physical therapy, 20-30 min/session LEMS\
Level of injury: T8-T10, T12-L1 WISCI II\
FIM\
CGS\
Stride, cadence\
Right and left leg swing time\
Hip and knee joint angle\
BI, mRS and adverse effects
Wirz et al., \[[@B42]\] Single group *n* = 20\ Lokomat (DGO)\ Primary: WISCI II, 10MWT, 6MWT, TUG\
AISA = C or D\ 8 weeks, 3 to 5 sessions each week, 45 min Secondary: at 1 center, *n* = 10\
Level of injury: L1 or higher LEMS, AS, SCATS
---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
AS: Ashworth scale; MAS: modified Ashworth scale; 10MWT: 10-meter walking test; 6MWT: 6-minute walk test; WISCI: Walking Index for Spinal Cord Injury; FIM: Functional Independence Measure; LEMS: lower extremity motor score; HR: heart rate; BP: blood pressure; SCIM: Spinal Cord Independence Measure; Penn: Modified Penn Spasm Frequency Scale; TUG: timed up and go test; SCATS: Spinal Cord Assessment Tools for Spasticity; PSFS: Penn Spasm Frequency Scale; ISCI: International Spinal Cord Injury Pain Data Set; PGIC: patient\'s global impression of change; CGS: comfortable gait speed; BI: Barthel index; mRS: modified Rankin Scale.
######
Assessment of study quality with Newcastle-Ottawa scale.
Selection Comparability Outcome
--------------------- ------ ----------- --------------- --------- --- -- -- --- --- --- ---
Aach et al. 2014 ★ ★ ★ ★ ★ ★ 6
del-Ama et al. 2014 ★ ★ ★ ★ ★ 5
Ekelem and Goldfarb 2018 ★ ★ ★ 3
Esquenazi et al. 2012 ★ ★ ★ ★ ★ ★ 6
Ikumi et al. 2017 ★ ★ ★ ★ ★ ★ 6
Lemaire et al. 2017 ★ ★ ★ ★ ★ ★ 6
Manella et al. 2010 ★ ★ ★ ★ ★ ★ 6
Mazzoleni 2017 ★ ★ ★ ★ ★ ★ 6
Stampacchia 2016 ★ ★ ★ ★ ★ 4
Watanabe 2017 ★ ★ ★ ★ ★ 4
Wirz et al. 2005 ★ ★ ★ ★ ★ ★ 6
[^1]: Academic Editor: Redha Taiar
| {
"pile_set_name": "PubMed Central"
} |
All original data from this study have been uploaded to Figshare, and can be found using the following links: Study 1: <https://doi.org/10.6084/m9.figshare.5692852>; Study 2: <https://doi.org/10.6084/m9.figshare.5692933>.
Introduction {#sec001}
============
In quiet standing the ankles play a crucial role in connecting the long, nearly vertical body to the feet, which are used to apply gravitational counteractive torque against the ground. This torque is used to stabilize the inherently unstable body. Ankle torque is produced by passive and active mechanisms. The passive mechanism consists of the visco-elastic forces produced by the stretch of the muscles, tendons and ligaments acting around the ankle, and this operates with zero delay. This is the intrinsic ankle stiffness. The active mechanism is the modulation of ankle muscle activity by the nervous system. Responses to unforeseeable disturbances will be delayed, but usually in quiet standing neural prediction minimizes the delay \[[@pone.0193850.ref001]\]. Intrinsic ankle stiffness is not normally sufficient to stabilize the body by itself \[[@pone.0193850.ref002],[@pone.0193850.ref003]\]. However, by providing passive instantaneous resistance to falling, it supplements actively generated torque and increases the time constant of the unstable body, giving more time for neural intervention \[[@pone.0193850.ref004]\].
Here we investigate intrinsic ankle stiffness for anterior-posterior sway (i.e. in the sagittal plane). The two main contributors to intrinsic ankle stiffness are the Achilles tendon and the triceps surae muscles, which act as springs arranged in series. During quiet standing, the stretch sizes are normally very small and ankle torque is relatively low \[[@pone.0193850.ref002]\]. Ankle stiffness is therefore determined by the combination of a very long and compliant tendon and nearly stationary muscles with short fibers \[[@pone.0193850.ref005]\]. In normal standing, the muscle is typically \~ 15 times stiffer than the tendon \[[@pone.0193850.ref006],[@pone.0193850.ref007]\]. As it is the weaker link, in this condition the tendon sets the maximal value of ankle stiffness. Consequently, muscle stiffness only becomes relevant to ankle stiffness if tendon stiffness increases, as it does under conditions of high levels of torque (e.g. walking, running or jumping).
The muscle can be stiffened either by the active contraction of its fibres or by passive means. Passive means include changes in muscle length through stretch and altering the immediate history of muscle movement (e.g. leaving the muscle still for 5--10 s), which can produce thixotropic increases in stiffness \[[@pone.0193850.ref005],[@pone.0193850.ref007]--[@pone.0193850.ref017]\].
Tendon stiffness is dependent on the linkage between the collagen molecules, cross-linked end-to-end within a fibril \[[@pone.0193850.ref018]\]. This is a completely passive mechanism. This 'netted' distribution, often compared to the behaviour of a knitted sock which stiffens as it is stretched, makes it possible for the tendon to be lengthened and to support high loads of tension without rupturing. Its stiffness is defined by the amount of deformation of its fibers, ranging from a slack region, when the fibers are crimped, to a linear region, when the fibers are relatively parallel, and finally microscopic and macroscopic failure regions, when the fibers start to snap, leading to rupture \[[@pone.0193850.ref019]\]. Thus the tendon stiffness increases with the tension that it transmits \[[@pone.0193850.ref019]--[@pone.0193850.ref027]\]. *In vivo* measurement of stiffness of whole bundles of tendons and muscles acting in combination was much facilitated with the introduction of the ultrasound technique \[[@pone.0193850.ref028]\]. For the measurement of the Achilles tendon stiffness, for example, ultrasound probes are used to track the change in position of the distal myotendinous junction of the muscle and the insertion point of the Achilles tendon. Tendon stiffness is then expressed as the slope between this change in tendon length and tendon force \[[@pone.0193850.ref028]--[@pone.0193850.ref033]\].
The function of the Achilles tendon is to connect the triceps surae muscles to the calcaneous bone. Therefore, it can be elongated in two different ways; either by shortening of the calf muscle fibers connected to it through active muscle contraction, or by passive dorsiflexion of the ankles. Both modes of elongation can act to increase tendon stiffness. This implies that, in standing, ankle stiffness should increase as ankle torque increases. Ankle torque can be increased by leaning forward or by dorsiflexion of the foot. However, it is important to note that in standing the underlying ankle torque is directly proportional to the angle of inclination of the body so that ankle torque cannot be arbitrarily altered without changing body lean. A further possible way of increasing ankle stiffness may be by co-contraction, although co-contraction is unusual in standing.
Two contradictory results for ankle stiffness have been reported for upright human subjects. Loram & Lakie \[[@pone.0193850.ref002]\] applied brief and very small (0.05 deg amplitude, 140 ms duration, squared-sine shaped) perturbations to individuals strapped to a vertical support while standing on footplates. At this fixed ankle angle, the participants were asked to maintain a constant mean level of bias plantarflexing ankle torque for 40 s. During this task mean ankle angle did not change, therefore, the tendon stiffness could only be altered by the contracting muscles pulling the tendon. Over a relatively large range of plantarflexing ankle torque (5--25 Nm in one leg only), the researchers found only a small and insignificant rise in ankle stiffness (from \~5 to \~6 Nm deg^-1^). In freely standing individuals, mean plantarflexing ankle torque can only be increased by leaning forwards. Casadio et al. \[[@pone.0193850.ref034]\] applied larger perturbations (1 deg, 150 ms, ramps) to freely standing individuals. They studied only two subjects, but in both there was a substantial increase in ankle stiffness as the subjects leaned forward. Because the subjects were leaning forward, there was an inevitable slight dorsiflexion.
In an attempt to resolve these conflicting observations, we carried out two experiments to further investigate how intrinsic standing stiffness can be affected by ankle torque or angle. In study 1, intrinsic stiffness was measured at various mean levels of forward lean using a range of perturbation amplitudes. In study 2, intrinsic stiffness was measured while standing normally with increased ankle dorsiflexion (that is, standing with toes raised).
The intention was to compare the effects of two conditions in which tendon tension was varied. In one it was varied mainly by muscle activation (forward body leaning) and in the other it was varied mainly by passive stretch (ankle dorsiflexion). Of course, active and passive sources of ankle torque are inevitably entangled during quiet stance, but our objective here was to maximize the contribution of either source using these two interventions. We were also able to investigate whether changes in stiffness were critically dependent on the perturbation amplitude used to make the measurement. In addition, we investigated how sway size and velocity were altered by leaning and dorsiflexion. To our knowledge, no previous study has investigated the effect of Achilles tendon tension on postural sway.
Methods {#sec002}
=======
Participants {#sec003}
------------
For each study, 10 healthy volunteers (Study 1: six female; age 28.1±4.4 years (mean±SD); height 1.68±0.1 m; weight 65.9±8.3 kg) (Study 2: six female; age 29.1±10.5 years) gave written informed consent and participated in this study, which was approved by the local human ethics committee at the University of Birmingham ([Table 1](#pone.0193850.t001){ref-type="table"}).
10.1371/journal.pone.0193850.t001
###### Participant anthropometric data.
{#pone.0193850.t001g}
----------------- ---------- --------------- ---------------------------------------------
Study 1
**Participant** Gender Age (years) Toppling torque per unit angle (Nm deg^-1^)
P01 M 31 8.47
P02 F 26 10.58
P03 F 37 8.26
P04 F 30 8.99
P05 F 22 8.92
P06 F 24 9.15
P07 M 31 13.27
P08 M 24 12.58
P09 M 29 9.81
P10 F 27 9.74
**Mean ± SD** **F(6)** **28.19±4.4** **9.98±1.7**
**Study 2**
**Participant** Gender Age (years) Toppling torque per unit angle (Nm deg^-1^)
P01 M 39 12.42
P02 F 25 12.25
P03 F 37 7.44
P04 F 23 10.13
P05 F 23 14.76
P06 F 22 8.84
P07 M 25 9.92
P08 F 23 8.06
P09 M 53 14.97
P10 M 21 17.31
**Mean ± SD** **F(6)** **29.1±10.5** **11.6±3.3**
----------------- ---------- --------------- ---------------------------------------------
Procedure and apparatus {#sec004}
-----------------------
A full description of the footplate apparatus used to measure ankle stiffness was given elsewhere \[[@pone.0193850.ref005]\]. In brief, the participants were asked to stand on top of motorized footplates, coaxially aligned with their ankles. Ankle torque and angle and left footplate acceleration were recorded, as well as lower limb EMG responses from the *medial gastrocnemius* and *tibialis anterior* muscles (Delsys Bagnoli). Only measurements of the left lower limb were used for the stiffness estimation. The methodology specific to the two studies is described below ([Fig 1](#pone.0193850.g001){ref-type="fig"}).
{#pone.0193850.g001}
We chose to record medial gastrocnemius rather than soleus activity because responses due to reflex or higher level activity are more prominent and easier to identify in that muscle \[[@pone.0193850.ref002],[@pone.0193850.ref035],[@pone.0193850.ref036]\]. However, previous research reports cross-talk between the triceps surae muscles when using surface EMG \[[@pone.0193850.ref037]\], so it is possible that the recording is representative of activity in the entire calf muscle. For clarification, we suggest recording the activity of both medial gastrocnemius and soleus muscles in future studies.
### Study 1 {#sec005}
The main objective of this experiment was to determine how passive ankle stiffness changes with increasing ankle torque caused by forward lean. Participants stood for approximately 3 minutes, during which small and brief perturbations were applied at a variable gap of 4--5 seconds. Each perturbation was a raised cosine curve of 140 ms duration with a randomly varied maximum amplitude of 0.1, 0.3 or 0.7 deg, in a randomly varied toes-up and toes-down direction. The three amplitudes were chosen to establish whether the effect of forward lean on ankle stiffness was critically dependent on the amplitude of the perturbation used to measure it. The strong perturbation repeatability was confirmed when calculating the standard deviation of its peak angular rotation (potentiometer recordings). Within the data of a single participant, it varied by as little as 0.013 deg SD for 0.1 deg perturbation, 0.013 deg SD for 0.3 deg perturbation and 0.017 deg SD for 0.7 deg perturbation. The whole experimental procedure consisted of one session of approximately 1 ½ hours.
We asked the participants to manipulate the amount of forward shift of their COP by monitoring the average baseline torque applied by the feet against the footplate. Before the experiment, participants did at least one familiarization trial per condition to establish the level of torque that they could sustain at each level of leaning. Due to individual variation of weight and height, it was not possible to establish a common target torque for all participants. For example, during the forward leaning condition, we asked the participants to lean as far forward as they could, comfortably for the duration of the trial. We averaged these torque traces and established this value as an individual target trace they had to maintain. It was displayed on a screen located at eye level during the actual trials. They performed three different levels of forward leaning:
1. *Normal*: standing at their spontaneously chosen position;
2. *Vertical*: standing with the COP shifted backwards in relation to their normal condition. Participants were asked to reduce torque applied against the footplate, as much as possible without compromising their free standing balance control;
3. *Lean*: standing with the COP shifted forwards in relation to their normal condition. Participants were asked to increase torque applied against the footplate to a level that was still comfortable and sustainable for the duration of the 3 min trials.
For obvious reasons, it was relatively easier for individuals to follow a target torque in normal condition (0.67±0.53 Nm, mean±SD, difference from target torque). It was more difficult to maintain proximity with the target torque in vertical and lean conditions (1.02±0.92 Nm and 1.91±1.69 Nm, respectively). Nevertheless, the participants were able to maintain a reasonable amount of accuracy, considering that they were freely standing.
There were 3 conditions of COP shift (normal, vertical and lean) and 3 different perturbation amplitudes (0.1, 0.3 and 0.7 deg), resulting in a total of 9 conditions. Given that 30 perturbations were applied per condition, altogether 270 events were recorded for each participant. The timing, conditions, perturbation amplitudes and directions were pseudo-randomized so that they could not be predicted by the participants.
### Study 2 {#sec006}
For this experiment the main objective was to increase ankle dorsiflexion with minimal change in baseline ankle torque. Participants performed standing trials with perturbations of the same shape and time-window intervals as in the previous experiment, but only with amplitudes of 0.1 or 0.7 deg. There were two different conditions:
1. *Normal*: standing at their spontaneously chosen position with footplates horizontal;
2. *Dorsiflexion*: standing with the footplate rotated upwards by 15 deg. As it was an angular measure, it was common to all subjects (unlike the torque in study 1). 15 deg was chosen because it was close to the maximum that the participants could comfortably maintain.
The whole experimental procedure consisted of one session of approximately 1 hour. There were 2 ankle positions and 2 perturbation amplitudes and 32 perturbations were recorded for each condition, resulting in a total of 128 events from each participant. Normal standing was investigated prior to the dorsiflexion trials. Perturbation amplitudes and directions were pseudo-randomized.
Data analysis {#sec007}
-------------
### Determination of mechanical intrinsic ankle stiffness {#sec008}
We assumed that the calf muscles, the Achilles tendon, aponeurosis and foot act as a mass-spring-damper system \[[@pone.0193850.ref038],[@pone.0193850.ref039]\] responsible for generating the corrective torque applied by the feet against the ground to stabilize position \[[@pone.0193850.ref040],[@pone.0193850.ref041]\]. The spring component (combination of muscles, tendon, aponeurosis and foot modulating stiffness of the ankles), damper component (viscosity of the ankle joint and associated tissues) and mass component (moment of inertia of the foot and moving muscle with respect to the medial malleolus acting as the axis of rotation) were estimated with a fitting equation in which the torque measured over the first 70 ms of the perturbation was compared with the torque generated by a simple second-order model. The three inputs to this model were the measured ankle position, velocity and acceleration \[[@pone.0193850.ref002],[@pone.0193850.ref042]\]: $$T = K\theta + B\overset{˙}{\theta} + I\overset{¨}{\theta}$$
Where: T = torque (Nm), *θ* = angle (deg) $\overset{˙}{\theta}$ = angular velocity (deg s^-1^), $\overset{¨}{\theta}$ = angular acceleration (deg s^-2^), K = stiffness (Nm deg^-1^), B = viscosity (Nm s deg^-1^) and I = moment of inertia of the foot (kg m^2^).
### Determination of toppling torque per unit angle {#sec009}
During normal standing, the gravitational torque exerted by the body COM is closely related to the COM rotation around the ankle joint \[[@pone.0193850.ref040],[@pone.0193850.ref043],[@pone.0193850.ref044]\]. This is because the body above the ankles is kept relatively aligned in the vertical position and its sway amplitude is below 6 deg \[[@pone.0193850.ref045]\]. Toppling torque per unit angle is a representation of this relationship. It is defined as m × g × h (for convenience, referred as ´mgh´), where m is the participant mass above the ankles, g is the gravitational acceleration and h is the height of the COM above the ankles. Here it is used only as a reference to normalize data from all participants, regardless of their height and body mass. The intrinsic ankle stiffness was then estimated as a percentage of mgh--if equal or higher, it would potentially stabilize the body alone. Mgh was indirectly calculated as the slope of the linear fit between ankle torque (load cell data) and body angle (umbilical laser-reflex sensor data), recorded during 180 s trials of voluntary sway, in which subjects were instructed to sway very gently about the ankle joint, minimizing any hip or knee motion.
### Determination of baseline ankle torque, ankle and body angle, body sway and EMG activity {#sec010}
We assessed the amount of leaning as the increase in total ankle torque. This was calculated from the mean total ankle torque during a 70 ms time window prior to each perturbation onset. The amount of ankle dorsiflexion and body inclination were calculated as the mean ankle angle (calculated from laser signal reflecting shin minus footplate angle) and body angle (laser signal reflecting movement of the waist) over a 2 s time window prior to each perturbation onset. We normalized each participant's data to their normal standing condition, described here as 0 deg for body angle and 90 deg for ankle angle. Thus we discounted their actual elected standing position, which corresponds to a variable 1.5--4 deg forward leaning in relation to the earth \[[@pone.0193850.ref002]\].
The effect of the different conditions upon stability and control of movement was assessed with measurements of anterior-posterior body sway and muscle activity. Body sway was quantified as the average root-mean-square (RMS) ankle angle and velocity. The mean value over a 2 s time window prior to each perturbation onset was subtracted. Muscle activity was calculated as the integral of the rectified EMG activity of the medial gastrocnemius and tibialis anterior muscles over a 70 ms time window envelope prior to each perturbation onset. To compare changes within conditions, we normalized all participants' EMG data as a ratio of normal standing data.
### Statistical analysis {#sec011}
Repeated-measures ANOVA was used to determine effects of condition (normal, vertical and lean or normal and dorsiflexion, for 1^st^ and 2^nd^ studies) and stimulus amplitude (0.1, 0.3 and 0.7 deg or 0.1 and 0.7 deg, respectively) upon ankle stiffness. Two-tailed paired samples t-tests and one-way ANOVA were used to verify differences in mean ankle and body angle, ankle torque, body sway and EMG activity between conditions. P\<0.05 was considered statistically significant for all tests.
Results {#sec012}
=======
Representative data of one participant during two different studies is shown in [Fig 2](#pone.0193850.g002){ref-type="fig"}. The upper panel is from study 1, while the lower panel is from study 2.
{#pone.0193850.g002}
Absolute ankle torque and relative ankle and body position {#sec013}
----------------------------------------------------------
As intended, there was a significant increase in mean ankle torque between each of the conditions within study 1, ranging from 4.9±2.4 Nm (*vertical*, mean±SD), 16.4±4.7 Nm (*normal*) and 31±5.7 Nm (*lean*). A one-way ANOVA revealed a significant difference between different leaning conditions (F~2,27~ = 85.5; p\<0.001) ([Fig 3](#pone.0193850.g003){ref-type="fig"}, left graph). In study 2, our main concern was to change the amount of ankle dorsiflexion by means of tilting the standing platform by 15 deg. There was an unintended but significant torque increase from 17.6±4.7 to 22.4±5.7 Nm (t~(9)~ = -2.4; p\<0.05) ([Fig 3](#pone.0193850.g003){ref-type="fig"}, right graph), a result of the participants leaning slightly more forwards when the toes were raised.
{#pone.0193850.g003}
We next describe the changes in ankle and body angle in each experiment. In study 1, both ankle and body angle increased as the participants leaned forward. However, they did not increase by the same amount and it was clear that the subjects did not act entirely as a rigid body ([Fig 4](#pone.0193850.g004){ref-type="fig"}). Nevertheless, one-way ANOVA analysis has shown significant difference between conditions in body (F~2,27~ = 5.9; p = 0.007) and ankle (F~2,27~ = 3.5; p = 0.043) mean angles ([Fig 4](#pone.0193850.g004){ref-type="fig"}, left graphs).
{#pone.0193850.g004}
Rotating the platform toes-up by 15 deg was enough to produce a large increase in ankle dorsiflexion in study 2 (t~(9)~ = -17.6; p\<0.001) ([Fig 4](#pone.0193850.g004){ref-type="fig"}, bottom right graph). With the analysis of the body average position, we verified that even though instructed otherwise, the participants leaned forward by a slight amount (0.4±0.7 deg), but this was not significant (t~(9)~ = -1.7; p = 0.12) ([Fig 4](#pone.0193850.g004){ref-type="fig"}, top right graph). In leaning forward there was considerable departure from a rigid body with the body arching (convex anteriorly) so that ankle angle moved forward by 3.7 deg and body angle by only 0.4 deg. An imposed dorsiflexion of 15 deg combined with an unintentional forward rotation of the ankle of 3.7 deg generated a shift in ankle angle from the nominal 90 deg to a mean of 71.3 deg.
Intrinsic ankle stiffness {#sec014}
-------------------------
For both studies, average intrinsic ankle stiffness is presented as a percentage of toppling torque per unit angle (% mgh) in [Fig 5](#pone.0193850.g005){ref-type="fig"}. A 3-way ANOVA analysis showed that there was no effect of perturbation direction (toes-up *versus* toes-down) on stiffness, for either study (F1,9≤0.37; p = ≥0.56). Therefore, we combined perturbations of both directions for the final calculation. For the first experiment (left graph), values ranged from 37% to 97%. There was a systematic increase in stiffness (approximately 30% overall) from vertical to normal and from normal to lean (effect of condition: F~2,18~ = 18.5; p\<0.001; effect of amplitude: F~2,18~ = 170.2; p\<0.001). There was no interaction between condition and amplitude (F~4,36~ = 0.63; p = 0.64). Ankle dorsiflexion also produced a significant increase in stiffness (mean 29%) (condition: F~1,9~ = 18.4; p = 0.002; amplitude: F~1,9~ = 40.2; p\<0.001), also without any interaction between condition and amplitude (F~1,9~ = 0.31; p = 0.59) (right graph). The results obtained from the normal condition in both experiments were very similar (51% - 50% for 0.7 deg perturbation and 82% - 77% for 0.1 deg perturbation, respectively), making comparisons between both experiments easier to interpret.
{#pone.0193850.g005}
As the duration of all perturbation sizes was the same (140 ms), the peak velocity was higher for larger perturbations, and we found a significant increase in viscosity with perturbation amplitude (first study: F~2,18~ = 13.4; p\<0.001; second study: F~1,9~ = 12.8; p = 0.006). Viscosity was also significantly affected by the degree of forward leaning (F~2,18~ = 13.7; p\<0.001) but not by passive stretch (F~1,9~ = 2.1; p = 0.177). Moment of inertia was dependent on degree of forward leaning and amplitude in the first study (condition: F~2,18~ = 4.0; p = 0.037; amplitude: F~2,18~ = 10.6; p = 0.001); it was dependent on passive stretch but not amplitude in the second study (condition: F~1,9~ = 42.4; p\<0.001; amplitude: F~1,9~ = 1.6; p = 0.233).
Body sway and muscle activity {#sec015}
-----------------------------
Body sway size is described in [Fig 6](#pone.0193850.g006){ref-type="fig"}. Assessment of sway (body angle RMS, top left panel) and sway velocity (body angular velocity RMS, bottom left panel) show no significant difference between conditions of forward leaning (one-way ANOVA F~2,27~ = 2.2; p = 0.133; F~2,27~ = 1.8; p = 0.189, respectively). Dorsiflexion also has no significant effect on sway size or velocity (t~(9)~ = -1.3; p = 0.225 and t~(9)~ = 1.2; p = 0.267, respectively) (right panels).
{#pone.0193850.g006}
The analysis of EMG activity confirms that changing the level of body leaning either backwards or forwards from normal stance will alter the neural activation. [Fig 7](#pone.0193850.g007){ref-type="fig"} shows normalized data. As expected, by leaning backwards (*vertical* condition), the *medial gastrocnemius* (GM) activity is reduced, but there is a large increase in *tibialis anterior* (TA) activity. When leaning forwards, there is a considerable increase in GM activity accompanied by a trivial increase in TA activity (one-way ANOVA, GM F~2,27~ = 11.7; p\<0.001; TA F~2,27~ = 12.6; p\<0.001) (left graph). In the dorsiflexion condition (right graph), there is a significant increase in TA activity and decrease in GM activity (GM t~(9)~ = 3.6; p = 0.006; TA t~(9)~ = -2.5; p = 0.032).
{#pone.0193850.g007}
Discussion {#sec016}
==========
In this paper we attempt to understand how intrinsic ankle stiffness can be altered by changes in normal standing conditions. To complement previous research on standing individuals \[[@pone.0193850.ref002],[@pone.0193850.ref034]\], two different techniques were used.
Firstly, active ankle torque was altered by asking the participants to lean at different angles. Intrinsic stiffness was shown to correlate positively with increasing levels of forward leaning which produced increasing levels of active ankle torque (increase of \~ 26.1 Nm from *vertical* to *lean*, [Fig 3](#pone.0193850.g003){ref-type="fig"}). These results support Casadio et al.'s \[[@pone.0193850.ref034]\] observation that stiffness increases with active torque in standing. This effect remained similar as we reduced the perturbation size from 0.7 to 0.1 deg. In opposition to the results presented here, for even smaller perturbations (0.05 deg), Loram & Lakie \[[@pone.0193850.ref002]\] have found little alteration in intrinsic stiffness with an increase in active torque.
Secondly, ankle angle was altered by rotating the standing surface by 15 deg toes-up while intrinsic stiffness was assessed. Because this caused an inadvertent postural alteration and a slight forward lean, ankle dorsiflexion increased by more than the imposed 15 deg (18.7±3.4 deg, [Fig 4](#pone.0193850.g004){ref-type="fig"}). We found that dorsiflexion also increased intrinsic stiffness substantially. As the participants did lean forwards a little, their ankle torque was slightly higher than normal (increase of \~ 4.8 Nm from normal to dorsiflexion, [Fig 3](#pone.0193850.g003){ref-type="fig"}). However, stiffness increased very considerably despite the modest torque increase. Consequently, intrinsic ankle stiffness must be increased in different ways by leaning and by dorsiflexion.
Finally, although both procedures caused significant changes in intrinsic ankle stiffness, there were only trivial alterations in the size or speed of sway. We now discuss each of these findings in turn.
In standing, intrinsic ankle stiffness has a positive dependency on active ankle torque {#sec017}
---------------------------------------------------------------------------------------
The results shown here (study 1), confirm Casadio et al.'s \[[@pone.0193850.ref034]\] significant positive correlation between active ankle torque and standing intrinsic ankle stiffness. This relationship is clearly shown in [Fig 5](#pone.0193850.g005){ref-type="fig"}. For an average active ankle torque ranging from 5 to 16 to 31 Nm, the average intrinsic stiffness (mean of the three perturbation sizes) increased from 53 to 66 to 81% mgh.
Thus, a change from normal standing position to forward lean produced an average torque rise of \~ 15 Nm ([Fig 3](#pone.0193850.g003){ref-type="fig"}) and an average intrinsic stiffness increase of 15% mgh ([Fig 5](#pone.0193850.g005){ref-type="fig"}). With a mean increase in ankle torque of 13.9 Nm from normal to leaning condition, Casadio et al. \[[@pone.0193850.ref034]\] found an increase in intrinsic stiffness of 33.5% mgh. Thus, with slightly less increase in ankle torque, they found a larger increase in intrinsic stiffness. However, their results were derived from only two subjects, and would certainly fit comfortably into the range of values we recorded. They used a larger perturbation (ramp of 1 deg), but as we show in [Fig 5](#pone.0193850.g005){ref-type="fig"}, the effect of body lean upon stiffness was seen for all perturbation amplitudes.
The results from study 1 were also comparable to those previously derived from recumbent individuals \[[@pone.0193850.ref046]\], which also showed a significant increase in intrinsic ankle stiffness with an increase in active ankle torque. The experimental technique was very different, but the results were strikingly similar. Mirbagheri et al.'s \[[@pone.0193850.ref046]\] relationship between torque and stiffness is curvilinear and variable between subjects, but their [Fig 6](#pone.0193850.g006){ref-type="fig"} suggests that an increase from 12 to 24 Nm would increase intrinsic stiffness by approximately 30% on average, an increase quite similar to the one reported here. Increasing stiffness with ankle torque can be simply and economically explained on the assumption that intrinsic ankle stiffness is a function of tendon stiffness. Ankle stiffness is the sum of the compliances of the structures that are deformed when the ankle is rotated. For low tensions the tendon is the weakest spring, so it dictates the upper limit of intrinsic stiffness. In effect, it is immaterial how stiff the muscles lying in series become, because the overall stiffness can never exceed that of the tendon. However, as tension increases (as in leaning forward) the tendon stiffness increases and, as a consequence, the intrinsic stiffness becomes larger.
Because the different results found by Loram & Lakie \[[@pone.0193850.ref002]\] and Casadio et al. \[[@pone.0193850.ref034]\] were possibly related to the different perturbation size used to assess stiffness, we expected to find less difference of stiffness for the smallest perturbation sizes with increase in active ankle torque. However, we did not see this interaction. The perturbation size (0.1--0.3--0.7 deg) used in study 1 was lower than in Casadio et al. \[[@pone.0193850.ref034]\] (1 deg), but larger than the perturbation size used by Loram & Lakie \[[@pone.0193850.ref002]\] (0.05 deg). In the present experiments the extreme difference of stiffness was 27.5% mgh for 0.1 deg perturbation, 22.0% for 0.3 deg and 33.1% for 0.7 deg ([Fig 5](#pone.0193850.g005){ref-type="fig"}, left graph). Thus there is no indication that reducing the size of the perturbation over the range that we used here causes a reduction in the stiffening associated with raised ankle torque.
Unusually, in the paper by Loram and Lakie, the authors measured both foot stiffness and ankle stiffness. They reported (see [Fig 6D](#pone.0193850.g006){ref-type="fig"} from Loram & Lakie \[[@pone.0193850.ref002]\]) that foot stiffness tended to decrease and become more consistent with increasing torque, and that this partly offset a rise in true ankle stiffness. It may be that, in using perturbations of less than a certain critical size, increasing compliance of the foot and soft tissues conceals the rise in stiffness associated with torque-induced tendon stiffening, setting a rather high and constant level of stiffness. This may be particularly relevant to quiet standing where many of the spontaneous sways tend to be very small in size. In this regimen, ankle stiffness may be effectively independent of torque level (muscle activity). Perhaps for very tiny perturbations or sways, stiffness maintains a constant level because the compliance of the foot and soft tissues acts as a relatively constant stiffness buffer. Interestingly, even when our participants were inclined as vertically as possible and applying an average ankle torque of as little as 2.2 Nm against the ground, the intrinsic stiffness was substantial, at 70% mgh when perturbed by 0.1 stimulus amplitude and 36% mgh when perturbed by 0.7 deg stimulus amplitude ([Fig 5](#pone.0193850.g005){ref-type="fig"}, left graph). It shows that even at conditions close to the vertical equilibrium position when there is minimal ankle torque and the ankle is as plantar flexed as far as possible commensurate within normal standing, intrinsic ankle stiffness is still relatively high.
In standing, intrinsic ankle stiffness has a positive dependency on dorsiflexion {#sec018}
--------------------------------------------------------------------------------
To our knowledge, there have not been any previous investigations into the effects of passive stretch on the intrinsic ankle stiffness of standing individuals. Here, we show that this relationship is significantly positive. With a large increase in passive stretch by 18.7 deg ankle dorsiflexion and an associated small increase in active ankle torque of \~ 4.8 Nm, stiffness increased from 50 to 77% mgh, and 77 to 109% mgh (for 0.7 and 0.1 deg perturbations, respectively) ([Fig 5](#pone.0193850.g005){ref-type="fig"}, right graph).
In freely standing subjects, as here, the rise in ankle torque must be associated with a forward lean and this is confirmed by [Fig 4](#pone.0193850.g004){ref-type="fig"}. Subjects adopted an unusual body configuration, moving the knees forward considerably and inclining the body forward a little. This arching of the body is presumably to counteract the sense of instability that arises when the toes are raised, tending to tip the body posteriorly. This rise in torque is very likely to have produced some rise in intrinsic stiffness as described above, however the increase that we observed seems much too large to be produced solely by this mechanism.
It is possible to compare our data with those from Mirbagheri et al. \[[@pone.0193850.ref046]\]. In that study, the authors have also shown in seated individuals that passive ankle stiffness strongly depends on passive stretch. Participants maintained a constant mean 5 Nm plantarflexion torque against the footplate and stiffness was measured in a range of angles from maximum plantarflexion to maximum dorsiflexion. When the ankle is taken from the neutral position to 14.4 deg of dorsiflexion the intrinsic stiffness increases by approximately 40% (see Fig 8 from Mirbagheri et al. \[[@pone.0193850.ref046]\]). Thus, these authors also report a substantial rise in intrinsic stiffness of about the same magnitude as the one we find here, and in their case, it is uncomplicated by associated torque increase. This strong relationship between ankle stiffness and passive stretch was also suggested by earlier work by Kearney and colleagues \[[@pone.0193850.ref047]\]. They showed that when the ankle is passively dorsiflexed from neutral until the end of the range of movement, there is relatively small change in ankle torque (ranging from 6 to 12.4 Nm plantarflexing torque), but intrinsic stiffness increases considerably (1.3 to 3.6 Nm deg^-1^).
Consequently, this increase is stiffness is associated with ankle angle rather than ankle torque so an alternative explanation for the rise in stiffness is necessary. This could be a consequence of either coactivation of antagonist muscles, increase of calf muscle moment arm, unmeasured change in knee angle acting on the biarticular gastrocnemius muscle or increased resistance of other tissues crossing the ankle.
Activity of the muscles is shown in [Fig 7](#pone.0193850.g007){ref-type="fig"}. In normal standing the TA muscle is virtually silent. Dorsiflexion does increase activity in the TA. However, although significant, the degree of coactivation is very slight. It can, for example, be compared to the greatly increased TA activity which is observed when the body is inclined vertically. Furthermore, in this condition of extreme dorsiflexion the TA is operating in a region of mechanical disadvantage (short fibres, reduced moment arm. Thus, the degree of coactivation seems inadequate to produce the large increase in intrinsic stiffness that we observed.
If dorsiflexion increases the moment arm of the calf muscles, a given rotation of the ankle will cause a larger linear displacement of the tendon and muscle and, consequently, measured angular stiffness will rise. However, as has been pointed out by \[[@pone.0193850.ref048]--[@pone.0193850.ref050]\], moment arm actually decreases with dorsiflexion. This was quantified in vivo with magnetic resonance imaging (MRI) and real-time ultrasonography by Maganaris et al. \[[@pone.0193850.ref049]\], who found that ankle dorsiflexion of 15 deg produces \~ 0.5 cm decrease in moment arm ([Fig 2](#pone.0193850.g002){ref-type="fig"}). Consequently, dorsiflexion should produce a decrease in stiffness rather than the increase that we and others observe.
In leaning forward, there might have been a slight, unintended and unmeasured extension of the knee joint. This would pull on the biarticular gastrocnemius muscle and potentially contribute to increased ankle stiffness. Similarly, during dorsiflexion the toes were moved upwards and there was an alteration in body conformation, as indicated by the sketched image in [Fig 2](#pone.0193850.g002){ref-type="fig"}. The shin was driven posteriorly, and this might have also acted to extend the knee slightly. Hence, this could contribute towards the increased stiffness that we measured in the dorsiflexion condition. However, the knee contribution is likely to be slight. Herbert et al. \[[@pone.0193850.ref030]\] showed that the relationship between gastrocnemius muscle tendon unit lengthening and joint rotation was much smaller for the knee than for the ankle. On average, ankle rotation stretches gastrocnemius by 0.83 mm per degree, whereas knee rotation stretches gastrocnemius by only 0.23 mm per degree (see [Fig 1](#pone.0193850.g001){ref-type="fig"} from Herbert et al. \[[@pone.0193850.ref030]\]). The unmeasured knee rotation in our experiments is very unlikely to have exceeded 10 degrees. With the possibly extreme degree of extension of the knee experienced by the participants of both studies, the gastrocnemius would be lengthened by much less than 3 mm. This is very small compared to the intended lengthening produced by ankle dorsiflexion, which would be nearly 17 mm.
Given that changes in co-contraction, moment arm and knee angle do not fully explain changes in stiffness found by these studies, a remaining possibility is that the rise in stiffness is due to stretching or compression of other tissues which bridge the ankle joint.
Tendon and aponeurosis strain as cause of increased intrinsic stiffness {#sec019}
-----------------------------------------------------------------------
The calf muscle-tendon complex can generate tension in two ways. First, with active contraction (as in the lean condition) there is a tendency for muscle to shorten and expand. This has the effect of straining the serially-arranged tendon. Series elastic structures in muscles are not limited to the extramuscular free tendon. Most muscles with a long free tendon also have a sheet-like aponeurosis that serves as a broad tendinous insertion for muscle fibers \[[@pone.0193850.ref051]\]. Thus, there is an associated complex biaxial straining of the aponeurosis which partly encircles the muscle \[[@pone.0193850.ref052]\]. On the well-established assumption that with low levels of force the tendon is less stiff than the muscle \[[@pone.0193850.ref007],[@pone.0193850.ref053],[@pone.0193850.ref054]\], the effect of the progressively increasing muscle activity required in leaning forward, is to increase tendon strain. This increases tendon stiffness and consequently ankle stiffness. Muscle stiffness is also likely to rise progressively with increased activity in leaning forward, but this is inconsequential because ankle stiffness is limited to the much lower level dictated by the tendon.
Second, the calf muscle can be passively pulled to a long length (as in the dorsiflexion experiment). This has the effect of considerably straining the aponeurosis, and because it now generates substantial passive tension at this longer length, the free tendon is similarly strained \[[@pone.0193850.ref053]\]. The effect of these changes is again to increase the stiffness of the tendon and consequently the ankle. In this case there is likely to be a small rise in muscle stiffness, which once again would have minimal effect on overall ankle stiffness. Direct observation of muscle/tendon motion with ultrasound would help confirm this in future studies.
Even though significant, the change in inertia in study 1 caused by condition and amplitude was very small (8.1%). Such changes are much smaller than the observed changes in stiffness, and could be accounted for simply by errors in the estimation process, since all three parameters are allowed to vary (K, B & I). For study 2, the changes were higher (14.1%), with dorsiflexion being associated with greater inertia. These apparent changes in inertia may also be due to variation in the model fitting process. Alternatively, it is concievable they are attributable to additional muscle mass being moved during the perturbation. This would be consistent with our assertion that dorsiflexion increases the stiffness of the tendon which would, in turn, cause greater muscle movement. Whether this added moving mass would be sufficient to cause the increased inertia is uncertain. Again, measurement with ultrasound would help confirm this.
Sway size changes little in response to altered intrinsic stiffness {#sec020}
-------------------------------------------------------------------
In a previous paper \[[@pone.0193850.ref005]\], we showed that ankle intrinsic stiffness is reduced when there is increased movement amplitude (more sway). Because this reduction in intrinsic stiffness was seen best with small perturbations, we attributed it to a movement induced reduction in muscle stiffness (thixotropy). Large sways are associated with low muscle stiffness and consequently low intrinsic ankle stiffness. We suggested that the converse might be true--small sways might be associated with high muscle stiffness and higher intrinsic ankle stiffness. That suggestion is not strongly supported by the present findings which show no clear relationship between ankle intrinsic stiffness and sway size ([Fig 6](#pone.0193850.g006){ref-type="fig"}). It seems inevitable that the tendon stiffness sets an upper limit on ankle stiffness. Because of the series arrangement of muscle and tendon, and the fact that at modest levels of torque the tendon is much less stiff than the muscle, the effect of muscle stiffness change is highly asymmetric. Movement can easily decrease muscle (and hence ankle) stiffness. Lack of movement might, indeed, increase muscle stiffness considerably but this does not increase intrinsic ankle stiffness which is limited to the value set by the tendon.
Intrinsic stiffness is a stabilizing feature because in partly offsetting gravitational acceleration it reduces the size and the speed of necessary neural interventions. However, the changes that we produced did not systematically alter sway size or velocity. It is possible that by requiring subjects to stand in unusual configurations, as here, the normally stabilizing role of intrinsic stiffness is vitiated by the unfamiliarity of the task and standing is dominated by active neural control so that any departure from normal tends to increase sway size.
Contributions to the ankle torque required to stand {#sec021}
---------------------------------------------------
Because both processes stretch elastic tendon structures, muscle activity and dorsiflexion can, in principle, contribute to the ankle torque required to stand. Weiss et al. \[[@pone.0193850.ref047],[@pone.0193850.ref055]\] showed that with large passive dorsiflexion their subjects generated an ankle torque ranging from 6 to 12.4 Nm. They suggested that as normal standing requires a total torque of the order of \~ 50 Nm the passive torque from both ankles might make a material contribution. We confirm this suggestion. We show that, with extreme dorsiflexion, muscle activity in standing can be decreased, even though ankle torque demand is higher because dorsiflexion causes an inadvertent slight extra lean forward. However, with the very small degree of dorsiflexion that occurs in standing on level ground the passive torque contribution to normal standing is likely to be very small. Our conclusion is that in normal standing ankle stiffness will be much more affected by modulation of ankle torque (which varies a lot) rather than ankle angle (which varies only a little).
Conclusion {#sec022}
==========
Our results show that intrinsic ankle stiffness in standing individuals increases considerably in conditions of increased forward lean and dorsiflexion. Even though stiffness is changed, body sway is little affected. Since during normal standing there is relatively small potential for changes in ankle angle compared with larges changes in ankle torque, we believe that intrinsic ankle stiffness is mostly affected by the latter.
Thanks to Steve Allen and Dr David McIntyre for technical assistance.
[^1]: **Competing Interests:**The authors have declared that no competing interests exist.
| {
"pile_set_name": "PubMed Central"
} |
Background {#Sec1}
==========
Chronic hepatitis C virus (HCV) infection, with a worldwide prevalence of 2%-3% \[[@CR1]\], causes substantial loss of life and reduces quality of life in those who are infected \[[@CR2]\]. Although the screening of blood products has reduced the incidence of HCV infection in developed countries, there is a long latency period before the disease becomes symptomatic and thus large numbers of new cases will continue to be identified over the coming years. Estimates suggest that about half of the 3.1 million US patients infected with HCV are unaware of their infection and only a small fraction have been treated \[[@CR3]--[@CR5]\].
Under optimal conditions, treatment of chronic HCV infection with pegylated interferon alpha (PEG-IFN-alfa) and ribavirin dual therapy produces sustained virologic response (SVR) of above 50% in patients with HCV infection \[[@CR6], [@CR7]\]. For patients with the most common HCV genotype (GT) in the US, GT-1, 48 weeks of PEG-IFN-alfa/ribavirin results in an SVR in 45%-50%, whereas in patients with GT-2 and GT-3, 24 weeks of PEG-IFN-alfa/ribavirin results in an SVR in 80%. Nearly all patients who achieve a SVR are cured of infection \[[@CR8]\] and have a reduced risk of hepatocellular carcinoma and death \[[@CR9]\]. However, treatment with PEG-IFN-alfa/ribavirin causes numerous adverse events including fatigue, flulike symptoms, gastrointestinal disturbances, psychological symptoms, and hematologic abnormalities \[[@CR10]\]. These adverse events lead to decreased adherence, dose reduction, and early discontinuation. Patients who are able to maintain at least 80% adherence to their drug regimen have the highest likelihood of achieving a SVR, but treatment-emergent comorbid events (TECs) commonly limit adherence \[[@CR10]--[@CR12]\].
In the initial treatment of HCV infection, the addition of a protease inhibitor (telaprevir or boceprevir) to PEG-IFN-alfa/ribavirin (triple therapy) significantly improved SVR \[[@CR13], [@CR14]\]. However, adverse events were reported at higher rates among users of triple therapy than among users of PEG-IFN-alfa/ribavirin \[[@CR13], [@CR14]\]. Although a small subset of patients receiving triple therapy who have a rapid response can receive a shorter course of treatment, the 47%-77% premature discontinuation rates of PEG-IFN-alfa/ribavirin reported in clinical care settings \[[@CR15], [@CR16]\] would likely be similar with triple therapy.
Premature discontinuation of treatment and nonadherence may cause viral resistance and increase costs. Although combination therapies to treat HCV may cost between \$23,000 and \$78,000 per year (depending on GT, patient weight, and drug selected) \[[@CR17]\], treatment with PEG-IFN-alfa/ribavirin has been shown to be cost effective when patients are cured \[[@CR18]\]. In addition, nonadherent patients have higher HCV-related medical (i.e., excluding medication) costs than adherent patients \[[@CR19]\]. However, despite their frequency and effect on treatment success, little is known about the cost associated with adverse events due to treatment with PEG-IFN-alfa/ribavirin itself \[[@CR20]\]. The objective of this study was to estimate the incidence of TECs and the incremental costs of treating these events in insured patients initiating PEG-IFN-alfa/ribavirin treatment for chronic HCV infection. Secondary objectives were to explore, in a managed care population, the rate of discontinuation of PEG-IFN-alfa/ribavirin therapy and the temporal pattern of the costs of these TECs.
Methods {#Sec2}
=======
Study design and data source {#Sec3}
----------------------------
This study was a retrospective cohort analysis that used data from the i3 Ingenix LabRx database spanning the 4-year period from 7/1/05 to 6/30/09. This database is a Health Insurance Portability and Accountability Act (HIPAA)-compliant administrative claims database of 8--10 million covered lives, representing all major regions of the US. The database contains deidentified adjudicated pharmacy and medical claims submitted for payment by providers, healthcare facilities, and pharmacies and includes information on physician visits, medical procedures, hospitalizations, drugs dispensed, and tests performed. Healthcare charges are reported (medical, inpatient, and pharmacy) in this database, but paid claims and costs are not. Also available are member enrollment and benefit information as well as limited patient, provider, and hospital demographic information. Since the study did not involve direct contact with human subjects, did not involve an intervention, and did not involve collection of any identifiable patient information, Institutional Review Board (IRB) approval was not required.
Study population {#Sec4}
----------------
The study included treatment-naive patients with HCV infection who began treatment with PEG-IFN-alfa/ribavirin during a 2-year period between 7/1/06 and 06/30/08. PEG-IFN-alfa and ribavirin were identified using National Drug Codes. The date of the first medication fill for PEG-IFN-alfa/ribavirin within this period was defined as the index date. The 24-month study period included 12 months before and 12 months after the index date. Treatment-naive HCV-infected patients were defined as those who did not fill any prescriptions for PEG-IFN-alfa prescription for at least 12 months before the first such fill. HCV infection was defined by the presence of at least 1 medical claim with an ICD-9-CM (International Classification of Diseases, 9th Revision, Clinical Modification) code for HCV (070.41x, 070.44, 070.51, 070.54, 070.7x) during the preindex period.
Patients were excluded if they were \<18 years of age or if they were not continuously enrolled during the 24-month study period. We also excluded individuals who initiated therapy at a dose not recommended by the manufacturer \[[@CR21]--[@CR24]\]. Finally, we excluded patients with certain medical claims during the preindex period, including those with an ICD-9-CM code for hepatitis B (070.2x, 070.3x),or hematologic malignancies for which interferon may have been indicated \[leukemia (204.xx-208.xx), Hodgkin's lymphoma (201.xx), non-Hodgkin's lymphoma (200, 202.0-202.2, 202.8), multiple myeloma (203.0-203.1, 238.6), acute lymphocytic leukemia (204.0), chronic lymphocytic leukemia (204.1), acute nonlymphocytic leukemia including acute myeloid leukemia (205.0), acute monocytic leukemia (206.0), chronic myeloid leukemia (205.1), or other leukemias (204.2, 204.8-204.9, 205.2, 205.8-205.9, 206.1-206.2, 206.8-206.9, 207.8, 208.0-208.2, 208.8-208.9)\].
Outcome measures {#Sec5}
----------------
The primary outcome variable was net incremental cost, which was calculated as the difference between preindex and postindex cost for a prespecified list of TECs (see following paragraph) and their treatments, excluding the cost of PEG-IFN-alfa/ribavirin therapy. Charges for TECs were taken from medical claims consistent with one of these events or pharmacy claims with National Drug Codes for medications to treat the events. Charges for visits lacking a code for one of the listed TECs were not included. New TECs were defined by a medical or pharmacy claim in the postindex period that was not present in the preindex period.
TECs were grouped as blood disorders (anemia, neutropenia, thrombocytopenia), gastrointestinal disorders (nausea/vomiting, diarrhea), endocrine disorders (diabetes, hyperthyroidism, hypothyroidism), psychiatric disorders (depression, anxiety disorders, bipolar disorders, insomnia), skin and subcutaneous disorders (alopecia, skin rash), and other disorders (dyspnea, fatigue, headache). Medications used to treat TECs were grouped similarly and included those to treat blood disorders (epoetin alfa, darbepoetin, filgastrim, and eltrombopag), endocrine disorders (antidiabetes medications and thyroid agents), psychiatric disorders (anxiolytics, antidepressants, antipsychotics/antimanics, and hypnotics), skin disorders (topical steroids), and other (antimigraine).
Secondary outcomes included the incidence of new TECs, timing of such TECs, and the rate of discontinuation of PEG-IFN-alfa/ribavirin therapy. PEG-IFN-alfa/ribavirin therapy was considered to be discontinued if there were no prescription fills for both medications for at least 60 days. The date of discontinuation was defined as the last day the patient had both PEG-IFN-alfa and ribavirin available, based on the days of supply as reported in the pharmacy claims. If PEG-IFN-alfa and ribavirin were discontinued on different dates, the earlier date was used.
Covariates {#Sec6}
----------
Other measures included patient age, gender, race, geographic region, and the presence of human immunodeficiency virus (HIV) infection/acquired immune deficiency syndrome (AIDS). Because physician specialty may impact cost and resource use, we identified the specialty of each patient's usual care physician using a validated method. The specialty of the physician prescribing therapy was not directly identifiable in the claims database. Instead, the method counts all claims for evaluation and management services and identifies the physician specialty with the largest plurality of such claims \[[@CR25]\]. Physician specialty was assigned using claims from the preindex period.
Sensitivity analyses {#Sec7}
--------------------
The recommended PEG-IFN-alfa/ribavirin treatment duration is either 24 or 48 weeks, depending on HCV GT. Claims data are used to process payments and generally do not contain clinical information such as test results. However, GT data were available for a small subset of study subjects and we thus explored discontinuation rates for this subset in a sensitivity analysis. In another sensitivity analysis, we restricted the group of patients analyzed to those who were treated beyond 28 weeks of PEG-IFN-alfa/ribavirin treatment, under the assumption that they had GT-1 infection.
Statistical analysis {#Sec8}
--------------------
For descriptive analysis, percentages, medians, means, and standard deviations were calculated for all baseline variables. We reported the percentage of patients with any new TEC as well as the percentage with a new event for each TEC. We reported the mean and standard deviations for TEC charges in both the pre- and postindex periods and for net incremental TEC charges. Treatment discontinuation was reported as the proportion of patients who no longer had PEG-IFN-alfa/ribavirin available at successive 4-week intervals. Charges for TECs were also stratified by duration of therapy. All data transformations and statistical analyses were performed using SAS version 9.2 (SAS Institute, Cary, NC).
Results {#Sec9}
=======
We identified 3,795 patients with HCV infection who were newly treated with PEG-IFN-alfa/ribavirin during the study period. Of these, 1,269 met the inclusion criteria (Table [1](#Tab1){ref-type="table"}), The most common reason for exclusion (2,274 patients) was lack of continuous enrollment for 24 months. One hundred thirty-eight patients were excluded because of a co-occurring diagnosis of hepatitis B or a hematologic malignancy, 16 because they lacked an HCV diagnosis in the pre-index period, and 85 because they had an initial dose of PEG-IFN-alfa/ribavirin other than one of those recommended by the manufacturer. Most of the included subjects were male (63.8%), with 85.1% between the ages of 40 and 59 years old (mean age, 50.2 years). Over half (55.6%) the patients were from the South. Only 3.5% of patients in this sample were coinfected with HIV. The most common treating physician specialty was gastroenterology (32.4%), followed by family practice (28.5%) and internal medicine (25.2%).Table 1**Characteristics of insured patients initiating PEG-IFN-alfa and ribavirin treatment for chronic HCV infection**CharacteristicNo. (%) of insured patients (n = 1269)**Age, mean (SD), y**50.2 (7.7)**Age group (y)**18-2931 (2.4)30-3961 (4.8)40-49405 (31.9)50-59675 (53.2)60-6991 (7.2)70+6 (0.5)**Female**459 (36.2)**Region**East123 (9.7)Midwest242 (19.1)South705 (55.6)West199 (15.7)**Type of health benefits**Commercial1,244 (98.0)Medicaid-HMO25 (2.0)**HIV/AIDS**44 (3.5)AIDS, acquired immune deficiency syndrome; HCV, hepatitis C virus; HIV, human immunodeficiency virus; PEG-IFN-alfa, pegylated interferon alpha.
Table [2](#Tab2){ref-type="table"} shows the proportion of patients who had a diagnosis recorded for each TEC before beginning treatment (preindex period) and the proportion with a diagnosis recorded for the TEC while they were receiving treatment (postindex period). In addition, the proportion of patients who had a new diagnosis while receiving treatment (e.g., no such diagnosis in the preindex period, followed by a diagnosis in the postindex period) is shown. These newly diagnosed conditions are considered to be TECs. Disorders of blood were observed in 14.3% of patients in the preindex period and in 41.2% of patients in the postindex period. In the postindex period, 35.2% of patients had a new diagnosis of a blood disorder and new psychiatric events were observed in 20.1%, endocrine events in 12.1%, and gastrointestinal events in 12.0%. Most (61.6%) patients had at least 1 newly diagnosed TEC. The most common TECs were anemia (29.2%), fatigue (16.4%), depression (11.5%), neutropenia (11.0%), insomnia (8.98%), nausea/vomiting (8.35%), skin rash (7.64%), hypothyroidism (6.93%), thrombocytopenia (6.70%), dyspnea (5.99%), and headache (5.91%).The mean net incremental charge for the predefined TEC in the postindex period was \$6,377, comprising \$2,782 for medical and \$3,595 for pharmacy claims. The largest component of the incremental increase in pharmacy costs was for medications to treat anemia and neutropenia (mean increase, \$3226; Figure [1](#Fig1){ref-type="fig"}). These medications included epoetin alfa, darbepoetin, filgastrim, and eltrombopag. The increase in non-drug-related charges was greatest in patients who completed only 12 weeks of PEG-IFN-alfa/ribavirin treatment (mean: \$6,015; range: -28,831 to 291,838) and least in patients who completed the 48-week treatment (mean: \$291; range: -208,060 to 95,782).Two hundred two patients (15.9%) discontinued PEG-IFN-alfa/ribavirin treatment by 12 weeks. The mean incremental TEC charge for this group was \$7,509 (range: -29,273 to 292,904). The mean incremental charge was \$8,202 (range: -57,384 to 226,864) in the 423 patients who continued treatment for between 25 and 47 weeks and \$6,249 (range: -207,548 to 130,297) in the 372 patients who completed 48 weeks of PEG-IFN-alfa/ribavirin treatment (Figure [2](#Fig2){ref-type="fig"}).Table 2**Frequency of various treatment-emergent comorbid events** ^**a**^ **in insured patients initiating PEG-IFN-alfa and ribavirin treatment for chronic HCV infection**VariableNo. (%) with treatment-emergent comorbid event in the preindex periodNo. (%) with treatment-emergent comorbid event in the postindex periodNo. (%) diagnosed with a new treatment-emergent comorbid event in the postindex period ^b^(n = 1269)(n = 1269)(n = 1269)**Any treatment-related comorbid event**803 (63.3)923 (72.7)782 (61.6)**Blood**182 (14.3)523 (41.2)446 (35.2)Anemia147 (11.6)461 (36.3)371 (29.2)Neutropenia18 (1.42)144 (11.4)139 (11.0)Thrombocytopenia46 (3.62)101 (7.96)85 (6.70)**Gastrointestinal**114 (8.98)175 (13.8)152 (12.0)Nausea/vomiting78 (6.15)124 (9.77)106 (8.35)Diarrhea55 (4.33)75 (5.91)63 (4.96)**Endocrine**317 (25.0)354 (27.9)154 (12.1)Diabetes214 (16.9)217 (17.1)61 (4.81)Hyperthyroidism20 (1.58)30 (2.36)26 (2.05)Hypothyroidism117 (9.22)164 (12.9)88 (6.93)**Psychiatric**262 (20.7)375 (29.6)255 (20.1)Depression181 (14.3)266 (21.0)146 (11.5)Anxiety disorders48 (3.78)54 (4.26)31 (2.44)Bipolar disorders22 (1.73)27 (2.13)17 (1.34)Insomnia82 (6.46)143 (11.3)114 (8.98)**Skin and subcutaneous**59 (4.65)119 (9.38)111 (8.75)Alopecia11 (0.87)17 (1.34)15 (1.18)Skin rash49 (3.86)103 (8.12)97 (7.64)**Other disorders**383 (30.2)423 (33.3)311 (24.5)Dyspnea91 (7.17)98 (7.72)76 (5.99)Fatigue270 (21.3)319 (25.1)208 (16.4)Headache96 (7.57)100 (7.88)75 (5.91)HCV, hepatitis C virus; PEG-IFN-alfa, pegylated interferon alpha.^a^ICD-9-CM diagnosis code for listed events appearing in any diagnosis field.^b^Event appears during period when patient is being treated but not in pretreatment period.Figure 1**Increase in non-medication-related and medication-related charges for treatment-emergent comorbid events between preindex and postindex periods.**Figure 2**Increase in charges from 1-year preindex to 1-year postindex periods for treatment-related comorbid events, by treatment duration.** Data on four mutually exclusive groups are presented: those who discontinued PEG-IFN-alfa/ribavirin between weeks 1--12, weeks 13--24, weeks 25--47, and after week 47. For all groups, annual charges are shown.
Considering all study subjects, 14.2% discontinued treatment before week 12, 32.8% before week 24, and 70.7% before week 48 (Table [3](#Tab3){ref-type="table"}). In a sensitivity analysis restricted to 238 patients for whom HCV GT was known, 81.1% (n = 193) had GT-1/4/6 and therefore these patients should have received 48 weeks of PEG-IFN-alfa/ribavirin. In this group of 193 patients, 14.5% discontinued treatment before week 12, 31.6% before week 24, and 62.7% before week 48. In a second sensitivity analysis restricted to patients treated beyond 28 weeks, and therefore presumed to have predominately GT-1 infection, 39.8% discontinued PEG-IFN-alfa/ribavirin treatment before week 48.Table 3**Duration and discontinuation of PEG-IFN-alfa/ribavirin therapy in insured patients receiving treatment for chronic HCV infection**Patient groupDuration of PEG-IFN-alfa/ribavirin therapy (days)Discontinue before 12 wkDiscontinue before 24 wkDiscontinue before 48 wkNo. of patientsMean(SD)Mediann(%)n(%)n(%)All1,269214.8(113.8)195180(14.2)416(32.8)897(70.7)Patients with known GT238223.1(115.3)21132(13.4)79(33.2)165(69.3)GT 2/345161.3(60.4)1684(8.9)18(40.0)44(97.8)GT 1/4/6193237.5(120.2)28228(14.5)61(31.6)121(62.7)Patients treated \>28 weeks^a^618315.6(57.1)339--------246(39.8)GT, genotype; HCV, hepatitis C virus; PEG-IFN-alfa, pegylated interferon alpha.^a^Presumed to have predominately GT-1 infection.
Discussion {#Sec10}
==========
In an insured US cohort with chronic HCV infection who were treated with PEG-IFN-alfa/ribavirin, most patients experienced one or more TEC. The most commonly identified TECs were anemia, fatigue, depression, and neutropenia. Anemia was identified in 29.2% of patients in this study, compared with in 22% of patients in clinical trials. Rates of several adverse events including fatigue (16%), insomnia (9%), and depression (12%) were lower than those observed in clinical trials (54%, 37%, 22%, respectively) \[[@CR10]\]. This may be expected since patients in clinical practice are typically monitored less aggressively than those in clinical trials. We used ICD-9-CM codes to identify TECs, and thus we only identified events that warranted an interaction with the healthcare system in the form of an office visit, hospitalization, or prescription. In addition, even when identified clinically, ICD-9-CM codes may not be recorded for less-severe events. As a result, the frequency and cost of TECs we report may be underestimated.
Despite this likely downward bias, we estimated that TECs increased direct treatment costs by 25% (\$6,377), with just over half of charges from prescription medications and the rest from office visits, hospitalizations, and other nonprescription charges. Although we did not assess indirect costs in this study, several of the TECs (e.g., anemia, depression, and insomnia) experienced by patients who receive PEG-IFN-alfa/ribavirin therapy have associated indirect costs. For example, in studies of patients with anemia due to chronic kidney disease, hemoglobin values were positively correlated with days worked \[[@CR26], [@CR27]\]. In a study that reviewed healthcare claims from employees at a major US corporation, nearly 20% of the costs of depressive illness were related to disability, and this study did not take into account lost productivity or sick leave \[[@CR28]\]. Depressed individuals took significantly more time off work than those with diabetes, heart disease, or back problems \[[@CR28]\]. In addition, absenteeism and disability expenditures were higher in individuals with insomnia than those without it \[[@CR29]\]. Together these studies suggest that indirect costs of these TECs are likely to be significant.
Shortened treatment duration was not associated with a reduced cost of treating TECs. Nonprescription costs accounted for most of the total costs in patients who stopped treatment by 12 weeks and between 24 and 48 weeks, suggesting that these groups of patients frequently experience TECs. The lowest cost of treating TECs was for patients who stopped treatment between 13 and 24 weeks; most of these patients probably discontinued treatment because they completed treatment for GT 2/3 infection rather than because of TECs. Patients who completed therapy at 48 weeks had the lowest nonprescription costs but high prescription costs, possibly indicating stable management of TECs such as anemia or neutropenia with costly growth factors.
Although comparisons to other studies are difficult because of differences in methodology, the mean treatment duration of 215 days is consistent with estimates from other studies of 172--240 days \[[@CR15], [@CR16], [@CR19], [@CR30]\]. In a retrospective cohort study of Veterans Affairs patients, two of the most common TECs we identified, anemia and neutropenia, were associated with lower persistence with PEG-IFN-alfa/ribavirin treatment (mean 172 days) \[[@CR30]\]. Similar to patients in our study, the investigators identified patients with anemia and neutropenia through both ICD-9-CM codes and by identifying medication use (specifically the use of growth factors). Growth factors may improve treatment tolerability and thus enhance persistence. Dissimilarities in data sources, clinical care settings, data analysis techniques, and patient characteristics may explain some of the differences in the treatment discontinuation estimates. Although we could not distinguish between patients who discontinued therapy because of adverse events and those who discontinued therapy because of lack of virologic response, our findings support the concept that adverse events lead to high rates of therapy discontinuation.
Limitations of this study include those common to claims analysis. A commercially insured population may not be representative of the entire US population nor of treatment patterns in other countries. Lack of clinical data can confound interpretation. In particular, HCV treatment duration is dictated by GT, which was unavailable for most patients in this study. HCV GT was available for a subset of 238 patients. In this subset, 81% had GT 1/4/6, which is similar to other US cohorts, suggesting that our data are representative of the US population \[[@CR31], [@CR32]\]. According to standard treatment recommendations, these patients with predominately GT-1 would be expected to complete 48 weeks of treatment \[[@CR33]\]. This subset of patients had discontinuation rates that were similar to the overall group, which suggests that our assumption that treatment duration was indicative of GT was reasonable. Furthermore, response guided therapy would lead some patients to have treatment recommended beyond 48 weeks. If these patients continued treatment up to 48 weeks, they might be clinically discontinuing therapy prematurely, whereas in our analysis they would not be considered to have done so. Conversely, discontinuation of therapy may be recommended when virologic response is poor, and we could not distinguish between this and TECs as a cause of discontinuation*.* This study was intended to examine only one aspect of cost, not total treatment costs or cost effectiveness. HCV treatment with PEG-IFN-alfa/ribavirin costs between \$23,000 and \$78,000 per year, and newer treatments are substantially more expensive \[[@CR17]\]. Given the timing of this study, none of the patients were treated with newer direct acting antivirals.
In an additional sensitivity analysis of the patients who stopped therapy after 28 weeks and were thus presumed to have GT 1/4/6, 40% stopped before completing the recommended duration of therapy. Patients treated over 12 weeks with GT-1 would be expected to have had a virologic response \[[@CR33]\] and thus discontinuations after 24 weeks would most likely be due to adverse events in those with GT-1. Overall, these sensitivity analyses further support the idea that discontinuation rates in patients treated with PEG-IFN-alfa/ribavirin are high and frequently due to TECs.
Additional limitations include miscoding or undercoding of claims, which may affect the accuracy of cost estimates. Finally, although we were only able to study patients who were receiving double therapy, the cost of TECs may rise with the use of triple therapy (PEG-IFN-alfa/ribavirin plus a protease inhibitor) because gastrointestinal events, skin rash, and anemia are more common with triple therapy than with PEG-IFN-alfa/ribavirin alone \[[@CR10], [@CR14], [@CR34]\].
Conclusion {#Sec11}
==========
Treatment of chronic HCV infection with PEG-IFN-alfa/ribavirin led to frequent TECs and that these events were associated with significant costs. Furthermore, nearly half of treated patients discontinued therapy. It is likely that adding a protease inhibitor to the PEG-IFN-alfa/ribavirin treatment discussed in this study will result in similar or increased TECs. Thus, better-tolerated therapies associated with fewer TECs that reduce healthcare system costs and improve patient continuation rates are needed.
**Competing interests**
Sandhya Sapra and Gilbert L'Italien are employees of Bristol-Myers Squibb Company, which provided funding for this study. Eunice Chang and Michael Broder are employees of Partnership for Health Analytic Research, which was paid by Bristol-Myers Squibb Company to conduct the research described in this manuscript.
**Authors' contributions**
SS was involved in conception and design, analysis and interpretation, drafting and revision of the manuscript, and final approval of the manuscript. EC served as a statistician and was involved in conception and design, statistical analysis and interpretation, revision of the manuscript, and final approval of the manuscript. MSB was involved in conception and design, analysis and interpretation, drafting and revision of the manuscript, and final approval of the manuscript. GL'I was involved in conception and design, drafting and revision of the manuscript, and final approval of the manuscript.
This study was funded by Bristol-Myers Squibb Company. Partnership for Health Analytic Research, LLC (PHAR, LLC) was paid by Bristol-Myers Squibb Company to conduct the research described in this manuscript.
| {
"pile_set_name": "PubMed Central"
} |
Introduction {#S1}
============
Focal epilepsies, in which the seizures originate from a region limited to a part of one cerebral hemisphere ([@B1]), are common and account for more than 50% of all epilepsies ([@B2]). Despite the great improvement in pharmacological research, approximately 30% of patients with focal epilepsies experience seizures that are resistant to anti-epileptic drugs (AEDs) ([@B3]). In these patients the surgical resection of the epileptogenic zone (EZ), the cortical region responsible for the onset, early organization, and propagation of seizures, may be the only way to suppress or reduce seizures. The EZ represents the minimum amount of cortex that must be resected (inactivated or completely disconnected) in order to achieve seizure freedom ([@B4]).
The EZ can sometimes be adequately localized by means of non-invasive investigations including clinical neurological examination, detailed description of ictal signs and symptoms, Magnetic Resonance Imaging (MRI), Positron Emission Tomography (PET), and interictal and ictal scalp video-EEG recordings. However, when the region of seizure onset cannot be precisely identified non-invasively, or when the obtained information is insufficient to exclude the involvement of "eloquent cortical areas" in seizure generation, invasive electrophysiological exploration can be performed from brain structures in order to evaluate their potential involvement in the epileptogenic process using stereo-EEG (SEEG) recordings by means of the stereotactic surgical placement of intracranial electrodes ([@B5], [@B6]), associated with video recordings, with the aim of planning a tailored resection based on individual anatomical and electroclinical characteristics. After invasive recordings, the EZ is currently identified by visually inspecting the video-SEEG recordings, a procedure that requires the involvement of specifically trained neurophysiologists and is inevitably affected by the drawback of subjectivity. Moreover, the rather high rate of failure in epilepsy surgery of extra-temporal epilepsies ([@B7]) underlines that the precise identification of the EZ is still an unsolved problem that requires more sophisticated methods of investigation.
Over the last years, great efforts have been made to develop and implement advanced signal analysis techniques able to improve the identification of the EZ. Particular attention has been paid to those methods aimed at quantifying and characterizing the interactions and causal relationships between neuronal populations, since is nowadays well assumed that epileptic phenomena are associated with abnormal changes in brain synchronization mechanisms, and a number of studies have shown that seizures are associated with the abnormal synchronization of distant structures ([@B8]--[@B10]). The aim of this review is to provide an overview of the different intracranial EEG signal processing methods used to identify the EZ, with particular attention being given to the methods aimed at characterizing the brain connectivity. In addition, the approach based on graph theory will be described, since the study of the topological properties of the networks has strongly improved the study of brain connectivity mechanisms.
Brain Connectivity {#S2}
==================
It is well known that human brain can be considered as a hierarchical complex structure, and that most of the brain functions are based on interactions among neuronal assemblies distributed within and across distinct cerebral regions ([@B11]). The concept of brain connectivity can be subdivided into three main categories ([@B12]--[@B15]) (a) anatomical (or structural) connectivity, which indicates the set of physical or structural connections linking neurons; (b) functional connectivity, defined as the temporal correlation expressed in terms of the statistical dependence between spatially remote neuronal populations; and (c) effective connectivity, which refers to the influence that one neural system exerts over another, thus taking into account the direction of the information flow from one region toward another.
Several techniques have been developed to study interactions in time and/or frequency domain, and in both linear and non-linear contexts. An important distinction between methods aimed at estimating connectivity is that between bivariate and multivariate measures ([@B16], [@B17]). Bivariate methods are able to evaluate the existence of interactions by considering only couple of signals; this can lead to the identification of spurious connections, as these methods do not allow to distinguish between direct and indirect relationships. Many recent papers have extensively described the advantages and disadvantages of bivariate measures, both in the field of functional and effective connectivity ([@B18]--[@B21]). On the contrary, multivariate measures allow the exploration of the whole network, by considering the entire set of signals in the same model.
A second important point is that functional connectivity, although useful in characterizing brain networks, does not provide any information concerning the direction of interactions, an issue very important in the field of EZ identification. To overcome these limitations, directed connectivity measures have been developed with the aim of estimating effective causality.
A large group of effective connectivity measures are based on the concept of Granger causality ([@B22]), initially defined for bivariate processes *x*(*t*) and *y*(*t*): *x*(*t*) Granger causes *y*(*t*) if the knowledge of the past of both *x*(*t*) and *y*(*t*) reduces the variance of the prediction error of *y*(*t*), in comparison with the knowledge of the past of *y*(*t*) alone. The lack of reciprocity makes it possible to evaluate the direction of the information flow between the signals. Granger's definition of causality can be easily implemented by means of autoregressive (AR) models, and extended to the case of multivariate systems ([@B23], [@B24]).
Several directed connectivity measures have been developed on the basis of the concept of Granger causality derived from the multivariate autoregressive (MVAR) modeling. Among them, directed transfer function (DTF) ([@B25]) and partial directed coherence (PDC) ([@B26]) are effective connectivity measures based on MVAR coefficients transformed into the frequency domain. PDC is particularly interesting because of its ability to distinguish direct and indirect causality flows. Particular care must be employed in using PDC, DTF, and other Granger causality-based measures of effective connectivity since they require that the signals are stationary. In epilepsy research this condition is often not matched, especially for ictal discharges. Recently, time varying versions of effective connectivity estimation methods, such as the ADTF (Adaptive DTF) ([@B27], [@B28]), have been proposed and successfully applied on intracranial signals of patients suffering from refractory epilepsy ([@B29]).
The main limitations of these approaches, however, are that the MVAR modeling assumes the hypothesis of the linearity of the underlying process and that it does not allow to study systems with high number of channels, because the number of parameters to be estimated must be lower than the number of samples ([@B30]).
To overcome these drawbacks non-linear, mainly bivariate, approaches have been developed ([@B16], [@B31]). Among these, one method largely applied in the field of intracranial EEG is the non-linear regression index *h*^2^, introduced by Pijn and Lopes da Silva ([@B32]) to analyze EEG signals, and subsequently extended to SEEG by Wendling et al. ([@B33]). This approach has mainly been used to investigate the connectivity between the temporal neocortex and limbic structures in patients with temporal lobe epilepsy (TLE) ([@B34], [@B35]).
Graph Analysis {#S3}
==============
The study of the functional or effective connectivity by means of linear or non-linear methods may not be enough to grasp the full complexity of the brain due to the huge amount of data obtained. Indeed, evaluation of interconnections between all possible pairs of EEG electrodes or SEEG contacts in a particular frequency band will produce huge matrices of correlation data, difficult to interpret and to handle statistically. To overcome these difficulties an approach based on graph theory ([@B36]), derived from the theory of complex networks, could provide useful measures to characterize the topological properties and the functional organization of the brain networks involved both in normal brain functioning and diseases.
According to this approach, the brain is represented as a graph consisting of a set of nodes, or vertices (the EEG electrodes, SEEG contacts, MEG sensors, fMRI voxels, ...) and edges, or links, indicating the presence of an interaction between pairs of nodes. Edges can be directed, as in the case of effective connectivity methods, putting in evidence that the activity of one node may depend on the other and not vice-versa, or undirected when direction was not evaluated (e.g., functional connectivity approach). If the value of the edges are taken into account the graph or network is called weighted, otherwise unweighted.
The procedure to study (intracranial) EEG using a graph theory approach usually includes the following steps. After estimating brain connectivity, an association matrix is generated whose elements are the estimated values between each pair of nodes. A binary or weighted adjacency matrix of undirected (symmetric) or directed graphs is then produced by applying a threshold to each element of the associated matrix. The choice of this threshold is a crucial issue, since it determines the number of the existing connections; therefore different thresholds will produce different graphs, and may affect several network indices ([@B37]).
Graphs can be characterized by various measures, each aimed at describing different topological properties of the network, allowing to characterize the network organization (i.e., random, regular, small-world; see ([@B38]) for a definition of these topological properties) and/or the global or local properties of each node of the network.
Since the main aim of this review is to characterize specific properties of the EZ, we will focus on the indexes representing node's specific properties. For a more detailed description of these indexes, including their mathematical representation, and other graph measures, see ([@B12], [@B31], [@B39], [@B40]).
Graph indexes can be broadly divided into three main categories -- measures of centrality, segregation, integration -- according to the network properties that they better describe.
Centrality measures the structural and functional importance ([@B40]) of each node with respect to the rest of the network; it is one of the main measures used to identify the hubs of a network, that is, nodes that interact with many other regions, playing a key role in functional integration.
Main measures of centrality are: Degree: the total number of edges connected to a node. In directed networks, is possible to distinguish between in-degree, the number of inward links, and out-degree, the number of outward edges. The mean degree of a graph is the average degree over all vertices.Density: the actual number of edges divided by the total number of possible edges in a graph.Betweenness centrality: the ratio between the number of shortest paths (defined as the smallest number of edges between two nodes) passing through a specific node and the total number of shortest paths in the network. It accounts for the importance of a node in facilitating interactions between other nodes in a network.Eigenvector centrality. This index also measures the importance of a node on the basis of the number (and strength) of links to other nodes, but it takes also into account if the node has strong connections with others having themselves a central position within the network.
Segregation refers to the existence of specialized brain areas where specific processes occur such as those responding to specific sensory inputs. Main measures of segregation are: Clustering coefficient: it measures the degree to which nodes in a graph tend to cluster together and is defined as the fraction of triangles around a node over the total number of possible triangles; it represents the fraction of a node's neighbors that are also neighbors of each other ([@B38]).Modularity is a more sophisticated measure able to detect and quantify the presence of segregated activity ([@B40]). A module represents a cluster of densely interconnected groups of nodes. In order to partition the graph into modules, it is necessary to find the optimal community structure, defined as a subdivision of the network into non-overlapping groups of nodes in a way that maximizes the number of within-group edges, and minimizes the number of between-group edges. Once a graph has been partitioned into modules, other two indexes can be defined.Participation coefficient: measure of diversity of inter-modular connections of individual nodes.Within degree: within-module version of degree centrality.
Integration refers to the ability to combine specialized processes distributed into different brain regions. Main measures of integration include: Shortest Path length: it measures the shortest path (or distance) between each pair of nodes, where the path is a sequence of distinct nodes and edges representing potential routes of information flow between pairs of regions. The path length is infinite in case of disconnected pairs of nodes. The average shortest path length between all pairs of nodes is called the characteristic path length of the network and is the most commonly used measure of functional integration.Efficiency: the efficiency of a connection between two nodes is defined as the inverse of the path length. Global efficiency is the average of all the inverse shortest path lengths ([@B12]), while local efficiency measures the efficiency of the connections between the neighbors of a node when the node itself is removed. The global efficiency may be computed on disconnected networks, since paths between disconnected nodes have zero efficiency.
Graph theory has been widely used to investigate epilepsy. It is commonly assumed that epileptic seizures are due to excessive neuronal firing and synchronization, but the underlying mechanisms are still not clear. For a recent review in this field see ([@B41]). The first indication that network analysis might be helpful in understanding epilepsy came from the simulation study of Netoff and others ([@B42]) on models of hippocampal networks. The authors showed that changes in the network topology could induce transitions from normal behavior to seizure. After this study, several works investigated the global network topology in patients with different types of epilepsy. Ponten and others ([@B43]), studying EEG recordings reported that absences are characterized by an increase in synchronization and that the functional network topology changed toward a more ordered pattern when compared to pre-ictal network configuration. Chavez and others ([@B44]) analyzed the connectivity topology of brain networks extracted from MEG signals of patients with absence seizures recorded at rest and found the brain networks of patients to display a richer connectivity with a clear modular structure with respect to controls. van Dellen and others ([@B45]), investigating the effects of on-going TLE, showed that functional connectivity was lower in patients with longer TLE history, and that longer TLE duration was correlated with more random network configuration.
Bartolomei and colleagues ([@B46]) investigated the topological properties of the epileptogenic networks of 11 patients presenting drug-resistant mesial temporal lobe epilepsy (MTLE) using as control group 8 patients with neocortical epilepsy. They found that the network organization of the interictal SEEG activity in both groups were characterized by a small-world model; network topology was however more altered in the MTLE group with respect to non-MTLE patients and corresponded to a more regular (less random) configuration, interpretable as an increase of local and a decrease of long distance connections. The authors suggested that these changes in topology could be a potential biomarker of epileptogenic network.
Overall, these studies indicate that a complex network approach may offer new methodologies not only for improving our understanding on the basic mechanisms of epilepsy, but also could potentially yield quantitative measures to be used as indicators or biomarkers for early detection and diagnosis of epilepsy, the assessment of its course, and for the evaluation of the effect and the efficacy of treatments.
EZ Localization {#S4}
===============
So far, only a few studies have investigated network alterations in relation to the EZ using graph theory in patients undergoing surgery for intractable epilepsies. Wilke et al. ([@B29]) used DTF and betweenness centrality to identify critical network nodes during ictal and interictal electrocorticogram recordings. They found that the betweenness centrality correlated with the location of the resected cortical regions in patients who were seizure-free following epilepsy surgery. Furthermore, the better outcome was associated with the resection of brain regions showing the highest betweenness centrality. This finding suggested that critical highly interconnected nodes played a crucial role in seizure onset and spreading. Importantly, the authors observed patterns of altered network interactions (mainly in the gamma band) not only during seizures, but also during interictal spike activity and random non-ictal periods. This result supported the hypothesis that functional brain networks of patients with focal epilepsies may be altered even during the seizure-free periods ([@B47]).
In a very recent paper, van Mierlo et al. ([@B48]) studied the temporal changes in effective connectivity during the first 20 s of ictal rhythmic intracranial EEG activity in the 3--20 Hz band, in eight patients suffering from focal intractable epilepsy who underwent successful resective surgery. In all the patients, the authors found that the SEEG contact showing maximal ictal out-degree was among those visually classified by clinicians as belonging to the seizure onset zone ([@B49]), and it was always included within the resected brain region. Furthermore, the patient-specific connectivity patterns were consistent over the majority of seizures, suggesting that most of the seizures originated from the same area and that the same driving hubs play a leading role into the epileptogenic network.
We ([@B50]) studied the changes in connectivity patterns in 10 patients with Taylor-type focal cortical dysplasia (type II FCD) under interictal, pre-ictal, and ictal conditions to characterize the network properties of the SEEG signals recorded from inside the dysplasia and distinguish them from those of other regions involved in ictal activity or not. We selected this type of focal epilepsy in order to validate the appropriateness of our approach because this type of dysplasia is intrinsically epileptogenic ([@B51]), the EZ normally overlaps the dysplasia. We applied the PDC to study the effective connectivity in the frequency domain, and characterized the epileptogenic network by means of out-density and betweenness centrality measures.
Our findings indicated that the region inside the dysplasia was characterized by abnormal out-going connectivity in comparison with the other examined areas. The main changes were observed in the gamma band, between 30 and 70 Hz. This specific connectivity pattern was also present in the interictal state, temporally distant from seizures several minutes, and even in the absence of obvious epileptiform activity. This suggests that key information that can help clinicians in localizing the EZ are included in the interictal SEEG activity and that effective connectivity and graph theory indexes are useful tools capable of disclosing them.
As a further step, we compared the differences in connectivity patterns occurring between the interictal, pre-ictal, and ictal conditions. The time course of the increase in synchronization of SEEG activity recorded from the dysplasia and other regions showing epileptiform activity was different; indeed a significant increase in out-going synchronization between the contacts within the EZ and from the EZ toward other areas occurred during the pre-ictal period with respect to the interictal state, whereas an increase of interactions in regions outside the dysplasia but involved in the ictal events occurred later on, at the time of seizure onset, marking the significant differences in connectivity between the ictal and pre-ictal state.
Altogether, our findings suggest that, in patients with type II FCD, the region inside the dysplasia plays a leading role in generating and propagating ictal EEG activity, and in recruiting other distant areas to become involved in the seizure; this area acts as an abnormal hub of the epileptic network that originates and sustains the seizures. The cortical areas outside the dysplasia involved in the SEEG ictal activity, act essentially as "secondary" generators of synchronous activity. The leading role of the dysplasia may account for the good post-surgical outcome of patients with type II FCD because the resection of dysplastic tissue removes the entire EZ responsible for seizure onset.
Conclusion {#S5}
==========
Accurate localization of the EZ is the goal of pre-surgical work-up in patients with drug-resistant focal epilepsies. Over the last years, much research has been dedicated to develop advanced signal processing techniques with the aim of studying the topological properties of functional brain networks in patients with epilepsy. Under the hypothesis that the brain tissue associated with the EZ is differently connected within an epileptic network compared to other regions, in the last years few groups have begun to investigate how the EZ gives rise to network alterations in patients with focal epilepsy, using an approach based on brain connectivity and graph theory. These studies indicate that this network-based approach may add new and valuable information, providing quantitative measures useful for localizing the EZ or for greatly reducing the number of contacts.
Interesting and promising is the finding that network alterations related to EZ are present and could be detected also during EEG interictal periods, since it suggests that, in the future, network-based algorithms could potentially be used to reduce the need for long-term monitoring in patients with drug-resistant focal epilepsy.
More studies in larger and possibly multicentre patient populations that include patients with drug-resistant focal epilepsy of different (or unknown) etiology, however, are expected to clarify which network measures work best in localizing the epileptogenic network and to validate them.
Conflict of Interest Statement {#S6}
==============================
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
This work was supported by the EU Project Grant FP7-ICT-2009-6-270460 ACTIVE.
[^1]: Edited by: Emanuela Formaggio, Foundation IRCCS San Camillo Hospital, Italy
[^2]: Reviewed by: Peter Halasz, Hungarian Sleep Society, Hungary; Marino M. Bianchin, Universidade Federal do Rio Grande do Sul, Brazil
[^3]: This article was submitted to Epilepsy, a section of the journal Frontiers in Neurology.
| {
"pile_set_name": "PubMed Central"
} |
This study funded by AHRQ examines the implementation and scale-up of Lean methodology for enhancing value in a large, ambulatory care delivery system. While evidence indicates that Lean techniques can lead to higher quality care at lower cost, its success or failure is inextricably tied to the views and activities of frontline care providers who are the daily implementers of intervention components. This study explores how such views and activities impacted system-wide efforts to redesign primary care.
Our analysis was guided by a modified version of the Consolidated Framework for Implementation Research, which articulates various \"measures\" of success when implementing process redesigns. Drawing on over 100 in-depth interviews with physicians and staff, we sought to understand the extent to which new Lean workflows were accepted and adopted into practice, and the contextual factors that impacted implementation success.
Frontline perceptions of Lean\'s potential to enhance value were impacted in part by: local dynamics of the care team; perceived skill or competency of team members (namely, medical assistants and licensed vocational nurses) in taking on new roles or scopes of work; and physicians\' own perceived efficiency, or lack thereof, prior to the introduction of Lean redesigns. The implementation strategy used by the organization was also critical. Physicians and staff who viewed the effort as \"top-down\" or \"inflexible\" were less likely to comply with changes. Even those who expressed positive views about Lean as an overall strategy for redesigning care, but who found the implementation process excessively top-down, were less likely to adopt more efficient workflows.
Gaining \"buy-in\" from frontline providers is critical to implementing care processes that are designed to improve the delivery of health care. Understanding how clinical insiders\' views inform their decision to embrace or reject changes may be instructive for organizations attempting to implement similar initiatives.
| {
"pile_set_name": "PubMed Central"
} |
Background {#Sec1}
==========
Exercise practice is a non-pharmacological treatment for a series of diseases \[[@CR1]\]. Along with dietary control, appropriate exercise programs are proposed to treat and attenuate obesity \[[@CR2]\] and its associated cardiovascular disorders such as hypertension \[[@CR3]\]. It is known that hypertension and obesity frequently coexist \[[@CR4]\], affecting millions of people worldwide \[[@CR5]\].
Gut microbiota have been recently indicated as having a close relationship with obesity, where the microbiota of an obese subject presents an enhanced ability to harvest energy and accumulate fat \[[@CR6]\]. The gut harbors the greatest density of these microorganisms in the body (e.g. \~up to 1.5 kg in the human gut) \[[@CR7]\] with Firmicutes, Bacteriodetes and Actinobacteria constituting the dominant phyla \[[@CR8]\]. Moreover, obesity is associated with reduced microbiota diversity at phylum level \[[@CR9]\], seen in rodents and in humans \[[@CR10], [@CR11]\].
The gut is a dynamic environment, highly exposed to environmental factors such as diet \[[@CR12]\], antibiotics \[[@CR13]\], pathogens \[[@CR14]\] and lifestyle \[[@CR15]\], which constantly interact with microbial communities. In addition, gut microbiota is also shaped throughout life by host-related factors such as host-genotype \[[@CR16]\]. Disturbance within gut microbiota has been reported to influence host susceptibility to pathogens and pathological conditions such as gastrointestinal inflammatory diseases and obesity \[[@CR17]\]. Moreover, hypertension induction was also seen to alter gut microbiota \[[@CR18]\].
It has been proposed that dysbiosis and pathologies associated with unbalanced gut microbiota may be prevented or treated with prebiotics, probiotics and fecal microbiota transplantation \[[@CR19], [@CR20]\]. In addition, controlled exercise intensities are related to protective effects on the gastrointestinal tract, including a reduced risk of colon inflammation and cancer \[[@CR21]\]. It is proposed that exercise may reduce intestinal transit time, diminishing the contact between the colon and cancer-promoting agents \[[@CR22]\].
Recently, exercise has also been shown to induce alterations within microbiota composition \[[@CR22], [@CR23]\], which suggests that exercise may be included as a possible therapeutic factor along with diet, prebiotics and other treatments. Since exercise plays a prominent role in metabolic regulation and energy expenditure, it might modulate host-microbiota interaction, affecting the host metabolism. Although these relations are still unknown, exercise may enhance the strategies for obesity control, along with other actions such as microbiota transplant \[[@CR20]\].
Although voluntary exercise was shown to alter microbiota in non-pathological animals \[[@CR22]--[@CR24]\], its effects on gut microbiota still need to be further investigated in pathologic phenotypes and through controlled parameters such as exercise volume and intensity. Therefore, in the present study, we proposed to examine the effect of controlled moderate exercise intensity on gut microbial status in rats with different phenotypes, by using pyrosequencing of 16S RNAs genes from fecal microbiota samples. To our knowledge, this is the first study to use controlled exercise parameters and distinct animal strains with different obesity and hypertension genotypes.
Analyzing 16S rRNA sequences revealed a similar microbiota profile shared between Wistar and Hypertensive rats, with both being divergent from Obese rats. Exercise was shown to enhance bacterial diversity and to alter microbial communities at the species level in all animal lineages. Thus, these data contribute to the emerging knowledge regarding the effect of exercise on gut microbiota, but further studies should be performed to establish the mechanism by which exercise signals in the bacterial community and to determine the impact of these modulations on host homeostasis.
Methods {#Sec2}
=======
Animals {#Sec3}
-------
Animals were obtained from the Federal University of São Paulo, Brazil (UNIFESP) and started the experiment at \~18 weeks of age. Three different strains from two different genotypes were used: an obese genotype, homozygous (fa/fa) obese Zucker rats \[[@CR25]\] (Obese rats; n = 5, 389.4 ± 21 g) and a hypertensive genotype, spontaneous hypertensive rats (Hypertensive rats; n = 5, 227.4 ± 29.3 g), a strain obtained by the selective breeding of Wistar-Kyoto rats with high blood pressure \[[@CR26]\]. A strain of Wistar rats (WR; n = 5, 223.2 ± 27.3 g) was used as normotensive control for SHR and as a non-obese phenotype \[[@CR26]\] (Additional file [1](#MOESM1){ref-type="media"}). All animals were allocated to collective cages according to their lineages, being kept in a 12 h light--dark cycle environment, with food and water ad libitum*.* All experimental procedures and interventions in the present study, involving animal-welfare, were carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health and were approved by the local ethics committee for standards in animal use, at the Institute of Biological Sciences, University of Brasilia, Brazil (UnBDOC no. 48695/2010); these were also in accordance with international standards. After experimental procedures, all animals were deeply anesthetized with 2% Xylazine (50 mg.kg^−1^) and 10% Ketamine (80 mg.kg^−1^) and euthanized by cervical dislocation. Through the entire experiment, all efforts were made to minimize animal suffering.
Exercise training {#Sec4}
-----------------
Before the training period, all animals underwent a familiarization period on a treadmill device (Li 870, Letica Scientific Instruments, Barcelona, Spain) for 2 weeks to reduce external stress. Duration and speed on treadmill were increased progressively (up to 12.5 m.min^−1^ for Obese rats; 20 m.min^−1^ for Hypertensive and Wistar rats) as previously described \[[@CR27], [@CR28]\]. In addition, blood pressure of Wistar and Hypertensive rats was measured by the tail-cuff method \[[@CR29]\] at the beginning of the experiment to characterize the hypertensive phenotype from the SHR strain (171.4 ± 7.7 mm.Hg^−1^ for Hypertensive rats and 128 ± 5.9 mmHg^−1^ for Wistar rats) (Additional file [2](#MOESM2){ref-type="media"}). As regards the Obese rats group (obese (fa/fa) Zucker rat), these animals are homozygous for the *fa* allele and are one of the most common models of genetic obesity where rats normally exhibit hyperlipidemia, hyperinsulinemia, hyperphagia and a significant weight gain by the 3rd to the 5th week of age \[[@CR25]\]. In this model, conflicting results were reported on whether these obese Zucker rats are hypertensive compared to the lean control rats, showing that blood pressure is not elevated in this model \[[@CR30], [@CR31]\]. However, the obesity phenotype may enhance arterial peripheral resistance and the animal may develop hypertension secondary to obesity and to secondary mechanisms \[[@CR32], [@CR33]\]. Considering that obese Zucker rats do not present a regular pattern of blood pressure as reported by different studies, the systolic blood pressure was not measured in this group.
After the adaptation period, all animals were trained for 30 min per day, 5 days per week for 4 weeks. Running intensity was set corresponding to maximal lactate steady state (MLSS), previously identified in obese Zucker rats \[[@CR28]\] and SHR rats \[[@CR27]\]. Therefore, for Obese rats, running velocity was set at 12.5 m.min^−1^ and for Hypertensive and Wistar rats at 20 m.min^−1^. A new MLSS identification was performed after the fourth week of exercise training to assess each animal's cardiovascular adaptation.
Blood lactate analysis {#Sec5}
----------------------
Capillary blood samples (10 μL) were collected through a small incision in the distal portion of the tail of the animals at rest and every 5 min during the MLSS test. Capillary blood samples were placed in microtubes (0.6 mL) containing 20 μL of 1% sodium fluoride and stored at −20°C. Analyses were performed by electro-enzymatic method with YSI 1500 Sports (Yellow Springs, OH, USA). The MLSS was considered when there was no increase over 1 mmol.L^−1^ of blood lactate from 10 to 25 min of exercise tests \[[@CR34]\].
Fecal DNA extraction and barcoded pyrosequencing of the 16S rRNA gene {#Sec6}
---------------------------------------------------------------------
Fecal content from all animals was individually collected in three replicates before exercise training period and in vivo 24 h after the last session of exercise (Additional file [1](#MOESM1){ref-type="media"}). Fecal samples were stored in RNA later® (Life technologies) until samples were frozen at −80°C. Three out of five individual fecal samples from each animal lineage were randomly selected. Fecal microbial DNA was extracted from \~0.25 g using the PowerFecal DNA Isolation Kit (MoBio, Carlsbad, CA, USA) according to the manufacturer's instructions. The triplicate DNA extractions were not pooled.
The bacterial community partial 16S rRNA gene was amplified with primer pair 787 F-1492R \[[@CR35]\]. For pyrosequencing analysis, the primer set was modified, where the forward primer included the Roche 454-A pyrosequencing adapter and a 12-bp barcode (unique to each sample), while the reverse primer included only the Roche 454-B pyrosequencing adapter. The 20 μl reaction mixture for the PCR contained approximately 10 ng of metagenomic fecal DNA, 1X PCR buffer (Invitrogen), 3.0 mM MgCl2, 10 pmol of each primer, 0.25 mM dNTP, and 1.5 U Taq DNA polymerase (Invitrogen). The cycling protocol started with an initial denaturation step of 3 min at 95°C, followed by 25 cycles of denaturation for 30 s at 95°C, annealing for 30 s at 58°C, and extension for 1.40 min at 72°C, followed by a final extension for 7 min at 72°C and cooling to 10°C. Finally, the rRNA amplicons from bacterial communities were purified with the QIAquick PCR Purification Kit (Chatsworth, CA). The concentrations of the rRNA amplicons were measured by Qubit fluorometer (Invitrogen), and subsequently the massively parallel GS FLX Titanium technology was performed in the Roche 454 Life Sciences Corporation, Branford, CT, USA.
Analysis of 16S rRNA sequences {#Sec7}
------------------------------
A total of 1,398,681 16S rRNA sequences were obtained by the 454 GS FLX Titanium sequencer. All the 16S amplicons were processed by the quantitative insights into microbial ecology (QIIME) pipeline version 1.6.0-dev \[[@CR36]\]. Briefly, all the 16S rRNA amplicons were sorted by their barcodes, and subsequently the reads with a length less than 180 bp, ambiguous sequences, bases with Phred values of \<30 and their primers, barcodes and adaptor sequences were removed. The remaining sequences were submitted to Denoiser algorithm \[[@CR37]\] to remove pyrosequencing errors. Operational taxonomic units (OTUs) were clustered at 97% similarity using an 'open-reference' OTU picking protocol, where sequences are clustered against the Greengenes database \[[@CR38]\] using Uclust \[[@CR39]\]. One of the most-abundant reads from each OTU was aligned using the PyNAST algorithm \[[@CR36]\]. The chimeric OTUs were detected with ChimeraSlayer \[[@CR40]\], and taxonomic classifications were assigned with the naïve Bayesian classifier of the Ribosomal Database Project (RDP) classifier \[[@CR41]\] applying 80% of confidence threshold. Shannon indices and observed richness were used to evaluate community richness, and the unweighted Unifrac algorithm was performed to generate principal coordinate plots (PCoA).
Statistical analysis {#Sec8}
--------------------
Statistical differences between the groups pre-exercise and after exercise were tested using the analysis of similarities (ANOSIM) by permutation of group membership with 999 replicates \[[@CR42]\], and bivariate relationships were measured with Pearson correlations and regression analysis available through QIIME \[[@CR36]\]. Statistical tests on the taxonomic differences between samples were calculated by Welch's *t*-test combined with Welch's inverted method for calculating confidence intervals (nominal coverage of 95%), using the Statistical Analysis of Metagenomic Profiles (STAMP) software version 2.0.0 (STAMP) \[[@CR43]\].
Accession number {#Sec9}
----------------
The 454 FLX Titanium flowgrams (sff files) have been submitted to the National Center for Biotechnology Information (NCBI) Sequence Read Archive database, project number: PRJNA246617.
Results {#Sec10}
=======
Effect of exercise training on aerobic capacity {#Sec11}
-----------------------------------------------
After four weeks of treadmill running exercise at moderate intensity, a novel MLSS assessment evidenced that all animals from each group had enhanced their aerobic capacity as demonstrated by improvement in the MLSS corresponding velocity (15 m.min^−1^ for Obese rats and 30 m.min^−1^ for Hypertensive and Wistar rats) (Figure [1](#Fig1){ref-type="fig"}A). Furthermore, when comparing the initial and final velocity of exercise, a significant reduction was evidenced in blood lactate concentration (BLC) of \~49% for Wistar rats \~39% for Hypertensive rats and \~33% for Obese rats. This reduction in BLC (before vs. after exercise training) demonstrates the effectiveness of the proposed exercise training intensity for all animals (p \< 0.01) (Figure [1](#Fig1){ref-type="fig"}B).Figure 1**Training parameters.** Comparison between exercise training velocity from MLSS (vMLSS) before and after four weeks of treadmill running exercise at moderate intensity, indicating that all animals from each group enhanced their aerobic capacity as demonstrated by improvement in the MLSS corresponding to velocity (15 m.min^−1^ for Obese rats and 3 0 m.min^−1^ for Hypertensive and Wistar rats) **(A)**. When the initial and final velocity of exercise training was compared, a significant reduction was evidenced in BLC of \~49% for Wistar rats, \~39% for Hypertensive rats and \~33% for Obese rats (p \< 0.01) **(B)**.
Composition of fecal microbiota in rat lineages {#Sec12}
-----------------------------------------------
After quality filtering, 889,124 out of 1,398,681 sequences were obtained from fecal samples collected pre-exercise and post-exercise, after 4 weeks of moderate exercise training (Additional file [1](#MOESM1){ref-type="media"}). An average of 49,951 denoised sequences per animal were obtained (average read length of 524.8) which composed an average of 583.1 distinct observed OTUs. Post-training samples presented a higher Shannon index compared to pre-training samples (6.4 ± 0.5 vs. 6.8 ± 0.2) (Additional file [3](#MOESM3){ref-type="media"}). Detailed sequencing information from all individual rats is presented in (Additional file [3](#MOESM3){ref-type="media"}).
Bacterial diversity was assessed by rarefaction measure of observed species against the number of sequences per sample, and the observed OTUs were identified with 97% of identity. Here, the rarefaction measure showed that species diversity (Additional file [4](#MOESM4){ref-type="media"}: A; Wistar rats; B; Hypertensive rats and C; Obese rats) reached a plateau tendency in all samples as the number of sequences increased, indicating that in the present study more sequences are unlikely to yield many additional species. As demonstrated in Additional file 4, the rarefaction curves revealed that OTU richness in post-exercise fecal samples is more species-rich than those found in the pre-exercise fecal samples. This was further evidenced in hypertensive rats and obese rats samples (Additional file [4](#MOESM4){ref-type="media"}B and C).
The relative abundance of the main dominant bacterial phylum from all fecal samples collected before and after exercise training is shown in Additional file [5](#MOESM5){ref-type="media"}A. Here, Firmicutes and Bacteroidetes are the most dominant phyla, followed by Proteobacteria. Firmicutes was shown to be enhanced after exercise training (1.1 fold change, p \< 0.05) (Additional file [5](#MOESM5){ref-type="media"}B), thus being more evidenced in Obese rats (Obese rats; 0.69 ± 0.03 vs. Exercised Obese rats; 0.78 ± 0.04, p \< 0.05). On the other hand, Proteobacteria were shown to be 1.8 fold reduced after exercise training (p \< 0.05) (Additional file [5](#MOESM5){ref-type="media"}C). The Bacteroidetes phylum was shown to be 1.3 fold reduced after exercise only in Wistar rats (Wistar rats; 0.23 ± 0.04 vs. Exercised Wistar rats; 0.17 ± 0.03, p \< 0.05).
Composition of bacterial communities before, during and after exercise training {#Sec13}
-------------------------------------------------------------------------------
The relative abundance at bacterial genus level for all animal lineages in response to exercise training was assigned only to those that presented a minimum variation at significant level (p \< 0.05) (Figure [2](#Fig2){ref-type="fig"}). In Wistar rats (A), Streptococcus was the only genus that presented a significant alteration within its abundance, while untrained rats were more enriched with Streptococcus when compared to post-exercise (p \< 0.05) (Figure [2](#Fig2){ref-type="fig"}A). In hypertensive rats, three genera (Allobaculum, Aggregatiobacter and Sutterella) were shown to be altered by exercise training. Despite minimal variation in the relative abundance of Allobaculum between pre-exercise and post-exercise samples, this genus was enriched by exercise training (p \< 0.05) (Figure [2](#Fig2){ref-type="fig"}B). This was in contrast to Allobaculum, Aggregatibacter and Suturella where both were more abundant in pre-exercise samples (Figure [2](#Fig2){ref-type="fig"}B). Aggregatibacter presented a minimal variation in their relative abundance in fecal samples pre- and post-exercise training; however, exercise was shown to reduce the abundance of this genus (p \< 0.05) (Figure [2](#Fig2){ref-type="fig"}B). The Suturella genus was also shown to be more enriched pre-exercise, with a greater relative proportion of all in this genus in hypertensive rats (p \< 0.05). In Obese rats, Pseudomonas and Lactobacillus were both significantly altered after exercise training (Figure [2](#Fig2){ref-type="fig"}C). Minimal variation in Pseudomonas relative abundance was observed between samples (P \< 0.05), while Lactobacillus presented the higher relative abundance after exercise from all identified genera (p \< 0.05) (Figure [2](#Fig2){ref-type="fig"}C).Figure 2**Effect of exercise on Genus relative abundance.** Proportion of relative abundance for the statistical analyses of Genus level profiles distributed in Wistar rats **(A)**, Hypertensive rats **(B)** and Obese rats **(C)**. The genera were altered between without exercise training (white bar) and with exercise training (black bar). Features with a *α-*value of \< 0.05 were considered significant.
The proportion of sequences (%) of the main bacterial species from fecal samples collected before and after exercise training is shown in box plots (Figure [3](#Fig3){ref-type="fig"}). In pre-exercise fecal samples (Figure [3](#Fig3){ref-type="fig"}A and B), only two species (*Bacteroides acidifaciens* and *Ruminococcus flavefaciens*) presented a significant differential abundance, in contrast to fecal samples post-exercise training, where six species (*Streptococcus alactolyticus, Bifidobacterium animalis, Ruminococcus gnavus, Aggregatibacter pnemotropica* and *Bifidobacterium pseudolongum*) presented a differential abundance (Figure [3](#Fig3){ref-type="fig"}C-G). From all samples (pre and post-exercise) only one species (*Ruminococcus flavefaciens*) was less abundant in obese animals (Figure [3](#Fig3){ref-type="fig"}B), with all other species being significantly more enriched in the obese animals (Figure [3](#Fig3){ref-type="fig"}C-G).Figure 3**Species abundance profile of fecal sample before and after exercise training.** Box plot showing the distribution in the proportion of sequences (%) of main species of each rat lineage (**A**; *Bacteroides acidifaciens*, **B**; *Ruminococcus flavefaciens*, **C** *; Streptococcus alactolyticus*, **D**; *Bifidobacterium animalis*, **E**; *Ruminococcus gnavus*, **F**; *Aggregatibacter pneumotropica*, **G**; *Bifidobacterium pseudolongum*) without exercise training (white box) and with exercise training (black box). The median value is shown as a line within the box and the mean value as a star, p-value for statistical significance was defined as p ≤ 0.05.
Regarding pre-exercise samples, the proportion of sequences from the *Bacteroides acidifaciens* species was significantly more abundant in Obese rats than in Wistar and Hypertensive rats (p \< 0.05) (Figure [3](#Fig3){ref-type="fig"}A). However, an opposite profile is observed in sequences attributed to the *Ruminococcus flavefaciens* species, where greater abundance is seen in Wistar rats followed by Hypertensive rats, with no abundance seen in Obese rats (p \< 0.05) (Figure [3](#Fig3){ref-type="fig"}B). After exercise training, the proportion of sequences indicated that all listed species (*Streptococcus alactolyticus, Bifidobacterium animalis, Ruminococcus gnavus, Aggregatibacter pneumotropica* and *Bifidobacterium pseudolongum*) were more abundant in Obese rats compared to Wistar and Hypertensive rats lineages (Figure [3](#Fig3){ref-type="fig"}C-G respectively). The relative abundance of *Streptococcus alactolyticus* in Obese rats diverged significantly from Hypertensive and Wistar rats (p \< 0.05), while a diminished proportion of sequences was seen in both rat strains (Figure [3](#Fig3){ref-type="fig"}C). The *Bifidobacterium animalis* species was seen to be highly enriched in Obese rats (p \< 0.05) and absent in Wistar and Hypertensive rats (Figure [3](#Fig3){ref-type="fig"}D). In relation to R*uminococcus gnavus,* this species was poor in Wistar and almost absent in Hypertensive rats, but more abundant in Obese rats **(**Figure [3](#Fig3){ref-type="fig"}E). The *Aggregatibacter pneumotropica* species presented a similar profile to the previous species, being also more abundant in Obese rats compared to Wistar and Hypertensive rats (p \< 0.05) (Figure [3](#Fig3){ref-type="fig"}F). Lastly, from the Actinobacteria phylum, *Bifidobacterium pseudolongum* abundance was almost exclusive to Obese rats (p \< 0.05), being almost absent in Hypertensive rats and completely absent in Wistar rats group (Figure [3](#Fig3){ref-type="fig"}G).
Principal coordinates analysis (PCoA) {#Sec14}
-------------------------------------
Principal coordinates analysis (PCoA) of unweighed UniFrac distances was calculated and compared between all fecal samples collected pre and post-exercise from the three rat lineages in order to observe similarity in microbiota composition and the effect of exercise training (Figure [4](#Fig4){ref-type="fig"}). All three biological replicates from each animal lineage (Wistar, Hypertensive and Obese rats) were shown to cluster with a high correlation between them (R = 0.79, P \< 0.001). UniFrac (PCoA) analysis showed that Wistar and Hypertensive rats share a similar bacterial composition, thus clustering far from Obese rats, indicating a distinct bacterial community composition between these rat lineages (Figure [4](#Fig4){ref-type="fig"}). It also indicated that microbiota from Wistar, Hypertensive and Obese rats were significantly altered by exercise training, where pre-exercise samples clustered significantly far from fecal samples collected after four weeks of exercise training (Figure [4](#Fig4){ref-type="fig"}). However, even though exercise altered microbial community composition in every animal lineage, Wistar and Hypertensive rats still maintained a similar bacterial composition, still clustering far from Obese rats microbiota (Figure [4](#Fig4){ref-type="fig"}).Figure 4**Effect of exercise training on bacterial community.** Principal coordinates analysis (PCoA) of unweighted UniFrac distances generated from fecal samples in Wistar rats (squares), Hypertensive rats (circles) and Obese rats (diamonds) collected from triplicate rats without exercise training (white symbols) and with exercise training (black symbols). The result of the ANOSIM similarity analyses confirmed that samples harbor a distinct bacterial community.
Correlation of bacterial abundance and lactate concentration {#Sec15}
------------------------------------------------------------
As shown in Figure [1](#Fig1){ref-type="fig"}A, when comparing the initial and final velocity of exercise training, a significant reduction in BLC was evidenced in all rat lineages, where lower BLC (means of 2.3 mmol.L^−1^, Figure [1](#Fig1){ref-type="fig"}B) is associated to an improved aerobic capacity from trained status when compared to higher BLC (3.8 mmol.L^−1^) from untrained rats (pre-exercise samples from all rat lineages). Therefore, fecal bacterial communities were plotted against BLC to establish a correlation between microbial abundance and training status (Figure [5](#Fig5){ref-type="fig"}). OTUs from two bacterial families (Clostridiaceae and Bacteroidaeae) and two genera (Oscillospira and Ruminococcus) were found to be significantly correlated with BLC. The OTU abundance from both bacterial families was negatively correlated to BLC (Clostridiaceae, R = −0.82 P \< 0.01; Bacteroidaceae, R = −0.73 P \< 0.01). In both cases, higher abundance in OTUs was observed to correlate with lower lactate concentrations, indicating that exercise training may be favorable to the proliferation of these OTUs from both bacterial families (Figure [5](#Fig5){ref-type="fig"}A and B). The relative abundance of OTUs from Bacteroidaceae family was shown to be close to zero when BLC reached \~4 mmol.L^−1^ (Figure [5](#Fig5){ref-type="fig"}B), being associated with untrained status. Regarding the genera, the abundance of OTUs from Oscillospira and Ruminococcus genera presented divergent correlations with BLC. OTUs from Oscillospira were shown to be positively correlated to BLC (R = 0.78 P \< 0.01) and Ruminococcus was negatively correlated (R = −0.75 P \< 0.01) (Figure [5](#Fig5){ref-type="fig"}C and D). The OTUs from Oscillospira genus were shown to be almost absent in lower lactate concentrations, increasing their abundance with higher concentrations over 3.5 mmol.L^−1^, which indicates that exercise training may be seen as an unfavorable factor for a specific OTU from Oscillospira genus and its proliferation in gut environment (Figure [5](#Fig5){ref-type="fig"}C). However, the OTUs from Ruminococcus microbial genus were shown to be more abundant at lower lactate concentration and with almost no abundance at higher BLC, indicating that exercise training may also influence proliferation in this genus.Figure 5**Microbial abundance and blood lactate concentration correlation.** Correlations between the relative abundances of the bacterial communities (OTUs) and blood lactate concentration (mmol.L^−1^). Gut microbiota profile was determined by 16S rRNA pyrosequencing for Wistar rats, Hypertensive rats and Obese rats in fecal samples. Pearson correlation coefficients (*r*) are shown for each taxon (**A**; Clostridiaceae, **B**; Bacteroidaceae, **C**; Oscillospira, **D**; Ruminococcus), with the associated FDR-corrected *P* values.
Discussion {#Sec16}
==========
Several environmental \[[@CR44]\] and host-related factors \[[@CR16]\] are known to affect gut microbiota composition. This dynamic ecosystem, highly susceptible to external agents, has a symbiotic link with host health homeostasis. In this context, imbalanced gut microbiota has been associated with the development of inflammatory gastrointestinal diseases, obesity and altered metabolic status \[[@CR19]\].
Recently, physical activity was shown to modulate gut microbiota in diet induced obesity \[[@CR24]\], and in healthy rodents, altering the microbiota diversity and composition \[[@CR23]\] and increasing n-Butyrate concentration in the cecum \[[@CR22]\]. In contrast to some of these previous studies that used PCR-TGGE \[[@CR22]\] and PCR-DGGE of bacterial 16S rRNA genes \[[@CR23]\], here a robust pyrosequencing of the 16S rRNA genes was used, along with controlled exercise training parameters to investigate this relationship in non-pathologic and pathologic rat models.
Rarefaction measurements and Shannon index indicated that exercise training enhances bacterial diversity in non-pathological Wistar rats and in Obese and Hypertensive rats (Additional file [3](#MOESM3){ref-type="media"} and Additional file [4](#MOESM4){ref-type="media"}). Here, Firmicutes and Bacteroidetes were found to be the most predominant phyla in all animal lineages (Additional file [5](#MOESM5){ref-type="media"}A). This predominance was also seen in mice cecum \[[@CR6]\], and in exercised rats \[[@CR23]\]. Considering all rat lineages, exercise was shown to enhance Firmicutes abundance and to diminish Proteobacteria content (Additional file [5](#MOESM5){ref-type="media"}B, C). Thus, Firmicutes were more abundant in post-exercise samples compared to pre-exercise in obese rats (Obese rats; 0.69 ± 0.03 vs. Exercised Obese rats; 0.78 ± 0.04, p \< 0.05), while Bacteroidetes was shown to be reduced after training only in WR (Wistar rats; 0.23 ± 0.04 vs. Exercised Wistar rats; 0.17 ± 0.03, p \< 0.05). Moreover, Bacteroidetes have been reported to be diminished in obese mice \[[@CR10]\], while the ratio of Firmicutes to Bacteroidetes was shown to change in favor of Bacteroidetes in overweight and obese subjects, compared to the lean group \[[@CR45]\]. Thus, as previously stated by Harris et al., \[[@CR7]\] studying gut microbiota and their relation with metabolic disorders revealed that there is no difference between obese and lean individuals at phylum level. However, data here reported showed a significant alteration in bacterial community abundance at phylum as well as at genus levels (Figure [2](#Fig2){ref-type="fig"}), which could be associated with the effects of exercise and/or pathological conditions.Furthermore, our study revealed a significant alteration in bacterial community abundance at genus and species level as an effect of exercise and/or pathological stimuli (Figure [2](#Fig2){ref-type="fig"}).
In accordance with the PCoA analysis presented in this study (Figure [4](#Fig4){ref-type="fig"}), other studies have also reported a distinction between non-obese and obese microbiota from Zucker^fa/fa^ rats \[[@CR46]\] and ob/ob mice \[[@CR10]\]. We also reported that Wistar and Hypertensive rats share a similar microbiota composition (Figure [4](#Fig4){ref-type="fig"}). In a similar way, it was reported that rats treated with nitric oxide synthase inhibitor NG-nitro-L-arginine methyl ester (L-NAME) develop hypertension, with a variation in cecal microbiota compared to control normotensive rats \[[@CR18]\]. Regarding the effect of exercise, PCoA analysis demonstrated that four weeks of moderate exercise training significantly altered microbiota composition in all rat lineages (Figure [4](#Fig4){ref-type="fig"}). According to our results, different exercise volumes, 6 days \[[@CR23]\], and 5 \[[@CR22]\] and 12 weeks \[[@CR24]\] of voluntary running exercise, were shown to alter microbiota composition, indicating that the microbial community is affected even by a few days of exercise. Together, these data suggest that besides other well-known factors, exercise may be seen as a potential environmental factor capable of modulating gut microbiota.
Here, exercise was also shown to significantly alter six bacterial genera (Figure [2](#Fig2){ref-type="fig"}). Fecal samples were more enriched with Allobaculum (Hypertensive rats), Pseudomonas (Obese rats) and Lactobacillus (Obese rats) after exercise training, while Streptococcus (Wistar rats), Aggregatibacter (Hypertensive rats) and Sutturela (Hypertensive rats) were shown to be more abundant before exercise training was performed **(**Figure [2](#Fig2){ref-type="fig"}). The Lactobacillus genus presented higher abundance after exercise only in Obese rats (13.4% p \< 0.05) (Figure [2](#Fig2){ref-type="fig"}C). In agreement with our data, the recent study of Queipo-Ortuno et al. \[[@CR23]\] revealed that Lactobacillus was also enhanced with exercise in lean rats, with a longer exercise stimulus (6 weeks). Lactic acid bacteria (LAB), represented in our study by Lactobacillus (enriched after exercise), are associated with the mucosal surface of the small intestine and colon in animals, where they produce lactic acid though homo or heterofermentative metabolism \[[@CR47]\]. In this second process, besides lactic acid, CO~2~, acetic acid and/or ethanol are produced \[[@CR48]\], which may contribute to a more acidic environment \[[@CR48]\]. It has been reported that LAB in the gastrointestinal tract leads to positive health benefits with influence on microflora, modulation of mucosal immunity and exclusion of pathogens \[[@CR47]\]. The enrichment of Lactobacillus in Obese rats after exercise may have some influence on gastrointestinal acidity trough the production of acidic compounds (e.g. lactic acid, acetic acid); however, this parameter was not measured in the present study. The capillary blood lactic acid was measured in order to establish aerobic capacity and thus to be used as a parameter for adaptation to exercise. Moreover, the lactate produced by Lactobacillus bacteria is converted into butyrate in the gut through bacteria such as *B. Coccoides and E. rectale,* also found to be enhanced after exercise \[[@CR23]\]. Furthermore, butyrate is shown to be related to mucin synthesis and gut epithelium protection \[[@CR49]\]. In another study, Matsunomo et al. \[[@CR22]\] showed that exercise altered microbiota and enhanced n-butyrate concentration in rats' cecum. Therefore, enhanced Lactobacillus found in Obese rats group may possibly have a positive role in the gastrointestinal environment of these animals. It has been reported that obesity-associated gut microbiota is enriched in some species of Lactobacillus (e.g. *Lactobacillus reuteri*) \[[@CR50]\] while other species (e.g. *Lactobacillus gasseri* BNR17) are involved in metabolism regulation \[[@CR51]\], presenting anti-obesity effects \[[@CR52]\].
In our study, the Sutterela genus was more abundant before exercise training in Hypertensive rats (Figure [2](#Fig2){ref-type="fig"}B). The role of this genus in inflammatory bowel disease has been recently investigated with no relation being found \[[@CR53]\]. Nevertheless, more research is needed to understand the relation of Sutturela with exercise and its possible gastrointestinal protective effect.
It is observed that these alterations in genera are not consistent across all host phenotypes. We believe that part of this inconsistency may be related to the biologic differences peculiar to each host genotype used in the study. Obesity has been shown to modulate gut microbiota \[[@CR10], [@CR11]\], while no study has shown this yet in a hypertensive phenotype. Furthermore, as a first exploratory study to use different genotypes with different pathologies, it is interesting to note that the gut microbiota of 3 different phenotypes (and 2 genotypes) was possibly altered by an external factor such as exercise.
Seven bacterial species were shown to have a significant differential abundance altered between the three animal lineages (Figure [3](#Fig3){ref-type="fig"}). From fecal samples collected pre-exercise, only *Bacteroides acidifaciens* and *Ruminococcus flavefaciens* presented a significant differential abundance (Figure [3](#Fig3){ref-type="fig"}A and B). While *Bacteroides acidifaciens* was more enriched in Obese rats compared to Wistar and Hypertensive rats (Figure [3](#Fig3){ref-type="fig"}A), *Ruminococcus flavefaciens* showed an opposite profile, being more enriched in Wistar rats followed by Hypertensive rats and significantly depleted in Obese rats (Figure [3](#Fig3){ref-type="fig"}B). *B acidifaciens* has recently been shown to have an important role in the production of imunoglobulin (IgA) in the large intestine of mice \[[@CR54]\]. This production plays an adaptive role in the intestinal mucosal immune system \[[@CR55]\]. Since IgA is enhanced in metabolic disorders \[[@CR56]\], the relative abundance of *Bacteroides acidifaciens* in Obese rats (Figure [3](#Fig3){ref-type="fig"}A) may be associated with the role of gut microbiota in the inflammatory signalling peculiar to obesity \[[@CR57]\]. Otherwise, *Ruminococcus flavefaciens* is a cellulolytic bacterium present in the rumen of mammals, and it has been shown to be inhibited by probiotic supplementation (*L. acidophilus* NCFM) in young rats \[[@CR58]\]. Our data indicate that the obesity phenotype from obese rats may suppress this particular species.
In samples collected after training, *Streptococcus alactolyticus, Bifidobacterium animalis, Ruminococcus gnavus, Aggregatibacter pnemotropica* and *Bifidobacterium pseudolongum* were all more abundant in Obese rats (Figure [3](#Fig3){ref-type="fig"}C-G). *Streptococcus alactolyticus* and *Bifidobacterium animalis* species were shown to be present in the gut of obese rats. In contrast to our data, Bifidobacterium is often associated with lean phenotypes \[[@CR19]\]; however, our study showed that this species was completely absent in non-obese Wistar rats and Hypertensive rats (Figure [3](#Fig3){ref-type="fig"}D). Regarding the relative abundance of *Ruminococcus gnavus* in Obese rats*,* this species is known to have an antibacterial effect and to protect the host from pathogens \[[@CR59]\], also found to be reduced in colon cancer tissue \[[@CR60]\]. However, this species was also shown to be enhanced in diverticulitis \[[@CR61]\], which is commonly associated with obesity \[[@CR62]\]. *Bifidobacterium pseudolongum* was another species almost exclusively in Obese rats (Figure [3](#Fig3){ref-type="fig"}G). Moreover, the content of this species was shown to be enhanced in obese mice induced by diet and probiotic administration, when compared to a group of mice without probiotic supplementation \[[@CR63]\].
In the present study, the MLSS was used to assess aerobic improvement as a result of four weeks of exercise training at moderate intensity \[[@CR2]\]. Thus, after exercise training, a significant reduction in BLC was observed in all rat lineages (Figure [1](#Fig1){ref-type="fig"}B), which is associated with an improved aerobic capacity when compared to higher BLC from untrained rats (Figure [1](#Fig1){ref-type="fig"}B). The OTUs from Oscillospira and Ruminococcus families were found to be negatively correlated with BLC (Clostridiaceae, R = −0.82, P \< 0.01; Bacteroidaceae, R = −0.73 P \< 0.01) as were the OTUs from Ruminococcus genus (R = −0.75 P \< 0.01). In these three cases, the greatest relative abundance of OTUs in both families is correlated with lower BLC, indicating that exercise training may be favorable to the proliferation of these specific OTUs (Figure [5](#Fig5){ref-type="fig"}A, B and D). Otherwise, OTUs from Oscillospira presented a positive correlation with BLC (R = 0.78, P \< 0.01) (Figure [5](#Fig5){ref-type="fig"}C). The relative abundance within these OTUs is seen to increase when the concentration of lactate goes over \~3.5 mmol.L^−1^. Since lower BLC during the MLSS test was associated with a more trained status, this result may indicate that exercise training may affect the abundance of the OTUs from this genus.
Conclusions {#Sec17}
===========
These findings suggest that exercise training is capable of altering gut microbiota at genus level, with significant alteration in bacterial composition and diversity in obese, non-obese and hypertensive rats. Exercise was shown to enhance the relative abundance of three genera, with Lactobacillus being the most abundant, while another three genera were shown to be more abundant before exercise training (Streptococcus, Aggregatibacter and Sutterella). Non-obese Wistar rats and spontaneously hypertensive rats were shown to share similar microbiota, unlike Obese rats. Rat lineages were also shown to harbor a differential abundance at species level, and six species were shown to be significantly more abundant in obese rats. Two bacterial families (Clostridiaceae and Bacteroidaceae) and two genera (Oscillospira and Ruminococcus) were also shown to significantly correlate with blood lactate accumulation, while exercise was shown to be favorable to the two families and Ruminococcus genus in opposition to Oscillospira. In conclusion, this is the first study to use controlled exercise parameters to assess gut bacterial community modification in three different animal lineages, which may reflect the potential of exercise to alter gut microbial community. However, the effect of exercise on the acidity of lumen or fecal samples was not measured, which limits us in establishing a direct link between exercise and gut alteration by acidic induction. Thus, more studies are necessary to establish these modifications as possible therapeutic implications for obesity or hypertension treatment through the modulation of gut microbiota.
Electronic supplementary material
=================================
{#Sec18}
######
Additional file 1: **Experimental design.** Obese rats (n = 3), Wistar rats (n = 3) and Hypertensive rats (n = 3) were used to verify the effect of four weeks of moderate exercise training on gut microbiota. Fecal samples were collected before and after exercise training. After DNA extraction, barcoded pyrosequencing of the rRNA genes was used to determine the gut microbiota modifications. (TIFF 909 KB)
######
Additional file 2: **Characterization of the hypertensive phenotype.** Histogram of blood pressure profile from Wistar rats and spontaneously Hypertensive rats measured by the tail-cuff method at the beginning of the experiment. Hypertensive rats showed a significantly higher systolic blood pressure (171.4 ± 7.7 mmHg) when compared to Wistar rats (128 ± 5.9 mmHg), indicating the hypertensive phenotype of this rat group. (TIFF 65 KB)
######
Additional file 3: **Table with the frequency of bacterial communities pre and post exercise training revealed by 16S rRNA pyrosequencing analysis.** (DOCX 15 KB)
######
Additional file 4: **Microbial alpha diversity.** Rarefaction curves for fecal samples, each with at least 23,000 16S rRNA sequences. Each line connects an average number (±SD) of observed 97% OTUs for **(A)** Wistar rats; **(B)** Hypertensive rats and **(C)** Obese rats. The color blue indicates the richness of bacterial communities without training and red indicates with training. (TIFF 150 KB)
######
Additional file 5: **Composition of fecal microbiota population of Wistar, Hypertensive and Obese rats before and after exercise training.** Bacterial distribution evaluated at the main phylum taxonomical level in fecal samples from Wistar, Hypertensive and Obese rats, collected from triplicate rats before and after four weeks of exercise training **(A)**. Results are shown for the independent samples described in Additional file [3](#MOESM3){ref-type="media"}, where the letter "E" represents the samples with exercise training. Boxplots of Firmicutes **(B)** and Proteobacteria **(C)** abundance pre and post-exercise. (TIFF 225 KB)
**Competing interests**
The author(s) declare that they have no competing interests.
**Authors' contributions**
BAP: Conceived the study, and carried out all experimental procedures and drafted the manuscript, APC: carried out the bioinformatics design, performed the statistical analysis and revised the manuscripts, JAA: carried out the experimental procedures and revised the manuscript, CPCG: carried out the experimental procedures and revised the manuscript, GRF: carried out the bioinformatics and statistics design and revised the manuscript, RHK: participated in the study design and revised the manuscript, RWP: participated in the study design and revised the manuscript, OLF: participated in the study design, revised all experimental procedures and drafted and revised the manuscript. All authors read and approved the final manuscript.
The authors acknowledge grant support from CNPq, CAPES, FAPDF and Catholic University of Brasilia -- Brazil.
| {
"pile_set_name": "PubMed Central"
} |
Biopolymers are polymers that are biodegradable. The input materials for the production of these polymers may be either renewable (based on agricultural plant or animal products) or synthetic. Current and future developments in biodegradable polymers and renewable input materials focus relate mainly to the scaling-up of production and improvement of product properties. Larger scale production will increase availability and reduce prices. Prunus is a genus of trees and shrubs, including the plums, cherries, peaches, apricots and almonds. The biopolymer derived from prunus avium has found various applications in pharmaceuticals.
Materials and Methods {#sec1-1}
=====================
To formulate indomethacin emulsion using bio-polymer as emulsifier.
Preparation method {#sec2-1}
------------------
2 ml bio-polymer was taken in a test tube than 10 ml of accetone was added to it with continuous shaking. 10 mg of Indomethacin drug was added in the 5 ml of soyabeen oil, 2.5 ml of water and different concentration of polymer in the morter pastle. To this mixture 10% pectin was added and was vigoursly shacked.
The different formulations containing different proportions of ingredients from which films were formed. Release of drug was carried out using egg shell membrane.
Egg shell membrane was separated by using conc. hydrochloric acid. Then measured quantity of film (i.e.1 cm) was attached to membrane. The egg shell membrane with drug (Indomethacin) was tied around one end of an open tube. Further the membrane with drug was dipped in the phosphate buffer solution (pH-2.3). Then after every 30 min. 5 ml solution was taken and 5 ml buffer solution was added to make the volume, the reading was taken until 3 hrs and than it was observed spectrophotometrically. Same procedure was repeated for different formulated films of Indomethacin emulsion with phosphate buffer (pH-2.3) \[[Figure 1](#F1){ref-type="fig"}\].
{#F1}
Results and Discussions {#sec1-2}
=======================
The stable formulation was successfully prepared using different concentration of biopolymer. [Table 1](#T1){ref-type="table"} shows the three formulations prepared using different concentration of biopolymer as an emulsifier. The results of evaluation parameters and *in vitro* diffusion studies were also suggests the formulation F3 shows the better performance than the other formulations.
######
Formulations
Conclusions {#sec1-3}
===========
From above three formulations, the best formulation was F3. Based on the above study a conclusion was drawn that in the *Prunus avium* bio-material can serve as a promising film forming agent for formulating various drug.\[[@ref1][@ref2]\]
**Source of Support:** Nil
**Conflict of Interest:** None declared.
| {
"pile_set_name": "PubMed Central"
} |
{#sp1 .4}
{#sp2 .5}
| {
"pile_set_name": "PubMed Central"
} |
Introduction {#s1}
============
Epidemiology studies from mainland China spanning 2000--2014 have shown that the prevalence of hyperuricemia is 13.7% in urban Chinese and 12.3% in rural Chinese individuals (Liu et al., [@B11]). Hyperuricemia causes gouty arthritis (Zamudio-Cuevas et al., [@B34]), kidney stones (Mirheydar et al., [@B16]) and chronic renal failure (Tsai et al., [@B25]), which significantly impact quality of life (Scire et al., [@B22]). Hyperuricemia has also been considered to be a risk factor for metabolic syndrome (Cibicková et al., [@B6]; Rubio-Guerra et al., [@B21]), diabetes mellitus (Wang et al., [@B27]), hypertension (Lyngdoh et al., [@B13]), stroke (Wu et al., [@B32]), chronic kidney disease (Ceriello et al., [@B4]) and cardiovascular disease (Qin et al., [@B20]; Amin et al., [@B1]; Moulin et al., [@B18]). Hyperuricemia is therefore a serious public health problem that should be detected early and treated. However, the predictors of hyperuricemia are not well-known.
Waist circumference, an indicator of abdominal obesity, has been reported to be associated with hyperuricemia in Chinese (Wang et al., [@B27]; Zhang et al., [@B35]), African Americans (McAdams-DeMarco et al., [@B15]) and Japanese individuals (Suma et al., [@B24]). The association of waist circumference and hyperuricemia was shown in a recent large study from China (Chen et al., [@B5]). The underlying mechanism linking waist circumference to hyperuricemia may be attributed to the excess free fatty acids released from the visceral fat, causing cellular fat accumulation and consequent lipotoxicity-mediated injury to multiple organs, which in turn would contribute to the metabolic disorder inflicted by uric acid on the kidney and liver (Weinberg, [@B30]; Yamada et al., [@B33]). Circulating free fatty acids have been postulated to emanate from the subcutaneous fat of the upper body (Martin and Jensen, [@B14]). Here, neck circumference was considered a predictive anthropometric measure of upper body fat distribution (Wang et al., [@B29]; Luo et al., [@B12]). Taken together, this finding serves as the rationale for our hypothesis that large neck circumference is associated with hyperuricemia. To the best of our knowledge, there has not been a study that has focused on assessing the association between neck circumference and hyperuricemia. A major aim of this study is therefore to examine whether neck circumference is associated with hyperuricemia and serum uric acid levels in non-hyperuricemia subjects. This study is also aimed at comparing the strength of association of neck circumference and waist circumference with hyperuricemia and their ability to detect hyperuricemia.
Materials and methods {#s2}
=====================
Subjects
--------
The data for this cross-sectional study were derived from a single city center, the Beijing sub-center, which is part of a much larger multi-center research study called the Risk Evaluation of Cancers in Chinese Diabetic Individuals: a lONgitudinal (REACTION) study (Ning, [@B19]). The REACTION project is a China-wide survey aimed at investigating the influence of metabolic diseases on cancer. The Beijing sub-center included 10 administrative regions, divided into 5 urban areas and 5 suburban areas. A total 10,276 participants aged 40 years or over were enrolled between April and October, 2015. After excluding subjects for any pathology or medical intervention that could alter the neck circumference, including neck malformation, prior neck surgery, thyromegaly, thyroid dysfunction and incomplete information for anthropometric parameters or laboratory examination, 8,971 subjects were finally included in our study. All subjects gave written informed consent in accordance with the Declaration of Helsinki. The protocol was approved by the Committee on Human Research of the Chinese People\'s Liberation Army General Hospital.
Data collection
---------------
A standard questionnaire was completed by an in-person interview (between the survey interviewer and each study participant). Information on demographics, history of disease and corresponding medication use, smoking status and drinking status were collected during this process.
Physical examination was performed to obtain information on weight, height, neck circumference, waist circumference and blood pressure. Weight was measured to the nearest 0.5 kg by an electronic weight scale while participants wore light clothing (Beijing Jianmin); height was measured with a vertical height meter to the nearest 0.1 cm. The participants had to remove their shoes, hats, jackets, overcoats, and empty their pockets. Neck circumference was measured with a measuring tape along the inferior margin of the laryngeal prominence and perpendicular to the long axis of the neck, with the subject remaining standing and the head in the horizontal plane position. Waist circumference was measured at the horizontal plane between the inferior costal margin and the iliac crest on the mid-axillary line. All circumferences were recorded to within 0.1 cm. Blood pressure was measured three times at 1-min intervals using a mercury sphygmomanometer after at least a 5-min rest, and the mean value of the three readings was recorded.
The blood samples at fasting status were taken to determine fasting plasma glucose and uric acid levels and lipid profile. Participants were then subjected to a glucose load. Participants without a validated history of diabetes underwent a 75-g oral glucose tolerant test (OGTT), and participants with a history of diabetes underwent a 100-g carbohydrate diet test. Two hours after the OGTT or carbohydrate diet test, blood samples were taken again to determine the glucose tolerance of the subjects.
Definitions
-----------
Body mass index (BMI) was calculated as weight (kilograms) divided by the square of the height (meters). Hypertension was defined as systolic blood pressure (SBP) ≥ 140 mm Hg and (or) diastolic blood pressure (DBP) ≥ 90 mm Hg or being on treatment for hypertension. Diabetes mellitus was diagnosed as fasting plasma glucose (FPG) ≥ 7.0 mmol/L or 2-h plasma glucose (2 hPG) ≥ 11.1 mmol/L, or being on treatment for diabetes. Hyperuricemia was defined as a serum uric acid level \> 420 μmol/L in males and \> 360 μmol/L in females, or subjects with gout or on hyperuricemia treatment. Current alcohol drinking status was classified into 3 categories: never drinking, occasional drinking (drinking less than once a week) and often drinking (drinking once or more a week). Current smoking status was classified into 3 categories: never smoked, occasional smoking (smoking less than one cigarette a day or seven cigarettes a week) and often smoking (smoking one or more cigarettes a day). Cardiovascular and cerebrovascular events included disease history of stroke and (or) myocardial infarction and (or) heart failure.
Statistical analysis
--------------------
All statistical analyses were performed with the SPSS software version 23 for Windows (SPSS Inc., Chicago, IL, USA, [RRID:SCR_002865](https://scicrunch.org/resolver/RRID:SCR_002865)). Graphs were created using R version 3.3.1 (R development core team; available from <http://www.r-project.org/>, [RRID:SCR_001905](https://scicrunch.org/resolver/RRID:SCR_001905)). The data were expressed as the means and standard deviations (SD) for continuous variables or numbers and percentages for categorical variables. The two-sample *t*-test and chi-square test were used to compare the differences in baseline characteristics for continuous and categorical variables, respectively. Correlations between variables and neck circumference or hyperuricemia were assessed by Pearson and Spearman correlation tests. The association of neck circumference and waist circumference with hyperuricemia was assessed by logistic regression analysis, while the association of neck circumference with serum uric acid levels was assessed by multivariable linear regression analysis. In the above regression analysis, the potential confounding factors were adjusted as indicated in the Results section. Receiver operating characteristic (ROC) cure analyses were used to identify hyperuricemia by neck circumference and waist circumference. All analyses were performed separately for gender. *P*-values less than 0.05 were considered statistically significant.
Results {#s3}
=======
Characteristics of the participants
-----------------------------------
In this cross-sectional study, 3,369 males and 5,604 females with a mean age of 60.0 ± 7.8 years were included. The mean serum uric acid was 341.5 ± 77.1 μmol/L (range from 112.0 to 755.2 μmol/L) and 281.5 ± 66.0 μmol/L (range from 106.8 to 664.2 μmol/L) for male and female participants, respectively. The prevalence of hyperuricemia in the male group (14.4%, *n* = 486) was 2.8% higher than that in the female group (11.6%, *n* = 651). Demographic information, indexes of body measurements and metabolic parameters of the subjects are shown in Table [1](#T1){ref-type="table"}.
######
Characteristic of participants categorized by gender and hyperuricemia status.
**Total (*n* = 8,971)** **Male (*****n*** = **3,369)** **Female (*****n*** = **5,062)**
-------------------------------------------------------------------------------------------------- ------------------------- -------------------------------- ---------------------------------- -------------- --------------
Age (y)[^\*^](#TN1){ref-type="table-fn"}[^\#^](#TN2){ref-type="table-fn"} 60.0 ± 7.8 60.9 ± 8.4 61.8 ± 7.9 60.7 ± 7.8 58.8 ± 7.5
Body mass index (kg/m^2^)[^\*^](#TN1){ref-type="table-fn"}[^\#^](#TN2){ref-type="table-fn"} 25.5 ± 3.5 26.7 ± 3.1 25.4 ± 3.2 27.5 ± 3.9 25.1 ± 3.6
Waist circumference (cm)[^\*^](#TN1){ref-type="table-fn"}[^\#^](#TN2){ref-type="table-fn"} 86.1 ± 9.5 93.2 ± 8.3 89.1 ± 8.6 88.7 ± 9.4 83.3 ± 9.2
Neck circumference (cm)[^\*^](#TN1){ref-type="table-fn"}[^\#^](#TN2){ref-type="table-fn"} 35.5 ± 3.3 39.1 ± 2.6 38.0 ± 2.6 35.1 ± 2.5 33.7 ± 2.4
Fasting glucose level (mmol/L)[^\*^](#TN1){ref-type="table-fn"}[^\#^](#TN2){ref-type="table-fn"} 5.9 ± 1.7 5.9 ± 1.5 6.1 ± 1.9 5.9 ± 1.3 5.7 ± 1.7
Glucose tolerance level (mmol/L)[^\#^](#TN2){ref-type="table-fn"} 9.6 ± 3.8 9.4 ± 3.5 9.4 ± 4.2 9.8 ± 3.4 8.7 ± 3.7
Glycated hemoglobin (%) 6.1 ± 1.0 6.1 ± 1.0 6.1 ± 1.1 6.2 ± 1.1 6.1 ± 1.0
Systolic blood pressure (mm Hg)[^\#^](#TN2){ref-type="table-fn"} 130.4 ± 17.0 134.2 ± 17.1 133.2 ± 16.9 131.9 ± 16.4 128.2 ± 16.8
Diastolic blood pressure (mm Hg)[^\*^](#TN1){ref-type="table-fn"} 76.9 ± 9.9 79.8 ± 11.0 78.3 ± 10.2 76.5 ± 9.7 75.8 ± 9.5
Total cholesterol (mmol/L) 4.9 ± 1.7 4.7 ± 1.0 4.6 ± 2.2 5.1 ± 1.0 5.1 ± 1.4
Triglycerides (mmol/L)[^\*^](#TN1){ref-type="table-fn"}[^\#^](#TN2){ref-type="table-fn"} 1.6 ± 1.1 2.1 ± 1.6 1.5 ± 1.0 2.0 ± 1.3 1.6 ± 1.1
High density lipoprotein (mmol/L)[^\*^](#TN1){ref-type="table-fn"} 1.5 ± 1.8 1.2 ± 0.3 1.3 ± 0.3 1.5 ± 2.8 1.6 ± 2.2
Low density lipoprotein cholesterol (mmol/L) 3.1 ± 1.4 3.1 ± 2.4 2.9 ± 1.4 3.2 ± 0.9 3.2 ± 1.3
Alanine transaminase(u/L)[^\*^](#TN1){ref-type="table-fn"}[^\#^](#TN2){ref-type="table-fn"} 20.7 ± 14.3 24.5 ± 15.6 20.8 ± 12.0 23.9 ± 17.4 19.9 ± 14.7
Aspartate transaminase(u/L)[^\*^](#TN1){ref-type="table-fn"} 21.5 ± 28.6 23.5 ± 26.2 20.7 ± 18.4 22.3 ± 10.9 21.6 ± 34.6
γ-glutamyltranspeptidase(u/L)[^\*^](#TN1){ref-type="table-fn"}[^\#^](#TN2){ref-type="table-fn"} 27.7 ± 32.0 46.9 ± 71.7 31.4 ± 33.8 29.9 ± 23.5 23.4 ± 23.5
Creatinine (umol/L)[^\*^](#TN1){ref-type="table-fn"}[^\#^](#TN2){ref-type="table-fn"} 70.1 ± 16.6 90.7 ± 22.6 80.3 ± 13.6 70.0 ± 17.9 62.1 ± 11.4
Serum uric acid (mmol/L)[^\*^](#TN1){ref-type="table-fn"}[^\#^](#TN2){ref-type="table-fn"} 304.0 ± 76.1 473.6 ± 51.9 319.2 ± 55.3 403.8 ± 39.8 265.4 ± 49.9
Diabetes, *n* (%)[^\#^](#TN2){ref-type="table-fn"} 2,554 (28.5) 155 (31.9) 954 (33.1) 242 (37.2) 1,203 (24.3)
Hypertension, *n* (%)[^\*^](#TN1){ref-type="table-fn"}[^\#^](#TN2){ref-type="table-fn"} 4,354 (48.5) 316 (65.0) 1,571 (54.5) 376 (57.8) 2,092 (42.2)
Cardiovascular and cerebrovascular events, *n* (%)[^\#^](#TN2){ref-type="table-fn"} 447 (5.0) 31 (6.4) 197 (6.8) 40 (6.1) 179 (3.6)
**CURRENT SMOKING STATUS**, ***n*** **(%)**
Never 7,335 (81.8) 284 (58.4) 1,590 (55.2) 632 (97.1) 4,829 (97.5)
Occasional 153 (1.7) 16 (3.3) 102 (3.5) 6 (0.9) 29 (0.6)
Often 1,483 (16.5) 186 (38.3) 1,191 (41.3) 13 (2.0) 93 (1.9)
**CURRENT DRINKING STATUS**, ***n*** **(%)**[^\*^](#TN1){ref-type="table-fn"}
Never 6,430 (71.7) 174 (35.8) 1,165 (40.4) 590 (90.6) 4,501 (90.9)
Occasional 1,345 (15.0) 127 (26.1) 813 (28.2) 44 (6.8) 361 (7.3)
Often 1,196 (13.3) 185 (38.1) 905 (31.4) 17 (2.6) 89 (1.8)
*p \< 0.05, comparison of hyperuricemia group to non-hyperuricemia group in male*.
*p \< 0.05, comparison of hyperuricemia group to non-hyperuricemia group in female*.
*Glucose tolerance level, blood glucose level for 2 h after the oral glucose tolerant test or carbohydrate diet test*.
Neck circumference was associated with hyperuricemia
----------------------------------------------------
To precisely assess the association between neck circumference and hyperuricemia, we first determined the confounding factors between them. We identified the confounding factors as those that are both correlated with neck circumference and hyperuricemia. The results demonstrated that age, hypertension, diabetes, cardiovascular and cerebrovascular events, triglycerides (TG) and high-density lipoprotein (HDL) met the criteria for confounding factors (Table [2](#T2){ref-type="table"}). Since current smoking and alcohol drinking status had been considered as confounding factors in previous studies (Zhang et al., [@B35]; Chen et al., [@B5]), we considered them to be additional confounding factors. We adjusted for these factors in the association analysis.
######
Correlation of variables with neck circumference and hyperuricemia stratified by gender.
**Male** ***(n*** = **3,369)** **Female (*****n*** = **5,062)**
------------------------------------------- -------------------------------- ---------------------------------- ------- ---------- ------- ---------- ------- ----------
Age −0.09 \< 0.001 −0.04 0.01 0.05 \< 0.001 0.08 \< 0.001
Fasting glucose level 0.11 \< 0.001 0.01 0.60 0.19 \< 0.001 0.09 \< 0.001
Glucose tolerance level 0.12 \< 0.001 0.03 0.07 0.21 \< 0.001 0.14 \< 0.001
Glycated hemoglobin 0.01 0.52 0.02 0.21 0.03 0.036 0.02 0.19
Systolic blood pressure 0.16 \< 0.001 0.02 0.17 0.23 \< 0.001 0.07 \< 0.001
Diastolic blood pressure 0.15 \< 0.001 0.05 0.01 0.16 \< 0.001 0.03 0.02
Total cholesterol 0.01 0.47 0.06 \< 0.001 −0.03 0.03 0.004 0.74
Triglycerides 0.18 \< 0.001 0.19 \< 0.001 0.18 \< 0.001 0.17 \< 0.001
High-density lipoprotein −0.31 \< 0.001 −0.10 \< 0.001 −0.05 0.001 −0.15 \< 0.001
Low-density lipoprotein cholesterol 0.01 0.52 0.03 0.05 0.02 0.11 0.01 0.47
Alanine transaminase 0.16 \< 0.001 0.10 \< 0.001 0.15 \< 0.001 0.11 \< 0.001
Aspartate transaminase −0.02 0.22 0.08 \< 0.001 0.01 0.49 0.06 \< 0.001
γ-glutamyltranspeptidase 0.08 \< 0.001 0.16 \< 0.001 0.12 \< 0.001 0.17 \< 0.001
Creatinine 0.08 \< 0.001 0.20 \< 0.001 0.07 \< 0.001 0.19 \< 0.001
Diabetes 0.12 \< 0.001 −0.01 0.60 0.18 \< 0.001 0.09 \< 0.001
Hypertension 0.22 \< 0.001 0.08 \< 0.001 0.22 \< 0.001 0.10 \< 0.001
Cardiovascular and cerebrovascular events 0.05 0.002 −0.01 0.71 0.06 \< 0.001 0.04 0.002
Current smoking status 0.02 0.24 0.02 0.18 0.04 0.002 0.01 0.49
Current drinking status 0.04 0.011 0.05 0.007 0.03 0.024 0.004 0.77
*Glucose tolerance level, blood glucose level for 2 h after the oral glucose tolerant test or carbohydrate diet test*.
To analyze the influence of neck circumference on hyperuricemia, we divided the subjects into 4 groups based on the quartiles of neck circumference. After adjusting for all of the confounding factors, the risk for male subjects in the second, third and fourth quartile for neck circumference being inflicted with hyperuricemia was 1.77-, 1.81-, and 2.61-fold the risk of males in the first quartile, respectively (Table [3](#T3){ref-type="table"}). This finding indicated that the corresponding risk for the prevalence of hyperuricemia increased 0.77-, 0.81-, and 1.61-fold. For waist circumference, the corresponding risk was 1.83-, 2.10-, and 3.03-fold the risk of males in the first quartile, respectively (Table [3](#T3){ref-type="table"}). The results of female subjects (Table [4](#T4){ref-type="table"}) were similar to those of males for both neck circumference and waist circumference (Table [3](#T3){ref-type="table"}).
######
Association of body adiposity measures with hyperuricemia in male subjects.
***n*** **Events, *n* (%)** **Model I** **Model II** **Model III**
------------------------- --------- --------------------- ------------- ------------------- --------------- ------------------- --------- -------------------
**NECK CIRCUMFERENCE**
Q1 (≤36.0) 880 68 (7.7) Ref. Ref. Ref.
Q2 (36.1--38.0) 1,001 144 (14.4) \<0.001 2.01 (1.48--2.72) \<0.001 2.00 (1.48--2.72) \<0.001 1.77 (1.29--2.43)
Q3 (38.1--39.0) 898 142 (15.8) \<0.001 2.24 (1.65--3.04) \<0.001 2.20 (1.62--2.98) \<0.001 1.81 (1.31--2.50)
Q4 (≥39.1) 590 132 (22.4) \<0.001 3.44 (2.51--4.71) \<0.001 3.36 (2.45--4.61) \<0.001 2.61 (1.86--3.67)
**WAIST CIRCUMFERENCE**
Q1 (≤83.9) 765 49 (6.4) Ref. Ref. Ref.
Q2 (84.0--89.9) 895 116 (13.0) \<0.001 2.18 (1.54--3.09) \<0.001 2.15 (1.52--3.05) 0.001 1.83 (1.28--2.62)
Q3 (90.0--94.9) 802 121 (15.1) \<0.001 2.60 (1.83--3.68) \<0.001 2.58 (1.82--3.66) \<0.001 2.10 (1.47--3.02)
Q4 (≥95.0) 907 200 (22.1) \<0.001 4.13 (2.97--5.75) \<0.001 4.07 (2.93--5.66) \<0.001 3.03 (2.13--4.31)
*OR, odd ratio; CI, confidence interval; Ref, reference. Model I, unadjusted; Model II, Model I+ adjusted for age, current smoking status and current drinking status. Model III, Model II+ adjusted for Triglycerides, High-density lipoprotein, hypertension, diabetes, cardiovascular and cerebrovascular events*.
######
Association of body adiposity measures with hyperuricemia in female subjects.
***n*** **Events, *n* (%)** **Model I** **Model II** **Model III**
------------------------- --------- --------------------- ------------- ------------------- --------------- ------------------- --------- -------------------
**NECK CIRCUMFERENCE**
Q1 (≤32.0) 1,639 94 (5.7) Ref. Ref. Ref.
Q2 (32.1--34.0) 998 79 (7.9) 0.028 1.42 (1.04--1.93) 0.033 1.40 (1.03--1.91) 0.127 1.28 (0.93--1.75)
Q3 (34.1--35.0) 1,603 202 (12.6) \<0.001 2.37 (1.82--3.06) \<0.001 2.33 (1.81--3.01) \<0.001 1.99 (1.53--2.58)
Q4 (≥35.1) 1,362 276 (20.3) \<0.001 4.18 (3.27--5.35) \<0.001 4.07 (3.18--5.22) \<0.001 3.27 (2.53--4.22)
**WAIST CIRCUMFERENCE**
Q1 (≤76.9) 1,199 57 (4.7) Ref. Ref. Ref.
Q2 (77.0--82.9) 1,317 112 (8.5) \<0.001 1.87 (1.34--2.59) \<0.001 1.84 (1.32--2.56) 0.004 1.63 (1.17--2.27)
Q3 (83.0--89.9) 1,621 196 (12.1) \<0.001 2.76 (2.04--3.75) \<0.001 2.63 (1.94--3.58) \<0.001 2.20 (1.61--3.01)
Q4 (≥90.0) 1,465 286 (19.5) \<0.001 4.87 (3.62--6.54) \<0.001 4.53 (3.37--6.11) \<0.001 3.50 (2.58--4.77)
*OR, odd ratio; CI, confidence interval; Ref., reference. Model I, unadjusted; Model II, Model I+ adjusted for age, current smoking status and current drinking status Model III, Model II+ adjusted for Triglycerides, High-density lipoprotein, hypertension, diabetes, cardiovascular and cerebrovascular events*.
To compare the diagnostic value of neck circumference and waist circumference in identifying the presence of hyperuricemia, we performed a ROC analysis. The strengths of neck circumference and waist circumference in identifying hyperuricemia were very close in both genders, with the area under the curve (AUC) analysis as 0.61 vs. 0.63 in male subjects, respectively, and 0.66 vs. 0.66 in female subjects, respectively (Figure [1](#F1){ref-type="fig"}).
{#F1}
Neck circumference was associated with serum uric acid levels in non-hyperuricemia subjects
-------------------------------------------------------------------------------------------
To examine whether neck circumference was also associated with the process of uric acid metabolism, we assessed the association between neck circumference and serum uric acid levels in the non-hyperuricemia subjects, which included 2,883 males and 4,951 females; the corresponding mean uric acid was 319.2 ± 55.3 μmol/L (range from 112.0 to 419.9 μmol/L) and 265.4 ± 49.9 μmol/L (range from 106.8 to 359.9 μmol/L), respectively. The characteristics of the non-hyperuricemia subjects are shown in Table [1](#T1){ref-type="table"}. After adjusting for all confounding factors as indicated above, the neck circumference was positively associated with serum uric acid levels (*P* \< 0.001). The strength of association was 2.58 (1.76--3.39) and 4.27 (3.70--4.84) for males and females, respectively (Figure [2](#F2){ref-type="fig"}). These results indicate that the neck circumference increased 1 cm each, and the plasma uric acid concentration increased 2.58 and 4.27 μmol/L in male and female subjects, respectively.
{#F2}
Discussion {#s4}
==========
In this study, we demonstrated that neck circumference was significantly associated with hyperuricemia and that neck circumference was also positively associated with serum uric acid levels in non-hyperuricemia subjects. Additionally, these associations were independent of metabolic status and cardio-metabolic disease. To the best of our knowledge, this study is the first to demonstrate that neck circumference is correlated with hyperuricemia.
Both epidemiologic and clinical evidence indicated a close interrelation between hyperuricemia and obesity (McAdams-DeMarco et al., [@B15]; Shi et al., [@B23]; Wang et al., [@B27]; Suma et al., [@B24]). Waist circumference, an indicator of abdominal obesity, had already been determined to be associated with hyperuricemia (McAdams-DeMarco et al., [@B15]; Wang et al., [@B27]; Suma et al., [@B24]; Zhang et al., [@B35]). Large neck circumference, a marker of upper body adiposity (Wang et al., [@B29]; Luo et al., [@B12]), is similar to large waist circumference in its association with metabolic syndrome and cardiovascular disease (Hingorjo et al., [@B10]; Assyov et al., [@B2]; Luo et al., [@B12]). However, there has not been a study to assess the association between hyperuricemia and neck circumference. In this study, we demonstrated that neck circumference was similar to waist circumference in terms of the strength of association (OR, 3.03 for waist circumference vs. 2.61 for neck circumference in males and 3.50 vs. 3.27 for females) with hyperuricemia and their similar ability to predict hyperuricemia (AUCs of 0.63 for waist circumference vs. 0.61 for neck circumference in males, and corresponding AUCs of 0.66 vs. 0.66, respectively, for females). In this study, we also found that neck circumference was associated with the plasma levels of uric acid in non-hyperuricemia subjects. This finding indicated that neck circumference was not only a marker of hyperuricemia but was also associated with the process of uric acid metabolism.
We found that the strength of association between hyperuricemia/serum uric acid levels and neck circumference in females was stronger than that in males. This finding was similar to the result from Luo\'s study (Luo et al., [@B12]). The underlying mechanisms might be attributed to the different lipid metabolism and fat distribution between the two genders, which might be due in part to differences in sex hormones (Gu et al., [@B8]; Luo et al., [@B12]; Vashishta et al., [@B26]).
Visceral fat has been associated with hyperuricemia in a manner independent of the total fat area, total subcutaneous fat area, and abdominal subcutaneous fat area (Yamada et al., [@B33]). This finding, taken with the neck circumference as being positively associated with visceral obesity (Wang et al., [@B29]), led us to infer that visceral fat, but not total or subcutaneous fat, may mediate the linkage between neck circumference and hyperuricemia. This postulate is supported by our findings, which showed that after adjusting for triglycerides, a non-specific indicator of fat distribution, neck circumference was still significantly associated with hyperuricemia \[OR, 2.61 (1.86--3.67)\], although the strength of association was slightly decreased. Ectopic distribution of fat in or around the visceral organs, such as the liver and kidney, may cause fat infiltration and dysfunction of these organs. However, it is not clear how fat accumulation in these organs contribute to increased uric acid production in the liver or reduced excretion by the kidney, which are mechanisms that need to be explored. This finding emphasizes a limitation of this study, which is that we did not examine the visceral fat, especially with respect to hepatic and renal fat accumulation, which, if concurrently present with the high neck circumference, would confer upon the latter a possible mechanistic explanation for the association with hyperuricemia.
Hyperuricemia was reported as an independent and important risk factor not only for gout and hyperlipidemia (Hikita et al., [@B9]) but also for hypertension, diabetes (Miyagami et al., [@B17]) and cardiovascular morbidity and mortality rates (Grassi et al., [@B7]). There is therefore urgency in elucidating the more clinically apparent markers for hyperuricemia, in part as predictors of these diseases. Waist circumference measurements may not be suitable for a number of clinical situations, such as for patients on bed rest or who are pregnant or who are in a number of disease states affecting waist circumference, such as ascites and abdominal tumors. Furthermore, methods for measuring waist circumference have not been clinically standardized (Wang et al., [@B28]; Willis et al., [@B31]; Bernritter et al., [@B3]). Thus, the utilization of waist circumference has been restricted, making the uniform and accurate measurement of neck circumference a better and more convenient clinical alternative.
Conclusion {#s5}
==========
Our findings showed neck circumference to be positively and independently associated with serum uric acid levels and hyperuricemia in both genders. Neck circumference could be used as a predictive indicator to trigger the screening for and treatment of hyperuricemia.
Nomenclature {#s6}
============
Resource identification initiative
----------------------------------
The SPSS software version 23 for Windows (SPSS Inc., Chicago, IL, USA, [RRID:SCR_002865](https://scicrunch.org/resolver/RRID:SCR_002865)).
R version 3.3.1 (R development core team; available from <http://www.r-project.org/>, [RRID:SCR_001905](https://scicrunch.org/resolver/RRID:SCR_001905)).
Author contributions {#s7}
====================
YH and JD conceived the study. AW, YM, LD collected population data. SW and HG performed laboratory assays. XY and JC performed statistical analyses. JJ wrote and revised the manuscript. JJ and JC interpreted the data and contributed equally to this work. All authors critically read and approved the manuscript.
Conflict of interest statement
------------------------------
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
We appreciate the efforts of all participants who contributed to sample measurements and data collections.
**Funding.** This study was funded by National Natural Science Foundation of China (31672375) and Key Projects in the National Science & Technology Pillar Program (No. 2015BAI09B01) to YH.
[^1]: Edited by: Elisabeth Lambert, Swinburne University of Technology, Australia
[^2]: Reviewed by: Ayo Priscille Doumatey, National Institutes of Health (NIH), United States; Marli Maria Knorst, Federal University of Rio Grande do Sul (UFRGS), Brazil
[^3]: This article was submitted to Integrative Physiology, a section of the journal Frontiers in Physiology
[^4]: †These authors have contributed equally to this work.
| {
"pile_set_name": "PubMed Central"
} |
Introduction {#Sec1}
============
The exponential increase in data traffic requires high-capacity optical links. A fast, compact, energy-efficient, broadband optical modulator is a vital part of such a system. Modulators integrated with silicon (Si) or silicon nitride (SiN) platforms are especially promising, as they leverage complementary-metal-oxide-semiconductor (CMOS) fabrication techniques. This enables high-yield, low-cost, and scalable photonics, and a route towards co-integration with electronics^[@CR1]^. SiN-based integrated platforms offer some added advantages compared to silicon-on-insulator, such as a broader transparency range^[@CR2]^, a lower propagation loss^[@CR3],[@CR4]^, significantly lower nonlinear losses^[@CR2],[@CR5]^, and a much smaller thermo-optic coefficient^[@CR2]^. Therefore, phase modulators on SiN in particular would open new doors in other fields as well, such as nonlinear and quantum optics^[@CR5]--[@CR7]^, microwave photonics^[@CR8]^, optical phased arrays for LIDAR or free-space communications^[@CR9]^, and more.
State-of-the-art silicon modulators rely on phase modulation through free carrier plasma dispersion in p--n^[@CR10]^, p--i--n^[@CR11]^, and MOS^[@CR12]^ junctions. Despite being relatively fast and efficient, these devices suffer from spurious amplitude modulation and high insertion losses. Alternative approaches are based on heterogeneous integration with materials such as III--V semiconductors^[@CR13],[@CR14]^, graphene^[@CR15],[@CR16]^, electro-optic organic layers^[@CR17]^, germanium^[@CR18]^, or epitaxial BaTiO~3~ (BTO)^[@CR19]--[@CR21]^.
Most of these solutions are not viable using SiN. Due to its insulating nature, plasma dispersion effects and many approaches based on co-integration with III--V semiconductors, graphene, and organics, which rely on the conductivity of doped silicon waveguides, cannot be used. The inherent nature of deposited SiN further excludes solutions using epitaxial integration. Finally, SiN is centrosymmetric, hampering Pockels-based modulation in the waveguide core itself, in contrast to aluminum nitride^[@CR22]^, or lithium niobate^[@CR23]^. Nonetheless, modulators on SiN exist. Using double-layer graphene, Phare et al.^[@CR24]^ achieved high-speed electro-absorption modulation, and using piezoelectric lead zirconate titanate (PZT) thin films, phase modulators based on stress-optic effects^[@CR25]^, and geometric deformation^[@CR26]^, have been demonstrated, albeit with sub-MHz electrical bandwidth.
In this work, we use a novel approach for co-integration of thin-film PZT on SiN^[@CR27]^. An intermediate low-loss lanthanide-based layer is used as a seed for the PZT deposition, as opposed to the highly absorbing Pt-based seed layers used conventionally^[@CR25],[@CR26]^, enabling direct deposition of the layer on top of SiN waveguides fabricated using front-end-of-line CMOS processes.
We demonstrate efficient high-speed phase modulators on a SiN platform, with bias-free operation, modulation bandwidths exceeding 33 GHz in both the O-band and C-band, and data rates up to 40 Gbps. We measure propagation losses down to 1 dB cm^−1^ and half-wave voltage-length products *V*~*π*~*L* down to 3.2 Vcm for the PZT-on-SiN waveguides. Moreover, based on simulations we argue that the *V*~*π*~*L* can be considerably reduced by optimizing the waveguide cross-section, without significantly increasing the propagation loss. Hence, the platform provides an excellent trade-off between optical losses and modulation efficiency. According to simulations, the product *V*~*π*~*Lα* can be as low as 2 VdB in optimized structures. Pure phase modulation also enables complex encoding schemes (such as QPSK), which are not easily achievable with absorption modulation. These results, especially in terms of the achieved modulation bandwidths, strongly improve upon what is currently possible in SiN^[@CR25],[@CR26]^. In terms of *V*~*π*~*Lα*, this platform can furthermore improve upon carrier dispersion modulators in silicon-on-insulator, which suffer from inherent carrier-induced losses absent in Pockels modulators^[@CR28]^.
Results {#Sec2}
=======
Device design and fabrication {#Sec3}
-----------------------------
The waveguides are patterned using 193 nm deep ultraviolet lithography in a 330-nm-thick layer of low pressure chemical vapor deposited SiN on a 3.3-μm-thick buried oxide layer, in a CMOS pilot line. Subsequently, plasma-enhanced chemical vapor deposited SiO~2~ (thickness ≈1 μm) is deposited over the devices and planarized, either using a combination of dry and wet etching, or by chemical--mechanical polishing (CMP). This step is performed so that the top surface of the SiN waveguide and the surrounding oxide are coplanar. The PZT films are deposited by chemical solution deposition (CSD), using a lanthanide-based intermediate layer (see Methods and ref.^[@CR27]^). Finally, Ti/Au electrical contacts are patterned in the vicinity of the waveguides using photolithgraphy, thermal evaporation, and lift-off. For the samples planarized through CMP, propagation losses of 1 dB cm^−1^ are measured on PZT-covered waveguides without metallic contacts (see Supplementary Note [2](#MOESM1){ref-type="media"}).
Figure [1a, b](#Fig1){ref-type="fig"} show the top view and waveguide cross-section of a C-band ring modulator, and for images of the other fabricated modulators (O-band ring, C-band Mach--Zehnder), see Supplementary Note [1](#MOESM1){ref-type="media"}. Figure [1c](#Fig1){ref-type="fig"} shows a schematic of the cross-section. An electric field is applied through in-plane electrodes, changing the refractive index in the PZT and hence the effective index of the waveguide mode. The PZT thin films exhibit a higher refractive index (*n* ≈ 2.3) than SiN (*n* ≈ 2), so a significant portion of the optical mode is confined in the PZT. A grating coupler is used for incoupling and outcoupling, into the fundamental quasi-transverse electric (quasi-TE) optical mode. The combined loss of a grating coupler and the transition between a bare and PZT-covered waveguide section is ≈12 dB at the optimum, with a 3 dB bandwidth of ≈90 nm. However, this is currently not optimized and can still be improved by design.Fig. 1Design and static response of a C-band ring modulator. **a** Top view of a PZT-on-SiN ring modulator. **b** Cross-section of a PZT-covered SiN waveguide. The image contrast was enhanced for clarity. **c** Schematic of the PZT-covered SiN waveguide. The fundamental TE optical mode is plotted in red. The quiver plot shows the applied electric field distribution between the electrodes. PZT thickness, waveguide width, and gap between the electrodes are, respectively, 150 nm, 1200 nm, and 4 μm. **d** Normalized transmission spectrum of a C-band ring modulator. **e** Transmission spectra for different DC voltages. **f** Resonance wavelength shift versus voltage applied across the PZT, including a linear fit
DC characterization and poling stability {#Sec4}
----------------------------------------
Figure [1d](#Fig1){ref-type="fig"} shows the transmission spectrum of a C-band (1530--1565 nm) ring modulator. The ring has a loaded *Q* factor of 2230 and a free spectral range Δ*λ*~FSR~ ≈1.7 nm. The ring radius, the length of the phase shifter *L*, and the electrode spacing are, respectively, 100, 524, and 4.4 μm. The relatively low *Q* factor is caused by sub-optimal alignment of the electrodes.
After deposition, the PZT crystallites have one crystal plane parallel to the substrate, but no preferential orientation in the chip's plane. To obtain a significant electro-optic response for the quasi-TE optical mode, a poling step is performed by applying 60--80 V (≈150 kV cm^−1^) for 1 h at room temperature, followed by several hours of stabilization time.
The transmission spectrum is measured for different direct current (DC) voltages applied across the PZT layer (Fig. [1e](#Fig1){ref-type="fig"}). The voltage-induced index change shifts the resonance. In Fig. [1f](#Fig1){ref-type="fig"}, the resonance wavelength shift is plotted as a function of voltage, and the slope gives the tuning efficiency Δ*λ*/Δ*V* ≈ −13.4 pm V^−1^. From this, we estimate the half-wave voltage-length product to be *V*~*π*~*L* = \|*Lλ*~FSR~Δ*V*/(2Δ*λ*)\| ≈ 3.3 Vcm. Through simulation of the optical mode and DC electric field, the effective electro-optic coefficient *r*~eff~ of the PZT layer is estimated to be 61 pm V^−1^ (see Methods). Measurements on other modulator structures yield consistent values for *r*~eff~ (67 and 70 pm V^−1^ for, respectively, the C-band Mach--Zehnder and O-band ring), and the smallest *V*~*π*~*L* value (≈3.2 V cm) was measured on an O-band ring (Supplementary Note [1](#MOESM1){ref-type="media"}).
The PZT was poled prior to the measurements, after which no bias voltage was used. To demonstrate longer term stability of the poling, the DC tuning efficiency was periodically measured (sweeping the voltage over \[−2, +2\] V) on a C-band ring over a total time of almost 3 days. In Fig. [2](#Fig2){ref-type="fig"}, the absolute value of the resulting tuning efficiency Δ*λ*/Δ*V* is plotted as a function of time, decaying towards a stable value of about 13.5 pm V^−1^ over the course of several hours. The poling stabilized and there have been no indications of decay over much longer periods of time, hence modulation is possible without a constant bias, as opposed to similar materials like BTO^[@CR19]--[@CR21]^.Fig. 2Poling stability of the electro-optic film. Tuning efficiency (C-band ring) as a function of time after poling. The axis on the right shows the estimated corresponding *V*~*π*~*L*
High-speed characterization {#Sec5}
---------------------------
For many applications, high-speed operation is essential. In Fig. [3a](#Fig3){ref-type="fig"} the setup used for high-speed characterization is shown. In Fig. [3b](#Fig3){ref-type="fig"}, the $\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document}$$\left| {S_{21}} \right|$$\end{document}$ measurement for different modulators is plotted. The measured 3 dB bandwidths of both rings are around 33 GHz, and the Mach--Zehnder has a bandwidth of 27 GHz. The bandwidths are not limited by the intrinsic material response of PZT, but by device design and/or characterization equipment, as the dominating contributions to the Pockels effect are expected to have a bandwidth which is almost two orders of magnitude larger^[@CR29],[@CR30]^. We furthermore demonstrate that our platform can be used for high-speed data transmission. In Fig. [3c](#Fig3){ref-type="fig"}, eye diagrams are plotted for different bitrates, a non-return-to-zero (NRZ) binary sequence (4.2 V peak-to-peak) is used. The eye remains open up until about 40 Gbps, limited by the arbitrary waveform generator (AWG) (25 GHz bandwidth), rather than by the modulator itself. The bit error rates were estimated from the measured eye diagrams^[@CR31]^, and are below the hard-decision forward error coding limit of 3.8 × 10^−3 [@CR32],[@CR33]^ for bitrates up to 40 Gbps (see Supplementary Note [3](#MOESM1){ref-type="media"}). At 10 Gbps, an extinction ratio of 3.1 dB is measured (see Supplementary Note [4](#MOESM1){ref-type="media"}).Fig. 3High-speed measurements. **a** Sketch of the setup used for small signal measurements (solid path in the switches) and for the eye diagram measurements (dashed path). VNA: vector network analyzer, AWG: arbitrary waveform generator, OTF: optical tunable filter. **b** Electro-optic small signal ($\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document}$$\left| {S_{21}} \right|$$\end{document}$ parameter) measurement of several modulators. **c** Eye diagrams of a C-band ring modulator, measured with a non-return-to-zero scheme (2^9^ − 1 pseudorandom binary sequence) and a peak-to-peak drive voltage of 4.2 V
Device optimization {#Sec6}
-------------------
The presented devices were not fully optimized in terms of electro-optic modulation parameters. Primarily the PZT thickness could be increased. Sub-optimal thicknesses were used to reduce bend losses and coupling losses into PZT-covered waveguide sections. These limitations can be alleviated by device design. In Fig. [4](#Fig4){ref-type="fig"}, simulation results of the most important figures of merit are plotted as a function of the PZT layer thickness and of the electrode spacing. Waveguide height, width, and the wavelength are, respectively, 300 nm, 1.2 μm, and 1550 nm. The waveguide propagation loss *α* (Fig. [4a](#Fig4){ref-type="fig"}) is calculated as the sum of a contribution caused by the electrodes, and a constant intrinsic propagation loss of 1 dB cm^−1^, a realistic value if the samples are planarized using CMP (see Supplementary Note [2](#MOESM1){ref-type="media"}). The half-wave voltage-length product *V*~*π*~*L* (see Methods) and the product *V*~*π*~*Lα* are shown in Fig. [4b, c](#Fig4){ref-type="fig"}, respectively. *V*~*π*~*L* represents a trade-off between drive voltage and device length, *V*~*π*~*Lα* also takes into account loss, and is arguably more important for many applications^[@CR26]^. The loss increases with decreasing electrode spacing, but also with increasing PZT thickness, since the mode expands laterally. Due to the increasing overlap between the optical mode and the PZT, *V*~*π*~*L* decreases with increasing thickness. *V*~*π*~*L* also increases with increasing electrode spacing. An optimization of the waveguide width is given in the Supplementary Note [5](#MOESM1){ref-type="media"}. From Fig. [4b](#Fig4){ref-type="fig"} it is clear that *V*~*π*~*L* can go well below 2 Vcm. The interplay between these different dependencies can be seen in the plot of *V*~*π*~*Lα* (Fig. [4b](#Fig4){ref-type="fig"}), which has an optimum for which *V*~*π*~*Lα* ≈ 2 VdB.Fig. 4Numerical optimization of a PZT-on-SiN phase modulator. Simulation of the waveguide loss *α* (**a**), the half-wave voltage-length product *V*~*π*~*L* (**b**), and their product *V*~*π*~*Lα* (**c**) of a PZT-covered SiN waveguide modulator of the type shown in Fig. [1c](#Fig1){ref-type="fig"}, for a wavelength of 1550 nm. Waveguide height, width, and intermediate layer thickness are, respectively, 300 nm, 1.2 μm, and 20 nm. The intrinsic waveguide loss (in the absence of electrodes) was taken to be 1 dB cm^−1^, and the effective electro-optic Pockels coefficient was 67 pm V^−1^. The circles show the approximate parameters used in this work, and the diamonds show the optimal point with respect to *V*~*π*~*Lα*
Discussion {#Sec7}
==========
To conclude, we have demonstrated a novel platform for efficient, optically broadband, high-speed, nanophotonic electro-optic modulators. Using a relatively simple chemical solution deposition procedure, we incorporated a thin film of strongly electro-optic PZT onto a SiN-based photonic chip. We demonstrated stable poling of the electro-optic material, and efficient and high-speed modulation, in the absence of a bias voltage. O-band and C-band operation was shown; however, we expect the platform to be operational into the visible wavelength range (\>450 nm)^[@CR2],[@CR34],[@CR35]^. From simulations it is clear that the devices characterized in this paper do not yet represent the limitations of the platform and *V*~*π*~*Lα* ≈ 2 V dB is achievable. Moreover, our approach is unique in its versatility, as the PZT film can be deposited on any sufficiently flat surface, enabling the incorporation of the electro-optic films onto other guided-wave platforms.
Methods {#Sec8}
=======
PZT deposition and patterning {#Sec9}
-----------------------------
While the details of the lanthanide-assisted deposition procedure have been published elsewhere^[@CR27]^, a short summary is given here. Intermediate seed layers based on lanthanides are deposited prior to the PZT deposition. The intermediate layer acts as a barrier layer to prevent the inter-diffusion of elements and as a seed layer providing the lattice match to grow highly oriented thin films. A critical thickness of the intermediate layer needs to be maintained (\>5 nm) to avoid diffusion and secondary phase formations. However, on samples with considerable surface topology, thicker intermediate layers are necessary to provide good step coverage and to avoid any issues associated with the conformity in spin-coating. On our samples planarized through etching, step heights between oxide and SiN waveguides varied. We typically used an intermediate layer of thickness ≈24 nm to avoid issues. Both the intermediate layer and the PZT thin films are deposited by repeating the spin-coating and annealing procedure, which allows easy control of the film thickness. The PZT layer is deposited and annealed at 620 °C for 15 min in a tube furnace under an oxygen ambient. This CSD method, also called sol--gel, provides a cheap and flexible alternative to achieve high-quality stoichiometric PZT thin films regardless of the substrate material. A reactive ion etching procedure based on SF~6~ chemistry is used to pattern the PZT layer. The PZT film was removed selectively over the grating couplers used for the optical measurements.
High-speed measurements {#Sec10}
-----------------------
The small-signal response measurements were performed using an Agilent PNA-X N5247A network analyzer and a high-speed photodiode (Discovery Semiconductors DSC10H Optical Receiver). For the eye diagram measurements, an arbitrary waveform generator (Keysight AWG M8195A) and radio frequency (RF) amplifier (SHF S807) are used to apply a pseudorandom NRZ binary sequence, the modulator output is measured with a Keysight 86100D oscilloscope with 50 GHz bandwidth and Discovery Semiconductors DSC-R409 PIN-TIA Optical Receiver.
Calculation of the electro-optic parameters {#Sec11}
-------------------------------------------
Using COMSOL Multiphysics^®^, several parameters can be calculated that strongly influence the performance of the modulators. To obtain efficient phase modulation, it is essential to maximize the overlap between the optical mode and the RF electrical signal, quantified by the electro-optic overlap integral^[@CR36]^,$$\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document}$${\Gamma} = \frac{g}{V}\frac{{\varepsilon _0cn_{\mathrm{PZT}}{\int\!\!\!\int}_{\mathrm{PZT}} {\kern 1pt} E_x^{`\mathrm{e}}\left| {E_x^{\mathrm{op}}} \right|^2{\mathrm{d}}x{\mathrm{d}}y}}{{{\int\!\!\!\int} {\kern 1pt} {\mathrm{Re}}\left( {{\bf{E}}^{{\mathrm{op}}} \times {\bf{H}}^{{\mathrm{op}}^ \ast }} \right) \cdot \widehat {\bf{e}}_z{\mathrm{d}}x{\mathrm{d}}y}},$$\end{document}$$where *g* is the spacing between the electrodes, *V* the applied voltage, *ε*~0~ the vacuum permittivity, *c* the speed of light in vacuum, and *n*~PZT~ the refractive index of PZT. $\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document}$$E_x^{\mathrm{e}}$$\end{document}$ is the in-plane (*x*-)component of the RF electric field, and $\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document}$$E_x^{{\mathrm{op}}}$$\end{document}$ represents the in-plane transversal component of the optical field. When used as a phase shifter, an important figure of merit is the half-wave voltage-length product *V*~*π*~*L*. This product relates to the electro-optic coefficient *r*~eff~ of the PZT films and to *Γ*^[@CR36]^,$$\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document}$$V_\pi L = \frac{{\lambda g}}{{n_{{\mathrm{PZT}}}^3{\it{\Gamma }}r_{{\mathrm{eff}}}}},$$\end{document}$$where *λ* is the free-space wavelength. Another important parameter is the propagation loss of the optical mode, consisting of an intrinsic contribution (scattering, material loss in the PZT, intermediate layer, nitride and oxide) and a contribution caused by the vicinity of the electrical contacts. The former can be estimated based on cut-back measurements on unmetalized waveguides (see Supplementary Note [2](#MOESM1){ref-type="media"}), and the latter can be numerically calculated.
Data availability {#Sec121}
-----------------
All data that support the findings of this study are available from the corresponding authors upon reasonable request.
Electronic supplementary material
=================================
{#Sec14}
Supplementary Information Peer Review File
**Publisher\'s note:** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
These authors contributed equally: Koen Alexander, John P. George.
**Change history**
9/6/2018
This Article was originally published without the accompanying Peer Review File. This file is now available in the HTML version of the Article; the PDF was correct from the time of publication.
Electronic supplementary material
=================================
**Supplementary Information** accompanies this paper at 10.1038/s41467-018-05846-6.
We thank Stéphane Clemmen for his overseeing role in the SiN chip fabrication, Philippe F. Smet for help with processing, Joris Van Kerrebrouck for help with data analysis, Liesbet Van Landschoot for operating the electron microscope, and Yoko Ohara for help with the figures. K.A. is funded by FWO Flanders. This work was funded by the European Commission through grant agreement no. 732894 (FET-Proactive HOT).
K.A. and J.P.G. designed the devices. K.A. performed the chip planarization. J.P.G. performed the PZT deposition, patterning, and metalization. K.A. performed the static optical measurements. J.V. and K.A. carried out the high-speed measurements. K.A. analyzed the data and performed device optimization simulations. K.N., B.K., D.V.T., and J.B. provided general advice and feedback. K.A. and J.P.G. wrote the manuscript, and all authors reviewed the manuscript and agree to its content.
Competing interests {#FPar1}
===================
The authors declare no competing interests.
| {
"pile_set_name": "PubMed Central"
} |
Introduction {#S0001}
============
Stroke is the leading cause of death and permanent disability in adults worldwide, especially in low-and middle-income countries, with important clinical and socio-economic implications.[@CIT0001] Among them, ischemic stroke accounts for about 87% of stroke cases.[@CIT0002] In the acute phase of the disease, neurons of the ischemic lesion die quickly, while other neuron groups in the ischemic penumbra are vulnerable to secondary injury.[@CIT0003] Unfortunately, despite advances in technology and pharmacology, there are still few drugs or treatments for stroke. To date, tissue plasminogen activator (t-PA) is the only FDA-approved treatment for stroke. However, due to narrow treatment window (\<4.5h) and safety concerns, less than 5% of patients benefit from the drug.[@CIT0004] Therefore, the development of safe and effective treatment methods with long treatment windows and effective drugs is imminent.
Danshen is a very important ingredient in traditional Chinese medicine and it is extracted from the dried root or rhizome of salvia miltiorrhiza. Its main ingredient is tanshinone IIA (TSA), which has traditionally been used to treat cardiovascular and cerebrovascular diseases, including ischemic stroke.[@CIT0005],[@CIT0006] It has been reported that TSA can penetrate the blood-brain barrier when brain content is limited to 30% of plasma concentration, possibly due to its role in reducing infarct volume and maintaining neuronal function.[@CIT0007] In addition, TSA has been shown to protect against focal cerebral ischemia/reperfusion (I/R) injury in animal models and has been used as a potential therapeutic agent in the treatment of heart, liver and cancer.[@CIT0008] However, there have been no reports on the neuroprotective effect after focal cerebral ischemia of TSA by promoting axonal regeneration.
Studies have shown that the central nervous system (CNS) in adults can repair itself after cerebral ischemia, but to a limited extent. The key factor is that it is difficult to regenerate axons due to the increased expression of axonal growth inhibitors after ischemic injury.[@CIT0009] Neurite outgrowth inhibitor-A (Nogo-A) is a well-known myelin-associated axon growth-inhibitory protein that has been shown to inhibit the migration and spread of nerve cells and play an important role in preventing axon regeneration and reconnection after stroke.[@CIT0010],[@CIT0011] It binds to the receptor Nogo receptor 1 (NgR1), triggering the downstream RhoA/ROCKII/MLC signaling pathway, inducing collapse of the growth cone and obstructing nerve repair and regeneration.[@CIT0012] A previous study has shown anti-Nogo-A therapy improves neurological deficits and enhances neuronal plasticity, suggesting that Nogo-A-targeted therapies are expected to be regenerative after stroke.[@CIT0013] However, it is unclear whether the neuroprotective effect of TSA is related to the inhibition of Nogo-A signaling pathway. Herein, the purpose of this study was to evaluate the neuroprotective effects of TSA and to explore whether TSA promotes axonal regeneration and protects damaged nerves by inhibiting Nogo-A/NgR1/RhoA/ROCKII/MLC signal activation.
Methods and Materials {#S0002}
=====================
Animals and Groups {#S0002-S2001}
------------------
Ninety-six Sprague-Dawley (SD) rats, weighing 250--280 g, provided by Liaoning Changsheng biotechnology co., Ltd., were included in the study \[SCXK (Liaoning) 2015-0001\].
All rats were maintained in the house with a background of temperature 21 ± 1°C and humidity (55±10%). The SD rats were randomly divided into four groups: Sham group (Sham, n=24), vehicle group (vehicle, n=24), Low TSA group (TSA-L, n=24) and High TSA group (TSA-H, n=24). The sham group received the same surgery and was injected with 10mL/kg PBS including 1% DMSO; The MCAO model was established in the remaining three groups, the difference was that the vehicle group was injected intraperitoneally with 10 mL/kg PBS including 1% DMSO, while the TSA-L group was 10 mg/kg TSA, and the TSA-H group was 20 mg/kg TSA. TSA was administered to each rat by intraperitoneal (i.p.) injection 15 mins after surgery, and every 24 hrs for 7 consecutive days. The structural formula of TSA is shown in [Figure 1](#F0001){ref-type="fig"}. This study was approved by the ethics committee of Hebei University of Traditional Chinese Medicine. All procedures are in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. The schedule of the experimental procedures is presented in [Figure 2](#F0002){ref-type="fig"}.Figure 1Chemical structures of tanshinone IIA (TSA).Figure 2The schedule of the experimental procedures.
Focal Cerebral Ischemia Animal Model {#S0002-S2002}
------------------------------------
Focal brain ischemia was induced by the permanent middle cerebral artery occlusion (pMCAO) method as previously described.[@CIT0014] Briefly, the rats were anesthetized by intraperitoneal injection of 10% chloral hydrate (350 mg/kg). The rat's right common carotid artery (CCA), external carotid artery (ECA), and internal carotid artery (ICA) were exposed, and then a 4-0 nylon suture was inserted into the ICA to occlude the middle cerebral artery (MCA). The temperature of the rats was adjusted by a thermostatic heating plate after surgery, and the rats were placed in the supine position until awakening and fed separately after resuscitation. The Zea-Longa scores test was used to evaluate whether the pMCAO was successful.[@CIT0015] More specifically, neurological deficits were scored with the following 5-point scale: 0, no neurological deficits; 1, failure to fully extend the left forelimb; 2, rotation to the left; 3, falling to the left during walking; and 4, no spontaneous walking. The rats of neurologic deficit score of 1--3 were considered to have a successful pMCAO and were included for subsequent evaluation.
Assessment of Neurological Deficits {#S0002-S2003}
-----------------------------------
Researchers who were blind to the experimental groups measured neurological scores of rats on the 3rd and 7th day after surgery to assess neurological deficits. The neurological impairment score system is based on the mNSS scoring system (0--18 points) developed by Chen et al.[@CIT0016] Score of 1 refers to an unfinished task or lack of response. Therefore, the higher the score, the more severe the injury.
Measurement of Cerebral Infarction Volume {#S0002-S2004}
-----------------------------------------
To measure the infarct volume, rats were deeply anesthetized and sacrificed on the 3rd and 7th day after surgery, the brain was dissected and cut into 2 mm sections. Subsequently, the sections were incubated in a 37°C 2% 2, 3, 5-triphenyltetrazolium chloride (TTC) solution for 20 mins and fixed with 4% paraformaldehyde. The staining images were recorded using a digital camera (Canon Oxus 950IS) and then quantified using an Image J (ver 1.51k, NIH).
Hematoxylin-Eosin Staining {#S0002-S2005}
--------------------------
On day 7 after modeling, the rats were decapitated immediately after deep anesthesia and heart perfusion with normal saline and 4% paraformaldehyde, and the brain was fixed in 4% paraformaldehyde for 24 hrs. Subsequently, the brain tissue was embedded in paraffin and cut into 5 μm sections with a microtome. Dewaxing and rehydration were performed with a xylene and ethanol aqueous solution, followed by HE staining by a conventional method. Sections were stained with hematoxylin for 5 mins and eosin for 3 mins.
Nissl Staining {#S0002-S2006}
--------------
After 7 days of modeling, we conducted Nissl staining. Briefly, after dewaxing and rehydration, the brain sections were stained with 0.5% cresol violet at 37°C for 10 mins. Then, we use 0.25% glacial acetic acid ethanol to distinguish. The cell morphology of the cerebral cortex was observed under an optical microscope (Olympus BX53). Four high-power fields were randomly selected and Nissl positive cells were calculated using Image J, which was expressed as the number of complete neurons per 1 mm^2^ length in the ischemic penumbra.
Immunofluorescence Staining {#S0002-S2007}
---------------------------
Immunofluorescence staining was performed 7 days after surgery. After the fixation with 4% paraformaldehyde, the brain sections were permeabilized with xylene, and blocked with goat serum for 20 mins. Then, the brain sections were immunolabeled with antibodies: NF200 (1:100; Proteintech) and GAP-43 (1:100; Proteintech). After incubation, the sections were washed with PBS followed by Cy3-conjugated secondary antibody. After the nuclei were counterstained with DAPI, the samples were observed under a fluorescence microscope (Olympus, Japan).
Western Blotting Analysis {#S0002-S2008}
-------------------------
On day 7 after surgery, the cortical tissue of the ischemic penumbra of the experimental rats was homogenized, and the protein concentration was measured using a BCA protein assay kit. Proteins were separated by 10% sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) and transferred to a PVDF membrane. After blocking with TBST, 5% nonfat dry milk for 1 h, the membrane was incubated overnight at 4°C with antibodies against NF200 (1:1000; Proteintech), GAP-43 (1:5000; Proteintech), Nogo-A (1:500; Proteintech), NgR1 (1:1000; Bioss), RhoA (1:1000; Proteintech), ROCKII (1:500; Proteintech), MLC (1:5000; Bioss), p-MLC (1:5000; Proteintech) and β-actin (1:1000; Wanleibio) followed by incubation with a HRP-conjugated secondary antibody for 45 min at room temperature. The blot was observed with ECL reagent, and the optical density of the target band was measured using a gel image processing system (Gel-Pro-Analyzer software).
Real-Time PCR Analysis {#S0002-S2009}
----------------------
On the 7th day after the surgery, total RNA was extracted from cortical tissue of ischemic penumbra of the experimental rats using TRIpure reagent (BioTeke, BEIJING) according to the standard instructions. After it was reverse transcribed to cDNA using a Sensiscript RT kit (ThermoFisher Scientific Inc., USA). Subsequently, RT-qPCR was performed. The thermocycling conditions were as follows: 94°C for 10 s, 60°C for 20 s; 40 cycles of 72°C for 30 s and extension at 72°C for 2.5 mins. The primer sequences are shown in [Table 1](#T0001){ref-type="table"}.Table 1Primers Used for Real-Time PCRGenesForward (5ʹ--3ʹ)Reserve (3ʹ--5ʹ)NF200GGAGGCACTGAAAAGCACCAGCCATCTCCCATTTGGTGTTGAP-43AGGGAGATGGCTCTGCTACCACATCGGCTTGTTTAGGCNogo-AAGTCTTGGGAAGGATAGTGAGGTGCTTTCGGTTGCNgR1GAAAGAACCGCACCCGTAGGGCCCAAGCACTGTCCAARhoATCGGAATGATGAGCACACAAGCTTCACAAGATGAGGCACROCK IITCATAAGGCATCACAGAAAACCACCCACGGACTMLCCTAAGGGACACGTTTGCTGAAGGCGTTGAGAATGGβ-actinGGAGATTACTGCCCTGGCTCCTAGCGGCCGGACTCATCGTACTCCTGCTT[^1]
Statistical Analysis {#S0002-S2010}
--------------------
SPSS 20.0 (SPSS Inc, USA) was applied to analyze all data. The survival rates were analyzed by the Kaplan--Meier method, with comparisons performed using the Log-rank test. The neurological deficit scores were analyzed with a nonparametric Kruskal--Wallis test with pairwise comparisons and post hoc comparisons . Differences among multiple groups were statistically analysed using one-way ANOVA and post hoc comparisons (Bonferroni test). Values of *P* \< 0.05 were considered statistically significant.
Results {#S0003}
=======
TSA Improves the Survival Rate of Rats After MCAO {#S0003-S2001}
-------------------------------------------------
As shown in [Figure 3B](#F0003){ref-type="fig"}, in the sham group, there was no death of rats during the test. Compared with sham group, the survival rate of rats in the vehicle group was 61.54%, lower than that in the sham group (*P*˂0.001). Further comparison found that there was no significant difference in survival between the vehicle group and the TSA-treated group (TSA-L group: 75.00%; TSA-H group: 82.76%), but the survival rate in the TSA-H group did increase by 21.22%.Figure 3The effects of TSA on the survival rate, neurological deficit scores and infarct volume in pMCAO rats. (**A**) TTC staining of rats in different groups 3rd and 7th days. (**B**) The survival rate of rats in each group was expressed as the percentage. (**C**) Neurological deficit scores of rats in different groups 3rd and 7th days. (**D**) Infarct volume of rats in different groups 3rd and 7th days; ^\#\#\#^P\< 0.001 vs the sham group; \*P\< 0.05, \*\*P \< 0.01, \*\*\*P \< 0.001 vs the vehicle group.**Abbreviation:** pMCAO, permanent middle cerebral artery occlusion.
TSA Reduces Neurological Deficit Score and Cerebral Infarction Volume {#S0003-S2002}
---------------------------------------------------------------------
To test whether TSA has a neuroprotective effect, we measured the neurological score and cerebral infarct volume of rats after MCAO. In [Figure 3C](#F0003){ref-type="fig"}, the sham group had a score of 0, showing no dysfunction, while the other groups all had neurological deficits of varying degrees. Compared with the vehicle group, the TSA-H group decreased the scores on both the 3rd and 7th days (*P*˂0.001), while the TSA-L group reduced the scores only on the 7th day (*P*˂0.05). 2,3,5-triphenyltetrazolium chloride (TTC) staining showed that each model had different degrees of ischemic changes. No infarction was observed in the sham group, but extensive lesions developed in the vehicle group ([Figure 3A](#F0003){ref-type="fig"}). More specifically, cerebral infarct volumes were both significantly reduced after treatment with 20 mg/kg TSA on the 3rd and 7th days, while only decreased on 7th day after treatment with 10 mg/kg TSA ([Figure 3D](#F0003){ref-type="fig"}).
TSA Alleviates Neuronal Injury After MCAO {#S0003-S2003}
-----------------------------------------
The HE staining in [Figure 4A](#F0004){ref-type="fig"} shows the morphological changes of cerebral cortex. From the figure, we can clearly observe that the neurons in the sham group are neatly arranged without necrotic neurons. In the vehicle group, neurons were arranged disorderly and degenerate cells increased, the nucleus was deeply stained, and the cytoplasm shrank. Administration of TSA significantly improved neurons\' morphology and reduced neurons' damage, especially group TSA-H. Further, we evaluated changes in the number of neurons with and without TSA. Nissl staining results indicated that compared with sham group, the neurons in the vehicle group showed lighter staining, looser cell arrangement and fewer neurons (*P*˂0.001). When TSA was added, it significantly reduced the number of degenerated neurons and increased the number of intact neurons ([Figure 4B](#F0004){ref-type="fig"} and [C](#F0004){ref-type="fig"}; *P*˂0.001).Figure 4TSA alleviated neuronal damage. (**A**) The results of HE staining in different groups. (**B**) The results of Nissl staining in different groups. (**C**) The changes in the number of intact neurons in different groups. Scale bar= 100μm. ^\#\#\#^P\< 0.001 vs the sham group; \*\*\*P\< 0.001 vs the vehicle group.
TSA Increases Axon Length and Promotes NF200 Expression {#S0003-S2004}
-------------------------------------------------------
To detect the effect of TSA on axonal regeneration, we first observed changes in the expression of NF200 and the length of the axons labeled with NF200 by immunofluorescence staining. In [Figure 5A](#F0005){ref-type="fig"} and [B](#F0005){ref-type="fig"}, the axons in the sham group were intact, while the axon structure was broken and the axon length was reduced after MCAO (*P*˂0.001). Compared with the vehicle group, the addition of TSA remarkably improved axon structure and increased axon length, especially group TSA-H (*P*˂0.001). In addition, immunofluorescence staining showed a strong expression of NF200 in the sham group, while the fluorescence intensity in the carrier group decreased significantly in the vehicle group. TSA treatment enhanced NF200 immunofluorescence when compared with the vehicle group ([Figure 5C](#F0005){ref-type="fig"}).Figure 5TSA increased axonal length and promoted NF200 expression. (**A**) Immunofluorescence staining for NF200, scale bar= 50μm. (**B**) The changes in the length of axons in different groups. (**C**) The mean optical density of NF200 in different groups. (**D**) The protein expression of NF200 in different groups. (**E**) The mRNA expression of NF200 in different groups; ^\#\#\#^P \< 0.001 vs the sham group; \*P\< 0.05, \*\*P \< 0.01, \*\*\*P \< 0.001 vs the vehicle group.**Abbreviation:** NF200, neurofilament protein 200.
To further verify the above results, we conducted Western blotting and qRT-PCR. The results showed that after MCAO, the protein expression of NF200 was significantly decreased when in comparison with the sham group (*P*˂0.001). However, after treatment with 10 mg/kg or 20 mg/kg TSA, NF200 expression was significantly increased compared with the vehicle group ([Figure 5D](#F0005){ref-type="fig"}; *P*˂0.05 TSA-L group, *P*˂0.001 in TSA-H group). Similarly, the transcription results of NF200 genes were consistent with the protein results, that is, MCAO decreased the mRNA expression of NF200, whereas TSA promoted the mRNA expression of NF200 ([Figure 5E](#F0005){ref-type="fig"}).
TSA Up-Regulates GAP-43 Expression {#S0003-S2005}
----------------------------------
As presented in [Figure 6A](#F0006){ref-type="fig"}, at 7th day after MCAO, GAP-43 immunofluorescence was evidently enhanced in comparison to the sham group (*P*˂0.001). Interestingly, after TSA treatment, the protein expression of GAP-43 was up-regulated more significantly in comparison to vehicle group ([Figure 6B](#F0006){ref-type="fig"}; *P*˂0.001). Next, we performed Western blotting and qRT-PCR to detect the protein expression and mRNA expression of GAP-43, respectively. The Western blotting results showed that MCAO significantly increased GAP-43 protein expression of vehicle group, while the addition of TSA promoted the induction of MCAO to GAP-43 ([Figure 6C](#F0006){ref-type="fig"}; *P*˂0.05 in TSA-L group, *P*˂0.001 in TSA-H group). qRT-PCR results also confirmed that TSA promoted the mRNA expression levels of GAP-43, which was consistent with previous protein results ([Figure 6D](#F0006){ref-type="fig"}).Figure 6TSA up-regulated GAP-43 expression. (**A**) Immunofluorescence staining for GAP-43, scale bar= 50μm. (**B**) The mean optical density of GAP-43 in different groups. (**C**) The protein expression of GAP-43 in different groups. (**D**) The mRNA expression of GAP-43 in different groups; ^\#^P \< 0.05, ^\#\#\#^P \< 0.001 vs the sham group; \*P \< 0.05, \*\*\*P \< 0.001 vs the vehicle group.**Abbreviation:** GAP-43, growth associated protein-43.
TSA Inhibits Expression of Nogo-A/NgR1/RhoA/ROCKII/MLC Signaling Pathways {#S0003-S2006}
-------------------------------------------------------------------------
To investigate the mechanism of axonal regeneration with TSA treatment, the expression of the vital target genes of the Nogo-A signaling pathway, Nogo-A, NgR1, RhoA, ROCKII and MLC, was investigated. As presented in [Figure 7A](#F0007){ref-type="fig"}, compared with sham group, the protein expression of Nogo-A, NgR1, RhoA, ROCKII and p-MLC were increased significantly after MCAO in vehicle group. However, after treatment with TSA, the expression of the above proteins was markedly down-regulated, especially in group TSA-H. Accordingly, the optical of Nogo-A, NgR1, RhoA, ROCKII and p-MLC increased significantly when compared with the sham group, respectively (*P*\<0.001). However, they decreased significantly when TSA was added as compared with vehicle group (all *P*\<0.001). Consistently, the qRT-PCR results also showed that MCAO induced the mRNA expression of Nogo-A, NgR1, RhoA, ROCKII and p-MLC, while TSA inhibited the induction of these proteins by MCAO ([Figure 7B](#F0007){ref-type="fig"}). In addition, there were no significant changes in MLC at the transcription and translation levels in each group.Figure 7TSA inhibited the expression of the Nogo-A/NgR1/RhoA/ROCKII/MLC signaling pathway. (**A**) The protein levels of Nogo-A, NgR1, RhoA, ROCKII, MLC and p-MLC. (**B**) The mRNA levels of Nogo-A, NgR1, RhoA, ROCKII and MLC; ^\#\#\#^P\< 0.001 vs the sham group; \*\*P \< 0.01, \*\*\*P \< 0.001 vs the vehicle group.**Abbreviations:** Nogo-A, Neurite outgrowth inhibitor-A; NgR1, Nogo receptor 1; RhoA, Recombinant Ras Homolog Gene Family, Member A; ROCK II, Rho-associated protein kinase 2; MLC, myosin light chain.
Discussion {#S0004}
==========
Ischemic stroke is one of the diseases with the highest mortality and disability rate in the world, and it often leaves severe neurological deficit symptoms. Stroke treatment usually focuses on promoting nerve recovery to improve nerve function. Previous studies have shown that the mechanisms of nerve recovery are complex, including angiogenesis, neurogenesis and synaptic plasticity.[@CIT0017] Among these mechanisms, axon regeneration is an important link and basis for nerve recovery. Therefore, promoting axon regeneration is a key way to promote nerve recovery and improve nerve function recovery. One form of axon regeneration is the regeneration of axon buds in the injured area. However, due to severe self-injury, axons grow slowly, and the growth process is easily blocked by glial scars. Another form is the regeneration of axons around the injury. The regeneration of axon buds of neurons in the area around the infarction and even the contralateral area after cerebral infarction is enhanced.[@CIT0018] The axial protuberant buds in the area surrounding the infarct grow and extend to the injured area and establish connections to facilitate nerve recovery in the injured area. However, due to the presence of a large number of inhibitory factors on axonal regeneration in the microenvironment, neural self-recovery is very limited, which cannot effectively promote the improvement of neurological dysfunction symptoms. Therefore, exogenous drug therapy is used to promote axonal regeneration, thereby promoting neurological recovery and improving neurological function recovery.
As a crude herbal medication isolated from the dried root or rhizome of salvia miltiorrhiza bunge, Danshen has been used in Asian countries for multiple therapeutic remedies including myocardial infarction, stroke and atherosclerosis.[@CIT0007] As one of the most pharmacologically active components isolated from Danshen, TSA can easily cross the blood-brain barrier due to its lipophilic properties.[@CIT0019] Numerous experimental studies have shown that TSA can prevent damage from neurodegenerative diseases, including ischemic stroke.[@CIT0007],[@CIT0020] Consistent with previous studies, our study showed that TSA exerted neuroprotective effect, as indicated by the ameliorated neurological outcome, reduced neurological deficit score, cerebral infarct volume, and neuronal damage.
NF200 is a neuron-specific structural protein, which is mainly expressed on the axons of neurons and is important for the growth, maintenance and regeneration of axons. A previous study reported that NF200 reflects the morphological changes of neuron axons and the state of axon growth or repair.[@CIT0021] Therefore, NF-200 has been used as a marker for assessing axon regeneration in central neurons.[@CIT0022] In addition, GAP-43 is a membrane-associated phosphorylated protein, and the expression of GAP-43 is very low in the brain tissue of healthy adult mammals. However, after cerebral ischemic injury, GAP-43 becomes concentrated mainly at the growth cone terminus and the presynaptic terminal, and it is expressed at high levels during axonal growth and synapse formation.[@CIT0023] Studies have shown that axons exhibit regeneration and remodeling after cerebral ischemia, as reflected by the re-expression of GAP-43.[@CIT0024],[@CIT0025]
In the current study, our results indicated that TSA treatment significantly improved axon morphology in MCAO rats, increased the length of NF200-labeled axons, and up-regulated the expression of NF200 and GAP-43, suggesting that TSA promoted axon regeneration after ischemic stroke.
Membrane protein Nogo-A is a key factor that inhibits axon regeneration and nerve repair. In the adult central nervous system, Nogo-A acts as a negative regulator of neuron growth, preventing abnormal fiber growth under physiological conditions to stabilize the central nervous system wiring. At the same time, axonal regeneration was inhibited under pathological conditions.[@CIT0026] Studies have shown that ischemic stimulation can activate Nogo-A in central neurons, which binds to NgR1 and further activates the downstream protein RhoA to phosphorylate Rho kinase (ROCK). The activation of Rho protein leads to the formation of fibroblast actin stress fibers, which eventually leads to axon retraction and collapse of the growth cone.[@CIT0007] ROCKII is mainly distributed in the cortex and hippocampus of the central nervous system. Phosphorylated ROCKII promotes the phosphorylation of the target protein myosin light chain (MLC), leading to rearrangement of the cytoskeleton, collapse of the growth cone, and ultimately inhibiting sprouting and growth of the axons.[@CIT0027] Existing reports indicated that Rho/ROCK/MLC pathway-specific inhibitors such as Fausudil and Y27632 promote axonal regeneration and functional recovery after stroke by blocking this pathway.[@CIT0028],[@CIT0029] In the present study, we found that the expression of Nogo-A, NgR1, RhoA, ROCKII, and p-MLC increased in rats after the establishment of a cerebral ischemia model, which is consistent with previous studies. After treatment TSA, the expressions of Nogo-A, NgR1, RhoA, ROCKII and p-MLC were significantly reduced, indicating that the neuroprotective effect of TSA and promotion of axon regeneration may be related to inhibiting Nogo-A/NgR1/RhoA/ROCKII/MLC pathway related ([Figure 8](#F0008){ref-type="fig"}).Figure 8Schematic diagram of tanshinone IIA promoting axonal regeneration and neuroprotective effect on focal cerebral ischemia in rats by inhibiting the Nogo-A/NgR1/RhoA/ROCKII/MLC signaling pathway.**Abbreviations:** Nogo-A, neurite outgrowth inhibitor-A; NgR1, Nogo receptor 1; RhoA, Recombinant Ras Homolog Gene Family, Member A; ROCKII, Rho-associated protein kinase 2; MLC, myosin light chain.
There are some limitations in our manuscript. First, we did not conduct further studies on pathways, such as the use of pathway-specific inhibitors for interventions or gene knockout techniques. Second, the time window for TSA intervention is single, and no studies have been conducted on the delayed treatment of TSA (for example, starting administration at different time periods after modeling). In future studies, we will further explore the effects of TSA on Nogo-A/NgR1/RhoA/ROCKII/MLC signaling pathway of MCAO rats by using pathway-specific inhibitors or gene knockout techniques. At the same time, for the delayed treatment of TSA, we will explore the effective time window of TSA by administering drugs at different time periods after modeling, so as to make the experimental study more suitable for clinical application.
Conclusion {#S0005}
==========
TSA plays an active role in promoting neurological recovery and axonal regeneration after ischemic stroke. The realization of this effect is related to the inhibition of Nogo-A/NgR1/RhoA/ROCKII/MLC signaling pathway, which may be one of the neuroprotective mechanisms induced by TSA.
Data Sharing Statement {#S0006}
======================
The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.
Disclosure {#S0007}
==========
The authors declare no conflicts of interest in this work.
[^1]: **Abbreviations:** NF200, neurofilament protein 200; GAP-43, growth associated protein-43; Nogo-A, neurite outgrowth inhibitor-A; NgR1, Nogo receptor 1; RhoA, Recombinant Ras Homolog Gene Family, Member A; ROCK II, Rho-associated protein kinase 2; MLC, myosin light chain; β-actin, β-non-muscle.
| {
"pile_set_name": "PubMed Central"
} |
Although laparoscopic treatment for appendicitis has been documented as a feasible and safe alternative to conventional open appendectomy with minimal complications and short hospital stay,[@b1-asm-2-100],[@b2-asm-2-100] there remains skepticism in the surgical community with respect to its increased technical difficulty and hospital cost.[@b3-asm-2-100]--[@b5-asm-2-100] Laparoscopic appendectomy is most frequently performed as "in" appendectomy with three trocars.[@b6-asm-2-100] In an attempt to overcome the criticism, a new technique," laparoscope-assisted"[@b7-asm-2-100],[@b8-asm-2-100] and using two trocars has gained acceptance in children.[@b9-asm-2-100] After a careful *Medline* internet search we found no studies of the new two-trocar technique in adults from the Kingdom of Saudi Arabia; only one study in adults in the English literature was found.[@b10-asm-2-100] We report our experience with 129 cases using this technique at Ohud Hospital Al Medinah Al Munawarah, Saudi Arabia. The aim of this study was to assess the efficacy and safety of this modified technique.
Patients and Methods
====================
During the period starting from July 2002 till December 2003,129 cases of acute appendicitis were done at Ohud Hospital Medinah Al-Munawarah, Saudi Arabia using the 2-trocar technique. Only patients older than 12 years of age were included in this study. Inclusion criteria included---pain in the right iliac fossa with muscle guarding, vomiting, fever, leukocytosis and localized ileus.[@b9-asm-2-100] All the patients had a plain abdominal radiograph and routine blood and urine analysis. Only female patients had a sonographic examination to exclude pregnancy and/or adnexal pathology. Patients diagnosed with diffuse peritonitis were excluded from the study. Informed and written consent for laparoscopic and/or open appendectomy was taken. The procedures were done by consultants and surgeons-in-training who attended an advanced course in laparoscopy and had good experience in the open technique. All the patients were asked to void before going to the operating room. Prophylactic antibiotics in the form of intravenous cefoxitin 1 g (Tabuk Pharmaceutical, Saudi Arabia) plus 500 mg metronidazole were given at the time of induction of anesthesia and were continued for the next 24 hours (3 doses). The same antibiotic regimen was given for 5 days in those cases with localized pus collection. All cases were done using general anesthesia with the patient in the supine position, the operating surgeon standing on the left side and the assistant on the right side and the TV monitor at the foot side. The pneumoperitoneum was established using a Veress needle. A 10-mm trocar was introduced in the supraumblical region and a 10-mm 00 telescope (Stryker 4000 Santa Clara, CA, USA) was introduced to view the intra-abdominal cavity using different table positions. The appendix was identified easily in most of the cases. In case of difficulty a slight lateral tilt to the left with the patient in the Trendelenburg position would solve this problem. The diagnosis of acute appendicitis was confirmed. The tip of the telescope was used to sweep away the omentum and/or intestinal loops obscuring the scene, but sometimes this did not work. In that situation another 10-mm trocar would be introduced in the right iliac fossa under direct vision at the shortest vertical distance between the caecum and the anterior abdominal wall. This access would be used as a working port for further dissection and in case of conversion the same wound would be enlarged to perform open appendectomy. The pull/push and strip and tease technique was used to skeletonize the appendix. Blood vessels up to 1 mm in diameter were controlled by carrying dissection along the mesoappendix border where the vessels are free up to the base. The proposed site of the appendix ligation was marked by briefly touching the appendix by diathermy,[@b11-asm-2-100] with the jaws of the endo-Babcock soaked in India ink and/or methylene blue. The tip of the appendix was grasped gently with the endo-Babcock and pulled out by screwing movements into the sleeve of the right iliac fossa trocar, lifting the wall of the caecum towards the anterior abdominal wall. Once the tenting of the caecal base was seen, the pneumoperitoneum was released to shorten the distance between the caecal wall and anterior abdominal wall, which laxed the opening in the abdominal wall of the 10-mm trocar at the right iliac fossa. With gentle screwing movements and a steady pull on the endo-Babcock, holding the tip of the appendix and simultaneously counter pushing on the skin surrounding the trocar entrance site by gauze, the appendix along with the trocar was pulled out from the peritoneal cavity through this incision. This whole controlled maneuver of traction and push helped to deliver the whole length of the appendix along with the trocar outside the peritoneal cavity. While the endo-Babcock held the appendix in position outside the abdominal cavity, the mesoappendix outside on the abdominal wall was held by an ordinary Babcock allowing the mesoappendix to fan out. The ordinary Babcock replaced the endo-Babcock and the trocar was removed. A conventional appendectomy was then performed extraabdominally. The cut end was painted with Alphadine (povidone iodine 10%W/V Riyadh Pharma, Riyadh, Saudi Arabia) and the pneumoperitoneum was reestablished. During this procedure care was taken to avoid any parietal contact during the delivery of the appendix from the peritoneal cavity and when the stump was reinserted in the peritoneal cavity, by surrounding the outside incision with Alphadine-soaked gauze. The cut end of the appendix with the base was returned to the peritoneal cavity as the intraabdominal pressure built up. A final check for hemostasis, and abdominal lavage, if necessary, was carried out afterwards. The linea alba was closed using 2-0 interrupted absorbable sutures and the skin was closed with 4-0 subcuticular absorbable sutures. Before skin closure wound lavage with normal saline was performed. In cases with peritonitis, a peritoneal lavage using 3 to 5 liters of normal saline was performed and a vacuum drain was left in the pelvis and brought through a side port. All the appendix specimens were sent for histopathological examination. In some cases it was difficult to mobilize the appendix because of a short or dense mesentery or adhesions. In those cases another 5-mm port in the left iliac fossa was introduced medial to the anterior iliac spine just above the bikini line and away from the urinary bladder, avoiding the inferior epigastric vessels. This additional port further facilitated the use of instruments for mobilization or skeletonization of the appendix and clipping the appendicular artery with an endodissector, hook scissors, and a knot pusher or endoclip. In cases where the appendix could not be delivered due to friability, the base of the appendix was ligated first by the locally made 2-0 vicryl endoloop and then the base was cut with scissors inside the peritoneal cavity. To avoid contact with the parietal peritoneum the thumb end of a sterile glove and/or a plastic cover of a nasogastric tube was used to deliver the appendix in pus-laden gangrenous cases. The operation time was measured at the time from skin incision to closure. The results were analyzed using Microsoft Access.
Results
=======
In the 18-month period, 129 appendectomies were performed. Fifty-nine patients (45.7%) were females and 70 patients (54.2%) were males. Patient ages ranged from 13 to 65 years ([Table 1](#t1-asm-2-100){ref-type="table"}). All the specimens were sent for histopathological examination. The diagnosis of acute appendicitis was made in 105 (81.4%) cases, gangrenous appendicitis in 6 (4.6%), and perforated appendicitis in 11 (8.5%). Six specimens (4.6%) were reported as normal. There was one case of appendiceal cancer detected during this study and 10 (7.7%) had concomitant adnexal pathology. The two-trocar technique was successful in 101 (78.3%) cases, while 14 (10.8%) needed a 5-mm third trocar in the left iliac fossa, beneath the bikini line to complete the laparoscope-assisted operation. In five cases (3.8%) the appendix was difficult to deliver intact through the right iliac fossa port because of a friable appendix, gangrenous and/or perforated, and autoamputated with localized pus collection. In these cases a laparoscopic "in" operation was performed. In all these cases of peritonitis a Redivac drain through the third 5-mm trocar incision in the left iliac fossa was left in place until it stopped draining. Six cases (4.6%) had appendicular mass, and they were all converted to the open tecnique extending the right iliac fossa port wound. The right iliac fossa trocar incision had to be enlarged in 3 (2.4%) cases to permit easy delivery of a turgid appendix with an edematous and inflamed mesoappendix even after using the third port site to ease out the adhesions. The mean operation time was 35 minutes (range, 30--90 minutes) and the mean hospital stay was 2.8 days (range 2--7 days).
A few postoperative complications were encountered ([Table 2](#t2-asm-2-100){ref-type="table"}). These included one case of a port abscess diagnosed on the sixth postoperative day in a patient with perforated appendicitis, which was drained. Six patients (4.6%) had minor wound infections, including one patient with a caecal perforation complicating electrocautery of the appendicular artery, which was detected on the second postoperative day. This patient had open drainage and primary closure of her caecal perforation. She had an uneventful recovery. One case of appendiceal carcinoma with peritoneal metastases was detected. The operation was completed by converting to the open technique. No case of port site hernia was seen throughout the follow-up period (range14 to 30 months).
Discussion
==========
Conventional open appendectomy is still the most common method of treatment for acute appendicitis and has stood the test of time even when performed by the surgeon-in-training at odd hours when senior surgeons are not available. Two hundred forty-three appendectomies were done in our institution during the study period. Although the technique of laparoscopic appendectomy using three trocars is gaining acceptance and popularity,[@b2-asm-2-100],[@b12-asm-2-100]--[@b13-asm-2-100] there are still reservations on its technical difficulties and cost.[@b5-asm-2-100],[@b14-asm-2-100]--[@b16-asm-2-100] Inspired by these arguments, we thought of using the 2-trocar laparoscope-assisted appendectomy in adults. The procedure is simple, cost effective and has all the advantages of minimally invasive and open surgery.[@b9-asm-2-100],[@b10-asm-2-100] Using this technique, we managed to spare the cost of two endoloops and one trocar saving up to US \$220 per case compared with the three-trocar technique. In case of conversion to the open procedure the wound for the trocar in the right iliac fossa can be used for a conventional ("McBurney") incision. Like other laparoscopic procedures we noticed that the operative time is related to the learning curve. Identification of the ceco-appendicular junction is very important so as not to leave a big stump of the appendix. This is done by accurate marking of the base, using methylene blue, India ink and/or electric cautery.[@b11-asm-2-100] Diagnostic laparoscopy has been advocated to clarify the diagnosis in equivocal cases and has been shown to reduce the rate of unnecessary appendectomy. Our figure of 4.6% was far less than the reported series (25.4%) using the open technique.[@b17-asm-2-100] It is most effective in female patients of childbearing age since a gynecological cause of pain is easily identified, as visualization of the pelvis is superior. [@b18-asm-2-100] Despite the restricted selection criteria in females, 10% of our patients still had adnexal problems detected intraoperatively. Although having access to the abdomen is an opportunity and advantage in dealing with adnexal problems in the same sitting, we suggest that there is a further scope for improvement in the preoperative screening of females in childbearing age. Six patients (4.6%) had minor wound infection, which is slightly more than the 2.3% reported in the laparoscopic technique and less than the 6.1% reported in the open technique.[@b19-asm-2-100] We had one major complication related to direct injury of the caecum caused by excessive use of electrocautery in the vicinity of the ceco-appendicular junction. Electrocautery should be discouraged. There is no obvious mention of caecal injury in laparoscopic appendectomy apart from the 0.2% rate of bowel injury reported in a study from Switzerland of 2179 cases.[@b20-asm-2-100] The laparoscope-assisted appendectomy using the two-trocar technique combines the advantages of the minimally invasive and open methods besides being economical and technically easy. For these reasons, the two-trocar technique is suggested as an alternative to other appendectomy procedures in adults. One of the restricting factors in doing this technique is that it needs to be done by fairly senior doctors in surgical training and this may exclude junior residents who are doing the majority of these cases by the open technique. For this reason, we encourage basic training in laparoscopy to be an integral part of the surgical training programme.
We wish to thank Dr. I Valioulis for his kind help.
######
Characteristics, diagnosis and outcome in 129 patients who underwent the two-trocar technique for appendectomy.
Number of patients (%)
---------------------------------------------------- ------------------------
Male 59 (45.7)
Female 70 (54.2)
Age range 13 to 65 years
Diagnosis
Acute appendicitis 105 (81.4)
Gangrenous appendicitis 6 (4.6)
Perforated appendicitis 11 (8.5)
Normal 6 (4.6)
Outcome
Technique successful 101 (78.3)
Third trocar needed 14 (10.8)
Appendix difficult to deliver intact 5 (3.8)
Appendicular mass converted to the open technique 6 (4.6)
######
Postoperative complications in 129 patients who underwent appendectomy.
-------------------------------------------------- ----------
Port abscess 1 (0.7%)
Minor wound infections 6 (4.6%)
Appendiceal carcinoma with peritoneal metastases 1 (0.7%)
-------------------------------------------------- ----------
| {
"pile_set_name": "PubMed Central"
} |
Background {#s1}
==========
As the Covid-19 pandemic evolves, acute care surgeons, intensivists and other surgical specialists increasingly may be asked to perform a tracheostomy in patients with known or suspected coronavirus-19 infection. Practitioners must be prepared for this inevitability while taking measures to perform the procedure safely for patients in altered or suboptimal conditions and protecting themselves and other healthcare personnel from undue risk of exposure and infection. This document provides a brief overview for those considering performing tracheostomy in known or suspected Covid-19. The information provided here is not intended to supersede clinical judgment. As the current pandemic evolves, some or all of the data and recommendations may not be applicable to future conditions.
Current severity of disease in the Covid-19 population {#s2}
======================================================
As of 26 March 2020, the Centers for Disease Control and Prevention (CDC) reported 68 440 total confirmed plus presumptive cases of Covid-19 in the USA, with 994 deaths.[@R1] These numbers are expected to change daily as more data are collected and more testing for the virus is performed. As of 16 March 2020, the last report of outcome data by the CDC,[@R2] 508 patients were known to have been hospitalized in the USA, with 121 (23.8%) admitted to an intensive care unit (ICU). ICU admissions were highest among adults 75--84 years old and lowest among adults 20--44 years old. Among the 44 cases with a known outcome, 80% of deaths have been in patients 65 years of age or older and 20% among adults 20--64 years of age. The largest percentage of severe outcomes are in those 85 years of age or older.
The early experience from Wuhan, China, on 138 hospitalized patients reported that 36 (26.1%) were admitted to ICU for complications, 22 (61.1%) of whom were diagnosed with ARDS and 17 (47.2%) of whom were placed on mechanical ventilation.[@R3] Discharge data were incomplete, with some patients still hospitalized at the time of the report. Six (4.3%) of the admitted patients died. Of the 47 (34.1%) who were discharged, the median hospital stay duration was 10 days.
Utility and benefits of tracheostomy in the general critical care population {#s3}
============================================================================
Tracheostomy has many known benefits in the critically ill and injured, but its utility in the recovery of patients with Covid-19 is unknown. In previous studies, early tracheostomy has been associated with reductions in the duration of mechanical ventilation[@R4] and short-term mortality and in specialized populations such as those with traumatic brain injury, reduced ICU and hospital days and risk of nosocomial pneumonia.[@R5]
In the trauma population, percutaneous bedside tracheostomy is common and safe. In patients with respiratory failure due to coronavirus, transport out of the ICU for open tracheostomy (OT) may be limited or restricted due to risk of viral exposure to staff and to physiological instability, making bedside percutaneous tracheostomy (PT) necessary. However, surgeons must also be prepared to perform OT and urgent cricothyroidotomy under safe conditions should the need arise.
Risks to providers during tracheostomy {#s4}
======================================
Tracheostomy poses a significant risk of viral transmission because it is an aerosol-generating procedure.[@R6] This risk pertains not only to the operating surgeon but to all team members in the room during the procedure. A systematic review estimated the odds of transmission from tracheostomy (OR 4.2) as second only to intubation (OR 6.6).[@R6] However, there was a paucity of studies on tracheostomy that prohibited a direct comparison of the procedures. In our opinion, current data vastly underestimate the risk of tracheostomy, in which droplet and blood splatter is virtually guaranteed. Healthcare practitioners performing tracheostomy are an at-risk population. In the 2002 severe acute respiratory syndrome (SARS) epidemic in Toronto, many of those hospitalized with illness were healthcare workers[@R8] who were exposed prior to major infection control measures were instituted. In Wuhan, China, 40 of 138 hospitalized patients were healthcare providers who were infected from presumed hospital spread.[@R6]
With the current pandemic, significant attention has been focused on the safety of healthcare workers, and many organizations have published guidance on infection prevention and control for these essential personnel.[@R9]
Considerations for indications and timing {#s5}
=========================================
Surgeons should consider both short-term and long-term outcomes of tracheostomy along with the risks of exposure of the clinical team. In many cases tracheostomy should be deferred until the patient has ceased viral shedding. In some circumstances, tracheostomy may accelerate ventilator weaning,[@R4] which might improve throughput of patients with Covid-19 in the hospital system, making room for new patients if ICU resources and ventilators become scarce. This is important since, depending on the trajectory of the pandemic in the USA, it is projected that the need for ventilators may far exceed the number of devices available, currently estimated at approximately 75 000 including those available in the Strategic National Stockpile[@R13] and another 98 000 ventilators that can perform only basic functions.[@R14]
The long-term outcomes after tracheostomy in non-surgical patients are poor, with a 1-year mortality of 46.5% overall and 54.7% for those over age 65 years,[@R15] and these data may be useful in having goals of care discussions with families. In the absence of large-scale triage, the decision to perform a tracheostomy in a patient with Covid-19 currently should be made on a case-by-case basis and with multidisciplinary input, maintaining a patient-centered and family-centered and caregiver safety-focused approach. It should be emphasized that data on tracheostomy in this population are very limited. Patients with severe disease likely are not physiologically stable enough to undergo the procedure, and patients who are recovering from the disease may benefit from traditional ventilator weaning and liberation strategies. At this time, we recommend against performing tracheostomy in patients with active Covid-19 disease.
We recommend using one of the following strategies {#s5-1}
--------------------------------------------------
- Delayed tracheostomy
- Consider pharmacological pretreatment and perform viral load testing first to confirm non-transmissibility of the disease.[@R16] If testing is negative for Covid-19, proceed with tracheostomy. While some have recommended using standard precautions after a patient with coronavirus tests negative,[@R17] full personal protective equipment (PPE) may be prudent unless testing has an acceptably low false negative rate.
- Not performing tracheostomy
- Continue standard ventilator weaning until extubation.
The high-risk surgical airway {#s6}
=============================
For urgent cricothyroidotomy, patients are often purposely not paralyzed to avoid removing any residual respiratory drive until a definitive airway is in place. However, in patients with known or suspected Covid-19 infection, risk of wide dissemination and droplet spray on surgical airway entry makes neuromuscular blockade prior to cricothyroid membrane incision justified.
- Despite the urgency of the situation, it is essential that providers wear appropriate PPE prior to any intervention.
- Hold ventilation prior to opening the cricothyroid membrane and until placement of the definitive airway.
- If difficulty is encountered in placing the airway, and the patient needs to be ventilated again by bag-valve mask, occlude the cricothyroidotomy opening with a finger to prevent air leak. Hold ventilation again prior to reattempting placement.
In other high-risk populations such as the very obese, transport to the optimal environment of the operating room is often preferred. If transportation is not desirable because of risk of viral spread, the surgeon should consider the feasibility of performing a safe bedside procedure or delaying the procedure. In all patients with Covid-19 who need a tracheostomy, an acceptable strategy is to wait for the disease to become non-transmissible\* prior to performing a high-risk aerosol-generating procedure such as tracheostomy.
\*See CDC recommendations for discontinuation of transmission-based precautions.[@R16]
Procedural guidance for Open and Percutaneous Tracheostomy {#s7}
==========================================================
Here we provide practical guidance for the performance of OT and PT in patients with known or suspected Covid-19 infection. The risk of complications and death are similar between OT and PT, except that stoma site infections are more common with OT.[@R4] The choice of OT or PT may be made based on an individual patient's clinical condition, anatomy, the operator's experience with each technique and logistical considerations such as the risk of transportation (if necessary) to the operating room for OT.
Preparation and procedural safety {#s7-1}
---------------------------------
1. Perform in a negative pressure airborne infection isolation room (AIIR).
a. If an AIIR is not available, avoid entry into room by non-essential personnel for up to 3 hours due to persistence of viable virus in aerosols.[@R18]
2. Limit the number of participants in the room to essential personnel only.[@R19]
3. An experienced attending surgeon or other experienced practitioner should perform the procedure.
a. Trainees should not be involved unless absolutely necessary[@R19] to expedite the procedure and avoid unnecessary risk.
4. Post a runner outside the room to aid communication and to obtain new equipment as needed.
5. Take only essential equipment into the room, including an oversupply of any medications that will be needed. Have potentially necessary and backup equipment immediately outside the room.
a. It is important to avoid delays or interruptions after starting the procedure due to lack of equipment or sedative medications.
6. Ensure presence of a HEPA viral filter on the ventilator and suctioning equipment.
7. Perform standard hand hygiene and use a double glove technique, which has been recommended to reduce risk of viral transfer during doffing of PPE.[@R20] Wear a fluid-resistant gown. Double gowning has also been recommended by some.[@R21]
8. Use a powered air purifying respirator (PAPR) with standard donning as recommended by the CDC. Use an N95 mask under the PAPR hood as backup in the event of PAPR mechanical failure.
a. According to the CDC, not using a respirator mask during an aerosol-generating procedure upgrades the risk of healthcare personnel from low risk to medium risk.[@R9]
b. According to the Anesthesia Patient Safety Foundation, PAPR is superior to a mask for protection from viral transmission.[@R11]
c. Respirator efficacy is measured by the assigned protection factor (APF). PAPRs used in healthcare typically have an APF of 25, while N95 masks filter 95% of particles and have an APF of 10.[@R22]
d. PAPR has been recommended based on experience with the SARS epidemic in Asia.[@R21]
e. During non-procedural situations, in the event of failure of PAPR gear, healthcare personnel are instructed to leave the room immediately since they are no longer protected from airborne viral transmission. Suddenly aborting a tracheostomy procedure at a critical moment could result in airway loss and death. Therefore, we recommend wearing an N95 mask under the PAPR as a backup to allow completion of the procedure should the PAPR fail.
f. If situations where a PAPR is not available, personnel should use an N95 or higher mask, along with a fluid shield and full eye protection.
9. Use neuromuscular blockade in addition to full sedation/analgesia to prevent coughing and resultant particulate spread.
a. Routine paralysis has been recommended by practitioners involved with tracheostomy during the SARS outbreak of 2002 in Asia[@R21] and other specialty societies.[@R24]
10. Fully drape the entire patient and bed to avoid any contamination of the bed, pillow, sheets or equipment.
a. Use a double layer of impervious draping to prevent soak-through.
b. Place instruments on a flat tray or table instead of on the patient to avoid equipment rolling or falling off the bed.
11. Place a cuffed, non-fenestrated tracheostomy tube.[@R24] Inflate cuff after placement and check to ensure absence of a cuff leak (see *Technical considerations* for PT and OT below).
12. Doff PPE as recommended by the CDC.
Technical considerations {#s7-2}
------------------------
### Open tracheostomy {#s7-2-1}
- Avoid electrocautery to prevent aerosolization of viral particles.
- Stop mechanical ventilation after an exhalation just prior to tracheal entry.
- Hold ventilation until intratracheal placement of the tracheostomy tube and inflation of the cuff, if the patient's condition will allow (no critical hypoxemia).
- Resume ventilation through the tracheostomy after cuff inflation.
- Remove endotracheal tube from the mouth, placing it directly into a plastic bag for disposal.
### Percutaneous tracheostomy {#s7-2-2}
- Use of bronchoscopy
a. Bronchoscopy is often used in PT to localize the insertion site, aid visualization and avoid damage to the back wall of the trachea. Bronchoscopy itself is an aerosol-generating procedure and could pose an addition risk of exposure. While there is insufficient evidence that its use decreases the number of complications during tracheostomy,[@R4] many surgeons use bronchoscopy as a standard component of the PT procedure. If performing PT in a patient with active Covid-19 infection, surgeons should consider their individual expertise and experience with performing PT without bronchoscopy to decide on its use.
b. If available, use of disposable, single-use bronchoscopes is recommended.
c. If opting not to use bronchoscopy, consider alternative methods to determine withdrawal of the endotracheal tube above the tracheotomy site, including but not limited to:
1. Palpation with a finger on the trachea while the endotracheal tube is being withdrawn; the surgeon can feel the trachea become softer and more pliable as the tube is withdrawn above the proposed tracheotomy site.
2. Use of a Doppler over the incision site; when the endotracheal tube is withdrawn above the proposed tracheotomy site, the audible volume from air flow through the end of the tube will be much louder.
3. Blind placement of the needle, using aspiration of air or bubbles in a fluid-filled syringe to confirm intratracheal placement.
- Avoid electrocautery to prevent aerosolization of viral particles.
- Stop mechanical ventilation after an exhalation, after placing the guidewire and just prior to tracheal dilation.
- Hold ventilation until intratracheal placement of the tracheostomy tube and inflation of the cuff, if the patient's condition will allow (no critical hypoxemia).
- Resume ventilation through the tracheostomy after cuff inflation.
- Remove endotracheal tube from the mouth, placing it directly into a plastic bag for disposal.
The authors would like to thank the members of the Critical Care and Acute Care Surgery Committees of the American Association for the Surgery of Trauma for their review of the document and for providing valuable insight and personal experience.
**Contributors:** CM provided the conception and design. All authors contributed to manuscript preparation, interpretation of data and critical revision.
**Funding:** The authors have not declared a specific grant for this research from any funding agency in the public, commercial or not-for-profit sectors.
**Competing interests:** None declared.
**Patient consent for publication:** Not required.
**Provenance and peer review:** Not commissioned; internally peer reviewed.
| {
"pile_set_name": "PubMed Central"
} |
Morelli F, Benedetti Y, Møller AP, Fuller RA. Measuring avian specialization. Ecol Evol. 2019;9:8378--8386. 10.1002/ece3.5419
**Data Availability Statement:** The datasets generated during and/or analyzed during the current study is provided in the Appendix.
1. INTRODUCTION {#ece35419-sec-0001}
===============
Measuring the extent to which a species is specialized is a major challenge in ecology, with important repercussions for fundamental and applied research, including conservation (Clavero, Brotons, & Herrando, [2011](#ece35419-bib-0011){ref-type="ref"}; Futuyma & Moreno, [1988](#ece35419-bib-0026){ref-type="ref"}). Ecologically specialist species are those occupying a relatively narrow niche or a restricted range of habitats (Clavel, Julliard, & Devictor, [2011](#ece35419-bib-0010){ref-type="ref"}), or using only a portion of the resources available in a habitat. In contrast, ecologically generalist species are able to thrive in a wide variety of environmental conditions, exploiting a large variety of available resources across space or time (Ducatez, Clavel, & Lefebvre, [2015](#ece35419-bib-0020){ref-type="ref"}; Irschick, Dyer, & Sherry, [2005](#ece35419-bib-0036){ref-type="ref"}).
Measuring the degree of specialization of a species is important for assessing extinction risk, since specialist species are considered more prone to the processes that lead to extinction than generalist species (Colles, Liow, & Prinzing, [2009](#ece35419-bib-0012){ref-type="ref"}; Devictor, Julliard, & Jiguet, [2008](#ece35419-bib-0018){ref-type="ref"}; McKinney, [1997](#ece35419-bib-0044){ref-type="ref"}). This is mainly because species with a broader niche have been hypothesized to have greater capacity to respond to or tolerate anthropogenic disturbances (Devictor et al., [2008](#ece35419-bib-0018){ref-type="ref"}; Hammond, Palme, & Lacey, [2018](#ece35419-bib-0032){ref-type="ref"}; Vázquez & Simberloff, [2002](#ece35419-bib-0058){ref-type="ref"}). There is also empirical evidence that specialist bird species are declining throughout Europe (Bowler, Heldbjerg, Fox, Jong, & Böhning‐Gaese, [2019](#ece35419-bib-0006){ref-type="ref"}; Julliard, Jiguet, & Couvet, [2004](#ece35419-bib-0038){ref-type="ref"}).
The IUCN Red List of threatened species is currently the most comprehensive tool for extinction risk classification (Webb, [2008](#ece35419-bib-0061){ref-type="ref"}), yet it could be further enhanced if a measure of ecological specialization were incorporated into the assessment criteria. This would add another dimension to determining which species are more vulnerable to anthropogenic threats (Devictor et al., [2008](#ece35419-bib-0018){ref-type="ref"}). All else being equal, a specialist species is more likely to be at a higher risk of extinction than a generalist species, for example, specialists forage on a narrower variety of food items or are able to live in a smaller range of habitats than generalist species (Julliard, Clavel, Devictor, Jiguet, & Couvet, [2006](#ece35419-bib-0037){ref-type="ref"}). Information about change in species\' environmental preferences or changes in niche size has recently been incorporated into IUCN Red List assessments (Breiner, Guisan, Nobis, & Bergamini, [2017](#ece35419-bib-0007){ref-type="ref"}), and since specialist species tolerate a narrower range of environmental conditions than generalists, adding a metric of specialization might also shed light on the capacity of species to respond to environmental challenges.
To explore the causes and consequences of ecological specialization, researchers often classify species as generalists or specialists, often by focusing on the strength of species\' affinities for particular habitats (Barnagaud, Devictor, Jiguet, & Archaux, [2011](#ece35419-bib-0002){ref-type="ref"}; Chazdon et al., [2011](#ece35419-bib-0009){ref-type="ref"}; Dondina, Orioli, D\'Occhio, Luppi, & Bani, [2016](#ece35419-bib-0019){ref-type="ref"}; Dufrene & Legendre, [1997](#ece35419-bib-0022){ref-type="ref"}; Vázquez & Simberloff, [2002](#ece35419-bib-0058){ref-type="ref"}). Indeed, a dichotomous distinction between "specialization" and "generalism" dates back more than 150 years in parasitology (Combes, [2004](#ece35419-bib-0014){ref-type="ref"}). In recent decades, more nuanced attempts have been made to estimate the level of ecological specialization of different bird species (Clavero et al., [2011](#ece35419-bib-0011){ref-type="ref"}; Julliard et al., [2006](#ece35419-bib-0037){ref-type="ref"}). While researchers have continued to produce classifications based on a binomial categorization as habitat "specialist" or "generalist" (Gregory et al., [2005](#ece35419-bib-0030){ref-type="ref"}), others have begun to arrange species along a gradient of specialization, for example, habitat, diet, or foraging substrate plasticity (Luck, Carter, & Smallbone, [2013](#ece35419-bib-0042){ref-type="ref"}; Moreira, Ferreira, Rego, & Bunting, [2001](#ece35419-bib-0045){ref-type="ref"}). The habitat specificity or species‐habitat specialization has been quantified by measuring the breadth of use of a particular habitat type by an individual and hence by implication for a given species (Devictor et al., [2010](#ece35419-bib-0017){ref-type="ref"}). Additionally, methods are becoming available to construct continuous measures of habitat generalism--specialism, known as the Species Specialization Index (SSI; Julliard et al., [2006](#ece35419-bib-0037){ref-type="ref"}), an approach now applied in many studies (Devictor et al., [2008](#ece35419-bib-0018){ref-type="ref"}; Reif, Hořák, Krištín, Kopsová, & Devictor, [2016](#ece35419-bib-0050){ref-type="ref"}; Reif, Jiguet, & Šťastný, [2010](#ece35419-bib-0051){ref-type="ref"}). The SSI is relatively easy to calculate, because it is based only on the frequency of occurrence of each species in each habitat or land use type available in the study area (Devictor et al., [2008](#ece35419-bib-0018){ref-type="ref"}). However, for the same reason, the resulting SSI has a limited value if based on few sample sites, or when there is significant bias in sampling (Fraser, Pichancourt, & Butet, [2016](#ece35419-bib-0024){ref-type="ref"}). Additionally, and perhaps more fundamentally, specialism can vary along multiple dimensions, and thus one cannot determine the extent of ecological specialization by considering only a single dimension. For example, a species could be highly specialized in a particular type of diet, while at the same time be generalist in the selection of breeding habitat or nesting site. In other words, specialization is a syndrome‐like modification of some characteristics of a phenotype to allow efficient exploitation of specific resources (Devictor et al., [2010](#ece35419-bib-0017){ref-type="ref"}). For this reason, measures of ecological specialization must span multiple dimensions, using data on multiple traits of species, such as behavior or diet. Yet many existing metrics of avian specialization are focused on just one dimension (e.g., diet type and specificity, habitat breadth; Devictor et al., [2008](#ece35419-bib-0018){ref-type="ref"}; Luck et al., [2013](#ece35419-bib-0042){ref-type="ref"}; Moreira et al., [2001](#ece35419-bib-0045){ref-type="ref"}). Because the degree of specialization can vary intraspecifically among traits, considering only one ecological dimension is incomplete and such measures must incorporate other attributes or ecological traits.
Here, we develop a multidimensional index of specialization, based on a set of ecological characteristics of species. We then test the phylogenetic distribution of the specialization indices, and determine how large a proportion of variance (or deviance) of such indices is shared throughout the phylogeny, by calculating the phylogenetic signal for each specialization index. As a case study, we use two recent databases of species traits of European birds based on foraging ecology, habitat, and breeding characteristics (Pearman et al., [2014](#ece35419-bib-0048){ref-type="ref"}; Storchová & Hořák, [2018](#ece35419-bib-0056){ref-type="ref"}). We expect that the methods for constructing and evaluating the multidimensional index could be readily adaptable to other taxa and regions, depending on the availability of information on species traits.
2. METHODS {#ece35419-sec-0002}
==========
2.1. Avian species traits {#ece35419-sec-0003}
-------------------------
We formulate a definition of ecological specialization of species referring to a set of multidimensional species traits, well studied in European birds. We choose a set of species traits of European breeding birds focusing on diet, foraging behavior, foraging substrate, general habitat, and nesting site characteristics for each European bird species, by compiling data from two recent publications (Pearman et al., [2014](#ece35419-bib-0048){ref-type="ref"}; Storchová & Hořák, [2018](#ece35419-bib-0056){ref-type="ref"}). The species‐trait approach is traditionally used to focus on the functional aspects of biodiversity (de Bello, Lavorel, Gerhold, Reier, & Pärtel, [2010](#ece35419-bib-0016){ref-type="ref"}; Violle et al., [2007](#ece35419-bib-0060){ref-type="ref"}). The list of the groups of species traits and the corresponding sources for each dataset are given in Table [1](#ece35419-tbl-0001){ref-type="table"}. The complete list of traits is provided in Table [S1](#ece35419-sup-0001){ref-type="supplementary-material"}. All variables are binomial, scored as 0 or 1.
######
Species traits used for the estimation of specialization indices in European birds, including the number of variables for each group and sources of data
Group of species traits No. variables Source
------------------------- --------------- ------------------------------------------------------------------
Diet (all year) 9 Storchová and Hořák ([2018](#ece35419-bib-0056){ref-type="ref"})
Diet (breeding season) 9 Storchová and Hořák ([2018](#ece35419-bib-0056){ref-type="ref"})
Foraging behavior 9 Pearman et al. ([2014](#ece35419-bib-0048){ref-type="ref"})
Foraging substrate 9 Pearman et al. ([2014](#ece35419-bib-0048){ref-type="ref"})
Habitat 15 Storchová and Hořák ([2018](#ece35419-bib-0056){ref-type="ref"})
Nesting site 18 Pearman et al. ([2014](#ece35419-bib-0048){ref-type="ref"})
John Wiley & Sons, Ltd
2.2. Specialization indices and overall specialization {#ece35419-sec-0004}
------------------------------------------------------
We estimated the degree of specialization in diet, foraging behavior, foraging substrate, habitat, and nesting site for each bird species using the Gini index of inequality (Colwell, [2011](#ece35419-bib-0013){ref-type="ref"}; Gini, [1921](#ece35419-bib-0029){ref-type="ref"}). The index is based on the Gini coefficient, a measurement of statistical dispersion on a scale between 0 and 1, representing low to high specialization, respectively. This measure was developed by the Italian statistician Corrado Gini in 1921 and is probably the best single measure of inequality (Gastwirh, [1972](#ece35419-bib-0028){ref-type="ref"}). It is commonly used in the study of economic inequalities (Lerman & Yitzhaki, [1984](#ece35419-bib-0041){ref-type="ref"}), and also for measuring the evenness of coverage of protected areas among habitat types (Barr et al., [2011](#ece35419-bib-0003){ref-type="ref"}).
The Gini coefficient is estimated with the following formula:$$G = \frac{\sum_{i = 1}^{n}\sum_{j = 1}^{n}{\lbrack x_{i} - x_{j}\rbrack}}{2n^{2}\overline{x}}$$where "*x*" is an observed value, \"*n*" is the number of values observed and "$\overline{x}$" is the mean value.
In the specific case of our table of avian traits, if every variable in a group (e.g., diet specialism) has exactly the same value or weight, the index would equal 0, indicating the maximum generalism for that trait. In contrast, the Gini coefficient would equal 1, indicating perfect inequality (high specialization), when a species has a diet entirely composed of a single type. Applying this procedure, we obtained five different specialization indices: diet specialism, foraging behavior specialism, foraging substrate specialism, general habitat specialism, and nesting site specialism.
Finally, to explore the consequences of reducing the index to a single number, an overall "specialization index" was estimated for each species, calculated as the mean, maximum, and minimum values of the five single specialization indices based on diet, foraging behavior, substrate, habitat, and nesting site, subsequently standardized between 0 (generalist species) and 1 (specialist species).
2.3. Phylogenetic signal of specialization {#ece35419-sec-0005}
------------------------------------------
The phylogenetic signal can be briefly defined as the tendency for related species to resemble each other, more than they resemble species drawn at random from a phylogenetic tree (Blomberg, Garland, & Ives, [2003](#ece35419-bib-0005){ref-type="ref"}). This is because all organisms descend from common ancestors and hence are related in a hierarchical fashion (Futuyma & Agrawal, [2009](#ece35419-bib-0025){ref-type="ref"}). A high phylogenetic signal indicates species traits that are more similar in close relatives than distant relatives, while traits that are more similar in distant than close relatives or randomly distributed species across a phylogeny suggest a low phylogenetic signal (Kamilar & Cooper, [2013](#ece35419-bib-0039){ref-type="ref"}). Some studies have focused on quantifying these differences in phylogenetic signal among species and traits (Blomberg et al., [2003](#ece35419-bib-0005){ref-type="ref"}; Münkemüller et al., [2012](#ece35419-bib-0046){ref-type="ref"}). However, further studies have to clarify the nature of phylogenetic signal in biological or functional traits, mainly in behavioral and ecological characteristics of species (Kamilar & Cooper, [2013](#ece35419-bib-0039){ref-type="ref"}). Here, we calculated the phylogenetic signal for all specialization indices, to test whether the indices appear to be describing an ecological phenomenon underpinned by evolution.
Considering that bird species are evolutionarily related, they cannot be treated as independent sampling units in comparative analyses (Harvey & Purvis, [1991](#ece35419-bib-0033){ref-type="ref"}). Thus, we modeled interspecific variation across a phylogeny, obtaining the phylogenetic relationships from "[www.birdtree.org](http://www.birdtree.org)". We downloaded 1,000 phylogenetic trees from the backbone tree based on Ericson et al. ([2006](#ece35419-bib-0023){ref-type="ref"}) for the 365 bird species that were the focus of this study. The consensus tree was obtained applying the 50% majority rule (i.e., the proportion of a split to be present in all trees). In order to manage phylogenetic trees, we used the following R packages: "ape" (Paradis, Claude, & Strimmer, [2004](#ece35419-bib-0047){ref-type="ref"}), "phangorn" (Schliep, [2011](#ece35419-bib-0053){ref-type="ref"}), and "Rphylip" (Revell & Chamberlain, [2014](#ece35419-bib-0052){ref-type="ref"}).
2.4. Statistical analysis {#ece35419-sec-0006}
-------------------------
The Gini coefficient for each group of species traits (specialization indices) was calculated using the package "DescTools" for R (Signorell, [2019](#ece35419-bib-0055){ref-type="ref"}). Associations among the specialization indices for diet, foraging behavior and substrate, habitat, and nesting site were explored using correlation coefficients. A Shapiro--Wilk normality test was used to test the normality of the distribution of each specialism index, and a Spearman correlation test was used when the distribution of the specialization indices was not normal (Triola, [2012](#ece35419-bib-0057){ref-type="ref"}).
To measure the strength of the phylogenetic signal (Blomberg & Garland, [2003](#ece35419-bib-0004){ref-type="ref"}) in the five specialization indices and the overall specialization index for 365 European bird species, we used Blomberg\'s *K* statistic and statistic *K* ^\*^ (Blomberg et al., [2003](#ece35419-bib-0005){ref-type="ref"}). The *K* statistic works as a mean square ratio, where the numerator is the error assuming that the trait evolves independently of the phylogenetic structure, and the denominator is corrected by the phylogenetic covariances. When *K* approaches 1, trait evolution follows a mode of evolution that is consistent with Brownian motion. If *K* \> 1 and \<1, close relatives are more similar and less similar, respectively, than expected under Brownian motion, indicating a strong phylogenetic signal, while *K*‐values closer to zero it is concluded that the trait has no phylogenetic signal (Blomberg et al., [2003](#ece35419-bib-0005){ref-type="ref"}). Blomberg\'s *K* statistic was estimated using the R package "phylosignal" (Keck, Rimet, Bouchez, & Franc, [2016](#ece35419-bib-0040){ref-type="ref"}). Moran\'s correlograms were used to assess how phylogenetic autocorrelation changes across different phylogenetic distances. Moran\'s correlograms were plotted using the function "phyloCorrelogram" from the package "phylosignal" (Keck et al., [2016](#ece35419-bib-0040){ref-type="ref"}).
All statistical tests were performed with R software version 3.2.4 (R Development Core Team, [2017](#ece35419-bib-0049){ref-type="ref"}).
3. RESULTS {#ece35419-sec-0007}
==========
We calculated six specialization indices for each bird species, considering different functional dimensions (diet all year, diet during the breeding season, foraging behavior, foraging substrate, habitat, and nesting site) by estimating the Gini coefficient (Table [S2](#ece35419-sup-0001){ref-type="supplementary-material"}). All specialism indices showed a non‐normal distribution (Shapiro--Wilk normality test for all specialism indices, *p*‐values \< 0.05). The most strongly correlated specialism indices among species traits were the indices for diet all year and diet during the breeding season, followed by habitat specialism with nesting site specialism (Figure [1](#ece35419-fig-0001){ref-type="fig"}, Table [S3](#ece35419-sup-0001){ref-type="supplementary-material"}). Foraging behavior specialism was also correlated with foraging substrate specialism and foraging substrate specialism with nesting site specialism (Figure [1](#ece35419-fig-0001){ref-type="fig"}, Table [S3](#ece35419-sup-0001){ref-type="supplementary-material"}). Nesting site specialism was significantly correlated with all the other specialism indices, while other specialism indices were not statistically significantly correlated among themselves (Figure [1](#ece35419-fig-0001){ref-type="fig"}, Table [S3](#ece35419-sup-0001){ref-type="supplementary-material"}). Considering the strong correlation between diet all year and diet during the breeding season (correlation coefficient = 0.833, *p* \< 2.2e−16), we use only diet during year for further analysis.
{#ece35419-fig-0001}
Analyzing specialization separately for each functional dimension, 111 species were classified as diet specialists (30.4%), 143 as foraging behavior specialists (39.2%), 68 as foraging substrate specialists (18.6%), 96 as habitat specialists (26.3%), and two as nesting site specialists (0.5%; Table [S2](#ece35419-sup-0001){ref-type="supplementary-material"}).
Additionally, we calculated the overall specialization index by normalizing the mean values of the five specialization indices between 0 and 1 (Table [S2](#ece35419-sup-0001){ref-type="supplementary-material"}). Overall, specialization varied markedly among taxa, with centers of specialization apparent for example in some shorebird clades, as well as raptors, Galliformes and Coraciiformes (Figure [2](#ece35419-fig-0002){ref-type="fig"}). The five species with the highest degree of overall specialism were great gray owl *Strix nebulosa*, bearded vulture *Gypaetus barbatus*, Eurasian crag martin *Ptyonoprogne rupestris*, sociable lapwing *Vanellus gregarius*, and boreal owl *Aegolius funereus* (Table [S2](#ece35419-sup-0001){ref-type="supplementary-material"}). Marked generalism occurred in several clades such as tits, thrushes, and crows (Figure [2](#ece35419-fig-0002){ref-type="fig"}). The five most generalist species were common chaffinch *Fringilla coelebs*, European pied flycatcher *Ficedula hypoleuca*, common crane *Grus grus*, carrion crow *Corvus corone*, and European robin *Erithacus rubecula* (Table [S2](#ece35419-sup-0001){ref-type="supplementary-material"}).
{#ece35419-fig-0002}
Analysis of the phylogenetic signal in all five specialization index values returned the following statistically significant *K* and *K* ^\*^ values (all *p* \< 0.01): *K* = 1.082 for diet specialism, *K* = 0.917 for foraging behavior specialism, *K* = 0.879 for foraging substrate specialism, *K* = 0.753 for habitat specialism, and *K* = 0.777 for nesting site specialism, suggesting a generally high degree of phylogenetic signal (Table [2](#ece35419-tbl-0002){ref-type="table"}, Figure [3](#ece35419-fig-0003){ref-type="fig"}). For habitat specialism and nesting site specialism, the *K*‐values were lower than 1 (*K* = 0.753--0.777) and statistically significant, suggesting that a model similar to Brownian motion is likely, although closely related species are slightly less similar in the two specialization indices than expected based on phylogenetic relatedness alone (Table [2](#ece35419-tbl-0002){ref-type="table"}). Also the index of overall specialism was characterized by a statistically significant phylogenetic signal (Table [2](#ece35419-tbl-0002){ref-type="table"}, Figure [3](#ece35419-fig-0003){ref-type="fig"}).
######
Phylogenetic signal of five specialization indices based on diet, foraging behavior, foraging substrate, habitat, and nesting site and the overall specialization index for 365 European bird species included in this study
Specialism index *K* statistic *p* value *K* ^\*^ statistic *p* value
------------------------------- --------------- ----------- -------------------- -----------
Diet specialism 1.082 \<0.01 1.081 \<0.01
Foraging behavior specialism 0.917 \<0.01 0.919 \<0.01
Foraging substrate specialism 0.879 \<0.01 0.872 \<0.01
Habitat specialism 0.753 \<0.01 0.755 \<0.01
Nesting site specialism 0.777 \<0.01 0.780 \<0.01
Overall specialism 0.892 \<0.01 0.889 \<0.01
The table shows *K* statistic, *K* ^\*^ statistic, and associated *p*‐values for each index.
John Wiley & Sons, Ltd
{#ece35419-fig-0003}
4. DISCUSSION {#ece35419-sec-0008}
=============
The use of niche or functional dimensionality in the study of wildlife ecology dates back more than 100 years to the classical work by Grinnell ([1917](#ece35419-bib-0031){ref-type="ref"}). Species diversification, changes in species traits, and niche evolution across the tree of life are mainly due to the process of adaptive radiation (Castiglione, Mondanaro, Carotenuto, & Passaro, [2017](#ece35419-bib-0008){ref-type="ref"}; Schluter, [2000](#ece35419-bib-0054){ref-type="ref"}). As a result, traits of species provide a tool for understanding---and potentially classifying---such species in terms of a specialization gradient. Here, we have provided and tested a simple framework for calculating specialization indices based on species traits.
We calculated five different indices of specialization, focusing on five different groups of readily available species traits or "natural history" dimensions of European birds and applying the Gini coefficient to each set of traits. We also explored how the specialization indices in different functional dimensions are correlated. Among the five specialization indices estimated for European birds, diet specialism calculated for the entire year and diet specialism calculated for the breeding season were the most tightly correlated (Table [S3](#ece35419-sup-0001){ref-type="supplementary-material"}). This result could be interpreted in ecological terms as confirming a relatively constant diet composition through the year in European breeding birds, but it could be also interpreted in methodological terms, suggesting that just one dimension (e.g., diet throughout the year) is sufficient to characterize dietary specialization in this group of birds. However, although many indices were positively correlated with one another, only a few were strongly related, highlighting the importance of assessing specialism in a number of different dimensions without reducing specialization to a single overall index value. Using a diverse set of traits permits a better description of each dimension characterizing the species, as well as the overall level of specialism. This is also important for conservation since different sets of traits can help identify a broader range of species\' vulnerabilities, and hence which species might be most sensitive to which anthropogenic threats (Allan et al., [2019](#ece35419-bib-0001){ref-type="ref"}; Hatfield, Orme, Tobias, & Banks‐Leite, [2018](#ece35419-bib-0034){ref-type="ref"}; Henle, Davies, Kleyer, Margules, & Settele, [2004](#ece35419-bib-0035){ref-type="ref"}).
All indices estimated in this study showed a strong phylogenetic signal, indicating that more closely related species tended to show more similar levels of specialization. This is further confirmation that the specialization indices calculated in this study, by applying the Gini coefficient on groups of species traits, are describing ecological phenomena congruent with evolutionary principles. The continuous traits of closely related species in a phylogeny tend to be similar, mainly because such traits are derived from a common ancestor and because they were shaped by selection originating from the environment (Keck et al., [2016](#ece35419-bib-0040){ref-type="ref"}). The Brownian motion model assumes that the correlation among trait values is proportional to the extent of shared ancestry for pairs of species, or, in other words, that "members of lineages that have only recently diverged will necessarily (on average) tend to be similar, as compared with more distantly related lineages" (Blomberg et al., [2003](#ece35419-bib-0005){ref-type="ref"}). Our results suggest that the five specialization indices estimated for European birds operate in a similar manner, even if for some specialism indices close relatives were more similar (diet specialism) or less similar (other specialism indices) than expected under a Brownian motion model of trait evolution (Blomberg et al., [2003](#ece35419-bib-0005){ref-type="ref"}).
The use of specialization indices based on species traits raise the possibility of robustly comparing results across studies and regions, and updating the indices as additional information becomes available. In comparison with previous approaches (e.g., methods for calculating the Species Specialization Index; Julliard et al., [2006](#ece35419-bib-0037){ref-type="ref"}), our method does not need de novo data collection.
In conservation ecology, a deep understanding of the characteristics that make a species susceptible to extinction is essential. Ecological specialization is generally thought to be a key contributor to a species\' risk of extinction, although while paleoecological studies investigating longer term survival have confirmed this hypothesis, other comparative studies focusing on the history of entire lineages, suggest that specialist species could be more ecologically "plastic" than expected, sometimes able to become generalists (Clavel et al., [2011](#ece35419-bib-0010){ref-type="ref"}; Colles et al., [2009](#ece35419-bib-0012){ref-type="ref"}). With the tool presented in this study, we expect to forge a deeper understanding of the level of specialization of species, by focusing on the relative specialism in different trait dimensions and pointing out how this multidimensional gradient of specialization can be used to assess the overall conservation status of different species. However, although the proposed methodology is useful for measuring the level of specialization of species, we highlight potential drawbacks and finally provide some thoughts for optimizing the potential of this approach.
An important point is that species classified as a specialist in at least one category (e.g., diet specialist and habitat specialist) could, and perhaps should, be considered a specialist species overall. For example, extreme specialism in just one category of species traits (e.g., diet) could determine the level of extinction risk for a species, much more so than the value of the overall specialization index, in which extreme values are averaged away. While we recognize the convenience of deriving a single index of specialization (e.g., overall specialism index, created in our study only for reference), we consider it preferable to work with the five constituent specialization indices. So, we suggest assessing the level of specialization of species by considering separately each dimension of specialization or bundle of traits. We also suggest treating the specialization indices as a package or bundle, in a similar way as proposed for the multidimensional indices for estimating functional diversity (Villéger, Mason, & Mouillot, [2008](#ece35419-bib-0059){ref-type="ref"}). Loss of information that is potentially useful for conservation will occur if we only consider the reduced subset of dimensions of the overall specialization index.
An index of specialization is only as reliable as the underlying data. The quality of information about traits varies from species to species, might be incomplete or inaccurate in some cases, depending on the quality and number of studies conducted on each species (Ducatez & Lefebvre, [2014](#ece35419-bib-0021){ref-type="ref"}; Garamszegi & Møller, [2012](#ece35419-bib-0027){ref-type="ref"}; McKenzie & Robertson, [2015](#ece35419-bib-0043){ref-type="ref"}). Furthermore, the type of variable used to fill out the trait‐features can also influence the index. In this study, we estimated the Gini coefficient using binomial traits (based on a characterization of the trait initially made explicit as yes/no. For example, diet specialism was assessed using nine categories (folivore, frugivore, granivore, arthropods, etc.) and such categories were filled out by determining whether at least 10% of the diet during the year is composed of each type of food. It would also be possible to directly estimate the percentage of the diet made up of each food type, and this data structure would be even better suited for summarizing the Gini coefficient, which works best on continuous data. However, significant uncertainty could exist across such a large number of possible categories. For example, the diet of *Sylvia atricapilla* changes over time, the species being more insectivorous during the breeding season and frugivorous during autumn and winter. The habitat of *Fringilla coelebs* could be more variable than is easily expressed by these traits, because it inhabits forests during the breeding season and more open‐country habitats during autumn and winter. A parameter could be devised to take into account this temporal variability in some species, when calculating the specialism indices. Also, other indices could be applied to estimate specialization, as has been done for size and fecundity specialization in plant communities, where the Lorenz asymmetry coefficient has been used to understand how inequality is distributed across a set of communities or species (Damgaard & Weiner, [2000](#ece35419-bib-0015){ref-type="ref"}).
Finally, although we focused on five bundles of species traits of avian species, we recognize that specialization can also be measured in other dimensions. For example, further studies on degree of specialization could introduce gradients of specialization in brood parasitic species, by considering the number of host species, host preferences, or interspecific relationships between pollinator species and plants.
In conclusion, we propose the more widespread use of multidimensional gradients of species specialization, especially for the assessment of the conservation status of species. For example, a metric indicating the level of species\' specialization based on a trait‐based approach could be included in the protocol for IUCN Red List assessments. In the same way that niche size change was recently incorporated in such assessments (Breiner et al., [2017](#ece35419-bib-0007){ref-type="ref"}), we propose that information on species specialism is also included, because it might predict other dimensions of extinction risks, as suggested in many studies (Colles et al., [2009](#ece35419-bib-0012){ref-type="ref"}; Devictor et al., [2008](#ece35419-bib-0018){ref-type="ref"}; McKinney, [1997](#ece35419-bib-0044){ref-type="ref"}).
CONFLICT OF INTEREST {#ece35419-sec-0009}
====================
None declared.
AUTHOR CONTRIBUTIONS {#ece35419-sec-0010}
====================
F.M., Y.B. and R.A.F. conceived the idea and designed methodology; F.M. and Y.B. prepared the data and performed data analyses. All authors contributed critically to the drafts and gave final approval for publication.
Supporting information
======================
######
######
Click here for additional data file.
We are grateful to Luis Maria Carrascal for fruitful and stimulating discussions during the initial stage of this study. F.M. and Y.B. were financially supported by the Czech Science Foundation GAČR (project number 18‐16738S).
DATA ACCESSIBILITY {#ece35419-sec-0012}
==================
The datasets generated during and/or analyzed during the current study is provided in the Appendix.
| {
"pile_set_name": "PubMed Central"
} |
A 78-year-old man with a history of type 2 diabetes mellitus and arterial hypertension presented to the emergency department with right groin pain and fever. Just two days before, he had undergone laparoscopic transperitoneal inguinal hernia repair (TAPP) of a right-sided indirect inguinal hernia with fixation of mesh. Physical examination revealed swelling and painful palpation of the right groin. The patient had a total white blood cell (WBC) count 22,100 per microliter, neutrophilia and hemoglobin level of 16.7 g per deciliter. Inguinal ultrasonography demonstrated enlargement of the right spermatic cord with inflammation of the fat (Figure [1](#F1){ref-type="fig"}, arrows). A non-contrast computed tomography (CT) of the pelvis (Figure [2 A, B](#F2){ref-type="fig"}) revealed right-sided thickening of the spermatic cord and edema of the inguinal canal (blue arrows), both indicative of vasitis. Postoperative subcutaneous emphysema was noted (red arrows). The patient was treated non-invasively with broad spectrum antibiotics and analgesics. After two days, he was discharged with pain relief and without fever. As in the literature there is no report of infectious vasitis as a complication of TAPP, and we assume this is the first.
{#F1}
{#F2}
Comment
=======
Vasitis is an uncommon condition that can be misdiagnosed as incarcerated inguinal hernia due to a similar appearance at ultrasound, which leads to unnecessary surgeries. CT helps in the differentiation of vasitis from inguinal hernia because of the latter is clearly identifiable in multiplanar reconstructions \[[@B1]\].
Competing Interests
===================
The authors have no competing interests to declare.
| {
"pile_set_name": "PubMed Central"
} |
1. Introduction {#sec1-ijms-20-00193}
===============
Metformin is a biguande class compound that is the most widely prescribed and well-tolerated drug for type II diabetes. In recent years, it has also received attention as a potential anticancer agent. A retrospective study done by Evan et. al. was the first to show a correlation between metformin treatment for diabetes and a decreased risk of various cancer types \[[@B1-ijms-20-00193]\]. Furthermore, a more recent clinical study reported that metformin improves response to chemotherapy in breast cancer patients with diabetes \[[@B2-ijms-20-00193]\]. The ability of metformin to act as an anticancer drug has been attributed to both the indirect effects of lowered insulin levels and the direct targeting of tumor cells. Hirsch et al. demonstrate that metformin, at clinically relevant doses, is able to preferentially target breast cancer stem cells (CSCs) \[[@B3-ijms-20-00193]\]. CSCs are a small subpopulation of cells within a tumor that share the stem cell capabilities of self-renewal and differentiation. Unlike the majority of cancer cells, these stem cells are resistant to conventional chemotherapy treatment and can regenerate tumors, which can lead to relapse of disease. Therefore, drugs that directly target CSCs offer substantial promise for the complete eradication of a tumor.
In order to effectively target both non-stem and stem cell populations, drugs that selectively target cancer stem cells are often tested in combination with conventional chemotherapy treatments that can deplete the bulk of the tumor. The in vivo studies of Hirsch et. al. revealed that low doses of metformin in combination with doxorubicin synergistically reduced tumor mass and prolonged remission in xenograft mouse models more effectively than either drug alone \[[@B3-ijms-20-00193]\]. In other studies, metformin showed comparable synergy with paclitaxel and carboplatin in breast, lung, and prostate cancer \[[@B4-ijms-20-00193]\]. In ovarian cancer stem cells, metformin was able to decrease cellular proliferation and increase the cytotoxic effects of cisplatin both in vitro and in vivo \[[@B5-ijms-20-00193]\].
Metformin's mechanism of action has mainly been attributed to the activation of AMP-activated protein kinase (AMPK) and the inhibition of Complex I of the mitochondria \[[@B6-ijms-20-00193]\]. AMPK inhibits the mammalian target of rapamycin (mTOR), a key regulator of cell growth and proliferation \[[@B7-ijms-20-00193]\]. In HNSCC, mutations in the Akt/mTOR pathway are considered some of the strongest oncogenic drivers \[[@B8-ijms-20-00193]\], which makes metformin an attractive therapeutic agent to study. Several other targets of metformin, such as mitochondrial glycerophosphate dehydrogenase (mGPD) and ataxia telangiectasia mutated (ATM), have been proposed and validated to some degree, suggesting that metformin's mechanism is complex and multi-faceted \[[@B6-ijms-20-00193]\]. Despite multiple hypotheses of metformin action, many aspects of its physiological effects, such as the alteration of gut microbiome composition, are not well understood \[[@B6-ijms-20-00193]\]. Furthermore, there are few molecular studies that validate the binding of metformin to proposed targets. For example, although metformin has been observed to inhibit complex I of the mitochondria in multiple studies, no definitive mechanism of the inhibition exists, and multiple groups claimed that they could not obtain direct evidence of complex I inhibition \[[@B9-ijms-20-00193]\].
In this study, we sought to determine the effects of metformin on HNSCC CSCs and non-stem cell cancer cells. In contrast to the reported effects of metformin on CSCs of other cancers, we demonstrated that metformin does not target HNSCC CSCs but instead promotes expression of stem cell markers, an indication of elevated stemness. Furthermore, when treated in combination with cisplatin, metformin significantly protected against chemotherapy-induced cell death. However, non-stem HNSCC cell populations were successfully reduced with metformin. We next investigated potential targets of metformin that could explain our results. Using a computational small molecule-to-protein docking software, we uncovered that the mitochondrial complex III interacts strongly with metformin. Since complex III is known as a major site of reactive oxygen species (ROS) production \[[@B10-ijms-20-00193]\], and metformin has been demonstrated to reduce ROS levels in several studies \[[@B11-ijms-20-00193],[@B12-ijms-20-00193]\], we offer the hypothesis that metformin reduces ROS levels through complex III inhibition to cause differential effects on HNSCC CSCs and non-stem cancer cells. Our hypothesis is supported by the fact that low ROS levels are required for CSC maintenance and self-renewal and that high ROS levels cause CSC differentiation or eradication \[[@B13-ijms-20-00193],[@B14-ijms-20-00193]\]. On the other hand, an increased ROS level is a hallmark of non-stem cancer cells and is likely to be a driver of increased cancer cell proliferation and cancer progression \[[@B15-ijms-20-00193],[@B16-ijms-20-00193]\]. The decrease in ROS levels mediated by metformin would promote stemness but reduce the ability of non-stem cancer cells to proliferate. Collectively, our results demonstrate that metformin is not effective for treating HNSCC in combination with cisplatin but could be an attractive treatment option if a third drug is used to selectively reverse the ROS-reducing effects of metformin in CSCs.
2. Results {#sec2-ijms-20-00193}
==========
2.1. Metformin Mitigates Cisplatin-Mediated Cell Death in HNSCC CSCs but Reduces Cell Proliferation in Non-Stem HNSCC Cells {#sec2dot1-ijms-20-00193}
---------------------------------------------------------------------------------------------------------------------------
We have previously created the JLO-1 CSC cell line by isolating HNSCC cells from a fresh laryngeal tumor and culturing them under conditions that favored the growth of stem cells. Flow cytometry confirmed the culture to be almost entirely CD44+ compared to the nonspecific IgG antibody used as a control \[[@B17-ijms-20-00193]\]. CD44 is an identified cell-surface marker for HNSCC stem cells, and in vivo studies showed that the enriched CD44+ population could give rise to new tumors while the CD44− population could not \[[@B18-ijms-20-00193]\]. We have previously verified the stemness of JLO-1 by demonstrating significantly higher levels of aldehyde dehydrogenase (ALDH1), Oct-4, and Nanog compared to established HNSCC cell lines \[[@B17-ijms-20-00193]\], and by demonstrating self-renewal. In addition, using the Aldefluor stem cell detection kit, we were able to sort the HNSCC cell line HN30 into ALDH+ and ALDH− populations using fluorescence-activated cell sorting (FACS) ([Figure 1](#ijms-20-00193-f001){ref-type="fig"}A). The ALDH marker isolates a subpopulation that is purer than using only CD44+ and has been utilized as a single-marker identifier of HNSCC CSCs \[[@B19-ijms-20-00193]\].
Using the doses previously reported to target breast CSCs \[[@B20-ijms-20-00193]\], MTS proliferation assays indicated that metformin alone has negligible effects on proliferation for both our putative CSC culture JLO-1 and the ALDH+ subpopulation of HN30 ([Figure 1](#ijms-20-00193-f001){ref-type="fig"}B). We next assessed the effectiveness of metformin on the separate ALDH+/− subpopulations. Using our FACS sorted populations, 72 h of metformin treatment decreased the proliferation of the ALDH- population, while causing little to no change in the ALDH+ fraction ([Figure 1](#ijms-20-00193-f001){ref-type="fig"}C). It is of note that only higher concentrations of metformin were able to produce a decrease in cellular proliferation in the non-stem cell ALDH− population.
Cisplatin is one of the most potent and commonly used chemotherapy drugs for treating head and neck cancer. However, our results indicate that metformin protects against the cytotoxic effects of cisplatin in both JLO-1 and the ALDH+ subpopulation of HN-30 ([Figure 1](#ijms-20-00193-f001){ref-type="fig"}D,E). At doses of 10 µM and 20 µM cisplatin, cell proliferation of JLO-1 decreased to 66% and 48% respectively ([Figure 1](#ijms-20-00193-f001){ref-type="fig"}E). However, even in combination with low doses of metformin, the cytotoxic effects of cisplatin were abrogated, and cell viability levels returned to values similar to the control ([Figure 1](#ijms-20-00193-f001){ref-type="fig"}E). A similar protection against cisplatin was not observed in the ALDH- population of HN30, and there was slight synergism between the two drugs at their respective highest concentrations ([Figure 1](#ijms-20-00193-f001){ref-type="fig"}D). Cisplatin causes cell death primarily through DNA crosslinking, so the effects of metformin on JLO-1 were further validated by measuring amounts of DNA strand breaks with a TUNEL assay. 20 µM of cisplatin causes 23.7% of the cells to undergo DNA strand breaks, while a combination treatment with 0.75 mM of metformin reduced the amount of DNA strand breaks to 10.5% ([Figure 1](#ijms-20-00193-f001){ref-type="fig"}F). To gain insight into a possible molecular mechanism behind the protective effect of metformin, we explored the well-studied cell survival pathway of Akt. Interestingly, our immunoblot showed a dose-dependent decrease of phosphorylated Akt in response to metformin treatment in JLO-1 cells, suggesting that metformin does not confer chemoprotective effects in CSCs through the Akt pathway ([Figure S1](#app1-ijms-20-00193){ref-type="app"}).
2.2. Metformin Increases Stem Cell Characteristics in HNSCC CSC Population {#sec2dot2-ijms-20-00193}
--------------------------------------------------------------------------
To gain a more thorough understanding of metformin's effects on HNSCC CSCs, we performed RT-qPCRs for JLO-1 cells to measure changes in expression of stem cell markers CD44, BMI-1, Oct-4, and Nanog after exposure to metformin. BMI-1 is a gene necessary for the stem cell property of self-renewal, and was shown to be differentially expressed in the CD44+ population in HNSCC \[[@B18-ijms-20-00193]\]. Oct-4 and Nanog are transcription factors that are required to maintain pluripotency in embryonic stem cells \[[@B21-ijms-20-00193]\]. Our results demonstrated an increase in all of the described stem cell genes, but most significantly in CD44 and BMI-1, where gene expression is increased up to 5 and 12-fold, respectively ([Figure 2](#ijms-20-00193-f002){ref-type="fig"}A). Interestingly, the strong expression increase is only observed at 0.75 mM of metformin applied, suggesting the existence of a possible threshold concentration for metformin to take effect. A verification of the increase in expression of CD44 was performed with immunofluorescence, in which the protein was observed to be highly localized to the cell surface ([Figure 2](#ijms-20-00193-f002){ref-type="fig"}B).
2.3. Computational Binding Analysis Reveals Strong Interaction of Metformin with Mitochondrial Complex III {#sec2dot3-ijms-20-00193}
----------------------------------------------------------------------------------------------------------
To elucidate the mechanism of metformin action that could explain its observed effects in HNSCC, we used the small molecule-to-protein docking program AutoDock Vina to explore metformin's binding interactions. Vina was demonstrated to be highly accurate in predicting position of binding and binding energies for small molecules that have low numbers of active rotatable bonds, such as metformin, which has two rotatable bonds \[[@B22-ijms-20-00193]\]. We observed that metformin binds next to the B~L~ heme, near the Q~o~ site, of complex III with a binding energy of −6.2 kJ, suggesting high stability of binding ([Figure 3](#ijms-20-00193-f003){ref-type="fig"}A--C). The binding site is predicted to be within cytochrome b and on the side of the B~L~ heme away from the Rieske Protein ([Figure 3](#ijms-20-00193-f003){ref-type="fig"}B). From a molecular surface visualization of the binding, it can be seen that metformin and the B~L~ heme are located within the same pocket in cytochrome b ([Figure 3](#ijms-20-00193-f003){ref-type="fig"}C).
To validate that metformin binds more strongly to complex III at the proposed site than to other binding sites and that AutoDock Vina is capable of predicting a diverse range of binding interactions, we explored the binding possibility of metformin with the top 250 proteins, and any proteins that they complex with, that are most associated with high stemness in HNSCC tissue samples. The fpocket program was used to identify possible sites for small molecule to bind to these proteins, and a total of 127,635 putative binding sites were screened using AutoDock Vina for binding with metformin. The binding of metformin to complex III near the B~L~ heme is within the top 20 most stable interactions we found ([Figure 3](#ijms-20-00193-f003){ref-type="fig"}D). The mode binding energy to metformin of the binding sites screened is −3.9 kJ, whereas metformin binds to complex III with a binding energy of −6.2 kJ ([Figure 3](#ijms-20-00193-f003){ref-type="fig"}D). The lower the binding energy is, the more stable the interaction. Our results thus demonstrated that metformin binds to complex III with enough exclusivity to justify further investigation of complex III as a major target of metformin.
2.4. Expressions of Complex III Genes Correlate with Clinical Variables and Stem Cell Marker Expressions {#sec2dot4-ijms-20-00193}
--------------------------------------------------------------------------------------------------------
Using gene expression data of HNSCC patient samples downloaded from The Cancer Genome Atlas (TCGA), we next correlated the expression of genes coding for complex III protein subunits to tumor histologic grade and patient survival to explore the clinical relevance of complex III activity. Interestingly, we discovered that high expressions of complex III genes correlated strongly with lower histologic grade but also lower patient survival, even though higher histologic grade suggests a more aggressive tumor and usually correlates with lower rate of survival ([Figure 4](#ijms-20-00193-f004){ref-type="fig"}A,B). Histologic grade is a measure of how well the tumors are differentiated, where a high grade indicates poor differentiation, and high histologic grade correlates strongly with the presence of CSCs \[[@B23-ijms-20-00193],[@B24-ijms-20-00193]\]. Therefore, we interpret our result to suggest that lower complex III activity is associated with higher CSC presence. Since we have demonstrated that metformin increases stemness, we hypothesize that metformin's interaction with complex III is inhibitory rather than activating. Following this hypothesis, we can further hypothesize that metformin's inhibition of complex III decreases the proliferation of non-stem cancer cells. Our data are supportive of these hypotheses because the duration for which a patient can survive is correlated more strongly with cancer proliferation than with CSC presence, which is more associated with treatment resistance and tumor recurrence. Therefore, suppression of complex III would lead to a better survival rate by decreasing cancer cell proliferation but also lead to higher histologic grade by increasing the stemness of CSCs, which explains the strong correlation of low expressions of complex III genes to better survival prognosis and higher histologic grade at the same time.
To validate our hypothesis that lower complex III activity is associated with higher CSC presence, we proceeded to explore the relationship between complex III and HNSCC stem cell markers. Using HN-30 cells, we knocked down expression of the Rieske protein subunit in complex III with siRNA and measured changes in the expression of HNSCC stem cell markers with RT-qPCR. We investigated eight of the most significant HNSCC stem cell markers described by Major et. al. \[[@B25-ijms-20-00193]\], including NANOG, ALDH1A1, BMI-1, CD44, LGR5, CD133 (PROM1), ABCG2, and OCT-4 (POU5F1) (for primer sequences, see [Table 1](#ijms-20-00193-t001){ref-type="table"}). We observed that the expressions of stem cell markers NANOG, CD133 (PROM1), ALDH1A1, and LGR5 are elevated by 1.77-, 2.49-, 2.93-, and 7.32-folds, respectively, after HN-30 cells are treated with the siRNA ([Figure 4](#ijms-20-00193-f004){ref-type="fig"}C). Using gene expression data of HNSCC patient samples downloaded from TCGA, we also found strong negative correlations of the expressions of complex III genes with the expressions of several stem cell markers. In particular, the expressions of CD44 and BMI-1, the aforementioned stem cell markers that are highly elevated by metformin, exhibit the most consistent negative correlation with complex III genes' expressions ([Figure 5](#ijms-20-00193-f005){ref-type="fig"}A,F). The expressions of stem cell markers or stemness markers c-Met (MET), SLC2A13, PDPN, and ALDH1A3 are also inversely correlated with the expressions of one or more complex III genes ([Figure 5](#ijms-20-00193-f005){ref-type="fig"}B--E) \[[@B25-ijms-20-00193]\]. Collectively, our in vitro and in silico results suggest that lowered complex III activity levels are correlated with higher stemness and CSC presence.
3. Discussion {#sec3-ijms-20-00193}
=============
Metformin gained attention as a promising potential anticancer therapy as some studies demonstrated a correlation between metformin use and decreased incidence of cancer, while other studies reported its ability to selectively target CSCs. To date, the CSC-inhibiting ability of metformin has been demonstrated in a variety of tumor types, including breast, pancreatic, lung, skin, and ovarian \[[@B3-ijms-20-00193],[@B4-ijms-20-00193],[@B7-ijms-20-00193],[@B26-ijms-20-00193]\]. However, to the best of our knowledge, this study is the first to test the effects of metformin on HNSCC stem cells. This study is also the first to demonstrate that metformin has negligible effects on the proliferation of a CSC population and even protects against cisplatin. In direct contrast to previous studies, our data suggests that metformin potentiates stem cell genes and self-renewal capabilities in our HNSCC stem cell line, JLO-1. Therefore, the effects of metformin are most likely highly dependent on the tumor cell type, so metformin may not be a viable option for targeting HNSCC stem cells. However, our data do suggest that metformin decreases the proliferation of non-stem HNSCC cells. Several studies have indicated that metformin treatment alone can decrease cancer proliferation using HNSCC cell lines, although each study describes a different mechanism of action, including AMPK-independent downregulation of the mTOR pathway or global inhibition of protein translation \[[@B27-ijms-20-00193],[@B28-ijms-20-00193]\]. These studies are consistent with our data, which indicate that the non-stem cell (ALDH-) fraction of HN-30 decreases in viability after treatment of metformin. Collectively, our results indicate that metformin may be a valuable drug against HNSCC, but only if another drug is used to mitigate its protective effects on HNSCC CSCs. Since metformin is much better tolerated by the body than traditional chemotherapy drugs, it is an attractive therapeutic option that can be used to reduce the amount of chemotherapy drugs needed for the same anti-tumor effects. However, since metformin's chemoprotection of CSCs will prevent complete elimination of the tumor and render treatment ineffective in the long term, we sought to determine the mechanism with which metformin acts on CSCs to explore the possibility of using a drug to mitigate this effect.
Through computational modelling of metformin's binding to proteins with the docking software AutoDock Vina, we discovered evidence of a strong binding interaction between metformin and complex III of the mitochondria. Complex III, also known as the cytochrome bc1 complex or coenzyme Q--cytochrome c reductase, is a complex within the electron transport chain of the mitochondria and is known as a major site of ROS production \[[@B10-ijms-20-00193],[@B29-ijms-20-00193]\]. It conducts the Q cycle, in which ubiquinol (QH~2~) is oxidized into ubiquinone (Q, or coenzyme Q). When QH~2~ enters the complex, it binds to the Q~o~ reactive site within the cytochrome b subunit of the complex, where two electrons are extracted from it. One would be transferred to the 2Fe/2S center located within the nearby Rieske protein, while the other would be transferred to the nearby B~L~ heme. The latter electron would flow from the B~L~ heme to the B~H~ heme then to a ubiquinone molecule within the complex, reducing it to the free radical ubisemiquinone, which has been reported to transfer the electron to oxygen, forming ROS \[[@B30-ijms-20-00193]\]. We discovered that metformin binds near the B~L~ heme, suggesting that it is potentially able to block the flow of electrons to ubisemiquinone, thereby preventing the formation of ROS. Indeed, complex III inhibitors that bind near the Q~o~ site, including myxothiazol and stigmatellin, have been demonstrated to reduce the amount of ROS generated by complex III \[[@B29-ijms-20-00193],[@B30-ijms-20-00193]\].
The results of this study could be well-explained under the supposition that metformin inhibits complex III and lowers ROS levels as result. Through qPCR assay of siRNA knocked-down cells and TCGA gene expression correlations, our results suggest that lowered complex III activity correlates with higher stem cell marker expressions and higher histologic grade. Therefore, if metformin inhibits complex III activity, it would be able to induce higher stem cell marker expressions, as was observed in our experiments. Additionally, it is previously known that metformin decreases the amount of ROS in cells, and low ROS levels are essential for the preservation of CSC self-renewal and other pro-survival properties \[[@B12-ijms-20-00193],[@B25-ijms-20-00193]\]. We thus hypothesize that metformin lowers ROS production in complex III to elevate the stemness of HNSCC CSCs and enable their maintenance after application of cisplatin. Moreover, we can also attribute the observed anti-proliferative effect of metformin on non-stem HNSCC cells to metformin's ability to decrease ROS levels, since elevated ROS levels in cancer cells can drive cancer progression through the activation of HIF-1 \[[@B31-ijms-20-00193]\].
Interestingly, the current literature has not reported that metformin targets complex III. Instead, a well-known potential target of metformin is mitochondrial complex I, a complex upstream of complex III in the electron transport chain that generates the QH~2~ consumed by complex III \[[@B6-ijms-20-00193]\]. However, many outstanding questions and concerns were raised by this hypothesis of complex I inhibition. We thus propose that complex III may be a more direct target of metformin and that this hypothesis could resolve some of the theoretical issues surrounding complex I inhibition by metformin. Many studies have characterized metformin's inhibition of complex I, but few have provided direct evidence of complex I inhibition due to the difficulties involved in isolating complex I from the mitochondrial membranes \[[@B9-ijms-20-00193]\]. Some labs even claimed failure to directly observe complex I inhibition \[[@B9-ijms-20-00193]\]. Furthermore, no molecular binding site of metformin to complex I has been proposed. The greatest controversy in the theory of complex I inhibition is the fact that extremely high levels of metformin, 1000-fold higher than serum concentration, are needed to induce complex I inhibition \[[@B6-ijms-20-00193]\]. It was proposed that this concentration can be achieved within the mitochondria when driven by the mitochondrial membrane potential, but this hypothesis has not been tested \[[@B32-ijms-20-00193]\]. A recent study that performed whole-body PET scan of radioactive-carbon labeled metformin demonstrated that no such concentration of metformin was reached in head and neck tissues \[[@B33-ijms-20-00193]\]. Additionally, we observed the chemoprotective effects of metformin against CSCs and increased expressions of stem cell genes when only 0.5--0.75 mM of metformin was applied. This is many times lower than the IC50 value (concentration of drug where target activity is inhibited by half) of 19.4 mM reported by one study for metformin inhibition of complex I \[[@B32-ijms-20-00193]\]. According to the dosage-response curve reported, metformin concentration of 0.75 mM would not lead to any inhibition of complex I \[[@B32-ijms-20-00193]\]. Therefore, it is likely that metformin acts on some target other than complex I, such as complex III. One study observed that metformin, unlike the classical complex I inhibitor rotenone, does not lead to an increase in ROS production due to forward electron flux through complex I but only decreases ROS production due to reverse electron flux \[[@B11-ijms-20-00193]\]. The fact that a complex I inhibitor would increase ROS production from forward electron flux and decrease ROS production from reverse electron flux at the same time is more intuitive because when complex I is inhibited, the electron that normally travels through complex I to reduce Q is blocked and would leak out of the complex to react with oxygen, forming ROS. Because the reaction of complex I never occurred, it would be less likely that the reverse reaction would occur where the electron travels backwards to reduce the substrate of complex I. Therefore, the ROS production associated with reverse electron flux would decrease. The fact that metformin only causes a decrease in reverse electron flux may suggest that a downstream target, such as complex III, was inhibited instead of complex I. Finally, it was also reported that biguanide drugs only inhibit complex I when mitochondria are in state 3 (active respiration) but not when they are in state 4 (termination of active respiration), which also hints at the possibility of a target downstream of complex I \[[@B34-ijms-20-00193]\].
Besides exploring the specific molecular mechanism, we have also investigated the effects of metformin on important pathways that are known to significantly influence the CSC phenotype. We demonstrated that metformin induces an increase in the expression of stem cell genes, most notably for CD44 and BMI-1. It is well accepted that BMI-1 is necessary for self-renewal in both cancer and normal stem cells, and CD44 is also one of the most well-known CSC biomarkers \[[@B25-ijms-20-00193]\]. We next explored the effects of metformin on Akt in JLO-1. The PI3K/Akt signaling pathway is frequently dysregulated in cancers, and Akt activation has been shown to contribute to chemotherapeutic resistance \[[@B26-ijms-20-00193]\]. Metformin has been reported to protect glioma cell lines against cisplatin via activation of Akt \[[@B5-ijms-20-00193]\]. However, our results indicate that metformin causes a decrease in Akt levels in CSCs, which implies that the chemo-protective actions of metformin are not conferred through Akt.
Interestingly, we did not observe an increase in tumorsphere formation after application of metformin to JLO-1 (data not shown), most likely because the self-renewal capabilities of the JLO-1 CSCs have already reached the maximum threshold. However, metformin rescues the decrease in cell proliferation caused by cisplatin, suggesting that metformin does not observably affect the activities of normal CSCs and only prevents these cells from being damaged by cisplatin. This observation suggests that cisplatin must act on CSCs in a manner that can be reversed by metformin. One attractive mechanism consistent with our hypotheses would be that cisplatin causes the generation of ROS, which has been reported in some studies, that would cause the CSCs to lose self-renewal capabilities \[[@B35-ijms-20-00193],[@B36-ijms-20-00193]\]. Metformin could then reverse this effect of cisplatin by lowering ROS levels.
In summary, our results suggested that metformin results in the chemoprotection of HNSCC stem cells but decreases the ability of non-stem cancer cells to proliferate. However, we emphasize that our results were purely derived from in vitro and in silico assays and that in vivo experiments are needed to further validate applicability of our results. We propose the novel mechanism that mitochondrial complex III inhibition by metformin causes a reduction in the ROS levels of these cells to yield these observed effects, as low ROS levels are required for stemness, but high ROS levels are drivers of tumor progression. While extensive in vitro binding analyses and functional assays are needed to validate whether complex III is a target of metformin and whether metformin binding to complex III reduces ROS levels, our proposed mechanism raises exciting possibilities for treating HNSCC using metformin in combination with another drug that could mitigates its effects on HNSCC stem cells. One such drug combination can be a molecule that inhibits complex I in CSCs, as it was reported that inhibition of both complex I and complex III of the mitochondria would lead complex II to produce a large amount of ROS \[[@B37-ijms-20-00193]\], which could debilitate HNSCC CSCs.
4. Materials and Methods {#sec4-ijms-20-00193}
========================
4.1. Cell Lines and Cultures {#sec4dot1-ijms-20-00193}
----------------------------
The JLO-1 cell line was derived from a fresh laryngeal tumor of a patient undergoing tumor resection. A stem cell selective cultivation condition was used to generate JLO-1, as described in our previous study \[[@B38-ijms-20-00193]\]. Briefly, flow cytometry was performed to select for CD44+ cells, which were then grown on laminin-coated plates and cultured in keratinocyte serum-free media (Invitrogen, Carlsbad, CA, USA) containing 2 mM [l]{.smallcaps}-glutamine (Invitrogen), 50 μg/mL gentamycin (Invitrogen), and 20 ng/mL EGF and FGF (R&D Systems, Minneapolis, MN, USA), supplemented daily. We also used the HN-30 cell line, a gift from Dr. J.S. Gutkind, University of California San Diego. Cell lines were routinely cultured in DMEM supplemented with 10% fetal bovine serum (FBS), 2% streptomycin sulfate (Invitrogen), and 2% [l]{.smallcaps}-glutamine (Invitrogen) and incubated at 37 °C in 5% CO~2~ and 21% O~2~.
4.2. FACS Identification of ALDH+ and ALDH- Cell Populations {#sec4dot2-ijms-20-00193}
------------------------------------------------------------
HN-30 cells were stained with the Aldefluor stem cell detection kit (STEMCELL technologies, Vancouver, BC, Canada), which will lead to fluorescence of cells with high ALDH activity. The ALDH-bright cells were sorted from ALDH-dim cells using a fluorescence-activated flow cytometer.
4.3. Cell Proliferation Assay {#sec4dot3-ijms-20-00193}
-----------------------------
MTS assays were performed using the CellTiter 96 Aqueous non-radioactive cell proliferation assay (Promega, Madison, WI, USA). Cells were trypsinized, counted, and replated into a 96-well plate at 5000 cells per well. Cells were allowed to adhere overnight. To generate a dose--response curve for cell proliferation vs. metformin concentration, indicated doses of metformin were added to the corresponding wells for an incubation period of 72 h. For synergistic assays involving the combination of cisplatin and metformin, HN-30 cells were treated with 8 or 12 mM metformin for 48 h, followed by co-treatment with cisplatin at a range of doses (1, 2, 5, 10, 20 μM) for an additional 48 h; while JLO-1 cells were treated with 0.5 or 0.7 mM of metformin for 48 h before co-treatment with cisplatin. Each permutation was performed in triplicates. Following the indicated incubation periods for the above assays, 20 μL of the MTS reagent was added into each well followed by a 1--3 h incubation period. The plates were then read at an absorbance of 490 nm.
4.4. TUNEL Assay {#sec4dot4-ijms-20-00193}
----------------
JLO-1 cells were treated with metformin 4 days prior to fixing in 70% ethanol. Media and growth factors were not replenished throughout the treatment. Using the APO-BRDUTMKit (Phoenix Flow Systems, Inc., San Diego, CA, USA), the cells undergoing apoptosis were labeled with bromolated deoxyuridine triphosphate nucleotides (BrdUTP). These cells were then identified and binded to a fluorescein labeled antiBrdU monoclonal antibody. After the required incubation times, the samples analyzed for the proportion of apoptotic cells by flow cytometry.
4.5. Western Blot Analysis {#sec4dot5-ijms-20-00193}
--------------------------
JLO-1 cells were harvested and lysed with lysis buffer containing 20 mM Tris (pH 7.5), 150 mM NaCl, 1 mM ethylenediaminetetraacetic acid (EDTA), 1 mM egtazic acid (EGTA), 1% Triton-x, 2.5 mM sodium pyrophosphate, 1 mM β-glycerophosphate, 1 mM Na~3~VO~4~, and 1 μg/mL leupeptin. Cell lysates were separated on 12% NuPAGE^®^ Novex Bis-Tris Gels (Invitrogen, Carlsbad, CA, USA) and transferred electrophoretically to a PVDF membrane (Immobilon-P membrane, 0.45 μm; Millipore, MA, USA). The membrane was blocked in 5% milk and probed with antibodies for phosphorylated-Akt (p-Akt) (Cell Signaling, Beverly, MA, USA), followed by a secondary antibody. Membranes were visualized with chemiluminescence detection system (Pierce, Rockford, IL, USA). The membranes were probed with antibody against Erk (Abcam, Cambridge, MA, USA) to ensure equal protein loading.
4.6. Quantitative Real-Time PCR and siRNA Knockdown {#sec4dot6-ijms-20-00193}
---------------------------------------------------
The cultured cells were treated with metformin (0--0.75 mM) for 48 h. Total cell lysate was collected and mRNA was extracted using the RNeasy kit (QIAGEN, Venlo, The Netherlands). cDNA was then synthesized from 1.5 μg of total mRNA using reverse transcriptase (Invitrogen, Carlsbad, CA, USA), as per the manufacturer's instructions. Real-time quantitative PCR was performed by combining 2.5 μL of the RT with 22.5 μL of SYBR green (Roche, Basel, Switzerland). The reaction was run using System 7300 (Applied Biosystems, Foster City, CA, USA) and results were analyzed by the relative quantity method. Experiments were performed in triplicates with GAPDH expression as the endogenous control. siRNA for the Rieske protein (UQCRFS1) was obtained from Dharmacon, Lafayette, CO, USA. Primers were custom designed by the authors and created by Eurofin Genomics, Louisville, KY, USA. The following sequences were used:
4.7. Immunofluorescence {#sec4dot7-ijms-20-00193}
-----------------------
HN-30 cells were cultured on cover slips under 0.75 mM of metformin. The cells were fixed with 4% paraformaldehyde and blocked in goat serum in Dulbecco's phosphate buffered saline at room temperature prior to incubation with mouse monoclonal to anti-human CD44 (Sigma Aldrich, St. Louis, MO, USA). Cells were then incubated with a goat anti-mouse FITC conjugated secondary antibody (Chemicon, Temecula, CA, USA) and counterstained with DAPI. Finally, SlowFade Gold antifade reagent (Invitrogen, Carlsbad, CA, USA) was used to mount the cover slips onto slides. Fluorescent images were obtained at 40× using the DMIRE2 inverted fluorescence microscope (Leica Microsystems, Buffalo Grove, IL, USA) and computer program Simple PCI (version 6.6, Hamamatsu Photonics, Sewickley, PA, USA) was used for image capture.
4.8. Computational Prediction of Metformin Binding Energy {#sec4dot8-ijms-20-00193}
---------------------------------------------------------
The crystallographic protein structure of mitochondrial complex III was downloaded from the Protein Data Bank (PDB) ([www.rcsb.org](www.rcsb.org)) under the ID 5OKD, which was contributed by the study of Amporndanai et al. \[[@B39-ijms-20-00193]\]. Metformin molecular structure was downloaded from the ZINC database (<http://zinc15.docking.org/>). fpocket (version 2.0, University of Paris-Diderot, Paris, France) was used to determine potential binding pockets of complex III, and AutoDock Vina (version 1.1.2, The Scripps Research Institute, San Diego, CA, USA) was used to uncover the position of metformin binding with each pocket that would result in the lowest (most favorable) binding energy \[[@B22-ijms-20-00193],[@B40-ijms-20-00193]\]. The position of binding was then visualized with UCSF chimera (version 1.12, San Francisco, CA, USA) \[[@B41-ijms-20-00193]\]. The favorability of metformin binding was assessed for 250 other proteins most associated with high stemness to determine whether the binding strength of metformin to complex III is relatively large compared to the binding strength of metformin to other binding pockets. To identify the 250 most abundant proteins in tumors with high stemness, we obtained the stemness scores of HNSCC tumors analyzed by TCGA from Malta et. al. \[[@B42-ijms-20-00193]\]. Normalized RNA-seq gene expressions for each HNSCC patient were downloaded from the GDC portal (<https://portal.gdc.cancer.gov/>) and correlated with these stemness scores using Gene Set Enrichment Analysis (GSEA) (version 3.0, Broad Institute, Inc., Cambridge, MA, USA) \[[@B43-ijms-20-00193]\]. GSEA generated a ranked gene-list that orders genes based on the degree to which expression correlates with stemness scores. The top 250 genes with positive correlation of expression to stemness were chosen, and any available crystallographic structures of proteins and protein complexes associated with these genes were downloaded from PDB. These structures were analyzed with fpocket, which predicted 127,635 binding pockets in total. AutoDock Vina virtual screening was used to determine the lowest possible energy of metformin binding with each pocket. Virtual screening was performed with the Comet supercomputer in the San Diego Supercomputer Center, with access provided through an allocation from the Extreme Science and Engineering Discovery Environment (XSEDE) \[[@B44-ijms-20-00193]\].
4.9. Correlation of Complex III Gene Expressions to Survival and Histologic Grade {#sec4dot9-ijms-20-00193}
---------------------------------------------------------------------------------
TCGA mRNA expression read counts downloaded for HNSCC samples were paired with the clinical data of corresponding patients, which include histologic grade and time to death or last follow up. The clinical data were downloaded from the Broad Institute GDAC Firehose (<https://gdac.broadinstitute.org/>). Survival analyses were performed using the Kaplan-Meier Model, with gene expression designated as a binary variable based on expression above or below the median expression of all samples. Univariate Cox regression analysis was used to identify candidates significantly associated with patient survival (*p* \< 0.05). The Kruskal-Wallis test was used to correlate gene expressions to histologic grade (*p* \< 0.05). The 11 human genes that encode complex III subunits were included in analysis: MT-CYB, CYC1, UQCRFS1, UQCRC1, UQCRC2, UQCRH, UQCRB, UQCRQ, UQCR9, UQCR10, and UQCR11.
4.10. Correlation of Complex III Gene Expressions to Expressions of Stem Cell Markers Using TCGA Data {#sec4dot10-ijms-20-00193}
-----------------------------------------------------------------------------------------------------
The expressions of the complex III genes listed above were correlated with the gene expressions of stem cell markers, including CD44, BMI-1, MET, SLC2A13, PDPN, ALDHA3, NANOG, OCT4, etc. The same expression data described in the above section are used. Correlations were assessed using Spearman's correlation test (*p* \< 0.05) and visualized as scatter plots.
This work used the Extreme Science and Engineering Discovery Environment (XSEDE) resource Comet at the San Diego Supercomputer Center (SDSC) through allocation TG-BIO170107.
Supplementary materials can be found at <http://www.mdpi.com/1422-0067/20/1/193/s1>. Figure S1: Western blot assay of phosphorylated-Akt levels after JLO-1 cells were treated with 0--0.75 mM of metformin. Erk level was probed as loading control.
######
Click here for additional data file.
Conceptualization, W.M.O.; methodology, W.M.O., W.T.L.; formal analysis, W.T.L., E.A.; investigation, S.Z.K., C.O.H., W.T.L., T.K.H., E.A.; resources, J.W.-R., W.M.O.; writing---original draft preparation, S.Z.K., E.K., W.T.L.; writing---review and editing, W.M.O., X.A., J.W.-R.; visualization, S.Z.K., W.T.L.; supervision, W.M.O.
This research received no external funding.
The authors declare no conflict of interest.
CSCs
Cancer stem cells
HNSCC
Head and neck squamous cell carcinoma
ROS
Reactive oxygen species
{#ijms-20-00193-f001}
{#ijms-20-00193-f002}
{#ijms-20-00193-f003}
{#ijms-20-00193-f004}
{#ijms-20-00193-f005}
ijms-20-00193-t001_Table 1
######
Primer sequences used for quantitative PCR.
--------- ---------- -------------------------------
CD44 forward: 5′-AGAAGAAAGCCAGTGCGTCT-3′
CD44 reverse: 5′-TGACCTAAGACGGAGGGAGG-3′
GAPDH forward: 5′-TTCTTTTGCGTCGCCAGCC-3′
GAPDH reverse: 5′-CGTTCTCAGCCTTGACGGTG-3′
BMI1 forward: 5′-CGAGACAATGGGGATGTGGG-3′
BMI1 reverse: 5′-AAATGAATGCGAGCCAAGCG-3′
ALDH1A1 forward: 5′-CACGCCAGACTTACCTGTCC-3′
ALDH1A1 reverse: 5′-TTGTACGGCCCTGGATCTTG-3′
NANOG forward: 5′-AATGGTGTGACGCAGGGATG-3′
NANOG reverse: 5′-ACCTCGCTGATTAGGCTCCA-3′
POU5F1 forward: 5′-TCCCGAATGGAAAGGGGAGA-3′
POU5F1 reverse: 5′-GGCTGAATACCTTCCCAAATAGA-3′
ABCG2 forward: 5′-TTACGCACAGAGCAAAGCCA-3′
ABCG2 reverse: 5′-GCAAGGGGCTAGAAGAAGGG-3′
PROM1 forward: 5′-GAATCCTTTCCATTACGGCGG-3′
PROM1 reverse: 5′-CCTGAAAAGGAGTTCCCGCA-3′
LGR5 forward: 5′-GGAGTTACGTCTTGCGGGAA-3′
LGR5 reverse: 5′-CAGGCCACTGAAACAGCTTG-3′.
--------- ---------- -------------------------------
[^1]: These authors contributed equally to this work.
| {
"pile_set_name": "PubMed Central"
} |
Introduction {#Sec1}
============
NOTCH proteins are trans-membrane receptors that transduce signals from cell-bound JAGGED (JAG) and Delta-like (DLL) families of ligands to mediate cell-cell interactions in processes as diverse as fetal development, heart disease and cancer^[@CR1],[@CR2]^. Upon interaction with ligand, NOTCH is cleaved by γ-secretase to release an intracellular domain (NICD) that binds the transcriptional effector RBPJ \[recombination signal-binding protein for immunoglobulin κ J region, also known as CSL and CBF1\]^[@CR3],[@CR4]^. There is only one small molecule reported to selectively inhibit NOTCH signaling and none known to target RBPJ^[@CR5]^. γ-secretase inhibitors (GSIs) have been used widely to block the proteolytic activation of NOTCH, but are inherently unselective since they also block the processing of \>90 other substrates, including the amyloid precursor protein (APP), ErbB4, and E-cadherin^[@CR6]--[@CR8]^. Although newer generation GSIs exhibit a biased inhibition of APP over NOTCH^[@CR9],[@CR10]^, there are no NOTCH-selective GSIs. Clinical use of GSIs cause numerous side effects, notably intestinal crypt cell proliferation^[@CR11]^, and skin rashes and tumors^[@CR12]^. These undesirable effects have led to the early termination of a phase III clinical trial of Semagacestat for treatment of Alzheimer's Disease^[@CR13]^.
The NOTCH ICD:RBPJ complex contains co-activators MAML (mastermind-like protein) and histone acetyltransferases (HATs) to activate downstream genes including members of the structurally related HES (Hairy/Enhancer of Split) and HEY/HRT (Hairy/Enhancer of Split-related) family^[@CR14]^. The transcriptional effects of NICD are mediated primarily through the interaction with RBPJ, which recruits NICD to recognition sites in promoter regions as well as to more distally located superenhancers^[@CR15]^. Thus, antagonism of RBPJ would be a desirable point to modulate NOTCH signaling, making it a useful probe and potential clinical candidate since it could provide additional selectivity over targeting γ-secretases or NOTCH itself. To develop a novel chemical inhibitor of NOTCH, we evaluated the feasibility of selectively targeting the RBPJ protein to perturb its interaction with the NOTCH ICD.
Although protein-protein interactions can be challenging to inhibit with small molecules, the binding interface between RBPJ and NOTCH is small^[@CR16],[@CR17]^ hence we reasoned that it might be possible to disrupt it with a small molecule. To increase the likelihood of selectively targeting RBPJ (as opposed to NOTCH directly, or mediators of its expression, trafficking or cleavage), we developed a primary high throughput screen (HTS) aimed at disrupting the interaction between RBPJ and an unrelated scaffold protein, SHARP, that binds to the same region of RBPJ to mediate its repressive effects^[@CR18]^. Hits from this screen were triaged by secondary screens that identified inhibitors of RBPJ in the context of NOTCH ICD. One molecule, termed **[R]{.ul}**BPJ **[IN]{.ul}**hibitor-1 (**RIN1**), potently disrupted the functional interaction of RBPJ with NOTCH and functioned as a probe of RBPJ function in cancer and skeletal myogenesis models. **RIN1** is the first inhibitor of RBPJ and can be exploited for research and possible therapeutic applications.
Results {#Sec2}
=======
Identification of an RBPJ inhibitor {#Sec3}
-----------------------------------
To identify selective inhibitors of RBPJ, we developed a primary screen to detect inhibitors of a functional interaction between RBPJ and the scaffold protein SHARP that was followed by secondary assays to establish efficacy against NOTCH. The primary screen was adapted from a cell-based 2-hybrid assay that probed the interaction of RBPJ with a minimal RBPJ-binding fragment of SHARP (amino acid residues 2770--3127)^[@CR18]^ (Fig. [1a](#Fig1){ref-type="fig"}, upper schematic). To minimize unwanted genomic effects of RBPJ, we used a mutated form of RBPJ that cannot bind to DNA (see Supplemental Methods). A functional interaction between the two proteins activated the UAS-Luciferase reporter gene in stably transfected AD-293 cells. We used siRNA against RBPJ to determine that the assay had an acceptable dynamic range (Z' = 0.84) (Fig. [1b](#Fig1){ref-type="fig"}). Z' scores between 0.5 and 1.0 are considered excellent^[@CR19]^.Figure 1Identification of small molecule RBPJ inhibitor, RIN1. (**a**) Schematic of the primary and counter screens. The screen was a cell-based two hybrid assay in which an active compound (stars) would disrupt the SHARP:RBPJ interaction and decrease activity of the Luciferase reporter. A minimal RBPJ-interacting domain of SHARP and a DNA-binding mutant of RBPJ were used (see Methods). (**b**) Assay validation using RBPJ siRNA transfection and a small molecule Luciferase inhibitor, Data is mean ± standard deviation, n = 5 wells. Z' is a metric of dynamic range^[@CR19]^. (**c**) Workflow schematic. (**d**) Screen flowchart and structure of **RIN1**. (**e**,**f**) Inhibition of NOTCH2 ICD (**e**) and RBPJ-VP16myc fusion protein (**f**) activity on the *Hes1-*Luciferase reporter in transient transfections, n = 4 wells. (**g**) Effect on cell viability, n = 4 wells. (**h**) Effect on CMV promoter activity, n = 4 wells. Data in b-h are presented as mean ± standard deviation; experiments were repeated \> 3 times. (**i**--**k**) Effect of Cycloheximide on **RIN1** inhibition of RBPJ-VP16. AD-293 cells were transfected with RBPJ-VP16myc and 48 hours later were treated ± **RIN1** (2 µM), ± Cycloheximide (CHX, 10 µg/ml) for an additional 17 hours and then assayed for *Hes1*-Luciferase activity (**i**) (n = 10 wells), *Luciferase* mRNA (**j**) and endogenous *HES5* mRNA (**k**), (n = 3 samples). Western blot of RBPJ-VP16myc fusion protein under identical conditions (**l**) and its quantification (**m**), n = 3 wells. Data are presented as mean ± standard deviation. Incrementing number of symbols (\* and \#) denote P \< 0.05, P \< 0.01, P \< 0.001 and P \< 0.0001 respectively, using two-tailed unpaired Student's T-test relative to empty vector (\*) or to RBPJ-VP16myc + DMSO vehicle control (\#) conditions. Experiments were repeated twice.
The primary assay was screened in 1536-well format against 1,780,000 compounds from the Sanofi Tucson combinatorial collection and the Sanofi Screening Collection (SASC1) at a single concentration (10 µM) (Fig. [1c](#Fig1){ref-type="fig"} and Supplementary Table [1](#MOESM1){ref-type="media"}). Using a single hit compound from a pilot screening at the Prebys Center for Drug Discovery (La Jolla, California, USA), we determined that the mean Z' for the entire screen was 0.64 ± 0.05. The hit selection cutoff was set at 4 standard deviations of the mean of all control wells across the entire screen, which corresponded to a threshold of ≥40% inhibition of UAS-Luc activity and yielded 18,047 compounds. Some of these were either duplicates or unavailable for retesting, therefore 17,086 primary positives were retested through a dose range to confirm activity (Fig. [1d](#Fig1){ref-type="fig"}). To rule out inhibition of the Luciferase reporter or other non-specific effects on the cell assay system, hits were also tested in a counter screen that was identical to the primary assay except that a fusion protein consisting of the Gal4 DNA binding domain covalently linked to the VP16 activation domain was used instead of the separate Gal4-SHARP and RBPJ-VP16 fusion proteins in the primary assay (Fig. [1a](#Fig1){ref-type="fig"} bottom schematic). Of the primary assay hits, 530 compounds confirmed activity (≥50% inhibition at 10 µM or lower concentration) in the confirmatory assay *and* were considered inactive (≤25% inhibition) in the counter screen. Of these compounds, 130 showed dose responsive inhibition comprising 14 distinct chemical families plus 17 singletons.
To distinguish compounds that inhibit RBPJ (as opposed to SHARP), we designed a secondary assay to test whether the molecules would block the function of RBPJ in the context of activated NOTCH signaling. In this assay (Fig. [1e](#Fig1){ref-type="fig"}), NOTCH2 ICD was transiently expressed in AD-293 cells harboring a *Hes1-*Luciferase reporter construct. One compound, **RIN1**, inhibited *Hes1*-Luciferase activity with an IC~50~ of 0.18 µM and E~max~ of 82% (Fig. [1e](#Fig1){ref-type="fig"}). **RIN1** also inhibited NOTCH3 ICD with similar potency and efficacy (0.19 µM and E~max~ = 88%). We tested whether **RIN1** would inhibit a RBPJ-VP16 fusion protein that, because of the Herpes Simplex Virus VP16 transactivation domain, induces transcription independently of NICD and co-activators. Again, **RIN1** inhibited RBPJ-VP16-dependent *Hes1-*Luciferase with the same potency and efficacy (IC~50~ = 0.20 µM and E~max~ = 81% inhibition; Fig. [1f](#Fig1){ref-type="fig"}). In the same experiment, **RIN1** had only a minimal effect on cell viability (Resazurin) (12% inhibition at 10 µM, Fig. [1g](#Fig1){ref-type="fig"}) or on the CMV promoter (22% reduction at 10 µM, Fig. [1h](#Fig1){ref-type="fig"}) that was used to direct transgene expression in the primary and secondary assays. Thus, these data suggest that **RIN1** is a potent inhibitor of RBPJ in the contexts of NICD and SHARP signaling.
Next, we examined if **RIN1** would inhibit pre-existing RBPJ or if it must be present during synthesis for inhibition to occur. If **RIN1** were required during RBPJ synthesis, its inhibitory activity should be abrogated by treatment with cycloheximide, which blocks translation of nascent proteins. **RIN1** (2 µM) decreased RBPJ-VP16-dependent *Hes1-*Luciferase activity and caused a corresponding decrease in *Luciferase* reporter mRNA and endogenous RBPJ target gene expression (Fig. [1i--k](#Fig1){ref-type="fig"}) mRNAs. Cycloheximide did not abrogate the ability of **RIN1** to inhibit either *Luciferase* or endogenous gene expression (Fig. [1j,k](#Fig1){ref-type="fig"}). Furthermore, **RIN1** did not alter the abundance of RBPJ-VP16 protein (Fig. [1l,m](#Fig1){ref-type="fig"}) under these conditions. Similarly, when new synthesis was blocked by siRNA silencing in cells, RIN1 did not alter the decay kinetics of endogenous RBPJ (Fig. [S1](#MOESM1){ref-type="media"}). These data suggesting that RIN1 does not alter the turnover of RBPJ. Instead, the ability to inhibit in the presence of cycloheximide suggests that **RIN1** disrupts the function of pre-synthesized RBPJ.
RIN1 inhibits NOTCH-dependent tumor cell proliferation {#Sec4}
------------------------------------------------------
NOTCH plays a role in the carcinogenesis and tumor progression, including leukemia, breast and lung cancers^[@CR2]^. As a bioassay of NOTCH inhibition, we assayed the effect of small molecule NOTCH inhibitors on the proliferation of two cell lines immortalized from T-cell acute lymphoblastic leukemia (T-ALL) patients (Jurkat and KOPT-K1) and in the mantle cell lymphoma (MCL) line REC-1, all of which have activating mutations in the NOTCH1 heterodimerization and/or PEST domain common in hematologic malignancies^[@CR20],[@CR21]^. **RIN1**, the γ-secretase inhibitor DAPT, and CB-103, a recently described small molecule NOTCH inhibitor with no reported mechanism^[@CR22]^, all decreased cell proliferation in the three cancer cell lines but with markedly different potencies and efficacies (Fig. [2a--c](#Fig2){ref-type="fig"}). The tumor lines had comparable levels of RBPJ protein (Fig. [2d](#Fig2){ref-type="fig"}); therefore, the varying effects of **RIN1** on tumor cell proliferation might reflect differential reliance on RBPJ-dependent versus RBPJ-independent NOTCH signaling^[@CR23]^.Figure 2Comparative effects of RIN1, DAPT and CB-103 on hematologic tumor cell proliferation. (**a**--**c**) Acute T cell leukemia cell lines Jurkat (**a**) and KOPT-K1 (**b**) and non-Hodgkin's mantle cell lymphoma Rec-1 line (**c**) were treated with small molecule NOTCH inhibitors during their logarithmic growth phase as indicated for 96 hours. n = 7, assay repeated twice. (**d**) Western blot showing levels of RBPJ in the tumor cell lines.
RIN1 promotes muscle differentiation {#Sec5}
------------------------------------
As a second bioassay of NOTCH inhibition, we tested the function of the small molecule inhibitors on the differentiation of muscle progenitor cells into mature myofibers, which is blocked by endogenous NOTCH activation^[@CR24]^. C2C12 myoblasts were induced to differentiate by passaging from low density, high serum culture conditions into high density, low serum conditions causing fusion of the myoblasts into multinucleated myofibers that expressed structural muscle proteins such as α-actinin and myosin heavy chain (MHC). Fusion was quantified by image analysis measuring the number of MHC^+^ cells and the number of nuclei per cell (Fig. [3](#Fig3){ref-type="fig"}). Relative to treatment with DMSO vehicle alone, **RIN1** (0.6 µM, corresponding to 3 × IC~50~) decreased the number of MHC^+^ cells and increased the number of nuclei per cell, indicating that it induced the formation of multinucleated myofibers. DAPT (0.6 µM, \~3 × IC~50~) was less potent in this assay, and CB-103 (0.6 or 2.5 µM) did not affect the formation of myofibers. Thus, **RIN1** was active in both the cancer cell anti-proliferation and myoblast differentiation assays.Figure 3Effect of RIN1 on C2C12 myoblast differentiation. Structured illumination photomicrographs of C2C12 cells at 4 days under permissive differentiation conditions and drug treatment as indicated. Upper panels: Cells were stained for myosin heavy chain with the MF20 antibody (green) and labeled with DAPI to identify nuclei (blue) (upper panels). Lower panels: Cell body and nuclei image masks for quantification. (**b**,**c**) Image analysis, n = 3 wells, quantified the number of cell mask (green) objects (**b**) and the ratio of nuclei per cell mask object (**c**). Assay repeated 3 times.
RIN1 treatment resembles RBPJ silencing {#Sec6}
---------------------------------------
We expected that inhibition of NOTCH signaling at the level of RBPJ would have a different effect on gene expression than would disruption at the level of γ-secretase since RBPJ can both activate and repress NOTCH target genes^[@CR25]^ and γ-secretase cleaves numerous proteins in addition to NOTCH^[@CR6]--[@CR8]^. Therefore, we compared the transcriptomic effects of treating Jurkat cells with **RIN1** (2 µM) to treatment with DAPT (2 µM), CB-103 (10 µM) (all compounds were at \~10x IC~50~) and siRNA against *RBPJ*, which reduced Jurkat cell proliferation consistent with the effect of **RIN1** (Fig. [S2](#MOESM1){ref-type="media"}). To compare gene expression profiles, we analyzed transcripts that did not vary (fold change \<1.4, p \< 0.05) between negative control groups (DMSO vehicle for the small molecules versus control siRNA for siRNA against RBPJ) to remove from analysis any genes that were influenced by the treatment differences *per se* (siRNA versus small molecule). Transcripts that were differentially expressed (fold change ≥ 2, p \< 0.05) are shown in the heatmap (Fig. [4a](#Fig4){ref-type="fig"}**)** and representative examples were confirmed by qRT-PCR (Fig. [4b--e](#Fig4){ref-type="fig"}). DAPT repressed known target genes as expected, including *HES1*, *HEY1* and *DTX*. CB-103 resembled DAPT suggesting that it functions at the level of inhibiting NOTCH. In contrast, **RIN1** resembled siRNA silencing of RBPJ, albeit often with a greater response, suggesting that it acts at the level of RBPJ. The differential response of NOTCH target genes to **RIN1** and RBPJ siRNA compared to DAPT and CB-103 suggests that NOTCH target gene expression in Jurkat cells is simultaneously sustained by RBPJ-independent signaling and repressed by RBPJ. Figure [4f](#Fig4){ref-type="fig"} positions **RIN1** in the context of NOTCH signaling and in relationship to other NOTCH pathway modulators.Figure 4Gene expression changes induced by RIN1. Jurkat cells were treated with either 8 hours with small molecules or 48 hours with transfected *RBPJ* siRNA and DMSO-vehicle or control (inert sequence) siRNA, respectively. The heatmap represents changes in the levels of transcript (fold change \> 2, P \< 0.05) that were induced by either the small molecules or RBPJ siRNA relative to their respective controls (DMSO vehicle or control siRNA) and varied \< 40% between techniques (DMSO-vehicle vs. control siRNA. (**b**--**e**) qRT-PCR analysis of RBPJ target gene expression. Data are presented as mean ± standard deviation. Incrementing symbols (\* and †) denote P \< 0.05, P \< 0.01, P \< 0.001 and P \< 0.0001 respectively, using two-way ANOVA with Dunnett's post-test. \*between groups indicated; ^†^relative to respective control treatments (DMSO or control siRNA). ns, not significant. (**f**) Schematic showing NOTCH pathway inhibitors (red text) in relationship to signaling and summary of RIN1 effects on downstream gene expression.
Discussion {#Sec7}
==========
Here we describe **RIN1** as the first small molecule inhibitor of RBPJ signaling. RBPJ can either activate genes by forming a complex with the NOTCH ICD when NOTCH is active, or silence an overlapping but non-identical set of genes by recruiting co-repressors in the absence of NOTCH signaling^[@CR25]^. In the context of NOTCH signaling, RBPJ mediates many processes, including progenitor renewal and cell fate selection during embryogenesis and tissue homeostasis, as well as pathological processes such as tumorigenesis and pulmonary hypertension^[@CR1],[@CR2],[@CR26],[@CR27]^. In addition to mediating NOTCH signaling, RBPJ can act independently, for instance to attenuate hypoxia signaling through direct physical interaction with HIF1α and 2α proteins^[@CR28]^. **RIN1**, therefore, is a unique chemical tool to probe this biologically important protein.
**RIN1** treatment induced a profile of transcript changes that was distinct from the changes induced by DAPT, including opposing effects on bonafide NOTCH target genes *HES1*, *HES5*, *HEY1* and *DTX* (Fig. [4](#Fig4){ref-type="fig"}). Activation of some NOTCH target genes by RBPJ inhibition was not unexpected since RBPJ functions as both a repressor (in the absence of NOTCH signaling) and as an activator in response to NOTCH (Fig. [4f](#Fig4){ref-type="fig"}). **RIN1** was identified based on the functional inhibition of RBPJ complexed with SHARP (primary screen) and secondarily based on inhibition of NOTCH (secondary screen), with which it forms repressing and activating complexes, respectively. Furthermore, siRNA-mediated knockdown of RBPJ induced a similar profile of transcript changes as did **RIN1**. Together, we conclude that **RIN1** inhibits RBPJ in both its activating (NOTCH) and inhibiting (SHARP) complexes. In contrast, CB-103 induced a profile of gene expression changes that differed from that of **RIN1** but resembled the changes induced by DAPT, suggesting that CB-103, whose target is unknown, might act at or near the level of the NOTCH receptor itself. Consistent with targeting RBPJ, **RIN1** inhibited the activity of RBPJ-VP16 fusion protein that is a constitutively active transcriptional activator (Fig. [1f](#Fig1){ref-type="fig"}). Moreover, **RIN1** was not selective for NOTCH isoform, also consistent with it targeting RBPJ. Thus, the transcriptomic and functional data suggest that **RIN1** targets RBPJ itself or a closely interacting protein. Studies to evaluate the possible physical interaction with RBPJ are underway.
As examples of its utility as a chemical probe, **RIN1** suppressed the proliferation of three hematologic tumor cell lines (Jurkat and KOPT-K1 T-ALL, and REC-1 MCL). Interestingly, the potencies and efficacies of **RIN1** relative to DAPT and CB-103 varied across lines (Fig. [2](#Fig2){ref-type="fig"}). **RIN1** and siRNA against RBPJ effectively blocked proliferation of all cells; however, DAPT modestly blocked proliferation of Jurkat cells (Fig. [2](#Fig2){ref-type="fig"}), suggesting that RBPJ plays a relatively more important role in controlling Jurkat cell proliferation than does NOTCH cleavage and release of ICD. Also, CB-103 was more effective than either DAPT or **RIN1** in Jurkat and KOPT-K1, but not Rec1 lines, suggesting differential roles of the NOTCH pathway components across cancer lines with activating NOTCH mutations. **RIN1** also promoted the differentiation of the C2C12 skeletal myoblasts into multinucleated myofibers (Fig. [3](#Fig3){ref-type="fig"}). CB-103 was ineffective in this context suggesting that its target (which is unknown) is not involved, again illustrating the variable effects of inhibiting different NOTCH pathway components.
In summary, **RIN1** inhibits the functional association of RBPJ with SHARP and NOTCH ICD, thereby blocking the transcriptional repression and activation, respectively, of downstream genes. **RIN1** is structurally and functionally distinct from existing small molecule NOTCH inhibitors. The only previously known selective inhibitor, IMR-1, acts at the level of NOTCH itself^[@CR5]^ (Fig. [4f](#Fig4){ref-type="fig"}). Other small molecule inhibitors (Fig. [4f](#Fig4){ref-type="fig"}) are γ-secretase inhibitors (GSIs), which act unselectively to block NOTCH processing, dihydropyridine (DHP) inhibitors of NOTCH trafficking^[@CR29],[@CR30]^ and CB-103^[@CR22]^, which appears to function at the level of NOTCH (Fig. [4](#Fig4){ref-type="fig"}) and has recently entered clinical development for treatment of NOTCH-dependent cancers^[@CR31]^. As the first small molecule RBPJ inhibitor, **RIN1** could be exploited for chemical genetics and therapeutic applications.
Methods {#Sec8}
=======
HTS for small molecule RBPJ inhibitors {#Sec9}
--------------------------------------
### Screen and counter screen cell lines {#Sec10}
The cell line AD-293 (Agilent \# 240085) was used to generate a stable cell line carrying the UAS-Luciferase reporter plasmid pGL4.35 (Promega) by hygromycin selection (50 µg/ml) termed AD-293-UAS-Luc. Subsequently, this line was transfected (followed by clone selection) with the plasmids indicated below to generate the primary screen and counter screen lines. AD-293-UAS-Luc and its derivatives were grown in DMEM 4.5 g/ml glucose, sodium pyruvate, glutamine, pen/strep, 10% fetal bovine serum and selection antibiotic. Assay media was DMEM (Cellgro \#17-205-CV) 4.5 g/ml glucose, glutamine, pen/strep, 5% fetal bovine serum, without phenol red.
To generate the primary screen cell line, AD-293-UAS-Luc cells were transfected with plasmid pBI-CMV (Clonetech) that contained two expression cassettes. Multiple cloning site 1 (ClaI-EcoRV) contained the Gal4 DNA binding domain (Gal4DBD) in frame with the nucleotide sequence corresponding to SHARP amino acids 2770--3127, which is a minimal fragment that retains RBPJ binding^[@CR18]^. Multiple cloning site 2 (BglII-XbaI) contains the sequence of full-length human RBPJ containing 4 mutations that block binding to recognition sites in DNA^[@CR32]^ (to avoid any effects of RBPJ directly interacting with DNA) fused to the transactivating domain of Herpes simplex virus VP16 and a MYC epitope tag.
To generate the counter screen cell line, AD-293-UAS-Luc cells were transfected with a plasmid that contained the CMV promoter driving the GAL4 DBD fused to the VP16 transactivation domain. This plasmid was made by replacing the DsRed2 coding region (AgeI-NotI fragment) in pDsRED2-C1 (Clontech) with a bicistronic DNA fragment encoding Gal4 DBD-VP16 separated from a UAS-eGFP cassette by a stop codon and poly adenylation sequence. This sequence was derived from pBSEGVUG obtained from Scott Fraser, Caltech.
### High throughput (primary) screening {#Sec11}
HTS was performed in white tissue culture treated 1536 well plates (Corning Cat. \#3727), 500 cells per well in 5 µl volume. Compounds were dispensed by either Labcyte acoustical dispenser (100,000 compound pilot screen) or pintool (1.78 MM compound screen) and incubated 17 hours at 37 °C and 5% CO~2~. Luciferase substrate was Britelite Plus (Perkin Elmer). The assay had excellent properties (Z' = 0.74 ± 0.06; S/B = 153 ± 19; hit rate of 0.16%, Supplementary Table [S1](#MOESM1){ref-type="media"}) determined in a single point determination (10 µM) pilot screen of \~100,000 diverse small molecules at the Prebys Center for Drug Discovery (PCDD, La Jolla, CA) that yielded 15 structurally similar compounds after counter screening. After transfer to Sanofi-Tucson, the primary screen was performed identically against 1,780,000 compounds (10 µM) using the PCDD hit as a positive control and DMSO vehicle as negative control. The mean Z' of the entire screen was 0.64 ± 0.05 with a 1% primary hit rate. Hits were evaluated through two confirmatory screens (Supplementary Table [S1](#MOESM1){ref-type="media"}**)**, first 4-point (10, 2.5, 0.625, and 0.156 µM) and then 8-point (20, 10, 5, 2.5, 1.25, 0.625, 0.313, 0.156 µM) in the same assay as used for primary screening, and concurrently in counter screens that measured inhibition of luciferase constitutively driven by Gal4-VP16 (same concentration ranges).
Secondary screening {#Sec12}
-------------------
### *Hes1*-luciferase reporter line {#Sec13}
The UAS sequence in pGL4.35 was replaced by the murine *Hes1* promoter (−194 to +160 relative to TSS of mouse Hes1 gene)^[@CR33]^ to generate the Hes1-Luciferase reporter construct that was then stably transfected into AD-293 cells to yield AD-293 *Hes1*-Luciferase line. This cell line was transiently transfected in 384 well plates (reverse transfection) with plasmids to direct expression of NOTCH2 ICD (pAdloxN2ICD^[@CR28]^), NOTCH3 ICD (pCDNA3.1 + Hygro N3ICD-HA^[@CR28]^) and RBPJ-VP16myc fusion protein (pCDNA3.1 RBPJ-VP16-myc^[@CR28]^). Luciferase signal was read out as for primary screening using the steadylite plus substrate (Perkin Elmer).
### Secondary screen controls {#Sec14}
The parental AD-293 cell line was transfected with plasmid pCNDA3.1-Firefly Luciferase \[made by transferring the Firefly Luciferase gene from pGL3 (Sigma-Aldrich)\]. For cell viability, resazurin (Sigma R7017) was added to the cells at 5 µg/ml final concentration and incubated 2 hours at 37 °C before fluorescence signal detection (Perkin Elmer Envision plate reader).
### AD-293 transfections and western blot analyses {#Sec15}
AD-293 were transfected with RBPJ siRNA (Dharmacon J-007772-06) or non-targeting control (D-001810-01) using Lipofectamine RNAiMax with the following protocol per 4 cm^2^ well (12 well format): 100 µl Optimem containing 1.2 ul siRNA at 25 µM was combined with 100 µl Optimem containing 3 µl Lipofectamine and briefly vortexed. After 5 minutes incubation the mixture was added to 550 ul complete culture media containing 330,000 cells and 250 µl media 8 µM compound (or DMSO). Final siRNA concentration was 30 nM and final compound concentration was 2 µM. After 12, 24 or 48 hours of culture, cells in wells were rinsed with cold PBS and lysed with 175 µl RIPA buffer containing protease inhibitor cocktail. Protein concentration was measured with BCA protein assay (Pierce). 6 µg of total protein was loaded per well in polyacrylamide gel (Miniprotean TGX 10% gel) and transferred to PVDF membrane. Antibodies used were: anti-RPBJ (Cell Signaling 5313, rabbit) and anti-α-Tubulin (T5168, mouse). Secondary antibodies were 780 nm anti-rabbit and 680 nm anti-mouse Ig. Imaging and band quantification were done using an Odyssey Licor System.
Cellular assays {#Sec16}
---------------
### Tumor cell proliferation assays {#Sec17}
Jurkat, Koptk-1 and Rec-1 cells were grown in RPMI 1640 supplemented with glutamine, pyruvate, 10% fetal bovine serum and pen/strep. The proliferation assay was performed in black wall, clear bottom 384 well plates. Each well received 20 µl of cell suspension and 20 µl of media with compound (at 2x final concentration). 48 hours later, an additional 40 µl of media with 1x compound was added to each well. At 96 hours of culture, Resazurin was added to each well and cell number was calculated from a standard curve of Resazurin fluorescence intensity as a function of cell number for each cell type.
### C2C12 myoblast differentiation assay {#Sec18}
C2C12 cells were grown at low density in DMEM 4.5 g/l glucose, pen/strep and 10% FBS. Cells were seeded in 96 well plate black wall clear bottom at 10,000 cells per well in DMEM 20%FBS. Media was removed on the following day, and wells were rinsed once with PBS and replaced with 200 µl differentiation media (DMEM 2% horse serum) plus treatment (considered day 0). Three wells were assigned to each treatment and DMSO concentration was 0.1% in all samples. Cells were incubated under 3% oxygen (5% CO~2~, 92% N2). Media (+treatment) was replaced at day 1 and 2 (media changes for hypoxia conditions were done under 5% oxygen). On day 4 (90 hours from initial treatment) cells were fixed with 4% paraformaldehyde for 10 minutes. MHC immunostaining was done using MF20 antibody concentrate (Developmental Studies Hybridoma Bank) overnight at 1:100 dilution in blocking buffer (3% BSA, 0.1% TritonX100, 0.2% Glycine PBS). Secondary antibody anti mouse Alexa 488 was used at 1:200 dilution. Nuclei were counter-stained using DAPI and cells were imaged using 20× (0.75 N.A.) objective using an IC200 automated microscopy system (Vala Sciences, San Diego, CA).
Image analysis {#Sec19}
--------------
DAPI and Alexa488 image Z-stacks image stacks consisting of 20 images with step size of 1 µm were obtained by structural illumination microscopy using an IC200 automated microscopy system (Vala Sciences, San Diego, CA) at 20× (0.75 N.A.). Each stack was then projected (maximum intensity projection) and analyzed to calculate the following parameters in ImageJ/Fiji. Nuclei count: A nuclear mask and count was created from the DAPI images by size thresholding (400 pixels) using the Analyze Particle function in ImageJ. MHC^+^ object count: The MHC^+^ object mask (to identify cells) was created as for the nuclear mask but using the Alexa488 images and a minimum size threshold of 3000 pixels and a constant signal intensity threshold for all images. This value is reported in Fig. [3b](#Fig3){ref-type="fig"} as the "Number of cells". MHC^+^ nuclei count: MHC^+^ object and nuclei masks were overlaid. Nuclear objects with \>80% MHC signal were counted as MHC^+^ nuclei. Average nuclei count per MHC^+^ object: This value is calculated as the ratio MHC^+^ nuclei/MHC^+^ objects) and reported in Fig. [3c](#Fig3){ref-type="fig"} as "Nuclei/cell".
qRT-PCR {#Sec20}
-------
Total RNA was extracted using Quick-RNA miniprep kit (Zymo Research) following kit protocol including a DNAse treatment step. RNA was quantified using a Nanodrop and 500 ng of total RNA was used for reverse transcription, which was performed with the Quantitect RT klt (Qiagen), which includes a genomic DNA removal step. cDNA was diluted to avoid PCR inhibition by contaminants. 0.2 µl of cDNA was used per 10 µl of qRT-PCR reaction for all genes tested except for *HES5* that required 2 µl of cDNA per 20 µl qRT-PCR reaction. qRT-PCR reactions were performed using iTaq Universal SYBR Green (BioRad) in ABI 7900HT (Applied Biosystems) following manufacturers' protocols.
RNA sequence analysis {#Sec21}
---------------------
Jurkat cells were transfected with siRNAs by electroporation (Neon, Thermo Fisher Scientific) using the following conditions: 100 µl cells at 20 million cells/ml, 10 µl siRNA at 100 µM, 3 pulses of 10 msec duration at 1350 V. siRNAs used were: Dharmacon J-007772-06 (CUCCCAAGAUUGAUAAUUA; for RBPJ NM_203283), non-targeting control D-001810-01 and siGLO (D-001630-02) to assess transfection efficiency.
RNA sequencing was performed by Novogene (Illumina HiSeq) to obtain 20 million paired-end \~150 bp reads per sample in biological duplicate. FasQC (v0.11.5) and MultiQC (v1.3) were used to assess read quality. Adapter and quality trimming of reads were performed with Trimmomatic (v0.36). Reads were mapped to genome GRCh37 (hg19) using STAR (v2.5.3a) with UCSC gene annotations, and on average 19.1 million uniquely mapped reads were counted. Uniquely mapped reads were summarized at the gene level with featureCounts (v1.28.1) from the Rsubread module. Differential expression was performed with DESeq2 (v1.17.39). Genes having a fold-change of greater than 1.5 and a p-value of less than 0.05 were considered significant. Analysis filters are described in the text and legend to Fig. [4](#Fig4){ref-type="fig"}.
Statistical analyses {#Sec22}
--------------------
Statistical analyses were performed on GraphPad Prism software using two way ANOVA with Dunnett's post-test or by an unpaired Student's T-test for calculation of P-values as indicated in the figure legends. The primary and counter screens were performed at a single dose and determination for each compound. The initial confirmatory screen was performed at 4 doses, single determination each, per compound. The second confirmatory screen was performed at 10 doses, single determination for each compound. The secondary screens and subsequent studies using **RIN1** were repeated with multiple determinations each (n indicated in the figure legends) and a minimum of three times with similar results. All experiments were repeated as indicated in the figure legends.
Supplementary information
=========================
{#Sec23}
Supplementary Information
**Publisher's note:** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supplementary information
=========================
**Supplementary information** accompanies this paper at 10.1038/s41598-019-46948-5.
We thank Dr. Thomas Look (Harvard University) for providing the KOPT-K1 cell line, and Mr. Matthew Greenhaw for RNAseq analysis. The pilot screen was performed at the Sanford-Burnham-Prebys Medical Discovery Institute with the support of P30CA030199. This research was supported by the NIH R01HL113601 and R01HL132225 to MM and Stanford Cardiovascular Institute and Stanford School of Medicine funds to MM and the DFG (German Research Foundation) through a collaborative research grant (SFB 1074/A03) to FO. CH was supported by a California Institute for Regenerative Medicine postdoctoral fellowship (TG2-01162).
C.H., A.S., M.S., R.C., A.A.N.B., J.G.G., T.S., F.O., J.R.C., P.R.L. and M.M. designed and performed experiments. F.O. provided reagents and data on the RBPJ:SHARP interface. J.R.C., P.J., T.S., F.O. and M.M. obtained funding and supervised the inter-institutional collaboration. C.H., M.M., J.R.C., P.R.L. and P.J. wrote the manuscript. All authors edited the manuscript.
The Jurkat cell RNAseq datasets are available on NCBI GEO (Accession number GSE134401). All other data generated or analyzed during this study are included in this published article (and its supplementary information files).
The authors declare no competing interests.
| {
"pile_set_name": "PubMed Central"
} |
INTRODUCTION
============
Oligodendrocytes are myelin-producing cells of the CNS (central nervous system) essential for the health and proper function of neural circuitry. Alterations in oligodendrocyte numbers and in myelination are associated with a number of neurological and psychiatric disorders that include multiple sclerosis, spinal cord injury, adult stroke, schizophrenia and several neurodegenerative diseases (Almad et al., [@B2]; Lee et al., [@B32]; Matute and Ransom, [@B35]; Bankston et al., [@B6]; Philips et al., [@B52]). It has been established that there is a maturation-dependent vulnerability of oligodendroglia to injury (Oka et al., [@B47]; Back et al., [@B3]; McTigue and Tripathi, [@B40]; Bradl and Lassmann, [@B11]) with the immature progenitors significantly more vulnerable than mature oligodendrocytes. For example, using a neonatal rat model of hypoxic-ischemic brain injury, Back and colleagues found that the white matter injury involved apoptotic death of the late OPCs (oligodendrocyte progenitor cells) (Back et al., [@B3], [@B5]). Late OPCs also were found to be vulnerable to cell death in other models of neonatal white matter injury (Vottier et al., [@B68]; Falahati et al., [@B19]). Oligodendrocyte progenitors are especially vulnerable to increased levels of glutamate, which has been implicated in white matter damage after neonatal H--I (hypoxia--ischemia), spinal cord injury and multiple sclerosis (Petito et al., [@B51]; Ness et al., [@B45]; Park et al., [@B49]). Axons, and rather surprisingly, oligodendroglia themselves, are major source of extracellular glutamate in white matter (Back et al., [@B4]). Oligodendroglia express AMPA/kainate receptors (Patneau et al., [@B50]) which mediate the excitotoxic death of the OPCs in developing white matter (Yoshioka et al., [@B76], [@B76]; Matute et al., [@B36]; McDonald et al., [@B38], [@B39]; Yoshioka et al., [@B77]; Alberdi et al., [@B1]; Ness and Wood, [@B43]; Sanchez-Gomez et al., [@B58]). Whereas oligodendroglia also express NMDA receptors, NMDA receptor expression appears to be most prominent in mature stages of the lineage (Karadottir et al., [@B28]; Salter and Fern, [@B57]; Micu et al., [@B41]).
Several studies have begun to define the death pathway initiated in OPCs by high levels of extracellular glutamate. In a prior study, we demonstrated that excess glutamate activates AMPA/kainate receptors and Ca^2+^ influx in late OPCs resulting in translocation of the pro-apoptotic protein Bax to the mitochondria, cytochrome *c* release, activation of caspase 9 and caspase 3, nuclear fragmentation and cell death (Ness et al., [@B44]). Moreover, we showed that IGF (insulin-like growth factor)-I sustains Akt phosphorylation in OPCs and prevents Bax mitochondrial translocation and apoptosis in excitotoxic conditions. Subsequently, Pang and colleagues demonstrated that IGF-I also protects OPCs from TNFα (tumor necrosis factor α) cytotoxicity in a dose-dependent manner by preventing Bax translocation from the cytosol to mitochondria via activation of PI3K (phosphoinositide 3-kinase)/Akt (Pang et al., [@B48]). Deletion of Bax in mice attenuated oligodendrocyte death after spinal cord injury *in vivo* (Dong et al., [@B17]) and after staurosporine, cyclosporine A or AMPA/kainite-mediated insults *in vitro* (Dong et al., [@B17]; Sanchez-Gomez et al., [@B59]). Moreover, blocking Bax translocation to mitochondria significantly protected rat and mouse oligodendrocytes from AMPA- and kainate-induced damage (Sanchez-Gomez et al., [@B59]).
The mechanisms by which Bax is regulated in healthy cells and activated during apoptosis appears varied depending on cell type and type of toxicity. Both direct activation and indirect Bax activation in apoptosis have been described (Labi et al., [@B31]; Hacker and Weber, [@B25]; Willis et al., [@B71]; Chipuk and Green, [@B14]; Giam et al., [@B22]). In the direct activation model, an 'activator' BH3-only domain protein is required to directly activate Bax or its related multidomain pro-apoptotic protein, Bak. In healthy cells the pro-survival Bcl-family members Bcl2 or Bcl-xL sequester the BH3 protein. In the indirect model of Bax activation, BH3 proteins activate apoptosis by binding and neutralizing the pro-survival proteins, allowing Bax/Bak to homo-oligomerize and permeabilize the mitochondria.
In previous studies we determined that glutamate-mediated Bax translocation and apoptosis, as well as IGF-I protection from these events in OPCs, was downstream of calcium entry following activation of AMPA/kainite receptors and independent of changes in Bcl-2, Bcl-xL, or Bax protein levels (Ness et al., [@B44]). The goal of the present study was to determine Bax suppressors in healthy OPCs and Bax activators in glutamate-induced excitotoxicity *in vitro* and in white matter following H--I in the immature brain.
MATERIALS AND METHODS
=====================
MEM (minimal essential media), DMEM (Dulbecco\'s modified Eagle\'s medium)/F-12, FBS (fetal bovine serum) and trypsin were purchased from Gibco/Invitrogen, and papain from Worthington. Cell culture supplements, and [L]{.smallcaps}-glutamic acid monosodium salt hydrate were purchased from Sigma Chemical Compan. Recombinant rat IGF-I was purchased from Upstate Biochemicals. Recombinant human FGF-2 (fibroblast growth factor-2) was purchased from R&D Systems. Rabbit polyclonal antibodies against Bax (N20), Bcl-xL, and Bid were obtained from Santa Cruz Biotechnology, against Bax6A7 from BD Transduction Labs, and against Bad and p-Bad(Ser^155^) from Cell Signaling Technology. Antibodies against β-actin were from Sigma--Aldrich; cleaved caspase 3 and VDAC (voltage-dependent anion channel) antibodies were from Cell Signaling Technology. Cofilin antibodies were from Santa Cruz Biotechnology and Cell Signaling Technology. Chemical reagents Chaps and NP-40 (Igepal) were from Sigma--Aldrich. Common laboratory chemicals were purchased from either Sigma or VWR International.
Cell culture and treatment conditions
-------------------------------------
All experiments were performed in accordance with research guidelines set forth by the Society for Neuroscience Policy on the use of animals in neuroscience research. Animal protocols were reviewed and approved by the IACUC committee of NJMS/UMDNJ. Mixed glia were isolated from newborn Sprague--Dawley rat forebrain cortices as previously described (McCarthy and de Vellis, [@B37]; Ness and Wood, [@B43]; Ness et al., [@B44]). Briefly, tissues were enzymatically digested with trypsin and DNase I and then mechanically dissociated and plated in MEM containing 10% FBS and 30% glucose with antibiotics. The mixed glial cells were grown in T75 flasks until they were confluent (10 days). Microglia were separated from the cultures by shaking the flasks on a rotary shaker for 1.5 h at 260 rev./min. OPCs were isolated following an additional 18 h. OPCs were seeded on to poly-[D]{.smallcaps}-lysine-coated T75 flasks at density of 2×10^4^ /cm^2^ in N2S, composed of 66% N2B2 media \[DMEM/F12, supplemented with 0.66 mg/ml BSA, 10 ng/ml [D]{.smallcaps}-Biotin, 20 nM progesterone, 100 μM putrecsine, 50 μg/ml apo-transferrin, 100 units/ml penicillin/streptomycin, 5 μg/ml insulin and 34% B104-conditioned medium (N2B2 preconditioned by B104 neuroblastoma cells), 5 ng/ml FGF-2 and 0.5% FBS\]. OPC cultures were amplified for 4 days and passaged once using papain (Young and Levison, [@B78]) prior to experiments. To obtain highly enriched cultures of late OPCs for experiments, early OPCs were replated at 2 × 10^4^ cells/cm^2^ on to poly-[D]{.smallcaps}-lysine-coated dishes in N2S with 0.5% FBS and 10 ng/ml FGF-2 for 48 h. These conditions produced cultures that contain 90--95% 04^+^/Ranscht^−^ late OPCs, 6--8% R24^+^/04^−^ early OPCs, 1--2% Ranscht^+^ immature oligodendrocytes, 2% GFAP^+^ astrocytes and 0.01% OX42^+^ microglia (Ness and Wood, [@B43]). Treatment medium was N2B2 medium without insulin, FGF-2 and FBS. Cells were treated for various time points with 500 μM glutamate and/or 20 ng/ml of IGF-1; media was replaced every 18 h.
Immunocytochemistry
-------------------
OPCs were plated on to round 12 mm cover glasses (Fisherbrand) coated with 0.1 mg/ml poly-[L]{.smallcaps}-ornithine in N2S media at a density of 2×10^4^ cells/cm^2^. Cells were differentiated to late OPCs and treated with glutamate or IGF-I as described above. Following treatments, cells were rinsed in PBS and fixed in 4% paraformaldehyde for 15 min at room temperature. OPCs were then rinsed three times in PBS and blocked for 30 min in 2% normal goat serum, 0.1% Triton X-100 in PBS at room temperature. Primary antibodies against Bax (1:50) or cofilin (1:50) were incubated in diluent (2% normal goat serum, 0.1% Triton X-100, 5 mg/ml BSA in PBS) overnight at 4°C. After three washes in PBS, primary antibodies were detected with fluorescein-conjugated goat-anti-mouse IgG (H+L) AF546 (Alexa Fluor 546) (red) and anti-rabbit IgG (H+L) AF488 (Alexa Fluor 488) (green) (1:2000) (Invitrogen) secondary antibodies for 45 min at room temperature. Incubation for 5 min with DAPI (1:1000) was used to identify nuclei. Coverslips were washed three times in PBS and mounted on slides with Fluoro-Gel (EMS).
Immunoprecipitation (IP) and Western blotting
---------------------------------------------
Late OPCs were treated with glutamate and/or IGF-I for 24 h. Cells were harvested in ice-cold PBS with PIC (protease inhibitor cocktail, 1:100; Sigma) and PMSF (1 μM; Sigma). For Bax(6A7) IP, cells were lysed in 1% CHAPS lysis buffer \[10 mM Hepes, 150 mM NaCl, PIC (1:100), 20 μM PMSF, 5 μM NaF, 1 μM Na~3~VO~4~, 50 μM NaH~2~PO~4~\]. For all other IPs, cells were lysed in 1% NP-40 lysis buffer \[150 μM NaCl, 5 μM EDTA, PIC (1:100), 20 μM PMSF, 1 μM Na~3~VO~4~, 5 μM NaF\]. After sonification, samples were quantified by Protein DC Assay (Bio-Rad Laboratories). For IPs 500--1000 μg of protein (or 2 mg for protein from white matter dissections) was incubated with the appropriate optimized amount of antibody overnight at 4°C on a rotating shaker according to the protocol from Pierce. The antigen--antibody complex was incubated with immobilized Protein A/G with gentle mixing for 2 h at 4°C on a rotating shaker. The complex was washed three times in 25 μM Tris/HCl (pH 7.5), 150 μM NaCl. For all Western blot analyses, 15--25 μg of protein was heated in electrophoresis loading buffer to 85°C for 5--10 min and separated on 12% mini-SDS polyacrylamide gels (Invitrogen). Proteins were then electrotransferred on to nitrocellulose membranes (Whatman). Membranes were blocked with appropriate primary and secondary antibodies (GAR-HRP or GAM-HRP conjugated IgG from Jackson ImmunoResearch) in 5% milk in TBS/0.1% Tween 20. Secondary antibody was detected using an enhanced chemiluminescence system (New England Nuclear). Images were digitized on an Ultra-lum Gel Imager (BioVision). Controls for IP reactions included isotope specific IgG (mouse, rabbit; Santa Cruz Biotechnology) and sample minus antibody incubation.
Amnis ImageStream imaging flow cytometry
----------------------------------------
Late OPCs (1×10^6^) in suspension were prepared by harvesting cells using 0.05% Trypsin/EDTA (Cellgrow, Mediatech), followed by centrifugation at 1000 ***g*** for 5 min. The cell pellet was washed with 1×PBS, resuspended in 500 μl of 4% paraformaldehyde and fixed overnight at 4°C. Cells were again collected by centrifugation and the cell pellet permeabilized in buffer (0.1%Triton X-100 in PBS), washed with 2% FBS in PBS and incubated with primary antibody against Bax (1:50; Santa Cruz Biotechnology) and CoxIV (1:50; Cell Signaling Technology) for 30 min at 4°C. Cells were washed and incubated with secondary antibody (AF488-green, AF647-red, 1:2000) for 30 min. Cells were rinsed with 2% FBS and resuspended in 1% paraformaldehyde. Samples were acquired on the Amnis Imagestream 100 with the EDF element. Samples were then analyzed with IDEAS 4.0. Briefly, single cells were identified by graphing the area of the BF (brightfield) vs Aspect Ratio Intensity. Single cells were then analyzed for double-positive AF488 and AF647 by graphing Intensity AF488 vs AF647. Double-positive cells were finally analyzed for co-localization by the Bright Detail Similarity feature for the two stains. A value of 1.75 or greater denoted co-localized events.
Subcellular fractionation
-------------------------
Mitochondrial/cytoplasmic fractionation was performed using the Mitochondria Isolation Kit for Cultured Cells (Pierce) with minor modifications. Late OPCs (2 × 10^7^) were harvested by centrifuging at 1000 ***g*** for 2 min. Following separation, the mitochondrial and cytoplasmic fractions were lysed in 2% CHAPS in Tris-buffered saline (25 mM Tris, 0.15 M NaCl, pH 7.2) and protein concentration was determined by the DC protein Assay (Bio-Rad Laboratories).
Mass spectrometry and protein identification
--------------------------------------------
OPCs were differentiated to the late OPC stage and then treated with glutamate or IGF-I as for prior experiments for 28 h and lysed in 1% CHAPS lysis buffer. Immunoprecipitation was performed for total Bax (N20) and then proteins were separated by SDS/PAGE. After electrophoresis, the gel was fixed and subjected to SYPRO Ruby staining, and gel bands were excised, diced into 1 mm^3^ pieces and washed with 30% ACN (acetonitrile) in 50 mM ammonium bicarbonate followed by reduction with DTT and alkylation by iodoacetamide. Digestion was initiated by adding trypsin and incubating at 37°C overnight. The resulting peptides were extracted with 30 μl of 1% TFA (trifluoracetic acid) followed by C~18~ Ziptip (Millipore) desalting according to the manufacturer\'s protocol. The peptides were dried in a Speedvac, and re-suspended in 10 μl of solvent A \[2% ACN, 0.1% FA (formic acid)\] for ESI-LC-MS/MS analysis on a API-US QTOF tandem MS system equipped with a nano-ESI source (Waters Corporation) as described previously (Wu et al., [@B73]). Protein identification was performed by searching the MS/MS spectra against rat protein sequences in the NCBInr database using a Mascot search engine (version 2.4.1). Precursor mass tolerance was set at 100 p.p.m. and fragment mass tolerance was set at 0.6 Da. Oxidized methionine and carbamidomethyl-labelled cysteine were set as variable modifications as the search parameters. Mascot search results were further exported to Scaffold software (version 3.6.5) for visualization and validation. Peptide identifications were accepted if they could be established at greater than 95.0% probability by the Peptide Prophet algorithm (Keller et al., [@B29]). Protein identifications were accepted if they could be established at greater than 95.0% probability and contained at least 1 identified peptides. Protein probabilities were assigned by the Protein Prophet algorithm (Nesvizhskii et al., [@B46]). Proteins that contained similar peptides and could not be differentiated based on MS/MS analysis alone were grouped to satisfy the principles of parsimony.
Perinatal H--I and white matter dissections
-------------------------------------------
Timed pregnant Wistar rats were obtained from Charles River Laboratories. After normal delivery litter size was adjusted to 12 pups per litter. Two different groups of animals were used: sham-operated controls and experimental H--I animals. Cerebral H--I was produced in male and female 6-day-old rats (day of birth=P0) by a permanent unilateral common carotid ligation followed by hypoxia (variation of the Vannucci Model) (Rice et al., [@B54]; Romanko et al., [@B55]). Briefly, pups were lightly anesthetized with halothane (4% induction and 1.5% maintenance). Once fully anesthetized, a midline neck incision was made and the right CCA (common carotid artery) was identified. The CCA was separated from the vagus nerve and then cauterized at two distinct locations with a bipolar cauterizer at power setting of 10 (Codman & Shurtleff, model number 80-1140). Animals were returned to the dam for 1.5 h. The pups were pre-warmed in jars for 20 min submerged in a 37°C water bath and then exposed to 75 min of hypoxia (8% O~2~/92% N~2~). After hypoxia the pups were returned to their dam for recovery periods of 24 h, at which time they were anesthetized and killed by intracardiac perfusion. Sham-operated animals were anesthetized and the surgery was performed without ligation of the right CCA followed by exposure to hypoxia.
All brain tissue samples (*n*=6 per experiment for H--I and sham) were dissected fresh and pooled \[IL (ipsilateral) and CL (contralateral) were pooled separately for H--I brains\] and homogenized on ice in 500 μl of homogenization buffer with a Tissue-Tearor (rotor/stator type) homogenizer (BioSpec Products). Homogenization buffer contained 1% NP-40 lysis buffer \[150 μM NaCl, 5 μM EDTA, PIC (1:100), 20 μM PMSF, 1 μM Na~3~VO~4~, 5 μM NaF\] and 1% CHAPS lysis buffer \[10 mM Hepes, 150 mM NaCl, PIC (1:100), 20 μM PMSF, 5 μM NaF, 1 μM Na~3~VO~4~, 50 μM NaH~2~PO~4~\]. Samples were used for IPs or Western blotting as described above.
RESULTS
=======
Time course of Bax activation and mitochondrial translocation in late OPCs after exposure to excitotoxic levels of glutamate *in vitro*
---------------------------------------------------------------------------------------------------------------------------------------
Our *in vitro* model for glutamate-mediated excitotoxicity of OPCs utilizes media where the IGF-1R (IGF type 1 receptor) is not stimulated (by eliminating micromolar levels of insulin present in most medium formulations) and it does not contain cyclothiazide, the glutamate-densensitizing blocker (Ness and Wood, [@B43]; Ness et al., [@B44]; Wood et al., [@B72]). Under these conditions, apoptosis occurs between 24 and 36 h following exposure to glutamate (Ness et al., [@B44]). This paradigm also has allowed us to study glutamate excitotoxicity as well as the effects of IGF-I in preventing glutamate-induced apoptosis. Thus, in our previously published and the current studies, we use low-insulin media supplemented with IGF-I for healthy OPCs, and the same media plus glutamate either in the presence or absence of IGF-I (Ness et al., [@B44]; Wood et al., [@B72]). Here, we use this culture paradigm to further define molecules involved in the apoptotic cascade in OPCs and to determine how a trophic factor such as IGF-I interferes with the cascade.
In healthy cells, endogenous Bax is a monomeric protein in an inactive conformation (Vogel et al., [@B66]). Following a death stimulus, Bax is activated, translocates to the mitochondria and oligomerizes (Gross et al., [@B24]; Edlich et al., [@B18]; Soriano and Scorrano, [@B64]; Walensky and Gavathiotis, [@B69]) ([Figure 1](#F1){ref-type="fig"}). In our prior studies we showed Bax associated with mitochondria in OPCs exposed to glutamate, but not in the presence of IGF-I (Ness et al., [@B44]); however, we did not assess the activation state of Bax. After exposure to apoptotic stimuli, an N-terminal epitope of Bax is exposed and involved in Bax membrane insertion (Sharpe et al., [@B61]). To determine whether Bax undergoes conformational changes in OPCs during excitotoxicity, we determined whether conformation-specific Bax was present in OPCs prior to apoptotic death. We detected activated Bax in OPCs exposed to glutamate at 24 h using the conformation specific antibody 6A7 ([Figure 2](#F2){ref-type="fig"}A, upper panels; Glut, arrowheads) but not in cells cultured in IGF-I either in the absence (IGF-I) or presence (Glut+IGF) of glutamate. Total Bax was detected in cells cultured under all conditions ([Figure 2](#F2){ref-type="fig"}A, lower panels).
{#F1}
{#F2}
To confirm the presence of active Bax in OPCs exposed to glutamate, we immunoprecipitated active Bax using the Bax conformation-specific antibody followed by Western blotting to detect total Bax protein. Consistent with the immunofluorescence results, there was more active Bax present after 24 h in OPCs exposed to glutamate and less active Bax present in OPCs stimulated with glutamate and IGF-I ([Figure 2](#F2){ref-type="fig"}B). Similar results were obtained as early as 18 h after exposure to glutamate (data not shown).
In our prior studies, we demonstrated Bax co-localization with mitochondria using immunofluorescence microscopy (Ness et al., [@B44]). To quantify Bax translocation to mitochondria in OPCs exposed to excitotoxic levels of glutamate, we used using Amnis Imagestream 100 multispectral imaging flow cytometry ([Figure 3](#F3){ref-type="fig"}), a quantitative method performed on thousands of individual cells with each cell considered an independent event and used to determine a similarity score indicative of co-localization. OPCs treated with glutamate and/or IGF-I were stained for Bax and CoxIV, a mitochondrial outer membrane marker. Co-localization was determined by calculating the similarity of two channels and similarity above a threshold indicated by R2 ([Figures 3](#F3){ref-type="fig"}A and [3](#F3){ref-type="fig"}B). In glutamate-treated OPCs, there was an approximately 3**--**4-fold increase in the percentage of cells where Bax and CoxIV co-localized at 18 h versus IGF-I- or IGF-I+glutamate-treated OPCs ([Figure 3](#F3){ref-type="fig"}C). Similar results were obtained at 24 h, however, since CoxIV levels begin to decrease once mitochondria are disrupted, these analyses are shown at the 18 h timepoint, which precedes cytochrome *c* release, caspase 9 activation and the appearance of apoptotic nuclei (Ness et al., [@B44]).
{#F3}
Using subcellular fractionation, we further investigated the accumulation of pro-apoptotic and anti-apoptotic proteins in mitochondria and cytoplasmic fractions after 24 h of glutamate or IGF-I treatment ([Figure 4](#F4){ref-type="fig"}). We observed higher levels of Bax protein in the mitochondrial fraction compared with the cytoplasmic fraction in glutamate-treated cells ([Figure 4](#F4){ref-type="fig"}A) consistent with our prior immunodetection of Bax translocation to the mitochondria 24 h after exposure to glutamate (Ness et al., [@B44]). In IGF-I-treated cells, Bax levels were higher in the cytoplasmic fraction ([Figure 4](#F4){ref-type="fig"}A). Bcl-xL, a member of the pro-survival Bcl-family that is abundant in OPCs (Itoh et al., [@B26]), partitioned preferentially to the mitochondrial fraction in IGF-I-treated OPCs whereas it translocated to the cytoplasmic fraction after glutamate treatment ([Figure 4](#F4){ref-type="fig"}A). This pattern was the reverse of what was observed for Bax in glutamate vs IGF-I conditions. These data on Bax activation and translocation demonstrate that OPC exposure to glutamate results in Bax activation and mitochondrial translocation, all of which can be inhibited by IGF-I. Moreover, the enrichment of Bcl-xL in the mitochondrial fraction in IGF-I/survival conditions supports the hypothesis that Bcl-xL antagonizes Bax-mediated apoptosis in OPCs. However, based on their inverse localization, Bcl-xL may antagonize Bax in OPCs, at least in part, through indirect mechanisms as reported in other cell types (Willis and Adams, [@B70]; Willis et al., [@B71]; Billen et al., [@B8]; Lovell et al., [@B33]).
{#F4}
Pro-apoptotic BH3-domain partners of Bax identified in late OPCs
----------------------------------------------------------------
One of the best-studied Bax activators is the BH3-only protein Bid, which binds pro-apoptotic Bax and Bak as well as anti-apoptotic Bcl-2 and Bcl-xL (Willis and Adams, [@B70]; Willis et al., [@B71]). Bid is cleaved by caspase 8 or caspase 3 (Bossy-Wetzel and Green, [@B9]; Slee et al., [@B62]; Yin, [@B74]) and the resulting tBid (truncated Bid) translocates from the cytoplasm to mitochondrial membranes. Indeed, Bax oligomerization and insertion into the outer mitochondrial membrane can be triggered by tBid (Lovell et al., [@B33]). Real-time PCR analyses have shown that Bid is expressed in oligodendrocytes at all stages of development (Itoh et al., [@B26]). Our analyses of Bid revealed that Bid partitioned to the cytoplasm in healthy cells (IGF-I conditions) but that it translocated to the mitochondrial fraction after glutamate-treatment ([Figure 3](#F3){ref-type="fig"}A). Furthermore, tBid co-immunoprecipitated with Bax after glutamate-treatment, whereas it was undetectable in cells treated with either IGF-I or IGF-I plus glutamate ([Figure 4](#F4){ref-type="fig"}B), suggesting that IGF-I blocks glutamate-mediated apoptosis upstream of Bid cleavage and association with Bax.
We further analyzed the association of the anti-apoptotic protein Bcl-xL with Bax and Bid in OPCs by co-immunoprecipitation. OPCs stimulated with glutamate had reduced Bax--Bcl-xL association as well as reduced association of Bid--Bcl-xL vs cells treated with IGF-I ([Figure 4](#F4){ref-type="fig"}C). These data support a model where tBid enhances Bax-mediated apoptosis in OPCs and Bcl-xL inhibits Bax activation under survival conditions through both direct and indirect mechanisms.
Identification of cofilin as a Bax-binding partner in OPCs
----------------------------------------------------------
To identify novel binding partners of Bax in both glutamate and IGF-I-treated late OPCs, we immunoprecipitated Bax and utilized a LC-MS/MS proteomic approach to identify associated proteins. One protein we identified as a Bax interacting protein in IGF-I-treated, but not glutamate-treated, OPCs was cofilin, a member of the ADF (actin depolymerizing factor)/cofilin family of actin-binding proteins, that prevents reassembly of actin filaments (Bernstein and Bamburg, [@B7]) ([Figure 5](#F5){ref-type="fig"}A). Across several independent experiments, we confirmed that cofilin co-immunoprecipitated with Bax in IGF-I-treated, but not in glutamate-treated, OPCs ([Figure 5](#F5){ref-type="fig"}B). Prior studies have demonstrated that cofilin associates with mitochondria during apoptosis (Chua et al., [@B15]; Klamt et al., [@B30]). Specifically, cofilin was shown to translocate to the mitochondria in several human cell lines prior to cytochrome *c* release following induction of apoptosis (Chua et al., [@B15]). Moreover, perturbations that reduced cofilin prevented apoptosis; cofilin localization to mitochondria promoted apoptosis in these studies (Chua et al., [@B15]). Therefore, we investigated whether cofilin shifted localization from the cytoplasm to the mitochondria, similar to Bax in OPCs exposed to glutamate. These studies established that cofilin preferentially partitioned to the cytoplasm in both IGF-I- and glutamate-treated cells ([Figure 5](#F5){ref-type="fig"}C). Since others have shown that phosphorylation of cofilin at Ser^9^ (p-cofilin) is associated with its inactivation (Klamt et al., [@B30]; Posadas et al., [@B53]), we determined whether the phosphorylation status of cofilin changed in glutamate-stimulated OPCs. These studies revealed that p-cofilin increased in glutamate-treated OPCs vs IGF-I-treated OPCs at both 18 h and 24 h ([Figure 5](#F5){ref-type="fig"}D), suggesting it is in an active state in the presence of IGF-I but is inactivated during excitotoxic death. Immunostaining for cofilin and Bax in OPCs revealed that cofilin co-localized with Bax in the cell body in the IGF-I condition, but was dispersed into cell processes and showed less co-localization with Bax after glutamate stimulation ([Figure 5](#F5){ref-type="fig"}E). Taken together, these data are inconsistent with the ascribed function of active, dephosphorylated cofilin in Bax translocation and mitochondrial dysfunction. Rather, our data are more consistent with the hypothesis that cofilin in late OPCs regulates actin reorganization during process outgrowth, which may include regulating mitochondria and/or Bax indirectly.
{#F5}
Bax regulation in white matter following hypoxia--ischemia in the immature brain
--------------------------------------------------------------------------------
To determine mechanisms of apoptosis and Bax regulation in OPCs *in vivo*, we used the Vannucci neonatal rat model of hypoxic--ischemic injury. In prior studies, we documented increased apoptosis of OPCs in white matter of the corpus callosum within 24 h after inducing H--I in 6-day-old neonatal rat pups (Ness et al., [@B45]). We also demonstrated that IGF-I rescued these cells from cell death in a separate set of studies (Wood et al., [@B72]). Extending these earlier studies, we micro-dissected the subcortical white matter at 24 h of recovery from H--I and evaluated Bid and Bcl-xL association with Bax. Since the Vannucci H--I model produces unilateral hemispheric damage, we examined the ipsilateral hemisphere (IL-damaged, right) and contralateral hemisphere (CL-spared, left), in addition to sham operated brains (controls).
To validate the purity of our white matter dissections, we analyzed expression of MAP2, a dendrite-restricted protein that is enriched in gray matter and absent from white matter (Bradke and Dotti, [@B10]; Iwata et al., [@B27]) ([Figure 6](#F6){ref-type="fig"}A). We also evaluated levels of cleaved caspase 3 in the IL white matter to confirm the increase in apoptotic death in the H--I white matter ([Figure 6](#F6){ref-type="fig"}B). These analyses revealed low levels of MAP-2 in the micro-dissected white matter and an increase in active caspase 3 in the IL vs CL and sham-operated white matter. Bax immunoprecipitation revealed increased association with full length Bid in both IL and CL white matter; however, tBid was undetectable in either sample ([Figure 6](#F6){ref-type="fig"}B). In contrast, Bax association with Bcl-xL was readily apparent in the CL white matter and reduced in the IL white matter ([Figure 6](#F6){ref-type="fig"}C).
{#F6}
Based on the absence of any detectable tBid associated with Bax in the damaged white matter, we expanded our studies to include an analysis of the pro-apoptotic, BH3-only protein Bad. Bad is a positive regulator of cell death and like other BH3-only proteins, selectively displaces Bax from heterodimerizing with Bcl-2 or Bcl-xL (Willis and Adams, [@B70]; Willis et al., [@B71]). In the absence of survival stimuli, endogenous Bad is dephosphorylated and localized in the mitochondrial outer membrane; in the presence of survival factors, Bad is phosphorylated at serine sites 136, 112, and 155 (Datta et al., [@B16]). We detected decreased pBad (Ser^136^) in the IL white matter at 24 h after H--I whereas there was no difference between CL and sham-operated white matter; no difference in total Bad was detected between conditions ([Figure 6](#F6){ref-type="fig"}D). These data are consistent with a role for Bad in mediating apoptosis in the immature white matter following H--I.
DISCUSSION
==========
In OPCs, the pro-apoptotic Bcl2 protein Bax is a major mediator of apoptotic death due to excitotoxicity (Dong et al., [@B17]; Ness et al., [@B44]; Pang et al., [@B48]; Sanchez-Gomez et al., [@B59]). Despite numerous studies on Bax in a variety of cell types, there is still considerable debate about how Bax is activated, and there are no data to address mechanisms of Bax activation in OPCs. Here, we examined Bax regulation in glutamate excitotoxicity in late OPCs *in vitro* and following H--I in immature white matter. Our studies suggest potential roles for the BH3-only proteins Bid or Bad in activating Bax depending on the model of OPC death. However, excitotoxicity *in vitro* as well as white matter damage *in vivo* both correlate with loss of Bax association with Bcl-xL. Moreover, we identified cofilin as a Bax-associated partner in healthy OPCs and found that this association is disrupted in cells undergoing excitotoxic death.
Excitotoxicity in OPCs involves tBid--Bax association and loss of Bcl-xL--Bax
-----------------------------------------------------------------------------
Bax is mainly present in the cytosol of healthy cells, but after apoptotic stimuli it can shuttle from the cytoplasm to ER (endoplasmic reticulum), mitochondria and nucleus (Gill et al., [@B23]; Ghibelli and Diederich, [@B21]) ([Figure 1](#F1){ref-type="fig"}). Our studies demonstrate that excitotoxic levels of glutamate result in a Bax conformational change and increased association of Bax with mitochondria in OPCs. Bax oligomerization and insertion into the outer mitochondrial membrane can be triggered by tBid. Our studies demonstrate that both Bax and Bid translocate from the cytoplasm to mitochondria during the early stages of excitotoxic cell death and that tBid also associates with Bax in OPCs exposed to damaging concentrations of glutamate ([Figure 7](#F7){ref-type="fig"}B). Recent studies on Bax--Bid function in cell-free systems demonstrate that tBid and Bax complex formation requires membrane association which leads to the insertion of Bax into the membrane and formation of membrane pores (Lovell et al., [@B33]). Taken together, our current and past data support the hypothesis that tBid--Bax binding contributes to mitochondrial permeabilization, cytochrome *c* release and apoptosis in OPCs during excitotoxicity. These data support a direct model of Bax activation where a BH3-only protein directly activates Bax (Chipuk and Green, [@B14]). Based on our prior data showing active caspase 3 during glutamate-mediated apoptosis of OPCs (Ness et al., [@B44]) and on a proposed model for Bid function in apoptosis (Slee et al., [@B63]), we hypothesize that Bid is cleaved downstream of caspase 3 to promote a caspase 3/tBid-dependent amplification loop. The initial trigger for Bid cleavage is unknown, however, likely candidates are calpains. Other studies have shown that calpains are responsible for the initial cleavage of Bid (Yin, [@B74]), and that calpains have demonstrated functions in Bax-mediated death in OPCs downstream of calcium influx through AMPA/kainate receptor activation (Sanchez-Gomez et al., [@B59]).
{ref-type="fig"}).](an2013-0027i007){#F7}
Bcl-xL, an anti-apoptotic member of the Bcl-2 family, is highly expressed in OPCs (Itoh et al., [@B26]; Ness et al., [@B44]). Two models have been proposed to explain how Bcl-xL inhibits Bax: (i) Bcl-xL binds to BH3-only proteins such as Bid, preventing Bax activation, and (ii) Bcl-xL directly binds to activated Bax to inhibit it from forming membrane pores (Chipuk and Green, [@B14]). We observed no complex formation between Bcl-xL and the active form of Bax (data not shown). In healthy OPCs Bcl-xL preferentially attached to the mitochondria whereupon it shifted to the cytoplasm during excitotoxicity. Recently, it was proposed that inactive Bax preferentially associates with the mitochondrial outer membrane and that anti-apoptotic proteins are required to constitutively retrotranslocate Bax into the cytosol (Edlich et al., [@B18]) (see [Figure 1](#F1){ref-type="fig"}). Bcl-xL may regulate Bax in this fashion in OPCs, but the association must be transient since our cell fractionation studies indicate an inverse relationship between Bcl-xL and both Bax and Bid in healthy OPCs, such that Bax and Bid are predominantly cytoplasmic and Bcl-xL is predominantly mitochondrial ([Figure 7](#F7){ref-type="fig"}A). Our data also support an alternative model where Bcl-xL inhibits Bax by competing with it for binding to organelle membranes (Billen et al., [@B8]).
Cofilin binds Bax in healthy OPCs and binding is disrupted during excitotoxic death
-----------------------------------------------------------------------------------
Our identification of cofilin as a binding partner of Bax in OPCs is both novel and intriguing based on prior literature. Cofilin is a member of the cofilin/ADF family that regulates actin dynamics by increasing the rate of actin depolymerization. Cofilin sequesters G-actin, severs F-actin, and is abundant in growth cones and presynaptic terminals (Fox et al., [@B20]; Bernstein and Bamburg, [@B7]; Marsick et al., [@B34]; Mizuno, [@B42]). In mammalian cell lines treated with staurosporine, cofilin was identified by a proteomic approach as a factor that translocated from the cytosol into the mitochondria before cytochrome *c* release (Chua et al., [@B15]). In a recent study that evaluated the mechanisms of NMDA-induced cortical neuronal cell death, cofilin was found to be dephosphorylated and activated resulting in its translocation with Bax to the mitochondria (Posadas et al., [@B53]). However, our data demonstrate that in OPCs cofilin binds to Bax only in IGF-treated OPCs and is predominantly cytoplasmic in OPCs in both IGF-I- and glutamate-treated conditions. Moreover, the presence of phosphorylated cofilin in OPCs undergoing excitotoxic death is inconsistent with active, dephosphorylated cofilin mediating Bax translocation and mitochondrial dysfunction. However, cofilin also has known functions in actin reorganization, which may include regulating mitochondria and/or Bax indirectly in OPCs. For example, it has been shown that mitochondria associate with actin prior to Bax translocation (Tang et al., [@B65]), thus cofilin phosphorylation and inactivation may function to stabilize the actin cytoskeleton to promote this association.
Regulation of Bax after H--I in white matter
--------------------------------------------
White matter damage after H--I in the immature brain is believed to result from vulnerability of the immature oligodendrocyte (the late OPC) to factors elevated during ischemic damage, such as oxygen free radicals and glutamate (Volpe, [@B67]; Back et al., [@B4]; Segovia et al., [@B60]; Buser et al., [@B12], [@B13]). Similar to the glutamate-mediated pathway of Bax activation we observed in OPCs *in vitro*, Bax--Bcl-xL association was disrupted after neonatal H--I in white matter. Although neonatal white matter is not exclusively comprised of OPCs, these data support a model where Bax is bound to Bcl-xL in healthy OPCs and the release of Bax from Bcl-xL is a component of its activation. However, in contrast with the *in vitro* studies, we failed to detect tBid associated with Bax in the H--I white matter. This could be due either to low levels of Bax--tBid in the OPCs that were undetectable in extracts of total white matter or to a different BH3-only protein responsible for Bax activation in OPCs after H--I. In support of the latter possibility, we found reduced p-Bad (Ser^136^) in the damaged white matter. Bad functions to bind and inactivate anti-apoptotic proteins including Bcl-xL to promote the formation of the mitochondrial permeability transition pore (Chipuk and Green, [@B14]; Roy et al., [@B56]). These data suggest that Bad may be involved in Bax activation in OPCs after H--I. However, we were unable to convincingly detect changes in p-Bad in glutamate- vs IGF-I-treated OPCs *in vitro*, suggesting either that the two paradigms differ in which BH3-only proteins are involved in Bax activation and/or that the reduction in p-Bad in damaged white matter occurs in cells other than oligodendroglia.
In conclusion we have begun to define interactions of Bcl-2 family members with Bax in OPC apoptosis induced by glutamate excitotoxicity *in vitro* and in white matter damage after H--I *in vivo*. Bax activation and dissociation of Bax with Bcl-xL are common to both paradigms whereas the involvement of specific BH3-only proteins may vary contextually. Understanding the pathways and molecules involved in OPC cell death will provide insights towards developing new therapeutic interventions to promote survival of this lineage in brain and spinal cord injuries and in neurodegenerative diseases.
We thank Sukhwinder Singh for assistance with the Amnis ImageStream flow cytometry and Yuhui Jiang for technical assistance with surgeries for the H--I studies.
AUTHOR CONTRIBUTION
===================
Sophia Simonishvili conducted the experiments for the majority of the data shown with the exception of performing the proteomic analyses and the Amnis Imagestream flow cytometry. She also contributed the first draft of the manuscript. Mohit Raja Jain performed the proteomic analyses and contributed to the manuscript methods and results for these studies. Hong Li provided oversight and interpretation of the proteomic studies as well as the methods and results sections of the manuscript on these studies. Steven Levison provided oversight and assisted with the H--I studies and white matter dissections, as well as editorial contributions to the manuscript. Teresa Wood provided oversight of all studies and wrote the manuscript after the first draft.
FUNDING
=======
This work was supported by the National Institute of Neurological Disorders and Stroke \[grant numbers NS050742 (to T.L.W.) and P30NS046593 (to H.L.), the National Institute of Child Health and Development \[grant number HD052064\] and a grant from the Leducq Foundation to S.W.L.
| {
"pile_set_name": "PubMed Central"
} |
{.958}
{.959}
{.960}
| {
"pile_set_name": "PubMed Central"
} |
Introduction
============
Medical therapy is the initial treatment option for the vast majority of cases of the open angle glaucomas. The use of the topical medications exposes the patient to the side effects of the active components of the drops as well as those of the preservatives. Monotherapy with a single drop is the recommended initial treatment of glaucoma.[@b1-opth-12-2393] However, about half of these patients will require a second antiglaucoma agent after 2 years.[@b2-opth-12-2393] Multiple drops on the contrary may not only have a negative impact on adherence[@b3-opth-12-2393] but they can also expose the patients to higher amounts of preservatives, which play a fundamental role in the development of ocular surface disease in glaucoma patients.[@b4-opth-12-2393] Research has shown that as many as 60% of glaucoma patients receiving topical therapy can develop ocular surface disease.[@b5-opth-12-2393] The release of the fixed combinations has helped to improve adherence, reduce exposure to preservatives, and have similar hypotonic effect.[@b6-opth-12-2393]--[@b8-opth-12-2393]
The fixed combination of travoprost 0.004%/timolol 0.5% is marketed as DuoTrav (Alcon Laboratories, Fort Worth, TX, USA), and it has shown to have a more potent hypotensive effect than its individual constituents (travoprost monotherapy or timolol monotherapy).[@b9-opth-12-2393] DuoTrav also had a higher hypotensive effect when switching from previous prostaglandin monotherapy.[@b10-opth-12-2393] A study by Schuman et al[@b11-opth-12-2393] has shown that DuoTrav is as effective as the unfixed combination of travoprost 0.004% administered in the evening and timolol 0.5% administered twice daily.
Geltim is a long-acting gel-forming carbomer that contains timolol 0.1%. It was shown to be as effective as the standard aqueous timolol 0.5% with a safer cardiovascular profile.[@b12-opth-12-2393],[@b13-opth-12-2393]
In this study, we compared the efficacy of the travoprost-- timolol fixed combination (TTFC) with the concomitant use of travoprost and timolol 0.1% gel (Trav + Geltim).
Materials and methods
=====================
Study design
------------
This is a randomized prospective and comparative study that adhered to the tenets of the Declaration of Helsinki. The study was approved by the ethic committee of the University Hospital of Evros, Greece. All patients provided written informed consent before their participation in the study.
Study population
----------------
Thirty-three patients were enlisted, 16 (31 eyes) in the TTFC group (group 1) and 18 (31 eyes) in the Trav + Geltim group (group 2). The demographic characteristics of each group are summarized in [Table 1](#t1-opth-12-2393){ref-type="table"}.
Full ophthalmic examination was performed at the baseline visit, including medical history, distance best-corrected visual acuity (BCVA), IOP measurement, gonioscopy, slit lamp examination, fundoscopy, and Humphrey 24-2 white-on-white perimetry. At each subsequent visit, BCVA and IOP measurements were taken and the patients were asked about possible side effects of the eye drops. Patients on previous antiglaucoma drops were asked to stop their medication. The washout period was 14 days for a-agonists and 30 days for b-blockers and prostaglandins. Patients on previous treatment had the IOP measured 14 days after the discontinuation of the drops. If the IOP was ≥35 mmHg, the patient was started on the appropriate antiglaucoma medication and was removed from the study.
Exclusion criteria included age \<18 years, BCVA \<0.9 logMAR, glaucoma other than primary open angle, pseudoexfoliative and pigmentary, IOP at 09:00 \<23 and \>35 mmHg, and previous ocular surgery (except for uncomplicated phacoemulsification at least 6 months before the baseline visit).
Treatment-naïve patients and patients selected for enrollment after discontinuation of their previous medical therapy were randomly assigned to the TTFC or the Trav + Geltim group.
After the baseline examination, the patients were reviewed at 1 and 3 months. The IOP was measured by two examiners (VK, AK) with an electronic Goldmann tonometer at the following time points during the day: 09:00, 12:00, 15:00, and 18:00. Patients in the TTFC group were asked to instill the drops at 21:00, and the patients in the Trav + Geltim group were advised to instill the prostaglandin at 21:00 and the timolol 0.1% gel formulation at 08:00.
Statistical analyses of between-treatment group comparing the IOP responses with the drug regimens were performed using a paired *t*-test for both individual time points and the entire diurnal curve (average mean pressures measured throughout the day). The significance level was set at 0.05. Within-treatment group, changes for individual time point were assessed using repeated measures ANOVA with Bonferroni correction. All statistical analyses were performed using MS-Excel Professional Plus 2010 and MedCalc statistical program (version 9.6.2.0; MedCalc Software, Mariakerke, Belgium).
Results
=======
There was no significant difference in the mean IOP value between the two groups at baseline. The mean value was 26.83 mmHg for the TTFC group and 26.63 for the Trav + Geltim group (*P*=0.626). Both groups showed significant hypotensive effect from baseline at all time points at 1 and 3 months ([Table 2](#t2-opth-12-2393){ref-type="table"}). The mean IOP reduction from baseline was also significantly reduced ([Table 3](#t3-opth-12-2393){ref-type="table"}, [Figure 1](#f1-opth-12-2393){ref-type="fig"}). The mean IOP was calculated as the mean value of the measurements taken at the designated four time points.
When the two groups were compared, there was no significant reduction in the mean IOP at both 1 and 3 months from the baseline visit ([Table 4](#t4-opth-12-2393){ref-type="table"}, [Figure 2](#f2-opth-12-2393){ref-type="fig"}). The mean IOP for the TTFC group at 1 month was 16.74 mmHg and for the Trav + Geltim group 16.23 mmHg (*P*\<0.26). At 3 months, the mean IOP for the TTFC group was 16.39 mmHg and for the Trav + Geltim group 16.27 mmHg (*P*\<0.44).
The two groups showed similar efficacy in lowering the IOP at all time points at 1 and 3 months except for the 18:00 time point where the Trav + Geltim group was more potent in lowering the IOP ([Table 5](#t5-opth-12-2393){ref-type="table"}, [Figure 3](#f3-opth-12-2393){ref-type="fig"}). The mean IOP for the TTFC group at 18:00 was 17.13 mmHg at 1 month vs 16.25 mmHg for the Trav + Geltim group (*P*=0.014) and 16.27 mmHg vs 15.5 mmHg at 3 months for the two groups, respectively (*P*=0.002).
Discussion
==========
In this study, we compared the efficacy of the fixed combination of travoprost 0.004%/timolol 0.5% (administered in the evening) to its constituents travoprost 0.004% (administered in the evening) and timolol 0.1% (administered in the morning). The efficacy and tolerability of the two separate drugs (travoprost and timolol gel-forming solution) have been investigated in previous articles[@b14-opth-12-2393],[@b15-opth-12-2393] and were not part of the this study.
Research has shown that fixed combinations are generally less potent in their hypotensive effect when compared with their constituents[@b16-opth-12-2393] although this difference is insignificant and in some cases the fixed combination may provide a better hypotensive effect[@b17-opth-12-2393] presumably due to the better convenience and adherence and the elimination of the washout effect of the second drop.
Our results have shown that the unfixed combination achieved a slightly better hypotensive effect although it did not reach statistical significance except for the 18:00 time point for which the Trav + Geltim group had lower IOPs at 1 and 3 months. A similar study by Nucci et al[@b18-opth-12-2393] that compared the IOP-lowering effect of the fixed combinations of latanoprost, travoprost, and bimatoprost with timolol to the unfixed concomitant use of the prostaglandins and timolol 0.1% gel-forming carbomer showed that the concomitant use of the drops offered a significant hypotensive effect. The differences in the results can be explained by different methodology that we used in our study. The patients that were enrolled in our study were either treatment-naïve or if they had already been on drops they were asked to discontinue them before they were started on the fixed or the unfixed combination. In the study by Nucci et al, the patients were already on the fixed prostaglandin/timolol combination and were switched to the unfixed treatment without a washout period. Furthermore, Nucci et al did not have separate arms for the fixed and the unfixed groups, but the hypotensive effect of the three fixed prostaglandin/timolol groups was compared with the unfixed combinations after the switch in the treatment. Finally, we enrolled 33 eyes in total, whereas in the study be Nucci et al there were nine patients in the travoprost/timolol group.
Two more studies have compared the efficacy of the travoprost/timolol fixed combination with its separate components.[@b11-opth-12-2393],[@b19-opth-12-2393] Both studies compared the efficacy of the fixed travoprost 0.004%/timolol 0.5% with the concomitant use of travoprost 0.004% instilled in the evening and timolol 0.5% administered in the morning. These studies confirmed that both regimes have equal efficacy although the concomitant use of the two drops showed a slightly better hypotensive effect.
The efficacy of the timolol 0.5% aqueous solution has been compared with the timolol 0.1% gel formulation. Both showed similar effects in terms of IOP lowering. The timolol 0.1% gel, however, was less adsorbed in the systemic circulation and affected the cardiovascular system to a lesser degree. The pulmonary function remained unaltered with both formulations. Thus, the timolol 0.1% gel had a higher risk-to-benefit ratio.[@b13-opth-12-2393],[@b20-opth-12-2393]
In our study, the concomitant use of travoprost and timolol had slightly better hypotensive effect, which can be explained by the fact that the travoprost was instilled in the evening in the unfixed combination group although evidence has shown that travoprost has a similar hypotensive effect whether administered in the morning or in the evening.[@b21-opth-12-2393] On the contrary, the use of timolol in a gel formulation in the morning can have its peak effect several hours after instillation and can explain the significant hypotensive effect at the 18:00 time point. The single morning dosage of the timolol gel avoids the washout effect of the second evening drop in the concomitant regimes with morning and evening timolol dosing.
In conclusion, the fixed travoprost/timolol combination provides similar hypotensive effect as the separate use of the two active components. The concomitant use of the drops had significant effect only in the early evening time point.
**Disclosure**
The authors report no conflicts of interest in this work.
{#f1-opth-12-2393}
{#f2-opth-12-2393}
{#f3-opth-12-2393}
######
Demographic characteristics
TTFC Trav + Geltim
----------------------- -------- ---------------
Age
Mean 65.25 65.24
Range 57--79 59--77
Gender
Male 5 8
Female 11 10
Diagnosis (n=31 eyes)
OHT 13 15
POAG 18 16
Treatment-naïve eyes 8 6
Eyes receiving drops 23 25
Baseline IOP (mean) 26.83 26.63
**Abbreviations:** OHT, ocular hypertension; POAG, primary open angle glaucoma; TTFC, travoprost--timolol fixed combination; Trav + Geltim, the concomitant use of travoprost and timolol 0.1% gel.
######
Mean IOP before treatment and after treatment (1 and 3 months) for both groups at different time points
TTFC Trav + Geltim
------- --------------- ------- ------- --------- ------- ------- ------- ---------
09:00 26.83 17.6 17.34 \<0.001 26.63 16.93 17.5 \<0.001
12:00 26.23 16.1 15.9 \<0.001 26.23 15.93 16.2 \<0.001
15:00 26.04 16.13 16.03 \<0.001 26.04 15.96 15.86 \<0.001
18:00 25.76 17.13 16.27 \<0.001 25.8 16.25 15.5 \<0.001
**Note:** IOP: measured in mmHg.
**Abbreviations:** TTFC, travoprost--timolol fixed combination; Trav + Geltim, the concomitant use of travoprost and timolol 0.1% gel.
######
Mean IOP reduction from baseline
Baseline IOP SD IOP at 1 month SD *P*-value IOP at 3 months SD *P*-value
--------------- -------------- ------ ---------------- ------ ----------- ----------------- ------ -----------
TTGC 26.83 2.84 16.74 1.34 \<0.0001 16.39 0.68 \<0.0001
Trav + Geltim 26.63 2.84 16.23 1.45 \<0.0001 16.27 0.66 \<0.0001
**Abbreviations:** TTFC, travoprost--timolol fixed combination; Trav + Geltim, the concomitant use of travoprost and timolol 0.1% gel.
######
Mean IOP hypotensive effect of TTFC vs Trav + Geltim
Baseline IOP *P*-value 1 month *P*-value 3 months *P*-value
--------------- -------------- ----------- --------- ----------- ---------- -----------
TTFC 26.83 0.626 16.74 0.26 16.39 0.44
Trav + Geltim 26.63 16.23 16.27
**Abbreviations:** TTFC, travoprost--timolol fixed combination; Trav + Geltim, the concomitant use of travoprost and timolol 0.1% gel.
######
Comparison of the two groups at the individual time points
Time points 1 month from baseline 3 months from baseline
------------- ----------------------- ------------------------ ------- ------- ------- -------
9:00 17.6 16.93 0.27 17.34 17.5 0.39
12:00 16.1 15.93 0.94 15.9 16.2 0.25
15:00 16.13 15.86 0.93 16.03 15.86 0.27
18:00 17.13 16.25 0.014 16.27 15.5 0.002
**Abbreviations:** TTFC, travoprost--timolol fixed combination; Trav + Geltim, the concomitant use of travoprost and timolol 0.1% gel.
| {
"pile_set_name": "PubMed Central"
} |
{#sp1 .600}
{#sp2 .600-1}
{#sp3 .600-2}
{#sp4 .601}
{#sp5 .602}
{#f1 .600-1}
{#f2 .600-1}
{#f3 .600-1}
{#f4 .600-1}
{#f5 .600-2}
{#f6 .600-2}
| {
"pile_set_name": "PubMed Central"
} |
INTRODUCTION {#s1}
============
The family of 14-3-3 proteins plays a pivotal role in integrating cellular survival signaling and thereby are key players in determining the cell fate \[[@R1], [@R2]\]. Through their role in binding multiple phospho-client proteins within the cell, 14-3-3 proteins regulate many important signaling events. In particular, 14-3-3 proteins maintain the proliferative capacity of the cell by supporting the efficient activation of the Raf-MAPK signaling cascade and also support cell survival through the PI3K-AKT signaling cascade \[[@R1], [@R2]\], key requirements of cancer cells \[[@R3]\].
Enhanced expression of 14-3-3 proteins has been detected in many human cancers including lung \[[@R4]\], head and neck \[[@R5]\], breast \[[@R6]\] and ovarian cancer \[[@R7]\] and correlates with more aggressive tumors and poor prognosis \[[@R8]\]. Over-expression of 14-3-3 isoforms has been shown to contribute to neoplastic transformation by stimulating Raf-MAPK and PI3K signaling \[[@R9]\]. Down-regulation of 14-3-3ζ in head and neck cancer cells \[[@R10]\] and also lung cancer cells \[[@R11]\] renders cells more sensitive to chemotherapy, supporting the notion that cancer cells utilize mechanisms that are 14-3-3 dependent. These experimental and clinical observations suggest that 14-3-3 proteins represent an addiction for many cancers and consequently are an attractive target for anti-cancer therapy \[[@R8], [@R12]\].
In search of 14-3-3 inhibitors, several studies have identified small molecules that function as mimics of 14-3-3 binding partners, blocking the interaction of 14-3-3 proteins with phospho-clients such as c-Abl \[[@R13]\] and Raf-1 \[[@R14]\], and thus behaving as competitive antagonists. These small molecule inhibitors of 14-3-3 proteins exhibit anti-cancer activity in cell-based assays but are limited by their specificity, ability to penetrate cells and also the need for high concentrations to effectively compete with abundant endogenous 14-3-3-binding proteins.
We have taken a novel approach to developing 14-3-3-targeting molecules by exploiting the requirement of 14-3-3 proteins to function as dimers \[[@R15]\]. 14-3-3 proteins are intrinsically dimeric in nature, a characteristic which is obligatory for many of their biological functions \[[@R16]\]. We showed previously that the dimeric status of 14-3-3 proteins is subject to regulation, with the critical step being phosphorylation of Ser58 (numbering relates to the ζ isoform), a site otherwise buried in the dimer interface \[[@R17]\]. The Ser58 dimer interface phosphorylation site is conserved in five of the seven mammalian isoforms and is recognized by several kinases \[[@R18]-[@R20]\] but, importantly, we showed it only becomes accessible after the endogenous lipid sphingosine binds to the 14-3-3 protein \[[@R15]\]. Once Ser58 is phosphorylated, the dimeric structure of the 14-3-3 protein is disrupted and its function is inhibited \[[@R17]\]. Thus, through this mechanism, sphingosine serves as a key regulator of dimeric 14-3-3 protein function and induces apoptosis \[[@R15]\].
Similarly, we found that the synthetic sphingosine analogue, FTY720, also renders 14-3-3 phosphorylatable \[[@R15]\]. FTY720, also known as Fingolimod and Gilenya^TM^, is in clinical use to induce immune suppression in the treatment of multiple sclerosis, but has reported anti-cancer characteristics in many experimental systems \[[@R21]\]. As an immune suppressant, FTY720 is a pro-drug which relies on conversion to a phosphorylated form for its effects, predominantly as a sphingosine-1-phosphate receptor antagonist \[[@R21]\]. The anti-cancer effects of FTY720 however are associated with the unphosphorylated pro-drug form \[[@R21]\] and we have shown that these are mediated in part by its effect on 14-3-3 proteins \[[@R15]\]. Owing to its immunosuppressant properties, FTY720 is not a suitable anti-cancer therapy but understanding the molecular basis of its anti-cancer action has provided us with a rationale to identify more sphingosine-like compounds that target 14-3-3 specifically.
We have now surveyed other sphingosine-like molecules and demonstrate that *N*-alkylated trimethyl ammonium (TMA) molecules act as sphingo-mimics to disrupt 14-3-3 dimers and induce apoptosis in Jurkat cells. Because these compounds are unsuitable drug candidates, we have combined the chemical nature of the TMAs with the FTY720 backbone to generate a novel chemical series that exhibit apoptotic characteristics through a 14-3-3-mediated mechanism *in vitro*. Furthermore, we have demonstrated that the most potent of this chemical series induce mitochondrial-mediated apoptosis at low micromolar concentrations *in vitro* and rapidly elicit a signaling cascade that corresponds kinetically to the disruption of dimeric 14-3-3 functions. In a mouse xenograft model of human non-small cell lung cancer where 14-3-3 is over-expressed, RB-012, our most active compound significantly reduced tumor growth without adverse effects on the animals. Our data show that RB-012 is soluble, readily taken up by cells, effective at low concentration and unlike other previously reported 14-3-3-directed small molecules, acts via a non-competitive mechanism. This compound and the mechanism that underlies its activity provide proof-of-principle for our approach to developing a new class of 14-3-3-targeting small molecule therapeutics for cancer treatment.
RESULTS {#s2}
=======
Trimethylammonium compounds render the 14-3-3 dimer interface accessible to kinases and induce mitochondrial apoptosis {#s2_1}
----------------------------------------------------------------------------------------------------------------------
We previously established that non-acylated sphingolipids with a net positive charge are capable of rendering 14-3-3 phosphorylatable \[[@R15]\]. To identify new compounds that are capable of rendering 14-3-3 phosphorylatable but are not susceptible to sphingolipid metabolism, we assessed non-sphingoid cationic lipids such as quaternary ammonium compounds for effects on 14-3-3 phosphorylatability. In our *in vitro* system using recombinant 14-3-3ζ as substrate and PKA catalytic subunit as the phosphorylating enzyme, we found that trimethylammonium (TMA) molecules with an alkyl chain of 14 carbons or longer rendered 14-3-3 phosphorylatable, whereas molecules with a shorter alkyl chain were ineffective (Figure [1A & 1B](#F1){ref-type="fig"}).
![**A.** Structures of the trimethylammonium (TMA) compounds assessed for 14-3-3 modulating activity. **B.** Phosphorylation of 14-3-3 by PKA *in vitro* in presence or absence of TMA compounds at the concentrations shown. The upper panel is \[^32^P\]-phospho-labeled 14-3-3ζ and the lower panel is Coomassie stained 14-3-3 protein. **C.** Effect of TMA compounds on Jurkat cell after 20 h treatment at the concentrations shown. Cell viability is shown in open bars and TMRE negative staining cells are shown in black bars. The error bars show the range of duplicate determinations: and the results are representative of multiple experiments.](oncotarget-06-14522-g001){#F1}
We next tested whether the *N*-alkyl TMA series could induce cell death of Jurkat cells. Cell death was assessed using flow cytometry after 20 h of treatment by analysis of cell viability together with tetramethylrhodamine ethyl ester (TMRE) staining to monitor mitochondrial permeability transition (ΔΨ~M~), an event commonly associated with programmed cell death (Figure [1C](#F1){ref-type="fig"}). The longer chain *N*-alkyl TMA molecules induced mitochondrial permeability transition at 5μM, consistent with their ability to render 14-3-3 phosphorylatable. This suggests that like sphingosine, these longer chain TMA molecules can regulate dimeric 14-3-3 proteins to disrupt their functions in cells.
To elucidate the mechanism of the TMA compound\'s effect on 14-3-3 protein, we carried out dose-response studies with C~16~-TMA, (cetyltrimethylammonium bromide, denoted as CTAB) in 14-3-3 phosphorylation reactions. We observed a dose-dependent increase in 14-3-3 Ser58 phosphorylation with increasing concentration of CTAB (Figure [2A](#F2){ref-type="fig"}) and using recombinant S58A 14-3-3ζ protein, confirmed that Ser58 was the only site phosphorylated by PKA (Figure [2A](#F2){ref-type="fig"}), consistent with our previous studies with sphingosine, and FTY720 \[[@R15]\]. Additionally, we observed no effect of CTAB on PKA catalytic subunit activity (as determined using kemptide phosphorylation) over the same concentration range (data not shown), reinforcing that CTAB has a direct effect on the 14-3-3 protein, not on enzyme activity.
{#F2}
Long chain *N*-alkylated TMA molecules are amphiphilic with a cationic head group and a long alkyl chain and can therefore behave as cationic detergents. The reported critical micelle concentration for CTAB in water is 1 mM but in 20 mM Tris-Cl pH 7.0, 10 mM NaCl the CMC decreases to \~ 0.15 mM \[[@R25]\]. We therefore assessed CTAB\'s effect on 14-3-3 phosphorylation more closely to determine whether non-specific denaturation of 14-3-3 was contributing to the phosphorylatability of 14-3-3 by PKA. In extended dose-response studies, CTAB induced 14-3-3 phosphorylation at low micromolar concentrations, plateauing at 25 μM, but above 75 μM a further increase in 14-3-3 phosphorylation was observed (Figure [2B](#F2){ref-type="fig"}). This implies that at low micromolar concentrations, CTAB binds discrete sites in 14-3-3, thereby conformationally altering the dimer interface to reveal the Ser58 phosphorylation site accessible to PKA, whereas at higher concentrations of CTAB, approaching CMC values, a more cooperative effect is seen, consistent with generalized protein denaturation. Consistent with this, PKA catalytic subunit activity was significantly inhibited at higher CTAB concentrations (Figure [2B](#F2){ref-type="fig"}, kemptide phosphorylation was reduced by 90% in the presence of 100 μM CTAB compared with vehicle alone) presumably due to the denaturation of the enzyme. Similar effects have been observed for C~14~-TMA binding to bovine serum albumin, with saturable binding to specific high affinity sites at low C~14~-TMA concentrations and cooperative non-specific binding at high concentrations \[[@R26]\].
To test whether the mitochondrial permeability transition induced by CTAB is associated with apoptosis, akin to the effect of sphingosine and FTY720 \[[@R15]\], we examined caspase-3 activation, a characteristic marker of apoptosis. As shown in Figure [2C](#F2){ref-type="fig"}, CTAB does induce caspase-3 activation in Jurkat cells and, as with FTY720 \[[@R27]\], Jurkat cells over-expressing Bcl-2 were protected from CTAB-induced cell death, confirming that the CTAB-induced apoptosis is mediated by the mitochondria. Thus, these data show that at concentrations well below the CMC, *N-*alkyl TMA molecules are able to bind to 14-3-3 proteins in cells, leading to disruption of 14-3-3 dimers and subsequent induction of mitochondrial apoptosis.
Development of the novel RB sphingomimetics and their modulation of 14-3-3 dimerisation and induction of mitochondrial apoptosis {#s2_2}
--------------------------------------------------------------------------------------------------------------------------------
Our results with CTAB suggest that at concentrations below its CMC, this molecule can bind to discrete sites in 14-3-3 proteins, causing conformational changes at the dimer interface and thereby allowing kinases access to the Ser58 phosphorylation site. CTAB however, is not an ideal anti-cancer candidate owing to its strong detergent properties and known toxic effects in mice. Data generated by the Developmental Therapeutics Program at the National Cancer Institute indicate that CTAB is toxic in mice when dosed above 10 mg/kg over a five day treatment, (mined from \<<http://dtp.nci.nih.gov/dtpstandard/dwindex/index.jsp>\>). The clinical drug FTY720 has previously been used in mice at 10 mg/kg with no reported adverse effects, although its clinical application is as an immunosuppressant, an activity associated with the phosphorylated form of FTY720 \[[@R21]\]. In order to generate more 14-3-3-selective agents we sought to combine the charged quaternary ammonium group of the TMA molecules with the clinically approved FTY720 alkyl chain, and thereby generated a panel of new quaternary amine derivatives of FTY720, denoted here as RB-011, -012, -066, -067 and -068 (Figure [3A](#F3){ref-type="fig"}). Unlike sphingosine and FTY720, these analogues lack hydroxyl sites for phosphorylation by sphingosine kinases and therefore cannot be converted to immunosuppressive phospho-species.
![**A.** Structures of the TMA-FTY720 hybrid RB compounds and their effect on *in vitro* 14-3-3ζ phosphorylation by PKA at the concentrations as shown. The upper panel is \[^32^P\]-phospho-labeled 14-3-3ζ (\[^32^P\]) and the lower panel Coomassie-stained 14-3-3ζ protein (C). **B.** Effect of 5 μM RB molecules on viability (open bars) and caspase-3 activation (black) of Jurkat cells after 5 h treatment. **C.** Effect of 5 μM RB compounds on viability (open bars) and Annexin V staining (hashed bars) of Jurkat cells after 24 h treatment. The error bars show the range of duplicate determinations and the results are representative of multiple experiments.](oncotarget-06-14522-g003){#F3}
Initially we tested the five RB compounds in our *in vitro* 14-3-3 phosphorylation assay. In dose response studies we found RB-011 and RB-012 were the most effective at rendering 14-3-3 phosphorylatable by PKA, whereas the other compounds were much less effective (Figure [3A](#F3){ref-type="fig"}). We then assessed the ability of the RB compounds to induce apoptosis of Jurkat cells compared with FTY720, CTAB and RB-015 \[[@R22]\], a hydroxylated version of RB-012 (in which the hydroxyl group is at the 4 position of the piperidinium ring, ref. [@R22]). At 5 μM, RB-011 and -012 were readily able to activate apoptosis as determined by caspase-3 activation within 5 h (Figure [3B](#F3){ref-type="fig"}), and Annexin V presentation and loss of viability at 24 h (Figure [3C](#F3){ref-type="fig"}), consistent with their ability to elicit 14-3-3 phosphorylation at low concentration. Importantly RB-011 and -012 were more potent than either CTAB or FTY720. None of the other RB compounds were as effective at inducing Jurkat cell apoptosis consistent with their effect on 14-3-3 phosphorylation and indicating that ring substitution is not tolerated for these activities (Figure [3](#F3){ref-type="fig"}).
The potency and apoptotic effects of RB-011 and -012 on cells were examined more closely. We determined the ED~50~ for apoptosis induction in Jurkat cells by assessing the activation of caspase-3 at 5 h, before any significant loss of cell viability (Figure [3B](#F3){ref-type="fig"}). RB-012 was slightly more potent than RB-011 at initiating apoptosis, with an ED~50~ of 2 μM compared with 3 μM for RB-011 (Figure [4A](#F4){ref-type="fig"}). Biochemical characterization of RB-treated Jurkat cells revealed that after 4 h of treatment PARP cleavage had occurred, consistent with the commitment to apoptosis (Figure [4B](#F4){ref-type="fig"}, upper panels). Using phospho-specific antibodies, we detected active stress-activated protein kinases, p38 and JNK, after 4 h of treatment with RB-011 or -012 (Figure [4B](#F4){ref-type="fig"}, second and third panel). Immunoblotting of 14-3-3 revealed a slightly lower molecular weight form of 14-3-3 after 4 h of RB-011 or -012 treatment (Figure [4B](#F4){ref-type="fig"}, bottom panel) probably associated with caspase cleavage as 14-3-3 proteins have previously been shown to be susceptible \[[@R28]\].
{#F4}
These analyses indicate that the activation of apoptosis in response to the RB compounds is rapid, and detectable by 4 h of treatment. Additionally and importantly, over-expression of Bcl-2 completely protected the Jurkat cells from the RB compounds (Figure [4C](#F4){ref-type="fig"}) as determined by analysis of viability and caspase activation after 20 h of treatment, confirming that the compounds initiate signaling upstream of mitochondrial-mediated apoptosis.
The novel RB sphingomimetics cause rapid inhibition of PI3K-AKT and MAPK signaling {#s2_3}
----------------------------------------------------------------------------------
To characterize the signaling changes associated with RB treatment of Jurkat cells, we carried out time-course studies using RB-012 and prepared cytosolic extracts for immunoblotting. Strikingly, phospho-specific antibodies revealed rapid dephosphorylation of both ERK and AKT (within 1 h) upon RB-012 treatment (Figure [5A](#F5){ref-type="fig"} first and third panels), indicating down-regulation of the MAPK and PI3K signaling pathways respectively. Activation of SAPKs p38 and JNK was detected but not until 2 h post RB-012 treatment (Figure [5A](#F5){ref-type="fig"} fourth and fifth panels), after ERK and AKT inactivation.
{#F5}
In the same time-course studies, apoptotic markers were analyzed by immunoblotting. BID cleavage is associated with diverse apoptotic stimuli and has been shown to occur after mitochondrial permeability transition and apoptotic commitment via a caspase-3 mediated process in Jurkat cells \[[@R29]\]. We analyzed BID cleavage after RB-12 treatment and found that BID breakdown is detected at 3 h of RB-012 treatment (Figure [5B](#F5){ref-type="fig"}, first panel). Immunoblotting for 14-3-3 over the time-course revealed that the lower molecular weight form of 14-3-3 protein is detected at 2 h (Figure [5B](#F5){ref-type="fig"}, second panel), consistent with activation of caspase-3 (Figure [5B](#F5){ref-type="fig"}, third panel). Additionally, PARP cleavage was detectable at 2 h (Figure [5B](#F5){ref-type="fig"}, fourth panel). Thus commitment to apoptosis in response to RB-012 occurs within 2-3 h of treatment, after the initial effects on MAPK and PI3K signaling.
RB sphingomimetics inhibit A549 lung cancer cell growth *in vitro* and *in vivo* {#s2_4}
--------------------------------------------------------------------------------
RB-011 and -012 exhibit apoptotic activity on Jurkat cells. Their effects are mediated at least in part by disruption of functional 14-3-3 dimers and inactivation of AKT and MAPK signaling pathways. These characteristics are desirable for a new anti-cancer therapy. We therefore sought to validate these compounds on human cancer cells where 14-3-3 over-expression has been implicated in tumor aggression. Of the multiple cancer types where 14-3-3 over-expression has been detected, non-small cell lung cancer (NSCLC) has been identified as a cancer in which the degree of 14-3-3ζ over-expression correlates strongly with poor patient survival and disease severity \[[@R4]\]. We therefore assessed the effect of the RB-011 and -012 molecules on the NSCLC line A549 (Figure [6](#F6){ref-type="fig"}).
{#F6}
First, we performed survival assays using MTS and found that the RB compounds reduced cell viability in a dose-dependent manner (Figure [6A](#F6){ref-type="fig"}). As with the Jurkat cells, RB-012 was slightly more potent than RB-011 with an IC~50~ of 5.5 μM compared with 7 μM for RB-011 (Figure [6A](#F6){ref-type="fig"}). We assessed the ability of the RB compounds to induce A549 cells apoptosis by monitoring caspase-3 activation. Both compounds induced caspase-3 activation with RB-012 showing greater potency at 10 μM (Figure [6B](#F6){ref-type="fig"}). The compounds were then tested for their effect on A549 colony formation in soft agar, a measure of neoplastic growth. Both compounds reduced colony formation in a dose-dependent manner with RB-012 again showing slightly greater potency (Figure [6C](#F6){ref-type="fig"}).
A549 cells harbor oncogenic K-Ras and consequently constitutive MAPK signaling is characteristic of this cell line. We analyzed the effect of RB-012 on MAPK signaling in the A549 cells *in vitro* and observed a rapid reduction in phospho-ERK in response to RB-012 (Figure [6D](#F6){ref-type="fig"}). Thus given RB-012\'s activity *in vitro*, we then studied its effects *in vivo*.
To assess any potential toxicity, BALB/c nude mice were administered either RB-012 or FTY720, delivered in saline (0.9%), daily by i.p. injection at 5mg/kg or 10mg/kg body weight, doses previously reported for FTY720 administration \[[@R21]\]. Over a course of 28 days there were no adverse effects associated with RB-012 treatment with the exception of apparent pain and abdominal cramping in the mice immediately after injection which subsided within 10-20 min. Similar effects were also observed with 5mg/kg/day FTY720 treatment. Importantly there was no pathology associated with the abdominal cramping suggesting a physiological response and over time the mice developed a tolerance to this effect. The body weight of RB-012 treated mice was unaltered over the course of the toxicity study and histology of tissues (liver, lung, kidney, heart and spleen) collected during the treatment course was unaffected ([supplementary Figure 2A](#SD1){ref-type="supplementary-material"}) indicating that the RB-012 had no major toxic effects. Additionally, bone marrow samples from RB-012 treated mice exhibited no alterations indicating no changes in steady state haemopoiesis ([supplementary Figure 2B](#SD1){ref-type="supplementary-material"}). In the same toxicity study we found that FTY720 dosed at 10mg/kg body weight was poorly tolerated and caused cardiac arrhythmia with one unexplained fatality. Toxicity has previously been noted with i.p. injection of FTY720 at 10mg/kg/day \[[@R30]\] and we were therefore unable to use this dose of FTY720 in the subsequent xenograft study.
To assess the effect of RB-012 on human lung cancer growth *in vivo,* A549 cells were implanted subcutaneously on the flanks of BALB/c nude mice and tumors were allowed to grow until they had reached a volume of 100 mm^3^. RB-012, FTY720 or saline was then administered daily to assess the effect on tumor growth. Owing to the physiological abdominal cramping response observed in the toxicity study, we were required on ethical grounds to use a dose escalation regime for administering both RB-012 and FTY720 to allow the mice to develop tolerance. Initially, dosing was limited to 2mg/kg body weight for two days and then increased to 5mg/kg daily for two weeks which minimized the cramping. Tumor volume was monitored twice a week and after two weeks of dosing at 5mg/kg body weight, mice administered RB-012 had tumors that were 20% smaller than mice receiving saline (Figure [6E](#F6){ref-type="fig"}). Interestingly, mice receiving FTY720 showed no reduction in tumor size (Figure [6E](#F6){ref-type="fig"}) in contrast to a previous report employing A549 xenografts \[[@R31]\] albeit in a different mouse strain (Balb/c SCID) and using a different route of FTY720 administration (by oral gavage).
As RB-012 was better tolerated than FTY720, the dosing of RB-012 treated mice was increased to 10mg/kg body weight for a further week with no adverse effects on the mice. The tumors on the RB-012-treated mice continued to grow more slowly than those on saline treated mice and at the end of the treatment course were 30% smaller than those of saline-treated mice (Figure [6E](#F6){ref-type="fig"}). Thus RB-012 effectively reduced A549 tumor growth whereas FTY720 was ineffective in this study.
To provide evidence that the anti-cancer effect of RB-012 in the A549 xenograft study involved disruption of dimeric 14-3-3, we analyzed the excised tumor tissue for MAPK activity. Quantitative immunofluorescence analysis showed that tumors treated with RB-012 had significantly less phospho-ERK than saline-treated tumors (Figure [6F](#F6){ref-type="fig"}), consistent with the *in vitro* results on A549 cells (Figure [6D](#F6){ref-type="fig"}). Thus RB-012 effectively reduces A549 xenograft growth by a mechanism involving disruption of the oncogenic signaling in these cells, supporting the notion that RB-012 reduces tumor growth by disruption of dimeric 14-3-3 function.
DISCUSSION {#s3}
==========
The widespread over-expression of 14-3-3 proteins in human tumors has highlighted the significance of 14-3-3 proteins in cancer development \[[@R4]-[@R8]\]. Increased 14-3-3 expression provides cancer cells with enhanced protection against apoptotic mediators making 14-3-3 proteins an attractive target for anti-cancer drug development. Several groups have identified molecules with the ability to block the binding of 14-3-3 proteins to client proteins \[[@R13], [@R14], [@R32]\]. Without exception these molecules compete with client proteins for binding in the amphipathic groove \[[@R32]\]. We have developed a novel approach that exploits discrete sphingolipid binding site(s) on 14-3-3 that when occupied, causes dimer disruption and loss of function \[[@R15]\].
We have now identified *N*-alkylated trimethylammonium (TMA) molecules as modulators of 14-3-3 *in vitro* with a corresponding capacity to induce apoptosis of Jurkat cells (Figure [1](#F1){ref-type="fig"}). From a chemical series of trimethylammonium compounds we determined that the length of the alkyl chain is an important factor in determining the effect on 14-3-3 modulation and Jurkat cell apoptosis. Long-chain TMAs have the greatest potency both in *in vitro* 14-3-3 phosphorylation assay and inducing Jurkat cell apoptosis (Figure [1](#F1){ref-type="fig"}). We have demonstrated that at low micromolar concentrations (below the CMC), CTAB modulates 14-3-3ζ, allowing phosphorylation by PKA at Ser58 in the dimer interface, in an analogous fashion to sphingosine and its analogues \[[@R15]\]. Additionally, the long-chain TMAs induce apoptosis in Jurkat cells at concentrations well below CMC via the mitochondrial pathway (Figure [2](#F2){ref-type="fig"}). These data support the notion that long-chain TMAs induce apoptosis by interfering with dimeric 14-3-3 in the cell.
CTAB was previously identified in a high-throughput screen for anti-cancer agents as being able to inhibit the growth of a panel of head and neck cancer (HNC) cell lines and reduce mitochondrial membrane potential (ΔΨ~M~) but the mechanism of action was unknown \[[@R33]\]. Consistent with our results, the apoptotic effect of CTAB was related to the length of the alkyl group \[[@R33]\]. Furthermore, CTAB administration at 5 mg/kg retarded the growth of HNC xenografts *in vivo* \[[@R33]\]. 14-3-3ζ over-expression has been reported in HNCs and correlates with poor patient prognosis \[[@R5]\] and knock-down of 14-3-3ζ in HNC cell lines by RNAi inhibited cell growth and increased apoptosis in response to chemotherapy \[[@R10]\]. Our results strongly suggest that the effects of CTAB on HNC lines are attributable to 14-3-3 dimer disruption.
CTAB has also been assessed for anti-cancer activity in the National Cancer Institute Developmental Therapeutics Program ([www.dtp.nci.nih.gov](http://www.dtp.nci.nih.gov/)) in mouse cancer models of lymphocytic leukemia (P388 and L1210) and melanoma (B16). A modest therapeutic effect was seen in the P388 model at low doses, at 3 and 6 mg/kg body weight per day i.p. 20 % more of the mice survived to 30 days of treatment compared with untreated controls. At high dose (\>10 mg/kg body weight) CTAB was toxic (after 5 days treatment, data extracted from DTP at NCI) and given its strong surfactant properties is not suitable as a drug candidate. However, CTAB\'s ability to disrupt 14-3-3 dimers, allowing phosphorylation in the dimer interface, coincident with its ability to induce mitochondrial apoptosis and experimental anti-cancer activity, provide initial validation for our strategy to identify compounds with 14-3-3-dimer disrupting activity as potential anti-cancer agents.
In the present study we designed a series of new compounds that combine the quaternary ammonium group of the alkylated TMA molecules with the backbone of FTY720. The compounds exhibit varying ability to render 14-3-3 phosphorylatable *in vitro* coinciding with their ability to induce apoptosis of Jurkat cells (Figure [3](#F3){ref-type="fig"}). In particular RB-011 and RB-012 have a potent ability to induce apoptosis of Jurkat cells within 4 h of treatment and have ED~50~ values in the low micromolar range (Figure [4](#F4){ref-type="fig"}). Time-course studies revealed rapid inactivation of ERK and AKT signaling within 1 h of treatment of Jurkat cells with RB-012, followed by activation of SAPKs, JNK and p38 by 2 h (Figure [5A](#F5){ref-type="fig"}). The apoptotic cascade then proceeds with caspase-3 processing, PARP and BID cleavage (Figure [5B](#F5){ref-type="fig"}). These are important findings that reveal the cellular effects of RB-012 occur in a time-dependent manner with primary effects on cell signaling occurring prior to apoptotic commitment.
14-3-3 proteins are known to play a key role in integrating survival signaling within the cell \[[@R1]-[@R2]\]. When mutant forms of 14-3-3 that are non-competent for phospho-client binding are expressed in cells they heterodimerise with endogenous isoforms and act as functionally monomeric proteins, causing disruption of MAPK signaling and activation of SAPK \[[@R34], [@R35]\]. These effects phenocopy what we have observed with RB-012, and match the known differential regulation of upstream MAPK Raf-1 \[[@R36]\] and SAPKK, ASK-1 \[[@R37]\] and MEKK2 \[[@R38]\] by dimeric 14-3-3 proteins. Additionally, recent studies have confirmed that 14-3-3 proteins are required for PI3K activity \[[@R39]\]. The downstream inactivation of AKT signaling in response to RB-012 is therefore entirely consistent with 14-3-3\'s role in PI3K function. Our studies reveal that targeting 14-3-3 proteins using our sphingomimetic approach provides a potent means to disrupt functional 14-3-3 dimers which is desirable for an effective anti-cancer drug.
Several isoforms of 14-3-3 have been associated with the transformed phenotype, both experimentally; over-expression of 14-3-3γ, β and ζ have been shown to cause cellular transformation \[[@R9], [@R40], [@R41]\], and in clinical samples (ζ, β, ε, γ, η and τ) \[[@R4]-[@R8]\]. However, the 14-3-3ζ isoform is the most commonly up-regulated in cancer \[[@R4]-[@R8]\]. This may be due to a specific role for the zeta isoform although to date none are known. More likely, the specific regulation of 14-3-3ζ expression by a micro-RNA (miR-) 451 may be the underlying cause \[[@R42]\]. The negative regulation of 14-3-3ζ by miR-451 has been elegantly demonstrated in breast cancer cells where it was shown that tamoxifen treatment leads to loss of miR-451 expression and up-regulation of 14-3-3ζ, correlating with increased disease severity \[[@R42]\]. Intriguingly, miR-451 expression has also been shown to be low in NSCLC tissue compared with non-cancerous lung tissue, and up-regulation of miR-451 in A549 cells reduced cell growth and increased the cells susceptibility to cisplatin \[[@R43]\]. These results mirror the over-expression 14-3-3ζ seen in NSCLC tissue which correlates with disease severity, and similarly targeted knock-down of 14-3-3ζ using RNAi in A549 cells also increased the cells sensitivity to cisplatin \[[@R4]\]. Thus the reciprocal expression of miR-451 and 14-3-3ζ may be important prognostic markers for disease severity and cancer progression in NSCLC.
As 14-3-3ζ expression is closely linked with NSCLC disease, we tested RB-011 and RB-012 for anti-cancer effects on NSCLC A549 cells. We found that these compounds inhibited A549 cell survival and colony formation in soft-agar and induced apoptosis (Figure [6](#F6){ref-type="fig"}). 14-3-3ζ knock-down in A549 cells has previously been shown to reduce colony formation and increase anoikis (apoptosis induced by loss of cell adhesion) \[[@R11]\]. We also demonstrated the therapeutic potential of RB-012 in an *in vivo* A549-xenograft model (Figure [6](#F6){ref-type="fig"}) and showed that RB-012 reduces tumor growth by up to 30% over a three week course of treatment. The compound was well tolerated with fewer side effects than FTY720, which did not elicit a reduction in tumor growth. FTY720 has previously been demonstrated to reduce tumor growth in other experimental cancer models \[[@R21]\] but at higher doses (10 mg/kg/day), suggesting that the RB-012 is more potent. FTY720 is readily converted to a phospho-form by an endogenous sphingosine kinase (SK), and in this form acts as a sphingosine-1-phosphate analogue which mediates the drug\'s immunosuppressive action. RB-012 cannot be phosphorylated as it lacks a phosphate accepting hydroxyl group and unlike FTY720, does not affect SK activity \[[@R22]\]. Therefore compared to FTY720, RB-012 has a more selective anti-cancer action. Compared with other 14-3-3-directed small molecules \[[@R13], [@R14] & [@R32]\], the RB compounds shown here are non-competitive in that they do not compete with endogenous phospho-clients for binding in the amphipathic groove. These data provide valuable proof-of-principle for our 14-3-3 dimer disruption approach to cancer drug discovery.
MATERIALS AND METHODS {#s4}
=====================
Compounds {#s4_1}
---------
*N-*alkylated tri-methyl ammonium (TMA) compounds were purchased from Sigma. RB compounds were generated as mesylate salts and sodium mesylate was used in all relevant vehicle treatments.
RB-011, -012 and -015 were synthesized as described previously \[[@R22]\]. RB-066, RB-067, and RB-068 were prepared by the following procedures and characterized by ^1^H and ^13^C NMR spectroscopy and electrospray ionization high-resolution mass spectrometry (ESI-HRMS).
### 4-Methyl-4-(4-octylphenethyl)morpholin-4-ium methanesulfonate (RB-066) {#s4_1_1}
To a solution of 4-octylphenethyl methanesulfonate (10 mg, 0.032 mmol) in 3 mL of acetonitrile was added *N*-methylmorpholine (70.4 μL, 0.64 mmol). The reaction mixture was stirred at 50°C for 2 d and concentrated. The residue was washed with hexane to give 7 mg (52%) of RB-066 as a yellow liquid; ^1^H NMR (400 MHz, CDCl~3~) δ 0.88 (t, *J* = 6.9 Hz, 3H), 1.26--1.29 (m, 10H), 1.57 (t, *J* = 7.9 Hz, 3H), 2.56 (t, *J* = 7.7 Hz, 3H), 2.76 (s, 3H), 3.08--3.12 (m, 2H), 3.46 (s, 3H), 3.49--3.53 (m, 2H), 3.67--3.70 (m, 2H), 3.81--3.85 (m, 2H), 3.92--4.00 (m, 4H), 7.14 (d, *J* = 8.0 Hz, 2H), 7.21 (d, *J* = 8.0 Hz, 2H); ^13^C NMR (100 MHz, CDCl~3~) δ 14.1, 22.7, 28.1, 29.3 (2C), 29.4, 29.5, 31.5, 31.9, 35.6, 39.6, 42.8, 47.7, 60.7, 65.7, 128.9, 129.3, 136.1, 142.0; ESI-HRMS (M + H)^+^ *m*/*z* calcd for C~21~H~36~NO 318.2797, found 318.2796.
### 1-Methyl-1-(4-octylphenethyl)-4-oxopiperidinium methanesulfonate (RB-067) {#s4_1_2}
To a solution of 4-octylphenethyl methanesulfonate (10 mg, 0.032 mmol) in 3 mL of acetonitrile was added 1-methyl-4-piperidone (74.7 μL, 0.64 mmol). The reaction mixture was stirred at 50°C for 2 d and concentrated. The residue was washed with hexane to give 8 mg (59%) of RB-067 as a yellow liquid; ^1^H NMR (400 MHz, CDCl~3~) δ 0.81 (t, *J* = 6.7 Hz, 3H), 1.02 (s, 9H), 1.12--1.25 (m, 24H), 1.32--1.76 (m, 2H), 3.47--3.51 (m, 1H), 3.68--3.82 (m, 3H), 4.20--4.23 (m, 1H), 4.47 (dd, *J* = 11.2, 19.6 Hz, 2H), 7.16--7.20 (m, 4H), 7.26--7.40 (m, 6H), 7.57--7.62 (m, 5H); ^13^C NMR (100 MHz, CDCl~3~) δ 14.1, 19.1, 22.7, 25.5, 26.8, 26.8 (2C), 29.4 (2C), 29.5 (2C), 29.6, 29.7 (2C), 31.9, 55.6, 63.9, 65.0, 72.1, 79.9, 127.7, 127.8 (3C), 127.9, 128.4, 130.0, 135.6 (3C), 135.7, 137.9; ESI-HRMS (M + H)^+^ *m*/*z* calcd for C~22~H~36~NO 330.2797, found 330.2791.
### 8-Methyl-8-(4-octylphenethyl)-1,4-dioxa-8-azoniaspiro\[4.5\]decane methanesulfonate (RB-068) {#s4_1_3}
To a solution of 4-octylphenethyl methanesulfonate (10 mg, 0.032 mmol) in 3 mL of acetonitrile was added 8-methyl-1,4-dioxa-8-azaspiro\[4.5\]decane (100 mg, 0.64 mmol). The reaction mixture was stirred at 50°C for 24 h and concentrated. The residue was washed with hexane to give 13 mg (85%) of RB-068 as a yellow liquid; ^1^H NMR (400 MHz, CDCl~3~) δ 0.88 (t, *J* = 6.8 Hz, 3H), 1.26--1.29 (m, 10H), 1.56 (t, *J* = 7.2 Hz, 2H), 1.90--1.94 (m, 2H), 2.06--2.11 (m, 2H), 2.54 (t, *J* = 7.7 Hz, 2H), 2.73 (s, 3H), 3.08--3.21 (m, 2H), 3.39(s, 3H), 3.57--3.61 (m, 2H), 3.72--3.76(m, 4H), 3.94--4.00 (m, 4H), 7.12 (d, *J* = 8.0 Hz, 2H), 7.23 (d, *J* = 8.0 Hz, 2H); ^13^C NMR (100 MHz, CDCl~3~) δ 14.1, 22.7, 27.0, 28.6, 29.3(2C), 29.5, 30.0, 31.5, 31.6, 31.9, 35.5, 39.6, 59.8, 64.9, 65.0, 103.3, 128.9, 129.2, 131.9, 142.5; ESI-HRMS (M + H)^+^ *m*/*z* calcd for C~24~H~40~NO~2~ 374.3059, found 374.3055.
14-3-3 phosphorylation assays {#s4_2}
-----------------------------
Substrate 14-3-3 (0.5 μg of purified recombinant 14-3-3) was added to 15 μl of reaction mixture comprising 0.2U of the PKA catalytic subunit, in the presence or absence of compounds (delivered in 0.1% v/v ethanol) in PKA reaction buffer (10 mM Tris-HCl pH 7.4, 15 mM MgCl~2~, 3 mM DTT containing 25 μM ATP and 0.3 μCi \[^32^P\] γ-ATP). Reactions were incubated at 37°C for 15 min. After incubation, reactions were separated on 12.5% SDS-PAGE and Coomassie stained. 14-3-3 phosphorylation was analyzed using a Typhoon Phosphorimager and quantified using Molecular Dynamics Image Q 5.2 software.
PKA activity assays {#s4_3}
-------------------
Reactions were essentially identical to 14-3-3 phosphorylation assays except that 50 μM kemptide substrate was added in place of 14-3-3 protein. After incubation at 37°C for 15 minutes the reactions were spotted onto phosphocellulose filters (Whatman P81). Filters were washed 5 times in 0.75% phosphoric acid and once in acetone before liquid scintillation counting.
Cell lines and culture {#s4_4}
----------------------
Jurkat E6.1 cells were obtained from the ATCC and verified by short tandem repeat (STR) analysis in December 2014. The A549 cell line was purchased from ECACC in April 2012 (which also performs STR verification) and used within 6 months of resuscitation in the studies presented here. Jurkat cells were routinely cultured in RPMI with 10% FBS and A549 cells in DMEM with 10% FBS at 37ºC with 5% CO~2~. Jurkat Bcl-2 cells were generated by lentiviral transduction using a third generation lentiviral construct as described previously \[[@R15]\] containing a Bcl-2α-IRES-IL2Rα encoding cassette. Transduced cells were FACS sorted for expression of IL2Rα using anti-CD25-PE (BD Pharmingen, \#555433) ([Supplementary Figure 1A](#SD1){ref-type="supplementary-material"}) to enrich for Bcl-2 over-expressing cells, and protein expression was further confirmed by immunoblotting with anti-Bcl-2 antibody (BD Transduction Laboratories, \#610538) ([Supplementary Figure 1B](#SD1){ref-type="supplementary-material"}).
Apoptosis assays: TMRE, Caspase 3 and Annexin V {#s4_5}
-----------------------------------------------
Jurkat cells were routinely set up in apoptosis assays at 2 × 10^5^/ml in RPMI with 0.5% FBS. After treatment, cells were stained either with 200 nM TMRE, 5 μl/ml NucView^TM^ (Biotium), or Annexin V-FITC (Roche) for 15 min prior to analysis by flow cytometry. Forward- and side-scatter properties were used to exclude debris and a 'viable' gate corresponding to the intact PI negative cell population was used for fluorescence analysis. For A549, cells at 90% confluency were treated with the compounds for 48h in DMEM with 0.5% FBS. After treatment, cells were released with trypsin and then stained with NucView^TM^prior to analysis by flow cytometry.
Assessment of A549 cell viability by MTS assay {#s4_6}
----------------------------------------------
A549 cells (2500/well) were plated in 96-well trays in DMEM with 0.5% FBS and cultured at 37ºC with 5% CO~2~ overnight. The following day RB compounds were added and the cells were incubated for a further 48 h prior to removal of the medium and replacement with MTS reagent (Promega) diluted 1:6 in Dulbecco\'s PBS. The cells were incubated at 37ºC for a further 4 h and the conversion of MTS to the colored formazan compound was determined by measurement of absorbance at 490 nM.
A549 colony assay {#s4_7}
-----------------
A549 cells (7500) were plated in 0.33% low-melting point agarose in DMEM with 10 % FBS with and without RB compounds over a 0.5% low-melting point agarose base. Cells were incubated at 37ºC with 5% CO~2~ for 14 days and then colonies were photographed and analyzed using Image J.
Immunoblotting {#s4_8}
--------------
Jurkat cells were treated at 5 × 10^5^/ml in RPMI with 0.5% FBS as detailed. Jurkat cells were harvested by centrifugation at 1500 g for 5 min and washed with PBS prior to lysis in homogenization buffer (20 mM Tris-HCl pH 7.4, 0.5 mM EDTA, 0.5 mM EGTA, 5 % glycerol, protease inhibitor (Roche), 4 mM NaF, 2 mM sodium vanadate, 10 mM β-glycerophosphate, 1 mM sodium pyrophosphate, 1 mM sodium molybdate) for 15 minutes on ice followed by three rounds of freeze-thawing. For treatment A549 cells were plated in 10 cm dishes and treated when they reached approximately 75 % confluence in DMEM with 0.5% FBS. Cells were scraped from the dish and collected by centrifugation at 1500 g and washed with PBS prior to lysis in homogenization buffer as detailed above.
All lysates were clarified at 13,000 rpm for 20 minutes at 4°C and protein concentration was determined using the BCA assay (Pierce). Thirty to forty μg of lysate were run on Criterion-XT 4-12 % Bis-Tris gels (Bio-Rad), followed by transfer onto nitrocellulose. The filters were blocked for 30 min at room temperature in TNT buffer (10 mM Tris-HCl pH 8.0, 150 mM NaCl, 0.05 % Tween-20) containing blocking buffer (Roche), prior to incubation with antibodies over night. Antibodies used were from Cell Signaling Technology; phospho-p38 MAPK (\#4571), phospho-SAPK/JNK (81E11), phospho- ERK (\#9101), phospho-AKT S473 (\#9271), ERK (\#9102), AKT (\#9272), BID (\#2002), caspase-3 (\#9602), PARP (\#9542), and Santa Cruz Biotechnology; pan 14-3-3 K19 (sc-629). After antibody binding, filters were washed in TNT for 1h at room temperature, incubated with 1/12,500 anti-rabbit or anti-mouse HRP secondary antibody (Pierce) for 1h at room temperature and finally washed for 1 hr with TNT before incubation with Clarity western ECL reagent (Bio-Rad). Filters were exposed on a LAS 4000 imager. Blots were analyzed using Multi-gauge/Colony software FUJI FILM.
*In vivo* toxicity and xenograft studies {#s4_9}
----------------------------------------
All experimental procedures involving animals were conducted in accordance with the NHMRC Australian Code for the Care and Use of Animals for Scientific Purposes and with approval by the institutional animal ethics committee. BALB/c nude mice (Nu/Nu, female, 5-6 weeks old) were purchased from the Animal Resources Centre (Perth, WA) and maintained under pathogen-free conditions.
Toxicity studies were carried out with both RB-012 and FTY720 (in 0.9% saline) at 5 and 10 mg/kg body weight, administered daily by intraperitoneal injection and mice were monitored over a 28 day treatment course. Five mice were used per group for RB-012 treatment and three mice per group for FTY720 treatment.
For xenograft studies, A549 cells (5 × 10^6^) in 100 μl of PBS were injected subcutaneously into the flanks of the mice and the resulting tumors were measured using digital calipers. The tumor volume was calculated using the following formula: Volume = (larger diameter) × (small diameter)^2^/2. Once the tumors had reached 100 mm^3^, mice were divided into groups of 9 and to each group was administered with either RB-012, FTY720 (in 0.9% saline) or saline by intraperitoneal injection daily. Initially, RB-012 and FTY720 were administered at 2 mg/kg body weight for two days and then the dose was increased to 5 mg/kg body weight for two weeks, after which RB-012 was increased to 10 mg/kg body weight but FTY720 was maintained at 5 mg/kg body weight. During this dosing regime tumors were monitored twice a week. Differences between samples were analyzed using 2-way ANOVA and statistical significance was accepted at P \< 0.05.
Tissue preparation and analysis {#s4_10}
-------------------------------
At the end of the treatment courses, mice were euthanized and tissues were collected and fixed in neutral-buffered formalin for 12 hrs at 4°C. Tissues were processed by embedding in paraffin and cutting into 4 μm sections. Sections were dewaxed and rehydrated and the antigen retrieved by boiling for 20 minutes in 10 mM citrate buffer (pH 6.0) under pressure (65 kPa above atmospheric pressure). Sections were blocked with 10 % goat serum in PBS (pH 7.4) solution for 30 minutes and incubated with rabbit anti-phospho-ERK (1:100 diluted; CST \#9101) at 4°C overnight. Sections were washed three times in PBS containing 0.1 % Tween and incubated with anti-rabbit IgG conjugated to Alexa Fluor-488 (1:400 diluted; Invitrogen) at room temperature for one hour. Sections were washed as before and mounted in vectashield hard-set mounting medium (Vector) containing DAPI and imaged using a LSM 710 two-photon microscope (Zeiss). Images were analyzed using ImageJ software (NIH) to calculate percentage area coverage by fluorescence signal per image using a binary converted image based on a single manually determined threshold value applied across all images (as previously described) \[[@R23], [@R24]\]. Results are expressed as medians with ranges and quartiles across all data sets.
SUPPLEMENTARY MATERIAL FIGURE {#s5}
=============================
This paper is dedicated to the memory of our colleague, Professor Robert (Bob) Bittman, without whom this work would not have been possible. We are greatly indebted to him for his generosity and insight. This work has been supported by the National Health and Medical Research Council of Australia.
**CONFLICTS OF INTEREST**
The authors declare no potential conflicts of interest.
This paper has been accepted based in part on peer-review conducted by another journal and the authors\' response and revisions as well as expedited peer-review in Oncotarget.
| {
"pile_set_name": "PubMed Central"
} |
Introduction {#s1}
============
Due to rapid changes in climate and demography, vector-transmitted arboviral diseases pose an increasing threat to global health and welfare [@ppat.1004319-Mackenzie1]--[@ppat.1004319-Wilson1]. Among the most severe arboviral infections known to affect the human race are those caused by members of the Flavivirus genus of the *Flaviviridae*. As such, flaviviruses, including Japanese encephalitis (JE), West Nile (WN), dengue, and tick-borne encephalitis virus (TBEV), are major emerging human pathogens, affecting millions of individuals worldwide. In addition, neurological disease frequently occurs upon infection with emerging flaviviruses, such as JEV, WNV, and TBEV [@ppat.1004319-Mackenzie1]--[@ppat.1004319-Wilson1]. Among neurotrophic flaviviruses, JEV is the most prevalent cause of viral encephalitis in the world, with approximately 67,900 cases reported annually [@ppat.1004319-Center1]. Of these cases, about 25--30% are fatal and 50% result in permanent neuropsychiatric sequelae [@ppat.1004319-Center1], for which JE is considered to be more fatal than West Nile encephalitis resulting in a fatality of 3--5% (1,100 death/29,000 symptomatic infection) [@ppat.1004319-Center2]. Indeed, more than 60% of the world\'s population inhabit JE endemic areas which include eastern and southern Asia, and the virus is currently spreading to previously unaffected regions, such as Indonesia, Pakistan, and the northern area of Australia [@ppat.1004319-Solomon1], [@ppat.1004319-Ghosh1].
Considerable progress in understanding the kinetics and mechanisms of JEV dissemination and pathogenesis has been made in murine models [@ppat.1004319-Mackenzie1]--[@ppat.1004319-Wilson1]. However, the molecular pathogenesis of JE still remains elusive. After peripheral amplification of the virus in dendritic cells (DC) and macrophages as primary target cells, the virus gains entry into the CNS through blood-brain barrier (BBB). While JEV infects and kills neurons directly [@ppat.1004319-Chen1], viral replication within microglia/glia and infiltrated monocytes leads to indirect neuronal killing *via* the secretion of pro-inflammatory cytokines (such as IL-6 and TNF-α) and soluble mediators which cause neuronal death [@ppat.1004319-Ghoshal1]. Thus, it is believed that uncontrolled over-activation of microglia/glia and infiltrated monocytes during JE progression is one of the key factors in indirect neuronal cell death [@ppat.1004319-Ghoshal1]. JEV-specific T cells and virus-neutralizing IgM and IgG are considered in part to play a role in the clearance of virus from peripheral lymphoid tissues, as well as from the CNS [@ppat.1004319-Ghosh1]. However, innate immune responses are considered to play a more crucial role in the early control of JEV infection due to delayed establishment of adaptive immunity, and may also be responsible for generating pathological levels of inflammation. Type I IFN gene expression and signaling are essential components of innate immune programs and control various viral infections, and thus may be potentially required for host control of JEV infection [@ppat.1004319-LeBon1]--[@ppat.1004319-Paun1]. Studies in genetically deficient models suggest that type I IFN production after WNV infection is triggered by recognition of viral pathogenic-associated molecular patterns (PAMPs) through cytoplasmic helicases RIG-I and MDA5 as host PRRs [@ppat.1004319-Suthar1]--[@ppat.1004319-Loo1], and thus the ablation of these molecules, their downstream signaling molecules (IPS-1), or transcriptional activators (IRF-3 and IRF-7) results in a greatly enhanced susceptibility to WNV infection [@ppat.1004319-Suthar1]--[@ppat.1004319-Loo1]. However, type I IFN innate responses have also evolved through the recognition of membrane-bound cell-surface or intracellular Toll-like receptors (TLRs) [@ppat.1004319-Brennan1], [@ppat.1004319-Khoo1]. While RIG-I and MDA5 helicases recognize single- and double-stranded RNA in the cytosol and signal through IPS-1, TLRs on the cell surface or within endosomes recognize single- and double-stranded RNA and viral components, and subsequently transmit intracellular signals through adaptor molecules MyD88 and/or TRIF. The role of TLR signal pathway through MyD88 and/or TRIF in restricting flaviviral infection, as well as in modulating immune responses, remains less clear because of conflicting and intricate data [@ppat.1004319-Scholle1]--[@ppat.1004319-Daffis1]. Moreover, the impact of each TLR signal pathway on JE progression has not been addressed to date. We therefore became interested in identifying the key TLR molecule(s) which regulate JEV-induced neurological disease.
TLRs function as intermediates by interacting with products of viral replication, and transmitting signals to a cascade of adaptors and kinases that ultimately lead to the activation of transcription of cytokines and type I IFN genes. TLR3 recruits the adaptor molecule TRIF to induce type I IFN gene *via* interactions with TRAF3, TBK1, and IKKε, which, in turn, activate the latent transcription factors IRF-3 and IRF-7, whereas other TLRs associating with the adaptor protein MyD88 form a complex with TRAF6, IRAK1, and IRAK4 to activate kinases that regulate IRF-5 and IRF-7. Notably, TLR4 signal pathway uses both adaptor molecules MyD88 and TRIF to initiate the production of cytokine and type I IFN proteins. In viral infection, four TLRs, including TLR3, TLR7, TLR8 and TLR9, seem to play critical roles in the recognition of viral nucleic acid components, and TLR2 and TLR4 were shown to detect viral components such as envelope glycoproteins [@ppat.1004319-Arpaia1]--[@ppat.1004319-Yew1]. We have previously shown that JEV can modulate innate immune responses and subsequent adaptive responses in MyD88-dependent and independent pathways [@ppat.1004319-Aleyas1], which indicate that JEV may be recognized by certain TLR signal pathways, thereby affecting the outcome of JEV-induced neurological diseases. Therefore, we aimed to determine whether each TLR signal pathway modulated neurological disease caused by JEV infection, using several TLR-deficient mice (TLR2, TLR3, TLR4, TLR7, TLR9). Surprisingly, among the tested TLR-deficient mouse strains we found a contrasting result in TLR3^−/−^ and TLR4^−/−^ mice, *i.e.* TLR3^−/−^ mice were highly susceptible to JE, whereas TLR4^−/−^ mice showed markedly enhanced resistance to JE. Subsequently, we investigated the pathologic feature, type I IFN innate and adaptive immunity of TLR3^−/−^ and TLR4^−/−^ mice during JE progression. TLR3^−/−^ mice displayed severe neuroinflammatory reactions as well as enhanced BBB permeability by failure of the early control of viral replication, whereas TLR4^−/−^ mice elicited the effective regulation of viral replication and subsequent inflammatory reaction by inducing potent type I IFN innate immune responses against JEV. Notably, our data revealed that TLR4 ablation provided potent type I IFN innate responses through enhanced induction of antiviral ISG genes by alternative activation of IRF-3 and NF-κB in myeloid-derived DCs and macrophages. Also, TLR4^−/−^ mice showed an alteration of plasmacytoid DC subpopulation and CD4^+^Foxp3^+^ regulatory T cells, which were closely associated with enhanced type I IFN innate immune and JEV-specific CD4^+^ and CD8^+^ T cell responses. These results suggest that the balanced triggering of TLR array during JE progression plays a pivotal role in predicting the outcome of neurological disease.
Results {#s2}
=======
Contrasting regulation of JE by triggering TLR3 and TLR4 signal pathway {#s2a}
-----------------------------------------------------------------------
It is believed that intracellular signaling through TLRs regulates host responses against various bacterial and viral infections. However, to date, the impact of TLR signaling on JEV-induced neuroinflammatory diseases and how this response is propagated and regulates *in vivo* innate and adaptive immunity have not been defined. To this end, we assessed the impact of each TLR molecule on JE, by evaluating the susceptibility of TLR2^−/−^, TLR3^−/−^, TLR4^−/−^, TLR7^−/−^, and TLR9^−/−^ mice to JEV infection (1.4×10^7^ pfu) (**[Figure S1](#ppat.1004319.s001){ref-type="supplementary-material"}**). The ablation of TLR2, TLR7, and TLR9 molecules did not significantly affect the progression of encephalitis caused by JEV. However, somewhat surprisingly, TLR3- and TLR4- triggered molecular signaling pathways were observed to induce a completely contrasting regulation of JE. While all TLR3^−/−^ mice succumbed to neuroinflammatory diseases caused by JEV infection (*p* = 0.0153), TLR4^−/−^ mice showed enhanced resistance to JE, compared to wild-type mice (*p* = 0.0819). This contrasting regulation of JE by TLR3 and TLR4 molecules was more apparent (*p* = 0.0314 for TLR3 and *p* = 0.0342 for TLR4), when we evaluated the susceptibility of TLR3^−/−^ and TLR4^−/−^ mice to neuroinflammatory diseases after infection with a higher dose of JEV (2.8×10^7^ pfu) ([**Figure 1A**](#ppat-1004319-g001){ref-type="fig"}). Also, the ablation of both TLR3 and TLR4 molecules induced a highly increased susceptibility to encephalitis caused by JEV infection (*p* = 0.0122). Likewise, TLR3^−/−^ mice infected with JEV showed more rapid signs of neurological disorder starting from 3 days pi, whereas TLR4^−/−^ mice showed delayed signs of neurological disorder with a lower frequency of occurrence, compared to wild-type mice ([**Figure 1B**](#ppat-1004319-g001){ref-type="fig"}). To further examine the contrasting roles of TLR3 and TLR4 molecules, we assessed viral burden within lymphoid and the CNS tissues ([**Figure 1C**](#ppat-1004319-g001){ref-type="fig"}). TLR3^−/−^ mice were found to exhibit 100--1,000-fold elevated viral load in spleen, brain, and spinal cord, but TLR4^−/−^ mice retained significantly lower viral loads with 10--100-fold decreased levels in the spleen, brain, and spinal cord, compared to those of wild-type mice. In addition, since two genetic backgrounds of mouse strains used for TLR3^−/−^ and TLR4^−/−^ mice could complicate the comparison of susceptibility to JE, we directly compared the susceptibility to JE between TLR3^−/−^ mice and wild-type mice, using TLR3^−/−^ mice derived from the same genetic background (H-2^b^) as TLR4^−/−^ mice. As expected, all TLR3^−/−^ (H-2^b^) mice succumbed to JE after infection with two different doses of JEV (1.4×10^7^ and 2.8×10^7^ pfu), while wild-type (H-2^b^) mice showed similar 50% and 70% mortality to wild-type mice of mouse strain (H-2^d^) used for TLR3^−/−^ mice, respectively (**[Figure S2A](#ppat.1004319.s002){ref-type="supplementary-material"}** and **B**). This indicates that the genetic background of mouse strains used in this study did not affect the progression of neuroinflammation caused by JEV infection. Also, TLR3^−/−^ (H-2^b^) mice showed faster neurological disorder and severely reduced body weight by JEV infection. Supportively, TLR3^−/−^ (H-2^b^) mice retained higher viral burden within lymphoid and the CNS tissues (**[Figure S2C](#ppat.1004319.s002){ref-type="supplementary-material"}**). Collectively, these results clearly indicate that triggering signal pathways through TLR3 and TL4 molecules differentially affect the outcome of neuroinflammatory disease caused by JEV infection and *in vivo* viral replication.
{#ppat-1004319-g001}
TLR3, but not TLR4, is essential for the control of CNS inflammation following JEV infection {#s2b}
--------------------------------------------------------------------------------------------
To further characterize the CNS inflammation caused by JEV infection, we assessed the infiltration of CD11b^+^Ly-6C^high^ cells into the CNS, as it has been demonstrated that CD11b^+^Ly-6C^high^ cells have properties of inflammatory monocytes [@ppat.1004319-Getts1]. Our results revealed that nearly identical percentage of CD11b^+^Gr-1^high^ neutrophil was retained in the brain of TLR3^−/−^ and wild-type mice 3 days following JEV infection, whereas a markedly higher frequency of infiltrated CD11b^+^Ly-6C^high^ monocytes in TLR3^−/−^ mice was observed with 10--20-fold increased levels 3 days after JEV infection, as compared to wild-type mice ([**Figure 2A**](#ppat-1004319-g002){ref-type="fig"}). However, there were no significant changes in the proportion of CD11b^+^Gr-1^high^ neurophils and CD11b^+^Ly-6C^high^ inflammatory monocytes infiltrated in the brain of TLR4^−/−^ mice, following JEV infection. Also, the absolute number of inflammatory CD11b^+^Ly-6C^high^ monocytes infiltrated in the brain of TLR3^−/−^ mice increased 100--200-fold, whereas TLR4^−/−^ mice showed no significant changes in the absolute number of infiltrated monocytes or neutrophils following JEV infection ([**Figure 2B**](#ppat-1004319-g002){ref-type="fig"}). To further determine whether the activation of infiltrated CD11b^+^Ly-6C^high^ monocytes could be affected by the ablation of TLR3 and TLR4 molecules, we characterized the phenotypes of infiltrated CD11b^+^Ly-6C^high^ monocytes. However, we found that there were no significant changes in phenotypic levels (CD40, CD80, CD86, MHC I, MHC II, F4/80) of brain infiltrated CD11b^+^Ly-6C^high^ monocytes between TLR3^−/−^ and TLR4^−/−^ mice (data not shown). It has been shown that microglia cells contribute to the pathogenesis of encephalitis caused by some neurotrophic viruses such as WNV [@ppat.1004319-Getts1], [@ppat.1004319-Szretter1]. Thus, triple-color staining (CD11c/CD11b/CD45) was used to distinguish the resting and activated microglia. Based on the CNS myeloid cell classification of Ford et al. [@ppat.1004319-Ford1], equivalent percentages and absolute numbers of resting microglia (CD11b^int^CD45^int^CD11c^−^) were observed in brains of TLR3^−/−^ and TLR4^−/−^ mice following JEV infection. However, the frequency and absolute number of activated microglia (CD11b^high^CD45^high^CD11c^−^) were increased 4--5-fold in TLR3^−/−^ mice ([**Figure 2C and D**](#ppat-1004319-g002){ref-type="fig"}). To confirm the effect of TLR3 and TLR4 molecules on patterns of leukocyte accumulation within the CNS, histological and confocal examinations were performed. Histological examination revealed that increased BBB permeability in JEV-infected TLR3^−/−^ mice was associated with perivascular cuffing, while JEV infection of TLR4^−/−^ mice elicited reduced numbers of infiltrating foci ([**Figure 2E**](#ppat-1004319-g002){ref-type="fig"}). Similarly, significantly higher numbers of infiltrated CD11b^+^ cells were detected in TLR3^−/−^ mice by confocal microscopy, and a small subset of CD11b^+^ myeloid cells co-stained positive with JEV antigen ([**Figure 2F**](#ppat-1004319-g002){ref-type="fig"}). Taken together, these results demonstrate that TLR3-induced signal pathway is essential for the control of neuroinflammation caused by JEV infection, while TLR4 molecules may be dispensable to provide resistance to fatal encephalitis.
{#ppat-1004319-g002}
In terms of severe neuroinflammation in TLR3^−/−^ mice, the expression levels of cytokines and chemokines within the CNS can be required for further explain encephalitis, because encephalitis caused by neurotrophic viruses is indirectly derived from CNS degeneration caused by robust immunological responses, such as the uncontrolled secretion of cytokines and chemokines, and resultant activation of microglia and astrocytes [@ppat.1004319-Ghosh1]--[@ppat.1004319-Ghoshal1]. Therefore, we examined the expression of cytokines and chemokines in inflammatory tissues. We found that JEV infection of TLR3^−/−^ mice induced a highly enhanced expression of IL-6 and TNF-α in the CNS, including brain and spinal cord, whereas moderate changes in the expression of pro-inflammatory cytokines were observed in TLR4^−/−^ mice ([**Figure 3A and B**](#ppat-1004319-g003){ref-type="fig"}). Also, the expression levels of chemokines including CCL2, CCL3, CCL4, CCL5, and CXCL10, which are involved in the migration of leukocytes into the CNS, was increased 10--1,000-fold in the brain and spinal cord of TLR3^−/−^ mice ([**Figure 3C and D**](#ppat-1004319-g003){ref-type="fig"}). To further characterize how TLR3 and TLR4 molecules modulate the inflammatory reaction to JEV infection, we measured the levels of systemic IL-6 in serum of JEV-infected mice at 4 and 6 days pi. A trend towards more rapid induction and increased levels of IL-6 were observed in serum of TLR3^−/−^ mice compared to those of the wild-type mice ([**Figure 3E**](#ppat-1004319-g003){ref-type="fig"}). However, no detectable differences in serum TNF-α levels were observed in TLR3^−/−^ or wild-type mice, since all samples clustered near the limit of detection by ELISA. Also, it was note worthy that TLR4 ablation induced no significant induction of systemic IL-6 and TNF-α. These results demonstrate that in the absence of TLR3, but not TLR4 molecules, greater pro-inflammatory cytokine and chemokine responses are induced during JE progression.
{#ppat-1004319-g003}
TLR3, but not TLR4, regulates JEV-induced BBB disintegrity {#s2c}
----------------------------------------------------------
Since BBB integrity is known to be damaged by neurotrophic virus-induced inflammation, such as WNV infection [@ppat.1004319-Wang2], [@ppat.1004319-Daffis1], we assessed whether the ablation of TLR3 and TLR4 molecules could modulate BBB permeability and, possibly, allow for the earlier entry of virus and leukocytes within the CNS. Changes in BBB integrity over time following JEV infection, as revealed by extravasated Evans blue dye, showed that JEV infection of TLR3^−/−^ mice gave rise to increased BBB permeability 3 days pi ([**Figure 4A**](#ppat-1004319-g004){ref-type="fig"}). In contrast, TLR4^−/−^ mice showed no significant change in BBB permeability following JEV infection. Supportively, the ablation of TLR3 molecules was found to induce increased BBB permeability by JEV infection, when the amount of extravasated Evans blue dye within the brain was measured by photometric analysis ([**Figure 4B**](#ppat-1004319-g004){ref-type="fig"}). Notably, TLR3^−/−^ mice apparently retained increased amounts of extravasated Evans blue dye in the brain 3 days pi, compared to those of wild-type mice. This demonstrates that the ablation of TLR3, but not TLR4 molecule, is able to regulate BBB integrity following JEV infection.
{#ppat-1004319-g004}
The spread of JEV in the brain of TLR3^−/−^ and TLR4^−/−^ mice after intracranial inoculation {#s2d}
---------------------------------------------------------------------------------------------
TLR3^−/−^ and TLR4^−/−^ mice showed distinct viral burdens in the CNS, which were closely associated with lethality to JE. This phenotype could be due to differential dissemination from the periphery and/or an independent antiviral effect in the CNS. To test this, wild-type, TLR3^−/−^, and TLR4^−/−^ mice were inoculated with 10^3^ pfu of JEV directly into the cerebral cortex *via* the intracranial (IC) route, and viral burdens in sub-tissues of brain (cortex, olfactory bulb, hippocampus, brain stem, cerebellum, and spinal cord) were monitored ([**Figure 5A--F**](#ppat-1004319-g005){ref-type="fig"}). Wild-type as well as TLR3^−/−^ and TLR4^−/−^ mice showed rapid and complete mortality following IC infection of JEV, and there was no significant difference in the average survival time between wild-type and KO mice following IC infection of JEV (data not shown). Interestingly, TLR3^−/−^ and TLR4^−/−^ mice showed slightly lower levels of median viral burden in several sub-tissues of the brain. These data suggest that TLR3 and TLR4 molecules had no regulatory function on viral dissemination within the CNS following introduction, but appeared to have a subtle role in regulating viral replication in the CNS. In addition, we examined the expression of pro-inflammatory cytokine (IL-6 and TNF-α), chemokine (CCL2), and type I IFN (IFN-α and IFN-β). The expression of such cytokines in sub-tissues of brain following IC infection of JEV was consistently the same between wild-type and KO mice ([**Figure 5G**](#ppat-1004319-g005){ref-type="fig"}). The accumulation of CD11b^+^Ly-6C^high^ leukocytes in the brain was slightly, but not significantly, higher in TLR3^−/−^ mice following IC infection of JEV, as compared to wild-type mice, and TLR4^−/−^ mice showed no significant change in leukocyte accumulation by IC infection of JEV ([**Figure 5H**](#ppat-1004319-g005){ref-type="fig"}). Collectively, these results imply that TLR3 and TLR4 molecules have different roles in controlling the dissemination of JEV from the periphery into the CNS, rather than a regulatory role(s) on viral dissemination within the CNS after CNS invasion.
{#ppat-1004319-g005}
Type I IFN responses are not blunted in TLR3^−/−^ mice following JEV infection {#s2e}
------------------------------------------------------------------------------
It has been demonstrated that TLR3 molecules, in concert with RIG-I, MDA5, and TLR7, recognize viral RNA and induce type I IFNs through activation of adaptor molecule TRIF and subsequent transcription regulators IRF-3 and IRF-7. Also, triggering signal pathway by TLR4 molecule can activate IRF-3, IRF-5, and IRF-7 through adaptor molecules TRIF and MyD88, thereby inducing the production of type I IFNs (IFN-α and β) [@ppat.1004319-Kondo1]--[@ppat.1004319-Kawai1]. Therefore, since TLR3 and TLR4 molecules contribute to the generation of a normal IFN response through activation of IRF-3, IRF-5, and IRF-7 after infection with neurotrophic virus [@ppat.1004319-Suthar1], [@ppat.1004319-Lazear1], we tested whether the ablation of TLR3 and TLR4 molecules affected type I innate responses in JEV infection. Our data revealed that the expressions of IFN-α and β mRNA were increased in inflammatory and lymphoid tissues of TLR3^−/−^ mice with the levels peaked at 4 days pi, compared to those of wild-type mice ([**Figure 6A and B**](#ppat-1004319-g006){ref-type="fig"}). In contrast, TLR4^−/−^ mice showed no significant increase of IFN-α or β mRNA expression in inflammatory or lymphoid tissues after JEV infection, compared to wild-type mice. Thus, the expression of type I IFN mRNA was not blunted in lymphoid and inflammatory tissues of TLR3^−/−^ mice following JEV infection, which indicates that alternate signal pathways *via* innate immune receptors, such as TLR7, RIG-I, and MDA5, can contribute to type I IFN responses in the absence of the TLR3 molecule. Similarly, TLR3^−/−^ mice showed delayed but slightly increased production of systemic IFN-β with peak levels attained at 48 h pi, compared to those of wild-type mice ([**Figure 6C**](#ppat-1004319-g006){ref-type="fig"}). However, paradoxically and surprisingly, TLR4 ablation induced more rapid and markedly increased production of systemic IFN-β in serum, as compared to production rates in TLR3^−/−^ or wild-type mice. This result indicates that a deficiency of TLR4 molecules can modify the systemic production of type I IFNs. Importantly, it is worthy to note that this markedly enhanced production of systemic type I IFN-β protein in TLR4^−/−^ mice might contribute to the early control of viral replication in the periphery, thereby ultimately preventing viral dissemination into the CNS.
{#ppat-1004319-g006}
Enhanced virus control and type I IFN responses in myeloid cells derived from TLR4^−/−^ mice after JEV infection {#s2f}
----------------------------------------------------------------------------------------------------------------
Myeloid cells, including tissue and lymphoid DCs and macrophages, are primary target cells of JEV infection and function to regulate the spread of virus to distant tissues such as the CNS [@ppat.1004319-Ghosh1]. Also, diverse cell populations can differentially utilize PRRs to induce innate immune responses upon viral infection. Therefore, these subtle functions may affect viral dissemination in the body and subsequent viral diseases. In addition, since our data showed that TLR4 ablation provided rapid and increased production of systemic IFN-β, we assessed whether TLR3 and TLR4 molecules affect JEV replication and type I IFN responses in myeloid-derived cells as primary target cells, in order to further define the differential roles of TLR3 and TLR4 molecules in controlling the progression of JE. Bone marrow-derived DCs (BMDC) and macrophages (BMDM) of TLR3^−/−^ and TLR4^−/−^ mice were infected with JEV and used to evaluate viral replication and the induction of pro-inflammatory cytokines and type I IFNs. TLR3^−/−^ BMDC sustained significantly higher JEV replication throughout the examination period compared to those of wild-type BMDC infected with JEV, whereas TLR4^−/−^ BMDC and BMDM showed delayed JEV replication at 24 and 48 h pi ([**Figure 7A**](#ppat-1004319-g007){ref-type="fig"}). Also, a rapid and increased induction of IL-6 mRNA in TLR3^−/−^ BMDC and BMDM was observed, while TNF-α expression was increased 2-fold in JEV-infected TLR4^−/−^ BMDC and BMDM ([**Figure 7B and C**](#ppat-1004319-g007){ref-type="fig"}), implying that the ablation of each TLR molecule could cause to trigger differential signal pathways to compensate for the production of pro-inflammatory cytokines. Surprising data was obtained from type I IFN innate responses of TLR4^−/−^ BMDC and BMDM after JEV infection. TLR4^−/−^ BMDC and BMDM induced rapid expressions of type I IFNs (IFN-α and β) mRNA with 10--100-fold increase in response to JEV infection, compared to wild-type BMDC and BMDM ([**Figure 7D and E**](#ppat-1004319-g007){ref-type="fig"}). In contrast, IFN-α and β expression by TLR3^−/−^ BMDC and BMDM was virtually identical to those of wild-type BMDC and BMDM following JEV infection, except at an early time point (24 h pi), where levels were notably lower in TLR3^−/−^ BMDC. In support, TLR4^−/−^ BMDC and BMDM showed rapid secretion of IFN-β protein with 5--10-fold increase in response to JEV infection, while TLR3^−/−^ BMDC and BMDM showed slightly higher or identical levels of secreted IFN-β protein, compared to wild-type BMDC and BMDM ([**Figure 7F**](#ppat-1004319-g007){ref-type="fig"}). Importantly, levels of IFN-β secretion in TLR3^−/−^ BMDC and BMDM were much lower than those of TLR4^−/−^ BMDC and BMDM. Conceivably, it is possible that potent type I IFN innate responses in TLR4^−/−^ myeloid-derived cells provides rapid and increased production of *in vivo* systemic type I IFNs, thereby contributing to the early control of viral replication in the absence of the TLR4 molecule. Collectively, these results indicate that TLR4 molecules are dispensable to induce rapid and increased response of type I IFN innate immunity in myeloid-derived cells upon JEV infection, and that a deficiency of TLR3 molecules does not virtually compromise type I IFN production in BMDC and BMDM after JEV infection.
{#ppat-1004319-g007}
Enhanced induction of type I IFNs and ISGs in the absence of TLR4 is associated with alternative IRF3 phosphorylation and IκBα degradation in myeloid-derived DCs and macrophages {#s2g}
---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Since myeloid-derived DCs and macrophages of TLR4-ablated mice showed highly enhanced production and expression of antiviral type I IFNs upon JEV infection, we measured the induction levels of antiviral ISG genes to define this finding in greater detail. We specifically focused on PRRs (RIG-I \[DDX1\], MDA5 \[IFITH1\]), their transcription factors (IRF3, IRF5, IRF7), and IFNAR transcription factor (STAT1) as well as IFN-independent (ISG49 \[IFIT3\], ISG54 \[IFIT2\], ISG56 \[IFIT1\], CXCL10) and dependent genes (PKR, Mx1, Oas1, Oasl-1). Our results revealed that TLR3^−/−^ BMDC and BMDM showed differential responses of antiviral ISG expression upon JEV infection ([**Figure 8A and B**](#ppat-1004319-g008){ref-type="fig"}). TLR3^−/−^ BMDC showed less induction of PRR genes (RIG-I and MDA5) and their transcription factors (IRF-3 and IRF-7), but member of genes (ISG49, ISG54, ISG56, CXCL10) that are induced in IFNAR^−/−^ cells (*i.e.*, are IFN-independent) [@ppat.1004319-Grandvaux1], [@ppat.1004319-Scherbik1] were expressed in TLR3^−/−^ BMDC with slightly higher levels, compared to those of wild-type BMDC. This result was consistent with the fact that TLR3^−/−^ BMDC showed slightly higher or identical secretion of IFN-β compared to wild-type BMDC ([**Figure 7F**](#ppat-1004319-g007){ref-type="fig"}), because IFN-independent ISG genes can also be induced through ISRE binding of ISGF3 complex initiated by type I IFN receptor [@ppat.1004319-Lazear1]. In contrast, TLR3^−/−^ BMDM showed less induction of IFN-independent ISG genes (ISG49, ISG54, ISG56, CXCL10) as well as IFN-dependent ISG genes (PKR, Mx1, Mx2) and members of the 2′-5′-oligoadenylate synthetase family (Oas1, Oasl-1), compared to wild-type BMDM. This result implies that macrophages could be more compromised in the inductiveness of type I IFN innate responses than DCs, if the TLR3 molecule was ablated. The prominent induction of antiviral ISG genes was observed in TLR4^−/−^ BMDC and BMDM after JEV infection ([**Figure 8A and B**](#ppat-1004319-g008){ref-type="fig"}). TLR4^−/−^ BMDC showed enhanced expression of PRR genes (MDA-5) and its transcription factors (IRF-3, IRF-5, IRF-7), and IFN-dependent genes (PKR, Oasl-1), as well as IFN-independent genes (ISG49, ISG 54, ISG 56, CXCL10). Also, TLR4^−/−^ BMDM showed much more apparently and highly induced expression of antiviral ISG genes after JEV infection compared to those of wild-type BMDM and other cells, because TLR4^−/−^ BMDM induced the expression of all tested ISG genes (PRRs, transcription factors, IFN-dependent and independent genes) with higher levels than other cells. Notably, TLR4^−/−^ BMDM showed markedly enhanced induction of both IFN-dependent (PKR, Oas1, Oasl-1, Mx1, Mx2) and independent genes (ISG49, ISG54, ISG56, CXCL10), compared to TLR3^−/−^ BMDM that showed less induction of such genes. Therefore, these results support that a deficiency of TLR4 molecule provides more efficient type I IFN innate immune responses in DCs and macrophages following JEV infection.
{#ppat-1004319-g008}
To further define the induction of antiviral IFN-independent and dependent ISG genes in JEV-infected TLR3^−/−^ and TLR4^−/−^ DCs and macrophages, the activation state of associated transcription factors was examined by western blot. In line with antiviral ISG induction data, TLR3^−/−^ BMDC displayed decreased expression of IRF-3 and IRF-7 at 6--48 h and 48 h pi, respectively, and phosphorylated form of IRF-3 was not detected in both TLR3^−/−^ and TLR4^−/−^ BMDC ([**Figure 8C**](#ppat-1004319-g008){ref-type="fig"}), which supports that enhanced induction of antiviral IFN-independent ISG genes (ISG49, ISG54, ISG 56, CXCL10) in TLR3^−/−^ and TLR4^−/−^ BMDC may be caused by stimulation of IFNAR signal through increased IFN-β secretion [@ppat.1004319-Lazear1]. Since slightly delayed phosphorylation of STAT1, an IFNAR transcription factor, was observed in TLR3^−/−^ and TLR4^−/−^ BMDC, other pathways to activate NF-κB were also considered to contribute to enhanced induction of IFN-independent ISG genes. Interestingly, this notion can be explained by the result that faster degradation of IκBα was detected in TLR4^−/−^ BMDC. IκBα proteins are phosphorylated *via* IκB kinase (IKK) activated by signal transducers, and are subsequently degraded after release of NF-κB [@ppat.1004319-DiazMeco1]. Therefore, these results suggest that TLR4^−/−^ BMDC could have evolved as yet unknown pathway(s) to activate NF-κB upon JEV infection, thereby inducing enhanced expression of type I IFNs and ISG genes. In addition, somewhat interestingly, transiently phosphorylated form of IRF-3 was strongly detected in TLR4^−/−^ BMDM, but not TLR3^−/−^ BMDM, as early as 6 and 12 h pi ([**Figure 8D**](#ppat-1004319-g008){ref-type="fig"}). Also, TLR4^−/−^ BMDM showed prolonged and strong phosphorylation of STAT1 after JEV infection, compared to wild-type BMDM. Therefore, it was considered that activation of IRF3 and STAT1 in TLR4^−/−^ BMDM derived potent type I IFN production as well as the induction of broad antiviral IFN-independent and IFN-dependent ISG genes.
Induction of type I IFN and ISGs in primary cortical neurons derived from TLR3 and 4-deficient mice after JEV infection {#s2h}
-----------------------------------------------------------------------------------------------------------------------
Neurons may be the main target cell of JEV infection in the CNS, and their death is a key factor in pathogenesis and neurological sequelae [@ppat.1004319-Chen1]. To examine whether TLR3 and TLR4 molecules can regulate JEV replication in neurons, primary cortical neurons generated from wild-type as well as TLR3^−/−^ and TLR4^−/−^ mice were infected with JEV, and virus yield, type I IFN responses and ISG expression were evaluated. It was likely that wild-type neurons were more permissive to JEV infection than DCs or macrophages, because infection of neurons with 10-fold less virus (MOI 0.1 versus 1.0) produced over ∼10^5^ viral RNA within 24 h ([**Figure 7A**](#ppat-1004319-g007){ref-type="fig"} and [**Figure 9A**](#ppat-1004319-g009){ref-type="fig"}). In the absence of TLR3 molecule, JEV replicated faster, resulting in a 1.5--2.0-fold increase in infectious virus production between 24 h and 48 h pi, as compared to infected wild-type neurons. The ablation of TLR4 molecule showed earlier replication of JEV at 24 h pi, but the levels of virus were similar in both wild-type and TLR4^−/−^ neurons at 48 h pi ([**Figure 9A**](#ppat-1004319-g009){ref-type="fig"}). Biphasic type I IFN mRNA induction was observed, with slightly higher levels at 24 h pi but much lower at 48 h pi in TLR3^−/−^ neurons, compared to wild-type neurons ([**Figure 9B**](#ppat-1004319-g009){ref-type="fig"}). In contrast, TLR4^−/−^ neurons showed transient induction of IFN-β at 24 h pi, after which IFN-β mRNA levels were comparable in both wild-type and TLR4^−/−^ neurons. The secretion of IFN-β protein in culture media was markedly lower in TLR3^−/−^ neurons at 48 h pi, while TLR4^−/−^ neurons showed increased expression and production of IFN-β at both 24 h and 48 h pi, as compared to those of wild-type neurons ([**Figure 9B**](#ppat-1004319-g009){ref-type="fig"}). Also, it seemed that the expression of antiviral ISGs in TLR3^−/−^ neurons followed type I IFN responses; hence, ISG49 and ISG56 showed transient increases at 24 h pi but much lower expression at 48 h pi ([**Figure 9C**](#ppat-1004319-g009){ref-type="fig"}). Also, a higher expression of RIG-I and MDA-5, a cytosolic PRRs of viral RNA, was observed in TLR3^−/−^ neurons, but their transcription factor IRF-3 was shown with decreased expression levels, as compared to wild-type neurons. It was thought that this caused the reduction in IFN-β production in TLR3^−/−^ neurons at 48 h pi. TLR4^−/−^ neurons showed transiently higher expression of ISG54 and MDA5 at 24 h pi, but the decreased levels of RIG-I and IRF-7 expression was observed at 48 h pi. Collectively, these results suggest that TLR3 may have an independent and subordinate role in triggering type I IFN innate responses in cortical neurons, because type I IFN responses and ISGs expression were much decreased at a later time point (48 h pi). Also, TLR4^−/−^ cortical neurons appeared to induce less potent type I IFN innate immune responses than TLR4^−/−^ DCs and macrophages, which indicates that specific types of cells differentially trigger innate immune responses following JEV infection.
{#ppat-1004319-g009}
Enhanced type I IFN innate and JEV-specific CD4^+^/CD8^+^ T cell responses in TLR4^−/−^ mice are associated with altered number of plasmacytoid DCs and CD4^+^Foxp3^+^ Treg cells in lymphoid tissues {#s2i}
-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
TLR signal pathway through MyD88 and/or TRIF adaptor molecules is required in some cases for antigen-specific antibody responses [@ppat.1004319-Desmet1], [@ppat.1004319-Michallet1], which may contribute to the control of JEV dissemination and replication in the brain. Our data revealed that TLR3 ablation showed slightly, but not significantly, increased level of IgM and IgG, while TLR4^−/−^ mice showed significantly increased levels of JEV-specific IgM and IgG, compared to wild-type mice (**[Figure S3A](#ppat.1004319.s003){ref-type="supplementary-material"}**). Also, JEV infection showed marginally increased numbers of CD4^+^, CD8^+^ T, and CD19^+^ B cells with activated phenotypes, as corroborated by the expression of surface markers, such as CD69, CD44, and CD80; however TLR3 and TLR4 molecules did not show apparently regulatory functions in T and B lymphocytes (**[Table S1](#ppat.1004319.s004){ref-type="supplementary-material"}**). Since effector antigen-specific CD4^+^ and CD8^+^ T cell responses are also required for the control and clearance of JEV in the CNS as well as in peripheral tissues [@ppat.1004319-Ghosh1], we evaluated whether the ablation of TLR3 and TLR4 molecules altered JEV antigen-specific CD4^+^ and CD8^+^ T cell responses. A deficiency of TLR3 molecules resulted in a similar percentage and absolute number of CD4^+^ and CD8^+^ T cells expressing IFN-γ and TNF-α, whereas TLR4^−/−^ mice showed an increased percentage and absolute number of IFN-γ and TNF-α-producing CD4^+^ and CD8^+^ T cells (**[Figure S3B--E](#ppat.1004319.s003){ref-type="supplementary-material"}**). Along with potent type I IFN innate responses, these data indicate that TLR4 ablation could provide enhanced antigen-specific responses, thereby contributing in part to the control of virus replication and dissemination. Therefore, to further characterize the immunological parameters associated with potent type I IFN innate and adaptive immune responses in JEV-infected TLR4^−/−^ mice, we analyzed the immune cellular components related to type I IFN innate and adaptive immune responses. TLR3^−/−^ and TLR4^−/−^ mice were challenged with JEV, and spleens were harvested at 3 and 5 days pi. At the early phase of infection, analysis of the spleen provides an insight into how TLR3 and TLR4 molecules modulate innate immune and inflammatory responses immediately after infection, because JEV was administered intraperitoneally. Analysis of lymphoid CD8α^+^ and myeloid CD11b^+^ DC subsets revealed that JEV-infected TLR3^−/−^ and TLR4^−/−^ mice exhibited similar increases in both DC subsets, compared to those of infected wild-type mice ([**Figure 10A**](#ppat-1004319-g010){ref-type="fig"}). However, somewhat surprisingly, the ablation of TLR4 molecule resulted in a highly increased number of CD11c^int^PDCA-1^high^ plasmacytoid DC (pDC) subset, which is known as a potent cellular component to produce type I IFNs in response to viral infection [@ppat.1004319-Wang3]. Thus, it was considered that highly increased pDC number might contribute in part to enhanced production of systemic IFN-β in TLR4^−/−^ mice. TLR4^−/−^ mice also showed a decreased frequency of inflammatory CD11c^−^CD11b^+^Ly-6C^high^ monocytes and no significant changes in the absolute number, whereas a significant increased number, but not frequency, of inflammatory monocytes was observed in TLR3^−/−^ mice, compared to that in wild-type mice ([**Figure 10B and C**](#ppat-1004319-g010){ref-type="fig"}). This implies that TLR4^−/−^ mice exhibit a mild inflammatory reaction in the spleen. In addition, a deficiency of TLR4 molecule provided an increased number of NK cells at 5 days pi, but TLR3 molecule had no modulatory function on NK cell number ([**Figure 10D**](#ppat-1004319-g010){ref-type="fig"}). Moreover, since CD4^+^CD25^+^Foxp3^+^ Treg cells contribute to the dampening of innate and adaptive immune responses during acute viral infection [@ppat.1004319-Rowe1], we addressed the frequency and number of CD4^+^CD25^+^Foxp3^+^ Treg cells in the spleen. We found that the frequency and absolute number of CD4^+^CD25^+^Foxp3^+^ Treg cells were increased 1.5--2-fold in response to JEV infection in wild-type mice ([**Figure 10E and F**](#ppat-1004319-g010){ref-type="fig"}). TLR3^−/−^ mice showed identical increase of CD4^+^CD25^+^Foxp3^+^ Treg cells to wild-type mice, while in TLR4^−/−^ mice a reduced frequency and absolute number of CD4^+^CD25^+^Foxp3^+^ Treg cells was observed, which indicates that TLR4 molecule could be involved in the increase of CD4^+^CD25^+^Foxp3^+^ Treg cell numbers following JEV infection. Collectively, these results suggest that increased number of CD11c^int^PDCA-1^high^ pDC subpopulation and reduced CD4^+^CD25^+^Foxp3^+^ Treg cells are closely associated with enhanced type I IFN innate immunity and JEV-specific CD4^+^ and CD8^+^ T cell responses in TLR4^−/−^ mice.
{#ppat-1004319-g010}
Discussion {#s3}
==========
Although recognition of ssRNA virus, such as flavivirus, *via* cytosolic helicase RIG-I and MDA5 may be dominant to induce type I IFN innate responses [@ppat.1004319-Suthar1]--[@ppat.1004319-Loo1], the role of TLRs as first-front line of innate immune receptors in the extracellular space, including the cell membrane and endosome, remains still undefined in flaviviral infections, due to conflicting and intricate data. Furthermore, despite the pathological importance of JE as a major cause of acute encephalitis, the role of TLR signal pathways in JE progression has not been fully explored to date. Here, we observed strikingly contrasting regulation of JE *via* TLR3 and TLR4 signal pathways; TLR3 ablation elicited highly enhanced susceptibility to JE, whereas TLR4 ablation provided significantly enhanced resistance to JE rather than inducing increased susceptibility. In the present study, interesting clues to such contrasting regulation of JE by TLR3 and TLR4 molecules were derived from the differential induction of type I IFN innate responses in TLR3^−/−^ and TLR4^−/−^ mice. Notably, TLR4 ablation induced potent type I IFN innate responses through enhanced induction of antiviral ISG genes by alternative activation of IRF-3 and NF-κB in DCs and macrophages. Additionally, altered CD11c^int^PDCA-1^high^ pDC and CD4^+^CD25^+^Foxp3^+^ Treg number in TLR4^−/−^ mice appeared to contribute in part to enhanced type I IFN innate as well as JEV-specific T cell responses. Collectively, potent type I IFN innate and adaptive immune responses generated in peripheral lymphoid tissues after JEV infection were closely coupled with a reduced JE lethality in TLR4^−/−^ mice. These findings imply that the balanced triggering of TLR signal array by viral components during JE progression could be responsible for determining the outcome of disease through negative and positive regulatory factors.
There are several conflicting reports on the role of TLR3 signaling pathway in neurological diseases caused by viral infection [@ppat.1004319-PeralesLinares1], [@ppat.1004319-Zhang1]. A deficiency of TLR3 in humans predisposes to a genetic risk factor for herpes simplex virus encephalitis [@ppat.1004319-Zhang2] and influenza A virus-induced encephalopathy [@ppat.1004319-Hidaka1], but TLR3^−/−^ mice infected with influenza [@ppat.1004319-LeGoffic1], punta toro [@ppat.1004319-Gowen1], and vaccinia viruses [@ppat.1004319-Hutchens1] showed improved survival and decreased production of inflammatory cytokines. Strikingly conflicting role of TLR3 signal pathway was derived from an infection model with WNV [@ppat.1004319-Wang2], [@ppat.1004319-Daffis1]. While TLR3 ablation protected mice from WNV lethal infection by decreased systemic TNF-α and IL-6 production and BBB permeability [@ppat.1004319-Wang2], there is a report demonstrating that TLR3 molecules are essential in protecting from WNV infection [@ppat.1004319-Daffis1]. Our results favor the latter report. TLR3^−/−^ BMDC, but not to BMDM, showed defective type I IFN innate responses at an early time (24 h pi), which may allow early viral replication. This result is in contrast to that of WNV infection, where TLR3 molecule did not modulate WNV replication and IFN induction in primary myeloid cells [@ppat.1004319-Daffis1]. Although TLR3^−/−^ BMDC is more permissive to JEV replication, JEV-infected TLR3^−/−^ BMDC elicited similar levels of type I IFN responses to wild-type BMDC with delayed kinetics, and TLR3^−/−^ mice also showed no blunted type I IFN responses in lymphoid and local tissues. This suggests that enhanced tissue tropism and rapid viral entry into the CNS is not affected by locally induced type I IFN responses. Type I IFN responses of TLR3^−/−^ mice were considered not to be attenuated since cytosolic RIG-I and MDA5 molecules are intact. However, TLR3 molecule appeared to play more important role in inducing type I IFN responses of neuron cells than BMDC and BMDM, because TLR3^−/−^ neuron cells showed a markedly reduced expression and production of type I IFN and ISGs at a late time (48 h pi), thereby promoting viral replication. This implies that TLR3 molecule had differential modulatory functions on type I IFN innate responses and JEV replication in a cell-type restricted manner. However, considering that TLR3^−/−^ and TLR4^−/−^ mice showed no difference in CNS replication of JEV following IC infection, subtle changes of CNS system, such as innate responses of microglia and astrocyte, appear to modulate the *in vivo* spread of directly inoculated JEV in the CNS.
Increased BBB permeability by systemic TNF-α and IL-6 appears to promote an earlier entry of virus into the CNS. In contrast to WNV infection, where TLR3^−/−^ mice showed no change in BBB permeability [@ppat.1004319-Daffis1], TLR3^−/−^ mice, but not TLR4^−/−^ mice, elicited increased BBB permeability associated with a huge production of systemic IL-6. Also, this result was in contrast with a previous study in which TLR3^−/−^ mice showed reduced cytokine (e.g., TNF-α and IL-6) responses, BBB permeability, neuroinvasion, and mortality following infection with mammalian cell-passaged WNV [@ppat.1004319-Wang2]. Nonetheless, our results showed some similarities with previous reports using WNV, such as increased viral burden in peripheral tissues. Although the impact of TLR3 molecule on BBB permeability is likely to differ, depending on the context and details of the model, virus-culture conditions, and the viral strain being tested, the failure of early viral clearance in the periphery of TLR3^−/−^ mice may ultimately cause enhanced inflammatory reactions, thereby increasing BBB permeability and viral load in the CNS. One intriguing result in this study was that TLR7^−/−^ mice showed no change in susceptibility to JE, since TLR7 molecule can recognize ssRNA of JEV. This result was in contrast with the report that the TLR7 molecule is involved in modulating the progression of WNV encephalitis *via* an IL-23-dependent accumulation of leukocytes in the CNS [@ppat.1004319-Town1]. Although systemic levels of proinflammatory cytokines and type I IFNs were higher in TLR7^−/−^ than in wild-type mice [@ppat.1004319-Town1], it is expected that splenic pDC or circulating pDC from TLR3^−/−^ mice may also have contributed to the type I IFN responses, because TLR7 signal pathway was intact in TLR3^−/−^ mice. In addition, we previously found that TLR2 molecule had modulatory function in cross-presentation of OVA protein using JEV-infected TLR2^−/−^ mice, suggesting that JEV infection may be also be recognized by TLR2 molecule [@ppat.1004319-Aleyas2]. However, in this study, TLR2 signal pathway had no impact on the progression of JE. One trivial explanation of this result is that TLR2 signal pathway was not involved in inducing pathologic disease by JEV infection, no matter what OVA cross-presentation is regulated by JEV infection in a TLR2-dependent manner.
The most intriguing result in this study was that TLR4^−/−^ mice showed markedly enhanced resistance to JE. To date, the role of TLR4 signal pathway in inducing innate and adaptive immune response against JEV and other flaviviruses has not been defined. Our results demonstrate that TLR4 ablation strongly induces *in vivo* systemic type I IFN innate responses, as well as type I IFN expression and production from myeloid-derived cells upon JEV infection. This presumably promotes early clearance of virus. In spite of the existence of TLR4 prototype ligand, LPS, a growing number of reports suggest that TLR4 molecule is biologically relevant, and is responsive to viral proteins, including those of Ebola virus [@ppat.1004319-Okumura1], hepatitis C virus [@ppat.1004319-Duesberg1], and respiratory syncytial virus [@ppat.1004319-KurtJones1], leading to the induction of proinflammatory cytokines. We are not sure whether the induction of potent type I IFN innate responses in the absence of TLR4 signal pathway was mediated directly by enhanced signal transduction of other PRRs, such as TLR3, RIG-I, and MDA5, and/or indirectly by soluble factors produced from host cells by viral infection, i.e. DAMPs. However, our results provide one explanation as to how TLR4^−/−^ myeloid-derived cells induce potent type I IFN innate responses, *i.e.* enhanced activation of NF-κB through unknown pathway(s) in DCs, and transient activation of IRF3 at 6--12 h pi and prolonged activation of STAT1 in macrophages. The expression of antiviral ISG genes in myeloid-derived cells after JEV infection was induced by both direct (by IRF-3) and indirect (by IFN-β production and IFNAR signaling) pathways. Considering that only small faction (10--20%) of myeloid-derived cells is infected by JEV [@ppat.1004319-Aleyas2], uninfected myeloid-derived cells are thought to substantially contribute to antiviral ISG induction through stimulation of IFNAR signal after binding with secreted IFN-β proteins. This notion was supported by two results, *i.e.* 1) induction of IFN-dependent genes (PKR, Mx1, Oas1) in TLR3^−/−^ BMDC, TLR4^−/−^ BMDC and BMDM with increased secretion of IFN-β after JEV infection, and 2) no detection of phosphorylated IRF-3 except in TLR4^−/−^ BMDM. Also, transient activation of IRF3 and prolonged activation of STAT1 explains strong induction of both IFN-independent ISG (ISG49, ISG54, ISG56, CXCL10) and dependent genes (PKR, Oas1, Mx1, Mx2) in TLR4^−/−^ macrophages. Although NF-κB activation in DCs and IRF-3 and STAT1 activation in macrophages after JEV infection support potent type I IFN innate responses in the absence of TLR4 molecule, how these signal molecules are activated remains still undefined. Therefore, future studies will be required to delineate the mechanistic and functional intermediates that link and regulate NF-κB, IRF-3 and STAT1 signal pathway in the absence of TLR4 molecule.
In addition, our results is strengthened by a recent report that TLR4^−/−^ or TLR4 antagonist-treated mice are highly refractory to influenza-induced lethality, due to blocking inflammation by host-derived, oxidized phospholipid that potently stimulates TLR4 [@ppat.1004319-Nhu1], [@ppat.1004319-Shirey1]. One similarity with our data is that mice treated with TLR4 antagonist, Eritoran, or TLR4^−/−^ mice had reduced lung pathology to infection with influenza virus, which is characterized by the reduction of viral burden and proinflammatory cytokine expression. However, it is not certain whether Eritoran-treated or TLR4^−/−^ mice displayed rapid and enhanced type I IFN innate responses after infection with influenza virus. Thus, it is worthwhile identifying whether blocking TLR4 signal pathway by antagonists such as Eritoran, affects JE progression through the induction of potent type I IFN innate responses. This study will provide valuable insights into developing therapeutic strategies to viral encephalitis caused by neurotrophic virus such as JEV and WNV. Analogously, in the absence of TLR4 molecule, the enhanced expansion of CD11b^+^Ly-6C^high^ "inflammatory monocytes" was not observed in comparison with TLR3^−/−^ mice, which was suggestive that in TLR4^−/−^ mice mild inflammatory responses were elicited in the spleen. This monocyte subset migrates to the site of infection, secretes pro-inflammatory cytokines, and thereby exacerbates immunopathologic diseases [@ppat.1004319-Getts1]. Thus, the aberrant recruitment and expansion of these CD11b^+^Ly-6C^high^ inflammatory monocytes may also contribute to JE immunopathogenesis in TLR3^−/−^ mice.
The production and response of type I IFN is considered to be a major linkage point between innate and adaptive immunity, because IFN-α/β sustains B cell activation and differentiation [@ppat.1004319-Coro1], [@ppat.1004319-Fink1], expands antigen-specific CD8^+^ T cells [@ppat.1004319-Kolumam1], CD4^+^ T cells [@ppat.1004319-HavenarDaughton1], and activation of NK cells [@ppat.1004319-Gerosa1]. Therefore, another intriguing finding of this study was the global alteration of immune responses in TLR4^−/−^ mice. This suggests that TLR4 molecule is largely dispensable for the efficient link between innate and adaptive immunity in JEV infection. Infection of TLR4^−/−^ mice with JEV exhibited the expansion of pDC and NK cells, and enhanced JEV-specific CD4^+^ and CD8^+^ T cell responses, which are involved in viral clearance at early and late phases of infection, respectively. Also, it is likely that increased number of pDCs contributed in part to the potent induction of type I IFN innate responses in TLR4^−/−^ mice. In addition, TLR4^−/−^ mice showed limited expansion of CD4^+^CD25^+^Foxp3^+^ Tregs, which have been known to suppress innate and effector T cells, thus preventing or controlling reactivity to self-antigen and pathogens, and thereby blunting severe inflammation and maintaining antigen-specific T cell homeostasis [@ppat.1004319-Rowe1]. The role of CD4^+^CD25^+^Foxp3^+^ Tregs in acute viral diseases is still debatable [@ppat.1004319-Lanteri1], [@ppat.1004319-Stross1]. Recent work implicates CD4^+^Foxp3^+^ Tregs in the control of WNV pathogenesis, wherein peripheral expansion of Treg was associated with mild inflammation, but reduced Treg levels were associated with WNV encephalitis [@ppat.1004319-Lanteri1]. However, while CD4^+^Foxp3^+^ Tregs that were adoptively transferred 2 days prior to JEV infection made the recipients vulnerable to JE, CD4^+^Foxp3^+^ Tregs that were adoptively transferred 2 days after infection provided resistance to JE (unpublished personal data). This suggests that CD4^+^Foxp3^+^ Tregs elicit dual-phased roles during the progression of JEV-induced neurological disorders. More importantly, Treg induction during a viral infection is considered to be a detrimental response that promotes virus persistence without benefits to the host [@ppat.1004319-Molling1], [@ppat.1004319-Perrella1]. One trivial explanation of CD4^+^Foxp3^+^ Treg role is that initially low number of CD4^+^Foxp3^+^ Tregs in TLR4^−/−^ mice may promote the expansion of effector CD4^+^ and CD8^+^ T cells specific for JEV antigen as well as innate immune responses, thereby inducing enhanced anti-viral response and virus-specific CTL to promote early viral clearance.
JE pathogenesis in the murine model may be altered by the route of peripheral administration, virus-propagation condition, and viral strains [@ppat.1004319-Ghosh1], [@ppat.1004319-Wang2], [@ppat.1004319-Daffis1]. It is also possible that the genetic background of mice affects the immunopathogenesis of JE. However, we found that two backgrounds of mouse strains used for TLR3^−/−^ and TLR4^−/−^ mice showed comparable mortality and similar clinical signs after JEV infection, which indicates that JE pathogenesis is unaffected by genetic background of mouse strains used in this study. Although JEV infected *via* i.p. route does not directly reflect natural infection mediated by intradermal or intramuscular route after biting of mosquitoes, JEV infected *via* i.p. route displays entirely similar pathogenesis to natural infection, due to peripheral amplification in the spleen. Also, since mice infected i.p. with JEV usually exhibited neurological disorder at 4--5 days pi, rapid innate immune responses are more critical to control JE progression than adaptive T cell responses, which take time to develop. Indeed, the role of T cells in flavivirus encephalitis is less clear. This is, in part, due to variation of virus strain, the infection dose, the route of administration, mouse strain and age of the mice. Therefore, considering that the character of CD4^+^ and CD8^+^ T cells specific for JEV is also governed by innate immune responses initiated by recognition of PRRs, triggering of each PRR by direct viral components and/or host factors derived from infection could affect innate immune responses to shape adaptive immune responses, thereby influencing JE pathogenesis. A better understanding of the mechanisms that govern the induction of protective immunity plays a critical role in developing novel therapeutic strategies against JE.
Materials and Methods {#s4}
=====================
Animals and ethics statement {#s4a}
----------------------------
C57BL/6 (H-2^b^) and BALB/c (H-2^d^) mice (4--6 weeks old) were purchased from Samtako (O-San, Korea). TLR2 (H-2^b^), TLR3 (H-2^d^ and H-2^b^), TLR4 (H-2^b^), TLR7 (H-2^d^), and TLR9 (H-2^b^)-deficient mice have been described elsewhere [@ppat.1004319-Kawai1]. TLR3/4^−/−^ mice that are deficient in both TLR3 and TLR4 molecules were generated by backcrossing with TLR3 and TLR4-deficient mice. All mice were genotyped and bred in the animal facilities of Chonbuk National University. All experimental procedures were pre-approved and adhered to the guidelines set by the Institutional Animal Care and Use Committees (IACUC), Chonbuk National University (Permission code 2013-0028), which is fully accredited by the Korea Association for Laboratory Animal Sciences (KALAS).
Cells and viruses {#s4b}
-----------------
JEV Beijing-1 strain was obtained from Green Cross Research Institute (Suwon, Korea) and propagated in the mosquito cell line (C6/36) using DMEM supplemented with 2% FBS, penicillin (100 U/ml), and streptomycin (100 U/ml). The C6/36 cultures were infected with JEV Beijing-1 at a multiplicity of infection (MOI) of 0.1, and were incubated in a humidified CO~2~ incubator for 1 h at 28°C. After absorption, the inoculum was removed, and 7 ml of a maintenance medium containing 2% FBS was added. Approximately 6--7 days pi, cultures of the host cells showing an 80--90% cytopathic effect were harvested. The virus stocks were titrated by conventional plaque assay or focus-forming assay, and were stored in aliquots at −80°C until use.
Antibodies and reagents {#s4c}
-----------------------
The mAbs used for the flow cytometric analysis and other experiments were obtained from eBioscience (San Diego, CA) or BD Biosciences (San Diego, CA) which include: fluorescein isothiocyanate (FITC) conjugate-anti-CD3ε (145-2C11), CD4 (RM4-5), CD8 (53-6.7), CD44 (IM7), CD62L (MEL-14), CD69 (H1.2F3), Ly-6G (1A8), anti-rabbit IgG, phycoerythrin (PE) conjugate-anti-mouse-CD11b (M1/70), Foxp3 (FJK-16s), IFN-γ (XMG1.2), goat anti-mouse IgG, peridinin chlorophyll protein complex (PerCP) conjugate-anti-Ly-6C (HK1.4), PE-Cyanine dye (Cy7)-anti-mouse NK1.1 (PK136), allophycocyanin (APC) conjugate-anti-mouse-CD25 (PC62.5), Ly-6G (Gr-1), TNF-α (MP6-XT22). The peptides of the defined I-A^b^-restricted epitopes JEV NS1~132--145~ (TFVVDGPETKECPD), H-2D^b^-restricted epitope JEV NS4B~215--223~ (SAVWNSTTA) [@ppat.1004319-Trobaugh1], and H-2^d^-restricted epitope JEV E~60--68~ (CYHASVTDI) [@ppat.1004319-Takada1] were chemically synthesized at Peptron Inc. (Daejeon, Korea). Poly(I:C) was purchased from Sigma-Aldrich (St. Louis, MO). JEV-specific primers for the detection of viral RNA (JEV10,564--10,583 forward, 5′-CCC TCA GAA CCG TCT CGG AA-3′ and JEV10,862--10,886 reverse, 5′-CTA TTC CCA GGT GTC AAT ATG CTG T-3′) [@ppat.1004319-Aleyas1] and primers specific for cytokines, type I IFNs (IFN-α/β), and ISGs (**[Table S2](#ppat.1004319.s005){ref-type="supplementary-material"}**) were synthesized at Bioneer Corp. (Daejeon, Korea) and used for PCR amplification of target genes.
Quantitative real-time RT-PCR for viral burden and cytokine expression {#s4d}
----------------------------------------------------------------------
Viral burden, cytokine (IL-1β, IL-6, TNF-α, IFN-α, and IFN-β) and chemokine (CCL2, CCL3, CCL4, CCL5, and CXCL10) expression in inflammatory and lymphoid tissues were determined by conducting quantitative SYBR Green-based real-time RT-PCR (real-time qRT-PCR). Mice were infected intraperitoneally (i.p.) with JEV (1.4 × 10^7^ PFU) and tissues including brain, spinal cord, and spleen were harvested at 2, 3, 4, and 6 days pi following extensive cardiac perfusion with Hanks balanced salt solution (HBSS). Total RNAs extracted from tissues using easyBLUE (iNtRON, INC., Daejeon, Korea) were employed for real-time qRT-PCR using a CFX96 Real-Time PCR Detection system (Bio-Rad Laboratories, Hercules, CA). Following reverse transcription of total RNAs with High-Capacity cDNA Reverse Transcription Kits (Applied Biosystems, Foster, CA), the reaction mixture contained 2 µl of template cDNA, 10 µl of 2× SYBR Primix Ex Taq, and 200 nM primers at a final volume of 20 µl. The reactions were denatured at 95°C for 30 s, and then subjected to 45 cycles of 95°C for 5 s, and 60°C for 20 s. After the reaction cycle was completed the temperature was increased from 65°C to 95°C at a rate of 0.2°C/15 s, and the fluorescence was measured every 5 s to construct a melting curve. A control sample that contained no template DNA was run with each assay, and all determinations were performed at least in duplicate to ensure reproducibility. The authenticity of the amplified product was determined by melting curve analysis. The relative ratio of viral RNA in the infected samples to uninfected samples was determined. All data were analyzed using the Bio-Rad CFX Manager, version 2.1 analysis software (Bio-Rad Laboratories).
Cytokine and type I IFN ELISA {#s4e}
-----------------------------
### (i) IL-6 cytokine ELISA {#s4e1}
Sandwich ELISA was used to determine the levels of IL-6 cytokines in sera and culture supernatants. ELISA plates were coated with IL-6 (MP5-20F3) anti-mouse Ab purchased from eBioscience, and then incubated overnight at 4°C. The plates were washed three times with PBS containing 0.05% Tween (PBST), after which they were blocked with 3% nonfat-dried milk for 2 h at 37°C. The sera and culture supernatant and standards for recombinant cytokine proteins (PeproTech, Rocky Hill, NJ) were added to the plates, which were then incubated for 2 h at 37°C. The plates were washed with PBST again and biotinylated IL-6 (MP5-32C11) Ab was added. Next, the plates were incubated overnight at 4°C, followed by washing with PBST and subsequent incubation with peroxidase-conjugated streptavidine (eBioscience) for 1 h. Color development was then performed by the addition of a substrate (ABTS) solution. Cytokine concentrations were determined with an automated ELISA reader and SoftMax Pro3.4 according to comparisons with two concentrations of standard cytokine proteins.
### (ii) Type I IFN (IFN-β) ELISA {#s4e2}
A commercial ELISA kit (PBL Biomedical Laboratories, Piscataway, NJ) was used to measure levels of secreted IFN-β protein in sera and cell culture supernatants, according to the manufacturer\'s protocol.
Analysis of leukocytes infiltrated into the CNS {#s4f}
-----------------------------------------------
Mice infected with JEV were perfused with 30 ml of HBSS on day 3 pi *via* cardiac puncture of the left ventricle. Brains were then harvested, and homogenized by gently pressing them through 100-mesh tissue sieve, after which they were digested with 25 µg/ml of collagenase type IV (Worthington Biochem, Freehold, NJ), 0.1 µg/ml trypsin inhibitor *Nα*-*p*-tosyl-L-lysine chloromethyl ketone, 10 µg/ml DNase I (Amresco, Solon, OH), and 10 mM HEPE in HBSS for 1 h at 37°C with shaking. Cells were separated by using Optiprep density gradient (18/10/5%) centrifugation at 400 ×g for 30 min (Axis-Shield, Oslo, Norway), after which cells were collected from 18% to 10% interface and washed twice with PBS. Cells were then counted and stained for CD11b, Gr-1, Ly6G, Ly6C, CD3, CD4, CD8, and NK1.1 with directly conjugated antibodies (eBioscience) for 30 min at 4°C. Finally, the cells were fixed with 10% formaldehyde. Data collection and analysis were performed with a FACSCalibur flow cytometer (Becton Dickson Medical Systems, Sharon, MA) and FlowJo (Tree Star, San Carlos, CA) software.
Histological examinations and confocal microscopy {#s4g}
-------------------------------------------------
Brain derived from mock and JEV-infected mice were embedded in paraffin and 10-µm sections were prepared and stained with hematoxylin and eosin (H&E). Sections were analyzed using a Nikon Eclipse E600 microscope (Nikon, Tokyo, Japan). For confocal microscopy of brain tissue, brains were collected and frozen in optimum cutting temperature (OCT) compound (Sakura Finetechnical Co., Tokyo, Japan) following vigorous perfusion with HBSS. 6--7 µm thick sections were then cut, air-dried, and fixed in cold solution (1∶1 mixture of acetone and methanol) for 15 min at −20°C. Non-specific binding was blocked with 10% normal goat serum, and samples were then permeabilized with 0.1% Triton X-100. Staining was performed by incubating sections overnight in moist chamber at 4°C with anti-JEV and biotin conjugated anti-mouse CD11b (BD Biosciences, San Diego, CA) antibody. Primary antibodies were detected with secondary FITC-conjugated streptavidin and PE-conjugated goat anti-mouse Ab. Nuclei were counterstained with DAPI (4′6-diamidino-2-phenylindole) (Sigma-Aldrich). Finally, the fluorescence was observed by confocal laser scanning microscope (CalZeiss, Zena, Germany).
Evaluation of BBB permeability {#s4h}
------------------------------
Blood--brain barrier (BBB) permeability was determined by visualizing and quantifying extravasated Evans blue dye into the brain, as described earlier with some modification [@ppat.1004319-Wang2], [@ppat.1004319-Daffis1]. Briefly, JEV-infected mice were injected i.p. with 800 µl of 1% (w/v) Evans Blue dye (Sigma-Aldrich) 2 and 3 days pi, and perfused *via* intracardiac puncture with HBSS 1 h later. Brains were subsequently removed, weighed, and stored at −80°C following visualization with a high resolution digital camera. For Evans blue quantification, brain tissues were homogenized in 1 ml of PBS, and 1 ml of 100% trichloroacetic acid (TCA) (Sigma-Aldrich) was then added to the homogenate to precipitate proteins. The mixture was then vigorously shaken to precipitate the proteins for 2 min and cooled for 30 min at 4°C. After centrifugation (30 min at 4,000 ×g), the absorbance of the supernatant was measured at 620 nm using a spectrophotometer. The content of Evans blue was valued as micrograms of dye per gram of brain tissue by using a standard curve.
Primary cell culture and infection {#s4i}
----------------------------------
### (i) Myeloid-derived DCs and macrophages {#s4i1}
Myeloid-derived DCs (BMDC) and macrophages (BMDM) were prepared from bone marrow cells of TLR3^−/−^, TLR4^−/−^, and WT mice, as described earlier with some modification [@ppat.1004319-Aleyas1]. Briefly, for BMDC, bone marrow cells (3×10^5^ cells/ml) from femurs and tibiae were cultured in RPMI 1640 supplemented with 2 ng/ml GM-CSF and 10 ng/ml IL-4. On day 3, another 6 ml of fresh complete medium containing 2 ng/ml GM-CSF and 10 ng/ml IL-4 was added, and half of the medium was changed on day 6. On day 8, non-adherent and loosely adherent DCs were harvested by vigorous pipetting. Cells were then characterized by flow cytometric analysis, which revealed that the culture generally consisted of \>85% CD11c^+^ cells (25% CD11c^+^CD11b^+^ and 65% CD11c^+^CD8α^+^). BMDM was prepared by culturing bone marrow cells in DMEM supplemented containing 30% L929 cell-conditioned medium (LCCM) as a source of macrophage-colony stimulating factor (M-CSF). On day 3, another 6 ml of fresh complete medium containing 30% LCCM was added, and half of the medium was changed on day 6. The cultured cells were harvested following 8-day incubation and were analyzed by flow cytometry. The prepared BMDMs were composed of \>85% F4/80^+^ cells that consisted of 99.2% F4/80^+^CD11b^+^ and ∼1% F4/80^+^CD11c^+^ cells. Prepared BMDC and BMDM were infected with JEV at a MOI of 1.0 for viral replication and 10 MOI for cytokine expression.
### (ii) Primary cortical neurons {#s4i2}
Primary cortical neurons were prepared from 15-day-old embryo as described previously [@ppat.1004319-Szretter1]. Cortical neurons were seeded in 12-well poly-D-lysine/laminin-coated plates in DMEM containing 5% FBS and 5% horse serum for 24 h, and then cultured for 4 days with Neurobasal medium containing B27 supplement and L-glutamine (Invitrogen, Carlsbad, CA). Primary cortical neurons were infected with JEV at a 0.1 MOI for viral replication and type I IFN responses.
Western blotting {#s4j}
----------------
BMDC and BMDM infected with JEV were lysed in RIPA buffer (10 mM Tris, 150 mM NaCl, 0.02% sodium azide, 1% sodium deoxycholate, 1% Triton X-100, 0.1% SDS, pH 7.4) supplemented with protease inhibitors (iNtRON Biotech, Daejeon, Korea). Samples (15 µg) were resolved by electrophoresis on 10 to 12.5% SDS-polyacrylamide gels. After proteins were transferred to PVDF Immobilon-P Transfer Membrane (Millipore, Billerica, MA), blots were blocked with 5% non-fat dried milk or 3% BSA overnight at 4°C, and probed with the following panel of primary antibodies: rabbit anti-IRF-3, phospho-IRF-3 (Ser396), STAT1, phospho-STAT1 (Tyr701), mouse anti-IκBα (Amino-terminal Antigen) antibodies (Cell Signaling, Danvers, MA), and rabbit anti-IRF-7, phospho-IRF-7 (Ser471+Ser472) antibodies (Bioss Inc, Woburn, MA). Western blots were incubated with peroxidase-conjugated secondary antibodies (SouthernBiotech, Birmingham, AL) and visualized with WEST-ZOL Plus Immunoblotting detection reagents (iNtRON Biotech) using chemi-documentation system (Fusion Fx7, Vilber Lourmat, Cedex1, France).
JEV-specific antibody and CD4^+^/CD8^+^ T-cell responses {#s4k}
--------------------------------------------------------
JEV-specific IgM and IgG levels in sera of infected mice were determined by conventional ELISA using JEV E glycoprotein antigen (Abcam, Cambridge, UK). JEV-specific CD4+ and CD8+ T cell responses were determined by intracellular IFN-γ and TNF-α staining in response to antigen stimulation. Mice were infected i.p. with 2.8 ×10^6^ PFU of JEV and were sacrificed at day 7 pi and splenocytes were prepared. The erythrocytes were depleted by treating single-cell suspensions with ammonium chloride-containing Tris buffer (NH~4~Cl-Tris) for 5 min at 37°C. The splenocytes were cultured in 96-well culture plates (5×10^5^ cells/well) in the presence of synthetic peptide epitopes (NS1~132--145~ and NS4B~215--225~ or E~60--68~) for 12 h and 6 h in order to observe CD4^+^ and CD8^+^ T cell responses, respectively. CD4^+^ responses of BALB/c genetic background (H-2^d^) strain groups were evaluated by stimulation with UV-irradiated virus for 12 h at 37°C. Monensin at the concentration of 2 µM was added to antigen-stimulated cells 6 h before harvest. The cells were washed twice with PBS, and surface stained for FITC-anti-CD4 or CD8 antibodies for 30 min at 4°C, and then washed twice with PBS containing monensin. After fixation, the cells were washed twice with permeabilization buffer (eBioscience) and then stained with PE-anti-IFN-γ, or APC-anti-TNF-α in permeabilization buffer for 30 min at room temperature. Finally, the cells were washed twice with PBS and fixed using fixation buffer. Sample analysis was performed with FACS Calibur flow cytometer (Becton Dickson Medical Systems, Sharon, MA) and FlowJo (Tree Star, San Carlos, CA) software.
Analysis of myeloid-derived and immune cell subsets in lymphoid tissues {#s4l}
-----------------------------------------------------------------------
Splenocytes from infected mice were isolated and digested with collagenase (Roche) and type I DNase in serum-free RPMI media at 37°C for 40 min with mechanical disruption. Cells were washed twice with RPMI media containing 10% FBS before FACS staining. Myeloid-derived and immune cells were stained antibodies specific for CD11c, CD11b, B220, CD3, CD4, CD8α, NK1.1, DX5, Gr-1, Ly-6C, PDCA-1, CD25, and intracellular Foxp3. Finally, the cells were fixed with 10% formaldehyde, and analyzed with FACS Calibur flow cytometer and FlowJo software.
Statistical analysis {#s4m}
--------------------
All data were expressed as the average ± standard deviation, and statistically significant differences between groups were analyzed by unpaired two-tailed Student\'s *t*-test for *in vitro* experiments and immune cell analysis or ANOVA and post-hoc test for multiple comparisons of the mean. The statistical significance of viral burden and *in vivo* cytokine gene expression were evaluated by Mann-Whitney test or unpaired two-tailed Student\'s t-test. Kaplan-Meier survival curves were analyzed by the log-rank test. A *p*-value ≤0.05 was considered significant. All data were analyzed using Prism software (GraphPadPrism4, San Diego, CA).
Supporting Information {#s5}
======================
######
Susceptibility of TLR2^−/−^, TLR3^−/−^, TLR4^−/−^, TLR7^−/−^, TLR9^−/−^ mice to lethal encephalitis caused by JEV infection. TLR2^−/−^ (A), TLR3^−/−^ (B), TLR4^−/−^ (C), TLR7^−/−^ (D), and TLR9^−/−^ (E) mice (*n* = 7--13) were infected with JEV (1.4×10^7^ pfu) and were then monitored for mortality over 15 days. Data represent the proportion of surviving mice relative to challenged mice. Survival differences were statistically significant (B, *p = *0.0153; C, *p* = 0.0819).
(PDF)
######
Click here for additional data file.
######
Susceptibility and viral burden of TLR3^−/−^ mice (C57BL/6 genetic background) in lethal encephalitis caused by JEV infection. TLR3^−/−^ mice derived from C57BL/6 genetic background (H-2^b^) were infected with two doses, 1.4×10^7^ pfu (A) and 2.8×10^7^ pfu (B), of JEV, and then monitored for mortality, paralysis rate and body weight. The survival rate was examined over 15 days, and ratio of mice showing neurological disease during JE progression was examined every 6 h from 4 to 7 days pi. The change of body weight was expressed as the average percentage ± SD of weight relative to the time of challenge (*n* = 9--13). (C) Viral burden in lymphoid and inflammatory tissues during JE progression. Viral burden in spleen, brain, and spinal cord of TLR3^−/−^ (C57BL/6 genetic background) mice infected with JEV was assessed by real-time qRT-PCR at the indicated days pi. The viral RNA load was expressed by viral RNA copy number per microgram of total RNA (*n* = 5). Each symbol represents the level of an individual mouse; horizontal line indicates the median of each group.
(PDF)
######
Click here for additional data file.
######
TLR4 is dispensable to induce adequate adaptive immune responses specific for JEV antigen. (A) JEV-specific IgM and IgG levels. The levels of JEV-specific IgM and IgG in sera of TLR3^−/−^ and TLR4^−/−^ mice infected with sub-lethal dose of JEV (2.8×10^6^ pfu) were determined by ELISA at the indicated days pi. (B and C) JEV antigen-specific CD4^+^ T cell responses. Splenocytes were stimulated with UV-inactivated JEV (5 moi) for TLR3^−/−^ mice and NS1~132--145~ peptide (2 µg/ml) for TLR4^−/−^ mice at 7 days pi. The frequency (B) and total number (C) of JEV-specific CD4^+^ T cells were enumerated by the combined staining of surface CD4 and intracellular cytokines (IFN-γ and TNF-α). (D and E) JEV antigen-specific CD8^+^ T cell responses. Splenocytes were stimulated with E~60--68~ and NS4B~215--225~ peptides for TLR3^−/−^ and TLR4^−/−^ mice at 7 days pi, respectively. The frequency (D) and total number (E) of JEV-specific CD8^+^ T cells were enumerated by the combined staining of surface CD8 and intracellular cytokines (IFN-γ and TNF-α). The values in representative dot-plots denote the average of the indicated cell populations obtained from three individual experiment (*n* = 3--4). The bar in the graph represents the average ± SD of total number of the indicated cell population. \*, *p*\<0.001; \*\*, *p*\<0.01; \*\*\*, *p*\<0.05 compared with the levels of the indicated group.
(PDF)
######
Click here for additional data file.
######
Summary of the percentage and activation of splenic lymphocyte subsets in TLR3 and TLR4-deficient mice following JEV infection. TLR3 and TLR4-deficient (KO) mice (*n* = 3) were infected i.p. with JEV (1.4×10^7^ pfu/mouse). Splenocytes were prepared 3 days pi, stained with the indicated Abs, and were then analyzed by flow cytometry. Results are expressed as the average ± SD of positive cells (CD69^+^, CD62L^Low^, CD44^high^, or CD80^+^) in a given cell subset (CD4^+^, CD8^+^, or CD19^+^).
(PDF)
######
Click here for additional data file.
######
Specific primers for cytokines, chemokines, and type I IFNs, and ISGs used in real-time qRT-PCR.
(PDF)
######
Click here for additional data file.
We would like to thank Dr. Md Masudur Rahman for discussion and his help to write manuscript.
[^1]: The authors have declared that no competing interests exist.
[^2]: Conceived and designed the experiments: YWH JYC SKE. Performed the experiments: YWH JYC EU SBK JHK. Analyzed the data: YWH JYC SKE. Contributed reagents/materials/analysis tools: KK. Wrote the paper: YWH JYC SKE. Contributed histological examination: BSK.
| {
"pile_set_name": "PubMed Central"
} |
Introduction {#s0005}
============
Kupffer cells (KCs), the resident tissue macrophages of the liver have a crucial role in both the pathogenesis and the resolution of various liver diseases and inflammatory states including alcohol-induced liver injury [@b0005], non-alcoholic fatty liver disease associated with obesity [@b0010], ischemia reperfusion injury [@b0015], immune tolerance to organ transplantation [@b0015] and infectious disease [@b0020].
Resident tissue macrophages, including KCs, were historically considered a hematopoietic population, with replenishment of the tissue reservoir from monocyte-derived precursors in the steady state. This view has now been challenged with the majority of tissue macrophages shown to develop independently of haematopoietic stem cells, being seeded in the tissues prior to birth from a population of yolk sac (YS) derived macrophages [@b0025], [@b0030]. These cells show some level of radiation resistance [@b0035] and are independent of replenishment by monocytes in the steady state [@b0040], [@b0045]. Parallel studies that identified the transcription factors MafB and c-Maf as the factors that control the self-renewal of differentiated macrophages [@b0050], and the observation that tissue macrophages are capable of self-renewal in models of acute inflammation [@b0055] and under T-helper cell 2 conditions in the presence of interleukin (IL)-4 [@b0060], have confirmed that mature tissue macrophages are capable of proliferation and self-renewal. Finally, recent landmark studies have demonstrated that macrophage identity is unique to each macrophage population and is plastic, with phenotype conferred by microenvironment rather than cellular origin [@b0065], [@b0070], [@b0075]. Together, the above studies represent a paradigm shift in the field of tissue macrophage biology.
The studies described above all examined the origin of tissue macrophages only in the steady state. However, there is clear evidence that infection or tissue injury, and the associated inflammatory response, promotes the recruitment of myeloid cells, mostly monocytes into peripheral tissues. It is unclear when, and if, the newly infiltrated monocytes undergo differentiation into macrophages *in situ*. In fact, some authors have described the infiltrating cells as tissue macrophages [@b0080], sometimes as early as 24 h post infiltration [@b0085] without defining any phenotypic or functional changes in the cells. Although once infiltrated, monocytes begin differentiating into cells that are similar to the macrophages in the tissue that they reside in [@b0090], the function of these bone marrow (BM)-derived monocytes once they are present in the tissue has not been fully investigated. Given that KCs are implicated in the pathogenesis and the resolution of a number of liver diseases [@b0010], [@b0015], [@b0020], [@b0095], [@b0100], [@b0105] and their phagocytic capacity makes them an easy target for particle based therapeutics [@b0110], a greater understanding of KC biology and heterogeneity will facilitate the development of targeted liver therapeutics. Understanding whether distinct functions can be attributed to KCs of different origin will also be important for the design of new anti-infective strategies.
Here, we have used an irradiation bone marrow chimera model to enforce loss of YS-derived KCs and repopulation of the KC niche from BM-derived precursors. Using intravital microscopy to characterise the morphological and dynamic properties of YS- and BM-derived KCs *in situ*, and microarray analysis to examine gene expression profile, we show that after 6 weeks of differentiation in the liver, BM-derived "KCs" closely resemble but are not fully identical to the YS-derived KCs they have replaced. Whilst uptake of acetylated low density lipoprotein (Ac-LDL) was more prominent in YS-derived macrophages and the converse was true for bacterial uptake, for most of the functional studies we performed, these populations were functionally similar. This was particularly notable in their capacity to exert early control of and direct granuloma formation in response to infection with the KC-tropic intracellular protozoan parasite *Leishmania donovani*. These findings demonstrate that in the context of enforced liver inflammation, BM derived monocytes transition into KCs, which are as capable of protecting the host from infectious challenge as their YS-derived counterparts.
Materials and methods {#s0010}
=====================
Mice and infection {#s0015}
------------------
C57BL6 or B6.CD45.1 mice were obtained from Charles River (UK) or the Australian Resource Centre (WA). mT/mG [@b0005], lysMcre [@b0010] and B6.MacGreen [@b0015] mice have been previously described. Mice were bred and housed under specific pathogen-free conditions and used at 6--12 weeks of age. The Ethiopian strain of *Leishmania donovani* (LV9) and tandem Tomato fluorescent protein expressing LV9 (tdTom.LV9) [@b0020] were maintained by serial passage in *Rag1*^−/−^ mice. Amastigotes were isolated from infected spleens [@b0025], and mice were infected with 3 × 10^7^ *L. donovani* amastigotes intravenously (i.v.) via the tail vein in 200 μl of RPMI 1640 (GIBCO, UK). For the generation of chimeras, mice were placed on acidified water for at least 2 days prior to irradiation. Donor mice were irradiated with 1100 rads on a split-dose regimen (550 rads per dose, 24 h apart) and were then reconstituted with 2--5 × 10^6^ donor bone marrow cells via tail vein injection. Reconstituted mice were treated with oral antibiotics (Baytril) for 4 weeks post-reconstitution.
Liver enzyme analysis {#s0020}
---------------------
Heparinized blood was immediately centrifuged for 10 min at 300 g. Plasma was stored at −80 °C until analysis. Alanine aminotransferase (ALT) and aspartate transaminase (AST) levels were determined using a Beckman Unicell DxC800 analyzer in a single batch.
Confocal microscopy {#s0025}
-------------------
Confocal microscopy was performed on 20 μm frozen sections. For tissue containing tdTom expressing parasites, tissues were fixed in 4% paraformaldehyde (PFA) for two h before overnight incubation in 30% sucrose and embedding in OCT medium (Sakura). Antibodies were conjugated to Alexa488 or Alexa647 (eBioscience, UK). Slides were blinded before imaging on a Zeiss LSM510 axioplan microscope (Carl Zeiss Microimaging). Data were rendered and analysed using Volocity software (Improvision).
Ethics statement {#s0030}
----------------
All experiments were approved by the University of York Animal Welfare and Ethical Review Body and performed under UK Home Office license ('Immunity and Immunopathology of Leishmaniasis' Ref \# PPL 60/3708) or approved by the Queensland Institute of Medical Research Berghofer (QIMRB) animal ethics committee Ref \#P2076 (A1412-614).
Intravital imaging {#s0035}
------------------
Mice were anaethetised and surgery performed similar to previously described [@b0115] except that anaethesia was maintained by inhalation of 4% isofluorane (Abbott laboratories, UK). Images were acquired on an inverted LSM 780 multiphoton microscope (Carl Zeiss Microimaging), maintained at 36 °C by a blacked-out environmental chamber (Solent Scientific, UK). Images were acquired with a 40x 1.1 water immersion objective and fluorescence excitation provided by a Chameleon XR Ti:sapphire laser (Coherent) tuned to 870 nm.
Whole genome array {#s0040}
------------------
RNA was isolated from purified KC and amplified via Agilent low-input Quick Amp labelling kit (Agilent Technologies). Amplified RNA was then assayed with Agilent SurePrint G3 mouse GE 8 × 60 k microarray chips that were scanned with an Agilent C Scanner with Surescan High Resolution Technology (Agilent Technologies). The data were normalized using the percentile shift method to the 75^th^ percentile. Comparison of the gene expression data between liver resident and BM-derived KCs was performed using the Benjamini and Hochberg false discovery rate (FDR) correction [@b0120]. This analysis was performed with GeneSpring software (version 9; Agilent) as a standard 5% FDR, with the variances assessed by the software for each *t* test performed. A 2-fold expression criterion was then applied to each gene list. Gene ontology analysis was performed using the GeneSpring (Agilent) and Ingenuity pathway systems analysis software packages (Ingenuity Systems). Gene expression data is available from European Bioinformatics Institute ArrayExpress (accession number E-MTAB-4954).
Further methodology may be found in the [Supplementary materials and methods](#s0110){ref-type="sec"}.
Results {#s0045}
=======
Radiation-induced liver injury causes loss of a proportion of liver resident KCs and their replenishment from the bone marrow {#s0050}
-----------------------------------------------------------------------------------------------------------------------------
To study KCs, we used (LysM-Cre x mT/mG)~F1~ mice. To confirm that the majority of KCs expressed Cre recombinase and were therefore green fluourescent protein (GFP) positive in this system, we performed immunofluorescent staining on fixed liver tissue. By this assay, 95% ± 2.04% of F4/80^+^ KCs expressed GFP. Therefore, we used these mice in subsequent experiments. We generated BM chimeras by irradiating (LysM-Cre x mT/mG)~F1~ mice and reconstituting them with wild-type C57BL/6 BM ([Fig. 1](#f0005){ref-type="fig"}A). This process resulted in transient liver damage and inflammation as assessed by a small but non-significant increase in ALT and AST levels in the serum of irradiated mice compared to control, non-irradiated mice ([Fig. 1](#f0005){ref-type="fig"}B, C). Histological examination of H&E sections revealed detectable portal and lobular inflammation at 24 h ([Fig. 1](#f0005){ref-type="fig"}E) and 3 days post-irradiation in one out of 3 mice in each group ([Fig. 1](#f0005){ref-type="fig"}F), but this was no longer visible after 7 days ([Fig. 1](#f0005){ref-type="fig"}G).
A proportion of the YS-derived macrophages that express GFP were lost as a result of irradiation, and these cells were replaced by F4/80^+^ GFP^-^ BM-derived KCs in the chimeric mice at 6 weeks post-irradiation ([Fig. 1](#f0005){ref-type="fig"}H). Conversely, reciprocal (LysM-Cre x mT/mG)~F1~ →C57BL/6 chimeras resulted in a loss of a proportion of GFP^-^ YS-derived KCs and their replacement with BM-derived GFP^+^ F4/80^+^ KCs at 6 weeks post-irradiation ([Fig. 1](#f0005){ref-type="fig"}H). Flow cytometric analysis of digested livers ([Fig. 1](#f0005){ref-type="fig"}I) with gating on complement receptor of the immunoglobulin superfamily (CRIg)^+^ F4/80^hi^ cells ([Fig. 1](#f0005){ref-type="fig"}I) demonstrated that both GFP^+^ and GFP^-^ KCs were present in chimeric mice. Importantly, the generation of mixed BM chimeras in this way allows for analysis of the function of each population *in situ* under the same microenvironmental conditions.
We then studied the kinetics of KC depletion and repopulation after irradiation in (LysM-Cre x mT/mG)~F1~ → C57BL/6 chimeras. The percentage of GFP^+^ cells that express CRIg and F4/80 and the total number of CRIg^+^ F4/80^+^ GFP^+^ cells ([Fig. 1](#f0005){ref-type="fig"}J) increased over time after irradiation to reach approximately 50% of the KC population by 6 weeks post-irradiation. We next used 2-photon real-time *in vivo* imaging of these reciprocal chimeras ([Fig. 1](#f0005){ref-type="fig"}K and L) to examine the morphology and motility of each KC population *in situ*. In C57BL/6 → (LysM-Cre x mT/mG)~F1~ chimeras, GFP^+^ YS-derived KCs were large, interdigitating cells that were active in their membrane movements, but did not travel along the sinusoids ([Fig. 1](#f0005){ref-type="fig"}L; [Supplementary movie 1](#s0110){ref-type="sec"}). BM-derived KCs observed in (LysM-Cre x mT/mG)~F1~ → C57BL/6 chimeras also showed similar morphology and remained stationary within the sinusoids, confirming that they had become resident within the liver ([Fig. 1](#f0005){ref-type="fig"}L; [Supplementary movie 2](#s0110){ref-type="sec"}). Analysis of KC volume and surface area demonstrated that YS-derived KCs were somewhat larger than BM-derived KCs ([Fig. 1](#f0005){ref-type="fig"}K). Taken together, these data demonstrate that following genotoxic damage BM-derived cells differentiate morphologically into KCs and become resident in the liver.
An ion homeostasis gene signature distinguishes tissue-resident from BM-derived KCs {#s0055}
-----------------------------------------------------------------------------------
We next sought to determine whether BM-derived KCs had indeed acquired global characteristics of YS-derived KCs via a comparative gene expression approach. BM-derived and YS-derived KCs were isolated from chimeric mice by fluorescence activated high speed cell sorting according to size, granularity and expression of CRIg ([Fig. 2](#f0010){ref-type="fig"}A). KCs were then further separated into GFP^+^ and GFP^-^ cell fractions. Sorted populations were typically \>90% pure for the population of interest. Giemsa stained cytopsin preparations of sorted cells demonstrated that BM-derived and YS-derived KCs had similar macrophage-like morphology ([Fig. 2](#f0010){ref-type="fig"}B). Gene expression analysis by microarray demonstrated that of the annotated genes that were represented on the chips, BM-derived and YS-derived KCs shared expression of \>99% of the probes, with only 42 probes meeting criteria for differentially binding of the cDNA between the two populations (5% FDR with 2-fold cut-off, [Table 1](#t0005){ref-type="table"}, [Fig. 2](#f0010){ref-type="fig"}C). Of note, the binding of these 42 transcripts was present for the resident KC population and lacking in BM-derived KCs. There were no transcripts uniquely present in BM- derived KCs. KCs showed similar transcript profiles whether isolated from chimeras that resulted in them being GFP^+^ or GFP^-^, with no significant differences in expression observed within the GFP^+^ and GFP^-^ fractions within the same KC population (data not shown).
Validation of differential expression of a proportion of the genes listed in [Table 1](#t0005){ref-type="table"} was then performed by measuring mRNA abundance by real-time PCR, using RNA isolated from independently sorted samples of each KC population. mRNA abundance for *Cd163*, *Marco*, *Ric3*, *Colec12* and *Timd4* ([Fig. 2](#f0010){ref-type="fig"}D) was higher in YS-derived KCs than in BM-derived KCs, confirming the microarray data ([Fig. 2](#f0010){ref-type="fig"}C). Abundance of *Clec4f* mRNA was similar between the populations, also confirming the microarray data and validating that a transition to KC, as defined by *Clec4f* (Kupffer Cell Receptor), had occurred in the BM-derived KC population.
Macrophage Receptor with Collagenous Structure (MARCO) was one of the most differentially expressed genes found in YS-derived KCs, with 15-fold higher mRNA abundance than seen in BM-derived KCs. We therefore analysed the expression of MARCO protein in KCs in C57BL/6 → (LysM-Cre x mT/mG)~F1~ chimeras. Clear co-expression of MARCO and GFP was observed in YS-derived KCs, but very little detectable expression of MARCO in GFP^-^ BM-derived KCs ([Fig. 2](#f0010){ref-type="fig"}E). In further validation experiments, flow cytometry on isolated hepatic mononuclear cells from B6.MacGreen → C57BL/6 chimeras demonstrated that Tim4 was expressed in 73.4 ± 4.21% of YS-derived KCs and only expressed in 5 ± 0.26% of BM-derived KCs confirming that Tim4 expression is enriched within the YS-derived KC population.
To gain a better understanding of the implications of the transcriptomic differences between YS-derived and BM-derived KCs, we performed gene ontology (GO) analysis. 3 of the 4 GO terms that showed highly significant enrichment (*p* \<0.001) were associated with ion homeostasis ([Supplementary Table 1](#s0110){ref-type="sec"}). This included GO:0055065; metal ion homeostasis, GO:0055080; cation homeostasis and GO:0050801; ion homeostasis. This enrichment was associated specifically with the expression of *Ric3*, *Ank2*, *Slc22a17*, *Epor* and *Hmox1* in YS-derived KCs ([Fig. 2](#f0010){ref-type="fig"}C). Resistant to inhibitors of cholinesterase 3 (Ric3) is a transmembrane protein that controls expression of nicotinic acetylcholine receptors, which are gated ion channels [@b0125]. Ankyrin 2 (Ank2) is also associated with ion channels, being an adaptor protein for connection of ion channels to the actin cytoskeleton [@b0130]. Slc22a17 (also known as 24p3R) is associated with iron uptake and apoptosis [@b0135]. In addition, mRNA for the erythropoietin receptor (*Epor*), hemeoxygenase 1 (*Hmox1*), both associated with red blood cell homeostasis and the haemoglobin scavenger receptor *Cd163* [@b0140] were more abundant in YS-derived KCs, the latter being one of the most differentially expressed genes ([Fig. 2](#f0010){ref-type="fig"}C).
BM-derived and YS-derived KCs exhibit comparable clearance of effete red blood cells {#s0060}
------------------------------------------------------------------------------------
We next set out to determine if both KC populations were comparable in a range of functional assays. First we studied another essential function of KCs, namely the clearance of effete red blood cells (RBCs). We injected neuraminidase treated PHK-26-labelled RBCs [@b0145] intravenously into chimeric mice and assessed their uptake by BM-derived or YS-derived KC populations by flow cytometry. These data showed that at 2 weeks post-irradiation, although still at low frequencies in the liver ([Fig. 1](#f0005){ref-type="fig"}), differentiated BM-derived KCs were capable of phagocytosing RBCs with the same efficiency as their YS-derived counterparts ([Fig. 3](#f0015){ref-type="fig"}A). A similar pattern was observed at 6 weeks post-irradiation ([Fig. 3](#f0015){ref-type="fig"}B). Hence, differentiation to allow for this essential KC function occurs rapidly once BM-derived cells enter the liver microenvironment.
YS-Derived macrophages more efficiently accumulate acetylated low density lipoproteins {#s0065}
--------------------------------------------------------------------------------------
Given that YS-derived macrophages had a scavenger phenotype, we next utilised the BM chimera model to examine the uptake of acetylated low density lipoprotein (Ac-LDL) by BM- and YS-derived liver macrophages. Hepatic mononuclear cells from BM chimeric mice were isolated and incubated with fluorescently labelled Ac-LDL. The proportion of YS-derived macrophages that were positive for LDL accumulation was significantly higher than that for BM-derived macrophages ([Fig. 3](#f0015){ref-type="fig"}C), consistent with the differential expression of scavenger receptors in these two populations.
*In vivo* response of KC to LPS challenge {#s0070}
-----------------------------------------
KCs play a major role in the innate immune response to infection and are continually conditioned by endotoxin draining from the intestinal tract [@b0150], [@b0155], [@b0160]. Previous *in vitro* studies have demonstrated that multiple exposure to LPS may lead to a state of LPS tolerance, wherein a second exposure to LPS fails to induce gene expression to a similar extent as the primary exposure [@b0165]. Furthermore, Medzhitov and colleagues have argued that LPS inducible genes can be classified as "tolerizable" or non-tolerizable", reflecting different epigenetic regulation of transcription [@b0170]. It might be expected that KCs had mechanisms to avoid loss of function due to LPS tolerance and that this might represent one aspect of tissue specific conditioning [@b0075]. We therefore asked whether genes that were not expressed by BM-derived KCs in the steady state were LPS inducible. We treated bone marrow chimeric mice with LPS, sorted the BM-derived and YS-derived KCs 24 h later and used real-time PCR to analyse mRNA abundance for a selected panel of genes. *Marco* is an example of a non-tolerizable gene and is expressed exclusively in YS-derived KCs ([Fig. 4](#f0020){ref-type="fig"}A). LPS treatment had no effect on *Marco* mRNA levels in either YS-derived or BM-derived KCs ([Fig. 4](#f0020){ref-type="fig"}A) suggesting that the differences observed in baseline levels of *Marco* mRNA were not a result of differences in long-term LPS exposure between the two populations. LPS treatment resulted in a down-regulation of *Cd163* mRNA abundance in YS-derived KCs, but had no effect on the minimal abundance of *Cd163* mRNA in BM-derived KCs ([Fig. 4](#f0020){ref-type="fig"}B). Similarly, LPS treatment reduced the abundance of *Ric3* and *Timd4* mRNA in YS-derived KCs and decreased *Clec4f* mRNA abundance in both populations of KCs ([Fig. 4](#f0020){ref-type="fig"}C--E). These data indicate that even high dose LPS exposure does not drive the final differentiation of BM-derived KCs to match the gene expression profile of YS-derived KCs, at least for the genes assayed here, and it appears that both populations are capable of responding to LPS challenge.
YS-derived and BM-derived KCs show differential uptake of bacteria pathogens {#s0075}
----------------------------------------------------------------------------
Given that KCs are central to the clearance of systemic bacteria and MARCO has been associated with bacterial uptake [@b0175], [@b0180], [@b0185], we next assessed the ability of YS- and BM-derived macrophages to take up three different bacterial pathogens. For these experiments, we generated B6.MacGreen → B6.CD45.1 BM chimeras and injected fluorescently labelled, heat-killed *Salmonella enterica* subspecies *enterica* serovar typhimurium (*S*. typhimurium), *Neisseria meningitidis* or *Listeria monocytogenes*. Phagocytosis was assessed 2 h after intravenous injection of bacteria by flow cytometry on isolated hepatic monocytes (MNCs) or by fluorescence microscopy on whole liver tissue ([Fig. 5](#f0025){ref-type="fig"}). Unexpectedly, our data showed that a greater proportion of BM-derived KCs phagocytosed *N. meningitidis* and *L. monocytogenes*, compared to YS-derived KC, with a similar trend also observed for *S. typhimurium* ([Fig. 5](#f0025){ref-type="fig"}E). Collectively with our data above on erythrophagocytosis, we conclude that whilst phagocytosis per se is not a unique property of either KC population, differences in phagocytic clearance rate based on ligand specificity exist.
YS-derived and BM-derived KCs respond similarly to *Leishmania* infection {#s0080}
-------------------------------------------------------------------------
Finally, we evaluated the ability of these two KC populations to respond to and contain infectious challenge by a KC-tropic intracellular parasite. *Leishmania donovani* infection of mice via the intravenous route leads to rapid KC infection and the subsequent T cell-dependent generation of inflammatory foci termed granulomas [@b0115], [@b0190], [@b0195]. We infected reciprocal BM chimeric mice with transgenic tdTomato-expressing *L. donovani* amastigotes, and determined the proportion of KCs of each origin that were infected. Unlike the situation observed with bacterial uptake ([Fig. 5](#f0025){ref-type="fig"}), YS-derived and BM-derived KCs did not differ in their ability to phagocytose *L. donovani*, ([Fig. 6](#f0030){ref-type="fig"}A). Both populations were capable of killing parasites, as judged by a decrease in the percentage of infected cells observed by 48 h post-injection ([Fig. 6](#f0030){ref-type="fig"}A), though this difference only reached significance for BM-derived KCs. The difference between BM-derived and YS-derived KCs at 48 h post-infection was not significant ([Fig. 6](#f0030){ref-type="fig"}A) suggesting that both populations are capable of controlling infection with *L. donovani*. By 7 days post-infection, both KC populations showed a reduction in percentage of infected cells when compared to the 2 h time point ([Fig. 6](#f0030){ref-type="fig"}A), indicating a similar level of control of infection.
Furthermore, quantifying the number of parasites within each infected KC showed that BM-derived and YS-derived KCs phagocytosed the same number of parasites at 2 h post-injection and had equal numbers of parasites per cell by 7 days post-infection ([Fig. 6](#f0030){ref-type="fig"}B). These data, in conjunction with those shown in [Fig. 6](#f0030){ref-type="fig"}A suggest that those KCs that do not clear infection are able to similarly support parasite multiplication and an increase in the mean number of parasites present per cell ([Fig. 6](#f0030){ref-type="fig"}B).
As an additional measure of function, we examined the ability of these two populations of KCs to become the focus for granuloma formation. Inflammatory foci were scored into 5 categories: foci that contained only YS-derived KCs, mostly YS-derived KCs a 50:50 mix of YS- and BM-derived KCs, mostly BM-derived KCs and only BM- derived KCs. By this analysis, we were able to show that inflammatory foci were more likely to be formed around BM-derived KCs than around YS-derived KCs ([Fig. 5](#f0025){ref-type="fig"}C), although both KC populations were capable of seeding granuloma formation. Examples of granulomas from YS-derived KCs ([Fig. 5](#f0025){ref-type="fig"}D), BM-derived KCs ([Fig. 5](#f0025){ref-type="fig"}E) and mixed granulomas ([Fig. 5](#f0025){ref-type="fig"}F), were evident in all infected mice.
In summary, our data collectively argue that BM-derived KC are not only capable of differentiating in the liver to look and act like YS-derived KCs but are also capable of responding to an infectious challenge in a similar way.
Discussion {#s0085}
==========
The recent investigations into the origins of tissue macrophages in mice have demonstrated that in the lung, spleen, skin, pancreas, kidney and liver, tissue-resident macrophages are not derived from haematopoietic precursors in the steady state, but are seeded in embryonic tissues (YS-derived) and self-maintain locally in adult tissues [@b0025], [@b0040], [@b0045]. We have now shown that following irradiation induced liver damage, YS-derived KCs are partially replaced by BM-derived precursors, and that these cells differentiate in the liver into mature KCs, where they become resident and share \>99% of their gene expression with YS-derived KCs. Although functionally distinct regarding Ac-LDL uptake (YS \>BM) and phagocytosis of bacteria (BM \>YS), these newly differentiated, BM-derived KCs are equally capable of responding to soluble (LPS) and parasitic insult as YS-derived KCs and to equally perform essential housekeeping functions like RBC clearance. We have also identified a combination of phenotypic markers including Marco and Tim4 that along with well characterised molecules including CRIg and F4/80 can be used to phenotypically separate liver macrophages that are recently BM-derived or of YS-origin.
The recent investigations and characterisation of Clec4f through gene knockout and sequencing approaches [@b0075], [@b0200] have confirmed Clec4f as a KC-restricted molecule, making it a good candidate molecule for defining KC differentiation. These investigations have now been further expanded by Scott *et al.* to include specific depletion of KC in the liver through the use of KC-DTR mice in which the human DTR gene was inserted into the 3' untranslated region of the *Clec4f* gene. Administration of diphtheria toxin (DT) resulted in specific depletion of KCs in the liver [@b0090]. We found that BM-derived KCs up regulate *Clec4f* gene expression and take up their characteristic stationary sinusoidal position, becoming KCs by all morphologic and dynamic definitions that are currently available. In their recent study, Lavin *et al.* described steady state (and presumably mostly YS-derived) KC as most closely associated with red pulp macrophages of the spleen, in terms of gene expression in these two populations. It is noteworthy, therefore, that YS-derived KCs, but not BM-derived KCs in our study have abundant mRNA for *Ric3*, *CD163*, *St6galnac2*, *Epor*, *Hmox1* and *Ecm1*, genes also reported in the ImmGen database as highly expressed in red pulp macrophages. These data suggest that full acquisition of housekeeping properties conserved between splenic and YS-derived KCs has not yet occurred in the BM-derived KCs we have studied.
Whilst this manuscript was under review, Scott *et al.* published gene signatures distinguishing YS-derived KCs from BM-derived KCs obtained at different times post DTR-mediated depletion of YS-derived KCs [@b0090]. Their data show many similarities, but some differences with that presented here. Of note, however, five (*Timd4*, *Colec12*, *Cd163*, *Snrpn* and *Xlr*) of the twelve genes described as differentially expressed in YS but not BM-derived KCs in their study [@b0090] were also identified by us, providing confidence in their assignment independent of methodology used.
In addition to the 5 commonly identified genes associated with YS-derived KCs, we also identified a further 37 transcripts that were differentially expressed. Analysis of these showed enrichment in YS-derived KCs for GO terms related to ion homeostasis, and differential expression of genes associated with red blood cell phagocytosis and turnover. However, our data using an *in vivo* phagocytosis assay demonstrated no difference in the ability of BM-derived and YS-derived KCs to phagocytose labelled red blood cells, suggesting that the two populations were equally capable of contributing to this important KC function [@b0205]. Nevertheless, further studies will be required to determine whether differences exist between the two populations in the scavenging of haemoglobin and the break down and recycling of red blood cells.
It was of significant note that there were no genes expressed by BM-derived cells that were not expressed by YS-derived KCs. This is in contrast to the findings of Scott *et al.* who found two genes (*Ccr3 and Tspan32*) that were uniquely expressed by bone marrow-derived KCs [@b0090]. Together, the similarity in gene expression profiles in these two studies suggest that BM-derived monocytes are malleable in nature and responsive to the microenvironmental cues they receive, enabling them to differentiate and perform functions required by the tissue that they become resident in. It would follow that this is not hindered by extensive residual expression of a BM-specific gene signature that cannot be reprogrammed when required.
These studies confirm and further extend the observations of others, that irradiation induces a loss of a proportion of YS-derived macrophages in a number of tissues [@b0035], [@b0040], [@b0210], triggering repopulation from BM-derived precursors. In non-radiation-induced depletion models, loss of a proportion of the tissue-resident macrophage population, induces local proliferation and repopulation by self-renewal in the lungs, liver, epidermis and brain [@b0025]. In contrast, the depletion of tissue-derived macrophages via irradiation appears to trigger a different mechanism of repopulation, with long-term seeding of hematopoietic stem cell-derived cells being a predominant feature of repopulation in the spleen, lungs, peritoneum and bone marrow [@b0040]. This difference is likely associated with the absence (conditional gene targeting) *vs.* presence (irradiation) of collateral damage and the induction of stress induced responses in a multitude of different cell types in any given organ. However, in the liver, complete loss of KCs via clodronate liposome mediated depletion also results in repopulation of KCs from BM precursors [@b0035], implying that there may be a threshold that triggers differentiation of BM-derived cells when the niche cannot be sufficiently filled by the proliferation of the YS-derived population. This phenomenon may be a secondary safety net for the function of the liver, as the BM-derived KCs appear to be capable of differentiating *in situ* to have a very similar gene expression profile to the cells they replace, with a similar ability to respond to an infectious challenge.
The use of the BM chimera model allowed us to evaluate the functional capacity of YS- and BM-derived KCs in a uniform microenvironment, obviating any differences in global liver function that might otherwise have arisen if we had used a total replacement strategy [@b0090]. Uptake of *E. coli in vitro* by isolated YS- and BM-derived KCs has been reported to be similar [@b0090]. In contrast, to study KC function we used an extended battery of *in vivo* assays that retain natural microenvironmental context and that employed both soluble and particulate ligands. In addition, we used a well- characterised parasitic model that allowed us to monitor innate microbicidal activity as well as T cell-dependent host protective immunity (expressed as granulomatous inflammation). Whereas uptake of Ac-LDL was more efficient in YS-derived KCs, we found that BM-derived KCs were more phagocytic towards *L. monocytogenes*, *N. meningitides* and *S. typhimurium.* As MARCO has been implicated in uptake of each of these bacteria, this result was unexpected. The differences were not absolute, but may suggest that with increasing ligand complexity, there is greater scope for receptor redundancy and/or cooperativity, a level of complexity that is not fully captured by gene expression analysis. For example, both *N. meningitides* and *L. monocytogenes* utilise SR-A in addition to MARCO to interact with macrophages [@b0175], a molecule that showed no difference in expression between the two macrophage populations at an RNA level in our gene expression studies.
Our data are not without clinical relevance, where radiation-induced liver damage (radiation hepatitis) can result in hepatic fibrosis. BM transplantation for treatment of haematological malignancy uses total body irradiation as a pre-conditioning strategy prior to transplantation. In this system, the remaining host macrophages [@b0215] including KCs in the liver [@b0220] have been implicated in the suppression of the pathological graft *vs.* host disease (GVHD) response. Whether a similar level of radiation resistance is observed in human KCs requires further investigation. Given that inflammatory events and the resultant cellular turnover are likely common in the liver *in vivo*, it is essential that BM-derived KCs have host protective functions as well as functional competence to maintain the homeostatic functions of the liver. In spite of some selected functional differences, our data provide strong evidence that this is indeed the case.
Financial support {#s0090}
=================
This work was funded by the UK Medical Research Council (Grant \#G0802620) and the Australian National Health and Medical Research Council (Grant \#APP1105817).
Conflict of interest {#s0095}
====================
The authors who have taken part in this study declared that they do not have anything to disclose regarding funding or conflict of interest with respect to this manuscript.
Authors' contribution {#s0100}
=====================
LB conceived and designed the study, performed experiments, analysed data and wrote the manuscript. AS, JM, TCMF, BT, FLR, KW-B, JWJM, EKL and JED performed and analysed some of the experiments. SM analysed the gene expression array data. Cre and KPM contributed to data analysis. PMK conceived and designed the study, analysed data, and wrote the manuscript.
Supplementary data {#s0110}
==================
Supplementary data 1Supplementary data 2Supplementary movie 1Supplementary movie 2
All microarray data are deposited in EBI ArrayExpress (tbc on publication). We wish to thank the staff of the Biological Services Facility and P. O'Toole, K. Hodgkinson, K. Hogg, G. Park, J. Marrison, and C. Whalley (Technology Facility, Department of Biology, University of York) for technical assistance. We wish to thank Genentech Inc for supplying the anit-CRIg antibody. We wish to thank Dr Cheryl-Lynn Ong, Dr Jayde Gawthorne, Saiyuri Naicker, Dr Nick West, and Professor Alastair McEwan (School of Chemistry and Molecular Biosciences, University of Queensland) for the supply of the heat killed bacterial strains. Thank you to Professor Andrew Clouston (Envoi Pathology, and the University of Queensland) for scoring pathology in the liver post-irradiation. This work was supported by the UK Medical Research Council (Grant \#G0802620) and the Australian National Health and Medical Research Council (Grant \#APP1105817).
Supplementary data associated with this article can be found, in the online version, at [http://dx.doi.org/10.1016/j.jhep.2016.05.037](10.1016/j.jhep.2016.05.037){#ir005}.
{#f0005}
{#f0010}
{#f0015}
{#f0020}
{#f0025}
{#f0030}
######
**The genes differentially expressed between liver resident and BM derived KC.**
--------------
--------------
| {
"pile_set_name": "PubMed Central"
} |
1. Introduction
===============
Recently, an increasing number of patients with critical limb ischemia (CLI) have been undergoing medical care at clinical practice sites. Improving the outcome of treatment for these patients is an important and urgent issue. Since 2013, the Japanese Society for Vascular Surgery (JSVS) initiated a nationwide CLI registration and tracking database project to obtain CLI epidemiological data that can be shared among medical staff. The background of CLI limbs, contents of treatment, early outcome, and long-term outcome until five years after surgery, including non-surgical limbs, are registered in this database. The database was named JAPAN Critical Limb Ischemia Database (JCLIMB) and established on the National Clinical Database (NCD). The JCLIMB project's primary objective is to clarify the current status of CLI treatment performed by vascular surgeons in Japan and inform physicians at practice sites, thus improving the quality of medical care. The initial registration data and their tracking data one month after registration in 2013-2016, have already been published.^[@R1]--[@R8])^ In this paper, for the purpose of clarifying the picture of clinical practice of CLI in Japan, we have compiled these data over the past four years. In addition, a part of data is represented by means of graphs to enhance convenience for readers. Graphs were made using overall data.
2. JCLIMB
=========
Registration details, including the definition of CLI, have already been described in the 2013 annual report. CLI to be registered was defined according to TASC II^[@R9])^: chronic ischemic rest pain, ulcers, or gangrene attributable to objectively-proven arterial occlusive disease. CLI diagnosis should be confirmed by ankle pressure (AP) below 50 mmHg or toe pressure (TP) below 30 mmHg in limbs with rest pain, and done by AP below 70 mmHg or TP below 50 mmHg in limbs with ulcer or gangrene.
The same limb can be registered in JCLIMB only once within a five-year tracking period. When the registered limb is treated at different times or different institutions, such data should be added only to the tracking items of each limb in JCLIMB, avoiding registration overlap. However, details of the procedure are registered each time in NCD apart from the registration in JCLIMB. On the other hand, the patient with bilateral CLI can be registered twice for each limb. Based on NCD regulations, fixing JCLIMB data is done as follows:
1. Initial registration data: Early April in the following year,
2. Tracking data early after treatment (one month)/six months after treatment: End of December in the following year
3. Tracking data one year after treatment: End of December after two years
4. Tracking data two years after treatment: End of December after three years
5. Tracking data three years after treatment: End of December after four years
6. Tracking data four years after treatment: End of December after five years
7. Tracking data five years after treatment: End of December after six years
As a general rule, the timing of tracking data registration is accepted within a ±2-month range until 12 months after treatment, and within a ±3-month range thereafter. Although the day for tracking data fixing is specified, it is made flexible because, in some limbs, follow-up data might be revealed later.
It is very difficult to require facilities participating in NCD to register CLI data since a great number of registration items in JCLIMB would put too much burden on them. Thus, facilities wishing to participate were recruited. Facilities which registered CLI limbs in each year were listed in the appendix in each annual report.
The audit of data registered in JCLIMB and the audit of data regarding vascular surgery registered in NCD was started in 2018. Committee members visit the selected institutes to collate the registered data to the data of clinical chart.
Since JCLIMB is positioned as a registry study on NCD, patient consent to participate in the study, and the ethical review of the study at the time of participation in NCD were adopted.
3. Comments on the Aggregated Data in 2013--2016
================================================
This paper shows the aggregated date which have already been reported in annual reports from 2013 to 2016. The date fixing the data was described in each annual report. For 4 years, 4,784 limbs, those of 3,361 males (70%) and 1,423 females (30%), were registered. All data and extracted data on arteriosclerosis obliterans (ASO) were collected according to the registered items. Since ASO accounted for 98% of all limbs, the overall and ASO data showed similar tendencies. In the comments, ASO data were presented in parentheses. In addition, because the Society for Vascular Surgery (SVS)'s WIfI classification was reported in 2014,^[@R10])^ JCLIMB made several changes and additions to the registered items and the aggregated data on WIfI classification were compiled from the data of 2015 and 2016 annual reports. The site of wound (either gangrene or ulcer) was registered in the item of "Sites of ulcer/gangrene" until 2014, while the ulcer and gangrene have been registered separately since 2015. Accordingly, the numerical values in "Main sites of ulcer/gangrene to be treated" in 2015 and 2016 were used for the 4-years' aggregated data of "Sites of ulcer/gangrene." For the aggregated data of "Vein usage and vein quality," the data in 2015 and 2016 were used. The total figure was not always consistent, mostly due to missing values. The comments to these problems were different depending on each annual report, which should be referred.
(1) Pretreatment patients' background
-------------------------------------
Pretreatment patients' background is shown in [**Tables 1-1**](#table1-1){ref-type="table"} to [](#table1-2){ref-type="table"}[](#table1-3){ref-type="table"}[](#table1-4){ref-type="table"}[](#table1-5){ref-type="table"}[**1-6**](#table1-6){ref-type="table"}. Good blood pressure control was defined as below 140/90 mmHg, without diabetes and renal failure, or below 130/80 mmHg with these diseases. Diabetes control was considered good when hemoglobin A1c was below 7.0% (national glycohemoglobin standardization program value). Dyslipidemia control was considered good when low-density lipoprotein was below 100 and 80 mg/dL in the absence and presence of other arteriosclerotic diseases, respectively. The presence of heart failure was judged clinically. The patient was regarded as having heart failure based on a past history of admission due to heart failure, clinical symptoms of heart failure, a diagnosis of heart failure was confirmed by echocardiography, or reduced cardiac function on echocardiography even with no clinical heart failure symptoms. Renal dysfunction was graded following the new chronic kidney disease severity classification of the "Clinical Practice Guidebook for Diagnosis and Treatment of Chronic Kidney Disease 2012"^[@R11])^: Renal dysfunction was absent when the estimated glomerular filtration rate (eGFR) (mL/min/1.73 m^2^) was 60 or higher, and graded as G3a, G3b, G4, and G5 when eGFR was 45--59, 30--44, 15--29, and below 15, respectively. eGFR below 15 in hemodialysis patients was graded as G5D.
###### Table 1 Patients' background Table 1-1 Patients' background 1
a\. Total
-------------- -------- ------- ------------ -------------- --------------------- ------- ------------- ---- ---- ---- ------------- ------------- ------------- ------------- -------------
Rutherford 4 1,079 742 337 549 530 21.08 1,054 6 2 7 73.4 (9.7) 51.0 (21.6) 57.0 (17.0) 55.1 (8.1)
Rutherford 5 2,982 2,098 884 1,513 1,469 21.05 2,907 28 36 11 73.3 (10.1) 52.4 (16.0) 68.3 (11.9) 75.2 (9.4)
Rutherford 6 723 521 202 350 373 21.14 712 3 5 3 71.0 (10.4) 54.3 (4.5) 70.8 (4.9) 60.0 (6.2)
Total 4,784 3,361 1,423 2,412 2,372 21.08 4,683 37 43 21 73.0 (10.1) 52.4 (16.1) 68.1 (11.6) 66.3 (12.7) 72.7 (10.3)
b\. ASO
n Sex Laterality BMI (Median) Age at registration
Male Female Right Left Mean (±SD)
Rutherford 4 1,064 732 332 541 523 21.08 73.4 (9.7)
Rutherford 5 2,907 2,056 851 1,478 1,429 21.08 73.3 (10.1)
Rutherford 6 712 514 198 346 366 21.17 71.0 (10.4)
Total 4,683 3,302 1,381 2,365 2,318 21.09 73.0 (10.1)
Vasculitis: Takayasu's arteritis, collagen disease, Behcet disease, FMD etc., excluding TAO Others: others (including debranch bypasses for TEVAR or EVAR) ASO: arteriosclerosis obliterans, TAO: thromboangiitis obliterans, FMD: fibromuscular dysplasia, BMI: body mass index, TEVAR: thoracic endovascular aortic/aneurysm repair, EVAR: endovascular aortic/aneurysm repair
###### Table 1-2 Patients' background 2
a\. Total
-------------- ------------ ------------------ -------------- ----------------- --------- ------- ------- ------- ----- ------- ------- ----- ------- ------- -----
Rutherford 4 522 450 107 88 306 163 274 698 107 656 342 81 423 433 223
Rutherford 5 950 1,590 442 271 928 833 705 1,921 356 1,833 984 165 1,228 1,289 465
Rutherford 6 172 399 152 84 202 265 221 415 87 448 229 46 295 325 103
Total 1,644 2,439 701 443 1,436 1,261 1,200 3,034 550 2,937 1,555 292 1,946 2,047 791
b\. ASO
Diabetes Diabetes therapy Hypertension Dyslipidemia Smoking
(−) (+) Diet therapy Medication Insulin therapy (−) (+) (−) (+) (−) (+)
Management Management Management Ex-smoker Current smoker
Good Poor Good Poor Good Poor
Rutherford 4 511 446 107 87 305 161 264 693 107 645 338 81 420 426 218
Rutherford 5 884 1,584 439 271 922 830 664 1,891 352 1,779 966 162 1,199 1,255 453
Rutherford 6 163 398 151 84 201 264 213 412 87 440 226 46 291 321 100
Total 1,558 2,428 697 442 1,428 1,255 1,141 2,996 546 2,864 1,530 289 1,910 2,002 771
Blood pressure management good: diabetes or renal failure (−) \<140/90 mmHg, (+) \<130/80 mmHg. Diabetes management good: HbA1c\<7.0%(NGSP). Dyslipidemia management good: other sclerotic lesions (−) LDL\<100 mg/DL, (+) LDL\<80 mg/DL. HbA1c: hemoglobin A1c, LDL: low-density lipoprotein, NGSP: national glycohemoglobin standardization program
###### Table 1-3 Patients' background 3
a\. Total
------------------- ------------------------ --------------- ------------------------- ------------------- ------- ----- ------- ------- ------- ----- ----- ----- ---- -------
Rutherford 4 703 134 142 100 975 104 874 205 498 115 86 43 7 330
Rutherford 5 1,671 383 553 375 2,519 463 2,238 744 909 250 224 126 29 1,444
Rutherford 6 364 127 135 97 604 119 560 163 236 63 42 31 10 341
Total 2,738 644 830 572 4,098 686 3,672 1,112 1,643 428 352 200 46 2,115
b\. ASO
Ischemic heart disease Heart failure Cerebrovascular disease Renal dysfunction
(−) (+) (−) (+) (−) (+) (−) (+)
Medical treatment PCI CABG G3a G3b G4 G5 G5D
Rutherford 4 688 134 142 100 960 104 860 204 486 114 86 43 7 328
Rutherford 5 1,611 373 550 373 2,451 456 2,168 739 845 246 221 123 29 1,443
Rutherford 6 356 126 133 97 593 119 549 163 227 63 42 31 10 339
Total 2,655 633 825 570 4,004 679 3,577 1,106 1,558 423 349 197 46 2,110
PCI: percutaneous coronary intervention, CABG: coronary arterial bypass grafting Heart failure (+): history of admision due to heart failure, clinical symptoms due to heart failure confirmed by ultrasound examination, apparently decreased cardiac function by ultrasound examination without clinical symptoms. Renal dysfunction: (−) (60≦), G3a (45--59), G3b (30--44), G4 (15--29), G5 (\<15), G5D (\<15 with hemodialyais). New CKD risk stratification by eGFR (mL/min/1.73 m^2^) in "Clinical Practice Guidebook for Diagnosis and Treatment of Chronic Kidney Disease 2012" eGFR: estimated glomerular filtration rate, CKD: chronic kidney disease
###### Table 1-4 Patients' background 4
a\. Total
------------------- -------------------- ----------------------------- ----------- ------ --------- ------------------------ ------- -------- -------- --------- ---------- -------- --- ---- ----
Rutherford 4 962 82 33 2 5 2 20 19 14 30 4 7 1 6 27
Rutherford 5 2,727 180 67 8 10 7 35 49 15 65 13 7 0 13 51
Rutherford 6 676 34 10 3 2 2 10 7 1 11 2 5 0 3 9
Total 4,365 296 110 13 17 11 65 75 30 106 19 19 1 22 87
b\. ASO
Malignant neoplasm Sites of malignant neoplasm
(−) (+) Head and neck Esophagus Lung Stomach Hepatobiliary pancreas Colon Breast Uterus Ovarian Prostate Others
History of cancer Under treatment\* Unknown
Rutherford 4 951 80 31 2 4 2 20 19 14 30 4 5 1 6 26
Rutherford 5 2,653 179 67 8 10 7 35 49 15 64 13 7 0 13 51
Rutherford 6 666 34 9 3 2 2 10 7 1 11 2 4 0 3 9
Total 4,270 293 107 13 16 11 65 75 30 105 19 17 1 22 86
\*Including palliative therapy or recurrence
###### Table 1-5 Patients' background 5
a\. Total
-------------- -------------------------------------- -------------------------------------- ---------------- -------- ----- -------- ----- -------- --------------------- ---------------------------- ------------------ -------- ------ ------- ---- ----- ---- ----- -----
Rutherford 4 304 241 177 168 35 1 153 781 0.8 78 0.45 302 39 970 1 44 7 35 22
Rutherford 5 654 797 238 90 578 30 595 2,069 0.77 261 0.42 1,274 35 2,626 20 79 15 169 73
Rutherford 6 152 196 51 15 52 81 176 371 0.76 24 0.36 258 35.5 657 2 12 3 35 14
Total 1,110 1,234 466 273 665 112 924 3,221 0.78 363 0.43 1,834 35 4,253 23 135 25 239 109
b\. ASO
Contralateral limb occlusive lesions Vascular lesions excluding occlusion
(−) (+)
Asymptomatic Intermittent claudication CLI Post-treatment ABI TBI SPP (−) TAA AAA (including IAA) Peripheral artery aneurysm Carotid stenosis Others
R4 R5 R6 n Median n Median n Median
Rutherford 4 297 238 176 164 35 1 153 771 0.8 77 0.45 300 39 957 1 43 7 35 21
Rutherford 5 629 789 233 90 554 29 583 2,017 0.76 251 0.41 1,249 35 2,566 19 77 10 168 67
Rutherford 6 149 195 51 14 51 79 173 367 0.76 24 0.36 256 35.5 647 2 12 3 34 14
Total 1,075 1,222 460 268 640 109 909 3,155 0.77 352 0.42 1,805 35 4,170 22 132 20 237 102
ABI: ankle brachial (pressure) index, TBI: toe brachial (pressure) index, SPP: skin perfusion pressure, CLI: critical limb ischemia, TAA: thoracic aortic aneurysm, AAA: abdominal aortic aneurysm, IAA: iliac artery aneurysm
###### Table 1-6 Patients' background 6
a\. Total (=ASO)
----------------------- ----------------------------- ---------------------------- -------- ------ -------- ------- -------- ------
Rutherford 4 31 156 31 62.6 31 113.7 31 0.35
Rutherford 5 69 153.6 69 50.2 69 111.2 69 0.34
Rutherford 6 18 129.2 18 50.1 18 100.6 18 0.31
Total 118 153.8 118 51.6 118 107.2 118 0.33
b\. ASO
Fatty acid
Arachidonic acid (AA) Eicosapentaenoic acid (EPA) Docosahexaenoic acid (DHA) EPA/AA
n Median n Median n Median n Median
Rutherford 4 31 156 31 62.6 31 113.7 31 0.35
Rutherford 5 69 153.6 69 50.2 69 111.2 69 0.34
Rutherford 6 17 125 17 54.5 17 104.3 17 0.33
Total 117 153.6 117 51.6 117 107.2 117 0.34
The causes of the arterial occlusion of the limb were ASO in 4,683 (98%) limbs, thromboangitis obliterans (TAO) in 37, vasculitis (Takayasu's arteritis, collagen disease, Behçet's disease, and fibromuscular dysplasia excluding TAO) in 43, and others in 21. Patients comorbidities consisted of diabetes in 66% (67%) of the limbs, hypertension in 75% (76%), dyslipidemia in 39% (39%), ischemic heart disease in 43% (43%), heart failure 14% (14%), cerebrovascular disease in 23% (24%), dialysis for renal failure in 44% (45%), past medical history of malignant neoplasm or that being treated in 8% (9%), arterial occlusive lesions in the contralateral limb in 77% (77%), and CLI in the contralateral limb (Rutherford 4--6) in 22% (22%).
(2) Conditions of limb ischemia
-------------------------------
Limb ischemia pretreatment conditions are shown in [**Tables 2-1**](#table2-1){ref-type="table"} to [](#table2-2){ref-type="table"}[](#table2-3){ref-type="table"}[](#table2-4){ref-type="table"}[](#table2-5){ref-type="table"}[**2-6**](#table2-6){ref-type="table"}. Regarding the walking function (Taylor's classification),^[@R12])^ patients who could walk outdoors or indoors independently, including with a cane, were regarded as "ambulatory," and those unable to walk but able to stand on their own legs during transfer from the bed to a wheelchair were designated as "ambulatory/homebound."
###### Table 2 Pretreatment condition Table 2-1 Pretreatment condition 1
a\. Total
-------------- ----------------------------------------------- --------------------------------------------------------- ----------------------------------------------------------- ----------------------------- ----- ------- ------- ------- ------------------------- --------------------------- ------ ------- ----------- -----
Rutherford 4 782 168 129
Rutherford 5 1,679 708 595 1,798 568 616 2,104 882 2,301 392 53 163 30 43
Rutherford 6 237 216 270 126 174 423 237 487 152 190 151 139 29 62
Total 2,698 1,092 994 1,924 742 1,039 2,341 1,369 2,453 582 204 302 59 105
b\. ASO
Ambulatory function (Taylor's classification) Tissue loss (University of Texas classification: grade) Tissue loss\* (University of Texas classification: stage) Sites of ulcer/gangrene^\#^
Ambulatory Ambulatory/homebound Nonambulatory I II III C D Toe Foot: distal metatarsal Foot: proximal metatarsal Heel Ankle Lower leg
Rutherford 4 769 166 129
Rutherford 5 1,618 697 592 1,747 551 609 2,051 859 2,249 373 52 161 30 42
Rutherford 6 233 211 268 125 170 417 233 480 149 186 150 136 29 62
Total 2,620 1,074 989 1,872 721 1,026 2,284 1,339 2,398 559 202 297 59 104
University of Texas classification: grade (I: superficial, not involving tendon, capsule, or bone, II: penetrating to tendon/capsule, III: penetrating to bone or joint) University of Texas classification: stage (C: ischemia without infection, D: ischemia with infection) \*Data in "infection" in [**Table 3-2**](#table3-2){ref-type="table"} were used in 2015 and 2016. ^\#^Data in "main sites of ulcer/gangrene to be treated" were used in 2015 and 2016.
###### Table 2-2 Pretreatment condition 2
a\. Total
-------------- ------------------- ------------ -------------- ------- -------- ------ -------- ------ -------- ------ -------- ------ -------- ------ ------- ----
Rutherford 4 1,056 23 1,040 6,600 976 0.42 964 3.7 1,030 1.09 586 0.54 43 0.36 385 20
Rutherford 5 2,855 127 2,919 7,100 2,814 1.03 2,755 3.4 2,922 2.39 1,914 0.6 159 0.31 1,801 22
Rutherford 6 618 105 707 8,800 692 4.51 672 3 704 2.18 337 0.61 12 0.18 348 21
Total 4,529 255 4,666 7,150 4,482 1.1 4,391 3.4 4,656 1.6 2,837 0.59 214 0.31 2,534 21
b\. ASO
Temperature \>38° Blood test Hemodynamics
(−) (+) WBC CRP Alb Cr ABI TBI SPP
n Median n Median n Median n Median n Median n Median n Median
Rutherford 4 1,042 22 1,026 6,600 962 0.42 952 3.65 1,016 1.1 579 0.54 43 0.36 382 20
Rutherford 5 2,784 123 2,846 7,100 2,745 1.05 2,686 3.4 2,849 2.75 1,864 0.6 152 0.3 1,760 22
Rutherford 6 608 104 697 8,800 682 4.58 662 2.95 694 2.3 332 0.61 12 0.18 343 21
Total 4,434 249 4,569 7,130 4,389 1.1 4,300 3.4 4,559 1.7 2,775 0.59 207 0.3 2,485 21
WBC: white blood cell, CRP: C reactive protein, Alb: albumin, Cr: creatinine, ABI: ankle brachial (pressure) index, TBI: toe brachial (pressure) index, SPP: skin perfusion pressure
###### Table 2-3 Pretreatment condition 3
a\. Total
-------------- -------------------- -------------------- ----------------------------------- ---------------------------------------- ---------------- ------- ----- ----- ----- ----------- ---- ----- ----- ----- ----------- -----
Rutherford 4 663 609 94 363 715 446 62 57 41 162 10 65 117 132 481 52
Rutherford 5 2,186 1,360 102 574 1,807 1,973 191 109 61 186 13 331 387 366 1,201 426
Rutherford 6 551 279 36 146 404 507 39 38 12 43 4 54 61 75 328 110
Total 3,400 2,248 232 1,083 2,926 2,926 292 204 114 391 27 450 565 573 2,010 588
b\. ASO
Diagnostic imaging Sites of occlusion TASC II classification aortoiliac TASC II classification femoropopliteal
IADSA CTA Others Aortoiliac Femoropop Lower leg/foot A B C D No lesion A B C D No lesion
Rutherford 4 655 600 94 358 708 437 61 56 39 161 10 65 116 132 472 52
Rutherford 5 2,124 1,331 100 569 1,781 1,912 191 108 61 182 13 328 385 358 1,170 400
Rutherford 6 542 274 36 144 399 500 39 38 12 42 3 54 61 74 324 108
Total 3,321 2,205 230 1,071 2,888 2,849 291 202 112 385 26 447 562 564 1,966 560
IADSA: intra-arterial digital subtraction angiography, CTA: computed tomography angiography
###### Table 2-4 Pretreatment condition 4
a\. Total
---------------- ----------------- ------------------------------- ----------------------------- --------------------- ------------------- --------------------- -------- ----- -------- ----- -------- ----- -------- -----
Rutherford 4 657 2.0 654 2.0 648 6.0 647 6.0 645 3.0 644 3.0 630 3.0
Rutherford 5 2,088 1.0 2,085 1.0 2,090 4.0 2,086 4.0 2,093 3.0 2,096 2.0 2,059 3.0
Rutherford 6 476 1.0 473 1.0 479 3.0 475 4.0 476 3.0 477 2.0 476 3.0
Total 3,221 1.0 3,212 1.0 3,217 4.0 3,208 5.0 3,214 3.0 3,217 2.0 3,165 3.0
b\. ASO
Bollinger score
Common femoral Deep femoral Superficial femoral: proximal Superficial femoral: distal Popliteal: proximal Popliteal: distal Tibioperoneal trunk
n Median n Median n Median n Median n Median n Median n Median
Rutherford 4 650 2.0 647 2.0 641 6.0 640 6.0 638 3.0 637 3.0 623 3.0
Rutherford 5 2,047 1.0 2,040 1.0 2,045 4.0 2,040 4.0 2,047 3.0 2,049 2.0 2,013 3.0
Rutherford 6 472 1.0 469 1.0 475 3.0 471 4.0 472 3.0 473 2.0 472 3.0
Total 3,169 1.0 3,156 1.0 3,161 4.0 3,151 5.0 3,157 3.0 3,159 2.0 3,108 3.0
###### Table 2-5 Pretreatment condition 5
a\. Total
---------------------------- -------------------------- --------------------------- ------------------------- -------------------- ------------------ ------ -------- ------ -------- ----- -------- ----- -------- ------
Rutherford 4 624 6.0 614 6.0 621 7.0 602 13.0 620 5.0 601 5.0 512 4.0
Rutherford 5 2,039 13.0 2,003 13.0 2,037 13.0 2,006 13.0 2,032 6.0 2,000 6.0 1,735 13.0
Rutherford 6 473 13.0 466 13.0 474 13.0 464 13.0 475 6.0 467 6.0 426 13.0
Total 3,136 13.0 3,083 13.0 3,132 13.0 3,072 13.0 3,127 6.0 3,068 6.0 2,663 6.0
b\. ASO
Bollinger score
Posterior tibial: proximal Posterior tibial: distal Anterior tibial: proximal Anterior tibial: distal Peroneal: proximal Peroneal: distal Foot
n Median n Median n Median n Median n Median n Median n Median
Rutherford 4 617 6.0 608 6.0 614 6.0 595 13.0 613 5.0 595 5.0 507 4.0
Rutherford 5 1,991 13.0 1,955 13.0 1,989 13.0 1,958 13.0 1,984 6.0 1,953 6.0 1,694 12.0
Rutherford 6 469 13.0 462 13.0 470 13.0 460 13.0 471 6.0 463 6.0 412 13.0
Total 3,077 13.0 3,025 13.0 3,073 13.0 3,013 13.0 3,068 6.0 3,011 6.0 2,613 6.0
Regarding the state of local tissue defect (University Texas classification),^[@R13])^ the most severe lesion, the main treatment target, was evaluated. Skin perfusion pressure (SPP) was measured on the foot (base of the toe, dorsum of the foot, or sole) and a lower value was adopted. To perform WIfI classification, the sites of ulcer and gangrene were registered separately. Although SPP is widely used as an objective index for evaluating ischemia in Japan, ischemic grading criteria using SPP is not shown in WIfI classification, wherein TP is given top priority. Therefore, in JCLIMB, the SPP value was converted to TP using the conversion equation SPP=0.6853 TP+14.48 from the correlation data of SPP and TP reported in Japan,^[@R14])^ and applied for WIfI ischemic grading.
The lesion was considered infected when it showed two or more of the following findings: local swelling or induration, erythema \>0.5 cm around the ulcer, local tenderness or pain, local warmth, and purulent discharge (thick, opaque to white, or sanguineous secretion). In addition, local infections involving only the skin and the subcutaneous tissue, and those involving structures deeper than the skin and subcutaneous tissues, were registered separately. Local infections involving only the skin and the subcutaneous tissue were differentiated based on the size of the erythema around the ulcer, ≦2 or \>2 cm.
Systemic inflammatory response syndrome, indicating systemic infection, was manifested by two or more of the following signs: temperature \>38°C or \<36°C, heart rate \>90 beats/min, respiratory rate \>20 breaths/min or PaCO~2~ \<32 mmHg, white blood cell count \>12,000 or \<4,000 cu/mm or 10% immature (band) forms. The arteries in the ankle joint region were classified as foot arteries.
On Taylor's classification, 56% (56%) of the patients were ambulatory, 23% (23%) were ambulatory/homebound, and 21% (21%) were non-ambulatory. On the Rutherford classification (R),^[@R15])^ limbs with categories R4, R5, and R6 accounted for 23% (23%), 62% (62%), and 15% (15%) of the limbs, respectively. The occlusive legion was located in the aortoiliac artery in 23% (23%) of the limbs, in the femoropopliteal artery in 61% (62%) of the limbs, and in the crural or foot artery in 61% (61%) of the limbs.
We were able to apply the WIfI classification with sufficient data to 1,724 limbs (1,689 limbs). On the WIfI classification, limbs with the stages 1, 2, 3, and 4 accounted for 11% (11%), 20% (19%), 27% (27%), and 43% (43%) of the limbs, respectively.
###### Table 2-6 SVS WIfI classification\*
a\. Total
-------------- ------- ---------- ---------------- ------- ----- ----- ----- ------- ------- ----- ----- ---- ----- ----- ----- -----
Rutherford 4 476 0 0 0 35 55 55 177 435 20 18 3 86 218 16 2
Rutherford 5 0 542 680 210 124 151 150 751 943 312 147 30 92 117 420 547
Rutherford 6 0 21 77 224 23 27 18 158 103 60 131 28 5 4 30 187
Total 476 563 757 434 182 233 223 1,086 1,481 392 296 61 183 339 466 736
b\. ASO
Wound Ischemia Foot infection Stage
0 1 2 3 0 1 2 3 0 1 2 3 1 2 3 4
Rutherford 4 469 0 0 0 35 55 53 173 429 19 18 3 86 213 15 2
Rutherford 5 0 530 661 207 116 150 147 738 921 306 145 26 88 112 416 535
Rutherford 6 0 21 75 221 23 27 17 155 102 60 127 28 5 4 30 183
Total 469 551 736 428 174 232 217 1,066 1,452 385 290 57 179 329 461 720
\*Data registered only in 2015 and 2016
(3) Treatment
-------------
[**Tables 3-1**](#table3-1){ref-type="table"} to [](#table3-2){ref-type="table"}[](#table3-3){ref-type="table"}[](#table3-4){ref-type="table"}[](#table3-5){ref-type="table"}[**3-6**](#table3-6){ref-type="table"} show the CLI treatment data. Revascularizations of the affected limbs were performed in 95% (95%) of the registered limbs, and primary major amputations were performed in 1.9% (2.0%) of the registered limbs. Among the surgical reconstruction procedures, distal bypass, a bypass to the crural or foot artery, accounted for 46% (45%). Endovascular treatment (EVT) applied to the crural or foot artery accounted for 38% (37%) of the total EVT.
###### Table 3 Treatment Table 3-1 Treatment 1
a\. Total
------------------------- -------------------- ------------------------- ------------------ ---------------------- ------------- ------------------ -------- --------- ----- ------- ----- ----- -----
Rutherford 4 307 5 1,011 9 0 0 1 1 16 788 169 53 53
Rutherford 5 821 12 2,868 32 0 0 1 3 42 2,161 476 138 165
Rutherford 6 173 2 671 52 2 0 0 0 11 513 111 35 53
Total 1,301 19 4,550 93 2 0 2 4 69 3,462 756 226 271
b\. ASO
Treatment Angiogenic therapy Reoperation
Pharmacological therapy Angiogenic therapy Arterial reconstruction Major amputation Lumber sympathectomy Bone marrow Peripheral blood Others Unknown (−) (+)
1X 2X 3X≦
Rutherford 4 305 5 997 9 0 0 1 1 16 775 169 53 51
Rutherford 5 799 12 2,796 32 0 0 1 3 42 2,107 465 135 158
Rutherford 6 171 2 662 52 2 0 0 0 11 507 108 35 51
Total 1,275 19 4,455 93 2 0 2 4 69 3,389 742 223 260
###### Table 3-2 Treatment 2
a\. Total
-------------- ------------------------------- ---------------- ---------------------------- -------------------------- --------------------- ----------------------- ------------------- ------------------ ----------------- ------------------------- ------------- ------------------- -------- ---- -------
Rutherford 4 4 2 30 151 64 133 67 13 31 67 14 7 94 18 489
Rutherford 5 0 2 42 246 161 384 486 22 48 63 11 12 210 25 1,529
Rutherford 6 0 0 6 57 40 106 111 4 13 17 6 2 35 7 330
Total 4 4 78 454 265 623 664 39 92 147 31 21 339 50 2,348
b\. ASO
Bypass TEA EVT
Aorta--aorta Aorta (with suprarenal clamp) Aorta--femoral Femoral-proximal popliteal Femoral-distal popliteal Femoral-crural/foot Popliteal-crural/foot Anatomical others Axillary-femoral Femoral-femoral Extra-anatomical others Aorta/iliac Fomoral/popliteal Others
Rutherford 4 4 2 30 151 63 129 64 12 31 65 13 6 94 17 487
Rutherford 5 0 2 42 243 157 373 462 19 47 62 11 12 208 24 1,501
Rutherford 6 0 0 6 55 40 104 108 4 12 17 5 2 35 7 328
Total 4 4 78 449 260 606 634 35 90 145 29 20 337 48 2,316
TEA: thromboendarterectomy, EVT: endovascular treatment/therapy
###### Table 3-3 Treatment 3
a\. Total
-------------- ------------------- --------------------- -------------- ---------------- ------- ------ -------- ----- --------- -------------- ---------- --------- ------ ------ ----
Rutherford 4 179 226 152 25 65 191 301 25 43 29 53 51 14 123 16
Rutherford 5 374 709 769 44 106 282 1,102 4 86 102 149 210 43 426 70
Rutherford 6 81 145 169 20 23 58 266 2 38 18 50 42 6 102 10
Total 634 1,080 1,090 89 194 531 1,669 31 167 149 252 303 63 651 96
b\. ASO
EVT Vascular prosthesis Vein usage\* Vein quality\*
Aorta/iliac Fomoral/popliteal Tibioperoneal/foot Others Polyester ePTFE Vein Others (−) In-situ Non-reversed Reversed Spliced Good Poor
Rutherford 4 179 226 150 25 63 189 294 25 42 27 53 48 14 118 16
Rutherford 5 373 706 747 42 105 277 1,066 4 85 101 141 201 43 411 68
Rutherford 6 81 144 167 20 23 57 260 2 37 18 47 41 5 98 10
Total 633 1,076 1,064 87 191 523 1,620 31 164 146 241 290 62 627 94
ePTFE: expanded polytetrafluoroethylene \*Data registered only in 2015 and 2016
###### Table 3-4 Treatment 4
a\. Total
---------------------- -------------------- -------------------------------------------- ------------------------------------------ -------------------- ------------------ -------- -------- -------- ------ --------------------- ------------------ ----------------- ---------- ------------------ ----------------- ---------- ---------------- --------- ----
Rutherford 4 1 73 10 47 26 34 5 4 129 71 12 76 22 20 24 7 2 33 6
Rutherford 5 7 202 27 163 111 305 34 19 324 544 17 173 99 36 143 54 4 268 76
Rutherford 6 2 50 4 42 35 70 14 1 89 128 7 40 33 11 36 18 2 60 13
Total 10 325 41 252 172 409 53 24 542 743 36 289 154 67 203 79 8 361 95
b\. ASO
Distal bypass
Proximal anastomosis Distal anastomosis Distal anastomosis: sites of crural artery Distal anastomosis: sites of foot artery
External iliac Common femoral Deep femoral Superficial femoral Proximal popliteal Distal popliteal Crural Others Crural Foot Tibioperoneal trunk Posterior tibial Anterior tibial Peroneal Posterior tibial Anterior tibial Peroneal Dorsalis pedis Plantar
Rutherford 4 1 69 10 47 25 32 5 4 124 69 12 73 20 20 22 7 2 33 6
Rutherford 5 6 199 26 154 107 289 33 19 310 523 17 171 94 29 140 53 3 263 65
Rutherford 6 2 49 4 41 34 68 14 1 86 126 7 39 31 11 36 18 2 59 12
Total 9 317 40 242 166 389 52 24 520 718 36 283 145 60 198 78 7 355 83
###### Table 3-5 Treatment 5
a\. Total
-------------- ------------------------- --------------- --------- -------- -------- -----
Rutherford 4 501 56 44 34 69 31
Rutherford 5 1,413 175 181 115 162 56
Rutherford 6 290 30 43 27 28 15
Total 2,204 261 268 176 259 102
b\. ASO
Pharmacological therapy
Antiplatelet ATA Prostaglandin Heparin Statin Others
Rutherford 4 499 56 44 34 69 31
Rutherford 5 1,376 168 172 108 158 51
Rutherford 6 287 30 42 26 28 15
Total 2,162 254 258 168 255 97
Antiplatelet: aspirin, cilostazol, beraprost, sarpogrelate, ticlopidine, clopidogrel, ethyl icosapentate ATA: antithrombotic agent
###### Table 3-6 Treatment 6
a\. Total
----------- ----------------------------------- --------------------------------- ---------------------------- ------------------------------
Polyester 56 13 9 9
ePTFE 251 58 30 39
Vein 160 210 588 623
Artery 5 1 18 26
Others 14 0 2 1
(−) 5 4 5 6
Total 491 286 652 704
b\. ASO
Femoral-proximal popliteal bypass Femoral-distal popliteal bypass Femoral-crural/foot bypass Popliteal-crural/foot bypass
Polyester 55 13 9 9
ePTFE 251 57 29 39
Vein 156 206 573 597
Artery 5 1 16 22
Others 14 0 2 1
(−) 5 1 5 6
Total 486 281 614 674
ePTFE: expanded polytetrafluoroethylene
(4) Outcomes early (one month) after treatment
----------------------------------------------
[**Tables 4-1**](#table4-1){ref-type="table"} to [](#table4-2){ref-type="table"}[](#table4-3){ref-type="table"}[](#table4-4){ref-type="table"}[](#table4-5){ref-type="table"}[](#table4-6){ref-type="table"}[](#table4-7){ref-type="table"}[**4-8**](#table4-8){ref-type="table"} show the outcomes early (one month) after treatment. Follow-up data one month after treatment were obtained in 3,188 limbs (67%), including 3,115 limbs (67%) with ASO, which included 105 limbs (102 limbs) without arterial reconstruction. Data were collected according to the severity of the local limb conditions (Rutherford classification) and treatment measures (EVT alone or surgical reconstruction with/without EVT). The mortality was 3.2% (3.2%) in the whole series, and 3.0% (3.1%) and 3.3% (3.3%) treated by EVT alone and by surgical reconstruction with/without EVT, respectively. The most common cause of death was cardiac disease, accounting for 31% (31%) of all deaths. Postoperative complications were cardiac disease in 2.9% (2.9%), cerebrovascular disease in 1.4% (1.4%), pneumonia in 1.9% (1.9%), and wound complication in 5.3% (5.0%). Complications at the puncture site were noted in 0.4% (0.4%) of the limbs treated by EVT.
###### Table 4 One month after the treatment (data collection at July 2015) therapeutic measures: EVT (only EVT without surgical reconstruction), surgical reconstruciton (surgical reconstruction with or without EVT) Table 4-1 Life prognosis/causes of death
a\. Total
------------------------- -------------------- ----------------- ----------------- ------------------------- -------------------- ---------------------------- ----------- -------------------- --------------------------- -------- --------- --- --- ---- ---- ---
Local condition Rutherford 4 637 20 6 3 0 1 0 2 0 1 1 1 0 8 3
Rutherford 5 1,976 64 23 21 0 3 0 4 0 9 2 5 0 13 7
Rutherford 6 440 17 5 7 1 0 0 0 0 1 4 1 0 1 2
Therapeutic measures Non-reconstruction 102 3 0 1 1 0 0 0 0 1 0 0 0 0 0
EVT 1,343 43 24 16 0 1 0 5 0 8 0 2 0 8 3
Surgical reconstruction 1,608 55 10 14 0 3 0 1 0 2 7 5 0 14 9
Total 3,053 101 34 31 1 4 0 6 0 11 7 7 0 22 12
b\. ASO
Life prognosis Causes of death
Alive Dead Unknown Cardiac disease Cerebrovascular disease Malignant neoplasm Aortic aneurysm/dissection Infection Ischemic enteritis Gastrointestinal bleeding Others Unknown
Hemorrhage Infarction Unknown Diseased limb Others
Local condition Rutherford 4 627 19 6 3 0 1 0 1 0 1 1 1 0 8 3
Rutherford 5 1,921 64 23 21 0 3 0 4 0 9 2 5 0 13 7
Rutherford 6 434 17 5 7 1 0 0 0 0 1 4 1 0 1 2
Therapeutic measures Non-reconstruction 99 3 0 1 1 0 0 0 0 1 0 0 0 0 0
EVT 1,322 43 23 16 0 1 0 5 0 8 0 2 0 8 3
Surgical reconstruction 1,561 54 10 14 0 3 0 0 0 2 7 5 0 14 9
Total 2,982 100 33 31 1 4 0 5 0 11 7 7 0 22 12
###### Table 4-2 Perioperative complications 1
a\. Total
------------------------- --------------------- ---------------------------- ----------------------- -------------------- --------------------- --------------------- ----- ----- ------- ----- ------- ----- ------- ----- ---- ---
Local condition Rutherford 4 595 12 4 7 616 0 0 2 606 12 595 23 614 3 1
Rutherford 5 1,910 29 7 15 1,923 6 15 17 1,931 30 1,841 120 1,940 13 8
Rutherford 6 427 5 3 7 440 0 0 2 426 16 424 18 438 1 3
Therapeutic measures Non-reconstruction 12 1 0 1 14 0 0 0 14 0 14 0 14 0 0
EVT 1,336 30 4 12 1,364 3 8 7 1,367 15 1,361 21 1,369 6 7
Surgical reconstruction 1,584 15 10 16 1,601 3 7 14 1,582 43 1,485 140 1,609 11 5
Total 2,932 46 14 29 2,979 6 15 21 2,963 58 2,860 161 2,992 17 12
b\. ASO
Cardiac disease Cerebrovascular disease Pneumonia Wound complication Peripheral embolism
(−) Angina Serious arrhysmia Myocardial infarction (−) TIA Cerebral infarction (−) (+) (−) (+) (−) (+)
Functional loss (−) Functional loss (+) Minor (including blue toe) Major
Local condition Rutherford 4 584 12 4 7 605 0 0 2 595 12 587 20 603 3 1
Rutherford 5 1,859 28 7 15 1,871 6 15 17 1,879 30 1,796 113 1,889 12 8
Rutherford 6 421 5 3 7 434 0 0 2 421 15 420 16 433 1 2
Therapeutic measures Non-reconstruction 12 1 0 1 14 0 0 0 14 0 14 0 14 0 0
EVT 1,316 29 4 12 1,343 3 8 7 1,346 15 1,341 20 1,348 6 7
Surgical reconstruction 1,536 15 10 16 1,553 3 7 14 1,535 42 1,448 129 1,563 10 4
Total 2,864 45 14 29 2,910 6 15 21 2,895 57 2,803 149 2,925 16 11
TIA: transient ischemic attack
###### Table 4-3 Perioperative complications 2
a\. Total
------------------------- -------------------- ------------------- --------------------- ------------------------------------- ------------------------------- ------- --------- ------ ----- ------- ----- ------- ----- ---
Local condition Rutherford 4 608 10 0 0 3 14 9 0 1 617 1 315 5
Rutherford 5 1,926 33 2 0 8 17 28 3 2 1,953 8 1,089 12
Rutherford 6 436 6 0 0 3 5 3 3 0 440 2 221 1
Therapeutic measures Non-reconstruction 14 0 0 0 0 0 0 0 0 14 0 32 1
EVT 1,368 14 0 0 4 10 11 2 0 1,376 6 1,371 17
Surgical reconstruction 1,588 35 2 0 10 26 29 4 3 1,620 5 222 0
Total 2,970 49 2 0 14 36 40 6 3 3,010 11 1,625 18
b\. ASO
Hemorrhage Sites of bleeding Outcome of bleeding Complication due to contrast medium Complication at puncture site
(−) (+) Unknown Brain GI tract Others Cured Uncured Dead (−) (+) (−) (+)
Local condition Rutherford 4 597 10 0 0 3 14 9 0 1 606 1 314 5
Rutherford 5 1,874 33 2 0 8 17 28 3 2 1,901 8 1,069 12
Rutherford 6 430 6 0 0 3 5 3 3 0 434 2 220 1
Therapeutic measures Non-reconstruction 14 0 0 0 0 0 0 0 0 14 0 32 1
EVT 1,347 14 0 0 4 10 11 2 0 1,355 6 1,350 17
Surgical reconstruction 1,540 35 2 0 10 26 29 4 3 1,572 5 221 0
Total 2,901 49 2 0 14 36 40 6 3 2,941 11 1,603 18
GI: gastrointestinal
###### Table 4-4 Hemodynamics
a\. Total
------------------------- ------------------------------- ------------------------------- -------- ---------------- -------- ----- -------- ------ -------- ----- -------- ------ ----
Local condition Rutherford 4 376 0.89 347 114 158 33.5 292 0.9 267 115 63 39
Rutherford 5 937 0.86 871 114 779 39 663 0.88 592 120 359 42
Rutherford 6 145 0.91 134 124 130 36 106 1 99 128 68 45.5
Therapeutic measures Non-reconstruction 46 0.79 34 103.5 31 32 35 0.83 27 110 18 30
EVT 674 0.85 633 112 522 35 476 0.87 436 116 299 42
Surgical reconstruction 738 0.89 685 118 514 40 550 0.93 494 122 180 43
Total 1,458 0.87 1,352 115 1,067 37 1,061 0.9 958 120 490 42
b\. ASO
Immediate after the treatment One month after the treatment
ABI Ankle pressure SPP ABI Ankle pressure SPP
n Median n Median n Median n Median n Median n Median
Local condition Rutherford 4 368 0.89 341 114 156 33 286 0.9 262 115.5 63 39
Rutherford 5 913 0.85 848 114 758 39 653 0.88 584 120 353 43
Rutherford 6 143 0.91 132 124.5 127 36 105 0.99 98 128 67 45
Therapeutic measures Non-reconstruction 45 0.77 33 101 30 29.5 34 0.87 27 110 17 32
EVT 668 0.85 628 112 521 35 472 0.87 433 116 298 42.5
Surgical reconstruction 711 0.89 660 118 490 40 538 0.93 483 123 174 43.5
Total 1,424 0.87 1,321 115 1,041 38 1,044 0.9 944 120 483 43
ABI: ankle brachial (pressure) index, SPP: skin perfusion pressure
###### Table 4-5 One month after the treatment
a\. Total
------------------------- ---------------------------- ------------------------------- ---------------- ------------------------------------------------------------ ----------- -------- ---------- ----------- -------------- ------- --------- --------- ------------ ---------------------- --------------- ----- ----- ----
Local condition Rutherford 4 560 12 23 0 0 4 2 563 51 18 416 142 38 27 461 108 88
Rutherford 5 1,747 37 90 0 5 14 19 1,612 234 94 426 1,210 278 26 1,086 459 499
Rutherford 6 360 15 29 0 1 2 12 321 53 28 47 277 73 4 126 137 195
Therapeutic measures Non-reconstruction 0 0 0 0 0 0 0 49 11 4 20 31 8 4 62 16 24
EVT 1,201 41 67 0 0 8 28 1,027 217 77 338 736 219 28 639 334 419
Surgical reconstruction 1,466 23 75 0 6 12 5 1,420 110 59 531 862 162 25 972 354 339
Total 2,667 64 142 0 6 20 33 2,496 338 140 889 1,629 389 57 1,673 704 782
b\. ASO
Bypass graft/EVT condition Clinical symptoms of the limb Ischemic wound Ambulatory function at discharge (Taylor's classification)
Good Stenosis Occlusion Deterioration Anastomosis disruption (aneurysm) Infection Others Improved No change Deteriorated Cured Uncured Unknown Ambulatory Amburatory/homebound Nonambulatory
Improved Deteriorated
Local Condition Rutherford 4 551 12 23 0 0 3 2 553 51 18 408 140 38 27 451 107 88
Rutherford 5 1,705 37 81 0 5 14 18 1,573 229 86 417 1,181 265 25 1,045 449 495
Rutherford 6 357 14 27 0 1 2 12 318 50 28 46 272 73 4 124 134 194
Therapeutic Measures Non-reconstruction 0 0 0 0 0 0 0 46 11 6 20 29 8 3 60 16 27
EVT 1,186 40 63 0 0 8 27 1,019 212 69 336 727 210 28 624 330 414
Surgical reconstruction 1,427 23 68 0 6 11 5 1,379 107 57 515 837 158 25 936 344 336
Total 2,613 63 131 0 6 19 32 2,444 330 132 871 1,593 376 56 1,620 690 777
###### Table 4-6 Revision one month after the treatment
a\. Total
--------------------------- -------------------------------------------------------------- ---------------------------------------------- ----------------------------------------------- ------------------ -------- ----- ------------------------------ -------------- ----- ----------- ------------- --------------- -------- ----- ------- ----- ---- ---
Local condition Rutherford 4 27 19 593 2 10 3 582 10 0 2 10 4 0 0 623 9 3
Rutherford 5 97 75 1,848 2 60 9 1,818 18 0 10 35 18 2 18 1,889 72 6
Rutherford 6 41 22 395 0 16 2 375 9 0 2 15 2 0 10 373 49 2
Therapeutic measures Non-reconstruction 0 0 11 0 1 1 12 0 0 0 1 0 0 0 81 6 0
EVT 85 67 1,273 1 58 3 1,245 8 0 8 40 13 2 19 1,263 73 2
Surgical reconstruction 80 49 1,552 3 27 10 1,518 29 0 6 19 11 0 9 1,541 51 9
Total 165 116 2,836 4 86 14 2,775 37 0 14 60 24 2 28 2,885 130 11
b\. ASO
Revision for those excluding good bypass graft/EVT condition Minor reintervention (revision for stenosis) Major reintervention (revision for occlusion) Major amputation
(+) (−) (−) Patch plasty EVT Others (−) Thrombectomy (±patch plasty) Thrombolysis EVT Re-bypass Jump bypass Interposition Others (−) (+)
Due to preoperative wound Due to new wound
Local condition Rutherford 4 26 19 583 2 10 3 573 9 0 2 10 4 0 0 614 9 2
Rutherford 5 91 71 1,796 2 59 9 1,772 17 0 10 32 18 2 15 1,837 70 5
Rutherford 6 38 22 391 0 15 1 371 8 0 2 15 2 0 9 367 49 2
Therapeutic measures Non-reconstruction 0 0 10 0 1 0 11 0 0 0 1 0 0 0 78 6 0
EVT 80 66 1,253 1 57 3 1,229 8 0 8 38 13 2 16 1,243 72 2
Surgical reconstruction 75 46 1,507 3 26 10 1,476 26 0 6 18 11 0 8 1,497 50 7
Total 155 112 2,770 4 84 13 2,716 34 0 14 57 24 2 24 2,818 128 9
###### Table 4-7 Contralateral limb one month after the treatment
a\. Total
------------------------- -------------------------------------- ---------------------------------- ---------------- ------------------------- -------------------- ----- ----------------- ------------------ ------------------ ---------------------- ---------------------------- -------- ---- ----- --- ---- ---- ---
Local condition Rutherford 4 234 192 56 29 9 2 135 41 284 3 73 63 4 9 0 5 1
Rutherford 5 569 668 127 44 162 14 460 191 867 6 206 204 33 90 0 33 11
Rutherford 6 127 156 14 10 31 12 107 58 192 0 53 33 8 23 0 6 4
Therapeutic measures Non-reconstruction 49 26 5 4 2 2 18 3 39 0 11 7 1 2 0 2 0
EVT 399 415 81 40 103 17 333 124 601 7 215 68 17 75 0 21 8
Surgical reconstruction 482 575 111 39 97 9 351 163 703 2 106 225 27 45 0 21 8
Total 930 1,016 197 83 202 28 702 290 1,343 9 332 300 45 122 0 44 16
b\. ASO
Contralateral limb occlusive lesions Treatment for contralateral limb
(−) (+) Unnecessary (+)
Asymptomatic Intermittent claudication CLI Post-treatment Pharmacological therapy Angiogenic therapy EVT Surgical bypass Minor amputation Major amputation Lumber sympathectomy Necessary but no treatment Others
R4 R5 R6
Local condition Rutherford 4 227 192 55 27 9 2 134 41 280 3 73 63 4 9 0 5 0
Rutherford 5 544 654 127 43 157 14 450 186 850 6 205 201 31 86 0 33 11
Rutherford 6 124 155 14 10 31 12 105 56 192 0 52 33 8 23 0 6 4
Therapeutic measures Non-reconstruction 46 26 5 4 2 2 15 2 39 0 11 7 1 2 0 2 0
EVT 387 412 81 40 101 17 331 123 596 7 214 68 17 73 0 21 8
Surgical reconstruction 462 563 110 36 94 9 343 158 687 2 105 222 25 43 0 21 7
Total 895 1,001 196 80 197 28 689 283 1,322 9 330 297 43 118 0 44 15
CLI: critical limb ischemia
###### Table 4-8 Malignant neoplasm one month after the treatment
a\. Total
------------------------- ------------------------------------ --------------------------------------------- --------------- ----------- ------ --------- ------------------------ ------- -------- -------- --------- ---------- -------- --- ---
Local condition Rutherford 4 646 3 8 0 0 1 0 0 0 0 0 0 0 2
Rutherford 5 2,013 9 23 2 1 3 1 0 1 1 0 0 0 1
Rutherford 6 454 1 3 0 0 0 1 0 0 0 0 0 0 0
Therapeutic measures Non-reconstruction 101 0 5 0 0 0 0 0 0 0 0 0 0 0
EVT 1,369 7 14 1 0 0 2 0 1 1 0 0 0 2
Surgical reconstruction 1,643 6 15 1 1 4 0 0 0 0 0 0 0 1
Total 3,113 13 34 2 1 4 2 0 1 1 0 0 0 3
b\. ASO
Newly diagnosed malignant neoplasm Sites of newly diagnosed malignant neoplasm
(−) (+) Unknown Head and neck Esophagus Lung Stomach Hepatobiliary pancreas Colon Breast Uterus Ovarian Prostate Others
Local condition Rutherford 4 635 3 8 0 0 1 0 0 0 0 0 0 0 2
Rutherford 5 1,959 9 22 2 1 3 1 0 1 1 0 0 0 1
Rutherford 6 448 1 3 0 0 0 1 0 0 0 0 0 0 0
Therapeutic measures Non-reconstruction 98 0 5 0 0 1 0 0 0 0 0 0 0 0
EVT 1,348 7 14 1 0 0 2 0 1 1 0 0 0 3
Surgical reconstruction 1,596 6 14 1 1 3 0 0 0 0 0 0 0 0
Total 3,042 13 33 2 1 4 2 0 1 1 0 0 0 3
Stenosis, occlusion, infection, or other trouble occurred after revascularization by EVT in 10.7% (10.4%) and by surgical reconstruction in 7.6% (7.3%). Secondary major amputation rate was 4.7% (4.6%). When ambulatory function at discharge was compared to that before surgery, the rate of patients with ambulatory changed from 56% (56%) to 53% (52%), ambulatory/homebound from 23% (23%) to 22% (22%), and non-ambulatory from 21% (21%) to 25% (25%).
4. Conclusions
==============
Vascular surgeons' contribution in participating facilities registered a sufficient amount of detailed data during busy clinical practice, which has clarified the current status of CLI treatment in Japan from 2013 to 2016. The JCLIMB Committee is planning to continue publishing an annual report and try to clarify the real clinical status of CLI treatment in Japan. Additionally, clinical studies using these data began in 2018. Facilities can participate in JCLIMB at any time and can get detail information about clinical research by contacting the JSVS secretariat for details.
5. JCLIMB Committee, NCD JCLIMB Analytical Team
===============================================
(1) JCLIMB Committee
--------------------
Shinsuke Mii (Chairman), Kunihiro Shigematsu (Vice Chairman), Nobuyoshi Azuma, Atsuhisa Ishida, Yuichi Izumi, Yoshinori Inoue, Hisashi Uchida, Masamitsu Endo, Takao Ohki, Sosei Kuma, Koji Kurosawa, Akio Kodama, Hiroyoshi Komai, Kimihiro Komori, Takashi Shibuya, Shunya Shindo, Ikuo Sugimoto, Juno Deguchi, Naomichi Nishikimi, Katsuyuki Hoshina, Hideaki Maeda, Hirofumi Midorikawa, Tetsuro Miyata, Terutoshi Yamaoka, Hiroya Yamashita, and Yasuhiro Yunoki
(2) NCD JCLIMB Analytical Team
------------------------------
Arata Takahashi and Hiroaki Miyata
Supplementary Data
==================
###### Supplementary Information
The authors have no conflict of interest.
This report was authorized by the institutional review board of Saiseikai Yahata General Hospital (Authorization No. 132).
Figures are available as supplementary information at the online article pages on J-STAGE and PMC.
The original Annual Report was published in Japanese Journal of Vascular Surgery Vol. 28 (2019) No. 3; however, errors in numerical data and a table were detected after the publication. The erratum was published in the Vol. 28 (2019) No. 4 of the same journal. This translation reflects those corrections.
| {
"pile_set_name": "PubMed Central"
} |
Introduction {#Sec1}
============
Cellular compartmentalization is a hallmark of eukaryotic cells and transport between compartments is required to maintain cellular homeostasis and energy production^[@CR1]^. This cellular organization is based on the interaction between actin filaments, microtubules and intermediate filaments with cellular organelles, such as mitochondria^[@CR1]^. Both actin filaments and microtubules can dynamically assemble and disassemble polar filaments that contribute to cell polarity and cell division, whereas intermediate filaments form non-polar structures that are disassembled during mitosis^[@CR2]^.
An important regulator of microtubule dynamics during cell division is the protein γ-tubulin^[@CR3],\ [@CR4]^. We and others have recently reported that γ-tubulin forms a cellular meshwork of γ-strings^[@CR5]--[@CR7]^ and γ-tubules^[@CR8]^. While γ-tubules are polar cytosolic filaments within the γ-string meshwork, γ-strings are detected in both the cytoplasm and the nucleus and are formed of non-polar protein threads that cross the double membrane of the nuclear envelope. The γ-string meshwork forms a boundary around chromatin, which coordinates cytosolic and nuclear events during mitosis by assuring that a nuclear envelope forms around daughter chromosomes^[@CR5]^. Furthermore, the γ-string meshwork formed by the C-terminal DNA-binding region of γ-tubulin forms a cytosolic network as well^[@CR9]^. These observations together suggest that the γ-tubulin meshwork may be a dynamic network that contributes to cellular homeostasis.
In the present study, we characterize the dynamics of the γ-tubulin meshwork and its implication in cellular homeostasis. We show that γ-strings are a mitochondrial structural component that associates with both mitochondrial DNA and membranes. In addition, we demonstrate that the GTPase domain of γ-tubulin facilitates the organization of a mitochondria-associated γ-string meshwork and that γ-tubulin depletion disrupts the meshwork. Our findings highlight an essential role for γ-tubulin in mitochondrial structure and ultimately mitochondrial function.
Results {#Sec2}
=======
C-terminal γ-tubulin^336--451^ associates with mitochondria {#Sec3}
-----------------------------------------------------------
We found that in close vicinity to the nuclear envelope, endogenous γ-tubulin formed a network of strings, γ-strings, which grew from the nuclear compartment and towards the plasma membrane (Fig. [1a](#Fig1){ref-type="fig"}). The immunofluorescence staining of γ-strings was abolished following gene editing, using a single-guide (sg)RNA targeting the γ-tubulin genes, *TUBG1* and *TUBG2* (*γTubulin* hereafter, unless a specific isoform is referenced) , demonstrating that γ-strings consist of γ-tubulin (Fig. [1b](#Fig1){ref-type="fig"}). Notably, we found that sgRNA-induced reduction of γ-tubulin was cytotoxic when γ-tubulin protein levels dropped below 50% (Fig. [1c](#Fig1){ref-type="fig"}). After *γTubulin* sgRNA transfection, cells divided during the subsequent three to four days. Thereafter, cells remained in interphase for several days before dying (Fig. [1d](#Fig1){ref-type="fig"}). Immunofluorescence staining of *γTubulin* sgRNA expressing cells with an anti-cytochrome c antibody and a chromatin dye, showed the reduction of γ-tubulin expression caused the mitochondrial release of cytochrome c and chromatin condensation, both apoptotic markers (Supplementary Fig. [1](#MOESM1){ref-type="media"})^[@CR10]^.Fig. 1γ-Tubulin forms protein strings and γ-tubulin knockdown is cytotoxic. **a**, **b** Confocal images of fixed U2OS or U2OS expressing *γTubulin* sgRNA (Cas9-crispGFP; green) that were immunostained with an anti-γ-tubulin (γTubulinAb) antibody originated in mouse. **a** The white box shows the magnified area displayed in the inset. **a**, **b** Yellow and white arrows show γ-strings or the indicated cell, respectively (*N* = 5). **c** U2OS expressing *γTubulin* sgRNA (Cas9-crispGFP) at day 0 were incubated for the indicated time before fixation. Cells were stained as in **a**. Within samples, quantification of γ-tubulin was done with ImageJ software by comparison of immunofluorescently labelled γ-tubulin in cells expressing Cas9-crispGFP with non-expressing cells (control; *N* = 7--11 cells). Graph represents the relative percentage of cells that expressed Cas9-crispGFP at the indicated period of time. To adjust for differences in transfection efficiency, the sample containing the largest number of cells expressing Cas9-crispGFP was defined as 100% and values at other time points were compared with that sample (*N* = 3--7). **d** Schematic representation of the time-lapse experiments. Cells were transfected (day 0) and experiments started three or five days after, as indicated. Cell populations were monitored for three days before fixation and subsequently immunostained with an anti-γ-tubulin (γTub) antibody that originated in mouse. The levels of endogenous γ-tubulin were determined as in **c**. The differential interference contrast (DIC)/fluorescence images show time-lapse series from U2OS cells expressing Cas9-crispGFP (green; dashed lines). Images were collected every 8 min. Top images show four cells and their respective daughter cells (yellow, blue, orange and white) before dying (white, orange, yellow and blue). Bottom images show ten cells that either remained in interphase (yellow, light blue, dark blue, orange, black, grey, brown, magenta, green, and white) or died (black, light blue, and white) during the course of the experiment. White boxes show the live cells displayed in the insets. Insets show γ-tubulin in fixed cells (Fixed) expressing Cas9-crispGFP. Numbers in images indicate the remaining protein levels of γ-tubulin relative to control in the indicated cells (*N* = 3). Scale bars are 10 μm in images. Please, see Supplementary Fig. [1](#MOESM1){ref-type="media"}
In contrast to cells expressing *γTubulin* sgRNA that undergo apoptosis (Fig. [1c, d](#Fig1){ref-type="fig"}), we instead used stable *γTubulin* shRNA expressing cells (*γTubulin*sh-U2OS). The *γTubulin* shRNA expression reduced the endogenous γ-tubulin pool by \~50% (Supplementary Fig. [2a](#MOESM1){ref-type="media"})^[@CR5],\ [@CR11]^, and we compensated for this reduction by stably co-expressing a sh-resistant human GFP-C-terminal region (residues 334--449, GFP-γ-tubulin^334--449^) in U2OS cells (*γTubulin*sh-U2OS-γ-tubulin^334--449^; Fig. [2a](#Fig2){ref-type="fig"}, Supplementary Fig. [2a](#MOESM1){ref-type="media"}). We found that the endogenous γ-strings (Fig. [1a](#Fig1){ref-type="fig"}) were similar to those structures formed by *γTubulin*sh-U2OS-γ-tubulin^334--449^, as shown by the immunofluorescence staining of *γTubulin*sh-U2OS-γ-tubulin^334--449^ with an anti-γ-tubulin antibody (Fig. [2b](#Fig2){ref-type="fig"}). Super-resolution microscopy showed that various GFP-γ-strings^334--449^ merged forming a dense protein meshwork (Fig. [2a](#Fig2){ref-type="fig"}).Fig. 2Endogenous cytosolic γ-tubulin associates with mitochondria. **a** Structure of human wild-type γ-tubulin (h-γTubulin) and the γ-tubulin C terminus (C-γTubulin^336--451^), depicting the GTPase domain and the C-terminal region of γ-tubulin. U2OS cells stably expressing *γTubulin* shRNA and sh-resistant GFP-γ-tubulin~resist~^334-449^ fragment were imaged by structured illumination microscopy. The yellow arrows show γ-strings. **b**, **c** Confocal fluorescence images of fixed or live U2OS cells stably expressing *γTubulin* shRNA and GFP-γ-tubulin~resist~^334--449^. Mitochondria were stained with the fluorescent dye MitoTracker or with the mitochondrial marker cytochrome c oxidase subunit II (MTCO2) and the total pool of γ-tubulin was immunofluorescence stained with an anti-γ-tubulin (γTubulinAb) antibody originated in rabbit. **c** Co-localization pixel maps (CM) of the red and green (blue) channels of images are shown. White areas denote colocalized pixels between channels (MitoTracker (life), Person's *R* = 0.7, fraction of red (MitoTracker) overlapping blue (GFP-γ-tubulin~resist~^334--449^) M1 = 1.0, fraction of blue overlapping red M2 = 0.9; MTCO2, Person's *R* = 0.5, fraction of red (MTCO2) overlapping blue (GFP-γ-tubulin~resist~^334--449^) M1 = 1.0, fraction of blue overlapping red M2 = 0.8). **d** Confocal fluorescence images of U2OS cells stably expressing γ-tubulin^336--451^. The mitochondria were stained with MitoTracker and the total pool of γ-tubulin with a γTubulinAb originated in rabbit. **e** Immunofluorescent staining of endogenous γ-tubulin in U2OS cells transiently expressing pmTurquoise2-mito (mito) with a γTubulinAb originated in rabbit. **d**, **e** Co-localization pixel maps (CM) of the red and green (blue) channels of the magnified areas displayed in the inset (the yellow box). White areas denote colocalized pixels between channels (**d**, MitoTracker, Person's *R* = 0.7, fraction of red (MitoTracker) overlapping blue (γTubulinAb) M1 = 1.0, fraction of blue overlapping red M2 = 0.9; **e**, Mito, Person's *R* = 0.5, fraction of red (γTubulinAb) overlapping blue (Mito) M1 = 0.9, fraction of blue overlapping red M2 = 1.0). **a**, **d**, **e** The white box shows the magnified areas displayed in the inset. **f** Fixed U2OS cells transiently expressing *γTubulin* sgRNA (Cas9-crispGFP) were immunofluorescence stained with an anti-MTCO2 antibody and a γTubulinAb originated in mouse. (**a**--**f**) The figure shows representative images from at least six experiments. Scale bars are 10 μm in images. Please, see Supplementary Fig. [2](#MOESM1){ref-type="media"}
By associating with the nuclear envelope, γ-strings give support to the double nuclear membrane structure and connect the cytoplasm with the chromatin^[@CR5]^. Similar to the nuclear compartment, mitochondria contains both DNA and is surrounded by a double membrane^[@CR12]^. Recent work demonstrates an association of γ-tubulin with mitochondrial membranes^[@CR13]^. In addition, the γ-tubulin's DNA binding motif is encoded in the γ-tubulin C terminus^[@CR14]^. Thus, we hypothesized that γ-strings may connect the cytoplasm with the mitochondrial DNA (mtDNA). Consequently, the γ-string meshwork in close proximity to the nuclear envelope, formed by either endogenous γ-tubulin (Fig. [1a](#Fig1){ref-type="fig"}) or GFP-γ-tubulin^334--449^ (Fig. [2a, b](#Fig2){ref-type="fig"}), looked similar in structure, and its positioning resembled the mitochondrial network formed in close proximity to the nuclear envelope^[@CR15]^. To analyze a possible association between γ-strings and mitochondria, we used immunofluorescence to stain *γTubulin*sh-U2OS-γ-tubulin^334-449^ with a fluorescent dye that identified mitochondria in live cells (MitoTracker, Fig. [2c](#Fig2){ref-type="fig"}), or with an antibody that recognized the mitochondrial marker cytochrome c oxidase subunit II (MTCO2; Fig. [2c](#Fig2){ref-type="fig"}). We also expressed γ-tubulin^336-451^ in U2OS to ascertain that the GFP did not affect the cellular positioning of the C-terminal region (Fig. [2b, d](#Fig2){ref-type="fig"}, Supplementary Fig. [2b](#MOESM1){ref-type="media"}). In addition, in U2OS cells, we ectopically expressed the pmTurquoise2-tagged mitochondrial-targeting signal from cytochrome c oxidase subunit VIII A (amino acids 1--29; Fig. [2e](#Fig2){ref-type="fig"}, Supplementary Fig. [2c](#MOESM1){ref-type="media"}), which from now on will be referred as mito^[@CR15]^. Confocal microscopy confirmed that endogenous γ-strings, recombinant GFP-γ-strings^334--449^, and γ-tubulin^336--451^ formed a dense meshwork in mitochondria-rich areas (Fig. [2c--e](#Fig2){ref-type="fig"}). Finally, sgRNA mediated reduction of γ-tubulin disrupted the mitochondrial network (Fig. [2f](#Fig2){ref-type="fig"}). These data suggest that γ-tubulin affects the morphology of the mitochondrial meshwork.
γ-Tubulin is a mitochondrial protein {#Sec4}
------------------------------------
Interestingly, despite evidence that the mitochondrial proteome is derived from endosymbiotic bacteria^[@CR12]^, most mitochondria lack the expression of the γ-tubulin homologue FtsZ^[@CR16]^, suggesting that γ-tubulin may replace FtsZ function in mitochondria. To confirm a possible role of endogenous γ-tubulin in mitochondria, we investigated the distribution of endogenous γ-tubulin by immunoelectron microscopy of high-pressure frozen U2OS cells (Fig. [3a, b](#Fig3){ref-type="fig"}). As γ-tubulin is distributed throughout the cell, we optimized the antibody concentration so that cellular structures were only partially stained. We found that mitochondria were connected to each other, and to the nuclear membrane by protein strings, which were 4--6 nm in diameter (Fig. [3a, b](#Fig3){ref-type="fig"}). Furthermore, a γ-tubulin antibody recognized these strings (Fig. [3b](#Fig3){ref-type="fig"})^[@CR5]^ and the immunostaining was negated following *γTubulin* sgRNA expression, confirming the specificity of the antibody (Supplementary Fig. [3a](#MOESM1){ref-type="media"}). Finally, the γ-tubulin antibody was able to stain γ-strings both inside and outside the mitochondria (Fig. [3b](#Fig3){ref-type="fig"}). Together, the data presented here implies that γ-strings form a network that harbours mitochondria.Fig. 3γ-Strings are associated with mitochondria. **a**, **b** Immunoelectron microscopy detection of endogenous γ-tubulin using three different conditions in high-pressure frozen (HPF) U2OS cells: first, no antibody (**a**), second, gold conjugated protein A (**a**) and third, an anti-γ-tubulin antibody originated in rabbit, and gold conjugated protein A (**b**, γTubulinAb). Images show the plasma membrane (PM), the nuclear envelope (NE), cytosol (C), mitochondria (MT) and nucleus (N) of a U2OS cell. Blue arrows show γ-strings and arrowheads show immunolabelled γ-strings. White arrows show the nuclear envelope or the plasma membrane, as indicated. White dashed boxes show the magnified areas displayed in the inset (*N* = 5). **c** The crude mitochondria fraction from U2OS and MCF10A cells was biochemically prepared. Each sample was subjected to immunoprecipitation (IP) with a control (C), an anti-γ-tubulin (γ; originated in mouse) or an anti-α-tubulin (α) antibody, as indicated, and developed by Western blotting (WB) with an anti-MTCO2 antibody (top, arrowhead), and then reprobed with γ-tubulin (originated in rabbit) and α-tubulin. Aliquots of the cytosolic lysates used in the immunoprecipitations were run as loading controls (lys. and Total lys.) and analyzed by Western blotting. Graph shows the mean content of γ-tubulin and α-tubulin found in their respective immunoprecipitates in the mitochondrial crude fraction. To adjust for differences between WBs, the protein content in control immunoprecipitates was defined as 1 and values of the other immunoprecipitates were compared with that level (mean ± SD; *N* *=* 4, \**P* \< 0.05). **d** The biochemically isolated crude mitochondria fraction from a MCF10A cell population was fixed and immunostained with anti-γ-tubulin originating in mouse (M) or rabbit (R) and anti-MTCO2, anti-α-tubulin or anti-GCP2 antibody, as indicated. Scale bars are 10 μm in images. The electron microscopy image shows mitochondria in the crude mitochondria fraction. Arrowheads show γ-strings between and in mitochondria (*N* = 4). Please, see Supplementary Fig. [3](#MOESM1){ref-type="media"}
Previous work demonstrates that microtubules are necessary for both mitochondrial activity and mitochondrial intracellular motor-driven transport^[@CR17],\ [@CR18]^. To compare the amount of mitochondria-associated γ-tubulin with the amount of mitochondria-associated microtubules, we analyzed immunoprecipitates of endogenous γ- and α-tubulin from the crude mitochondria fraction. The immunoprecipitated complexes highlighted an association of γ-tubulin with the mitochondrial membrane marker MTCO2 in U2OS and MCF10A cells (Fig. [3c](#Fig3){ref-type="fig"}, Supplementary Fig. [3b](#MOESM1){ref-type="media"}). Moreover, we noticed that in comparison to α-tubulin, γ-tubulin was highly enriched in the crude mitochondrial fraction (Fig. [3c](#Fig3){ref-type="fig"}), confirming the presence of a mitochondria-associated γ-tubulin pool that is not microtubule related. Accordingly, Western blot analysis of lysates from mitochondria from MCF10A cells prepared by density gradient centrifugation using Percoll confirmed the presence of γ-tubulin in the mitochondria (Supplementary Fig. [3c](#MOESM1){ref-type="media"}).
In the cytoplasm, γ-tubulin together with various γ-tubulin complex proteins (GCP) is able to nucleate microtubules by forming the γ-tubulin ring shaped complex (γTURC)^[@CR3],\ [@CR4],\ [@CR19]^. To study a possible association of the γTURC with the mitochondria, we performed immunofluorescent staining of the isolated crude mitochondria fraction from MCF10A cells, which confirmed that, in comparison to α-tubulin and GCP2, γ-tubulin is enriched in mitochondria (Fig. [3d](#Fig3){ref-type="fig"}). Finally, electron microscopy analysis of the isolated crude mitochondrial fraction showed the presence of γ-strings between and inside mitochondria (Fig. [3d](#Fig3){ref-type="fig"}). Thus, the association of γ-strings with mitochondrial membranes suggests that γ-tubulin forms a membrane-associated meshwork that provides mitochondria with a structural scaffold.
γ-Tubulin binds to mitochondrial DNA {#Sec5}
------------------------------------
A nuclear localization signal (NLS) mediates γ-tubulin translocation to the nucleus^[@CR14]^. To study the effect of mutations in the γ-tubulin NLS, we stably co-expressed *γTubulin* sgRNA (depleted the endogenous γ-tubulin pool), and either a wild-type sg-resistant human *TUBG1* or a sg-resistant *TUBG1* containing a mutated NLS. In comparison to cells regularly expressing γ-tubulin, we found that mutations of R399A, K400A, and R409A in the NLS^[@CR14]^ caused the formation of tubular structures (Fig. [4a](#Fig4){ref-type="fig"}, Supplementary Fig. [4](#MOESM1){ref-type="media"}), suggesting that the reduced import of γ-tubulin to the nuclear compartment enhances the binding of γ-tubulin to mtDNA. To determine whether γ-tubulin binds to mtDNA, we performed a chromatin immunoprecipitation (ChIP) assay using γ-tubulin antibodies and found that endogenous γ-tubulin was present on mtDNA (Fig. [4b](#Fig4){ref-type="fig"}). To identify underlying mitochondrial processes associated with γ-tubulin binding to DNA, we synchronized cells in S-phase (Fig. [4c](#Fig4){ref-type="fig"}) and mapped the location of γ-tubulin on the mtDNA of MCF10A and of MCF10A cell populations that stably expressed *γTubulin* shRNA (*γTubulin*sh-MCF10A) by sequencing the DNA associated with chromatin immunoprecipitates (ChIP-seq) from γ-tubulin (Fig. [4d--f](#Fig4){ref-type="fig"}). Also, we evaluated γ-tubulin's effect on RNA expression of mitochondrial-related mRNA by massive sequencing of the purified RNA (RNA-seq) from the studied cell populations (Fig. [5a](#Fig5){ref-type="fig"}). Differential protein peaks on DNA were called with MACS2 (version 2.1.1)^[@CR20]^. The identified differential peaks associated with nuclear chromatin were more numerous in γ-tubulin immunoprecipitates from MCF10A cells, than those from *γTubulin*sh-MCF10A cells, suggesting that shRNA-mediated reduction of γ-tubulin protein levels led to a reduced binding of γ-tubulin to genomic DNA (Fig. [4d](#Fig4){ref-type="fig"}).Fig. 4γ-Tubulin binds to mitochondrial DNA. **a** Structure of human wild-type γ-tubulin and the γ-tubulin DNA-binding domain (DnaBD), depicting residues R399, K400 and R409 in the nuclear localization signal of γ-tubulin. Confocal fluorescence microscopy of fixed U2OS stably expressing *γTubulin* sgRNA (Cas9-crispGFP) and co-expressing a *γTubulin* sgRNA resistant transcript (γTubulin) or a mutant form, γTubulin^R399A-K400A-R409A^ (γTubulin^399-400-409^). The recombinant proteins were immunostained with an anti-γ-tubulin antibody (γTubulinAb) that originated in mouse. **b** U2OS cells were analyzed by ChIP using an anti-γ-tubulin antibody (γTubChIP) that originated in rabbit. PCR primers amplified the indicated regions of the mitochondrial DNA (*N* = 3). **c** MCF10A and *γTubulin* shRNA stably expressing MCF10A (γ*Tubulin* sh MCF10A) cells were synchronized in early S-phase (0 h) by double thymidine block and released for 1 h and 2 h. Cell cycle progression was monitored by determining the DNA content of cells with a nucleocounter (graphs; *N* = 3). Total lysate from MCF10A (Control) and MCF10A cells stably expressing *γTubulin* shRNA (γ*Tub* sh) were analyzed by Western blot (WB) for the expression of endogenous γ-tubulin. An α-tubulin loading control is shown (*N* = 3). The number on the WB indicates the level of depletion of γ-tubulin relative to control. To adjust for differences in protein loading, the protein concentration of γ-tubulin was determined by its ratio with endogenous α-tubulin. The protein ratio in control extracts was set to 1. **d**, **e** To map the location of γ-tubulin in the chromatin of MCF10A and of MCF10A cells stably expressing *γTubulin* shRNA (γ sh), we sequenced the DNA associated with chromatin immunoprecipitates from γ-tubulin. Immunoprecipitations were performed in early S-phase synchronized cell populations using an anti-γ-tubulin antibody. Graphs show the number of binding sites (peaks called) found in the human genome (**d**) or mitochondrial chromosome (**e**) to which γ-tubulin binds at the indicated period of time. **f** The graphs show ChIP-seq analysis of γ-tubulin distribution on mitochondrial chromosome (ChrM). The entire chromosome M is presented. Black arrows indicate areas loaded with γ-tubulin. In grey is the schematic representation of the called peaks (*N* = 2). Please, see Supplementary Fig. [4](#MOESM1){ref-type="media"}Fig. 5The γ-tubulin meshwork controls mitochondrial activity. **a** GSEA of mitochondrial-upregulated gene set performed on a ranked gene list of differentially expressed genes between synchronized MCF10A (non shRNA) and *γTubulin*sh-MCF10A (*Tubulin* shRNA) cells. **b** Western blots (WB) show total lysates (Tot. lys.) of U2OS or MCF10A cells that stably expressed *γTubulin* shRNA (γ*TUB* sh) using the indicated antibodies. Anti-α-tubulin antibody was used as loading control (*N* = 3). Arrowheads indicate proteins whose expression is affected by the expression of *γTubulin* shRNA. The numbers on the WBs indicate variations in MTCO2, ATP6, HTATIP2, SLC25A6, and γ-tubulin expression relative to *γTubulin* shRNA non-expressing cells, as indicated. To adjust for differences in protein loading, the protein concentration of the various proteins was determined by their ratio with α-tubulin for each sample. The protein ratio in control extracts was set to 1. **c** The mean values of the relative basal oxygen consumption was determined using Seahorse analyser in U2OS cells, U2OS cells stably expressing *γTUB* sh and stably co-expressing GFP-γ-tubulin~resist~ (γTubGFP) or GFP-A^13^γ-tubulin~resist~ (γTub13A) and U2OS cells pre-treated with CDA for 2 h. Note that CDA is present during the Seahorse assay, which takes 3 h. The data were normalized to the total number of cells. The oxygen consumption rate activity of U2OS cells was set as 1, and relative activities were calculated (mean ± SEM; *N* = 4--6, \*\*\**P* \< 0.001). **d** Structure of human wild-type γ-tubulin (h-*γTubulin*), depicting residue Cys13 in the GTPase domain of γ-tubulin. Graph shows seahorse assay of the respiratory capacity after addition of glucose mixture (20 mM glucose, 20 μg/ml insulin), 4 μg/ml oligomycin, 1 μM FCCP and 0.5 μM Rotenone in 28 × 10^3^ of the indicated cells. Note the very low basal oxygen consumption rate of CDA-treated U2OS cells (mean ± SEM; *N* *=* 4--6). **e** WST-1 assay showing the metabolic activity of U2OS cells transfected with the indicated construct or after 4 h CDA pre-treatment (mean ± SEM; *N* = 6, \*\*\**P* \< 0.001)
In S-phase, γ-tubulin accumulates in the chromatin of cells^[@CR14],\ [@CR21],\ [@CR22]^. Accordingly, we found an increase in the number of peaks associated with genomic- and mitochondrial DNA as the cells progressed through S-phase (2 h; Fig. [4d--f](#Fig4){ref-type="fig"}). The identified differential peaks associated with mtDNA at 2 h were more abundant in γ-tubulin immunoprecipitates from MCF10A cells than those from *γTubulin*sh-MCF10A cells. This indicated that shRNA-mediated reduction of γ-tubulin protein levels led to reduced binding of γ-tubulin to mtDNA (Fig. [4e](#Fig4){ref-type="fig"}). However, we found mtDNA in γ-tubulin immunoprecipitates from *γTubulin*sh-MCF10A cells in early S-phase (1 h) that was not found in the ones from MCF10A cells. We think that this may be caused by the limited reduction of γ-tubulin protein levels in *γTubulin*sh-MCF10A cells, which affects γ-tubulin dynamics during S-phase progression. Taken together, the data presented here demonstrate that γ-tubulin binds to mtDNA.
γ-Tubulin regulates the expression of mitochondrial genes {#Sec6}
---------------------------------------------------------
Given that γ-tubulin associates with both nuclear- and mitochondrial DNA, we postulate that the protein levels of γ-tubulin might synchronize gene expression of mitochondrial-related genes with on-going biological processes, such as mtDNA replication. To elucidate the effect of *γTubulin* reduction on mitochondrial-related gene signatures, we made use of mitochondrial-regulating gene signatures from 58 pre-defined gene sets containing mitochondrial-associated genes (Supplementary Data [1](#MOESM3){ref-type="media"}). Also, we created a custom gene set using all genes in the MitoCarta^[@CR23]^. The detailed listing of all 59 gene sets is presented in Supplementary Data [1](#MOESM3){ref-type="media"}. The effect of *γTubulin* reduction on mitochondrial target gene signatures were examined by performing a gene set enrichment analysis (GSEA)^[@CR24],\ [@CR25]^ between RNA-seq samples from MCF10A and from *γTubulin*sh-MCF10A cells. Among the top 10 gene sets found to be affected, we noted biological process such as regulation of mitochondrial membrane permeability and reactome RNA, POL I, RNA POL III, as well as mitochondrial transcription (Supplementary Tables [1](#MOESM1){ref-type="media"} and [2](#MOESM1){ref-type="media"}). The detailed results of the GSEA that shows an example of an enrichment plot produced for one of the used gene set is presented in Fig. [5a](#Fig5){ref-type="fig"} and Table [1](#Tab1){ref-type="table"}. In concert, the data presented here suggest that alterations in the γ-tubulin meshwork affect the expression of mitochondrial-related genes.Table 1Upregulated mitochondria-related genesNo.GeneP valueFDRNo.GeneP valueFDR1.GTPBP36.8 10^-14^7.9 10^-10^16.FAM210A2.6 10^-5^9.6 10^-3^2.HTATIP22.0 10^-10^7.2 10^-7^17.CMC22.9 10^-5^1.0 10^-2^3.RECQL41.7 10^-9^4.9 10^-6^18.SLC25A253.0 10^-5^1.1 10^-2^4.EYA25.3 10^-9^1.2 10^-5^19.UCP25.6 10^-5^1.7 10^-2^5.COX5B2.2 10^-8^4.0 10^-5^20.OXLD11.3 10^-4^3.2 10^-2^6.SLC25A63.1 10^-7^3.3 10^-4^21.SLC25A292.2 10^-4^4.9 10^-2^7.NME31.1 10^-6^8.7 10^-4^22.CKMT1A2.2 10^-4^4.8 10^-2^8.TP733.2 10^-6^2.0 10^-3^23.MPC11.7 10^-4^4.1 10^-2^9.UQCRFS14.2 10^-6^2.4 10^-3^24.HIST1H3J2.9 10^-5^1.0 10^-2^10.PMAIP14.9 10^-6^2.7 10^-3^25.PINK12.3 10^-5^9.0 10^-3^11.FAM195A8.3 10^-6^4.1 10^-3^26.AK21.8 10^-5^7.5 10^-3^12.H2AFX1.6 10^-5^6.7 10^-3^27.CKMT1B1.1 10^-5^5.2 10^-3^13.BRF11,6 10^-5^6.7 10^-3^28.HIST3H2BB3.7 10^-7^3.6 10^-4^14.FASN1.8 10^-5^7.4 10^-3^29.HIST1H2BJ7.5 10^-8^9.9 10^-5^15.MRPL432.0 10^-5^7.8 10^-3^Genes are displayed in order of most positively enriched (genes 1--21) to the most negatively enriched (genes 22--29) in the *γTubulinsh*-MCF10A cells. *P* values and FDR values were produced using edgeR^[@CR53]^, using a generalized exact binomial test
To further elucidate the effect of γ-tubulin on mtDNA dynamics, we analyzed changes in the expression of the mitochondrial-related genes *HTATIP2* (HIV-1 TAT-Interacting Protein 2) and *SLC25A6* (Solute Carrier Family 25 Member 6), and of the mitochondrial-encoded genes *MTCO2* and *ATP6* (ATP synthase F0 Subunit 6) in U2OS and MCF10A cell populations that stably expressing *γTubulin* shRNA. In comparison to control cells, reduced expression of γ-tubulin in *γTubulin*sh-U2OS and *γTubulin*sh-MCF10A cells caused an increased protein expression of MTCO2, HTATIP2 and SLC25A6, whereas the expression of ATP6 was decreased (Fig. [5b](#Fig5){ref-type="fig"}). Furthermore, it can also be noted that HTATIP2, and SLC25A6 were amongst the most positively enriched genes in *γTubulin*sh-MCF10A cells (Table [1](#Tab1){ref-type="table"}). These observations confirm that the protein levels of γ-tubulin affect the expression of mitochondrial-related genes.
The γ-tubulin meshwork controls mitochondrial function {#Sec7}
------------------------------------------------------
To establish the metabolic effect of the interaction of mitochondria with γ-tubulin, we analyzed mitochondrial respiration by determining the cellular oxygen consumption rate with the Seahorse Extracellular Flux analyser (Methods section) in cells with variable concentrations of γ-tubulin. We measured the oxygen consumption rate in U2OS cells, *γTubulin*sh-U2OS and in *γTubulin*sh-U2OS cells stably co-expressing human GFP-tagged sh-resistant γ-tubulin (*γTubulin*sh-U2OS-γ-tubulin~resist~; Fig. [5c](#Fig5){ref-type="fig"}, Supplementary Fig. [5a, b](#MOESM1){ref-type="media"}).
Moreover, considering that the GTPase domain of β-tubulin affects microtubule dynamics^[@CR3],\ [@CR26]--[@CR28]^, we postulated that the N-terminal region of γ-tubulin might control the γ-string meshwork associated with mitochondria, as it encloses γ-tubulin's GTPase domain^[@CR11]^. To test this, we mutated Cyst^[@CR13]^ to an Ala (GFP-A^13^γ-tubulin~resist~), as this mutation impairs GTP binding to the GTPase domain^[@CR11],\ [@CR28]^ and stably co-expresses the mutated recombinant protein in *γTubulin*sh-U2OS cells (*γTubulin*sh-U2OS-A^13^γ-tubulin~resist~; Supplementary Fig. [5b](#MOESM1){ref-type="media"}). In addition, we monitored the oxygen consumption rate in citral dimethyl acetyl treated U2OS cells (CDA; impairs the GTPase activity of γ-tubulin^[@CR11]^; Fig. [5c, d](#Fig5){ref-type="fig"}, Supplementary Fig. [5c](#MOESM1){ref-type="media"}). In comparison to control U2OS cells, only 100 µM CDA treatment significantly decreased the basal oxygen consumption rate per cell. Oligomycin, an inhibitor of the ATP synthase; FCCP, the uncoupler of mitochondrial oxidative phosphorylation carbonyl cyanide 4-(trifluoromethoxy) phenylhydrazone; and rotenone (interferes with the electron transport chain in mitochondria) had effects on the oxygen consumption rate in all the cell lines tested, by decreasing (oligomycin and rotenone) or increasing (FCCP) the oxygen consumption rate. However, neither oligomycin nor FCCP had an effect on the low basal oxygen consumption rate levels of CDA-treated U2OS cells (Fig. [5d](#Fig5){ref-type="fig"}). These data were further confirmed by a WST-1 assay, which measured the succinate-tetrazolium reductase system in the respiratory chain of the mitochondria, and confirmed that only CDA-treated cells exhibited an impaired mitochondrial respiratory capacity (Fig. [5e](#Fig5){ref-type="fig"}).
To investigate whether the contradictory effects of CDA treatment and *γTubulin*sh-U2OS and *γTubulin*sh-U2OS-A^13^γ-tubulin~resist~ cells depended on an adaptive mechanism of these cells to an impaired mitochondrial respiration, we estimated variations in the expression levels of the mitochondrial-encoded protein MTCO2 (Fig. [6a--c](#Fig6){ref-type="fig"}). CDA treatment did not affect the expression of MTCO2 most likely owing to the short-term exposure of the cells to the drug (4 h; Fig. [6a](#Fig6){ref-type="fig"}). However, longer exposure of U2OS cells to either CDA or to another inhibitor of the GTPase activity of γ-tubulin, dimethyl fumarate (DMF; Supplementary Fig. [5c](#MOESM1){ref-type="media"})^[@CR11]^ increased the expression of MTCO2 (Fig. [6b](#Fig6){ref-type="fig"}) and further confirmed that the γ-tubulin meshwork controls the expression of mitochondrial-related genes.Fig. 6Reduced protein levels of γ-tubulin increase mitochondrial mass. **a**, **b** Total lysate (Tot. lys.) of U2OS cells, U2OS cells stably expressing *γTubulin* shRNA (*γTUB* sh) and U2OS treated with CDA or DMF for 4 h (**a**) or 24 h (**b**) before harvested. Protein levels of the mitochondrial marker MTCO2 and γ-tubulin were analyzed by Western blotting (WB) with an anti-MTCO2 and anti-γ-tubulin antibody, both originated in rabbit. Anti-α-tubulin was used as loading control (*N* = 4). **c** U2OS cells were transiently transfected with *γTUB* sh (day 0) or control shRNA (control sh) and changes in the protein levels of γ-tubulin, MTCO2, Drp1, Opa1 and Mfn1 and in the metabolic activity were analyzed by Western blotting (*N* = 3). **a**--**c** Anti-α-tubulin antibody was used as loading control. Numbers on the WBs indicate variations in MTCO2, Drp1, Opa1 and Mfn1 expression relative to day 0, as indicated. To adjust for differences in protein loading, the protein concentration of the various proteins was determined by their ratio with α-tubulin for each sample. The protein ratio in control extracts was set to 1. **d** WST-1 assay showing the metabolic activity of U2OS cells transfected with the indicated construct and treated as in **c** (mean ± SEM; *N* = 6). **e** Ratio between mitochondrial and nuclear DNA content measured by quantitative PCR in U2OS and U2OS cells stably expressing *γTub* sh and stably co-expressing GFP-γ-tubulin~resist~ (γTubGFP) or GFP-A^13^γ-tubulin~resist~ (γTub13A) and U2OS cells pre-treated with CDA for 4 h (mean ± SEM; *N* = 3, \**P* \< 0.05; each sample the qPCR reaction was performed in triplicates). **f** Extracts from the mitochondria crude fraction from synchronous and non-synchronous (ns) MCF10A and MCF10A cells stably expressing *γTUB* sh were examined by WB using an antibody against DNA polymerase gamma (*N* = 3). DNA content was determined with a nuclear counter (percentage of S-phase cells indicated). The remaining mitochondrial crude depleted lysate (Tot. lys.) was run as loading control and examined with an anti-α-tubulin antibody. Numbers on the WBs indicate variations in γ-tubulin expression and are calculated as in **a**. Please see Supplementary Fig. [5](#MOESM1){ref-type="media"}
Accordingly, a transient *γTubulin* shRNA-mediated reduction of γ-tubulin protein levels caused an immediate increase of the levels of MTCO2 (Fig. [6c](#Fig6){ref-type="fig"}), whereas no significant changes of the mitochondrial capacity were observed (Fig. [6d](#Fig6){ref-type="fig"}). This agrees with a rapid adaptation of the mitochondrial respiratory capacity to variations in the γ-tubulin meshwork. Finally, the protein levels of optic atrophy 1 (OPA1, regulates mitochondrial shape and morphology) and mitofusin 1 (Mfn1, mediates mitochondrial fusion) were transiently increased after 1--2 days and were returned to normal levels after 4 days, whereas the protein levels of dynamin-related protein 1 (Drp1, mediates mitochondrial fission) were unaffected. This suggests that as adaptive changes are necessary for the adjustment of mitochondrial function and morphology to variations in the protein levels of γ-tubulin, the mitochondrial cellular mass adjusts (Fig. [6c](#Fig6){ref-type="fig"}).
γ-Tubulin affects the replication of mitochondrial DNA {#Sec8}
------------------------------------------------------
To confirm that changes in mitochondrial mass are an adaptive mechanism for the adjustment of mitochondrial function to variations in the protein levels of γ-tubulin, we estimated alterations in mtDNA copy number per cell by measuring the ratio between mitochondrial and nuclear DNA with quantitative PCR (Fig. [6e](#Fig6){ref-type="fig"}). Indeed, decreased levels of γ-tubulin resulted in increased amount of mtDNA in *γTubulin*sh-U2OS and *γTubulin*sh-U2OS-A^13^γ-tubulin~resist~ cells (Fig. [6e](#Fig6){ref-type="fig"}), confirming that to maintain a normal basal respiratory capacity, these cell lines need to adjust the mitochondrial mass (Fig. [5](#Fig5){ref-type="fig"} and Fig. [6](#Fig6){ref-type="fig"}). However, inhibition of γ-tubulin with CDA did not affect the number of mitochondria, which is most likely due to the short time exposure of the cells to the drug (4 h; Fig. [6e](#Fig6){ref-type="fig"}).
Finally, to evaluate the effect of γ-tubulin protein levels on mtDNA replication, we altered γ-tubulin levels in a synchronized cell population. We found that in the crude mitochondria fraction of *γTubulin*sh-MCF10A cells, reduced protein levels of γ-tubulin had enhanced the protein levels of the mitochondrial-specific DNA polymerase gamma^[@CR29]^ as the cells progressed through S-phase (Fig. [6f](#Fig6){ref-type="fig"}), which provides us with a potential molecular mechanism accounting for the observed rise in mtDNA.
γ-*Tubulin* knockdown affects mitochondrial function {#Sec9}
----------------------------------------------------
In healthy mitochondria, the action of enzymes of the electron transport chain in the inner membrane of the mitochondria generates a proton gradient across the membrane, which drives the production of ATP. To determine whether endogenous γ-tubulin is essential for mitochondrial function, we measured in single cells the membrane potential by recording the accumulation of the cell-permeant dye tetramethylrhodamine methyl ester (TMRM). In cells with a polarized and intact inner mitochondrial membrane (∆ψ~m~), TMRM signal is bright, and the signal decreases upon hyperpolarization of the inner mitochondrial membrane, as a result of increased proton transport over the membrane. To examine these processes, we reduced the endogenous levels of γ-tubulin in U2OS cells by expression of *γTubulin* sgRNA (*γ*sgRNAU2OS). We added back γ-tubulin by stably co-expressing *γTubulin* sgRNA and a sg-resistant human *γTubulin* gene (*γ*sgRNAU2OS-γ-tubulin~resist~; Fig. [7a](#Fig7){ref-type="fig"}, Supplementary Fig. [4](#MOESM1){ref-type="media"}). In comparison to U2OS and *γ*sgRNAU2OS-γ-tubulin~resist~ cells, *γ*sgRNAU2OS cells responded to increased glucose levels by less polarization (lower decrease of signal; grey trace in Fig. [7b](#Fig7){ref-type="fig"}) of the inner mitochondrial membrane. This implies that the mitochondrial function in *γ*sgRNAU2OS cells is negatively affected in the absence of γ-tubulin. Inhibition of ATP synthase with oligomycin did not differentially affect the ∆ψ~m~ in any of the cells. These data confirm that γ-tubulin is necessary for robust function of the inner mitochondrial membrane. Similarly, the uncoupler of mitochondrial oxidative phosphorylation FCCP disrupted the ∆ψ~m~ of U2OS, *γ*sgRNAU2OS and *γ*sgRNAU2OS-γ-tubulin~resist~ cells equally (Fig. [7b](#Fig7){ref-type="fig"}), demonstrating that the mitochondrial transport and ATP phosphorylation are robustly coupled.Fig. 7Sg-mediated knockdown of γ-*Tubulin* affects the activity of the mitochondria, but not the structure of the endoplasmic reticulum. **a** Confocal fluorescence microscopy of fixed U2OS stably expressing *γTubulin* sgRNA (Cas9-crispGFP) along or with a *γTubulin* sgRNA resistant gene (γTubulin). The protein levels of γ-tubulin in *γTubulin* sgRNA expressing cells were immunostained with an anti-γ-tubulin antibody (γTubulinAb) originating in mouse (*N* = 5). **b** U2OS were transfected with *γTubulin* sgRNA (Cas9Crisp-*γTubulin* sg) at day 0 and incubated for 7 days. Graphs show the mitochondrial membrane potential changes and the maximal hyperpolarization (maximal drop in emission) before oligomycin treatment (Δmin TMRM emission 590 nm) of single U2OS cells or single U2OS cells transiently expressing *γTubulin* sgRNA alone or with a *γTubulin* sgRNA resistant gene (γTubulin sg-resist.). The mitochondrial membrane potential was analyzed after addition of 20 mM glucose, 4 μg/ml oligomycin and 1 μM FCCP by recording the mitochondrial accumulation of the cell-permeant dye tetramethylrhodamine methyl ester (TMRM; mean ± SEM; unpaired two tailed Student's *t*-test, Cas9Crisp-*γTubulin*sg vs. Cas9Crisp-*γTubulin*sg-γTubulin sg-resist, \*\*\**P* \< 0.001, \*\*\*\**P* \< 0.0001; U2OS, *N* = 107 cells; Cas9Crisp-*γTubulin*sg *N* = 46 cells; Cas9Crisp-*γTubulin*sg-γTubulin sg-resist. *N* = 128 cells). **c** Immunoelectron microscopy detection of endogenous γ-tubulin using an anti-γ-tubulin antibody that originated in rabbit, and gold conjugated protein A (γTubulinAb) in high-pressure frozen (HPF) U2OS cells. Images show the cytosol (C), mitochondria (MT) and endoplasmic reticulum (ER) of a U2OS cell. White dashed box shows the magnified area displayed in the inset (*N* = 5). **d** Confocal fluorescent microscopy of fixed U2OS or fixed U2OS transiently expressing *γTubulin* sgRNA (Cas9-crispGFP) that were immunostained with an anti-γ-tubulin (γTubulinAb) antibody that originated in rabbit, and with an antibody that recognized the ER marker calnexin (*N* = 5). Please, see Supplementary Fig. [4](#MOESM1){ref-type="media"}
The endoplasmic reticulum is not affected by γ-tubulin {#Sec10}
------------------------------------------------------
Considering the regulatory role of γTURC's association with the Golgi membrane-associated GMAP-210 protein in the proper positioning and biogenesis of the Golgi apparatus^[@CR30]^, we investigated the possibility that γ-tubulin also interacts with and affects the morphology of the endoplasmic reticulum (ER). Immunoelectron microscopy of high-pressure frozen U2OS cells with an anti-γ-tubulin antibody showed that the anti-γ-tubulin antibody recognized the membrane of the ER (Fig. [7c](#Fig7){ref-type="fig"}). Immunofluorescent analysis of fixed U2OS cells co-stained with an antibody that recognized the ER marker calnexin and an anti-γ-tubulin antibody confirmed that the anti-γ-tubulin antibody recognized the membrane of the ER (Fig. [7d](#Fig7){ref-type="fig"}). To analyze the possible effects of the association between γ-strings and ER, we immunofluorescently co-stained endogenous γ-tubulin and the ER, in U2OS and *γ*sgRNAU2OS cells (Fig. [7d](#Fig7){ref-type="fig"}). We found that *γTubulin* sgRNA mediated depletion of γ-tubulin did not affect the morphology of the ER (Fig. [7d](#Fig7){ref-type="fig"}). Together these data suggest that γ-tubulin is associated with the endoplasmic reticulum, but these interactions do not have an impact on the morphology of the ER.
The metabolite fumarate affects the γ-string meshwork {#Sec11}
-----------------------------------------------------
A membrane-associated protein meshwork that regulates mitochondrial homeostasis may provide a cell with a tool to connect the cellular metabolic status with the mitochondrial respiratory chain. To test this hypothesis, we increased the endogenous levels of the metabolite fumarate by treating U2OS cells with DMF^[@CR11],\ [@CR31]^, a γ-tubulin inhibitor (Supplementary Fig. [5c](#MOESM1){ref-type="media"}) that is a cell-permeable derivative of fumarate. We then measured the activity of the succinate-tetrazolium reductase system in the respiratory chain of the mitochondria, as well as monitoring DMF's effect on the mitochondrial network by immunofluorescence. Similar to CDA treatment, DMF impaired mitochondrial respiratory capacity in the treated cells (Fig. [8a](#Fig8){ref-type="fig"}). Moreover, the effect of both DMF and CDA on the mitochondrial succinate-tetrazolium reductase system was attenuated by reduced levels of γ-tubulin (Fig. [8b](#Fig8){ref-type="fig"}), which confirm that the effects of CDA and DMF on mitochondrial respiration are γ-tubulin dependent. Finally, immunofluorescent microscopy showed that the γ-tubulin inhibitor DMF disassembled γ-strings (Fig. [8a](#Fig8){ref-type="fig"}) and with the higher the amount of DMF, the more disassembled the γ-tubulin meshwork became (Fig. [8a](#Fig8){ref-type="fig"}). These findings provide a novel cellular mechanism that may synchronize the metabolic status of a cell with its mitochondrial activity.Fig. 8The cellular metabolite fumarate and γ-tubulin controls the shape of the mitochondrial network. **a** Confocal fluorescence images of fixed U2OS cells treated for 4 h with the indicated concentrations of DMF. The images show immunofluorescence stained endogenous γ-tubulin and MTCO2 using an anti-γ-tubulin (γTubulinAb) that originated in mouse, and an anti-MTCO2 antibody. Arrowheads and white boxes show cytosolic areas with discontinuous γ-strings and the magnified areas are displayed in the insets, respectively. WST-1 assay (relative mitochondrial succinate-tetrazolium reductase activity) showing the metabolic activity of DMF-treated U2OS cells (mean ± SD; *N* *=* 4--16, \*\*\**P* \< 0.001). The graph shows the percentage of cells with more than two areas containing discontinuous cytosolic γ-strings. A minimum of 100 cells was counted in each sample, and the percentage of cells was calculated (bottom; mean ± SD; *N* = 3, \**P* \< 0.05, \*\*\**P* \< 0.001). **b** WST-1 assay showing the metabolic activity of DMF and CDA-treated U2OS and U2OS cells stably expressing *γTubulin* shRNA (mean ± SD; *N* = 3, \**P* \< 0.05). **c** Confocal fluorescent microscopy of fixed U2OS cells that stably expressed *γTubulin* shRNA and co-expressed GFP-γ-tubulin~resist~ or GFP-A^13^γ-tubulin~resist~. The fluorescent images show representative areas of immunostained cells with an anti-γ-tubulin antibody, which recognized endogenous γ-tubulin and an endogenous anti-MTCO2, as indicated (*N* = 5). **a**, **c** The white boxes show the magnified areas displayed in the insets. Scale bars are 10 μm in images. Please, see Supplementary Figs. [5](#MOESM1){ref-type="media"} and [6](#MOESM1){ref-type="media"}
We therefore investigated whether the γ-tubulin meshwork plays a role in the maintenance of the mitochondrial network. Similar to sgRNA and treatment with DMF (Figs. [2f](#Fig2){ref-type="fig"} and [8a](#Fig8){ref-type="fig"}), sh-RNAi-induced reduction of γ-tubulin disrupted the mitochondrial network, which became shorter and disorganized, in both *γTubulin*sh-U2OS and *γTubulin*sh-MCF10A cell lines (Fig. [8c](#Fig8){ref-type="fig"}, Supplementary Fig. [6a, b](#MOESM1){ref-type="media"}). This effect was reversed in U2OS cells by the expression of a shRNA-resistant GFP-γ-tubulin protein (Fig. [8c](#Fig8){ref-type="fig"}), but not by the expression of the A13γ-tubulin~resist~ GTPase mutant (Fig. [8c](#Fig8){ref-type="fig"}). Finally, treatment of U2OS cells with CDA also altered the mitochondrial network, which became disorganized (Supplementary Fig. [6c](#MOESM1){ref-type="media"}). These findings indicate that γ-tubulin expression and its GTPase domain are necessary for the organization of mitochondria in tubular structures.
Transport to mitochondria affects the γ-tubulin meshwork {#Sec12}
--------------------------------------------------------
Consistent with a dynamic γ-tubulin meshwork, we saw that the γ-tubulin meshwork associated with mitochondria became more distinct upon increased transport to the mitochondria triggered by the transient expression of the mitochondrial-targeting signal enclosed in mito (Fig. [2e](#Fig2){ref-type="fig"}, Supplementary Fig. [2c](#MOESM1){ref-type="media"}), suggesting that an increased protein transport to mitochondria may trigger rearrangements in the γ-tubulin meshwork. To address this possibility, we ectopically expressed mito in *γTubulin*sh-U2OS-γ-tubulin~resist~ (Fig. [9a](#Fig9){ref-type="fig"}) and *γTubulin*sh-U2OS-A^13^γ-tubulin~resist~ (Fig. [9b](#Fig9){ref-type="fig"}) cells, or mito and GFP in U2OS cells (Fig. [9c](#Fig9){ref-type="fig"}). Live imaging of *γTubulin* sh-U2OS-γ-tubulin~resist~ and GFP-expressing U2OS cells transiently expressing mito showed that in all mito-expressing cells, only GFP-γ-tubulin~resist~ was associated with mito. (Fig. [9a, c](#Fig9){ref-type="fig"}). By contrast, we found that the mito localization became dispersed throughout the cells in 30% of *γTubulin*sh-U2OS-A^13^γ-tubulin~resist~ cells, which expressed a γ-tubulin mutant with a defective GTPase domain (Fig. [9b](#Fig9){ref-type="fig"}). These data show that increased protein transport to the mitochondria induces the enrichment of GFP-γ-tubulin at mitochondria in a GTP-dependent manner.Fig. 9Increased mitochondria protein transport and low cellular mitochondria content affect the γ-tubulin meshwork. **a**--**c** Confocal fluorescent microscopy of fixed and live U2OS cells that stably expressed *γTubulin* shRNA and co-expressed the sh-resistant GFP-γ-tubulin~resist~ (**a**), the Cyst^[@CR13]^ to Ala^[@CR13]^ GFP-A^13^γ-tubulin~resist~ (**b**) or GFP (**c**) and transiently expressed pmTurquoise2-mito (mito), as indicated. Endogenous MTCO2 was stained with an anti-MTCO2 antibody (*N* = 3). **d** Average intensity projection of three Z-stack images (Z-stack) of fixed human clear cell renal carcinoma (ccRCC) and human normal kidney cells. Cells were imaged by confocal immunofluorescent staining with an anti-γ-tubulin antibody (γTubulinAb) that originated in mouse, and anti-MTCO2 antibody (*N* = 3). The yellow box shows co-localization pixel-map (CM) of the red and green (blue) channels of the magnified area displayed in the inset. White areas denote colocalized pixels between channels (ccRCC, Person's *R* = 0.3, fraction of red (MTCO2) overlapping blue (γTubulinAb) M1 = 0.9, fraction of blue overlapping red M2 = 0.8; normal kidney, Person's *R* = 0.12, M1 = 0.9, M2 = 0.7. **a**--**d** Scale bars 10 μm. **e** The GTPase domain of γ-tubulin is necessary for organizing a meshwork of mitochondria-associated strings that regulates mitochondrial respiratory capacity and gene expression and connects these organelles to the nuclear envelope. The metabolite fumarate and mitochondrial-targeted protein transport regulate the γ-tubulin meshwork. Please, see Supplementary Fig. [2](#MOESM1){ref-type="media"}
The fewer mitochondria, the more pronounced the γ-tubulin tubules {#Sec13}
-----------------------------------------------------------------
To further characterize the function of the γ-tubulin meshwork, we hypothesized that in a cell population with a need to enhance its respiratory capacity due to low mitochondrial content, for example, the γ-tubulin meshwork could be utilized as a tool to optimize mitochondrial respiration. We subsequently examined primary human clear cell renal carcinoma (ccRCC), as those cells have low mitochondrial content, and normal human kidney cells^[@CR32]^. Immunofluorescent staining revealed a clear accumulation of mitochondria in γ-tubulin enriched areas in ccRCC (Fig. [9d](#Fig9){ref-type="fig"}), indicating that low mitochondrial content affects the organization of the γ-tubulin meshwork. Together, these data demonstrate that γ-strings form a cellular infrastructure component that may provide the cell with a tool to synchronize cellular metabolism by modulating the mitochondrial respiratory capacity.
Discussion {#Sec14}
==========
Although the functions of γ-tubulin have been extensively studied over the past decades, there are still functions that remain to be unravelled. Here, we show that γ-tubulin forms γ-strings and these are enriched between and in mitochondria. The γ-string meshwork interacts with both the mtDNA, and with the inner mitochondrial membrane protein MTCO2, and also affects the replication of mtDNA and expression of mitochondria-related genes. Manipulation of the cellular protein levels of γ-tubulin, or of γ-tubulin GTP binding by either CDA/DMF treatment, or by mutating Cys^[@CR13]^ on its GTPase domain, disrupts the structure of the γ-tubulin meshwork^[@CR11]^ and affects mitochondrial respiratory capacity and mass. As the γ-tubulin protein levels drop below 40 % in *γTubulin* sgRNA expressing cells, the mitochondrial membrane potential is impaired and finally the mitochondrial integrity is lost, subsequently realising cytochrome c. In contrast, increased transport to the mitochondria, the expression of the C-terminal DNA-binding domain of γ-tubulin or disruption of γ-tubulin's NLS, relocates γ-tubulin to the mitochondrial meshwork. Moreover, the cellular mitochondrial content and the metabolite fumarate affect the γ-string meshwork. In view of these observations, we propose that the mitochondria-associated γ-tubulin meshwork provides mitochondria with a structural scaffold that regulates mitochondria function (Fig. [9e](#Fig9){ref-type="fig"}). Together, these data provide a novel mechanistic explanation on how the microtubule-independent functions of γ-tubulin affect the metabolic status of a cell.
Changes in mitochondrial network, mass and function occur in response to cellular stress. Fusion, fission and movement of mitochondria maintain the mitochondrial dynamics and morphology. In addition, formation of a mitochondrial network is important for maintaining mtDNA integrity and function^[@CR33]^. In a previous study, the protein kinesin family member 5B (KIF5B) and polymerised microtubules were shown to mediate the formation of a dynamic mitochondrial network at the cellular periphery^[@CR34]^. However, how the structure of the mitochondrial network affects the integrity of mtDNA is currently unknown. In this study, we demonstrate that γ-tubulin structurally organises the mitochondrial network and binds to mtDNA. Using the formation of mitochondrial γ-strings, γ-tubulin may link the structure of the mitochondrial network with mtDNA integrity, also potentially mediating the interactions of mitochondria with microtubules^[@CR17],\ [@CR18],\ [@CR34]^.
*TUBG1* knockout mice survive only to morula/blastocyst stages, because a redundant function is activated by expression of *TUBG2* during the first stage of embryonic development^[@CR35]^. In human neuroblastoma cell lines, both neuronal development and mitochondrial-induced oxidative stress results in upregulation of *TUBG2*, which is considered to be a pro-survival signal^[@CR13]^. Here we show that *γTubulin* sgRNA mediated depletion of γ-tubulin protein kills cell lines, whereas cell lines stably expressing *γTubulin* shRNA survive with approximately 50% of the γ-tubulin pool, demonstrating an important role of γ-tubulin in cellular homeostasis and cell survival.
γ-Tubulin depletion in *Xenopus laevis* egg extracts prevents nuclear formation, as a border of γ-strings around chromatin is necessary for the formation of the nuclear membrane^[@CR5]^. Furthermore, the C-terminal region of γ-tubulin contains the DNA-binding domain and stable expression of γ-tubulin^336-451^ co-localizes with the mitochondrial meshwork and nuclear compartment^[@CR9],\ [@CR14]^. Both the nuclear compartment and the mitochondria contain chromatin and a double membrane, so accordingly, γ-tubulin associates to both compartments. Taking into consideration that the maintenance and expression of mtDNA depends on the mitochondrial import of many nuclear-encoded proteins that control mitochondrial function^[@CR29]^, we hypothesize that the accumulation of γ-strings at the nuclear and mitochondrial membranes may connect the cytosolic and DNA-associated γ-tubulin pools. This association may establish a path for transport of proteins to the mitochondria^[@CR29]^ and for converting cytosolic signals into a gene response^[@CR5],\ [@CR14],\ [@CR36]^.
Proliferating tumour cells have a reprogrammed cell metabolism to sustain cell growth and proliferation that is driven by genetic and non-genetic alterations. In the presence of oxygen, glucose is metabolized in the cytoplasm to pyruvate and then pyruvate is oxidized in the mitochondria^[@CR37]^. Thus, cell metabolism requires the synchronous production and coordinated transport of metabolites between cell compartments that might be assisted by the γ-tubulin meshwork. Notably, in various tumours and cell lines, the localization and expression of γ-tubulin are altered^[@CR38]--[@CR40]^. Thus, the function of γ-tubulin as a regulator of the metabolic status of a cell may be one of various mechanisms that provide tumour cells with metabolic advantages that favour tumour growth.
Interactions between mitochondria and various cytoskeleton networks are known to influence mitochondrial respiration, morphology and cellular localization^[@CR18]^. Nonetheless, to our knowledge there is no description of the cytoskeletal element that shapes the mitochondria network and synchronizes cytosolic and nuclear events with mitochondrial function. We therefore propose that a network of γ-strings forms a cytosolic meshwork that organises mitochondria and provides a structural basis for the mitochondrial machinery. In this model, the degree of association of γ-strings with mitochondria creates a cellular infrastructure component that may provide cells with a tool to synchronize the cellular metabolism with cellular function (Fig. [9e](#Fig9){ref-type="fig"}).
Our results demonstrate the existence and the regulation and function of a novel cytoskeletal element, the γ-string meshwork and provide a logical explanation for the mechanism underlying the location of mitochondria in close vicinity to the nuclear envelope. Furthermore, these findings uncover a novel regulatory mechanism that controls mitochondrial homeostasis.
Methods {#Sec15}
=======
cDNA and reagents {#Sec16}
-----------------
PmTurquoise2-tagged- mitochondrial-targeting signal from COX8A (amino acids 1--29; plasmid 36208; Addgene, Cambridge, MA, USA) was provided by Dr. D. Gadella^[@CR15]^. Human pEGFP-γ-tubulin^334--449^ (γ-tubulin (334--449)), pEGFP- sh-resistant *TUBG1* gene, pEGFP-A13-γ-tubulin~resist~, and *γTubulin* shRNA were prepared as previously described^[@CR4],\ [@CR11],\ [@CR14]^. pSpCas9(BB)-2A-GFP *γTubulin* sgRNA, and pcDNA3-sg-resistant *TUBG1* gene were prepared as previously reported^[@CR8]^. The γ-tubulin fragment γ-tubulin (336-451) was prepared using a Quickchange Mutagensis Kit (Stratagene) and the primers listed in Supplementary Table [3](#MOESM1){ref-type="media"}. Finally, excision of the N-terminal region of γ-tubulin (γ-tubulin^336--451^) with *Hin*dIII was performed before re-ligation of the final construct. The Arg399-to-Ala--Lys400-to-Ala--Arg409-to-Ala substitutions in pcDNA3-sg-resistant *TUBG1* gene were prepared using a Quickchange Mutagensis Kit (Stratagene) and the primers listed in Supplementary Table [3](#MOESM1){ref-type="media"}. The mutations and constructs were verified by sequencing.
The following antibodies and reagents were used: anti-GFP (1:500), anti-GCP2 (1:500), anti-DRP1 (1:500), anti-Mfn1 (1:500), anti-OPA1 (1:500), anti-cytochrome c (1:400) and anti-calnexin (1:400, all from Santa Cruz Biotechnology, Dallas, USA); anti-γ-tubulin (1:500, mouse, T6557 recognizes the N-terminal amino acids 38 to 53 of γ-tubulin and rabbit, T3320, recognizes the C-terminal amino acids 437 to 451 of γ-tubulin, antibodies, Sigma-Aldrich, Saint Louis, USA), anti-α-tubulin (1:1000, Millipore, California, USA); anti-MTCO2, anti-HTATIP2, anti-ATP6, anti-SLC25A6, (1:400, Abcam, Cambridge, UK); anti-cox8 (1:500, Atlas antibodies, Stockholm, Sweden); anti-histone (1:400, Merck); MitoTracker Red CMXRos (Molecular Probes); citral dimethyl acetal and dimethyl fumarate (Sigma-Aldrich, Munich, Germany). All other reagents were obtained from Sigma-Aldrich.
Total lysates from cells, and Western blot analysis were prepared as reported^[@CR4],\ [@CR14],\ [@CR41]^.
Cell culture {#Sec17}
------------
Human osteosarcoma U2OS, human retinoblastoma Y79 and human mammary gland epithelia MCF10A cells were cultured as described^[@CR4],\ [@CR11]^. Stably or transient transfected *γTubulin* shRNA, pEGFP-γ-tubulin~resist~, pEGFP-A13-γ-tubulin~resist~, γ-tubulin~sgrest~, γ-tubulin^R399A-K400A-R409A^~sgrest~ and γ-tubulin^336--451^ U2OS, and MCF10A cells were obtained as described in Supplementary Table [4](#MOESM1){ref-type="media"} and elsewhere^[@CR11]^.
Primary human kidney epithelial and renal carcinoma cells were isolated after informed consent was obtained from participants. Primary cells were cultured from patient nephrectomies and subsequent analyses were performed in accordance with the ethical approval from Lund University ethical committee (LU680-08 and LU 289-07). An experienced pathologist classified tumours as ccRCCs. Normal samples were collected from healthy kidney cortex farthest from the tumour. Excised tissue was cut in pieces and incubated overnight with 300 U/ml Collagenase type I (Gibco) and with 200 U/ml DNAse I (Sigma) in full media. The following day, cells were collected and incubated for 5 min with trypsin. The remaining cell suspension was serially filtered through 40 µm and 20 µm filcons. Isolated cells were grown in DMEM high glucose supplemented with 10% FBS and 1% penicillin/streptavidin (Thermo Scientific). All cell lines were routinely tested for the presence of mycoplasma.
U2OS cells were transfected with sgRNA and examined on day seven or for the indicated period of time^[@CR9],\ [@CR11]^. Cas9-crispGFP expressing cells (GFP-expressing cells) were counted in a fluorescence microscope in each sample.
Fluorescent imaging microscopy {#Sec18}
------------------------------
U2OS cells were cultured on coverslips and fixed as described previously^[@CR9],\ [@CR11],\ [@CR42]^. Coverslips were then incubated in PBS staining buffer (PBSB; PBS, 1% Fetal Calf Serum and 0.5% BSA) to prevent non-specific antibody binding. Cells were incubated (1 h) with primary antibody (in PBSB), washed with PBSB, and incubated with Alexa488-labelled, Cy3-labelled, Dylight-labelled or Alexa647-labelled secondary antibody (Jackson). Slides were washed and mounted in Vectashield with diamidino-2-phenylindole (DAPI; Vector laboratories, Burlingame, California) or in slowFade Gold reagent (super resolution, ThermoFisher Scientific). Cells were treated with DMF or CDA for 20 h before imaging. A minimum of 100 cells was counted in each sample, and the percentage of cells was calculated. Super-resolution images were captured with an ELYRA PS.1 SIM/PALM super-resolution structured illumination (SR--SIM; Carl Zeiss) with an alpha Plan-Apochromat ×100 NA 1.46 oil immersion objective. The SR--SIM performed a multiple image acquisition procedure with varying illumination patterns and then used Zen software to reconstruct the acquired images into one super-resolved image that had double the spatial resolution in compared to a wide-field image.
Confocal and fluorescence imaging were performed using a Zeiss LSM 700 Axio Observer microscope with a Plan-Apochromat ×63 NA 1.40 oil immersion objective. Sequential images were collected at 0.2-µm or 0.34-µm intervals. All images captured with the Zeiss LSM 700 Axio Observer microscope that are presented in this article were subjected to a rolling ball background subtraction (Fiji). Co-localization analysis, 3D projections, and processing of images were carried out with ImageJ (Fiji) software. The plug-ins "just another co-localization" in Fiji and "colocalization threshold" were used to determine co-localization between two channels^[@CR43]^. In short, a cote's background subtraction was applied^[@CR44]^ before determining the Pearsons's correlation coefficient (Pearson's *R*), Manders split coefficients above cote's threshold (Manders' M1 & M2), and colocalized pixel maps.
Near simultaneous Cas9-crispGFP/DIC imaging sequences were collected and analyzed as described elsewhere^[@CR9],\ [@CR11]^. Time-lapse images were captured every 8 min.
Cell fractionation, Immunoprecipitation and Western blot analysis {#Sec19}
-----------------------------------------------------------------
Purification of the crude mitochondrial fraction from MCF10A and U2OS cells (20 × 10^6^/sample) were prepared as follows. Cells were first lysed in buffer containing 0.1% triton X-100 (BADT^[@CR9]^) and the resulting supernatant was the total cytosolic lysate. The remaining pellet containing the crude mitochondria fraction and nuclei was resuspended in 300 ml of cold BAD buffer and drawn slowly into a 21 g ½ needle and ejected with one stroke 10 times. The chromatin fraction was removed by centrifuged the nuclei at 1,700×*g* for 5 min at 4 °C. The resulting supernatant was pooled together with the total cytosolic lysate and further centrifuged at 10,000×*g* for 10 min at 4 ^o^C. The pelleted crude mitochondria fraction was resuspended in immunoprecipitation buffer (100 mM tris, pH 7.5, 300 mM NaCl, 2 mM dithiothreitol \[DTT\], 2 mM EGTA, 2 mM MgCl~2~, 1% Triton X-100, 250 mM phenylmethylsulfonyl fluoride \[PMSF\] and 100 mM Na~3~VO~4~, 0.5 mg/ml aprotinina, 0.5 mg/ml leupeptin and 0.5 mg/ml pepstatin) and further solubilized by a short sonication (3 × 5 s). The final extracts were thereafter divided into three samples and subjected to immunoprecipitation as described^[@CR9],\ [@CR14]^. Total cell lysates^[@CR41]^, and the different immunoprecipitates were analyzed by Western blotting using MTCO2 as molecular marker for inner mitochondrial membrane, as described^[@CR41]^.
For immunofluorescence staining, mitochondria were prepared as described above. The crude mitochondria fraction was resuspended in 300 µl network assembly buffer (NAB: 40 mM hepes, pH 7.2, 150 mM NaCl, 1 mM DTT, 1 mM EGTA, 1 mM MgCl~2~, 250 mM sucrose, 2 mM GTP, 250 mM phenylmethylsulfonyl fluoride (PMSF) and 100 mM Na~3~VO~4~, 0.5 mg/ml aprotinina, 0.5 mg/ml leupeptin and 0.5 mg/ml pepstatin) and incubated for 1 h at 22 °C. Network assembly was ended by addition of 3% formaldehyde. Immunostaining of assembled mitochondrial network were performed by first air-drying 10 µl of the network on a Superfrost Plus glass slide (Thermo Scientific, USA) and thereafter were permeabilized for 3 min with methanol/acetone (1:1; v/v) at -80 °C before immunofluorescence stained as described elsewhere^[@CR4]^.
Purification of the mitochondrial fraction from MCF10A cells (20 × 10^6^/sample) was prepared as follows. To remove possible cytoskeletal elements attached to mitochondria, cells were pre-incubated for 15 min with culture medium containing 100 ng/ml colcemid and 5 µg/ml cytochalasin B (37 °C, 5% CO~2~)^[@CR4],\ [@CR5],\ [@CR8]^. Thereafter, purification of the mitochondrial fraction by a Percoll density gradient centrifugation was prepared as described elsewhere^[@CR45],\ [@CR46]^, with the following modifications. Cells were lysed on ice in a dounce homogenizern in 1 ml cold homogenisation buffer (HB: 5 mM hepes, pH 7.2, 210 mM mannitol, 70 mM sucrose, phenylmethylsulfonyl fluoride (PMSF) and 100 mM Na~3~VO~4~, 0.5 mg/ml aprotinina, 0.5 mg/ml leupeptin and 0.5 mg/ml pepstatin). The cell debris, unbroken cells, and nuclei were removed by centrifugation (1300×*g* for 5 min at 4 °C). The nuclear membrane fraction was obtained by resuspending and preparing lysates of the first pellet as described elsewhere^[@CR14]^. The obtained nuclear membrane fraction was lysed in sample buffer (SB)^[@CR41]^. The total cytosolic homogenate was further centrifuged at 10,000×*g* for 10 min at 4 °C and the pellet containing the crude mitochondrial fraction was resuspended in HB supplemented with 1 mM EGTA (HBE) and layered on preformed gradient consisting of 2 ml 30% Percoll layered over 0.8 ml 50% Percoll in HBE buffer in a 4 ml centrifuge tube (Supplementary Fig. [3c](#MOESM1){ref-type="media"}). The gradient was subjected to ultracentrifugation for 15 min at 95,000×*g* in a swinging-bucket rotor. Mitochondrial and endoplasmic reticulum fractions were carefully collected with a needle and diluted with HBE buffer (1:5) followed by centrifugation at 17,000×*g* for 10 min. The final pellets were resuspended in SB. The microsome fraction was pelleted by centrifugation of the remaining total cytosolic homogenate at 100,000×*g* for 15 min. Both the final microsome pellets and the cytosolic supernatant were dissolved in SB. The purified fractions were boiled and analyzed by Western blotting.
Uncropped images of all Western blots presented in the study are shown in Supplementary Figs. [7](#MOESM1){ref-type="media"}--[10](#MOESM1){ref-type="media"}.
Electron microscopy {#Sec20}
-------------------
For high-pressure freezing, U2OS cells were seeded to 100% confluence onto carbon coated (10 nm) 6 mm sapphire discs (Leica) in 12-well dishes (Nunc). Cells were cryo-preserved with high-pressure freezing (HPM100, Leica) followed by freeze substitution (Leica AFS2, Leica) for 48 h at −90 degrees in Acetone with 0.1% Uranyl Acetate and embedded in Lowicryl with polymerization at -25 degrees for 48 h. The 60 nm sections were cut with a Leica Ultracut UC7 (Leica, Wienna, Austria) and collected on one whole formvar coated carbon grids and 200 mesh Nickel grids. The sections were pre-incubated for 30 min with pre-incubation buffer (50 mM glysine, 0.1% sodium borhydride NaBH~4~, 0.05 M Tris pH 7.4, 0.1% Triton), before incubation with polyclonal anti-γ-tubulin antibody (T3320) in TBST (0,05 M Tris pH 7.4, 0.1% Triton, 1% BSA) for 2 h at room temperature, followed by 1 h incubation with gold conjugated protein A (1:100, 10 nm gold; Agar Scientific, Essex, UK) in TBST. Final staining was performed with filtered 0.5% uranyl acetate for 10 min.
For preparation of the crude mitochondrial fraction, mitochondria were prepared as described above but in the absence of triton X-100. Pellets of mitochondria were embedded in low-melting point agarose and fixed 2% v/v glutaraldehyde in 0.05 M sodium phosphate buffer (pH 7.4). Samples were rinsed three times in 0.15 M sodium cacodylate buffer (pH 7.4) and subsequently post-fixed in 1% w/v osmium tetroxide and 0.05 M potassium ferricyanide in 0.12 M sodium cacodylate buffer (pH 7.4) for 2 h. The specimens were dehydrated in graded series of ethanol, transferred to propylene oxide and embedded in Epon according to manufacturer's instructions. Sections were prepared as above and stained with uranyl acetate and lead citrate. High-pressure frozen and mitochondrial samples were examined with a Philips CM 100 TEM (Philips, Eindhoven, The Netherlands), operated at an accelerating voltage of 80 kV. Digital images were recorded with an OSIS Veleta digital slow scan 2k × 2k CCD camera and the ITEM software package.
Chromatin immunoprecipitation and expression analyses {#Sec21}
-----------------------------------------------------
Chromatin immunoprecipitation (ChIP) was described elsewhere^[@CR14]^. ChIPs were performed using rabbit polyclonal antibodies: a mixture (1:1) of anti-γ-tubulin T3320 and affinity purified anti-γ-tubulin T5192 (Sigma). Coprecipitated chromatin from U2OS cells was analyzed by PCR for the presence of mitochondrial DNA between the following base pairs: 1880--2186, 2423--2640, 10,654--10,854 and 15,102--15,370, with the oligos listed in Supplementary Table [3](#MOESM1){ref-type="media"}. Uncropped images of ChIP analysis presented in the Fig. [4b](#Fig4){ref-type="fig"} are shown in Supplementary Figures [8](#MOESM1){ref-type="media"}.
Alternatively, coprecipitated chromatin from S-phase synchronized MCF10A and MCF10A cells stably expressing γTubulin shRNA were sequenced using the Proton system (ThermoFisher) according to the manufactures. In brief, a DNA sample was fragmented using the S2 system from Covaris. End repair and adaptor ligation were performed by the AB Library Builder System (ThermoFisher). Samples were amplified according to the Ion Xpress™ Plus and Ion Plus Library Preparation for the AB Library Builder™*System* protocol and size selected with a target range of 220-310 bp (Blue Pippin™, Sage Science). Library size and concentration were assessed by a Bioanalyzer High Sensitivity Chip (Agilent Technologies) and by the Fragment Analyzer system (Advanced Analytical). Samples were pooled together in sets of two, followed by template preparation on the Ion Chef™ System using the Ion PI Hi-Q Chef Kit (ThermoFisher). Samples were then loaded on Ion PI™ v3 chips and sequenced on the Ion Proton™ System using the Ion PI™ Hi-Q Sequencing 200 Kit chemistry (200 bp read length, ThermoFisher).
Total RNA from MCF10A and *γTubulin*sh-MCF10A cells was prepared as previous described^[@CR4],\ [@CR14]^. Thereafter, 50 ng of total RNA was reverse transcribed according to Ion AmpliSeq™ Transcriptome Human Gene Expression Kit Preparation protocol (ThermoFisher). The cDNA was amplified using Ion AmpliSeq™ Transcriptome Human Gene Expression core panel (ThermoFisher) and the primer sequences were then partially digested. Adaptors (Ion P1 Adapter and Ion Xpress™ Barcode Adapter, Life Technologies) were then ligated to the amplicons. Adaptor ligated amplicons were purified using Agencourt® AMPure® XP reagent (Beckman Coulter) and eluted in amplification mix (Platinum® PCR SuperMix High Fidelity and Library Amplification Primer Mix, ThermoFisher) and amplified. Size-selection and purification was conducted using Agencourt® AMPure® XP reagent (Beckman Coulter). The amplicons were quantified using the 2100 Bioanalyzer® (Agilent) with High Sensitivity DNA Kit (Agilent.) Samples were then pooled, six per pool, followed by emulsion PCR on the Ion OneTouch™ 2 System using the Ion PI™ Hi-Q™ OT2 Kit (ThermoFisher). The pooled samples were loaded on Ion PI™ v3 chips and sequenced on the Ion Proton™ System using the Ion PI™ Hi-Q Sequencing 200 Kit chemistry (200 bp read length, ThermoFisher).
RNA-seq and ChIP-seq analysis {#Sec22}
-----------------------------
RNA-seq reads were mapped to the reference genome (including the mitochondria genome) using TopHat^[@CR47]^ (version 2.0.9, parameters: read mismatches: 2, -read gap length: 2, -max insertion length: 3, -max deletion length: 3). The python package HTSeq^[@CR48]^ (version: 0.6.0) was used to produce the count table (a table counted number of reads mapped to each gene). We calculated the RPKM value for each gene based on count table using locally developed Perl code. Differential gene expression analysis was performed using edgeR^[@CR49]^. An FDR cutoff of 0.05 was used to select significantly differentially expressed genes. ChIP-seq reads were mapped to the reference genome Bowtie^[@CR50]^ (version 1.0.0, parameters: -m: 1). The mitochondria genome was downloaded from Ensembl (ftp://ftp.ensembl.org/pub/release-75/fasta/homo_sapiens/dna/) and included in the reference genome for mapping. Differential peak analysis was performed using the bggdiff tool as part of MACS2 ChIP-seq analysis package^[@CR20]^ (version: 2.1.1, parameter: -g: 100, -cutoff: 3, -min len: 200, -depth1: 1, -depth2: 1).
Gene set enrichment analysis (GSEA) was performed using software obtained at <http://software.broadinstitute.org/gsea/index.jsp>, using 58 mitochondria-related gene sets provided in MSigDB (<http://software.broadinstitute.org/gsea/msigdb/index.jsp>) and a customized gene set constructed using MitoCarta^[@CR23]^.
Measurement of mitochondrial respiration and membrane potential (∆ψ~m~) {#Sec23}
-----------------------------------------------------------------------
Measurement of oxygen consumption rate per cell was performed in a Seahorse XF24 Extracellular Flux Analyzer (SeaHorse Bioscience, North Billerica, MA, USA) according to the manufacturer's instructions. In brief, before measurement the medium was changed to non-buffered XF Assay medium and cells were treated with 100 µM CDA or DMF for 2 h in a non-CO~2~ chamber after which the oxygen consumption rate was analyzed. Oxygen consumption rate was determined following injection of 4 mg/ml oligomycin (ATP synthase inhibition), 1 mM FCCP (uncoupler of mitochondrial oxidative phosphorylation; Abcam, Cambridge, UK) and 0.5 mM Rotenone (interferes with the electron transport chain in mitochondria by inhibiting the electron transferring activity of Complex I) according to instructions for the Cell Mito Stress Test (SeaHorse Bioscience). Each sample was run in four replicates.
For ∆ψ~m~ measurements, cells were pre-loaded with low K buffer (135 mM NaCl, 3.6 mM KCl, 1.5 mM CaCl~2~, 0.5 mM MgSO~4~, 0.5 mM Na~2~HPO~4~, 10 mM HEPES, 5 mM NaHCO~3~, pH 7.4) supplemented with 2.5 µM cyclosporin A (prevents permeability transition pore opening and dye leakage from mitochondria), 2.8 mM glucose and 100 nM tetramethylrhodamine methyl ester (TMRM; Invitrogen) for 2 h. After washing the cells once with low K buffer, cells were resuspended in low K buffer supplemented with 2.8 mM glucose. TMRM fluorescence measurements were performed using 543-nm excitation, 585-nm long pass emission filter settings on a Zeiss LSM510 inverted confocal fluorescence microscope. ∆ψ~m~ measurements were performed in quench mode^[@CR51]^.
Cellular thermal shift assay {#Sec24}
----------------------------
To demonstrate that CDA and DMF directly bind to γ-tubulin we monitor the effect of the drugs on the thermodynamic stabilization of γ-tubulin upon ligand binding, as described elsewhere^[@CR52]^. In brief, Y79 were treated with 100 µM CDA, 150 µM DMF or vehicle (DMSO) for 1 h. After removing the supplemented culture media, cells were resuspended in PBS supplemented with protease inhibitors (2 × 10^6^cells/ml) and heat-treated as described^[@CR52]^. Cell lysates were prepared by freeze-thaw the cells three times. The resulting cell lysates were analyzed by Western blotting using an anti-γ-tubulin and an anti-α-tubulin antibody^[@CR41]^.
Proliferation assay and cell cycle analysis {#Sec25}
-------------------------------------------
Metabolic activity was analyzed with WST-1 cell proliferation assay (Roche) according to the manufacturer's instructions. In short, 8 h before measurement, 100 μl of resuspended U2OS cells or U2OS cells transiently expressing human *γTubulin* shRNA or resuspended U2OS cells supplemented with either 100 µM CDA or various concentrations of DMF were plated at a concentration of 180 cells/μl in 96-well plates. 4 h before measurement, 10 μl of WST-1 reagent was added into each well, and the absorbance was measured at 450 nm using Fluostar omega Microplate Reader (BMG labtech).
For cell synchronization, cells were arrested at early S-phase by presynchronization with thymidine as previously described^[@CR9]^ and released for different periods. Cells were ethanol-fixed and cell cycle progression was analyzed in a Nucleocounter^®^ NC-3000™ by measuring cell DNA content with diamidino-2-phenylindole (DAPI) as described by the manufacturer (ChemoMetec, Denmark). Cell cycle profiles were analyzed with FlowJo (Tree Star, Inc.).
Statistical analysis {#Sec26}
--------------------
All data are expressed as ±SEM or SD, as indicated, and statistical significance of the differences between two groups or more was analyzed by paired Student's *t*-test unless otherwise indicated (\**P* \< 0.05, \*\**P* \< 0.01, \*\*\**P* \< 0.001, \*\*\*\**P* \< 0.0001). Western blotting bands were quantitated with ImageJ software.
Availability of data and material {#Sec27}
---------------------------------
Raw sequence data and processed files are publicly available at NCBI Gene Expression Omnibus (GEO) (<http://www.ncbi.nlm.nih.gov/geo>) under the following accession numbers: GSM2884568, GSM2884569, GSM2884570, GSM2884571, GSM2884576, GSM2884577, GSM2884578, GSM2884579, GSM2884584, GSM2884585, GSM2884586, GSM2884587, GSM2884588, GSM2884589, GSM2884590, and GSM2884591. All other data not present in the manuscript or supporting materials are available from the corresponding author upon request.
Electronic supplementary material
=================================
{#Sec28}
Supplementary Information Description of Additional Supplementary Files Supplementary Data 1
**Publisher\'s note:** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Electronic supplementary material
=================================
**Supplementary Information** accompanies this paper at 10.1038/s42003-018-0037-3.
We thank Dr. Gadella for reagents, and the Core Facility for Integrated Microscopy, Faculty of Health and Medical Sciences, University of Copenhagen for support with electron microscope, the Centre for cellular imaging at the Sahlgrenska Academy, University of Gothenburg for support with 3D super-resolution structured illumination microscope, the National Genomics Infrastructure (NGI)/Uppsala Genome Center and UPPMAX for providing assistance in massive parallel sequencing and computational infrastructure (funded by RFI/VR and Science for Life Laboratory, Sweden), and Elevate Scientific and Benjamin Duell for editorial assistance. This work was supported by the Skane University Hospital in Malmö Cancer Research Fund (20151209), the Swedish Cancer Society (CAN 2016/3669), the Swedish Childhood Cancer Fund (PR2016-0084), Crafoordska foundation (20170530) and Novo Nordisk foundation (12759).
M.A.K. involved in study design and wrote the paper; T.L. analyzed ChIP-seq and RNA-seq data; L.L., D.M., A.K., H.N., N.V., H.M., M.J., C.A.R., and M.A.K. experiments and analyzed the data.
Competing interests {#FPar1}
===================
The authors declare no competing interests.
| {
"pile_set_name": "PubMed Central"
} |
Background {#Sec1}
==========
Livestock farming is considered a major contributor to anthropogenic methane (CH~4~) emissions, which is mainly attributed to ruminants \[[@CR1]\]. Methane production is also energetically wasteful for ruminants resulting in a loss of 2--12% of the ingested feed energy \[[@CR2]\]. Several dietary strategies based on additives and supplements have been proposed to mitigate rumen methanogenesis but only few of them have shown persistent effect *in vivo* without negative impacts to the host animal and the environment \[[@CR3], [@CR4]\]. Use of direct-fed microbials (DFM) is one possible option that could be sustainable and easily acceptable by both consumers and producers \[[@CR5]\].
Direct-fed microbials are used in the dairy sector to improve animal productivity and health \[[@CR6], [@CR7]\]. *Propionibacterium* and *Lactobacillus* spp. alone or in combination are the most common bacterial DFM used in ruminant production \[[@CR7]\]. A metabolic aspect that characterizes these bacterial species is the production of propionate, which is a H~2~-consuming reaction \[[@CR8]\]. Promoting this pathway is expected to produce less H~2~ and consequently less CH~4~ in the rumen. However, *in vivo* studies using *Propionibacterium* and/or *Lactobacillus* spp. as modulators of enteric CH~4~ production showed contrasting results with decreases, no effect or even increases in CH~4~ emissions \[[@CR4], [@CR9]--[@CR11]\]. These differences could be originated from several factors such as type of ruminant, physiological stage, and diet, but also due to differences in the strains of DFM used.
The bacterial DFM used in this study: *Propionibacterium freudenreichii* 53-W, *Lactobacillus pentosus* D31 and *Lactobacillus bulgaricus* D1 were previously selected for their CH~4~-decreasing effect *in vitro* \[[@CR9]\]. They were also tested in adult wethers fed a hay-based diet (70% natural grassland hay and 30% concentrate) at maintenance with contrasting results \[[@CR9]\]. Whereas *L. pentosus* reduced CH~4~ emissions (g/kg DMI), no effect was observed for *L. bulgaricus* and *P. freudenreichii* increased CH~4~ emissions (g/kg DMI). The efficacy of DFM may differ depending on the animal species, physiological stage and diet \[[@CR4], [@CR10]--[@CR12]\]. The objective of this study was to examine the potential of three selected bacterial DFM to modulate ruminal fermentation in lactating primiparous cows. The effect on milk production and composition, more particularly fatty acid (FA) composition, was also monitored. As efficacy of bacterial DFM has been shown to be affected by diet a high-starch diet (HSD) and a high-fiber diet (HFD) were used in this study.
Methods {#Sec2}
=======
This study was conducted using the animal facilities at the French National Institute for Agricultural Research (INRA) in Theix. Procedures on animals used in this study complied with the guidelines for animal research of the French Ministry of Agriculture and all other applicable National and European guidelines and regulations.
Animals, experimental design, and diets {#Sec3}
---------------------------------------
Eight lactating primiparous Holstein cows (age of 2.9 ± 0.4 years, mean ± SD) were housed in individual stalls during the study. The cows were randomly allocated into two balanced groups of four animals and fed two different basal diets: one based on corn silage, hereafter called high-starch diet (HSD), and the second based on grass silage, hereafter called high-fiber diet (HFD; Table [1](#Tab1){ref-type="table"}). At the start of the study, average daily milk production was 22.8 ± 4.9 and 22.6 ± 1.1 kg/cow, days in milk 83.2 ± 11.3 and 91 ± 15.6 days, and body weight 587.5 ± 51.1 and 585.7 ± 32.3 kg for cows fed HSD and HFD, respectively.Table 1Ingredients and chemical composition of the high-starch and high-fiber control diets used in this studyItemsControl diets^a^High-starch dietHigh-fiber dietIngredients, % of DM Corn silage44.0\_^b^ Grass silage_55.0 Hay11.0\_ Grain mix^c^34.2\_ Citrus pulp_12.0 Dehydrated beet pulp_20.0 Molasses, beet_5.0 Soybean meal8.78.0 Urea1.0\_ Cane molasses1.1_Chemical composition, % of DM OM92.285.1 CP12.512.2 NDF35.448.4 ADF19.529.3 Starch27.41.8 Ether extract2.32.3Fatty acids (FA), g/100 g of total FA 12:00.170.34 14:00.350.89 16:020.721.7 *cis*-9 16:10.722.03 18:02.572.13 *cis*-9 18:119.49.8 18:2*n*-644.928.5 18:3*n*-37.328.0 GE, MJ/kg DM16.816.9^a^Each cow was fed 250 g mineral mix comprising (g/kg): P, 2.5; Ca, 20; Mg, 4.5; Na, 3.5 (Galaphos Midi Duo GR, CCPA, Aurillac, France)^b^Ingredients not included^c^Composition: barley (14.1% of DM), wheat (10.9% of DM) and corn (9.2% of DM)
Cows in each group were randomly assigned to four treatments in a 4 × 4 Latin square design that were run in parallel. The treatments were 1) Control without DFM (CTL), 2) *Propionibacterium freudenreichii* 53-W (2.9 × 10^10^ colony forming units (CFU)/cow per day), 3) *Lactobacillus pentosus* D31 (3.6 × 10^11^ CFU/cow per day) and 4) *Lactobacillus bulgaricus* D1 (4.6 × 10^10^ CFU/cow per day). The dose of each DFM (CFU/mL rumen fluid) was chosen considering cost of production and the results from an earlier study with the same DFM preparations administered to sheep fed a hay-based diet \[[@CR9]\]. *Propionibacterium freudenreichii* 53-W (DSM 20271) was obtained from DSMZ (Deutsche Sammlung von Mikroorganismen und Zellkulturen GmbH, Braunschweig, Germany) and both *Lactobacillus* species were obtained from Danone culture collection (Danone Research, Palaiseau, France). The DFM preparations used in this study were obtained from Danone Research (Palaiseau, France) in a frozen pellet form. Their viability was checked prior to the study. Weighed pellets were thawed in 0.1% sterile peptone solution, serially diluted and inoculated onto agar plates (DSMZ medium 91 for *P. freudenreichii* and MRS medium for both *Lactobacillus* species). Plates were incubated at 39 °C for 48 h before colony counts. Results were in agreement with the quantity of CFU stated by the manufacturer.
Diets were formulated at the beginning of the study to meet the energy and protein requirements for maintenance and lactation of dairy cows based on INRA nutritional recommendation for ruminants \[[@CR13]\]. Diets were free from antibiotics, chemical buffer and yeast to avoid potential interfering effect with the effect of bacterial DFM tested in this study. Two weeks before starting the study, cows in both groups were fed CTL diet ad libitum. Then, throughout the study, feeds were restricted to 90% of their ad libitum intake to ensure complete consumption of the diet. Each experimental period (5 weeks) consisted of 4 weeks of treatment and 1 week of washout, without DFM supplementation. Cows were fed twice daily with 60% of the daily ration at 07:00 h and 40% at 16:00 h. During the treatment period, DFM preparations were administered during the morning feeding. Each day, just before feed distribution, the appropriate amount of pellets were thawed in 50 mL of 0.1% sterile peptone solution at room temperature. To ensure the entire DFM consumption, the 50-mL doses were mixed with a small portion of silage (about 500 g sampled from their diet) and offered before feeding. The amounts of feed offered and refused were weighed daily to estimate DMI. Cows were allowed continuous access to water and water intake was measured for each cow. The body weight of each animal was recorded at the end of each experimental period, 3 h after morning feeding.
Measurements and analyses {#Sec4}
-------------------------
### Feed analysis {#Sec5}
The dry matter content of each feed ingredient was determined (103 °C for 24 h, ISO 6496 \[[@CR14]\]) weekly for hay and concentrates and twice per week for silages throughout the experimental period. During the last week of each experimental period (week 4), silage, hay and concentrates were sampled (about 100 g) daily and were pooled at the end of the week. Samples of silage were stored at − 20 °C and samples of hay and concentrates were stored at 4 °C. At the end of the study, all feed samples were dried in an oven and ground (1-mm screen) before chemical analyses (InVivo Labs, Saint Nolff, France). Organic matter was determined by ashing samples at 550 °C for 6 h (method 942.05; \[[@CR15]\]). Fiber (NDF and ADF) was determined by sequential procedures \[[@CR16]\] after pre-treatment with amylase and expressed exclusive of residual ash. Total N was analyzed by combustion according to the Dumas method (method 968.06; \[[@CR15]\]) and CP content was calculated as N × 6.25. Ether extract was determined after acid hydrolysis (method 954.02; \[[@CR15]\]). Starch content was analyzed using an enzymatic method \[[@CR17]\]. Briefly, samples are incubated in a shaking water bath with pancreatic α-amylase and amyloglucosidase for 16 h at 37 °C, during which starch is hydrolyzed to *D*-glucose by the combined action of the enzymes. Then, the *D*-glucose is measured with glucose oxidase/peroxidase reagent. The gross energy (GE) was analyzed by isoperibolic calorimetry (C200 model; IKA, Staufen, Germany).
### Enteric methane {#Sec6}
In the last week of the experimental period (week 4, days 2--4) enteric CH~4~ emission was determined using individual open circuit respiration chambers (1 cow/chamber) for 3 consecutive days as described in Guyader et al. \[[@CR18]\]. Cows were allocated to the same chamber so that the DFM effect was not confounded with the chamber effect. Air leaks from the chambers were examined before the start of the experiment using water-based smoke machines (Kool Light-FOGGER 1500E; EPICAP, Saint-Symphorien d'Ozon, France). The chambers operated at a slightly negative pressure, with an air flow averaging 743.6 ± 19.61, 792.1 ± 17.89, 771.7 ± 14.40 and 756.6 ± 18.43 m^3^/h for periods 1, 2, 3 and 4 respectively. Continuous air sampling was performed in each chamber at a 0.1-Hz sample frequency for 5 min every 25 min and analyzed for CH~4~ gas concentrations with an infrared gas analyzer (Ultramat 6, Siemens, Karlsruhe, Germany). The chambers were opened twice daily at 07:00 h and 15:00 h for about 20 min for milking and subsequent feeding. The gas analyser was calibrated at the start of every CH~4~ measurement period with pure N~2~ and a certified standard gas mixture of CO~2~ (1.36 g/m^3^) and CH~4~ (0.459 g/m^3^). Real time gas emissions in a chamber were calculated by the difference between chamber and ambient gas concentrations multiplied by the airflow corrected for temperature, relative humidity, and pressure according to the Wexler equation \[[@CR19]\]. Calculations of CH~4~ yield (g CH~4~/kg DMI) and intensity (g CH~4~/kg milk) were done using data on DMI and milk production when cows were in chambers.
### Ruminal fermentation and microbes {#Sec7}
In the last week of the experimental period (week 4) rumen samples (approximately 500 mL) were collected 3 h after the morning feeding for two non-consecutive days (day 1 and 5) using a stomach tube fitted with a vacuum pump. The samples were subjected to visual examination to ensure that they were not contaminated with saliva. Values of pH were also used as an additional control. Samples suspected to be contaminated were removed, and fresh samples were taken.
The pH of each sample was recorded immediately with a portable pH-meter (CG840, electrode Ag/AgCl, Schott Gerate, Hofhein, Germany). One aliquot of rumen contents (about 200 mL) was strained through a polyester monofilament fabric (mesh size 250 μm) and the filtrate was sampled for analysis of VFA, ammonia-N (NH~3~-N), and protozoa counts. Samples for VFA were prepared by transferring 0.8 mL filtrate into a micro-tube containing 0.5 mL of a crotonic-metaphosphoric acid solution (crotonic acid 0.4% *wt*/*vol*, metaphosphoric acid 2% *wt*/*vol*, in HCl 0.5 mol/L) and stored at − 20 °C until analysis. For NH~3~-N, 1 mL of rumen filtrate was mixed with 0.1 mL of 5% H~3~PO~4~ and stored at − 20 °C until analysis. For protozoa counts, 2 mL of the rumen filtrate was mixed with 2 mL of methyl-green-formalin and saline solution (MFS) and preserved from light until counting. For quantitative microbial analysis, another aliquot (about 200 mL) of rumen contents was frozen immediately at − 80 °C and subsequently lyophilized. Lyophilized samples were then ground and stored at − 80 °C until DNA was extracted. For each sampling time, unfiltered rumen contents were dried at 103 °C for 24 h for DM determination.
Volatile fatty acid concentrations were determined by gas chromatography \[[@CR20]\] on a Perkin-Elmer Clarus 580 GC (Perkin Elmer, Courtaboeuf, France) equipped with a column Stabilwax -- DA (30 m × 0.53 mm i.d.) and using crotonic acid as the internal standard. The concentration of NH~3~-N in rumen fluid was determined using the Berthelot reaction \[[@CR21]\]. Rumen fluid/MFS solutions were diluted in an equal volume of phosphate buffer saline solution (PBS) and protozoa were enumerated in a Neubaeur chamber \[[@CR22]\].
Total genomic DNA was extracted from ground lyophilized rumen samples using a bead beating and column purification (QIAamp DNA stool mini kit, Qiagen, Valencia, CA) method \[[@CR23]\]. The yield and purity of the extracted DNA was determined using a NanoDrop spectrophotometer (Thermo Scientific, Wilmington, DE) and stored at − 20 °C. The primers used in this study are listed in Additional file [1](#MOESM1){ref-type="media"}: Table S1.
Quantitative real-time PCR assays were performed on a StepOne™ system (Applied Biosystems, Courtabeuf, France) using SYBR Ex Taq™ pre mixture (Takara Bio Inc., Otsu, Japan). Amplification of 16S rRNA genes of *P. freudenreichii* and *L. bulgaricus*, and intergenic spacer regions (16S--23S) of *L. pentosus* were performed as described in Jeyanathan et al. \[[@CR9]\]. Quantification of bacterial 16S rRNA and methanogenic *mcrA* genes were performed as previously described \[[@CR24]\].
### Milk production and composition {#Sec8}
Cows were milked twice daily at 07:00 h and 15:00 h, and milk production of individual animals was recorded electronically throughout the study except for the last week of treatment period (week 4) when cows were in chambers. In week 4, milking and weighing were done manually. Samples of milk for the measurement of fat, protein, and lactose were collected individually once per week and treated with preservative (bronopol-B2; Trillaud, Surgeres, France). Samples of unpreserved milk were also collected at each milking over 2 non-consecutive days (Tuesday and Thursday) of week 4 of the experimental period and stored at − 20 °C until analysis for FA composition.
Milk fat and protein contents were determined by mid-infrared spectrophotometry using a Milkoscan 4000 (Foss Electric, Hillerød, Denmark). The FA of the lyophilized milk samples were methylated and analyzed as before \[[@CR25]\] with some modifications: 2 mL of 0.5 mol/L sodium methanolate and 1 mL hexane were mixed with the lyophilised milk at 50 °C for 15 min, followed by the addition of 1 mL 12 mol/L HCl 5% in methanol (*v*/*v*) at 50 °C for 15 min. The fatty acid methyl esters (FAME) were washed with a saturated K~2~CO~3~ solution and recovered with 1.5 mL hexane. The FAME were injected (0.6 μL) by auto-sampler into a gas chromatograph equipped with a flame ionisation detector (Agilent Technologies 7890A, Wilmington, USA) and separated on a 100 m × 0.25 mm i.d. fused-silica capillary column (CP-Sil 88, Chrompack, Middelburg, The Netherlands). A reference standard butter (CRM 164, Commission of the European Communities, Community Bureau of Reference, Brussels, Belgium) was used to estimate correction factors for short-chain FA (C4:0 to C10:0). Identification of FAME was accomplished by comparison to a standard mixture purchased from Nu-Chek-Prep, Inc. (Elysian, MN 56028 USA). Mixtures of *cis*/*trans* (9--12) isomers of linoleic acid methyl ester and *cis* and *trans* (9--11) and (10--12) isomers of CLA methyl esters purchased from Sigma-Aldrich Corporation (38297 Saint Quentin Fallavier, France) were used for complete identification.
Statistical analysis {#Sec9}
--------------------
Data were averaged per period and per animal and analyzed using the Mixed procedure of SAS version 9.4 (SAS Institute, 2004). Data from HSD and HFD were analyzed separately as comparison between diets was not the objective of the study. The following model was used: Y~*ijk*~ = μ + T~*i*~ + P~*j*~ + C~*k*~ + e~*ijk*~*, w*here: Y~*ijk*~ are observations for dependent variables; μ is the overall mean; T~*i*~ is the fixed effect of DFM (control, *Propionibacterium*, *L. pentosus* and *L. bulgaricus*); P~*j*~, is the fixed effect of period (*j* = 1--4); C~*k*~ is the random effect of cow; and e~*ijk*~ is the random residual error. The effect of individual DFM supplementation was tested using Dunnett's test, whereas orthogonal contrasts were performed to evaluate the effect of CTL versus all DFM treatments. Data were considered significant at *P* \< 0.05, and trends were discussed at 0.05 \< *P* ≤ 0.10.
Results and discussion {#Sec10}
======================
In this study, we tested the effects of bacterial DFM on enteric CH~4~, ruminal fermentation parameters, milk production and composition and the quantity of ruminal microbes in lactating primiparous dairy cows fed two contrasting diets differing in starch and fiber contents. Differences induced by diets (shown in Tables [2](#Tab2){ref-type="table"}, [3](#Tab3){ref-type="table"} and supplementary Tables) were as expected for diets of similar composition \[[@CR10]\] and are not further discussed as they were not the aim of the study. Additionally, the effects of these type of diets on ruminal fermentation and production are well documented \[[@CR10]\].Table 2Enteric methane (CH~4~) emissions of lactating cows fed high-starch or high-fiber diets (CTL) supplemented with bacterial direct-fed microbials (DFM) *Propionibacterium freudenreichii* 53 W (PF), *Lactobacillus pentosus* D31 (LP), and *Lactobacillus bulgaricus* D1 (LB)CH~4~ emissionsTreatmentSEM^a^*P*-valueCTLPFLPLBTreatmentCTL vs DFM^b^CH~4~, g/d High-starch diet286.4327.8303.9271.421.120.330.56 High-fiber diet290.8310.0301.4292.39.790.510.38CH~4~, g/kg DMI High-starch diet20.022.821.318.72.190.620.72 High-fiber diet23.924.824.024.01.160.930.81CH~4~, g/kg milk High-starch diet13.116.7^\*^14.612.60.780.020.12 High-fiber diet18.918.718.219.01.530.980.87CH~4~, g/kg ECM^c^ High-starch diet12.615.614.412.51.020.150.22 High-fiber diet18.117.417.018.91.110.640.80^a^SEM-standard error of the means^b^*P*-value for control vs all direct-fed microbials (DFM) within each diet^c^ECM-energy corrected milk \[(0.327 × kg of milk) + (12.95 × kg of fat) + (7.65 × kg of protein)\]^\*^Significantly (*P* ≤ 0.05) different from CTL groupTable 3Intake, milk production, milk composition and body weight (BW) gain of lactating cows fed high-starch or high-fiber diets (CTL) supplemented with bacterial direct-fed microbials (DFM) *Propionibacterium freudenreichii* 53 W (PF)*, Lactobacillus pentosus* D31 (LP), and *Lactobacillus bulgaricus* D1(LB)ItemsTreatmentSEM^a^*P*-valueCTLPFLPLBTreatmentCTL vs DFM^b^Dry matter intake, kg/d High-starch diet14.314.514.514.60.690.990.80 High-fiber diet12.212.512.612.30.720.970.74Water intake, L/d High-starch diet53.162.962.059.25.780.650.25 High-fiber diet56.550.747.551.63.250.330.11Milk, kg/d High-starch diet22.119.720.921.60.790.230.18 High-fiber diet16.316.917.015.61.400.890.90Fat, g/kg milk High-starch diet37.040.136.136.12.570.660.88 High-fiber diet38.741.441.137.72.850.750.68Protein, g/kg milk High-starch diet29.730.630.129.92.100.990.83 High-fiber diet27.127.827.926.31.380.820.91ECM^c^, kg/d High-starch diet22.721.221.322.10.700.410.17 High-fiber diet16.417.918.015.81.160.470.55Efficiency^d^, High-starch diet1.601.461.481.520.080.650.27 High-fiber diet1.351.451.421.280.050.200.62Body weight gain, kg High-starch diet−6.011.7^¶^13.0^¶^5.74.980.080.02 High-fiber diet−3.04.20.5−3.27.850.890.71^a^SEM-standard error of the means^b^*P*-value for control vs all direct-fed microbials (DFM) within each diet^c^ECM-energy corrected milk \[(0.327 × kg of milk) + (12.95 × kg of fat) + (7.65 × kg of protein)\]^d^Efficiency = ECM/Dry matter intake^¶^0.05 \< *P* ≤ 0.10 from CTL group
Enteric methane and ruminal fermentation {#Sec11}
----------------------------------------
Cows supplemented with *Propionibacterium* numerically emitted more CH~4~ than CTL particularly with HSD (Table [2](#Tab2){ref-type="table"}). When calculated as CH~4~ intensity expressed in g/kg milk, *Propionibacterium* increased emission by 27% (*P* \< 0.05). Supplementation of *L. bulgaricus* or *L. pentosus* did not affect daily CH~4~ emission (g/d), yield or intensity (*P* \> 0.05). Concentrations of total VFA and NH~3~-N, and VFA profile were similar among DFM treatments for both diets (Additional file [2](#MOESM2){ref-type="media"}: Table S2).
Previous studies have shown that the effect of bacterial DFM in the rumen can vary depending on the type of DFM strain, physiological conditions of the animal \[[@CR26]\], and composition of diet \[[@CR4], [@CR10], [@CR11]\]. In studies using *Propionibacterium acidipropionici* strains P169 and P5 and *Propionibacterium jensenii* P54, reduced CH~4~ emissions (g CH~4~/kg DMI) were reported in beef steers fed a high-forage diet \[[@CR4]\], whereas the same strains failed to show any effect on beef heifers fed a high-grain diet \[[@CR11]\]. A similar observation was reported by Philippeau et al. \[[@CR10]\] using a combination of *P. jensenii* and *Lactobacillus plantarum*. The combined DFM decreased CH~4~/kg DMI in lactating cows fed low starch diet but was ineffective with a high starch diet. It was suggested that the efficacy of *Propionibacteria* to increase propionate levels in the rumen and subsequently reduce CH~4~ emissions might not be observed with high-grain diets where propionate concentration is naturally high \[[@CR11]\]. The increases in CH~4~ emissions observed with the supplementation of *P. freudenreichii* 53-W in our study cannot be explained by the above hypothesis as, as mentioned above, there were no changes in VFA profiles (Additional file [2](#MOESM2){ref-type="media"}: Table S2). However, in our previous study with wethers, this strain also increased CH~4~ emissions (g CH~4~/kg DMI) \[[@CR5]\] and a similar observation (increased trend in g CH~4~/kg DMI) was reported by Vyas et al. \[[@CR27]\] in beef heifers fed a mixed diet (60:40 forage to concentrate ratio on DM basis) with *Propionibacterium* supplementation (*P. freudenreichii* T114, T54 and *P. thoenii* T159). In the present study, the starch level of HSD was similar to the study of Vyas et al. \[[@CR27]\]. This can partly explain the similar results between these studies.
A possible reason of DFM failure is that added bacteria were not active or not present in sufficient numbers to have a detectable effect. The viability of the bacterial DFM inocula was tested before utilization and their presence was assessed in the rumen 3 h after administration. The abundance of all three DFM 3 h after administration was higher (tenfold or more) when compared to CTL cows (*P* \< 0.05; Fig. [1](#Fig1){ref-type="fig"}). However, it can not be excluded that these concentrations were not high enough to modulate ruminal functions. The doses used for the three DFM was chosen based on our previous study in wethers \[[@CR9]\] and also for practical considerations of industrial production. These doses are comparable and rather in the high end of the range found in the literature \[[@CR4], [@CR10], [@CR11], [@CR27], [@CR28]\]. For the effect of DFM supplementation on the numbers of other ruminal microbial groups, there was no effect on 16S rRNA copy numbers of total bacteria and *mcrA* copy numbers of total methanogens. Similarly, no treatment effect was observed in total protozoal counts or protozoal profile (Additional file [3](#MOESM3){ref-type="media"}: Table S3).Fig. 1Average abundance of 16S rRNA copies of *Propionibacterium freudenreichii* (PF) and *Lactobacillus bulgaricus*, and 16S--23S intergenic region copies of *Lactobacillus pentosus* in the rumen of dairy cows fed high-starch (HSD) and high-fiber diets (HFD), collected 3 h after administration of direct-fed microbials. CTL-Control cows (in white), TRT-DFM treated cows (in black). Please note that *Y* axis starts at 3 and not 0. \* significantly (*P* ≤ 0.05) different from CTL group
Dry matter intake, milk production and composition {#Sec12}
--------------------------------------------------
DFM supplementation did not influence DMI, milk production or protein and fat yields (Table [3](#Tab3){ref-type="table"}). Improvements in milk production (4.6%) were reported in multiparous dairy cows (with 3 or more lactations) fed a high-grain diet supplemented with *Propionibacterum* strain P169 \[[@CR28]\]. In the same study no difference in milk production was observed with *Propionibacterium* supplementation in younger dairy cows (up to 2 lactations). A similar finding on milk production was reported in dairy cows fed a total mixed ration supplemented with *Propionibacterium* strain P169 \[[@CR26]\], in which the positive effect of the DFM was more marked in multiparous than in primiparous cows. The studies cited above suggest that parity may have an influence on the response to DFM, with primiparous cows, like the ones used in our study, being less reactive. More *in vivo* studies needed to confirm this suggestion. Notwithstanding, strain particularities and other factors might also be involved.
Although we did not find any effect of individual DFM on milk performance, BW increased when HSD was supplemented with DFM (*P* \< 0.05; Table [3](#Tab3){ref-type="table"}). These changes in BW were mainly driven by *Propionibacterium* and *L. pentosus*. Although not statistically significant, a similar numerical trend in BW was observed when HFD was supplemented with DFM. Improved energy balance and increased BW in *Propionibacterium*-treated cows were observed previously by Francisco et al. \[[@CR29]\] in early lactation cows. In our study, the restriction of DMI to 90% of the ad libitum intake may have exacerbated a potential influence of *Propionibacterium* and *L. pentosus* on the energy balance and partitioning in cows fed HSD. Numerically lower milk production in cows fed HSD with *Propionibacterium* supplementation resulted in increased CH~4~ intensity expressed in g/kg milk (*P* \< 0.05). The metabolic shift that may have been induced by these bacterial DFM could be due to the physiological status of primiparous dairy cows that mobilize significantly less body reserves than second- and third-parity cows \[[@CR30]\]. This mode of action beyond the gastrointestinal tract should be further explored using a larger number of both primiparous and multiparous lactating cows.
Milk fatty acid composition {#Sec13}
---------------------------
Milk fatty acids were determined because they can be used as proxies to estimate CH~4~ emissions \[[@CR31]\]. Also, several strains of *Propionibacterium* and *Lactobacillus* species have been identified as potential producers of conjugated linoleic acids (CLA) \[[@CR32]\]. In this study, the milk FA composition was affected by diet as expected (statistics not presented) but DFM induced almost no effect (Additional file [4](#MOESM4){ref-type="media"}: Table S4).Additional file [4](#MOESM4){ref-type="media"}: Table S4 shows also some other minor changes that were particularly detected using orthogonal contrasts.
Apas et al. \[[@CR33]\] showed that supplementation of a mixture of *Enterococcus, Lactobacillus* and *Bifidobacterium* strains modified milk FA composition of goats with increases in *cis*-9, *trans*-11 CLA content. In contrast, we did not see any changes in milk *cis*-9, *trans*-11 CLA concentration due to DFM supplementation. The absence of clear changes in the FA profile of milk is in line with other observations.
Conclusions {#Sec14}
===========
The bacterial DFM *P. freudenreichii* 53-W increased CH~4~ emissions intensity (g CH~4~/kg milk) when cows were fed a high starch diet, whereas, none of the DFM used (*P. freudenreichii* 53-W, *L. pentosus* D31 or *L. bulgaricus* D1) affected ruminal fermentation and production parameters in lactating primiparous dairy cows irrespective of diet.
Most information on the effect of DFM on ruminal fermentation and CH~4~ reduction has been obtained*in vitro*. The results of this work should be taken as a cautionary note as bacteria selected for their modulating activities *in vitro* were not able to induce similar effects *in vivo* and for one DFM the opposite effect was observed for CH~4~ emission. Although discrepancy between *in vitro* and *in vivo* studies is generally known, published studies on this aspect are scarce. Reporting these kinds of studies, where the original hypothesis was not supported by the results, is necessary for an unbiased body of information. To explain this discrepancy, it is important that in future work, strains should be clearly identified, and doses and mode of administration stated.
Additional files
================
{#Sec15}
Additional file 1:**Table S1.** Primers used in this study. (DOCX 33 kb) Additional file 2:**Table S2.** Ruminal fermentation parameters of lactating cows fed high-starch (HSD) or high-fiber diets (HFD) supplemented with bacterial direct-fed microbials (DFM) *Propionibacterium freudenreichii* 53 W (PF), *Lactobacillus pentosus* D31 (LP), and *Lactobacillus bulgaricus* D1 (LB). (DOCX 33 kb) Additional file 3:**Table S3.** Ruminal concentration of bacteria, archaea, and protozoa (per mL rumen fluid) of lactating cows fed high-starch (HSD) or high-fiber diets (HFD) supplemented with bacterial direct-fed microbials (DFM) *Propionibacterium freudenreichii* 53 W (PF), *Lactobacillus pentosus* D31 (LP), and *Lactobacillus bulgaricus* D1 (LB). (DOCX 31 kb) Additional file 4:**Table S4.** Major milk fatty acid (FA) composition of cows fed high-starch (HSD) or high-fiber diets (HFD) supplemented with bacterial direct-fed microbials (DFM) *Propionibacterium freudenreichii* 53 W (PF), *Lactobacillus pentosus* D31 (LP), and *Lactobacillus bulgaricus* D1 (LB). (DOCX 69 kb)
ADF
: Acid detergent fibre
CFU
: Colony forming units
CP
: Crude protein
CTL
: Control without DFM
DFM
: Direct-fed microbials
DMI
: Dry matter intake
FA
: Fatty acid
FAME
: Fatty acid methyl esters
GE
: Gross energy
h
: Hour
HFD
: High-fibre diet
HSD
: High-starch diet
NDF
: Neutral detergent fibre
We thank D. Roux, J. Bourdassol, F. Rosa and S. Rudel (UE1414 Herbipôle) and P. G. Toral (UMR Herbivores) for animal care and assistance in sample collection; and Y. Rochette, D. Graviou and C. Delavaud (UMR Herbivores) for their help in laboratory analysis.
Funding {#FPar1}
=======
Funding for the study was from Danone Research, Palaiseau, France. MP and DM acknowledge support from METHLAB a FACCE ERA-GAS project in collaboration with the French National Research Agency (ANR).
Availability of data and materials {#FPar2}
==================================
All data analysed during this study are included in this article and its supplementary files.
DM, CM and JJ conceived and designed the experiments. JJ, CM and ME carried out the experiments. JJ, ME, AF and MP analysed the samples. DM and JJ wrote the manuscript. CM, ME, AF and MP provided critical discussions during revision. All authors read and approved the final manuscript.
Ethics approval {#FPar3}
===============
The study was approved by the Auvergne regional ethic committee for animal experimentation, approval number CE05--12.
Consent for publication {#FPar4}
=======================
Not applicable.
Competing interests {#FPar5}
===================
The authors declare that they have no competing interests.
| {
"pile_set_name": "PubMed Central"
} |
Background
==========
Alzheimer\'s disease (AD) is an age-related progressive neurodegenerative disorder that causes impairments in memory and thinking. The strongest genetic risk factor for AD is apolipoprotein E (*APOE*) genotype \[[@B1]\]. In comparison to people who are homozygous for the common ε3 allele, people who carry the ε4 allele are at higher risk for AD and generally have an earlier age of onset, while people who carry the ε2 allele are at lower risk and have a later age of onset \[[@B2]-[@B6]\]. ApoE is a chaperone for amyloid-β (Aβ) peptide, which deposits in the brain and is thought to initiate a cascade of events that causes AD \[[@B7],[@B8]\]. Mouse models have shown that the time of onset and amount of Aβ deposition depends not only on *APOE*genotype but also on apoE levels. Interestingly, higher expression of mouse apoE increases the amount of Aβ deposition \[[@B9],[@B10]\], while higher expression of the human ε3 isoform of *APOE*knocked into the mouse *Apoe*locus decreases levels of amyloid deposition \[[@B11]\]. Additionally, expression of human apoE in mice delays the onset of Aβ deposition in an isoform-specific fashion, with ε2 expression decreasing Aβ deposition the most and ε4 expression decreasing Aβ deposition the least \[[@B12],[@B13]\].
Despite evidence from animal studies suggesting that apoE levels affect Aβ deposition, there is no consensus regarding levels of apoE expression and its effects on Aβ deposition in human studies. The examination of whether apoE levels affect AD risk in humans has focused on *APOE*promoter polymorphisms. Over 50 studies listed on the Alzforum website tested for an association between AD and one or more polymorphisms within the *APOE*promoter \[[@B14]\]. Meta-analyses on this website support the notion that *APOE*promoter variation is associated with risk for AD. However, it is unclear whether this association is due to linkage disequilibrium with the coding polymorphisms or whether there are independent effects on risk due to the level of *APOE*expression. Some studies have examined the effect of *APOE*promoter polymorphisms on *APOE*expression *in vitro*\[[@B15],[@B16]\]. More recently, allele specific gene expression has been used in post-mortem brain samples to measure the relative expression of *APOE*ε3 and ε4 isoforms \[[@B17]\]. However, even these studies do not directly examine the effect of the promoter polymorphisms on levels of apoE protein.
Previous studies of CSF apoE levels in humans have reached varying conclusions. Some report that CSF apoE levels are lower in AD subjects than in control subjects \[[@B18]-[@B20]\], other studies find no association between CSF apoE levels and AD \[[@B21],[@B22]\], and one study shows that CSF apoE levels are higher in AD subjects than in control subjects \[[@B23]\]. Multiple studies found that the *APOE*genotype was not associated with differing CSF apoE levels \[[@B19]-[@B22]\]. In contrast, plasma apoE levels are clearly dependent on *APOE*genotype \[[@B24],[@B25]\], which suggests that apoE is metabolized differently in the CSF and plasma. Gender and age do not appear to affect CSF apoE levels \[[@B22]\].
Recently, our laboratory and others reported that apoE levels were greatly reduced in mice lacking functional ATP-binding cassette A1 transporter (ABCA1) \[[@B26]-[@B28]\]. Within the CNS of ABCA1 knock-out mice, CSF apoE was 2% of normal levels and apoE in the cortex was 20% of normal levels \[[@B26]\]. ABCA1 transfers cholesterol and phospholipids from the cell membrane to apolipoproteins (including apoE) to form nascent high density lipoproteins (HDL). In the rare case that both copies of *ABCA1*are non-functional, as occurs in Tangier\'s disease, apoE and other lipoproteins do not receive normal amounts of lipid and are rapidly degraded \[[@B29]\]. Multiple studies have shown that levels of plasma HDL-C and associated apolipoproteins are affected by single nucleotide polymorphisms (SNPs) in *ABCA1*\[[@B30]-[@B34]\]. In particular, studies have implicated the following SNPs in affecting levels of plasma HDL-C: rs2230806 (R219K) \[[@B33]\], rs2066718 (V771M) \[[@B31],[@B32]\], rs2066715 (V825I) \[[@B31]\], rs4149313 (I883M) \[[@B34]\], rs2230808 (R1587K) \[[@B31]\]. Since ABCA1 appears to have a similar role in the CNS and in the periphery, we hypothesized that these *ABCA1*SNPs would also have an effect on CSF apoE levels since apoE is the major apoprotein component of HDL produced in the CNS. Additionally, studies by others have reported that the *ABCA1*SNP rs2230806 (R219K) affects risk for AD \[[@B35]-[@B38]\]. This is particularly interesting because *ABCA1*falls within a region of chromosome 9 that is linked to late-onset AD \[[@B39]-[@B43]\]. The profound effect of ABCA1 levels on CNS apoE levels in mice, in addition to reports that an *ABCA1*SNP may affect risk for AD, suggested that *ABCA1*may be involved in the genetic control of CNS apoE levels in humans.
Given the contrasting results and small sample sizes used in some studies of apoE levels in human CSF, we chose to begin our study by characterizing CSF apoE levels in a relatively large sample of 168 individuals with respect to AD status, *APOE*genotype, gender, race and age. We next examined whether ten *ABCA1*SNPs, including five SNPs shown to affect plasma HDL-C, affected levels of apoE in the CSF. Finally, in a large sample of 1225 AD cases and 1431 controls, we attempted to replicate the previously reported association between the *ABCA1*SNP rs2230806 and AD.
Results
=======
ApoE levels and stability in human CSF
--------------------------------------
ApoE levels were measured in CSF samples from 168 subjects who were 43 to 91 years old (Table [1](#T1){ref-type="table"}). We included all samples available without regard to AD status, *APOE*genotype, gender, race or age. ApoE values were sorted into 1 μg/ml bins and the number of subjects with apoE values within each bin from 0 to 16 μg/ml was tallied (Fig. [1A](#F1){ref-type="fig"}). The mean apoE level was 9.09 μg/ml with a standard deviation of 2.70 μg/ml. The number of individuals per bin was in a normal distribution according to a Kolmogorov-Smirnov test (p \> 0.10).
######
Characteristics of subjects who underwent lumbar puncture.
**CDR 0, \<65** **CDR 0, ≥65** **CDR 0.5** **CDR 1+**
-------------- ----------------- ---------------- ------------- ------------
**n =** 70 55 26 17
**Male** 29% 28% 54% 47%
**Female** 71% 72% 46% 53%
**Age\*** 54 ± 6 76 ± 8 75 ± 8 76 ± 6
**ε2 freq.** 0.11 0.13 0.06 0.03
**ε3 freq.** 0.64 0.73 0.56 0.74
**ε4 freq.** 0.25 0.14 0.38 0.24
\*Age is mean ± standard deviation
{#F1}
To determine the intra-individual stability of CSF apoE levels sampled over time, lumbar puncture was performed on five subjects at two different times that were two weeks apart. CSF apoE levels within the same individual were strongly correlated (r^2^= 0.93, p \< 0.01). In contrast, CSF apoE levels between different individuals showed large variation (coefficient of variation = 46%) (Fig. [1B](#F1){ref-type="fig"}). This demonstrates that CSF apoE levels are relatively stable within an individual during a short time interval, but vary widely between individuals. Furthermore, this suggests that CSF apoE levels may be influenced by stable individual differences, such as genetic sequence variation.
Effects of AD status, APOE genotype, gender or age on CSF apoE levels
---------------------------------------------------------------------
There are varying reports in the literature on whether CSF apoE levels are affected by AD status, *APOE*genotype, gender or age. In our relatively large sample, we investigated whether these variables, as well as race, modified CSF apoE levels. The levels of CSF apoE were not significantly different between subjects who were cognitively normal who had a clinical dementia rating (CDR) score of 0 and those who had very mild (CDR 0.5) or mild-moderate dementia believed to be due to AD (CDR 1+) (Fig. [2A](#F2){ref-type="fig"}). Since a recent study reported that apoE levels may be affected by *APOE*genotype \[[@B44]\], we examined whether *APOE*genotype affects CSF apoE levels in our sample. Despite large numbers of patients, we found no significant differences in CSF apoE levels in subjects with different *APOE*genotypes (Fig. [2B](#F2){ref-type="fig"}). Next, we looked for gender effects on CSF apoE levels and found none (Fig. [2C](#F2){ref-type="fig"}). We also found no significant difference in CSF apoE levels between subjects who identified themselves as Caucasians and African Americans (Fig. [2D](#F2){ref-type="fig"}). Additionally, we studied whether age affects CSF apoE levels (Fig. [2E](#F2){ref-type="fig"}). Average apoE levels increased by a small but significant extent, \~0.5 μg per 10 years (r^2^= 0.05, p = 0.003). Finally, to test the possibility that AD status, *APOE*genotype, gender and age interact to influence apoE levels in the CSF, we performed a multivariate ANOVA and found no significant interactions. We conclude that CSF apoE levels are not greatly affected by AD status, *APOE*genotype, gender or race, but do increase with age.
{#F2}
Effects of ABCA1 SNPs on CSF apoE levels and risk for AD
--------------------------------------------------------
We sought to determine whether SNPs in *ABCA1*affect CSF apoE levels. The subjects for whom we had CSF apoE data were genotyped for the following *ABCA1*SNPs: rs2230806 (R219K), rs2066718 (V771M), rs2066715 (V825I), rs4149313 (I883M), rs2230808 (R1587K), rs1883025 (intron), rs2275544 (intron), rs2777799 (intron), rs3904999 (intron) and rs6479283 (intron). The numbers of subjects for which we obtained conservatively called (high quality) genotypes, as well as the frequencies of the minor and major alleles, are listed in Table [2](#T2){ref-type="table"}. We found no association between CSF apoE levels and any of the *ABCA1*SNPs, including the five coding SNPs that were previously associated with alterations in plasma HDL-C levels.
######
The number of subjects with high quality genotypes and the frequency of the minor and major ABCA1 SNP alleles.
**n =** **minor allele freq.** **major allele freq.**
--------------- --------- ------------------------ ------------------------
**rs2230806** 123 0.309 0.691
**rs2066718** 124 0.040 0.960
**rs2066715** 144 0.073 0.927
**rs4149313** 124 0.185 0.815
**rs2230808** 124 0.315 0.685
**rs1883025** 102 0.358 0.642
**rs2275544** 122 0.131 0.869
**rs2777799** 123 0.126 0.874
**rs3904999** 123 0.203 0.797
**rs6479283** 119 0.223 0.777
We also attempted to reproduce the finding, reported by some groups but not others, that the *ABCA1*rs2230806 SNP is associated with altered risk for AD \[[@B35]-[@B38],[@B45]\]. We combined information on 794 subjects from Washington University with 1,862 additional subjects from the University of California-San Diego and the United Kingdom to yield the maximum power. The subjects from Washington University had previously been analyzed and it was found that risk for AD in this group did not depend on the rs2230806 SNP \[[@B36]\]. The 1,862 additional subjects had not previously been used to examine the rs2230806 SNP. In this large group of 1225 case and 1431 control subjects, there was no effect of the rs2230806 SNP on risk for AD (Table [3](#T3){ref-type="table"}). Analysis of sub-groups based on *APOE*genotype and gender also failed to show an effect of the rs2230806 SNP on risk for AD.
######
The distribution of the rs2230806 polymorphism in subjects with Alzheimer\'s disease and control subjects.
*\#* \# AA \# AG *\#*GG freq. A freq. G AD vs. Control
------------- --------- ------ ------- ------- -------- --------- --------- ----------------
**Total** AD 1225 81 476 668 0.260 0.740 **p = 0.76**
n = 2656 Control 1431 105 548 778 0.265 0.735
**E3/E3** AD 437 31 170 236 0.265 0.735 p = 0.93
n = 1316 Control 879 63 351 465 0.271 0.729
**E4/E3** AD 555 32 227 296 0.262 0.738 p = 0.10
n = 832 Control 277 18 92 167 0.231 0.769
**E4/E4** AD 125 8 40 77 0.224 0.776 p = 0.86
n = 150 Control 25 1 9 15 0.220 0.780
**Females** AD 267 26 105 136 0.294 0.706 p = 0.99
n = 505 Control 238 23 94 121 0.294 0.706
p values are caculated by Chi Square tests with 2 degrees of freedom
Discussion
==========
A notable finding in this study was that CSF apoE levels vary widely between individuals, with a range in our sample from 2 μg/ml to 16 μg/ml, but are stable within individuals during an interval of 2 weeks. This suggests the presence of stable factors within individuals, which may be genetic or environmental, that regulate CSF apoE levels. Recently, it was reported that levels of Aβ vary according the time of day and it is possible that apoE could vary in a similar fashion \[[@B46]\]. However, since all of our samples were obtained at the same time of day (8:00 am), any diurnal variation of apoE levels in this study should be minimal.
We examined whether AD status, *APOE*genotype, gender, race or age affected CSF apoE levels, but only age was significantly correlated. It is interesting that levels of apoE are not elevated in carriers of the ε2 allele. ApoE3 and apoE4 both bind with high affinity to LDLR resulting in receptor-mediated endocytosis and degradation of apoE. ApoE2 does bind to LDLR, but much less effectively than apoE3 and apoE4 \[[@B47]\]. In mice, the decreased affinity of apoE2 for LDLR leads to elevated levels of CSF apoE in mice in which the human *APOE*ε2 gene is knocked-in to the mouse *Apoe*gene locus \[[@B48]\]. The lack of a difference in apoE levels according to genotype in human CSF samples suggests that LDLR may not have as large of an effect on human CSF apoE levels. It will be important to assess this issue in future studies in *APOE*ε2 homozygous individuals as there may be a much smaller effect in individuals with one copy of the *APOE*ε2 gene.
We hypothesized that genetic variation in certain genes may contribute to CSF apoE levels and examined whether SNPs in *ABCA1*, especially SNPs that have been reported to affect plasma HDL-C levels, affect CSF apoE levels. We did not find a significant association between CSF apoE levels and any of the ten *ABCA1*SNPs we examined, including the five coding SNPs thought to be associated with altered HDL-C levels. Perhaps this is because the metabolism of apoE is different in the plasma and CSF. Alternatively, these changes in ABCA1 may not affect HDL in the CNS as much as occurs as with HDL in the plasma. This may be due to apoAI being the main apoprotein in plasma HDL whereas apoE is the most abundant apoprotein produced in the CNS in CSF HDL. The effects of the SNPs may also be too small to significantly affect CSF apoE levels. However, it remains possible that rare sequence variations that strongly influence *ABCA1*function could contribute to variation in CSF apoE levels. Recent studies demonstrate that several rare polymorphisms in *ABCA1*collectively affect overall levels of plasma HDL-C in the population \[[@B30],[@B31]\]. Since ABCA1-mediated lipid transport is critical in the formation of both HDL-C in plasma and apoE-containing lipoproteins in CSF, it is possible that the same rare *ABCA1*polymorphisms that have large effects on plasma HDL-C levels would also affect CSF apoE levels.
Additionally, we failed to replicate the finding of other groups that the *ABCA1*rs2230806 SNP is associated with altered risk for AD \[[@B35]-[@B38]\]. We suggest three possible reasons for the differing results: 1) the *ABCA1*rs2230806 SNP does affect risk for AD, but the effect is small so that the association cannot be reproduced regularly in samples of \~2500 subjects; or 2) the population we examined was genetically different from the populations in the other studies assessed; or 3) the *ABCA1*rs2230806 SNP does not affect risk for AD. Since the populations that we and others examined are similar and consisted primarily of Caucasians with Northern European heritage, we believe that it is most likely that the *ABCA1*rs2230806 SNP contributes either a very small amount or not at all to overall risk for AD.
It seems likely that many different genes modulate levels of apoE in the CSF. Studies suggest that LDLR and LRP influence levels of CSF apoE in mice \[[@B48],[@B49]\]. Given the animal data, it is possible that variations in *LDLR*or *LRP*could affect CSF apoE levels in humans, but this has not yet been examined. Further investigation of the genetic control of apoE levels in the CNS could uncover new information on apoE metabolism. This research would not only be relevant to AD, but also to a number of other neurological diseases that may be modulated by apoE such as stroke \[[@B50],[@B51]\], multiple sclerosis \[[@B52]\] and traumatic brain injury \[[@B53]\]. Ultimately, an understanding of the regulation of CSF apoE levels could lead to novel apoE-based treatments for AD and other neurological disorders.
Conclusion
==========
We found that CSF apoE levels vary widely between individuals, but are stable within individuals over a two-week interval. Secondly, AD status, *APOE*genotype, gender and race do not affect CSF apoE levels, but CSF apoE levels do increase with age. Additionally, ABCA1 SNPs that have been reported to affect plasma HDL-C levels do not affect CSF apoE levels in our sample. Finally, any association that exists between the *ABCA1*SNP rs2230806 and AD is very weak.
Methods
=======
Subjects
--------
Subjects in the Washington University sample were community-living participants in the Alzheimer\'s Disease Research Center (ADRC) registry. All research subjects underwent a clinical evaluation to determine their Clinical Dementia Rating (CDR), as well as a 2-hour psychometric test battery. A medical history was taken to exclude participants that might have confounding medical disorders. Details of the assessment have been described previously \[[@B54]-[@B56]\]. Additional case control DNA samples were from the University of California-San Diego and the United Kingdom.
CSF was obtained via lumbar puncture (L.P.) from 168 subjects at Washington University in the General Clinical Research Center after obtaining informed consent. The study protocol was approved by the Human Studies Committee at Washington University. All L.P.s were performed at 8 am after an overnight fast with a 22 gauge atraumatic needle. 25--30 ml of CSF was obtained from each subject and was free of blood contamination. After collection, CSF samples were briefly centrifuged at 1,000 × *g*to pellet any cell debris, frozen, and stored in polypropylene tubes at -80°C in 0.5 ml aliquots until analysis.
ApoE ELISA
----------
ApoE ELISAs were performed on CSF apoE as previously described \[[@B48]\]. Briefly, plates were coated overnight with WUE4, a monoclonal antibody to human apoE \[[@B57]\]. The plates were washed, blocked with 1% dry milk and washed again. ApoE standards were purified from human β-VLDL (BioDesign, Sako, ME). Standards and samples were diluted and loaded onto the plate, then incubated overnight. The plate was washed and incubated with a polyclonal goat anti-apoE antibody (Calbiochem, San Diego CA). The plate was washed again and incubated with anti-goat-HRP (Vector Laboratories, Burlingame, CA). The plate was washed once more, then developed with TMB (Sigma, St. Louis, MO).
Genotyping
----------
The following SNPS in *ABCA1*were genotyped in the Washington University sample of 168 subjects: rs2230806 (R219K), rs2066718 (V771M), rs2066715 (V825I), rs4149313 (I883M), rs2230808 (R1587K), rs1883025 (intron), rs2275544 (intron), rs2777799 (intron), rs3904999 (intron) and rs6479283 (intron). Genotyping was performed using a modified single nucleotide extension reaction with allele detection by mass spectrometry (Sequenom MassArray system; Sequenom, San Diego, CA, USA). PCR primers, termination mixes and multiplexing capabilities were determined with Sequenom Spectro Designer software v2.00.17. Genotyping of rs2230806 in the large group of 2,656 subjects was performed using allele specific real-time PCR \[[@B58]\]. For all SNPs, genotypes were tested and found to be in Hardy-Weinberg equilibrium.
Statistical analyses
--------------------
Frequency distributions, correlation analysis, ANOVAs, T-tests and Kolmogorov Smirnov tests of normality were performed using GraphPad Prism, Version 4.00 (GraphPad, San Diego, CA). Multivariate ANOVAs were performed using SAS Version 9.0 for Windows XP (SAS Institute Inc., Cary, NC).
Abbreviations
=============
Aβ, amyloid-β peptide; ABCA1, ATP-binding cassette transporter A1; AD, Alzheimer\'s disease; apoE, apolipoprotein E; CDR, clinical dementia rating; CNS, central nervous system; CSF, cerebrospinal fluid; ELISA, enzyme-linked immunosorbent assay; HDL, high density lipoprotein; LDLR, low density lipoprotein receptor; LP, lumbar puncture; LRP, low density lipoprotein related protein; SNP, single nucleotide polymorphism.
Competing interests
===================
The author(s) declare that they have no competing interests.
Authors\' contributions
=======================
SEW performed the primary writing and editing of the manuscript and was involved in experimental design, genotyping and data analysis. ARS processed CSF samples and assayed them for levels of apoE. AMF was involved in coordinating CSF collection and experimental design. SS, AG, KM, and HJ were involved in genotyping and experimental design.
JSKK and AH were involved in experimental design and statistical analysis. LJT provided samples from the UCSD collection. AMG and DMH were involved in experimental design, data analysis, and manuscript writing. All authors approved the manuscript.
Acknowledgements
================
The authors gratefully acknowledge the Genetics, Clinical, Psychometric, and Biostatistics Cores of the Washington University ADRC for subject APOE genotyping and clinical, cognitive and psychometric evaluation and data management. We also acknowledge the contributions of our LP physicians at Washington University (Dept. of Neurology): David Holtzman, MD; Randall Bateman, MD; David Brody, MD, PhD; B. Joy Snider, MD, PhD; and Beth Ann Ward, MD. Grants: This work was supported by grants from the National Institute on Aging (P01 AG03991, P01 AG026276, P50 AG05681, RO1 AG16208), a pilot grant from the Genome Sequencing Center at Washington University, and the Washington University General Clinical Research Center funded by the US Public Health System (M01 RR00036). J.S.K.K. is funded by a Ford Foundation Predoctoral Fellowship.
| {
"pile_set_name": "PubMed Central"
} |
Background {#Sec1}
==========
Breast cancer (BC) is the most common cancer in women and the 5-year survival rate in Europe is 82% suggesting a considerable residual risk \[[@CR1]\]. BC is associated also with other chronic conditions including type two diabetes which may increase the risk of BC recurrence \[[@CR2]--[@CR4]\] and many BC patients are at increased risk for cardiovascular disease (CVD) \[[@CR5], [@CR6]\]. Mechanisms of BC appear to be linked to sex hormones, impairment in glucose metabolism, hyperglycemia, hyperinsulinemia, insulin-like growth factors (IGF), inflammation, oxidative stress and impaired cell apoptosis \[[@CR7], [@CR8]\]. Current advice to BC survivors suggests adopting cancer prevention strategies \[[@CR8], [@CR9]\], however there is no consensus on the effectiveness of lifestyle programs in women with BC \[[@CR10]--[@CR12]\] mostly due to lack of sufficient evidence \[[@CR13]\]. The rationale for the study is to target several mechanisms of cancer suppression or proliferation, with a healthy diet and exercise program, to avoid low vitamin D levels in order to obtain maximal efficacy of the lifestyle program, and concomitantly to offer maximal CVD and diabetes protection. Dietary carbohydrates are the main food components to affect glycemia and insulinemia. The glycemic index (GI) is able to capture the difference between those that increase glycemia the most (high GI foods) and those that increase it the least (low GI foods) \[[@CR14]\]. Low GI foods lower the glycemic and insulinemic potential of the diet and have been shown to reduce the risk of several cancers particularly diabetes-related cancers including BC \[[@CR15]--[@CR18]\] and some evidence suggests they also correlate with lower recurrence \[[@CR19]\]. Furthermore, low GI diets have been inversely associated with risk of type two diabetes \[[@CR15], [@CR20], [@CR21]\] and CVD \[[@CR22], [@CR23]\] and favourably modified blood glucose \[[@CR24]--[@CR26]\], blood lipids \[[@CR27]\], inflammatory markers \[[@CR28]\], oxidative damage \[[@CR29]\], body weight \[[@CR30]\] and IGF binding proteins \[[@CR31]\], all factors relevant to carcinogenesis, diabetes and CVD. Our previous studies in Italy have shown significant risk reduction of 40% with a low GI diet compared to a high GI diet in BC primary prevention \[[@CR18]\]. Physical activity is one of the mainstays of primary prevention of cancer and it is also included in guidelines for BC survivors (at least 150 minutes per week) mainly to reduce complications such as lower muscle strength and risk of depression \[[@CR13]\]. However, physical activity after BC diagnosis has also been shown to reduce the risk of BC mortality by 40--50% particularly when it is of moderate intensity such as 30 minutes of brisk walking per day \[[@CR32]--[@CR35]\] an effect possibly modulated partly by insulin economy improvements in reduced insulin, insulin-like growth factors and estrogen levels \[[@CR36]\].
To maximize lifestyle changes it may be useful to avoid vitamin deficiencies, particularly vitamin D deficiency which has been linked to higher breast cancer risk \[[@CR37], [@CR38]\]. Vitamin D alters genes implicated in cellular growth, affecting proliferation, apoptosis, differentiation, angiogenesis, invasion and metastasis \[[@CR37]\]. Preliminary studies suggest that normal-high ranges of serum vitamin D levels improve BC survival \[[@CR39], [@CR40]\].
Low glycemic index diet, exercise and vitamin D intake, in addition to determining metabolic changes and benefits for cancer patients, may be able to change the tumor microenvironment and lead to epigenetic modifications \[[@CR41]--[@CR44]\]. In this context, microRNAs (miRNAs), small noncoding RNA molecules, may play a fundamental role in modulating gene expression and breast cancer progression \[[@CR45], [@CR46]\]. The evaluation of circulating miRNAs is useful to identify the change of specific miRNAs involved in cancer pathways and predict the development of recurrence in breast cancer patients.
A lifestyle program that targets all the above mechanisms may be warranted. To our knowledge no trial has evaluated the combined effect of a lifestyle program with a healthy diet focusing on low GI, additional physical activity and supplemental vitamin D on BC recurrence and complications in the context of a Mediterranean dietary setting.
Methods/Design {#Sec2}
==============
Aims and objectives {#Sec3}
-------------------
DEDiCa study aims primarily at reducing BC recurrence in women with BC using a lifestyle approach with additional vitamin D. The primary objective of DEDiCa study is to determine the effect of a 33-month program combining advice on diet, exercise and supplemental vitamin D, on reducing BC recurrence rates or increasing disease-free survival (DFS). Secondary objectives are to improve markers of diabetes risk and management for those who already have diabetes, to improve cardiometabolic health and quality of life (QoL) and to investigate whether changes in microRNA correlate with changes in lifestyle.
Study design {#Sec4}
------------
This is a randomized clinical trial targeting women with BC stages I-III within 12 months from BC surgery (see Tables [1](#Tab1){ref-type="table"} and [2](#Tab2){ref-type="table"} for details). The study involves at least five cancer centers in Italy (Table [3](#Tab3){ref-type="table"}) and it started on Oct 10^th^, 2016. Consenting participants are randomized to either a high intensity program (HIT) or a lower intensity program (LITE, positive control). The endpoints will be assessed at baseline and every three months until the end of the study (Fig. [1](#Fig1){ref-type="fig"}).Table 1Inclusion and exclusion criteriaInclusion criteriaExclusion criteria1. Women with primary diagnosis of histologically confirmed breast cancer (T1 with Ki67 ≥ 30%, T2, T3 without metastasis) within 12 months from diagnosis.\
2. Age ≥ 30 and \< 75 years.\
3. Patients who are able to comprehend and are willing to sign the consent form and are able to adhere to the protocol including scheduled clinic visits and assigned treatment.1. Patients who do not possess the inclusion criteria for this study.\
2. Patients with sarcoidosis or other granulomatous diseases or with hypercalcemia (Ca \> 11 mg/dL).\
3. Patients with any previous or current concomitant malignant cancer.\
4. Pregnant or lactating women.\
5. Patients with AIDS diagnosis\
6. Patients with severe renal insufficiency\
7. Patients with kidney stones (nephrocalcinosis or nephrolithiasis)\
8. Patients participating in other lifestyle clinical trials Table 2Details of inclusion criteria n. 1StagePrimary TumorLymph nodesMetastasisKi-67IT1b, T1cN0M0≥30%IIAT1a, T1b, T1c\
T2N1\
N0M0\
M0anyIIBT2\
T3N1\
N0M0\
M0anyIIIAT1a, T1b, T1c\
T2\
T3N2\
N2\
N1, N2M0\
M0\
M0anyIIICT1a, T1b, T1c\
T2\
T3N3\
N3\
N3M0\
M0\
M0any Table 3List of recruiting centres in ItalyRecruiting centres• Coordinating Centre: National Cancer Institute Fondazione G. Pascale (Napoli);\
○ Via Mariano Semmola -- 80131 Napoli\
○ Tel: 081/5903395\
○ Email: epidemiologia\@istitutotumori.na.it\
○ [www.istitutotumori.na.it/](http://www.istitutotumori.na.it/)\
• Clinica Mediterranea, Senology Unit; Via Orazio, 2--80122 Napoli; [www.clinicamediterranea.it/](http://www.clinicamediterranea.it/)\
• Cannizzaro Hospital, Senology Unit; Via Messina, 829--95126 Catania; [www.aocannizzaro.it/](http://www.aocannizzaro.it/)\
• San Vincenzo Hospital of Taormina, Oncology Unit; Via Sirina, 98039 Taormina (Messina); [www.oncologiataormina.it/](http://www.oncologiataormina.it/)\
• National Cancer Institute CRO Aviano; Via Franco Gallini, 2--33081 Aviano (Pordenone); [www.cro.it](http://www.cro.it) Fig. 1Schematic representation of DEDiCa study protocol
To improve compliance participants on both sides will be followed equally with blood tests and clinic visits and will be equally offered general advice on a healthy Mediterranean diet and physical activity, step counters and vitamin D supplements to avoid insufficiency. Participants in the HIT group will additionally receive dietary counselling on how to reduce the GI of their diet, packages of pasta, more in-depth advice on exercise and more vitamin D supplements compared to the LITE group.
This study was approved by the ethic board of the Italian Medicine Agency (AIFA), and of each recruiting hospital: National Cancer Institute "Fondazione Giovanni Pascale" in Naples, Azienda Ospedaliera per l'Emergenza Ospedale Cannizzaro in Catania, Azienda Ospedaliera Universitaria Policlinico "G. Martino" in Messina for San Vincenzo Hospital of Taormina, Comitato Etico Campania Centro ASL NA1 Centro for Clinica Mediterranea in Naples. The study has been registered with ClinicalTrials.gov (registration date and number: May 11, 2016; NCT02786875).
Participants {#Sec5}
------------
Women who had undergone surgery for primary histologically confirmed BC, stages I-III (see Tables [1](#Tab1){ref-type="table"} and [2](#Tab2){ref-type="table"} for details), within the previous 12 months in cancer centres in Italy, who are between the ages 30--74 years and have no contraindications to participate in this study will be eligible to participate. Eligibility is confirmed by central reviewing of medical records and pathology reports. The inclusion and exclusion criteria are summarized in Table [1](#Tab1){ref-type="table"}. To participate in the study each potential participant is required to read and sign the study information/consent form at baseline.
Recruitment and randomization {#Sec6}
-----------------------------
Eligible participants are recruited in five oncologic centres in Italy (four in Southern Italy and one in Northern Italy). Details of the recruiting centres can be found in Table [3](#Tab3){ref-type="table"}. Eligible participants are contacted either by phone or during one of their follow-up visits and offered to hear more about the study during an information session (either group session or a one-to-one session to accommodate all needs). Eligible participants are explained the study details and are given an informed consent document to take home which they bring back at the baseline visit. At the baseline visit, after obtaining written consent the participant is sent for a blood test and other measurements, is randomized to one of the two arms of the study, receives the advice on the program according to her randomization arm and is given the next appointment slip. Participants are contacted by phone monthly for the first three months and once between visits afterwards. Study visits are scheduled every 3 months until end of study (33 months).
Randomization is done electronically in real time for all recruiting centres and is stratified by stage (I/IIA vs. IIB/IIIA) and age (\<50 yrs or ≥50 yrs) at diagnosis and is based on a permuted block design with block size of 4. Stratification by stage and age is done because we expect these variables to affect the outcomes. To prevent any possible study bias the randomization sequence will be done off-site by a third party statistician (Contract Research Organization, Naples, Italy) who will not have contacts with the study participants. The PI is blinded to the randomization of the study participants but not the staff involved in the clinic visits.
Interventions {#Sec7}
-------------
Eligible and consenting subjects are randomized to either one of the two treatment programs (higher or lower intensity):**Higher intensity (HIT) arm (test)**: low GI diet + exercise + vitamin D.All carbohydrate foods advised will be low GI choices (GI \< 70, on the bread scale), e.g. legumes, pasta al dente, barley, low GI rice, low GI bread, oat, apples, oranges, berries, avocado and nuts.Brisk walk of at least 30 min per day (or approximately 5000 steps) more than the habitual physical activity.Vitamin D supplement (cholecalciferol) up to 4000 IU/day to reach the upper end of normal blood levels of 25(OH)D (60 ng/ml).[Lower intensity (LITE) arm (positive control)]{.ul}: general recommendations for a healthy diet and physical activity. Vitamin D (cholecalciferol) will be given only if hypovitaminosis D is detected to bring blood levels up to normal ranges of 30 ng/mL.
Both groups will be counselled to follow a healthy Mediterranean diet (≥5 servings vegetable/fruit per day, ≤1 serving red meat/cold cuts per week, \<7% saturated fat).
Treatment evaluations are conducted every three months at each clinic visit and include all three components of the treatment (diet, exercise, vitamin D). Seven-day food records is collected from each patient which is filled a week before the clinic visit. The information in the food record is reviewed by the nutritionist staff with the patient and subsequently inserted in a diet analysis program (WinFood Medimatica, Version 3.9.0). Daily means of GI, energy intake, macro- and micro-nutrients and food groups will be obtained from WinFood and periodically evaluated to ensure adherence to the dietary advice. The physical activity component is measured with a step counter with 7-day memory (OMRON Walking Style IV) and with a questionnaire. Participants bring the step counter back at each clinic visit and the 7-day values are recorded by the research staff. Vitamin D is evaluated by blood analysis of 25(OH)D every three months, the dose reviewed at each clinic visit and changed if necessary to reach the group target (Table [4](#Tab4){ref-type="table"}). QoL is measured with questionnaires specifically made for cancer patients (EQ-5D-3 L, EORTC QLQ-C30 e EORTC QLQ-BC23) \[[@CR47]--[@CR49]\].Table 4Vitamin D algorithmBlood levels (ng/ml)Oral dose (IU)Treatment durationGroup A (target: 60 ng/ml): \< 1075 000 at study visit\
+4000/day3 months\
then re-evaluate \> 10-2050 000 at study visit\
+4000/day3 months\
then re-evaluate \> 20-3025 000 at study visit\
+4000/day3 months\
then re-evaluate \> 304000/day3 months\
then re-evaluate 60-80change to 1000 IU/day3 months\
then re-evaluateGroup B (target: 30 ng/ml): ≤ 10100 000 at study visitRe-evaluate after 3 months ≤ 2075 000 at study visitRe-evaluate after 3 months \> 20-2550 000 at study visitRe-evaluate after 3 months \> 25-2925 000 at study visitRe-evaluate after 3 months ≥ 300Re-evaluate after 3 months
Outcome measures {#Sec8}
----------------
Figure [1](#Fig1){ref-type="fig"} depicts the timing and frequency of all study measures. Blood analyses, blood pressure, anthropometric measurements and 7-day food records are taken at baseline and every 3 months afterwards until end of study (up to 33 months). Complete blood analyses are done at baseline, 1 year and end of study while blood analysis pertinent to the treatment are done every three months (Table [5](#Tab5){ref-type="table"}). The primary outcome is the percentage of patients alive at end of study without BC recurrence (disease in the same or opposite breast or any metastasis). The primary outcome is assessed at each collaborating centre by the collaborating oncologist and confirmed by hospital pathological results which are communicated to the coordinating centre. Secondary end points include glycemic control including blood glucose and glycated hemoglobin (HbA1c), cardiometabolic variables including body weight, waist circumference, body mass index (BMI), blood pressure, C-reactive protein (CRP) and blood lipids, hormonal measures including insulin, insulin-like growth factor-1 (IGF-1), estradiol, testosterone and sex hormone binding globulin (SHBG), and epigenetic markers (microRNA). Program adherence and any difficulty noticed, medications and medication changes, as well as any unusual or adverse events, including illness or stressful issues, that occurred since the last clinic visit are recorded in detail.Table 5Blood analyses performed during the studyParametersBaseline, 12 mo and study endEvery 3 months25(OH)DxxCalciumxxGlucosexxHbA1cxxInsulinxTriglyceridesxTotal Cholesterolx LDL-Cx HDL-CxAST/ALTxCRPxEstradiolxTestosteronexSHBGxIGF-1xmicroRNAxFor future analysesxx*AST/ALT*: aspartate transaminase/alanine transaminase, *CRP*: C-reactive protein, *HbA1c*: hemoglobin A1c, *HDL-C*: high density lipoprotein cholesterol, *IGF-1*: insulin-like growth factor-1, *LDL-C*: low density lipoprotein cholesterol, *SHBG*: sex hormone binding globulin, *25(OH)D*: 25-hydroxyvitamin D.
Blood samples and all information regarding the patient are sent to the coordinating centre where blood samples are centrally analyzed and information entered electronically and statistically analyzed for the interim and end of study reports.
Biochemical analyses {#Sec9}
--------------------
Serum vitamin D and IGF-1 are analyzed using DiaSorin kits on Liaison XL analyzer (DiaSorin) by chemiluminescent method (CLIA). The HbA1c value is analyzed using whole blood collected in EDTA Vacutainer tubes (Vacutainer; Becton, Dickinson and Co) by a turbidimetric inhibition latex immunoassay (TINA QUANT Roche Diagnostics) on Cobas C6000 analyzer (Roche). Serum glucose, total cholesterol, triglycerides, high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), are measured using reagents and analyzer (Cobas C6000) by Roche Diagnostics according to the manufacturer's instructions. Serum insulin, estradiol, testosterone and SHBG are performed on the same analyzer by electro-chemiluminescent method (ECLIA). Nephelometric quantification of CRP is performed on BNP ProSpec nephelometer (Siemens Healthcare Diagnostics) according to the manufacturer's instructions. Serum samples are obtained by blood collected in Vacutainer tubes without anticoagulant (Becton, Dickinson and Co) and analyzed within 24 hours. All analytes are measured in the coordinating hospital routine analytical laboratory undergoing quality control procedures.
MicroRNA Analysis {#Sec10}
-----------------
Previous studies have demonstrated that microRNAs (miRNAs) are frequently dysregulated in human cancers, including BC \[[@CR50], [@CR51]\] and may be modified by the glycemic load of the diet \[[@CR41]\]. Computational models are important for the understanding of biological systems \[[@CR52]\]. The following are the procedures to identify and analyze miRNA. These analyses will be conducted at the Laboratory of the Biomedical Sciences Department at the University of Catania (Italy).
### Identification of plasma miRNA in BC {#Sec11}
Plasma samples are randomly selected for the analysis by the Human Serum & Plasma miScript miRNA PCR Array (Qiagen) that profiles the expression of 84 miRNAs. In this phase, miRNA expression patterns are analyzed in BC patients according to the lifestyle and dietary intake.
### Circulating microRNA analysis {#Sec12}
miRNAs can be easily purified from a number of patient body fluids. Several studies shown that miRNAs were present in serum and plasma and easily detectable by a sample of peripheral blood. Circulating miRNAs may be important players in the formation of the tumor microenvironment and metastatic evolution by promoting epithelial to mesenchymal transition (EMT) of tumor cells. RNA and miRNAs fraction are extracted from 200 μL of plasma are isolated by miRNeasy Serum/Plasma Kit (Qiagen, Hilden, Germany) according to the manufacturer's recommendations Extracted miRNAs are reverse transcribed into cDNA and analyzed by miScript SYBR Green PCR Kit (Qiagen).
### Validation of circulating miRNAs in BC tissues {#Sec13}
Data obtained in the previous two steps above are validated in tumor tissue samples to confirm the origin source of the circulating miRNA previously identified.
### miRNAs and epithelial mesenchymal transition (EMT) {#Sec14}
miRNAs, a class of small non-coding RNA molecules that post-transcriptionally regulate gene expression, are attractive candidates for regulating stem cell self-renewal, cell fate decisions and cell plasticity. For this reason, miRNAs analysis is carried out to shed light on the post-transcriptional control of EMT and stemness.
### Analysis of EMT genes {#Sec15}
Publicly available gene expression datasets are analyzed to identify pathways involved in EMT. For this reason a comparative analysis of expression of master regulators of EMT genes (Beta-Catenin, OPN, Twist1/2, Snail1/2, Zeb1/2, N-cad, Vim, E-cad and NGAL) will be performed. GSEA (Gene Set Enrichment Analysis <http://broad.mit.edu/gsea>) will assess whether the potential pathways involved in EMT, identified in the previous analysis, will be confirmed using a different computational approach. The key genes, representing the highly connected hubs in the identified networks, are assessed by qRT-PCR and IHC on selected archival tumor series to determine their diagnostic and prognostic value according to clinical information, lifestyle and dietary intake.
### MiRNAs and their role in modulating EMT {#Sec16}
Validated miRNAs are analyzed by in silico approaches (miRanda, miRbase, TargetScan) to evaluate if EMT genes previously analyzed are included among their putative target genes. miRNAs here identified and their putative target genes are analyzed in BC tumor tissue, positive for canonical mesenchymal markers, such as Vimentin and N-cadherin. Evaluation of EMT markers are carried out by IHC followed by laser micro-dissection. Further functional experiments are performed to validate in silico considerations.
### Risk recurrence in BC by computational modeling {#Sec17}
The development of a computational model based on agent based modelling, differential equations or Petri nets could lead to a validated tool able to predict the efficacy of the dietary regimen. The model is fit with data coming from the diet regiment components, diet effects, life style, and from plasma miRNAs analysis. miRNAs expression allows the model to be tuned to find evidences that may be relevant to predict the time and the probability of recurrence risk.
Sample size {#Sec18}
-----------
Considering a 20% recurrence rate within 3 years in most collaborating centres for BC cases specified in Table [2](#Tab2){ref-type="table"}, and a predicted rate of 10% in the high intensity arm, with power of 80% and two-sided alpha of 0.05, the number of subjects are 506 (*n* = 253 in each arm).
Statistical analyses {#Sec19}
--------------------
All randomized patients will be analysed considering the "intention-to-treat" principle (ITT analysis). Results will be expressed as percentages or means ± SEM or 95% confidence intervals (CI).
### Primary analyses {#Sec20}
The primary analyses will assess the between-treatment difference in BC recurrence measured as disease-free survival (DFS), calculated as the percentage of patients alive without recurrence of disease at study end (up to 33 months from randomization). BC recurrence is defined as the relapse of the disease or metastasis either in the same breast (including new positive lymph nodes), or the opposite breast or in distant organs. The duration of DFS in patients lost at follow-up will be censured at the date of the last day the patient was considered free of disease. Between-treatment differences in DFS will be analyzed by *log-rank* test. Kaplan-Meier curves will be provided to estimate median DFS and 95% CI. To address the impact of potential imbalance in prognostic factors we will repeat the primary analysis using the *log-rank* test stratified by stage, age, family history of BC, time since diagnosis, molecular subtype, medication use, smoking, baseline waist circumference and dietary variables (baseline GI, dietary fiber, saturated fat, vegetables/fruit, meat, sweets/desserts). Missing data for covariates will be handled by using the missing indicator method.
### Sensitivity analyses {#Sec21}
To assess the robustness of our ITT primary analysis with possible missing data we will repeat the primary analysis using completers data only as well as per-protocol data only, using multiple imputation method to generate missing data in the stratification step. To assess the impact of participant-level factors on the primary outcome we will examine DFS separately in those who, at study end, have reached normo-glycemia (\<110 mg/dl) and normal levels of HbA1c \<6.0% versus those who have higher levels, in those who reached circulating 25(OH)D ≥60 ng/ml versus those who reached ≤30 ng/ml, in those who have lower versus those who have higher insulin levels (median will be used as cut-offs), in those who have good overall compliance on the three treatment components that is in the highest quantile of: low GI + number of daily steps + levels of 25(OH)D, versus those who have lower compliance. For these analyses study end is defined as the mean of the last three visits.
### Secondary analyses {#Sec22}
The mean and standard error for each of the following variables will be determined for each study group. The change from baseline to end of study will be compared between groups using repeated measures ANOVA and using nonparametric tests if necessary, for the following variables: body weight, waist circumference, BMI, blood pressure, serum levels of 25(OH)D, blood glucose, HbA1c, insulin, IGF-1, lipids, CRP, estradiol, testosterone, SHBG, specific microRNAs, dietary glycemic index, number of steps per day and quality of life. We will also test treatment differences in medication use and medication side effects. Chi-square test will be used to compare categorical variables and Student t-test or Wilcoxon test for continuous variables. The association between each variable and prognosis will be analyzed using Cox proportional hazard model and logistic regression. Finally, we will assess whether these secondary analyses are different at year 1 compared to end of study. All statistical analyses will be conducted with SPSS statistical software version 23.0 (SPSS Inc., Chicago IL, USA).
Potential toxicity and adverse events {#Sec23}
-------------------------------------
Toxicity due to diet, moderate exercise or vitamin D supplementation are not expected. The diet is an enhanced Mediterranean diet with one arm consuming more low GI foods and physical activity involves moderate exercise of 30 min of daily brisk walk and both respond to lifestyle principles suggested by cancer guidelines such as the American Cancer Society \[[@CR8]\] and the World Cancer Research Fund \[[@CR9]\]. Oral vitamin D (cholecalciferol) supplementation will be given at safe dosages (up to 4000 IU/day) to reach safe blood levels (60 ng/ml for the test arm and 30 ng/ml for the control arm) \[[@CR53], [@CR54]\]. In Italy the normal range of 25(OH)D is between 30-100 ng/ml. Excess vitamin D levels (\>100 ng/ml) could induce excess calcium absorption from the intestine and potentially result in hypercalcemia (\>11 mg/dl). Both serum 25(OH)D and calcium levels are monitored throughout the study at each clinic visit and any signs or symptom of hypervitaminosis D (hypercalcemia, excess thirst, etc.) will be recorded and immediately communicated to the participant's physician and supplementation stopped. All adverse events occurring after signing the consent form will be recorded in a specific Adverse Event Form.
Ethical considerations {#Sec24}
----------------------
The study will determine which group will benefit the most however both treatments are expected to gain some health benefits since both treatments are based on a healthy diet and lifestyle and sufficient vitamin D levels. An interim analysis will be conducted to evaluate whether there are excessive disadvantages for one group over the other. Should this happen the study would be terminated and both groups allowed to follow the most beneficial treatment program. Participants health complications will be dealt by the research team of doctors and by contacting patients' physicians. Should any complication put the participant at risk by continuing the study the participant will be invited to withdraw from the study. The study participants will also have the right to withdraw from the study anytime and for any personal reason without jeopardizing their health care in the study institution or any other institution.
Discussion {#Sec25}
==========
The purpose of this study is to reduce BC recurrence and hence increase disease-free survival through a lifestyle program that includes a low glycemic index diet, physical activity and vitamin D supplementation in women with BC living in a Mediterranean country. It is expected that the higher intensity program of low GI diet, exercise and vitamin D will reduce BC recurrence by 50% compared to the lower intensity program.
Dietary clinical trials to reduce BC recurrence have been conducted previously. In the USA two large studies were conducted, the Women's Intervention Nutrition trial (WIN) and the Women's Healthy Eating and Living trial (WHEL). The WIN study which focused on a low-fat diet found a 24% reduction in BC recurrence after 5 years compared to a control of minimal dietary counselling \[[@CR10]\], whereas the WHEL study focusing on a combination of low-fat and high-fruit and -vegetable diet did not reduce BC recurrence after a 7-year intervention compared to a lower intensity fruit and vegetables advice \[[@CR11]\]. This null result may be partly explained by methodology aspects as most women had early stage BC and one of the inclusion criteria was the diagnosis of BC within 4 years. It is possible that such a diet may be protective if consumed earlier, possibly within 1 year of diagnosis as in the WIN study \[[@CR10]\]. However, in the WHEL study, women who at baseline consumed more than 3 servings of vegetables per day showed a 30% reduction in BC recurrence and an even greater reduction (up to 52%) in tamoxifen users \[[@CR55]\]. Furthermore, women who at baseline consumed at least 5 portions of vegetables/fruit per day and additionally walked for at least 30 min a day had a 44% higher chance of survival, independently of obesity, suggesting that obesity did not impact on lowering survival if people adhered to a healthy lifestyle pattern \[[@CR56]\]. Overall these large clinical trials suggest the importance of intervening early after diagnosis and that plant-based diets and half hour daily walk may be protective from future recurrence. The low fat advice may however be appropriate only in countries with high intakes of saturated fat. In Mediterranean countries where diets are characterized by high intakes of olive oil, reducing fat intakes may not be beneficial. In the PREDIMED intervention trial, women without BC but following the intensive Mediterranean dietary advice with high olive oil intakes showed up to 68% lower risk of BC compared with those on lower olive oil intake \[[@CR57]\]. The traditional Mediterranean diet, rich in plant food and olive oil, has been associated with protection from BC risk \[[@CR58], [@CR59]\] CVD events \[[@CR60]\], type two diabetes risk \[[@CR61]\] and complications \[[@CR62], [@CR63]\] and favour weight loss \[[@CR64]\]. However, adherence to a traditional Mediterranean diet has halved from the 60's to 2003 in Mediterranean countries including Italy \[[@CR65]\], hence there may be health benefits in improving the current Mediterranean style dietary pattern. The Italian diet is very rich in carbohydrates, particularly bread and other fast absorbing carbohydrates (high GI foods) which have been associated with higher risk of BC as well as other chronic conditions \[[@CR15], [@CR16], [@CR18], [@CR66]\]. Since factors influencing the metabolism of glucose may play a relevant role in the development of chronic diseases including BC \[[@CR15]\], it is possible that lowering the GI of the Mediterranean diet of Italian women with BC through guided dietary advice, a lower recurrence may be achieved. Chronic elevation of insulin concentrations may be one of the mechanisms explaining the positive association between dietary GI and cancer risk \[[@CR67]\]. High GI diets increase blood glucose and insulin levels more than low GI diets and hence may be involved in increasing IGF-1 bioavailability \[[@CR31]\]. Insulin is an anabolic hormone able to increase IGF-1 synthesis and activity and IGF-1 in turn may promote cancer development by inhibiting apoptosis, stimulating cell proliferation and sex-steroid synthesis \[[@CR68], [@CR69]\]. Another mechanism for high GI-related increased cancer risk may be through hyperglycemia-induced oxidative stress \[[@CR70], [@CR71]\] which has been implicated in free radical-dependent DNA damage, known as a contributor to carcinogenesis \[[@CR72], [@CR73]\]. Within normal ranges of glycemia higher levels of normal have been directly associated with BC risk in previously healthy women \[[@CR74]\] and with higher recurrence rates in women with BC \[[@CR75]\]. Another potential mechanism may be through lowering the availability of circulating glucose following low GI diets and exercise. Cancer cells are avid consumers of glucose due to their altered metabolism characterized by insufficient oxidative phosphorylation and compensatory glucose fermentation (the Warburg effect) \[[@CR76]\]. This results in lactic acid production and higher protons which acidify the external cellular microenvironment reaching a pH of 6.5-6.9 \[[@CR77]\]. A lower pH may provide a competing survival and metastatic advantage for cancer cells (e.g. greater spreading capacity) over normal cells unable to survive below a pH of 7.2 \[[@CR77], [@CR78]\] and may induce drug resistance of weak base-anticancer drugs (e.g. doxorubicin) \[[@CR78]\].
Physical activity after BC diagnosis has also been shown to reduce the risk of BC mortality by 40-50% particularly when it is of moderate intensity such as 30 minutes of brisk walking per day \[[@CR32]--[@CR35]\]. This effect may be modulated partly by reduced insulin, insulin-like growth factors and estrogen levels \[[@CR36], [@CR79], [@CR80]\] which are associated with BC recurrence and death \[[@CR81], [@CR82]\]. Furthermore, physical activity can improve insulin sensitivity, reduce blood triglycerides, blood pressure and body fat \[[@CR83]--[@CR85]\]. The lifestyle changes proposed in this study (lower dietary GI, enhanced Mediterranean diet, physical activity) may also induce weight loss which in turn may improve insulin sensitivity, IGF profile and reduce aromatase activity in adipose tissue with consequent reduction in estrogen levels \[[@CR86]\], particularly relevant in postmenopausal women \[[@CR87]\]. Although weight loss is not a goal of the DEDiCa study, participants will be allowed to lose weight in situations of overwheight or obesity should they wish to. The secondary statistical analyses will take into consideration this aspect.
Lifestyle changes may be more efficacious in a setting of vitamin D sufficiency. Vitamin D alters genes implicated in cellular growth, through upregulation of E-cadherin thereby stimulating cell differentiation and apoptosis \[[@CR39]\]. Higher serum vitamin D levels (\>30 ng/ml) in BC patients have been associated with 50% less fatality rates compared to lower levels (\<20 ng/ml) \[[@CR88]\]. The vitamin D dose--response relationship for BC appears inverse and linear up to 60 ng/ml \[[@CR39]\]. These levels could achieve a 25% lower BC incidence and could be reached by supplementing 2000--4000 IU/day \[[@CR53], [@CR89]\]. The upper dose recommended by the National Academy of Sciences is 4000 IU/day \[[@CR54]\]. The Italian guidelines for the prevention and treatment of hypovitaminosis D (SIOMMMS) indicate that in Italy this condition is present in 50% of young adults and at higher rates in older individuals \[[@CR53]\]. The updated SIOMMMS guidelines for vitamin D deficiency (25-OH-D \< 10 ng/ml) suggest up to 600,000 as a cumulative dose \[[@CR90]\]. In Italy the normal range for circulating 25-OH-D is set between 30 ng/ml and 100 ng/ml while toxicity levels are considered above 150 ng/ml \[[@CR53]\]. Vitamin D can also protect against bone loss and risk of fractures as a consequence of osteoporosis typically seen after postmenopause or after estrogen deprivation therapy used in BC treatment. Hence vitamin D supplementation may also reduce adverse skeletal effects \[[@CR53]\]. Furthermore, higher vitamin D levels have also been associated with reduced risk of developing diabetes \[[@CR91], [@CR92]\]. Continuous improvements in survival rates will have an impact on comorbidities and quality of life of BC survivors. Co-morbid conditions have been found at higher prevalence in cancer survivors than in age-matched controls \[[@CR93]\] and in the subgroup of cancer patients presenting with two comorbidities, the most frequent combination of diseases appeared to be CVD in men and diabetes in women \[[@CR94]\]. There are downstream effects of cancer therapy (e.g. radiation and chemotherapy) causing heart, respiratory, kidney and memory problems but also of hormone suppressing therapy in BC patients \[[@CR95]\]. Compounding on this problem, cancer survivors fail to receive the same level of care for their comorbid condition compared to the general population \[[@CR96]\] and this is especially true for type two diabetes \[[@CR97]\]. Hence it may become even more relevant to implement cost-effective lifestyle risk reduction strategies. Two international cancer organizations have published lifestyle guidelines for cancer survivors, The American Cancer Society (ACS) \[[@CR8]\] and the World Cancer Research Fund (WCRF) \[[@CR9]\], however there is no agreement on a specific lifestyle program that could reduce BC recurrence and complications including recommendations on vitamin D supplementation.
DEDiCa study includes treatment components such as low GI, traditional Mediterranean foods, high dietary fiber and physical activity which have been shown to reduce risk factors for type two diabetes and CVD including HbA1c, blood glucose and lipids, and inflammatory factors \[[@CR15], [@CR20], [@CR22], [@CR23], [@CR25]--[@CR28], [@CR30], [@CR60], [@CR98]--[@CR100]\].
This study will contribute to understanding the efficacy of lifestyle changes in a Mediterranean population of BC survivors with respect to several novel aspects: testing a lifestyle modification (diet and exercise) within normal vitamin D levels on disease-free survival; investigate lifestyle changes in relation to BC staging, molecular subtypes, menopausal status, body weight, CVD risk factors and events, diabetes control, new diabetes cases, quality of life, response to medication. It will also allow to understand breast carcinogenesis and the role of microRNA in BC and whether they are modulated by dietary and other lifestyle aspects. It will allow to investigate best time of treatment adherence since women will be enrolled within 12 months of surgery, some will have just started their cancer therapy and some will have ended.
Given the supporting evidence of important health effects and safety of the components of DEDiCa intervention we believe it is feasible and urgent to test this program in BC patients.
25(OH)D
: 25-hydroxyvitamin D
AST/ALT
: Aspartate transaminase/alanine transaminase
BC
: Breast cancer
CRP
: C-reactive protein
CVD
: Cardiovascular disease
GI
: Glycemic index
HbA1c
: Hemoglobin A1c
HDL-C
: High density lipoprotein cholesterol
HIT
: High intensity treatment
IGF-1
: Insulin-like growth factor-1
LDL-C
: Low density lipoprotein cholesterol
LITE
: Low intensity treatment
RNA
: Ribonucleic acid
SHBG
: Sex hormone binding globulin
We thank Barilla Spa for providing pasta, Abiogen Pharma for providing vitamin D, and Lega Italiana per la Lotta Contro i Tumori (LILT Nazionale) Progetto Cinque Per Mille 2013 and LILT Catania for their support.
Funding {#FPar1}
=======
This trial is funded by a grant of the Italian Ministry of Health (grant no. PE-2013-02358099). The funding body will not play any role in the study.
Availability of data and material {#FPar2}
=================================
None declared.
Authors' contributions {#FPar3}
======================
LSA, ML, AC, MG and MM contributed to the protocol and grant proposal. LSA, CWCK, DJAJ and GR contributed to the dietary aspect of the intervention program design, MP and LSA contributed to the exercise aspect of the intervention program design, DG and LSA contributed to the vitamin D aspect of the intervention program design. MDL, MR, MDA, FC, GB, FF, RR, DS, SM, GT, GC contributed to patient selection criteria. EC, LSA, MM and ML contributed to the biochemical analysis plan. AC, EV and EB were primarily involved in developing the statistical analysis plan. LSA wrote the manuscript. The manuscript was amended based on comments from all authors. All authors read and approved the final manuscript.
Competing interests {#FPar4}
===================
LSA has received honoraria from the Nutrition Foundation of Italy (NFI, Milan) and from Lega Italiana per la Lotta Contro i Tumori (LILT-Catania). DG has received speaking and/or consulting fees from Abiogen Pharma, Amgen, Eli-Lilly, Janssen-Cilag, Merck and Mundipharma. MP has received research support from Amgen. GR has received research grants from Barilla Spa to his University Department and speaker and travel fees from Sanofi, Astra Zeneca and Takeda. CWCK has received research support from the Advanced Foods and Material Network, Agrifoods and Agriculture Canada, the Almond Board of California, the American Pistachio Growers, Barilla, the California Strawberry Commission, the Calorie Control Council, CIHR, the Canola Council of Canada, the Coca-Cola Company (investigator initiated, unrestricted grant), Hain Celestial, the International Tree Nut Council Nutrition Research and Education Foundation, Kellogg, Kraft, Loblaw Companies Ltd., Orafti, Pulse Canada, Saskatchewan Pulse Growers, Solae and Unilever. He has received travel funding, consultant fees and/or honoraria from Abbott Laboratories, the Almond Board of California, the American Peanut Council, the American Pistachio Growers, Barilla, Bayer, the Canola Council of Canada, the Coca-Cola Company, Danone, General Mills, the International Tree Nut Council Nutrition Research and Education Foundation, Kellogg, Loblaw Companies Ltd., the Nutrition Foundation of Italy, Oldways Preservation Trust, Orafti, Paramount Farms, the Peanut Institute, PepsiCo, Pulse Canada, Sabra Dipping Co., Saskatchewan Pulse Growers, Solae, Sun-Maid, Tate and Lyle, and Unilever. He has served on the scientific advisory board for the Almond Board of California, the International Tree Nut Council, Oldways Preservation Trust, Paramount Farms and Pulse Canada. He is a member of the International Carbohydrate Quality Consortium (ICQC), Executive Board Member of the Diabetes and Nutrition Study Group (DNSG) of the European Association for the Study of Diabetes (EASD), is on the Clinical Practice Guidelines Expert Committee for Nutrition Therapy of the EASD and is a Director of the Toronto 3D Knowledge Synthesis and Clinical Trials foundation. DJAJ has received research grants from Saskatchewan Pulse Growers, the Agricultural Bioproducts Innovation Program through the Pulse Research Network, the Advanced Foods and Material Network, Loblaw Companies Ltd., Unilever, Barilla, the Almond Board of California, Agriculture and Agri-food Canada, Pulse Canada, Kellogg's Company, Canada, Quaker Oats, Canada, Procter & Gamble Technical Centre Ltd., Bayer Consumer Care, Springfield, NJ, Pepsi/Quaker, International Nut & Dried Fruit (INC), Soy Foods Association of North America, the Coca-Cola Company (investigator initiated, unrestricted grant), Solae, Haine Celestial, the Sanitarium Company, Orafti, the International Tree Nut Council Nutrition Research and Education Foundation, the Peanut Institute, the Canola and Flax Councils of Canada, the Calorie Control Council (CCC), the CIHR, the Canada Foundation for Innovation and the Ontario Research Fund. He has been on the speaker's panel, served on the scientific advisory board and/or received travel support and/or honoraria from the Almond Board of California, Canadian Agriculture Policy Institute, Loblaw Companies Ltd, the Griffin Hospital (for the development of the NuVal scoring system, the Coca-Cola Company, EPICURE, Danone, Saskatchewan Pulse Growers, Sanitarium Company, Orafti, the Almond Board of California, the American Peanut Council, the International Tree Nut Council Nutrition Research and Education Foundation, the Peanut Institute, Herbalife International, Pacific Health Laboratories, Nutritional Fundamental for Health, Barilla, Metagenics, Bayer Consumer Care, Unilever Canada and Netherlands, Solae, Kellogg, Quaker Oats, Procter & Gamble, the Coca-Cola Company, the Griffin Hospital, Abbott Laboratories, the Canola Council of Canada, Dean Foods, the California Strawberry Commission, Haine Celestial, PepsiCo, the Alpro Foundation, Pioneer Hi-Bred International, DuPont Nutrition and Health, Spherix Consulting and WhiteWave Foods, the Advanced Foods and Material Network, the Canola and Flax Councils of Canada, the Nutritional Fundamentals for Health, Agri-Culture and Agri-Food Canada, the Canadian Agri-Food Policy Institute, Pulse Canada, the Saskatchewan Pulse Growers, the Soy Foods Association of North America, the Nutrition Foundation of Italy (NFI), Nutra-Source Diagnostics, the McDougall Program, the Toronto Knowledge Translation Group (St. Michael's Hospital), the Canadian College of Naturopathic Medicine, The Hospital for Sick Children, the Canadian Nutrition Society (CNS), the American Society of Nutrition (ASN), Arizona State University, Paolo Sorbini Foundation and the Institute of Nutrition, Metabolism and Diabetes. He received an honorarium from the US Department of Agriculture to present the 2013 W.O. Atwater Memorial Lecture. He received the 2013 Award for Excellence in Research from the International Nut and Dried Fruit Council. He received funding and travel support from the Canadian Society of Endocrinology and Metabolism to produce mini cases for the Canadian Diabetes Association (CDA). He is a member of the International Carbohydrate Quality Consortium (ICQC). His wife is a director and partner of Glycemic Index Laboratories, Inc., and his sister received funding through a grant from the St. Michael's Hospital Foundation to develop a cookbook for one of his studies. All other authors declare that they have no competing interests.
Consent for publication {#FPar5}
=======================
Not applicable.
Ethics approval and consent to participate {#FPar6}
==========================================
This study was approved by the ethic board of: the Italian Medicine Agency (AIFA), the National Cancer Institute "Fondazione Giovanni Pascale" in Naples, Azienda Ospedaliera per L'Emergenza Ospedale Cannizzaro in Catania, Azienda Ospedaliera Universitaria Policlinico "G. Martino" in Messina for San Vincenzo Hospital of Taormina, Comitato Etico Campania Centro ASL NA1 Centro for Clinica Mediterranea in Naples. Written informed consent will be obtained from each recruited patient before the intervention and questionnaire survey.
| {
"pile_set_name": "PubMed Central"
} |
[Fig. 5](#F5){ref-type="fig"} inadvertently appeared in black and white. The correct color version appears below.
{#F5}
| {
"pile_set_name": "PubMed Central"
} |
DNA double-strand break (DSB) repair is critical for chromosomal maintenance. There are several proposed pathways for homology-directed DSB repair including single-strand annealing, synthesis-dependent strand annealing, break-induced replication and Holliday junction pathways[@b1]. All recombinational DSB repair pathways have at least one step in common: the pairing of a broken DNA end with an intact part of a homologous chromosome. This central DNA strand invasion step is carried out by a family of enzymes that include the bacterial RecA and the eukaryotic Rad51 proteins[@b2][@b3]. While many of the functions of gene products involved in various DSB pathways are known, there are still key players, even in bacteria, which remain enigmatic. One such mysterious protein critical to DSB repair in bacteria is the RecN protein.
The RecN protein is a member of the structural maintenance of chromosomes (SMC) family of proteins. SMC proteins have important functions in a variety of housekeeping DNA processes including chromosomal condensation, sister chromatid cohesion and recombinational DNA repair[@b4]. The *Deinococcus radiodurans* RecN protein was shown previously to act as a cohesin mediating the intermolecular tethering of DNA molecules[@b5], confirming an SMC protein-like function for a bacterial recombination enzyme. The SMC-like architecture of *D. radiodurans* RecN has further been confirmed by X-ray structure analysis[@b6].
*Escherichia coli* RecN is involved in the recombinational repair of DNA damage and is likely functioning in both the RecFOR- and RecBCD-dependent pathways[@b7]. RecN forms discrete foci in response to DSBs in bacteria[@b8][@b9][@b10] and the repair of DBSs is severely hampered in the absence of RecN[@b11][@b12][@b13]. But, what is the function of the RecN protein in recombinational DNA repair pathways? *Bacillus subtilis* RecN protein is thought to coordinate damaged DNA and recombination proteins early in the DNA damage response[@b14][@b15] and RecN appears important for ultraviolet-induced compaction of the *E. coli* nucleoid[@b16]. Further, strains harbouring *recN* mutants are more sensitive to ionizing radiation or multiple, site-specific DSBs than to a single site-specific DSB in *E. coli.*[@b11] These observations are consistent with a pre-recombination, chromosomal maintenance role for RecN protein.
The DNA tethering activity of RecN likely provides an important utility leading to DNA repair. However, evidence also suggests that RecN acts with or affects the RecA protein. First, *E. coli recN recA* double mutants are no more sensitive to DSBs than *recA* single mutants[@b11], consistent with RecN involvement in RecA-mediated recombinational DNA repair. And further, it appears that RecA recruits RecN to DSBs in *E. coli*[@b9] suggesting a role for RecN after RecA accumulates at a the site of damage. There are no published accounts that the *E. coli* RecN protein has been purified for biochemical studies, although not for a lack of trying. *E. coli* RecN is highly susceptible to proteolytic breakdown and is generally insoluble in aqueous solutions[@b17][@b18]. In the current study, we demonstrate the effect of the RecN protein on RecA-mediated recombination using a biochemical system and the purified *D. radiodurans* proteins. Here we report a significant stimulation of the RecA-mediated DNA strand invasion step of recombination by the RecN protein. Further, we establish the order of assembly of RecN and RecA proteins that leads to a functional interaction between these proteins. And, finally, we describe the influence of RecA on the ATP hydrolysis activity of the RecN protein.
Results
=======
RecN protein stimulates RecA-mediated DNA strand exchange
---------------------------------------------------------
The RecN protein can actively reduce the in-solution distance between DNA molecules[@b5][@b19][@b20]. We reasoned that intermolecular DNA tethering might stimulate RecA-mediated DNA strand exchange under dilute, suboptimal reaction conditions. Therefore, the effect of *D. radiodurans* RecN protein on RecA activity was assayed *in vitro* using a modified DNA strand exchange reaction ([Fig. 1a](#f1){ref-type="fig"}). The reaction is carried out in the order typically used to assess RecA-promoted DNA strand exchange. However, under the dilute DNA conditions (see Methods) used for reactions shown in [Fig. 1c](#f1){ref-type="fig"}, RecA protein only promotes a small amount of nicked, circular duplex product compared to the reaction carried out under standard (undiluted) DNA conditions (compare lane 2 of **c** with lane 4 of **b** in [Fig. 1](#f1){ref-type="fig"}). We found that the addition of RecN protein to this reaction greatly stimulates product formation in a concentration-dependent manner ([Fig. 1c](#f1){ref-type="fig"}). The stimulation is observed with as little as 250 nM RecN protein and saturates with the addition of 1 μM RecN protein. The addition of 500 nM RecN protein stimulates product formation by RecA protein more than threefold under these conditions ([Fig. 1d](#f1){ref-type="fig"}). RecN protein alone does not promote DNA strand exchange (lane 9, [Fig. 1c](#f1){ref-type="fig"}).
RecN protein stimulates RecA-mediated D-loop formation
------------------------------------------------------
As described in the introduction, evidence suggests that RecN may function at an early step in the DSB repair pathway. [Figure 1](#f1){ref-type="fig"} shows that RecN indeed stimulates RecA-promoted DNA strand exchange *in vitro*. This enhancement is likely accomplished by RecN bringing the recombining substrates closer together enabling RecA to find and pair the homologous DNA molecules. To clearly show that RecN acts early in the reaction, we confirmed the stimulatory effect of adding RecN to a RecA-mediated displacement loop (D-loop) reaction ([Fig. 2](#f2){ref-type="fig"}). D-loops form when RecA filament promotes synapsis of a 3′-overhang on the probing DNA sequence with a complementary region of a homologous target duplex DNA molecule ([Fig. 2a](#f2){ref-type="fig"}). This type of reaction is thought to mimic the initial pairing step of recombinational DSB repair[@b3]. Products are observable independent of the extensive DNA branch migration required to detect the products of the DNA strand exchange reaction shown in [Fig. 1](#f1){ref-type="fig"}. The inefficient D-loop formation mediated by the RecA protein alone (lane 4, [Fig. 2b](#f2){ref-type="fig"}) is significantly enhanced by the addition of the RecN protein (lane 5, [Fig. 2b,c](#f2){ref-type="fig"}). RecN does not promote DNA pairing in the absence of RecA (see [Fig. 3b](#f3){ref-type="fig"} below). Interestingly, the RecN stimulation of RecA-mediated D-loop formation is not observed when a RecN ATPase-deficient mutant (RecN K67A) is used in the reaction (lane 6, [Fig. 2b](#f2){ref-type="fig"}). This indicates that RecN ATP hydrolysis is required for the stimulation of RecA under these conditions. The requirement for ATP hydrolysis will be explored further below.
DNA pairing stimulated by RecN protein is species-specific
----------------------------------------------------------
The RecN-dependent increase in DNA-pairing products promoted by the RecA protein shown in [Figs 1](#f1){ref-type="fig"} and [2](#f2){ref-type="fig"} can be explained by a simple model in which the cohesin-like activity of RecN[@b5] reduces the distance between substrate DNA molecules. This model predicts that RecN protein should enhance the DNA-pairing activity of any RecA recombinase. We tested the DNA strand exchange activity of the *E. coli* RecA protein in the presence of RecN ([Fig. 3](#f3){ref-type="fig"}). Surprisingly, we observed no stimulation of the *E. coli* RecA protein by the RecN protein using either the diluted DNA strand exchange reaction ([Fig. 3a](#f3){ref-type="fig"}) or the D-loop assay ([Fig. 3b](#f3){ref-type="fig"}). This negative result may reflect the minor mechanistic differences between the *E. coli* RecA and *D. radiodurans* RecA proteins previously described[@b21][@b22]. It may also reflect a species-specific interaction. We explored the intriguing possibility of a direct interaction between purified RecA and RecN proteins using a pull-down strategy ([Fig. 3c](#f3){ref-type="fig"}). The RecA and RecN proteins (2 μM each) were incubated in the presence or absence of DNA. Diluted mixtures were loaded onto RecN or RecA antibody covalently coupled resin, and unbound proteins were removed. On elution, RecA and RecN co-eluted suggesting a physical interaction. The RecA and RecN proteins do not interact nonspecifically with RecN and RecA antibodies, respectively ([Supplementary Fig. 1](#S1){ref-type="supplementary-material"}). A direct interaction may explain the fact that *D. radiodurans* RecN does not stimulate the DNA strand exchange reactions catalysed by *E. coli* RecA protein ([Fig. 3a,b](#f3){ref-type="fig"}), but does stimulate its cognate RecA protein ([Figs 1](#f1){ref-type="fig"} and [2](#f2){ref-type="fig"}).
RecA bound to DNA stimulates RecN ATP hydrolysis
------------------------------------------------
The role of ATP hydrolysis by the RecN protein and cohesins in general is not well understood. Robert Lloyd and colleagues have shown that *recN* ATPase-deficient mutant bacterial strains (replacing lysine with alanine in the Walker A motif) have phenotypes that mimic that of a *recN* null mutant[@b23]. The purified RecN K67A mutant protein from *D. radiodurans* exhibits no measurable ATPase activity but is not defective in DNA binding or cohesin-like DNA-bridging activity ([Supplementary Fig. 2](#S1){ref-type="supplementary-material"}). We were intrigued by the observation that RecN K67A mutant protein does not stimulate RecA-mediated DNA pairing (lane 6, [Fig. 2b](#f2){ref-type="fig"}) especially given the fact ATP nucleotide is not required for the wild-type RecN protein to carry out cohesin-like DNA-bridging activity[@b5]. In an attempt to understand the role of ATP hydrolysis, we explored the effect of RecA protein on the ATP hydrolysis activity of RecN ([Fig. 4](#f4){ref-type="fig"}). We showed previously that the rate of ATP hydrolysis catalysed by the RecN protein is not stimulated by single-stranded DNA (ssDNA), but is stimulated by linear, and relaxed or supercoiled circular duplex DNA[@b5]. The experiments shown in [Fig. 4](#f4){ref-type="fig"} are carried out at a sub-optimal concentration ratio of RecN to DNA such that the DNA-independent and -dependent ATP hydrolysis rates are similar (reactions 1 and 2 of [Fig. 4](#f4){ref-type="fig"}, [Table 1](#t1){ref-type="table"} and ref. [@b5]). Since RecA also hydrolyses ATP, we utilized the ATPase-deficient *D. radiodurans* RecA Walker A mutant protein (RecA K83R) to test the effect of added RecA protein under the conditions previously optimized for RecN activity[@b5]. We observed a three- to fivefold increase in the RecN ATP hydrolysis rate when a ternary mixture of RecN (2 μM), RecA K83R (2 μM) and duplex DNA (50 μM) are present in the reaction mixture (compare reactions 3, 4 and 5 with reactions 1 and 2, [Fig. 4](#f4){ref-type="fig"} and [Table 1](#t1){ref-type="table"}). This indicates that RecN ATPase activity is enhanced in the presence of RecA protein. Data in [Supplementary Fig. 3](#S1){ref-type="supplementary-material"} confirm that the stimulation of RecN ATP hydrolysis is species-specific and occurs under conditions optimized for RecA protein activity.
The RecN ATPase rate increase is more instructive when we measure the time required for RecN protein to reach a steady-state rate of ATP hydrolysis (lag time) as a function of protein order of addition. When RecN is allowed to interact with the DNA for 20 min before the addition of RecA, a lag of 25 min is observed (reaction 3, [Fig. 4](#f4){ref-type="fig"} and [Table 1](#t1){ref-type="table"}). The lag increases to more than 30 min when RecN and RecA are pre-incubated before the addition of duplex DNA (reaction 4, [Fig. 4](#f4){ref-type="fig"}). In this case, there is a small but reproducible decrease in the DNA-independent ATPase rate of RecN before the ternary mixture is complete by the addition of DNA (compare reactions 1 and 4, [Fig. 4](#f4){ref-type="fig"}). This may be due to the interactions between RecA and RecN proteins. Finally, the lag in RecN ATP hydrolysis described above is almost eliminated if RecN protein is added to DNA pre-incubated with RecA protein (reaction 5, [Fig. 4](#f4){ref-type="fig"}). The final, steady-state rate of RecN-catalysed ATP hydrolysis is similar in all three orders of addition ([Table 1](#t1){ref-type="table"}).
The most likely explanation for the lag measured when either RecN is pre-incubated with DNA before the addition of RecA (reaction 3, [Fig. 4](#f4){ref-type="fig"}) or RecN is pre-incubated with RecA before the addition of DNA (reaction 4, [Fig. 4](#f4){ref-type="fig"}) is that a reorganization of proteins is occurring such that RecN is dissociating from DNA or RecA so that RecA is free to bind the DNA. And, the increase in RecN activity only occurs when RecN interacts with DNA that is bound with RecA. Consequently, we observe no lag to RecN ATP hydrolysis when RecN is added to preformed RecA filaments. These data suggest that RecN ATPase activity may be needed for a step in the DSB repair pathway after RecA is loaded onto the DNA.
RecN ATPase is stimulated under RecA D-loop conditions
------------------------------------------------------
To gain insight into the role of ATP hydrolysis, RecN ATPase activity was measured under D-loop assay reaction-optimized conditions (see Methods) using the RecA K83R ATPase-deficient mutant ([Fig. 5b](#f5){ref-type="fig"}[](#f6){ref-type="fig"}). The RecA K83R mutant protein is proficient in mediating D-loop formation, as has been shown for other RecA Walker A mutant homologues ([Supplementary Fig. 4](#S1){ref-type="supplementary-material"} and ref. [@b24]). Under the conditions used, the background RecN ATPase activity is very low (∼2 μM min^−1^, [Table 2](#t2){ref-type="table"}) in the absence of RecA when the D-loop probe DNA, the target DNA or both DNA substrates ([Fig. 5b](#f5){ref-type="fig"}, reactions 1, 2 and 3, respectively) are included in the reaction. For reactions containing RecA, the DNA and proteins are assembled as indicated in [Fig. 5a](#f5){ref-type="fig"}, and individual reactions in [Fig. 5b](#f5){ref-type="fig"} indicate the protein(s) and the DNA substrate(s) added. Reaction assemblies are also noted in [Table 2](#t2){ref-type="table"}. The rate of RecN ATP hydrolysis ([Table 2](#t2){ref-type="table"}) increases in the presence of RecA and either D-loop probe DNA (reaction 4) or D-loop target DNA (reaction 5) and the highest rate is observed when the complete D-loop assay components of both DNA substrates are present (reaction 6). The RecN ATPase rate measured under complete D-loop reaction conditions (reaction 6) represents a 20-fold stimulation ([Table 2](#t2){ref-type="table"}) compared to the same reaction in the absence of RecA protein (reaction 3). This enhanced rate of RecN ATP hydrolysis under the D-loop reaction conditions is largely dependent on both RecA protein bound to the probe DNA and the presence of the target DNA.
We asked whether the ATPase activity of RecN is dependent on homology. The RecA-dependent D-loop assay was repeated with non-homologous target DNA (supercoiled phage *φ*X174 RF1) and RecN ATPase activity was monitored (reaction 7, [Fig. 5b](#f5){ref-type="fig"}). Interestingly, the measured rate of RecN ATP hydrolysis is the same whether the target DNA is homologous or heterologous to the probe DNA ([Table 2](#t2){ref-type="table"}). It appears that the enhancement of RecN ATPase observed during the D-loop reaction may not be concomitant with the RecA-mediated DNA-pairing reaction. This suggests that the RecN ATPase activity is required for a presynaptic step of the reaction after RecA protein binds to the probe DNA.
Target DNA concentration and length affects RecN ATPase
-------------------------------------------------------
The difference in the measured rate of RecN hydrolysis in RecA--RecN--DNA ternary complexes with the D-loop probe DNA compared to the D-loop target DNA ([Table 2](#t2){ref-type="table"}) was unexpected (compare reactions 4 and 5, [Fig. 5b](#f5){ref-type="fig"}). While the two DNAs are present at the same concentration (in micromolar nucleotides), the probe DNA contains a 150-nucleotide ssDNA overhang and the target DNA is supercoiled circular duplex DNA. The RecN ATPase rate in the absence of RecA is the same for both of these DNA molecules (reactions 1 and 2, [Fig. 5b](#f5){ref-type="fig"}). We have previously determined that the double-stranded DNA-dependent rate of ATP hydrolysis is similar whether the duplex DNA cofactor is supercoiled, linearized or relaxed circular, even under conditions optimized for RecN ATP hydrolysis[@b5]. Therefore, we only observe different RecN ATPase rates with the probe DNA and the target DNA when RecA is present in the reaction. A further difference between reactions 4 and 5 of [Fig. 5b](#f5){ref-type="fig"} is the presence of DNA at the time of RecA addition. In reaction 4, RecA is pre-incubated with probe DNA. In reaction 5, RecA is not incubated with DNA before the addition of RecN. Since the RecN hydrolysis rate observed in reaction 5 is much greater than in reaction 4 where RecA is bound to probe DNA, it appears that RecN is not activated for hydrolysis until additional DNA is added to the preformed RecA filaments in the presence of RecN. The only DNA present in reaction 5 is supercoiled DNA and it is possible that the RecA protein is unable to saturate the DNA, so there is free DNA present for RecN to bind to or for RecA/RecN to conduct some unproductive searching. Since the rate of RecN ATP hydrolysis is higher when additional DNA is added as in reaction 6 of [Fig. 5b](#f5){ref-type="fig"}, it is possible that RecN is interacting both with RecA protein bound to probe DNA and to the second DNA strand added (target), as we suggest below in the model of [Fig. 7](#f7){ref-type="fig"}. In reaction 4 of [Fig. 5b](#f5){ref-type="fig"}, the probe DNA (with ssDNA extensions) should be saturated by the RecA protein added and there is much less opportunity to pseudo-search another DNA strand not bound by RecA protein, although there is probably a bit of unbound probe DNA at any point in time. It is possible that to some extent the DNA can act as both probe and target, albeit unproductive due to the four strands present. RecA might not be searching in this scenario, due to the four strands, but RecN may be bringing the two DNA molecules into juxtaposition nonetheless. The idea that RecN--RecA--DNA complexes engaged in unproductive, pseudo-searching also stimulates the RecN ATP hydrolysis is further supported by the fact the two DNAs need not be homologous (as in [Fig. 5b](#f5){ref-type="fig"}, reaction 7).
To determine whether the stimulation of hydrolysis is indeed related to a second DNA strand addition, we carried out DNA titration experiments in the presence of RecA K83R protein. The D-loop probe DNA concentration was held constant at 10 μM and the D-loop target DNA was titrated from 0 to 10 μM producing a clear dependency of the rate of ATP hydrolysis on the target DNA concentration ([Fig. 6a](#f6){ref-type="fig"} and [Table 2](#t2){ref-type="table"}) and less concentration dependence on RecN ATPase rates when the probe DNA was titrated.
In addition to titrating the concentration of target DNA as described above, we also investigated the effect of changing the length of the target DNA ([Fig. 6b](#f6){ref-type="fig"}) added to the same D-loop reaction in [Figs 5](#f5){ref-type="fig"} and [6a](#f6){ref-type="fig"}. When the concentration of target DNA molecules was held constant at 2 nM (closed squares, [Fig. 6b](#f6){ref-type="fig"}), the RecN ATPase activity measured was proportional to the length of target DNA added to the reaction. Since the concentration of DNA base pairs is also increasing as the DNA length increases (see legend to [Fig. 6b](#f6){ref-type="fig"}), we repeated the DNA target length experiment with a constant concentration of DNA (10 μM) of nucleotides (open squares, [Fig. 6b](#f6){ref-type="fig"}). In this case, the measured RecN ATPase rate with DNA target lengths 2.4 kilobase pairs (kbp) or longer were approximately the same (∼50 μM min^−1^). However, the RecN ATP hydrolysis rate is again higher in the presence of these longer target DNAs than when the shorter target DNAs are added, even though the number of target DNA molecules decreases as the length increases. Again, the RecN ATPase activity measured was proportional to the length of target DNA added to the reaction for DNA lengths below 2.4 kbp.
RecN protein is activated for ATP hydrolysis by RecA bound to DNA ([Fig. 4](#f4){ref-type="fig"}) and the rate of that hydrolysis during a D-loop reaction is largely dependent on the concentration of the target DNA ([Fig. 6a](#f6){ref-type="fig"}). This suggests an activation of RecN protein ATPase where the stimulation occurs as a function of the addition of the second DNA strand (the target) to the RecA protein bound to the probe DNA. The target DNA used in [Fig. 5b](#f5){ref-type="fig"}, reactions 5 and 6 is supercoiled plasmid duplex DNA fully homologous to the probe DNA. We also find that the stimulation of RecN ATP hydrolysis under D-loop conditions (reaction 6, [Fig. 5b](#f5){ref-type="fig"}) is observed even when the target DNA is linearized ([Table 2](#t2){ref-type="table"} and [Fig. 6b](#f6){ref-type="fig"}). Therefore, the stimulation of RecN ATPase activity is not unique to a RecA--RecN--supercoiled DNA complex.
Discussion
==========
We conclude that the SMC-like RecN protein has a presynaptic role in RecA-mediated homologous DNA pairing. We present several lines of evidence suggesting the function of RecN is important after RecA has bound DNA and before the DNA-pairing step. Our results show that (i) RecN stimulates the RecA-dependent DNA strand invasion reaction *in vitro*, (ii) the RecN and RecA proteins physically interact, (iii) the stimulation of RecA requires the ATPase activity of RecN protein, (iv) RecA bound to DNA stimulates the ATPase activity of RecN protein more than 20-fold, (v) the rate of RecN-mediated ATP hydrolysis during the D-loop reaction is sensitive to the concentration and the length of a second DNA strand added (target) and (vi) the stimulation of RecN ATP hydrolysis observed during the D-loop reaction is not homology-dependent.
We propose a role for RecN in the search for homology between two DNA strands, one bound by RecA ([Fig. 7](#f7){ref-type="fig"}). First, RecA binds to a resected DSB end and recruits the RecN protein. We believe the recruitment of RecN is mediated by a direct interaction between RecN and RecA protein. Keyamura *et al*.[@b9] has shown that green fluorescent protein-labelled RecN protein fails to localize to the sites of DSBs in the absence of RecA in *E. coli*. Alonso\'s group has determined the localization kinetics of several *B. subtilis* recombination proteins and reports that RecN is among the first responders to a DSB, followed by the RecA protein[@b8][@b14][@b25]. Further, the *B. subtilis* RecA protein was shown to promote the disassembly of *B. subtilis* RecN--DNA complexes *in vitro*[@b20]. In the current study, we demonstrate that an ordered assembly of ternary complex occurs such that RecA binds to DNA followed by the binding of RecN protein (as measured by RecN ATP hydrolysis; [Fig. 4](#f4){ref-type="fig"}). In fact, a measurable lag in the engagement of RecN is observed if RecN is pre-incubated either with the DNA or RecA suggesting a reorganization of proteins is occurring such that RecN is dissociating from DNA or RecA so that RecA is free to bind the DNA.
We have presented several lines of evidence that suggest a direct interaction between RecA and RecN proteins: (i) the stimulation of RecA-mediated DNA pairing and strand exchange by RecN is species-specific ([Fig. 3](#f3){ref-type="fig"}); (ii) the kinetic enhancement of RecN ATP hydrolysis by RecA bound to DNA is species-specific ([Supplementary Fig. 3](#S1){ref-type="supplementary-material"}); (iii) the rate of RecN ATP hydrolysis is lower when RecA and RecN are incubated together in the absence of DNA than when RecN is bound to DNA in the absence of RecA (reaction 4, [Fig. 4](#f4){ref-type="fig"}); and (iv) the RecN and RecA proteins co-elute from antibody-coupled resins ([Fig. 3](#f3){ref-type="fig"}).
Once the RecN protein has been recruited to the DNA-bound RecA protein, the ATPase activity of RecN is stimulated ([Figs 4](#f4){ref-type="fig"} and [5](#f5){ref-type="fig"}). Monitoring the rates of RecN ATP hydrolysis during the RecA-mediated D-loop reaction allowed us to detect an ordered activation of RecN activity since the stimulation is sensitive to the concentration of the second DNA strand added to the reaction (the target), but is less sensitive to the concentration of the probe DNA bound by RecA protein ([Fig. 6a](#f6){ref-type="fig"}). Further, the increased activity of RecN is likely important for a step of the reaction before the DNA base-level, Watson--Crick sampling that is thought to be the initial steps of homologous pairing[@b26][@b27][@b28][@b29] since the rate increase is observed even when the target DNA is not homologous to the probe DNA ([Fig. 5b](#f5){ref-type="fig"}). It is likely that the RecA--DNA--RecN--DNA complex illustrated in [Fig. 7](#f7){ref-type="fig"} represents an initial bridging of two DNA molecules, a first step in the process of homology search. Although the function(s) of this RecA-dependent increase in ATP hydrolysis by RecN (more than 20-fold under some conditions) is not fully elucidated, the rate increase elicits the intriguing possibility that RecN possesses a motor activity. The high level of ATP hydrolysis by RecN protein indicates that RecN may be affecting the dynamics of RecA-mediated pairing of two DNA strands. It is interesting to note that the eukaryotic Rad54 protein stimulates Rad51 homologous DNA pairing by remodelling DNA topology[@b30]. RecN could conceivably generate negative topological stress that would stimulate DNA strand invasion by RecA[@b31]. Although we observe the same activation of RecN ATP hydrolysis whether the target DNA is supercoiled or linearized, a detailed study of the effect of RecN on the topological state of duplex DNA will be required to determine whether this mechanism is a contributing factor.
The ATPase activity of RecN may power movement of RecN or RecN--RecA complexes along the target DNA or facilitate movement between DNA segments. The enhanced rate of RecN ATP hydrolysis under the D-loop reaction conditions ([Fig. 5b](#f5){ref-type="fig"}) is largely dependent on both RecA protein bound to the probe DNA and the concentration of the target DNA. Further, the rate of RecN ATP hydrolysis is sensitive to the length of target DNA ([Fig. 6b](#f6){ref-type="fig"}). This may be a reflection of a higher RecN affinity for longer DNA molecules. RecN likely binds DNA with some degree of cooperativity since the DNA-dependent rate of ATP turnover depends on RecN protein concentration[@b5]. Although more work is necessary to clearly delineate the mechanism of these effects, another interpretation is that shorter DNA limits the rate of ATP hydrolysis because RecN movement is limited in range on a single DNA-binding event.
Studies by several groups[@b32][@b33][@b34] have argued that SMC complexes (both cohesins and condensins) translocate, presumably by sliding, to relocate over large distances along a chromosome (reviewed in refs [@b35], [@b36]). Therefore, it is possible that RecN translocation is at least contributing to the high rates of ATP hydrolysis we have observed in the current study. In the model proposed in [Fig. 7](#f7){ref-type="fig"}, translocation powered by ATP hydrolysis would allow RecN to move RecA presynaptic filaments (through a direct interaction) along target duplex DNA. At least two previous studies have proposed a three-dimensional diffusion model for the RecA homology search and provide evidence that RecA--DNA filaments do not slide with respect to target duplex DNA[@b37][@b38]. However, Ragunathan *et al*.[@b39] observed one-dimensional sliding of RecA presynaptic filaments with respect to target duplex DNA. It is possible that both sliding and intersegmental transfer contribute to the search for homology[@b39][@b40]. The RecN movement of RecA presynaptic filaments could contribute to both movement along the chromosome and intersegment transfer between chromosomes as part of a global search for homology in the cell. This idea of a cellular-wide homology search, although not new, is frequently overlooked when describing recombination events[@b28]. In eukaryotic cells, chromatin mobility increases at sites of DSBs[@b41][@b42][@b43], and movement of DNA DSB ends has been visualized in live *E. coli* cells[@b10] indicating a presynaptic, long-range homology search in bacterial cells.
The data presented here represent a major step forward in understanding the biochemical role of RecN in recombinational DNA DSB repair pathways. To the best of our knowledge, this study is the first observation of a RecN protein stimulating the DNA strand exchange activity ([Fig. 1](#f1){ref-type="fig"}), and in particular, the formation of D-loops catalysed by the RecA protein ([Fig. 2](#f2){ref-type="fig"}). Further, we have provided evidence for a role of RecN ATP hydrolysis during the presynaptic steps of DSB repair. The elucidation of the bacterial RecN protein activities may further the understanding of recombinational DNA repair in eukaryotic cells. The enhancement of the DNA-pairing activity of RecA by the RecN protein is strikingly similar to the stimulation of Rad51 by the Rad54 protein or of Dmc1 by Hop2--Mnd1 proteins shown by several groups[@b44][@b45][@b46][@b47][@b48][@b49][@b50][@b51][@b52]. The activity of RecN as a first responder to DSBs has also been compared to the Rad50 component of the Mre11--Rad50--Nbs1 complex[@b14][@b53]. The model presented in [Fig. 7](#f7){ref-type="fig"} supposes a direct link between the assembly of RecN and RecA at a DNA DSB and the RecN facilitation of a long-range homology search by RecA protein that leads to the critical homologous DNA-pairing event of the recombinational DNA repair pathway.
Methods
=======
Protein expression and purification
-----------------------------------
*Wild-type RecN protein*. The *D. radiodurans* RecN protein was expressed in EAW3 (*E. coli* strain MG1655 *ΔrecN*; ref. [@b5]) cells co-transformed with pT7Pol26 and pEAW309 (ref. [@b5]). A 10 l culture was grown in Luria broth (LB) broth (10 g l^−1^ tryptone, 5 g l^−1^ yeast extract and 10 g l^−1^ NaCl, with pH adjusted to 7.0) to an OD~600~ of 0.5. RecN protein expression was induced by the addition of isopropyl β-[D]{.smallcaps}-1-thiogalactopyranoside (IPTG) to a final concentration of 0.4 mM. Following a 6 h incubation at 37 °C, cells were collected by centrifugation, flash-frozen in liquid N~2~ and stored at −80 °C. All subsequent steps were carried out at 4 °C. Cell paste was thawed and fully resuspended to 20% cell weight per volume ratio in Tris-sucrose solution (25% sucrose and 250 mM Tris-HCl 80% cation, pH 7.5) supplemented with the protease inhibitors 4-(2-aminoethyl)benzenesulfonyl fluoride-HCl and pepstatin A to final concentrations of 0.5 mg ml^−1^ and 0.7 μg ml^−1^, respectively. The cells were lysed by 60 min incubation with lysozyme in 250 mM Tris-HCl (80% cation, pH 7.5) to 2.5 mg ml^−1^, followed by the addition of 0.4 ml of 25 mM EDTA per ml of lysed cell suspension, sonication and centrifugation for 1 h. The lysate was precipitated with polyethyleneimine, pH 7.5 (0.5% final concentration) and centrifuged. The pellet was washed with R buffer (20 mM Tris-HCl (80% cation, pH 7.5), 1 mM dithiothreitol (DTT), 0.1 mM EDTA and 10% glycerol)+150 mM ammonium sulfate and extracted two times with R buffer+300 mM ammonium sulfate. The protein solution was precipitated by the addition of 0.23 g solid ammonium sulfate per ml of solution (40% saturation). The precipitant was washed with R buffer+2.3 M ammonium sulfate, resuspended in R buffer+300 mM KCl and dialysed extensively versus R buffer+50 mM KCl. The protein was loaded onto DEAE-Sepharose resin (GE Healthcare), washed with 1 column volume of R buffer+50 mM KCl and eluted with a linear gradient of KCl from 50 mM to 1 M KCl over 10 column volumes. The RecN protein eluted at ∼300 mM KCl. Peak fractions were analysed by SDS--PAGE and fractions containing RecN were pooled and dialysed versus P buffer (20 mM potassium phosphate (pH 6.8), 1 mM DTT, 0.1 mM EDTA and 10% glycerol). Pooled protein was loaded onto ceramic hydroxyapatite resin (BioRad), washed with 1 column volume of P buffer and eluted with a linear gradient from 20 mM to 1 M potassium phosphate buffer (pH 6.8) over 10 column volumes. The RecN protein eluted at ∼ 400 mM phosphate. The fractions containing RecN were pooled and concentrated using a Source Q column. The pure RecN protein was determined to be free of nuclease contamination and was dialysed extensively versus storage buffer (R buffer+50 mM KCl), flash-frozen in liquid N~2~ and stored at −80 °C. The concentration of the RecN protein (molecular weight 59,798 Da) was determined from the absorbance at 280 nm using the calculated extinction coefficient 29,160 M^−1^ cm^−1^.
*RecN K67A mutant protein*. Cloning of pPLP01 containing *D. radiodurans recN K67A* was carried out via site-directed mutagenesis, using pEAW309 (wild-type *recN* expression plasmid[@b5]) as a template, according to the Stratagene Quick Change site-directed mutagenesis kit manual. DNA sequencing confirmed the desired point mutation. The *D. radiodurans* RecN K67A mutant protein was expressed and purified as described above for wild-type RecN protein except that the protein required flow through passage in Heparin FF resin (GE Healthcare) to remove trace nuclease contamination before the final concentration step.
*D. radiodurans RecA proteins*. The wild-type RecA and the RecA K83R mutant proteins were purified following a modified version of a procedure previously described[@b54]. Briefly, the *D. radiodurans* wild-type RecA or RecAK83R proteins were expressed by growing a 10 l culture of *E. coli* strain STL2669 (ref. [@b55]) co-transformed with either pEAW158, (wild-type *recA*) or pEAW244 (*recA K83R*) and pT7Pol26 in LB broth to an OD~600~ of 0.6. The pEAW158 and pEAW244 expression plasmids were gifts from Michael Cox (University of Wisconsin-Madison). Expression was induced by addition of IPTG to a final concentration of 0.4 mM. Following a 3 hour incubation with IPTG at 37 °C, the cells were collected by centrifugation, flash-frozen in liquid N~2~ and stored at −80 °C. All subsequent steps of this purification were carried out at 4 °C. Cells were thawed and fully resuspended to a final 20% cell weight per volume ratio in Tris-sucrose solution. Cell suspensions were lysed by 60 min incubation with lysozyme in 250 mM Tris-HCl (80% cation, pH 7.5) to 2.5 mg ml^−1^, followed by the addition of 0.4 ml of 25 mM EDTA per ml of lysed cell suspension, sonication and centrifugation for 1 h. The lysates were precipitated with polyethyleneimine (Sigma), pH 7.5 (0.5% final concentration) and centrifuged. The resulting pellets were washed with 50 ml of R buffer+50 mM ammonium sulfate and then extracted two times with 50 ml of R buffer+300 mM ammonium sulfate. The protein solutions were precipitated by adding 0.33 g solid ammonium sulfate per ml of solution (55% saturation). The precipitants were washed with 50 ml R buffer+3 M ammonium sulfate, resuspended in 50 ml of R buffer+300 mM KCl and dialysed versus R buffer+150 mM KCl and then extensively dialysed into R buffer+50 mM KCl. The proteins were loaded onto a DEAE-Sepharose column and washed with two column volumes of R buffer+50 mM KCl. Flow-through fractions were identified by SDS--PAGE, pooled and dialysed into P buffer. The dialysed proteins were then loaded onto Bio-Gel hydroxyapatite resin (BioRad), washed with two column volumes of P buffer and eluted with two column volumes of 500 mM potassium phosphate buffer (pH 6.8). Pooled fractions were dialysed verses R buffer+50 mM KCl and loaded onto a PBE-94 column (GE Healthcare), washed with one column volume of R buffer+50 mM KCl and eluted with a linear gradient from 50 mM to 1 M KCl. The RecA proteins eluted from this column at ∼500 mM KCl. Pooled protein fractions were determined to be free of nuclease contamination and were concentrated by ammonium sulfate precipitation at 55% saturation as described above. Pellets were resuspended in 5 ml of R buffer+300 mM KCl and dialysed extensively into R buffer (storage buffer). Concentrations of *D. radiodurans* RecA proteins were determined from the absorbance at 280 nm, using the determined extinction coefficient 0.372 mg ml^−1^ cm^−1^ and molecular mass 38,013 Da (ref. [@b54]).
*E. coli RecA protein*. The *E. coli* wild-type RecA protein was purified as described[@b55]. Briefly, the RecA protein was overexpressed by growing a 10 l culture of STL2669 (ref. [@b5]) co-transformed with pAIR79 (wild-type *recA*) and pT7pol26 in LB broth to an OD~600~ of 0.8. Protein expression was induced by the addition of IPTG to a final concentration of 0.4 mM. Following a 3 h incubation with IPTG at 37 °C, the cells were collected by centrifugation, flash-frozen in liquid N~2~ and stored at −80 °C. All subsequent steps of this purification were carried out at 4 °C. Cell paste was thawed and fully resuspended to a final 20% cell weight by volume ratio in Tris-sucrose solution. Cell suspension was lysed by a 60 min incubation with lysozyme (2.5 mg ml^−1^ final) in 250 mM Tris-HCl (80% cation, pH 7.5) followed by the addition of 0.4 ml of 25 mM EDTA per ml of fraction, sonication and centrifugation. The lysate was precipitated with polyethyleneimine, pH 7.5 (0.5% final concentration) and centrifuged. The pellet was washed once with 100 ml of R buffer+150 mM ammonium sulfate and then extracted two times with 100 ml of R buffer+300 mM ammonium sulfate. The protein solution was precipitated by adding 0.28 g of solid ammonium sulfate per ml of solution (47% saturation) followed by centrifugation. The precipitated protein was washed twice with 50 ml R buffer+2.8 M ammonium sulfate, resuspended in 100 ml of R buffer+100 mM KCl and dialysed extensively into R buffer+100 mM KCl. The protein was loaded onto DEAE-Sepharose resin and washed with two column volumes of R buffer+100 mM KCl. Flow-through fractions were identified by SDS--PAGE, pooled and dialysed into P buffer and loaded onto ceramic hydroxyapatite resin, washed with two column volumes of P buffer. Protein was eluted with a linear gradient from 20 to 350 mM potassium phosphate over 10 column volumes. Pooled protein fractions were determined to be free of nuclease contamination and were concentrated by ammonium sulfate precipitation at 47% saturation as described above, resuspended in R buffer and dialysed extensively into R buffer (storage buffer). The concentration of purified RecA protein (37,842 Da) was determined from the absorbance at 280 nm using the extinction coefficient 2.23 × 10^4^ M^−0^ cm^−c^ (ref. [@b55]).
*Single-strand binding proteins*. The *E. coli* single-strand binding (SSB) protein was purified as described[@b56]. Briefly, the *E. coli* SSB protein was overexpressed by growing a 10 l culture of BL21(DE3) transformed with pEAW134 (a gift from Michael Cox) in LB broth to an OD~600~ of 0.5. Protein expression was induced by the addition of IPTG to a final concentration of 0.4 mM. Following a 3 h incubation with IPTG at 37 °C, the cells were collected by centrifugation and flash-frozen in liquid N~2~. All steps of this purification were carried out at 4 °C. Cell paste was thawed and fully resuspended to a final 25% cell weight to volume ratio in lysis buffer (50 mM Tris-HCl, pH 8.3, 0.2 M NaCl, 15 mM spermidine tri-Cl, 1 mM EDTA and 10% sucrose). Cell suspension was lysed by a 30 min incubation with lysozyme (200 μg ml^−1^ final lysozyme concentration). The cell suspension was also supplemented with phenylmethylsulfonyl fluoride (0.1 mM final concentration). Cell lysate was incubated for 30 additional minutes with sodium deoxycholate (0.05% final concentration) followed by sonication and centrifugation. The lysate was precipitated with polyethyleneimine, pH 7.5 (0.4% final concentration) for 30 min and centrifuged. The pellet was resuspended for 45 min in 100 ml of TGE buffer (50 mM Tris-HCl, pH 8.3, 1 mM EDTA and 20% glycerol)+0.4 M NaCl and centrifuged. The protein solution was supplemented with 0.15 g ml^−1^ ammonium sulfate (27% saturation) overnight and centrifuged. The precipitant was washed in 100 ml of TGE buffer+0.15 g ml^−1^ ammonium sulfate and centrifuged. Washing step was repeated twice. Protein pellet was resuspended in 20 ml of TGE buffer+300 mM NaCl and subjected to step dialysis into TGE buffer+50 mM NaCl. The solution was loaded onto DEAE-Sepharose resin, washed with 2 column volumes of TGE buffer+50 mM NaCl and eluted with a linear gradient from 50 to 800 mM NaCl over 10 column volumes. Pooled fractions were dialysed into TGE buffer+50 mM NaCl, loaded onto Heparin resin, washed with 2 column volumes of TGE buffer+50 mM NaCl and eluted with a linear gradient from 50 to 300 mM NaCl over 10 column volumes. Pooled fractions were determined to be free from nuclease contamination and dialysed extensively into storage buffer (20 mM Tris-HCl, pH 8.3, 0.5 M NaCl, 1 mM EDTA, 1 mM β-mercaptoethanol and 50% glycerol). The concentrations of *E. coli* SSB (18,843 Da) was determined by absorbance measurements at 280 nm using the extinction coefficient of 2.38 × 10^4^ M^−1^ cm^−1^ (ref. [@b56]). The purified *D. radiodurans* SSB protein[@b57] was a gift from Michael Cox.
Biochemicals
------------
Unless otherwise noted, all of the reagents were purchased from Fisher. Phosphoenolpyruvate was from Spectrum. Pyruvate kinase, lactate dehydrogenase, NADH, ATP, polyethyleneimine and bromophenol blue were purchased from Sigma. All restriction endonucleases and T7 exonuclease were purchased from New England Biolabs. DTT was from Soltec Ventures.
DNA substrates
--------------
Bacteriophage *φ*X174 circular ssDNA (virion) and *φ*X174 RFI supercoiled circular duplex DNA (5,386 bp) were purchased from New England Biolabs. Plasmid DNA substrates pEAW324 (8,716 bp) and pEAW3 (2,431 bp) were gifts from Michael Cox[@b58][@b59], and prepared using CsCl-ethidium bromide gradients[@b60]. Unless otherwise noted, full-length linear duplex DNA substrates were generated by the digestion at unique restriction sites on the *φ*X174 RFI DNA with PstI, pEAW324 with ApaI and pEAW3 with SspI restriction endonucleases, using conditions suggested by the enzyme supplier. The digested DNA was extracted with phenol/chloroform/isoamyl alcohol (25:24:1), followed by ethanol precipitation. D-loop substrate described as target DNA is supercoiled pEAW3 plasmid, unless otherwise indicated. D-loop substrate described as probe DNA ([Figs 2a](#f2){ref-type="fig"} and [5](#f5){ref-type="fig"}) was prepared by incubation of SspI treated pEAW3 linear duplex DNA with T7 exonuclease at 30 °C for 90 s. T7 exonuclease catalyses the removal of 5′-mononucleotides from duplex DNA in the 5′--3′ direction at a rate of ∼150 nucleotides per 90 s in buffer conditions suggested by enzyme supplier. Various-length linear duplex DNA substrates used as target for the experiments of [Fig. 6b](#f6){ref-type="fig"} were prepared by digesting pEAW3 (2,431 bp) with NdeI, XbaI and ScaI to produce 1,450, 650 and 303 bp and pEAW324 (8,716 bp) with BsaXI to produce 4,205 and 1,074 bp, using conditions suggested by the enzyme supplier. DNA substrates were gel-purified using Agarase (Thermo Scientific) following the protocol for recovery of DNA from low-melting agarose gels provided by the supplier. The concentrations of ssDNA and double-stranded DNA were determined by absorbance at 260 nm, using 36 and 50 mg ml^−1^ A~260~^−1^, respectively, as conversion factors. Unless otherwise specified, all DNA concentrations are given in micromolar nucleotides.
Buffers
-------
Buffer A is 25 mM Tris acetate (80% cation, pH 7.4), 1 mM DTT, 3 mM potassium glutamate, 10 mM Mg(OAc)~2~ and 5% (weight per volume) glycerol. Buffer N is 25 mM Tris-OAc (80% cation, pH 7.4), 1 mM DTT, 3 mM potassium glutamate, 17.5 mM Mg(OAc)~2~, 40 mM KOAc, 5% (weight per volume) glycerol and 1% buffered polyethylene glycol 8000. Buffer T is 20 mM Tris-Cl (80% cation, pH 7.4), 20 mM EDTA. 0.5% SDS. 2 × loading buffer contains 15% Ficoll, 0.24% bromphenol blue and 0.24% xylene cyanole. 2 × loading stop buffer solution contains 15% Ficoll, 4% SDS, 0.24% bromophenol blue and 0.24% xylene cyanole. TBE buffer is 90 mM Tris borate and 2 mM EDTA, pH 8. 5 × SDS loading buffer contains 250 mM Tris-Cl (pH 6.8), 4% SDS, 20% glycerol, 10% 2-mercaptoethanol and 0.1% bromophenol blue.
RecA-mediated DNA three-strand exchange
---------------------------------------
The DNA strand exchange reactions were carried out at 37 °C in buffer A and an ATP regeneration system (10 units per ml of pyruvate kinase and 2.5 mM phosphoenolpyruvate). Protein and DNA concentrations are described in figure legends. Experiments measuring the DNA strand exchange activity of the *E. coli* or *D. radiodurans* RecA protein utilized the *E. coli* or *D. radiodurans* SSB protein, respectively. RecA protein was incubated with *φ*X174 circular ssDNA for 10 min. SSB protein, 3 mM ATP and the RecN protein (where indicated) were added, followed by another 10 min incubation. The reaction was initiated by the addition of *φ*X174 linear duplex DNA and incubated for 45 min, or the times indicated in the figure legend. Dilute condition reactions ([Figs 1c](#f1){ref-type="fig"} and [3a](#f3){ref-type="fig"}) were stopped by addition of 80 μl buffer T plus 5 μl of proteinase K (to 1.25 mg ml^−1^). DNA was extracted with phenol/chloroform/isoamyl alcohol (25:24:1), followed by ethanol precipitation. DNA was resuspended in TE (10 μl) plus 2 × loading buffer. Normal DNA concentration reactions ([Fig. 1b](#f1){ref-type="fig"}) were stopped by adding 2 × loading stop buffer and proteinase K (to 1.25 mg ml^−1^). Samples were subjected to electrophoresis in 0.8% agarose gels with TBE buffer, stained with ethidium bromide and exposed to ultraviolet light. The inverted gel images were obtained using a digital charge-coupled device camera with Foto/Analyst PC Image software version 10.21 (Fotodyne).
RecA-dependent D-loop formation
-------------------------------
The D-loop formation reactions were carried out at 37 °C in buffer A and an ATP regeneration system (10 units per ml of pyruvate kinase and 2.5 mM phosphoenolpyruvate). Protein and DNA concentrations are described in figure legends. RecA protein was incubated with probe DNA for 10 min. ATP (3 mM) and the RecN protein were added, followed by another 10 min incubation. The reaction was initiated by the addition of target DNA and incubated for 45 min. Reactions were stopped by adding 2 × loading stop buffer and proteinase K (to 1.25 mg ml^−1^). Samples were subjected to electrophoresis as described above.
ATPase assay
------------
The ATPase activity of RecN protein was measured using an enzyme-coupled spectrophotometric enzyme assay as described[@b5]. All reactions were carried out at 37 °C. The RecN ATPase activity measured in experiments represented in [Figs 5](#f5){ref-type="fig"} and [6](#f6){ref-type="fig"}, and were carried out in buffer A. Buffer N was used in experiments represented in [Fig. 4](#f4){ref-type="fig"}. The concentration of proteins and DNA, RecN, as well as the order of protein additions are indicated in the figure legends. Reactions were initiated by the addition of 3 mM ATP.
Co-elution of RecA and RecN proteins
------------------------------------
Pull-down experiments were done using purified *D. radiodurans* RecA and RecN proteins. Antibodies against these proteins were raised in chicken (Chicken IgY) and affinity-purified (GeneTel Laboratories, LLC, Madison, WI). The stock concentration for the RecA and RecN antibodies are 13.2 and 1 mg ml^−1^, respectively. A unit of 50 μg antibody for RecA or RecN (diluted directly from the stock in AminoLink Plus (Pierce) coupling buffer) was coupled to 100 μl of AminoLink Plus coupling resin (50% slurry). All concentrations given are final concentrations. Reactions (40 μl) were carried out in buffer N with an ATP regeneration system (10 units per ml pyruvate kinase and 3.5 mM phosphoenolpyruvate) in the presence or absence of 25 μM *φ*X174 3′-linear duplex DNA. RecA protein (2 μM final) was added to the above reaction mixture and incubated at 37 °C for 20 min. Binding reactions were initiated by the addition of ATP (to 2.5 mM) and allowed to proceed for 20 min before the addition of 4.8 μg RecN protein (2 μM final) and further incubated for 15 min at 37 °C. Samples (except for the input sample) were diluted 1:3 in 1 × Dulbecco\'s modified PBS buffer (Sigma). Diluted mixtures were loaded on RecA or RecN antibody-coupled resin and incubated for 2 h at 4 °C on a turn-table. Resins were washed with 150 μl IP lysis/wash buffer (Pierce). Protein complexes were eluted with 50 μl of elution buffer (Pierce). Protein samples were mixed directly with 5 × SDS loading buffer, and 10 μl were loaded and separated by 14% SDS--PAGE. The inverted gel images were obtained using a digital charge-coupled device camera with Foto/Analyst PC Image software version 10.21 (Fotodyne).
Data availability
-----------------
All relevant data are available from the authors.
Additional information
======================
**How to cite this article:** Uranga, L. A. *et al*. The cohesin-like RecN protein stimulates RecA-mediated recombinational repair of DNA double-strand breaks. *Nat. Commun.* **8,** 15282 doi: 10.1038/ncomms15282 (2017).
**Publisher\'s note**: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supplementary Material {#S1}
======================
###### Supplementary Information
Supplementary figures and supplementary references.
The work was supported by National Institutes of Health grants R01GM104375 and P20GM103451 (to S.L.L). This work was also supported by Howard Hughes Medical Institute Undergraduate Science Education Grant 52005881 (to L.N.R.). We thank Michael Cox for plasmids and antibodies and Madison Spence for technical assistance.
The authors declare no competing financial interests.
**Author contributions** L.A.U. designed, conducted and analysed the experiments in [Figs 2](#f2){ref-type="fig"}, [3b](#f3){ref-type="fig"}, [4](#f4){ref-type="fig"}, [5](#f5){ref-type="fig"} and [6](#f6){ref-type="fig"} with assistance from E.D.R. and L.N.R.; E.D.R. designed and conducted the experiments in [Figs 1](#f1){ref-type="fig"} and [3a](#f3){ref-type="fig"}; P.L.P. designed and conducted the experiment in [Fig. 3c](#f3){ref-type="fig"} and cloned the pPLP01 plasmid; L.A.U., E.D.R., P.L.P. and L.N.R. all purified enzymes used; L.A.U., P.L.P. and S.L.L. prepared figures and wrote the manuscript. All authors edited and reviewed the manuscript.
{#f1}
{#f2}
{ref-type="fig"}. (**b**) EcRecA (6.7 μM) was incubated with 20 μM probe DNA for 10 min. ATP (3 mM) and 1 μM RecN, as indicated at the top of each lane, were added and incubated for an additional 10 min. The reactions were initiated by the addition of 20 μM target DNA. All reactions were incubated for 45 min. See [Fig. 2](#f2){ref-type="fig"} for target and probe DNA description. This experiment was repeated three times with no measurable difference between + and -- RecN conditions. (**c**) Purified *D. radiodurans* RecA (38 kDA) and RecN (60 kDa) proteins co-elute, in the presence (+dsDNA) or absence (−dsDNA) of linear duplex DNA, from a RecN antibody-coupled resin (top) or from a RecA antibody-coupled resin (bottom). Lane M indicates a protein size marker. The input lanes contain an 8 μl load of a mixture of 0.12 μg RecN per μl and 0.08 μg RecA per μl. Excess protein complex was removed during the early wash steps, and 8 μl of the final 50 μl wash and 8 μl of the 50 μl elution were loaded directly onto the gel.](ncomms15282-f3){#f3}
{ref-type="table"} for steady-state rates of RecN ATP hydrolysis and lag times.](ncomms15282-f4){#f4}
{ref-type="fig"} legend) for 10 min before the addition of 3 mM ATP and RecN (1 μM where indicated). Target DNA (see [Fig. 2](#f2){ref-type="fig"} legend) was added 10 min later. For each reaction described, components omitted from reactions were compensated for by protein storage buffers or TE, in the case of DNA. All reactions were carried out under buffer A conditions and followed the reaction scheme shown. ATP hydrolysis was measured after the addition of ATP. (**b**) Controls measuring RecN ATP hydrolysis in the absence of RecA K83R are shown with 10 μM probe DNA and no target DNA (reaction 1), no probe DNA and 10 μM target DNA (reaction 2), and 10 μM probe DNA plus 10 μM target DNA (reaction 3). Reaction 4: RecN ATP hydrolysis when RecA K83R protein was incubated with 10 μM probe DNA in the absence of added target DNA. Reaction 5: RecN ATP hydrolysis when RecA K83R protein was incubated in the absence of probe DNA followed by 10 μM target DNA. Reaction 6: RecN ATP hydrolysis when RecA K83R protein was incubated with 10 μM probe DNA followed by 10 μM target DNA. Reaction 7: RecN ATP hydrolysis when RecA K83R protein was incubated with 10 μM probe DNA followed by 10 μM non-homologous, supercoiled RF1 *φ*X174 DNA. See [Table 2](#t2){ref-type="table"} for steady-state RecN ATP hydrolysis rates.](ncomms15282-f5){#f5}
{ref-type="fig"}, except as noted. (**a**) The steady-state RecN ATP hydrolysis rate was measured when added to RecA K83R protein incubated with probe DNA (where indicated) followed by target DNA (where indicated). The probe DNA concentration is held constant at 10 μM and the target DNA concentration is titrated (0--10 μM), and white bars represent the rate of ATP hydrolysis catalysed by RecN protein at each DNA concentration relative to the rate measured at zero target DNA. The target DNA concentration is held constant at 10 μM and the probe DNA concentration is titrated (0--10 μM), and grey bars represent the rate of ATP hydrolysis catalysed by RecN protein at each DNA concentration relative to the rate measured at zero probe DNA. Error bars represent the s.d. of the relative rate (the s.d. of the average rate divided by the average rate) from 3 to 15 independent experiments (see [Table 2](#t2){ref-type="table"} for steady-state rates). (**b**) The steady-state RecN ATP hydrolysis rate was measured when added to RecA K83R protein incubated with 2 nM molecules (10 μM nt) probe DNA followed by 2 nM molecules linearized, target DNA (target lds DNA, black square) or 10 μM of nucleotides linearized, target DNA (target lds DNA, white square) of different lengths, as indicated. The linear duplex target DNA substrate length in kilobase pairs (kbp) and concentration in μM nt and nM molecules for the two sets are as follows: 0.65 kbp (2.6 μM, black square; 7.7 nM, white square); 1.1 kbp (4.3 μM, black square); 1.5 kbp (5.8 μM, black square; 3.5 nM, white square); 2.4 kbp (9.6 μM, black square; 2.1 nM, white square); 4.2 kbp (16.8 μM, black square); 5.4 bp (21.5 μM, black square; 0.9 nM, white square); and 8.7 kbp (34.8 μM, black square; 0.6 nM, white square). The error bars represent the s.d. of four independent experiments. nt, nucleotide.](ncomms15282-f6){#f6}
{#f7}
###### Kinetic measurements for experiments illustrated in [Fig. 4](#f4){ref-type="fig"}.
**Order of addition** **Average ATP hydrolysis rate (μM min**^**−1**^**)±s.d.** **Average lag time (min)±s.d.**
-------------------------------- ----------------------------------------------------------- ---------------------------------
1---RecN, no DNA, no RecA K83R 8.3±1.7 0
2---RecN+DNA, no RecA K83R 13.7±4.1 0
3---RecN+DNA→RecA K83R 53.4±4.2 25.0±5.0
4---RecN+RecA K83R→DNA 47.5±8.5 34.3±5.3
5---RecA K83R+DNA→RecN 56.3±15.6 8.3±6.7
RecA K83R+DNA, no RecN 0.20±0.17 0
RecN ATP hydrolysis rate (in μM min^−1^) and lag times (in min) for the experiments illustrated in [Fig. 4](#f4){ref-type="fig"}. The order of additions match the numbered reactions. The ATP hydrolysis rates were measured after the reaction reached the steady state. Lag times, where applicable, represent the time required for the reaction to reach a steady-state rate of hydrolysis after the addition of all reaction components ([Fig. 4](#f4){ref-type="fig"}). All averages and s.d.\'s were calculated from 4 independent trials except that data for reaction 5 were calculated from 10 independent trials.
###### Steady-state RecN ATP hydrolysis rates under D-loop assay conditions.
**Order of addition** ***n*** **Probe DNA (μM nt)** **Target DNA (μM nt)** **Average ATP hydrolysis rate (μM min**^**−1**^**)±s.d.**
---------------------------------------------------------------------------------------------------------------- --------- ----------------------- ------------------------ -----------------------------------------------------------
RecN+target DNA control (reaction 1, [Fig. 5b](#f5){ref-type="fig"}) 3 0 10 2.4±0.5
RecN+probe DNA control (reaction 2, [Fig. 5b](#f5){ref-type="fig"}) 3 10 0 2.4±0.4
RecN+target and probe DNA control (reaction 3, [Fig. 5b](#f5){ref-type="fig"}) 3 10 10 2.3±0.5
RecA K83R+probe DNA→RecN (reaction 4, [Figs 5b](#f5){ref-type="fig"} and [6a](#f6){ref-type="fig"}) 9 10 0 4.5±1.8
RecA K83R→RecN+target DNA (reaction 5, [Figs 5b](#f5){ref-type="fig"} and [6a](#f6){ref-type="fig"}) 9 0 10 30.9±3.2
RecA K83R+probe DNA→RecN+target DNA ([Fig. 6a](#f6){ref-type="fig"}) 5 1 10 27.3±1.7
8 2.5 10 28.5±1.7
5 5 10 36.1±4.2
3 7.5 10 39.7±2.4
5 10 1 7.7±4.7
8 10 2.5 14.5±7.4
3 10 5 29.2±9.3
5 10 7.5 38.6±2.5
RecA K83R+probe DNA→RecN+target DNA (reaction 6, [Figs 5b](#f5){ref-type="fig"} and [6a](#f6){ref-type="fig"}) 15 10 10 49.2±6.4
RecA K83R+probe DNA→RecN+heterologous target DNA (reaction 7, [Fig. 5b](#f5){ref-type="fig"}) 3 10 10 48.3±1.4
RecA K83R+probe DNA→RecN+linearized, homologous target DNA ([Fig. 6b](#f6){ref-type="fig"}, 2.4 kb target DNA) 6 10 10 47.9±5.0
Reaction conditions are described in the legend to [Fig. 5](#f5){ref-type="fig"}. Averages and s.d.\'s were calculated from the number of independent trials indicated (*n*). Order of addition describes the experimental condition and the relevant figure containing representative data. The probe and target DNAs are derived from 2.4 kbp plasmid DNA and are described in the legend to [Fig. 2](#f2){ref-type="fig"}. Heterologous DNA is non-homologous, supercoiled RF1 *φ*X174 DNA.
| {
"pile_set_name": "PubMed Central"
} |
Introduction
============
Among the different forms of endometrial cancer, endometrioid endometrial adenocarcinoma (EEA) is the most common type (85%) ([@b1-ol-0-0-10504]). Although the prognosis of EEA is good, extensive heterogeneity has been reported in a number of studies, particularly in patients with an early stage of disease, exposing women to recurrent disease ([@b2-ol-0-0-10504]--[@b4-ol-0-0-10504]). Clinically, certain patients with EEA with an advanced stage of disease have a good prognosis, whereas certain patients with an early stage of disease can still relapse and succumb ([@b5-ol-0-0-10504],[@b6-ol-0-0-10504]). All these features indicate that traditional clinical features are not sufficient to accurately predict the prognosis of EEA. Molecular biological characteristics and traditional clinical features are particularly important in the prognosis of EEA. Tumor occurrence and development are driven by genetic alterations, and the phenotypic diversity may be accompanied by the corresponding diversity in the pattern of gene expression ([@b7-ol-0-0-10504]). Therefore, establishing a predictive prognostic model on the basis of gene expression profiles and traditional clinical features, which are different from the traditional criteria, is of great clinical value.
Machine-learning methods have been used to predict the prognosis of numerous types of cancer ([@b8-ol-0-0-10504],[@b9-ol-0-0-10504]). In the machine-learning area of research, the prognosis of cancer is a typical classification problem. When training a machine-learning model to undergo a prediction task, the factors relevant to the prognosis of cancer can be regarded as the features of the data, and the prognosis results are the class labels. Random Forest (RF) is a type of machine-learning method, which has been experimentally proven to be the best classifier ([@b10-ol-0-0-10504]). RF has a number of advantages and has already been successfully applied to microarray data classification ([@b11-ol-0-0-10504],[@b12-ol-0-0-10504]) and numerous other disease classifications ([@b13-ol-0-0-10504],[@b14-ol-0-0-10504]). Among the different variable selection methods, variable selection using RF (VSURF) has demonstrated the best predictive performance thus far ([@b15-ol-0-0-10504]). VSURF can handle thousands of input variables and identify the most significant variables ([@b10-ol-0-0-10504]); thus, it is considered a feature selection method and has been used to select the genes relevant to the type of cancer in question ([@b11-ol-0-0-10504],[@b16-ol-0-0-10504]).
However, to the best of our knowledge, there is currently no RF for predicting EEA prognosis by combining gene expression and traditional clinical characteristics. Therefore, the aim of the present study was to establish a prediction model combining genes and clinical features via RF for the prognosis of EEA. First, the state-of-the-art method VSURF was used to select informative factors that are relevant to the prognosis of EEA. The selected factors were then used to design an accurate predictive model via RF.
Materials and methods
=====================
### Patient selection
The present study was performed in the Department of Obstetrics and Gynecology, Peking University People\'s Hospital (PKUPH; Beijing, China). In the training cohort, 154 primary EEA (PE) samples without neoadjuvant therapy, RNA-sequencing (RNAseq) expression (combining level 3 data) and clinical data of female patients with uterine cancer were obtained from The Cancer Genome Atlas (TCGA) data portal ([cancergenome.nih.gov](cancergenome.nih.gov)) on January 7, 2018. These data included 64 PE samples without relapse (≥3 years of clinical follow-up), without radiation therapy and without additional pharmaceutical treatment, and 90 samples from relapsed or deceased PE (R/D-PE) patients with or without postoperative adjuvant therapy. TCGA samples were sub-stratified into four molecular subgroups: i) Copy number low (CN-L), ii) copy number high (CN-H), iii) microsatellite instability (MSI) and iv) catalytic subunit of DNA polymerase ε involved in nuclear DNA replication and repair (POLE) ultra-mutated, with different prognoses. Of the 154 cases in TCGA training cohort, 20.13% were CN-L, 5.19% were CN-H, 33.12% were MSI and 5.84% were POLE ultra-mutated; in 35.72% of the cases, the molecular typing information was lacking. The detailed inclusion or exclusion criteria and information on the selection of these 154 TCGA participants are presented in [Fig. 1](#f1-ol-0-0-10504){ref-type="fig"}. In the testing cohort, 21 PE samples without neoadjuvant therapy, as well as RNAseq expression and clinical data, were obtained from 21 surgically treated patients at the Department of Obstetrics and Gynecology PKUPH. All 21 samples were from patients without neoadjuvant therapy and who underwent surgical resection between January 2008 and December 2012. The cohort included 13 PE samples from patients without relapse (≥3 years of clinical follow-up) and 8 R/D-PE samples from patients with or without adjuvant therapy. The EEA samples were divided into two groups according to the prognosis. The group with a good prognosis contained the samples from non-relapsed EEA patients, and the group with a poor prognosis contained the samples form relapsed or deceased EEA patients. All deceased patients had succumbed to EEA. The study was approved by the Institutional Ethics Committee (Human Research) of the PKUPH.
### RNA isolation, RNAseq library preparation and sequencing of the 21 EEA samples
The total RNA was extracted with TRIzol^®^ (Tiangen Biotech Co., Ltd., Beijing, China) and assessed with an Agilent 2100 BioAnalyzer instrument (Agilent Technologies, Inc., Santa Clara, CA, USA) and a Qubit^™^ 4 Fluorometer (Invitrogen; Thermo Fisher Scientific, Inc., Waltham, MA, USA). The total RNA samples that met the following requirements were used in subsequent experiments: RNA integrity number \>7.0 and a 28S/18S ratio \>1.8. RNAseq libraries were generated and sequenced by CapitalBio Technology Co., Ltd. (Beijing, China). Triplicate samples of all assays were constructed in an independent library. The NEB Next Ultra RNA Library Prep kit for Illumina (New England BioLabs, Inc., Ipswich, MA, USA) was used to construct the DNA libraries for sequencing. The NEB Next Poly(A) mRNA Magnetic Isolation Module kit (New England BioLabs, Inc.) was used to enrich the poly(A)-tailed mRNA molecules from 1 µg total RNA. The mRNA was fragmented into \~200-bp pieces. The first-strand cDNA was synthesized from the mRNA fragments using reverse transcriptase and random hexamer primers (New England BioLabs, Inc.), and the second-strand cDNA was synthesized using DNA polymerase I and RNaseH (New England BioLabs, Inc.). The end of the cDNA fragment was subjected to an end repair process that included the addition of a single 'A' base, followed by ligation of the adapters, according to the instructions of the NEB Next Ultra RNA Library Prep kit (New England BioLabs, Inc.). The end Repair/dA-tail program was: i) 20°C for 30 min; ii) 65°C for 30 min; iii) Hold at 4°C. The products were purified using Agencourt AMPure XP Beads (Beckman Coulter, Inc., Brea, CA, USA) according to the manufacturer\'s protocol and enriched by polymerase chain reaction (PCR) to amplify the library DNA. Universal Primer Mix (New England BioLabs, Inc.) was used for amplification. The thermocycling conditions were as follows: 98°C for 30 sec; 12 cycles of 98°C for 10 sec, 65°C for 30 sec and 72°C for 30 sec; 72°C for 5 min. The final libraries were quantified using the KAPA Library Quantification kit (KAPA Biosystems; Roche Diagnostics, Basel, Switzerland) and an Agilent 2100 Bioanalyzer (Agilent Technologies, Inc.). The libraries were validated using reverse transcription-quantitative PCR, and the thermocycling conditions were as follows: 95°C for 5 min; 40 cycles of 95°C for 30 sec and 60°C for 45 sec. The libraries were subjected to paired-end sequencing with a pair-end 150-bp reading length on an Illumina HiSeq sequencer (Illumina, Inc., San Diego, CA, USA) ([@b17-ol-0-0-10504]).
### Data processing
In total, 18,669 coding genes were included in TCGA RNAseq data. Fragments per kilobase of exon model per million mapped fragments (FPKM) gene expression values were used for the statistical analysis. The format of RNAseq data downloaded from the TCGA was log~2~(FPKM+1); thus, the RNAseq data of the TCGA into FPKM was transformed for the follow-up study. There were a number of clinical features in TCGA clinical data, including age at initial pathological diagnosis (age), International Federation of Gynecology and Obstetrics (FIGO) stage ([@b18-ol-0-0-10504]), grade ([@b19-ol-0-0-10504]), peritoneal wash status and lymph node status. However, only the data for age, FIGO stage and grade were complete in TCGA cohorts. Thus, of all the clinical features, only age, FIGO stage and grade were included in the present study. To improve the generalization of the study results, a numerical value was given to age, FIGO stage and grade, according to a prior published study ([@b20-ol-0-0-10504]) and clinical experience. The numerical values of age, FIGO stage and grade were as follows: Age (\<60 years, 1; and ≥60 years, 2.55), grade (I--II, 1; and III, 2.43), and FIGO stage (Ia, 1; Ib, 1.5; II, 2.75; IIIa-b, 4; IIIc1, 4.21; IIIc2, 4.5; and IV, 6). These numerical values were used for the establishment of the RF prognostic prediction model of EEA.
### RF
RF is an ensemble of decision trees, which forms multiple decision trees and then aggregates them to provide a final prediction. When a new object from an input vector is to be predicted, the input vector is placed on each of the trees in the forest simultaneously. Each tree gives a prediction, then, the forest chooses the classification that has the most votes (out of all the trees in the forest). RF uses the bagging technique and the random feature selection technique. RF has two parameters, which are the number of trees (ntree) and the number of variables randomly sampled as candidates at each split (mtry).
### VSURF
VSURF is a three-step feature selection method based on RF. The first step is dedicated to removing irrelevant features from the dataset. The second step aims to select important features relevant to the class labels for interpretation purposes. The third step refines the selection by removing redundancy in the set of features selected by the second step, for prediction purposes. The ntree parameter was set to its default value of 2,000, and the mtry parameter was set to its default value (if nvm, the number of variables in the model is not greater than the number of observations; otherwise it is set to nvm/3). The VSURF results of the genes and clinical features are summarized in [Fig. 2](#f2-ol-0-0-10504){ref-type="fig"}, with [Fig. 2Aa-b](#f2-ol-0-0-10504){ref-type="fig"}, and [2Ba-b](#f2-ol-0-0-10504){ref-type="fig"} corresponding to the 'thresholding step', [Fig. 2Ac](#f2-ol-0-0-10504){ref-type="fig"} and [2Bc](#f2-ol-0-0-10504){ref-type="fig"} corresponding to the 'interpretation step', and [Fig. 2Ad](#f2-ol-0-0-10504){ref-type="fig"} and [2Bd](#f2-ol-0-0-10504){ref-type="fig"} corresponding to the 'prediction step'. The features from the 'interpretation step' had a strong association with EEA prognosis and were determined as the most important factors that affect the prognosis of EEA.
### Prediction experiment
RF parameter setting. The ntree parameter was set to 2,000, i.e., the RF included 2,000 decision trees, and the mtry parameter was set to its default value.
### Statistical analysis
All the model-associated data analyses were performed using R software (version 3.2.4; <http://www.r-project.org>). The VSURF method was used to select the most relevant prognostic genes and clinical characteristics. RF was used to build the predictive model for separating (relapsed or deceased) and unrelapsed patients. SPSS software (version 13.0; SPSS Inc., Chicago, IL, USA) was used to perform the statistical analysis. The associations between clinicopathological characteristics and outcomes were calculated using the χ^2^ test and Fisher\'s exact test. Survival curves for the 154 PE samples from TCGA cohort ([Fig. 3](#f3-ol-0-0-10504){ref-type="fig"}) were plotted using the Kaplan-Meier method and the differences between survival curves were calculated using a log-rank test. P\<0.05 was considered to indicate a statistically significant difference.
Results
=======
### Patient characteristics
In total, 154 PE samples meeting the inclusion criteria from TCGA cohort were selected for the training set, and 21 PE samples from the PKUPH cohort were included in the testing set. The median age of diagnosis for the samples in TCGA cohort was 64 years (range, 35--90 years). No significant difference was observed in the diagnostic age and menopause status between the PE samples and the R/D-PE samples, whereas significant differences existed in the grade, FIGO stage, lymph node status, adjuvant radiotherapy and body mass index. The median age of diagnosis for the samples in the PKUPH cohort was 55 years (range, 31--75 years). There was no significant difference observed in all the stated clinical characteristics, perhaps due to the limited sample size in the PKUPH cohort. The detailed data are presented in [Table I](#tI-ol-0-0-10504){ref-type="table"}.
### Establishing an RF prediction model on the basis of the selected genes
The VSURF method was used to select genes from 18,669 coding genes of TCGA RNAseq data for the establishment of RF prediction models, and ultimately, 11 genes were selected ([Fig. 2Ab](#f2-ol-0-0-10504){ref-type="fig"}). First, 19 genes that had the most relevance to the prognosis of EEA were selected ([Fig. 2Ac](#f2-ol-0-0-10504){ref-type="fig"}). To further reduce the number of genes for the RF models, 11 genes ([Table II](#tII-ol-0-0-10504){ref-type="table"}) were selected from these 19 genes as the input factors. For seven of the 11 genes \[low density lipoprotein receptor class A domain-containing 2 (LDLRAD2) (OS, P\<0.05; RFS, P\>0.05), 24-dehydrocholesterol reductase (DHCR24) (OS, P\<0.05; RFS, P\<0.05), EF-hand calcium-binding domain-containing protein 6 (EFCAB6) (OS, P\<0.05; RFS, P\<0.05), epithelial-splicing-regulatory-protein 1 (ESRP1) (OS, P\<0.05; RFS, P\<0.05), apolipoprotein L2 (APOL2) (OS, P\>0.05; RFS, P\<0.05), derlin-1 (DERL1) (OS, P\<0.05; RFS, P\<0.05) and mediator complex subunit 8 (MED8) (OS, P\<0.05; RFS, P\<0.05), the gene expression was significantly associated with the survival of EEA (P\<0.05; [Fig. 3](#f3-ol-0-0-10504){ref-type="fig"}). The classification ability of the 11 genes ([Fig. 2Ad](#f2-ol-0-0-10504){ref-type="fig"}) was approximately equal to the 19 genes ([Fig. 2Ac](#f2-ol-0-0-10504){ref-type="fig"}). In the training set, the out-of-bag (OOB) error of RF model-1 established by the 11 genes was 15% ([Fig. 2Ad](#f2-ol-0-0-10504){ref-type="fig"}). In the testing set, when RF model-1 was used to validate the 21 EEA samples from the PKUPH cohort, its classification accuracy was 71.43%.
### Establishing an RF prediction model on the basis of the clinical features
The VSURF method was used to select clinical features for establishing RF prediction models, and the grade was selected ([Fig. 2B](#f2-ol-0-0-10504){ref-type="fig"}). The results indicated that grade and FIGO stage were the most relevant to the EEA prognosis ([Fig. 2Bc](#f2-ol-0-0-10504){ref-type="fig"}). To further reduce the number of clinical factors in the RF models, grade was finally chosen as the input factor ([Fig. 2Bd](#f2-ol-0-0-10504){ref-type="fig"}). Grade had an almost equal ability to assign a classification compared with the 'grade combined with FIGO stage' ([Fig. 2Bc, 2Bd](#f2-ol-0-0-10504){ref-type="fig"}). In the training set, the OOB error of the RF model-2 established by grade was 0.39 ([Fig. 2Bd](#f2-ol-0-0-10504){ref-type="fig"}). When RF model-2 was used to validate the 21 EEA samples from the PKUPH cohort, the classification accuracy was 66.67% ([Fig. 4A](#f4-ol-0-0-10504){ref-type="fig"}).
### Establishing a RF combined model on the basis of the 'genes and clinical features'
Molecular biological characteristics and traditional features serve important roles in EEA prognosis. Thus, a RF-combined model-3 for EEA prognosis was established by combining '11 genes and grade'. In the training set, the OOB error of RF model-3 established by '11 genes and grade' was 0.15. When RF model-3 was used to validate the 21 EEA samples from the PKUPH cohort, its classification accuracy was 80.95% ([Fig. 4B](#f4-ol-0-0-10504){ref-type="fig"}). The classification accuracy of the RF model established by '11 genes, grade and stage' was 80.95% ([Fig. 4C](#f4-ol-0-0-10504){ref-type="fig"}), further proving that grade alone had equal classification ability compared with 'grade combined with FIGO stage'.
Discussion
==========
Although the prognosis of EEA is good, extensive heterogeneity can expose patients to recurrent disease and poor prognosis ([@b3-ol-0-0-10504],[@b4-ol-0-0-10504]). Treatments for EEA have become more complicated, as the histological classification, adjuvant therapies, indications and modalities for lymphadenectomy, and the classifications used to predict relapse risk factors have all changed ([@b21-ol-0-0-10504]). Traditional clinical criteria are not enough to predict EEA prognosis accurately, although studies have demonstrated that a number of clinical factors, including tumor grade, age, comorbidities, tumor diameter, American Society of Anesthesiologists score ([@b22-ol-0-0-10504]), lymphovascular space involvement and postoperative complications at 30 days, serve important roles in the prognosis of endometrial cancer ([@b23-ol-0-0-10504]--[@b25-ol-0-0-10504]). For the limits of conventional traditional methods used for histological classification of endometrial cancer subtypes, Barlin *et al* suggested a combination of molecular and conventional characteristics as classifications for better appraisal of prognostic and predictive factors ([@b26-ol-0-0-10504]). Combining traditional clinical factors and molecular biological characteristics for the prognosis of EEA is important.
Machine-learning methods ([@b27-ol-0-0-10504],[@b28-ol-0-0-10504]) can provide increased prediction accuracy and can account for complex interactions among predictors. In addition, machine-learning approaches tend to be more suitable than traditional statistical methods for certain situations, such as cancer prognostic prediction, which involves a certain number of potential predictors ([@b28-ol-0-0-10504]). In machine learning, traditional classifiers are usually desired for prediction accuracy and easily fit in with clinical norms, whereas RF stands out for its own inherent characteristics, which include a better generalization performance and excellent classification results ([@b10-ol-0-0-10504],[@b29-ol-0-0-10504]). RF has also been demonstrated to be highly suitable for reducing the dimensionality of the data ([@b29-ol-0-0-10504]); it has been successfully used in numerous scientific realms, such as evaluating cancer-associated cognitive impairment, disease prediction, genetics, proteomics and informatics ([@b29-ol-0-0-10504]--[@b31-ol-0-0-10504]), but currently has no application in the prediction of EEA prognosis. Not only are RF good classifiers, but they are also increasingly used as feature-selection methods. In the present study, the VSURF method was used to identify informative factors that were relevant to the prognosis of EEA. The selected factors were then used to design a good RF predictive model.
In the present study, grade and 11 genes were selected for the establishment of an RF model. The selected 11 genes were involved in a number of important biological processes and potentially affect the prognosis of EC. LDLRAD2 is an integral component of the cell membrane. The present study indicated that LDLRAD2 was associated with the prognosis of EEA, but the definite biological significance of LDLRAD2 remains to be investigated. In addition, DHCR24 serves important roles in anti-apoptosis, cell cycle arrest, the negative regulation of cell death and the regulation of caspase activity; these biological processes are associated with poor prognosis ([@b32-ol-0-0-10504]). Dai *et al* ([@b33-ol-0-0-10504]) identified that insulin-induced cholesterol synthetase DHCR24 aggravates the invasion of cancer and the resistance to progesterone in endometrial carcinoma. A previous study also demonstrated that DHCR24 is able to predict poor clinicopathological features of patients with bladder cancer, and that its expression may promote bladder cancer cell proliferation via several oncogenesis-associated biological processes (for example, via estrogen response, heme metabolism, the p53 pathway, cholesterol homeostasis, mammalian target of rapamycin complex 1 signaling, peroxisomes, xenobiotic metabolism, glycolysis and protein secretion) ([@b34-ol-0-0-10504]). EFCAB6 and MED8 genes serve important roles in the transcription of genes, including certain prognosis-associated genes. ESRP1 and embryonic lethal abnormal visual system-like neuron-specific RNA-binding protein 4 (ELAVL4) participates in RNA processing, mRNA processing and mRNA metabolic processing. Li *et al* ([@b35-ol-0-0-10504]) demonstrated that ESRP1 inhibited the invasion and metastasis of lung adenocarcinoma, and served a role in regulating proteins involved in the epithelial-to-mesenchymal transition. ESRP1 was associated with prognosis in epithelial ovarian cancer ([@b36-ol-0-0-10504]) and human colorectal cancer ([@b37-ol-0-0-10504]). Expression of the ELAVL4 gene was demonstrated to be a diagnostic and prognostic marker of bone marrow lesions in patients with neuroblastoma and male patients with meningioma ([@b38-ol-0-0-10504],[@b39-ol-0-0-10504]). APOL2 serves important roles in the acute inflammatory response, lipid transport, the steroid metabolic process and the cholesterol metabolic process. APOL2 was found to be overexpressed in ovarian/peritoneal carcinoma and may provide a molecular basis for therapeutic target discovery ([@b40-ol-0-0-10504]). DERL1 participated in the endoplasmic reticulum (ER)-nuclear signaling pathway, the ER-associated protein catabolic process and the ER to cytosol process. The results of a previous study have indicated that the expression of DERL1 distinguishes malignant from benign canine mammary tumors ([@b41-ol-0-0-10504]). Post-glycosylphosphatidylinositol attachment to proteins 3 (PGAP3) is involved in protein amino acid lipidation, the glycerophospholipid metabolic process, the lipid biosynthetic process, the lipoprotein metabolic process and the lipoprotein biosynthetic process. Previous studies demonstrated that lipid metabolism disorders serve an important role in endometrial cancer ([@b42-ol-0-0-10504],[@b43-ol-0-0-10504]). PGAP3 may affect the prognosis of EEA by regulating the lipid metabolism process. Transmembrane protein 27 (TMEM27) serve important roles in proteolysis. Javorhazy *et al* ([@b44-ol-0-0-10504]) demonstrated that a lack of TMEM27 expression in conventional renal cell carcinoma defines a group of patients as at a high risk of cancer-associated mortality.
The use of the RF model for the prediction of EEA prognosis when deciding whether to recommend adjuvant therapies is of great importance, particularly for patients with FIGO stage I disease. Those who have a low risk of relapse according to traditional clinicopathological risk factors may not have to receive postoperative adjuvant chemoradiotherapy. The results from previous studies have indicated that a large proportion of patients with EEA, who were at a low risk of relapse according to the traditional criteria and had not received postoperative adjuvant chemoradiotherapy, eventually relapsed or deceased ([@b5-ol-0-0-10504],[@b6-ol-0-0-10504]). The RF prediction model derived on the basis of clinical features and gene expression is promising for providing an individualized and more accurate prediction for patients with EEA. Combining the predictive results of the RF model and traditional criteria could also be used for better stratification of patients in clinical trials, as well as for providing more accurate counseling ideas for patients.
Two nomograms ([@b45-ol-0-0-10504],[@b46-ol-0-0-10504]) established by traditional characteristics for the predictive survival of EC have been produced, and their training accuracies were between 0.71 and 0.78. The first nomogram consists of five simple clinical features, including FIGO stage, age at diagnosis, final histological grade, negative lymph nodes and histological subtype ([@b45-ol-0-0-10504]). The second nomogram was validated in randomly selected patients ([@b46-ol-0-0-10504]) and indicated that tumor grade, age and lymphovascular space involvement were highly predictive for all outcomes. The establishment of the two nomograms was based on Cox regression analyses. Previous studies have demonstrated that machine-learning approaches appear to be more suitable than traditional statistical methods for some situations, such as the prediction of cancer prognosis, which involves a certain number of potential predictors ([@b10-ol-0-0-10504],[@b11-ol-0-0-10504]); thus, it may be more suitable to build such nomograms with machine-learning methods such as RF. In addition, biological characteristics and clinical features were particularly important in the prognosis of EEA. Using a combination of molecular and conventional characteristics as classifications would provide a better appraisal of prognostic and predictive factors, and the combination of traditional clinical factors and molecular biological characteristics is very important for the prognosis of EEA.
The classification accuracy of the RF prediction model combined with traditional clinicopathological features and gene expression was markedly higher than that of the RF models that were based on the traditional clinicopathological features or gene expression alone, indicating that traditional clinicopathological features and gene expression were important factors for the prognosis of EEA. The inclusion of numerous patients and prognosis-associated clinical features in the establishment of the RF prediction models is vital, and unfortunately, the number of samples in the present study was limited. In future research, more clinical samples and more clinical features will be collected for RF model establishment. The RF prediction model presented within the present study could provide a more individualized and accurate estimation of relapse and/or mortality for patients diagnosed with EEA following primary therapy. The RF model could also be used for better stratification of patients in clinical trials and for providing more accurate counseling ideas for patients.
To the best of our knowledge, the present study is the first to establish an EEA predictive model that combines genes and traditional features using RF. The RF model derived on the basis of the '11 genes and grade' achieved better predictive performances than RF models established by either the 11 genes or grade alone, indicating that the RF model derived on the basis of the 'genes and clinical features' had a stronger predictive ability for the prognosis of EEA.
Not applicable.
Funding
=======
The present study was supported by the National Natural Science Foundation of China (grant nos. 81502237, 81272869 and 81672571), the National Key Technology Research and Development Program of the Ministry of Science and Technology of China (grant no. 2015BAI13B06), and the Basic research project of Peking University (grant no. BMU2018JC005).
Availability of data and materials
==================================
The datasets used and/or analyzed during the present study are available from the corresponding authors upon reasonable request.
Authors\' contributions
=======================
FY and XS wrote the manuscript, collected the clinical information and performed the statistical analyses. LZ, XL, YC, JZ, XH and JL designed the study and revised the manuscript. SL analyzed the data. JW conceived and supervised the study and approved the final manuscript. All authors read and approved the manuscript and agree to be accountable for all aspects of the research to ensure that the accuracy or integrity of any part of the work is appropriately investigated and resolved.
Ethics approval and consent to participate
==========================================
The present study was approved by the Ethics Committee of Peking University People\'s Hospital (Beijing, China). All participating patients received and provided written informed consent prior to joining the study.
Patient consent for publication
===============================
Not applicable.
Competing interests
===================
The authors declare that they have no competing interests.
EEA
: endometrioid endometrial adenocarcinoma
TCGA
: The Cancer Genome Atlas
PKUPH
: Peking University People\'s Hospital
VSURF
: variable selection method using Random Forests
PE
: primary EEA
R/D-PE
: relapsed or deceased primary EEA
RF
: Random Forest
{#f1-ol-0-0-10504}
{#f2-ol-0-0-10504}
{#f3-ol-0-0-10504}
{#f4-ol-0-0-10504}
######
Clinicopathological characteristics of patients with EEA in TCGA and PKUPH cohorts.
TCGA cohort PKUPH cohort
-------------------------- ------------- ------------- -------------- --------- ------------- ------------- ------------- -------
Age, years 0.09 0.631
Median (range) 64 (35--90) 62 (35--89) 65 (35--90) 55 (31--75) 51 (41--75) 56 (31--63)
\<60, n 55 28 27 13 10 5
≥60, n 99 36 63 8 3 3
Grade, n 0.005 0.930
1--2 80 42 38 18 11 7
3 74 22 52 3 2 1
FIGO, n 0.005 0.203
I 105 52 53 12 9 3
II--IV 49 12 37 9 4 5
Menopause status, n 0.789 0.131
Premenopausal 12 4 8 4 4 0
Postmenopausal 129 54 75 17 9 8
Unknown 13 6 7 0 0 0
ER status, n 0.133
Positive NA 19 13 6
Negative 2 0 2
PR status, n 0.381
Positive NA 20 13 7
Negative 1 0 1
Lymph node status, n 0.008 0.716
Positive 22 4 18 2 1 1
Negative 46 25 21 19 12 7
Unknown 86 35 51 0 0 0
Adjuvant radiotherapy, n \<0.001 0.67
Yes 41 0 41 11 6 5
No 110 64 46 9 6 3
Unknown 3 0 3 1 1 0
Adjuvant chemotherapy, n 0.599
Yes NA 13 7 6
No 6 4 2
Unknown 2 2 0
BMI, n \<0.001 0.659
\<28 40 40 0 10 7 3
≥28 108 24 84 11 6 5
Unknown 6 0 6 0 0 0
EEA, endometrioid endometrial adenocarcinoma; TCGA, The Cancer Genome Atlas; PKUPH, Peking University People\'s Hospital; PE, primary EEA samples; R/D-PE, relapsed or deceased primary EEA samples; FIGO, International Federation of Gynecology and Obstetrics; BMI, body mass index.
######
Genes selected by Random Forest feature selection that may contribute to the prognosis of endometrial adenocarcinoma.
Gene Chromosome no. Definition
--------- ---------------- ------------------------------------------------------------------------------------
LDLRAD2 1 Low density lipoprotein receptor class A domain containing 2
DHCR24 1 24-dehydrocholesterol reductase
EFCAB6 22 EF-hand calcium-binding domain 6
ESRP1 8 Epithelial splicing regulatory protein 1
APOL2 22 Apolipoprotein L2
DERL1 8 Derlin 1
MED8 1 Mediator complex subunit 8
PGAP3 c7 Post-GPI attachment to proteins 3
ELAVL4 1 Embryonic lethal abnormal visual system-like neuron-specific RNA binding protein 4
TMEM27 X Transmembrane protein 27
ATF7IP2 16 Activating transcription factor 7 interacting protein 2
[^1]: Contributed equally
| {
"pile_set_name": "PubMed Central"
} |
{#sp1 .34}
| {
"pile_set_name": "PubMed Central"
} |
Introduction {#s1}
============
MicroRNAs are small, single-stranded RNAs that have been highly conserved during evolution and function by repressing target gene expression. miRNAs have recently emerged as critical modulators of gene expression networks in mammals, and their impaired expression or function has been linked to a variety of human diseases [@pone.0063074-OConnell1]. A wide range of human cancer cell types display dysregulated miRNA expression patterns, and there is overwhelming evidence that some miRNAs are functionally relevant in malignancies by playing imperative tumor suppressor roles or by acting as aggressive oncogenes. In addition to cancer, miRNAs are also perturbed in cardiovascular disease [@pone.0063074-Ono1], neurological disorders [@pone.0063074-MezaSosa1], and autoimmunity [@pone.0063074-OConnell1], [@pone.0063074-Hu1], and this is consistent with miRNAs having obligatory regulatory roles in a variety of human organ systems. Beyond disease, miRNAs can also participate in the formation of induced pluripotent stem (iPS) cells [@pone.0063074-Mallanna1], which hold significant promise in the field of regenerative medicine.
Despite these important and clinically significant roles for miRNAs, our ability to manipulate miRNA expression and function in human cells remains a challenging task. For instance, unlike protein coding mRNAs, siRNAs cannot be used to reduce miRNA levels within cells. Delivery of oligonucleotides antisense to target miRNAs has had some success, but is limited to certain cell types that can uptake these oligonucleotides with high efficiency, such as hepatocytes, and requires constant delivery of fresh inhibitor [@pone.0063074-Filipowicz1]. Thus, novel approaches to regulating miRNA expression or function in human cells are clearly needed and should have a substantial impact on our ability to study human physiology, combat human diseases, and regenerate damaged tissues. Following their transcription in the cell nucleus, miRNAs undergo a series of processing steps before reaching maturity [@pone.0063074-Filipowicz2]. The fully processed miRNA is loaded into RISC and then mediates target gene repression by directing the RISC complex to specific mRNA 3′ UTRs containing cognate binding sites for the miRNA. This interaction between the miRNA and 3′ UTR is dependent upon a 6--8 nucleotide sequence found in the 5′ end of the miRNA called the "seed" sequence. This sequence must have perfect complementarity with its 3′ UTR binding site for repression to occur. Disruption of the seed region of the miRNA or cognate binding site abolishes repression and thus miRNA function. Therefore, miRNAs can be modulated by controlling their expression levels or by disrupting their seed:target interaction.
Recently, a novel class of DNA-binding proteins from *Xanthomonas* plant pathogens, called Transcription Activator-Like Effectors (TALEs), have been shown to bind DNA in a highly sequence specific manner and mediate gene modifications based upon their fusion to trans-activation, repression or nuclease domains [@pone.0063074-Bogdanove1]. Importantly, because TALE proteins are made up of modules, with each interchangeable module recognizing specific DNA bases, TALEs can theoretically be engineered to bind virtually any DNA sequence. Just recently TALE proteins have been shown to function in human cells indicating that this technology can be used to modify specific human genes [@pone.0063074-Cermak1], [@pone.0063074-Miller1].
In the present study, we have developed custom TALENs that have been engineered to target 4 specific miRNAs with established functional importance, and these include miR-155, miR-155\*, miR-146a and miR-125b. We demonstrate that in all cases we can achieve sequence deletions within these genes that include disruptions to the miRNA seed sequence, and achieve complete miR-155 hairpin removal by using two TALEN pairs together. Furthermore, we observe bi-allelic modifications indicating that TALENs can disrupt both miRNA gene alleles within a human cell. This work describes a novel approach to targeting and disrupting miRNA genes in the human genome, and has important implications for both basic and translational research involving miRNAs.
Results {#s2}
=======
Design and construction of miRNA-targeting TALENs {#s2a}
-------------------------------------------------
To develop TALE proteins with the capacity to modify specific miRNA gene sequences, we first used an *in silico* approach to identify promising TALE protein pair binding sites that flank the sequences of a specific human miRNA of interest, miR-155\*. Our analysis was carried out using TALE-NT software (<https://boglab.plp.iastate.edu/node/add/talen>), and followed the parameters described in our Materials and Methods. This led to the identification of a putative TALE binding site flanking the miR-155\* sequence. Next, TALE-repeat variable dinucleotides (RVDs) corresponding to the specific target DNA sequences were cloned into expression plasmids using the Golden Gate assembly system as described in the Materials and Methods section. The expression plasmids were comprised of the TALE DNA binding modules, a short linker, and a modified FokI nuclease domain ([Fig. 1](#pone-0063074-g001){ref-type="fig"}). To increase target specificity, we used modified FokI nuclease domains that function as obligate heterodimers. Consequently, both the 'Left' and 'Right' TALE nuclease (TALEN) proteins that make up the pair must bind simultaneously to their DNA sites in the genome for Fok1 to heterodimerize and become an active enzyme. With this design, the DNA spacer sequence between the bound TALENs is cut, and deletions or other mutations are introduced during non-homologous end joining (NHEJ) DNA repair of the cut DNA.
{#pone-0063074-g001}
The miR-155\* targeting TALEN pair facilitates deletions in the miR-155\* sequence {#s2b}
----------------------------------------------------------------------------------
After construction of the miR-155\* TALEN pair (TALEN A) expression plasmids, we subjected this TALEN pair to a functional analysis pipeline to assess its ability to target and mutate the intended sequence in the human genome ([Fig. S1](#pone.0063074.s001){ref-type="supplementary-material"}). Equal amounts of the miR-155\* TALEN pair expressing plasmids, along with a GFP expressing vector, were transfected into human HEK 293T cells. 48 hours later, GFP+ cells were isolated using FACS ([Fig. 2A](#pone-0063074-g002){ref-type="fig"}) and the gDNA was extracted from the cells and analyzed for the presence of mutations to the desired miRNA gene region.
{#pone-0063074-g002}
To initially screen for the presence of DNA sequence modifications, short PCR amplicons (90--150 bp) that included the region of interest were generated from the gDNA samples ([Fig. 2A,B](#pone-0063074-g002){ref-type="fig"}). The PCR product was next subjected to a High Resolution Melt Analysis (HRMA) analysis, described previously [@pone.0063074-Dahlem1]. If TALEN-induced mutations were present in the template gDNA, the thermostability of the dsDNA population of renatured PCR amplicons would be different from amplicons produced using wild-type gDNA samples ([Fig. 2B](#pone-0063074-g002){ref-type="fig"}). Consistent with this, amplicons from gDNA taken from the miR-155\* TALEN A pair transfected cells had an altered thermostability compared to control cells not receiving the TALEN pair. Furthermore, the difference was more significant when we compared the sorted GFP+ cells with unsorted cells, while there were no thermostability differences between Wt cells and sorted GFP- cells from the TALEN transfected cell culture ([Fig. 2C](#pone-0063074-g002){ref-type="fig"}). Upon generating fluorescence difference plots, we found that the curves for the Wt and sorted GFP- samples were clustered around the baseline while the curves for the TALEN transfected samples (GFP+) were clearly above background ([Fig. 2D](#pone-0063074-g002){ref-type="fig"}). These data indicated that the miR-155\* TALEN A pair caused sequence modifications, potentially within the desired region.
The PCR products containing TALEN-targeted mutations were next subjected to TOPO cloning. Individual TOPO clones containing single PCR DNA fragments were analyzed by gel electrophoresis or sequenced ([Fig. 2E,F](#pone-0063074-g002){ref-type="fig"}). Several clones were found to be of different sizes, indicating the presence of deletions within the targeted genomic region ([Fig. 2E](#pone-0063074-g002){ref-type="fig"}). The relative sizes of the deletions were also consistent with sequencing data, which confirmed the presence of deletions of varying lengths ([Fig. 2F](#pone-0063074-g002){ref-type="fig"}). These findings demonstrate that the miR-155\* TALEN A pair causes targeted deletions in the mature human miR-155\* sequence. Although these deletions are specific to the miR-155\* sequence, they will also disrupt the stem-loop structure containing miR-155 and miR-155\*. This is expected to inhibit processing and production of both mature miR-155 and miR-155\*.
The miR-155\* TALEN pair elicits both bi-allelic and mono-allelic mutations {#s2c}
---------------------------------------------------------------------------
Although able to successfully target the miR-155\* sequence using our TALEN A pair, we next assessed if mono- or bi-allelic modifications were occurring ([Fig. 2B](#pone-0063074-g002){ref-type="fig"}). Because each cell has two alleles of an individual miRNA gene, it was important to determine if we could disrupt both copies, which would completely abolish the function of the target miRNA. To make this determination, FACS-sorted TALEN-transfected GFP+ cells were used to generate clonal populations ([Fig. 3A](#pone-0063074-g003){ref-type="fig"}). gDNA was extracted from clonal populations and subjected to PCR to amplify the targeted sequence. The amplicons from 3 different cell clones with TALEN-mediated deletions were then subjected to TOPO cloning followed by sequencing to decipher the precise mutations that had occurred in the targeted region in each clonal population. The sequencing results revealed that some cell clones had two unique deletions and no Wt alleles, consistent with a bi-allelic alteration, while other clones had one Wt allele and a deleted sequence, consistent with a mono-allelic modification ([Fig. 3B,C](#pone-0063074-g003){ref-type="fig"}). These data indicate that bi-allelic modifications to miRNA genes can be mediated by TALENs.
{#pone-0063074-g003}
Complete deletion of the miR-155 hairpin by using two TALEN pairs together {#s2d}
--------------------------------------------------------------------------
In order to target miR-155, we designed another TALEN pair (TALEN C) that bind just upstream from the 8-nucleotide miR-155 seed region. Sequencing showed that the miR-155 TALEN C caused deletions in the desired locus that included the miR-155 seed region in some cases ([Fig. 4A](#pone-0063074-g004){ref-type="fig"}).
{#pone-0063074-g004}
Since both miR-155 TALEN pairs A and C target each end of the miR-155 hairpin sequence, we tested whether the combination of these two TALEN pairs could delete the entire sequence that constitutes pre-miR-155 ([Fig. 4B](#pone-0063074-g004){ref-type="fig"}). Both TALEN A and C pairs were co-transfected into 293T cells and gDNA was analyzed by PCR and TOPO cloning as described in [Fig. 2](#pone-0063074-g002){ref-type="fig"}. Using gel electrophoresis we found that several clones were approximately 80 bps smaller than the Wt sequence, indicating the presence of larger deletions within the targeted miR-155 locus ([Fig. 4C](#pone-0063074-g004){ref-type="fig"}). Sequencing data confirmed the presence of deletions that span the entire miR-155 hairpin sequence ([Fig. 4D](#pone-0063074-g004){ref-type="fig"}). These findings demonstrate that the combination of the two TALEN pairs caused complete deletion of human pre-miR-155.
Targeting of human miR-146a and miR-125b1 using engineered TALENs {#s2e}
-----------------------------------------------------------------
We next used this same overall approach to design, build and test 2 other TALEN pairs against additional human miRNAs of interest, including miR-146a and miR-125b1. For miR-125b1, we designed the TALENs to target sequences just upstream instead of flanking the miRNA seed region, and did so to stay within the design parameters. However, for miR-146a, we identified promising TALEN sites that flanked its seed sequence ([Fig. 5A](#pone-0063074-g005){ref-type="fig"}). Like the miR-155\* and miR-155 TALENs, each new TALEN pair successfully targeted and mutated the expected miRNA gene sequences, albeit at different efficiencies ([Fig. 5A,B](#pone-0063074-g005){ref-type="fig"} and [Table 1](#pone-0063074-t001){ref-type="table"}). Interestingly, all of the TALEN pairs caused at least some deletions that disrupted the seed sequences of each respective miRNA, even if their binding sites did not flank, but were near, the seed. These results indicate that TALENs can be routinely used to disrupt human miRNA seed sequences, and that the design parameters consistently allow for successful targeting of a DNA sequence as small as an 8-nucleotide miRNA seed found in specific DNA locations and contexts.
{ref-type="fig"}) to detect mutations in the targeted regions. (A, B) The upper panel shows the schematics of the TALEN pair binding sites adjacent to (A) miR-146a and (B) miR-125b1. The lower panel shows the sequence alignments between the Wt and mutated miRNA genes. The left and right TALEN binding sites are highlighted in yellow and the miRNA seed regions are boxed in red.](pone.0063074.g005){#pone-0063074-g005}
10.1371/journal.pone.0063074.t001
###### Mutation rate mediated by TALEN pairs in transfected 293T cells.
{#pone-0063074-t001-1}
Talen targeting miRNA Total Topo clones sequenced Topo clones with mutations Mutation rate
----------------------- ----------------------------- ---------------------------- ---------------
miR-155\* Talen A 19 10 52.6%
miR-155 Talen C 12 5 41.7%
miR-146a 34 5 14.7%
miR-125b1 18 2 11.1%
Discussion {#s3}
==========
Our study set out to determine whether TALEN technology could be used to target miRNA genes in human cells. Unlike protein coding genes that are typically made up of thousands of nucleotides from which optimal TALEN binding sites can be found, relevant miRNA gene sequences are considerably smaller, which limits the likelihood of finding well positioned TALEN sites. However, our results indicate that TALENs can be repeatedly designed to target specific miRNA loci and achieve miRNA seed disruption, or miRNA hairpin removal when two TALEN pairs are used together. Because miRNA-mediated gene repression is dependent upon its seed sequence, this approach can be used to permanently block the function of specific human miRNAs.
Although our TALENs successfully disrupted miRNA gene seeds, the resulting deletions were heterogeneous in nature as reported by others [@pone.0063074-Cermak1], [@pone.0063074-Miller1]. Thus, although one can disrupt miRNA function using this method, the resulting modification is variable. It has recently been demonstrated that DNA editing can be achieved using TALENs and a single-stranded donor DNA molecule with homologous arms [@pone.0063074-Bedell1]. Future work should use this method to edit miRNA seed sequences in a manner that prohibits, or alters, their targeting capacity or specificity in a controlled manner. Furthermore, such an editing approach could also be used to modify miRNA-binding sites in the 3′ UTRs of specific target genes, or polymorphisms within *cis* regulatory elements that influence miRNA expression. A recent example of such a polymorphism is found in miR-146a, which has a G/C polymorphism within in pre-miRNA sequence that reduces its expression and contributes to a predisposition to papillary thyroid cancer [@pone.0063074-Jazdzewski1]. Polymorphisms within the miR-155 gene have also been associated with its altered expression in human Multiple Sclerosis patients [@pone.0063074-Paraboschi1].
We also found that each TALEN pair had a different functional efficiency as determined by the rate of target allele mutations ([Table 1](#pone-0063074-t001){ref-type="table"}). This indicates that despite following established design guidelines, additional factors are able to influence TALEN function. These may include chromatin structure, DNA modifications such as methylation, or other DNA sequence variations that influence TALEN binding dynamics. However, such determinants are presently being investigated, as this is a relatively new field of study. The capacity to deliver TALENs to precise cell types is also a challenging endeavor. Similar to other studies, we have demonstrated that transfection of cells with plasmids encoding the TALEN pair can be used to express TALENs in target cells [@pone.0063074-Miller1]. However, the development of viral vector systems that enable transient expression of TALENs in specific cell types is necessary for many important applications *in vivo* [@pone.0063074-Holkers1].
As we continue to understand how miRNAs regulate mammalian biology, both in physiological and pathological contexts, it is becoming increasingly necessary to develop tools with the ability to specifically target and modify human miRNA genes *in vivo*. Based upon our findings here, TALENs make excellent candidates to achieve miRNA gene targeting and manipulation in a variety of relevant human cell types, including those with important therapeutic applications, such as stem cells, neurons and primary tumors.
Materials and Methods {#s4}
=====================
Cell culture and transfection {#s4a}
-----------------------------
HEK 293T cells were obtained from the American Type Culture Collection (Rockland, MD) and were cultured with DMEM supplemented with antibiotics and 10% FBS. Cells were maintained at 37°C in a humidified incubator supplied with 5% CO~2~. For transfection of plasmids, TransIt-293 transfection reagent (Mirus, WI) was used to transfect 293T cells according to the manufacturer\'s protocol. The TALENs were transfected at a molar ratio of 1∶1.
TALEN target site design {#s4b}
------------------------
TALEN target sites were designed as described in Dahlem et al [@pone.0063074-Dahlem1]. Briefly, the TALEN Targeter (old version) program at <https://boglab.plp.iastate.edu/node/add/talen> was used to scan the sequences flanking the miRNAs of interest (including miR-155\*, miR-155, miR-146a and miR-125b1) for potential TALEN pair target sites. Site selection was restricted using the following parameters: 1) spacer length between TALENs: 14--17; 2) TALE repeat array length of 16--21 and 3) by applying all additional options that restrict target choice. Preference was given to target sites where the spacer centered on or near the seed regions of the miRNAs and therefore would likely induce loss of function deletions. The uniqueness of potential TALEN target sequences was determined using the Target Finder (<https://boglab.plp.iastate.edu/node/add/talen>) and a BLAST analysis, ensuring that highly similar Left and Right binding sites in close proximity did not exist at other regions in the human genome. The designed TALEN pair and spacer sequences are as follows: miR-155\*-Talen A-left: GCCTCCAACTGACTCCT; miR-155\*-Talen A-right: AGTGTATGATGCCTGTTACT; miR-155\*-Talen A-spacer: ACATATTAGCATTAAC; miR-155-Talen C-left: ATGCCTCATCCTCTGAGT; miR-155-Talen C-right: AGGCTGTATGCTGTTAATGCT; miR-155-Talen C-spacer: GCTGAAGGCTTGCTGT; miR-146a-Talen-left: GTGTATCCTCAGCTT; miR-146a-Talen-right: ATGGGTTGTGTCAGTGTCAG; miR-146a-Talen-spacer: TGAGAACTGAATTCC; miR-125b1-Talen-left: CCCTGAGACCCTAACTTGTGAT; miR-125b1-Talen-right: ACGGGTTAGGCTCTTGGG; miR-125b1-Talen-spacer: GTTTACCGTTTAAATCC.
TALEN assembly and expression plasmid construction {#s4c}
--------------------------------------------------
TALEN pairs were designed and constructed by the Mutation Generation and Detection Facility at the University of Utah (<http://www.cores.utah.edu/>). The TALEN Golden Gate kit described by Cermark et al [@pone.0063074-Cermak1] was used and TALENs were assembled as described by Dahlem et al [@pone.0063074-Dahlem1]. Each nucleotide of a target site is recognized by one repeat module of the TALEN protein. Two amino acids within each module, called the Repeat Variable Di-residues (RVDs), are responsible for nucleotide recognition. RVDs NI, NN, NG, and HD bind to nucleotides A, G, T, and C, respectively. The TALEN Golden Gate Kit contains plasmids containing each of the individual RVD modules along with intermediate cloning plasmids and a final expression vector. These reagents were used to construct specific TALEN expression vectors. The TALEN Golden Gate Kit (\#1000000024) was obtained from Addgene. Briefly, successive rounds of Golden Gate cloning assembly were used to generate TALEN expressing plasmids with n RVD repeat modules. First, two arrays corresponding to repeat modules 1--10 and 11-n-1 were assembled into separate intermediate vectors and those acquiring RVD arrays were screened on IPTG/X-gal plates. Correct assembly was determined first by XbaI and AflII restriction enzyme digestion followed by sequencing of plasmids with the correct insert sizes. Second, the two arrays and sequences encoding the n^th^ motif were assembled into the final expression backbone vectors pCS2TAL3-DDD and pCS2TAL3-RRR to generate a left and right TALEN expressing plasmids, respectively, followed by screening on IPTG/X-gal plates. Correct assembly was determined first by SphI and BamHI restriction enzyme digestion followed by sequencing of plasmids with the correct insert sizes.
The pCS2TAL3-DDD and pCS2TAL3-RRR final expression vectors were modified from pCS2TAL3-DD and pCS2TAL3-RR plasmids described in Dahlem et al [@pone.0063074-Dahlem1] with extra mutations in the FoKI nuclease domain. In pCS2TAL3-DDD the amino acid change of H496D was introduced into the Fok I domain and in pCS2TAL3-RRR the amino acid change H537R was introduced [@pone.0063074-Szczepek1], [@pone.0063074-Doyon1]. The DDD and RRR mutations within the Fok I nuclease domains require that the TALENs function as obligate heterodimers, requiring both 'Left' and 'Right' monomers to simultaneously recognize their cognate binding sites to achieve nuclease activity. The pCS2TAL3 final backbone vectors contain the simian IE94 cytomegalovirus eukaryotic enhancer/promoter (CMV), the recognition sequence used by the prokaryotic SP6 RNA polymerase (SP6), and the polyadenylation signal sequence derived from SV40 (SV40pA) taken from the pCS2+ plasmid (<http://sitemaker.umich.edu/dlturner.vect> ors). Other domains include a nuclear localization signal (NLS); the FLAG epitope (Flag); and truncated TAL protein N-terminus and C-terminus (TAL-N′ and TAL-C′) sequences derived from pTAL3 [@pone.0063074-Cermak1].
Genomic DNA extraction {#s4d}
----------------------
Genomic (g) DNA was extracted from transfected 293T cells or single cell clones by using DNeasy Blood & Tissue Kit (QIAgen, MD). The average gDNA concentration was adjusted to 30 ng/ul for PCR reactions.
High Resolution Melt Analysis (HRMA) {#s4e}
------------------------------------
To detect TALEN-induced mutations by HRMA, a ∼100--150 bp amplicon that included the entire genomic target site was amplified by PCR. Primers flanking the target site were used to amplify the genomic region in a 10 ul PCR reaction containing: 1 ul gDNA (30 ng/µl), 1× LightScanner Master Mix (containing the LC Green Plus dye, Idaho Technology), 200 µM dNTP, and 200 nM each Forward and Reverse primers. Amplification/duplex formation conditions were: 94°C, 3 min; 50 cycles \[94°C, 30 s; 70°C, 17 s\]; 94°C, 30 s; 25°C, 30 s; 10°C. HRMA data was collected on a LightScanner (Idaho Technology) and analyzed using LightScanner Call-IT Software. The primer sequences are available upon request.
TALEN-induced mutation screening {#s4f}
--------------------------------
Amplicons were cloned using the TOPO TA Cloning Kit (Invitrogen). TOPO plasmids were digested and the inserted PCR DNA fragments were analyzed by gel electrophoresis. In parallel, TOPO plasmids containing the amplified DNA clones were sequenced at the DNA sequencing core facility at University of Utah.
Supporting Information {#s5}
======================
######
**Experimental plan used to develop miRNA-targeting TALENs.** TALEN pairs targeting different miRNAs were designed, constructed and transfected into 293T cells along with a GFP expression vector. After transfection, 293T cells were subjected to FACS sort to isolate cells with the TALEN pairs. gDNA was extracted from Wt, unsorted, GFP- or GFP+ cells. The TALEN pair-targeted regions were amplified by PCR and subjected to HRMA or TOPO cloning and sequencing to determine the presence of mutations. Moreover, GFP+ cells were plated in 96 well plates to obtain single cell clones. Single cell clones were subjected to PCR and TOPO cloning analyses to determine if bi- or mono-allelic mutations were being generated within the TALEN targeted regions.
(TIF)
######
Click here for additional data file.
We would like to thank Dr. Dana Carroll and the Mutation Generation and Detection Facility at University of Utah for assisting with TALEN design and construction.
[^1]: **Competing Interests:**The authors have declared that no competing interests exist.
[^2]: Conceived and designed the experiments: RH TJD RMO. Performed the experiments: RH JW. Analyzed the data: RH RMO. Contributed reagents/materials/analysis tools: RH JW TJD DJG RMO. Wrote the paper: RH RMO.
| {
"pile_set_name": "PubMed Central"
} |
Background
==========
Sepsis is a major health challenge. Despite improved treatment options, sepsis remains a leading cause of death in intensive care units \[[@B1]\]. Lipopolysaccharide (LPS), or endotoxin, the major outer membrane component of gram-negative bacteria, is a potent inflammatory response stimulator \[[@B2]\]. In addition, LPS triggers inflammation in gram-negative sepsis \[[@B3]\]. Excessive amounts of gut-derived LPS released during intestinal hypo-perfusion have also been implicated in sepsis caused by gram-positive and fungal infections \[[@B4],[@B5]\]. LPS signaling is initiated by the activation of the myeloid differentiation factor 2 and toll-like receptor 4 (TLR4) complexes on myeloid cells \[[@B2],[@B6]\]. TLR4 has recently been shown to recognize endogenous danger-type, or 'alarmin,' factors, thereby implicating TLR4 as a tissue injury and microbial invasion sensor \[[@B7]\]. Studies using mouse strains deficient in TLR4 signaling \[[@B8],[@B9]\] or expression \[[@B10]-[@B13]\] or those using TLR inhibitors in wild-type mice \[[@B14],[@B15]\] confirmed that TLR4 contributes to bacterial clearance and the host inflammatory response in the infection setting \[[@B16]\].
Two missense single nucleotide polymorphisms (SNPs) in the *TLR4* gene, Asp299Gly/Thr399Ile, have been reported to be associated with endotoxin hypo-responsiveness to inhaled LPS \[[@B17]\]. This investigation was followed by a series of studies that explored the potential impact of these SNPs on the incidence and course of infectious diseases \[[@B18]\], such as septic shock with gram-negative bacterial infection \[[@B19]\]. Although some studies have shown a relevance of the Asp299Gly/Thr399Ile SNPs in gram-negative infections, others did not confirm this association \[[@B20]-[@B22]\]. Furthermore, recent studies using primary cells isolated from individuals bearing these mutations have indicated that the Asp299Gly/Thr399Ile haplotype has little or no effect on LPS responsiveness \[[@B23]\].
Recently, Sato et al. demonstrated the biological significance of a genetic variation of the *TLR4* gene called rs11536889. Functional analyses revealed that *TLR4* rs11536889 contributes to the translational regulation of TLR4 expression and has some influence on the response to LPS, possibly by binding to microRNAs, which act in post-transcriptional regulation \[[@B24]\]. A large study that included prostate cancer patients and age-matched controls from Sweden revealed an association between *TLR4* rs11536889 and prostate cancer \[[@B25]\]. Later, Hishida et al. observed that *TLR4* rs11536889 genotypes are associated with severe gastric atrophy in *helicobacter pylori*-seropositive Japanese subjects \[[@B26]\]. Zhou et al. found that the *TLR4* rs11536889 SNP is significantly associated with *hepatitis type B* virus recurrence after liver transplantation \[[@B27]\]. In addition, Miedema et al. found that this SNP is associated with an increased risk of chemotherapy-induced neutropenia in children with acute lymphoblastic leukemia \[[@B28]\]. These observations suggest that the rs11536889 genetic variation of the TLR4 gene may influence human inflammatory and/or malignant diseases \[[@B24]\].
This study aimed at exploring whether the putative regulatory *TLR4* rs11536889 genotypes relate to organ failure severity in critically ill patients with sepsis during their time in the intensive care unit. The outcomes of wild-type GG were compared to those of GC/CC.
Methods
=======
Patients
--------
Adult Caucasian patients admitted to the University Medical Center Goettingen (UMG) intensive care units (ICUs) between April 2012 and May 2013 were screened daily according to the American College of Chest Physicians/Society of Critical Care Medicine (ACCP/SCCM) criteria for sepsis, severe sepsis, or septic shock \[[@B29],[@B30]\]. Caucasian origin was assessed by questioning the patients, their next of kin or their legal representatives. The patient exclusion criteria were as follows: 1. age younger than 18 years; 2. pregnancy, nursing an infant, or planning to become pregnant or nurse an infant; 3. receiving an immunosuppressive therapy (e.g., cyclosporine or azathioprine) or cancer-related chemotherapy; 4. a documented or suspected acute myocardial infarction within the previous six weeks; 5. a history of New York Heart Association functional class IV chronic heart failure: 6. human immunodeficiency virus infection or end-stage process (e.g., progressive multifocal leukoencephalopathy or systemic *Mycobacterium avium* infection); 7. morbidity and death were considered imminent, the patient was classified as "do not resuscitate" or "do not treat", or the patient and/or a legally authorized representative was not committed to aggressive management; 8. the patient was not expected to survive the observation period of 28 days or was not likely to be given life support because of a preexisting, uncorrectable medical condition, including a poorly controlled neoplasm, end-stage lung disease, or home oxygen requirement; 9. the patient was in a chronic vegetative state or had a similar long-term neurologic condition; 10. participation in any other investigational study (drug or device); 11. the patient was unwilling or unable to be fully evaluated during the study period; and 12. the patient was a study-site employee or was an immediate family member of a study-site employee involved in the study. The study was approved by the University of Goettingen ethics committee, Goettingen, Germany (15/1/12) and conformed to the Declaration of Helsinki ethical principles (Seoul, 2008). Written informed consent was obtained either from patients or their legal representatives.
Data collection
---------------
The Sequential Organ Failure Assessment (SOFA) \[[@B31]\] and Acute Physiology and Chronic Health Evaluation (APACHE) II \[[@B32]\] scores were evaluated at the onset of sepsis. Organ function was assessed subsequently on days 2, 3, 5, 7, 14, 21 and 28, and organ failure was quantified, with the SOFA score as the primary outcome variable. Patients were followed up for a maximum of 90 days, and their deaths were recorded as a secondary outcome variable. The necessity of mechanical ventilation, vasopressor administration, or renal-replacement therapy as well as the ICU duration was recorded as secondary variables.
TLR4 rs11536889 genotyping
--------------------------
Peripheral blood monocytes (PBMCs) from approximately 30 ml of heparinized peripheral blood were isolated through Ficoll density gradient centrifugation according to standard procedures described previously \[[@B33]\]. Cell preparations were routinely assessed for viability (\>95%) by trypan blue dye exclusion.
Genomic DNA (gDNA) was purified from PBMCs using the AllPrep DNA Mini Kit (Qiagen, Hilden, Germany) according to the manufacturer's protocol. The isolated nucleic acid concentration and purity were determined by 260 and 280 nm optical density readings. DNA integrity was evaluated through 0.6% agarose gel electrophoresis.
Genotyping was performed using 4 ng of PBMC-derived gDNA in a commercially available genotyping assay (Assay ID C_31784034_10, Applied Biosystems, Darmstadt, Germany) in a total volume of 10 μl. The reactions were performed in a StepOnePlus sequence detection system (Applied Biosystems, Darmstadt, Germany) according to the supplier's instructions.
Statistical analyses
--------------------
The Hardy-Weinberg equilibrium exact test for deviation was performed using an online calculator, which was provided by the Institute of Human Genetics, Helmholtz Center Munich, Germany (<http://ihg.gsf.de/cgi-bin/hw/hwa1.pl>). Statistical analyses were performed with the Statistica (StatSoft, Tulsa, Oklahoma, USA, version 10) or R (The R Foundation for Statistical Computing, version 3.0.0) software. Significance, based on contingency tables, was calculated using two-sided Fisher's exact or chi-square tests, as appropriate. Two continuous variables were compared using the Mann-Whitney test. To estimate the significance of the clinical observations over the 28-day period, we fitted a linear regression model with the parameters day, genotype, and genotype-day interaction. The results were visualized by calculating the means and 95% confidence intervals (CIs) from normal distributions at each time point. Time-to-event data were compared using the log-rank test from the Statistica package survival. A p-value less than 0.05 was considered significant.
Results
=======
Study population
----------------
A total of 212 adult Caucasian patients with sepsis were enrolled into this study. Two patients were excluded; one patient fulfilled an exclusion criterion, as a B cell lymphoma diagnosis became known, and the other patient was excluded because informed consent was withdrawn by his legally authorized representative. Subsequently, the study population comprised 210 patients, in which 134 were male (64%), and 76 were female (36%; Table [1](#T1){ref-type="table"}). The patient ages ranged from 19 to 91 (median, 65). The sepsis subtypes were sepsis/severe sepsis (n = 100) and septic shock (n = 110). At baseline, the patient disease severity SOFA and APACHE II scores were 8.6 ± 4.1 and 21.3 ± 7.4, respectively (Table [1](#T1){ref-type="table"}). The comorbidities comprised hypertension, myocardial infarction history, chronic obstructive pulmonary disease (COPD), renal dysfunction, diabetes mellitus, chronic liver diseases, cancer history, and stroke history (Table [1](#T1){ref-type="table"}).
######
**Patient baseline characteristics with regard to the*TLR4*rs11536889 genotypes**
**All n = 210** **GG n = 146** **GC/CC n = 64** ***p*value**
---------------------------------- ----------------- ---------------- ------------------ --------------
Age \[years\] 63 ± 15 63 ± 16 63 ± 15 0.9116
Male, % 64% 46% 58% 0.2752
Body-mass index, mean ± SD 28 ± 9 28 ± 7 30 ± 13 0.1507
Severity of sepsis 48% 46% 52% 0.4576
Sepsis/severe sepsis, % 52% 54% 48%
Septic shock, %
SOFA score, mean ± SD 8.6 ± 4.1 8.9 ± 4.3 8.0 ± 3.7 0.2449
APACHE II score, mean ± SD 21.3 ± 7.4 21.6 ± 7.4 20.6 ± 7.6 0.4053
Comorbidities, %
Hypertension 58 59 1.0000
History of myocardial infarction 5 6 0.7586
COPD 18 17 1.0000
Renal dysfunction 11 20 0.0834
Diabetes mellitus (NIDDM) 11 8 0.6202
Diabetes mellitus (IDDM) 11 5 0.1938
Chronic liver diseases 7 9 0.7839
History of cancer 18 19 0.8481
History of stroke 6 8 0.7647
Recent surgical history, % 0.0631
Elective surgery 28 31 19
Emergency surgery 44 44 44
No history of surgery 28 24 37
Site of infection, % 0.1516
Lung 50 46 56
Abdomen 30 31 28
Bone or soft tissue 7 7 6
Surgical wound 2 3 0
Urogenital 1 0 3
Primary bacteremia 6 7 3
Other 4 5 3
Organ support, %
Mechanical ventilation 83 83 83 0.8933
Use of vasopressor 52 54 48 0.4576
Renal-replacement therapy 10 11 9 0.7300
The data are presented as the mean ± SD or percentages.
Disease severity at sepsis onset
--------------------------------
*TLR4* rs11536889 was successfully genotyped in all subjects. The genotype distribution was 146:62:2 (GG:GC:CC), which was consistent with the Hardy-Weinberg equilibrium (p = 0.12). The resulting 0.16 minor allele frequency was similar to that given for Caucasians in public databases. The rs11536889 GC and CC genotypes were pooled together because the size of the CC genotype group was too small (n = 2). Subsequently, patients of genotype GG were compared to that of CG/CC. There was no difference regarding age, gender, or body mass index related to the *TLR4* rs11536889 genotype. A comparison between the G homozygous patients and C allele carriers revealed no significant difference between the proportion of patients with sepsis/severe sepsis and septic shock at baseline (day 1 of sepsis; p = 0.4576). Furthermore, there were no SOFA and APACHE II score differences regarding the *TLR4* rs11536889 genotypes at sepsis onset, and there were no significant preexisting conditions differences between the *TLR4* rs11536889 genotypes (Table [1](#T1){ref-type="table"}). Moreover, the recent surgical histories and primary infection sites showed no significant difference with respect to the genotype distribution (Table [1](#T1){ref-type="table"}).
Disease progression and mortality
---------------------------------
Disease progression was monitored by SOFA score changes during the patient ICU stays. The scores and the need for organ support were recorded on days 1, 2, 3, 5, 7, 14, 21, and 28. Although no differences in disease scores were observed at sepsis onset, the TLR4 rs11536889 GG patients experienced significantly higher SOFA scores over time (p = 0.0005) compared with the C allele carriers (Table [2](#T2){ref-type="table"}). The three organ-specific SOFA scores were significantly different between the two groups; the GG patients presented higher SOFA-renal scores (p = 0.0005), SOFA-coagulation scores (p = 0.0245), and SOFA-hepatic scores (p \< 0.0001; Table [2](#T2){ref-type="table"}, Figure [1](#F1){ref-type="fig"}). An overall linear model was fitted to the values, which included the various time points. This model also revealed a significant genotype effect; the GG patients presented higher SOFA scores than did the GC/CC subjects (p = 0.015; Figure [2](#F2){ref-type="fig"}). The mean SOFA scores of the two CC patients were 11.6 ± 3.5 (mean ± SD). Among all SOFA scores, the minimum and maximum were 0 and 23, respectively.
######
**Disease progression with regard to the*TLR4*rs11536889 genotypes**
**All n = 210** **GG n = 146** **GC/CC n = 64** ***p*value**
----------------------------------- ----------------- ---------------- ------------------ --------------
SOFA 7.6 ± 4.5 7.9 ± 4.5 6.8 ± 4.2 0.0005
SOFA-Respiratory score 2.1 ± 1.1 2.1 ± 1.1 2.0 ± 1.1 0.1950
SOFA-Cardiovascular score 1.6 ± 1.5 1.7 ± 1.5 1.4 ± 1.4 0.1296
SOFA-Central nervous system score 2.0 ± 1.5 2.1 ± 1.5 1.9 ± 1.4 0.0802
SOFA-Renal score 0.9 ± 1.4 1.0 ± 1.4 0.7 ± 1.2 0.0005
SOFA-Coagulation score 0.3 ± 0.8 0.4 ± 0.8 0.3 ± 0.7 0.0245
SOFA-Hepatic score 0.3 ± 0.7 0.4 ± 0.8 0.2 ± 0.5 \<0.0001
Mortality analysis, %:
Death at day 28 18 18 17 1.0000
Death at day 90 30 30 28 0.8698
Length of stay in ICU (days) 18 ± 16 18 ± 17 20 ± 14 0.0720
Organ support^\*^, %
Mechanical ventilation 75 74 0.6286
Use of vasopressor 38 37 0.8426
Renal replacement therapy 14 9 0.4486
The data are presented as the mean ± SD or percentages.
^\*^Based on the total number of observations in follow-up days.
{#F1}
{#F2}
The 28-day and 90-day mortality analyses yielded no significant result between the groups (p = 1.0000 and p = 0.8698, respectively; Table [2](#T2){ref-type="table"} (Additional file [1](#S1){ref-type="supplementary-material"}: Figure S1)). Both at the beginning and over the observational period, there was no significant difference between the GG patients and C allele carriers regarding organ support requirement (mechanical ventilation, vasopressor use, and renal replacement therapy; Tables [1](#T1){ref-type="table"} and [2](#T2){ref-type="table"}). The mean ICU stay duration of the GG survivors did not differ significantly from that of the GC/CC survivors (Table [2](#T2){ref-type="table"}). Additionally, there was a significantly higher gram-negative infection incidence rate among the C allele carriers (81%) compared with that in the GG patients (62%; p = 0.0062) (Table [3](#T3){ref-type="table"}).
######
Infection types over the observational period
**GG n = 146** **GC/CC n = 64** ***p*value**
---------------- ---------------- ------------------ --------------
Infection type
Gram-negative 62% 81% 0.0062
Gram-positive 83% 83% 1.0000
Fungal 52% 58% 0.4565
Viral 18% 17% 1.0000
Parasitic 0% 0%
Other 1% 3%
Discussion
==========
The present study addressed the question of whether the putative regulatory *TLR4* rs11536889 genotypes are related to organ failure in critically ill patients with sepsis.
The primary endpoint, organ failure, was quantified using SOFA scores as a specific clinical marker in patients with sepsis and was significantly higher in *TLR4* rs11536889 GG patients compared with those of *TLR4* rs11536889 GC and CC patients (Table [2](#T2){ref-type="table"}).
The *TLR4* rs11536889 genotype distribution among the septic patients was similar to database entries regarding healthy Caucasians and also followed the Hardy-Weinberg equilibrium. The *TLR4* rs11536889 genotypes, however, were not associated with any of the recorded baseline characteristics. As assessed by scoring the sepsis type (sepsis and severe sepsis vs. septic shock) and SOFA and APACHE II scores, we found that the *TLR4* rs11536889 genotypes were also not related to the septic disease severity at sepsis onset (Table [1](#T1){ref-type="table"}). We believe that the SOFA scores at sepsis onset did not differ between the GG versus GC and CC genotypes because of the phenotypic heterogeneity of the sepsis syndrome. This heterogeneity is influenced by many factors, including the pathogenic organism responsible for the infection, its location, and the amount of time passed since the onset of infection, as well as other individual parameters. To detect genotypic differences, a longitudinal observation involving SOFA scores quantified over the study period is much more promising (Table [2](#T2){ref-type="table"}).
As shown by Sato et al., monocytes from *TLR4* rs11536889 CC subjects expressed significantly higher levels of TLR4 compared with those from TLR4 rs11536889 GG and GC subjects. When PBMCs were stimulated with LPS, a TLR4 ligand, the cells from the *TLR4* rs11536889 CC and GC subjects secreted significantly higher levels of the proinflammatory cytokine IL-8 compared to cells from the GG subjects \[[@B24]\]. Accordingly, these previous investigations support the assumption that *TLR4* rs11536889 GG sepsis patients present severe organ dysfunction (as measured using SOFA scores) because of attenuated TLR4 proinflammatory signaling in response to LPS compared to C allele carriers.
These significant results, with respect to organ dysfunction, reveal severe morbidity among *TLR4* rs11536889 GG patients (according to SOFA scores) and together with the fact that GG patients are assumed to present attenuated TLR4 expression \[[@B24]\], offer an explanation why synthetic TLR4 antagonists have failed to produce a clinical benefit in patients with severe sepsis \[[@B14],[@B15]\]. These agents may alter the inflammatory response via TLR4 to pathogens, thereby contributing to organ dysfunction in these patients.
The SOFA-renal score was higher in *TLR4* rs11536889 GG patients, indicating severe renal dysfunction in this group. Because *TLR4* rs11536889 GG patients may exhibit decreased *TLR4* expression, our results are consistent with former observations indicating that decreased TLR4 expression in chronic kidney disease patients was associated with attenuated proinflammatory cytokine synthesis during infection \[[@B34]\]. The observed severe hepatic dysfunction measured using the SOFA-hepatic score among *TLR4* rs11536889 GG subjects, indicating severe hyperbilirubinemia in this group, is in agreement with recent findings reported by Deng et al. \[[@B16]\] that TLR4 signaling is essential for LPS clearance by hepatocytes during sepsis. The absence of an association between the rs11536889 genotypes and the SOFA respiratory score may be attributed to the fact that the SOFA respiratory score is somewhat weak because it only refers to the oxygenation index. This score depends on several factors, such as ventilator settings during mechanical ventilation and different ventilator settings that result in different oxygenation indices, which lead to score variation. We believe that there was no significant difference in the SOFA-CNS score between the genotypes mainly because sepsis patients are treated with sedating medication, which impacts the CNS and thus affects the SOFA-CNS score. The SOFA-Cardiovascular score most likely did not differ between GG patients and C allele carriers because this score is only based on the catecholamine therapy needed, which depends on volumetric status and cardiac function.
Analysis of the 28-day and 90-day mortality revealed no significant differences among the *TLR4* rs11536889 genotypes. The severe organ failure observed in G homozygous patients may not have contributed to higher mortality rates because the patients received sufficient intensive care treatment, which allowed their organ failures to be managed appropriately. The patients were treated according to current guidelines for the treatment of sepsis (Surviving Sepsis Campaign) \[[@B35]\].
Additionally, there was a significantly higher gram-negative infection rate among the C allele carriers (81%) compared with the rate observed in the GG patients (62%; p = 0.0062) (Table [3](#T3){ref-type="table"}). This result is in agreement with previous observations linking this polymorphism with an increased susceptibility to infection \[[@B24],[@B26],[@B27]\]. This observation of higher susceptibility of C allele carriers to gram-negative infections should be thoroughly examined in future studies to detect any causality between the SNP and susceptibility to gram-negative infections.
A possible limitation of this study is the possibility that the studied *TLR4* rs11536889 SNP associated with organ failure in patients with sepsis is in linkage disequilibrium with SNPs in another nearby gene and that these latter genes are responsible for the observed phenotypic effects.
To the best of our knowledge, this is the first investigation evaluating this putative regulatory polymorphism in this key innate immune receptor in adult Caucasian sepsis patients, showing a significant association between the *TLR4* rs11536889 GG genotype and the severity of organ failure (renal, coagulation and hepatic). According to these results, it would be worthwhile to further assess the *TLR4* rs11536889 polymorphism for its relevance to sepsis in larger and independent cohorts.
Conclusions
===========
This study assesses the validity of the assumption that a well-known regulatory *TLR4* polymorphism influences the outcome of sepsis among adults. An analysis of organ-specific SOFA sub-scores revealed significant differences in three organ systems: renal, coagulation and hepatic. These results offer the first evidence that *TLR4* rs11536889 is a useful marker of organ failure in patients with sepsis. This polymorphism should be assessed for its organ failure relevance in sepsis in larger, independent cohorts.
Abbreviations
=============
APACHE: Acute physiology and chronic health evaluation; CNS: Central nervous system; COPD: Chronic obstructive pulmonary disease; ICU: Intensive care unit; LPS: Lipopolysaccharide; PBMCs: Peripheral blood monocytes; SOFA: Sepsis-related organ failure assessment; TLR4: Toll-like receptor 4.
Competing interests
===================
The authors declare no competing interests.
Authors' contributions
======================
All authors have contributed to the study design, the acquisition of data (clinical and experimental), or the analysis and interpretation of data. Specifically, LVG and MS primarily performed sample and clinical data collection and TLR4 laboratory analyses. MS, AP, IB, DR, MG and MB participated in study design and supervised patient enrollment and clinical data monitoring. TB contributed to the study design and conception, performed the bioinformatics and performed and approved statistical analyses. AM and JH designed the study, supervised sample and data collection, performed analyses and drafted the manuscript. All authors were involved either in manuscript drafting or its revision. All authors have approved the final version of the manuscript.
Supplementary Material
======================
###### Additional file 1
**Kaplan-Meier survival analysis.** The Kaplan-Meier curve shows the survival curves censored at day 90 for the TLR4 rs11536889 GG and GC/CC genotypes. A mortality risk among the patients under study did not differ between the two groups.
######
Click here for file
Acknowledgments
===============
The authors thank the staff of the ICUs of the Department of Anesthesiology and the Department of General and Visceral Surgery, all of whom were involved in patient care and control. The authors also thank Benjamin Liese, Simon Wilmers, Yvonne Klee, Chang Ho Hong and Sebastian Gerber for their continuous and devoted help with performing analyses and collecting data.
| {
"pile_set_name": "PubMed Central"
} |
INTRODUCTION {#sec1-1}
============
Worm infection, particularly helminthiasis is one of the most common chronic infections of humans. Worldwide more than 200 million people harbor these infections.\[[@ref1]\] It is major global health problem mainly in tropical countries. These infections are common risk factor to community health in the developing countries and responsible for several conditions such as pneumonia, anemia, eosinophilia, and malnutrition.\[[@ref2]\] Expelling or killing of helminths is the two primary mechanisms by which drugs produces anthelmintic effects.\[[@ref3]\] Helminths mainly reside in the gastrointestinal tract, but some are known to invade tissues. The harmful effects of infection manifest as food deprivation, blood loss, local organ injury, intestinal or lymphatic obstruction. Sometimes, the same effects are also produced by worm secreted toxins. Although helminthiasis is not fatal, it is a principal etiological factor responsible for morbidity that in turn influences personal and social health and productivity.\[[@ref2]\] The main aim of anthelmintic treatment is to control morbidity and eradicate infection burden, interrupt transmission and complete eradication of the parasite reservoir.\[[@ref4]\] WHO has recommended only four drugs: Albendazole, mebendazole, levamisole, and pyrantel pamoate for soil-transmitted helminthiasis and these have been in use for several decades.\[[@ref5]\] The main challenges in animal and human health are a wide range of parasitic diseases and the emergence of resistance to anthelmintic drugs. Resistance to anthelmintic drugs is a persistent problem worldwide and needs urgent solution.\[[@ref6]\]
Drugs of natural origin have gained attention as a potential source of new therapeutic agents. Most of the clinically active drugs are either natural products or pharmacophore of the natural substance. It indicates the importance of drugs having natural sources in drug discovery process.\[[@ref7]\] The use of plant as a source of medicine is as old as mankind. Traditional Indian medicine system is based on plants.\[[@ref8]\]
*Cordia dichotoma* is an edible plant belonging to family *Boraginaceae*. The plant is widely distributed across India with references in traditional literature of Ayurveda and materia medica. In this, the *Cordia* species has been recommended in the treatment of several diseases. The plant is richly supplied with diverse phytochemicals including flavonoids, glycosides, alkaloids, terpenes and sterols, carbohydrates and proteins, and minerals like chromium.\[[@ref9][@ref10]\] Various parts of *C. dichotoma* such as fruit, seed, bark, and leaves have been screened for antidiabetic, antiulcer, anthelmintic, immunomodulatory, hepatoprotective, and antilarvicidal activity. The fruit and pulp have been reported as possessing significant anthelmintic activity.\[[@ref9][@ref11]\] Although roots of the plant have already screened for different phytochemicals, the anthelmintic activity is yet to be explored.
MATERIALS AND METHODS {#sec1-2}
=====================
Collection and identification of plant material {#sec2-1}
-----------------------------------------------
The roots of *C. dichotoma* were collected from Kendra (Budruk) area of Hingoli district, Maharashtra, India in the month of December and it was authenticated by Dr. B. D. Gachande of Department of Botany, N. E. S. Science College, Nanded, Maharashtra, India.
Preparation of extract {#sec2-2}
----------------------
The shade-dried root material was powdered with the help of a grinder (sieving was manually done), and coarse powder material was obtained. The coarse powder material (500 g) was extracted with methanol using soxhlet apparatus. The extract was filtered, concentrated by evaporating the solvent in a rotary evaporator and kept in the refrigerator.
Preliminary phytochemical screening {#sec2-3}
-----------------------------------
The freshly prepared root extract of *C. dichotoma* was subjected to preliminary phytochemical screening for the identification of major chemical constituents according to the standard procedures.\[[@ref12][@ref13]\]
Collection and authentication of worm {#sec2-4}
-------------------------------------
The Indian earthworms *Pheretima posthuma* (*Annelida*) employed in the present study were collected from moist and muddy soil of Vishnupuri village, Nanded district, Maharashtra, India. Worms were washed with normal saline to remove all the fecal matter and authenticated by Dr. P. B. Deshmukh, Department of Zoology, N. E. S. Science College, Nanded, Maharashtra, India.
Anthelmintic activity {#sec2-5}
---------------------
The anthelmintic potential of methanolic root extract was screened using Indian earthworms *P. posthuma* because they anatomically and physiologically resemble the human intestinal roundworm parasite.\[[@ref3][@ref14]\] The different extract concentrations (10, 25, 50, and 75 mg/ml) were obtained by dissolving 100 mg, 250 mg, 500 mg, and 750 mg/ml of crude extract in 1 ml of tween 80 and adjusted volume up to 10 ml with normal saline solution. Albendazole was used as a standard drug for this study.\[[@ref15][@ref16]\] Nine groups each containing six earthworms of approximately of equal size were used. Four groups of earthworms were tested with methanolic extract at concentrations 10 mg/ml, 25 mg/ml, 50 mg/ml, and 75 mg/ml and other four groups were treated with standard anthelmintic drug, albendazole at same concentrations. A group of earthworms was exposed normal saline and tween 80 and considered as control.\[[@ref16]\] Each group was observed for worm motility, and the time required to produce paralysis, or complete inactivity and mortality was recorded. The worms were considered paralyzed when no movement were observed except after vigorous shaking. The death time was recorded after ascertaining that worms neither moved when given external stimuli nor dipped in warm (50°C) water. Fading of body color was considered as a sign of complete mortality. Inhibition of worm motility was the factor considered for anthelmintic activity.\[[@ref2][@ref3]\]
Prediction of activity spectra for substances computer program {#sec2-6}
--------------------------------------------------------------
The software-based program, prediction of activity spectra for substances (PASS) was used to obtain biological activity spectra including anthelmintic activity of phytoconstituents. Software estimates predicted activity spectrum of a compound as probable activity (P~a~) and probable inactivity (P~i~). The prediction of activity is based on structure-activity relationship analysis of the training set containing more than 205,000 compounds exhibiting more than 3750 kinds of biological activities. The values of P~a~ and P~i~ vary between 0.000 and 1.000. Only activities with P~a~ \> P~i~ are considered as possible for a particular compound. If P~a~ \> 0.7, the probability of experimental pharmacological action is high and if 0.5 \< P~a~ \< 0.7, probability of experimental pharmacological action is less. If the value of P~a~ \< 0.5, the chance of finding the activity experimentally is less, but it may indicate a chance of finding a new compound.\[[@ref17][@ref18][@ref19]\]
Statistical analysis {#sec2-7}
--------------------
All data were expressed as the mean ± standard error of mean data were subjected to one-way ANOVA followed by Tukey test. The statistical analysis performed with Graphpad Instat (Version 3, USA) software. *P* \< 0.05 was considered statistically significant.
RESULT {#sec1-3}
======
Phytochemical screening of the extract {#sec2-8}
--------------------------------------
Preliminary phytochemical screening of *C. dichotoma* root extract revealed presence of several phytoconstituents listed in [Table 1](#T1){ref-type="table"}.
######
Phytochemical investigation of *cordia dichotoma* root extract
Anthelmintic activity {#sec2-9}
---------------------
The extract exhibited significant dose-dependent anthelmintic activity at all concentrations as compared to standard, albendazole \[[Table 2](#T2){ref-type="table"}\]. At higher concentration, loss of motility and mortality was more pronounced against *P. posthuma*. The time required for causing paralysis was (3.00 ± 0.11) min and death (9.00 ± 0.46) min at 75 mg/ml by the extract which was almost equal to the results obtained with albendazole. Among all the concentrations tested, the significant results for time to produce paralysis were observed at 10, 25 and 50 mg/ml whereas significant mortality was shown at all concentrations.
######
Anthelmintic activity of control, methanolic extract and albendazole
Prediction of activity spectra for substances computer program {#sec2-10}
--------------------------------------------------------------
The phytoconstituents of *C. dichotoma* root were evaluated for their biological activity spectra, and results were used in a flexible manner. All the compounds showed greater P~a~ than P~i~ \[[Table 3](#T3){ref-type="table"}\]. The P~a~ of caffeic acid, lupeol, and hentricontanol was found to be highest as 0.674, 0.622, 0.600, respectively. The P~a~ of octacosanol and epigenine was more than 0.5 whereas taxifolin, rutin, hesperidin, α-amyrin, β-sitosterol glycoside, β-sitosterol, botulin, and chlorogenic acid showed less P~a~ \< 0.5.
######
PASS predictions of main phytoconstituents for anthelmintic activity
DISCUSSION {#sec1-4}
==========
Several bioactive phytoconstituents such as alkaloids, tannins, glycosides, saponins, flavonoids, and phenolics were found predominantly during preliminary phytochemical screening of the extract of *C. dichotoma* which have been associated with an anthelmintic potential.\[[@ref11][@ref19]\] The plant, *C. dichotoma* has been already found to be rich in various secondary metabolites.\[[@ref9][@ref20]\]
Infection of helminths is a major problem in human as well as in animals that in turn adversely affects the health and also causes drug resistance to other diseases.\[[@ref6][@ref21]\] To overcome these problems, there is a need for studies focusing on natural sources such as plants which give new biologically active agents having no or fewer side effects and more compatible with human physiology.\[[@ref22]\] Phytochemical screening of the extract of *C. dichotoma* revealed the presence of different phytochemicals that produces anthelmintic effects by acting on metabolic pathways or physiological functioning of the worms. Alkaloids have ability to produce paralysis by acting on central nervous system whereas polyphenols and tannins selectively bind to free proteins present in the gastrointestinal tract and eventually cause mortality. Anthelmintic potential of saponin is due to its membrane permeabilizing property.\[[@ref23][@ref24]\] The anthelmintic efficacy of the methanolic extract of *C. dichotoma* may be due to a single compound or combined effect of these phytochemicals.
Previous studies of *C. dichotoma* reported the presence of caffeic acid, β-sitosterol, taxifolin, rutin, octacosanol, lupeol, hesperidin, α-amyrin, and hentricontanol.\[[@ref10]\] PASS computer program was used to screen biological activity of these compounds as anthelmintics. Caffeic acid, lupeol, and hentricontanol showed highest P~a~ value indicating a high probability of these compounds acting as anthelmintic agents. In humans, caffeic acid produces reactive oxygen species\[[@ref25][@ref26]\] whereas epigenine shows healing and platelet aggregation effect.\[[@ref27]\] Lupeol is a multi-target agent with immense anti-inflammatory potential and also has an inhibitory effect on microtubule formation while octacosanol rapidly blocks the intracellular gap junction channels.\[[@ref28]\]
In the present study, methanolic root extract of *C. dichotoma* has demonstrated potent anthelmintic effect on *P. posthuma*. *C. dichotoma* may offer an alternative drug source to control worm infection. However, more detailed studies are needed to confirm the results of anthelmintic activity and PASS predictions by various *in vitro* and *in vivo* methods. Further, there is a need to identify and evaluate the active components, their mechanisms of action, and toxicity profile so as to explore *C. dichotoma* as an anthelmintic agent.
We are very thankful to Director Prof. S. G. Gattani and Head of Department Prof. S. C. Dhawale School of Pharmacy, Swami Ramanand Teerth Marathwada University, Nanded, Maharashtra, India for providing laboratory facilities for this research work.
**Source of Support:** The present work has been supported by School of Pharmacy, Swami Ramanand Teerth Marathwada University, Nanded, Maharashtra, India.
**Conflict of Interest:** None declared.
| {
"pile_set_name": "PubMed Central"
} |
1. Introduction {#sec1-ijms-18-01711}
===============
The burden of major depressive disorders (MDD) and schizophrenia (SCZ) is on the rise globally, reflecting ongoing population aging and demographic growth. In 2015, ≈300 million people were affected by MDD; this corresponds to 4.4% of the global population and an increase by 18% across the last 10 years \[[@B1-ijms-18-01711]\]. At the same time, ≈21 million were affected by SCZ \[[@B2-ijms-18-01711]\]. For comparison, ≈35 million people were affected by cancer.
MDD is the leading cause of life with disability and is a major contributor to the overall burden of disease \[[@B3-ijms-18-01711],[@B4-ijms-18-01711]\]. More women are affected by depression than men \[[@B1-ijms-18-01711]\]. In contrast to usual mood fluctuations in daily life, MDD is a serious health condition characterized by a depressed mood, loss of interest and enjoyment, and reduced energy, leading together to poor function at work, at school, and in the family \[[@B5-ijms-18-01711],[@B6-ijms-18-01711]\]. Additional symptoms are feelings of guilt or low self-worth, anxiety, disturbed appetite and sleep, poor concentration, and even unexplained physical symptoms. Depressive episodes can last over an extended period of time or manifest remission and recurrent relapses \[[@B6-ijms-18-01711]\]. At its worst, depressed people commit suicide, with close to 800,000 victims every year. Therefore, MDD represents the second leading cause of death in 15 to 19 year olds \[[@B1-ijms-18-01711]\].
SCZ is not as common as MDD but is more common among males (12 million) than females (nine million) and also develops earlier among men \[[@B2-ijms-18-01711]\]. Distortions in perception, thinking, language, emotions, sense of self, and behavior are characteristic for SCZ and give rise to the ample delusions that can associate with acoustical (hearing voices), optical, and sensory hallucinations \[[@B6-ijms-18-01711]\]. Behavioral abnormalities include self-neglect, wandering aimlessly, strange appearance, incoherent speech, mumbling, or laughing to oneself. SCZ causes great disability that may interfere with educational and occupational performance. Overall, individuals with SCZ are 2 to 2.5 times more likely to die early than the general population due to cardiovascular, metabolic, and infectious diseases \[[@B2-ijms-18-01711]\].
Research on MDD/SCZ has not identified a single causal factor. Instead, psychiatric diseases are hypothesized to result from complex interactions of social, psychological, and biological factors during neurodevelopment and beyond. The neurodevelopmental hypothesis of psychiatric diseases dates back to the latter part of the nineteenth century, when authorities such as Clouston \[[@B7-ijms-18-01711]\] posited that at least some insanities were 'developmental' in origin. With the spread of Kraepelin's concept of dementia praecox as a degenerative disease (see pp. 426--441 \[[@B8-ijms-18-01711]\]), this view passed largely into oblivion until the 1980s when several research groups again began to speculate that SCZ might have a significant neurodevelopmental component \[[@B9-ijms-18-01711],[@B10-ijms-18-01711],[@B11-ijms-18-01711]\]. Likewise, more than 100 years have passed since Sigmund Freud \[[@B12-ijms-18-01711]\] postulated the important role of early traumatic experiences on the development of depression and major disorders. Still, it was not before the late 1980s that the critical role of early life events and parenting for the development of MDD was re-addressed at the biological scale \[[@B13-ijms-18-01711],[@B14-ijms-18-01711],[@B15-ijms-18-01711],[@B16-ijms-18-01711]\].
With respect to the neurodevelopmental origin of SCZ/MDD, epigenetic mechanisms (see below) can serve as an interface between the genetic blueprint and the environment and were originally recognized to mediate the interactions between intrinsic and extrinsic clues during cellular and organismal development \[[@B17-ijms-18-01711]\]. More recently, epigenetic mechanisms have been hypothesized to mediate also between genes and social experiences in translational rodent studies as well as in men \[[@B18-ijms-18-01711],[@B19-ijms-18-01711]\]. Especially, experiences during early life can influence lastingly the expression of genes controlling neuronal cell numbers, neuronal activity, and connectivity through epigenetic mechanisms; all of these processes are well-known to regulate behavior, cognition, and mood, among other higher brain functions. While studies on gene-environment interactions have centered mostly on single genes known to act in pathways that are thought to be involved in mental diseases \[[@B20-ijms-18-01711],[@B21-ijms-18-01711]\], recent genome-wide approaches have provided new insights into epigenomic changes in MDD/SCZ and their potential interaction with genetic variation.
Here, we refer to previous discoveries on the genetic architecture of MDD/SCZ and the emerging role of epigenomic studies ([Section 2](#sec2-ijms-18-01711){ref-type="sec"}). We also evaluate the eminent role of early life for future mental health ([Section 3](#sec3-ijms-18-01711){ref-type="sec"}) and recent insights into the dynamic role of DNA methylation during early brain development ([Section 4](#sec4-ijms-18-01711){ref-type="sec"}). In this context, we explore the hypothesis that early life experience, particularly adversity, can lead to enduring epigenetic changes that increase the risk of later MDD/SCZ. With respect to MDD, findings on epigenetic changes in rodent models and postmortem human brains will be discussed ([Section 5](#sec5-ijms-18-01711){ref-type="sec"}). Further, we will consider the latest results on epigenetic changes in SCZ and how they intersect with genetic risk variants ([Section 6](#sec6-ijms-18-01711){ref-type="sec"}). Concluding, we will discuss future steps to be taken to address current limitations and to advance insight into the functional implications from these findings by using human pluripotent stem cell models.
2. The Genetic Architecture of MDD and SCZ {#sec2-ijms-18-01711}
==========================================
The advent of high-throughput genotyping technologies has pioneered insight into common genetic variation contributing to psychiatric diseases. More than 85 million single nucleotide polymorphisms (SNPs) have been identified in the human population, accounting for 95% of all known sequence variants \[[@B22-ijms-18-01711]\]. For practical reasons, genome-wide association studies (GWAS) include only a few tag SNPs that represent all SNPs in the same linkage disequilibrium (LD) block. Multiple SNP associations within the same LD block are considered to detect a single causal variant, the physical boundaries of which are statistically inferred for the identified SNP associations. Since tag SNPs capture all the other SNPs localizing to the risk-associated haplotype block, they are unlikely to encode on their own the causal genetic variant that underlies the disease association.
Family history is a strong and well-replicated risk factor that associates with different heritability estimates for MDD (0.37) and SCZ (0.81) \[[@B23-ijms-18-01711],[@B24-ijms-18-01711],[@B25-ijms-18-01711]\]. In the case of MDD, however, the combination of high prevalence and moderate heritability poses a major challenge to the genetic analysis of MDD. Consistent with this concern, a seminal study four years ago on more than 70,000 MDD cases and controls from combined data sets did not obtain any evidence for any variant significant at genome-wide association thresholds \[[@B26-ijms-18-01711]\].
Likewise, people manifesting the same MDD or SCZ symptoms may not share the same etiology, and no robust laboratory tests exist to distinguish between subtypes within either condition. Until now, neuroscience and genetic studies into psychiatric disorders rely generally on disease definitions that are based on the influential "Diagnostic" and Statistical Manual of Mental Disorders; (DSM) that was designed as a purely diagnostic tool \[[@B27-ijms-18-01711]\]. Although DSM considers different disorders as distinct entities, the boundaries between disorders are often not as strict as the DSM suggests. To develop an alternative framework for research into psychiatric disorders, the US National Institute of Mental Health (NIMH) introduced in 2009 its Research Domain Criteria (RDoC) project to develop a research classification system for mental disorders based upon dimensions of neurobiology and observable behavior \[[@B28-ijms-18-01711]\]. RDoC supports research to explicate fundamental biobehavioral dimensions that cut across current heterogeneous disorder categories, with the goal to transform the approach to the nosology of mental disorders. While progress in the diagnostic validity of major psychiatric disorders is thus still looked for, a number of more recent studies have sought to advance insight into the genetics of psychiatric disorders by increasing case control numbers or by narrowing the range of clinical phenotypes.
First, the CONVERGE consortium \[[@B29-ijms-18-01711]\] minimized genetic heterogeneity by including only women of Han Chinese ethnicity, among whom depression is thought to be under-diagnosed and among whom those who were diagnosed with two or more episodes of MDD are likely to represent more severe forms of the disease. Two regions in which genetic variants associate with MDD were identified; one mapping near *SIRT1* (encoding Sirtuin 1, an NAD(+)-dependent histone deacetylase) and the other mapping in an intron of *LHPP* (encoding a protein phosphatase that cleaves phospholysine and/or phosphohistidine bonds). These variants were corroborated with a second method and in an independent sample, suggesting that the proximity of one of the variants to *SIRT1* could contribute to deregulated mitochondrial energy metabolism \[[@B30-ijms-18-01711]\]. Since these two risk allele variants are extremely rare in Europe \[[@B26-ijms-18-01711]\], they may be actually restricted to severe cases of MDD from China.
Second, Hyde et al. \[[@B31-ijms-18-01711]\] dramatically increased the cohort size (including 75,607 cases with self-reported or clinical diagnosis or treatment for depression and 231,747 controls) and performed meta-analysis of these data with published MDD genome-wide association study results. This approach identified 17 MDD-associated variants in 15 regions of the genome in people of European descent. While the question arises as to how well self-reports correspond to the kind of depression physicians see in the clinic \[[@B32-ijms-18-01711]\], evidence for shared polygenic risk between their MDD cases and published SCZ cases \[[@B33-ijms-18-01711]\] provides initial support for this phenotyping approach.
Third, Power et al. \[[@B34-ijms-18-01711]\] reanalyzed data from a previously published meta-analysis conducted by the Psychiatric Genomics Consortium (PGC) \[[@B26-ijms-18-01711]\] by acknowledging that earlier onsets of MDD show greater familial loading \[[@B35-ijms-18-01711],[@B36-ijms-18-01711]\]. Discovery case-control studies (8920 cases and 9521 controls) were stratified using increasing/decreasing age-at-onset cutoffs and led to the identification of one replicated genome-wide significant locus with adult-onset MDD that had not reached significance in the original unstratified PGC meta-analysis. Interestingly, polygenic score analysis additionally showed that earlier-onset cases of MDD share a greater genetic overlap with SCZ and bipolar disorder (BIP) than adult-onset cases.
Contrary to MDD, higher heritability in SCZ enabled the identification of common variants encoding subtle effects as well as rare but highly penetrant copy number variations and possibly exome variants \[[@B25-ijms-18-01711]\]. Ripke et al. \[[@B37-ijms-18-01711]\] recently estimated that 6333 to 10,200 independent and mostly common SNPs may underlie the risk for SCZ, with each conferring a small increase in risk. Incrementally, these SNPs are thought to account for around 50% of the total variance in liability to SCZ and indicate that common genetic variation is a major factor in SCZ heritability. Since the first GWAS for SCZ was published in 2009 \[[@B38-ijms-18-01711],[@B39-ijms-18-01711]\], the size of the studies and the number of loci associated with the condition has expanded, whereby the latest study \[[@B40-ijms-18-01711]\] included more than 150,000 people and identified 108 genomic regions containing genetic risk factors. Since the identified risk variants are common, they will contribute to most, if not all, cases. The authors estimate that the 108 loci (hereafter referred to as Psychiatric Genomics Consortium (PGC) risk variants) collectively implicate a total of 305 genes, including "plausible" candidates such as *DRD2* (a well-known dopamine receptor gene), a locus on chromosome 6 that harbors the major histocompatibility complex, and several genes encoding calcium channel subunits and proteins involved in synaptic plasticity. Importantly though, further studies are needed to map the genetic variation-to-genes-to-function to corroborate the role of these and others candidates in SCZ.
Collectively, genetics is a major cause of MDD/SCZ. Despite considerable progress on the genetic architecture of SCZ, causal variants from GWAS still remain to be verified. Presently, a substantial fraction of the risk of MDD/SCZ still remains unexplained and points to the role of environmental factors.
3. Early Life Events Preset Adult Behavior {#sec3-ijms-18-01711}
==========================================
From conception onwards, the physical and social environment acts on the genetic blueprint to adjust development and lifelong programs of somatic and mental functions. These interactions are particularly important during early life and can elicit long-lasting "anticipatory" changes in phenotype, referred to as "programming" \[[@B41-ijms-18-01711],[@B42-ijms-18-01711]\]. Epigenetic mechanisms can mediate the effects of early life through sustained changes in (neuronal) gene expression that prepare the organism to effectively cope with changing environments. However, early adaptations may also result in inefficient responses due to inherent (genetic) constrains or misguided adjustments that can increase the risk of future disease \[[@B19-ijms-18-01711],[@B43-ijms-18-01711]\].
The embryonic and early postnatal brain is highly plastic due to extensive cell proliferation that passes over to progressive differentiation \[[@B44-ijms-18-01711]\]. For example, the embryonic human brain produces some 250,000 new cells per minute \[[@B45-ijms-18-01711]\] and forms about 40,000 synapses per minute in the last trimester \[[@B46-ijms-18-01711]\]. Early-born neurons dynamically organize themselves into functional networks and undergo extensive pruning and reorganization from early postnatal life through early adulthood \[[@B16-ijms-18-01711]\]. Various environmental cues and life events can act on these cellular processes and trigger lasting (epigenetic) changes in gene regulatory networks.
Adverse events, especially in early life, are the greatest risk factor for the development of MDD \[[@B47-ijms-18-01711]\]. Early adversity can comprise interpersonal loss (e.g., parental death), parental maladjustment (e.g., mental illness or substance abuse), low socioeconomic status in childhood, and maltreatment (physical or sexual abuse and neglect), among other threats \[[@B48-ijms-18-01711]\]. Maltreatment is the leading cause of early life adversity in Westernized societies (prevalence 0.9%), with 80% of the cases that associate with the highest rates of increased risk for MDD corresponding to neglect \[[@B49-ijms-18-01711]\]. Recurrent episodes of early adversity increase the risk for depression four-fold, whereby the severity of exposure correlates with the risk of life-long recurrent depression \[[@B50-ijms-18-01711]\] and completed suicide \[[@B49-ijms-18-01711]\]. This disease process suggests that the initial event leads to sustained changes in gene regulatory and/or neuronal networks that can be reactivated with new exposure and thereby facilitate disease development. At the same time, early life adversity is also an important risk factor for the development of SCZ \[[@B24-ijms-18-01711],[@B51-ijms-18-01711],[@B52-ijms-18-01711],[@B53-ijms-18-01711]\]. A growing body of evidence based on epidemiological and translational studies further shows that prenatal adversity, in the form of stress and trauma, mood and anxiety disorders, or infections and severe physical illness in the mother, is a shared risk factor for MDD and SCZ \[[@B54-ijms-18-01711],[@B55-ijms-18-01711],[@B56-ijms-18-01711]\].
A common response to early life adversity is the activation of the hypothalamic-pituitary-adrenal axis (HPA axis), a major mediator of the stress response \[[@B57-ijms-18-01711]\] and the subsequent secretion of glucocorticoids (GC) that normally serve to restore physiological and behavioral homeostasis \[[@B58-ijms-18-01711]\]. The action of GCs is mediated by binding to high- and low-affinity glucocorticoid receptors (MR and GR, respectively) that operate as nuclear transcription factors and through membrane bound mechanisms \[[@B59-ijms-18-01711]\]. However, the adaptation-promoting action of GCs can turn into the opposite in case the type, strength, or duration of the stressor overwhelms the regulatory mechanisms that act to restrain GC secretion \[[@B58-ijms-18-01711],[@B59-ijms-18-01711]\].
Taken together, early-life adversity is a major risk factor for MDD/SCZ and frequently leads to sustained deregulation of the HPA axis. Such deregulation, in turn, associates with structural brain and epigenetic changes within individual cells that can confer an increased risk of psychiatric disease.
4. From Epigenetics to Epigenomics {#sec4-ijms-18-01711}
==================================
Major epigenetic mechanisms \[[@B60-ijms-18-01711]\] comprise DNA methylation, posttranslational modifications of core histones, nucleosome positioning, and non-coding RNA (ncRNA). All of these mechanisms are thought to act jointly in "the structural adaptation of chromosomes so as to register, signal, or perpetuate activity states" (see p. 398 in \[[@B61-ijms-18-01711]\]). Historically, DNA methylation has been the most studied and is of particular relevance to this review.
Canonical DNA methylation (mCG) refers to the transfer of a methyl group to a cytosine-guanine dinucleotide (CG) ([Figure 1](#ijms-18-01711-f001){ref-type="fig"}A) and is broadly distributed across the genome. In the human brain ≈80% of all CGs are methylated similarly to other tissues \[[@B62-ijms-18-01711]\]. mCG exists in the neural and glial cells of all brain tissues from prenatal life to old age \[[@B63-ijms-18-01711]\] and is thought to fulfill an important role in cell differentiation and cellular identity \[[@B17-ijms-18-01711]\]. The de novo DNA methyltransferases (DNMTs) DNMT3A and DNMT3B catalyze mCG, which is reestablished by DNMT1 following genome replication ([Figure 1](#ijms-18-01711-f001){ref-type="fig"}B). Both DNMT1 and DNMT3A are also critically involved in neuronal plasticity, learning, and memory through their joint role in DNA methylation and its effect on neuronal gene expression \[[@B64-ijms-18-01711]\].
Mammalian genomes are commonly depleted of CG residues, except for enrichment in so-called CG islands (CGIs) that occur in less than half of all human gene promotors and remain usually methylation-free. While only few promoter CGIs undergo DNA methylation \[[@B65-ijms-18-01711],[@B66-ijms-18-01711]\], DNA methylation is commonly found in CGI-free promoters modulating gene expression in undifferentiated and differentiated cells \[[@B67-ijms-18-01711]\]. Historically, DNA methylation is thought to confer lasting, or even irreversible, gene repression during development and beyond. More recent reports suggest, however, a broader role in enhancing transcription through the inhibition of spurious transcription initiation or the promotion of prolongation efficiency \[[@B65-ijms-18-01711]\]. In any case, the overall effects of DNA methylation depend critically on the genomic position, primary sequence, and pre-existing transcriptional activity.
Rapid progress in high-throughput sequencing has provided an unprecedented genome-wide view on DNA methylation by charting exact sites and sequence contexts but has also evidenced unique features of the human and mice brain methylome when compared to the respective peripheral tissues.
First, the brain methylome contains a high amount of 5-hydroxymethylcytosine (5hmCG), which gradually accumulates during mouse development ([Figure 1](#ijms-18-01711-f001){ref-type="fig"}C) \[[@B68-ijms-18-01711],[@B69-ijms-18-01711],[@B70-ijms-18-01711]\]. This modification is catalyzed by the family of ten-eleven translocation enzymes (TET) that oxidize mCG to 5hmCG and further derivatives ([Figure 1](#ijms-18-01711-f001){ref-type="fig"}B). The final product 5-carboxylcytosine serves as a substrate for DNA glycosylase-mediated-base excision and replacement by unmodified cytosine through the base excision and/or nucleotide excision repair machinery (BER/NER) \[[@B71-ijms-18-01711]\]. Recent findings indicate that, in mature neurons, 5hmCG occurs as a transient modification that is replaced in a temporally and spatially confined manner in small regulatory regions of the genome \[[@B72-ijms-18-01711],[@B73-ijms-18-01711],[@B74-ijms-18-01711],[@B75-ijms-18-01711]\].
Second, methylation in a CH context (mCH, where H corresponds to A, C, or T) is found in neuronal cells as well as in embryonic stem cells \[[@B63-ijms-18-01711],[@B76-ijms-18-01711],[@B77-ijms-18-01711],[@B78-ijms-18-01711]\] but rarely in peripheral differentiated tissues \[[@B79-ijms-18-01711],[@B80-ijms-18-01711],[@B81-ijms-18-01711]\]. This so-called non-canonical DNA methylation is barely detectable in the fetal and early-infant brain methylome but also rapidly accumulates postnatally across two years ([Figure 1](#ijms-18-01711-f001){ref-type="fig"}C). Thereafter, mCH gradually reaches the level of mCG by adulthood and accounts then for more than half of all neuronal methylcytosines. Since the absolute number of C residues exceeds by far the absolute number of CG dinucleotides, mCH still remains a relatively rare event in the neural methylome. Notwithstanding this reservation, mCH has been hypothesized to associate with the rapid rise in postnatal synaptogenesis and synaptic pruning that shapes neural network formation. However, a role as substrate for epigenomic changes in response to early life events has not been reported so far.
In view of an environmental causation of MDD/SCZ, dynamic changes in DNA methylation ([Figure 1](#ijms-18-01711-f001){ref-type="fig"}B,C) offer an intriguing interface for the interaction of societal risk factors with the genome. Experience driven neuronal activity connects to various transcriptional regulators that in turn recruit the epigenetic machinery. These factors can confer local \[[@B18-ijms-18-01711],[@B19-ijms-18-01711],[@B43-ijms-18-01711]\] and genome-wide (see [Section 5](#sec5-ijms-18-01711){ref-type="sec"} and [Section 6](#sec6-ijms-18-01711){ref-type="sec"}) epigenetic changes that last beyond the initial stimulus and cause changes in the gene expression underlying various brain functions.
Epigenomics refers to the study of epigenetic mechanisms at the genome-scale \[[@B82-ijms-18-01711]\] and enables important insights into the functional relationships of genes in health and disease through the identification of regulatory mechanisms that are sensitive to environmental and lifestyle factors. Hereby, complex phenotypes are analyzed for genetically induced epigenetic alterations and/or environmentally induced epigenetic alterations that in turn can be controlled by genetic effects. Since the epigenome is highly dynamic, the extent of interindividual phenotypic variation needs to be assessed by large-scale, systematic epigenome-wide association studies (EWAS). By approach, EWAS are equivalent to GWAS, with variation at a single CG site corresponding formally to a SNP \[[@B83-ijms-18-01711]\]. Measurements of CpG methylation average, however, thousands of DNA copies at the tissue level and represent therefore aggregate information on the effect sizes between cases and controls, as opposed to the categorical nature of SNP information.
5. Early Life Adversity-Dependent Epigenomic Responses as Risk Factor for MDD {#sec5-ijms-18-01711}
=============================================================================
The quality of maternal care critically influences brain function through lasting effects on stress regulation, emotion, learning, and memory \[[@B84-ijms-18-01711],[@B85-ijms-18-01711]\]. In rats, naturally occurring variations in perinatal maternal care associate with changes in offspring's behavior and hippocampal gene expression (\>900 genes) that persist into adulthood \[[@B86-ijms-18-01711],[@B87-ijms-18-01711]\]. These changes can be reversed by cross-fostering \[[@B86-ijms-18-01711]\], pharmacological, or dietary (methyl supplementation) treatments \[[@B88-ijms-18-01711]\] and indicate that maternal care leads to the epigenetic programming of gene expression. In support of this hypothesis, a pioneering study by Weaver et al. \[[@B89-ijms-18-01711]\] showed that differences in early maternal care triggered differential DNA methylation at the proximal glucocorticoid receptor gene (*NR3C1*) promoter in hippocampal cells, which was prevented by co-treatment with a histone deacetylase inhibitor.
Moving beyond single gene analysis, the researchers further analyzed gene expression, histone acetylation, and DNA methylation in a large region of chromosome 18 by customized tilling arrays \[[@B90-ijms-18-01711]\]. Differences in maternal care were associated with both increased and decreased peaks of histone acetylation and DNA methylation over one hundred kilobase pairs, covering promoters, exons, and genes. The response to maternal care appeared to be specific since not all genes were affected, whereas differences in epigenetic marks co-clustered over large distances, which was indicative of widespread epigenetic effects on multiple genes in the same genomic region. In general, increased transcription was associated with reduced DNA methylation at upstream regulatory sites, increased exonic histone acetylation, and DNA methylation. The co-clustering of epigenetic responses was highly developed at the protocadherin gene cluster (*PCDH*), which regulates synaptic development and neuronal function \[[@B91-ijms-18-01711]\]. Taken together, these findings indicate that large groups of functionally related genes or gene networks may be coordinately regulated by early life events.
Another line of evidence for the profound effects of stress on early brain development stems from animal studies, which showed that maternally administered synthetic GCs (e.g., dexamethasone) in late gestation can induce lasting changes in HPA-axis function and behavior in adult offspring of different species, including guinea pigs, mice, sheep, and non-human primates \[[@B92-ijms-18-01711],[@B93-ijms-18-01711]\]. Similarly, recent studies in humans have reported an increased risk of emotional and behavioral abnormalities in children exposed to elevated glucocorticoid concentrations in utero by either antenatal dexamethasone treatment (to advance lung maturation) or maternal stress \[[@B94-ijms-18-01711],[@B95-ijms-18-01711]\]. To obtain insight into GC effects on the brain methylome, Crudo et al. \[[@B96-ijms-18-01711]\] investigated guinea pigs, whose fetal pattern of brain development more closely resembles that of the human \[[@B97-ijms-18-01711]\]. Genomic DNA was purified from fetal hippocampi that were isolated immediately before the fetal plasma cortisol surge (gestational day 52, GD52) or in late gestation (GD65). In parallel, genomic DNA was isolated for 24 h or 14 days (GD52 and GD65, respectively) after the completion of a serial dexamethasone application to the mother (GD40, GD41, GD50, and GD51). All of these samples were analyzed for genome wide changes in DNA methylation using promoter tilling arrays containing ≈43,000 genes. Compared to GD52, extensive genome-wide promoter hypomethylation was detected on GD65 after the fetal cortisol surge or 24 h after the last dexamethasone application. However, 14 days after dexamethasone application, these differences did not persist and a different set of promoters became hypermethylated or hypomethylated when compared to the untreated fetus on GD65 \[[@B96-ijms-18-01711]\].
In general, changes in DNA methylation correlated negatively with genome-wide changes in transcription following the cortisol surge (1086 genes) or dexamethasone application (1126 genes) \[[@B98-ijms-18-01711]\]. Among the 173 genes shared between both conditions, 159 genes showed the same directional change. However, none of the dexamethasone regulated genes remained similarly different from the controls at GD65, indicating that dexamethasone triggered precocious changes in expression at GD52.
Beyond the methylome, GCs also affected GR DNA binding. When comparing GD65 to GD52, 1245 gene promoters exhibited differential GR DNA-binding, with 627 promoters showing an increase and 618 promoters a decrease. Twenty four hours after dexamethasone treatment only 94 gene promoters showed differential GR DNA binding (58 increases versus 46 decreases), as compared to 690 promoters after 14 days (279 increases versus 411 decreases). This indicates that development and GC treatment regulate GR DNA binding largely differently, whereby the dexamethasone-induced precocious maturation of GR DNA-binding is confined to a rather small set of gene promoters \[[@B98-ijms-18-01711]\].
Overall, these results show that the fetal cortisol surge drives the genome-wide reconfiguration of the hippocampal methylome, altered transcription, and GR DNA-binding during late hippocampal development. Immediate effects from GC application mostly anticipate changes from the fetal cortisol surge; however, in the long run they induce profound changes in developmental trajectories that may underlie in part the lasting endocrine and behavioral phenotypes associated with antenatal GC treatment.
These results raise the important question as to whether they may also extend to men. Since experimental glucocorticoid application to a human fetus is ethically inacceptable, such studies have to rely on postmortem brain analyses of people who were exposed to different stress conditions during early life. In a hypothesis-driven approach, McGowan et al. \[[@B99-ijms-18-01711]\] originally detected differential hippocampal *NR3C1* promoter methylation and gene expression between suicided subjects with histories of childhood abuse or severe neglect relative to controls (victims of sudden, accidental death with no history of abuse or neglect). In light of previous findings in rats \[[@B90-ijms-18-01711]\], the researchers extended their analysis to postmortem hippocampal tissue from humans with and without a history of early life adversity (12 individuals in each group) \[[@B100-ijms-18-01711]\]. Methylation profiles covered the genomic region from 3.25 Mb upstream to 3.25 Mb downstream of *NR3C1* at 100-bp spacing, which was compared to the homologous regions in rats that had experienced differential maternal care. Methylation profiles showed hundreds of differentially methylated regions (DMR) associated with differences in early life care that were unevenly spread across the *Nr3c1* locus in rats. In humans, 281 regions were differentially methylated between individuals without and with a history of early life adversity. Among these, 126 DMRs were hypermethylated in non-exposed individuals versus 155 hypermethylated DMRs in exposed individuals. In comparison, the rat profiles showed twice as many DMRs, possibly reflecting the more homogenous study group, of which 373 and 350 DMRs were associated with high and low maternal care, respectively.
Moreover, DMRs showing the same direction of change in response to early life experiences clustered in large genomic regions, indicating a high level of organization connecting distant sites. For example, DMRs associated with early adversity clustered at the *PCDH* locus in both species. Taken together, these findings suggest that epigenomic responses to early life adversity are conserved between rats and humans and can affect broad regions in the hippocampus, including the *NR3C1* and *PCDH* loci.
Following this, the research teams \[[@B101-ijms-18-01711]\] extended their investigations to the genome-wide analysis of DNA methylation in individuals (*n* = 25) exposed to early life adversity (i.e., severe abuse during childhood) in comparison to non-exposed controls (*n* = 16). Neuronal and non-neuronal cells from hippocampal tissues were sorted by fluorescence-assisted cell sorting, and methylated DNA fractions were isolated by immunoprecipitation and subsequently hybridized to a custom-designed 400K promoter tilling array containing 23,551 proximal promoter regions. The methylation profiles were also compared with the corresponding genome-wide expression profiles derived from RNA microarrays. This approach led to the identification of 362 differentially methylated promoters in individuals exposed to early abuse when compared to controls. Among these promoters, 248 were hypermethylated and 114 hypomethylated, whereby methylation differences occurred mostly in the neuronal cell fraction. Abuse associated epigenetic alterations were evenly distributed throughout the genome, and most of the methylation changes correlated inversely with gene expression levels. Functional annotation clustering analysis evidenced enrichment in genes associated with neuronal plasticity, including cell adhesion and signaling. For example, the most differentially methylated gene corresponded to *ALS2* (Alsin), a member of the guanine nucleotide exchange factors for the small GTPase RAB5 (RAS-associated protein), which plays a crucial role in intracellular endosomal trafficking.
In a further study, the researchers also analyzed genome-wide methylation changes in the hippocampus of suicide completers (*n* = 46 subjects) versus non-psychiatric sudden-death subjects (*n* = 16) by means of their customized 400K tilling array \[[@B102-ijms-18-01711]\]. Predisposition to suicide partially overlaps with vulnerability to depression and strongly associates with depressive psychopathology \[[@B103-ijms-18-01711]\]. Similar to depression, suicide results from the interactions between the genetic, developmental, and social risk factors \[[@B104-ijms-18-01711]\] that are thought to act through lasting mechanisms on brain function. In order to assess the role of epigenomic changes in suicide, Labonté et al. \[[@B102-ijms-18-01711]\] performed an analysis of methylated DNA fractions from neuronal and non-neuronal hippocampal cells, as described above. Additionally, the effects from epigenomic changes on gene expression were assessed by expression profiling on a substantial subgroup of the same tissue samples. The researchers detected 366 differentially methylated promoters in suicide completers relative to comparison subjects, which were evenly distributed across the genome. Among these, 273 promoters were hypermethylated and 93 promoters were hypomethylated, whereby DNA methylation anti-correlated in general with gene expression differences. Functionally, DMRs were enriched in promoters of genes involved in behavioral and cognitive processes, including learning, memory, and synaptic transmission (e.g., *CHRNB2*, encoding neuronal acetylcholine receptor subunit beta-2; *GRM7*, encoding metabotropic glutamatergic receptor 7; and *DHB*, encoding dopamine beta-hydroxylase), which have been associated with vulnerability to suicide \[[@B105-ijms-18-01711]\].
In sum, these results suggest an important role for epigenomic changes in genes regulating behavioral and cognitive processes in the hippocampus of suicide completers. While these findings resemble those on epigenomic effects in response to early life adversity, the DMRs identified in the two studies were different and are known to affect different pathways. Hence, epigenomic changes due to early life adversity seem to recapitulate specific early events rather than a general response to near-term psychopathology. If this is the case, further studies on genome wide epigenomic changes in MDD associated with early life adversity are looked for to deepen our understanding of the molecular pathways affected. It is important to caution, however, that future studies need also to address the cell type specificity of epigenetic effects \[[@B106-ijms-18-01711]\] in face of the high diversity of neuronal cell types in individual brain regions, which can dilute epigenetic marks in bulk tissue preparations and confound any analysis of pathway specific effects. This issue is particularly relevant for the assessment of differentially methylated regions and CpGs across brain development and in disease states in which cell type composition is known to change, possibly due to the disease condition. Such changes include the transition from mitotically active cells to differentiated cells, adjustments in developmental trajectories in response to various environmental insults, and the immigration of immune cells in case of inflammatory and neurodegenerative processes, among other events. Thus, it is critical to properly control for potential differences in cell type composition between cases and controls by e.g., purifying specific populations.
6. Epigenomics in SCZ {#sec6-ijms-18-01711}
=====================
Over the last years, an increasing number of studies have investigated epigenomic changes in rodent brains treated with neuroleptics or in peripheral tissues in medicated and non-medicated schizophrenic patients \[[@B107-ijms-18-01711],[@B108-ijms-18-01711],[@B109-ijms-18-01711]\]. Here, we will focus on recent reports that examined epigenomic changes in postmortem brain tissues from people diagnosed SCZ and will discuss how these findings relate to early brain development and genetic variation.
6.1. Epigenomic Changes in SCZ Are Enriched at Neurodevelopmental Loci {#sec6dot1-ijms-18-01711}
----------------------------------------------------------------------
The first methylome analysis of postmortem frontal brain tissues from individuals diagnosed with SCZ (*n* = 35) or bipoloar disorder (*n* = 35) and from matched controls (*n* = 35) was carried out by Mill et al. \[[@B110-ijms-18-01711]\] by hybridization of enriched unmethylated DNA fractions to CGI-arrays. This analysis showed disease-associated DNA methylation differences in multiple loci, particularly in genes regulating glutamatergic and GABAergic neurotransmission, which are thought to be dysregulated in SCZ. Epigenetic dysregulation also affected genes involved in neuronal development (e.g., *WNT1*, encoding a secreted glycoprotein), learning and motor functions (e.g., *LMX1* and *LHX5*, encoding homeobox transcription factors). The researches further assessed the number of connections between nodes representing correlated methylation observed between different genomic loci. This network approach suggested a lower connectivity of epigenomic changes in major psychosis, pointing to a systemic epigenetic dysfunction in SCZ.
A more recent study on genome-wide DNA methylation changes in postmortem frontal cortex from 24 patients with SCZ and 24 unaffected controls used the Illumina Infinium HumanMethylation450 Bead Chip array, containing 485,000 CpG sites (hereafter referred to as 450K array) \[[@B111-ijms-18-01711]\]. Among these, 4641 probes corresponding to 2926 unique genes were differentially methylated, including *NOS1* (encoding neuronal nitric oxidase 1), *AKT1* (encoding protein kinase B involved in neuronal proliferation, survival, and differentiation), *DNMT1*, and *SOX10* (encoding a transcription factor from neural crest development), among other genes previously associated with SCZ and neurodevelopment. Since the patients diagnosed SCZ were not medication-free, these results raise the question of drug effects \[[@B109-ijms-18-01711]\] and point to the need for longitudinal studies on medication free subjects.
The necessity for longitudinal studies to evaluate epigenomic changes in SCZ, was firstly approached by Pidsley et al. \[[@B112-ijms-18-01711]\], who investigated epigenomic changes in postmortem prefrontal cortex (PFC) and cerebellum from 20 schizophrenic patients and 23 matched controls by the 450K array and subsequently assessed the disease-associated regions in human fetal cortex samples, spanning 23 to 184 days post-conception. Highly significant differentially methylated CpGs were detected in the PFC; specifically, probes in four genes were associated with SCZ at a false discovery rate (FDR) ≤ 0.05: *GSDMD*, promoting programmed necrosis that occurs upon the activation of inflammatory caspases; *RASA3*, encoding GTPase-activating protein-1; *HTR5A*, a human serotonin receptor subtype, and *PPFIA1*, a tyrosine-phosphatase interacting with protein regulation neuronal arborization, spine, and synapse numbers.
The correlation of DNA methylation across adjacent sites \[[@B113-ijms-18-01711],[@B114-ijms-18-01711]\] allows us to aggregate single CpGs in regions and thereby to reduce complexity in the comparison of large sample sizes. This approach corroborated SCZ-associated DNA methylation differences and evidenced additionally hypomethylation in *Neuritin 1* (*NRN1*) that plays a well-established role in neurodevelopment, synaptic plasticity, and the prevention of the effects from chronic stress \[[@B115-ijms-18-01711],[@B116-ijms-18-01711]\].
In order to establish a system-level view of DNA methylation differences associated with SCZ, the researchers carried out additionally a weighted gene co-methylation network analysis \[[@B117-ijms-18-01711]\]. This approach led to the identification of 100 prefrontal modules, with each representing discrete networks of co-methylated sites. Such SCZ-associated co-methylated modules were enriched in neurodevelopmental pathways and loci previously implicated in SCZ. On the other hand, no SCZ-associated modules were identified in the cerebellum, which was consistent with region-specific DNA methylation differences in SCZ.
In further support of these findings, SCZ-associated differentially methylated CpGs were enriched for CpGs that underwent dynamic methylation changes during fetal neocortical development. Specifically, 44% of SCZ-associated CpGs were associated with post-conception age in the developing fetal brain indicating a highly significant enrichment for neurodevelopmental differentially methylated CpGs.
Taken together, this study identified discrete modules of co-methylated loci in the PFC associated with SCZ that are significantly enriched for genes guiding neurodevelopment. Additionally, the methylomic profiling of fetal cortices evidenced that SCZ-associated differentially methylated CpGs undergo dynamic changes in DNA methylation during development. Conclusively, these data strengthen a neurodevelopmental component in SCZ with epigenetic mechanisms, possibly mediating between neurodevelopment dysregulation and risk of disease.
6.2. Epigenomic Changes in SCZ Are Genetically Controlled {#sec6dot2-ijms-18-01711}
---------------------------------------------------------
Epigenetic mechanisms can act as an interface between a genetic blueprint and environmental exposures/developmental cues and are on their own controlled by genetics. In light of SCZ's high heritability, a line of recent studies has examined the impact of genetics on epigenomic changes in SCZ.
The conceptual groundwork for these studies was established in 2001, when Jansen and Nap \[[@B118-ijms-18-01711]\] first proposed the term "genetical genomics" for the identification of genes regulated by genetic variation. Similar to other quantitative trait loci (QTL) that can influence any given trait of interest (e.g., growth, body weight, and disease risk), expression or methylation QTLs (eQTLs and meQTLs, respectively) are identified by measuring gene expression or DNA methylation in panels of genetically different, genotyped people ([Figure 2](#ijms-18-01711-f002){ref-type="fig"}A). Statistical association tests are used to compare expression or methylation levels with the respective genotype of each subject in order to infer eQTLs and meQTLs, respectively. A meQTL is therefore defined as a genomic region that contains one or more DNA sequence variants that regulate the methylation level of other DNA regions harboring regulatory or genic sequences or sequences of unknown functions \[[@B119-ijms-18-01711]\]. Genetically controlled changes in DNA methylation can, but do not necessarily, translate into persistent changes in gene expression. For example, meQTLs may act only in a spatiotemporal or context dependent manner by regulating transcription during sensitive developmental time windows or in the presence of neuronal activation.
Formally, eQTLs and meQTLs are distinguished according to their relative location to the affected genic or non-genic region(s). Local QTLs reside near the site(s) that they regulate and occur to a similar degree (10,000--20,000 QTLs) in human peripheral \[[@B120-ijms-18-01711],[@B121-ijms-18-01711],[@B122-ijms-18-01711],[@B123-ijms-18-01711]\] and brain \[[@B124-ijms-18-01711],[@B125-ijms-18-01711],[@B126-ijms-18-01711]\] tissues. The allele encoding the QTL operates in *cis* by regulating the copy of the gene that localizes on the same physical chromosome ([Figure 2](#ijms-18-01711-f002){ref-type="fig"}B). Frequently, *cis*-eQTLs encode allele-specific differences in regulatory DNA elements, e.g., SNPs in the DNA-binding site of a transcription factor that lead to changes in gene expression and DNA methylation. Both expression and methylation *cis*-QTLs are in general of large effect size and can be identified in fewer than one hundred samples \[[@B120-ijms-18-01711],[@B127-ijms-18-01711],[@B128-ijms-18-01711],[@B129-ijms-18-01711]\].
Alternatively, QTLs can operate in *trans* through altering the expression, structure, or function of a diffusible factor \[[@B130-ijms-18-01711]\]. *Trans*-QTLs are of smaller effect size and do not show the allele-specific differences in gene expression or DNA methylation that are typical for *cis*-QTLs.
Both kinds of QTLs can also reside further away from the gene(s) they control. Traditionally, such distant QTLs were thought to operate in *trans*, a view that needs, however, to be reconsidered given the highly dynamic and topologically structured nucleus \[[@B131-ijms-18-01711]\].
The role of meQTLs in psychiatric diseases was firstly addressed by Gamazon et al. \[[@B132-ijms-18-01711]\] through the analysis of postmortem brain tissues from people diagnosed with BIP (bipolar disorder), a condition sharing substantial genetic overlap with SCZ \[[@B33-ijms-18-01711]\]. Initially, the researchers re-assessed previously published meQTL data from 153 cerebella collected from BIP and control individuals \[[@B133-ijms-18-01711]\] by the inclusion of imputed genotype data. This analysis detected 5974 different genes associated with a *cis*-meQTL. Further, they found that genetic variants regulating DNA methylation levels are enriched in top-ranking risk variants from BIP GWAS. Specifically, 132 *cis*-meQTLs fulfilled these conditions and matched ≈14% of the most significant associations from two previous BIP GWAS. About half of these *cis*-meQTLs corresponded additionally to a *cis*-eQTL, raising the possibility that BIP risk variants jointly regulate DNA methylation and gene expression. However, only a few SNPs among those that were most significantly associated with BIP seemed actually to control both DNA methylation and gene expression (hereafter referred to as combined SNPs) in postmortem brain tissues. This result is in accord with previous reports \[[@B120-ijms-18-01711],[@B124-ijms-18-01711],[@B134-ijms-18-01711],[@B135-ijms-18-01711]\], indicating that the effects of a substantial proportion of meQTLs on gene expression may be context-dependent (see above). Noteworthy, one combined BIP-associated SNP localized in *DLG5*, encoding a scaffolding protein that regulates precursor cell division and proliferation, epithelial cell polarity, cell migration, and adhesion; all of these processes are key to early neural stem cells and neurodevelopment \[[@B136-ijms-18-01711]\].
Following this, Numata et al. \[[@B137-ijms-18-01711]\] conducted a comprehensive meQTL analysis on dorsolateral prefrontal cortices (DLPFC) collected from 106 individuals diagnosed with SCZ and 110 matched controls. Although the researches still used the Illumina 27K array (containing 27,578 CpG sites spanning 14,495 genes), they detected in their *cis*-analysis (*cis* identified as within 1 Mb of a CpG site) that 36,366 SNP-CpG pairs were significantly correlated, independent from the case-control status, and corresponded to 18,452 *cis*-meQTLs.
Taken together, these studies corroborate the existence of abundant meQTLs in the human brain and indicate that epigenomic changes in SCZ are at least in part genetically controlled by GWAS risk variants.
6.3. A Role for Fetal meQTLs in Mediating the Genetic Risk for SCZ {#sec6dot3-ijms-18-01711}
------------------------------------------------------------------
Extending the above findings further, two recent landmark studies on SCZ have comprehensively studied the role of CpG methylation and meQTLs in human fetal and adult brains and how they intersect with genetic risk variants from SCZ GWAS \[[@B125-ijms-18-01711],[@B126-ijms-18-01711]\].
Both studies first sought to investigate the role of CpG methylation and meQTLs during fetal development. Jaffe et al. \[[@B126-ijms-18-01711]\] assessed genome-wide DNA methylation at 230,000 CpGs in DLPFCs collected from 335 individuals ranging in age from the 14th week of gestation to 80 years of age, who were unaffected by psychiatric disease. This approach led to the identification of 6480 statistically significant DMRs that emerged during the transition from the 2nd fetal trimester to postnatal life and that were mapped to 4557 genes ([Figure 3](#ijms-18-01711-f003){ref-type="fig"}). Most of these genes shared a crucial role in brain development and morphogenesis.
At the same time, Hannon et al. \[[@B125-ijms-18-01711]\], detected ≈16,000 *cis*-meQTLs within a 1 Mb sliding window in 166 fetal brains ranging from 56 to 166 days post-conception ([Figure 4](#ijms-18-01711-f004){ref-type="fig"}). In agreement with a previous report \[[@B124-ijms-18-01711]\], the effect sizes of these meQTLs were small, with a median change in DNA methylation per allele of ≈7%. Likewise, only a few *trans*-meQTLs (*n* = 5) of smaller effect size were found.
Functionally, fetal brain meQTLs fulfilled the criteria of regulatory domains \[[@B134-ijms-18-01711],[@B138-ijms-18-01711]\]; they were characterized by the presence of DNase I hypersensitive sites, regulatory histone marks, transcription factor binding sites, and eQTLs. Interestingly, fetal brain meQTLs were strongly enriched in DNA-binding sites for the architectural protein CTCF ([Figure 4](#ijms-18-01711-f004){ref-type="fig"}) that connects higher-order chromatin structure to lineage-specific, but also to aberrant, gene expression \[[@B139-ijms-18-01711]\]. Furthermore, a recent integrated approach for pathways and genes disturbed in SCZ has pointed to a role for CTCF \[[@B140-ijms-18-01711]\]. By connecting genetic variation to genomic function, CTCF may operate as an important organizational factor for fetal meQTLs.
Extending beyond fetal development, both research teams analyzed next the relationship between CpG methylation, meQTLs, and SCZ, regardless of genetic changes identified by previous GWAS.
Interestingly, Hannon et al. \[[@B125-ijms-18-01711]\] found that 2903 CpGs located in PGC risk loci for SCZ were more likely to show differential methylation during the transition of prenatal to postnatal life than non-SCZ risk loci ([Figure 3](#ijms-18-01711-f003){ref-type="fig"}). Additionally, fetal brain meQTLs were four-fold enriched for genome-wide significant PGC risk variants, suggesting further a developmental role for SCZ risk variants ([Figure 4](#ijms-18-01711-f004){ref-type="fig"}). Consistent with these findings, PGC risk variants are hypothesized to play a role in neurotransmission (glutamatergic-, calcium-, and G-protein coupled receptor signaling), synaptic plasticity, and neurodevelopment. Moreover, 83% of the fetal meQTLs persisted also into adulthood and were detected in at least one of the three brain regions (prefrontal cortex, striatum, and cerebellum) analyzed ([Figure 4](#ijms-18-01711-f004){ref-type="fig"}).
In an independent approach, Jaffe et al. \[[@B126-ijms-18-01711]\] performed additionally a comprehensive meQTL analysis in a large set of adult cortices (191 individuals diagnosed with SCZ and 240 matched controls). The researchers found that 62 out of 104 genome-wide significant PGC loci contained a meQTL within 20 kb of tag SNPs and those in LD (*R*^2^ \> 0.6) ([Figure 3](#ijms-18-01711-f003){ref-type="fig"}). While none of these PGC meQTLs was specific to control or disease status, they could still influence SCZ development in response to environmental exposures.
At the same time, Jaffe et al. identified 2104 CpGs in the adult brain that were differentially methylated between SCZ cases and controls. These CpGs were weakly but significantly enriched in PGC risk loci (40 CpGs out of 2104 CpGs). On the other hand, only 97 diagnosis-associated CpGs corresponded to genome-wide significant meQTLs, suggesting that diagnosis-associated CpGs are distinct from SCZ risk loci-associated meQTL.
Taken together, the key findings from these two pioneering studies are that a substantial fraction of PGC risk loci contain a meQTL (62 out of 104 loci) \[[@B126-ijms-18-01711]\] and that fetal meQTLs, which mostly persist (83%) into the adult brain, are four-fold enriched with PGC risk loci \[[@B125-ijms-18-01711]\]. Therefore, genetic variation directing differential DNA methylation in neurodevelopmental genes may constitute an import risk factor in SCZ.
7. Conclusions and Outlook {#sec7-ijms-18-01711}
==========================
Early brain plasticity provides a unique substrate for epigenetic mechanisms in the mediation between a genetic blueprint and adverse environments. Such interactions can trigger sustained epigenomic changes that preserve early life events and thereby contribute to the development of MDD/SCZ. Genome-wide changes in DNA methylation in response to early adversity have been detected in MDD, which is consistent with a major environmental component in this disease. Moreover, genetic variations controlling dynamic DNA methylation in early life are recognized to influence later epigenomic changes in SCZ. This finding is consistent with SCZ's high genetic load and neurodevelopmental origin and strengthens the concept that epigenetic changes in response to the environment are, at least in part, genetically controlled. At the same time, meQTLs are enriched in GWAS risk variants, supporting their role in SCZ. In both MDD and SCZ, epigenomic changes localize to regions containing genes important to different aspects of early brain development, including neural proliferation, differentiation, and synapse plasticity, among others.
The genomic distribution of DNA methylation encodes important biological information and is central to ongoing efforts to understand its role in development and disease. DNA methylation analysis technology has rapidly progressed over the past decade \[[@B141-ijms-18-01711]\], whereby the implementation of array hybridization techniques greatly facilitated early genome-scale analysis of DNA methylation. Endonuclease-treated or affinity-enriched DNA methods (used in references \[[@B96-ijms-18-01711],[@B98-ijms-18-01711],[@B100-ijms-18-01711],[@B101-ijms-18-01711],[@B102-ijms-18-01711],[@B110-ijms-18-01711]\]) are particularly well suited for array hybridization. More recently, high density CpG array systems have been broadly used for large clinical samples (used in references \[[@B111-ijms-18-01711],[@B112-ijms-18-01711],[@B125-ijms-18-01711],[@B126-ijms-18-01711],[@B132-ijms-18-01711],[@B137-ijms-18-01711]\]). This technology enables content selection independent of the bias-associated limitations often associated with methylated DNA capture methods and combines comprehensive coverage and high-throughput capabilities. In this regard, the recently introduced MethylationEPIC BeadChip covers more than 850,000 methylation sites, enabling a pan-enhancer and coding region view of the methylome.
While whole genome bisulfite sequencing has the advantage of theoretically capturing all cytosines in the genome at single-nucleotide resolution, it has also a number of significant practical drawbacks as its cost and inefficiency are limiting its broad use despite decreasing sequencing costs. As an alternative, targeted bisulfite sequencing of the dynamic DNA methylome \[[@B142-ijms-18-01711]\] maintains the ability to link cytosine methylation to genetic differences, the single-base resolution, and the analysis of neighboring cytosines, while notably reducing the cost per sample by focusing the sequencing effort on the most informative and relevant regions of the genome. Since no single technique covers at present all aspects, sample numbers and characteristics, as well as the desired accuracy, coverage, and resolution, continue to influence the choice of technique until single-molecule and nanopore sequencing approaches are likely to catalyze the next transformation in high-throughput DNA methylation analysis \[[@B143-ijms-18-01711]\].
At this time, postmortem epigenomic studies on MDD and SCZ have still to overcome a number of limitations to unfold their full potential. For MDD, larger sample sizes are needed to extend the present findings and to assess the possibility of genetic variation in vulnerability to early life adversity. Since our knowledge on epigenomic changes in MDD is still based on the hybridization of DNA recovered from immunoprecipitation to customized tilling arrays, a pan-enhancer and coding region view of the methylome in MDD is urgently looked for. In this respect, postmortem epigenomic studies in SCZ are more advanced compared to those in MDD, both in terms of sample size and comprehensive genome-wide coverage, although the potential relationship between genetically controlled methylation changes and early life events, if at all, needs to be explored in greater depth.
It is also important to note that accumulating evidence from human, animal, and in vitro studies that indicates that antipsychotic and antidepressant agents can influence the epigenetic machinery and lead to changes in DNA methylation, histone modifications, and possibly micro RNA expression \[[@B109-ijms-18-01711],[@B144-ijms-18-01711]\]. Likewise, the gradually developing but persistent therapeutic effects of antidepressant medications may be achieved in part via epigenetic mechanisms. Although current studies on epigenetic deregulation in SCZ and MDD have accounted for various demographic variables such as age, sex, and post-mortem interval, the effects of medication are still incompletely acknowledged. This failure refers to dosage regimen, combination therapy, and the duration of therapy. Moreover, treatment responses to psychopharmacological therapy depend as well on genetic variation \[[@B145-ijms-18-01711]\]. Therefore, future studies on post-mortem epigenomic changes in major psychosis need to take into account that genotypes, epigenetic mechanisms, and psychopharmacology may interact at multiple layers.
Given the scarcity of well-documented and suitable brain samples (e.g., from suicide with a history of early life adversity) cross-sectional and longitudinal studies on genome-wide methylation changes in peripheral blood cells have been increasingly adopted for practical reasons \[[@B146-ijms-18-01711]\]. These approaches can deliver biomarkers for disease onset, progression, and therapeutic responses; however, they appear less suitable to elucidate the molecular and cellular processes underlying the actual disease processes in MDD and SCZ. Cell-type specific differences in the response to the environment and the genetic control of DNA methylation are likely to exist between peripheral blood cells and the brain but also within each tissue, particularly in the brain \[[@B147-ijms-18-01711]\]. Challenges from tissue heterogeneity can confound spatiotemporal effects from environmental exposures and from genetic variation, and ultimately, conceal functional causality. For example, the epigenetic programming of *Nr3c1* in response to early life stress in mice shows a high degree of cell-type specificity in the paraventricular hypothalamic nucleus that associates with distinct endocrine and behavioral phenotypes upon renewed stress exposure \[[@B106-ijms-18-01711]\]. In view of the brain's high specialization, epigenomic changes have to be mapped to cells to function to promote insight into complex disease processes and drugable targets.
Relatedly, SCZ-associated SNPs account for rather small changes in methylation differences, which correspond to 1.3% of average methylation differences in diseased and control prefrontal cortex samples \[[@B126-ijms-18-01711]\] and a 6.7% difference per allele for the average meQTLs \[[@B125-ijms-18-01711]\]. Both studies do not investigate the effects on gene expression in homogenized tissues that are very likely to dilute cell-type specific effects and biologically relevant variability at the level of single cells. Refined biostatistical methods can resolve, at least in part, such limitations by correcting for differences in cell type compositions \[[@B148-ijms-18-01711],[@B149-ijms-18-01711]\]. Moreover, recently developed single-cell assays for genome, epigenome, and transcriptome analysis provide new opportunities to advance cellular resolution in healthy and diseased brains \[[@B150-ijms-18-01711],[@B151-ijms-18-01711]\].
Beyond technical improvements, an intriguing question is whether epigenomic changes in MDD/SCZ are to be expected to translate into gene expression changes in postmortem brain to be classified as functional, and, by inference, disease-relevant. Studies on postmortem brains from suicide completers with and without a history of early life adversity have focused on epigenomic changes in regions associated with gene expression changes in neurodevelopmental pathways. In case these epigenomic changes are established preferentially in early life, they can, but must not, underlie changes in gene expression in adult life. In the absence of longitudinal studies, primary and secondary epigenomic changes remain undistinguishable, as are their potential effects on gene expression.
Similarly, the origin and functional consequences of fetal meQTLs remain presently unanswered. Conceptually, meQTLs map genome-wide DNA methylation levels to genetic variation but do not pinpoint causal variants, a caveat that applies as well to conventional GWAS. Hence, meQTLs can affect DNA methylation at multiple genes that may or may not translate in early and/or adult gene expression changes, irrespective of a causative role. Furthermore, the effects of meQTLs on gene expression may be confined to spatiotemporal time windows or encode gene expression potential that depends on (renewed) neuronal activation to manifest \[[@B152-ijms-18-01711]\]. Taken together, constraints in gene regulation during development and beyond could well explain the low correlation between meQTLs and gene expression levels reported so far.
In this context, induced pluripotent stem cells (iPSCs) \[[@B153-ijms-18-01711]\] from case and control subjects could provide an interesting opportunity to address part of these questions. iPSCs recapitulate major features of fetal brain cells and can be differentiated into various cell types \[[@B154-ijms-18-01711]\]. Moreover, organoids derived from 3D culture show gene expression profiles \[[@B155-ijms-18-01711]\], epigenomic features \[[@B156-ijms-18-01711]\], and structural self-organization into various regions \[[@B157-ijms-18-01711]\], closely mimicking fetal tissues. The role of non-coding variants of fetal meQTLs in DNA methylation and gene expression can be further assessed within such cellular models by using programmable nucleases (e.g., RNA-guided engineered nucleases derived from the bacterially clustered, regularly interspaced short palindromic repeat (CRISPR-Cas) associated system). These tools enable us to evaluate the effects of fetal meQTLs through high precision genome editing in an early neurodevelopmental context. While offering potential insight into the role of genetically controlled methylation, it has to be kept in mind that DNA methylation in iPSCs is often aberrant and incompletely reset with differentiation \[[@B158-ijms-18-01711]\]. Such failure can critically confound the interpretation of disease-specific epigenetic mechanisms in iPSC derived neural systems and makes the investigation of multiple, independently generated iPSC clones or populations mandatory.
Overall, recent technical advancements can throw new light on the question of to which degree neuronal epigenomes encode past and present gene expression profiles and how these intersect with genetic risk variants from GWAS. Current findings from MDD/SCZ raise the possibility that neuronal epigenomes trace expression patterns from developmental time windows that are particularly susceptible to environmental or genetic disturbances that may influence vulnerability to disease. If this is the case, timely therapeutic intervention may help to attenuate these processes. The reversibility of epigenetic processes lends wings to this perspective and may offer hope for improving the lives of patients and their families.
We are thankful to the members of our groups for their thoughtful comments and advice. Michael Ziller is supported by BMBF grant 01ZX1504.
Anke Hoffmann, Vincenza Sportelli, Michael Ziller, and Dietmar Spengler jointly wrote this manuscript.
The authors declare no conflicts of interest.
{#ijms-18-01711-f001}
{#ijms-18-01711-f002}
![CpG methylation and meQTLs in fetal and adult brains from controls and people diagnosed with schizophrenia (SCZ). Jaffe et al. \[[@B126-ijms-18-01711]\] assessed 230,000 CpGs during fetal development. Among these, they identified 6480 differentially methylated regions (DMRs) during the transition from the second fetal trimester to postnatal life. These DMRs mapped to 4557 genes. Furthermore, the genetic risk loci for SCZ, previously identified by the Psychiatric Genetics Consortium (PGC), contained 2903 CpGs that were differentially methylated during the perinatal transition phase. In an independent study on adult brains, 2104 CpGs were differentially methylated between controls and people diagnosed with SCZ in the dorsolateral prefrontal cortices (marked in light green and orange, respectively). Moreover, 97 CpGs corresponded to adult meQTLs but not another set of 40 CpGs that mapped to PGC loci. Interestingly though, Hannon et al. found in an independent study \[[@B125-ijms-18-01711]\] as indicated by the dashed line that 62 out of 104 genome-wide significant PGC loci contained an adult meQTL.](ijms-18-01711-g003){#ijms-18-01711-f003}
![A role for fetal *cis*-meQTLs in schizophrenia. Hannon et al. \[[@B125-ijms-18-01711]\] detected 16,000 *cis*-meQTLs in fetal brains aged 56 to 166 days post-conception. Fetal brain *cis*-meQTLs are strongly enriched in DNA binding sites (DBS) for the transcriptional regulator CTCF. Additionally, fetal brain *cis*-meQTLs are four-fold enriched for genetic risk variants identified previously by the Psychiatric Genetics Consortium (PGC). Lastly, 83% of the fetal *cis*-meQTLs are maintained beyond fetal life in the adult brain.](ijms-18-01711-g004){#ijms-18-01711-f004}
| {
"pile_set_name": "PubMed Central"
} |
Sabidi S, Koh SP, Abd Shukor S, Adzni Sharifudin S, Sew YS. Safety assessment of fermented jackfruit (*Artocarpus heterophyllus*) pulp and leaves in Sprague‐Dawley rats. Food Sci Nutr. 2020;8:4370--4378. 10.1002/fsn3.1734
1. INTRODUCTION {#fsn31734-sec-0001}
===============
The increase of health awareness due to change of eating habits and living standards has shifted consumer preference into nutritious, healthy, and disease preventive food which promises wider health benefits. This phenomenon has changed the primary role of food as main energy source to a new concept of biologically active food components for maintaining good health (Granato, Branco, Nazzaro, Cruz, & Faria, [2010](#fsn31734-bib-0007){ref-type="ref"}). The presence of various phytochemicals, antioxidant, vitamin, mineral, and dietary fiber contents confer fruits as an ideal functional food. Unfortunately, the short shelf life of fruits needs proper processing techniques in order to maintain their nutritious market value (Panghal et al., [2017](#fsn31734-bib-0020){ref-type="ref"}).
Jackfruit (*Artocarpus heterophyllus*) has been widely accepted by consumers as highly nutritious fruits and a desirable fruit crop to be grown due to its rich bioactive compounds (Swami, Thakor, Haldankar, & Kalse, [2012](#fsn31734-bib-0023){ref-type="ref"}). Past findings have reported that jackfruit is rich in phenolic compounds which contributing to its multiple pharmacological properties for the treatment of antidiabetic, inflammatory, wound healing, and fungal infection. (Baliga, Shivashankara, Haniadka, Dsouza, & Bhat, [2011](#fsn31734-bib-0001){ref-type="ref"}; Biworo, Tanjung, Iskandar, Khairina, & Suhartono, [2015](#fsn31734-bib-0004){ref-type="ref"}; Jagtap & Bapat, [2010](#fsn31734-bib-0009){ref-type="ref"}). Besides possessing a unique flavor, jackfruits also rich in phytochemical compounds serve as vital components for various health‐promoting benefits (Jagtap, Panaskar, & Bapat, [2010](#fsn31734-bib-0010){ref-type="ref"}). Currently, jackfruit pulp is processed to make chips, chutneys, jellies, candies, or added to favor food products like ice cream, beverages, or desserts. However, overproduction of jackfruits particularly during harvest season and its short shelf life might cause economic lose to farmers. The short shelf life of jackfruits can be improved by manipulating high sugar content of the fruit pulp which is a potential substrate for fermentation process (Kumoro, Sari, Pinandita, Retnowati, & Budiyati, [2012](#fsn31734-bib-0015){ref-type="ref"}). On the other hand, jackfruit leaves are commonly used as a traditional folk medicine for maintaining good health. However, its awful taste is still unfavorable by consumers. Therefore, abundant of jackfruit leaves are disposed as agricultural waste due to lack of product development study. Past toxicological evaluation had reported no toxicity symptoms observed in Albino mice either fed with methanolic or aqueous extract of jackfruit leaves on a dose of 2,000 mg/kg (Bhattacharjee & Dutta, [2013](#fsn31734-bib-0003){ref-type="ref"}).
Microbial fermentation is simple, natural yet powerful techniques to preserve food and improve the biological functionality of raw materials with the aid of microbial works. Fermented foods are more palatable and enriched with aroma and taste, as a consequence of the bio‐fermentation process that capable to hydrolyze macronutrient molecular into easily digestible phytonutrients as being reported in many fermented foods (Kang, Yang, Dominy, & Lee, [2010](#fsn31734-bib-0011){ref-type="ref"}; Marotta, Celep, Cabeca, & Polimeni, [2012](#fsn31734-bib-0017){ref-type="ref"}). In Malaysian Agricultural Research and Development Institute (MARDI), we have developed a new bio‐fermentation process to produce fermented jackfruit pulp (JP) and leaves (JL) extracts with improved functional bioactivities using pure symbiotic culture of bacteria and yeast (SCOBY) from MARDI\'s Collection of Functional Food Cultures (Koh et al., [2018](#fsn31734-bib-0012){ref-type="ref"}). The traditional fermented sugared black tea (Kombucha tea) was produced using symbiotic aggregate of various yeast and bacteria. Even though this fermented tea has been claimed as healthy elixir with medicinal properties, however, there is limited evidence available which raise safety concern that it may cause health risk. Hyperthermia, hepatic dysfunction, lactic acidosis, respiratory failure, or fatal are few toxicity cases had been reported upon Kombucha tea ingestion (Kole, Jones, Christensen, & Gladstein, [2009](#fsn31734-bib-0014){ref-type="ref"}). Our SCOBY jackfruit extracts taste acidic due to the presence of multiple organic acids produced as a result of microbiological activities as shown in Table [1](#fsn31734-tbl-0001){ref-type="table"}. To date, there is no safety assessment carried out for this newly developed fermented JP and JL extracts. Because of this reason, we were focused on the safety aspect evaluation of fermented jackfruit JP and JL extracts in rats fed continuously at a high dosage of 4,000 mg/kg. The body weight, hematological, biochemical, and histopathological parameters were analyzed as recommended in the Nurul et al.\'s study ([2018](#fsn31734-bib-0018){ref-type="ref"}). The main objective of this experiment was to determine the possible toxic effect of fermented JP and JL extracts on both female and male Sprague‐Dawley rats after oral administration for 28 consecutive days.
######
The pH, brix value and organic acids profile of fermented jackfruit extracts
Sample pH value Brix value Acetic acid Citric acid Oxalic acid (µg/ml) Malic acid Kojic acid Quinic acid Succinic acid
-------- ------------- -------------- ---------------------- ---------------- --------------------- -------------- -------------- ---------------- ---------------
JL 3.15 ± 0.02 8.23 ± 0.05 16,892.80 ± 1,269.00 276.05 ± 24.74 16.38 ± 0.27 86.58 ± 4.99 7.21 ± 0.43 478.88 ± 20.23 0.00 ± 0.00
JP 3.06 ± 0.01 11.18 ± 0.04 16,012.58 ± 3,576.00 871.63 ± 14.04 17.33 ± 0.74 84.92 ± 8.24 11.04 ± 0.41 201.21 ± 6.29 84.09 ± 4.56
Each value in Table represents the mean ± standard deviation from triplicate sample analyses
Abbreviations: JL, fermented jackfruit leaves; JP, fermented jackfruit pulp.
John Wiley & Sons, Ltd
2. MATERIALS AND METHODS {#fsn31734-sec-0002}
========================
2.1. Preparation of fermented jackfruit extracts {#fsn31734-sec-0003}
------------------------------------------------
The jackfruit (*Artocarpus heterophyllus* L.) with variety of Tekam Yellow J33 was purchased from a local jackfruit plantation in Lanchang, Pahang. The ripened jackfruits pulp was separated from the seeds and thoroughly cleaned with filtered water before minced into smaller chunks and oven‐dried (45°C) to produce pulp granule. The jackfruit leaves need to wash thoroughly to remove all dirts before oven‐dried to make leaves powder. Both dried substrates were packed in the aluminum bag and stored in a chiller (4°C) for future use.
Both jackfruit leaves and pulp suspension at the concentration of 5% (w/v) were prepared as a growth medium for the production of fermented jackfruit pulp (JP) and leaves (JL) extracts, respectively. Both jackfruit pulp and leaves suspension were inoculated with two types of microorganisms: (a) yeast (*Dekkera* sp.) and (b) acetic acid bacteria (*Komagataiebacter* sp.) were selected from MARDI\'s Collection of Functional Food Cultures. After a week of fermentation process, the supernatant was collected after centrifuged at 11,200 *g* for 10 min to remove microbes and substrate residue. The extracts were named as fermented JP and JL, which were produced from the substrate of jackfruit pulp and leaves, respectively.
2.2. Organic acids analysis {#fsn31734-sec-0004}
---------------------------
Analyses of organic acids profile of fermented jackfruit pulp and leaves extracts were carried out with a high‐performance liquid chromatography (HPLC), Alliance Separation Module (Waters, 2695), equip with a diode array detector (Waters, 2996) as described in Koh et al.\'s study ([2019](#fsn31734-bib-0013){ref-type="ref"}). Quantification of organic acids was determined from a calibration curve obtained by injecting known amount of each organic acid as external standard. All analyses were conducted in triplicate.
2.3. Ultra high‐performance liquid chromatography (UHPLC)---quadrupole time‐of‐flight (QTOF) mass spectrometry analysis {#fsn31734-sec-0005}
-----------------------------------------------------------------------------------------------------------------------
The UHPLC system was performed on ACQUITY UPLC I‐Class system, consisting of a binary pump, a vacuum degasser, an auto‐sampler, and a column oven and was coupled to a Vion IMS QTOF hybrid mass spectrometer from Waters, equipped with a lock spray ion source. The compounds of fermented jackfruit extracts were chromatographically separated using a column ACQUITY UPLC HSS T3 (100 mm × 2.1 mm × 1.8 μm) and with column oven maintained at 40°C. A linear binary gradient of water with 0.1% formic acid and acetonitrile was used as mobile phase A and B, respectively. The mobile system was run according to gradient profile as follows: 0 min, 1% B; 0.5 min, 1% B; 16.00 min, 35% B; 18.00 min, 100% B; 20.00 min, 1% B. The flow rate was set to 0.6 ml/min and the injection volume was 1 μl. The ion source was operated in negative electrospray ionization (ESI) mode under the following specific conditions: capillary voltage, 1.50 kV; reference capillary voltage, 3.00 kV; source temperature, 120°C; desolvation gas temperature, 550°C; desolvation gas flow, 800 L/hr, and cone gas flow, 50 L/hr. Nitrogen (\>99.5%) was employed as desolvation and cone gas. Data were acquired in high‐definition MSE (HDMSE) mode in the range m/z 50--1,500 at 0.1 s/scan. The scan with different collision energies (CE) was acquired during the run: a low‐energy (LE) scan at a fixed CE of 4 eV, and a high‐energy (HE) scan where the CE was ramped from 10 to 40 eV. Argon (99.999%) was used as collision‐induced dissociation (CID) gas.
2.4. Animals {#fsn31734-sec-0006}
------------
Six weeks old of male and female Sprague‐Dawley rats were purchased from A‐Sapphire Enterprise (001303794‐M) which supplied healthy animal for laboratory experiment. All rats were placed in the Animal Metabolism, Toxicology and Reproductive Centre (MARDI). The rats were maintained under standard condition at 25 ± 2°C with 12 hr light/dark cycles. All animals were acclimatized for 2 weeks in cage, provided with standard sawdust bedding, distilled water, and commercial pellet Gold Coin, Malaysia (crude protein 21%, crude fiber 5%, crude fat 3%, moisture 13%, ash 8%, calcium 0.8%, and phosphorus 0.4%). The study was conducted according to the guidelines and was approved by the Animal Ethics Committee of MARDI (20170420/R/MAEC 00008).
2.5. Safety assessment study {#fsn31734-sec-0007}
----------------------------
Male (*n* = 9) and female Sprague‐Dawley rats (*n* = 9) at the approximate average weight of 140 g were separated into three groups; which were normal control rats group with access of food pellet and distilled water ad libitium and another two treated rat groups were fed with 4,000 mg/kg body weight of fermented JP and JL extract, respectively. The jackfruit fermented extracts were administered orally by oral gavage technique for 28 consecutive days, and the observation of morbidity and mortality was carried out at least twice daily. The rats\' body weights gained and glucose level were measured once a week. On day 29, rats were sacrificed in the CO~2~ chamber and blood was drawn afterward via brachial artery. The sacrificed rats were then dissected and the organs (stomach, liver, kidney, and spleen) were harvested and weighed.
2.6. General behavior and mortality {#fsn31734-sec-0008}
-----------------------------------
All rats were individually observed daily for abnormal behavior and appearance especially in changes in skin, fur, eyes, nose, and fecal. Changes in hyperactivity, tremors, ataxia, salivation, diarrhea, lethargy, and sleep of all rats were examined as well for 28 consecutive days.
2.7. Body weight and blood glucose measurement {#fsn31734-sec-0009}
----------------------------------------------
Each body weight and blood glucose of rats was recorded at the initiation treatment and once weekly until the day of necropsy. Necropsy of all rats was carried on Day 29 and individual organ of liver, spleen, kidney, and stomach for each rat group were harvested.
2.8. Hematology analysis {#fsn31734-sec-0010}
------------------------
All rats\' blood samples were withdrawn from the brachial artery and stored in tubes contain anticoagulant EDTA‐2K. Red blood cell, platelets, white blood cell, and lymphocytes\' count were obtained by running automated hematology analyzer (Exigo Blood Hematology Analyzer).
2.9. Serum biochemistry {#fsn31734-sec-0011}
-----------------------
Blood samples were collected from the brachial artery of all rats. Serum samples were obtained by centrifugation (Centrifuge S417R, Eppendorf) at 1,792 *g* for 10 min under the controlled temperature of 4°C. Then, serums samples were analyzed using a clinical chemistry autoanalyzer (DIRUI CS‐300) to identify alanine aminotransferase (ALT), alkaline phosphatase (ALP), aspartate aminotransferase (AST), total bilirubin, total protein, creatinine, urea, and cholesterol readings.
2.10. Histopathology examination {#fsn31734-sec-0012}
--------------------------------
The liver, spleen, kidney, and stomach organ were fixed in neutral buffered 10% formalin solution. All fixed organs were embedded in paraffin, sectioned, stained with hematoxylin and eosin (H&E), and examined under microscope (Leica).
2.11. Statistical analysis {#fsn31734-sec-0013}
--------------------------
Results are presented as mean ± standard deviation of replicated samples. Data were subjected to one‐way analysis of variance (ANOVA) and multiple comparisons were performed by Duncan\'s test. Statistical significance was set at the level of *p* \< .05. All analyses were performed using statistical analysis software, IBM SPSS Statistic 22.0 (IMB Corp.).
3. RESULTS {#fsn31734-sec-0014}
==========
3.1. General behavior and mortality observation {#fsn31734-sec-0015}
-----------------------------------------------
Daily oral administration of 4,000 mg/kg body weight of fermented JP and JL extracts for 28 consecutive days did not induce any obvious symptom of toxicity in rats. There was an absence of hyperactivity, tremors, ataxia, salivation, diarrhea, and lethargy detected. Most importantly, no incidence of mortality and no abnormal behavior were recorded throughout the experiment.
3.2. Net weight gain and glucose measurement {#fsn31734-sec-0016}
--------------------------------------------
Body weight of both male and female rats was found gain gradually in every week. Surprisingly, both female and male treated rats (JL and JP group) showed significantly lower (*p* \< .05) in weight gain over a time when compared to the control group (Figure [1](#fsn31734-fig-0001){ref-type="fig"}). Generally, the glucose level reading (nonfasting state) showed no significant difference (*p* \> .05) between control and treated rat group (Figure [2](#fsn31734-fig-0002){ref-type="fig"}).
{#fsn31734-fig-0001}
{#fsn31734-fig-0002}
3.3. Organ weight, macroscopic and microscopic observation {#fsn31734-sec-0017}
----------------------------------------------------------
The relative organ weight was calculated as (organ/body weight) × 100%. Most of the organ weight data shown no statistically significant differences (*p* \> .05) were observed, except for liver for both male and female rats treated with fermented JP and JL extracts (Table [2](#fsn31734-tbl-0002){ref-type="table"}). However, no abnormal physical appearances or swelling symptom on liver were observed. Furthermore, histopathology analysis on liver, kidney, spleen, and stomach indicated normal cells with no inflammatory symptoms (Figure [3](#fsn31734-fig-0003){ref-type="fig"}).
######
Percentage of relative organ per weight of rats after 28 days treatment with fermented jackfruit pulp (JP) and leaves (JL) extracts
Organ Relative organ body weight (%)
--------- -------------------------------- ---------------- ---------------- ---------------- ---------------- ----------------
Liver 4.57 ± 0.50^b^ 3.79 ± 0.07^a^ 3.86 ± 0.15^a^ 4.75 ± 0.23^B^ 4.20 ± 0.08^A^ 3.94 ± 0.01^A^
Spleen 0.27 ± 0.09^a^ 0.25 ± 0.04^a^ 0.27 ± 0.05^a^ 0.29 ± 0.02^A^ 0.25 ± 0.07^A^ 0.27 ± 0.04^A^
Kidney 0.73 ± 0.04^a^ 0.76 ± 0.01^a^ 0.80 ± 0.02^a^ 0.92 ± 0.04^A^ 0.84 ± 0.04^A^ 0.85 ± 0.03^A^
Stomach 0.61 ± 0.02^a^ 0.62 ± 0.08^a^ 0.64 ± 0.06^a^ 0.52 ± 0.01^B^ 0.50 ± 0.01^B^ 0.42 ± 0.05^A^
The results of both male and female Sprague‐Dawley rats (n=6) of each group are expressed as mean±standard deviation. Means within the control and treated rats in female group with different small letter are significantly (*p* \<0.05) different. Means within the control and treated rats in male group with different capital letter are significantly (*p* \<0.05) different.
Abbreviations: JL, fermented jackfruit leaves; JP, fermented jackfruit pulp.
John Wiley & Sons, Ltd
{#fsn31734-fig-0003}
3.4. Hematology analysis {#fsn31734-sec-0018}
------------------------
Results of the hematological parameters on red blood cell, white blood cell, and lymphocytes\' count displayed slightly significant differences (*p* \< .05) between control and both treatment (JL and JP) groups except platelets (Table [3](#fsn31734-tbl-0003){ref-type="table"}). Nevertheless, all hematology readings were falls within the normal reference range for all hematology parameters (Lillie, Temple, & Florence, [1996](#fsn31734-bib-0016){ref-type="ref"}; Petterino & Argentino‐Storinob, [2006](#fsn31734-bib-0021){ref-type="ref"}).
######
Hematological profile of control and treated rats with fermented jackfruit extracts
Parameter Male Female
---------------------------- --------------------- --------------------- --------------------- ---------------------- -------------------- ----------------------
Red blood cell (10^12^/L) 7.37 ± 0.07^a^ 8.10 ± 0.04^ab^ 8.20 ± 0.62^b^ 7.56 ± 0.35^AB^ 8.02 ± 0.70^A^ 8.02 ± 1.00^B^
Platelets (10^9^/L) 1,071.00 ± 20.51^a^ 1,195.00 ± 20.51^a^ 1,128.00 ± 39.60^a^ 1,031.00 ± 240.42^A^ 825.00 ± 370.52^A^ 1,095.00 ± 151.32^A^
White blood cell (10^9^/L) 6.20 ± 0.57^a^ 5.00 ± 0.85^b^ 8.35 ± 2.05^ab^ 16.10 ± 0.57^AB^ 13.35 ± 0.35^A^ 16.30 ± 5.09^B^
Lymphocytes (10^9^/L) 3.90 ± 0.71^a^ 6.20 ± 0.99^b^ 3.40 ± 0.71^b^ 11.40 ± 2.56^AB^ 9.35 ± 0.64^A^ 9.35 ± 0.64^B^
The results of both male and female Sprague‐Dawley rats (n=6) of each group are expressed as mean±standard deviation. Means within the control and treated rats in female group with different small letter are significantly (*p* \<0.05) different. Means within the control and treated rats in male group with different capital letter are significantly (*p* \<0.05) different.
Abbreviations: JL, fermented jackfruit leaves; JP, fermented jackfruit pulp.
John Wiley & Sons, Ltd
3.5. Serum biochemistry {#fsn31734-sec-0019}
-----------------------
Serum biochemistry data for both female and male rats are shown in Table [4](#fsn31734-tbl-0004){ref-type="table"}. No significant differences (*p* \> .05) were observed in serum biochemistry parameter for ALT, ALP, AST, total bilirubin, total protein, creatinine, urea, and cholesterol for all rat group except for ALT reading for male rats and cholesterol level for female rats which were treated with fermented JP and JL products, whereby it exhibited slight higher reading when compared to control group. Nevertheless, all serum biochemistry parameters were fell within the normal physiological range of Sprague‐Dawley rats ages 8--16 weeks (Petterino & Argentino‐Storinob, [2006](#fsn31734-bib-0021){ref-type="ref"}).
######
Serum biochemistry parameters of control and treated rats with fermented jackfruit extracts
Parameter Female Male
------------------------- ------------------- ------------------- ------------------- ------------------- ------------------ -------------------
ALT reading (U/L) 51.00 ± 3.27^a^ 56.50 ± 4.49^a^ 55.50 ± 9.39^a^ 53.50 ± 2.86^A^ 60.50 ± 1.22^B^ 70.50 ± 0.41^AB^
ALP reading (U/L) 183.00 ± 15.51^a^ 201.50 ± 28.99^a^ 205.50 ± 0.41^a^ 269.50 ± 42.87^A^ 275.00 ± 4.90^A^ 326.00 ± 60.42^A^
AST reading (U/L) 139.50 ± 2.86^a^ 141.50 ± 2.04^a^ 142.50 ± 28.17^a^ 95.00 ± 11.43^A^ 122.50 ± 6.94^A^ 144.50 ± 4.49^A^
Total bilirubin (mg/dl) 0.44 ± 0.01^a^ 0.39 ± 0.00^a^ 0.42 ± 0.00^a^ 0.39 ± 0.00^A^ 0.40 ± 0.00^A^ 0.43 ± 0.01^A^
Total protein (g/L) 78.15 ± 0.78^a^ 80.30 ± 2.20^a^ 84.80 ± 2.78^a^ 75.85 ± 1.18^A^ 88.20 ± 2.12^A^ 87.70 ± 2.94^A^
Creatinine (µmol/L) 47.00 ± 4.08^a^ 50.50 ± 2.86^a^ 56.50 ± 4.49^a^ 49.50 ± 2.86^A^ 51.50 ± 1.22^A^ 54.50 ± 2.04^A^
Urea (mmol/L) 7.25 ± 0.53^a^ 6.65 ± 0.37^a^ 7.50 ± 0.00^a^ 6.35 ± 0.20^A^ 6.70 ± 0.33^A^ 7.50 ± 0.33^A^
Cholesterol (mmol/L) 1.54 ± 0.22^a^ 1.79 ± 0.16^b^ 1.75 ± 0.26^ab^ 1.19 ± 0.00^A^ 1.42 ± 0.03^A^ 1.44 ± 0.13^A^
The results of both male and female Sprague‐Dawley rats (n=6) of each group are expressed as mean±standard deviation. Means within the control and treated rats in female group with different small letter are significantly (*p* \<0.05) different. Means within the control and treated rats in male group with different capital letter are significantly (*p* \<0.05) different.
Abbreviations: ALP, alkaline phosphatase; ALT, alanine aminotransferase; AST, aspartate aminotransferase; JL, fermented jackfruit leaves; JP, fermented jackfruit pulp.
John Wiley & Sons, Ltd
4. DISCUSSION {#fsn31734-sec-0020}
=============
In this study, two types of fermented jackfruit pulp (JP) and leaves (JL) extracts were produced using selected mixed SCOBY strains under controlled bio‐fermentation process. As expected, both fermented jackfruit extracts were found to be more palatable with improved functional benefits after gone through microbial fermentation process. Both fermented JP and JL extracts were acidic with the pH value of 3 because of the presence of multiple organic acids with the predominant content of acetic acid, citric acid, and quinic acid as shown in Table [1](#fsn31734-tbl-0001){ref-type="table"}. Acetic acids plays important role in our body health as it contributes to the multiple biological function, particularly in reducing obesity, anti‐inflammatory effect and improving glucose tolerance in type 2 diabetic rats (Beh et al., [2017](#fsn31734-bib-0002){ref-type="ref"}; Yamashita, [2015](#fsn31734-bib-0024){ref-type="ref"}). However, prolonged consumption of acidic food may have adverse effect and limited evidence available could rise concern on possible serious health risk as being reported by Kole et al. ([2009](#fsn31734-bib-0014){ref-type="ref"}). Therefore, it is crucial to investigate the toxicity aspect of consuming fermented jackfruit extracts.
Throughout 28 days oral administration of fermented jackfruit JP and JL extracts at the high dose of 4,000 mg/kg, the data presented in the results have supported the safety consumption of fermented jackfruit extracts. Raza, Al‐Shabanah, El‐Hadiyah, and Al‐Majed ([2002](#fsn31734-bib-0022){ref-type="ref"}) reported that the body weight of rats might change as a consequence of adverse effect due to the presence of chemical or toxic compounds. The changes will consider as statistically significant if the current body weight loss is more than 10% from the initial body weight. However, in this study, all control and treated rats groups showed a gradually rise in total weight gained throughout 1 month toxicity study which signed a good safety response.
Surprisingly, there was a significant lower (*p* \< .05) in body weight gained in both fermented jackfruit treated male and female rats observed after 1‐month treatment when compared to control group (Figure [1](#fsn31734-fig-0001){ref-type="fig"}). This phenomenon indicated that the fermented jackfruit extracts may possess anti‐obesity effects due to the presence of acetic acid content (\~1.6%), a bioactive metabolite produced from SCOBY strains fermentation process. This data was supported by the scientific findings which revealed that the presence of acetic acid content in vinegar contributing to its anti‐obesity and antiglycemic effects via increasing satiety (Beh et al., [2017](#fsn31734-bib-0002){ref-type="ref"}; Ostman, Granfeldt, Persson, & Bjὃrck, [2005](#fsn31734-bib-0019){ref-type="ref"}). Besides acetic acid, the presence of quinic acids in both fermented jackfruit extracts also reported to have anti‐adipogenic and lipolytic properties as described by Dungjai et al. ([2018](#fsn31734-bib-0006){ref-type="ref"}). The UHPLC‐QTOF mass spectrometry analysis has identified other potential anti‐obesity compounds presents in both fermented jackfruit extracts such as caffeic acid, apigenin, baicalin, catechin, epicatechin‐3‐gallate, galactose, kaempferol glycoside, naringin, neomangiferin, pectolinarin, quercetin, tiliroxide, and toosendanin that have been confirmed by other researchers. The presence of these compounds further supporting the potential of fermented jackfruit extracts as a new anti‐obesity therapeutic agent. Prolong consumption of fermented jackfruit extracts may change in gut microbiota, indirectly leading to lower body net weight gain and its involvement on energy homeostasis (Clarke et al., [2012](#fsn31734-bib-0005){ref-type="ref"}). Therefore, future studies will be focused on anti‐obesity effects of fermented jackfruit extracts and its influence on the gut microbiota profile.
The nonfasting state of glucose level reading showed no significant difference (*p* \> .05) between the control and both treatment groups during 28‐day oral administration (Figure [2](#fsn31734-fig-0002){ref-type="fig"}), indicating neither hyperglycemia nor hypoglycemia effect of consuming fermented jackfruit products. The internal rats organ weight can be used as one of the relevant toxicity indicators as it could lead to morphological changes. In general, the organs weight of both female and male treated rats was comparable to control group. The study also revealed no mortality and adverse effect recorded on the rats in the treatment group as shown in the histopathology findings (Figure [3](#fsn31734-fig-0003){ref-type="fig"}) and relative organ weight data (Table [2](#fsn31734-tbl-0002){ref-type="table"}). In comparison to control rats group, the slightly lower liver weight observed in both treated male and female rat groups might due to the smaller rat size as a consequence of loss weight after 1 month of consuming fermented jackfruit extracts. Nevertheless, all liver, kidney, spleen, and stomach cell of both treated rats group remained in a healthy cell condition even though these rats have been subjected to prolonged consumption of high dose fermented jackfruit extracts.
Toxicity of a test material can be measured using biochemical parameters as an indicator. The serum biochemical data can be used to detect whether the fermented jackfruits extracts may affect the hematopoiesis or leukopoiesis in rats. Generally, hematological profile for both control and treated rats group were in normal reference range even though there was a significant difference recorded between treated group and control group for both gender (Table [3](#fsn31734-tbl-0003){ref-type="table"}). Hence, fermented jackfruit extracts did not interfere with hematological profile in rat blood, supporting the evidence that it is safe to be consumed. The serum biochemistry of control and treated rats was scrutinized to investigate the possible health risk and toxicity effect of consuming fermented jackfruit extracts. As expected, there were no abnormalities observed in serum biochemistry parameters of rats treated with fermented jackfruit extracts when compared to the control group (Table [4](#fsn31734-tbl-0004){ref-type="table"}). The normal levels of ALT, ALP, AST, total bilirubin, total protein, creatinine, urea and cholesterol are good indicators of liver and kidney functions (Hilaly, Israili, & Lyoussi, [2004](#fsn31734-bib-0008){ref-type="ref"}). The normal serum parameters level showed that the oral administrations of high dose fermented jackfruit extracts did not change the physiology function of liver and kidneys rats. This finding was further supported by the histopathology evaluation on liver, kidney, spleen, and stomach which indicating that high dose consumption of fermented jackfruit extracts did not show adverse effect on the morphology of both male and female rat organs (Figure [3](#fsn31734-fig-0003){ref-type="fig"}).
5. CONCLUSIONS {#fsn31734-sec-0021}
==============
In conclusion, there was no toxicological significant symptom observed throughout the 28 days oral administration of high dose fermented JP and JL extracts in Sprague‐Dawley rats as supported by the evidence data from blood glucose, hematology, histopathology, and serum biochemistry profile of both control and treated rats groups. All data collected are found within the reference range reading in normal features of rat. The normal rat behaviors and gradually increase body weight gained throughout 28 days safety assessment study indicating no harmful side effect of consuming fermented jackfruit extracts.
CONFLICT OF INTEREST {#fsn31734-sec-0023}
====================
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
This study was supported by Horticulture Research Centre, MARDI, and financially funded by Malaysia Government Development Fund RMK‐11 (P21003004050001).
| {
"pile_set_name": "PubMed Central"
} |
1. Introduction {#sec1-nanomaterials-06-00144}
===============
The depletion of fossil fuels and the effects of global warming and pollution caused by their combustion is a major challenge of modern societies. In recent times, hydrogen (H~2~) production and storage has been investigated as a potential renewable energy concept. By mimicking the process of natural photosynthesis by plants, an artificial photosynthesis device can be constructed. An artificial photosynthesis device consists of a photoanode and a photocathode, respectively, for the oxidation and reduction of water molecules to O~2~ and H~2~, respectively \[[@B1-nanomaterials-06-00144]\].
Suitable photoelectrodes for photoelectrochemical (PEC) water splitting are titanium dioxide (band gap 3.0 eV), cadmium selenides (band gap 1.7 eV), cadmium sulphides (band gap 2.4 eV) and silicon (band gap 1.1 eV) \[[@B2-nanomaterials-06-00144],[@B3-nanomaterials-06-00144]\]. Silicon is considered to be optimal photoelectrode because of its low band gap, abundant availability in the earth's crust and broad solar absorption spectrum leading to high solar energy conversion efficiency \[[@B4-nanomaterials-06-00144],[@B5-nanomaterials-06-00144]\]. Nanostructured forms of silicon namely porous silicon (pSi) \[[@B6-nanomaterials-06-00144],[@B7-nanomaterials-06-00144]\], pSi micro/nanoparticles \[[@B8-nanomaterials-06-00144],[@B9-nanomaterials-06-00144],[@B10-nanomaterials-06-00144]\] and silicon nanowires (SiNWs) \[[@B11-nanomaterials-06-00144],[@B12-nanomaterials-06-00144],[@B13-nanomaterials-06-00144]\] have demonstrated its potential in solar water splitting. The features demonstrating the potential of SiNWs includes the tunable band gap and antireflective surfaces \[[@B14-nanomaterials-06-00144],[@B15-nanomaterials-06-00144],[@B16-nanomaterials-06-00144],[@B17-nanomaterials-06-00144],[@B18-nanomaterials-06-00144],[@B19-nanomaterials-06-00144],[@B20-nanomaterials-06-00144],[@B21-nanomaterials-06-00144],[@B22-nanomaterials-06-00144]\]. A reproducible and well-regulated fabrication of SiNWs can be achieved through a metal-assisted electroless chemical etching (MACE) approach \[[@B23-nanomaterials-06-00144]\]. High surface area pSi can also be fabricated using MACE which has inward pores instead of vertically standing wires \[[@B24-nanomaterials-06-00144],[@B25-nanomaterials-06-00144]\]. Well-ordered silicon pores/wires can be fabricated using a lithographical etching technique \[[@B26-nanomaterials-06-00144],[@B27-nanomaterials-06-00144]\]. Both SiNWs and pSi have been widely used as photoelectrodes, in PEC water splitting. However, in order to achieve an improvement in the proton reduction using SiNWs/pSi, an electrocatalyst is required.
Platinum (Pt) is an excellent electrocatalyst for proton reduction, but its scant availability in the earth's crust and its high price limits widespread use \[[@B28-nanomaterials-06-00144]\]. A non-noble earth abundant electrocatalyst would be an alternative to overcoming this issue and help in the up-scaling of solar water splitting technologies \[[@B29-nanomaterials-06-00144],[@B30-nanomaterials-06-00144]\].
For SiNWs coated with Pt nanoparticles (NPs), photocurrent densities of −20.7 mA/cm^2^ (*p-*type silicon) \[[@B31-nanomaterials-06-00144]\] and \~25 mA/cm^2^ (0 V vs. Pt counter electrode, *n*-type silicon) were achieved \[[@B12-nanomaterials-06-00144]\]. Similarly, *p-*type SiNWs coated with more abundant molybdenum sulphide (MoS~3~) catalyst showed a photocurrent density of −24.9 mA/cm^2^ \[[@B32-nanomaterials-06-00144]\]. In the case of *p-*type pSi fabricated using MACE coated with Pt NPs, a photocurrent density of \~−22.5 mA/cm^2^ was reached \[[@B25-nanomaterials-06-00144]\]. The conditions used to measure current densities for the above reports were 0 V vs. RHE and light intensity (100 mW/cm^2^) unless otherwise stated. More recently, we have fabricated a pSi photocathode using an electrochemical anodisation technique for solar hydrogen production \[[@B7-nanomaterials-06-00144]\] where in the presence of indium phosphide nanocrystals (InP NCs) and iron sulphur carbonyl (Fe~2~S~2~(CO)~6~) a photocurrent density of −1.2 mA/cm^2^ (light intensity 100 mW/cm^2^) at −500 mV bias potential was achieved \[[@B7-nanomaterials-06-00144]\].
In this work, we successfully demonstrated the fabrication of a SiNWs working electrode coated with Fe~2~S~2~(CO)~6~ catalyst. The fabricated working electrode was assembled into a three electrode electrochemical cell for photocurrent density measurements and the H~2~ produced in the headspace was analysed by gas chromatography (GC).
2. Results and Discussion {#sec2-nanomaterials-06-00144}
=========================
*p-*type planar silicon (resistivity 10--20 mΩ·cm) was etched by means of the MACE technique with 10 min of etching time. [Figure 1](#nanomaterials-06-00144-f001){ref-type="fig"}A show a representative top-view scanning electron microscopy (SEM) image of etched SiNWs sample (10 min) and the inset show the corresponding cross-sectional view. The length of the SiNWs was measured to be approximately 4 µm. [Figure 1](#nanomaterials-06-00144-f001){ref-type="fig"}B shows the transmission electron microscopy (TEM) image of an individual SiNW with a diameter of approximately 120 nm.
[Figure 2](#nanomaterials-06-00144-f002){ref-type="fig"}A--D shows the current density measurements as a function of time at different bias potentials for planar silicon, SiNWs, planar silicon coated with Fe~2~S~2~(CO)~6~ catalyst and SiNWs coated with Fe~2~S~2~(CO)~6~ catalyst photocathodes, respectively. The current density measurements were acquired by ramping the bias potential between −100 to −500 mV for 5 min under AM 1.5 solar illumination (one sun) in 0.1 M H~2~SO~4~. The bare SiNWs photocathode gave a current density of −5 mA/cm^2^ which is 1.3-fold greater than the planar silicon (−4 mA/cm^2^) at a bias potential of −500 mV. An enhanced current density of −17 mA/cm^2^ was observed for SiNWs photocathode coated with Fe~2~S~2~(CO)~6~ catalyst which is 4.3-fold greater than the planar silicon photocathode coated with Fe~2~S~2~(CO)~6~ catalyst (\~−5 mA/cm^2^) at a bias potential of −500 mV. We hypothesise that the high surface area of the nanostructured SiNWs improved the catalyst loading and reduced the reflection of the incoming light for improved solar energy conversion.
Therefore, attachment of a bioinspired Fe~2~S~2~(CO)~6~ catalyst on SiNWs increased the current density to 3.4 and 4.3-fold higher when compared to the bare SiNWs and planar silicon, respectively. These results, obtained with a synthetic \[FeFe\]-hydrogenase mimic, compare well with similar work using hydrogenase enzymes on semiconductor electrodes \[[@B24-nanomaterials-06-00144],[@B33-nanomaterials-06-00144]\]. Very recently, a \[FeFe\]-hydrogenase enzyme was directly adsorbed onto high surface area pSi fabricated through a MACE technique \[[@B24-nanomaterials-06-00144]\]. A current density of 1 mA/cm^2^ at a bias potential of −500 mV vs. Ag/AgCl in phosphate buffer (light intensity of 10 mW/cm^2^) was achieved. This current density is several orders lower than the performance of SiNWs described in this work.
A long run measurement was performed to test the stability of the catalyst-loaded planar silicon and SiNWs photocathode for 7 h in 0.1 M H~2~SO~4~ at a bias potential of −500 mV ([Figure 3](#nanomaterials-06-00144-f003){ref-type="fig"}). The planar silicon coated with catalyst (black trace) showed an unstable current density and then decreased with time presumably due to surface oxidation and catalyst being desorbed from the planar silicon surface. However, the current density of SiNWs coated with catalyst (red trace) was stable over 7 h. The sample gas from both the photocathodes were quantified by collecting 500 µL of the sample gas after 1 h from the headspace of the PEC setup and injected into the GC. The peak value at \~0.5 min from the GC ([Figure 4](#nanomaterials-06-00144-f004){ref-type="fig"}) confirmed the presence of H~2~ gas in both the photocathodes \[[@B34-nanomaterials-06-00144]\]. The peak value at \~0.9 min from the GC ([Figure 4](#nanomaterials-06-00144-f004){ref-type="fig"}) corresponds to the atmospheric O~2~. The amount of H~2~ gas was determined to be approx. 315 µmol/h for SiNWs coated with catalyst (red trace) which is 4-fold greater than the planar silicon coated with catalyst (\~85.7 µmol/h, black trace).
3. Materials and Methods {#sec3-nanomaterials-06-00144}
========================
3.1. Materials {#sec3dot1-nanomaterials-06-00144}
--------------
SiNWs were fabricated from *p-*type silicon wafers (Czochralski, Silicon Quest Intl. Ltd., San Jose, CA, USA) with resistivity of 10--20 mΩ·cm, orientation (100). Hydrofluoric acid (HF) (48%) was purchased from Scharlau Chemie (Chem-Supply Pty. Ltd. Australian representation, South Australia, Australia). Silver nitrate (AgNO~3~) and hydrogen peroxide (H~2~O~2~) (30%) were purchased from Merck (Victoria, Australia).
3.2. SiNW Fabrication {#sec3dot2-nanomaterials-06-00144}
---------------------
A *p-*type silicon wafer with resistivity of 10--20 mΩ·cm was used as a starting wafer. The wafer was cleaned by ultrasonication in acetone, ethanol and deionised water for 5 min, respectively. The wafers were cut into 1 × 1 cm^2^ pieces and dipped in 1:1 HF and ethanol solution to remove the native oxide layers. The unpolished side of the wafers were masked using sticky tape to avoid etching of the surface. Firstly, the wafers were dipped in 4.8 M HF and 0.02 M AgNO~3~ solution for 30 s to deposit Ag on the polished side. The wafers were then immediately dipped in the etching solution of 4.8 M HF and 0.1 M H~2~O~2~ for 10 min. The etched wafers were then rinsed with de-ionised water and the sticky tape was removed from the unpolished side of the wafers. Finally, the etched wafers were dipped in concentrated nitric acid for 20 min to remove Ag coating, then washed with de-ionised water.
3.3. Electrode Fabrication {#sec3dot3-nanomaterials-06-00144}
--------------------------
The 1 × 1 cm^2^ planar silicon and SiNW array pieces were dipped in 1:1 HF:ethanol solution for 2 min to remove the native oxide layers. It was then dried in a stream N~2~ gas and quickly transferred to an argon purged glove box. Then, five layers of Fe~2~S~2~(CO)~6~ catalyst in toluene were drop casted on the samples at room temperature. A back contact to the samples (unpolished surface) was formed using In-Ga eutectic applied via a cotton swab. A copper plate was used as an electrical contact.
3.4. Surface Characterisation {#sec3dot4-nanomaterials-06-00144}
-----------------------------
SEM images were obtained on a FEI Quanta 450 environmental scanning electron microscope (Hillsboro, OR, USA). TEM images were obtained on a computer-controlled TEM JEM-2100F (Jeol Pty Ltd., Peabody, MA, USA), equipped with a field emission gun. SiNWs were scrapped from the etched wafer and were suspended in ethanol. The samples were dried on a 300 lines/mesh copper grid coated with a Formvar film (PST ProSciTech, Queensland, Australia). The instrument was operated at a 200 kV accelerating voltage and images were acquired with a Gatan Orius SC1000 CCD camera (Pleasanton, CA, USA) mounted at the bottom of the column.
3.5. Photocurrent and GC Measurements {#sec3dot5-nanomaterials-06-00144}
-------------------------------------
We used an Abet solar simulator (air mass 1.5--1 sun) to irradiate the samples. The solar simulator was calibrated against a silicon solar cell (New-Spec). Electrochemical measurements were then performed using a PG 310 potentiostat from HEKA Electronics (Lambrecht/Pfalz, Germany). For electrolysis a sealed three-electrode Teflon PEC cell was used consisting of a Pt counter electrode with a frit, a Ag/AgCl 3 M KCl reference electrode and the working electrode. The working electrode was illuminated with a light intensity of 100 mW/cm^2^ with 12 s dark and 12 s light cycle to measure the photocurrents over a period of 5 min. The potential between the working and reference electrodes was ramped between 0 and −500 mV in 100 mV steps. Photocurrent runs over 7 h were performed under a light cycle at a bias voltage of −500 mV. Gas in the headspace (500 µL) above the electrolyte was sampled after 1 h of electrolysis and analysed using a SRI 310C series GC (Torrance, CA, USA) equipped with a thermal conductivity detector and a column held at 70 °C in N~2~ as the carrier gas.
4. Conclusions {#sec4-nanomaterials-06-00144}
==============
SiNWs coated with catalyst showed a 3.4-fold increase in the current density when compared to bare SiNWs. We demonstrated a strategy of fabricating a robust SiNW array photocathode using a bio-inspired catalyst that was able to consistently produce H~2~ gas for up to 7 h. The attachment of hydrogenase synthetic mimics as catalysts onto high surface area SiNW photocathodes initiate an affordable solar energy conversion by replacing Pt-based catalysts.
S.C. thanks the University of South Australia for a PhD scholarship.
S.C., N.H.V., T.N. conceived and designed the experiments; S.C. performed the experiments; S.C, N.H.V and T.N analysed the data and wrote the paper.
The authors declare no conflict of interest.
{#nanomaterials-06-00144-f001}
{#nanomaterials-06-00144-f002}
{#nanomaterials-06-00144-f003}
{#nanomaterials-06-00144-f004}
| {
"pile_set_name": "PubMed Central"
} |
Findings {#Sec1}
========
Background {#Sec2}
----------
*Angiostrongylus vasorum* (Baillet, 1866) Kamensky, 1905, is a protostrongylidae parasite nematode of domestic dogs and wild canids, which causes angiostrongylosis, disease important in public health. In this aspect, stands out the presence of free larvae in the environment and thus the possibility of human infection, since other parasites of the genus *Angiostrongylus* are proven zoonotic. In dogs, the disease is associated with the occurrence of cough, dyspnea, exercise intolerance, weight loss, vomiting, neurological signs, heart failure and death. The infection of dog may occur when ingesting 1) infected paratenic hosts, such as frogs and small mammals, 2) infected intermediate hosts (molluscs) of the genera *Biomphalaria* and *Physa* and among others, 3) or food or water contaminated with free infective larvae in the environment \[[@CR1], [@CR2]\].
Despite some successful cases, currently the control of this parasite has been associated with use of anthelmintics that although routinely used, do not act very well on the parasite in the definitive host \[[@CR3]\]. Moreover, angiostrongylosis rarely develops acutely and in this sense the clinical signs are perceived later, allowing a continuous environmental dispersion of the parasite through the feces of infected dogs. Some authors have shown that the use of complementary measures to combat helminthoses that complete their development in the environment can be used as tools of control \[[@CR4], [@CR5]\]. Thus, the use of nematophagous fungi is cited here. These organisms use mechanical devices such as modified hyphae (traps) and enzymatic artifices to overcome the nematode larvae (production of hydrolytic enzymes, especially proteases) \[[@CR6]\].
According to Soares and colleagues \[[@CR5]\], one of the promising genera of nematophagous fungi is *Monacrosporium*. Those authors have developed some works that demonstrate their nematicidal activity. In this context, in a recent study, Soares and colleagues \[[@CR7]\] showed that the predatory nematophagous fungus *Monacrosporium sinense* (SF53) produces three proteases with nematicidal activity when grown on solid media culture. It is suggested that these extracellular proteases are important at various stages of infection, such as release of nutrients for growth of the microorganism penetration of the cuticle and digestion of the host tissue. However, the production of an enzymatic complex by this fungus remains unclear.
Thus, the objective of this work was to evaluate the production of proteases from nematophagous fungus *Arthrobotrys sinensis* in liquid medium and its nematicidal activity on first stage larvae of *A. vasorum*.
Materials {#Sec3}
---------
The nematophagous fungus *A. sinensis*, isolate SF53, was used for the production of proteases in liquid medium. This isolate is derived from Brazilian soil and has been kept under laboratory conditions through continuous transfer to solid medium. The fungus was cultivated for 10 days in the dark. Then fungal mycelia were transferred to previously autoclaved flasks containing 50 ml of liquid medium composed of (in grams per liter): glucose, 10; yeast extract, 10; K~2~HPO~4~, 5; MgSO~4~, 0.10; ZnSO~4~, 0.005; FeSO~4~, 0.001; CuSO~4~, 0.0005. The inoculum has grown in shaken flasks at 120 × g. After 6 days, proteases were obtained in its crude form, by filtration using Whatman no.1 filter paper, followed by centrifugation for 5 min at 10 × g and 4°C. The supernatant (crude proteases) was used in the subsequent assays \[[@CR8]\].
In this study, the strain of *A. vasorum* used has been maintained by the Department of Parasitology, Federal University of Minas Gerais and it is originated from naturally infected dogs, from the city of Caratinga, Minas Gerais \[[@CR9]\]. For its obtaining, faeces of infected dogs were collected and placed in a modified Baermann apparatus for the recovery of L~1~. The faeces remained in the apparatus for 12 hours. After this period, the tube was removed, centrifuged at 200 × g for 2 min, the supernatant was discarded and the pellet containing the *A. vasorum* L~1~ was resuspended in 5 ml of 0.85% NaCl. The content present in the tube was homogenized, and from this three aliquots were taken of 10 μL, distributed in glass plate of 7.5 × 2.5 cm. The larvae were counted using a stereomicroscope at increase of 25× \[[@CR1]\].
The proteolytic activity was measured by the method of Soares and colleagues \[[@CR7]\]. A standard curve of tyrosine was built for the quantification of enzyme activity. One unit of protease was defined as the amount of enzyme required to liberate 1.0 μg of tyrosine per minute under the assay conditions.
A zymogram with co-polymerized casein in an acrylamide gel \[[@CR10]\] as substrate (casein-SDS-PAGE) was performed as described by Soares and colleagues \[[@CR7]\]. The proteolytic activity was observed by the formation of white halos. The halos of digestion were excised and analyzed by SDS-PAGE (Laemmli, 1970) in order to verify the presence of enzymes.
An *in vitro* assay was conducted to evaluate the nematicidal action of the proteases of *A. sinensis* (SF53) produced in liquid medium on *A. vasorum* L~1~ following the methodology of Soares and colleagues \[[@CR8]\]. Two groups were formed in sterile tubes, a treated group containing the crude proteases and a control group (without enzyme), which were then incubated at 26°C in the dark for 24 h. A total of 100 *A. vasorum* L~1~ were poured into sterile tubes containing the crude enzyme. The control group consisted of only 100 *A. vasorum* L~1~ in distilled water. Six replicates were performed for each group. After 24 h, the number of *A. vasorum* L~1~ present in each tube of the treated and control groups was counted.
The data obtained in this experiment were interpreted by analysis of variance in significance levels of 1 and 5% probability. The efficiency of L~1~ predation compared to control was assessed by the Tukey test at 1% probability \[[@CR11]\]. Subsequently, the percentage reduction of average larvae (L~1~) was calculated according to the following equation:
Results and discussion {#Sec4}
----------------------
It was observed that the fungus *A. sinensis* (SF53) has produced proteases, when grown in an inducer liquid medium. However, proteolytic activity (15.78 U/mL) was lower than that obtained when the same fungus was grown on solid culture medium (38.0 U/mL) \[[@CR7]\]. Moreover, a single halo was observed at the beginning of digestion of the gel, suggesting that the three proteases of SF53 are produced in an enzymatic complex of large molecular weight (Figure [1](#Fig1){ref-type="fig"}). When grown on solid culture medium, three evident halos were observed in the zymogram of the same fungus \[[@CR7]\]. Probably this difference is due to the enzymatic extraction by stirring, a more brute technique, in the case of solid medium, which may have ruptured the enzymatic complex, "releasing" each protease.Figure 1**Zymogram of the proteases of** ***Arthrobotrys sinensis*** **(SF53) in liquid medium.** Zymogram of the proteases produced by *Arthrobotrys sinensis* (SF53) in liquid medium. Through analysis of the zymograms, a single halo of digestion at the beginning of the gel was observed, suggesting that the three proteases of SF53 are produced in an enzymatic complex of large molecular weight.
Regarding nematicidal activity, within 24 hours, the proteases produced in liquid medium of *A. sinensis* (SF53) showed a percentage reduction of 64.3% on the L~1~ of *A. vasorum*. Also, difference (p \<0.01) was observed in nematicidal action in relation to larvae present in the control group, in the same studied interval.
The predatory activity of fungi of *Monacrosporium* genus has been tested on *A. vasorum* L~1~. However, only one early work \[[@CR12]\] had demonstrated its capture and subsequent *in vitro* destruction, in culture medium WA2%. Braga and colleagues \[[@CR12]\] demonstrated that isolate SF53 was effective (p \< 0.05) in the capture and destruction of *A. vasorum* L~1~ under laboratory conditions, registering at the end of seven days the percentage reduction of 74.2%. However, in the present work, it was noted that the obtained percentage reduction was 64.3%, what is interesting from a biological point of view, since only enzyme was used (no fungi).
Furthermore, our results suggest that the nematicidal activity of SF53 was due to the action of enzymes on the cuticle of the L~1~ of *A. vasorum*, since the cuticles of larvae are especially rich in proteic components that hinder the action of antagonist organisms (Figure [2](#Fig2){ref-type="fig"}). Accordingly, another work developed by the present group showed that the use of crude enzyme extract of nematophagous fungus on *Ancylostoma caninum* L~3~ (a geohelminth) showed good efficacy \[[@CR13]\]. In that work, it was observed the hydrolysis of the cuticle by enzymatic action, which also has acted inside the nematode, causing its destruction.Figure 2**Nematicidal activity of** ***Arthrobotrys sinensis*** **(SF53) proteases. (a-b)** Photomicroscopy of nematicidal activity of proteases from nematophagous fungus *Arthrobotrys sinensis* (SF53) on first stage larvae of *Angiostrongylus vasorum* after 24 hours (treated group **(a)** and control group **(b)**).
Soares and colleagues \[[@CR7]\] reported that extracellular proteases are an important virulence factor for the species *M. sinense*. Its nematicidal activity was also evaluated on *Panagrellus redivivus* larvae (free-living nematode), and at the end of the experiment the average percentage was 79% of reduction in the number of recovered larvae. In the present work, another enzyme production was tested using liquid culture medium, and the results were interesting both for proteolytic and nematicidal activity. Furthermore, it is also suggested that the use of *A. vasorum* can probably contribute to further research about its control.
Three proteases were produced in both cases (solid and liquid medium), however, probably because of differences in the extraction process (much more gentle in the case of liquid medium, using only filtration and centrifugation, than in the case of the solid medium in which there is an intense mechanical agitation), we can find at beginning of the gel a halo that suggests the presence of an enzymatic complex of large molecular mass, which has failed to migrate into the gel, due to this large mass.
In the present work, three proteases of the isolate SF53 were produced in an enzymatic complex and was also demonstrated that these enzymes were effective in destroying *A. vasorum* L~1~ under laboratory conditions.
Authors' information {#Sec5}
====================
Jackson Victor de Araújo CNPq scholarship.
**Competing interests**
The authors declare that they have no competing interests.
**Authors' contributions**
JHQ; WSL; JVA contributed in the designing and coordinating the experiment and also in the preparation and revision of the manuscript. FEFS, FRB and TTZ carried out the experiments. All authors read and approved the final manuscript.
The authors thank FAPEMIG, CNPq and CAPES for financial support.
| {
"pile_set_name": "PubMed Central"
} |
1. Introduction {#sec1-ijms-20-01174}
===============
Chemerin, a novel adipokine highly expressed in the white adipose tissue, is associated with inflammation and adipogenesis, and also known as retinoic acid receptor responder protein 2 (*RARRES2*) \[[@B1-ijms-20-01174],[@B2-ijms-20-01174],[@B3-ijms-20-01174],[@B4-ijms-20-01174]\]. Chemerin not only regulates the expression of adipocyte genes linked with glucose and lipid homeostasis but also affects innate and adaptive immunity as well as cascades of fibrinolytic, coagulation, and other inflammatory \[[@B3-ijms-20-01174],[@B5-ijms-20-01174],[@B6-ijms-20-01174]\]. Plasma chemerin is increased in chronic inflammatory diseases, and elevated circulating chemerin levels is positively associated with detrimental effects in lipid, glucose and cytokine homeostasis, serving as a connection among obesity, metabolic disorders, and inflammation \[[@B7-ijms-20-01174],[@B8-ijms-20-01174],[@B9-ijms-20-01174],[@B10-ijms-20-01174],[@B11-ijms-20-01174]\]. Furthermore, by promoting the formation of vascular inflammation through recruiting macrophages to inflamed blood vessels, chemerin may develop atherogenesis \[[@B12-ijms-20-01174]\].
Using a genome-wide meta-analysis, Tönjes et al. \[[@B13-ijms-20-01174]\] highlighted the aspect of *RARRES2* genetic variants in the control of circulating chemerin. Two other genome-wide association studies (GWASs) have indicated no genome-wide significant association between *RARRES2* genotypes and chemerin levels \[[@B14-ijms-20-01174],[@B15-ijms-20-01174]\]. By Genotype-Tissue Expression (GTEx) data set, *RARRES2* SNPs were found associated with the expression quantitative trait loci of *RARRES2* and nearby genes, supporting the crucial roles of *RARRES2* genotypes \[[@B16-ijms-20-01174]\]. Our preliminary analysis revealed that promoter polymorphisms of *RARRES2* were more significantly associated with circulating chemerin levels in a Taiwanese population \[[@B7-ijms-20-01174]\]. The current study aimed to investigate the genetic basis of chemerin levels by conducting a GWAS in a Taiwan Biobank (TWB) population \[[@B17-ijms-20-01174]\] and to confirm the crucial role of circulating chemerin levels and *RARRES2* polymorphisms in the long-term outcome of patients with angiographically confirmed coronary artery disease (CAD), especially when combined with C-reactive protein (CRP) level.
2. Results {#sec2-ijms-20-01174}
==========
2.1. Clinical and Biochemical Characteristics of TWB Participants and CAD Patients {#sec2dot1-ijms-20-01174}
----------------------------------------------------------------------------------
[Table 1](#ijms-20-01174-t001){ref-type="table"} provides a summary of the baseline characteristics of the TWB participants and CAD population stratified by survival status in the follow-up period. Compared with the surviving CAD patients, those who died were older and have higher incidences of diabetes mellitus (DM), initial presentation other than stable angina pectoris, and multiple vessel disease; higher serum creatinine, CRP, and chemerin levels; higher leukocyte counts; and lower hematocrit and estimated glomerular filtration rates (eGFR).
2.2. Results of GWAS and Replication Genotyping {#sec2dot2-ijms-20-01174}
-----------------------------------------------
In the present GWAS, we fitted a linear regression model for genotype trend effects. The peak of the --log~10~ *p* value for circulating chemerin was found on chromosome 7q36.1 where *RARRES2* is located. Eight SNPs passed the genome-wide significance threshold with each minor allele positively associated with circulating chemerin and rs3735167 was the most significant SNP (*p* = 2.35 × 10^−21^) ([Figure 1](#ijms-20-01174-f001){ref-type="fig"}A, [Supplementary Figure S1A and Supplementary Table S1](#app1-ijms-20-01174){ref-type="app"}). Conditional analysis with further adjustment of the rs3735167 genotypes showed none of the SNPs around the *RARRES2* locus had significance *p* \< 0.01 ([Figure 1](#ijms-20-01174-f001){ref-type="fig"}B, [Supplementary Figure S1B and Supplementary Table S1](#app1-ijms-20-01174){ref-type="app"}), indicating that, in this chromosomal region, variances in chemerin concentrations were mainly explained by rs3735167. For replication, we further genotyped rs1962004 using the TaqMan assay in a previously reported cardiovascular health examination population \[[@B10-ijms-20-01174]\] and by stepwise regression analysis, rs3735167 remained the only independent SNP associated with chemerin levels in this population ([Supplementary Tables S2 and S3](#app1-ijms-20-01174){ref-type="app"}).
2.3. Associations Between Chemerin and CRP Levels and Clinical and Biochemical Correlations in the CAD Patients {#sec2dot3-ijms-20-01174}
---------------------------------------------------------------------------------------------------------------
After Bonferroni correction for multiple testing, significant correlations were observed between chemerin levels and BMI; hematocrit, leukocyte, and platelet counts; eGFR; and creatinine and CRP levels ([Table 2](#ijms-20-01174-t002){ref-type="table"}). A positive association between chemerin and BMI was demonstrated in TWB participants (*p* = 1.0 × 10^−72^) and CAD population (*p* = 0.0004) respectively. Furthermore, associations between BMI and tertiles of circulating chemerin levels also showed consistent correlations in TWB participants (*p* = 1.17 × 10^−63^) and CAD population (*p* = 0.002). Significant correlations were also observed between CRP levels and hematocrit, leukocyte counts, and serum creatinine and chemerin levels. By analyzing the associations with risk factors for cardiovascular disease, plasma levels of chemerin were significantly higher in women, current smokers, those with hypertension, and those with DM ([Supplementary Table S4](#app1-ijms-20-01174){ref-type="app"}). Plasma CRP levels were significantly higher in current smokers and those with DM.
2.4. Circulating Chemerin Levels, RARRES2 Genotypes, and Long-Term Prognosis in Patients with CAD {#sec2dot4-ijms-20-01174}
-------------------------------------------------------------------------------------------------
In the CAD population, the follow-up time was 1022 ± 320 days (minimal: 5 days; maximum: 1460 days) with 27 patients died during the follow-up. Using ROC curve analysis and the Youden index, the best prognostic cutoff values were 163.8 ng/mL and 9.7 mg/L, respectively, for chemerin and CRP levels. Kaplan--Meier survival analysis showed that a high chemerin level was a strong predictor of mortality ([Figure 2](#ijms-20-01174-f002){ref-type="fig"}A, *p* = 7.61 × 10^−7^) and a secondary endpoint ([Figure 2](#ijms-20-01174-f002){ref-type="fig"}B, *p* = 2.26 × 10^−9^), as well as a high CRP level was a strong predictor of mortality and a secondary endpoint ([Figure 2](#ijms-20-01174-f002){ref-type="fig"}C,D). When the CAD patients were divided into three subgroups according to chemerin and CRP levels, the combination of high chemerin and CRP levels demonstrated by Kaplan--Meier survival curves was a powerful predictor of all-cause death and secondary endpoints (*p* = 4.74 × 10^−16^ and *p* = 4.64 × 10^−13^, respectively; [Figure 2](#ijms-20-01174-f002){ref-type="fig"}E,F). Cox regression analysis indicated that higher circulating chemerin and CRP levels were the independent predictors of both primary and secondary endpoints ([Table 3](#ijms-20-01174-t003){ref-type="table"}). When combined circulating chemerin and CRP levels were analyzed, a stepwise increase in poor clinical outcomes from low- to high-risk subgroups was noted. As shown in [Supplementary Table S5](#app1-ijms-20-01174){ref-type="app"}, stepwise and significant increases in age, leukocyte and platelet counts, serum creatinine level, and frequency of DM, as well as stepwise decreases in eGFR and hematocrit, were demonstrated for each additional risk of subgroups. We further genotyped the three polymorphisms of rs3735167, rs1962004, and rs7806429 in the CAD population and found borderline significance between *RARRES2* polymorphisms and chemerin levels (minimal *p* = 0.038 for rs3735167; [Table 2](#ijms-20-01174-t002){ref-type="table"}) and no significant difference between *RARRES2* genotypes and the long-term outcome of CAD patients ([Supplementary Figure S3](#app1-ijms-20-01174){ref-type="app"}).
3. Discussion {#sec3-ijms-20-01174}
=============
In this investigation, we confirmed that common variations near or within *RARRES2* were associated with plasma chemerin concentrations at a genome-wide significance with SNP rs3735167 to be the lead *RARRES2* polymorphism in a Taiwanese population. Furthermore, high chemerin and CRP levels and their combination are associated with the severity and a poor prognosis of CAD. This is the first report, to the best of our knowledge, to reveal that the synergistic effect of chemerin and CRP levels predict the long-term outcome of patients with angiographically confirmed CAD. By contrast, markedly decreased explained variance in chemerin levels in patients with CAD indicated that the effect of *RARRES2* polymorphisms was not large enough to alter the risk of mortality and secondary outcomes.
3.1. Chemerin Levels and the Long-Term Outcome of Various Disease States Including CAD {#sec3dot1-ijms-20-01174}
--------------------------------------------------------------------------------------
Chemerin has been suggested as a marker to predict cardiovascular risk \[[@B18-ijms-20-01174]\] and several studies have shown that circulating chemerin concentrations correlated with various cardio-metabolic parameters and with CAD and the severity of atherosclerosis \[[@B18-ijms-20-01174],[@B19-ijms-20-01174],[@B20-ijms-20-01174],[@B21-ijms-20-01174]\]. Gasbarino et al. \[[@B22-ijms-20-01174]\] showed circulating chemerin is associated with carotid plaque instability. Leiherer et al. \[[@B15-ijms-20-01174]\] found elevated plasma chemerin is correlated with renal impairment and is predictive for occurrence of cardiovascular episodes in patients that underwent angiography where half of their study patients had significant CAD. By contrast, all the study participants for the outcome study in our population were angiographically confirmed CAD patients. Our data demonstrated that a high chemerin level is associated with increased mortality and major adverse cerebral and cardiovascular events as well as multiple prognostic predictors of adverse outcomes.
3.2. The Role of Chemerin in the Pathophysiology of CAD {#sec3dot2-ijms-20-01174}
-------------------------------------------------------
Our data revealed that higher chemerin levels were found in CAD patients than in TWB population. Additionally, a recent prospective cohort study demonstrated a strong positive association with a clear dose-response trend between chemerin and myocardial infarction (MI), independent of established risk factors \[[@B23-ijms-20-01174]\]. Participants who developed MI during follow-up had higher concentrations of chemerin than at study baseline. Immune-inflammatory responses have been increasingly proposed in the pathogenesis of atherosclerosis, which is found to be the leading cause of CAD \[[@B24-ijms-20-01174]\]. Chemerin has been proposed to play a vital role in the pathophysiology of CAD by acting as a chemokine and an adipokine, involving mechanisms in more than one level of metabolic and immune-inflammatory processes \[[@B5-ijms-20-01174]\]. It participates in activation and migration of immune cells to sites of injury on endothelium and smooth muscle cells \[[@B20-ijms-20-01174],[@B25-ijms-20-01174]\]. Receptors of chemerin are identified on the endothelium of blood vessels and on their underlying smooth muscle layers \[[@B25-ijms-20-01174]\]. The damage endothelium may uncover chemerin receptors on smooth muscle cells and cause atherosclerosis \[[@B20-ijms-20-01174]\]. Chemerin activates the adhesion of macrophage to fibronectin and VCAM-1, and stimulates adhesion \[[@B12-ijms-20-01174]\]. Secretion of chemerin by perivascular adipose tissue can result in contraction of vascular smooth muscle cells and acts as a link between chemerin and the development of hypertension \[[@B25-ijms-20-01174]\]. Chemerin induces production of the adhesion molecules of ICAM1 and E-selectin and interacts with endothelium \[[@B26-ijms-20-01174]\] to promote the releases of MMP which may play a role on blood vessel remodeling and growing in vitro experiments \[[@B14-ijms-20-01174],[@B27-ijms-20-01174]\]. With the ability to regulate MMPs and other growth factors \[[@B27-ijms-20-01174],[@B28-ijms-20-01174]\], chemerin could involve in the progression and the development of thrombus or embolus. Furthermore, chemerin activates apoptosis in a time- and dose-dependent way in cultured cardiomyocytes, which plays a vital role in the pathophysiological development of diverse heart diseases including CAD, acute myocardial infarction and congestive heart failure \[[@B29-ijms-20-01174],[@B30-ijms-20-01174],[@B31-ijms-20-01174]\]. By acting as an adipokine, chemerin has an established detrimental role in metabolic disorders \[[@B32-ijms-20-01174]\]. Chemerin affects the lipid \[[@B3-ijms-20-01174]\] and glucose metabolism \[[@B33-ijms-20-01174]\] possibly by changing their infiltration into endothelium, these are additional properties of chemerin linked to the pathogenesis of CAD. These observations suggested that chemerin related metabolic and immune-inflammatory pathways are crucial in the pathogenesis of CAD.
3.3. Combining Biomarkers and Risk Scores for the Prognosis of CAD {#sec3dot3-ijms-20-01174}
------------------------------------------------------------------
Multiple marker approaches with or without biomarker scores have improved risk estimations for cardiovascular events in healthy cohorts and patients with acute coronary syndrome \[[@B34-ijms-20-01174],[@B35-ijms-20-01174],[@B36-ijms-20-01174]\]. By evaluating multiple biomarkers of cardiovascular stress, Sabatine et al. \[[@B35-ijms-20-01174]\] found that the approach helped to select those patients with stable coronary disease who were at a higher possibility of heart failure and cardiovascular death, which may be beneficial for identifying patients who obtain compelling advantages from angiotensin-converting enzyme inhibitor treatment. Wang \[[@B37-ijms-20-01174]\] suggested that finding "uncorrelated" biomarkers outside of an already characterized pathway may improve the performance of risk models. Our data showed that circulating chemerin and CRP levels are pathobiologically diverse biomarkers with fair correlations in CAD patients. A combination of these two biomarkers has been found to be associated with multiple risk pathways with synergistic effects in predicting the long-term outcome of angiographically confirmed CAD.
3.4. Lead SNP of RARRES2 Polymorphisms for Chemerin Levels {#sec3dot4-ijms-20-01174}
----------------------------------------------------------
Previous GWASs derived from diverse Caucasian populations have shown variable results on the association between *RARRES2* polymorphisms and circulating chemerin levels \[[@B13-ijms-20-01174],[@B14-ijms-20-01174],[@B15-ijms-20-01174]\]. Tönjes et al. \[[@B13-ijms-20-01174]\] provided the only report in Caucasians revealing genome-wide significant association between *RARRES2* locus and chemerin levels with the rs7806429 in the 3′ untranslated region as the lead SNP. This is in contrast with our GWAS from the TWB population, in which the rs3735167 polymorphism, located −781 base pair upstream of the transcriptional initiation site of *RARRES2*, is the lead SNP for chemerin levels. These differences may attribute to ethnic genetic heterogeneity in the association of *RARRES2* SNPs with chemerin levels; each ethnic group may present specific results. The associations were further confirmed in two other Taiwanese populations, one from a cardiovascular health examination and another from CAD patients. In this study, we also found a markedly diminished effect of *RARRES2* SNPs on chemerin levels in CAD patients when compared with the healthy populations ([Supplementary Table S6](#app1-ijms-20-01174){ref-type="app"}). This may at least partly explain why controversial results were noted in previous GWASs. The diminished effect of *RARRES2* SNPs may also explain why circulating chemerin levels, but not the lead *RARRES2* polymorphism, predict the long-term outcome of angiographically confirmed CAD. The results suggested that the GWAS result from a healthy population may not be directly applied to the disease population such as CAD.
3.5. Limitations of the Study {#sec3dot5-ijms-20-01174}
-----------------------------
This study has several limitations. First, only a medium-sized CAD population was analyzed with a follow-up of a moderate duration and low mortality. A larger population with a longer follow-up may further confirm the associations and roles of multiple markers, thereby facilitating predicting the risk of angiographically confirmed CAD. Second, more than 80% of the patients presented with stable angina pectoris, and only 12% presented with acute coronary syndrome or congestive heart failure. Thus, patients with chronic stable ischemic heart disease constituted most of the study population. Although significantly higher mortality was noted in patients with acute cardiac disease, the adjustment of the clinical presentation did not attenuate the significance of chemerin and CRP levels and their combination in the prognosis, suggesting the crucial role of both biomarkers in the long-term outcome of patients with CAD.
4. Materials and Methods {#sec4-ijms-20-01174}
========================
4.1. Participants {#sec4dot1-ijms-20-01174}
-----------------
The GWAS cohort consisted of participants from the TWB population. Information was gathered at recruitment centers across Taiwan between 2008 and 2015. A total of 2349 participants with no history of cancer, stroke, CAD, or systemic disease were recruited. Exclusion criteria were subjects who announced to withdraw the informed consent (*n* = 2), fasting for \<6 h (*n* = 38), no chemerin level available (*n* = 1), no rs3735167 data available (*n* = 1), and quality control (QC) for GWAS (*n* = 110); finally, 2197 participants were enrolled for the analysis. Ethical approval (approval number: 05-X04-007) was received from the Research Ethics Committee of Taipei Tzu Chi Hospital, Buddhist Tzu Chi Medical Foundation, and Ethics and Governance Council of the Taiwan Biobank (approval number: TWBR10507-02 and TWBR10611-03). Each participant signed an approved informed consent form.
Between July 2010 and September 2013, a total of 565 patients with CAD who presented with more than ≥50% stenosis of one major coronary artery and performed coronary angiography were enrolled from National Taiwan University Hospital. A flow chart of the study inclusion and exclusion criteria and the definition of baseline measurements were previously reported \[[@B38-ijms-20-01174]\] and finally 481 patients were enrolled. From the patients' medical records, all clinical data were collected. All-cause mortality was the primary endpoint. The secondary endpoint was defined as the combination of all-cause death, myocardial infarction, stroke, and hospitalization for heart failure. After a follow-up period of 1022 ± 320 days, 27 patients died. Seven patients who were lost during follow-up after recruitment were called by telephone before the completion of the research. Three of them had expired; the causes of death were confirmed by the family members. Approval was collected from the Institutional Review Board of National Taiwan University Hospital (No.201002015M). Written informed consent was provided to all of the participants.
4.2. Genomic DNA Extraction and Genotyping {#sec4dot2-ijms-20-01174}
------------------------------------------
For the TWB participants, DNA was isolated from blood samples using a PerkinElmer chemagic™ 360 instrument following the manufacturer's instructions (PerkinElmer, Waltham, MA, USA). SNP genotyping was conducted using custom TWB chips and performed on the Axiom Genome-Wide Array Plate System (Affymetrix, Santa Clara, CA, USA) \[[@B17-ijms-20-01174]\]. For the CAD population, genotyping was completed adopting TaqMan SNP Genotyping Assays of Applied Biosystems (ABI; Foster City, CA, USA) \[[@B38-ijms-20-01174],[@B39-ijms-20-01174]\].
4.3. GWAS Analysis {#sec4dot3-ijms-20-01174}
------------------
For GWAS analysis, each genomic DNA was genotyped using the Axiom TM-TWB genome-wide array comprising 642,832 single-nucleotide polymorphisms (SNPs) with minor allele frequencies of ≥5% in a set of 1950 samples from a Taiwanese Han Chinese population \[[@B17-ijms-20-01174]\]. Further, SNP rs3735167, previously reported to be the most significant SNP associated with chemerin levels \[[@B7-ijms-20-01174]\], was also genotyped with the Taqman Assay. In this investigation, all the samples enrolled for the analysis had a call rate of ≥97%. SNP QC was set as follows: An SNP call rate of \<3%, a minor allele frequency of \<0.05, and a violation of Hardy--Weinberg equilibrium (*p* \< 10^−6^); these were excluded from subsequent analyses. After QC, a total of 614,820 SNPs were enrolled for the GWAS analysis.
4.4. Laboratory Examinations {#sec4dot4-ijms-20-01174}
----------------------------
By adopting ELISA kits (R&D, Minneapolis, MN, USA), circulating plasma levels of chemerin were determined. Circulating plasma levels of CRP were measured using the particle-enhanced turbidimetric immunoassay technique (Siemens Healthcare Diagnostics Ltd., Camberley, UK). The increase in turbidity that accompanies aggregation is proportional to the CRP concentration.
4.5. Statistical Analysis {#sec4dot5-ijms-20-01174}
-------------------------
Continuous variables were examined utilizing analysis of variance or a two-sample t-test, and are presented as the mean ± standard deviation, whereas median and interquartile ranges are given when the distribution was strongly skewed. Differences in categorical data distribution were identified by adopting chi-squared test or chi-squared test for trend. To conform to a normality assumption, serum creatinine and fasting plasma glucose levels and fasting plasma CRP and chemerin levels were logarithmically transformed before investigation. A generalized linear model was adopted to examine the relationship of chemerin with the analyzed genotypes and confounders. The genetic effect was assumed to be additive, and adjustments were made for sex, age, body mass index (BMI), and present status of smoking. Genome-wide scans were calculated using the analysis software package PLINK. *p* values below the threshold of *p* = 5 × 10^−8^ were considered genome-wide significant. Conditional analysis in GWAS was conducted by adding the most strongly associated SNP into the regression model as a covariate and by testing the residual association with all remaining SNPs.
We compared CRP and chemerin levels and the rs3735167 genotypes to predict primary and secondary endpoints by plotting curves of receiver operating characteristic (ROC). Subsequently, the area under the ROC curve (AUC) for all variables of interest was compared non-parametrically. A survival curve was identified adopting the Kaplan--Meier estimate, and significance was examined adopting the log-rank method. All calculations were performed using SPSS version 22 (SPSS, Chicago, IL, USA).
5. Conclusions {#sec5-ijms-20-01174}
==============
Our data revealed rs3735167 to be the lead *RARRES2* polymorphism for chemerin levels in a Taiwanese population. Chemerin levels, but not the rs3735167 genotype, predict the long-term outcome of patients with angiographically confirmed CAD, especially when combined with CRP levels.
Supplementary materials can be found at <https://www.mdpi.com/1422-0067/20/5/1174/s1>. Supplementary Table S1. Genome-wide significance for the association between RARRES2 gene polymorphisms and chemerin levels. Supplementary Table S2. *RARRES2* gene polymorphisms and chemerin levels in a cardiovascular health examination population previously reported \[[@B40-ijms-20-01174]\]. Supplementary Table S3. Chemerin levels: Stepwise linear regression analysis, including genotypes, in a cardiovascular health examination population previously reported \[[@B7-ijms-20-01174]\]. Supplementary Table S4. Chemerin and C-reactive protein (CRP) levels according to the cardiovascular risk factors and severity of coronary artery disease (CAD). Supplementary Table S5. Combined chemerin and CRP levels associated with various clinical and biochemical parameters in CAD patients. Supplementary Table S6. Association between *RARRES2* genotypes and chemerin levels in previous genome-wide association studies (GWASs) and in our studies. Supplementary Figure S1. Manhattan plots of the genome-wide association study for chemerin levels. Supplementary Figure S2. Association between BMI and chemerin levels in TWB population and patients with coronary artery disease. Supplementary Figure S3. Kaplan--Meier curves of the cumulative incidence of primary and secondary endpoints.
######
Click here for additional data file.
Conceptualization, Y.-L.K. and L.-K.E.; methodology, Y.-L.K. and S.W.; software, M.-S.T.; validation, F.-T.C. and J.-F.L.; formal analysis, L.-K.E. and Y.-L.K. L.-A.H. and I.-S.T.; investigation, L.-K.E. and Y.-L.K.; resources, Y.-L.K. and J.-M.J.J.; data curation, M.-S.T.; writing--original draft preparation, L.-K.E.; writing--review & editing, Y.-L.K. and L.-A.H.; visualization, S.W. and J.-M.J.J.; supervision, F.-T.C.; project administration, Y.-L.K. and L.-K.E.; funding acquisition, Y.-L.K. and J.-M.J.J.
This research was sponsored by allocation from the Taipei Tzu Chi Hospital, Buddhist Tzu Chi Medical Foundation (TCRD-TPE-MOST-105-03, TCRD-TPE-MOST-106-01, TCRD-TPE-106-C1-1, TCRD-TPE-106-RT-3), grants from the Tzu Chi Medical Mission Project 104-06, Buddhist Tzu Chi Medical Foundation (TCMMP104-06-03), Buddhist Tzu Chi Medical Foundation Academic Advancement (TCMF-A 106-01-16), grants from the National Science Council (MOST 104-2314-B-303-013-MY3) to Y. L. Ko. We thank the staff and participants of the Core Laboratory of the Taipei Tzu Chi Hospital, Buddhist Tzu Chi Medical Foundation for their important contributions. JMJ Juang is assisted by research grants from NTUH-104-S2649, NTUH-104-S2671, NTUH104-2640, NTUH104-UN001, NTUH104L005, MOST-104-2314-B-002-193-MY3, MOST 103-2314-B-002-189 and MOST 107-2314-B-002 -009 and MOST 107-2314-B-002 -261-MY3 and we further greatly appreciate the team and participants of the Sixth Core Lab, Department of Medical Research, National Taiwan University Hospital for their contribution.
The authors have no conflicts of interest to disclose.
{#ijms-20-01174-f001}
{#ijms-20-01174-f002}
ijms-20-01174-t001_Table 1
######
Clinical and biochemical characteristics of the Taiwan Biobank (TWB) participants and coronary artery disease (CAD) patients according to their survival state.
TWB (2197) CAD
----------------------------- -------------------- --------------------- ---------------------- ----------
Baseline characteristics
Sex (male/female) 984/1213 370/84 18/9 0.65
Age (years) 48.4 ± 10.9 64.9 ± 11.0 77.1 ± 9.3 \<0.0001
Body mass index (kg/m2) 24.2 ± 3.5 26.0 ± 4.0 25.2 ± 4.2 0.56
Hypertension (%) 15.6 77.8 85.2 0.58
Diabetes mellitus (%) 5.9 43.2 63.0 0.02
Dyslipidemia (%) 48.5 61.7 48.1 0.90
Current smoker (%) 18.0 24.7 18.5 0.80
Initial presentation
Stable angina pectoris (%) 87.4 29.6 \<0.0001
ACS/MI (%) 5.7 40.7
CHF/lung edema (%) 3.5 22.2
Others (%) 3.3 7.4
CAD (S vs. D vs. T) (%) 29.3:28.6:42.1 3.7:18.5:77.8 0.004
Biochemistry
Serum creatinine (mg/dL) 0.7 (0.6--0.9) 1.1 (0.9--1.3) 1.4 (1.1--2.2) 0.007
eGFR 108.0 ± 25.0 71.0 ± 23.7 46.7 ± 26.0 0.0004
Blood cell counts
Leukocyte counts (10^3^/μL) 6.1 ± 1.6 6.5 ± 1.8 8.3 ± 4.8 0.0007
Hematocrit (%) 43.9 ± 4.6 41.1 ± 5.1 35.4 ± 7.2 0.0008
Platelet counts (10^3^/μL) 240.1 ± 56.4 213.5 ± 60.0 185.4 ± 70.3 0.29
Inflammation markers
C-reactive protein (mg/L) 2.4 (1.2--4.1) 4.2 (2.2--24.7) 0.0002
Chemerin (ng/mL) 96.6 (80.6--110.3) 123.3 (93.8--157.1) 176.2 (108.5--227.6) 0.001
ACS/MI: Acute coronary syndrome or myocardial infarction; CHF: Congestive heart failure; S vs. D vs. T: Single vs. double vs. triple vessel coronary artery disease; eGFR: estimated glomerular filtration rate; Data are expressed as mean ± SD, percentage, or median (interquartile range) as appropriate. A Comparison between CAD patients according to their survival state.
ijms-20-01174-t002_Table 2
######
Association between circulating chemerin and CRP levels and measurable cardiovascular risk factors in patients with coronary artery disease.
Chemerin CRP
--------------------- ----------------------------- ---------- ---------- ---------- -------- ---------- ----------
Anthropology Age (years) 0.011 0.803 0.091 0.046
Body mass index (kg/m^2^) 0.160 0.0004 0.003 0.054 0.242
Blood cell counts Leukocyte counts (10^3^/μL) 0.262 \<0.0001 \<0.0001 0.408 \<0.0001 \<0.0001
Hematocrit (%) −0.382 \<0.0001 \<0.0001 −0.173 0.0002 0.002
Platelet counts (10^3^/μL) 0.200 \<0.0001 0.0002 0.074 0.107
Renal function Serum creatinine (mg/dL) 0.470 \<0.0001 \<0.0001 0.148 0.001 0.009
eGFR (mL/min/1.86 m^2^) −0.553 \<0.0001 \<0.0001 −0.11 0.017
Inflammatory marker CRP (mg/L) 0.378 \<0.0001 \<0.0001
Chemerin (ng/mL) 0.378 \<0.0001 \<0.0001
Abbreviations as in [Table 1](#ijms-20-01174-t001){ref-type="table"}. ^a^ *p* value: Adjusted for sex and age. ^b^ Adjusted *p* value: After Bonferroni correction; a Bonferroni correction for multiple testing was used with α = 0.005 after the nine different tested laboratory variables were considered. Only significant *p* values of \<0.05 are shown.
ijms-20-01174-t003_Table 3
######
Predictors of primary and secondary endpoints in Cox regression analysis.
---------------------------------------------------------------------------------------------------------------------------------------------
Predictors Model 1 ^a^ Model ^b^ Model ^c^
--------------------- ------------------------------ ----------------------- --------------------- --------------------- --------------------
Primary end point Chemerin level subgroups ^d^ Hazard ratio (95% CI) 5.71 (2.62--12.48) 4.55 (1.86--11.16) 3.55 (1.46--8.68)
*p* value \<0.0001 0.001 0.005
CRP level subgroups ^e^ Hazard ratio (95% CI) 7.82 (3.66--16.71) 5.73 (2.39--13.75) 4.27 (1.72--10.61)
*p* value \<0.0001 \<0.0001 0.002
Combined risk subgroups\ Hazard ratio (95% CI) 2.61 (0.97--7.00) 2.72 (0.94--7.93) 1.85 (0.62--5.53)
(intermediate vs. low)
*p* value 0.057 0.063 0.275
Combined risk subgroups Hazard ratio (95% CI) 17.02 (7.04--41.13) 11.17 (3.84--32.47) 8.71 (2.99--25.31)
(high vs. low) *p* value \<0.0001 \<0.0001 \<0.0001
Secondary end point Chemerin level subgroups Hazard ratio (95% CI) 4.44 (2.59--7.60) 3.78 (2.11--6.76) 3.04 (1.69--5.47)
*p* value \<0.0001 \<0.0001 0.0002
CRP level subgroups Hazard ratio (95% CI) 4.84 (2.72--8.60) 3.78 (2.02--7.07) 2.76 (1.45--5.25)
*p* value \<0.0001 \<0.0001 0.002
Combined risk subgroups Hazard ratio (95% CI) 3.82 (2.07--7.05) 4.14 (2.17--7.89) 3.18 (1.64--6.18)
(intermediate vs. low) *p* value \<0.0001 \<0.0001 0.001
Combined risk subgroups Hazard ratio (95% CI) 9.47 (4.70--19.06) 5.87 (2.67--12.93) 4.52 (2.04--10.03)
(high vs. low) *p* value \<0.0001 \<0.0001 0.0002
---------------------------------------------------------------------------------------------------------------------------------------------
95% CI: 95% confidence interval ^a^ Model 1: Unadjusted. ^b^ Model 2: Adjusted for baseline data (sex, age, BMI, current smoking status, diabetes mellitus, hypertension, and dyslipidemia). ^c^ Model 3: Adjusted for baseline data and initial presentation (sex, age, BMI, current smoking status, diabetes mellitus, hypertension, dyslipidemia, and initial presentation). ^d^ Chemerin level subgroups: \>163.8 ng/mL vs. ≤163.8 ng/mL of chemerin level. ^e^ CRP level subgroups: \>9.7 mg/L vs. ≤9.7 mg/L of CRP level.
| {
"pile_set_name": "PubMed Central"
} |
Background {#Sec1}
==========
In 2016, there were approximately 36.7 million people living with HIV (PLHIV), 1.8 million new HIV infections, and 1 million AIDS-related deaths \[[@CR1]\]. In 2014, the Joint United Nations Programme on HIV/AIDS (UNAIDS) established the 90-90-90 goals for the HIV care cascade. The HIV care cascade and these goals aim for 90% of PLHIV to know their status, of whom 90% link to and initiate antiretroviral treatment (ART) and of whom 90% achieve virological suppression by 2020 \[[@CR2]\]. UNAIDS modeling projections estimate that new HIV infections will be reduced by up to 90% if the global community achieves the ambitious UNAIDS 90-90-90 goals by 2030 \[[@CR3]\].
A successful program for the 90-90-90 goals will achieve virological suppression in 73% (90% × 90% × 90%) of PLHIV, which will leave 27% of PLHIV---approximately 9.9 million people---without achieving virological suppression. Virological suppression occurs when an infected individual's HIV RNA level drops below a particular threshold and results in improved health outcomes for the individual and reduced risk of onward transmission. If those who do not achieve HIV virological suppression are the drivers of HIV transmission by engaging in high-risk behaviors, then there is a strong risk that achieving the 90-90-90 targets will not lead to ending the HIV epidemic \[[@CR4]\]. Therefore, incorporating data on the risk profile of the missing 27% who are not virologically suppressed may help to more accurately predict the impact of achieving the 90-90-90 targets \[[@CR5]\]. This will also inform HIV programs to develop strategies, in addition to universal testing and treatment (UTT), that are most beneficial to ending the HIV epidemic \[[@CR6]\].
Simulation studies, observational data, and randomized trials have had mixed results in estimating the population benefits of UTT. Mathematical modeling based on results of a cluster-randomized trial of UTT and circumcision in Zambia and South Africa estimated reductions of incidence between 25% and 62% over 10 years \[[@CR6]\]. A large population-based cohort in South Africa found a considerable reduction in HIV incidence across communities with high levels of ART coverage \[[@CR7]\]. However, completed randomized trials of UTT demonstrated limited population-level effectiveness, which supports the need for more research on the role that groups which are not virologically suppressed play in sustaining high levels of HIV incidence \[[@CR8]\]. Empirical evidence is needed to determine the characteristics of people who are not reached through the 90-90-90 strategy, which may inform future strategies to curb onward HIV transmission.
We report herein our protocol for a systematic review and meta-analysis of the demographic, epidemiologic, sexual-risk behavior, and geographic heterogeneity across the HIV care cascade throughout sub-Saharan Africa.
Aims {#Sec2}
====
The primary aim of this systematic review and meta-analysis is to characterize the HIV transmission potential (defined by demographic and other risk group characteristics) of populations not virologically suppressed in the era of UTT in sub-Saharan Africa. The second aim of this review is to quantify the demographic and risk group heterogeneity at each of the individual 90-90-90 goals---awareness of HIV infection, ART initiation, and HIV virological suppression. We will identify and use subpopulations and risk group strata that have been defined, find comparable strata by which to quantify heterogeneity across the 90-90-90 cascade, and calculate pooled point estimates of the proportion who do not achieve virological suppression within these strata.
Methods {#Sec3}
=======
This systematic review protocol follows the Preferred Reporting Items for Systematic Review and Meta-Analysis Protocols (PRISMA-P) guidelines \[[@CR9]\]. The PRISMA-P checklist can be reviewed as Additional file [1](#MOESM1){ref-type="media"}. This review protocol is registered on the International Prospective Register of Systematic Reviews database (PROSPERO CRD42018089505) \[[@CR10]\].
Eligibility criteria {#Sec4}
--------------------
### Study design and settings {#Sec5}
We will include all observational cohort studies, case-control studies, cross-sectional studies, HIV surveillance reporting, and randomized control trials that report on one or more elements of the 90-90-90 HIV care cascade. We will include studies that report aggregate estimates or stratum-specific estimates. We will exclude studies that provide modeled estimates of testing, treatment, or virological suppression, systematic reviews and meta-analyses, and data collected exclusively prior to January 2014.
### Population {#Sec6}
We will include any study of adult populations, defined as PLHIV aged 15 years or older, located in sub-Saharan Africa. We will include strata defined in Table [1](#Tab1){ref-type="table"}.Table 1Demographic and risk group strata of interestAgeSexSubnational administrative unitUrban, peri-urban, or rural residenceOccupationActive militaryDistance from facilityOut of pocket paymentsMarital statusAge at sexual debutCD4 countTiming of last HIV testEducation attainmentIncomeMigratory populationsRecency of migrationMobile populationsDepression statusAnxiety statusOpportunistic infectionsNumber of concurrent sexual partnersNumber of previous sexual partnersAge of sexual partnersCondom usePregnant womenDiscordant couplesInjection drug usersTransactional sex/commercial sex workersGay, bisexual, and other men who have sex with menTransgender individuals
### Outcomes {#Sec7}
The primary outcome is the prevalence of unsuppressed viral load by demographic and risk group, and risk factors associated with being unsuppressed. The labels and definitions of strata used in included studies will be recorded to document the diversity and similarity of strata categorization (e.g., mobile versus migratory populations). The secondary outcomes of interest are defined as the proportion, relative risk, or odds ratio of the following three groups: PLHIV who know their HIV status, PLHIV who know their status and are receiving ART, and PLHIV on ART who are virologically suppressed.
Information sources {#Sec8}
-------------------
### Electronic databases {#Sec9}
We will conduct four comprehensive literature searches, one for each of the 90-90-90 targets and a fourth search focusing on the entire HIV care cascade (Fig. [1](#Fig1){ref-type="fig"}). We will perform searches in PubMed and Embase and restrict the search to articles published in English between January 2014 and March 2018. We will limit our search to articles published after 2014 because the UNAIDS 90-90-90 targets were announced mid-2014 and the World Health Organization (WHO) universal test and treat guidelines were released in 2015 \[[@CR11]\].Fig. 1Search strategy
In addition to searching the published peer-review literature, we will conduct a comprehensive search of websites and databases for publicly available grey literature sources, such as WHO and UNAIDS reports, Ministries of Health websites and national surveillance reports, Médecins Sans Frontières (MSF) reports, the Integrated Bio-Behavioral Surveys (IBBS), U.S. President's emergency plan for AIDS Relief (PEPFAR) country operational plans, the Analyzing Longitudinal Population-based HIV/AIDS data on Africa (ALPHA) Network, the National Technical Reports Library (NTRL), the Population-based HIV Impact Assessments (PHIA), AIDS Indicator Surveys (AIS), Demographic and Health Surveys (DHS), and other similar national HIV surveys (e.g., the South African National HIV Prevalence, Incidence and Behavior Survey) \[[@CR12]--[@CR22]\]. We will also search HIV conference abstracts including the International AIDS Society (IAS), the Conference on Retroviruses and Opportunistic Infections (CROI), and the International Conference on AIDS and STIs in Africa (ICASA) \[[@CR23]--[@CR25]\]. For eligible grey literature sources, we will review listed citations for additional data sources.
### Search strategy {#Sec10}
A detailed search strategy is presented in Additional file [2](#MOESM2){ref-type="media"}. In brief, MeSH and keywords will be included for HIV, such as "Human Immunodeficiency Virus" and "AIDS", along with terms for each portion of the 90-90-90 cascade, and key terms for "Sub-Saharan Africa" and specific country names using Boolean "AND" and "OR" operators. For the first 90, we will include terms such as "testing", "diagnosis", and "serostatus". The search for the second 90 will include terms such as "antiretroviral" and "treatment", and the search for the third 90 will include terms such as "viral suppression" and "viral load". Lastly, for our fourth search, we will include terms to capture the whole cascade, including "90-90-90", "HIV care cascade", and "universal test and treat". Prior to publishing results, we will update our search to include any additional eligible papers that are published after March 2018.
Study records {#Sec11}
-------------
### Data management {#Sec12}
Each of the separate four searches will be conducted and merged into a reference manager. We will record the number of duplicate records, which will then be removed prior to the selection process. We will use Microsoft Excel and Covidence, a systematic review management software, to document the outcome of the title screening, abstract screening, full text review, and data abstraction processes \[[@CR26]\].
### Selection process {#Sec13}
We will first screen records for inclusion based on title only, duplicated independently by two reviewers. In cases of disagreement, the record in question proceeds to the abstract review stage. Abstract review will also include duplicate reviews. Full texts of the records will then be obtained and reviewed for inclusion and will be conducted independently and in duplicate. Discrepancies from the abstract and full text review will be resolved through consensus or in discussion with a third independent reviewer. Records will be excluded from consideration at title, abstract, and full text review stages if they satisfy any of the following exclusion criteria: study only among HIV-negative persons, study on persons outside of sub-Saharan Africa setting, study on children aged under 15, study of an inappropriate design, study data collected exclusively prior to January 2014, or study does not report on at least one of the 90-90-90 outcomes (Table [2](#Tab2){ref-type="table"}).Table 2Eligibility criteriaCriteriaVariablesInclusion criteriaStudy includes HIV-positive personsStudy set in sub-Saharan AfricaStudy on adults aged 15 and olderStudy design is interventional, cohort, cross-sectional, case-controlStudy data collected at least partially from January 2014 onwardsStudy reports results on at least one 90-90-90 targetExclusion criteriaStudy includes only HIV-negative personsStudy not set in sub-Saharan AfricaStudy includes children aged under 15Study design is qualitative, mathematical model, systematic review, or editorialStudy data collected exclusively prior to January 2014Study does not report on any of the 90-90-90 targets
### Data collection process {#Sec14}
We will develop a data extraction form and pilot test this form on ten randomly selected publications which have been selected for data abstraction. This form will guide the collection of strata-specific estimates of inclusion across the three 90-90-90 targets, as well as strata definitions. Data extraction will be done independently by two reviewers, with discrepancies resolved by consensus or in consultation with a third reviewer.
### Data items {#Sec15}
The data items for extraction are informed by the study aims. We will include the following information from all studies:Study characteristics (authors, year of publication, study period, study design, geographic location, duration of follow-up, key findings)Study setting, specifically differentiating between clinical, community, and population-based study populationsDefinition and methods used to define the reported levels of the cascade, such as the threshold for virologic suppressionDefinitions and inclusion criteria used for specific subgroups defined by age, sex, occupation, migratory/mobile status, and risk behavior as well as key population definitions usedFor each population strata for which data are reported, the sample size, 95% confidence intervals, proportions, relative risk or odds ratio of PLHIV who know their HIV status (first 90), PLHIV who know their status and are on ART (second 90), and PLHIV on ART who are virologically suppressed (third 90)
### Data synthesis {#Sec16}
We will summarize information on virological suppression, and strata associated with being virologically suppressed. Strata reported, definitions used to define strata and levels of the HIV care cascade, and study characteristics including geography and study design will be summarized in the narrative. Study characteristics such as year and study period will be reported separately for each included study. Our qualitative assessment of the comparability of these definitions will inform our quantitative analysis.
We will summarize information on strata associated with knowledge of HIV serostatus, enrollment on ART, and virological suppression. This will include synthesizing odds ratios, relative risks, and proportions of persons virologically suppressed. We will hand calculate measures of association, odds ratio and risk ratio point estimates without confidence intervals, for studies which do not report or calculate measures of association but provide sufficient data for their calculation. Where appropriate and feasible, we will use random effects meta-regression analyses to estimate the distribution of demographic and risk features among groups of combinable strata who are not virologically suppressed. We hypothesize that national-level epidemic characteristics such as HIV prevalence and progress towards the 90-90-90 goals will be critical contextual features to consider in quantitative meta-analyses.
### Missing or incomplete data {#Sec17}
In the case of missing or incomplete data, we will contact corresponding authors to request relevant information or for additional clarification. If the corresponding authors fail to respond, then additional authors will be contacted. A description of the missing data for each included study will be provided, along with the possible implications of missing data. We will assess potential publication bias through the use of funnel plots and a subjective assessment of asymmetry \[[@CR18]\].
### Risk of bias in included studies {#Sec18}
There will be three primary reviewers of included studies. At least two reviewers will assess the risk of selection bias, reporting bias, and attrition bias for each study included for data abstraction by using an adapted Cochrane Risk of Bias Tool including items for non-randomized studies \[[@CR27]\]. Using this tool, reviewers will grade studies as having low, moderate, serious, or critical risk of bias. Conflicting ratings will be resolved through consultation of a third reviewer. We intend to assess the heterogeneity of study results by calculation of the *I*^2^ statistic \[[@CR28]\]. Studies with critical risk of bias will not be considered for quantitative synthesis.
Discussion {#Sec19}
==========
The effectiveness of the UNAIDS 90-90-90 targets to curb the HIV epidemic by 2030 may depend on reaching individuals with the highest risk of onward HIV transmission. In addition, successful achievement of these targets will mean that 27% of PLHIV are not virologically suppressed. Little is known about the demographic profile of those accessing the HIV care cascade and perhaps more importantly those people who are falling off the HIV care cascade. This systematic review and meta-analysis will help identify key demographic, epidemiologic, and risk factor heterogeneity across the 90-90-90, with a specific focus on correlates of being among the 27% who are not virologically suppressed. This will include those who fail to test and get linked to care, in addition to those on ART who do not achieve suppression. Results from this review will be useful in more accurately predicting the impact of achieving the 90-90-90 targets for reducing global HIV incidence and guiding health intervention and prevention strategies to more effectively reach key populations.
Additional files
================
{#Sec20}
Additional file 1:PRISMA-P 2015 checklist. (125 kb) Additional file 2:Search Terms. (19 kb)
AIDS
: Acquired immune deficiency syndrome
AIS
: AIDS Indicator Surveys
ALPHA
: Analyzing Longitudinal Population-based HIV/AIDS data on Africa
ART
: Antiretroviral treatment
CD4
: Cluster of differentiation 4
CROI
: The Conference on Retroviruses and Opportunistic Infections
DHS
: Demographic and Health Surveys
HIV
: Human immunodeficiency virus
IAS
: The International AIDS Society
IBBS
: Integrated Bio-behavioral Surveys
ICASA
: The International Conference on AIDS and STIs in Africa
MSF
: Médecins Sans Frontières
NTRL
: National Technical Reports Library
PEPFAR
: U.S. President's Plan for AIDS Relief
PHIA
: Population-based HIV Impact Assessments
PLHIV
: People living with HIV
PRISMA-P
: Preferred Reporting Items for Systematic Review and Meta-Analysis Protocols
PROSPERO
: International Prospective Register of Systematic Reviews database
UNAIDS
: Joint United Nations Programme on HIV/AIDS
UTT
: Universal test and treat
WHO
: World Health Organization
None
Ethical approval and consent to participate {#FPar1}
===========================================
Not applicable
Funding {#FPar2}
=======
This work was funded by the Bill and Melinda Gates Foundation through an agreement with the University of Washington Strategic Analysis, Research and Training (START) Center. The funders had no role in the decision to publish or preparation of the manuscript.
Availability of data and materials {#FPar3}
==================================
Not applicable
AA and AB conceived the study. DG, BK, and DT prepared and registered the protocol and will conduct the literature search and data extraction under the supervision of PD, AD, AA, and AB. MM and GG provided content expertise. All authors contributed to the concept and design of the protocol. All authors provided critical revision of the protocol and read and approved the final manuscript.
Consent for publication {#FPar4}
=======================
Not applicable
Competing interests {#FPar5}
===================
The authors declare that they have no competing interests.
Publisher's Note {#FPar6}
================
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
| {
"pile_set_name": "PubMed Central"
} |
The first author's name is incorrect in the citation. The first author in the citation should be Nordlund LM. The correct citation is: Nordlund LM, Koch EW, Barbier EB, Creed JC (2016) Seagrass Ecosystem Services and Their Variability across Genera and Geographical Regions. PLoS ONE 11(10): e0163091. doi:[10.1371/journal.pone.0163091](http://dx.doi.org/10.1371/journal.pone.0163091).
| {
"pile_set_name": "PubMed Central"
} |
Background {#Sec1}
==========
Comorbid depression in patients with type 2 diabetes (DM2) and/ or coronary heart disease (CHD) is a major health issue. The risk of depression in these patients is approximately double compared to the general population \[[@CR1], [@CR2]\]. This comorbidity is associated with diminished self-care and medication adherence \[[@CR3], [@CR4]\], poorer quality of life \[[@CR5]\], and increased mortality \[[@CR6], [@CR7]\]. Similar negative effects are seen with comorbid subthreshold depression \[[@CR8]\], defined as clinically relevant depressive symptoms without fulfilling the criteria for major depressive disorder (MDD). Subthreshold depression is present in approximately one third of the patients with DM2 and/or CHD \[[@CR9]--[@CR11]\] and is the strongest predictor for the onset of MDD \[[@CR12], [@CR13]\].
Despite its negative impact, depression often remains under-recognized, under-discussed and undertreated in the general population \[[@CR14]\]. The detection of comorbid depression in patients with long term physical conditions, like DM2 and CHD, is even more challenging as symptoms can overlap \[[@CR15], [@CR16]\]. Therefore, in clinical guidelines, various organizations have suggested screening for depression to improve detection rates \[[@CR15], [@CR17], [@CR18]\]. However, at present, there is no substantial evidence that this approach is effective \[[@CR19], [@CR20]\]. Reducing the burden of depression by preventing the influx of new cases is a promising strategy, particularly through early recognition and treatment of patients at risk (indicated prevention), such as those with subthreshold depression. Meta-analyses have shown that preventative psychological interventions can overall reduce the incidence of MDD in comparison to control groups \[[@CR12], [@CR21]\].
Offering preventative psychological interventions in a stepped-care format could be an efficient approach, which also fits well with current task shifting and delegating trends. In primary care in the Netherlands, most GPs work with psychological practice nurses (those who provide low-intensity mental health care) and somatic practice nurses (those who largely focus on general physical care), who are generally located in the same building. This internationally unique integrated primary care team aims to provide local community-based continuity of care \[[@CR22], [@CR23]\]. In stepped-care, patients start with minimally intensive evidence-based treatments and progress is monitored systematically. Those who do not improve adequately, step up to a treatment of higher intensity \[[@CR24]\]. Many guidelines endorse this stepped-care principle for depression treatment \[[@CR15], [@CR25], [@CR26]\], but the evidence on the effectiveness of prevention is limited and conflicting. While effective in reducing the incidence of MDD in elderly or visually impaired populations \[[@CR27]--[@CR29]\], it was not superior to usual care in other elderly, diabetic or primary care populations \[[@CR9], [@CR30]--[@CR32]\]. Recently, we conducted a randomized controlled trial in which we evaluated whether a pragmatic, nurse-led stepped-care program was effective in reducing the incidence of MDD at 12-months of follow-up in comparison with usual care among patients with DM2 and/or CHD and subthreshold depression (Step-Dep study) \[[@CR33]\]. The stepped-care approach was not superior to usual care after one year \[[@CR34]\]. Consecutively, qualitative research was conducted to gain a deeper understanding of these results. A process evaluation was conducted in which we explored both patients' and practice nurses' experiences with the Step-Dep program using the RE-AIM model which assesses five dimensions of an intervention: reach, efficacy, adoption, implementation, and maintenance \[[@CR35]\]. We focused on barriers and facilitators of the implementation of the Step-Dep program \[[@CR36]\] next to a more conceptual exploration of how DM2/CHD patients and practice nurses perceived comorbid depression.
More insight into patients' perceptions of depression in long-term conditions is of great value. Recent systematic reviews have suggested that the limited understanding we currently have contributes to many of the encountered difficulties in depression care \[[@CR16], [@CR37]\]. Differences between patients' and health care providers' perceptions further add to these difficulties \[[@CR38]\]. As most patients with depression and DM2 and/or CHD are managed in primary care in the Netherlands, knowledge of these patients' and their health care providers' perceptions of this comorbidity is important, but only a few studies investigating this have been conducted \[[@CR16], [@CR39]--[@CR44]\]. The most used theoretical framework in such studies on perceptions or illness representations is the Common Sense Self-Regulation Model of Health and Illness by Leventhal et al. \[[@CR45]\]. This framework states that patients construct their own perceptions of the causes and consequences of the illness, its time-course, the feasibility of controlling or curing it, how it affects one's identity and emotions, and how well the illness is understood. This helps patients to make sense of their illnesses and serves as the basis for coping. Previous studies have mainly focused either on the patient perspective and their experienced relationship between these disorders (the 'cause' item of the model) \[[@CR16], [@CR39]--[@CR42]\], or on health care providers' views of managing depression in these long-term conditions ('cure-control') \[[@CR43], [@CR44]\]. Studies exploring aspects like illness perceptions ('identity') and perceived need for care ('cure-control'), or comparing patients' and health care providers' perceptions are lacking. Yet, to improve patient engagement in future indicated prevention programs, knowledge on whether targeted patients perceive themselves as 'ill' and if they perceive a need for care, seems crucial. Moreover, better understanding of both caregivers' and patients' perspectives, as well as the differences between them, may enable prevention programs to be more tailored to these perceptions, potentially improving depression care \[[@CR46]\]. Therefore, this study aimed to investigate patients' and practice nurses' perceptions of depression in patients with DM2 and/or CHD screened for subthreshold depression.
Methods {#Sec2}
=======
Step-dep study {#Sec3}
--------------
This qualitative study was part of the Step-Dep study, which consisted of both a pragmatic cluster randomized controlled trial with economic evaluation, whose design \[[@CR33]\], results \[[@CR34]\] and process evaluation \[[@CR36]\] have been described elsewhere. The qualitative process evaluation consisted of semi-structured face-to-face interviews with 24 participants of the intervention arm. As an extension to the 'reach' dimension of the RE-AIM model \[[@CR35]\], which describes study participants' characteristics and compares them to the target population, patients' and practice nurses' perceptions of depression were thoroughly explored and reported in this article.
Participants and recruitment {#Sec4}
----------------------------
We interviewed all the practice nurses involved in the implementation of the Step-Dep intervention. Amongst them were both psychological practice nurses and somatic practice nurses. Psychological practice nurses provide low-intensity mental health care for primary care patients. Somatic practice nurses provide chronic disease management in the GP practice of patients with physical long-term conditions like DM2, CHD, chronic obstructive pulmonary disease (COPD) and asthma. In the Netherlands, the educational programs for these two types of practice nurses are separate and generally take one year after an appropriate pre-registration education of four years at a University of Applied Sciences.
All Step-Dep study participants had a diagnosis of DM2 and/or CHD, hence the term 'patient' used in this paper. In addition, these patients screened positive on subthreshold or mild depression, which was defined as a Patient Health Questionnaire 9 (PHQ-9; range 0--27) score of six or more \[[@CR47], [@CR48]\] without evidence of a major depressive disorder according to the Mini International Neuropsychiatric Interview (MINI) \[[@CR49], [@CR50]\]. We used purposive sampling \[[@CR51]\] to recruit a diverse sample of patients in order to elicit as many different views as possible on the pre-specified topics of the Common Sense Self-Regulation Model of Health and Illness model \[[@CR45]\]. Based on a literature review of factors influencing depression incidence and outcome, we selected patients on: gender, age, presence of DM2 and/or CHD, self-reported history of depression, self-reported current depression, level of education, baseline depression severity (PHQ-9), baseline anxiety severity (HADS-a), baseline quality of life score (EQ5D), baseline social support scores, locus of control scores. In addition, we selected patients from different urban and rural residential areas.
Both patients and practice nurses were asked by an investigator (AP or DO) to participate in the interviews by phone. All initially selected participants agreed to be interviewed, except for three patients, who were either suffering from a terminal illness or had a terminally ill partner. Three other patients were then asked and all agreed to participate.
Data collection {#Sec5}
---------------
The interview topic guide ([Appendix 1](#Sec20){ref-type="sec"}) was both based on the study aims of the process evaluation \[[@CR36]\], and included the assessment of patients' and practice nurses' perceptions of (subthreshold) depression in DM2/CHD. For the latter, open-ended questions were formulated ([Appendices 2](#Sec21){ref-type="sec"} and [3](#Sec22){ref-type="sec"}) that drew upon the Common Sense Self-Regulation Model of Health and Illness \[[@CR45]\]. We focused on illness perception ('identity'), need for care ('control-cure') and causes of depression and the interplay with their DM2/CHD ('cause'). These topics were considered to be most clinically relevant by the research-team, because of both patients' and practice nurses' interim feedback during the Step-Dep study, and the research questions that remained after its effectiveness analyses \[[@CR34]\].
Two researchers (AP and DO) conducted all the interviews from September to November 2015. After consent, all interviews were anonymized, digitally recorded, transcribed verbatim and entered into Atlas.ti 5.7.1 for analysis and data management. Interviews took place at venues preferred by participants; at home (patients *n* = 11), at the GP practice (practice nurses *n* = 8) or at the VU University Medical Center in Amsterdam (patients *n* = 4, practice nurses *n* = 1) and lasted about 45 min each. AP kept at a reflective journal to be of aid in later analyses. A member check was performed and all participants but one (a patient who could not be reached despite multiple attempts) confirmed the content of the summary sent by mail to be representative of the interview \[[@CR51]\]. Data saturation was reached after interviewing 11 patients. Four more patients were subsequently interviewed to confirm this \[[@CR51]\]. All nine participating practice nurses in the intervention arm were interviewed, with data saturation reached after the eighth interview.
Data analysis {#Sec6}
-------------
The process of data collection and analysis was iterative, as data analysis was concurrent with data collection to enable the incorporation and validation of relevant emerging themes into subsequent interviews. Elements from a responsive evaluation were used. This approach provides the opportunity to explore the multiple perspectives of involved stakeholders and to create a rich and multi-layered understanding of a phenomenon \[[@CR52]\]. An important notion in responsive evaluations is that stakeholders are involved in the study and that the perspective of patients is taken into account \[[@CR52]\]. Along the research process, data were subject to a inductive thematic analysis \[[@CR53], [@CR54]\] .
First, codes were attached to citations related to specific (sub)topics (open coding), leading to a set of descriptive topics per transcript. Then, all codes of all transcripts were compared and redefined, and clustered into themes and subthemes (axial coding) and, overarching themes were formulated (selective coding). Next, similarities and differences between cases were identified (cross case analysis of constant comparison \[[@CR55]\]). Patient and practice nurse transcripts were analyzed separately, but comparisons were made across data sets. Two researchers (AP and DO) analyzed the data individually, and relevant themes were agreed upon.
Results {#Sec7}
=======
Participants {#Sec8}
------------
Table [1](#Tab1){ref-type="table"} shows the patient and practice nurse characteristics as measured at baseline of the Step-Dep study. Of the 15 participating patients, eight were female. The average age was 62, ranging from 48 to 84 years. PHQ-9 scores at inclusion varied from seven to 16 and were 10.9 on average. 11 patients reported a history of depression and five patients a current depression. Of the nine practice nurses interviewed, six were psychological practice nurses, three were somatic practice nurses and one of the latter had been a psychological practice nurse before. The average number of treated Step-Dep patients per practice nurse was 11 and varied from 3 to 24. Additional data can be found in Table [1](#Tab1){ref-type="table"}.Table 1Patient (*n* = 15) and practice nurse (*n* = 9) characteristics at inclusion Step-Dep studyPatients Gender (n)Female8Male7 AgeRange48--84Mean62 Chronic disease (n)DM29CHD10DM2 and CHD4 Number of long-term conditionsRange1--9Mean3 Level of education (n)Low4Average5High6 History of depression (n)Yes11No4 Self-reported depression (n)Yes5No10 Depression severity PHQ-9 at inclusionRange7--16Mean10,9 Anxiety HADS-ARange2--15Mean8 Quality of life EQ5DRange0,39-0,92Mean0,72 Social supportRange34--55Mean45 Locus of controlRange5--21Mean14Practice nurses Gender (n)female7 Type (n)Psychological practice nurse6Somatic practice nurse3 Number of patients treated during Step-DepRange3--24Mean11 Years of relevant professional experience as health-care providerRange3--30Mean16,3Abbreviations: *CHD* = Coronary Heart Disease, *DM2* = Type 2 Diabetes Mellitus, *PHQ-9* = Patients Health Questionnaire 9 score (range 0--27, higher scores indicating more severe depression), *HADS-A* = Hospital Anxiety and Depression Scale (range 0--21, with higher scores indicating more severe anxiety), *EQ5D* = EuroQol-5D (range 0--1, with higher scores indicating higher quality of life), social support (range 0--48, higher scores indicating more perceived social support), locus of control (range 0--20, higher scores indicating a more external locus of control)
Main themes {#Sec9}
-----------
The results of this study are presented by three main themes: 1) illness perception, 2) need for care and 3) causes of depressive symptoms. As the focus of this study was on perceptions of mental health, this is implied in both the concept of (mental) illness perception and need for (mental health) care. An overview of the main findings per theme and the corresponding interview questions can be found in Table [2](#Tab2){ref-type="table"}. For each theme, the most illustrative quotes were selected. Per quote, the main interviewee characteristics are described; P is used for patients and N for practice nurses. Full (anonymized) interviewee details can be found in [Appendix 4](#Sec23){ref-type="sec"}.Table 2Overview of themes, questions and resultsThemesQuestionsResultsIllness perception (identity**)**Patient\
• How would you describe your mental state before starting Step-Dep?\
• If not depressed: please tell more about it?\
• If depressed: please tell more about it? Did it influence your life?\
PN\
• How did you view their mental state/ depressive symptoms?\
• Did patients recognize themselves in the depressed profile?• Patients' and PNs' perceptions of depressive symptom severity varied from not to severely depressed and were not always congruent with PHQ-9 scores at inclusion\
• Almost all patients considered themselves at least mildly to moderately depressed\
• PNs frequently perceived their patients as 'not depressed'\
• Patients sometimes needed time to talk about and reflect on their mood\
• Work experience perhaps influenced PNs' perceptions of patients' depressive symptoms\
• Many patients did not initially realize that the mental state they were in was a level of depression\
• Patients preferred using their own words to describe their mental state, some terms were not connected to mood.\
• Sleeping was frequently pointed out as the most burdensome symptomNeed for care (cure/control)Patient\
• Were you in need of care/ a preventive program to improve depressive symptoms?\
• How would it have been, if you had not received an invitation for Step-Dep?\
• What were your expectations/ hopes from the program?\
• What would your care of choice have been like? And to improve depressive symptoms?\
PN\
• Were the patients in need for care for depression? Other need for care? Why? Why not?• Most interviewed patients experienced a need for care and preferred psycho-educational advice and talking therapy\
• PNs frequently said that patients had minimal need for specific care and mostly needed attention\
• In patients, perceived symptom severity corresponded with perceived need for care, but did not necessarily match help-seeking behaviour\
• Barriers to seek care:\
○ Not realizing that mental state is a level of depression\
○ Experienced stigma of depression\
○ Unfamiliarity with mental health care\
○ Experienced barriers discussing mental problems with GPDepression causes (cause)Patient\
• Is there a relationship with your chronic disease? How?\
• What do you think caused your depressive symptoms?\
• How is your mental state now? If improved: what are the reasons for that?\
PN\
• How do you view the relationship with the chronic disease? What coping strategies do patients have with a chronic disease?\
• What are causes of depressive symptoms?\
• If the depressive symptoms improved in your patients; what was the reason?• Most patients and PNs appointed a mix of causes of depression\
• Most were related to negative life events and circumstances\
• Many PNs and patients perceived indirect links with long-term conditions via:\
○ physical limitation\
○ changed future perspectives\
○ difficulties with acceptance of diagnosis of a long-term condition
### Illness perception {#Sec10}
In general, patients and practice nurses perceived patients' depressive symptom severity prior to the start of Step-Dep as varying widely, ranging from 'not depressed' to 'severely depressed'. Patients' perceptions of their symptom severity did not necessarily correspond with their individual PHQ-9 scores at inclusion. It was more common for the PHQ-9 scores to be higher than the perceived symptom severity than the other way around, but both occurred. When asked whether patients recognized themselves in the 'subthreshold depression' profile they were screened on, three patients responded at first that they had not felt depressed at all. One of them explained that she only screened 'positive' (with a relatively high score of 11) on the PHQ-9 by scoring on physical symptoms, unrelated to her mood, caused by her multiple chronic diseases. However, during the interviews, the other two patients eventually explained that they had been somewhat down or 'sombre'. In Dutch primary care depression guidelines and patient information, 'sombre' is the most frequently used term to describe all severity levels of depression \[[@CR56]\]. It seemed that these two patients needed time to feel comfortable enough to open up and reflect on their mood, which could be problematic in short consultations in primary care."***"*** *It was not as if I was in a sombre mood when I decided to participate. \[...\] Well yes, that was when I was not feeling too happy..." (P2, female, CHD)*"
All the other interviewed patients did experience some level of depression and indicated that this had a significant impact on their daily lives. About half the patients thought their mood matched the subthreshold depressed profile well and confirmed that they had felt mildly to moderately depressed. Yet many others described themselves as fully depressed."*"I was feeling really miserable. Too often feeling sombre and too tired. A complete lack of energy, just a wreck. I had trouble sleeping and concentrating. I was just not happy. Not a fun person anymore, in my opinion. (laughs). There was no room for anything else. I think I was actually barely hanging on. Yes, I was certainly depressed." (P5, female, DM2)*"
Whereas almost all patients would have labelled themselves as at least mildly to moderately depressed, practice nurses frequently perceived their patients as 'not depressed'. Two very experienced psychological practice nurses, who treated more Step-Dep patients than other practice nurses, even reported that virtually none of their patients were depressed."*"But I did not consider them depressed. That is something you can sense, or taste almost. No." (N7, psychological PN)*"
One of these practice nurses questioned whether her perception of her patients' depressive symptoms was influenced by her working experience: *"As I am used to working with some more severe problems, I thought: 'Am I missing something here?'" (N2, psychological PN).* In contrast, a few practice nurses called their patients chronically (mildly) depressed and one somatic practice nurse thought all her patients had severe depressive symptoms.
In terms of acknowledging, labelling and naming symptoms as part of 'depression', patients initially had difficulty realizing that their mental state was actually a level of depression. Filling out the PHQ-9 questionnaire as part of the screening process of Step-Dep and reflecting on that, seemed to help patients to identify their negative mental state as a (subthreshold) depression."*"Looking back, I wouldn't have thought that I was that... how should I phrase that...sombre. That actually shocked me at times. To realize that I seemed quite negative. And I actually was negative back then." (P9, male, DM2 & CHD)*"
The following extract illustrates how some practice nurses experienced this in their patients as well."*"Due to that questionnaire, they would say: 'My gosh, all this time, I have been depressed without knowing.' The best example was this one patient who had a massive score and was like: 'My goodness, what is the matter with me?' Well, she had been feeling miserable, but had not connected the dots." (N3, psychological PN)*"
Many patients however, who did recognize their mental state as a mild to moderate depression, rejected the term 'sombre' to describe it. It felt like a stigma for some or just too exaggerated for others. The following quote illustrates this."*"Sombre would be exaggerating, but I sure wasn't cheerful. Not a happy lad and at the same time seeing a psychologist." (P11, male, DM2)*"
Almost all patients would spontaneously use other words to describe their low mood, even the patients that labelled themselves as 'fully depressed'."*"took a bad turn" (P10, male, DM2) "rough times" (P11, male, DM2) "continuous sorrow" (P13, male, CHD) "down and out" (P12, female, DM2 & CHD) "wrecked" (P12, female, DM2 & CHD)*"
Some of the terms they used, were not even necessarily connected to a depressed mood."*"loss of self-confidence" "stress" (P11, male, DM2) " burdensome worries" (P1, female, CHD) "burn-out" (P12, female, DM2 & CHD)*"
A striking number of patients reported troubled sleeping and described this as their most burdensome symptom. The following quotation illustrates a perceived link between trouble sleeping and depression."*"Trouble sleeping. You fall into a downwards spiral, you get so tired, chronically tired I would say. It makes it so easy to stay underneath the covers in the morning, drifting off to depression." (P15, female, CHD)*"
### Need for care {#Sec11}
Most interviewed patients indicated that they experienced a need for (mental health) care prior to the start of the Step-Dep study. In general, the perceived symptom severity matched the level of perceived need for care. Many practice nurses explained that the need for care varied considerably between patients, and usually corresponded with their perception of the patients' symptom severity. However, the general opinion amongst practice nurses was that the majority of patients had minimal need for specific care and mostly needed attention."*"The majority did not have a need for care, no. And those who did, were so depressed that they needed clinical treatment." (N7, psychological PN)*"
The majority of patients cited practical advice and someone to talk to as their preferred modes of care. Practical advice entailed ways to improve their mood, for example through physical exercise and activity planning, and often concerned handling sleeping problems."*"Just talking to someone, every other week, for half an hour or an hour. To get some practical advice of (name practice nurse) on how to cope with trouble sleeping for example. For her to say: 'Why don't you try this', that really works." (P15, female, CHD)*"
Many patients did not have a clear idea who the person 'to talk to' would be, but psychotherapists were most frequently mentioned."*"I did not know much about it, except for the term 'psychotherapy'." (P7, male, DM2)*"
Whereas patients mainly emphasized their need for practical psycho-educational advice and talking therapy, practice nurses reported that patients predominantly needed attention."*"I often reckoned that maybe they just needed some attention. Not to be negative or anything. Just to have somewhere and someone to talk to without sparing that someone, like they would have to with a partner or family member. The freedom to just talk. A need for attention." (N2, psychological PN)*"
Such mismatches in patients' and practice nurses' views on how much and which care is needed potentially jeopardizes patient engagement in offered care.
An interesting finding from the patients' interviews was that the perceived symptom severity and the corresponding perceived need for care did not necessarily match patients' own predictions of or actual help-seeking behaviour. While most patients experienced a need for care, many did not and would not have asked for it. Patients explained that they experienced barriers that withheld them from seeking care. These appointed barriers were often also perceived by practice nurses. Many of the barriers indicated dealt with the taboo and social stigma of depression. The following quotation illustrate a variety of these."*"Being a true 'Twent' (Dutch word for someone from the eastern province of the Netherlands) I never reveal what I am truly feeling." (P10, male, DM2)*""*"I never would have asked for that kind of help myself. Growing up, I was taught not to complain. Especially not about mental problems, because that is just all in your head and therefor something you should resolve on your own. \[...\] To overcome the idea of 'You used to be normal, yet now you have become a psychiatric patient' \[...\] The stigma already completely surrounds you." (P7, male, DM2)*""*"So many of them were of a certain age, when society used to say 'Take it like a man, stop complaining.' And so many would lead their lives according to these social codes, bearing their problems in silence." (N1, psychological PN)*"
In addition, as patients would often not realize that they were depressed, they were unaware that they could ask for help."*"I wonder if I would have looked for any help, since I was just so used to feeling like that. I just feel so much better now. It makes me think: 'Darn, things were definitely not alright back then.' But, it was normal for me." (P5, female, DM2)*""*"But in the end, there were quite a few who did have a need for care. But apparently, they had not acted upon it yet. It had not reached their frontal lobe yet, so to say. Not up to the point where they would say: 'I need to do something about this, I should make an appointment.'"(N3, psychological PN)*"
Patients also mentioned barriers that practice nurses did not. Some patients explained that they were unfamiliar with mental health care and its possibilities for help. Other patients mentioned difficulties talking to their GPs about mental problems. Patients felt that GPs mainly focused on physical disease, or they experienced a lack of time and space, or a lack of continuity of care to discuss mental issues with their GP."*"I guess because I was unfamiliar with that area of health care, I would not have looked for it." (P9, male, DM2 & CHD)*""*"In my experience, GP's are always short on time. That makes it really difficult to discuss that kind of problems. Because GPs, like mine, are so busy already and work part-time too, that you always see a different one, which I find very disturbing." (P1, female, CHD)*"
### Causes of depressive symptoms {#Sec12}
Both patients and practice nurses indicated various causes of depressive symptoms, both related and unrelated to long-term conditions, and most said that a mix of these leads to depression. The most frequently mentioned causes were serious life-events like divorce, bereavement or childhood traumas, as well as negative circumstances like job loss or work pressure. Other, less frequently named causes were personal traits like personality, genes or character, aging and loneliness."*"Those life events obviously had an impact on their quality of life and appealed to their coping mechanisms." (N1, psychological PN)*""*"It was caused by job insecurity, financial problems or by thyroid medication that needed adjusting. They would appoint very specific problems and say: 'The way I felt, was a reaction to those problems.' Circumstances, yes." (N2, psychological PN)*"
Many patients and practice nurses experienced indirect links between long-term conditions and depression. Physical limitations caused by DM2, CHD or other chronic diseases, along with their impact on daily life, were seen as the most prominent indirect causes of depression. Interviewees did not necessarily presume a 'linear' relation between the severity of these limitations and depressive symptoms. Changed prospects of the future due to a chronic disease formed another important indirect cause. Further, both patients and practice nurses explained how 'mourning' the diagnosis of a chronic illness could lead to depression, in which acceptance problems played a dominant role."*"I used to walk 20 to 25 km with a friend every other week. That used to be so easy for me, but I can't anymore. The fact that we had to turn around, that I couldn't finish that specific walk and had to take a short-cut back... That had a considerable impact. It did not cheer me up at all, to the contrary." (P6, male, CHD)*""*"But even in those people with severe limitations, it would not necessarily have that much of an impact. I am remembering this lady who was severely limited, but was so incredibly active. (laughs) In her case, it did not influence her mood, per se." (N4, psychological PN)*""*"I don't really feel those glucose levels. I know the diabetes is there and I realize its consequences, which is possibly the most frightening aspect for me. People say that it is a secret assassin, and that is true, actually." (P10, male, DM2)*""*"It is a kind of 'mourning' process that you have to go through, to reach a state of acceptance of your losses, like your energy levels, at work, things you used to be able to do. You have to learn to accept that you won't be able to do all of that anymore. Well, that was my biggest problem." (P7, male, DM2)*"
Very few patients and practice nurses directly linked DM2 and/ or CHD to depression."*"That (her and her husband's chronic diseases) absolutely has it effect on the things you want to do or the way you feel. I do believe that." (P12, female, DM2 & CHD)*""*"Well, I have seen how being chronically ill just leads to a depressed mood." (N8, somatic PN)*"
There were also some patients and practice nurses who believed that depression is not related to DM2 or CHD."*"Well, it didn't even cross my mind, that is how important it is to me. I have a hint of diabetes. (laughs) I just use one pill a day. For me, it is such a none-issue, that it hadn't even occurred to me." (P11, male, DM2)*"
Only one practice nurse reckoned that the diagnosis of a long-term condition itself could have an anti-depressant effect."*"I did not see that presumed relation, or hardly. It is very well possible that people adjust their lifestyle, and realize the impermanence of life...that it is a wake-up call and acts as an anti-depressant." (N7, psychological PN)*"
Discussion {#Sec13}
==========
This qualitative study explored patients' and practice nurses' perceptions of the construct of 'depression' in patients with DM2 and/or CHD screened for subthreshold depression. Our overall analysis is that better understanding of how chronically ill patients make sense of depressive symptoms or illnesses, in view of their need for care and in view of how they see the symptoms in the context of their lives (the 'causes') is crucial for the implementation of mental health care into chronic disease care. Perhaps practice nurses can also be better trained for this.
Illness perception {#Sec14}
------------------
In general, the interviewed patients considered themselves at least mildly depressed, whereas practice nurses, interestingly, frequently perceived patients as not depressed. This discrepancy is perhaps partially caused by the fact that psychological practice nurses are used to working with patients with quite severe depression and a clear request for help. Step-Dep patients, on the other hand, were pro-actively selected on the presence of subthreshold depression on a self-report questionnaire. Furthermore, previous research suggests that somatic practice nurses sometimes experience a lack of competence to adequately recognize and handle mental problems in chronically ill patients \[[@CR36]\]. However, this could also be part of a more widespread phenomenon, as it is has been described before that many caregivers have the tendency to 'normalize' depression in patients with long-term conditions \[[@CR43], [@CR44]\]. In addition, some patients initially did not recognize their mental state as a level of depression, which might prohibit them from disclosing their depressive symptoms to caregivers. This has been observed in other studies as well \[[@CR37], [@CR39]\]. These studies suggest that patients might 'refuse' to recognize and acknowledge their depression due to an inner conflict of their ideal self-identity and perceiving themselves as a person with depression (ego dystonia in Freudian terms). In this study, we have also observed the opposite as in some patients sombre feelings would be present for so long, that they accepted these as normal (egosyntonic) and therefore failed to recognize these as a level of depression.
Perceived depressive symptom severity was not always congruent with PHQ-9 scores at inclusion. Both over- and underestimation by the PHQ-9 of depression severity was perceived. Even though the PHQ-9 is a validated instrument to screen for mild depression in the chronically ill using a cut-off of 6 \[[@CR47], [@CR48]\], our findings could indicate that, in these specific long-term conditions, the discriminative properties of this method were not optimal. A recent study in a population of patients with DM2/CHD, found optimal cut-off scores for minor and major depression to be within a small range of 8 and 10 respectively \[[@CR57]\]. This suggests that the PHQ-9 might not be specific enough to distinguish minor from major depression for scores in this range. A higher cut-off score of 8 might be necessary in order not to over-diagnose mild depression in patients with DM2/CHD, as symptom of the somatic diseases and depression, like fatigue and altered appetite, can overlap. Also, there is an association between depressive symptoms and distress related to long-term conditions \[[@CR58]\], such as diabetes distress \[[@CR59], [@CR60]\]. A complex finding of our study was that even though patients explained they felt mildly to moderately depressed, they independently labelled their mental state differently than 'depression'. It seems likely that patients actually do suffer from depressive symptoms, but prefer using different labels like 'stress' or 'sleeping disorders', as they perceive these as less stigmatizing than 'depression'. However, since the specificity of the PHQ-9 with a cut-off of 6 was found to be only 55% \[[@CR57]\], our findings raise questions over whether it discriminates enough between mild depression and mild forms of other psychological problems, like anxiety, burn-out or sleeping disorders.
In this study, many patients expressed both the heavy burden of sleeping problems and the wish to alleviate it. Problems with sleeping are classic symptoms of depression, but the associations between disturbed sleep and depression \[[@CR61]\] or long-term conditions \[[@CR62]\], like CHD and DM2, have also been well established. While the underlying mechanisms of the relationships between these conditions and their implications for rational therapeutics should be further explored \[[@CR63]\], addressing sleeping problems seems a promising starting point for the delivery of mental health care for most patients with depressive symptoms.
Need for care {#Sec15}
-------------
Perceived need for care coincided with perceived symptom severity, but often did not match help-seeking behaviour. Although most patients experienced a need for care, preferring psycho-educational advice and talking therapy, many would not have sought such care if it had not been offered pro-actively. Patients blamed several experienced barriers. The perceived stigma of depression was the most important barrier, but the initial lack of awareness about depression and mental health care options, and perceived difficulties to discuss mental health issues with GPs, were also mentioned. In previous studies, experienced stigma and taboo of depression were found to form important barriers to both help-seeking and disclosure of depressive symptoms \[[@CR16], [@CR37], [@CR39], [@CR43]\]. Whereas the appointed barriers apparently withheld patients from actively seeking care, it did not seem to withhold them from accepting care by participating in the Step-Dep program. Pro-actively offering care therefore appears to be an appropriate approach to overcome such barriers. However, we cannot exclude the possibility that the motivation to contribute to research was actually pivotal in their decision to participate in the program. The process evaluation of Step-Dep revealed that all patients appointed the contribution to research as (one of the) primary motivators to participate. Only less than half named the need to improve their mood as a primary motivation \[[@CR36]\]. Yet, given the importance and magnitude of perceived stigma of depression, it seems more likely that naming the contribution to research instead of experienced depression as the main motivator felt less stigmatizing for some and the pro-active offer of care facilitated the acceptance of care.
Causes of depression {#Sec16}
--------------------
The interviewees in this study cited a mix of causes leading to depression. The perceived importance of the contribution of negative life events and circumstances to the development of depression is in line with findings from the review by Anderson et al. \[[@CR37]\]. A direct causal link between long-term conditions and depression was largely not supported in our study. This is in contrast with the views of the elderly interviewed by Bogner et al. \[[@CR42]\], who perceived that their long-term condition directly caused depression and vice versa. Patients in this and several other studies reported that long-term conditions can lead to depression, not the other way around. In these patients' views, long-term conditions caused depression indirectly, via the burden of physical limitations \[[@CR16], [@CR41]\], diminished future perspectives \[[@CR16]\] and difficulties accepting the long-term condition diagnosis. The latter was frequently explained in our study as part of the 'mourning' process. Both patients and caregivers frequently referred to terms like 'mourning' and 'acceptance' when describing the response to chronic illness \[[@CR64]\], which are originally derived from the Kubler-Ross' grief model \[[@CR65]\]. These outcomes, however, are not supported by multiple studies showing that the diagnosis of DM2 by screening does not have significant psychological impact \[[@CR66], [@CR67]\]. Still, tuning into patients' perceptions of the causes might facilitate the conversation on depression. In chronic disease care, starting points could therefore be the impact of the diagnosis, physical limitations, or the impairment of future perspectives. However, patients who do not perceive any link between their long-term condition and depression might not disclose depressive symptoms in integrated care settings, which may be a barrier for such care.
### Implications {#Sec17}
It seems to be of great importance to better inform caregivers in chronic care about the risk of normalising depression and the magnitude of stigma patients experience about depression. Pro-actively educating patients in chronic care on possible comorbid depression and how to handle such symptoms might further help to diminish experienced stigma and create more patient awareness of depression. This might further facilitate integrated somatic and mental health care, as patients would get more acquainted with the concept of chronic caregivers discussing mental health, which is something patients do not necessarily expect, potentially interfering with the success of care integration \[[@CR23]\]. Additionally, exploring in practice which terms and settings individual patients relate most to seem very relevant to improve the acceptance of mental health care. Addressing sleeping problems, for example, might be an easily accepted starting point for patients with (subthreshold) depression. More research on how to best identify mild depressive disorders in patients with DM2/CHD and what prompts patients to accept and seek care could contribute to the success of future depression prevention programs.
### Strengths and limitations {#Sec18}
An important strength of this paper is that both patients and practice nurses were interviewed. Deeper understanding was gained of the caregivers' and patients' views, which led to valuable complementary and contrasting data. The utilisation of two analysts, the systematic development of codes and code definitions, the use of a qualitative computer program, and complete data saturation while still conducting interviews, enhanced the quality of the data.
As the organization of care, for example concerning the role of practice nurses in primary care, might be different outside the Netherlands, these findings might be less applicable in other settings. Furthermore, the results are based on the perceptions of patients who participated in the Step-Dep study. It would be of much added value to interview screened patients who did not consent to participate in Step-Dep, as they might have different perceptions of the investigated themes. Additionally, since our interviewees were mainly native Dutch, cultural differences that may influence depression perceptions were not explored.
Conclusion {#Sec19}
==========
Data of the interviewed patients and practice nurses suggest that they have different perceptions about (subthreshold) depressive illness and the need for care, although views on its causes seem to overlap more.
Appendix 1 {#Sec20}
==========
Table 3Topic listRE-AIMTopicReachAppropriateness Step-Dep patients (target population)Depression: recognition, severity, causes, improving factorsNeed for careMotivation to participateAccess mental health careEfficacyPerceived effectivenessPerceived usefulnessAdoptionInformation practices, caregiversImplementationBarriers & facilitatorsDeviations from protocolReasons for dropoutPrerequisites for implementationMaintenanceSatisfactionFeasibility for future
Appendix 2 {#Sec21}
==========
Table 4Patients interviewTopicQuestionGeneralHow was your experience participating in Step-Dep/ the program in your general practitioner practice?What was the best part for you?What was the weakest part for you?MotivationWhy did you decide to participate in Step-Dep?Mental stateHow would you describe your mental state before starting Step-Dep?If not depressed: please tell more about it?If depressed: please tell more about it? Did it influence your life? What do you think caused it? Is there a relationship with your chronic disease? How? How is your mental state now? If improved: what are the reasons for that improvement?Did you feel the PHQ-9 reflected your mental state correctly? Why? Why not?Need for careWere you in need of care/ a preventive program to improve depressive symptoms?How would it have been, if you had not received an invitation for Step-Dep?What were your expectations/ hopes from the program?Did the program match your needs?What would your care of choice have been like? And to improve depressive symptoms?How would it have been for you to be offered a program at the time of diagnosis of your chronic disease?Perceived effectivenessWas the offered program useful to improve your depressive symptoms? Why? Why not? What was most useful to you? How do you see that in the long-term?How were/was the consultations with the practice nurse/ self-help/ problem solving treatment/ referral to general practitioner for you?Suggestions for future careWould you recommend this program to others? Why? Why not? To whom?What would your suggestions be to improve Step-Dep?Is there anything you would like to add to the interview?
Appendix 3 {#Sec22}
==========
Table 5Practice nurses interviewTopicQuestionGeneralHow did you experience executing Step-Dep?What is your opinion on the Step-Dep program?What were the main facilitators?What were the main barriers?ReachWere the selected patients appropriate for this prevention program? Why? Why not?How did you view their mental state/ depressive symptoms? Did patients recognize themselves in the depressed profile? What are causes for depressive symptoms? How do you view the relationship with the chronic disease? What coping strategies do patients have with a chronic disease?Were the patients in need for care for depression? Other need for care? Why? Why not?EfficacyDid the program match their need for care?Was Step-Dep effective in your opinion on preventing depression/ improving depressive symptoms for these patients? Why? Why not? How?What is your view on the program elements: consultations, self-help, problem solving treatment, referral to general practitioner?If the depressive symptoms improved in your patients; what was the reason for this improvement? Did the program play a part?ImplementationWhy did you decide to participate in Step-Dep?How do you view your competences to execute the program?Was it necessary to deviate from the protocol? Why? Why not?How was using the PHQ-9 for you? And as a screening/ monitoring/ decision tool?How much time would you need for the consultations/ self-help/ problem solving treatment?MaintenanceIs this program (or elements) useful in daily practice for this group? Why? Why not?Would you use this program (or elements) in the future? Why? Why not?What would be necessary to implement this in your practice?How would you ideally see depression prevention?What is your opinion on offering a program like that at the time of diagnosis of the chronic disease?
Appendix 4 {#Sec23}
==========
Table 6Patient and practice nurse characteristicsPatientsInterview nrAgeSexDM2/ CHDEducational levelSelf-reported depression at baseline\*Self-reported History of depressionPHQ-9 score at inclusion P166fCHDhighnoyes7 P261fCHDhighnoyes7 P363fBothintermediateyesyes9 P484fCHDlowyesno10 P553fDM2highnoyes16 P672mCHDintermediatenoyes10 P756mDM2highnoyes10 P873fBothlownono11 P955mBothintermediatenoyes14 P1048mDM2intermediateyesyes12 P1161mDM2lowyesyes8 P1256fBothhighyesyes14 P1366mCHDhighnoyes7 P1457mDM2intermediatenono14 P1555fCHDlownono15Practice nursesInterview nrPractice nurse typeNumber of Step-Dep patients treated N1psychological practice nurse24 N2psychological practice nurse15 N3psychological practice nurse13 N4psychological practice nurse10 N5somatic practice nurse3 N6somatic practice nurse6 N7psychological practice nurse15 N8currently somatic practice nurse, previously psychological practice nurse3 N9psychological practice nurse7Abbreviations: *F* female, *M* male, *CHD* Coronary Heart Disease, *DM2* Type 2 Diabetes Mellitus, *PHQ-9* Patients Health Questionnaire 9 score. \*Scores do not equal inclusion PHQ-9 scores due to time between inclusion and baseline
CHD
: Coronary heart disease
DM2
: Type 2 Diabetes Mellitus
GP
: General practitioner
PHQ-9
: Patients Health Questionnaire 9
PN
: Practice nurse
The authors would like to thank Lotte Bakker for the transcriptions of the interviews. We also would like to thank all the participating general practices and the research networks of general practitioners (ANH, THOON and LEON) for their participation and collaboration in the implementation and execution of the Step-Dep study. Furthermore, this study has been possible thanks to all interviewed participants. We would like to extend our gratitude to all Step-Dep participants.
Funding {#FPar1}
=======
This study is funded by ZonMw, the Netherlands Organisation for Health Research and Development (project number 80--82310--97-12110). The sponsor had no role in the design and conduct of the present study or in the writing of the manuscript.
Availability of data and materials {#FPar2}
==================================
The data (interview transcripts in Dutch) generated and analyzed during the current study are not publicly available due to participant privacy reasons.
AP constructed the design of the study and drafted the manuscript. AP and DO performed all interviews and analyses. KS collaborated in constructing the design, supervised the analyses and revised the manuscript. MA, MvT and HvM collaborated in constructing the design and revised the manuscript. The final manuscript was read and approved by all authors.
Ethics approval and consent to participate {#FPar3}
==========================================
The Step-Dep project included several studies, including an RCT registered in the Dutch Trial Register (registration number 3715), and was approved by the Medical Ethics Committee of the VU Medical Center (Step-Dep study, NL39261.029.12, registration number 2012/223). The study was performed in accordance with the declaration of Helsinki (2008) and the Dutch Medical Research involving Human Subjects Act (WMO). Written informed consent was obtained from all participating patients. Additional spoken informed consent for the interviews was given by each participant. Confidentiality was maintained using restricted, secure access to the data, destruction of audio tapes following transcription and de-identifying the transcripts.
Consent for publication {#FPar4}
=======================
Not applicable.
Competing interests {#FPar5}
===================
The authors declare that they have no competing interests.
Publisher's Note {#FPar6}
================
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
| {
"pile_set_name": "PubMed Central"
} |
Introduction {#s1}
============
The increasing incidence of skin cancer is a global trend. Skin cancer, which was previously known to be a common disease in Western countries, is occurring more frequently in South Korea. According to the Korean Statistical Information Service[^1^](#fn0001){ref-type="fn"}, the number of patients with non-melanoma skin cancer in 2015 was 4,804 (9.4 people per 100,000), an increase over the 1,960 in 2005 and 3,270 in 2010. The increase in incidence rate is thought to be due to the aging population, the increased popularity of outdoor activities, increased ultraviolet exposure, improved access to medical services, and increased awareness of skin cancer among patients ([@B1]).
Skin biopsy and histopathologic evaluation are essential in confirming skin cancer. However, it is impossible to confirm all pigmented lesions by biopsy due to pain and scar development. Therefore, it is first necessary to establish whether or not a biopsy is required through a visual inspection performed by an experienced dermatologist. Furthermore, dermatologist needs a device that can detect changes over time in skin lesions and record the lesions in detail so that wrong-site surgery does not occur ([@B2], [@B3]).
With the development of imaging technologies, methods and devices for recording and analyzing what doctors see have progressed rapidly. Universally, dermoscopic imaging irradiates light onto the upper dermal layer, to observe and record more detailed pigment changes. In recent years, development of high-resolution non-invasive diagnostic devices (e.g., confocal microscopy, multiphoton microscopy, etc.) that can detect cellular levels of the skin lesions without biopsy has also been enriched ([@B4]--[@B6]). In addition, diagnoses of such skin images using artificial intelligence (AI) have been shown to outperform the average diagnosis performances of doctors. These developments are expected to have a significant impact on the diagnosis of skin cancer, the accurate recording of changes in suspicious lesions, and the effectiveness of follow-up skin cancer surgery. For user convenience, applications suitable for general smartphones have become available; however, these are not sufficiently supported by scientific evidence.
In this review, we introduce the basic concepts and clinical applications of AI via a literature review and discuss how these can be implemented in the field of dermatological oncology.
Basic Concepts of Artificial Intelligence {#s2}
=========================================
AI is a field of computer science that solves problems by imitating human intelligence, these problems typically require the recognition of patterns in various data. Conventional machine learning refers to machine learning methods that do not involve deep learning; these methods extract features such as those relating to colors, textures, and edges. In conventional machine learning, precise engineering knowledge and extensive experience are required to design feature extractors capable of extracting suitable features. Using these features, conventional machine learning can derive various results and identify correlations.
Deep learning uses deep neural networks to learn features, which are obtained by designing simple but non-linear modules for each layer. Using deep neural networks, very complex functions can be learned. For example, in the field of computer vision, a deep neural network\'s first layer typically learns the presence of edges at particular orientations and locations within the image. Larger combinations of such edges are identified in the next layer. As the layers become deeper, they learn larger and more specific features ([@B7]).
[Figure 1](#F1){ref-type="fig"} shows the relationship between AI, machine learning, and deep learning. Deep learning falls within the category of machine learning, which falls within the category of AI. In this figure, the examples for conventional machine learning and deep learning are classifications of acral lentiginous melanoma (ALM) and benign nevus (BN) in dermoscopy images. Conventional machine learning extracts specific features from dermoscopy images; for example, the gray-level co-occurrence matrix (GLCM) is used to extract texture features ([@B8]). The conventional machine learning method then trains classifiers, using the extracted features to classify ALM and BN. However, deep learning methods learn by extracting various features through deep neural networks. The main difference between conventional machine learning and deep learning is that deep learning extracts various features per layer, without human intervention ([@B9]).
{#F1}
We divided the cutaneous oncology publications into those evaluating malignant skin cancers and non-melanoma skin cancers. In addition, each publication was divided into machine learning (excluding deep learning), deep learning, and hybrid methods (a combination of machine learning and deep learning) ([Figure 2](#F2){ref-type="fig"}).
{#F2}
In terms of machine learning methods, most publications use a feature extractor to extract a feature from an image, they then train the classifier model using these features (e.g., malignant melanoma (MM) vs. BN). Recently, deep convolution neural network (DCNN) have been implemented in many medical-imaging studies ([@B10]--[@B12]). DCNN use convolution operations to compensate for the problems that arise through neglecting the correlations and pixel localities of multi-layer perceptron (MLP). Thus, deep learning can be used to train a robust classifier model with a variety of data. [Figure 3](#F3){ref-type="fig"} shows an example of a DCNN for classifying ALMs and BNs in dermoscopic images. The DCNN feature extractor repeatedly applies convolution and max-pooling (to obtain the largest activation for each region) operations to the layer input. This process generates a feature map. The feature map is inputted to a classifier via global average pooling for each channel. The classifier finally determines probabilities for ALM and BN. The result is then compared with the actual label, and the parameters are updated via backpropagation. However, DCNN operations require highly powerful graphics processing units to manage the complex computations and large datasets involved. Although DCNN learning capacities can be limited by insufficient medical-image data, it is possible to fine-tune state-of-art deep learning models that show high performance in ImageNet large-scale visual recognition challenge (ILSVRC), making them suitable for medical purposes ([@B13]). In the hybrid method, an ensemble classifier is designed by combining a conventional machine learning method and a deep learning method. For example, after extracting the features of an image using a conventional machine learning method, these extracted features are used as inputs for a DCNN. Another example is that of training a support vector machine (SVM) using a feature map obtained through a DCNN ([@B14]). One publication showed that hybrid models outperform both deep learning and conventional machine learning models ([@B15]), another publication highlighted the limitations of deep learning and stated a need for hybrid models to overcome these limitations ([@B16]). Thus, these two methods can be used effectively to create more accurate models.
{#F3}
Every year, the number of articles describing AI implementations in the field of cutaneous oncology increases. By observing the trends of the discipline, it can be seen that studies using conventional machine learning have been decreasing in popularity since 2015 (five publications in 2015, three publications from 2016 to 2017, and one publication after 2018); however, the number of studies conducted using deep learning methods has increased significantly since 2015 (zero publications in 2015, seven publications from 2016 to 2017, and nine publications after 2018). These tendencies are a result of the increasing availability of big data and powerful GPUs. Since 2015, state-of-art deep learning models such as ResNet have also been studied \[ResNet competed for the first time in the 2015 ILSVRC ([@B17]); it surpassed the human error rate of 5%, achieving an error rate of 3.6%\].
Application of Artificial Intelligence in the Diagnosis of Malignant Skin Cancers {#s3}
=================================================================================
Melanoma
--------
A total of 18 publications were identified, six of these described the use of conventional machine learning, nine publications showed the use of deep learning, and two publications presented the use of hybrid models. Among the total 18 publications, 14 used dermoscopic images as the dataset, and the remainder used unspecified or clinical images; nine used more than 500 datasets, and the remainder used \<500 datasets. Moreover, in five of the publications, other skin lesion data such as seborrheic keratosis (SK) and basal cell carcinoma (BCC) were used alongside malignant melanomas and nevus. Seven publications presented the area under the curve (AUC) as a performance indicator of the model and the remainder presented accuracy (Acc), sensitivity (Sen), and specificity (Spe) ([Tables 1](#T1){ref-type="table"}--[3](#T3){ref-type="table"}).
######
Melanoma skin cancer publications using deep learning method.
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
**Publication** **End-point** **Results** **Method** **Dataset**
------------------------------- ------------------------------------------------------------------------------------------------------------- --------------------------------------------- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ----------------------------------------------------------------------------------------------------------------------------------------------
Li et al. ([@B18]) -- Lesion segmentation (task1)\ Task 1: JA : 0.710 (LIN)\ -- Task 1 and Task 3\ ISIC 2017 dataset (*n* = 2000)\
-- Lesion dermoscopic feature extraction (task2)\ Task 2 :AUC : 0.848 (LFN)\ Preprocessing:\ The dataset contains melanoma, SK and nevus
-- Lesion classification (task3) Task 3 :AUC : 0.912 (LIN with LICU) Center crop + Resize(320\*320)\
Data augmentation\
Task 1 used LIN\
Task 3 used LIN with LICU-- Task 2\
Pre-processing:\
Superpixel extraction\
Data augmentation\
Random sample\
Patch rotation\
Using LFN
Pour et al. ([@B19]) -- Lesion segmentation\ Lesion segmentation:\ Data Augmentation\ ISBI 2016 challenge dataset\
-- Lesion dermoscopic feature segmentation (streak and globule features) Sen : 0.91\ -- Lesion segmentation : Deeper model with 16 conv. layers, augmentation by flipping and cropping (7200 training images)\ The dataset contains a representative mix of images of both MM and BN\
Spe : 0.95\ -- Lesion dermoscopic feature segmentation:\ -- Lesion segmentation:\
Acc : 0.94\ Similar convolutional layers initialized with a pre-trained model from lesion segmentation phase. This architecture is followed by two parts, each contains two convolutional layers and four deconvolutional layers to predict masks for both streak and globule features. Train_Images (*n* = 900)\
JA : 0.83\ Test_Images (*n* = 379)\
DI : 0.89\ -- Lesion dermoscopic feature segmentation:\
Lesion dermoscopic feature segmentation:\ Train_Images (*n* = 807)\
Sen: 0.119\ Test_Images (*n* = 335)
Spe: 0.997\
Acc: 0.991\
JA: 0.60\
DI: 0.108
Yu et al. ([@B20]) Classification (ALM and BN) Group A\ -- Training\ Dermoscopic images\
-- CNN\ Data augmentation : 12 patches cropping, rotation, and flipping\ ALM (*n* = 350) and BN (*n* = 374)\
Sen: 92.57%\ CNN: 5-layer convolution network + FC\ -- Group A\
Spe: 75.39%\ -- Testing\ ALM (*n* = 175) and BN (*n* = 187)
Acc: 83.51%\ Cropping 12 patches per test image and when one or more images were predicted as containing melanoma, the corresponding test image was interpreted as containing melanoma
PPV: 77.14%\
NPV: 91.88%\
-- Expert\
Sen: 94.88%\
Spe: 68.72%\
Acc.: 81.08%\
PPV: 73.13%\
NPV: 93.71%
Nasr-Esfahani et al. ([@B21]) Classification (melanoma and nevus) Sen: 81%\ Pre-processing : Removal of noise and illumination artifacts\ Clinical images\
Spe: 80%\ CNN: 2-layer convolution network (20 feature maps and 50 feature maps) + FC Melanoma (*n* = 70) and nevus (*n* = 100)\
Acc: 81%\ Train_Images (80%)\
PPV: 75%\ Test_Images (20%)
NPV: 86%
Premaladha et al. ([@B22]) Classification (MM and BN)\ Sen. 94.83%\ Pre-processing : CLAHE and median filter\ Dermoscopy images\
Best model : DLNN Spe. 90.46%\ Segmentation : Normalized Otsu\'s segmentation (NOS)\ Train_Images (85%)\
Acc. 92.89% Classifier : Compared DLNN and hybrid Adaboost-SVM. Best model was DLNN Test_Images (15%)
Matsunaga et al. ([@B23]) Classification (melanoma, nevus and SK) -- ISBI 2017 challenge dataset\ Pre-processing : luminance and color balance of input images are normalized exploiting color constancy\ ISBI 2017 challenge dataset\
AUC : 0.958\ CNN :\ Train_Images : provided data (374 MM, 254 SK, 1372 nevi) + external data (409 MM, 66 SK, 969 nevi)\
-- ISBI 2016 challenge dataset\ Fine-tuned 50-layer ResNet MM classifier and SK classifier\ Test_Images (*n* = 150)
AUC : 0.874 Ensemble classifier made by merging two classifiers
Tschandl et al. ([@B24]) Classification (melanoma, BCC, dermatofibroma, melanocytic naevi, seborrheic keratoses and vascular lesion) -- CNN\ All images from the students\' training session were also used to retrain the last layer of the "GoogLeNet Inception v3" neural network, without any kind of test-set augmentation (4,000 epochs, learning rate 0.001, batch size 50). Dermoscopic images (*n* = 348)\
Sen: 90%\ Train_Images (*n* = 298):\
Spe: 71%\ melanoma (*n* = 62), BCC (*n* = 40), dermatofibroma (*n* = 7), melanocytic naevi (*n* = 129), SK (*n* = 38), and vascular lesion (*n* = 22)\
AUC: 91%\ Test_Images (30%):\
-- StudentsSen: 86%\ melanoma (*n* = 10), BCC (*n* = 10), dermatofibroma (*n* = 2), melanocytic naevi (*n* = 14), SK (*n* = 9) and vascular lesion (*n* = 5)
Spe: 79%\
AUC: 85%
Esteva et al. ([@B25]) Classification (757 diseases) -- 3-way classification\ -- Training Using Google Inception v3 CNN architecture pretrained on the ImageNet dataset (1.28 million images of over 1,000 generic object classes) and fine-tuned on their own dataset of 129,450 skin lesions comprising 2,032 different diseases. The 757 training classes were defined using a novel taxonomy of skin diseases and a partitioning algorithm that maps diseases into training classes.\ Dermoscopic and conventional images (*n* = 129,450)\
Dermatologist 1 Acc: 65.6%\ -- Testing Author developed an algorithm to partition diseases into fine-grained training classes (for example, amelanotic melanoma and acral lentiginous melanoma). During inference, the CNN outputs a probability distribution over these fine classes. To recover the probabilities for coarser-level classes of interest (for example, melanoma) they summed the probabilities of their descendants Train_Images (*n* = 127,463)\
Dermatologist 2 Acc: 66.0%\ Test_Images (*n* = 1,942)
CNN Acc: 69.4 ± 0.8%\
CNN\
-- partitioning algorithm Acc: 72.1 ± 0.9%\
-- 9-way classification\
Dermatologist 1 Acc: 53.3%\
Dermatologist 2 Acc: 55.0%\
CNN Acc: 48.9 ± 1.9%\
CNN\
-- partitioning algorithm Acc: 55.4 ± 1.7%
Lee et al. ([@B26]) Classification (ALM and BN) -- CNN\ Data augmentation : four center-overlapping patches\ Dermoscopy images\
Sen: 90.2%\ CNN :\ ALM (*n* = 500), BN (*n* = 500) and intermediate tumor (*n* = 72)\
Spe: 93.8%\ Fine-tuned 50-layer ResNet\ Train_Images (*n* = 872):\
AUC: 97.6%\ Made an ensemble model \[merging Model 2 (intermediate tumor in BN set) and Model 3 (intermediate tumor in ALM set)\] ALM (*n* = 400), BN (*n* = 400) and intermediate tumor (*n* = 72)\
-- Board-certified dermatologists\ Test_Images (*n* = 200):\
Sen: 87.0%\ ALM (*n* = 100), BN (*n* = 100)
Spe: 71.4%\
Acc: 79.2%
Cui et al. ([@B27]) Classification (melanoma and non-melanoma) -- CNN\ CNN:\ Dermoscopy images\
(best model: Inception V3)\ Fine-tuned CNNs (AlexNet, VGG16, VGG19, Inception V3) and compared CNNs (best model was Inception V3) deep learning dataset (*n* = 2,200):\
Acc: 93.70%\ melanoma (*n* = 564) and non-melanoma (*n* = 1,636)
Sen: 95.30%\
Spe: 92.10%
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
######
Melanoma skin cancer publications using conventional machine learning method.
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
**Publication** **End-point** **Results** **Method** **Dataset**
--------------------------- ----------------------------------------------------------------------------- ------------------------------------------------------------------------------------------------------------------- ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- --------------------------------------------------------------------------------------------------
Sabouri et al. ([@B28]) Classification (MM and BN) Sen: 89.28%\ Pre-processing:\ Unspecified images\
Spe: 100%\ Artifact removal (hair artifact removal)\ Train_Images (*n* = 370):\
(best model: cascade classifier) Cropping (512\*512) and\ MM (*n* = 175) and BN (*n* = 195)\
lesion segmentation (border segmentation)\ Test_Images (*n* = 42):\
Feature extraction:\ MM (*n* = 16) and BN (*n* = 26)
Color features: RGB and HSV\
Texture features: using GLCM\
Classifier: compared many models (KNN, MLP, Naïve Bayes, RF, SVM). Best model was cascade SVM Classifier (SVM \#1 using normalized HSV, SVM \#2 using a combination of color and texture features)
Kaur et al. ([@B29]) Pink lesion classification within MM or BN AUC: 0.879 (all features) Relative color thresholds\ Dermoscopic images.\
Segment of 3 shades of pink (light, dark and orange pink)\ Train_Images (*n* = 60):\
Quintile overlays\ Only MM containing visible pink areas within the lesion\
Feature extraction:\ Test_Images (*n* = 132):\
Blob features (5 per shade)\ MM (*n* = 54), benign dysplastic, and congenital nevi (*n* = 78)
Color features for each pink shade over entire lesion (15 per shade)\
Texture features derived from lesion histogram (24 per shade)\
Location features (6 per shade)\
Classifier : multivariate analysis using linear regression was performed using the Proc Logistic function in SAS
Shimizu et al. ([@B30]) 4-class classification (melanoma, nevus, BCC, SK) Best model: Layered model Acc: 0.904\ Border detection : The core of the algorithm was color thresholding, removal of artifacts such as microscope border and hair, and inclusion of bright area seen specifically in NoMSLs (BCC and SK)\ Dermoscopic images\
AUC: 0.864\ Feature extraction:\ Train_Images (not described)\
(AUC denotes the area of the receiver-operating characteristic (ROC) curve between %M and min (%N, %B, %S).) Color-related Features : calculating ten statistics for the intensity of six color channels (RGB, HSV)\ Test_Images (*n* = 964):\
Subregion-related features : describing geometrical distribution of the color.\ melanoma (*n* = 105), nevi (*n* = 692), and SK (*n* = 98), BCC (*n* = 69)
Texture-related features : by adopting GLCM\
Classifier: compared layered model and flat model.
Abedini et al. ([@B31]) Classification (MM and BN) Acc: 0.90\ One feature of the system enables the domain expert to improve previously built models.\ Dermoscopic images\
(accuracy continues to improve with some fluctuation before converging at approximately 90% after 150 responses.) Classifier with a stochastic gradient descent SVM and a feedback mechanism.\ Train_Image (*n* = 100)\
Eventually, as more feedback is provided (more training examples), the SVM accuracy improves. Test_Image (*n* = 5):\
melanoma (*n* = 3) and normal skin (*n* = 2)
Marchetti et al. ([@B32]) Classification (MM and BN)\ -- Greedy fusion\ Compared five methods of fusing all automated predictions from the 25 participating teams in the ISBI challenge into a single prediction (three machine learning methods and two non-learned approaches) ISBI 2016 challenge dataset\
Best model: greedy fusion Sen: 58%\ The dataset contains a representative mix of images of both MM (*n* = 248) and BN (*n* = 1,031)\
Spe: 92%\ Train_Images (*n* = 900)\
AUC: 86%\ Test_Images (*n* = 379)\
-- Average dermatologist\ Reader study images (*n* = 100):\
Sen: 82%\ MM (*n* = 50) and BN (*n* = 50)
Spe: 59%\
AUC: 71%
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
######
Melanoma skin cancer publications using hybrid method.
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
**Publication** **End-point** **Results** **Method** **Dataset**
-------------------------- ------------------------------------- ------------------------------ ----------------------------------------------------------------------------------------------------------------------------- -------------------------------------------------------------------------
Jafari et al. ([@B33]) Lesion segmentation Sen: 95.2%\ -- Training\ Clinical images (*n* = 126)\
Spe: 99%\ Pre-processing: edge-preserving smoothing\ MM (*n* = 66) and non-MN (*n* = 60)\
Acc: 98.7% Patch selection:\ Train_Images (75%)\
Lesion patch selection\ Test_Images (25%)
Border patch selection\
Normal skin patch selection\
CNN: local texture analysis + general structure analysis\
-- Testing\
Pre-processing: edge-preserving smoothing\
Patch selection:\
Global and local patch definition\
CNN: local texture analysis + general structure analysis\
Post-processing : selecting largest connected component, dilation and hole filling
Xie et al. ([@B34]) Classification (melanoma and nevus) -- Xanthous race dataset\ Pre-processing : hair removal (using partial differential equation)\ Dermoscopy images\
Sen: 95%\ Segmentation: using SGNN\ -- Xanthous race dataset (*n* = 240): MM (*n* = 80) and BN (*n* = 160)\
Spe: 93.75%\ Feature extraction:\ -- Caucasian race dataset (n = 360): MM (*n* = 120) and BN (*n* = 240)
Acc: 94.17%\ Region division on dermoscopy images\
-- Caucasian race dataset\ Description of color, texture and border features\
Sen: 83.33%\ Feature normalization and dimensionality reduction\
Spe: 95%\ Classifier: meta-ensemble model of multiple neural network ensembles\
Acc: 91.11% Ensemble 1: single-hidden-layer BP nets with same structures\
Ensemble 2: single-hidden-layer BP nets and fuzzy nets\
Ensemble 3: double-hidden-layer BP nets with different structures
Sabbaghi et al. ([@B35]) Classification (MM and BN) Sen: 95.5%\ Each RGB dermoscopy image from a training set is converted to BoF mode\ Dermoscopic images.\
Spe: 94.9%\ Then, the generated BoF (scale-invariant feature transform (SIFT) + color) are fed into the stack auto-encoder for training MM (*n* = 174) and BN (*n* = 640)\
Acc: 95%\ Train_Images (*n* = 570)\
(Deep auto-encoder with BoF) Test_Images (*n* = 244)
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
### Deep Learning
Among the deep learning algorithms discussed in the literature, five were fine-tuned using pre-trained models. The remainder were fully trained with new models. In four publications, preprocessing was performed prior to model training. In addition, two publications performed lesion segmentation and classification or segmentation of dermoscopic features. To measure the model performance, one publication (Tschandl, Kittler et al.) compared the results of final-year medical students with those of the model; two publications (Yang et al. and Lee et al.) used the results of dermatological experts as the comparison. From these, Lee et al. showed that experienced dermatologists and inexperienced dermatologists improved their decision making with the help of deep learning models. One publication (Premaladha and Ravichandran) compared the conventional machine learning method \'Hybrid Adaboost-SVM\' and a deep learning-based neural network on the same dataset; they showed that the deep learning-based neural network delivered superior performance. Moreover, one publication (Cui et al.) demonstrated that when more data was used, the results of deep learning outperformed conventional machine learning methods.
### Conventional Machine Learning
From the conventional machine learning publications, four of the five publications performed feature extraction and then created a classifier. Two of these publications used SVM for the classifier, one used multivariable linear regression, and one used a layered model. In three publications, artifact removal or lesion segmentation were performed prior to feature extraction. On the other hand, one publication (Marchetti, Codella et al.) presented a new model using a fusion method, developed by 25 teams participating in International Symposium on Biomedical Imaging (ISBI) 2016.
### Hybrid (Deep Learning + Machine Learning)
In the publications using hybrid methods, one publication (Jafari, Nasr-Esfahani et al.) preprocessed the input images, extracted the patches, and performed segmentation using a convolutional neural network (CNN). In one publication (Xie, Fan et al.), segmentation was performed after preprocessing, using a neural network called self-generating neural network (SGNN); they then presented an ensemble network by designing a feature extractor and classifier. Furthermore, in one publication (Sabbaghi et al.), a deep auto-encoder combined with bag of features (BoF) outperformed the model using a BoF or deep auto-encoder alone.
Non-melanoma Skin Cancer: BCC, Squamous Cell Carcinoma (SCC)
------------------------------------------------------------
We identified seven deep learning publications, three machine learning publications and three hybrid publications on non-melanoma skin cancer. Several publications discussed MM; however, all of them discussed BCC and three publications discussed SCC, thus we classified the publications into these categories. The results are organized in [Tables 4](#T4){ref-type="table"}--[6](#T6){ref-type="table"}.
######
Non-melanoma skin cancer publications using deep learning method.
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
**Publication** **End-point** **Results** **Method** **Dataset**
------------------------------- -------------------------------------------------------------------------------------------------------------------------- ------------------------------- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ ------------------------------------------------------------------------------------------------------------
Rezvantalab et al. ([@B36]) Compare ability of deep learning with the performance of highly trained dermatologists -- Melanoma AUC\ Pre-trained Inception v3,\ *n* = 10,015 dermoscopic images.\
82.26% (Dermatologist)\ Inception ResNet v2,\ Melanoma (1,113 samples)\
93.80% (DenseNet 201)\ ResNet 152,\ Melanocytic nevi (6,705 samples)\
94.40% (ResNet 152)\ DenseNet 201 BCC (514 samples)\
93.40% (Inception v3)\ Actinic keratosis and intraepithelial carcinoma (327 samples)\
93.20% (Inception ResNet v2)\ Benign keratoses (1,099 samples)\
-- Basal cell carcinoma AUC\ Dermatofibroma (115 samples)\
88.82% (Dermatologist)\ Vascular lesions (142 samples)\
99.30% (DenseNet 201)\ n_train = 70%\
99.10% (ResNet 152)\ n_val, n_test = 15%
98.60% (Inception v3)\
98.60% (Inception ResNet v2)
Zhang et al. ([@B37]) Automatically classify dermoscopic images for clinical decision support -- Dataset A\ GoogLeNet Inception v3\ *n* = 1,067 dermoscopic images\
Acc: 86.54%\ Pre-trained on over 1.28 million images and adjusted the final layer to input own datasets using transfer learning Dataset A\
-- Dataset B\ 418 (Nevus)\
Acc: 85.86% 291 (SK)\
132 (BCC)\
226 (Psoriasis)\
Dataset B\
132 (Nevus, SK, BCC, Psoriasis)\
n_train = 80%\
n_val, n_test = 10%
Vander Putten et al. ([@B38]) Demonstrate a sensitivity and specificity that could make neural networks a realistic tool for dermatologists Classification layer\ 1\. Segmentation (deep residual network)\ Two independent sources (BCC)\
53 layers\ 2. Classification (very deep residual network) Dermoscopic images1. "Skin Lesion Analysis Toward Melanoma Detection" competition released with ISBI 2016\
AUC 0.92, Sen 0.98, Spe 0.95\ 2. International Skin Imaging Collaboration (ISIC) Archive
98 layers\
AUC 0.89, Sen 0.98, Spe 0.94\
152 layers\
AUC 0.93, Sen 0.97, Spe 0.96
Mandache et al. ([@B39]) Propose exploiting FFOCT images AUC: 95.93%\ **Feature extractor**\ *n* = 40 FFOCT images\
Sen: 95.2%\ -- Convolutional blocks\ 10 (BCC)
Spe: 96.54% -- Dropout layer\
-- ReLU\
**Classifier**\
-- Fully connected layer\
-- Dropout layer
Zhang et al. ([@B40]) Machine learning algorithms need to be combined with sufficient clinical expertise in order to achieve an optimal result -- Dataset A\ Developed algorithm based on pre-trained GoogLeNet Inception v3\ *n* = 1,067 dermoscopic images\
Acc: 87.25%\ In order to facilitate decision-making and improve the accuracy algorithm, this summarized classification/diagnosis scenarios based on domain expert knowledge and semantically represented them in a hierarchical structure Dataset A\
-- Dataset B\ 418 (Nevus)\
Acc: 86.63% 291 (SK)\
132 (BCC)\
226 (Psoriasis)\
Dataset B\
132 (Nevus, SK, BCC, Psoriasis)\
n_train = 80%\
n_val, n_test = 10%
Yap et al. ([@B41]) A method which combines multiple imaging modalities together with patient metadata -- Melanoma\ Used pre-trained modified ResNet-50 architecture (to extract image features)\ *n* = 2,917 (metadata + macroscopic images + dermoscopic images)\
AUC: 86.1%\ Using a late fusion technique\ 1,127 (Nevus)\
-- Cancer\ Image feature vectors were concatenated together with the metadata feature vectors and sent through the embedding network 727 (All cutaneous melanomas except mucosal and ocular)\
AUC: 88.8% 647 (BCC)\
273 (SCC)\
143 (BKL)
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
######
Non-melanoma skin cancer publications using conventional machine learning method.
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
**Publication** **End-point** **Results** **Method** **Dataset**
---------------------------- -------------------------------------------------------------------------------------------------------------------------------------------------------- ------------- ------------------------------------------------------------------------------------------------------------------------------------------------------ ---------------------------------
Marvdashti et al. ([@B42]) Fully automated procedure to detect BCC in *ex-vivo* human skin from PS-OCT images AUC: 97.2%\ Extracting image features from the two complementary image contrasts offered by PS-OCT, intensity and phase retardation (PR) using machine learning\ *n* = 520 PS-OCT\
Sen: 95.4%\ Then, classify image features using SVM with linear and Gaussian kernels, KNN, and RF 260 (Healthy, 26 patients)\
Spe: 95.4% 260 (BCC, 26 patients)
Kharazmi et al. ([@B43]) Detection and segmentation of cutaneous vasculature from dermoscopy images and extracted vascular features are explored for skin cancer classification -- BCC\ Segment vascular structures by decomposing the image using ICA, k-means clustering\ *n* = 659 dermoscopy images\
AUC: 96.5%\ Then, a vessel mask is generated as a result of global thresholding\ 299 (BCC)\
-- Non-BCC\ Vascular features fed into an RF classifier (decision tree) 360 (Non-BCC)
AUC: 96.5%
Kefel et al. ([@B44]) Automatic method for detection of pink blush\ Manual\ Border detection by GAC and modified Otsu\'s threshold\ *n* = 2,266 dermoscopic images\
(common feature in BCC)\ AUC: 87.8%\ Classification:\ manually created borders\
Manually created borders vs.\ Automatic\ logistic regression by Proc Logit of SAS (smoothness, brightness) n_train = 354\
automatic created borders AUC: 87.7% n_test = 1,024\
GAC\
n_train = 888\
n_test = 1,024
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
######
Non-melanoma skin cancer publications using Hybrid method.
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
**Publication** **End-point** **Results** **Method** **Dataset**
------------------------- ------------------------------------------------------------------------- ------------------------------------------- ------------------------------------------------------------------------------------------------------------------------ -------------------------------------------------------------- ------------ ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ------------------------------------------
Annan et al. ([@B45]) Propose BCC detection method Proposed method AUC (Best model : VGG-16) 1\. Graph based skin surface segmentation\ *n* = 5,040 OCT images 1,875 (lesion or irregular structure)
2. Surface flattening\
3. Deep feature extraction (pre-trained AlexNet, GoogLeNet, VGG-16, VGG-19)\
BCC classification (PCA, SVM)
**ConvNet** **Ori** **Dimension of PCA feature**
**100** **200** **500** **1,000**
AlexNet 0.916 0.897 0.915 0.897 0.917
GoogLeNet 0.744 0.744 0.744 0.744 0.744
VGG-16 0.935 0.858 0.913 0.928 0.931
VGG-19 0.891 0.798 0.824 0.863 0.894
Sarkar et al. ([@B46]) Novel state of the art deep neural network for skin carcinoma detection -- BCC\ Pre-processing:\ n = 700 dermoscopic images\
AUC: 97.9%\ Denoising of images by Gaussian blurring\ 300 (BCC positive)\
Spe: 97.5%\ Enhancement of images by CLAHE algorithm and use parallel deep residual network (RMSprop optimizer) for classification 100 (augmented set of SCC positive)\
Sen: 98.3%\ 300 (benign skin lesion)\
Precision: 96.7%\ n_train = 560\
F1 score: 97.5%\ n_val = 140
-- Benign\
AUC: 97.9%\
Spe: 98.6%\
Sen: 96.6%\
Precision: 98.3%\
F1 score: 97.5%
Dorj et al. ([@B47]) Focus on the task of the classifying skin cancer **Cancer** **AUC, %** **Sen, %** **Spe, %** Pre-trained AlexNet is used to extract training features and the obtained convolutional neural network features are classified into four groups using error-correcting output codes (ECOC), SVM *n* = 3,753 collected from the internet\
Actinic Keratoses\
-- 712 (Train), 185 (Test)\
BCC\
-- 728 (Train), 193 (Test)\
SCC\
-- 777 (Train), 200 (Test)\
Melanoma\
768 (Train), 190 (Test)
Actinic Keratoses 92.3 98.9 91.67
Basal cell carcinoma 91.8 97.7 86.73
Squamous cell carcinoma 95.1 96.9 94.17
Melanoma 94.2 97.83 90.74
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
The results of all publications were presented using an accuracy indicator, and some of these publications using a variety of indicators, such as specificity, sensitivity, precision, and F1 score. The datasets used in each publication were different, making it impossible to compare them directly.
### Deep Learning
Rezvantalab et al. compared the abilities of deep learning against the performances of highly trained dermatologists. This publication presented outcomes from various deep learning models. In the BCC classification, the highest AUC of the publication was reported as 99.3%, using DenseNet 201. When compared against dermatologists (AUC 88.82%), the results of deep learning were found superior.
Five publications used datasets of dermoscopic images. One used full-field optical coherence tomography (FFOCT) images, and Jordan Yap et al. used different forms of data including metadata, macroscopic images, and dermoscopic images. Next, they trained a deep learning model using fusion techniques, in which image feature vectors were concatenated with the metadata feature vectors. Two publications by Zhang et al. written in 2017 and 2018, showed interesting results; the 2018 publication improved the previous year\'s algorithm for utilizing medical information. Their results showed an average improvement of 0.7% over those of the previous year.
### Conventional Machine Learning
We identified four publications that used only machine learning techniques. Three publications used dermoscopic images and one used polarization-sensitive optical coherence tomography (PS-OCT) images. Each author used different methods and features.
Marvdashti et al. performed feature extraction and classification using multiple machine learning methods \[SVM, k-nearest neighbor (KNN)\]. Kharazmi et al. segmented vascular structures using independent component analysis (ICA) and k-means clustering, then classified them using a random forest classifier. Kefel et al. introduced automatically generated borders using geodesic active contour (GAC) and Otsu\'s threshold for the detection of pink blush features, known as a common feature of BCCs. Subsequently, they classified using logistic regression, based on features such as smoothness and brightness.
### Hybrid (Deep Learning + Machine Learning)
Three publications implementing hybrids were identified. Each publication used a different dataset. One publication used optical coherence tomography (OCT) images and another used dermoscopic images. Unusually, the third publication used data downloaded from the Internet, not directly taken.
Annan Li et al. used deep learning for feature extraction, then classified images using the principal component analysis (PCA) and SVM machine learning techniques. They compared deep learning models and assessed the differences in dimensions of the PCA features. Sarkar et al. applied Gaussian blurring to denoise the images and then used the contrast-limited adaptive histogram equalization (CLAHE) algorithm to enhance them. Unlike previous publications, deep learning was used for classification.
Implementation in Smartphones {#s4}
=============================
With the spread of smartphones, the mobile application market has expanded rapidly. Applications can be used in various fields, particularly in the field of dermatology through the use of smartphone cameras. In particular, due to the ubiquity of smartphones, easily accessible mobile apps can make it more efficient to detect and monitor skin cancers during the early stages of development. In addition, with the recent development of smartphone processors and cameras, machine learning techniques can be applied, and skin cancer diagnoses can be conducted through smartphones. [Table 7](#T7){ref-type="table"} shows that a lot of research and development on smartphone implementation is carried out. AI technology relevant to skin cancer diagnosis is anticipated to eventually be implemented in smartphones, enabling the reduction of unnecessary hospital visits. Many types of mobile health application are already available.
######
Smartphone applications.
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
**Application name** **Algorithm** **Evidence** **Performance** **References**
------------------------ ------------------------------------------------------------------------------------------- ---------------------------------------------------------------------------------- ----------------------- ----------------------- ------------------------
DermaCompare (removed) Machine learning Not found Not found \[1\]
Lubax (removed) Content-based image retrieval (compare), KNN (classification) One peer-reviewed supporting publication Sensitivity (95% CI)\ Specificity (95% CI)\ ([@B48]) \[2\]
90% (86--94) 92% (85--95)
MskinDoctor (removed) Grab cut algorithm (segmentation), SVM (classification) Not found Not found \[3\]
MySkinMap (removed) Machine learning Not found Not found \[4\]
SkinScan Image processing technique, ABCDE rule Not found Not found \[5\]
SkinVision Conditional generative adversarial neural network (segmentation) and SVM (classification) Two peer-reviewed supporting publications, evaluated in independent publications Sensitivity (95% CI)\ Specificity (95% CI)\ ([@B49]--[@B52]) \[6\]
iOS: 50% (22--78)\ iOS: 50% (22--78)\
Android: 72%(58--87) Android: 27% (1--56)
SpotMole Image processing techniques, ABCDE rule Evaluated in independent publications Sensitivity (95% CI)\ Specificity (95% CI)\ ([@B52]) \[7\]
43% (28--58) 80% (60--100)
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
*\[1\]AppAdvice. Derma Compare by Emerald Medical Applications. Available online at: <https://appadvice.com/app/derma-compare/982517772>*.
*\[2\]AppAdvice. Lubax - Skin Lesion ID Using Image Recognition by Lubax, Inc. Available online at: <https://appadvice.com/app/lubax-skin-lesion-id-using-image-recognition/956423382>*.
*\[3\]AppBrain. mSkin Doctor Mobile Application for Skin Cancer Detection by Aleem Technologies. Available online at: <https://www.appbrain.com/app/mskin-doctor/com.maleemtaufiq.mSkinDoctor>*.
*\[4\]AppAdvice. MySkinMap by Xyrupt Technologies, LLC. Available online at: <https://appadvice.com/app/myskinmap/1151655127>*.
*\[5\]AppAdvice. SkinScan by TeleSkin ApS. Available online at: <https://appadvice.com/app/skinscan/1025190936>*.
*\[6\]SkinVision. Available online at: <https://www.skinvision.com/>*.
*\[7\]Google Play. SpotMole. Available online at: <https://play.google.com/store/apps/details?id=com.spotmole&hl=nl>*.
Types and Accuracies of Diagnostic Applications Using a Smartphone
------------------------------------------------------------------
According to a recent review ([@B53], [@B54]), numerous applications have already been released, seven of which use image analysis algorithms. Four of the seven applications are not supported by scientific evidence, and these four have been deleted from the app store since the review was conducted; the other three apps are still available. [Table 7](#T7){ref-type="table"} provides a summary of the apps. SkinScan, SkinVision, and SpotMole are currently available. SkinVision uses machine learning algorithms and SkinScan and SpotMole use the ABCDE rule (that is asymmetry, border irregularity, color that is not uniform, diameter \>6 mm, and evolving size, shape or color). Only one application employs a machine learning technique. The sensitivity and specificity of these applications are shown in the table.
Most diagnosis applications are not accurate ([@B55]). Furthermore, only a few inform users using image analysis and machine learning. Most apps are not supported by scientific evidence and require further research.
Problems and Possible Solutions
-------------------------------
Inaccuracies in medical applications can result in problems of legal liability. In addition, the transmission of patient information may correspond to telemedicine practices, for which there are certain legal restrictions; these include information protection regulations to prevent third parties accessing data during the transmission process. Even if the accuracy is improved, the advertisements embedded in the application suggest that the technology could be used for commercial advertisements; for example, to attract patients.
To solve this problem, a supervisory institution in which doctors participate is required, along with a connection to remote medical care services. The United States has been steadily attempting to promote telemedicine in its early stages, to address the issue of access to healthcare. Since the establishment of the American Telemedicine Association (ATA)---a telemedicine research institute---in 1993, legislation, including the Federal Telemedicine Act, has been established. It has been applied to more than 50 detailed medical subjects, including heart diseases, and has been successfully implemented in rural areas, prisons, homes, and schools ([@B56]).
To obtain good results, it is necessary to focus on securing high-quality data, to form a consensus between the patient and the doctor, and to actively participate in development.
In summary, the evidence for the diagnostic accuracy of smartphone applications is still lacking because few mHealth apps offer services. In addition, because the rate of service or algorithm change is faster than the peer-review publishing process, it is difficult to compare different apps accurately.
Risks of Smartphone Applications
--------------------------------
Smartphone applications pose some risk to users, especially if the algorithm returns negative results and delays the detection and treatment of undiagnosed skin cancer. It is very difficult to study false-negative rates because there is no histological evidence. Users may not be able to assess all skin lesions, especially if they are located in areas difficult to reach or to see. Given the generally low specificity of current applications, there would be a few false positives. This would put unnecessary stress on the user and result in unnecessary visits to the dermatologist. Furthermore, through limited trust and awareness, the user may not follow the advice provided by the smartphone application.
Chao et al. described the ethical and privacy issues of smartphone applications ([@B57]). Whilst applications have the potential to improve the provisions of medical services, there are important ethical concerns regarding patient confidentiality, informed consent, transparency in data ownership, and protection of data privacy. Many apps require users to agree to their data policies; however, the methods in which patient data are externally mined, used, and shared are often not transparent. Therefore, if a patient\'s data are stored on a cloud server or released to a third party for data analysis, assessing liability in the event of a breach of personal information is a challenge. In addition, it is unclear how the responsibilities for medical malpractice will be determined if the patient is injured as a result of inaccurate information.
Conclusion {#s5}
==========
In this review, we analyzed a total of 35 publications. Studies on skin lesions were divided into those assessing malignant melanomas and non-melanoma skin cancers. In addition, studies involving clinical data and OCT images were used alongside those involving the dermoscopic images widely used in dermatology. Because the considered datasets differed between the publications, it was impossible to determine how best to perform the analysis. However, as seen in the publication by Cui et al. deep learning methods obtain better results than conventional machine learning methods if the dataset is large. Also, certain publications have reported comparable or superior results to dermatologist. In particular, recent publications have reported that dermatologists have improved diagnostic accuracy with the help of deep learning ([@B26], [@B58]). Therefore, in the future, computer-aided diagnostics in dermatology will show greater reliance on deep learning methods.
For the convenience of users, the use of a smartphone is necessary. However, an accuracy limitation occurs when applied to smartphones. This problem is due to the limitations of hardware, which used conventional machine learning techniques such as SVM rather than deep learning. However, MobileNet has recently made it possible to use deep learning methods in IoT devices, including smartphones ([@B59]). This enables deep learning to be applied to IoT devices for faster performances than large networks, which will lead to more active research into skin lesion detection using applications.
Application inaccuracies can lead to legal problems. To solve this problem, doctors and patients must participate together in the development stage, and an institution for managing and supervising this process is also required.
Author Contributions {#s6}
====================
BO and SY: contributed conception and design of the study, wrote sections of the manuscript, and contributed to manuscript revision. YC and HA: collected data and wrote the first draft of the manuscript. All authors read and approved the submitted version.
Conflict of Interest {#s7}
====================
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
^1^<http://kosis.kr/statHtml/statHtml.do?orgId=117&tblId=DT_117N_A00025&conn_path=I2>
**Funding.** This research was supported in part by the Global Frontier Program, through the Global Frontier Hybrid Interface Materials (GFHIM) of the National Research Foundation of Korea (NRF) funded by the Ministry of Science, ICT & Future Planning (2013M3A6B1078872), in part by the National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIT) (No. 2019R1F1A1058971) and in part by Leaders in Industry-University Cooperation + Project, supported by the Ministry of Education and National Research Foundation of Korea.
ACC
: accuracy
AI
: artificial intelligence
ALM
: acral lentiginous melanoma
AUC
: area under the curve
BCC
: basal cell carcinoma
BN
: benign nevus
BoF
: bag of features
CLAHE
: contrast-limited adaptive histogram equalization
CNN
: convolutional neural network
DCNN
: deep convolution neural network
DI
: dice score
DLNN
: deep learning-based neural network
FFOCT
: full-field optical coherence tomography
GAC
: geodesic active contour
GLCM
: gray-level co-occurrence matrix
ICA
: independent component analysis
ILSVRC
: ImageNet large-scale visual recognition challenge
ISBI
: International Symposium on Biomedical Imaging
ISIC
: International Skin Imaging Collaboration
JA
: Jaccard index
KNN
: k-nearest neighbor
LFN
: lesion feature network
LICU
: lesion index calculation unit
LIN
: lesion indexing network
MLP
: multi-layer perceptron
MM
: malignant melanoma
NPV
: negative predictive values
OCT
: optical coherence tomography
PCA
: principal component analysis
PPV
: positive predictive values
PS-OCT
: polarization-sensitive optical coherence tomography
RF
: random forest
SCC
: squamous cell carcinoma
SGNN
: self-generating neural network
SK
: seborrheic keratosis
Sen
: sensitivity
Spe
: specificity
SVM
: support vector machine.
[^1]: Edited by: Ivan V. Litvinov, McGill University, Canada
[^2]: Reviewed by: Pietro Rubegni, University of Siena, Italy; Serena Lembo, University of Salerno, Italy
[^3]: This article was submitted to Dermatology, a section of the journal Frontiers in Medicine
| {
"pile_set_name": "PubMed Central"
} |
Background
==========
Renal cell carcinoma (RCC) is the most common malignancy of the kidneys, with clear cell (ccRCC) subtype identified in 85% of these cases; one-third of these patients experience synchronous metastatic disease and 20--30% of the remaining patients have metachronous metastatic RCC \[[@b1-amjcaserep-19-1354],[@b2-amjcaserep-19-1354]\]. Tibia is a very rare site of RCC metastasis: Zekri et al. \[[@b3-amjcaserep-19-1354]\] in a recent study of 103 patients with advanced RCC and metastatic bone disease reported that the pelvis and ribs were involved in 48% of the patients, followed by the spine in 42%, followed by the long bones and skull. Fottner et al. \[[@b4-amjcaserep-19-1354]\] described only 3 cases of tibial metastasis in 101 cases (2.97%) with RCC and found that patients with solitary types, age \<65 years old, absence of pathologic fracture, and tumor-free resection margins had a better survival rate compared to patients with multiple metastases. Patients with solitary bone metastasis from RCC have the best prognosis, with a 5-year survival rate between 35% and 60% \[[@b5-amjcaserep-19-1354]\]. Due to the longer survival of these patients, some authors recommend a surgical approach aiming at curative rather than palliative outcome and implant stabilization to prevent local disease progression \[[@b6-amjcaserep-19-1354],[@b7-amjcaserep-19-1354]\]. Here, we present a rare case of solitary tibial metachronous metastasis from RCC in a 54-year-old male that appeared 8 years after nephrectomy without any evidence of disease until then. Segmental skeletal resection, intercalary allograft over locked reamed intramedullary nailing, and soleus flap coverage led to a successful clinical and radiological outcome without any evidence of recurrence for 5 years postoperatively.
Case Report
===========
A 54-year-old man presented in October 2013 to the Department of Orthopedic Oncology with a palpable mass at the middle of his left tibia. He had noticed it a month ago but did not complain of any difficulty weight-bearing or walking, nor was there nocturnal pain or systemic illness. The mass was painful on palpation without any signs of local inflammation. He had undergone a right nephrectomy 8 years ago due to clear cell RCC, but he remained asymptomatic without any evidence of recurrence according to his most recent computed tomography (CT) screening (brain, chest, abdomen) performed 1 year ago. Plain anteroposterior and lateral radiographs of the tibia revealed a small lucent lesion at the midshaft ([Figure 1A, 1B](#f1-amjcaserep-19-1354){ref-type="fig"}). A complete diagnostic imaging workup was done, including computed tomography (CT) of the tibia and staging protocol (brain-chest-abdomen), 3-phase bone scintigraphy, and magnetic resonance imaging (MRI) ([Figure 1C--1E](#f1-amjcaserep-19-1354){ref-type="fig"}). No other possible metastatic lesions were identified. The CT and MRI scans demonstrated a medullary osteolytic lesion of the middle tibia, measuring 1.5×1 cm in size, breaching the nearby anterior tibial cortex and involving the soft tissues of the anterior compartment. There was homogeneous enhancement with the use of paramagnetic material. Moderate recruitment on the tibia lesion was noticed on the bone scan but the rest of the skeleton was normal. Open biopsy of the lesion showed clear cell carcinoma, morphologically compatible with metastasis of the renal carcinoma. After detailed discussion of all available therapeutic options, the patient consented to biological surgical treatment, including wide resection, intercalary allograft over nailing, and soleus flap interposition. The operation took place 3 weeks after the biopsy under general anesthesia ([Figures 2](#f2-amjcaserep-19-1354){ref-type="fig"}, [3](#f3-amjcaserep-19-1354){ref-type="fig"}). Tibial traction was applied through a Steinmann pin placed at the calcaneus. A longitudinal oval incision incorporating the previous biopsy scar was performed. The tumor was recognized and a 7-cm bone segment was resected with transverse osteotomy using an oscillating saw; the surrounding soft-tissue mass was also excised to obtain healthy margins. With the tibia under traction, a guide wire was inserted through the patellar tendon and its central position was confirmed with C-arm fluoroscopy in both antero-posterior and lateral views. Both the proximal and distal parts of the tibia were reamed to a diameter of 12 mm and a flexible GK nail (Grosse & Kempf^®^ Locking Nail System -- Stryker, Kalamazoo, Michigan, USA), 11 mm in diameter and 345 mm in length was inserted. A matched femoral allograft that had been previously thawed in antibiotic solution was incorporated into the tibia. The allograft was trimmed at both edges with a burr to achieve a more cylindrical shape corresponding to the tibial cortex in both sides. Distal interlocking was performed first, and with a backslap stroke the intercalary allograft was further compressed. The nail was finally locked proximally with another 2 screws. The soft-tissue defect was covered with medial soleus flap and skin grafting from the ipsilateral femur. The duration of the operation was 2 hours and there was no significant blood loss or need of transfusion. The histological examination of the entire tibial specimen confirmed the presence of metastatic ccRCC being resected on "clear margins". After oncological consultation, no adjuvant chemotherapy or radiotherapy was proposed to the patient. Partial weight bearing was initiated on the 2^nd^ postoperative day, with instructions to the patient to increase weight bearing progressively and attain full weight bearing at 6 weeks. The patient was followed up regularly, having no clinical complaints and showed progressive healing of the allograft-host junction, especially in the proximal part; the distal part showed delayed union but the patient had no problems during activities of daily living. Thirty months later (March 2016), he experienced sudden pain at the distal tibia and inability to bear weight; radiological examination revealed hardware failure with breakage of both the nail and distal screws due to nonunion at the distal part of the allograft-host junction ([Figure 4A](#f4-amjcaserep-19-1354){ref-type="fig"}). The patient underwent reamed exchanged nailing using a GK nail (12-mm diameter and 330-mm length) with distal interlocking only, fibular osteotomy, and application of iliac bone cancellous autograft ([Figure 4B](#f4-amjcaserep-19-1354){ref-type="fig"}).
No major complications where noticed during the postoperative period and he was allowed to do full weight-bearing thereafter. At his last follow up 2 years later and 5 years after the first operation, the patient was free of tumor disease and showed solid union ([Figure 4C](#f4-amjcaserep-19-1354){ref-type="fig"}), unrestricted mobilization, no leg-length discrepancy, and a Revised Musculoskeletal Tumor Society Rating Scale of 27/30. This scale was introduced in 1993 by Enneking et al. \[[@b8-amjcaserep-19-1354]\] (*<https://faoj.files.wordpress.com/2009/03/fosstab1.pdf>*) and assigns numerical values (0--5) for each of 6 categories: pain, and function and emotional acceptance in upper and lower extremities; supports and walking and gait in the lower extremity; and hand positioning, and dexterity and lifting ability in the upper extremity.
Discussion
==========
Limb-salvage procedures without compromising fundamental oncological principles have become the rule rather than the exception in patients with solitary metastatic bone tumors; early diagnosis, advanced imaging modalities, refined surgical reconstructions, and multidisciplinary approaches have contributed to a significant increase of the long-term survival of these patients, who now demonstrate survival rates up to 80% \[[@b9-amjcaserep-19-1354]\].
Metastasis in RCC occurs most commonly to the lungs, followed by bone involvement in 20--35%, lymph nodes, liver, adrenal glands, and brain. In metastatic disease, the median survival rate of patients is about 8 months, with 50% mortality rate within the first year, while the 5-year survival rate is only 10% \[[@b10-amjcaserep-19-1354]\]. Skeletal involvement is usually an aggressive, lytic process which causes substantial morbidity through skeletal related events (SREs: pain, impending fracture, spine cord compression, hypercalcemia, and pathological fracture). The tibia and the diaphysis of long bones in general are a very rare site of involvement \[[@b3-amjcaserep-19-1354],[@b4-amjcaserep-19-1354]\]. In a recent case report (2012) of synchronous metastatic tibial diaphysis fracture in the presence of bilateral renal cancer with liver deposits, a 75-year-old male patient was treated with prophylactic intramedullary nailing \[[@b11-amjcaserep-19-1354]\]. In a more recent (2016) case report \[[@b12-amjcaserep-19-1354]\] of concurrent tibial and ankle metastasis in a 67-year-old male who presented 1 year after radical nephrectomy and was treated with above knee amputation, the authors mentioned 23 similar cases in their literature review. Laitinen et al. \[[@b13-amjcaserep-19-1354]\] reported the survival and complication rates of skeletal prosthetic reconstruction in 206/253 patients with metastatic RCC and performed this kind of treatment in only 2 tibial diaphyseal cases (1.3%). Our patient had a small lucent lesion in the tibial diaphysis that presented 8 years after nephrectomy. In general, prognostic factors for a good clinical outcome include young age, solitary metastasis, no pathologic fracture, tumor-free resection margins, and long interval from initial RCC appearance, which were all met in our case \[[@b4-amjcaserep-19-1354]--[@b6-amjcaserep-19-1354],[@b13-amjcaserep-19-1354]\].
There are several options to achieve reconstruction and stabilization of segmental intercalary diaphyseal defects: (1) allografts \[[@b14-amjcaserep-19-1354]\], (2) free or pedicled vascularized fibula grafts \[[@b15-amjcaserep-19-1354]\], (3) combined allograft and vascularized fibula \[[@b16-amjcaserep-19-1354]\], (4) extracorporeal devitalized autograft \[[@b17-amjcaserep-19-1354]\], (5) distraction osteogenesis \[[@b18-amjcaserep-19-1354]\], and (6) segmental intercalary endoprosthesis \[[@b19-amjcaserep-19-1354],[@b20-amjcaserep-19-1354]\].
The goal of any type of reconstruction is to achieve local control of the disease while maintaining limb function. Our decision to apply biological reconstruction with intercalary allograft was mainly based on tumor location (tibial diaphysis), size (small, inside the medullary canal with limited soft-tissue compromise) and type (solitary RCC), the young age of the patient, the long interval of metastatic emergence (8 years), and the absence of metastatic disease in other organs. Wide excision (7 cm) was performed in accordance with the study by Fortner et al. \[[@b4-amjcaserep-19-1354]\], which found a better Kaplan-Meier survival curve in patients with a tumor-free resection margin. For the same reason, we did not use preoperative embolization; the latter provides tumor devascularization, controls hemorrhage, reduces intra-operative blood loss, and facilitates curettage, but if wide resection is planned, there is no indication because it would lead to marked hypervascularity in the area surrounding the tumor, which would result in heavy bleeding during surgery \[[@b21-amjcaserep-19-1354]\].
Segmental endoprosthesis is another non-biological alternative for intercalary reconstructions that offers early weight bearing, rapid rehabilitation, and immediate stability. Nevertheless, healing is ignored and a significant risk of infection, periprosthetic fracture, aseptic loosening, and mechanical wear has been reported \[[@b13-amjcaserep-19-1354],[@b19-amjcaserep-19-1354],[@b20-amjcaserep-19-1354]\]. The 10-year survival of segmental endoprostheses ranges from 63% to 80%, with a reported 17--33% rate of implant failure \[[@b20-amjcaserep-19-1354],[@b22-amjcaserep-19-1354]\]; therefore, in our opinion that method should be applied in elderly patients with poor healing capacity, patients with metastatic bone disease, or those with a very short life expectancy, in whom instant weight bearing and full function are more important than construct maintenance. The use of allograft reconstruction in oncological surgery was first reported 50 years ago and has been popular ever since, especially with the establishment of organized tissue banks and the minimization of justifiable concerns regarding immunogenicity, antigenicity, and potential disease transmission. Their main advantages are preservation of bone stock, biological graft incorporation, adequate attachment of salvaged soft tissues, and initial mechanical strength \[[@b23-amjcaserep-19-1354]\]. Five-year allograft survival rate is around 80% \[[@b14-amjcaserep-19-1354],[@b23-amjcaserep-19-1354],[@b24-amjcaserep-19-1354]\], but up to 70% of patients will require additional surgical procedures due to the common "triad" of complications -- infection, fracture, and nonunion -- that usually tend to occur within 3 years of the index procedure, as in our case (nonunion), with the construct becoming much more stable if it survives this crucial period of time. Nonunion rate varies from 8% to 44% (higher for diaphyseal junctions); fractures occur in 15--19%, and infection occurs in 11.5--16%, most commonly within the first year \[[@b14-amjcaserep-19-1354],[@b23-amjcaserep-19-1354]--[@b25-amjcaserep-19-1354]\]. Allografts unite with the host bone through external callus formation, which is directed to the surface of the allograft. As the allograft is only partially incorporated to the host, a stable fixation either with compression plating, intramedullary nailing, or both is of fundamental importance. Plate fixation allows for more controlled compression of the host osteotomy site but carries a higher risk for fracture due to screw penetration through the allograft, while intramedullary fixation is less invasive but can induce distraction at the host-allograft junction \[[@b26-amjcaserep-19-1354]\]. In an already compromised healing environment, a residual gap may lead to delayed union or nonunion, as in our case. Allograft fixation with IM nailing has been considered a negative factor for allograft union in comparison to plate fixation \[[@b27-amjcaserep-19-1354]\]. However, other studies \[[@b28-amjcaserep-19-1354]\] found no statistically significant difference between plate and nail fixation for host-allograft union. The use of "compressive nails" that allow internal compression of the junction site seems to promote healing, with a reported union rate of 87% \[[@b29-amjcaserep-19-1354]\]. The use of a larger IM nail with dynamic distal interlocking at the revision operation of our patient attained healing of the distal osteotomy site.
Apart from mechanical stability, biological enhancement of the allograft-host junction is another important factor to promote healing and can be achieved with the addition of cancellous bone autograft, bone morphogenic proteins, autologous bone marrow aspiration, bisphosphonate treatment, and muscular flaps, as in our case. Tumor resection in our case resulted in a soft-tissue deficit; the use of a soleus flap filled up the void and also covered the allograft, thus protecting it from exposure and infection. Muscle has been also found to promote fracture healing, not just because of the increased blood flow \[[@b30-amjcaserep-19-1354]\] but also due to the migration of muscle-derived stroma cells to the osteotomy site and their subsequent differentiation to osteoblastic cells \[[@b31-amjcaserep-19-1354]\].
Conclusions
===========
Our patient was an ideal candidate for biological reconstruction as he presented with all favorable prognostic factors for aggressive surgical treatment. Nonunion or delayed union, which is a common complication of allografting, can be successfully treated with exchange nailing, leading to a good outcome.
**Conflict of interest**
None.
{#f1-amjcaserep-19-1354}
{#f2-amjcaserep-19-1354}
{#f3-amjcaserep-19-1354}
{#f4-amjcaserep-19-1354}
[^1]: Authors' Contribution:
[^2]: Study Design
[^3]: Data Collection
[^4]: Statistical Analysis
[^5]: Data Interpretation
[^6]: Manuscript Preparation
[^7]: Literature Search
[^8]: Funds Collection
[^9]: **Conflict of interest:** None declared
| {
"pile_set_name": "PubMed Central"
} |
INTRODUCTION
============
The resting-state network (RSN) is a term referring to functional networks of brain regions that are active when an individual is not focused on the outside world. The RSN is characterized by coherent oscillations at a rate lower than 0.1 Hz among brain regions. Generally, the default mode network (DMN) comprises a set of brain regions including the cingulate cortex, hippocampus, and medial prefrontal and inferior parietal cortices and includes members of several RSNs ([@B1], [@B2]). The DMN is active during sleep ([@B3], [@B4]) and under anesthesia ([@B5]), as well as when at rest ([@B6]), but is not activated during passive viewing with a natural scene ([@B7]). The DMN can be measured using various neurophysiological or functional neuroimaging tools such as electroencephalography (EEG) ([@B8]), magnetoencephalography (MEG) ([@B9], [@B10]), functional magnetic resonance imaging (fMRI), and functional near infrared (fNIR) ([@B11]). The activity of the DMN detected with functional neuroimaging is similar to that of the electric or optic signal measured directly from the brain cortex ([@B12], [@B13]). While performing a task, the DMN shows a reciprocal relationship with the task-related active network ([@B14]). This means that the activity of the DMN may be decreased while performing a task and increased during the resting state. A previous study reported that the more extensive the cognitive task is, the more active the subsequent resting state of the DMN becomes ([@B15]). Thus, the RSN, including the DMN, might form part of the backbone of brain behavior.
It is understood that the DMN plays a role in cognitive processes such as self-reflection, moral reasoning, recollection, or imagining the future ([@B6]). The DMN may also be associated with some psychopathologies, (e.g., self-disturbance in schizophrenia) ([@B16]), and inattentiveness in attention deficit hyperactivity disorder (ADHD) ([@B8], [@B17]). In previous neuroimaging studies, deficits in the DMN have been reported in various neuropsychiatric disorders ([@B18]), including schizophrenia ([@B10], [@B14]), autism ([@B19]), ADHD ([@B20]-[@B23]), Alzheimer\'s-type dementia, and mild cognitive impairment ([@B24]-[@B26]). From a clinical perspective, a question of interest is whether a deficit in the DMN can be restored or modulated. Prior studies have reported that deficits in the DMN can be modulated with antipsychotic medication in schizophrenia ([@B27]) and with psychostimulants in ADHD ([@B28], [@B29]). These results suggest that the characteristics of the DMN may provide a measure of treatment effectiveness as well as confirmation of diagnosis. To measure the treatment effect, however, the characteristics of the DMN need to be consistent across time. Previous studies using the fMRI data of healthy individuals have reported the test-retest reproducibility of the spatial characteristics of the RSN including the DMN ([@B30], [@B31]). However, the reproducibility of the functional connectivity of the RSN, including the DMN, has not yet been established using fMRI data, even though the functional connectivity among resting-state-related independent components (RSICs) has been used to explore clinical significance ([@B32]).
Usually, the DMN is constructed using a correlation map between the time course of a seeding point or region of interest (ROI) and those of all other voxels within the brain ([@B33], [@B34]). This seed-based correlation analysis using anatomical ROIs could suffer from the confounding factor of individual anatomical variability ([@B35]), and could also show only relationships of one-to-multiple ROIs and not relationships among all ROIs. An alternative ways of constructing the DMN is to extract raw time-course data from the ROI selected on the spatial map of the independent component (IC) ([@B32]). However, the raw time courses might also be affected by various forms of noise, even though the seeding point was chosen based on the IC map. In the present study, the time courses of RSICs, rather than raw time courses, were used for measuring functional connectivity. Our approach might be less affected by either individual anatomical variability, as with a seed-based approach, or by noise, as with an ICA approach using raw time courses. Also, all connections among RSICs could be explored with our approach using the time courses of RSICs, unlike the seed-based correlation analysis. In this resting-state fMRI (rs-fMRI) study, we addressed the consistency of functional connectivity among the time courses of RSICs and also confirmed the spatial characteristics of RSICs in healthy individuals.
MATERIALS AND METHODS
=====================
Subjects
--------
Enrollment for subjects was advertised on the message board at Eulji University Hospital, Daejeon, Korea. Thirteen healthy young male subjects (Age = 22.4 ± 2.5 years) underwent fMRI. Subjects were interviewed by a psychiatrist using DSM-IV criteria to exclude any possible psychiatric disorders. All subjects were right-handed and had no history of psychiatric or neurological disorders or head trauma. This study was approved by the institutional review board of the Eulji University Hospital. All subjects provided written informed consent before participation in the present study.
Functional MRI Experiments
--------------------------
Subjects underwent fMRI scan of 5 minutes and 30 seconds both on day one and after 4 weeks at rest with their eyes closed. Using 3.0 Tesla MRI (ISOL, Gwangju, Korea), fMRI was acquired with a BOLD-sensitive echo-planar gradient-echo (EPI) sequence with the following imaging parameters: repetition time (TR), 2000 ms; echo time (TE), 35 ms; flip angle (FA), 70°; spatial resolution, 3.4375 × 3.4375 × 5 mm^3^, imaging matrix, 64 × 64; field-of-view (FOV), 220 × 220 mm^2^, number of slices, 25, 165 volumes. Subjects were instructed to keep their eyes closed and not to think of anything particular during fMR scans.
Data Preprocessing
------------------
The fMRI data were processed and analyzed using the MELODIC (Multivariate Exploratory Linear Optimized Decomposition into Independent Components) module of the FSL software (<http://www.fmrib.ox.ac.uk/fsl/feat5/index>). The first five scans were discarded to account for T1 saturation effects. The remaining 160 images were spatially realigned using a rigid-body transformation and were subjected to slice-timing correction. Next, a brain mask from the first fMR data volume was created to eliminate signals outside of the brain in each subject. Spatial smoothing using a full-width Gaussian kernel at half-maximum, 6 mm was performed to reduce noise without reducing valid activation. The smoothed images were filtered with a 128-second high pass filter to effectively remove signal drift. The serial correlations were removed to validate the statistics and be maximally efficient. Finally, the functional images were registered to the Montreal Neurological Institute (MNI) T1 template using FLIRT (FMRIB\'s Linear Image Registration Tool) for FSL.
Spatial Consistency of RSIC Maps Across Time
--------------------------------------------
To test the spatial consistency, multisession temporal concatenation ([@B36]) was run on all 13 participants as a group at each time point where a standard (space × time) ICA decomposition was conducted. Both the principle-component analysis (PCA) and the subsequent ICA algorithm produced 54 ICs at week 0 and 63 ICs at week 4. Each IC map was divided by the standard deviation of the residual noise and was plotted onto a histogram of intensity values ([@B37]). MELODIC was used to carry out inference on the estimated maps using a Gaussian/gamma mixture model and an alternative hypothesis testing approach at a threshold level of 0.5. A threshold level of 0.5 in the case of alternative hypothesis testing means that a voxel survives thresholding as soon as the probability of it being in the active class exceeds the probability of it being in the background-noise class ([@B37]). The spatial cross-correlation analyses among 3402 IC pairs (54 × 63), between two time points, were performed with thresholded IC maps using FSL. This means that 63 correlation coefficients were produced for each of the 54 ICs on day 1. The IC pair showing the highest correlation coefficient among the 63 possible ICs was considered as the same IC between two time points and their correlation coefficients used for a value of spatial consistency. Among the 63 ICs, 41 were not RSICs were signals from white matter, cerebrospinal fluid, vessels, and motion artifacts. The remaining 22 pairs were selected as the RSIC and were ordered with their correlation coefficients ([Fig. 1](#F1){ref-type="fig"}). The spatial correlation coefficients were compared between RSICs and non-RSICs using a *p* value \< 0.05 as a significance threshold. While this approach provided the degree of spatial consistency of each IC, it did not provide its statistical difference between two time points.
Another method using a post-hoc regression analysis was employed to test the question about whether spatial maps of RSICs are consistent across time. Unlike the first ICA method performed for each time point individually, this multisession temporal concatenation ([@B37]) was run on both time points of all the 13 participant pairs as a group. Both the PCA and subsequent ICA algorithm yielded 48 spatial IC maps. Each IC map was thresholded by fitting a mixture model using the same method outlined above ([@B37]). Among 48 ICs, 26 that were not RSICs were discarded as signals from white matter, cerebrospinal fluid, vessels, and motion artifacts. The remaining 22 RSICs were selected for a pairwise group comparison ([Fig. 2](#F2){ref-type="fig"}). For the pairwise group comparison, to identify differences in spatial characteristics of RSICs according to time point, we specified a temporal-design matrix, a subject-design matrix, and corresponding contrast matrices. Finally, a post-hoc regression analysis was performed on estimated time courses and subject modes at a statistical threshold level of *p* \< 0.05.
The Consistency of Functional Connectivity Among Time Courses of RSICs Across Time
----------------------------------------------------------------------------------
To extract the RSIC time courses, we used the 22 ICs selected in the prior step for spatial consistency using the Gaussian/gamma model. To identify subject-specific temporal dynamics within each individual\'s fMRI dataset, the full set of group-ICA spatial maps was used in a linear model fit against the separate fMRI data sets ([@B38]). This process produced time courses for each component and subject. Time courses of the 22 RSICs of each subject were selected for further functional connectivity analysis. Here, both cross-correlation (CC) and partial-correlation (PC) coefficients were used as measures of functional connectivity. These correlation analyses produced CC and PC matrices of 22 by 22 in each subject. The CC referred to the covariance between time courses of members in each IC pair. The PC referred to the correlation of time courses between two ICs after accounting for the relationship of each time course to the other 20 reference time courses.
To test the consistency of functional connectivity across time, two approaches were applied. First, using correlation coefficients for all IC pairs, the intraclass correlation coefficient (ICC) was calculated between the two time points. The correlation coefficient of each IC pair was averaged across all subjects. Then, 462 pairs excluding 22 self-pairs in the group-averaged coefficient matrix (22 by 22) were used to calculate the ICC. The Psych statistical package \"R\" (<http://www.personality-project.org/r/html/ICC.html>), was used to calculate the ICC between time points identified as \"at day 1\" and \"after 4 weeks.\" The second approach was the comparison of the correlation coefficient of each IC pair between two time points. Fisher\'s r-to-z transformation was performed on the correlation coefficients. The z-transformed mean correlation coefficient of each IC pair was compared between \"at day 1\" and \"after 4 weeks\" using a paired *t* test at a statistical threshold level of *p* \< 0.05. Corrections for multiple comparisons and false-discovery rate (FDR) were applied to the *p* values resulting from the paired *t* test.
RESULTS
=======
Spatial Consistency
-------------------
As a result of spatial cross correlation using FSL, 22 IC pairs showing the highest correlation coefficient between two time points were identified as RSICs. The spatial correlation coefficients among 22 RSICs ranged from 0.20 to 0.74 (0.53 ± 0.13) across RSICs ([Fig. 1](#F1){ref-type="fig"}). The most spatially consistent RSIC was in the posterior cingulate cortex (PCC)/precuneus region known as a posterior DMN ([@B39]). Relatively high spatial consistency was also found in the bilateral inferior occipital gyri (0.69), bilateral calcarine gyri (posterior medial visual component, 0.69), and bilateral precentral/postcentral gyri (0.68). RSICs showing relatively low spatial consistency were found in the basal ganglia (0.20), a parietal region of the left frontoparietal network (0.36), and the supplementary motor area (SMA) (0.38). The remaining 26 non-RSICs including physiological noise, non-gray matter signals, and motion artifacts showed significantly lower correlation coefficients (0.38 ± 0.17, *t* test, t = 3.58, df = 51.65, *p* = 0.001) than did the RSICs. The spatial correlation coefficients and the map of each RSIC are presented in the [Figure 1](#F1){ref-type="fig"}.
In the post-hoc regression analysis with a mixed Gaussian/gamma model, the spatial characteristics of each RSIC pair were not significantly different between day 1 and after 4 weeks (z = 0.10-1.21, *p* = 0.10-0.90) ([Table 1](#T1){ref-type="table"}).
The Consistency of Functional Connectivity: ICC Across Time Interval
--------------------------------------------------------------------
Intraclass correlation analysis showed that the correlation coefficients among all pairs of IC time courses were significantly consistent across the time interval. The ICC of CC was 0.78 (F = 7.93, *p* \< 0.001, CI = 0.74-0.81), while the ICC of PC was 0.75 (F = 7.05, *p* \< 0.001, CI= 0.71-0.79).
The Consistency of Functional Connectivity: Comparison of Mean Correlation Coefficients Across a Time Interval ([Fig. 2](#F2){ref-type="fig"})
----------------------------------------------------------------------------------------------------------------------------------------------
CCs between several IC pairs were significantly different across a time interval. The CC was significantly decreased between IC3 (upper part of the precuneus) and IC4 (superior medial frontal gyrus: anterior DMN; t = 2.08, *p* = 0.05), IC9 (bilateral middle temporal gyri; t = 2.73, *p* = 0.01) or IC19 (bilateral middle temporal/lateral middle occipital gyri: V5; t = 2.13, *p* = 0.04). A decreased CC was also found for the IC12 (right angular and supramarginal gyri/intraparietal lobule) - IC19 pair (t = -2.20, *p* = 0.04). CC also significantly increased between IC17 (bilateral inferior occipital gyri) and IC6 (bilateral frontal poles/middle frontal gyri; t = -2.15, *p* = 0.04), IC7 (left inferior frontal gyrus; t = -2.23, *p* = 0.04), IC8 (bilateral supramarginal gyri; t =-2.48, *p* = 0.02), and IC12 (t = -2.29, *p* = 0.03). Increased CCs were also found for pairs IC9-IC21 (left precentral gyrus) (t = 2.49, *p* = 0.01) and IC15 (bilateral calarine gyri: anterior medial visual component) - IC20 (bilateral precentral/postcentral gyri) (t = 2.15, *p* = 0.04). In terms of PCs, the functional connectivity was significantly decreased in the IC12 - IC19 pair (t = -2.13, *p* = 0.04) and increased in the IC18 (bilateral cuneus/calcarine gyri: posterior medial visual component) - IC22 (right precentral/postcentral gyrus) pair (t = 2.63, *p* = 0.01). The functional connectivity, both for CC and PC, of the precuneus/posterior cingulate cortex component (IC 1, posterior DMN) with other RSICs, did not significantly change over 4 weeks. The differences in CC or PC disappeared after the correction for multiple comparisons using FDR.
DISCUSSION
==========
This rs-fMRI study evaluated the consistency of both functional connectivity and spatial characteristics between \'at day 1\' and \'after 4 weeks\' among RSICs. First, we confirmed the spatial consistency of RSICs ([@B30], [@B31]) using two different methods: correlation analysis and the Gaussian/gamma model. Also, we found that spatial consistency was variable across RSICs. Second, our high intraclass correlation coefficient (ICC) result suggested that the overall functional connectivity among RSICs is consistent across time. However, in the comparisons of the correlation coefficients of each IC pair across time, the functional connectivity of several RSIC pairs was variable across time. Finally, RSIC pairs showing differences were more frequent in the cross-correlation (CC) analysis using Pearson\'s coefficient than in the PC analysis. Thus, our results suggest that both the spatial map and functional connectivity are consistent across time, but that the degree of their consistencies is variable across RSICs or by the correlation analysis method.
Our results regarding the spatial consistency of RSICs are consistent with previous results. We should also note that the spatial consistency was variable across RSICs ([@B40]), although they were more spatially consistent than were non-RSICs. Relatively high consistency was found in the posterior DMN, bilateral inferior occipital gyri, bilateral calcarine gyri (posterior medial visual component), and bilateral precentral/postcentral gyri, whereas relatively low consistency was found in the basal ganglia, parietal region of the left frontoparietal network, and supplementary motor area (SMA) ([Fig. 1](#F1){ref-type="fig"}). A previous study using ICA also reported both variability of spatial consistency and relatively high spatial consistency of RSICs compared with non-RSICs ([@B31]). In that prior study, RSICs showing relatively high consistency were the bilateral frontoparietal network, bilateral occipital poles (bilateral inferior occipital gyri), posterior DMN, bilateral inferior frontal gyri, and bilateral supramarginal gyri, whereas the cerebellum, SMA, and basal ganglia had low consistency. The spatial consistency of most RSICs in the present study was similar to this prior study, in which the sessions were 45 min or 11 months apart. Both the present and the prior studies suggest that DMNs are spatially reproducible and that the low spatial consistency of RSICs in both SMA and basal ganglia should be further investigated in a longitudinal study. The low spatial consistency of the frontoparietal network, especially in the left parietal region, was not consistent with the prior study. However, there has been an rs-fMRI study supporting our results ([@B41]), and they reported a systematic impairment of associative frontoparieto-cingulate areas in altered states of consciousness. The discrepancy between these studies regarding spatial consistency in the frontoparietal network, especially in the left parietal region, should be investigated using a high-dimensional ICA analysis, which could separate the frontoparietal network into frontal and parietal regions.
We also note the variability of functional connectivity in several RSICs, as well as the overall consistency of RSICs across time. In our study, the RSIC pairs showing differences in functional connectivity were mainly the connections within visual RSICs or between visual and other RSICs ([Fig. 2](#F2){ref-type="fig"}). A MEG study has also shown that graph metrics of functional connectivity were generally consistent but that the reliability was variable across the frequency band ([@B42]). A recent rs-fMRI study using Pearson\'s CC analyses among raw time courses found that the correlation between the DMN and the \'anticorrelated\' network of the resting-state network can vary over time ([@B43]). Thus, the functional connectivity of the resting-state brain network may be dynamically changed, yet may be consistent with the exception of a few RSIC pairs when functional connectivity is averaged within a certain period. The RSIC pairs showing the differences across time were more frequent in the CC analysis using Pearson\'s coefficient than in the PC analysis. Moreover, functional connectivity of the anterior DMN was variable across time in CC, but not PC. Furthermore, in CC, but not in PC, the upper part of the precuneus and inferior occipital gyri (occipital pole) showed differences in functional connectivity with multiple RSICs across time. This suggests that a possible indirect effect of functional connectivity among other RSICs on the functional connectivity of a certain RSIC pair can be controlled for using a PC analysis. A functional connectivity study using simulated fMRI data reported that a PC analysis was more reliable than a CC analysis ([@B35]). Consequently, the PC results are preferable in the present study. Interestingly, in both the PC and CC analysis, functional connectivity decreased across time between IC 12 (right angular and supramarginal gyri/intraparietal lobule) and IC 19 (bilateral middle temporal gyri). IC 19 comprises bilateral V5 areas with connectivity to V1 and V2 and the inferior and medial temporal gyrus. IC 19 is a region of the extrastriate visual cortex and is thought to play a role in the perception of motion and the guidance of some eye movements ([@B44]). In an fMRI study on a blind cohort, dorsal occipito-temporal regions were activated during the detection of auditory motion ([@B45]). The activated area in blindness was close to the maximum intensity point of IC 19 in the present study. Also, IC 12 was connected to the angular and supramarginal gyri and the visual cortex, consisted primarily of the parietal component of the right fronto-parietal network. The functional connectivity between IC 18 (posterior medial visual component) and IC 22 (right precentral/postcentral gyrus) was also variable in the PC analysis. Thus, there are some possible explanations for the variable functional connectivity of these visual information-process-related RSICs. Even if subjects have been asked to close their eyes, they could sometimes open their eyes, roll their eyeballs, or engage in visuospatial processing of scanner-induced noise during fMRI scanning. Their possible brain activity could modulate the functional connectivity of visual information-process-related RSICs. When subjects underwent fMRI scanning again, they might be less exploratory and more comfortable with their environment. If so, their functional connectivity under the relatively unfamiliar environment at day one could be reduced at 4 weeks. This may be a reason that the negative (IC 12-IC 19: bluish) or positive (IC 18-IC 22: yellowish) functional connectivity among RSICs \'at day 1\' was not significant (whitish) at the \'after 4 weeks\' time point in the present study ([Fig. 2C](#F2){ref-type="fig"}). Our study showed no difference in functional connectivity across time after performing the correction for multiple comparisons. This should be not interpreted as implying that all functional connections are consistent. Rather, we argue that the correction for multiple comparisons should be applied in functional connectivity studies to reduce the possible false-positive error rate.
In our study, none of the RSICs showed inconsistency in either functional connectivity or spatial characteristics. However, in a disordered brain, the intensity of the change could be different between functional connectivity and the activation intensity of brain regions in a network. Thus, to understand the brain as a dynamic network, we would recommend the investigation of the modulation of functional connectivity in brain networks as well as the change in activation of the brain regions.
We should note some limitations of the present study. The spatial map of the identified RSICs could be slightly different at every ICA analysis, and thus their functional connectivity could be variable as well. Our results could be confirmed by ICA at various dimensions in a further study, as the number of RSICs produced by ICA differs according to the magnitude of the dimension used for data reduction. Inter-subject variance in areas such as mental status during scan, education, or intelligence might cause different functional connectivity among RSICs. Thus, our results require confirmation with a dataset from a large sample, even though one study verified ICA repeatability using 42 RSICs ([@B46]). The neuropsychological or clinical meaning of the functional connectivity of each IC pair should be explored in further studies.
We found that most of the RSICs were reproducible across time, whereas some RSICs were variable in both their spatial characteristics and functional connectivity. Functional connectivity might be affected by the correlation analysis method applied. Our results suggested that researchers should consider both the variability of functional connectivity among the RSICs across time and the influence of the correlation analysis method.
This research was supported by the Original Technology Research Program for Brain Science through the National Research Foundation of Korea (NRF) funded by the Ministry of Education, Science and Technology (No. 2010-0029266: B.S.J.).
{#F1}
{#F2}
######
Spatial Consistency of Resting State Related Independent Components (RSICs) Using Results of Post-Hoc Regression Analysis
**Note.**- N = network, S = system, DMN = default mode network, SMA = supplementary motor area, Inf = inferior, Sup = superior, C = cortex, G = gyrus, L = left, R = right, B = bilateral
| {
"pile_set_name": "PubMed Central"
} |
1. Introduction {#sec0005}
===============
Splenic cystic lesions are uncommon entity with an incidence rate of 0.07% reported in a review of 42,327 autopsies \[[@bib0005]\]. They are classified into true and false cysts based on the presence of cellular epithelial lining \[[@bib0010]\].True or primary cysts are epithelium-lined cysts and represent 25% of all splenic cysts. According to their etiology, they are classified as congenital (SEC), neoplastic or parasitic cysts \[[@bib0015]\]. SEC is benign, sporadic and occurs predominantly in females between the second and third decade of life \[[@bib0010],[@bib0020]\]. The clinical presentation is nonspecific and it varies from asymptomatic occurrence to symptoms related to the size, location and the presence of cyst complications \[[@bib0025]\].
Diagnostic imaging modalities such as abdominal ultrasound, scan and MRI demonstrate easily the splenic cyst but are unable to differentiate SEC from other types of primary splenic cyst such as parasitic and neoplastic \[[@bib0025],[@bib0030]\]. The definitive diagnosis is based on the histopathologic examination of the operative specimen \[[@bib0030]\].
We herein report an interesting case of large epithelial cyst with elevated serum total IgE level misdiagnosed as Hydatid cyst. This work has been reported in line with the SCARE criteria \[[@bib0035]\].
2. Case presentation {#sec0010}
====================
A 17 year old horse-rider girl was referred to our clinic for 2 weeks history of moderate continuous, crampy abdominal pain, starting in the epigastric region and shifted to the left upper quadrant. This pain was associated with fatigue, loss of appetite. Patient denies any nausea, vomiting, diarrhea, fever, night sweats. She reported a remoteleft shoulder pain with negative MRI. Her physical exam was positive for splenomegaly 9 cm below costal margins, and left upper quadrant tenderness, with no rebound tenderness. Her laboratory examination showed a Hg:12.8 g/dL, platelet: 124,000/mm[@bib0015], WBC: 6500/mm[@bib0015] with 62% neutrophils, 24% lymphocytes, and 7% eosinophils, platelet: 124,000/mm[@bib0015], a CRP:0.78 mg/L. The Liver enzymes, Bilirubin, Albumin, LDH, the Chemistry panel were all in normal range. Abdominal ultrasound showed a large splenic cyst of 15 cm containing homogenous internal debris ([Fig. 1](#fig0005){ref-type="fig"}). An abdominal computed tomography scan showed the same 15 cm splenic cyst with parietal calcifications, compressing the stomach, most likely of hydatid origin ([Fig. 2](#fig0010){ref-type="fig"}a, b). Abdominal MRI showed unilocular splenic cyst hypo-intense T1, hyper-intense T2 ([Fig. 3](#fig0015){ref-type="fig"}a, b). Differential diagnosis for described findings include; Splenic abscess, Hydatid cyst, epithelial cyst and post traumatic hemorrhage in pre-existing epithelial cyst. Based on clinical picture and endemic status for hydatid cyst differential can be narrowed.Fig. 1Abdominal ultrasound showing a large splenic cyst containing homogenous internal echoes/debris.Fig. 1Fig. 2(a) Axial image of non-contrast CT abdomen shows the large splenic cyst with focal peripheral calcification in wall (White arrow), (b) contrast enhanced CT shows no peripheral enhancement, no floating membranes or enhancing solid component.Fig. 2Fig. 3Axial T1 post contrast MR image shows a non-enhancing cystic lesion (3a), Coronal T2 image shows a hyper-intense fluid signal intensity splenic lesion (3b).Fig. 3
Serologic test for Hepatitis B virus (HBV), hepatitis C virus (HCV), Cytomegalovirus (CMV), HIV, Toxoplasmosis, Entamoeba histolytica, Leishmania brazilensis, donovani, and EBV IgM were all negative. EBV IgG was elevated showing prior immunization. The immune-diffusion test for Echinococus multilocularis was negative. The Indirect hemagglutination test and the Elisa test for Echinococcus granulosus were also negative.
Nevertheless, due to an elevated IgE level: 317 kU/L, the patient was considered as having splenic hydatid cyst and was treated by albendazole PO with meals in a dose of 400 mg twice daily for 28 days, and received the anti-pneumococcal vaccine.
Due to the severe continuous pain, the large size, the risk of spontaneous rupture and the patient's wishes to resume her hobby as a horse-rider as soon as possible, she was consented for operative exploration via a laparotomy incision for splenic cyst un-roofing.
Exploration was done, abdominal cavity was protected by hypertonic saline (3%NaCl) filled pads, cyst was punctured, 2 liters of dark green fluid was aspirated. Hypertonic saline was injected in the cyst, and then aspirated after 15 min. Un-roofing and partial resection was done afterward.
The postoperative course was un-eventful and the patient was discharged home on the post-operative day 5.
The pathology report showed stratified epithelium with fibro-inflammatory reaction in the pericystic zone compatible with splenic epithelial cyst.
The patient still symptom free after 5 years of follow up and her labs showed a WBC: 7700/mm^3^ normalization of eosinophils (2.5%).
3. Discussion {#sec0015}
=============
SEC are congenital true splenic cyst characterized by an epithelial lining. They are divided in 3 subgroups depend on the type of epithelial lining, Epidermoid cysts are covered with stratified squamous epithelial lining, Mesothelial cyst with cuboidal epithelial lining and Dermoid cyst with squamous epithelium with hair follicles, sebaceous glands and skin appendages \[[@bib0040]\]. The epidermoid subtype represents 10% of all SEC and they are strongly linked to elevated CA 19-9 level because inner epithelial cells secrete CA 19-9 \[[@bib0045]\]. By immunohistochemistry, the epidermoid cysts are CA 19-9, CEA and cytokeratin positive but show no immunoreactivity for calretinin; whereas mesothelial cysts are calretinin and cytokeratin positive but show no staining for CA 19-9 and CEA \[[@bib0050]\].
The clinical presentation of SEC is non-specific and it can be various depend on the size, location and the presence of complications. Uncomplicated cyst less than 8 cm in diameter are usually asymptomatic \[[@bib0055]\]. The increase in its size leads to distention of its capsule and therefore the development of pain and mainly in the left upper quadrant which is the most common symptom. Progressive symptoms are related to compression of adjacent organs including distension, early satiety, vomiting, flatulence, persistent cough, pleuritic pain and hydronephrosis due to local pressure on stomach, left hemi-diaphragm and left kidney respectively \[[@bib0060]\]. In very rare cases, complications of the cyst such as hemorrhage, infection or rupture induce peritoneal sign due to hemoperitoneum, peritonitis or even sepsis \[[@bib0045]\].
Regarding the pre-operative diagnosis of our case, the negative Echinococcosis detection test doesn't rule out the hydatid disease and is present only in case of microscopic rupture \[[@bib0065]\]. Accordingly, the association of elevated IgE level made the diagnosis more difficult. Although elevated IgE is not specific for hydatid disease, but is present in a large subset of affected individuals \[[@bib0070]\].
Nowadays, the expansion of medical screening systems increases the incidental detection of splenic lesions especially for SEC \[[@bib0075]\]. Epithelial cysts appear as well defined, thin wall, liquid containing lesions on Ultrasound, CT scan and MRI. They can contain debris in case of intra-cystic bleed or infection \[[@bib0080]\]. Abdominal MRI has a higher sensitivity in the identification of the septa and calcification \[[@bib0060]\]. But all of these diagnostic modalities are unable to differentiate the SEC from other splenic cysts. The definitive diagnosis is established by the anatomo-pathologic examination \[[@bib0010]\].
Concerning the indication of surgical treatment, there is a limited data to determine the appropriate time to interfere. Traditionally, it has been recommended to treat symptomatic cysts, or when they are larger than 5 cm due to an increased risk of spontaneous rupture, hemoperitoneum, chemical peritonitis or abscess formation \[[@bib0085]\]. However, some studies described that small cysts may be also at risk of rupture after a simple trauma or heavy cough in infants \[[@bib0090]\].
There are multiple surgical treatment modalities including aspiration, marsupialization, cystectomy, cyst de-roofing, cyst de-capsulation, partial splenectomy and splenectomy \[[@bib0095],[@bib0100]\].
Historically, open total splenectomy was the ideal surgical approach in front of SEC in order to decrease the risk of bleeding and complications from the cyst. Currently, with the awareness of immunologic role of the spleen and particularly in young age and the increased risk of post splenectomy sepsis, splenic preserving surgery with laparoscopic approach is advocated \[[@bib0095],[@bib0100]\]. Laparoscopic de-roofing is reported to be effective treatment of splenic epithelial cyst but it carries a risk of recurrence of about 22% of which only 3% needs re-intervention \[[@bib0105]\]. The choice of surgical procedure depends on several factors like: the amount of remaining healthy splenic tissue, the size, number and location of the cyst in relation to the hilum, pathogenesis of the cyst and patient's age \[[@bib0010]\].
In conclusion, epithelial splenic cyst is a rare entity, usually diagnosed incidentally in asymptomatic patients. Symptoms when present are related to size and location. SEC can be misdiagnosed as hydatid cyst. Imaging can be helpful in the diagnosis but cannot differentiate it from the other types of primary cyst. The definitive diagnosis is based on pathologic examination of operative specimen. SEC is best treated by parenchymal sparing surgery.
Conflicts of interest {#sec0020}
=====================
We have no conflict of interest to declare.
Funding {#sec0025}
=======
No funding source.
Ethical approval {#sec0030}
================
The submitted article is a case report, ethical approval has been exempted by our institution.
Consent {#sec0035}
=======
A Written informed consent was obtained from the patient for surgery and potential publication of this case report and any accompanying images.
Authors contribution {#sec0040}
====================
Youssef Sleiman and Ali Bohlok wrote the manuscript. Melody El-Khoury assisted in large part of the literature review and revised the manuscript for correction before submission. Marc Zalcman did the imaging diagnosis, and wrote the imaging part of the manuscript. Issam El Nakadi and operated the patient, wrote the part about the detailed operative technique and revised the manuscript for correction before submission. Peter Demetter and did the pathologic examination of the operative specimen and explained the pathologic diagnosis in the manuscript. All authors have read and approved the manuscript before submission to your journal.
Registration of research studies {#sec0045}
================================
Not applicable.
Guarantor {#sec0050}
=========
Dr Issam El Nakadi.
Availability of data and materials {#sec0055}
==================================
The data sets supporting the conclusions of this article are included within the article.
Compliance with ethics guidelines {#sec0060}
=================================
All procedures reported here were in accordance with the ethical standards of the Institut Jules-Bordet Committee and with the 1964 Declaration of Helsinki and its later amendments or comparable ethical standards. Informed consent was obtained from all the subject of the case report included in the study.
Provenance and peer review {#sec0065}
==========================
Not commissioned, externally peer reviewed.
[^1]: Both authors contributed equally in the writing of the manuscript.
| {
"pile_set_name": "PubMed Central"
} |
Introduction
============
Articular cartilage injury is increasing in incidence year by year, which is an important healthcare problem. Cell-based tissue engineering holds promise for restoring cartilage defects ([@b1-etm-0-0-4474]). To date, the most widely used cell sources in cartilage regeneration are mesenchymal stem cells (MSCs) and mature chondrocytes. Notably, MSCs have already been used to repair cartilage defects in clinical trials ([@b2-etm-0-0-4474],[@b3-etm-0-0-4474]). MSCs are easily obtained from various kinds of tissues, such as bone marrow, synovial tissue and muscle, and they would not be rejected by the immune system when used *in vivo* ([@b4-etm-0-0-4474]--[@b6-etm-0-0-4474]). However, the limited proliferation and differentiation potential has restrained the use of MSCs in regenerative medicine ([@b7-etm-0-0-4474]). In addition, the proliferative capability and differentiation potential of MSCs has been reported to decline with age ([@b8-etm-0-0-4474]).
Induced pluripotent stem cells (iPSCs) can be generated from well-differentiated somatic cells by introducing defined reprogramming transcription factors using retroviruses ([@b9-etm-0-0-4474]). iPSCs possess pluripotency, proliferation ability and multi-lineage differentiation potential similar to embryonic stem cells (ESCs) ([@b9-etm-0-0-4474]--[@b11-etm-0-0-4474]). In addition, a variety of new methods have been developed to generate iPSCs for the purpose of reducing the risk of tumor formation ([@b12-etm-0-0-4474],[@b13-etm-0-0-4474]). Therefore, iPSCs are regarded as alternative cell sources in regenerative medicine.
Undifferentiated iPSCs will form teratoma *in vivo* ([@b14-etm-0-0-4474]), which is the main obstacle to the use of iPSCs for tissue regeneration. The differentiation of iPSCs into MSCs has the promise to solve this problem. ESC markers (for example, Nanog and Sox2) were reported to no longer appear in iPSC-derived mesenchymal stem cells (iPSCs-MSCs), which may reduce the risk of tumorigenicity when used *in vivo* ([@b15-etm-0-0-4474],[@b16-etm-0-0-4474]). Studies have also showed that it is possible to induce the differentiation of iPSCs-MSCs into osteogenic, chondrogenic and vascular lineages *in vitro* ([@b15-etm-0-0-4474]--[@b19-etm-0-0-4474]). However, few studies have used iPSCs or iPSCs-MSCs to repair cartilage defects *in vivo*.
In the present study, mesenchymal progenitor cells were obtained from human iPSCs (hiPSCs) via embryoid body (EB) formation, a step that mimics embryonic development. The *in vivo* ability of hiPSCs-MSCs to repair cartilage defects was examined using a full-thickness cartilage defect rabbit model.
Materials and methods
=====================
### hiPSC culture
The hiPSC line (no. 0209-001; Sidan Sai Biotechnology Co., Ltd., Shanghai, China) was generated previously by introducing six reprogramming factors (Oct3/4, Sox2, Klf4, c-Myc, Nanog and Lin 28) into human newborn foreskin fibroblasts ([@b20-etm-0-0-4474]). The undifferentiated hiPSCs were maintained and expanded according to previous reported methods ([@b20-etm-0-0-4474]). In brief, chemically inactivated murine embryonic fibroblasts (MEFs) were used as feeder cells and were seeded on Matrigel-coated (Sigma-Aldrich; Merck Millipore, Darmstadt, Germany) dishes. hiPSCs were cultured on MEF feeder layers in ES medium (Sidan Sai Biotechnology Co., Ltd.) supplemented with 4 ng/ml basic fibroblast growth factor (bFGF) (Peprotech, Inc., Rocky Hill, NJ, USA). The medium was refreshed every day. Type IV collagenase (Sigma-Aldrich; Merck Millipore) was used to perform cell passage.
### hiPSCs-MSCs preparation
Undifferentiated hiPSCs were detached from culture dishes using 1 mg/ml type IV collagenase and were then plated onto low-attachment culture dishes at a density of 1,000--1,200 cell clusters per 100 mm dish. The cells were allowed to aggregate and form spheres in a humidified atmosphere at 37°C and 5% CO~2~ (Thermo Fisher Scientific, Inc., Waltham, MA, USA) in a maintenance medium containing Dulbecco\'s modified Eagle\'s medium (DMEM)/F12 and 10% fetal bovine serum (FBS; Invitrogen; Thermo Fisher Scientific, Inc.). EBs formed after 7 days\' suspension culture and were transferred to gelatin-coated (Sigma-Aldrich; Merck Millipore) dishes at 800--1,000 EBs/100 mm dish in expansion medium with DMEM/F12, 10% FBS, 100 U/ml penicillin, and 100 mg/m2 streptomycin (all from Invitrogen; Thermo Fisher Scientific, Inc.). The cells sprouted from EBs were harvested as hiPSC-MSCs and expanded in expansion medium. The hiPSC-MSCs were purified by removing non-adherent cells.
### Flow cytometry
The hiPSCs-MSCs at passage 3 were harvested. One million cells were suspended in 100 µl buffer that consisted of 0.5% bovine serum albumin (BSA; Sigma-Aldrich; Merck Millipore) and 2 mM EDTA (Sunshine Biotechnology Co., Ltd., Nanjing, China). Subsequently, 10 µl 1:10 diluted fluorescein isothiocyanate (FITC)-coupled antibodies recognizing CD11b (130-098-778), CD105 (130-098-778), CD90 (130-097-930), CD45 (130-098-043) and CD34 (130-098-142) (MACS; Miltenyi Biotec, Bergisch Gladbach, Germany) were added. In addition, 1:10 diluted mouse IgG1 (130-104-562) and mouse IgG2a (no. 130-098-877) antibodies (MACS; Miltenyi Biotec) were used as isotype controls. Incubation for 10 min incubation in the dark at 4°C was performed. The cells were then washed with buffer containing phosphate-buffered saline, pH 7.2, 0.5% BSA, and 2 mM EDTA by diluting MACS BSA Stock Solution (130-091-376) 1:20 with autoMACS Rinsing Solution (130-091-222) (MACS; Miltenyi Biotec). Then, the cells were centrifuged at 300 × g for 10 min at 4°C and resuspended in 500 µl of the aforementioned buffer for analysis by flow cytometry (BD FACSCalibur, BD Biosciences, Franklin Lakes, NJ, USA). The data was analyzed using Flowjo 7.6 sofrware (BD Biosciences).
### Animal model and transplantation procedure
A total of 36 skeletally mature female New Zealand white rabbits (age, 12 weeks; weight, 2.0--2.5 kg), purchased from the Jinling Farm, Nanjing, China were used in this study. Rabbits were fed a regular diet twice a day and allowed free access to water. They were housed under controlled conditions (temperature, 25±3°C; humidity, 45±5%; 12-h light/dark cycle). All surgical procedures were approved by the Institutional Rabbit Care and Use Committee of Drum Tower Hospital, Medical School of Nanjing University (Nanjing, China). A full-thickness cartilage defect model was made in the trochlear grooves of the rabbits as previously reported ([@b21-etm-0-0-4474]). In brief, the rabbits were anesthetized with an intramuscular injection of 20 mg/ml xylazine hydrochloride (Huamu Animal Health Care Co., Ltd., Jilin, China) at a dose of 3 ml/kg. The knee articular surface of the rabbits was exposed through a medial parapatellar approach. Whether the right or the left knee was used to perform the surgery was determined randomly. An osteochondral transplantation system (3.5 mm in diameter, 3.0 mm in depth) was used to create osteochondral defects. The rabbits were then divided into three groups according to implantation: Control group, scaffold implantation group and scaffold/hiPSCs-MSCs (experimental) group (n=12 per group).
The poly(lactic-co-glycolide) (PLGA) scaffold was purchased from Shandong Institute of Medical Instruments (Jinan, China). The average pore diameter of the PLGA scaffold was \~200 µm. The PLGA scaffold was cut to 3.5×3.5×3 mm dimensions with a razor blade. The prepared PLGA scaffolds were immersed in Matrigel for 24 h to enhance cell attachment. Then, 5×10^6^ hiPSCs-MSCs were seeded onto the prepared scaffold. After incubating in complete medium for 12 h, the PLGA/hiPSCs-MSCs complex was transplanted into the cartilage defect in the experimental group. The scaffold implantation group received only PLGA scaffold, and the control group was untreated. Six rabbits from each group were sacrificed at 3 and 6 weeks after surgery. The repair quality was evaluated by gross and histological examination.
### Histological analysis
The specimens were cut into 5-µm sections and stained with hematoxylin and eosin (H&E) and toluidine blue as previous reported ([@b21-etm-0-0-4474]). In brief, the specimens were fixed, decalcified, dehydrated and embedded in paraffin. The specimens were then cut into 5-µm sections and stained with H&E (Beyotime Institute of Biotechnology, Shanghai, China) and toluidine blue (Toyond Biotechnology Co., Ltd., Shanghai, China) staining according to the manufacturers\' instructions. The results were assessed independently by 3 different investigators.
Results
=======
### Generation of hiPSCs-MSCs
A multistep culture method consisting of spontaneous differentiation via a step of EB formation, cell outgrowth from EBs, and monolayer culture following cell dissociation was used in the present study to generate hiPSC-MSCs ([Fig. 1](#f1-etm-0-0-4474){ref-type="fig"}). hiPSC-MSCs originated from the mesoderm and neural crest of the EBs and exhibited a spindle-like shape ([Fig. 1D](#f1-etm-0-0-4474){ref-type="fig"}). Flow cytometric analysis was used to analyze the mesenchymal properties of the hiPSC-MSCs obtained in this study. The results showed that the majority of cells expressed CD90, and some expressed CD105, but most cells did not express CD34, CD11b or CD45 ([Fig. 2](#f2-etm-0-0-4474){ref-type="fig"}).
### Macroscopic evaluation of repair quality
Following transplantation ([Fig. 3A](#f3-etm-0-0-4474){ref-type="fig"}), in the control and scaffold implantation groups, little repair tissue was observed in the cartilage defect 3 weeks after surgery ([Fig. 3B and C](#f3-etm-0-0-4474){ref-type="fig"}). However, in the experimental group, repair tissue covering \>50% of the defects was observed ([Fig. 3D](#f3-etm-0-0-4474){ref-type="fig"}). At 6 weeks, the cartilage defect was only partially covered by fibrous tissue in the control and scaffold implantation groups ([Fig. 3E and F](#f3-etm-0-0-4474){ref-type="fig"}). At 6 weeks, repair tissue almost 100% filled the cartilage defect in the experimental group ([Fig. 3G](#f3-etm-0-0-4474){ref-type="fig"}).
### Histological evaluation of the repair quality
H&E staining showed better repair quality in the experimental group compared with that in the other two groups ([Fig. 4](#f4-etm-0-0-4474){ref-type="fig"}). Only fibrous tissue was observed in defects of the control and scaffold implantation groups at 3 and 6 weeks. In the experimental group, H&E staining showed cartilage-like tissue in the top layer of the defect at 6 weeks. However, subchondral bone formation was poor in all the groups. The newly formed tissue was stained slightly in the control and scaffold implantation groups at 6 weeks by toluidine blue staining ([Fig. 5A-D](#f5-etm-0-0-4474){ref-type="fig"}). The matrix of regenerated tissue in the top layer of the defect in the experimental group was stained intensely. However, native cartilage degeneration was also observed ([Fig. 5E and F](#f5-etm-0-0-4474){ref-type="fig"}).
Discussion
==========
The results revealed cartilage-like tissue formation in the top layer of the cartilage defect when hiPSCs-MSCs were used. An apparently better quality of *in vivo* cartilage defect repair in the experimental group compared with the control and scaffold implantation groups was demonstrated by gross and histological appearance. Another important finding was that there was no evidence of teratoma formation in the experimental group. Although the restoration of full-thickness cartilage defect was not totally satisfactory, the results of the present study indicated that iPSCs may be a new cell source for cartilage defect repair *in vivo*.
iPSCs have been considered as the optimal cell source for regenerative medicine because of their self-renewal and pluripotency capability ([@b22-etm-0-0-4474]). Few studies have examined *in vivo* cartilage defect repair using iPSCs. Ko *et al* ([@b19-etm-0-0-4474]) reported successful induction of chondrogenesis and repair of cartilage defect *in vivo* with hiPSCs in immunosuppressed rats. Yamashita *et al* ([@b23-etm-0-0-4474]) reported hyaline chondrogenesis from hiPSCs and showed neo-cartilage survival in joint surface defects following newly generated cartilage particle transplantation in immunosuppressed rats and mini pigs. The results of the present study appear to be inferior to those of the aforementioned studies. This might be the result of using hiPSCs-MSCs transplantation, rather than newly generated cartilage transplantation in the present study, as local environmental inductive effects would be inferior to those of exogenous growth factors. Similarly, Marquass *et al* ([@b24-etm-0-0-4474]) also demonstrated that differentiated MSCs showed better histological outcomes compared with undifferentiated MSCs. However, the rabbit model used in the present study was more appropriate for the examination of cartilage defect repair than a rat model, as the cartilage thickness of rats is much thinner and the endogenous healing potential in rats is greater ([@b25-etm-0-0-4474]).
No teratoma formation was observed in the present study. This suggests that iPSCs-MSCs may be safer than iPSCs when used *in vivo*, although the mechanism is not clear. In previous studies, Ko *et al* ([@b19-etm-0-0-4474]) and Yamashita *et al* ([@b23-etm-0-0-4474]) did not report teratoma formation *in vivo*, consistent with observations in the present study. The method used to get hiPSCs-MSCs in the present study consisted of three steps: i) EB formation; ii) cell outgrowth from EBs; and iii) monolayer cell culture to select cells that can adapt to MSC growth conditions. Numerous alternative approaches for the preparation of MSCs from ESCs or iPSCs, such as using co-culture methods ([@b26-etm-0-0-4474],[@b27-etm-0-0-4474]), gene transfection ([@b28-etm-0-0-4474]) or conditioned medium ([@b29-etm-0-0-4474]) have been reported. However, the use of other cells or exogenous genetic material may introduce contamination with animal pathogens or the risk of tumorigenicity. Thus, the culture protocol used in the present study, which is simple and reproducible, appears to be suitable for the generation of MSCs from hiPSCs.
This study also had some limitations. Firstly, no examination was conducted to confirm whether the newly generated repair tissue was induced from transplanted hiPSCs-MSCs, or whether the implanted hiPSCs-MSCs remained *in situ*. Some unexpected factors may play a role during cartilage defect repair *in vivo* with hiPSCs-MSCs. It is possible that the paracrine effect of implanted hiPSCs-MSCs contributed to the attraction of host chondrocytes and MSCs to the cartilage defects. Second, the follow-up period may have limited the repair quality in this study. Results were only observed at 3 and 6 weeks, as we were keen to avoid any rejection reactions in the xenotransplantation model used in this study. There have been a few studies concerning xenotransplantation for cartilage defect repair. Pei *et al* ([@b30-etm-0-0-4474]) demonstrated failure of xenoimplantation using porcine MSCs for rabbit cartilage defects at a follow up of 6 months. However, Jang *et al* ([@b31-etm-0-0-4474]) reported a successful result in xenoimplantation of human MSCs into rabbit cartilage defects at 4 and 8 weeks. Thus, although there is no consensus for the appropriate follow-up period in xenoimplantation, the follow-up period in the present study may have been too short to induce rejection reactions. Thirdly, the hiPSCs-MSCs were not purified by cell sorting, which may also limit the cartilage defect repair.
Although this study had some limitations, it suggested that full-thickness cartilage defects can be repaired using hiPSCs-MSCs. Further understanding of the differentiation of iPSCs and a long-term investigation of full-thickness cartilage defect regeneration with iPSCs are necessary.
The present study was supported by the Projects of International Cooperation and Exchanges Natural Science Foundation of China (NSFC) (grant no. 81420108021), National Key Technology Support Program (grant no. 2015BAI08B02), Excellent Young Scholars NSFC (grant no. 81622033), NSFC (grant no. 81572129), Jiangsu Provincial Key Medical Center Foundation, Jiangsu Provincial Medical Talent Foundation and Jiangsu Provincial Medical Outstanding Talent Foundation, Social Development Project of Jiangsu Provincial Science and Technology Department (grant no. BE2016609).
{#f1-etm-0-0-4474}
{#f2-etm-0-0-4474}
{#f3-etm-0-0-4474}
{#f4-etm-0-0-4474}
{#f5-etm-0-0-4474}
| {
"pile_set_name": "PubMed Central"
} |
Introduction
============
Despite the advances in medical care, bacterial infections remain very common in patients with liver cirrhosis and account for significant morbidity and mortality in them \[[@b1-cm-91-356],[@b2-cm-91-356]\]. Infections are more frequent in patients with decompensated cirrhosis than in those with compensated liver disease. The impaired immune responses such as decreased phagocytic activity, neutrophil dysfunction, decreased complement levels in serum and an impaired opsonic activity both in ascitic fluid and serum increase the susceptibility of cirrhotic patients to bacterial infections \[[@b3-cm-91-356]--[@b5-cm-91-356]\].
Spontaneous bacterial peritonitis (SBP), urinary tract infection (UTI), pneumonia and bacteremia are the commonly described bacterial infections among patients with liver disease \[[@b6-cm-91-356],[@b7-cm-91-356]\]. The majority of the bacterial infections in cirrhotic patients are caused by Gram-negative bacteria, whereas gram positive bacteria comprise about 20% and anaerobes only 3% of these infections \[[@b8-cm-91-356]\].
Once the infection occurs, excessive production of pro-inflammatory cytokines in cirrhosis further facilitates the development of serious complications such as shock, hepatic encephalopathy, multiple organ failure and death \[[@b1-cm-91-356],[@b4-cm-91-356],[@b5-cm-91-356]\].
Hence, early recognition of the infection and proper management is essential in order to minimize the complications and reduce mortality. We designed this study to identify the clinical characteristics and outcome of bacterial infections affecting various organ systems in patients with cirrhosis and to study the distribution and outcome of bacterial infections in relation to the severity of liver dysfunction as per Child Pugh classification.
Materials and methods
=====================
This was a cross sectional study conducted at a tertiary care hospital in Southern India. Only one admission of the patient during the study period was considered. Subjects aged ≥18 years with cirrhosis having microbiologically proven bacterial infection involving various organ systems were admitted to the hospital and those with culture-negative neutrocytic ascites (CNNA) were included. We excluded patients with human immunodeficiency virus (HIV) infection, secondary bacterial peritonitis and those on cancer chemotherapy and steroid therapy. The study protocol was approved by the Institutional Ethics Committee and an informed consent was taken from all the subjects.
The diagnosis of cirrhosis was based on the clinical and biochemical features suggestive of chronic liver disease. Abdominal ultrasound was performed in all the cases. A detailed history including symptoms indicative of bacterial infection was obtained from each patient. They were subjected to physical examination to detect the presence of fever, anemia, jaundice, ascites and encephalopathy. Blood, urine, ascitic fluid, sputum, endotracheal tube and wound cultures were sent according to site of infection.
Definitions
-----------
Community acquired infection was defined as infection present upon hospitalization or diagnosed within the first 48 hours of admission to the hospital. Infection occurring after 48 hours of admission to the hospital was considered as hospital acquired \[[@b9-cm-91-356]\]. SBP was defined as an ascitic fluid infection with PMN count \> 250 cells/mm^3^ and positive ascitic fluid culture without evidence of intra-abdominal surgically treatable source of infection. Culture-negative neutrocytic ascites (CNNA) was defined as an ascitic fluid infection with PMN count \>250 cells/mm^3^ and negative ascitic fluid culture. Monomicrobial nonneutrocytic bacterascites (MMBA) was defined as an ascitic fluid PMN count \< 250 cells/mm^3^ with a positive culture for a single organism \[[@b10-cm-91-356]\].
The diagnosis of UTI was made if the patient had at least one of the following symptoms or signs: fever (\>38° C), urgency, frequency of micturition, dysuria, or suprapubic tenderness and positive urine culture that were \>10^5^ microorganisms per cm^3^ with no more than two species of microorganisms. Asymptomatic bacteriuria was defined as growth of bacteria \>10^5^ microorganisms per cm^3^ of urine with no more than two species of microorganisms in the absence of urinary symptoms \[[@b11-cm-91-356]\]. Respiratory tract infection (RTI) were diagnosed by positive sputum/endotracheal culture with or without chest X-ray findings. Bacteremia was defined as the presence of bacteria in blood as evidenced by positive blood culture \[[@b12-cm-91-356]\]. Acute kidney injury (AKI) was defined as increased serum creatinine \>1.5 times from the baseline or decreased urine output of \<0.5 ml/kg/hour for 6 hours. Severity of liver dysfunction was graded using Child Pugh's classification \[[@b13-cm-91-356]\]. The patient outcome was defined as survivors or non-survivors during the particular hospital stay.
Statistical analysis
--------------------
Categorical variables are summarized by frequency and percentage. Continuous variables are summarized using mean and standard deviation (for normally distributed variables) or median and inter quartile range (for non-normally distributed variables). Chi-square test was used as test of association between two categorical variables. Independent sample t test or Mann Whitney U test was used to compare means across binary variable. Univariate analysis was performed to identify factors associated with mortality. Variables with p-value cutoff of \<0.2 on univariate analysis were considered for logistic regression analysis. Among the significant variables in the univariate analysis, independent variables with a sample size of \>5 in each category were included in the multivariate analysis. Forward selection criteria were used to select significant variables in multiple logistic regression analysis for predicting the mortality in cirrhotic patients. A p-value of \<0.05 is considered statistically significant. Statistical analysis was carried out using EZR plugin in R software version 3.4.1.
Results
=======
A total of 158 patients formed the study group with 143 (90.5%) males and 15 (9.5%) females. The most affected age group was 41--60 years, with 50 patients (31.6%) being in the age group 41--50 years and 53 patients (33.5%) in the age group 51--60 years. The mean age of the patients was 50.97 ± 10.3 years. The etiology of cirrhosis was alcoholic liver disease in 105 (66.4%), viral hepatitis in 27 (17%), cryptogenic in 18 (11.4%), autoimmune diseases in 4 (2.5%) and combined HBV infection and alcoholic liver disease in 3 (1.8%) cases. Co-infection by HBV and HCV was found in 1 case.
The most common presenting symptoms were abdominal distension 107 (67.7%) followed by jaundice 90 (56.9%), fever 75 (47.4%) and altered sensorium 45 (28.4%). Other symptoms included upper gastrointestinal (UGI) bleeding, abdominal pain and cough, which were noted in 20 (12.7%), 19 (12%) and 20 (12.7%) patients respectively.
Ascitic fluid infection (AFI) (38.3%) and bacteremia (24.3%) comprised majority of infections. Other infections included abscess, lower limb cellulitis and tuberculous cervical lymphadenitis. Out of 74 patients with ascitic fluid infections, 45 had CNNA, 19 had SBP and 10 had MMBA. Among 48 episodes of bacteremia, the source was detected as intra-abdominal in 15, UTI in 3 and RTI in 2 patients, while the source was not detected in 24 cases. Almost 63.9% of the patients were in Child class C, 32.2% in Child class B and 3.7% in Child class A. All types of infections except UTI occurred largely in Child class C, while UTI was predominantly seen in Child class B ([Table I](#tI-cm-91-356){ref-type="table"}). Among the complications, AKI was most common which was observed in 37, 13 and 28 patients with AFI, UTI and bacteremia respectively followed by hepatic encephalopathy which was noted in 21 patients with AFI and 17 cases with bacteremia.
Microbiological spectrum
------------------------
Community acquired infections were more frequent than hospital acquired infections \[120 (76%) vs 38 (24%)\]. Among the community acquired infections, AFI was most common (n=68, 41.4%) followed by bacteremia which was noted in 41 (24.8%) patients. A significant proportion (74.3%) of isolates comprised of Gram negative bacilli while 22% were Gram positive. *M. tuberculosis* was responsible for 11 episodes of bacterial infections. Special pathogens such as *Brucella* spp*., Burkholderia pseudomallei* and *Salmonella* Typhi were isolated from two and one each patient respectively. E. coli was frequently isolated from ascitic fluid, blood and urine. RTI occurred more frequently during hospitalization with *Klebsiella* and *Acinetobacter* being responsible for most of these infections ([Table II](#tII-cm-91-356){ref-type="table"}). It was noted that 78.8% (n=41) of *E. coli* and 66.6% (n=16) of *Klebsiella* were extended-spectrum β-lactamase (ESBL) producers.
Factors predicting mortality
----------------------------
The overall in-hospital mortality was recorded in 38 (24%) patients. The factors associated with mortality were type of infection, Child Pugh category, AKI, hepatic encephalopathy, UTI, creatinine and bilirubin levels. A high mortality rate was observed in patients with Child class C (36.7%). The patients without UTI had a poorer outcome than the patients with UTI (p-value 0.021); however, the number of patients who died with UTI was small. The mean creatinine was high (2.51±1.35) in the group of patients who died than in the group of patients with favorable outcome (1.48 ± 0.98) and the p-value was significant between the two groups (p-value \<0.001) ([Table III](#tIII-cm-91-356){ref-type="table"}). There was a significant difference in bilirubin value between the group of patients who had a favorable and poor outcome (p-value 0.004). Logistic regression analysis revealed that type of infection (OR: 0.33, 95% CI: 0.11--1.01), AFI (OR: 2.81, 95% CI: 1.11--7.12), hepatic encephalopathy (OR: 0.17, 95% CI: 0.070--0.422) and AKI (OR: 0.19, 95% CI: 0.077--0.502) were significantly associated with in-hospital mortality among cirrhotic patients ([Table IV](#tIV-cm-91-356){ref-type="table"}).
Discussion
==========
In the present study 193 episodes of bacterial infections were noted in 158 patients. 126 cases had single site infections whereas 32 had multiple site infections. The proportion of males was predominant (90.5% vs. 9.5% females). The higher incidence of bacterial infections in males can be explained by the higher incidence of cirrhosis in them compared to females. Chronic alcohol intake was identified as a frequent etiology of cirrhosis in our study. It is believed that alcohol increases the permeability of the intestinal mucosa and decreases the activity of Kupffer cells, thus favoring infection \[[@b14-cm-91-356],[@b15-cm-91-356]\].
Ascitic fluid was the most common site of bacterial infection (38.3%) similar to the observations recorded by Caly et al. and Preda et al \[[@b7-cm-91-356],[@b14-cm-91-356]\]. As we know, the main trigger for SBP is bacterial translocation. The possible pathophysiology for bacterial translocation in patients with cirrhosis are intestinal bacterial overgrowth, altered intestinal permeability and altered gut-associated lymphoid tissue (GALT) immune response \[[@b16-cm-91-356]\]. In our study, 24.8% of patients presented with bacteremia and was the second most common site of infection which appears to be higher than in the previous studies conducted by Borzio M et al and Caly et al. where the incidence was 21% and 13.8% respectively \[[@b7-cm-91-356],[@b17-cm-91-356]\].
UTI occurred in 14.5% of patients, this is in discrepancy with other studies \[[@b7-cm-91-356],[@b17-cm-91-356]\]. This difference may be explained by the fact that only culture positive cases were taken in the present study whereas in the other studies patients with leukocyturia alone were also included. It has been suggested that residual urinary volume and possible vesical dysfunction may be responsible for high incidence of UTI in cirrhotic patients especially in those with ascites \[[@b18-cm-91-356]\]. RTIs were seen in 15.5% of patients, frequency of which is consistent with other studies \[[@b7-cm-91-356],[@b17-cm-91-356]\]. This could be attributed to aspiration resulting from upper gastrointestinal bleeding and hepatic encephalopathy.
Positive cultures were obtained in 148 episodes of bacterial infections excluding CNNA. Gram negative bacteria constituted the largest group (74.3%) than Gram positive isolates. The high incidence of gram negative isolates may be related to the enhanced translocation of bacterial flora from the gut into the bloodstream and ascitic fluid and is generally encountered in patients with decompensated cirrhosis \[[@b17-cm-91-356],[@b6-cm-91-356]\]. We found that *E. coli* was frequently detected in patients with ascitic fluid infection, bacteremia and UTI which is similar to the results obtained by Fernandez et al \[[@b19-cm-91-356]\]. *Acinetobacter* and *Klebsiella* were isolated more commonly from patients with nosocomial pneumonia while *Mycobacterium tuberculosis* was the commonest organism implicated in community acquired RTI in the present series. All the nine patients from whom *Acinetobacter* was isolated from the tracheal aspirate were on ventilator support. *Streptococci* were isolated in patients with community acquired pneumonia and *pseudomonas* in patients with hospital acquired pneumonia. This difference in the organisms causing community acquired RTI may be related to the low prevalence of tuberculosis in western population.
Patients with decompensated cirrhosis are more prone for bacterial infections accounting for about 30%--50% deaths \[[@b19-cm-91-356],[@b20-cm-91-356]\]. We observed that infections occurred more frequently in patients with advanced cirrhosis. This is probably due to increased immunosuppression which puts them at greater risk of acquiring infections. The mortality was noted in 24% of the patients which was almost similar to the mortality rates of 23.3% and 22% reported previously by Caly et al. and Borzio et al. respectively \[[@b7-cm-91-356],[@b17-cm-91-356]\], suggesting that mortality from bacterial infections in cirrhotics has not changed in the current era. The high mortality rate was observed in subjects with RTI (38.7%) in the present study, this may be because majority of respiratory infections were hospital acquired and caused by multidrug resistant bacteria.
In a study conducted by Shih et al., in-hospital mortality was significantly high in patients with cirrhosis with bacteremia and was identified to be one of the independent risk factor for poor outcome (OD: 9.7; 95% CI: 1.9--50.6) in the presence of gastrointestinal bleeding \[[@b21-cm-91-356]\]. On the contrary, multivariate analysis in the present series identified AFI as a strong risk factor for reduced survival (OR: 2.81, 95% CI: 1.11--7.12). Most of the infections in our study were community acquired, while more deaths were noted among patients with hospital acquired infections. This could be attributed to infection with resistant gram-negative organisms and decreased efficacy of empirical antibiotic treatment in these patients. This finding is in agreement with a previous study by Fernandez et al. which included 669 infections from two series (2005--2007 and 2010--2011), the hospital mortality rate of nosocomial infections (25%--48% respectively) was significantly higher than observed in community acquired episodes (7%--21%) \[[@b19-cm-91-356]\]. It is therefore advisable to initiate appropriate antibiotics as early as possible according to the local epidemiological patterns in patients with hospital acquired infections.
Bacterial infections may also be a precipitating factor for hepatic encephalopathy and AKI. As described earlier, the possible mechanism is that bacterial infection induces excess cytokine production causing cardiovascular and endothelial dysfunction \[[@b22-cm-91-356],[@b23-cm-91-356]\]. In our study, both hepatic encephalopathy (P\<0.001) and AKI (P\<0.001) showed significant association with mortality. We observed that more deaths occurred among patients without UTI compared to the cases with UTI with a significant p-value which may be explained by the presence of infection at the other site in these patients that might have contributed to mortality.
Conclusion
==========
Based on these results, we conclude that ascitic fluid was found to be the most predominant site of infection. Gram negative bacterial infections were more common, with Enterobacteriaceae group being responsible for most of the infections. The type of infection, hepatic encephalopathy, ascitic fluid infection and acute kidney injury had a significant association with in-hospital mortality. Since most of the cirrhotic patients with bacterial infection are asymptomatic, high degree of suspicion is critical to avert complications and in the presence of above mentioned factors, early initiation of appropriate antibiotics would prevent further deterioration of liver disease.
Limitations
===========
There was selection bias where all patients during the study period were randomly selected and included. Only culture positive cases were taken except for culture negative variant of spontaneous ascitic fluid infection (CNNA). Subjects who had other evidence of infection with sterile cultures were not included in the study.
######
Distribution of infections according to Child Pugh class.
Type of infection (n, %) Child Class A Child Class B Child Class C Acquisition, n (%)
-------------------------- --------------- --------------- --------------- ---------------------------------------------------- ------------
AFI (74, 38.3% ) 1 (1.3%) 12 (16.2%) 61 (82.4%) 68 (92%)[\*](#tfn1-cm-91-356){ref-type="table-fn"} 6 (8%)
Bacteremia (48, 24.3%) 3 (6.1%) 14 (28.6%) 31 (65.3%) 41 (85.4%) 7 (14.6%)
RTI (31, 15.5%) 1 (3.2%) 10 (32.2%) 20 (64.5%) 21 (67.7%) 10 (32.3%)
UTI (28, 14.5%) 2 (7.1% ) 14 (50%) 12 (42.9%) 23 (82%) 5 (18%)
Others (12, 6.2%) \- \- \- 12 (100%) \-
Only 29 are culture confirmed infections;
AFI: Ascitic fluid infection; RTI: Respiratory tract infection; UTI: Urinary tract infection.
######
Common bacterial agents isolated from various infections.
Bacterial etiology Total (n) AFI Bacteremia RTI UTI Others
------------------------------ ----------- ----- ------------ ----- ----- --------
Escherichia coli 57 15 22 2 17 1
*Klebsiella pneumoniae* 27 2 9 9 4 3
*Acinetobacter* spp. 13 2 1 9 0 1
*Pseudomaonas aeruginosa* 5 1 0 4 0 0
*Enterobacter* spp. 3 0 1 1 0 1
*Aeromonas* spp. 2 2 0 0 0 0
MRSA 6 1 4 0 0 1
MSSA 7 1 2 2 1 1
*Enterococcus* spp. 10 2 1 1 5 1
BHS 8 3 5 0 0 0
AHS 3 1 2 0 0 0
*Streptococcus pneumoniae* 1 0 0 1 0 0
*Mycobacterium tuberculosis* 11 0 0 7 1 3
MRSA- Methicillin resistant *Staphylococcus aureus*; MSSA- Methicillin sensitive *Staphylococcus aureus*; BHS- Beta hemolytic streptococci, AHS- Alpha hemolytic streptococci; AFI: Ascitic fluid infection; RTI: Respiratory tract infection; UTI: Urinary tract infection.
######
Univariate analysis for factors predicting mortality in cirrhotic patients.
Variables Outcome p-value
------------------------------------------------------ ---------------- ----------------- ----------- ---------
Type of infection Community 94 (78.4) 26 (21.6) 0.001
Hospital 26 (59.4) 12 (31.6)
Bacteremia Present 34 (70.83) 14 (28.6) 0.376
Absent 96 (80) 24 (20)
AFI Present 51 (69) 23 (31) 0.053
Absent 69 (82.8) 15 (17.2)
RTI Present 19 (61.3) 12 (36.7) 0.073
Absent 101 (79.6) 26 (20.4)
UTI Present 26 (93) 2 (7) 0.021
Absent 94 (72.3) 36 (27.7)
Child Pugh Category A 6 (100) 0 \<0.001
B 50 (98) 1 (2)
C 64 (63.3) 37 (36.7)
Acute kidney injury Present 42 (58.3) 30 (41.7) \<0.001
Absent 78 (90.7) 8 (9.3)
Hepatic Encephalopathy Present 22 (48.9) 23 (51.1) \<0.001
Absent 98 (86.7) 15 (13.3)
UGI Bleed Yes 12 (60) 8 (40) 0.075
No 102 (77.3) 30 (22.7)
Albumin [\*](#tfn4-cm-91-356){ref-type="table-fn"} 2.38 ± 0.48 2.24 ± 0.53 0.163
Creatinine[\*](#tfn4-cm-91-356){ref-type="table-fn"} 1.48 ± 0.98 2.51 ± 1.35 \<0.001
Bilirubin^+^ 3.8 (1.8--6.2) 6.2 (3.3--12.3) 0.004
Expressed as mean ± SD,
**+** expressed as median (Q1, Q3); UGI bleed: Upper gastrointestinal bleed; AFI: Ascitic fluid infection; RTI: Respiratory tract infection; UTI: Urinary tract infection.
######
Multivariate analysis for factors predicting mortality in cirrhotic patients.
Variable p-value 95 % CI Adjusted Odds Ratio
------------------------ ---------- -------------- ---------------------
Type of infection p=0.052 0.11--1.01 0.33
AFI p=0.029 1.11--7.12 2.81
Hepatic encephalopathy p\<0.001 0.070--0.422 0.17
AKI p\<0.001 0.077--0.502 0.19
CI: Confidence interval, AFI: Ascitic fluid infection, AKI: Acute kidney injury.
| {
"pile_set_name": "PubMed Central"
} |
Introduction {#s1}
============
Exercise therapy has long been considered a promising strategy to ameliorate physical disability after stroke (Saposnik et al., [@B58]). However, neurological outcomes of post-stroke rehabilitation appear to differ according to the intensity of the exercise regimen that is used (Bell et al., [@B2]; Xing et al., [@B71]). Some previous studies have demonstrated that higher intensity exercise may yield better functional recovery and neuroplasticity (Linder et al., [@B40]; Luo et al., [@B43]; Andrews et al., [@B1]), while other studies have suggested that mild exercise results in superior neuroprotection and synaptic plasticity after stroke (Lee et al., [@B30]; Shih et al., [@B62]). These conflicting results underscore the principle that exercise intensity is an important determinant of post-stroke neurological outcomes. Therefore, clarifying the mechanisms that underlie an intensity-dependent effect of exercise on neurologic function may help direct the clinical application of exercise-based neurorehabilitation. Currently, these mechanisms have not been fully explored, and are consequently incompletely understood.
Recently, brain-derived neurotrophic factor (BDNF) has become the subject of increasing attention as a possible mediator of the neurological benefits of exercise. BDNF is an abundant growth factor that is involved in activity-induced neuroplasticity and is upregulated in the animal brain by exercise. The regulation of neuroplasticity depends on a complex set of interactions between a variety of neural proteins, including postsynaptic density 95 (PSD-95; Wang et al., [@B68]), synapsin I (SYN; Pan et al., [@B52]), growth-associated protein 43 (GAP-43), and microtubule-associated protein (also known as Tau; Biundo et al., [@B4]; Mercerón-Martínez et al., [@B47]; Pu et al., [@B55]). Changes in these neuroplastic factors are related to exercise-induced activation of BDNF (Kim and Leem, [@B27]; Belviranli and Okudan, [@B3]). Previous research has suggested a pivotal regulatory role for BDNF and its receptor, BDNF-tyrosine kinase B (TrkB), regarding neuroplasticity after physical exercise (Lee et al., [@B29]), mediated through the expression of the transcription factor cyclic AMP response element-binding protein (CREB; Hu et al., [@B24]). Moreover, activation of the BDNF/TrkB/CREB signaling pathway has also been shown to promote functional recovery after stroke (Liu H. et al., [@B41]). Taken together, these lines of evidence suggest that post-stroke exercise regimens such as the one used in this study may induce neuroplasticity and influence rehabilitative outcomes through the changes they provoke in the BDNF pathway.
Another factor that may be involved in determining the outcomes of post-stroke exercise regimens is hypoxia-inducible factor-1α (HIF-1α). Upregulation of HIF-1α by exercise has been reported to play a role in reducing infarct volumes following ischemia/reperfusion injury (Li C. et al., [@B32]), and in post-stroke neuroplasticity (Wu et al., [@B70]). Previous studies demonstrated that HIF-1α also induced the expression of BDNF (Shi et al., [@B61]; Nakamura et al., [@B49]; Helan et al., [@B22]), and thereby promoted neuroplasticity, reduced neuronal death, and improved neurological function in a rat model of ischemic stroke (Chen et al., [@B8]). HIF-1α has further been shown to stimulate the expression of the TrkB receptor (Martens et al., [@B45]), and the CREB receptor in various cancer cells (Yu et al., [@B72]). However, despite this considerable circumstantial evidence, previous studies have not yet explored the effect of HIF-1α on BDNF/TrkB/CREB pathway in improving synaptic plasticity following ischemia/reperfusion injury. Although the present study did not determine this relation, as the first step, we intended to assess the expression of HIF-1α and BDNF/TrkB/CREB proteins following ischemia/reperfusion injury. These results might suggest a potential association of these molecules and provide a base for our future study regarding the regulation of HIF-1α on the BDNF/TrkB/CREB pathway.
Materials and Methods {#s2}
=====================
Animals {#s2-1}
-------
A total of 150 adult male Sprague--Dawley rats (280--300 g, Vital River Laboratory Animal Technology Company Limited, Beijing, China) were used in this study. The protocol by which they were studied was approved by the Animal Care and Use Committee of Capital Medical University, and the study was conducted following the NIH Guide for the Care and Use of Laboratory Animals. Animals were randomly divided into three groups: middle cerebral artery occlusion (MCAO) without exercise (50), MCAO plus intense treadmill exercise (50), and MCAO plus mild treadmill exercise (50). Both exercise protocols were initiated after 24 h reperfusion, and animals in each group were sacrificed at days 3, 14, and 28 after reperfusion for further biochemical analysis.
Focal Cerebral Ischemia {#s2-2}
-----------------------
The animals were subjected to transient right MCAO according to the method we described previously (Li F. et al., [@B34]). Briefly, rats were anesthetized in a chamber using 3% isoflurane and a mixture of 70% nitrous oxide and 30% oxygen. Then rats were then transferred to a surgical table, where anesthesia was maintained with a facemask that delivered 1% isoflurane from a calibrated precision vaporizer, and poly-L-lysine-coated nylon (4.0) sutures were used to generate infarcts with minimal inter-animal variability. During the unilateral, 2-h MCAO procedure, cerebral blood flow (CBF), blood pCO~2~ and pO~2~, mean arterial pressure (MAP), and rectal temperature were monitored continuously. Rectal temperatures were maintained between 36.5°C and 37.5°C using a heating pad and a heating lamp. Ipsilesional ischemic cerebral hemispheres were used for molecular analysis.
Treadmill Exercise {#s2-3}
------------------
Animals were randomly assigned either to the exercise groups or the non-exercise control group. Exercise animals were run on a four-lane treadmill (ZS-PT-II, ZS Dichuang Instruments, Inc., Beijing, China), either at a constant speed of 30 m/min for 30 min each day (intense); or at 5 m/min for the first 10 min, 9 m/min for the second 10 min, and 12 m/min for the last 10 min on days 1 and 2, followed by 12 m/min on the third and subsequent days (mild). The mild exercise was begun at a shorter intensity (days 1--2) and ultimately ended with the final mild speed at 3 days and thereafter, such that low intensity was maintained throughout. This gradual start could not be achieved for the intense exercise group, however, as we would, in this case, have been unable to accurately assess the effects of high intensity in the short-term (3 days) when rats were sacrificed. Both exercise and non-exercise animals were housed in groups of three in standard cages for equal time.
Neurological Deficit {#s2-4}
--------------------
The modified scoring systems proposed by Zea Longa (5-point) and Belayev et al. ([@B80]) (12-point) were used to examine the severity of neurological deficits in rats before and after 24 h reperfusion (Li et al., [@B34]). After MCAO, rats with scores of 2 or below were considered to represent the unsuccessful establishment of the MCAO model and were consequently excluded (about 10%); exclusion was then confirmed on autopsy by lack of a core, indicating a faulty surgery.
Cerebral Infarct Volume {#s2-5}
-----------------------
At 3 days of ischemia and reperfusion in rats that underwent MCAO, brains were resected and cut into 2-mm-thick slices, which were then treated with 2,3,5-triphenyltetrazolium chloride (TTC; Sigma--Aldrich, St. Louis, MO, USA) for staining (Li et al., [@B36]), facilitating the use of an indirect method for calculating infarct volume to minimize error caused by edema.
Apoptotic Cell Death {#s2-6}
--------------------
For quantification of apoptosis-related DNA fragmentation, a commercial enzyme immunoassay was used to determine cytoplasmic histone-associated DNA fragments (Cell Death Detection ELISA; mlbio, Shanghai, China). The degree of apoptosis was quantified according to the amount of cytoplasmic histone-associated DNA fragments in the control and experimental groups at 3 days after physical exercise.
Neurobehavioral Tests {#s2-7}
---------------------
These tests included adhesive removal, beam balance, forelimb placing, grid walking, and Rota-rod performance (R03--1; Xin-Ruan Instruments, Inc., Shanghai, China), as assessed at days 1, 3, 7, 14, 21, and 28. The Morris water maze (ZS-II; ZS Dichuang Instruments, Inc., Beijing, China) was also employed, in our case at 24--28 days; this is following a previous report that showed no obvious suppressive effect on swimming at 24 days after the ischemic event (Ran et al., [@B57]). In the adhesive removal test, the tape was attached to the palmar surface of the forepaw, and the time taken for the first attempt to touch and to remove the tape was recorded. In the beam balance test, rats were placed on a narrow wooden beam (122 × 2.5 × 42 cm), and performance was scored from 0 to 6 (0 = no attempt to stay on the beam; 1 = attempted to stay on the beam but no movement; 2 = attempted to cross the beam but failed; 3 = crossed the beam with contralateral hindlimb slips \>50% of the time; 5 = crossed the beam with contralateral hindlimb slips \<50% of the time; 6 = crossed the beam without slips; see Ran et al. ([@B57]). In the forelimb placing test, rats were held gently with forelimbs close to the tabletop while the surface was lightly brushed using each side of their vibrissa. The ability of rats to place the preferred forelimb on the edge of the table in this context was recorded 10 times, and placing rates were calculated. In the grid walk test, rats were placed on a wire grid (100 × 25 × 50 cm) and allowed to walk from one end to the other; the total number of foot slips during this crossing was recorded. In the Rota-rod test, rats were placed on a rotating drum that accelerated from 4 to 40 rpm within 300 s; the time that the animals stayed on the rotating rod was then recorded. Finally, in the Morris water maze, rats were placed into a pool (diameter = 150 cm) at one of the four locations, and allowed to swim for 90 s to find a hidden platform (diameter = 10 cm); swim speed and the time taken to find the hidden platform were recorded using a camera positioned above the pool that transmitted data to an analysis system for calculation. Tests were conducted on non-consecutive days to mitigate possible confounding due to motor learning that might have occurred if these tests were performed in close succession.
Neuron Isolation and Flow Cytometry Assay {#s2-8}
-----------------------------------------
As described previously by our group (Chen et al., [@B6]), an adult brain dissociation kit (Miltenyi Biotec, Bergisch Gladbach, Germany) was used for neuron isolation. Briefly, rats were sacrificed at 3, 14, and 28 days after exercise and ipsilesional brains were finely cut, ground, and filtered through a 70-μm cell strainer (Miltenyi Biotec, Bergisch Gladbach, Germany) to obtain a single-cell suspension. Cell pellets (5 × 10^6^ cells) were stained with primary antibodies against Tau or GAP-43 (1 μg/1 × 10^6^ cells, rabbit anti-Tau, and rabbit anti-GAP43, Abcam, MA, USA) in darkness for 30 min at room temperature. Cells were washed three times with PBS and incubated with Alexa Fluor^®^ 488 fluorescein-conjugated secondary antibodies (Sigma, St. Louis, MO, USA) for 30 min at room temperature, and then washed again and analyzed on a FACS Calibur flow cytometer (Accuri C6, BD, San Jose, CA, USA) with Cell Quest software (BD, San Jose, CA, USA).
Protein Expression {#s2-9}
------------------
At 3, 14, and 28 days after initiation of the exercise regimens, rats were sacrificed for Western blot analysis. Tissue samples from the ipsilesional ischemic cerebral hemispheres of all experimental groups were harvested, and total protein extraction was performed using cell lysis solutions (Thermo Fisher Scientific, Inc., Waltham, MA, USA). Protein concentration was then determined by the BCA method. Electrophoresis (10% SDS-PAGE gel) was performed with 30 μg of protein per lane. Gel transfer to a PVDF membrane was performed under 200 V for 1 h. Membranes were blocked with 5% skimmed milk, followed by incubation with primary antibodies (1:1,000 rabbit anti-BDNF, rabbit anti-NGF, rabbit anti-PSD-95, rabbit anti-SYN, rabbit anti-Tau, and rabbit anti-GAP43, Abcam, MA, USA; 1:500 rabbit anti-HIF-1α, Santa Cruz Biotechnology, Inc., Santa Cruz, CA, USA) overnight at 4°C. The next day, membranes were washed three times and further incubated with a goat anti-rabbit IgG-HRP secondary antibody (1:1,000, Santa Cruz) at room temperature for 1 h. After washing, the ECL method was used to detect signals. Western blot images for each antibody were analyzed using an image analysis program (ImageJ 1.42, National Institutes of Health, Bethesda, MD, USA) to quantify protein expression according to relative image density.
Statistical Analysis {#s2-10}
--------------------
Statistical analyses were performed with SPSS Statistics for Windows, Version 17.0 (SPSS Inc., Chicago, IL, USA). Differences among groups were assessed using one-way ANOVA with a significance level of *p* \< 0.05. *Post hoc* comparison among groups was performed using the least significant difference method.
Results {#s3}
=======
Experimental Design and Physiological Parameters {#s3-1}
------------------------------------------------
Illustration of the experimental timelines ([Figure 1A](#F1){ref-type="fig"}). There were no significant differences in CBF ([Figures 1B,C](#F1){ref-type="fig"}), blood MAP, pO~2~, or pCO~2~ ([Table 1](#T1){ref-type="table"}) between these groups.
{#F1}
######
Physiological parameters during surgery.
Stroke Intense exercise Mild exercise
------------------------- ------------- ------------------ ---------------
MAP (mm Hg)
Prior to MCAO 86.8 ± 3.3 87.0 ± 3.5 87.5 ± 3.4
Onset of reperfusion 86.7 ± 2.4 86.8 ± 2.5 86.4 ± 3.5
After reperfusion 82.3 ± 2.9 85.3 ± 3.1 85.8 ± 4.0
pCO~2~ (mm Hg)
Prior to MCAO 44.8 ± 1.7 45.0 ± 2.2 47.2 ± 3.5
Onset of reperfusion 42.4 ± 2.7 45.4 ± 2.0 43.2 ± 2.1
After reperfusion 45.2 ± 4.0 44.3 ± 2.7 43.4 ± 4.4
pO~2~ (mm Hg)
Before MCAO 132.9 ± 5.9 139.5 ± 5.6 132.4 ± 5.7
Onset of reperfusion 135.2 ± 5.1 132.7 ± 5.6 134.7 ± 4.9
After reperfusion 134.1 ± 9.1 138.2 ± 6.4 132.4 ± 4.1
*MAP, mean arterial pressure; MCAO, middle cerebral artery occlusion*.
Brain Infarction and Correlates {#s3-2}
-------------------------------
A large infarct volume (42.5%) was seen following 2 h MCAO and 3 days reperfusion. Both mild (\*\**p* \< 0.01) and intense (\**p* \< 0.05) exercise significantly decreased infarct volumes (32.3% vs. 26.3%, respectively; [Figures 1D,E](#F1){ref-type="fig"}). Neurological deficits were detected by the 5- ([Figure 1F](#F1){ref-type="fig"}) or 12- ([Figure 1G](#F1){ref-type="fig"}) point score systems; compared to the control group, deficits were decreased significantly (\**p* \< 0.05) after either mild or intense exercise. Apoptotic cell death was detected at 3 days as described above; both mild and intense exercise significantly (\*\*\**p* \< 0.001) decreased cell death (0.07 and 0.14 ng/ml, respectively, vs. 0.22 ng/ml), but a further significant decrease was noted (\*\*\**p* \< 0.001) in the mild exercise group ([Figure 1H](#F1){ref-type="fig"}). Also, a significant (\*\*\**p* \< 0.001) increase in weight was seen in both exercise groups, with mild exercise rats demonstrating additional gain ([Figure 1I](#F1){ref-type="fig"}).
Functional Outcomes {#s3-3}
-------------------
As shown in [Figure 2A](#F2){ref-type="fig"}, the time taken to fall off the grid was significantly reduced after both mild (*p* \< 0.01) or intense (*p* \< 0.05) exercise as compared to rest at 3, 7, 14, 21, and 28 days ([Figure 2A](#F2){ref-type="fig"}); this reduction was significantly more pronounced in mild exercise rats on 7, 21, and 28 days. Similar results were observed in beam balance ([Figure 2B](#F2){ref-type="fig"}), Rota-rod ([Figure 2C](#F2){ref-type="fig"}), adhesive removal ([Figures 2D--E](#F2){ref-type="fig"}), and forelimb placing tests ([Figure 2F](#F2){ref-type="fig"}). On assessment using the Morris water maze ([Figures 2G--J](#F2){ref-type="fig"}) at 24--28 days, exercised rats demonstrated significantly shorter latency to locate the hidden platform as compared to rested controls, with mildly exercised rats attaining significantly better outcomes ([Figures 2G--H](#F2){ref-type="fig"}). Exercised rats spent more time (\**p* \< 0.05) in the target quadrant to find the hidden submerged platform than rested rats ([Figure 2I](#F2){ref-type="fig"}). In contrast, there was no significant difference between groups concerning swim speed, suggesting similar gross motor skills ([Figure 2J](#F2){ref-type="fig"}). These results demonstrate the significant role of exercise generally, and mild exercise in particular, in the long-term recovery of sensorimotor functions and spatial learning capability after ischemia/reperfusion injury.
{#F2}
Neuroplasticity {#s3-4}
---------------
Flow cytometry assay demonstrated that both mild (\*\**p* \< 0.01) and intense (\*\**p* \< 0.01) exercise increased expression of Tau ([Figures 3A,C](#F3){ref-type="fig"}) and GAP-43 ([Figures 3B,D](#F3){ref-type="fig"}) at 3, 14, and 28 days; significantly more Tau expression was seen in mildly exercised rats (\*\**p* \< 0.01). Also, compared to the control group, mild and intense exercise both significantly increased protein expression of Tau, GAP-43, and PSD-95 at 3, 14, and 28 days. Compared to the control group, levels of Tau (\*\**p* \< 0.01, [Figure 3E](#F3){ref-type="fig"}), GAP-43 (\*\*\**p* \< 0.001, [Figure 3F](#F3){ref-type="fig"}), PSD-95 (\**p* \< 0.05, [Figure 3G](#F3){ref-type="fig"}), and SYN ([Figure 3H](#F3){ref-type="fig"}) were found by Western Blot to be increased in mildly exercised rats at 3, 14, and 28 days; the same results were also seen in the intense exercise group. Taken together, these results demonstrate the capacity of exercise to augment neuroplasticity after ischemia/reperfusion injury.
{#F3}
Expression of HIF-1α, BDNF, TrkB, and CREB {#s3-5}
------------------------------------------
Both exercise protocols yielded a significant increase in levels of these proteins at 3, 14, and 28 days. Compared to the control group, levels of HIF-1α (3 days, \**p* \< 0.05; 28 days, \*\**p* \< 0.01, [Figure 4A](#F4){ref-type="fig"}), BDNF ([Figure 4B](#F4){ref-type="fig"}), NGF (14 days and 28 days, \*\*\**p* \< 0.001, [Figure 4C](#F4){ref-type="fig"}), TrkB (\**p* \< 0.05, [Figure 4D](#F4){ref-type="fig"}), and CREB (\*\**p* \< 0.01, [Figure 4E](#F4){ref-type="fig"}) were significantly increased in both the mild and intense exercise groups. Levels of HIF-1α (3 days, \**p* \< 0.05, [Figure 4A](#F4){ref-type="fig"}), BDNF (14 days, \*\**p* \< 0.01, [Figure 4B](#F4){ref-type="fig"}), NGF (3 days, \*\**p* \< 0.01, [Figure 4C](#F4){ref-type="fig"}), TrkB (14 days, \**p* \< 0.05, [Figure 4D](#F4){ref-type="fig"}), and CREB ([Figure 4E](#F4){ref-type="fig"}) were further increased in mildly exercised rats. These results demonstrate the alterations in HIF-1α, BDNF, TrkB, and CREB in association with synaptic plasticity following ischemia/reperfusion injury.
{#F4}
Discussion {#s4}
==========
The results obtained in this study, confirmed the augmentation in neuroplasticity in the ipsilesional hemisphere and functional outcomes provided by physical exercise after ischemic brain injury. Specifically, we showed that both mild and intense exercise regimens reduced brain infarct volume and apoptotic cell death, and improved motor and cognitive function at 3, 14, and 28 days after ischemia/reperfusion injury. The early improvement in infarct volume seen in these results aligned with a previous meta-analysis, in which infarct volume was reduced most effectively by exercise administered with the shortest delays after ischemia (Egan et al., [@B15]); data from our group derived from pre-conditioning experimentation suggest that this may be related to the capacity of exercise to mitigate inflammatory damage during reperfusion (Ding et al., [@B14]), with the caveat that the exercise initiation too early after ischemia may be detrimental (Li et al., [@B33]). More recent work by our group further substantiated these findings by demonstrating that exercise improved glycometabolism in the ischemic area and decreased neuroinflammation and apoptosis as early as 1 day post-stroke, and also at 3 days (Shen et al., [@B60]; Li et al., [@B33],[@B36],[@B37]). These findings suggest that it is beneficial to initiate exercise early after ischemia/reperfusion, as was done in the present study.
Furthermore, our biochemical analyses showed that the expression of synaptic plasticity proteins (Tau, GAP-43, and PSD-95) and their potential upstream regulators (HIF-1α, BDNF, NGF, TrkB, and CREB) were significantly increased after exercise. These findings suggest that long-term physical exercise may induce synaptic plasticity through the HIF-1α and BDNF/TrkB/CREB pathway. Brain synaptic regeneration may be related to elevated levels of GAP-43 or Tau proteins, and exercise has been shown to increase expression of GAP-43 in the ischemic area in rats with cerebral ischemia/reperfusion injury (Mizutani et al., [@B81]). Exercise-induced GAP-43 has been associated with augmented hippocampal neuroplasticity (Liu et al., [@B42]; Rahmati and Kazemi, [@B56]) in a process that appears to be dependent on BDNF maturation and the TrkB signaling promoted by mature BDNF (Ding et al., [@B12]). Similarly, exercise has been shown to promote axonal recovery as assessed by the upregulation of Tau and GAP-43 and is associated with functional improvement after cerebral infarction (Li et al., [@B31]). Short-term moderate exercise also appears to be capable of inducing the BDNF-regulated marker of hippocampal and structural plasticity known as SYN (Ferreira et al., [@B16]). One critical component of synaptic plasticity, the dynamic reorganization of the PSD protein scaffold (Coley and Gao, [@B9]), is augmented by yet another synaptic protein, PSD-95. PSD-95 is constituent of the postsynaptic membrane that plays a key role in the plasticity and structure of the excitatory chemical synapse (Wu et al., [@B69]), and multiple studies have associated physical exercise with its induction (Jung and Kim, [@B25]; Pan et al., [@B52]). A series of studies have shed light on the relationship between these factors by demonstrating that BDNF/NGF participate in promoting neuroplasticity for motor rehabilitation after focal cerebral infarction (Matsuda et al., [@B46]; Mizutani et al., [@B48]; Mang et al., [@B44]); BDNF was reported to be induced by exercise, and may regulate the expression of synaptic proteins including GAP-43 (Liu W. et al., [@B42]), SYN (Ferreira et al., [@B16]), PSD-95 (Li X. et al., [@B39]) and Tau (Kerling et al., [@B26]). This research indicates that synaptic plasticity after stroke is determined, at least in part, by the induction and upregulation of axonal or synaptic proteins that, in our study, were found to be increased in both exercise cohorts.
Also, our work helps to elucidate the role occupied by another factor: HIF-1α. The present results are consistent with previous studies showed that the upregulation of HIF-1α promoted synapse plasticity by mediating synaptic markers (Li G. et al., [@B38]), and played a beneficial role in post-stroke exercise inducing angiogenesis and neurogenesis (Li C. et al., [@B32]). Recent work demonstrates that exercise could activate the cerebral motor and cognitive circuits by increasing the expression of HIF-1α, suggesting a regulatory role of HIF-1α in exercise-enhanced neuroplasticity (Halliday et al., [@B19]). Additionally, previous studies also indicated that HIF-1α regulated BDNF (Chen et al., [@B8]), TrkB (Martens et al., [@B45]), and CREB (Yu et al., [@B72]). Furthermore, the activation of the BDNF/TrkB/CREB pathway has been reported to contribute to the reduction in cerebral ischemic injury and improvement in functional recovery after stroke (Liu H. et al., [@B41]). Therefore, the enhanced expression of HIF-1α and BDNF/TrkB/CREB proteins after ischemia/reperfusion injury in the present study suggest that HIF-1α might be involved in the BDNF pathway, known to promote synaptic plasticity. Although, we did not explicitly study the regulation of HIF-1α on the BDNF/TrkB/CREB pathway, our results suggest a potential link between the molecules. Our findings could be a basis to further clarify the participation of HIF-1α in BDNF-mediated synaptogenesis.
The results obtained in this experiment are in agreement with previous studies showing that exercise improves motor and cognitive function (Chen et al., [@B7]; Palasz et al., [@B50]; Tíglás et al., [@B64]). Our results also, by suggesting the adequacy of milder post-stroke exercise, address the controversy in the literature regarding the dependence of these beneficial effects on the intensity of the prescribed exercise regimens (Han et al., [@B20]). A previous investigation supports the findings in the present study, by reporting that higher intensity exercise can exacerbate brain injury after ischemia, whereas the effects of mild intensity training were found to be encouraging (Scopel et al., [@B59]). Additional work has shown that cell proliferation in the dentate gyrus (Kim et al., [@B28]), spatial memory function (Lee et al., [@B30]), and synaptic plasticity (Shih et al., [@B62]) were more remarkable with mild rather than with heavy exercise after ischemia. In contrast, high-intensity intermittent exercise (HIT) was reported to be superior to moderate-intensity continuous training (MCT) in improving neural plasticity after cerebral ischemia in rats (Pin-Barre et al., [@B54]; Luo et al., [@B43]). HIT had a similar effect on cardiac troponin-I as workload-matched continuous exercise in endurance runners, which could be considered as high intensity exercise (Li et al., [@B35]) and was reported to be acceptable in stroke patients (Boyne et al., [@B5]). In agreement with these studies, our present results support the beneficial effect of intense exercise, but also indicate that mild exercise is not necessarily worse; and may be adequate to augment neuroprotection and neuroplasticity after stroke. Further investigation is needed in order to optimize exercise intensity post-stroke and determine which intensity may be most beneficial for neurorehabilitation. Furthermore, a standardized definition of exercise intensities may be necessary in order to homogenize methodologies and better compare results between studies in the future.
The protocol utilized in the present study to define exercise intensities was based on prior works (Curry et al., [@B11]; Zhang et al., [@B73]). We employed as a standard of achieved exercise training intensity the speed at which rats could not run any longer due to fatigue within three minutes after the onset of exercise. The therapeutic doses of physical exercise training used in our study were calculated as 40% of this maximum velocity in the case of mild exercise training, which amounted to approximately 15 m/min, and 80% in the case of intense exercise training, which was about 32 m/min (Zhang et al., [@B73]). To further increase the difference between our categories, we reduced speed in the mild group to a maximum of 12 m/min as previous studies (Tian et al., [@B65]; Zhang P. et al., [@B74]; Zhang et al., [@B75]; Tang et al., [@B63]). For the high-intensity group, we selected 30 m/min because we have employed this speed in previous work, in which we found that it reduced brain damage (Ding et al., [@B13]), blood-brain barrier dysfunction (Guo et al., [@B18]), and brain inflammation in stroke (Curry et al., [@B11]). Recently, studies used physiological parameters, transferable to patients, to determine high and low intensity in rats (Pin-Barre et al., [@B54]; Luo et al., [@B43]). To agree with the principle that a physical exercise regimen is reproducible (Gronwald et al., [@B17]), our further exercise procedure would focus on a physiological indicator such as the lactate threshold. Another key finding from our previous work was the influence of initiation time on post-stroke rehabilitation outcomes: initiation 6 h post-stroke exacerbated brain damage, but this was avoided when exercise was deferred for 1--3 days (Li et al., [@B33],[@B36]). Therefore, the initiation time of 24 h after stroke was selected in this study.
In this study, we intended to observe the protective effects of post-stroke exercise on brain infarct at 3 days as previous studies did. Recent work by our group substantiates these findings by demonstrating exercise-improved glycometabolism, decreased neuroinflammation, and apoptosis (Li et al., [@B33],[@B36],[@B37]). The present results were largely supported by the findings of other groups, which demonstrated that exercise accelerated CBF (Pianta et al., [@B53]), decreased infarct volume (Tian et al., [@B65]; Zhang Y. et al., [@B76]; Pan et al., [@B51]) and improved functional outcomes (Pianta et al., [@B53]). In contrast, a few studies have reported no neuroprotective effects of exercise on the neurological deficit and infarct volume after stroke within 3 days (Matsuda et al., [@B46]; Cui et al., [@B10]). Future studies are necessary to fully elucidate the effect of post-stroke exercise on brain injury.
Some limitations are important to consider when interpreting the results of our study. The different constant-intensity regimens were used to demonstrate the concept of mild or intense exercise. Although the exercise protocol in rats cannot be directly transferred to patients, in the present study, as the first step, we intended to use these two exercise procedures to investigate the mechanism underlying the dose-dependent benefit of exercise on recovery after stroke. More careful design is on the way to develop a translational strategy that better applies to human stroke patients (Gronwald et al., [@B17]). For this purpose, our future work will focus on a connection between animal and clinical exercise procedures by controlling the workload of each training regimen as well as using lactate threshold or oxygen uptake as an indicator. Regarding the interpretation of the neurobehavioral test, possibly the multiple Rotarod tests served as a training procedure that may have influenced our results. Given our study's focus on the effect and mechanism of different exercise doses on rehabilitation, in each group, the rats received the same test. It is unlikely that this small amount of possible training would change the direction of our results.
In conclusion, this study demonstrates the positive effect on brain injury, functional outcome, and neuroplasticity conferred by both mild and intense long-term treadmill exercise. Additionally, our results suggest that intense exercise did not confer further benefit when compared with its milder counterpart, thus mild exercise may be adequate and sufficient to elicit rehabilitative benefits post-stroke. Moreover, the results may provide a base for our future study regarding the regulation of HIF-1α on the BDNF/TrkB/CREB pathway in the biochemical processes underlying post-stroke synaptic plasticity.
Data Availability Statement {#s5}
===========================
All datasets generated for this study are included in the article/[Supplementary Material](#SM1){ref-type="supplementary-material"}.
Ethics Statement {#s6}
================
The animal study was reviewed and approved by the Animal Care and Use Committee of the Capital Medical University.
Author Contributions {#s7}
====================
FL conducted the animal and biochemical experiments employed in this research. FL, XG, CH, CS, and YD were instrumental in preparing or revising the manuscript. YD was responsible for the experimental design, in addition to assisting with manuscript preparation and revision.
Conflict of Interest {#s8}
====================
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
We thank Sainan Wang, Menglei Liu, Yanlong Zhao, and Chencheng Zhang for technical assistance.
**Funding.** This work was partially supported by the National Natural Science Foundation of China (FL, 81802231 and XG, 81871838), the Organization Department of Beijing talents project (FL, 2018000082595G485), the Beijing Tongzhou District Financial Fund, the Science and Technology Plan of Beijing Tongzhou District (FL, KJ2020CX002 and KJ2019CX004-07).
Supplementary Material {#s9}
======================
The Supplementary Material for this article can be found online at: <https://www.frontiersin.org/articles/10.3389/fncel.2020.00186/full#supplementary-material>.
######
Click here for additional data file.
[^1]: Edited by: Claudio Rivera, Aix-Marseille Université, France
[^2]: Reviewed by: Jérôme Laurin, Aix-Marseille Université, France; Anne Sophie Tessier, INSERM U1093 Cognition, Action et Plasticité Sensomotrice, France
[^3]: **Specialty section**: This article was submitted to Cellular Neuropathology, a section of the journal Frontiers in Cellular Neuroscience
| {
"pile_set_name": "PubMed Central"
} |
Background {#Sec1}
==========
Hepatitis C virus (HCV) is a major cause of chronic liver disease that can potentially progress to cirrhosis and hepatocellular carcinoma (HCC) \[[@CR1]--[@CR3]\]. It has been estimated that about 130 million to 150 million people worldwide have HCV infection and that HCV causes about 500,000 deaths each year \[[@CR4]\]. In addition to liver damage, HCV infection is associated with several extrahepatic disorders \[[@CR5]\]; among them, the striking association between HCV infection and mixed cryoglobulinaemia (MC), an immune complex-mediated vasculitis of small and medium vessels, has been clearly established \[[@CR6], [@CR7]\]. On the basis of B-cell clonal expansion that can be detected in the majority of patients, MC can also be considered a low-grade, indolent lymphoproliferative disorder \[[@CR8]\].
Clinically, MC is characterised by the classic Meltzer's triad of symptoms (purpura, weakness and arthralgia). The clinical spectrum of MC is quite variable, however, ranging from skin lesions, including recurrent purpuric eruptions, leg dyschromatosis and torpid ulcers, to peripheral neuropathy and renal damage. In a minority of patients, multi-organ involvement may result in life-threatening conditions \[[@CR9], [@CR10]\]. In our experience, compared with patients with chronic HCV infection without cryoglobulinaemia, HCV-positive patients with MC exhibit a lower rate of liver fibrosis and a lower cumulative probability of developing cirrhosis and HCC, but a more frequent occurrence of renal failure, neurologic impairment and B-cell non-Hodgkin lymphoma (B-NHL). Even so, the 15-year survival rates in these groups have been found to be roughly similar \[[@CR11]\].
The therapeutic approach to MC has conventionally been based on the use of steroids and immunosuppressive drugs. Interferon (IFN)-based treatment was introduced in 1987, before the identification of HCV as the major etiologic factor \[[@CR12], [@CR13]\]. Subsequently, the antiviral combination of pegylated interferon-α plus ribavirin (pIFN-α + RBV) was employed to obtain a sustained virological response (SVR), and for many years, this combination has been considered the standard of care for the treatment of chronic HCV infection, capable of inducing SVR and clinical remission of vasculitis in about 78% of patients with MC \[[@CR14], [@CR15]\]. A multicentre, open-label study showed viral response rates ranging from 36% to 64% according to viral genotype and a clinical response in 88% of the patients \[[@CR16]\]. The association of rituximab (RTX), an anti-CD20 monoclonal antibody aimed at inducing an effective B-cell depletion, resulted in further benefits, both in patients refractory to antiviral therapy \[[@CR17], [@CR18]\] and in those previously untreated \[[@CR19]\]. In our experience, the addition of RTX to the pIFN-α + RBV combination has resulted in a 33% complete response (CR) (i.e., virological, immunological and molecular response) \[[@CR19]\].
However, the efficacy of therapy in terms of viral clearance has remained unsatisfactory and is burdened with remarkable side effects. The use of first-generation protease inhibitors, such as boceprevir or telaprevir, associated with pIFN-α + RBV in a triple-combination therapeutic regimen resulted in significantly higher response rates in patients with chronic HCV infection without and with MC. Serious adverse events were recorded in almost 50% of patients, however \[[@CR20]\].
The introduction of all-oral, direct-acting antiviral agents (DAAs) has dramatically changed the treatment of chronic HCV and consequently the prognosis of these patients. According to the viral molecular target, DAAs include inhibitors of non-structural proteins NS3/NS4 (simeprevir, paritaprevir and grazoprevir), NS5A (ledipasvir \[LDV\], daclatasvir \[DCV\], ombitasvir, elbasvir and velpatasvir) and NS5B (sofosbuvir \[SOF\] and dasabuvir) that are already available in Italy, except elbasvir, grazoprevir and velpatasvir, for which approval is pending. These drugs have been shown to be capable of eradicating the infection in over 90% of patients with chronic HCV, with limited variations related to viral genotype \[[@CR21]\]. An additional advantage of DAAs is the possibility of avoiding IFN-related side effects. However, owing to the high costs of these drugs, it has been necessary to fix priority criteria among patients to be treated. Both the American and the European associations for the study of liver diseases, as well as the Italian Medicines Agency (AIFA), have established that extrahepatic manifestations of HCV infection, such as cryoglobulinaemic vasculitis and lymphoproliferative disorders, should be recognised as priority indications for treatment with DAAs \[[@CR22], [@CR23]\].
So far, only a few papers have been published on the efficacy and safety of the use of DAAs in patients with MC. Saadoun et al. \[[@CR24]\] reported a 74% rate of sustained virological response 12 weeks after therapy completion (SVR12) and an 87.5% clinical response rate in 24 patients with HCV-related MC treated with an SOF plus RBV combination. In another study, carried out with 12 patients with MC (7 of whom had renal involvement) treated with SOF-based regimens, the SVR12 rate was 83% and was associated with improvement in serum creatinine and reduction in proteinuria in patients with glomerulonephritis. In addition, an overall reduction or disappearance of cryoglobulins was recorded in 89% of the patients \[[@CR25]\]. Mention should also be made of a prospective double-centre (Florence/Rome) study of a cohort of 44 consecutive patients with HCV-associated MC, 2 of whom had evolved into an indolent NHL with monoclonal B-cell lymphocytosis. Following guideline-tailored therapy with DAAs, an SVR was demonstrated 12 and 24 weeks post-treatment in parallel with clinical response in all patients. A striking reduction in the mean cryocrit value was revealed at SVR12, and an even greater reduction was observed for sustained virological response 24 weeks after therapy completion (SVR24). A partial vasculitis response and a roughly 50% reduction of cryocrit were detected in the two patients with lymphoma. Adverse events were rather frequent but usually mild \[[@CR26]\]. More recently, Bonacci et al. \[[@CR27]\] reported a large prospective study of 64 cryoglobulinaemic patients with HCV infection, divided into those with symptomatic (55% of cases) cryoglobulinaemia and those with asymptomatic cryoglobulinaemia (45% of cases). Sixteen percent of the patients received an IFN-based DAA combination, whereas IFN-free regimens were administered to the remaining patients. Overall, 94% of patients showed SVR12, and 71% achieved a clinical CR. In 10 of 13 patients, concomitant immunosuppressive therapy was reduced or withdrawn. Forty-eight percent of patients achieved an immunological response, defined as disappearance of circulating cryoglobulins and normalisation of complement and/or rheumatoid factor (RF). However, despite SVR12 in 52% of patients, cryoglobulins or complement consumption or RF activity persisted, a low baseline cryocrit level being the only factor associated with an immunological CR.
We report the results of our single-centre, prospective study of a cohort of patients with MC whose main features can be summarised as follows: (a) unresponsiveness to or relapse after the previous standard-of-care treatment consisting of pIFN-α + RBV combination, or previously untreated patients; (b) the assignment to receive variable combinations of DAAs according to the guidelines of the AIFA eligibility criteria; and (c) a post-treatment evaluation at weeks 12 and 24, with a prolonged follow-up until 12 months in some cases.
Methods {#Sec2}
=======
Patients {#Sec3}
--------
Twenty-two consecutive HCV-positive patients with MC were enrolled in this study, with the aim being to assess both the efficacy and the safety of all-oral, IFN-free DAAs. Thirteen patients were non-responsive to or relapsed after previous pIFN-α + RBV combination therapy, whereas the remaining nine patients were treatment-naïve. Written informed consent was obtained from all patients. Because treatments were administered on-label, no approval by the ethics committee was required, and the study was conducted according to the principles of the Declaration of Helsinki.
Clinical and virological evaluation {#Sec4}
-----------------------------------
All patients were thoroughly examined according to validated criteria reported elsewhere \[[@CR28]\]. Clinical features included the presence of purpura, fatigue and arthralgia, associated or not with skin ulcers. Renal damage and/or peripheral nervous system involvement were also looked for. Circulating cryoglobulins were detected as previously described \[[@CR29]--[@CR31]\] and immunochemically characterised by immunofixation, and their amounts were quantified as cryocrit (%). Serum HCV RNA levels were assessed using the AmpliTaq real-time polymerase chain reaction system with a detection threshold of 15 IU/ml (Roche Molecular Systems Inc., Pleasanton, CA, USA), whereas HCV genotyping was performed using the VERSANT HCV genotype 2.0 assay (LiPA) (Siemens Healthcare Molecular Diagnostics, Berkeley, CA, USA).
Treatment schedules {#Sec5}
-------------------
Patients admitted to treatment had previously been examined by transient elastography (FibroScan; Echosens, Paris, France) for the assessment of liver fibrosis grade. Antiviral therapy was prescribed according to the AIFA eligibility criteria, HCV genotype, grade of liver fibrosis and prior treatment experience. For patients with RBV treatment, dose reduction was considered in accordance with technical schedule. The following regimens were employed: (a) daily SOF (400 mg) plus RBV (1000 to 1200 mg, depending on whether the patient's body weight was ≤75 or \>75 kg); (b) ombitasvir/paritaprevir/ritonavir (12.5 mg/75 mg/50 mg) plus dasabuvir (250 mg) (3D combo) twice daily, with or without weight-based RBV; (c) SOF/LDV (400 mg/90 mg) once daily; and (d) SOF (400 mg) plus DCV (60 mg) (SOF/DCV) once daily.
Biochemical evaluation {#Sec6}
----------------------
In addition to clinical examination, baseline evaluation included aspartate transaminase/alanine transaminase (ALT) values, haemoglobin level, platelet count, RF activity, serum levels of C3 and C4 complement component fractions, and serum protein electrophoresis. The same parameters were evaluated every 4 weeks during treatment and 12 weeks after therapy completion to establish whether SVR12 had been achieved.
Endpoints {#Sec7}
---------
Primary efficacy endpoints were (a) SVR12 (virological response), (b) regression of clinical manifestations of vasculitis (clinical response) and (c) cryocrit reduction by at least 50% or disappearance of cryoglobulins (immunological response). CR was defined as the occurrence of all primary endpoints, partial response (PR) as the occurrence of SVR12 with either immunological or clinical response, and no response (NR) as the lack of all criteria.
Statistical analyses {#Sec8}
--------------------
To assess the statistical power and significance of CR proportion (π~1~), criterion power analysis was performed using G\*Power 3.1 (Heinrich Heine University, Düsseldorf, Germany) with a one-tailed binomial test (one-sample case), assuming as the constant proportion the frequency of CR to pIFN-α + RBV therapy in patients with cryoglobulinaemia (π~0~ = 0.33) \[[@CR19]\]. Sensibility (statistical power) was fixed at 0.80. This was a one-tailed test because we assumed a greater antiviral efficacy of DAAs as compared with pIFN-α + RBV therapy. To evaluate the frequencies and differences of virological and immunological responses, as well as C4 complement fraction, RF and ALT level normalisation, continuous variables were categorised as dichotomous variables and statistically analysed by cross-tabs and χ^2^ test.
Continuous variables were first subjected to descriptive statistical analyses to verify the normality of data (Kolmogorov-Smirnov test). The statistical significance of variable changes across the study was assessed by a general linear model (GLM) repeated measures analysis. Three factors (baseline, end of treatment \[EoT\] and SVR12 time points) and four variables (cryocrit, RF, C4 complement fraction and ALT levels) were taken into consideration. HCV RNA serum levels were excluded from the GLM because the test results for HCV were negative in all patients after treatment. The Huynh-Feldt test and group comparisons by simple contrasts were used for univariate repeated measures analysis of variance to evaluate EoT and SVR12 time point variations against the baseline. Two-tailed α probability (*p* value), estimation of the effect size in the sample (η^2^), and statistical power were reported, with *p* \< 0.05 being considered significant. Statistical analyses were performed by using IBM SPSS Statistics version 20.0 software (IBM, Armonk, NY, USA).
Results {#Sec9}
=======
Fourteen patients were treated with the SOF/RBV regimen for 12--24 weeks according to viral genotype and fibrosis stage, and seven patients with genotype 1 infection were treated with such a regimen before the availability of more effective combinations that all guidelines considered as optimal. Three patients received a 3D combo regimen, four with SOF/LDV and one SOF/DCV combination. Thirteen of twenty-two patients had experienced treatment failure in previous IFN-based therapeutic regimens. No statistically significant relationship was found between previous treatment failure and response to DAAs.
Table [1](#Tab1){ref-type="table"} summarises the main clinical, biochemical and virological characteristics of the patients enrolled. The subjects' mean age was 66.9 years, ranging from 46 to 84, and the male/female ratio was 8/14. All patients were viraemic with HCV RNA levels ranging from 2.1 to 7.7 logarithmic IU/ml. The mean cryocrit value was 1.8%, ranging from 0.5% to 4%. ALT levels ranged from 13 to 668 IU/L (mean ± SD 104.8 ± 144.7), and C4 complement fraction ranged from 0 to 24 mg/dl (mean ± SD 9.6 ± 7.3). RF activity ranged from 0 to 530 IU/ml (mean ± SD 69.3 ± 119.1). Cryoglobulins were immunochemically characterised as type 2 MC (monoclonal immunoglobulin M \[IgM\]/polyclonal IgG) in 21 patients and as type 3 MC (polyclonal IgM/polyclonal IgG) in the remaining 1 patient.Table 1Baseline clinical, virologic and laboratory parameters of 22 patients with chronic hepatitis C virus infection with mixed cryoglobulinaemiaParametersDataAge, years, mean ± SD (range)66.9 ± 11.2 (46--84)Female sex, *n* (%)14 (63.6)Serum HCV RNA, logarithmic IU/ml, mean ± SD6.02 ± 1.2 (2.1--7.7)HCV genotype, *n* (%) GT114 (63.6) GT27 (31.8) GT31 (4.6)Cryocrit, %, mean ± SD (range)1.8 ± 1.3 (0.5--4)Cryoglobulin type, *n* (%) II21 (95.4) III1 (4.6)C3, mg/dl (n.v. 90--180) (mean ± SD)93.4 ± 15.3 (51--118)C4, mg/dl (n.v. 10--40) (mean ± SD)9.6 ± 7.3 (0--24)RF, IU/ml (n.v. 10--15) (mean ± SD)69.3 ± 119.1 (0--530)Clinical features, *n* (%) Meltzer's triad22 (100) Glomerulonephritis4 (18.1) Peripheral neuropathy2 (9.1)Liver involvement ALT, IU/L (n.v. 12--78) (mean ± SD)104.8 ± 144.7 (13--668) Cirrhosis, *n* (%)12 (54.5)HCV-related tumours, *n* (%) Hepatocellular carcinoma2 (10) Non-Hodgkin lymphoma1 (5) Small lymphocytic lymphoma1 (5)*Abbreviations: ALT* Alanine transaminase, *C3* and *C4* Complement components, *GT* Genotype, *HCV* Hepatitis C virus, *RF* Rheumatoid factor, *n.v.* Normal values
Meltzer's triad of symptoms (purpura, arthralgia and weakness) was consistently present in all patients. Cryoglobulinaemic glomerulonephritis was recorded in four patients (18.2%) and severe peripheral neuropathy in two patients (9.1%). Autoimmune haemolytic anaemia and idiopathic thrombocytopenic purpura were diagnosed in two patients each. In addition, four patients (18.2%) showed a monoclonal IgG component (two with κ light chains and two with λ light chains) on serum protein electrophoresis, and three patients received a diagnosis of NHL, one with stage 4 small lymphocytic lymphoma (SLL) who received SOF/RBV therapy for 24 weeks followed by four RTX infusions at 1-week intervals, one with marginal zone B-NHL that extended to ocular adnexa previously treated with pIFN-α + RBV + RTX, and one with diffuse centrocytic B-NHL who had received chemotherapy 2 years before that had resulted in CR. Finally, two patients who had previously received a diagnosis of HCC were treated with locoregional ablation therapies with no evidence of recurrence on the basis of contrast-enhanced ultrasound and computed tomography, both before starting therapy and following its completion. At FibroScan evaluation, 12 patients showed advanced liver fibrosis (6 with F3 and 6 with F4 fibrosis and no signs of decompensation), whereas an F1 or F2 score was recorded in the remaining 10 patients.
Four weeks after starting therapy, HCV RNA determination was found to be almost invariably below the lower limit of detection. A rapid and progressive reduction of ALT levels until normalisation was also observed. These strikingly positive results were confirmed throughout the length of treatment. Consequently, all patients reached an EoT response and SVR12 (Fig. [1a](#Fig1){ref-type="fig"}) in step with persistent ALT normalisation (Fig. [1b](#Fig1){ref-type="fig"}). Interestingly, seven patients with genotype 1 infection who were given suboptimal SOF/RBV combination therapy also showed SVR12.Fig. 1Distribution of hepatitis C virus (HCV) RNA (**a**) and alanine transaminase (ALT) (**b**) levels in 22 HCV-positive patients with cryoglobulinaemia treated with direct-acting antiviral agents at baseline, at week 4, at the end of treatment (EoT) and 12 weeks after treatment. *SVR12* Sustained virological response 12 weeks after therapy completion
An immunological response was observed at EoT in 18 patients (81.8%). In 12 patients (54.5%), cryoglobulins completely disappeared, whereas in 6 patients (27.3%), a cryocrit reduction ≥50% of basal values was observed (Fig. [2a](#Fig2){ref-type="fig"}). A less remarkable reduction in cryocrit was calculated in two additional patients, and roughly unchanged levels were observed in the remaining two patients. At the time of SVR12, an overall immunological response was recorded in 17 patients (77.3%). More precisely, cryoglobulins were undetectable in 13 of them and were reduced to \<50% of basal value in 4 patients (Fig. [2a](#Fig2){ref-type="fig"}). Conversely, cryoglobulins were unchanged in three of the remaining five patients at SVR12, and two patients had relapsed at EoT evaluation.Fig. 2Distribution of cryocrit levels (**a**), rheumatoid factor activity (**b**) and complement component C4 levels (**c**) in 22 hepatitis C virus-positive patients with cryoglobulinaemia treated with direct-acting antiviral agents at baseline, at week 4, at the end of treatment (EoT) and 12 weeks after treatment. *SVR12* Sustained virological response 12 weeks after therapy completion
Additional immunological features also showed a remarkable improvement. As shown in Fig. [2b](#Fig2){ref-type="fig"}, test results for RF activity were negative in 12 patients at EoT (54.5%) and in 11 patients at SVR12 (50%). In two patients, RF levels were undetectable at EoT but were increased again at SVR12, whereas in one patient the RF test result was negative only at SVR12. Serum C4 levels increased to the normal range in 10 patients at EoT (45.5%) and in 13 patients at SVR12 (59.1%) (Fig. [2c](#Fig2){ref-type="fig"}).
Compared with basal parameters, cryocrit values were significantly decreased at EoT and SVR12 (*p* \< 0.0001 and *p* = 0.001, respectively) (Table [2](#Tab2){ref-type="table"}). The decrease in ALT levels (*p* \< 0.0001 and *p* \< 0.0001, respectively) and the normalisation of C4 levels (*p* = 0.003 and *p* = 0.002, respectively) also reached statistical significance. Conversely, RF activity was found to be significantly reduced at EoT but not at SVR12 (*p* = 0.003 and *p* = 0.083, respectively). The results of univariate analysis were statistically significant for all parameters. In addition, considering the effect size value (η^2^), a remarkable effect of therapy was observed for all biochemical parameters, the effect being particularly prominent for cryocrit and ALT levels (η^2^ = 0.470 and η^2^ = 0.572, respectively). These results were confirmed by cross-tabs and χ^2^ tests, in that the disappearance of cryoglobulins and ALT normalisation were strongly influenced by therapy (Fig. [3](#Fig3){ref-type="fig"}).Table 2Statistical analysis of the changes in the major biochemical parametersParametersBaselineEnd of treatmentSVR12Univariate analysisMean ± SD (range)Mean ± SD (range)*p* ValueMean ± SD (range)*p* Value*p* Valueη^2^PowerCryocrit, %1.8 ± 1.3 (0.5--4)0.4 ± 0.6 (0--2)\<0.00010.8 ± 1.4 (0--4.6)0.001\<0.00010.4700.999Rheumatoid factor, IU/L69.3 ± 119.1 (0--530)30.9 ± 58.8 (0--255)0.00340.1 ± 63.0 (0--244)0.0830.0020.2810.913C4, mg/dl9.6 ± 7.3 (0--24)13.6 ± 8.8 (4--36)0.00314.1 ± 9.4 (2--40)0.0020.0010.3130.962ALT, IU/L104.8 ± 144.7 (13--668)20.9 ± 5.6 (12--32)\<0.000121.1 ± 6.6 (8--32)\<0.0001\<0.00010.5721.000*Abbreviations: ALT* Alanine transaminase, *C4* Complement component C4, *SVR12* Sustained virological response 12 weeks after therapy completion Fig. 3Cross-tab analysis and χ^2^ tests showing modifications of cryocrit, rheumatoid factor (RF), C4 and ALT levels in relation to antiviral therapy. Besides hepatitis C virus RNA, cryocrit and ALT values were significantly influenced by therapy with respect to RF and C4 (*p* \< 0.0001 and *p* \< 0.0001, respectively). *ALT* Alanine transaminase, *C4* Complement component C4, *EoT* End of treatment, *SVR12* Sustained virological response 12 weeks after therapy completion
In terms of clinical manifestations, we observed regression of Meltzer's triad of symptoms (cryoglobulinaemic glomerulonephritis and peripheral neuropathy), which either were no longer present or were remarkably reduced in 20 patients (90.9%) and 16 patients (72.7%) at EoT and SVR12, respectively. Two patients with peripheral neuropathy had only a partial benefit during antiviral treatment and showed progressive deterioration after EoT. One patient with glomerulonephritis experienced a worsening of renal function at the time of SVR12, but progression of glomerulonephritis to end-stage kidney disease was not observed. Another patient complained of persistent purpuric rashes throughout the treatment and the follow-up period. The patient with SLL treated with DAAs experience a tumour progression that required chemotherapy with fludarabine and cyclophosphamide in addition to RTX. No recurrence or disease progression was recorded in the two patients with previously treated HCC.
According to the above-mentioned response criteria at the time of EoT and SVR12, a CR was observed in 17 patients (77.3%) and in 14 patients (63.7%), respectively, and a partial virologic plus clinical or immunological response was seen in 4 patients (18.2%) and 5 patients (22.7%), respectively. Also at the time of EoT and SVR12, one patient (4.5%) and three patients (13.6%), respectively, achieved only SVR. At SVR12, a CR was obtained in 8 of 14 patients receiving SOF/RBV, in all 4 patients receiving SOF/LDV, and in 2 of the 3 patients receiving 3D combo therapeutic regimens. At the same time point, a PR was observed in 3 of the 14 patients receiving SOF/RBV, in the only patient receiving SOF/DCV, and in 1 of the 3 patients receiving 3D combo. Three of the fourteen patients who were given SOF/RBV achieved SVR only. Given that all patients reached an SVR 12 weeks after stopping therapy, 86.4% of patients in our series obtained a clinical and/or immunological benefit.
Comparing these results with those of our previous study in which we assessed the response to pIFN-α + RBV + RTX combination therapy \[[@CR19]\], we found that the all-oral therapeutic approach was significantly better in terms of CR achievement (*p* = 0.0008 and *p* =0.030 at EoT and SVR12, respectively) (Table [3](#Tab3){ref-type="table"}).Table 3Response rates in 22 patients with mixed cryoglobulinaemia following treatment with direct-acting antiviral agents at the end of treatment and sustained virological response 12 weeks after therapyType of responsepIFN-α/RBV \[[@CR19]\], *n* (%)DAAs therapy, *n* (%)EoT*p* ValuePowerSVR12*p* ValuePowerComplete5 (33.3)17 (77.3)0.00080.89514 (63.7)0.0300.867Partial5 (33.3)4 (18.2)0.2160.8015 (22.7)0.5530.895SVR only5 (33.3)1 (4.5)0.0100.9263 (13.6)0.1020.830*Abbreviations: DAA* Direct-acting antiviral agent, *EoT* End of treatment, *pIFN-α* Pegylated interferon-α, *RBV* Ribavirin, *SVR12* Sustained virological response 12 weeks after therapy completionResults achieved with pIFN-α/RBV regimen are given by comparison
All patients enrolled in the present study were kept under follow-up, with the aim of assessing the persistence of SVR and the possible changes of clinical and immunological features. At the time of this writing, 20 patients (90.9%) have achieved SVR24. Among them, cryoglobulins reappeared in three patients (including the patient with SLL), despite the persistence of SVR24 and in the absence of clinical relapse. In one patient, the test for cryoglobulins was negative at EoT, was positive again at SVR12 and then disappeared at SVR24. So far, 6 of the 14 CR patients at SVR12 have been followed for 48 weeks after completion of therapy and have maintained a CR. In two of them, cryoglobulins were absent, although RF activity was still present at levels less than baseline. In two additional patients, the cryocrit was reduced at roughly 50% of basal value since EoT and remained unchanged throughout follow-up. In the last two patients, both cryocrit and RF remained undetectable. Moreover, one patient with PR and persistence of symptomatic neuropathy despite undetectable cryoglobulins and two SVR-only responsive patients with symptomatic cryoglobulinaemia reached 48 weeks of follow-up.
No severe adverse events related to the administration of DAAs were observed. Two patients receiving SOF/RBV developed progressive anaemia that required reduction of the RBV dosage and erythropoietin administration. Fatigue and mild pain of the dorsolumbar region were reported by several patients, but in none of them was treatment discontinuation required.
Discussion {#Sec10}
==========
Following the identification of HCV as the largely prevalent etiologic agent of MC, researchers in several studies have evaluated the therapeutic efficacy of antiviral drugs such as pIFN-α, with or without RBV, in this condition \[[@CR14]--[@CR16]\]. These studies consistently reported lower rates of SVR in HCV-positive patients with than in those without MC. In addition, IFN-based regimens were characterised by severe adverse events, often leading to therapy discontinuation. In consideration of the underlying B-cell clonal expansion that characterises MC, B-cell depletion therapy with RTX has also been employed after or in addition to the antiviral treatment \[[@CR17]--[@CR19]\]. MC can indeed be considered as an indolent, benign, lymphoproliferative disease potentially susceptible of evolving into malignant B-NHL \[[@CR8]\].
The introduction into clinical practice of the new DAAs for therapy of HCV infection offers for the first time the possibility of achieving SVR in \>90%, and virtually all, of the chronically infected patients \[[@CR21]\], with the coveted perspective of viral eradication and clinical benefits also in patients with extrahepatic manifestations of HCV infection such as MC.
In this study, we report a single-centre experience with the use of the new DAAs in the therapy of HCV-related MC. The first striking result to be emphasised is the high rate of SVR12. All patients reached this primary endpoint, a dramatic reduction of viraemia starting from the fourth week of therapy. A slightly lower response rate has been demonstrated in patients with MC than in those obtained in patients without cryoglobulinaemia with chronic HCV infection \[[@CR24], [@CR26]\]. The suboptimal efficacy of the SOF/RBV combination employed in those studies has been considered as a possible factor accounting for the reduced SVR rates. At variance from these observations, all seven of our patients with genotype 1 HCV infection who were treated with SOF/RBV regimen reached SVR12. Gragnani et al. \[[@CR26]\] similarly reported 100% SVR12 and SVR24 by using different SOF-based antiviral combinations, confirming the very high antiviral efficacy of these drugs. More recently, SVR12 was found in 94% of 64 patients with MC treated with DAAs \[[@CR27]\].
As reported above, at SVR12, cryoglobulins disappeared or decreased by at least 50% in 77.3% of our patients, whereas RF levels normalised in 50% and C4 in 63.6% of the patients. Furthermore, statistical analysis (i.e., univariate analysis, cross-tabs and χ^2^ test) showed that cryocrit and ALT levels were the biochemical parameters more significantly influenced by therapy. Because ALT is a biochemical marker of virus-induced hepatocytolysis, its normalisation is obviously expected to follow viral load reduction or disappearance. Similarly, because viral particles are constitutive components of the cryoprecipitating immune complexes, the formation of such complexes is clearly affected when viral particles are no longer present following the administration of DAAs.
On one hand, RF and C4 levels are less influenced by DAAs within the short period of 12 weeks after stopping therapy and may represent independent markers not only of clinical activity but also of persistently activated B-cell clones. On the other hand, on the basis of our experience, variable changes of RF and C4 levels may occur when a more prolonged follow-up is taken into consideration, in step with either cryoglobulin disappearance or recurrence.
The administration of DAAs was also shown to result in a remarkable improvement of the clinical manifestations, although symptoms persisted or worsened in six patients (27.3%) despite their achievement of SVR12 or more. Even so, the overall results (63.7% of CR plus 22.7% of PR) were significantly better than those obtained with pIFN-α + RBV therapy \[[@CR19]\] or with a triple-combination therapy that included first-generation protease inhibitors \[[@CR20]\], with the additional advantage of less frequent and less severe side effects in patients receiving DAAs.
Sollima et al. \[[@CR32]\] reported a clinical response at week 12 after treatment in two of seven HCV-positive patients with cryoglobulinaemia who were given all-oral antiviral therapy, but one of them had relapsed vasculitis 8 weeks later. It can be inferred that in some cases the pathogenetic process underlying MC progresses despite viral clearance and that, when a "point of no return" has been overstepped, the immune system keeps synthesising pathogenic virus-independent immune complexes. At the same time, when organ damage appears well established, regression cannot be reasonably expected following antiviral therapy. In our study, only a partial benefit was observed in patients with peripheral neuropathy. Renal function worsened but did not reach end-stage renal disease in one patient with cryoglobulinaemic glomerulonephritis, and vasculitic skin lesions persisted in another patient.
At variance from previous studies in which researchers reported NHL remission following antiviral therapy \[[@CR33], [@CR34]\], we observed disease progression in a patient with SLL. Similar observations were reported by Gragnani et al. \[[@CR26]\] and more extensively by Arcaini et al. \[[@CR35]\], who studied 46 HCV-positive patients with indolent NHL treated with IFN-free antiviral regimens and achieved SVR12 in 98% of the patients and lymphoproliferative disease response of splenic marginal zone lymphomas in 73% of the patients. On the contrary, NR was reported in patients with chronic lymphocytic leukaemia/SLL. Possible explanations for these findings are that DAAs lack the anti-proliferative and immunomodulatory effects of IFN-α and that chronic HCV infection has a different impact on the oncogenic process in this particular subset of NHLs.
Interestingly, new clinical scenarios are being opened following the treatment with DAAs of patients with chronic HCV infection who have hematologic malignancies \[[@CR36]\]. In addition to preventing the development of some NHLs, the use of DAAs in fact may avoid HCV reactivation and hepatic flare after chemotherapy, may induce a better outcome after allogeneic bone marrow transplantation, and may reduce the risk of hepatic failure due to cytotoxic drugs, especially in patients with liver cirrhosis \[[@CR36]\].
Conclusions {#Sec11}
===========
Although the worldwide experience with the use of DAAs in HCV-positive patients with cryoglobulinaemia is still limited, our single-centre study shows that variable combinations of DAAs are able to induce SVR12 and SVR24 in virtually all patients. At 12 weeks after EoT, CR and PR were diagnosed in as many as 86.4% of the patients. These results not only are strikingly better than those obtained with previous antiviral treatments \[[@CR14]--[@CR16], [@CR20], [@CR37]\] but also are closely comparable, in terms of SVR12, to the results reported in remarkably more numerous multicentre cohorts of patients with chronic HCV infection without MC \[[@CR22], [@CR23]\].
However, in step with the remarkably high rates of virus eradication induced by the growing cluster of DAAs, new and unforeseen scenarios are emerging. In a minority of patients, clinical and immunological features of cryoglobulinaemic vasculitis persist despite viral clearance. The long-term evolution of these features is totally unknown, and whether an early antiviral approach with DAAs, before the occurrence of severe organ damage, might be able to prevent the occurrence of these initially virus-induced and then virus-cleared vasculitides remains to be elucidated. In addition, at variance from patients with MC achieving SVR and a successful vasculitic response, in whom the regression of B-cell monoclonal expansion in the bone marrow has been described \[[@CR19]\], the presence of circulating RF may be associated with the over-representation of mature activated memory B cells, which persist at least during the early phase following eradication of HCV \[[@CR38]\]. To ascertain whether an HCV-independent, B-cell lymphoproliferative disorder may eventually appear in this particular group of patients, more prolonged follow-up after viral clearance is necessary.
3D combo
: Ombitasvir/paritaprevir/ritonavir plus dasabuvir
AIFA
: Italian Medicines Agency
ALT
: Alanine transaminase
B-NHL
: B-cell non-Hodgkin lymphoma
CR
: Complete response
DAA
: Direct-acting antiviral agent
DCV
: Daclatasvir
EoT
: End of treatment
GLM
: General linear model
GT
: Genotype
HCC
: Hepatocellular carcinoma
HCV
: Hepatitis C virus
IFN
: Interferon
Ig
: Immunoglobulin
LDV
: Ledipasvir
MC
: Mixed cryoglobulinaemia
NHL
: Non-Hodgkin lymphoma
NR
: No response
pIFN
: Pegylated interferon
PR
: Partial response
RBV
: Ribavirin
RF
: Rheumatoid factor
RTX
: Rituximab
SLL
: Small lymphocytic lymphoma
SOF
: Sofosbuvir
SVR
: Sustained virological response
SVR12
: Sustained virological response 12 weeks after therapy completion
SVR24
: Sustained virological response 24 weeks after therapy completion
Not applicable.
Funding {#FPar1}
=======
This study was supported in part by Associazione Italiana per la Ricerca sul Cancro (AIRC) (investigator grant 14095 \[to AV\]), the University of Bari "Aldo Moro" (to GL) and Fondo di Sviluppo e Coesione 2007-2013 -- APQ Ricerca Regione Puglia "Programma regionale a sostegno della specializzazione intelligente e della sostenibilità sociale ed ambientale - FutureInResearch" (to SR).
Availability of data and materials {#FPar2}
==================================
The datasets used and/or analysed during the present study are available from the corresponding author on reasonable request.
Authors' contributions {#FPar3}
======================
GL, SR, FP and FD substantially contributed to the conception of the study, the analysis and interpretation of the data, and the drafting of the manuscript. AV critically revised the manuscript for important intellectual content. All authors read and approved the final manuscript.
Competing interests {#FPar4}
===================
The authors declare that they have no competing interests.
Consent for publication {#FPar5}
=======================
Not applicable.
Ethics approval and consent to participate {#FPar6}
==========================================
Written informed consent was obtained from all patients. Treatments were administered on-label. Because of the nature of the study, the institutional independent ethical committee of "Azienda Ospedaliero-Universitaria Consorziale Policlinico di Bari" waived the need for ethical approval. The study was conducted according to the principles of the Declaration of Helsinki.
Publisher's Note {#FPar7}
================
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
| {
"pile_set_name": "PubMed Central"
} |
Introduction
============
The non-ambiguous perceptual organization of planar visual images into figure and ground requires the visual system to be able to generate a three-dimensional (3D) representation from a two-dimensional (2D) stimulus input. During viewing of a natural 3D scene, objects that are closer to the viewer may block or occlude the view of objects that are further away. Boundaries of these occluding objects are perceived as belonging to them, a property called *border ownership*. Because occluding objects occur closer in depth than the objects they occlude, border ownership in response to a 3D scene typically coexists with a percept of being closer in depth. The importance of surface border ownership to what may seem nearer to us was already noticed by Galileo (see the review by [@B23]). The borders of occluding surfaces generally occur in the foreground, while the borders of occluded surfaces generally occur in the background.
An important problem in visual perception concerns how border ownership assignment occurs in response to 2D pictures, and what role it may play in determining 3D percepts of such pictures. In response to 2D pictures, there are famous examples where the perceptual assignment of surface borders to 3D percepts of foreground and background may be reversible, leading to totally different interpretations of the objects in each representation (**Figure [1](#F1){ref-type="fig"}**). Such spontaneous changes in figure-ground perception occur only under particular circumstances due to competition between multiple, approximately balanced, 3D interpretations of the 2D image.
![**Two faces or a vase?** In these variations on the famous reversible figures of [@B80], with surface contrasts of opposite signs, the perceptual assignment of border ownership to foreground and background may be influenced by both shifts in spatial attention and prior learning of object categories.](fpsyg-07-01102-g001){#F1}
During the past half century, many perceptual displays and psychophysical data have described properties of figure-ground perception in response to 2D pictures and 3D scenes. The FACADE (Form-And-Color-And-DEpth) model of 3D vision and figure-ground perception, and its further development and extension by the 3D LAMINART laminar cortical model, have explained and predicted many data about how the brain consciously sees 3D surface percepts in response to 2D pictures and 3D scenes, including, but not restricted to, percepts that involve figure-ground perception ([@B36], [@B37], [@B38], [@B41],[@B42]; [@B46]; [@B66]; [@B63]; [@B49]; [@B53]; [@B95]; [@B8], [@B9]; [@B55]; [@B44]; [@B6]; [@B56]; [@B7]; [@B29]; [@B64]; [@B50]). Along the way, these models have also explained and predicted many anatomical and neurophysiological data about 3D vision and figure-ground perception in response to both static and moving images and scenes. These explanations involve multiple brain areas, including the lateral geniculate nucleus (LGN) and three parallel cortical streams interacting among cortical areas V1, V2, V4, MT, and MST.
Neurophysiological experiments have also been done to record properties of neurons those activities contribute to figure-ground percepts. In this series, Von der Heydt et al. have published important data in a series of neurophysiological experiments about the border ownership properties of neurons in cortical area V2 of monkeys. In particular, [@B97] reported data from neurons in cortical area V2 that tend to respond to borders with different firing rates depending on whether the border is owned by an occluding or an occluded surface. These neurons are often maximally excited by a preferred combination of direction-of-contrast and border ownership. [@B96] further studied the contribution of individual edges to border ownership assignment by decomposing figural contours into fragments. Fragments on the preferred side-of-figure produced facilitation, while fragments on the opposite side produced suppression of neural responses. Border-ownership signals also persist for about a second in the brain ([@B69], [@B70]). Border-ownership signals are generally consistent over multiple variations in shape geometry, configuration, and contrast ([@B90]; [@B79]; [@B78]). [@B28] furthermore used fMRI and found a border ownership BOLD signal in the human visual cortex.
The FACADE and 3D LAMINART anticipated a number of these V2 cell properties, but not all of them. By unifying results from the above-cited theoretical articles with results about how V1 cells that are sensitive to *absolute* binocular disparity are transformed into V2 cells that are sensitivity to *relative* binocular disparity---namely, the difference in absolute disparity of two visible features ([@B51]) --- [@B43] was able to propose a unified explanation of all the main von der Heydt et al. V2 data properties.
As noted above, the von der Heydt et al. data show that various neurons in V2 that are sensitive to border ownership also respond with a preferred contrast polarity. However, the same figure-ground properties can sometimes occur in a given configuration when contrast polarities are mixed, or are switched from one polarity to the opposite, across the stimulus fragments that induce 3D surface percepts (e.g., [@B65]), and the phenomenal "logic" of such shape percepts (see [@B73]) is indeed likely to involve a complex hierarchy of integration levels in the brain, as explicated by model explanations that involve cortical areas other than V2. The new psychophysical experiments that are reported in this article further probe these intercortical interactions, and illustrate the limitations of explanations that depend exclusively upon V2. Indeed, V2 has been predicted not to directly represent any consciously visible 3D surface qualia, but rather to support amodal object recognition of occluding and partially occluded objects ([@B36], [@B37], [@B42]). The Discussion section explains how and why this may happen as part of a focused summary of the cortical mechanisms that can explain the new data that are reported herein. Before turning to these new results, they are put into a larger historical context with the following partial survey of previous psychophysical and theoretical results.
The great pioneering work of [@B58], [@B59],[@B60], [@B61]) on subjective contours provided many compelling examples of how illusory surfaces can be induced by spatially sparse, albeit (approximately) colinear, and co-oriented inducers, including examples of figure-ground separation. [@B75], [@B77]) additionally noted that the phenomenal strength of surfaces standing out against uniform backgrounds appears as marked in configurations with inducers of opposite contrast polarites as in configurations with inducers of one and the same polarity. Quantitative data for the relative strength of these percepts were not made available in these earlier reports. They were, however, so compelling that they motivated theoretical accounts for boundary detection mechanisms that are insensitive to the local sign of contrast elements in the perceptual assignment of border ownership. [@B14], [@B47],[@B48]), [@B81], and [@B19] all noted, in particular, the conceptual importance of a reverse-contrast Kanizsa square (**Figure [2](#F2){ref-type="fig"}**, part 1) as an example of long-range grouping across opposite contrast polarities in response to polarity-specific inputs from spatially disjoint, oriented detectors. In addition to this boundary-grouping property, the percepts of filled-in surface brightness caused by different inducer configurations had also to be explained.
![**Four reverse-contrast Kanizsa configurations.** The two Kanizsa squares in the first column represent stimuli used in experiments on sign-invariant boundary detection by [@B81] and obey their criteria of sign-invariant boundary induction. The two Kanizsa squares in the second column show cases where single inducing elements are given locally opposing contrast signs, used as stimuli in experiments by [@B82] and [@B83]. These two configurations do not obey Shapley and Gordon's criteria of sign-invariant boundary induction. The strength of the illusory boundaries therein was reported to be less discriminable, and even more so when exposure duration was limited to less than 320 ms ([@B82]; [@B83]). (2) Six Ehrenstein configurations. The circular illusory surface in the center was reported less perceptible when the radial inducing lines are fragmented, as in the configuration in the right column, and given locally opposing contrast signs, as in the top display of the right column ([@B22]; [@B83]). When all fragments share the same contrast sign, (as in the bottom display of the left column), the 'O' illusion discovered by [@B57] is perceived. This percept is abolished when the local contrast signs are of the opposite polarity, \[as in the bottom display of the right column. (3) Reverse-contrast Kanisza square inducers can generate a percept of transparency (top row, left column) ir not (top row, right column)\]. When the pac men are removed (bottom row), a central rectangular background is perceived to be further away than a surrounding nearer surface with a rectangular aperture. See text for details. \[Reprinted with permission from [@B71]\].](fpsyg-07-01102-g002){#F2}
Both the boundary grouping and surface brightness properties were simulated in a series of neural modeling articles from Grossberg and his colleagues; e.g., [@B14], [@B34], [@B47],[@B48]) and [@B54], at around the same time that classical neurophysiological data about how opposite contrast polarity inputs are pooled at V1 complex cells ([@B86]), and about how illusory contour formation occurs in cortical area V2 ([@B89]), supported model predictions of how bipole grouping cells in V2 can pool inputs from V1 simple, complex, and hypercomplex cells, to form long-range groupings from either like, or opposite, contrast polarity inducers. These explanations, however, were restricted to explaining 2D boundary and surface properties. Neural explanations of 3D properties, including 3D figure-ground separation properties, began with the the FACADE model ([@B35], [@B36], [@B37]) as additional neurophysiological and psychophysical studies (e.g., [@B62]; [@B74]; [@B20], [@B21]; [@B93]) reported more properties of sign-invariant boundary grouping, sensitive to contrast intensities only, in response to inducers of either polarity \[see the recent reviews by [@B25] and [@B85]\].
The postulate that boundary grouping by the visual system is insensitive to the contrast polarity of its inducers was subsequently challenged by findings from studies by [@B57], [@B82] and [@B83], with new configurations where the contrast polarity varies repeatedly within one and the same inducing element. In these cases, the strength of induced perceptual boundaries, or illusory contours, was found to be significantly diminished, especially at stimulus durations shorter than 300 ms (e.g., [@B83]). In contrast to examples like the reverse-contrast Kanizsa square, these authors created patterns where the local signs cancel each other out locally, not globally, along an axis of boundary induction (**Figure [2](#F2){ref-type="fig"}**, part 2). These studies hark back to earlier observations on the Ehrenstein illusion ([@B22]), where the perceptual strength of the centrally induced surface does not depend on the contrast polarity of the inducing lines, provided the contrast sign is homogenous within a given inducing element. When the inducers are fragmented into several parts with variable contrast signs (e.g., **Figure [2](#F2){ref-type="fig"}**, part 2, upper right display), considerably weaker groupings are found. [@B57] reported a new ring-shaped illusion, the 'O' illusion (**Figure [2](#F2){ref-type="fig"}**, part 2, lower left display), which is only perceived in fragmented radial lines of one and the same polarity. These findings suggest that the ways in which contrast polarity variations are locally distributed, and the exposure duration of the stimuli, matter critically in the perceptual genesis of shape illusions. At identical physical luminance, opposite contrast signs within one and the same local inducing element may largely cancel each other out and become less effective in perceptual grouping when viewing durations are not long enough. Analogous effects of local contrast changes on long-range perceptual groupings may be observed in percepts of Glass patterns and reverse-contrast Glass patterns ([@B33]; [@B76]), and can be explained by simular boundary and surface interactions ([@B16]).
The general theme of different effects of spatially short-range vs. long-range effects of same-polarity vs. opposite-polarity inducers on perceptual grouping and figure-ground perception has a long history, both experimentally and theoretically. Such differences exist in response to both static and moving displays and have led to a large literature about how short-range and long-range filters and grouping mechanisms work together to generate percepts. In the case of static form perception, simple cells in cortical area V1 typically respond to one contrast polarity, but not its opposite, whereas complex cells pool signals from pairs of like-oriented but oppositely polarized simple cells to begin the process of contrast-invariant boundary grouping. Theoretical explanations of these interactions are by now well known in the literature (see Discussion below). Key classical data and neural explanations of how mixtures of contrast-dependent and contrast-invariant mechanisms influence percepts ranging from spatial location and hyperacuity ([@B3],[@B4]) to brightness perception ([@B94]; [@B15]) can, for example, be found in [@B47],[@B48]), [@B35], and [@B54].
[@B71] and [@B84] have presented additional displays in which same-contrast and opposite-contrast inducers can lead to different effects. In particular, the displays in **Figure [5](#F5){ref-type="fig"}** from [@B71] are shown in **Figure [2](#F2){ref-type="fig"}**, part 3. The display in the first row, left panel, generates a percept of unimodal transparency, with an emergent square surface lying in front of four partially occluded pac man figures. The two white pac men are more luminous than the gray background, whereas the two black pac men are less luminous than the background. The illusory Kanizsa square that emerges in this percept thus bridges between opposite-polarity inducers. Opposite polarity inducers also occur in the display in the first row, right panel, but no percept of transparency obtains.
[@B55] have explained the percepts that are generated by displays of this kind by simulating how, just by varying the relative contrasts of regions in a display, without changing their geometrical relationships, one can cause a percept of unimodal transparency, bistable transparency, or of a flat surface. **Figure [2](#F2){ref-type="fig"}** (part 3, top row) illustrates two of these possibilities. A key factor in determining whether such a display looks transparent or not is whether the curved pac man boundary segments at each side of a Kanizsa square boundasry segment have the same contrast, or opposite contrasts, relative to the background. In **Figure [3](#F3){ref-type="fig"}** (part 3, top row, left), the answer is "same" from pac man to background, and transparency is perceived. In **Figure [2](#F2){ref-type="fig"}** (part 3, top row, right), the answer is "opposite" and no transparency is perceived.
{#F3}
Many researchers have noted how contrast relations within an image can cause or eliminate a percept of transparency (e.g., [@B67]; [@B5]; [@B91],[@B92]; [@B2]; [@B1]). [@B55] explain these different percepts using the full machinery of cortical area V1, V2, and V4 interactions within the 3D LAMINART model, with a key role in these transparency vs. no-transparency percepts predicted to be played by a like-polarity competition among simple cells in layer 4 of cortical area V1. [@B55] also summarize neuroanatomical and neurophysiological evidence for all these interactions, but no experiments have yet been done to try to manipulate a transparency percept in animals by altering the strength of this V1 inhibitory interaction.
The displays in **Figure [2](#F2){ref-type="fig"}** (part 3, bottom row) are derived by removing the outer pac man shapes from the displays in **Figure [2](#F2){ref-type="fig"}** (part 3, top row). The resulting identical displays generate a percept of an inner background rectangle that is further away than the open rectangular figure that surrounds it. This kind of display is explained in the Discussion section in the same way that the percepts that are generated by **Figure [3](#F3){ref-type="fig"}** (top row) in the new experiments are explained.
The displays used in the current experiments (**Figure [3](#F3){ref-type="fig"}**) do not change polarity on a spatial scale within the size of individual simple cell receptive fields. Moreover, these displays are conceptually and mechanistically more challenging to explain than previously tested configurations, and the percepts that they generate are quantified in the *Experimental Results*. The displays used here specifically tested for figure-ground assignment in terms of what is seen as standing out "in front" and what is seen as as "lying behind" by creating configurations in which inducers of varying sign were displayed on either of two sides of a perceptual boundary while the contrast sign within one and the same inducing element was always homogenous. In these configurations, the orientation, direction and polarity of contrast are locally controlled, and may be mixed or switched from one direction and/or polarity to the opposite across the stimulus elements that produce the resulting figure-ground percept. The duration of presentation was not limited in time, as in natural free viewng conditions. An alternative forced choice task similar to that from earlier studies was employed (e.g., [@B18]; [@B26], [@B27]).
A key variable of the FACADE theory relative to the orientation of surface-inducing contrast edges was tested by presenting inducing elements with outward-oriented contrast edges (**Figure [3](#F3){ref-type="fig"}**, top row) as well as inducers with inward-oriented edges (**Figure [3](#F3){ref-type="fig"}**, bottom row). The former case induces percepts with the inducers seen "lying behind" within a closed rectangular background region. This situation requires a subtle analysis of the intercortical mechanisms that are responsible for the depth, surface, and persistence properties of such a background region.
Materials and Methods {#s1}
=====================
The psychophysical experiments were conducted in accordance with the Declaration of Helsinki (1964) and with the full approval of the corresponding author's institutional (CNRS) ethics committee. Informed written consent was obtained from each of the participants of the psychophysical experiments. Experimental sessions were run under laboratory conditions of randomized free trial-by-trial image viewing using a Dell PC computer equipped with a mouse device and a high resolution color monitor (EIZO LCD 'Color Edge CG275W'). This screen has an in-built calibration device which uses the Color Navigator 5.4.5 interface for Windows. The images were generated in Photoshop using selective combinations of Adobe RGB increments to generate contrast inputs (see also [@B24]). The luminance levels for each RGB triple could be retrieved from a look-up table after calibration and the values were also cross-checked on the basis of standard photometry using an external photometer and adequate interface software (Cambridge Research Instruments).
Subjects
--------
Ten unpracticed observers, mostly students in computational engineering who were unaware of the hypotheses of the study, participated in the experiments. All subjects had normal or corrected-to-normal visual acuity.
Stimuli
-------
The stimuli (**Figure [3](#F3){ref-type="fig"}**) consisted of six images with different edge contrast inputs. The luminance of the background was 50.5 cd/m^2^ (148,148,148 RGB) in all eight images. The luminance of the black contrast fragments was 1.5 cd/m^2^ (0,0,0 RGB) and the luminance of the white contrast fragments was 99.5 cd/m^2^ (255,255,255 RGB), yielding perfectly balanced Weber contrasts (L~feature~--L~background~/L~background~) of --0.97 and 0.97 for negative and positive polarities in the six images with the fragmented edge contrasts. The height of the central surfaces was 10 cm on the screen, whereas the width was 12 cm. In the six images with the ambiguous fragmented edge contours, about 50% of the inner surface contour was void of a contrast, so that 50% of the boundary contour had to be completed perceptually ([@B17]).
Task Instructions
-----------------
A classic psychophysical forced choice procedure with three response alternatives was used to measure perceptual decisions for relative depth (figure-ground). Observers were asked to indicate whether the central surface appeared to "stand in front" of, to "lie behind", or to be in the "same plane" as the surrounding surface. It was made sure that all observers understood the instructions correctly before an experimental session was initiated.
Procedure
---------
Subjects were seated at a distance of 1 m from the screen and asked to look at the center of the screen. The experiments were run in a dimmed room (mesopic conditions), with blinds closed on all windows. The six images were presented in random order for about one second each, and each image was presented four times in a session. Inter-stimulation intervals were measured. They typically varied from one to three seconds, depending on the observer, who initiated the next image presentation by striking a key on the computer keyboard. The experiment produced a total of 300 observations from 30 trials per subject in an individual session.
Results
=======
The individual data from this depth judgment experiment were analyzed in terms of conditional response frequencies, or the frequencies with which the different perceptual responses ("in front", "behind", "same plane") occurred within a given experimental condition. These frequency distributions, permit conclusions relative to event saliency, and allow plotting probabilities (e.g., [@B72]), based on the assumption that a similar frequency distribution is statistically likely to occur in any study population with the same characteristics as the sample population selected for this experiment. To assess whether the observed differences between the response frequencies reflecting the most salient events were statistically predictable, we fed the frequency distributions for "in front" and "behind", which reflect complementary dimensions of the underlying psychological decision, into analysis of variance (ANOVA) using *Systat 11* (see also [@B18], or [@B26], [@B27]). The balanced 2x3 factorial design, with stimuli presented in random order, allowed for generation of psychophysical judgements from an even number of independent forced-choice trials per factor level. Criteria for parametric testing, including normality and egality of variance of the frequency distributions, were met.
Experimental Results
--------------------
The results (**Figure [4](#F4){ref-type="fig"}**) show that the configurations generate a higher event probability for the central surface to be perceived as figure ("in front") when the local contrast edges of the fragmented contour elements are inward directed, as indicated by the distribution of the response frequencies *RF*, with the following average values: *RF*~(infront)~ = 0.83 (SEM = 0.05), *RF*~(behind)~ = 0.07 (SEM = 0.03), and *RF*~(same)~ = 0.10 (SEM = 0.04). The configurations generate a higher event probability for the central surface to be perceived as ground ("behind") when the local edges are outward directed: \[*RF*~(infront)~ = 0.06 (SEM = 0,02), *RF*~(behind)~ = 0.75 (SEM = 0,03), *RF*~(same)~ = 0.19 (SEM = 0,04)\]. These perceptual decisions do not depend on the contrast signs of the local edges. Configurations with negative like-contrasts, positive like-contrasts and mixed contrast polarities produced similar response frequency distributions, with average values as follows: *RF*~(infront)~ = 0.51 (SEM = 0,14), *RF*~(behind)~ = 0.48 (SEM = 0,12), and *RF*~(same)~ = 0.10 (SEM = 0,04) for negative like-contrasts; *RF*~(infront)~ = 0.42 (SEM = 0,14), *RF*~(behind)~ = 0.43 (SEM = 0,13), and *RF*~(same)~ = 0.15 (SEM = 0,04) for positive like-contrasts; *RF*~(infront)~ = 0.43 (SEM = 0,11), *RF*~(behind)~ = 0.42 (SEM = 0,10), *RF*~(same)~ = 0.15 (SEM = 0,05) for mixed polarities.
{#F4}
ANOVA on the response frequencies for "in front" and "behind" for the two levels of the factor "contrast edge direction" and the three levels of the factor "contrast sign" returned statistically significant effects of "contrast edge direction" on perceptual decisions for "in front" \[*F*(1,2) = 228.30, *p* \< 0.001\] and "behind" \[*F*(1,2) = 212,77, *p* \< 0.001\]. As expected (e.g., [@B18]), no effect of contrast sign on either type of perceptual decision \[*F*(1,2) = 2.58, NS on response frequencies for "in front" and *F*(1,2) = 0.25, NS on response frequencies for "behind"\] was observed.
Discussion
==========
A unified mechanistic explanation is here provide of the percepts induced by the **Figure [3](#F3){ref-type="fig"}** images using FACADE and 3D LAMINART model mechanisms (**Figures [5](#F5){ref-type="fig"}** and **[6](#F6){ref-type="fig"}**). In particular, model mechanisms are summarized with enough detail to achieve a self-contained explanation of the new data using boundary and surface stream interactions within and between cortical areas V1, V2, and V4, while also clarifying the insufficiency of V2 neurophysiological data about border ownership to explain the resulting conscious percepts.
{#F5}
![**3D LAMINART model circuit diagram.** This laminar visual cortical model consists of a boundary stream that includes V1 interblobs, V2 pale stripes (also called interstripes), and part of V4, and computes 3D perceptual groupings in different scales; and a surface stream that includes V1 blobs, V2 thin stripes, and part of V4, and computes 3D surfaces that are infused with lightness in depth. Both the boundary and surface streams receive illuminant-discounted signals from LGN cells with center-surround receptive fields, and both converge in V4, where visible 3D surfaces are consciously seen that are separated from their backgrounds. Models V2 and V4 also output to inferotemporal cortex (not shown), where object recognition takes place. Model V1 interblobs contain both monocular and binocular cells. Binocular simple cells become disparity-sensitive by binocularly matching left and right scenic contours with the same contrast polarity in layer 3B before pooling opposite polarity responses at complex cells in layer 2/3A. Monocular and binocular boundary cells control filling-in of monocular 3D surfaces within V1 blobs. Closed boundaries can contain the filling-in process, and can send feedback to V1 interblobs that selectively strengthens the closed boundary components. Monocular and binocular V1 boundaries are pooled in V2. V2 pale stripes can complete 3D perceptual groupings while inhibiting false binocular matches using the disparity filter to solve the correspondence problem. These completed boundaries form compartments in the V2 thin stripes within which filling-in of monocular 3D surfaces occurs. Closed boundaries can contain the filling-in process and send surface-to-boundary surrface contour feedback signals to enhance their generative boundaries, while also suppressing redundant boundaries at the same positions and frrther depths. These conmpleted boundaries and filled-in surfaces complete the representations of partially occluded objects. They do not generate visible percepts, but can be recognized by activating inferotemporal cortex. Visible surfaces in which figures are separated in depth from their backgrounds are formed in V4. Here, left and right eye feature contour signals from the LGN are binocularly matched, while redundant feature contour signals are pruned at further depths by inhibitory signals from the thin stripes. Then the pruned feature contour signals induce filling-in of a visible surface percept within enriched binocular boundaries. V4 emits output signals that lead to recognition and grasping of unoccluded parts of opaque surfaces -- Reproduced with permission from [@B29].](fpsyg-07-01102-g006){#F6}
Bipole Boundary Completion Can Pool Over Opposite Contrast Polarities
---------------------------------------------------------------------
In response to all of the images in **Figure [3](#F3){ref-type="fig"}**, boundaries can be completed inwardly between pairs of adjacent colinear inducers. The completion process uses the oriented long-range horizontal cooperation of *bipole grouping* cells in layer 2/3 of cortical area V2, balanced by shorter-range disynaptic inhibition (**Figures [6](#F6){ref-type="fig"}** and **[7A](#F7){ref-type="fig"}**). Bipole cells can complete boundaries in response to colinear inducers with the same relative contrasts with respect to the background, as in the leftmost two columns of **Figure [3](#F3){ref-type="fig"}**, as well as between inducers with opposite relative contrasts with respect to the background, as shown repeatedly in psychophysical experiments (e.g., [@B93]; [@B88]). This is true because bipole cells receive their inputs, after several stages of additional processing, from complex cells in layer 2/3 of cortical area V1 (**Figures [5](#F5){ref-type="fig"}** and **[6](#F6){ref-type="fig"}**). Complex cells, in turn, pool inputs from simple cells in layer 4 of V1 that have the same preferences for position and orientation, but opposite contrast polarities. As a result, bipole cells can complete boundaries around objects that lie in front of textured backgrounds whose relative contrasts reverse along the perimeter of the object. In the present cases, bipole cells complete rectangular boundaries that abut all their inducers.
![**(A)** T-Junction Sensitivity. (left) T-junction in an image; (middle). Bipole cells provide long-range cooperation (+), and work together with inhibitory interneurons that provide cells provide short-range competition (-); (right). An end gap in the vertical boundary arises because, for cells near where the top and stem of the T come together, the top of the T activates bipole cells along the top of the T more than bipole cells are activated along the T stem. As a result the stem boundary gets inhibited whereas the top boundary does not -- Reprinted with permission from [@B37] -- **(B)** Necker cube. This 2D picture can be perceived as either of two 3D parallelograms whose shapes flip bistably through time. **(C)** When attention switches from one circle to another, that circle pops forward as a figure and its brightness changes. See [@B55] for an explanation -- Reprinted with permission from [@B87].](fpsyg-07-01102-g007){#F7}
Bipoles Are Sensitive to T-junctions
------------------------------------
The long-range cooperation and short-range competition processes whereby bipoles complete boundaries are sensitive to any T-junctions that lie along the boundaries that they complete (**Figure [7A](#F7){ref-type="fig"}**). In the images with incomplete boundaries, there are no explicit T-junctions in the image. However, when a rectangular boundary is completed, T-junctions are created at the corners of the colinear inducing contrasts. The bipole cells that lie along the orientation of a completed boundary (the "head" of the T) get more excitatory input than do the bipole cells that lie near the head of the T, but whose orientational preference is along the perpendicular or oblique orientation of the inducing contrast (the "stem" of the T). This is true because the bipole cells that are activated along the head of the T receive strong excitatory inputs from both sides of their receptive fields, whereas the bipole cells that are activated along the stem of the T receive excitatory inputs from just one side of their receptive fields (**Figure [7A](#F7){ref-type="fig"}**). The more strongly activated bipole cells inhibit surrounding bipole cells more than conversely through a spatially short-range competitive network. As a result, the bipole cells near the head that are along the stem get inhibited. An *end gap* hereby forms in each boundary near where the stem of a T touches its head (**Figure [7A](#F7){ref-type="fig"}**).
Because the bipole cells can complete rectangular boundaries in response to spatially disjoint inducers with the same relative contrasts with respect to their surrounding regions, or in response to combinations of inducers with opposite relative contrasts, end gaps at the T-junctions can form in either case.
As originally explained in [@B36], [@B37]), and simulated in such articles as [@B63], [@B53], and [@B55], end gaps trigger a process of figure-ground perception and border ownership in which the rectangular boundaries are often perceived in front of the regions that they enclose, which are themselves perceived as a ground at a slightly further depth. For example, the percepts of the Necker cube (**Figure [7B](#F7){ref-type="fig"}**) can be explained in this way (see [@B53]), as can the way that shifts in attention can make an attended disk in **Figure [7C](#F7){ref-type="fig"}** look both nearer and darker ([@B55]; [@B87]). These concepts are reviewed and extended below in order to explain the conscious 3D surface percepts that are generated by the images in **Figure [3](#F3){ref-type="fig"}**, notably why the percepts of the completed rectangular surfaces in response to the **Figure [3](#F3){ref-type="fig"}** (bottom row) displays appear in front of their surrounding regions, but the percepts of the completed rectangular surfaces in response to the **Figure [3](#F3){ref-type="fig"}** (top row) displays look further away than their surrounding regions.
In order to motivate these theoretical explanations, it is useful to ask the following question: If it is indeed the case that these figure-ground relationships do not depend on having inducers with the same contrast polarity, then why do so many cortical area V2 cells that are sensitive to border ownership also exhibit a particular contrast preference; e.g., [@B97]. This can be understood by going into more detail about how end gaps trigger figure-ground perception and border ownership.
Feedback between Boundaries and Surfaces Achieves Complementary Consistency
---------------------------------------------------------------------------
The FACADE and 3D LAMINART models (**Figures [5](#F5){ref-type="fig"}** and **[6](#F6){ref-type="fig"}**) detail how the figure-ground perception process utilizes feedback between the boundary completion process in the interblob cortical stream and the surface filling-in process in the blob cortical stream within V1, V2, and V4 of visual cortex, This feedback enables boundaries and surfaces to generate a consistent percept, despite the fact that they obey computationally complementary laws. This property is called *complementary consistency*. As will be noted shortly, the mechanisms that ensure complementary consistency also contribute to 3D figure-ground separation. [@B43] explains in detail how the data of von der Heydt et al. about border ownership and related properties of V2 cells fit into this larger theory.
In particular, the completed boundaries with their end gaps are projected topographically from the interstripes, or pale stripes, of V2, at which boundaries are completed, to the thin stripes of V2, at which one stage of surface filling-in occurs. When surface filling-in occurs within these boundary inducers, brightness and color can flow out of the end gaps, thereby equalizing the filled-in brightness and color on both sides of the remaining boundaries near these gaps (**Figure [8](#F8){ref-type="fig"}**, bottom row). Only if the boundary of the rectangle is closed, with no significant gaps, can it fully contain its surface-filling in. In the percepts that are generated by the displays in **Figure [3](#F3){ref-type="fig"}**, the inducers that are inside or outside these rectangles are surrounded by closed boundaries, since the frame of the image provides another closed boundary that can contain filling-in between it and the bipole-generated rectangular boundary that lies within it. The significance of this fact will be discussed below.
![**The top row illustrates how, at a prescribed depth, a closed boundary contour abuts an illuminant-discounted feature contour.** When this happens, the feature contours can fill-in within the closed boundary. The bottom row (left) depicts how filling-in of the feature contours is contained by this closed boundary contour, thereby generating large contrasts in filled-in activity at positions along the boundary contour. Contrast-sensitive surface contour output signals can then be generated in response to these large contrasts. The bottom row (right) depicts a boundary contour that has a big hole in it at a different depth. Feature contours can spread through such a hole until the filled-in activities on both sides of the boundary equalize, thereby preventing contrast-sensitive surface contour output signals from forming at such boundary positions -- Reprinted with permission from [@B43].](fpsyg-07-01102-g008){#F8}
Closed Boundaries, Surface Contours, and Boundary Pruning
---------------------------------------------------------
As filling-in occurs, feedback can occur from the surfaces in the thin stripes to the boundaries in the pale stripes (**Figure [9](#F9){ref-type="fig"}**). These feedback signals occur from each active Filling-In DOmain, or FIDO. They are *surface contours* that are generated by contrast-sensitive on-center off-surround networks that act across position and within the depth represented by each FIDO. These contrast-sensitive networks sense sufficiently large and steep spatial discontinuities in the filled-in brightnesses or colors within their FIDO. They hereby generate surface contour output signals only at the surface positions that are surrounded by closed boundaries. In response to the incomplete inducers in the top row of **Figure [3](#F3){ref-type="fig"}**, these regions lie on both sides of the completed boundaries. However, due to the end gaps, surface contour signals are not generated at the boundary positions of the inducers themselves.
![**A closed boundary can form at Depth 1 by combining a binocular vertical boundary at the left side of the square with three monocular boundaries that are projected along the line of sight to all depths.** Surface contour output signals can thus be generated by the FIDO at Depth 1, but not the FIDO at Depth 2. The Depth 1 surface contours excite, and thereby strengthen, the boundaries at Depth 1 that controlled filling-in at Depth 1. These surface contours also inhibit the redundant boundaries at Depth 2 at the same positions. As a result, the pruned boundaries across all depths, after the surface contour feedback acts, can project to object recognition networks in inferotemporal cortex to facilitate amodal recognition, without being contaminated by spurious boundaries -- Reprinted with permission from [@B43].](fpsyg-07-01102-g009){#F9}
The surface contour output signals generate topographic feedback signals to a subset of the boundary representations that induced them (**Figure [9](#F9){ref-type="fig"}**). These feedback signals are delivered to the boundary representations via an on-center off-surround network whose inhibitory off-surround signals act within position and across depth (**Figure [9](#F9){ref-type="fig"}**). The on-center signals strengthen the boundaries that generated the successfully filled-in surfaces at the same depth, whereas the off-surround signals inhibit spurious boundaries at the same positions but further depths. This inhibitory process is called *boundary pruning*. Surface contour signals hereby strengthen consistent boundaries and prune, or inhibit, redundant boundaries.
Because surface contour signals are generated by the contrasts of a filled-in surface, they are sensitive to a particular contrast, but not to the opposite one. Their feedback to boundaries thus makes the responses of the recipient bipole cells also sensitive to this contrast, even though the bipole cells, in the absence of surface contour feedback signals, respond to both contrast polarities, due to their inputs from V1 complex cells, so that they can complete boundaries of objects in front of textured backgrounds. Thus, both surface contour signals and their target bipole cells also exhibit sensitivity to a particular contrast polarity, as in the neural data of [@B97].
In response to 3D scenes, boundary pruning is part of the process of *surface capture* whereby feature contours can selectively fill-in visible surface qualia at depths where binocular fusion of object boundaries can successfully occur. Boundary pruning helps to strengthen closed boundaries, while competitively eliminating boundaries with gaps, leaving the closed boundaries to contain the filling-in process and to thereby support depth-selective surface percepts. Surface contour and boundary pruning signals hereby work together to generate 3D percepts based on successfully filled-in surface regions.
For example, the open boundary at Depth 2 in V1 and the V2 pale stripes of **Figure [9](#F9){ref-type="fig"}** can be created due to a monocularly viewed vertical boundary that is seen by only one eye, as occurs during daVinci stereopsis ([@B68]; [@B32]; [@B8]), and by a pair of horizontal boundaries that do not give rise to strong binocular disparities. Such depth-non-selective boundaries are projected to all depth planes along the line of sight ([@B45]; [@B8]). The closed boundary at Depth 1 in **Figure [9](#F9){ref-type="fig"}** is due to these boundaries plus a left vertical boundary that is formed at that depth due to binocular disparity matching between the two eyes. As a result of surface filling-in Depth 1 of the V2 thin stripes, and the resultant formation of surface contours only at Depth 1, the closed boundary at Depth 1 is strengthened, whereas the spurious open boundary at Depth 2 is inhibited by the on-center off-surround surface contour feedback signals within position and across depth from V2 thin stripe surfaces to V2 pale stripe boundaries.
From Boundary Pruning to Figure-Ground Separation
-------------------------------------------------
Remarkably, by eliminating the spurious boundaries, the off-surround signals that are activated by surface contours also enable figure-ground separation to proceed. They do so by separating occluding and partially occluded surfaces onto different depth planes, after which partially occluded boundaries and surfaces can be amodally completed behind their occluders without interference from the now-inhibited spurious boundaries. For example, the three rectangles in **Figure [10A](#F10){ref-type="fig"}** are perceived as a vertical rectangle in front of a partially occluded horizontal rectangle. Due to the action of surface contours, the redundant copy of the vertical rectangle at a further depth (denoted by D2 in **Figure [10A](#F10){ref-type="fig"}**) is inhibited, thereby enabling the horizontal boundaries corresponding to the smaller rectangles to be colinearly completed within depth D2. In response to the picture in **Figure [10B](#F10){ref-type="fig"}**, the redundant vertical rectangular boundary is inhibited at depth D2, thereby restoring the boundary fragments at depth D2 that previously were inhibited by the D2 vertical boundaries at end gaps. For this reason, end gaps are not seen in the final depthful percept.
{ref-type="fig"}** 10 propose how end gap boundaries are eliminated. **(B)** This figure is composed of two abutting rectangles. Although there is no completion of the horizontal rectangle behind the vertical rectangle, a 3D percept can nonetheless be generated using mechanisms summarized in **Figure [11](#F11){ref-type="fig"}** and the surrounding text.](fpsyg-07-01102-g010){#F10}
The above interactions help to explain how, in response to the images in **Figure [3](#F3){ref-type="fig"}**, the inducers always appear to lie on a surface behind an occluding surface. Whether the end gaps form inside an illusory rectangle, as in response to the images in **Figure [3](#F3){ref-type="fig"}** (top row) or outside an illusory rectangle, as in response to the images in **Figure [3](#F3){ref-type="fig"}** (bottom row), they will be seen as further away than the surface that contains no end gap boundaries.
Further analysis is, however, needed to explain how any surfaces are consciously seen---since, as further explained below, V2 boundaries and surfaces are predicted to support recognition, but not conscious seeing, of completed occluders and their partially occluded objects---and also to explain how spurious end cut boundary fragments at both depths do not interfere with the recognition process.
How the Disparity Filter Eliminates Some Spurious Boundaries in the Near Depth
------------------------------------------------------------------------------
Although the boundaries containing end-gaps in response to the displays in **Figure [10A](#F10){ref-type="fig"}** are eliminated by surface contours at the further depth D2, they are not eliminated in this way from depth D1. These near-depth boundary fragments are eliminated by the disparity filter (**Figure [6](#F6){ref-type="fig"}**), an inhibitory circuit in layer 2/3 of V2 that operates along the line of sight and across depth to help solve the correspondence problem ([@B46]; [@B45]; [@B8]). In particular, the D1 near-depth end gap horizontal boundaries are inhibited by the D2 far-depth rectangular boundaries in **Figure [10](#F10){ref-type="fig"}** at corresponding positions by the disparity filter. This happens because the D2 far-depth rectangular boundary can be completed after surface contour signals act from the D1 closed vertical rectangular boundary to inhibit the spurious D2 vertical boundaries at the same positions. The completed D2 far-depth horizontal rectangular boundary can then contain an amodal surface filling-in process, and can generate its own surface contour signals. In contrast, the D1 end gap horizontal boundaries remain, and no boundary strengthening occurs along them. As a result, the D2 horizontal rectangle boundaries can inhibit the D1 end gap horizontal boundaries via the disparity filter, more than conversely.
Although the disparity filter can eliminate the near-depth end gap horizontal boundaries in response to the image in **Figure [10A](#F10){ref-type="fig"}**, it cannot do so in response to the image in **Figure [10B](#F10){ref-type="fig"}**. This is because the D2 far-depth boundary is not closed in this case after surface contour signals act from the D1 vertical rectangular boundary, and thus is not strengthened by its own surface contour feedback signals. The same kind of situation occurs in response to the fragmented inducers in **Figure [3](#F3){ref-type="fig"}**. How, then, are end gap near-depth D1 horizontal boundaries eliminated in this case?
From Unoccluded and Occluded Recognition in V2 to Unoccluded Seeing in V4
-------------------------------------------------------------------------
In order to explain how these spurious boundaries are also eliminated, it needs to be explained how additional mechanisms generate the *modal*, or consciously visible, percepts of the unoccluded parts of both occluding and occluded objects in depth. FACADE theory proposes how boundaries and surfaces may be *amodally* completed in V2 for purposes of recognition, but also that conscious qualia of the unoccluded surfaces of opaque objects are predicted to be represented in V4 due to a *surface-shroud resonance* that is triggered between V4 and the posterior parietal cortex (PPC); see [@B40] for a discussion of these resonant dynamics and the data that they help to explain. These proposed V2 and V4 representations enable the brain to complete the representations of partially occluded objects behind their occluders in V2 for purposes of object recognition, without forcing all occluders to appear transparent, which would be the case if the completed boundaries and surfaces that are illustrated in **Figure [10A](#F10){ref-type="fig"}** could generate visible surface qualia. How these V2 and V4 mechanisms may cooperate to achieve both effective recognition and seeing were first described in [@B36], [@B37]) and then further developed and simulated in many further articles; e.g., [@B63] and [@B29]. As noted above, [@B55] additionally explained and simulated how both opaque and transparent percepts can be generated using the same model cortical dynamics.
Before summarizing these V2-to-V4 mechanisms for conscious seeing, it is worth noting here that surface contour signals also help to control where the eyes look and to thereby help to regulate how the brain learns invariant object categories. The first role arises because surface contour signals are strongest at the distinctive features of an attended object, such as at high curvature positions along a boundary. In addition to the (thin stripe)-to-(pale stripe) feedback that enhances some boundaries while pruning others, a parallel pathway, that is predicted to occur through cortical area V3A, clarifies how these enhanced surface contour positions can also determine target positions of eye movements that explore an attended object's surface. These signals are proposed to determine where the eyes will look next on an attended surface, and thereby enable inferotemporal cortex to learn view-, size-, and positionally-invariant object categories as the eye movements explore this surface. Thus, the 3D LAMINART model is part of a more comprehensive 3D ARTSCAN Search architecture for active vision wherein 3D boundary and surface representations help to control eye movements for attending, seeing, searching, learning, and recognizing invariant object categories ([@B30]; [@B39]; [@B10]; [@B31]; [@B13]; [@B52]).
Boundary Enrichment and Surface Pruning in V4
---------------------------------------------
To set the stage for explaining these V2-to-V4 processes, keep in mind that the boundary pruning process spares the closest surface representation that successfully fills-in at a given set of positions, while removing redundant copies of the boundaries of occluding objects that would otherwise form at further depths. This process illustrates "the asymmetry between near and far". When the competition from redundant occluding boundaries is removed, the boundaries of partially occluded objects can be amodally completed behind them on boundary copies that represent further depths, as in the percept induced by the display in **Figure [10A](#F10){ref-type="fig"}**. Moreover, when the redundant occluding boundaries collapse, the redundant surfaces that they momentarily supported collapse as well. Occluding surfaces are hereby seen to lie in front of occluded surfaces.
These surface representations in V2 are depth-selective due to their depth-selective capture by binocular boundaries, but they do not combine brightness and color signals from both eyes (**Figure [5](#F5){ref-type="fig"}**). They are said to be computed within *monocular* Filling-In-DOmains, or FIDOs. The computation of binocular surfaces that combine brightness and color signals from both eyes is proposed to take place in V4 (**Figure [5](#F5){ref-type="fig"}**). These networks are called *binocular* FIDOs. Here monocular surface signals from both eyes are binocularly matched (pathways 8 in **Figure [5](#F5){ref-type="fig"}**). The successfully matched binocular signals are pruned by inhibitory signals from the monocular FIDOs (pathways 9 in **Figure [5](#F5){ref-type="fig"}**). These sur*face pruning* inhibitory signals eliminate redundant feature contour signals at at their own positions and further depths. As a result, occluding objects cannot redundantly fill-in surface representations at multiple depths. This surface pruning process is a second example of the "the asymmetry between near and far".
As in the case of the monocular FIDOs, the feature contour signals to the binocular FIDOs can initiate filling-in only where they are spatially coincident and orientationally aligned with binocular boundaries. Boundary pathways 10 in **Figures [5](#F5){ref-type="fig"}** and **[6](#F6){ref-type="fig"}** hereby carry out depth-selective surface capture of the binocularly matched feature contour signals that survive surface pruning. In all, the binocular FIDOs fill-in feature contour signals that: (a) survive within-depth binocular feature contour matching (via pathways 8) and across-depth feature contour inhibition (via pathways 9); (b) are spatially coincident and orientationally aligned with the binocular boundaries (pathways 10); and (c) are surrounded by a connected boundary, or fine web of such boundaries.
In addition, at the binocular FIDOs, the binocular boundaries of nearer depths are added topographically to those that represent further depths (e.g., **Figure [11B](#F11){ref-type="fig"}**). This third instance of the asymmetry between near and far is called *boundary enrichment*. When the vertical right boundary of the vertical rectangle at depth D1 in V4 enriches the boundaries at depth D2, a closed horizontal rectangular boundary is completed at D2, as shown in **Figure [11C](#F11){ref-type="fig"}**. This closed boundary can then modally fill-in its surface brightness at D2. These enriched boundaries prevent opaque occluding objects, such as the D1 vertical rectangle in **Figure [11C](#F11){ref-type="fig"}**, from looking transparent by duplicating its boundaries at further depths, and thereby blocking filling-in of occluded objects behind them, much as the horizontal rectangle at D2 is prevented from filling-in behind the vertical rectangle at D1 in **Figure [11C](#F11){ref-type="fig"}**.
{ref-type="fig"}**. In this case, the end gap boundaries at depth D1 in **(A)** cannot be eliminated, as they can in response to the percept generated by **Figure [10A](#F10){ref-type="fig"}**, by the disparity filter in V2 after surface contour feedback strengthens closed boundaries at the pale stripes from thin stripes. This is true because the boundary at depth D2 is not closed; see **(A)**. On the other hand, this boundary is closed by boundary enrichment in V4; see **(B)**. As a result, top-down attention from the filled-in surfaces in V4 (see **(C)**) can strengthen the boundaries of closed regions in V2 (see thicker lines in **D**). After this happens, the disparity filter in V2 can eliminate the end gap boundary at depth D1 in **(A)**.](fpsyg-07-01102-g011){#F11}
The total filled-in surface representation across all binocular FIDOs---after all three processes of boundary pruning, surface pruning, and boundary enrichment act---represents the visible surface percept. It is called a FACADE representation because it combines properties of Form-And-Color-And-DEpth. As to the three asymmetries between near and far, it is possible that they arise during development due to the asymmetric optic flows that are caused by moving forward much more than backward.
Top--Down Attention from V4 to V2 Eliminates End Gap Boundaries via Disparity Filter
------------------------------------------------------------------------------------
As noted above, although the disparity filter can eliminate the D1 near-depth end gap horizontal boundaries in response to the image in **Figure [10A](#F10){ref-type="fig"}**, it cannot do so in response to the image in **Figure [10B](#F10){ref-type="fig"}** because the D2 far-depth boundary is not closed in this case after surface contour signals act from the D1 vertical rectangular boundary, and thus is not strengthened by its own surface contour feedback signals. The same kind of situation occurs in response to the fragmented inducers in **Figure [3](#F3){ref-type="fig"}**. Given the above discussion about how V4 boundaries and surfaces form, it is now possible to explain how end gap near-depth D1 horizontal boundaries are eliminated in this case.
This is accomplished by top--down feedback from the V4 filled-in surfaces to their generative V2 boundaries (not shown in **Figure [6](#F6){ref-type="fig"}**). These top--down signals are contour-sensitive and obey the ART Matching Rule (e.g., [@B11], [@B12]), which predicts how top-down object attention works. The ART Matching Rule is defined by a modulatory on-center, off-surround network. The modulatory on-center can select and gain-amplify features within it, while the off-surround can inhibit features at other positions in the broad off-surround. The predicted properties of this network have been supported by many psychological and neurobiological data, and there is a convergence among models of attention about the mathematical form that the rule should take. See [@B40] for a review.
In the present instance, the modulatory on-centers of the completed rectangles at each depth, D1 and D2, in V4 can strengthen the corresponding boundaries at their respective positions and depths in V2, while inhibiting other boundaries in their off-surrounds, as illustrated in **Figure [11D](#F11){ref-type="fig"}**. The disparity filter can then eliminate the spurious end gap boundaries at depth D1 in V2 that are generated by the image in **Figure [10B](#F10){ref-type="fig"}**.
The 3D boundary and surface representations that are depicted in **Figures [10](#F10){ref-type="fig"}** and **[11](#F11){ref-type="fig"}** provide an explanation of how the fragmented images in **Figure [3](#F3){ref-type="fig"}**, each of whose inducers is caricatured by the image in **Figure [10B](#F10){ref-type="fig"}**, generate their depthful figure-ground percepts, notably why the relative depths of figure and ground depend on the positions of the T-junctions relative to the completed boundaries, but not on the relative inducer contrasts that caused them. In response to the fragmented images in **Figure [3](#F3){ref-type="fig"}**, these boundaries need to be completed by bipole grouping cells before T-junctions can be created at the fragmented inducers, unlike in response to the images in **Figure [10](#F10){ref-type="fig"}**. Once that happens, surface filling-in within closed boundaries ensues. **Figures [10](#F10){ref-type="fig"}** and **[11](#F11){ref-type="fig"}** clarify how the boundary and surface representations within V2 can lead to recognition of figure and ground objects in V2, without these representations also leading to visible surface qualia (see **Figure [9](#F9){ref-type="fig"}**). The filled-in surface representations within V4 are predicted to support conscious percepts of the qualia of the unoccluded parts of opaque surfaces, while their boundaries also enhance the strength of the boundary fragments at corresponding positions in V2.
Although the present exposition focuses on the perception of opaque surfaces in V4, both unique and bistable transparent percepts have also been explained by these FACADE and 3D LAMINART mechanisms ([@B55]).
Conclusion
==========
This article presents additional experimental evidence to complement the fact that many cells in cortical area V2 that are sensitive to border ownership, and thus implicated in the process of figure-ground perception, also exhibit a preferred contrast polarity. The experimental results here, with configurations that match previously established criteria for sign-invariant boundary grouping, show that contrast polarity is often unimportant in determining what part of a 2D picture generates a 3D percept of a closer figure, and what part generates a 3D percept of a further background. Both same-polarity and mixed-polarity sets of figural inducers, with either darker or lighter contrasts compared to the background, can generate the same percepts of relative depth.
The results support the hypothesis that V2 is just one stage in a cortical hierarchy that also includes V1 and V4 in the generation of surface percepts with figure-ground properties. Using model interactions among all of these cortical areas, FACADE theory and the 3D LAMINART model explain the psychophysical experimental data here, as well as many other psychophysical data about 3D vision and figure-ground perception in previously published articles. These models also explain many data about identified cells and circuits in these cortical areas, notably, in [@B43] all the key V2 data that have been reported in neurophysiological experiments about border ownership and related figure-ground properties by the von der Heydt laboratory.
Author Contributions
====================
All authors listed, have made substantial, direct and intellectual contribution to the work, and approved it for publication.
Conflict of Interest Statement
==============================
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
[^1]: Edited by: *Qasim Zaidi, State University of New York, USA*
[^2]: Reviewed by: *Greg Francis, Purdue University, USA; Baingio Pinna, University of Sassari, Italy*
[^3]: This article was submitted to Perception Science, a section of the journal Frontiers in Psychology
| {
"pile_set_name": "PubMed Central"
} |
This article is part of the Thematic Series \"Organic free radical chemistry\".
Introduction
============
Radical-trapping (chain-breaking) antioxidants are arguably the most important class of compounds used to protect organic materials from oxidative degradation from autoxidation ([Scheme 1](#C1){ref-type="fig"}) \[[@R1]--[@R2]\]. Phenolic compounds are almost universally used for this purpose -- for industrial/commercial applications as well as in nature -- since they possess inherently high reactivities to chain-carrying peroxyl radicals (ROO^•^) and are readily manipulated to adjust their physical properties for use under specific conditions. The mechanism of the reaction involves the formal transfer of an H-atom from the phenol (ArOH) to a peroxyl radical ROO^•^ (reaction 1 in [Scheme 2](#C2){ref-type="fig"}).
{#C1}
{#C2}
In general, the resultant phenoxyl radical (ArO^•^) is sufficiently unreactive toward the substrate (RH) that it reacts with a second peroxyl radical (reaction 2 in [Scheme 2](#C2){ref-type="fig"}), thereby breaking two oxidative chains per molecule of antioxidant -- a ratio commonly referred to as the stoichiometric factor (*n*). However, under some circumstances (e.g. when diffusion of the antioxidant is impeded and it has limited opportunity to encounter other radical species), it is possible for the antioxidant-derived phenoxyl radical to propagate the chain reaction (reaction 3 in [Scheme 2](#C2){ref-type="fig"}). The most relevant example of this is so-called 'tocopherol mediated peroxidation' (TMP), which occurs when α-tocopherol (the most biologically active form of vitamin E) is left alone to protect the lipid core of low-density lipoproteins (LDL). LDL is the particle responsible for the distribution of cholesterol in blood plasma and whose oxidation has been linked to the development of cardiovascular disease. Under these conditions, α-tocopherol is not an effective radical-trapping antioxidant \[[@R3]--[@R4]\].
For this (and other) reason(s), radical-trapping antioxidants are rarely used alone -- be it in nature or industrial/commercial applications. Instead, organic substrates are generally protected from oxidation by the addition of a combination of antioxidants (or co-antioxidants) that function in a synergistic fashion, i.e. they inhibit autoxidation more effectively together than would be expected from the simple additive contributions of their individual antioxidant activities. The interplay of α-tocopherol and ascorbate (vitamin C) in preventing the oxidation of LDL lipids is perhaps the best-known example of such synergism, since the regeneration of α-tocopherol by reduction of the α-tocopheroxyl radical by ascorbate prevents TMP, and effectively turns a water-soluble reducing equivalent into a lipid-soluble one \[[@R5]--[@R7]\].
In recent years, some of us have worked to understand the kinetic and thermodynamic basis for synergism among radical-trapping antioxidants in homogeneous solution, which is summarized in [Scheme 3](#C3){ref-type="fig"} \[[@R8]--[@R9]\]. When two (or more) antioxidants are present in a system, the principal antioxidant (AH) is identified as that which reacts most rapidly with peroxyl radicals than the other(s), the so-called co-antioxidant(s) (co-AH), i.e. *k* ~inh~' \> *k* ~inh~'' in reaction 4 and reaction 5 ([Scheme 3](#C3){ref-type="fig"}), respectively. As a result of its greater reactivity, AH must be consumed before co-AH. However, if the equilibrium in reaction 6 ([Scheme 3](#C3){ref-type="fig"}) is favourable, co-AH can regenerate AH for further reaction with ROO^•^. Of course, this is only true if equilibration is faster than consumption of the AH-derived radical by reaction with a second peroxyl radical, i.e. *k* ~r~\[co-AH\] \> *k* ~7~\[ROO^•^\] for reaction 6 and reaction 7 ([Scheme 3](#C3){ref-type="fig"}), respectively. However, this is a condition that is generally easily met since \[ROO^•^\]~ss~ in the presence of AH/co-AH must be very low and *k* ~r~ for phenol/phenoxyl couples is normally ≥10^4^ M^−1^s^−1^ \[[@R8]--[@R9]\].
{#C3}
Based on this model, in order for synergism to occur among equilibrating phenolic antioxidants it is necessary that the principal antioxidant has both a higher reactivity with peroxyl radicals (*k* ~inh~) and a higher O--H bond dissociation enthalpy (BDE), as compared to the co-antioxidant. Unfortunately, this is a very demanding requirement since *k* ~inh~ and the O--H BDE are inversely correlated according to well-established Evans--Polanyi relationships \[[@R2],[@R10]\].
Over the years, our research groups have developed novel air-stable and highly reactive radical-trapping chain-breaking antioxidants based on either 3-pyridinol (**1**) or 5-pyrimidinol (**2**) core structures ([Fig. 1](#F1){ref-type="fig"}) \[[@R11]--[@R17]\]. Compared to equivalently-substituted phenols, these compounds have been shown to possess stronger O--H bonds (e.g. +1.4 kcal/mol for **1** and +2.5 kcal/mol for **2** relative to **3**) while maintaining similar or higher reactivity toward peroxyl radicals \[[@R11],[@R13]\]. As is the case for phenols, 3-pyridinols and 5-pyrimidinols can be substituted with electron-donating groups to weaken their O--H bonds and increase their rates of reaction with peroxyl radicals in a predictable fashion \[[@R11]\]. Based on these facts, we surmised that 3-pyridinols and 5-pyrimidinols would be ideal principal antioxidants in synergistic co-antioxidant systems with phenols. Herein we describe the rational design and kinetic characterization of such systems based on the combination of suitably substituted 3-pyridinols and 5-pyrimidinols (**4**--**9**) with conventional phenolic antioxidants (**10**--**12**).
{#F1}
Results and Discussion
======================
**Synthesis** ***.*** The preparation of compounds **4a**, **4b** and **5**--**8** involved installation of the aryl alcohol moiety as the final step via a Cu-catalyzed benzyloxylation/hydrogenolysis sequence on the corresponding pyri(mi)dyl halides, whereas the preparation of **4c**, **4d** and **9** followed a route starting from pyridoxamine, wherein the aryl alcohol is present throughout the sequence. Details are provided in the Experimental section and/or the cited references.
**Reactivity with peroxyl radicals** ***.*** To set up a rational framework for the design of co-antioxidant systems, the rate constants for the reactions of various pyridinols and pyrimidinols with peroxyl radicals (*k* ~inh~) were measured by the well-established inhibited autoxidation of styrene (or cumene) in chlorobenzene at 303 K. These measurements also included experiments with three well-established phenolic antioxidants: 2,6-di-*tert*-butyl-4-methylphenol (BHT, **10**), 2,6-di-*tert*-butyl-4-methoxyphenol (DBHA, **11**), a hindered analogue of the widely employed BHA, and 2,2,5,7,8-pentamethylchroman-6-ol (PMHC, **12**), a synthetic analogue of α-tocopherol lacking its phytyl sidechain. While some of these rate constants have been reported in the literature, we felt it necessary to determine them all under the exact same conditions in order to be able to predict and/or rationalize observations made when the antioxidants are used in combinations. The results are given in [Table 1](#T1){ref-type="table"}. It must be pointed out that where previous data have been obtained under comparable conditions, our data are in excellent agreement -- with *k* ~inh~ values usually within a factor of two. As a result, the reactivity trends parallel those that have been observed before: the bicyclic naphthyridinol compounds **9a**--**c** generally possess the highest reactivities, followed by the aminopyridinols **4** and aminopyrimidinols **6**, and finally the alkoxypyridinols **5** and alkoxypyrimidinols **7**. A previously unstudied compound -- the 2,4-dimethylpyrrole-substituted pyrimidinol **8** -- was the least reactive pyri(mi)dinol we studied, with a rate constant almost 200 fold lower than that of the analogous dimethylamino-substituted pyrimidinol **6b** (*k* ~inh~ = 4.4 × 10^4^ versus 7.4 × 10^6^ M^−1^s^−1^, respectively). Clearly, the 2,4-dimethylpyrrole substituent is not as electron-donating as a dimethylamino substituent (the O--H bond in **8** is 6.6 kcal/mol stronger than that in **6b**, vide infra). These results provide an explanation for the significant differences in the radical scavenging activities of pyridinols bearing these substituents in recently reported cell-based assays \[[@R18]\].
######
Rate constants for the reactions of **4**--**12** with peroxyl radicals (*k* ~inh~) at 303 K obtained from AIBN-initiated inhibited autoxidations of styrene (50% v/v) in either chlorobenzene (PhCl) or acetonitrile (CH~3~CN). O--H Bond dissociation enthalpies calculated using CBS-QB3 are given along with available experimental data where possible.
------------- ------------------------ --------------------- -------------------------------------- --------------------------- ---- ----------------
*k* ~inh~ (PhCl) *k* ~inh~ (CH~3~CN) *k* ~inh~ (PhCl)/*k* ~inh~ (CH~3~CN) BDE~OH~ ^calc^(^exp^)^a^\
/kcal/mol
/M^−1^s^−1^ *n* /M^−1^s^−1^ *n*
**4a** (3.6 ± 0.6) × 10^6\ b^ 1.9 (5.4 ± 0.2) × 10^5^ 2.0 7 77.9
**4b** (1.4 ± 0.6) × 10^7\ c^ 1.9 (3.0 ± 0.3) × 10^6^ 2.0 5 74.8 (75.9)
**4c** (2.0 ± 1.0) × 10^6\ d^ 2.1 (3.1 ± 0.6) × 10^5^ 2.0 6 78.0
**4d** (8.5 ± 2.8) × 10^6\ d^ 1.9 (3.0 ± 0.4) × 10^6^ 2.0 3 74.5
**5a** (7.3 ± 0.4) × 10^4^ 2.2^e^ (4.1 ± 0.3) × 10^4^ 1.9^e^ 18 82.4
**5b** (4.4 ± 0.7) × 10^5\ c^ 2.1 (3.8 ± 0.9) × 10^4^ 1.9^e^ 12 78.9
**6a** (2.0 ± 0.6) × 10^6\ b^ 2.1 (3.0 ± 0.7) × 10^5^ 1.9 7 78.3
**6b** (7.4 ± 0.6) × 10^6^ 2.1 (1.0 ± 0.3) × 10^6^ 1.8 7 75.6 (77.1)
**7a** (3.1 ± 0.4) × 10^5^ 2.0 (1.1 ± 0.6) × 10^4\ f^ 2.1^e^ 28 80.9 (81.4)
**7b** (3.7 ± 0.3) × 10^5^ 2.0 (1.4 ± 0.5) × 10^4^ 2.0^e^ 26 80.9^g^
**8** (4.4 ± 1.0) × 10^4^ 2.0^e^ (1.3 ± 0.5) × 10^3^ n.d. 34 81.8
**9a** (5.5 ± 3.1) × 10^7\ h^ 1.3 (9.2 ± 1.9) × 10^6^ 1.7 6 74.9 (75.2^i^)
**9b** (7.8 ± 0.8) × 10^7\ h^ 1.5 (1.3 ± 0.3) × 10^7^ 1.7 6 75.0 (75.2^i^)
**9c** (1.5 ± 0.2) × 10^7\ j^ 2.0 (2.9 ± 1.4) × 10^6^ 2.0 5 75.4
**10** (1.1 ± 0.2) × 10^4\ k^ 2.0^e^ n.d. n.d. \- 78.7 (79.9^l^)
**11** (1.1 ± 0.2) × 10^5\ k^ 2.0 (2.5 ± 1.0) × 10^4\ f^ n.d. 4 75.5 (77.2^l^)
**12** (3.2 ± 0.5) × 10^6\ k^ 2 ^l^ (6.5 ± 0.8) × 10^5\ f^ 2^m^ 5 77.7 (77.1^l^)
------------- ------------------------ --------------------- -------------------------------------- --------------------------- ---- ----------------
^a^Experimental values (in benzene) obtained by REqEPR at 298 K are from \[[@R12]--[@R13]\] and have been corrected for the revised O--H BDE of phenol \[[@R19]\]. ^b^Values for **4a** and **6a** were previously determined as 4.8 × 10^6^ M^−1^s^−1^ and 1.1 × 10^6^ M^−1^s^−1^ at 303 K from the inhibited oxidation of styrene in PhCl and as 1.1 × 10^7^ M^−1^s^−1^ and 6.5 × 10^6^ M^−1^s^−1^ at 310 K in benzene by radical clock \[[@R16]\]. ^c^Values of 1.6 × 10^7^ M^−1^s^−1^ and 2.9 × 10^5^ M^−1^s^−1^ were previously reported for **4b** and **5b** from inhibited styrene oxidation in PhCl at 303 K \[[@R14]\]. ^d^Values for **4c** and **4d** of 3.3 × 10^6^ M^−1^s^−1^ and 8.7 × 10^6^ M^−1^s^−1^ measured by inhibited autoxidation of styrene in PhCl and of 1.6 × 10^6^ M^−1^s^−1^ and 1.4 × 10^7^ M^−1^s^−1^ in benzene at 310 K by radical clock \[[@R17]\]. ^e^Determined from the inhibited autoxidation of cumene at 303 K. ^f^Values of 7.9 × 10^2^ M^−1^s^−1^, 2.2 × 10^4^ M^−1^s^−1^ and 6.8 × 10^5^ M^−1^s^−1^were previously measured for **7a**, **11**, **12** from the autoxidation of styrene in acetonitrile at 303 K \[[@R20]\]. ^g^Assumed the same as **7a**. ^h^Values of 6.1 × 10^7^ M^−1^s^−1^ and 5.2 × 10^7^ M^−1^s^−1^ for **9a** and **9b** in benzene at 310 K were obtained by radical clock \[[@R15]\]. ^i^Measured for the analogue of **9a/b** with R = R' = H. ^j^The value of 3.1 × 10^7^ M^−1^s^−1^ in benzene at 310 K was obtained by radical clock \[[@R15]\] for an analogue of **9c**. ^k^Values of 1.4 × 10^4^ M^−1^s^−1^, 1.1 × 10^5^ M^−1^s^−1^ and 3.8 × 10^6^ M^−1^s^−1^ were previously determined for **10**, **11** and **12** in the inhibited autoxidation of styrene in PhCl at 303 K \[[@R21]\]. ^l^From \[[@R22]\]. ^m^Used as reference value.
To provide further insight into the relative reactivities of these compounds we also carried out measurements of *k* ~inh~ in acetonitrile as a representative polar solvent. We felt this was necessary since there is essentially no data available in the literature for the reactivity of the vast majority of these compounds in any media other than chlorobenzene (or benzene) and we wanted to examine the solvent-dependence of any synergism we observed (vide infra). The results demonstrate a significant kinetic solvent effect, which was most pronounced for the least reactive compounds. For example, *k* ~inh~ for the methoxy-substituted pyridinol **5a** dropped by a factor of 18 on going from chlorobenzene to acetonitrile, while the reactivity of the more reactive *N*,*N*-dimethylamino-substituted pyridinol **4a** dropped only a factor of 7. Likewise, while *k* ~inh~ for the 2,4-dimethyl-6-methoxy-3-pyridinol (**5b**) dropped 12-fold with the change in solvent, the reactivity of the equivalently-substituted, but less reactive, pyrimidinol **7a** dropped 28-fold.
Ingold has clearly demonstrated that formal H-atom transfer reactions of the type X--H + Y^•^ → X^•^ + H--Y, where X is an electronegative atom, can experience a large kinetic solvent effect (KSE). In fact, these reactions are slowed down in hydrogen-bond accepting (HBA) solvents as a result of H-bond formation between X--H and the solvent since the H-bonded complex is essentially unreactive to the abstracting radical; hence only the "free" fraction of X--H in solution can react \[[@R23]--[@R26]\]. This KSE (illustrated in [Scheme 4](#C4){ref-type="fig"}) is known to have major impact on the performance of phenolic antioxidants \[[@R2],[@R10],[@R27]--[@R28]\].
{#C4}
Since the formation of the H-bonded complex is driven by both the HBA ability of the solvent and the hydrogen-bond donating (HBD) ability of the H-atom donor, the KSEs evident in the data above reflect the H-bond acidity of the various radical-trapping antioxidants we have studied. On quantitative grounds, an empirical relationship between *K* ~solv~ and the HBD ability of the compound is provided by Abraham\'s equation ([Eq. 1](#FD1){ref-type="disp-formula"}), where  and  are empirical solvatochromic parameters (range: 0 to 1) quantifying the HBD and HBA ability of the two interacting partners (e.g. the phenol and the solvent), respectively, in the formation of a 1:1 H-bonded complex \[[@R29]--[@R30]\].
An  value of 0.37 \[[@R24]\] has been reported for α-tocopherol (expectedly identical to PMHC, **12**), while they have been estimated as 0.50 and 0.55 for compounds **4a** and **6a** respectively \[[@R2],[@R10]\], consistent with the larger KSEs on the reactions of the latter (ca. 7) relative to the former (ca. 5).
**FTIR measurements.** To put the HBD ability of the pyridinols and pyrimidinols on solid quantitative ground, we performed independent (non-kinetic) measurements of *K* ~solv~ for three representative compounds (**5b**, **6b** and **7b**) in three reference solvents of different HBA ability \[[@R30]\]: acetonitrile ( = 0.44), ethyl acetate ( = 0.45), and dimethyl sulfoxide ( = 0.78) using IR spectroscopy \[[@R25]\]. Representative results are shown in [Fig. 2](#F2){ref-type="fig"}.
{ref-type="disp-formula"} (b).](Beilstein_J_Org_Chem-09-2781-g003){#F2}
Addition of a HBA solvent to solutions of the pyri(mi)dinols in non-H-bonding CCl~4~ resulted in the progressive decrease of the IR signal corresponding to the "free" O--H stretch (\~3610 cm^−1^), accompanied by the growth of a broad intense band at lower frequency attributed to the O--H stretch of the H-bonded species ([Fig. 2](#F2){ref-type="fig"}). By fitting the data corresponding to the integrated IR signal for the free O--H versus the concentration of the HBA co-solvent to the expression in [Eq. 2](#FD2){ref-type="disp-formula"} as illustrated in [Fig. 2](#F2){ref-type="fig"}, the values of *K* ~solv~ collected in [Table 2](#T2){ref-type="table"} could be obtained.
It should be noted that, for any of the compounds that were investigated, there is good agreement between the  values obtained by [Eq. 1](#FD1){ref-type="disp-formula"} from equilibrium constants in ethyl acetate and DMSO, while the value measured for acetonitrile is consistently lower. Indeed, for each of these compounds, *K* ~solv~ measured for acetonitrile is lower than that for ethyl acetate despite the fact that the two solvents are attributed essentially the same HBA ability by Abraham's β~2~ ^H^ scale (0.44 versus 0.45). A similar trend is observed in available literature kinetic data; the rate constants for formal H-atom transfer from a variety of phenols to a variety of radicals (e.g*.* alkyl, alkoxyl, peroxyl and DPPH) is consistently higher in acetonitrile than in ethyl acetate \[[@R23]--[@R28]\] strongly suggesting that the  value for acetonitrile needs revision. As a result, we suggest averaging the  values determined for **5b**, **6b** and **7b** in EtOAc and DMSO, resulting in 0.55, 0.53 and 0.65, respectively. Such values are in line with other phenol-type antioxidants \[[@R2],[@R10]\]. The same values can then be used as inputs in [Eq. 1](#FD1){ref-type="disp-formula"} to obtain an average  of 0.39 for acetonitrile. This value is in accord with that obtained from kinetic measurements: for instance, using the reaction of α-tocopherol with *tert*-butoxyl radicals as model, it was shown that acetonitrile has the same HBA ability as water \[[@R31]\], which is attributed a reliable  = 0.38 \[[@R29]--[@R30]\]. As such, we recommend a value of  of 0.39 for acetonitrile.
######
FTIR measured equilibrium constants at 298 K for H-bonding of solvents with selected antioxidants (*K* ~solv~) and corresponding  values calculated by [Eq. 1](#FD1){ref-type="disp-formula"}.
-------- -------------- ------------------ -------------------------------------------- --------
Solvent *K* ~solv~/M^−1^  KSE^a^
**5b** CH~3~CN 3.1 ± 0.2 0.49
EtOAc 5.5 ± 0.3 0.55
DMSO 116.1 ± 11.2 0.55
average^b^ 0.55 12
**6b** CH~3~CN 3.0 ± 0.3 0.49
EtOAc 4.7 ± 0.5 0.53
DMSO 95.0 ± 5.9 0.53
average^b^ 0.53 7
**7b** CH~3~CN 6.9 ± 1.8 0.60
EtOAc 14.1 ± 0.9 0.68
DMSO 285.0 ± 9.6 0.62
average^2^ 0.65 26
-------- -------------- ------------------ -------------------------------------------- --------
^a^KSE = kinetic solvent effect, taken from data in [Table 1](#T1){ref-type="table"}. ^b^Average of the data in EtOAc and DMSO only, see text.
**Computational thermodynamics** ***.*** The rational design of synergistic co-antioxidant mixtures requires knowledge of not only the kinetics of the reactions of the antioxidants with peroxyl radicals, but also the relative stabilities of the antioxidant-derived radicals, since synergism relies on the position of the equilibrium of reaction 6 ([Scheme 3](#C3){ref-type="fig"}) \[[@R8]--[@R9]\], which is related to the difference in the O--H BDEs of the equilibrating antioxidants as in [Eq. 3](#FD3){ref-type="disp-formula"}.
In order to complete the necessary framework of kinetic and thermodynamic data, we computed the O--H BDEs of compounds **4**--**12** using quantum chemical methods. The calculations were carried out at the CBS-QB3 level of theory \[[@R32]\], since this approach has been shown to provide highly accurate O--H BDEs in phenols and related compounds \[[@R19],[@R33]--[@R35]\]. The results of these calculations are given in [Table 1](#T1){ref-type="table"} alongside the limited experimental data obtained using the radical equilibration EPR (REqEPR) technique \[[@R12]--[@R13] [@R22]\]. The calculated BDEs are in very good agreement with the experimental values, with systematic deviations between 0.2 and 1.7 kcal/mol and, most importantly, they allow insight into the position of equilibrium (reaction 6, [Scheme 3](#C3){ref-type="fig"}) unfettered by differing experimental conditions.
**Co-antioxidant systems** ***.*** Given the foregoing kinetic and thermodynamic data, we next set out to design and test representative co-antioxidant mixtures. For simplicity we investigated only binary AH/co-AH mixtures. The baseline strategy consisted of selecting a pyridinol or a pyrimidinol as principal antioxidant (AH) capable of providing maximum radical-trapping kinetics to the mixture (higher *k* ~inh~), but at the same time a sufficiently high O--H BDE to be regenerated by the co-antioxidant, co-AH (vide supra), which was selected among the conventional phenols **10**--**12**.
The autoxidation of an organic substrate (e.g. styrene) thermally initiated at a constant rate, *R* ~i~, by an azo-initiator will consume oxygen at a constant rate in the absence of an inhibitor. In the presence of a very effective antioxidant such as **4a** (*k* ~inh~ = 3.6 × 10^6^ M^−1^s^−1^, see [Table 1](#T1){ref-type="table"}) a plot of oxygen uptake versus time shows a clear inhibition period of length τ~0~ that depends on the concentration of AH and the stoichiometric factor (*n* \~ 2 for all tested antioxidants, see [Table 1](#T1){ref-type="table"}). During the inhibited period (cf. [Fig. 3](#F3){ref-type="fig"}), i.e. until AH has been consumed, the rate of oxygen consumption is almost completely suppressed, after which it resumes at the uninhibited rate. [Fig. 3](#F3){ref-type="fig"} also shows that an equivalent amount of a modest antioxidant such as **11** (*k* ~inh~ = 1.1 × 10^5^ M^−1^s^−1^, see [Table 1](#T1){ref-type="table"}) does not produce a neat inhibition of the autoxidation under the same conditions, but instead simply retards oxygen uptake, since the propagation of autoxidation can compete effectively with inhibition. However, when equimolar amounts of **4a** and **11** are present, a clear inhibited period is observed -- as was the case for **4a** alone, but its duration is twice what it was in the absence of the equivalent of **11**. This result implies that **11** can regenerate **4a** from its corresponding aryloxyl radical. This reaction is driven by the fact that **4a** has an O--H bond which is 2.4 kcal/mol stronger (77.9 kcal/mol) than the O--H bond in **11** (75.5 kcal/mol). The addition of another equivalent of **11** extends the inhibited period to three times that of **4a** alone, clearly demonstrating that it is effectively used as the sacrificial reductant during the inhibited period.
{#F3}
The duration of the inhibited period (τ) is related to the concentration ratio of the principal antioxidant and co-antioxidant by a proportionality constant α ([Eq. 4](#FD4){ref-type="disp-formula"}), which represents the efficiency with which AH is regenerated by co-AH, which can be written in terms of the rate constants of the relevant competing reactions in [Scheme 3](#C3){ref-type="fig"} as in [Eq. 5](#FD5){ref-type="disp-formula"}; thus, its value may lie between 0 (no regeneration of AH by co-AH) and 1 (complete regeneration of AH by co-AH).
In other words, α is a measure of synergism in the co-antioxidant system. Since the ratio *k* ~8~/*k* ~7~ (see [Scheme 3](#C3){ref-type="fig"}) for phenolic antioxidants is normally \~1 \[[@R8]\], the efficiency of regeneration depends almost entirely on *K* ~r~. Moreover, since H-atom transfer between phenols usually proceeds with a negligible change in entropy, α depends largely on the difference in the O--H BDEs of the two co-antioxidants ([Eq. 3](#FD3){ref-type="disp-formula"}). If *K* ~r~ \<\< 1, regeneration will be inefficient and no synergism will be observed; under those circumstances the co-antioxidants will simply behave in an additive fashion, as illustrated in [Fig. 3](#F3){ref-type="fig"} for the combination of **6b** (BDE = 75.6 kcal/mol) and **10** (BDE = 78.7 kcal/mol).
Although, in principle, synergism can occur with any AH/co-AH ratio (e.g. [Fig. 4](#F4){ref-type="fig"}), the efficiency often changes to some extent as a function of such ratio, as well as with the actual experimental conditions. For simplicity, all co-antioxidant mixtures were investigated under comparable settings in the low μM range with AH/co-AH ratios of 1:1 and 1:2. As can be seen from [Table 3](#T3){ref-type="table"}, several of the antioxidant mixtures investigated showed good synergism in chlorobenzene (α \> 0.5); particularly the couples **4a**/**11**, **6a**/**11**, **6a**/**12**, **6b**/**11**, **6b**/**12**, **7a**/**11**, **7a**/**10** (similar to **7b**/**11**, **7b**/**10**), **9c**/**12**. From our results we can conclude that, in general, ΔBDE needs to be \> −1 kcal/mol to expect synergism based on the equilibrium of reaction 6. Not surprisingly, regeneration was more efficient when a pyrimidinol was used as the principal antioxidant due to the higher O--H BDEs of the pyrimidinols relative to equivalently substituted pyridinols. On the other hand, it should be noted that the efficiency α is not the only relevant parameter in determining the overall efficacy of a co-antioxidant mixture, since the apparent *k* ~inh~ of the mixture will be identical to that of the most reactive antioxidant in the mixture \[[@R8]\]. For instance, the mixture **4a**/**11** ([Fig. 3](#F3){ref-type="fig"}) is a significantly better antioxidant system than mixtures of **7b**/**11** ([Fig. 4](#F4){ref-type="fig"}) and **7b**/**10** ([Fig. 4](#F4){ref-type="fig"}), despite all systems having α = 1.
{#F4}
######
Regeneration efficiency (α) of a principal antioxidant (AH) by a co-antioxidant (co-AH) in the inhibited autoxidation of styrene in chlorobenzene or acetonitrile (50% v/v) at 303 K.^a^
-------- -------- ---------------- ------------ -------------
AH co-AH ΔBDE (AH-CoAH) α (PhCl) α (CH~3~CN)
**4a** **11** +2.4 1.0 ± 0.1 0.5 ± 0.1
**4c** **11** +2.5 \~0.1 \~0
**4d** **11** −1.0 \~0.1 \~0.1
**10** −4.2 \~0 \~0
**6a** **11** +2.8 1.0 ± 0.1 0.8 ± 0.1
**12** +0.6 0.8 ± 0.2 0.4 ± 0.2
**6b** **10** −0.4 0.5 ± 0.1 \~0
**11** +0.1 0.7 ± 0.1 0.5 ± 0.1
**10** −2.1 \~0 \~0
**7a** **11** +4.4 0.9 ± 0.1 n.d.^b^
**10** +2.2 0.9 ± 0.1 0.3 ± 0.1
**7b** **11** +4.4 1.0 ± 0.1 0.7 ± 0.2
**10** +2.2 1.0 ± 0.1 n.d*.* ^b^
**9c** **12** −2.3 0.6 ± 0.2 0.5 ± 0.1
-------- -------- ---------------- ------------ -------------
^a^Values are averaged on at least three independent experiments with AH/co-AH ratios of 1:1 and 1:2, in the concentration range 2--10 μM both for AH and co-AH. ^b^n.d*.* = not determined.
Since synergistic activity requires a favourable ΔBDE (vide supra), it seems reasonable to expect that the values of α should correlate with ΔBDE. Such a correlation is shown in [Fig. 5](#F5){ref-type="fig"}, in which there appears to be a clear sigmoidal relationship between α and ΔBDE. The correlation is sigmoidal since below ΔBDE values of ca. −1 kcal/mol little to no regeneration is observed, whereas above ca. 1 kcal/mol, regeneration is essentially quantitative.
![Regeneration efficiencies (α) observed in autoxidations of styrene in chlorobenzene (50% v/v) at 303 K inhibited by co-antioxidant mixtures as a function of the difference in the O--H BDEs of the principal (AH) and co-antioxidant (Co-AH) \[ΔBDE = BDE(AH) -- BDE(Co-AH)\]. The data highlighted in red correspond to the combinations of **9c**/**12** and **4c**/**11**.](Beilstein_J_Org_Chem-09-2781-g006){#F5}
There are two data points that do not lie on the correlation: the combinations of **4c** with **11**, and **9c** with **12**. The latter has ΔBDE = −2.3 kcal/mol, implying that **9c** should not be regenerated by **12**. However, because the phenolic co-antioxidant used in this case (**12**) is a very reactive antioxidant itself (*k* ~inh~ = 3.2 × 10^6^ M^−1^s^−1^ compared to 1.5 × 10^7^ M^−1^s^−1^ for **9c**) it also gives a pronounced inhibited period under the reaction condition (see [Supporting Information File 1](#SD1){ref-type="supplementary-material"} for oxygen uptake plots). As such, the additive contributions of **9c** and **12** give the appearance of synergism where only additivity exists. It should also be pointed out that although the simple additive contributions of the highly reactive antioxidants should give α = 1, a value of only ca. 0.6 is observed. This is likely due to the consumption of some **12** by autoxidation at higher concentrations of **9c** due to the longer inhibition period, which is known to lead to lower stoichiometric factors for highly similar compounds (i.e. **9a**, **9b**) \[[@R17]\]. This leaves the **4c**/**11** data point as the only real outlier; since it has ΔBDE = +2.5 kcal/mol, the much more reactive **4c** should be regenerated by the less reactive **11**. This result is puzzling in light of the fact that the structurally similar pyridinol **4a**, which has an essentially identical O--H BDE, is fully regenerated by **11**. The only structural feature which distinguishes **4c** from the other pyridinols and pyrimidinols in [Table 3](#T3){ref-type="table"} (and [Fig. 5](#F5){ref-type="fig"}) that have positive ΔBDE values (and therefore α \~ 1) is in the conformation of the substituent at the 4-position relative to the reactive hydroxyl moiety. Due to steric interactions between the adjacent ring methyl, the dimethylamino substituent in **4c** is rotated out of the plane of the aromatic ring \[[@R17]\]. On the basis of theoretical calculations, H-atom transfer between phenol and the phenoxyl radical is believed to be a proton-coupled electron transfer reaction, occurring via an approximately planar transition state wherein the unpaired electron is delocalized across both phenyl rings \[[@R36]\]. As such, it is difficult to envision how the conformation of the dimethylamino substituent in **4c** may slow the analogous reaction between the radical derived from **4c** and **11** relative to the other reaction couples. However, it should be pointed out that theoretical calculations on the phenol/phenoxyl H-atom transfer reaction do not include substituents in the ortho positions relative to the phenolic oxygen, which may change the transition state structure substantially.
Regardless of whether the foregoing rationalization is in fact correct, the lower than expected regeneration efficiency observed for the combination of **4c** and **11** underscores the fact that the actual value of α depends not only on the equilibrium in reaction 6 ([Scheme 3](#C3){ref-type="fig"}) -- and hence the ΔBDE -- but also on the absolute rates of the other reactions depicted in [Scheme 3](#C3){ref-type="fig"} \[[@R8]\]. In this connection, it is important to note that the values of α we measured were generally lower in acetonitrile than in chlorobenzene despite the fact that ΔBDE is expected to increase on going from chlorobenzene to acetonitrile due to the stronger H-bonding of the pyridinols and pyrimidinols ( \~ 0.5--0.7) to acetonitrile (vide supra) as compared to the sterically hindered phenols such as **10** and **11** ( \~ 0.2) \[[@R37]\]. The drop in α can only be explained by considering [Scheme 3](#C3){ref-type="fig"} in more detail. For regeneration of the principal antioxidant AH (the pyridinol/pyrimidinol) to occur, it is necessary that equilibrium of reaction 6 is faster than reaction 7 ([Scheme 3](#C3){ref-type="fig"}), i.e. *k* ~7~ × \[ROO^•^\]~SS~ \< *k* ~r~ × \[co-AH\]. In the presence of a good antioxidant AH (rapidly trapping peroxyl radicals by reaction 4 ([Scheme 3](#C3){ref-type="fig"})) this condition is easily met \[[@R8]\]. However, if the reactivity of AH is hampered by H-bonding to the solvent, the steady state concentration of peroxyl radicals may grow sufficiently to react competitively with co-AH, thereby decreasing the efficiency of regeneration \[[@R38]\].
Conclusion
==========
Herein we have provided the kinetic and thermodynamic rationale for the design of synergistic co-antioxidant systems employing highly reactive 3-pyridinol or 5-pyrimidinol antioxidants in combination with less reactive, but much less expensive, phenolic antioxidants. In several cases, the approach has shown to equal the performance of the best co-antioxidant systems designed by nature, such as the tocopherol/ascorbate system \[[@R5]\] or the tocopherol/catechol system \[[@R8]\]. In general, the most effective individual antioxidants, e.g. the bicyclic pyridinols (**9a--c**), pyridinols (**4b**) and pyrimidinols (**6d**) are not good partners for co-antioxidant systems because their O--H BDEs (74.8--75.6 kcal/mol) are too low. Instead, the slightly less reactive pyridinols and pyrimidinols (e.g. **4a**, **6a**, **7a**/**7b**), which have much stronger O--H bonds (\>78 kcal/mol), are the ideal candidates to be used with abundant, persistent phenols such as BHT (**11**). We anticipate that this work will prompt the use of antioxidant mixtures based on 3-pyridinol and 5-pyrimidinol antioxidants, in order to take advantage of the greater reactivities of these compounds, but to minimize the cost of doing so by making use of the inexpensive phenolic antioxidants typically used in industrial/commercial applications to regenerate them in situ.
Experimental
============
**Materials** ***.*** Solvents were of the highest grade commercially available (Fluka/Aldrich) and were used as received. 2,2,5,7,8-Pentamethyl-6-chromanol (PMHC, **12**, 97%) was commercially available (Aldrich) and used without further purification. Commercial 2,6-di-*tert*-butyl-4-methylphenol (BHT, **10**, 98%) and 2,6-di-*tert*-butyl-4-methoxyphenol (DBHA, **11**, 97%) were re-crystallized from hexane. Commercially available 2,2\'-azodiisobutyronitrile (AIBN ≥98%) was recrystallized from hexane and stored at −20 °C. Cumene (98%) and styrene (≥99%) were distilled under reduced pressure and percolated twice through silica and alumina prior to use. All solutions were prepared fresh immediately prior to use.
**Synthesis** ***.*** Compounds **4a**, **4b**, **6a** and **6b** were prepared as described in \[[@R39]\]. Compounds **4c** and **4d** were prepared as described in \[[@R17]\]. Compound **5b** was prepared as in \[[@R14]\]. Compound **7a** was prepared as in \[[@R13]\]. Compounds **9a** and **9b** were prepared as in \[[@R15]\], whereas compound **9c** was prepared as in \[[@R40]\].
**3-Hydroxy-6-methoxypyridine (5a).** A solution of 3-benzyloxy-6-methoxypyridine \[[@R39]\] in MeOH was treated with 10% Pd/C and the resulting black suspension was stirred at room temperature under an atmosphere of H~2~ (1 atm) overnight. The catalyst was removed by filtration through a pad of celite and the filtrate was concentrated under reduced pressure. The crude residue obtained was subjected to flash chromatography on silica gel (eluent: ethyl acetate/hexanes) and the product isolated in quantitative yield. ^1^H NMR (CDCl~3~) δ 9.32 (br s, 1H, exchanges with D~2~O), 7.75 (s, 1H), 7.25 (d, *J* = 8.6 Hz, 1H), 6.56 (d, 8.6 Hz, 1H), 3.84 (s, 3H); ^13^C NMR (CDCl~3~) δ 54.2, 111.0, 128.7, 132.3, 148.2, 158.3; HRMS (EI^+^) *m*/*z*: calcd for C~6~H~7~NO~2~, 125.0477; found, 125.0484.
**5-Hydroxy-2-octyloxy-4,6-dimethylpyrimidine (7b).** *O*-Octylisouronium trifluoromethanesulfonate (5.1 g, 15.8 mmol) was dissolved in dry DMF (30 mL), and 3-acetoxy-2,4-pentanedione (2.5 g, 15.8 mmol) was added along with sodium acetate (1.14 g, 15.8 mmol), and the mixture stirred for 24 hours at 70 °C. Water (200 mL) was then added, the pH adjusted to \~5, and the organics extracted with EtOAc (3 × 100 mL). The organic layers were combined, dried over MgSO~4~ and concentrated under reduced pressure. The product was then recrystallized from CH~3~CN to yield 35% **7b**. ^1^H NMR (CDCl~3~) δ 4.15 (t, *J* = 6.6 Hz, 2H), 2.32 (s, 6H), 1.63 (m, 2H), 1.32 (br m, 2H), 1.17 (m, 8H), 0.78 (t, 6.5 Hz); ^13^C NMR (CDCl~3~) δ 14.1, 18.8, 22.6, 26.0, 29.0, 29.2, 29.3, 31.8, 67.4, 142.5, 156.4, 158.3; HRMS (EI^+^) *m*/*z*: calcd for C~14~H~24~N~2~O~2~, 252.1838; found, 252.1836.
**5-Hydroxy-2-(2,5-dimethyl-1** ***H*** **-pyrrol-1-yl)-4,6-dimethylpyrimidine (8).** A solution of 5-benzyloxy-2-(2,5-dimethyl-1*H*-pyrrol-1-yl)-4,6-dimethylpyrimidine \[[@R41]\] in MeOH was treated with 10% Pd/C and the resulting black suspension was stirred at room temperature under an atmosphere of H~2~ (1 atm) overnight. The catalyst was removed by filtration through a pad of celite and the filtrate was concentrated under reduced pressure. The crude residue obtained was subjected to flash chromatography on silica gel (eluent: ethyl acetate/hexanes) and the product isolated in quantitative yield. ^1^H NMR (CDCl~3~) δ 2.15 (s, 6H), 2.43 (s, 6H), 5.71 (s, 2H); ^13^C NMR (CDCl~3~) δ 13.2, 18.4, 107.4, 128.6, 145.7, 149.3, 155.3; HRMS (EI^+^) *m*/*z*: calcd for C~12~H~15~N~3~O, 217.1215; found, 217.1217.
**Autoxidation studies** ***.*** The chain-breaking antioxidant activity of the title compounds was evaluated by monitoring the course of thermally initiated inhibited autoxidations of either styrene or cumene (RH) in chlorobenzene or acetonitrile. The autoxidation experiments were performed in a oxygen-uptake apparatus already described elsewhere \[[@R42]--[@R44]\]. In a typical experiment, an air-saturated mixture of styrene or cumene in acetonitrile or chlorobenzene (50% v/v) containing AIBN (1--5 × 10^−2^ M) was equilibrated with the reference solution containing also an excess of PMHC (1 × 10^−2^ M) in the same solvent at 30 °C. After equilibration, a concentrated solution of the antioxidant (final concentration 1--10 × 10^−6^ M) was injected into both the sample flasks, and the oxygen consumption of the sample was measured. From the rate of oxygen consumption during the inhibited period (*R* ~inh~), *k* ~inh~ values were obtained by using [Eq. 6](#FD6){ref-type="disp-formula"} \[[@R44]\], where *R* ~0~ is the rate of oxygen consumption in the absence of antioxidants, *R* ~i~ is the initiation rate (in the range 2--10 × 10^−9^ Ms^−1^), 2*k* ~t~ is the bimolecular termination rate constant of styrylperoxyl or cumylperoxyl radicals (4.2 × 10^7^ and 4.6 ×10^4^ M^−1^s^−1^ respectively) \[[@R21],[@R43]\] and *n* is the stoichiometric coefficient of the antioxidant. The *n* coefficient was determined experimentally from the length of the inhibited period (τ) by [Eq. 7](#FD7){ref-type="disp-formula"}.
A similar procedure was employed to investigate the kinetics of the antioxidant mixtures. The efficiency, α, was determined from the oxygen uptake plots by the extension of the inhibition period according to [Eq. 4](#FD4){ref-type="disp-formula"}. In cases where no clear inhibition period was observed, α was obtained by fitting the experimental traces with numerical simulations based on [Scheme 3](#C3){ref-type="fig"} using Gepasi 3.0 software, as previously described \[[@R45]\].
**FTIR spectroscopy** ***.*** Spectra were recorded at 298 K in a Nicolet Protegé 460 FTIR spectrometer under nitrogen atmosphere using a sealed KBr cell with optical path of 0.5 mm. Solutions of the test compound (10 mM) in CCl~4~ and in CCl~4~/HBA-solvent mixtures were analyzed in absorbance mode and the blank spectrum of the corresponding solvent mixture was subtracted. The signal in the "free" O--H stretching region at ca. 3610 cm^−1^ was manually integrated after manual baseline correction and plotted versus the concentration of the HBA solvent and fit to [Eq. 2](#FD2){ref-type="disp-formula"} \[[@R25]\]. In the case of compound **6b**, similar analysis was repeated using IR peak height in place of peak area and essentially indistinguishable results were obtained. In order to confirm the absence of self-association of the test compounds and to calibrate the spectrometer response, linear regression plots (Absorbance versus \[ArOH\]) in CCl~4~ were preliminarily recorded in the range 1--10 mM. Deviation from linearity was observed only in the case of **7b**, allowing the determination of its self-association equilibrium constant as *K* ~self~ = 121 ± 10 M^−1^. Therefore, its H-bonding to the solvent was analyzed as described above using [Eq. 8](#FD8){ref-type="disp-formula"} (see [Supporting Information File 1](#SD1){ref-type="supplementary-material"} for further details).
Supporting Information
======================
######
Additional experimental details, oxygen-uptake plots and FTIR spectra, as well as cartesian coordinates for calculated structures.
This work was supported by grants from the Italian MIUR (PRIN 2010-2011 2010PFLRJR (PROxi project)) and the Natural Sciences and Engineering Research Council (NSERC) of Canada to L.V. and D.A.P., respectively. D.A.P. also acknowledges the support of the Canada Research Chairs program. The computational efforts in this work were made possible by generous access to the High Performance Computing Virtual Laboratory, a supercomputing facility funded by the Government of Ontario, the Canada Foundation for Innovation and NSERC Canada.
| {
"pile_set_name": "PubMed Central"
} |
{#sp1 .609}
| {
"pile_set_name": "PubMed Central"
} |
Subsets and Splits